# ANTIMICROBIAL RESISTANCE AS A GLOBAL PUBLIC HEALTH PROBLEM: HOW CAN WE ADDRESS IT?

EDITED BY : Ilana L. B. C. Camargo, Leonardo Neves de Andrade, Thierry Naas, Luciene Andrade Da Rocha Minarini, Raffaele Zarrilli, Eliana De Gregorio and Filipa Grosso PUBLISHED IN : Frontiers in Microbiology, Frontiers in Public Health and Frontiers in Medicine

#### 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 978-2-88966-284-5 DOI 10.3389/978-2-88966-284-5

## About Frontiers

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

## Frontiers Journal Series

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

## Dedication to Quality

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

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

### What are Frontiers Research Topics?

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

# ANTIMICROBIAL RESISTANCE AS A GLOBAL PUBLIC HEALTH PROBLEM: HOW CAN WE ADDRESS IT?

Topic Editors:

Ilana L. B. C. Camargo, University of Sao Paulo, Brazil Leonardo Neves de Andrade, University of São Paulo, Brazil Thierry Naas, Assistance Publique Hopitaux De Paris, France Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil Raffaele Zarrilli, University of Naples Federico II, Italy Eliana De Gregorio, University of Naples Federico II, Italy Filipa Grosso, University of Porto, Portugal

Citation: Camargo, I. L. B. C., De Andrade, L. N., Naas, T., Da Rocha Minarini, L. A., Zarrilli, R., De Gregorio, E., Grosso, F., eds. (2020). Antimicrobial Resistance As a Global Public Health Problem: How Can We Address It?. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-284-5

# Table of Contents

## *08 Editorial: Antimicrobial Resistance as a Global Public Health Problem: How Can We Address It?*

Luciene Andrade Da Rocha Minarini, Leonardo Neves de Andrade, Eliana De Gregorio, Filipa Grosso, Thierry Naas, Raffaele Zarrilli and Ilana L. B. C. Camargo

*14 Counter Clinical Prognoses of Patients With Bloodstream Infections Between Causative* Acinetobacter baumannii *Clones ST191 and ST451 Belonging to the International Clonal Lineage II*

Eun-Jeong Yoon, Dokyun Kim, Hyukmin Lee, Hye Sun Lee, Jong Hee Shin, Young Uh, Kyeong Seob Shin, Young Ah Kim, Yoon Soo Park, Jeong Hwan Shin and Seok Hoon Jeong

*23 Synergistic Microbicidal Effect of Auranofin and Antibiotics Against Planktonic and Biofilm-Encased* S. aureus *and* E. faecalis

Pengfei She, Linying Zhou, Shijia Li, Yiqing Liu, Lanlan Xu, Lihua Chen, Zhen Luo and Yong Wu

*35 Genomic Characterization of Prevalent* mcr-1, mcr-4, *and* mcr-5 Escherichia coli *Within Swine Enteric Colibacillosis in Spain* Isidro García-Meniño, Dafne Díaz-Jiménez, Vanesa García, María de Toro,

Saskia C. Flament-Simon, Jorge Blanco and Azucena Mora

*49* In vitro *Activity and Heteroresistance of Omadacycline Against Clinical*  Staphylococcus aureus *Isolates From China Reveal the Impact of Omadacycline Susceptibility by Branched-Chain Amino Acid Transport System II Carrier Protein, Na/Pi Cotransporter Family Protein, and Fibronectin-Binding Protein*

Bing Bai, Zhiwei Lin, Zhangya Pu, Guangjian Xu, Fan Zhang, Zhong Chen, Xiang Sun, Jinxin Zheng, Peiyu Li, Qiwen Deng and Zhijian Yu

*61 Increasing Frequencies of Antibiotic Resistant Non-typhoidal* Salmonella *Infections in Michigan and Risk Factors for Disease* Sanjana Mukherjee, Chase M. Anderson, Rebekah E. Mosci, Duane W. Newton,

Paul Lephart, Hossein Salimnia, Walid Khalife, James. T. Rudrik and Shannon D. Manning

*73 Constructing and Characterizing Bacteriophage Libraries for Phage Therapy of Human Infections*

Shelley B. Gibson, Sabrina I. Green, Carmen Gu Liu, Keiko C. Salazar, Justin R. Clark, Austen L. Terwilliger, Heidi B. Kaplan, Anthony W. Maresso, Barbara W. Trautner and Robert F. Ramig

*90 Characterization of Lytic Bacteriophages Infecting Multidrug-Resistant Shiga Toxigenic Atypical* Escherichia coli *O177 Strains Isolated From Cattle Feces*

Peter Kotsoana Montso, Victor Mlambo and Collins Njie Ateba

*103 Is Meropenem as a Monotherapy Truly Incompetent for Meropenem-Nonsusceptible Bacterial Strains? A Pharmacokinetic/Pharmacodynamic Modeling With Monte Carlo Simulation* Xiangqing Song, Yi Wu, Lizhi Cao, Dunwu Yao and Minghui Long

## *118 Multiple* Klebsiella pneumoniae *KPC Clones Contribute to an Extended Hospital Outbreak*

Carolina Ferrari, Marta Corbella, Stefano Gaiarsa, Francesco Comandatore, Erika Scaltriti, Claudio Bandi, Patrizia Cambieri, Piero Marone and Davide Sassera

*129 Over-Expression of IS*Aba1*-Linked Intrinsic and Exogenously Acquired OXA Type Carbapenem-Hydrolyzing-Class D-ß-Lactamase-Encoding Genes is Key Mechanism Underlying Carbapenem Resistance in*  Acinetobacter baumannii

Marcus Ho-yin Wong, Bill Kwan-wai Chan, Edward Wai-chi Chan and Sheng Chen

*138 Antibiotic Susceptibility, Virulence Pattern, and Typing of* Staphylococcus aureus *Strains Isolated From Variety of Infections in India*

Shifu Aggarwal, Smrutiti Jena, Sasmita Panda, Savitri Sharma, Benu Dhawan, Gopal Nath, N. P. Singh, Kinshuk Chandra Nayak and Durg Vijai Singh


Divakar Sharma, Anjali Garg, Manish Kumar, Faraz Rashid and Asad U. Khan


Elodie Olivares, Stéphanie Badel-Berchoux, Christian Provot, Gilles Prévost, Thierry Bernardi and François Jehl

*211 Colistin Resistance Gene* mcr-1 *Mediates Cell Permeability and Resistance to Hydrophobic Antibiotics*

Baiyuan Li, Fang Yin, Xuanyu Zhao, Yunxue Guo, Weiquan Wang, Pengxia Wang, Honghui Zhu, Yeshi Yin and Xiaoxue Wang

*218 CRISPR-Cas System in Antibiotic Resistance Plasmids in* Klebsiella pneumoniae

Muhammad Kamruzzaman and Jonathan R. Iredell

*231 Linezolid and Rifampicin Combination to Combat* cfr*-Positive Multidrug-Resistant MRSA in Murine Models of Bacteremia and Skin and Skin Structure Infection*

Yu-Feng Zhou, Liang Li, Meng-Ting Tao, Jian Sun, Xiao-Ping Liao, Ya-Hong Liu and Yan Q. Xiong


Zhichang Liu, Dun Deng, Huijie Lu, Jian Sun, Luchao Lv, Shuhong Li, Guanghui Peng, Xianyong Ma, Jiazhou Li, Zhenming Li, Ting Rong and Gang Wang


Yufei Lin, Xiaohong Dong, Jiao Wu, Dawei Rao, Lihua Zhang, Yousef Faraj and Kun Yang

*355 Molecular Analysis of Selected Resistance Determinants in Diarrheal Fecal Samples Collected From Kolkata, India Reveals an Abundance of Resistance Genes and the Potential Role of the Microbiota in Its Dissemination*

Rituparna De, Asish Kumar Mukhopadhyay and Shanta Dutta


Alexandra Vasilakopoulou, Polyxeni Karakosta, Sophia Vourli, Aikaterini Tarpatzi, Paraskevi Varda, Maria Kostoula, Anastasia Antoniadou and Spyros Pournaras

*386 Inactivation of Glutamine Synthetase-Coding Gene* glnA *Increases Susceptibility to Quinolones Through Increasing Outer Membrane Protein F in* Salmonella enterica *Serovar Typhi*

Ana R. Millanao, Aracely Y. Mora, Claudia P. Saavedra, Nicolás A. Villagra, Guido C. Mora and Alejandro A. Hidalgo

*396 Confronting Tigecycline-Resistant* Acinetobacter baumannii *via Immunization Against Conserved Resistance Determinants* Ming-Hsien Chiang, Ya-Sung Yang, Jun-Ren Sun, Yung-Chih Wang,

Shu-Chen Kuo, Yi-Tzu Lee, Yi-Ping Chuang and Te-Li Chen *409 Prevalence, Characterization, and Drug Resistance of* Staphylococcus

Aureus *in Feces From Pediatric Patients in Guangzhou, China* Xiaolan Ai, Fei Gao, Shuwen Yao, Bingshao Liang, Jialiang Mai, Zhile Xiong, Xiantang Chen, Zhuwei Liang, Hongling Yang, Zhiying Ou, Sitang Gong, Yan Long and Zhenwen Zhou


Thaysa Leite Tagliaferri, Natália Rocha Guimarães, Marcella de Paula Martins Pereira, Liza Figueiredo Felicori Vilela, Hans-Peter Horz, Simone Gonçalves dos Santos and Tiago Antônio de Oliveira Mendes

*457 Characterization of KPC-Producing* Serratia marcescens *in an Intensive Care Unit of a Brazilian Tertiary Hospital*

Roumayne L. Ferreira, Graziela S. Rezende, Marcelo Silva Folhas Damas, Mariana Oliveira-Silva, André Pitondo-Silva, Márcia C. A. Brito, Eduardo Leonardecz, Fabiana R. de Góes, Emeline Boni Campanini, Iran Malavazi, Anderson F. da Cunha and Maria-Cristina da Silva Pranchevicius

*471 Prediction of Antimicrobial Resistance in Gram-Negative Bacteria From Whole-Genome Sequencing Data*

Pieter-Jan Van Camp, David B. Haslam and Aleksey Porollo


Zahid Hayat Mahmud, Mir Himayet Kabir, Sobur Ali, M. Moniruzzaman, Khan Mohammad Imran, Tanvir Noor Nafiz, Md. Shafiqul Islam, Arif Hussain, Syed Adnan Ibna Hakim, Martin Worth, Dilruba Ahmed, Dara Johnston and Niyaz Ahmed

*533 Molecular Mechanisms and Epidemiology of Fosfomycin Resistance in*  Staphylococcus aureus *Isolated From Patients at a Teaching Hospital in China*

Wenya Xu, Tao Chen, Huihui Wang, Weiliang Zeng, Qing Wu, Kaihang Yu, Ye Xu, Xiucai Zhang and Tieli Zhou

*543 Alternative Therapeutic Options to Antibiotics for the Treatment of Urinary Tract Infections*

Paul Loubet, Jérémy Ranfaing, Aurélien Dinh, Catherine Dunyach-Remy, Louis Bernard, Franck Bruyère, Jean-Philippe Lavigne and Albert Sotto

*561 Coexistence of the Oxazolidinone Resistance–Associated Genes* cfr *and*  optrA *in* Enterococcus faecalis *From a Healthy Piglet in Brazil*

Lara M. Almeida, Anthony Gaca, Paulo M. Bispo, François Lebreton, Jose T. Saavedra, Rafael A. Silva, Irinaldo D. Basílio-Júnior, Felipe M. Zorzi, Pedro H. Filsner, Andrea M. Moreno and Michael S. Gilmore

# Editorial: Antimicrobial Resistance as a Global Public Health Problem: How Can We Address It?

Luciene Andrade Da Rocha Minarini <sup>1</sup> , Leonardo Neves de Andrade<sup>2</sup> , Eliana De Gregorio<sup>3</sup> , Filipa Grosso<sup>4</sup> , Thierry Naas <sup>5</sup> , Raffaele Zarrilli <sup>6</sup> and Ilana L. B. C. Camargo<sup>7</sup> \*

<sup>1</sup> Department of Pharmaceutical Sciences, Federal University of São Paulo, Diadema, Brazil, <sup>2</sup> School of Pharmaceutical Sciences of Ribeirão Preto, University of São Paulo, Ribeirão Preto, Brazil, <sup>3</sup> Department of Molecular Medicine and Medical Biotechnology, University of Naples Federico II, Naples, Italy, <sup>4</sup> Faculty of Pharmacy, University of Porto, Porto, Portugal, <sup>5</sup> Assistance Publique Hopitaux De Paris, Paris, France, <sup>6</sup> Department of Public Health, University of Naples Federico II, Naples, Italy, <sup>7</sup> Laboratory of Molecular Epidemiology and Microbiology (LEMiMo), São Carlos Institute of Physics, University of São Paulo, São Carlos, Brazil

Keywords: antimicrobial resistance, CRISPR-Cas, oxazolidinone-resistant *Enterococcus faecalis*, carbapenem-resistant *A. baumannii*, colistin-resistant *E. coli*, antibiotic-resistant non-typhoidal Salmonella, MRSA, VRE

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

#### *Edited by:*

Marc Jean Struelens, Université Libre de Bruxelles, Belgium

> *Reviewed by:* Sergey Eremin, World Health Organization, Switzerland

*\*Correspondence:* Ilana L. B. C. Camargo ilanacamargo@ifsc.usp.br

#### *Specialty section:*

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

*Received:* 30 September 2020 *Accepted:* 21 October 2020 *Published:* 12 November 2020

#### *Citation:*

Minarini LADR, Andrade LNd, De Gregorio E, Grosso F, Naas T, Zarrilli R and Camargo ILBC (2020) Editorial: Antimicrobial Resistance as a Global Public Health Problem: How Can We Address It? Front. Public Health 8:612844. doi: 10.3389/fpubh.2020.612844

#### **Antimicrobial Resistance as a Global Public Health Problem: How Can We Address It?**

Misuse of antibiotics in agriculture, food production, and especially among humans and animals is the predominant factor in the emergence and spread of antimicrobial resistance, which might further lead to deaths from infections worldwide (1). More specifically, in the European Union, attributable deaths due to antimicrobial-resistant microorganisms were estimated to be 33,110 per year (2). At the same time, it is now easier to isolate and characterize antimicrobial-resistant bacteria in clinical settings or the environment (1). In 2017, the WHO described the most critical multidrug-resistant bacteria for which novel therapeutics are urgently needed (3). Without surprise, they belonged to the already known group, ESKAPE (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.) (4), which causes most of the healthcare-associated infections nowadays.

On the top of the WHO list are the critical priorities: carbapenem-resistant A. baumannii, P. aeruginosa, and Enterobacteriaceae, plus ESBL-producing Enterobacteriaceae (3). As high priorities are vancomycin-resistant E. faecium, Staphylococcus aureus (methicillin-resistant, vancomycinintermediate and resistant), clarithromycin-resistant Helicobacter pylori, fluoroquinolone-resistant Campylobacter spp., fluoroquinolone-resistant Salmonellae, Neisseria gonorrhoeae (cephalosporinresistant, fluoroquinolone-resistant) (3). The third part of the WHO list ranked as medium priority consists of penicillin-non-susceptible Streptococcus pneumoniae, ampicillin-resistant Haemophilus influenzae, and fluoroquinolone-resistant Shigella spp. (3). In addition to the hospital (patients and healthcare workers) (5–8), bacteria of this list can also disseminate through the community (causing infections in humans or pets) (9–11), food chain (10, 12–14), and environment (15, 16).

Overuse of antibiotics prescribed to humans and animals plus the improper use in agriculture can select resistant bacteria that emerge after a spontaneous mutation or after acquiring resistance genes through horizontal transfer mediated by mobile genetic elements (17). This selection can occur in both the human and non-human gastrointestinal tracts as well as in contaminated environments. Once they become resistant to antibiotics, clonal bacterial cells are selected in the microbial population and disseminates its daughter cells. Alternatively, plasmids carrying resistance genes are transferred to other bacteria through conjugation or transformation (17). In Switzerland, a recent study on freshwater samples from rivers, inland canals, and streams showed environmental dissemination of high-risk carbapenemase-producing Enterobacteriaceae with plasmids composed of elements identical to those of resistance plasmids retrieved from clinical and veterinary isolates locally and worldwide (15). The authors concluded that these plasmids carrying carbapenemase genes replicate and evolve pollutants of river ecosystems and represent a threat to public health and environmental integrity (15). ESBL-producing Enterobacteriaceae were isolated from a Tunisian semi-industrial pilot plant with biological treatment (WWPP) and its receiving river with high rates of CTX-M-15 and high genetic diversity (16). Therefore, the water treatment and appropriate use of antibiotics may represent important infection control measures to avoid antimicrobial resistance spread in hospitals and communities (1, 18).

Bacteria may grow planktonically but most often form biofilms in nature or human body infections (19). These biofilms are common in the hospital environment, especially in polymeric clinical devices, such as catheters and cardiac pacemakers. The inhibition of biofilm growth is of great importance in infection control and treatment of healthcare-associated infections owing to their inherent tolerance and "resistance" to antimicrobial therapies (19). In addition, biofilms constitute a great challenge in clinical settings since there is no "gold standard" available to reveal the presence of microbial biofilms.

The prevention of bacterial resistance spread is a significant challenge among healthcare professionals, especially in hospital settings. Surveillance of clones and lineages spread in a hospital environment or among patients, and the knowledge of their susceptibility to antimicrobials are crucial data to initiate a proper empiric treatment of hospital-acquired infections and infection prevention and control.

In many cases, precision medicine is the key to therapy since different microbes are prone to different remedies. New antimicrobials might seem more effective and promising, but some traditional compounds, such as beta-lactamase or efflux pump inhibitors that could prevent enzymatic action on known antibiotics or efflux pump activity, respectively, are also beneficial.

This Research Topic harbors 48 published manuscripts, including original research, review, opinion, and brief research reports addressing these subjects. The 443 involved authors focused on tackling antimicrobial resistance in bacteria. According to Merlin, although the increasing occurrence of antibiotic resistance among bacteria due to the antimicrobial agent consumption has been recognized for decades, only recently, the seriousness of the situation has been considered by international, national, and local health agencies. Some recommendations were defined in order to educate and improve the practices of health professionals and consumers. Most recommendations proposed coordinated actions in public health and veterinary/agricultural domains, limiting the inappropriate exposure of bacteria to antibiotics, in order to delay the natural evolution toward resistance and its propagation in the downstream environment, in a One Health context.

The chapters of this Research Topic show that it is crucial to pay attention to antimicrobial resistance in microorganisms from wild animals, pets, livestock, food, and the environment, in addition to those isolated from humans, to understand how antimicrobial resistance spreads. Lin et al. used a genomic approach to study the horizontal transfer of colistin resistance between polluted rivers and wild birds. Metadata analysis of mcr-1- positive multidrug-resistant E. coli strains, which were isolated from the environment of egret habitat (polluted river) and egret feces, showed highly homologous mcr-1-bearing IncI2 plasmids among isolates from different regions along the East Asian-Australian Flyway, suggesting that migratory birds may mediate the intercontinental transportation of colistin resistance.

Mahmud et al. call the attention to the prevalence and characterization of MDR and ESBL-producing E. coli in drinking water samples collected from Rohingya camps, Bangladesh. The authors described that 66 out of 384 E. coli isolates in this study were ESBL-producers, of which 71% (47/66) were MDR. Sixtyfour percent (42/66) of the ESBL-producing E. coli carried one to seven plasmids ranging from 1 to 103 MDa. Only large plasmids with antibiotic resistance properties were found transferrable via conjugation. This study shows that drinking water samples could foster exposure and dissemination of MDR, ESBL-producing, and pathogenic E. coli lineages, posing a health risk to the people residing in the densely populated camps of Bangladesh.

Chen et al. investigated the antimicrobial resistance trends and characteristics of ESBL-producing E. coli isolates from pets and whether this correlates with antibiotic usage in the clinic. The authors observed high rates of resistance to β-lactams and fluoroquinolones and resistance to >3 antibiotic classes. ESBL-encoding genes blaCTX−<sup>M</sup> (n = 44, 34.65%) and plasmidmediated quinolone resistance genes qnrB (n = 119, 93.70%) were most common in 36 PFGEs types and 28 different sequence types (STs), including ST405 (7, 15.9%), ST131 (3, 6.8%), ST73, ST101, ST372, and ST827 (2, 4.5% each). Additionally, blaNDM−<sup>5</sup> was detected in three isolates (n = 3, 2.36%, two ST101 and one ST405), and mcr-1 was also observed in three colistin-resistant E. coli (ST6316, ST405, and ST46).

García-Meniño et al. performed molecular characterization of 35 prevalent strains of colistin-resistant E. coli from swine enteric colibacillosis in Spain. The resistome analysis showed six different mcr variant genes. In seven of them, the authors observed an association with ESBL. The summative presence of mechanisms associated with high-level resistance to quinolone/fluoroquinolones and colistin could confer adaptive advantages to prevalent pig E. coli strains. The authors also reported the co-occurrence of double colistin-resistance mechanisms in a significant number of E. coli isolates.

Mukherjee et al. observed increasing frequencies of antibioticresistant non-typhoidal Salmonella (NTS) infections in Michigan, which is not part of the FoodNet surveillance system in the USA. A total of 198 isolates, from 2011 to 2014, belonged to 35 different serovars with the predominance of Enteritidis (36.9%), followed by Typhimurium (19.5%), and Newport (9.7%). A total of 30 (15.2%) NTS isolates were resistant to ≥1 antibiotic, and 15 (7.5%) were resistant to ≥3 antimicrobial classes. They demonstrated the importance of surveillance, resistance frequency monitoring, and identification of risk factors that can aid in the development of new prevention strategies.

Almeida et al. described the coexistence of oxazolidinone resistance genes cfr and optrA in a MDR E. faecalis isolate belonging to ST29, obtained from a healthy piglet in Brazil in the context of surveillance study. Whole-genome sequencing revealed that the cfr gene was in a transposon-like structure of 7,759 nucleotides flanked by IS1216E and capable of excising and circularizing, distinguishing it from known genetic contexts for cfr in Enterococcus spp., while optrA was in an Inc18 broad-hostrange plasmid of >58 kb. Both genes were able to conjugate. This study highlights the need for monitoring the use of antibiotics in the Brazilian swine production system to control the spread and proliferation of antibiotic resistance.

This Research Topic also presents studies reporting the mechanisms of resistance and lineages of bacteria of clinical importance. Carbapenem-resistant A. baumannii (CRAB) often causes fatal infections among seriously ill patients. Wong et al. used whole-genome sequencing, gene expression studies, and enzyme kinetics analyses to investigate the underlying carbapenem resistance mechanisms in 14 clinical A. baumannii. A large majority of isolates belonged to the International Clone II (IC-II), ST208. CRAB harbored blaOXA−23/blaOXA−72, or overexpression of the blaOXA−<sup>51</sup> gene upon ISAba1 insertion, and had strong carbapenem-degrading activities. Over-production of carbapenem-hydrolyzing-class D-ß-lactamases (CHDLs) is the key mechanism of carbapenem resistance in the isolates studied.

Fosfomycin exhibits significant antimicrobial activity against a broad spectrum of pathogens, including S. aureus. Although widely described in gram-negative bacteria, fosfomycin's resistance mechanism is rarely reported among gram-positive bacteria (Xu et al.). The molecular mechanisms of 11 fosfomycinresistant S. aureus isolates from a Teaching hospital in China were determined. The authors showed that the mutations in the uhpT, glpT, and murA genes and the overexpression of the efflux pump gene tet38 under a subinhibitory concentration of fosfomycin might play a major role in conferring fosfomycin resistance in these isolates. In contrast, none of the resistant strains carried the fosA, fosB, fosC, fosD, and fosX genes, indicating that these genes might not be the primary factors mediating the resistance of S. aureus against fosfomycin. Furthermore, MLST analysis identified ST5708, a novel sequence type of S. aureus (Xu et al.).

The possible transference of resistance genes among different species and the consequences that these genes may cause in the cells were covered in this Research Topic, for instance, in the study by Xiang et al., which described the presence of blaNDM−<sup>1</sup> in 54 out of 1,735 carbapenem-non-susceptible strains, including K. pneumoniae, A. baumannii, and E. coli. After analyzing the blaNDM−<sup>1</sup> gene and its genetic environment, the authors hypothesized that, due to the presence of the ble and tnpA genes, blaNDM−<sup>1</sup> originates from A. baumannii, which is retained in K. pneumoniae over a long period by transposition of mobile elements.

Mobile genetic elements harboring resistance genes, in certain cases, may reduce bacterial fitness. In this Research Topic, Li et al. investigated the physiological function of the mcr-1 gene, which encodes an LPS-modifying enzyme, in E. coli K-12. The removal of the mcr-1 gene not only reduced resistance to colistin, but also increased cell viability under high osmotic stress conditions and led to increased resistance to hydrophobic antibiotics. Increased expression of mcr-1 also resulted in a decreased growth rate and changed the cellular morphology of E. coli (Li et al.).

Sharma et al. used a proteomic approach to study the growth and fitness of a carbapenem-resistant NDM-4-producing K. pneumoniae clinical isolate under meropenem stress. The authors showed the downregulation of flagellar, fimbriae, and pili proteins by mass spectrometry and system biology analysis of networking pathways. Sharma et al. suggested that these proteins might be used as targets for the development of novel therapeutics against antimicrobial-resistant K. pneumoniae infections.

This Research Topic also brings the search and analysis of the CRISPR-Cas system, a microbial adaptive immune system involved in defense against different types of mobile genetic elements, in whole plasmid and chromosomal sequences of multidrug-resistant K. pneumoniae from GenBank (Kamruzzaman and Iredell). The authors found a putative CRISPR-Cas system in 44 plasmids from Klebsiella species and identified an identical system in three plasmids from other Enterobacteriaceae, with CRISPR spacers targeting different plasmids and chromosome sequences. Additionally, the authors showed that plasmid type IV CRISPR may depend on the chromosomal type I-E CRISPRs for their competence (Kamruzzaman and Iredell). Different aspects are involved in the acquisition and transference of resistance determinants, and the presence of CRISPR-Cas in plasmids just showed us we still have a lot to understand.

In this Research Topic, several manuscripts described the diversity of the most important pathogenic bacterial species and their susceptibility profiles. For instance, in India, the antibiotic susceptibility and typing of 109 S. aureus strains isolated from a variety of infections showed that all S. aureus isolates were multidrug-resistant, virulent, and diverse irrespective of sources and place of isolation being assigned to 46 spa-types, 33 STs and eight clonal complexes (Aggarwal et al.).

Ai et al. studied the prevalence, characterization, and drug resistance of S. aureus fecal carriage among pediatric patient feces in Southern China and found that the fecal carriage rates were 20.0% for S. aureus and 4.5% for methicillin-resistant S. aureus (MRSA). Among the 76 STs, the authors found 25 new STs, and the most prevalent were ST188, ST6, and ST15 for methicillin-sensitive S. aureus (MSSA) and ST59 and ST45 for MRSA. The authors described high rates of penicillin, erythromycin, and clindamycin resistance among the isolates. However, CC59 (ST338 and ST59) and CC45 exhibited different antibiotic resistance patterns. Ai et al. indicated an urgency for strengthening the surveillance programs in China that could contribute to S. aureus infection prevention and treatment.

Bispo et al. analyzed the population structure of ocular MRSA isolates using a combination of MLST analysis, SCCmec typing, and detection of the panton-valentine leukocidin (PVL) coding gene. The 68 isolates (2014–2016) belonged to 14 different sequence types (STs) that grouped within two predominant clonal complexes: CC8 (47.0%) and CC5 (41.2%). Although the isolates were associated with lineages with community and hospital origins, the authors suggested that the niche specialization could act as a driver for the community structure observed.

In Iran, Goudarzi et al. conducted a study from August 2013 to July 2018 to investigate the prevalence and molecular epidemiology of inducible macrolide-lincosamide streptogramin B (MLSB) resistance in S. aureus. The prevalence of inducible MLSB (iMLSB) S. aureus increased from 7.5 to 21.7% during the study period. The authors observed a replacement of the CC22, predominant in 2013–2014, by CC8 in 2017–2018. Goudarzi et al. described for the first time the temporal shifts of iMLSB S. aureus isolates in Iran that identify predominant clones and treatment options for iMLSB S. aureus–related infections.

Vasilakopoulou et al. described the prevalence and risk factors of antimicrobial resistance by studying the gastrointestinal carriage of vancomycin-resistant enterococci (VRE) and carbapenem-resistant gram-negative bacteria (CRGN) in the fecal flora of the inpatients of a tertiary university hospital in Greece. The authors found high prevalence rates for VRE and CRGN carriage. Prolonged hospitalization and age were independent risk factors for VRE carriage, while CRGN carriage was associated with an increased risk of acquiring a resistant pathogen, prolonged hospital stay, and increased mortality.

In South Korea, Yoon et al. showed that most (84.5%) of the A. baumannii blood isolates belonged to the international clonal lineage II (ICLII), and 89.5% of the isolates were either multidrug- or extensively drug-resistant. The patient's old age, the sequential organ failure assessment score, and causative A. baumannii ST191 belonging to ICLII were risk factors for 30-day mortality (Yoon et al.).

Ferrari et al. identified the circulation of three distinct genome clusters of carbapenem-resistant KPC-producing K. pneumoniae in an intensive care unit in a hospital in Northern Italy. The authors described two clusters, one linked to ST512 and one to ST258; the ST258 isolates were also colistin-resistant and responsible for a high number of clinical infections (Ferrari et al.).

Ferreira et al. characterized KPC-producing Serratia marcescens strains isolated during an outbreak involving patients hospitalized in the intensive care unit (ICU) and neonatal intensive-care-unit (NIUC) of a Brazilian tertiary hospital. All isolates carried KPC-carbapenemase (blaKPC) and extendedspectrum beta-lactamase blaTEM genes. In addition to the MDR profiles exhibited by almost 25% of isolates, 98.2% of the isolates had the genes codifying to the pore-forming toxin (ShlA), phospholipase with hemolytic and cytolytic activities (PhlA), flagellar transcriptional regulator (FlhD), and positive regulator of prodigiosin and serratamolide production (PigP).

Due to increasing resistance to multiple antibiotics classes, the search for molecules displaying potent activity against multidrug-resistant (MDR) bacteria is mandatory. Additionally, the discovery of new therapeutic approaches and/or new pathways to be inhibited is an open door for a directed search for bioactive compounds. As an example, Barra et al. discussed the possibilities, using an in silico-based analysis, of enzyme inhibition of two new possible druggable biosynthesis pathways present in most bacteria, fungi, and plants but incomplete in humans: the thiamine (vitamin B1) and the pyridoxal 5′ phosphate and its vitamers (vitamin B6).

Loubet et al. discussed alternative therapies and preventative strategies for urinary tract infections caused by multidrugresistant uropathogens, focusing on the following phases of the pathogenesis: colonization, adherence of pathogens to uroepithelial cell receptors, and invasion. In this review, the authors discussed vaccines, small compounds, nutraceuticals, immunomodulating agents, probiotics, and bacteriophages, highlighting the challenges each of these approaches face Loubet et al.

Zhang et al. investigated the effects of pharmacological doses of copper, an essential microelement for animals, on the microbial communities in the hindgut, and the antimicrobial resistance profiles of E. coli in weaned piglets. The pharmacological dose of copper affected the composition of the microbial community (affected the microbial metabolic functions of energy metabolism, protein metabolism, and amino acid biosynthesis), increased antimicrobial resistance (to chloramphenicol and ciprofloxacin) of intestinal E. coli, and was most likely harmful to the health of piglets at the early stage after weaning (Zhang et al.).

The new anti-biofilm or antimicrobial agents' analysis, as well as their mode of action, were assessed in some studies. Murugan et al. described that arborine and skimmianine compounds from ethyl acetate extract of Gycosmis pentaphylla harbor a high antibacterial activity against MDR S. aureus. Wang et al. showed that the flavone glycoside baicalin inhibits biofilm formation and the quorum-sensing system in Staphylococcus saprophyticus, thus suggesting that it might be a potential therapeutic agent for S. saprophyticus biofilm-associated infections.

She et al. described the potent antimicrobial activity of auranofin against MRSA/MSSA and E. faecalis, both planktonic cells and biofilms, with minimum inhibitory concentrations ranging from 0.125 to 0.5 mg/L. Auranofin, in combination with linezolid or fosfomycin, showed synergistic antimicrobial activities against MSSA and MRSA both in vitro and in a mouse infection model. There was also a synergistic effect with chloramphenicol against E. faecalis. Additionally, auranofin improved the antibiofilm efficacy of chloramphenicol and linezolid, even on biofilms grown on a catheter surface. Therefore, they showed that the combination of auranofin with linezolid, fosfomycin, and chloramphenicol provides a synergistic microbicidal effect in vitro and in vivo, which rapidly enhances antimicrobial activity.

Lytic bacteriophage therapy is a potential antibacterial therapy. Montso et al. characterized lytic E. coli O177-specific bacteriophages isolated from cattle feces. A total of 31 lytic E. coli O177-specific bacteriophages were isolated and 71% of these phage isolates produced large plaques on 0.3% soft agar. Selected phage isolates had a similar morphotype (an icosahedral head and a contractile tail) and were classified under the order Caudovirales, Myoviridae family. Montso et al. demonstrated that lytic E. coli O177-specific bacteriophages isolated from cattle feces are highly stable and can infect different E. coli strains.

Among the papers, the authors analyzed new antimicrobials and different compounds against MDR bacteria. Cefoselis, a fourth-generation cephalosporin, was evaluated by Cheng et al. against bacterial pathogens in China. The authors observed that cefoselis showed good activity against non-ESBL-producing E. coli, K. pneumoniae, and P. mirabilis, MSSA, and was also potent against Enterobacteriaceae, P. aeruginosa, and Streptococcus.

Omadacycline, a new tetracycline-class broad-spectrum aminomethylcycline, exhibited excellent in vitro activity against both MRSA and MSSA clinical isolates (Bai et al.). Omadacycline susceptibility in S. aureus may be affected by efflux pump proteins (i.e., a branched-chain amino acid transport system II carrier protein and a Na/Pi cotransporter family protein) (Bai et al.).

Zheng et al. compared the effects of radezolid and linezolid on planktonic and biofilm cells of E. faecalis. After analyzing 302 E. faecalis, they observed that the MIC50/MIC90 values of radezolid (0.25/0.50 mg/L) were eight-fold lower than those of linezolid (2/4 mg/L). They observed that the radezolid MICs against the high-level linezolid-resistant isolates (linezolid MICs ≥64 mg/L) increased to ≥4 mg/L with mutations in the four copies of the V domain of the 23S rRNA gene. In addition, they showed the involvement of OG1RF\_12220 (mdlB2, multidrug ABC superfamily ATP-binding cassette transporter) in the alteration of radezolid and linezolid susceptibility. Zheng et al. described that Radezolid inhibited E. faecalis biofilm formation to a greater extent than linezolid. Therefore, radezolid is more effective than linezolid against E. faecalis.

Hao et al. showed a potent in vitro activity of apramycin, an antibiotic with good activity against a range of multidrugresistant pathogens, against clinical carbapenem-resistant, and hypervirulent K. pneumoniae (CR-hvKp) along with carbapenem-resistant non-hvKp (CR-non-hvKp) isolates, including those resistant to amikacin or gentamicin.

Dong et al. demonstrated that ebselen, an organo-selenium with well-characterized toxicology and pharmacology, has high bactericidal activity against multidrug-resistant (MDR) S. aureus based on taking TrxR as a major target and disruption of the

## REFERENCES


redox microenvironment. The authors showed that ebselen is an effective topical antibacterial agent in animal model of MDR S. aureus LT-1 skin infection, reducing the bacterial load and the expression of pro-inflammatory cytokines.

Murugan et al. focused on the isolation and structural characterization of bioactive compounds from the ethyl acetate extract of Gycosmis pentaphylla. The isolated compounds identified as arborine and skimmianine have high bactericidal effects against MDR S. aureus clinical isolates. These compounds induced intracellular molecular imbalance and cell membrane disturbances that caused cell death of all MDR S. aureus strains and the reference standard S. aureus MTCC-96 strain.

A bottleneck in clinical microbiology, the incubation time necessary to the susceptibility test is ready for the antibiotic decision by the clinician is something to overcome. Genomics and technology can work together to improve the turnaround time in the clinical laboratory. Liu et al. used 2 genotype-based machine learning methods named Support Vector Machine (SVM) and Set Covering Machine (SCM) to predict the resistance to tetracycline, ampicillin, sulfisoxazole, trimethoprim, and enrofloxacin. The correlation results between the phenotype and the model predictions of the five antimicrobial agents indicated that both SVM and SCM models could significantly discern the resistant isolates of the sensitive isolates and could be potential tools in antimicrobial resistance surveillance and clinical diagnosis in veterinary medicine.

Resistance mechanisms, surveillance of antimicrobialresistant bacteria, and new drug discovery studies offer different views of the antimicrobial resistance, which became a public health problem. There are more articles dedicated to the community focused on antimicrobial resistance in this Research Topic, and we hope you enjoy it.

## AUTHOR CONTRIBUTIONS

All authors wrote about the manuscripts they edited in the Research Topic. IC also compiled and edited the complete text.

## ACKNOWLEDGMENTS

We are grateful to all reviewers who put their efforts during the COVID-19 pandemic to analyze all the manuscripts.


tertiary hospital. Revista da Sociedade Brasileira de Medicina Tropical. (2016) 49:51–6. doi: 10.1590/0037-8682-0284-2015


Enterobacteriaceae in rivers in Switzerland. Environ Pollut. (2020) 265(Pt B):115081. doi: 10.1016/j.envpol.2020.115081


**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 Minarini, Andrade, De Gregorio, Grosso, Naas, Zarrilli and Camargo. 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.

# Counter Clinical Prognoses of Patients With Bloodstream Infections Between Causative *Acinetobacter baumannii* Clones ST191 and ST451 Belonging to the International Clonal Lineage II

#### *Edited by:*

*Marc Jean Struelens, European Centre for Disease Prevention and Control (ECDC), Sweden*

#### *Reviewed by:*

*Kwan Soo Ko, Sungkyunkwan University School of Medicine, South Korea Raffaele Zarrilli, University of Naples Federico II, Italy*

> *\*Correspondence: Seok Hoon Jeong kscpjsh@yuhs.ac*

#### *Specialty section:*

*This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health*

> *Received: 15 April 2019 Accepted: 02 August 2019 Published: 16 August 2019*

#### *Citation:*

*Yoon E-J, Kim D, Lee H, Lee HS, Shin JH, Uh Y, Shin KS, Kim YA, Park YS, Shin JH and Jeong SH (2019) Counter Clinical Prognoses of Patients With Bloodstream Infections Between Causative Acinetobacter baumannii Clones ST191 and ST451 Belonging to the International Clonal Lineage II. Front. Public Health 7:233. doi: 10.3389/fpubh.2019.00233* Eun-Jeong Yoon<sup>1</sup> , Dokyun Kim<sup>1</sup> , Hyukmin Lee<sup>1</sup> , Hye Sun Lee<sup>2</sup> , Jong Hee Shin<sup>3</sup> , Young Uh<sup>4</sup> , Kyeong Seob Shin<sup>5</sup> , Young Ah Kim<sup>6</sup> , Yoon Soo Park <sup>7</sup> , Jeong Hwan Shin<sup>8</sup> and Seok Hoon Jeong<sup>1</sup> \*

*<sup>1</sup> Department of Laboratory Medicine and Research Institute of Bacterial Resistance, Yonsei University College of Medicine, Seoul, South Korea, <sup>2</sup> Biostatistics Collaboration Unit, Yonsei University College of Medicine, Seoul, South Korea, <sup>3</sup> Department of Laboratory Medicine, Chonnam National University Medical School, Gwangju, South Korea, <sup>4</sup> Department of Laboratory Medicine, Yonsei University Wonju College of Medicine, Wonju-si, South Korea, <sup>5</sup> Department of Laboratory Medicine, Chungbuk National University College of Medicine, Cheongju-si, South Korea, <sup>6</sup> Department of Laboratory Medicine, National Health Insurance Service Ilsan Hospital, Goyang-si, South Korea, <sup>7</sup> Department of Internal Medicine, National Health Insurance Service Ilsan Hospital, Goyang-si, South Korea, <sup>8</sup> Department of Laboratory Medicine and Paik Institute for Clinical Research, Inje University College of Medicine, Busan, South Korea*

This study was conducted to evaluate the possible clinical and bacteriologic features associated with 30-day mortality from *Acinetobacter baumannii (A. baumannii)* bloodstream infections (BSIs). We conducted a prospective, multicenter, observational study of 181 entire episodes of *A. baumannii* BSI from six general hospitals between May 2016 and April 2017 in South Korea. Cox proportional-hazards regression model was used to estimate risks of the primary endpoint, i.e., all-cause mortality within 30 days from the initial blood culture. Most (84.5%) of the *A. baumannii* blood isolates belonged to the international clonal lineage II (ICLII) and 89.5% of the isolates were either multidrug- or extensively-drug resistant. We identified three risk factors including the old age of patient {hazard ratio, 1.033; [95% Confidential Interval (CI), 1.010–1.056]}, the sequential organ failure assessment score [1.133 (1.041–1.233)], and causative *A. baumannii* sequence type (ST) 191 belonging to ICLII [1.918 (1.073–3.430)], and three protective factors including causative *A. baumannii* ST451 belonging to ICLII [0.228 (0.078–0.672)], platelet count [0.996 (0.993–0.999)], and definitive therapy within 72 h [0.255 (0.125–0.519)]. Differing 30-day mortality rate in the dominant ICLII was observed by ST, which was much high in ST191 and low in ST451 and it was likely associated with the molecular traits, rather than the drug resistance.

Keywords: *Acinetobacter baumannii*, bloodstream infection, international clonal lineage II, ST191, ST451

## INTRODUCTION

Acinetobacter baumannii is dominant in clinical settings primarily targeting immunocompromised patients in intensive care units (ICUs) and has the potential to cause healthcareassociated infections (1). Infections caused by A. baumannii have repeatedly been reported to have grave clinical outcomes (2, 3). A. baumannii has an epidemic potential of emerging diseases because of its ability to persist in hospital environments by resisting desiccation and disinfectants (1). A propensity toward rapid acquisition of foreign DNA, including antimicrobial resistance determinants and virulence determinants, also contributes to its effective dissemination (4).

Previous studies have demonstrated that A. baumannii infections are accompanied by serious morbidity and mortality associated with a high sequential organ failure assessment (SOFA) score (5), comorbidities (6), immunosuppression (3), and antimicrobial treatment failure, mainly due to the multidrug resistance (MDR) of the causative A. baumannii (5, 7, 8). The rates of resistance to the major anti-Acinetobacter agents are increasing (9). In particular in South Korea, predominance of the multidrug-resistant international clonal lineage II (ICLII) in hospital settings is of great concern (10–12).

This study was performed to evaluate the possible clinical and bacteriologic features associated with 30-day mortality from A. baumannii bloodstream infection (BSI) for the entire 1-year incidence of A. baumannii BSI cases of multiple general hospitals in South Korea.

## METHODS

## Study Design

A prospective, multicenter, observational study was designed to monitor the entire A. baumannii BSI episodes between May 2016 and April 2017 occurring in six general hospitals with 715–1,050 beds participating in a national surveillance study of antimicrobial resistance in South Korea (13). Local ethics committee approvals were expedited or waived at each participating hospital since the study had a purely observational nature. The A. baumannii BSI cases were detected through reviewing daily blood culture results. One authorized investigator at each hospital reviewed patients' medical records, all of the raw data were recorded in a pre-formatted spreadsheet, and the datasheet was submitted via email to an analysis center without monitoring. Prognoses of patients were followed for at least 30 days and the cases of transfer and hospital discharge were recorded as they were. The Charlson comorbidity index (14) and the SOFA score (15) were calculated in the analysis center using the collected raw data for the day of initial blood culture. For the sequential isolation of A. baumannii from one patient, the first isolate was used for analyses and the ensuing isolates were discarded. Among a total of 67,803 patients subjected to blood culture, A. baumannii isolates were recovered from blood specimens of 188 patients, and following the elimination of 7 episodes of death on the day of the initial blood culture, 181 cases were included in the analysis. The microbiological assessment was carried out in the analysis center.

## Definitions

Laboratory-confirmed A. baumannii BSI was defined as an A. baumannii-positive blood culture from one or more blood specimens (16). Hospital-originated (HO) infection was specified by ≥2 calendar days of hospitalization at the day of initial blood culture, including the time spent at a previous health care facility before transfer. The 30-day mortality was defined as all-cause mortality within 30 days from the initial blood culture. Non-susceptibility (NS) to a drug class was defined as NS to at least one drug in a class and resistance phenotypes were categorized following Magiorakos et al. (17): MDR (NS to three or more drug classes), extensively drug resistance (XDR, NS to at least one agent in all but one or two drug classes), and pan-drug resistance (NS to at least one agent in all drug classes). An empirical antimicrobial therapy was defined as an initial antimicrobial treatment before the susceptibility testing results for the causative pathogen. A definitive antimicrobial therapy was defined as prescribing a revised antimicrobial regimen based on the antimicrobial susceptibility testing results and, if the revised regimen was administered after 72 h, it was deemed to be a delayed definitive therapy. Adequate antimicrobial therapy must include at least one antimicrobial agent active in vitro and both the dosage and route of administration should meet current medical standards.

## Microbiological Assessment

Bacterial species were initially identified by matrix assisted laser desorption/ionization time-of-flight mass spectrometry using a Bruker BiotyperTM system (Billerica, MA, USA) and the detailed species of Acinetobacter spp. were confirmed by PCR and sequencing the rpoB gene. For A. baumannii, multilocus sequence typing was followed through the Oxford scheme (18). Antimicrobial susceptibility was tested following the Clinical and Laboratory Standards Institute (CLSI) guidelines (19). Antimicrobial susceptibilities to 13 anti-Acinetobacter agents belonging to seven drug classes were tested by the disk diffusion method. For carbapenem-resistant isolates, the blaOXA−<sup>23</sup> gene and the ISAba1-blaOXA−51−like were identified by PCR and direct sequencing (20). Colistin MICs were determined by the broth microdilution method (21). Twenty-six virulence genes associated with eight virulence characteristics (22) were investigated by real-time PCR using gene-specific primers (23) and a KAPA SYBR Fast quantitative PCR master mix (Kapa biosystems, Inc., Wilmington, MA, USA). The resistance islands AbaR and AbGRI1 were specified by PCR using a set of primers, comMtr3\_145-126F (5′ - CAAGCGGTTGGCGTAAGACT-3′ ) and reverse primers tniB\_546-525R (5′ -CAAGTACATGCTCTGCAAGATG-3 ′ ) only for the full tniB gene and tniB\_37-16R (5′ - CAGACTCAATTCATTGCTGAGG-3′ ) for both the full and the truncated tniB genes.

## Statistical Analysis

Statistical analyses were performed in SPSS statistics (version 23, IBM Corp., Armonk, NY, USA). Data are expressed as the mean ± standard deviation if no special description is indicated. Categorical variables were compared by Pearson's X 2 test, and continuous variables were compared by an independent two sample t-test. Associations between 30-day mortality and risk factors were evaluated using univariable and multivariable Cox proportional hazard regression models. The proportional hazards assumption was evaluated by including an interaction term between variables, and variables were natural-log transformed by follow-up time and by log-minus-log survival plots. For multivariable analyses, variables with a P value <0.1 in univariable analyses were considered to include any possible variables. Multicollinearity between variables was diagnosed by calculating the variance inflation factors, and a multivariable analysis was conducted by the backward stepwise method. All tests of significance were two-tailed; P values <0.05 were considered to be significant.

## RESULTS

## Characteristics of Enrolled Patients

A total of 181 laboratory-confirmed A. baumannii BSI cases were enrolled from the six participating hospitals and the number of cases was ranged from 22 to 52 by hospital (**Figure 1**). Among patients with A. baumannii BSIs, the sex ratio of male vs. female was 115:66, and their mean age was 67.7 ± 14.3 years (**Table 1**). The mean Charlson comorbidity index score was 4.1 ± 2.2. Solid tumors (26.5%, 48/181), diabetes mellitus (23.2%, n = 42), cerebrovascular diseases (18.2%, n = 33), kidney diseases (13.8%, n = 25), and chronic pulmonary diseases (6.6%, n = 12) were the five most observed underlying diseases and dementia (n = 9) and chronic liver diseases (n = 7) were followed. Of the total, 88.4% (n = 160) of cases were HO infections, 95.0% (n = 172) were inpatients, and 68.0% of those (117/172) stayed in ICUs. The SOFA score had a mean value of 5.9 ± 3.5. Secondary BSIs were identified in 44.2% (80/181) of cases, and the majority (86.3%, 69/80) were originated from a pulmonary tract infection.

## The BSI-Causative *A. baumannii* Isolates Sequence Types

The majority of the A. baumannii blood isolates belonged to ICLII (84.5%, 153/181) and the ICLII mostly comprised sequence type (ST) 191 (allele numbers of gltA-gyrB-gdhB-recA-cpn60-gpirpoD, 1-3-3-2-2-94-3; 47.7%, 73/153), ST784 (1-3-3-2-2-107-3; 20.3%, n = 31), and ST451 (1-3-3-2-2-142-3; 19.0%, n = 29). The 30-day mortality rate of patients with A. baumannii ICLII BSIs was 1.83-fold higher than that with A. baumannii BSIs by non-ICLII strains (45.8 vs. 25.0%); and among the ICLII BSIs, the 30-day mortality rate with ST191 BSIs was the highest and, the lowest for ST451 BSIs (60.3 vs. 17.2%).

## Antimicrobial Resistance

Majority of the A. baumannii blood isolates (89.5%, 162/181) were either MDR or XDR, and the proportion was markedly greater in A. baumannii ICLII (98.7%, 151/153; 18 MDR and 133 XDR) than in non-ICLII (39.3%, 11/28; 5 MDR, and 6 XDR). All of the tested isolates were susceptible to colistin, and no pandrug resistant isolate was observed. All but one carbapenem-NS A. baumannii harbored the blaOXA−<sup>23</sup> gene (99.4%, 159/160) and the remaining one carried an insertion sequence ISAba1 upstream from the blaOXA−51−like gene. Two thirds and onequarter of A. baumannii possessed AbGRI1 (63.0%, 114/181) and AbaR (26.5%, 48/181), respectively.

## Virulence Factors

Factors for adhesion and membrane integrity, biofilm formation, carbohydrate metabolism, quorum sensing, and siderophore


FIGURE 1 | Participating sentinel hospitals and collected *Acinetobacter baumannii* blood isolates. Locations of the hospitals (in gray) and covering parts (in gray patterns) are indicated in a map of the South Korean peninsula, respectively, and the associated table represent information regarding blood isolates from each hospital including demographic information from 2016. HO, hospital-origin; ICLII, international clonal lineage II; ST, sequence type.

TABLE 1 | Thirty day mortality-associated factors for patients with an *Acinetobacter baumannii* BSI.


*(Continued)*

#### TABLE 1 | Continued


\**The percentage was calculated from the total number of AB-BSI.*

*† All-cause mortality within 30 days from the initial blood culture.*

*‡P value is either from Pearson's X<sup>2</sup> test for categorical data or from the t-test for continuous data. Bonferroni corrected significance level from 32 hypotheses was considered to be P* < *0.0016, and the significant P values are indicated in bold.*

*§Cox proportional hazards regression model was applied to calculate the hazard ratio (HR) and the 95% confidence interval (CI).*

*¶Underlying disease and primary site of infection could be multiple by patient.*

*#Ulcer (n* = *2), connective tissue disease (n* = *2), congestive cardiac insufficiency (n* = *2), and leukemia (n* = *2) are included.*

*††Multilocus sequence typing profiles are indicated with the combination of alleles (gltA-gyrB-gdhB-recA-cpn60-gpi-rpoD).*

*SD, standard deviation; ICLII, the international clonal lineage II.*

were detected in A. baumannii blood isolates (**Table 1**). The acinetobactin-associated basD (98.9%, 179/181), the glycokinase gntK (98.9%, n = 179), the adhesion-associated smpA (97.8%, n = 177), and the biofilm formation-associated pga genes (95.6%, n = 173) were harbored by most of the A. baumannii blood isolates, while the capsule protein mviM (3.9%, n = 7), the polysaccharide exporter epsA (5.0%, n = 9), lipopolysaccharide vipA (8.8%, n = 16), and the type IV pili pilA (9.9%, n = 18) genes were occasionally identified. Among the virulence factors tested, the pga gene was significantly associated with 30 day death and the adhesion-associated porB had a tendency of 30-day mortality-associated.

A different frequency of each gene by ST was occasionally identified (**Table 2**). The psaB gene, formerly the pab2 gene (24), was identified in all but one (96.6%, 28/29) ST451 isolates.

### Antimicrobial Treatment

The majority (95.0%, 172/181) of patients with A. baumannii BSIs received an empirical antimicrobial therapy with anti-Acinetobacter drugs and 28.5% (49/172) of those were adequate (**Table 3**). The other nine patients received either not-for-Acinetobacter drugs (n = 7) or none due to refusal of treatment (n = 2). Among the 132 patients who received an inadequate empirical therapy, 37.1% (n = 49) of patients were treated with a definitive antimicrobial therapy within 72 h and 20.4% (10/49) of those patients met death within 30 days, much less than that of patients who received a delayed definitive therapy (57.9%, 48/83).

## Risk Factors for 30-Day Mortality

Risk and protective factors associated with the 30-day mortality in patients with A. baumannii BSIs were searched for among the variables of P <0.1 through a Cox-regression univariable analysis. Most of the variables were omitted by stepwise selection procedure, and each three of risk factors and protective factors associated with the 30-day mortality in patients with A. baumannii BSIs were identified through a Cox-regression multivariable analysis (**Table 4**): the age of patients of hazard ratio (95% CI), 1.033 (1.010–1.056); the SOFA score of 1.133 (1.041–1.233); and the BSI-causative A. baumannii ST191 of 1.918 (1.073–3.430). Three protective factors were also identified: the platelet count of 0.996 (0.993–0.999), the BSI-causative A. baumannii ST451 of 0.228 (0.078–0.672), and the definitive therapy within 72 h of 0.255 (0.125–0.519).

## DISCUSSION

During the surveillance study period, the overall incidence of A. baumannii was the third-highest (1.1/10,000 patient-days) BSIcausative Gram-negative bacteria followed by Escherichia coli and Klebsiella pneumoniae. Besides, the incidence was strikingly higher in ICUs (6.6/10,000 patient-days) than in general wards (0.4/10,000 patient-days) (25). The 30-day mortality rate of patients with A. baumannii BSIs is pretty varied by study depending on the subjected patients, and we observed 42.5% of 30-day mortality for entire A. baumannii BSI cases.

Among the variables studied, the old age and the high SOFA score were identified as the patient-associated risk factors. Those were reasonable hazardous factors since both factors pronounced the severity of illness directly or indirectly. A patient-associated protective factor, platelet counts in peripheral blood reflecting the patients' condition also made sense. As well, a treatmentassociated protective factor, a proper definitive therapy within 72 h was comprehensible.

Curiously among the causative-pathogen-associated variables, the A. baumannii ST191 was a risk factor, while the ST451, another ICLII member, was a protective factor. A. baumannii

#### TABLE 2 | Characteristics of the *A. baumannii* ICLII blood isolates.


\**P value is either from Pearson's X<sup>2</sup> test for categorical data or from the t-test for continuous data. The Bonferroni corrected significance level from 33 hypotheses was considered to be P* < *0.0015, and the significant P values are indicated with asterisks.*

ICLII, is a well-known multidrug resistant clone (26, 27). The treatment options for patients infected by the clone are often limited to the very last resorts, such as carbapenems, tigecycline, minocycline, and polymyxins. For the infections cause by multidrug-resistant A. baumannii, the use of combination antimicrobial therapy is recommended (28) and two thirds (66.3%, 120/181) of the patients were treated empirically by antimicrobial combinations. Carbapenem resistance is already Yoon et al. *Acinetobacter baumannii* Bloodstream Infection

common in ICLII, and the remaining choices are always risky because tigecycline (29) and minocycline (30) have a bacteriostatic activity, which requires a combinational treatment; colistin has a severe dose-dependent neuro- and nephrotoxicity (31); and the colistin resistance is easily developed by prior use of the drug (32). The importance of an adequate antimicrobial therapy (33), including the proper use of colistin (34), has been addressed to diminish the 30-day mortality rate caused by A.


\**Patients received drugs not-for-Acinetobacter (n* = *7) or none due to refusal of treatment (n* = *2).*

baumannii infections. The present results confirmed the notion through a significant protective effect of the definitive therapy including colistin. Among ICLII members, ST191 was the most and both ST784 and ST451 are recently emerging ICLII clones (27, 32). In this study, ST784 and ST451 were the second and third most ICLII clones and the clones were evenly recovered from the six hospitals (**Figure 2**).

In this study, all but two (98.7%) ICLII blood isolates, one ST191 and the other ST357, exhibited MDR/XDR. Since drug resistance is one of the noted hazardous factors obstructing an adequate antimicrobial therapy (7, 8), similar mortality rates were presumed for BSIs caused by any STs belonging to the ICLII. However, a marked difference in 30-day mortality was observed in the three dominant STs: 60.3% in ST191 BSIs, 45.2% in ST784 BSIs, and 17.2% in ST451 BSIs. Curiously, the least proportion of patients with ST451 BSIs (10.3%) were treated by an adequate empirical therapy (**Table 2**), indicating that the drug resistance and the relevant adequacy of treatment were not related to the 30-day mortality rate at least to BSIs caused by ST191 and ST451. ST451 was more associated with a primary pulmonary tract infection (58.6%) than ST191 (32.9%), however the primary site of infection

TABLE 4 | Risk factors associated with the 30-day mortality in patients with *A. baumannii* BSI.


\**VIF, variance inflation factor. The collinearity was diagnosed for the variables including the multivariate analysis.*

*† A Cox-regression multivariate analysis was conducted by the backward method using the Wald model.*

FIGURE 2 | Mortality of BSI patients by causative pathogen. (A) Mortality of patients with BSIs by *A. baumannii*, (B) comparison of mortality of BSI patients by ICLII and by non-ICLII *A. baumannii*, (C) comparison of mortality of BSI patients by ST191, ST451, and other STs belonging to CC91. The Kaplan-Meier plot illustrated survival in patients with AB-BSI infected with the indicated bacterial clone. Differences between groups were compared by the Log-Rank test of the Mantel-Cox model, and the *P* values are presented. ICLII, international clonal lineage II; ST, sequence type.

was insignificantly associated with a 30-day death (**Table 1**). Similarly, ST451 harbored the Tn6019-associated AbaR (82.8%) more frequently than the Tn6022-associated AbGRI1 (17.2%) differently from ST191 (2.7 vs. 94.5%), however no difference was expected since both may harbor the blaOXA−<sup>23</sup> gene (12). The lipooligosaccharide biosynthesis-associated psaB gene was harbored strikingly more by ST451 than by ST191 (96.6 vs. 2.7%). The psaB gene is a part of the oligosaccharide gene clusters generating an ordinary side-branch sugar Gal, differing from the A. baumannii ST191 often carries the gene cluster producing FucNAc sugar composing the oligosaccharide (24). Since the lipooligosaccharides of pathogenic bacteria have long been recognized as a major virulence factor (35), the difference in oligosaccharides between the two clones is possibly associated with the 30-day mortality rate and the detailed impact should be further studied.

This study should be considered with the following caveats in mind. First, the enrolled cases had a lack of external validation due to a distinct epidemiology of A. baumannii clinical isolates in South Korea in terms of high carbapenem resistance, mostly conferred by OXA-23. Second, analyses of bacterial lipooligosaccharides differed by the clones are remained for further study. However, to the best of our knowledge, this is the first notification of the counter clinical prognoses of the patients with a BSI by dominant ICLII clones, drawing an attention to a further follow-up for the clonal exchange and expansion of ICLII clones in clinical settings.

This delicate prognosis assessment combined with an evaluation of the molecular epidemiology for the causative A. baumannii blood isolates confirmed (i) the predominance of ICLII in clinical settings, (ii) the hazardousness of antimicrobial resistance and the importance of an adequate antimicrobial therapy, and finally, (iii) we highlighted

## REFERENCES


the differing 30-day mortality rates in patients with BSIs between causative multidrug-resistant A. baumannii clones ST191 and ST451 belonging ICLII probably due to the differing lipooligosaccharides.

## 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.

## ETHICS STATEMENT

The research, which has no involvement of human subject but the clinical isolates, does meet the exempt category without approval from Ethics Committee on Human Research of the Health Ministry in South Korea and the study design has not been reviewed by the committee.

## AUTHOR CONTRIBUTIONS

SJ conceived the study. E-JY and SJ designed the study. E-JY, HSL, and SJ analyzed the data. E-JY and SJ wrote the manuscript. DK, HL, JongS, YU, KS, YK, YP, JeongS, and SJ collected the clinical data and bacterial isolates.

## FUNDING

This study was supported by the Research Programme funded by the Korea Centers for Disease Control and Prevention (2017E4400101#). We would like to thank to Yena Oh for technical support for microbiological experiments.

on clinical outcomes. Antimicrob Agents Chemother. (2013) 57:6270–5. doi: 10.1128/AAC.01520-13


May 2016 to April 2017: first one-year report from Kor-GLASS. Euro Surveill. (2018) 23:1800047. doi: 10.2807/1560-7917.ES.2018.23.42.18 00047


**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 Yoon, Kim, Lee, Lee, Shin, Uh, Shin, Kim, Park, Shin and Jeong. 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.

# Synergistic Microbicidal Effect of Auranofin and Antibiotics Against Planktonic and Biofilm-Encased S. aureus and E. faecalis

Pengfei She, Linying Zhou, Shijia Li, Yiqing Liu, Lanlan Xu, Lihua Chen, Zhen Luo and Yong Wu\*

Department of Clinical Laboratory, The Third Xiangya Hospital of Central South University, Changsha, China

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### Reviewed by:

Mariusz Stanislaw Grinholc, Intercollegiate Faculty of Biotechnology of University of Gdansk and Medical University ´ of Gdansk, Poland ´ Airat R. Kayumov, Kazan Federal University, Russia

> \*Correspondence: Yong Wu wuyong\_zn@csu.edu.cn

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 12 July 2019 Accepted: 11 October 2019 Published: 24 October 2019

#### Citation:

She P, Zhou L, Li S, Liu Y, Xu L, Chen L, Luo Z and Wu Y (2019) Synergistic Microbicidal Effect of Auranofin and Antibiotics Against Planktonic and Biofilm-Encased S. aureus and E. faecalis. Front. Microbiol. 10:2453. doi: 10.3389/fmicb.2019.02453 Methicillin-resistant/susceptible Staphylococcus aureus (MRSA/MSSA) and Enterococcus faecalis strains are often found in community- and hospital-acquired infections. The single use of conventional antibiotics hardly completely kills the bacterial cells of interest, especially in the form of biofilms. Thus, drug repurposing and antimicrobial combination are promising ways to solve this problem. Antimicrobial susceptibility assays against cocci in a suspension and in a biofilm mode of growth were performed with broth microdilution methods. Checkerboard assays and the cutaneous mouse infection model were used to examine the activity of auranofin and conventional antibiotics alone and in combination. In the present study, auranofin possesses potent antimicrobial activities against both planktonic cells and biofilms with minimum inhibitory concentrations ranging 0.125–0.5 mg/L. Auranofin in combination with linezolid or fosfomycin showed synergistic antimicrobial activities against S. aureus MSSA and MRSA both in vitro and in vivo. Similarly, auranofin also behaved synergistic effect with chloramphenicol against E. faecalis. Additionally, auranofin improved the antibiofilm efficacy of chloramphenicol and linezolid, even on the biofilms grown on a catheter surface. Though, S. epidermidis showed significant susceptibility to AF treatment, no synergistic antimicrobial effects were observed with antibiotics we tested. In all, the use of a combination of auranofin with linezolid, fosfomycin, and chloramphenicol can provide a synergistic microbicidal effect in vitro and in vivo, which rapidly enhances antimicrobial activity and may help prevent or delay the emergence of resistance.

Keywords: auranofin, biofilm, combination therapy, subcutaneous abscess model, Staphylococcus aureus, Enterococcus faecalis

## INTRODUCTION

Staphylococcus aureus and Enterococcus faecalis have been known to be responsible for most of healthcare- and nosocomial-associated infections. S. aureus could cause polymicrobial infections with many pathogens, such as enterococcus (Reyes et al., 2010; Hayakawa et al., 2013), Pseudomonas aeruginosa (Radlinski et al., 2017; Alves et al., 2018),

Peptostreptococcus anaerobius (Yamagishi et al., 2017), Streptococcus pyogenes (Gilmer et al., 2013), and even Candida species (Nash et al., 2015; Todd et al., 2019), which are hard to be eradicated and finally led to a striking mortality rate. According to the report by the SENTRY antimicrobial surveillance program (North America), the main pathogens isolated from skin as well as soft tissue infections (SSTIs) now include 45.9% S. aureus and 8.2% Enterococcus sp. (Rennie et al., 2003). SSTIs abscesses, for instance, create fluid, pus-filled pockets infiltrated by bacteria as well as inflammatory cells, and are frequently extremely resilient to conventional antibiotic therapy (Ki and Rotstein, 2008). In addition, abscesses are the utmost common sign for high-dose, recurrent and long-term intravenous broad-spectrum antibiotic administration (Ramakrishnan et al., 2015).

Biofilms are a widespread problem in healthcare facilities and hospitals. Indeed, the United States National Institutes of Health reported that 80% of chronic infections are related to biofilms (Monroe, 2007). The attachment of S. aureus, S. epidermidis, and E. faecalis onto tissues or the surface of medical apparatuses contributes to the pathogenesis of infection (Archer, 1998). The bacterial cells living in a biofilm are responsible for a number of chronic infections and become resilient to antibiotics as well as host-defense mechanisms (Gomes et al., 2009).

Recently, many studies have been conducted to address the repurposing of FDA-approved drugs as new antimicrobial agents. Auranofin (AF) is a gold-containing compound and prescribed for the treatment of rheumatoid arthritis (Glennas et al., 1997). The study of AF for its antimicrobial effects and inhibition of biofilm formation is an attractive possible treatment approach (Natsis and Cohen, 2018). Researchers found its antimicrobial efficacy against cocci (including Staphylococcus sp. and E. faecalis) and Mycobacterium tuberculosis. AF employs its effects via a distinctive process comprising the prevention of TrxR, and it maintains action against current antibiotic-resistant strains (Cassetta et al., 2014; Harbut et al., 2015; Fuchs et al., 2016). In addition, AF compared with most of the conservative medications available might be an appropriate feature in the fight against a dynamic as well as quickly altering microbial community such as biofilms.

AF shows good antimicrobial effects on cocci and AF in combination with topical antibiotics (mupirocin, retapamulin, and fusidic acid) exhibits additive antimicrobial activity against MRSA (Thangamani et al., 2016). However, to the best of our knowledge, there is no research reporting combinationary therapy with conventional systemic administration associated antibiotics in a subcutaneous abscess infections model. In the present study, we showed the antimicrobial and antibiofilm activities of AF alone or in combination with conventional antibiotics against S. aureus and E. faecalis strains in vitro and in vivo.

## MATERIALS AND METHODS

## Bacterial Strains

AF and antibiotics [fosfomycin (FOF); ciprofloxacin (CIP); tetracycline (TET); linezolid (LZD); chloramphenicol (CHL); levofloxacin (LVX); teicoplanin (TEC); clindamycin hydrochloride (CLI) hydro; gentamicin (GEN); vancomycin (VAN)] were purchased from the MedChemExpress company (Monmouth Junction, NJ, United States). E. faecalis ATCC 29212, S. aureus ATCC25923 and ATCC29213 were kindly provided by Juncai Luo (Tiandiren Biotech, Changsha, China).

TABLE 1 | Antimicrobial susceptibility testing of AF and VAN toward bacterial strains (mg/L).


<sup>a</sup>Methicillin resistant Staphylococcus aureus (MRSA). <sup>b</sup>Biofilm formation positive strains determined by CV staining method.

S. epidermidis RP62A and ATCC 12228 were given by Di Qu (Shanghai Medical College of Fudan University), S. aureus ATCC43300 (MRSA), Newman, and RJ-2 were given by Min Li (Renji Hospital, Shanghai Jiao Tong University School of Medicine). Other clinical strains were isolated from the wound secretion or sputum of inpatients at the Third Xiangya Hospital of Central South University. Staphylococcus spp. were grown in tryptic soy broth (TSB) broth medium (Solarbio, Shanghai, China), and E. faecalis was grown in brain heart infusion (BHI) broth medium (Solarbio, Shanghai, China) at 37◦C.

## Susceptibility Testing of Planktonic Bacteria

Bacterial strains were cultured in cationic corrected Mueller– Hinton (MH) broth (BD/Difco, United States). Susceptibility


tests were performed by twofold regular broth microdilution of the test compounds, as recommended by the Clinical and Laboratory Standards Institute (CLSI) (Harbut et al., 2015). After 16–18 h of incubation at 37◦C, the nominal concentration necessary to stop the development of test bacteria was defined as the MIC, and the minimum bactericide concentration (MBC) was identified depending on the lowermost concentration of antimicrobials that killed 99.9% of the test bacteria by spreading the bacterial culture out onto a suitable agar plate (CLSI, 2005).

## Susceptibility Testing of Biofilms

For S. aureus biofilm determination. The culture was grown overnight in TSB and successively diluted 1:50 in TSB to achieve an absorbance at 630 nm of ∼0.1. Two hundred microliter aliquots of the diluted culture were added to every well of a microtiter plate and incubated at 37◦C for 24 h. For E. faecalis biofilm determination, bacterial suspensions (18 µL) from overnight cultures were mixed with 162 µL of BHI in the wells, and biofilms were allowed to form on the plates for 24 h (Lee et al., 2012).

Following the incubation, the contents were removed and rinsed, 50 µL of medium and 50 µL of the specified drug were added to every well, and incubated at 37◦C for 24 h. Then, the contents were removed and the remaining biofilms were determined by crystal violet (CV), XTT staining or live cell count as follows:


MBEC30/MBEC50/MBEC70 were defined as the minimal concentration of the particular antimicrobial's ability to inhibit 30/50/70% growth of the biofilms, respectively, compared to the control group (Gomes et al., 2009).

(3) Biofilm viable count (Mataraci and Dosler, 2012). One hundred microliters of 1 × PBS was aliquoted into each well, and the contents were scrapped and mixed thoroughly with pipette tips. A sample volume of 100 µL was plated onto blood agar and successively diluted with a saline solution before plating onto additional agar plates.

## Checkerboard Assays for Planktonic Bacteria

The impacts of individual antibiotics and in combination with AF were evaluated using the broth microdilution checkerboard technique (Odds, 2003; Flamm et al., 2019). Each microtiter well-comprising the designated combination of antibiotics was inoculated with an overnight culture diluted to provide an absolute concentration of ∼5 × 10<sup>5</sup> CFU/ml. Following incubation, the optimal fractional inhibitory concentration index (FICI) was measured as the minimal inhibitory concentration of the combination divided by that of the single antibiotic (Odds, 2003): FICI ≤ 0.5 designates synergy; 0.5 < FICI ≤ 4.0 designates no interaction; FICI > 4.0 designates antagonism.

## Checkerboard Assays for Preformed Biofilms

The preparation of overnight biofilms was the same as explained earlier in this study. The biofilms were rinsed, twofold sequential dilutions of antibiotics and AF in a 96-well microtiter plate were prepared, and 100 µL of these mixtures were added to the biofilms. Concentration ranges, as recognized with susceptibility testing, were utilized for the antibiotics as well as the AF. Following an incubation for 24 h at 37◦C, the medium containing antimicrobials was removed, and 100 µL of XTT with PSM was

TABLE 3 | The antibacterial activity and combined effects of AF and selected antibiotics alone or in combination against MRSA and other clinical isolates.


staining. Overnight cultures of strains were diluted with AF to the designated concentrations. After 24 h incubation, planktonic cells were removed and stained with 0.25% CV. (B) Biofilm eradication by AF detected by XTT staining. Biofilms grown for 24 h were treated with AF at the designated concentrations. After incubation, planktonic cells were removed and stained with a solution of XTT/PMS. The red dashed line indicates 50% of the biofilm biomass of the control group. (C) Live cell counts. Twenty-four hour biofilms were treated with AF at a concentration of MBEC50 for 24 h, and serial dilutions and plate counts were performed to determine the live cells in the biofilms (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001).

added as described above (Koppen et al., 2019). The MBEC50 values were quantified.

## Antibiofilm Effect of AF on Catheters

To study the efficacy of AF combined with antibiotics against biofilms on catheters, overnight cultures of the biofilm-forming strains were diluted 1:40 in TSB (S. aureus) or BHI (E. faecalis) containing 5% rabbit plasma. Catheter (Jerry infusion set, Shandong, China) pieces (1 cm in size) were cut, divided into two halves, and added to the culture. Next, they were incubated at 37◦C for 24 h. Afterward, the catheters were removed and washed. The biofilms on catheters were challenged with AF alone or in combination with antibiotics for 24 h. The catheters were scratched by an inoculation loop and sonicated for 15 min. Then,


TABLE 4 | Activity of AF in combination with conventional antibiotic against preformed biofilms (mg/L).

the samples were vortexed carefully and plated on blood agar plates (Nair et al., 2016).

## Confocal Laser Scanning Microscopy (CLSM)

The above-treated bacteria were cultured on glass cover slides and incubated with 10 µL of 1000-fold diluted SYTO9 fluorescent staining solution and propidium iodide at a ratio of 1:1 (vol/vol) for 15 min in the dark. After rinsing, the stained biofilm was examined with a CLSM (Zeiss LSM 800, Jena, Germany) (Nair et al., 2016).

## Cutaneous Mouse Infection Model

Seven-week-old female mice CD-1 were purchased from Hunan Slake Jingda Experimental Animal, Co., Ltd. (Hunan, China). They weighed approximately 25 ± 3 g at the time of the experiments.

The high bacterial load abscess infection model was performed as defined earlier with slight adaptations (Pletzer et al., 2018). Before the injection, bacterial cells were rinsed resuspended in 1 × PBS. An injection of bacterial suspension was given to the dorsum to achieve the concentrations to generate reproducible abscesses and bacterial counts: S. aureus, 1 × 10<sup>8</sup> CFU/mice; and E. faecalis, 1 × 10<sup>9</sup> CFU/mice. Antimicrobial administration was given directly into the subcutaneous space of the infected area at 1 h post-infection. The development of the infection was observed every day. Abscesses were determined on day 2 using a caliper. Skin abscesses were removed (comprising all accrued pus) and regimented in sterile PBS by an automatic tissue homogenizer (Servicebio KZ-II, Wuhan, China). Bacterial counts were quantified by serial dilution. For histopathological analyses, hematoxylin and eosin (H&E) staining was performed.

## Statistical Analysis

Statistical evaluations were performed using GraphPad Prism 7.0. Checkerboard methods were performed at least in biological duplicates, and other experiments were performed in triplicate.

cells, red). (A) S. aureus LZB1, AF 8 mg/L, LZD 16 mg/L. (B) E. faecalis ATCC29212, AF 2 mg/L, CHL 8 mg/L. Scale bar: 40 µm.

## RESULTS

## Determination of the Susceptibility of Planktonic Cells

The MICs of AF and VAN against type strains and clinical isolates of S. aureus (MSSA/MRSA), S. epidermidis, and E. faecalis were 0.125–2 mg/L. The MBCs against S. aureus and S. epidermidis were 0.5–4 and 1–8 mg/L for AF and VAN, respectively. And the susceptibility of AF against MRSA and MSSA strains showed no difference. However, the MBCs of E. faecalis were > 32 mg/L for both AF and VAN (**Table 1**). In all, the strains we tested were more sensitive to the AF treatment than the VAN treatment.

## Synergistic Effect Between AF and Antibiotics Against Planktonic Cells

The synergistic effects of AF were investigated with some conventional systemic antibiotics (antibiotics with MIC values greater than 256 mg/L were excluded). The results of the combination screening assay are presented in **Table 2**. Synergistic interactions between AF and FOF (FICI = 0.375) or LZD (FICI = 0.375) were observed against S. aureus LZB1. For E. faecalis ATCC29212, synergistic interactions were observed between AF and CHL (FICI = 0.375). But no interactions between AF and antibiotics were observed against S. epidermidis RP62A (FICI > 0.5). Combinations with the lowest FICI values were selected for other representative strains. As shown in **Table 3**, the combination of AF and CHL still showed synergistic effect against the E. faecalis clinical isolates; combinations of AF + LZD/FOF still showed synergistic effects against S. aureus ATCC43300 (MRSA) and most of the clinical isolates, except for strain SA1435 which showed no interaction between AF and FOF with FICI of 0.5.

## Determination of the Susceptibility of Biofilms

Staphylococcus aureus LZB1 and E. faecalis ATCC29212 were selected to test the antibiofilm activities of AF due to their strong biofilm formation abilities (Kart et al., 2017). AF showed strong biofilm inhibitory effects against S. aureus, and E. faecalis at concentrations of 0.125 and 1 mg/L (p < 0.05), respectively, in a dose-dependent manner (**Figure 1A**), which were very close to its MICs, indicating that the biofilm inhibitory effect of AF could be mainly due to its bacteriostatic or bactericidal activity by targeting thiol-redox homeostasis (Harbut et al., 2015). Because biofilm formation strongly increased the antimicrobial resistance to AF, the lowest concentrations needed to eradicate preformed biofilms were up to 4 and 2 mg/L for S. aureus and E. faecalis, respectively (**Figure 1B**). MBEC50 was selected to detect the time kill efficacy of AF against biofilms. AF showed significant biofilm killing activity against these strains in a time-dependent manner. Compared to the control group, AF reduced the live biofilm cells of S. aureus from (1.38 ± 0.29) × 10<sup>9</sup> CFU/ml to (1.13 ± 0.90) × 10<sup>7</sup> CFU/ml (p < 0.001). Although statistical significance was only observed at 8 h after treatment (p < 0.05), AF killed E. faecalis ATCC29212 biofilm cells throughout the 24 h period (**Figure 1C**). In addition, AF could also effectively eradicate clinical isolates with low MBEC50 values (**Supplementary Table S2**).

## Synergistic Effect Between AF and Antibiotics Against Biofilms

Antibiotics that showed a synergistic effect on planktonic cells were tested against preformed biofilms in combination with AF (**Table 4**). AF significantly promoted the antibiofilm efficacy of CHL against E. faecalis ATCC29212 (4- and 8-fold decrease of MBEC50 for CHL and AF, respectively). Meanwhile, AF also increased the antibiofilm activity of LZD against S. aureus LZB1 and exhibited a 2- and > 8-fold decrease of MBEC50 for AF and LZD, respectively, but showed no interaction with FOF (**Supplementary Figure S1**). Similar observations were made by visualization of AF and/or antibiotic-treated biofilms by CLSM, when AF used in combination with FOF (S. aureus LZB1, **Figure 2A**) or CHL (E. faecalis ATCC29212, **Figure 2B**), the live cells in the biofilms were significantly reduced, although some intact patches of biofilm could still be visualized.

To simulate the in vivo conditions for biofilm formation in device-associated infections, we allowed strains to form biofilms on the surfaces of catheters. Treatment of biofilms with AF and in combination with antibiotics led to their synergistic eradication (**Figure 3**). A single dose of AF or antibiotics only showed moderate antibiofilm effects; however, combination treatment led to a 4.96- and 1.95-log reduction in CFUs for S. aureus (AF + LZD, **Figure 3A**) and E. faecalis (AF + CHL, **Figure 3B**), respectively, confirming that AF possesses antibiotic-promoting activity against preformed biofilms on catheters.

## Therapeutic Efficacy of AF Combined With Conventional Antibiotics in vivo

To optimize the treatment strategy, antimicrobials were chosen based on their moderate in vivo pharmacodynamics (**Figure 4**), and the concentrations used in the present study were equal or less than those empirically tested in vivo (CHL, 10 mg/kg; LZD, 60 mg/kg; and FOF, 100 mg/kg) (Shibl, 1982; Guo et al., 2013; Zykov et al., 2018) to determine an appropriate concentration that reduces abscess sizes just enough to observe the synergy between the AF and the antibiotics (**Figure 4**, red dashed line). A significant reduction in the mean bacterial load was observed for each combined treatment condition compared with the control (receiving DMSO or Tween-80) or single dose group.

Except for FOF, which reduced the abscess area of S. aureus LZB1 57.38 mm<sup>2</sup> , single use of AF (0.5 mg/kg), LZD (5 mg/kg) or FOF (20 mg/kg) showed no statistical significance in reducing abscess area or bacterial loads of S. aureus LZB1 (MSSA) (**Figure 5A**) and ATCC43300 (MRSA) (**Figure 5B**) infections; however, AF combined with LZD significantly decreased the abscess area and reduced the bacterial load for 4.51- (S. aureus LZB1, p < 0.01) and 2.45-fold log10 (S. aureus ATCC43300, p < 0.001). Similarly, AF or CHL could not inhibit the abscess growth of E. faecalis ATCC29212 individually, but when combined, the area of abscess was reduced by 74.14 mm<sup>2</sup> (p < 0.01). Single use of AF or CHL had no impact on bacterial load; however, combined therapy reduced the bacterial load by 0.61-fold log10 (p < 0.01) (**Figure 5C**). For in vivo observations, the abscesses caused by S. aureus were more obvious than those caused by E. faecalis. The ulcers were formed when infected with S. aureus LZB1 or ATCC43300 (**Figures 6A,B**); however, infection with a high load of E. faecalis ATCC29212 (**Figure 6C**) only caused subcutaneous lumps. In accordance with the in vitro observations, the representative pictures of abscesses and histological examinations showed that single use of AF, LZD, or FOF showed no/moderate activity against infections caused by S. aureus LZB1 (**Figure 6A**) or ATCC43300 (**Figure 6B**), and extensive inflammation with leukocyte infiltration emerged; however, drug combination (AF + LZD or AF + FOF) significantly reduced the size and inflammation of the abscesses,

which even eventually disappeared. The single use of AF or CHL had no influence on the abscesses caused by E. faecalis ATCC29212, the drug combination significantly diminished the abscess size and inflammation infiltration (**Figure 6C**). These important observations highlight that antimicrobial monotherapies are often ineffective when bacteria form highdensity infections for S. aureus and E. faecalis. In addition, drug combinations could significantly improve the efficacy.

## DISCUSSION

In the present study, the antimicrobial activity of AF was assessed against a panel of type strains and clinical isolates of Staphylococcus spp. and E. faecalis. In accordance with the previous studies reported by Cassetta et al. (2014), Harbut et al. (2015), Fuchs et al. (2016), the MICs for S. epidermidis, S. aureus (including MSSA and MRSA), and E. faecalis were 0.125–0.5 mg/L, which showed more susceptibility than with VAN treatment, with MICs ranging from 0.5 to 2 mg/L.

Drug combination is a promising way to improve the efficacy of drugs and reduce side effects and cytotoxicity. In our study, highly synergistic interactions between AF and CHL were observed against E. faecalis. CHL is a broad-spectrum antibiotic against many gram-positive/negative bacteria (Civljak et al., 2014). However, CHL is an old antimicrobial agent that is rarely used today mainly due to its most significant adverse effect of dose-related bone marrow suppression, according a meta-analysis by Eliakim-Raz et al. (2015), CHL is as safe a treatment alternatives as short antibiotic courses. In this way, drug combination could significantly diminish the dose required but achieve better antimicrobial efficacy, so that AF combined with CHL could be a better choice than CHL used alone in clinical therapy. Moreover, E. faecalis has shown many different metabolic responses from anaerobic to aerobic circumstances; these main metabolic cascades are related to the response to nutrients and may change the susceptibility of this bacterium to bactericidal drugs (Portela et al., 2014). However, even in the anaerobic condition, AF still showed a highly synergistic effect with CHL (**Supplementary Table S1**).

Highly synergistic interactions between AF and LZD/FOF were also observed against MSSA or MRSA strains of S. aureus. LZD has a wide spectrum of action against the mainstream of common gram-positive cocci. However, due to the development

of resistance to antibiotics as well as their unwanted side effects, combination therapy has evolved as an imperative novel treatment approach (Yang et al., 2018). FOF by itself has a bactericidal impact both in vitro as well as in vivo. Nevertheless, MRSA can easily develop tolerance, making utilization of FOF unattainable for medical situations (Roussos et al., 2009). FOF displays positive synergistic impacts on MRSA and its biofilms when utilized in combination with other antibiotics (Shi et al., 2014). Similarly, in our study, AF showed an excellent synergistic effect with FOF against MSSA and MRSA planktonic cells. In addition, different clinical isolates showed outcomes of different drug combinations, which indicates the importance of an in vitro synergistic test before clinical use.

Biofilms are easily formed on indwelling medical apparatus surfaces (Francolini and Donelli, 2010). During implantation of catheters, tissue damage might occur due to the buildup of platelets as well as fibrin at the suture site as well as on the devices. Microbial cells have enhanced capability to colonize these sites (Jamal et al., 2018). The formation of biofilms increases the antibiotic resistance and leads to persistent infections posing major healthcare challenges. AF showed modest biofilm inhibitory and eradicating effects against Staphylococcus aureus and E. faecalis both in type strains and clinical isolates with high values of MBEC70. Thus, our favorable outcomes of synergy among AF and antibiotics encouraged us to inspect the action of drug combinations against biofilms. AF combined with CHL showed synergistic antibiofilm effects for E. faecalis on cover slides or infusion catheters (**Supplementary Figure S1**), AF significantly improved the antibiofilm effects of LZD against S. aureus. As numerous antibiotics have diverse antibacterial mechanisms and several bacteria have diverse resistance mechanisms, to entirely eliminate the whole biofilmbacteria is a difficult challenge. Combination therapy comprising two or more antibiotics with diverse bactericidal mechanisms could synergistically eliminate biofilms (Simoes, 2011).

A high bacterial load-containing abscess model has rarely been studied for AF efficacy evaluation. In our in vivo subcutaneous abscess model study, single use of antimicrobials showed an extremely modest effect on abscess area or bacterial load. However, AF combined with LZD or FOF synergistically inhibited abscess and inflammation formation and reduced the bacterial load for both MSSA and MRSA strains. The safety of AF in in vivo animal studies and clinical use is welldocumented. AF is widely used in clinical settings for longterm treatment at the daily dosage at 6 mg/day, and a average blood concentration of 3.5 µM (∼2.38 µg/ml, which is far beyond the value of MICs) is reached in 12 weeks. Besides, the effectiveness and safety of AF at an dose of 12 mg/day is under Phase II clinical trial (Harbut et al., 2015). As reported by Aguinagalde et al. (2015), the dosage of AF used for murine model even reach to 10 mg/kg due to its safe toxicity profile and well-known pharmacokinetic/pharmacodynamic characteristics.

Similarly, the safety of antibiotics of FOF and LZD is wellstudied and documented. As reported by Pachón-Ibáñez et al. (2011) and Guo et al. (2013), the dosages of FOF and LZD used are reached to 100 and 60 mg/kg in murine models, respectively. And the FOF and LZD used in our study are only 20 and 5 mg/kg, respectively. Therefore, the inflammation caused in our animal models is not caused by the antimicrobials we used. In all, the combination therapy of AF plus LZD/FOF might be an effective option for treating patients with S. aureus-related subcutaneous abscess infection. Similarly, AF combined with CHL also showed synergistic antibacterial effects on E. faecalis abscesses and partially reduced inflammation formation.

## CONCLUSION

The present study provides a valuable effect of antimicrobial combination therapy against cocci in subcutaneous abscess infections. This type of synergistic combination of two medications is likely preferred in clinical situations. The rationality of the outcomes should be validated by future clinical trials.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## ETHICS STATEMENT

Ethical approval was obtained from the Animal Ethics Committee (certificate number 2017-S139), the Third Xiangya Hospital, Central South University, China.

## REFERENCES


## AUTHOR CONTRIBUTIONS

PS and YW designed and performed the experiments and wrote the manuscript. PS, YW, and LZ performed the experiments and data collection. SL, YL, LC, and ZL performed the experiments and revised the manuscript. All authors read and approved the final manuscript.

## FUNDING

This study was funded by the Hunan Provincial Natural Science Fund, China (Grant No. 2019JJ80029).

## ACKNOWLEDGMENTS

We are particularly grateful to the staff of the Department of Clinical Laboratory of the Third Xiangya Hospital of Central South University, Changsha, China.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02453/full#supplementary-material

FIGURE S1 | Representative CLSM images of biofilm eradication by AF and FOF mono-/combination treatment. S. aureus LZB1 biofilms on the surfaces of cover slides were treated with AF (16 mg/L) and/or FOF (128 mg/L) as described in Section "Materials and Methods," the stained with the SYTO9/PI fluorescent dye mixture.

TABLE S1 | The combinational antibacterial activities of AF and different antibiotics against E. faecalis ATCC29212 in anaerobic conditions.

TABLE S2 | Biofilm eradication activities of AF against other type and clinical strains (mg/L).


rheumatoid arthritis. Br. J. Rheumatol. 36, 870–877. doi: 10.1093/rheumatology/ 36.8.870


**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 She, Zhou, Li, Liu, Xu, Chen, Luo and Wu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# Genomic Characterization of Prevalent mcr-1, mcr-4, and mcr-5 Escherichia coli Within Swine Enteric Colibacillosis in Spain

Isidro García-Meniño<sup>1</sup>† , Dafne Díaz-Jiménez<sup>1</sup>† , Vanesa García<sup>1</sup>‡ , María de Toro<sup>2</sup> , Saskia C. Flament-Simon<sup>1</sup> , Jorge Blanco<sup>1</sup> and Azucena Mora<sup>1</sup> \*

<sup>1</sup> Laboratorio de Referencia de Escherichia coli, Departamento de Microbiología y Parasitología, Facultad de Veterinaria, Universidad de Santiago de Compostela, Lugo, Spain, <sup>2</sup> Plataforma de Genómica y Bioinformática, Centro de Investigación Biomédica de La Rioja, Logroño, Spain

#### Edited by:

Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil

#### Reviewed by:

Darren Trott, The University of Adelaide, Australia Maite Muniesa, University of Barcelona, Spain

> \*Correspondence: Azucena Mora azucena.mora@usc.es

†These authors have contributed equally to this work

### ‡Present address:

Vanesa García, Department of Veterinary and Animal Sciences, Faculty of Health and Medical Sciences, University of Copenhagen, Frederiksberg, Denmark

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 26 August 2019 Accepted: 15 October 2019 Published: 01 November 2019

#### Citation:

García-Meniño I, Díaz-Jiménez D, García V, de Toro M, Flament-Simon SC, Blanco J and Mora A (2019) Genomic Characterization of Prevalent mcr-1, mcr-4, and mcr-5 Escherichia coli Within Swine Enteric Colibacillosis in Spain. Front. Microbiol. 10:2469. doi: 10.3389/fmicb.2019.02469 Antimicrobial agents are crucial for the treatment of many bacterial diseases in pigs, however, the massive use of critically important antibiotics such as colistin, fluoroquinolones and 3rd–4th-generation cephalosporins often selects for co-resistance. Based on a comprehensive characterization of 35 colistin-resistant Escherichia coli from swine enteric colibacillosis, belonging to prevalent Spanish lineages, the aims of the present study were to investigate the characteristics of E. coli clones successfully spread in swine and to assess the correlation of the in vitro results with in silico predictions from WGS data. The resistome analysis showed six different mcr variants: mcr-1.1; mcr-1.10; mcr-4.1; mcr-4.2; mcr-4.5; and mcr-5.1. Additionally, blaCTX-M-14, blaCTX-M-<sup>32</sup> and blaSHV-<sup>12</sup> genes were present in seven genomes. PlasmidFinder revealed that mcr-1.1 genes located mainly on IncHI2 and IncX4 types, and mcr-4 on ColE10-like plasmids. Twenty-eight genomes showed a gyrA S83L substitution, and 12 of those 28 harbored double-serine mutations gyrA S83L and parC S80I, correlating with in vitro quinolone-resistances. Notably, 16 of the 35 mcrbearing genomes showed mutations in the PmrA (S39I) and PmrB (V161G) proteins. The summative presence of mechanisms, associated with high-level of resistance to quinolones/fluoroquinolones and colistin, could be conferring adaptive advantages to prevalent pig E. coli lineages, such as the ST10-A (CH11-24), as presumed for ST131. SerotypeFinder allowed the H-antigen identification of in vitro non-mobile (HNM) isolates, revealing that 15 of the 21 HNM E. coli analyzed were H39. Since the H39 is associated with the most prevalent O antigens worldwide within swine colibacillosis, such as O108 and O157, it would be probably playing a role in porcine colibacillosis to be considered as a valuable subunit antigen in the formulation of a broadly protective Enterotoxigenic E. coli (ETEC) vaccine. Our data show common features with other European countries in relation to a prevalent clonal group (CC10), serotypes (O108:H39, O138:H10, O139:H1, O141:H4), high plasmid content within the isolates and mcr location, which would support global alternatives to the use of antibiotics in pigs. Here, we report for first time a rare finding so far, which is the co-occurrence of double colistin-resistance mechanisms in a significant number of E. coli isolates.

Keywords: Escherichia coli, colistin, mcr, ESBL, fluoroquinolones, ST10, colibacillosis, swine

## INTRODUCTION

fmicb-10-02469 October 31, 2019 Time: 18:39 # 2

Multidrug-resistant Enterobacteriaceae, such as Escherichia coli, represent a threat to both human and veterinary health. E. coli has a great capacity to accumulate resistance genes, mostly through horizontal gene transfer. The major problematic mechanisms correspond to the acquisition of genes coding for extended-spectrum beta-lactamases (ESBL), carbapenemases, 16S rRNA methylases, plasmid-mediated quinolone resistance (PMQR) and mcr genes conferring resistance to polymyxins (Poirel et al., 2018).

Colistin has been widely used in Spain for the control of neonatal and post-weaning diarrhoea (PWD) in pigs caused by certain E. coli pathotypes: Enterotoxigenic E. coli (ETEC), defined by the presence of genes encoding enterotoxins (eltA, and/or estA, and/or estB); atypical Enteropathogenic E. coli (aEPEC), carriers of eae but negative for bfpA (aEPEC); Shiga toxin–producing E. coli (STEC), positive for stx2e; STEC/ETEC, positive for both shiga toxin type 2e and enterotoxin-encoding genes (stx2e and estB and/or estA) (García-Meniño et al., 2018). PWD results in significant economic losses for the pig industry due to costs derived of treatment and handling, decreased weight gain, and mortality. These circumstances have promoted the use and abuse of antibiotics in intensive farming (Luppi, 2017; Rhouma et al., 2017). However, specific regulations have been set up in Europe due to the concern that colistin resistance could be transmitted from food-production animals to humans which makes necessary the investigation of sustainable alternatives to antimicrobials (EUROPEAN COMMISSION, 2018).

In Spain, the rates of antibiotic resistance in pig farming were recently analyzed in a collection of 499 E. coli isolates from 179 outbreaks of enteric colibacillosis occurred during a period of 10 years (2006–2016) (García et al., 2018; García-Meniño et al., 2018). The results revealed a prevalence of colistin-resistant E. coli implicated in PWD in Spanish farms as high as 76.9% within 186 ETEC, STEC and STEC/ETEC isolates. Besides, PCR and sequencing identified the presence of mcr-4 in 102 isolates, mcr-1 in 37 isolates and mcr-5 in five isolates. Interestingly, almost all mcr-4 isolates belonged to the clonal group ST10- A (CH11-24) (García et al., 2018), which was shown to be highly present (more than 50%) within the mcr-1 diarrheagenic isolates of a second study (García-Meniño et al., 2018). Both studies reinforced other countries' findings that the pig industry is an important reservoir of colistin-resistant E. coli, as well as being carriers of other additional risk genes such as blaESBL genes (García et al., 2018; García-Meniño et al., 2018; Magistrali et al., 2018; Manageiro et al., 2019). Based on reported evidences (Beyrouthy et al., 2017; Gilrane et al., 2017), there is great concern about the in vivo acquisition of mcr- and blaESBL-bearing plasmids by human E. coli isolates following treatment with colistin, or via animal transmission through direct contact or via food chain. Particular attention is given to those named as high-risk clones of (ESBL)-producing bacteria, worldwide spread within humans and animals, including Escherichia coli sequence types ST10, ST131, ST405, and ST648 (Mathers et al., 2015; Sellera and Lincopan, 2019).

The aims of this study were (i) the characterization of resistances and plasmid profiles of successfully spread mcr-1, mcr-4, and mcr-5 E. coli in Spanish pig farming; (ii) the assessment of WGS-based approaches for the characterization of pathogenic E. coli, through the correlation of the in vitro results with in silico predictions using the bioinformatics tools of the Center for Genomic Epidemiology (CGE).

## MATERIALS AND METHODS

## E. coli Collection

Thirty-five swine E. coli, positive by PCR for mcr-genes, were fully sequenced. Specifically, the 35 E. coli were selected from 499 diarrheagenic isolates of different geographic areas of Spain (2006–2016) (García et al., 2018; García-Meniño et al., 2018), taking into account the results of prevalence and significant association observed between pathotypes, presence of mcr and certain serogroups. In brief, the serogroups O108, O138, O141, O149, O157 were found significantly associated with ETEC; serogroups O26, O49, O80, O111 with aEPEC; serogroups O138 and O141 with STEC/ETEC; serogroup O139 with STEC; and serogroups O2, O15, O26, O45, O111, O138, O141, O157 with mcr-positive isolates (García-Meniño et al., 2018). Therefore, the collection analyzed here included 27 ETEC isolates (of serogroups O7, O8, O15, O45, O108, O138, O141, O149, O157, ONT); four STEC (O2, O139); three STEC/ETEC (O138 and O141) and one aEPEC (O111). The 35 representative isolates were carriers of the three mcr-types (mcr-1, mcr-4, and mcr-5) detected so far in our E. coli collection of porcine origin. Conventional pheno- and geno-typing was performed to complete classical characterization of serotypes, phylogroups, pathotypes and resistance profiles.

## Conventional Typing

The H antigen was established for motile isolates by serotyping using H1 to H56 antisera, while non-motile isolates (HNM) were analyzed by PCR to determine their flagellar genes as described elsewhere (García-Meniño et al., 2018). The phylogroup was assigned by means of the quadruplex PCR of Clermont et al. (2013). Antimicrobial susceptibility was determined by minimal inhibitory concentrations (MICs) using the MicroScan WalkAway <sup>R</sup> -automated system (Siemens Healthcare Diagnostics, Berkeley, CA, United States) according to the manufacturer's instructions for: amikacin, ampicillin-sulbactam, aztreonam, cefepime, ceftazidime, ciprofloxacin, colistin, fosfomycin, gentamicin, imipenem, levofloxacin, meropenem, minocycline, nitrofurantoin, piperacillin-tazobactam, ticarcillin, tigecycline, and tobramycin. Additionally, resistance to ampicillin, amoxicillin/clavulanate, cefazolin, cefotaxime, cefoxitin, cefuroxime, chloramphenicol, doxycycline, nalidixic acid and trimethoprim-sulfamethoxazole was determined by disk (Becton Dickinson, Sparks, MD, United States) diffusion assays. All results were interpreted according to the CLSI break points (Clinical and Laboratory Standards Institute, 2019). Genetic identification of the ESBLs was performed by PCR using

the TEM, SHV, CTX-M-1 and CTX-M-9 group-specific primers followed by amplicon sequencing (García-Meniño et al., 2018).

## Whole Genome Sequencing (WGS) and Sequence Analysis

The libraries for sequencing were prepared following the instructions provided by the TruSeq Illumina PCR-Free protocol. Mechanical DNA fragmentation was performed with Covaris E220, and the final quality of the libraries assessed with Fragment Analyzer (Std. Sens. NGS Fragment Analysis kit 1- 6000 bp). Lastly, the libraries were sequenced in an Illumina HiSeq1500, obtaining 100–150 bp paired-end reads which were trimmed (Trim Galore 0.5.0) and filtered according to quality criteria (FastQC 0.11.7). The reconstruction of the genomes and plasmids in the genomes was carried out using the methodology PLAsmid Constellation NETwork (PLACNETw)<sup>1</sup> (Lanza et al., 2014). The assembled contigs, with genomic size ranging between 4.9 and 5.9 Mbp (mean size 5.5 Mbp), were analyzed using the bioinformatics tools of the Center for Genomic Epidemiology (CGE)<sup>2</sup> for the presence of antibiotic resistance (ResFinder V2.1.), virulence genes (VirulenceFinder v1.5.), plasmid replicon types (PlasmidFinder 1.3./PMLST 1.4.), and identification of clonotypes (CHTyper 1.0), sequence types (MLST 2.0) and serotypes (SerotypeFinder 2.0). All the CGE predictions were called applying a select threshold for identification and a minimum length of 95 and 80%, respectively. Phylogroups were predicted using the ClermonTyping tool at the iame-research center web<sup>3</sup> . The mcr gene location was determined using PlasmidFinder/ResFinder prediction, together with PLACNETw references, and automatic annotation with Prokka v1.13 (Seemann, 2014).

## RESULTS AND DISCUSSION

The phenotypic and genotypic traits of the 35 mcr-positive E. coli of swine origin, as well as their resistome and mobilome are summarized in **Table 1**. ResFinder confirmed that all genomes were mcr carriers. Likewise, VirulenceFinder predicted the acquired virulence genes encoding for the enterotoxins (sta1, stb, itcA), for fimbriae (fedF, k88), verotoxin (stx2) and intimin (eae), correlating in all cases with the pathotype assignation previously determined by PCR (García et al., 2018; García-Meniño et al., 2018).

## Serotype Identification

In most studies, there is lack of information on E. coli serotypes since serotyping is performed by very few laboratories worldwide, hindering epidemiological comparisons. Here, we not only proved that there is a very good correlation between serotyping and SerotypeFinder predictions, but also the advantage of in silico H-antigen identification for those non-mobile (HNM) isolates. It is of note that 15 of the 21 HNM isolates were predicted

<sup>1</sup>https://castillo.dicom.unican.es/upload/

<sup>2</sup>https://cge.cbs.dtu.dk/services/

as H39 (**Table 1**), namely O108:H39, O157:H39 and O45:H39 (five genomes, each). Given that the H39 is associated with the most prevalent O antigens within swine colibacillosis, such as O108 and O157, as well as ONT (García-Meniño et al., 2018), it would be probably playing a role in porcine colibacillosis to be considered as a valuable subunit antigen in the formulation of a broadly protective ETEC vaccine (Roy et al., 2009). The remaining six HNM isolates showed different O:H combinations: O138:H14, ONT:H5, O8:H20, O50/O2:H32, and O182:H19. SerotypeFinder also allowed the O45-antigen determination of two non-typeable (ONT) isolates (LREC-141 and LREC-146) and O182 of LREC-172; while LREC-147, belonging to O157 serogroup (**Table 1**), was predicted as ONT, probably due to the limitation of the assembly based on Illumina short reads (100–150 bp paired-end reads here) (Wick et al., 2017).

## Phylogroups, Sequence Types and Clonotypes

The phylogroups established for the 35 genomes were the common ones reported for porcine E. coli isolates (A, B1, D-E) (Shepard et al., 2012; Bosak et al., 2019). However, we found discrepancies in the assignation obtained with the quadruplex PCR of Clermont et al. (2013) in comparison with that predicted by ClermonTyping for seven isolates: phylogroup E by PCR, while phylogroup D in silico (**Table 1**).

MLST and CHTyper tools determined 12 different STs, but mostly belonging to CC10 (21 genomes) and clonotype CH11- 24 (18 genomes) (**Table 1**). The predominance of CC10, and specifically ST10, is in accordance with published data on E. coli isolates of swine origin, independently of the pathogenicity or antibiotic-resistance/susceptibility status (Shepard et al., 2012; Kidsley et al., 2018; Magistrali et al., 2018).

## Resistome, Plasmidome and Phenotypic Expression of Resistances

The resistome analysis revealed that 34 of the 35 genomes encoded mechanisms of antibiotic resistance for ≥three different antimicrobial categories (**Table 1**). Seven E. coli were carriers of blaESBL, namely blaCTX-M-<sup>14</sup> (four genomes), blaCTX-M-<sup>32</sup> (one) and blaSHV-<sup>12</sup> (two). Besides, six different mcr variants were identified within the 35 E. coli: mcr-1.1 (in 18 genomes, including two mcr-4.2 carriers); mcr-1.10 (one); mcr-4.1 (one); mcr-4.2 (13 genomes, including the two mcr-1.1 carriers); mcr-4.5 (two) and mcr-5.1 (two).

PlasmidFinder revealed a high plasmid diversity based on the identified replicons, with two to seven different plasmid types per genome (**Table 1**). Within this heterogeneity, mcr-1.1 genes were found mainly on plasmids of the IncHI2 and IncX4 types (six and four of the 12 mcr-1.1 plasmidlocated genes, respectively); however, mcr-1.1 was also found integrated in the chromosome of LREC-145, LREC-148, LREC-149 and LREC-164 genomes. The mcr-1.10 gene of LREC-151 was located on the chromosome, while mcr-4 and mcr-5 variants were on Col8282-like (mcr-4.1), ColE10 like (for all 13 mcr-4.2 and two mcr-4.5 carriers) and pKP13a-like (mcr-5.1) plasmids. Furthermore, we found that

<sup>3</sup>http://clermontyping.iame-research.center/

fmicb-10-02469 October 31, 2019 Time: 18:39 # 4

TABLE 1 | Features of the 35 colistin-resistantE. coli genomes of swine origin based onin silicocharacterization (light columns) and on conventional typing (gray columns).


(Continued)

WGS of

mcr-1, mcr-4,

and

mcr-5 E. coli


WGS of

mcr-1, mcr-4,

and

mcr-5 E. coli

fmicb-10-02469 October 31, 2019 Time: 18:39 # 5

TABLE 1 | Continued


TABLE 1 | Continued


fmicb-10-02469 October 31, 2019 Time: 18:39 # 6

WGS of

November 2019 | Volume 10 | Article 2469


(Continued)

fmicb-10-02469 October 31, 2019 Time: 18:39 # 7


#### TABLE 1 | Continued


(Continued)

WGS of

mcr-1, mcr-4,

and

mcr-5 E. coli

fmicb-10-02469 October 31, 2019 Time: 18:39 # 8


#### TABLE 1 | Continued


<sup>1</sup>Year of isolation of the WGS isolates recovered from pig colibacillosis. <sup>2</sup>Serotypes, <sup>4</sup>clonotypes, <sup>5</sup>sequence types, <sup>6</sup>replicon/plasmid STs, <sup>7</sup>acquired antimicrobial resistance genes and/or chromosomal mutations, <sup>9</sup>virulence genes were determined using SerotypeFinder 2.0, CHtyper 1.0, MLST 2.0, PlasmidFinder 2.0, pMLST 2.0, ResFinder 3.1 and VirulenceFinder 2.0 online tools at the Center of Genomic Epidemiology (https://cge.cbs.dtu.dk/services/), respectively; while <sup>3</sup>phylogroups were predicted using the ClermonTyping tool at the Iame-research Center web (http://clermontyping.iame-research.center/). <sup>2</sup>Serotypes: underlined those antigens that were non-typeable (ONT or HNM) by conventional serotyping but determined by SerotypeFinder. ∗LREC-147 was solved as O157 by conventional typing. <sup>3</sup>Phylogroups: " ∗D" indicates that LREC-133, LREC-134, LREC-143, LREC-151, LREC-170, LREC-171, LREC-174 revealed discrepancies between the assignation obtained with the quadruplex PCR of Clermont et al. (2013) and the in silico assignation using ClermonTyping tool, showing phylogroup E by PCR, but phylogroup D in silico. <sup>6</sup>Plasmid STs: " ∗" indicates alleles with less than 100% but >95% identity and 100% coverage; "/" indicates alleles with multiple perfect hits found. <sup>7</sup>Resistome: chromosomic and plasmid mechanisms associated to colistin resistance are highlighted in bold. Genes coding for ESBLs appear underlined. Acquired resistance genes: beta-lactam: blaTEM-1B, blaTEM-1A, blaOXA-1, blaCTX-M-14, blaCTX-M-32, blaSHV-12, aminoglycosides: aac(3)-II/IV, acc(6 0)-Ib3, aadA, aph(3 0)-I/IIa, aph(300)-Ib, aph(4)-Ia, aph(6)-Id; phenicols: catA1, catB3, cmlA1, floR; fluoroquinolones: aac(6 0)-Ib-cr, qnrS1; macrolides: erm(B), inu(F), inu(G), mdf(A), mph(B); sulfonamides: sul1, sul2, sul3; tetracycline: tet(A), tet(B), tet(M); trimethoprim: dfrA1, dfrA12. Point mutations: quinolones and fluoroquinolones: gyrA S83L: TCG-TTG, gyrA D87G: GAC-GGC, gyrA D87N: GAC-AAT, gyrA D87Y: GAC-TAT, parC S80I: AGC-ATC, parC S80R: AGC-AGG, parC E84G: GAA-GGA, parE L416F: CTT-TTT; colistin: pmrB V161G: GGG-GTG, pmrA S39I: AGC-ATC. <sup>8</sup>The mcr gene location was determined using PlasmidFinder/ResFinder predictions, together with PLACNETw building and references (https://castillo.dicom.unican.es/upload/), and Prokka annotations. <sup>9</sup>Highlighted in bold, those features defining E. coli pathotypes (STEC, ETEC, EPEC). Virulence genes: itcA, coding for heat labile enterotoxin A subunit; sta1, heat stabile enterotoxin ST-Ia; stb, heat stabile enterotoxin II; fedF, fimbrial adhesin AC precursor; K88, K88/F4 protein subunit; air, enteroaggregative inmunoglobulin repeat protein; astA, EAST-1; capU, hexosyltransferase homolog; cba, colicin B; celB, endonuclease colicin E2; cif, type III secreted effector; cma, colicin M; eae, intimin; efa1, EHEC factor for adherence; eilA, Salmonella HilA homolog; espA, type III secretions system; espB, secreted protein B; espF, type III secretions system; espJ, prophage -encoded type III secretion system effector; etpD, type II secretion protein; fedA, F107; gad, glutamate descarboxylase; iha, adherence protein; iroN, enterobactin siderophore receptor protein; iss, increased serum survival; lpfA, long polar fimbriae; nleA, non LEE encoded effector A; nleB, non-LEE encoded effector B; sepA, Shigella extracellular protein A; tccP, Tir cytoskeleton coupling protein; tir, translocated intimin receptor protein. bp, base pairs; CHType, clonotype (fumC-fimH); ST (CC), sequence type and clonal complex according to Achtman scheme; pMLST, plasmid sequence type. <sup>10</sup>Phenotypic resistances interpreted according to the CLSI (intermediate resistance is indicated with an asterisk ∗). AMC, amoxicillin/clavulanate; AMP, ampicillin; AMP/SAM, ampicillin-sulbactam; ATM, aztreonam; CAZ, ceftazidime; CHL, chloramphenicol; CIP, ciprofloxacin; CST, colistin; CTX, cefotaxime; CXM, cefuroxime; CFZ, cefazolin; DOX, doxycycline; FEP, cefepime; FOF, fosfomycin; GEN, gentamicin; LVX, levofloxacin; MI, minocycline; NAL, nalidixic acid; NIT, nitrofurantoin; SXT, trimethoprim/sulfamethoxazole; TIC, ticarcillin; TOB, tobramycin. <sup>11</sup>Pathotype of the isolates, established by conventional PCR, based on specific genes encoding toxins (LT, STa, STb, Stx1, Stx2, Stx2e), fimbriae (F4, F5, F6, F18, F41) and intimin (Eae).

fmicb-10-02469 October 31, 2019 Time: 18:39 # 9

there was no mcr plasmid co-occurrence in LREC-141 and LREC-163, but rather the mcr-1.1 and mcr-4.2 genes were located in independent plasmids (IncHI2 and ColE10-like types, respectively). The mcr location remained undetermined for four isolates.

Since the mcr-1 plasmid gene was first described (Liu et al., 2016), it has been identified in members of the Enterobacteriaceae family encoded in different plasmid types, including IncI2, IncX4, IncHI1, IncHI2, IncFI, IncFII, IncP, IncK (Sun et al., 2018). Different authors corroborate that large conjugative plasmids of types IncHI2, IncX4 and IncI2 would be the maximum responsible for the dissemination of the mcr-1 gene among E. coli isolates from different sources and geographical locations (Doumith et al., 2016; Li et al., 2017; Manageiro et al., 2019). To date, other mcr genes (2–9) have been described (Carroll et al., 2019); among them, the mcr-4 and mcr-5 genes appear mostly encoded in small and non-conjugative ColE-like type plasmids (Sun et al., 2018). Here we found similar results, since mcr-1.1 genes were located mainly on IncHI2 and IncX4 types, and mcr-4 on ColE10-like plasmids. It is of note that the mcr-5.1 gene, predicted in LREC-152 and LREC-177, was linked to a Kp13-like plasmid (CP003996.1), location previously described by Hammerl et al. (2018) for one mcr-5 isolate recovered from a fecal pig sample at farm. Chromosomally-encoded mcr-1 location remains rare, however, it was described soon after the discovery of this plasmid-borne gene (Falgenhauer et al., 2016; Veldman et al., 2016). Here, we determined chromosomal location in five genomes by means of PLACNETw, and according to the predictive annotation of the mcr-contigs, the only common element flanking the mcr-1 was a putative ORF, pap2, which is part of the Tn6360 and encodes a Pap2 superfamily protein. Thus, Pap2 was detected in LREC-145, LREC-148, LREC-149, and LREC-164, while the ISApI1 element typically associated with the initial mobilization of mcr-1, was missing within the five contigs (Snesrud et al., 2018).

Overall, our findings are in accordance with those reported by Magistrali et al. (2018) on 13 mcr-positive E. coli isolated from swine colibacillosis in Belgium, Italy and Spain. Both studies show common features in relation to a prevalent clonal group (CC10), serotypes (O108:H39, O138:H10, O139:H1, O141:H4), and mcr-plasmid types. The confirmation of these similarities are of interest for the global design of alternatives to antibiotics that would curb the dissemination of specific clones in the pig farming.

The in vitro analysis of resistances showed that 30 of the 35 E. coli were multidrug-resistant (MDR) according to Magiorakos et al. (2012) definition (**Table 1**). Phenotypic results corresponded broadly to those predicted by ResFinder (**Supplementary Table S1**) as detailed below.

The quinolones/fluoroquinolones (FQ), together with polymyxins and 3rd–4th-generation cephalosporins, all are included in Category B of restricted antimicrobials in the EMA categorization, considering that the risk to public health resulting from its veterinary use needs to be mitigated by specific restriction (EMA/CVMP/CHMP, 2019). Two major mechanisms are implicated in the resistance to FQ, namely, mutations in the genes for DNA gyrase and topoisomerase IV, and decreased intracellular drug accumulation. In addition, plasmid-mediated quinolone resistances also play a role but usually conferring lowlevel FQ resistance (van Duijkeren et al., 2018). Phenotypically, 17 of the 35 isolates showed resistance to both nalidixic acid and ciprofloxacin, and other eight resistance to nalidixic acid only (**Supplementary Table S1**). In the majority of cases, phenotypic results correlated with those predicted by ResFinder. Particularly, 28 of the 35 genomes carried the gyrA S83L substitution, with 12 of those 28 showing double-serine mutations (gyrA S83L and parC S80I). An additional substitution (gyrA D87N) was detected in two of the 12 gyrA S83L/parC S80I genomes. Thus, nalidixic acid resistance in vitro corresponded to one single substitution (gyrA S83L), and FQ resistance to double or triple substitutions (gyrA S83L/parC S80I/gyrA D87N). Plasmid-mediated quinolone resistant genes acc(6<sup>0</sup> )-Ib-cr and qnrS1 were also present together with chromosomal mutations in LREC-147 and LREC-169, respectively. Double-serine mutations in specific positions of the gyrA and parC genes have been reported as a dominant feature of MDR lineages within E. coli, S. aureus and K. pneumoniae, with favorable fitness balance linked to high levels of resistance to FQ (Fuzi et al., 2017). This finding, in 12 out of the 28 in silico predicted FQ-resistant could be conferring adaptive advantages to certain widely spread pig pathogenic clonal groups of E. coli, such as the ST10-A (CH11-24) (García et al., 2018). This hypothesis is presently assumed for ST131 and other risk clones linked to high FQ-resistance (Johnson et al., 2015; Fuzi et al., 2017).

On the other hand, colistin has been widely used for the control of enteric diseases, mainly in swine and poultry (Rhouma et al., 2016; Hammerl et al., 2018). Several mechanisms of resistance due to chromosomal mutations or acquired resistance genes have been described so far (Olaitan et al., 2014; Poirel et al., 2018). The 35 colistin-resistant E. coli of this study showed MIC values > 4 mg/L. As detailed above, ResFinder confirmed that all the analyzed genomes were mcr-carriers. In addition to the plasmid mechanism (mcr) of resistance, polymyxin resistance in E. coli can be due to genes encoding LPS-modifying enzymes, particularly to mutations in the two-component systems PmrAB and PhoPQ, or in the MgrB regulator. Quesada et al. (2015) detected two colistin-resistant E. coli recovered in 2011 and 2013 from the stools of two pigs, which showed mutations in PmrB V161G and PmrA S39I, reporting the finding as a rare event. Subsequently, Delannoy et al. (2017) analyzed 90 strains of E. coli isolated from diseased pigs: 81 were phenotypically resistant to colistin and 72 mcr-1 carriers (including two colistinsusceptible). Although different mutations were found in the amino acid sequences of the MgrB, PhoP, PhoQ, and PmrB proteins of eight isolates, only two of them were mcr-1 positive (but colistin-susceptible). Surprisingly, we found here the double mechanism of colistin resistance in 16 E. coli, harboring mcrgenes together with one amino acid substitution: PmrB V161G (14 genomes) or PmrA S39I (two genomes). In a recent study on Parisian inpatient fecal E. coli (Bourrel et al., 2019), the authors found 12.5% of colistin-resistant E. coli carriers among 1,217 patients; however, mcr-1 gene was identified in only seven of 153 isolates, while 72.6% harbored mutations in the PmrA and PmrB proteins. According to the authors, their findings

indicate two evolutionary paths leading to colistin resistance in human fecal E. coli, one corresponding to a minority of plasmid-encoded mcr-1 isolates of animal origin, and a second corresponding to a vast majority of human isolates exhibiting chromosomally encoded mechanisms (Bourrel et al., 2019). Thus, and given the limited data regarding the co-occurrence of double resistance mechanism, it is of note that 16 of the 35 mcr-bearing genomes of our study showed mutations in the PmrA and PmrB proteins. Furthermore, two E. coli (LREC-141 and LREC-163) shown to be carriers of two different mcr-bearing plasmids together with PmrB V161G mutation. An explanation for this rare finding is that these isolates would be reflecting a cumulative evolution to antibiotic pressure and, as a consequence, enhancing the transmission (vertical and horizontal) of colistin resistance. In any case, further investigation is needed to evaluate the implication of chromosomal mutations and mcr co-occurrence regarding colistin resistance phenotype.

In this study, 22 out of the 25 isolates showing phenotypic resistance to beta-lactams (**Supplementary Table S1**), were positive in the analysis in silico for the presence of blaTEM-<sup>1</sup> genes, alone (14 genomes), or in combination with other bla genes (blaCTX-M-14, blaSHV-12, blaCTX-M-<sup>32</sup> and blaOXA-1); additionally, two genomes showed the presence of blaCTX-M-<sup>14</sup> and blaSHV-12, respectively. With the exception of LREC-147, LREC-164, and LREC-170, which were phenotypically resistant to narrowspectrum beta-lactamases but negative for the presence of genes, a good correlation was observed between genes predicted and resistance shown in vitro. It is of note that blaTEM-135, determined in LREC-156 by conventional typing, was not identified in silico. Beta-lactams are the most widely used family in current clinical practice. Numerous genes in E. coli confer resistance to this group, being some of them, such as blaTEM-<sup>1</sup> widespread in E. coli from animals coding for narrow-spectrum beta-lactamases that can inactivate penicillins and aminopenicillins. However, genes encoding ESBLs/AmpCs have increasingly emerged in E. coli from humans and animals, including food-producing animals (Cortes et al., 2010).

Thirty out of the 35 genomes showed high frequency of resistance genes to aminoglycosides, specifically encoding AAC(3)-II/IV and AAC(6)-Ib, which are the most frequently encountered acetyltransferases among E. coli of human and animal origins. The subclass AAC(3)-II, which is characterized by resistance to gentamicin, netilmicin, tobramycin, sisomicin, 2 0 -N-ethylnetilmicin, 6<sup>0</sup> -N-ethylnetilmicin and dibekacin (Shaw et al., 1993), and AAC(6<sup>0</sup> ) enzymes that specify resistance to several aminoglycosides and differ in their activity against amikacin and gentamicin C1 (Ramirez and Tolmasky, 2010) seemed to correlate with the phenotypic detection of resistance to gentamicin and/or tobramycin (12 of the 17 resistant isolates) (**Supplementary Table S1**). We also detected high prevalence of genes encoding nucleotidyltransferases (aadA), which specify resistance to spectinomycin and streptomycin, alone or together with phosphotransferases (APHs) (Ramirez and Tolmasky, 2010), but they were not tested in the phenotypic antimicrobial susceptibility tests.

It is noteworthy that the 35 genomes of our study were carriers of mdf(A). Edgar and Bibi (1997) described that cells expressing MdfA from a multicopy plasmid are substantially more resistant to a diverse group of cationic or zwitterionic lipophilic compounds. Besides, the authors found that MdfA also confers resistance to chemically unrelated, clinically important antibiotics such as chloramphenicol, erythromycin, and certain aminoglycosides and fluoroquinolones. This capability could correlate with the in vitro resistance observed for some isolates to tetracyclines and aminoglycosides, in absence of other specific genes. In our collection, of the 24 isolates showing phenotypic resistance to minocycline and, or doxycycline (**Supplementary Table S1**), 20 showed carriage of tet genes: 12 tet(B), six tet(A), one tet(A) + tet(B) and one tet(B) + tet(M). However, two tet(A) isolates were susceptible to those antibiotics (LREC-163, LREC-169). Additionally, tet genes were not detected in silico in four phenotypically resistant isolates. In general, tet(A) and tet(B) are the most prevalent tetracycline resistance genes in E. coli of animal origin, and specifically in isolates from pigs (Tang et al., 2011; Holzel et al., 2012; Jurado-Rabadan et al., 2014).

Although the use of chloramphenicol was banned in the European Union in food-producing animals in 1994, fluorinated derivative florfenicol is allowed for the treatment of bacterial infection in these animals (Schwarz et al., 2004; OIE, 2019). In the present study, all 19 chloramphenicol-resistant isolates (**Supplementary Table S1**) correlated with the presence of genes catA1 (12 genomes),catB3 (one genomes),cmlA (ten genomes) or floR (three genomes) detected in silico. Travis et al. (2006) showed that chloramphenicol resistant genes are frequently linked to other antibioresistance genes. Thus, through transformation experiments conducted with E. coli from pigs demonstrated that aadA and sul1 were located with catA1 on a large ETEC plasmid, and plasmids carrying cmlA also carried sul3 and aadA. According to the authors, this linkage might partly explain the long-term persistence of chloramphenicol resistance in ETEC despite its withdrawal years ago. In our study, ResFinder also showed an association of genes cmlA, sul3 and aadA present in the same contig (7 of the 10 genomes positive for cmlA), and cmlA/aadA in all cases. Additionally, aadA and sul1 were located with floR in LREC-146.

In E. coli from food-producing animals, sulfonamide resistance is mediated by sul genes (sul1, sul2, sul3), widely disseminated, and frequently found together with other antimicrobial resistance genes, while dfr genes confer trimethoprim resistance in E. coli and other gramnegative bacteria (van Duijkeren et al., 2018). Within our collection, 20 of the 35 isolates were in vitro resistant to trimethoprim/sulfamethoxazole (**Supplementary Table S1**), and most of them correlated with the presence of sul + dfrA genes in their genomes, with the exception of LREC-133 and LREC-170 (negative for the in silico detection of sul, dfrA genes) and LREC-146 (in which only sul1 and sul2 genes were predicted). Besides, ResFinder showed that sul1 (present in 16 genomes), sul3 (14 genomes) and sul2 (three genomes) were located together with dfrA, and other resistance genes, as mentioned previously.

The fosfomycin resistance showed in vitro by two isolates of the study collection, was not predicted for LREC-144 and LREC-145 (**Supplementary Table S1**) by ResFinder, which

analyzes the presence of fos genes encoding for fosfomycinmodifying enzymes. The use of this antibiotic has been limited to the treatment of infections by Gram-positive and negative pathogens, included E. coli, mainly in pig and poultry farming (Poirel et al., 2018). However, phosphonic acid derivates such as fosfomycin, have been recently categorized by the EMA (EMA/CVMP/CHMP, 2019) as Category A (antimicrobial classes not currently authorized in veterinary medicine in EU).

## CONCLUSION

Swine colibacillosis control has been traditionally managed through the extensive use of antibiotics. Our results are a reflection of the situation within the industrial pig farming, where global hygiene procedures and vaccinations are essential for improvement in antimicrobial stewardship. The summative presence of antibioresistances could be conferring adaptive advantages to prevalent pig E. coli lineages, such as the ST10-A (CH11-24). Based on the different replicons identified by PlasmidFinder (up to seven), it is of note the high plasmid diversity found within these isolates; further research is needed to know mechanisms of maintenance and advantages conferred to them.

Here, we report for first time a rare finding so far, which is the co-occurrence of double colistin-resistance mechanisms (mcr-genes and chromosomal mutations in the PmrA and PmrB proteins) in a significant number of E. coli isolates. This fact could be increasing the risk of colistin resistance-acquisition by means of food transmission. Globally, we found a very good correlation between resistances determined in vitro and genes predicted using CGE tools, and the same observation applies to the E. coli pathotype determination.

## DATA AVAILABILITY STATEMENT

The nucleotide sequence of the 35 LREC genomes have been deposited in the NCBI sequence databases with accession codes

## REFERENCES


SAMN11523829 to SAMN11523863. These sequences are part of BioProject ID PRJNA540146.

## AUTHOR CONTRIBUTIONS

IG-M, DD-J, and SF-S undertook the laboratory work. AM and JB conceived and designed the study. IG-M, DD-J, VG, MT, and AM performed the data analysis. IG-M, DD-J, VG, MT, JB, and AM drafted the manuscript. All authors provided critical input and approved the final version.

## FUNDING

This study was supported by projects PI16/01477 from Plan Estatal de I+D+I 2013–2016, Instituto de Salud Carlos III (ISCIII), Subdirección General de Evaluación y Fomento de la Investigación, and FEDER; AGL2016- 79343-R from the Agencia Estatal de Investigación (AEI, Spain) and FEDER; ED431C 2017/57 from the Consellería de Cultura, Educación e Ordenación Universitaria (Xunta de Galicia) and FEDER. IG-M and VG acknowledge the Consellería de Cultura, Educación e Ordenación Universitaria, Xunta de Galicia for their pre-doctoral and post-doctoral grants (Grant Numbers ED481A-2015/149 and ED481B-2018/018, respectively). SF-S acknowledges the FPU programme from the Secretaría General de Universidades, Ministerio de Educación, Cultura y Deporte, Gobierno de España (Grant Number FPU15/02644).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02469/full#supplementary-material

specificity and detection of new phylo-groups. Environ. Microbiol. Rep. 5, 58–65. doi: 10.1111/1758-2229.12019



standard definitions for acquired resistance. Clin. Microbiol. Infect. 18, 268–281. doi: 10.1111/j.1469-0691.2011.03570.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 © 2019 García-Meniño, Díaz-Jiménez, García, de Toro, Flament-Simon, Blanco and Mora. 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.

fmicb-10-02546 November 6, 2019 Time: 17:34 # 1

## Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Agnese Lupo, ANSES Site de Lyon, France Eliana De Gregorio, University of Naples Federico II, Italy

#### \*Correspondence:

Qiwen Deng qiwendeng@hotmail.com Zhijian Yu yuzhijiansmu@163.com †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 24 May 2019 Accepted: 22 October 2019 Published: 08 November 2019

#### Citation:

Bai B, Lin Z, Pu Z, Xu G, Zhang F, Chen Z, Sun X, Zheng J, Li P, Deng Q and Yu Z (2019) In vitro Activity and Heteroresistance of Omadacycline Against Clinical Staphylococcus aureus Isolates From China Reveal the Impact of Omadacycline Susceptibility by Branched-Chain Amino Acid Transport System II Carrier Protein, Na/Pi Cotransporter Family Protein, and Fibronectin-Binding Protein. Front. Microbiol. 10:2546. doi: 10.3389/fmicb.2019.02546 In vitro Activity and Heteroresistance of Omadacycline Against Clinical Staphylococcus aureus Isolates From China Reveal the Impact of Omadacycline Susceptibility by Branched-Chain Amino Acid Transport System II Carrier Protein, Na/Pi Cotransporter Family Protein, and Fibronectin-Binding Protein

Bing Bai1,2† , Zhiwei Lin1,2† , Zhangya Pu<sup>3</sup>† , Guangjian Xu1,2† , Fan Zhang1,2, Zhong Chen1,2 , Xiang Sun<sup>1</sup> , Jinxin Zheng<sup>1</sup> , Peiyu Li1,2, Qiwen Deng1,2 \* and Zhijian Yu1,2 \*

<sup>1</sup> Department of Infectious Diseases and Shenzhen Key Lab of Endogenous Infections, Shenzhen Nanshan People's Hospital and the 6th Affiliated Hospital of Shenzhen University Health Science Center, Shenzhen, China, <sup>2</sup> Quality Control Center of Hospital Infection Management of Shenzhen, Shenzhen Nanshan People's Hospital, Guangdong Medical University, Shenzhen, China, <sup>3</sup> Key Laboratory of Viral Hepatitis of Hunan Province, Department of Infectious Diseases, Xiangya Hospital, Central South University, Changsha, China

Omadacycline (Omad), a new tetracycline (Tet)-class broad-spectrum aminomethylcycline, has been reported to exhibit excellent potency against Grampositive bacteria, including Staphylococcus aureus and Enterococci. The aim of this study was to evaluate the in vitro activity and heteroresistance characteristics of Omad in clinical S. aureus isolates from China and investigate Omad resistance mechanisms. A sample of 263 non-duplicate clinical S. aureus isolates [127 methicillin-resistant (MRSA) and 136 methicillin-sensitive (MSSA)] were collected retrospectively. Our data indicated that Omad exhibited excellent in vitro activity against both MRSA and MSSA. Omad heteroresistance frequencies were 3.17% (4/126) in MRSA and 12.78% (17/133) in MSSA. No mutations in Tet target sites, (five 16SrRNA copies and 30S ribosomal protein S10) were present in heteroresistance-derived clones, whereas Tet target site mutations contribute to induced Omad resistance in S. aureus in vitro. RNA sequencing (RNA-Seq) revealed that overexpression of branched-chain amino acid transport system II carrier protein and Na/Pi cotransporter family protein contributes to Omad heteroresistance emergence. Whole-genome sequencing demonstrated that the genetic mutation of fibronectin-binding protein (FnBP) could increase the Omad MIC. In conclusion, Omad heteroresistance risk should be considered in clinical isolates with MICs ≥ 0.5 mg/L and Omad susceptibility in S. aureus may be affected by efflux pump proteins (i.e., a branched-chain amino acid transport system II carrier protein and an Na/Pi cotransporter family protein), and FnBP.

Keywords: omadacycline, Staphylococcus aureus, antimicrobial activity, multilocus sequence typing, tetracycline specific resistance genes

## INTRODUCTION

fmicb-10-02546 November 6, 2019 Time: 17:34 # 2

Staphylococcus aureus is a pervasive human pathogen that causes infectious diseases ranging in severity from superficial skin abscesses to bacteremia and septic shock (Calfee, 2017). Although the incidence of methicillin-resistant S. aureus (MRSA) infection appears to be declining worldwide, the incidence of bacteremia and severe community-acquired infection caused by methicillin-susceptible S. aureus (MSSA) continues to be an important human health threat (Bal et al., 2017). Both MRSA and MSSA infections remain a major clinical problem exacerbated by the ongoing evolution and transmission of traits engendering resistance or reduced susceptibility to current last-line antimicrobial agents, including linezolid, daptomycin, tigecycline (Tig), and vancomycin. Thus, there remains an urgent need for the development of new antimicrobial agents (Bal et al., 2017; Calfee, 2017).

Omadacycline [7-dimethylamino, 9-(2,2-dimethylpropyl)-aminomethylcycline; Omad] is a recently developed semisynthetic aminomethylcycline belonging to the tetracycline (Tet) family (Pfaller et al., 2017a). It has extraordinarily broadspectrum antimicrobial activity against Gram-positive and -negative bacteria, including difficult-to-treat multidrug resistant bacteria, such as MRSA and vancomycin-resistant enterococci (Pfaller et al., 2017a). Like other Tet drugs, Omad is a potent inhibitor of the bacterial ribosome that inhibits bacterial protein synthesis by binding 30S ribosomal subunits during translation (Draper et al., 2014; Honeyman et al., 2015; Heidrich et al., 2016; Villano et al., 2016). Omad has the advantage of being minimally affected by classical Tet resistance mechanisms, including efflux pumps and ribosomal protection mechanisms. Omad also exhibits lower minimum inhibitory concentration (MIC) values against multidrug resistant bacteria than minocycline and Tig (Draper et al., 2014; Honeyman et al., 2015; Heidrich et al., 2016; Villano et al., 2016), making it a novel potential last-resort antibiotic for difficult-to-treat bacteria infections (Noel et al., 2012; Macone et al., 2014; Pfaller et al., 2017b; Zhang et al., 2018).

Heteroresistance means that there are population-wide variable responses to antibiotics. Several reports, including the earliest studies describing the phenomenon, applied this definition without specifying a particular antibiotic concentration range (El-Halfawy and Valvano, 2013, 2015). Previously, we obtained MICs and heteroresistance occurrence data for the new generation Tet-class drug erevacycline in clinical S. aureus isolates from China, underscoring the importance and necessity of investigating the characteristics of new-generation Tet derivatives (Zhang et al., 2018; Zheng et al., 2018). There are limited data regarding Omad activity against clinical S. aureus isolates from China. Heteroresistance development in last-resort antibiotics can hinder efficacy and, ultimately, lead to treatment failure (Claeys et al., 2016; Zhang et al., 2018; Zheng et al., 2018). Thus, it is important to establish potential factors associated with Omad heteroresistance development.

Reduced susceptibility to Tig, an archetype new-generation Tet-class drug, in several species of bacteria has been associated with genetic mutations affecting 30S ribosomal subunits, including altered copy numbers of genes encoding 16S rRNA and 30S ribosomal proteins S3 and S10 (Nguyen et al., 2014; Lupien et al., 2015; Grossman, 2016; Argudin et al., 2018). Tig resistance has been related to regulators of cell envelop proteins, including efflux pumps (e.g., SoxS, MarA, RamA, and Rob) in Gramnegative enterobacteria and MepR/MepA in S. aureus (Nguyen et al., 2014; Grossman, 2016; Linkevicius et al., 2016; Dabul et al., 2018). The possible influences of 30S ribosomal subunit mutations and the overexpression of efflux proteins on Omad heteroresistance and resistance in S. aureus has not been resolved and needs to be further studied.

The main purpose of this study was to investigate the in vitro antimicrobial activity of Omad and to use population analysis profile (PAP) analysis to evaluate the occurrence of Omad heteroresistance in S. aureus isolates from China. We examined Omad heteroresistance mechanisms in S. aureus by conducting polymerase chain reaction (PCR) experiments to detect genetic mutations in 30S ribosome units, administering efflux protein inhibitors (Zhang et al., 2018; Zheng et al., 2018), and conducting RNA sequencing (RNA-Seq) studies. Furthermore, we used in vitro induction of resistance under Omad pressure and next generation sequencing (NGS) to compare Omad-sensitive and -resistant isolates and uncover molecular factors involved in Omad resistance.

## MATERIALS AND METHODS

## Bacterial Isolates

A total of 263 non-duplicate clinical S. aureus isolates (127 MRSA and 136 MSSA) were collected from Shenzhen Nanshan People's Hospital, a tertiary hospital with 1,200 beds in China, between 2008 and 2016. The specimen sources are summarized in **Supplementary Figure S1**. S. aureus ATCC29213 was used as a quality control organism. All procedures involving human participants were performed in accordance with the ethical standards of Shenzhen University and the 1964 Helsinki declaration and its later amendments, or comparable ethical standards. For this type of study, formal consent is not required.

## Antimicrobial Susceptibility

fmicb-10-02546 November 6, 2019 Time: 17:34 # 3

Staphylococcus aureus antimicrobial susceptibilities to a panel of antibiotics (i.e., amikacin, erythromycin, ciprofloxacin, rifampicin, Tet, tobramycin, vancomycin, linezolid, nitrofurantoin, amoxicillin/clavulanate, and quinupristin) with the VITEK 2 system (BioMérieux, Marcy l'Etoile, France) and susceptibility breakpoints based on CLSI guidelines (2016). Omad was obtained from The Medicines Company (Med Chem Express, Monmouth Junction, NJ, United States). Omad MICs were determined with the agar dilution method according to CLSI guidelines (Klionsky et al., 2016). We employed three Omad MIC levels (≤0.25 mg/L, 0.5 mg/L, and ≥1 mg/L) in our antimicrobial susceptibility analysis. The following Acute Bacterial Skin and Skin Structure Infections Omad susceptibility breakpoints recommended by FDA criteria were adopted: ≤0.5 mg/L for susceptibility, 1 mg/L for intermediate status, and ≥2 mg/L for resistance.

## PAP Development

Omad heteroresistance in S. aureus was determined by PAP development as described previously (Zhang et al., 2018; Zheng et al., 2018) with a MIC cut-off criterion of ≤0.5 mg/L. Briefly, 50-µL aliquots (108 bacterial colony forming units) were spread onto Müller-Hinton broth plates containing serial dilutions of Omad (in mg/L): 0.5, 1, 2, and 3. Colonies were counted on Omad-containing plates after 24 h of incubation at 37◦C. According to the criteria described above, we defined 2 mg/L as the susceptibility breakpoint for PAP determination of S aureus. For Omad-resistant subpopulations detected among Omad-susceptible S. aureus isolate colonies grown on agar plates with 2 mg/L Omad with a detection limit of ≥5 colony forming units/ml, the parental isolates were considered to have Omad heteroresistance. Two heteroresistance-derived colonies were selected randomly from plates and their Omad and Tig MICs were measured by agar dilution according to CLSI guidelines and then subjected to PCRs, efflux inhibition, and transcriptional analysis (Zhang et al., 2018).

## Polymerase Chain Reaction

Genomic DNA was extracted from isolates with Lysis Buffer for Microorganisms to Direct PCR (Takara Bio Inc., Japan). Tet resistance genes encoding Tet(K), Tet(L), Tet(M), and Tet(O) were detected by PCR analysis as described previously (Bai et al., 2018). The presence of 30S ribosomal subunit mutations, including five separate copies of the 16S rRNA gene, the genes encoding the 30S ribosomal proteins S3 and S10, and the genes encoding recombinase (RecB) and fibronectinbinding protein (FnBP) were analyzed by PCR and sequence alignment (primer sequences listed in **Supplementary Table S1**). Multi-locus sequence typing (MLST) was conducted to identify the distributions of sequence types (STs) among MRSA and MSSA isolates. PCR conditions recommended for locus amplification<sup>1</sup> were employed.

## Efflux Inhibition

The role of efflux pumps in Omad heteroresistance was evaluated with the efflux pump inhibitors phenylalaninearginine-β-naphthylamide (PaβN) and carbonyl cyanidemchlorophenylhydrazine (CCCP; both from Sigma). Omad MICs were determined by the agar dilution method in the presence and absence of 50 mg/L PAβN or 16 mg/L CCCP. Inhibition was confirmed based on a ≥4-fold MIC reduction (Osei Sekyere and Amoako, 2017; Zhang et al., 2018).

## In vitro Induction of Omad-Resistance Under Omad Pressure

Seven parental S. aureus isolates, including six clinical isolates (MSSAs: CHS221, CHS165, and 149. MRSAs: CHS759, CHS810, and CHS820) and a well-characterized antibiotic-susceptible MS4 strain, were used to select Omad-resistant isolates. These isolates were subcultured serially in Mueller-Hinton broth containing gradual increasing Omad concentrations with the initial concentration being MIC equivalents followed by successive increases to 2×, 4×, 8×, and 16× MICs (Yao et al., 2018), with four passages at each concentration. Isolates from the passages of each concentration were stored at −80◦C in Mueller-Hinton broth containing 40% glycerol for subsequent Tet-target site genetic mutation detection, subsequent MIC assays, next generation sequencing, and PCR analysis. Killing curves were performed on the CHS221 (wild-type MSSA strain), CHS221- O (Omad heteroresistance MSSA strain), CHS221-11: (Omad resistance MSSA strain), CHS759 (wild-type MRSA strain), CHS759-O (Omad heteroresistance MRSA strain), and CHS759- 11: (Omad resistance MRSA strain). Tubes containing Omad at concentrations corresponding to 0 and 4 mg/L were inoculated with a suspension of each test strain, yielding to a final bacterial density of 8 × 106 cfu/ml. The killing curves shown Omadresistant strain could grow well at concentrations corresponding to 4 mg/L (**Supplementary Figure S2**).

## RNA-Seq

Wild-type strain CHS221 (S221) and its heteroresistancederived S. aureus isolate CHS221-O (S221-O1) were grown and prepared for total RNA extraction with TRIzol reagent (Invitrogen, Carlsbad, CA, United States) as described previously (Zheng et al., 2018). RNA-Seq of the aforementioned parental and heteroresistance-derived isolates was performed as reported previously (Lin et al., 2017). Raw data (raw reads) of fastq format were firstly processed through inhouse perl scripts. In this step, clean data (clean reads) were obtained by removing reads containing adapter, reads containing ploy-N and low quality reads from raw data. All the downstream analyses were based on the clean data with high quality. The raw data from the samples were analyzed in Subread software; raw counts for each group

<sup>1</sup>http://www.mlst.com/server

fmicb-10-02546 November 6, 2019 Time: 17:34 # 4

were normalized and processed in the EdgeR Bioconductor software package. 1.3-fold differences in expression level by RNA-Sequencing were considered to be differentially expressed genes (DEGs). The RNA sequencing outcomes for two strains were deposited in the NCBI database (BioProject accession number PRJNA505108).

## Quantitative Real Time (qRT)-PCR Analysis

We selected 30 DEGs based on our RNA-Seq results and employed qRT-PCR to compare transcriptional expression levels between the parental and heteroresistance-derived strains as described in detail previously (Zheng et al., 2018). The transcriptional expression levels of the eight candidates genes (USA300HOU\_RS00705, USA300 HOU\_RS03535, USA300 HOU\_RS01625, USA300 HOU\_RS00550, USA300HOU\_ RS13205, USA300HOU\_RS13945, USA300HOU\_RS10505, and USA300HOU\_RS00660) were further analyzed and compared among the CHS165 (MSSA), 149 (MSSA), CHS759 (MRSA), CHS810 (MRSA), and CHS820 (MRSA) parental strains and their derivative heteroresistant and resistant strains. Total RNA of bacteria was extracted using the RNeasyH Mini Kit (QIAGEN, Hilden, Germany) following the manufacturer's instructions. The extracted RNA was reverse transcribed into cDNA using iScript reverse transcriptase (Bio-Rad, Hercules, CA, United States) with incubation for 5 min at 25◦C, followed by 30 min at 42◦C and 5 min at 85◦C. Subsequently, qRT-PCRs were performed using SYBR green PCR reagents (Premix EX TaqTM, Takara Biotechnology, Dalian, China) in the Mastercycler realplex system (Eppendorf AG, Hamburg, Germany) with amplification conditions of 95◦C for 30 s, 40 cycles of 95◦C for 5 s and 60◦C for 34 s, followed by melting curve analysis. The control gene gyrB was used to normalize gene expression. Threshold cycle (Ct) numbers were determined by detection system software and analyzed with the 2−44Ct method and three replicates have been made. The qRT-PCR primers used are listed in **Supplementary Table S2**.

## Next Generation Sequencing

An Omad resistant S. aureus strain, MS4O8, was derived from an Omad-susceptible isolate, MS4. Chromosomal DNA was extracted from MS4O8 cells for NGS. Nextera shotgun libraries and whole genome sequencing were performed by Novogene Company (Beijing, China). Illumina PE150 sequencing data were mapped against the CP009828 S. aureus MS4 strain reference genome in bwa mem software (v0.7.5a)<sup>2</sup> with standard parameters. Small nucleotide polymorphisms and small insertions/deletions were detected in MS408 cells, relative to MS4, in MUMmer (version 3.23). A custom script was used to detect substitutions, insertions, and deletions that might be impacting protein coding regions. Binary alignment/map files of the sequenced strains were deposited in the NCBI database (BioProject accession number PRJNA511962).

Full-length candidate genes, including USA300HOU\_RS00550 (encodes a Na/Pi cotransporter family protein), USA300HOU\_ RS01625 (encodes a branched-chain amino acid transport system II carrier protein), USA300HOU\_RS03535, USA300HOU\_ Tet(K), NI36\_12460 (FnBP), and NI36\_00170 (RecB), were amplified with extra double enzyme sites from total DNA extracted from USA300HOU and MS4 isolates. RecB-M is RecB with a mutation (R10R, I23V, I23N, H24N, H24Q, V29L, and V35M) and FnBP-M is FnBP with a mutation (T672S and I665V). RecB-M and FnBP-M DNA fragments were amplified from MS4O8 by PCR. The candidate gene fragments were each integrated into separate pIB166 vectors, and their encoded target protein were induced with 2 mM chromium chloride (Wu et al., 2012). The primers used for vector constructs in this study are listed in **Supplementary Table S3**. Positively vector transformed S. aureus clones were selected with chloramphenicol and verified by PCR and Sanger sequencing. The overexpression plasmids were transformed separately into three to five Omad-sensitive isolates and their integrations was confirmed by PCR and Sanger sequencing. Candidate gene transcriptional levels were measured by qRT-PCR, as described above. Omad and Tig MICs for these derivatives were determined and heteroresistance occurrence in these derivatives was tested by PAP analysis under Omad pressure as described above.

## Statistical Analysis

Continuous data were analyzed with Student's t-tests and one-way factorial analyses of variance (ANOVAs) in SPSS software package (version 17.0, Chicago, IL, United States). P-values < 0.05 were regarded as statistically significant.

## RESULTS

## In vitro Activity of Omad Against Clinical S. aureus Isolates

Of the 127 MRSA isolates examined, 46 (36.22%), 80 (62.99%), and 1 (0.78%) were found to have Omad MIC levels of ≤0.25 mg/L (sensitive), 0.5 mg/L (sensitive), and 1 mg/L (intermediate), respectively. Of the 136 MSSA isolates examined, 23 (16.91%), 110 (80.88%), and 3 (2.20%) were categorized into these levels, respectively. Thus, there were higher frequencies of MSSA isolates than MRSA isolates with Omad MICs in the 0.5 and ≥1 mg/L levels. We analyzed the distribution of the above three Omad MIC levels among strains with sensitive and intermediate status relative to other common antibiotics (amikacin, erythromycin, ciprofloxacin, rifampicin, Tet, tobramycin, nitrofurantoin, quinupristin, and amoxicillin/clavulanate, vancomycin, and linezolid). The frequencies of isolates at each Omad MIC level found to be resistant to these antibiotics are reported in **Table 1**, together with the resistance breakpoints used. All S. aureus isolates in this study were susceptible to vancomycin and linezolid, and all of the MSSA isolates were susceptible to

Gene Overexpression

<sup>2</sup>http://biobwa.sourceforge.net/

fmicb-10-02546 November 6, 2019 Time: 17:34 # 5


TABLE 1 | Staphylococcus aureus antibiotic resistance and correspondence to Omad MIC level.

amoxicillin/clavulanate. Interestingly, as reported in **Table 1**, Tet-resistant MRSA were more frequent than Tet-resistant MSSA, and Omad MICs ≥ 0.5 mg/L were more frequent among MSSA isolates than among MRSA isolates, suggesting a non-conformity in the antimicrobial susceptibility dynamics of Tet and Omad. The characteristics of S. aureus with Omad MICs of 1 mg/L are summarized in **Supplementary Table S4**. Briefly, no genetic mutations in 30S ribosome units were detected and efflux pump inhibition reversed Omad resistance, as evidenced by significant reductions in MICs to ≤0.03 mg/L with CCCP and to 0.25–1 mg/L with PAβN.

## Omad MICs of S. aureus Isolates Harboring Tet-Resistance Genes

fmicb-10-02546 November 6, 2019 Time: 17:34 # 6

The frequencies of genes encoding the Tet-resistance factors Tet(M), Tet(L), Tet(K), and Tet(O), alone and in combination, in MRSA and MSSA isolates are reported in **Supplementary Table S5**. There were 109 S. aureus isolates harboring at least one Tet-resistance factor gene; their MIC<sup>90</sup> values were consistently 0.5 mg/L. Omad exhibited excellent in vitro activity against both Tet-resistance gene carrying and non-carrying S. aureus isolates. Omad MICs for both MRSA and MSSA harboring Tet-resistance factors were ≤0.5 mg/L for all isolates, with the exception of three Tet(K) gene-carrying isolates (1 MRSA and 2 MSSA), indicating that overexpression of Tet(K) might impact Omad susceptibility. It is noteworthy that the Omad MIC values obtained for all 46 MSSA isolates carrying the Tet(K) gene, Tet(L) gene, or both were ≥0.5 mg/L. Meanwhile, of the 63 MRSA isolates with the Tet(M) gene, Tet(K) gene, Tet(L) gene, or some combination of these genes, just 47 (74.60%) had Omad MIC values ≥0.5 mg/L.

## Clonality of Omad MIC Distribution in Clinical S. aureus Isolates

MLST results for the 263 isolates are summarized in **Supplementary Figure S3**. To evaluate the relationship of ST clonality with Omad MIC distribution, we examined ST distributions relative to Tet and Omad MICs (**Supplementary Table S6**). Omad MICs ≥ 0.5 mg/L were found for 72.58% (45/62) of ST239-MRSA, 57.5% (23/40) of ST59-MRSA, and 57.14% (4/7) of ST1-MRSA isolates. Meanwhile, Omad MICs ≥ 0.5 mg/L were found for 89.66% (26/29) of ST7-MSSA, 89.47% (17/19) ST59-MSSA, 76.92% (10/13) of ST398-MSSA, 85.71% (6/7) of ST88-MSSA, and 71.43% (5/7) of ST120-MSSA isolates. Hence, Omad sensitivity differed between MRSA and MSSA of the same ST (e.g., ST59-MRSA vs. ST59-MSSA). MLST indicated that 72/81 (88.89%) MRSA isolates with Omad MICs ≥ 0.5 mg/L belonged to the top three MRSA STs (ST239, ST59, and ST1), whereas only 52/113 (46.12%) of MSSA isolates with Omad MICs ≥ 0.5 mg/L belonged to the top three MSSA STs (**Supplementary Table S6** and **Table 1**), revealing a more pronounced clustering of Omad MIC creep in the top three MRSA STs than in the top three MSSA STs (nearly 90% vs. less than half).

## Omad Heteroresistance Frequency and Mechanism in S. aureus

Omad heteroresistance was identified in 0.0% (0/46) of MRSA with an Omad MIC ≤ 0.25 mg/L and 3.75% (3/80) of MRSA with an Omad MIC of 0.5 mg/L. Omad heteroresistance was identified in 0.0% (0/30) of MSSA with an Omad MIC ≤ 0.25 mg/L and 17.48% (18/103) of MSSA with an Omad MIC of 0.5 mg/L. We determined the Omad and Tig MICs of two clones from each heteroresistant subpopulation and found that their Omad MICs were in the range of 1–8 mg/L and their Tig MICs were in the range of 2–8 mg/L (shown in **Supplementary Table S7** and data for six isolates subjected to in vitro resistance induction are shown in **Table 2**). Moreover, no genetic mutations were found in 30S ribosomal subunit genes (five 16SrRNA gene copies, 30S ribosomal protein S3 and S10 genes) in heteroresistance-derived clones (**Table 2** and **Supplementary Table S7**). In efflux pump inhibition experiments, Omad MICs in heteroresistance-derived S. aureus clones were reduced to ≤0.03 mg/L by CCCP and reduced to 0.25–1 mg/L by PAβN (**Supplementary Table S7**).

## Association of Selected Candidate Genes With Omad Heteroresistance

Efflux pump inhibition experiments indicated that efflux pumps or membrane proteins might participate in the development of heteroresistance. Therefore, RNA-Seq was performed and unigene transcription levels were compared between the parental strain CHS221 (S221) and its heteroresistant derivative CHS221-O1 (S221-O). Ninety six DEGs were found by this approach between S221 and S221-O, including 58 upregulated and 38 down-regulated genes in the derivative strain (**Supplementary Figure S4**). KEGG pathway analysis showed that the pathways most frequently linked to DEGs were related to phosphate ion transport (3 DEGs), inorganic anion transport (3 DEGs), dihydrofolate reductase activity (2 DEGs), and glycine biosynthesis process (2 DEGs).

Subsequent qRT-PCRs for 30 candidate genes were carried out to test the accuracy of our transcriptomic analyses implicated eight efflux-pump encoding DEGs in heteroresistance. The expression levels of these eight candidate genes determined by RNA-Seq and qRT-PCR are shown in **Table 3**. The results of qRT-PCRs performed to quantitate transcription in six S. aureus strain groups—inclusive of parental strains, their heteroresistance derivatives, and resistant isolates (**Table 2**)—enabled us to probe the relationship of their expression levels with Omad susceptibility (**Figure 1**). The data suggest that transcription levels of the three candidate genes USA300HOU\_RS03535, USA300HOU\_ RS01625 (encodes a Na/Pi cotransporter family protein), and USA300HOU\_RS00550 (encodes a branched-chain amino acid transport system II carrier protein) may impact heteroresistance occurrence.

## Mechanism of Omad-Induced Resistance in S. aureus Under Omad Pressure

To evaluate Omad resistance mechanisms and Omad-Tig crossresistance, in vitro induction experiments were carried out under Omad pressure with the following Omad-resistant S. aureus: CHS221 (MSSA), CHS165 (MSSA), 149 (MSSA), CHS759 (MRSA), CHS810 (MRSA), CHS820 (MRSA), and MS4. The Omad-resistant isolates were characterized with respect to MICs and resistance mechanisms (**Table 2**). Importantly, increasing Omad MICs were accompanied by increasing Tig MICs in Omad-resistant S. aureus isolates, indicating that Omad-Tig cross-resistance can be induced under Omad pressure. Moreover, upregulation of Omad MICs was related to increasing numbers of 16SrRNA copies with a genetic mutation. The mutation sites varied among the five 16SrRNA copies, with high frequencies of the T170G polymorphism in RR1, A1124G in RR2, C810T in RR3, and G1036A in RR4. Leu47His and Tyr87His amino acid

fmicb-10-02546 November 6, 2019 Time: 17:34 # 7



Parental isolates are isolated with shading; ∗ and 1 represent heteroresistant and Omad-induced resistant strain, respectively (see qRT-PCR in Figure 1); RR1-7 are 16s rRNA gene copies; S3 and S10 are 30S ribosome proteins; W, wild-type (no mutation).

Omadacycline Activity to

substitutions in the 30S ribosomal protein S10 were relatively frequent in Omad-resistant bacteria.

## Candidate Genes Related to Omad Resistance in NGS

fmicb-10-02546 November 6, 2019 Time: 17:34 # 8

To identify the genetic mutations that correlate with Omad resistance, whole genome sequencing of MS4O8 was performed and variants relative to MS4 were detected by NGS in MUMmer, version 3.23 (**Supplementary Table S8** and **Table 4**). Nonsynonymous mutations were found in NI36\_11090 (encodes 30S ribosomal protein S10), NI36\_12460 (fnbp encoding FnBP protein), and NI36\_ 00170 (recB encoding recombinase). Mutations affecting these three genes also emerged in our induced Omad-resistance experiment above (**Table 2**). Notably, 30S ribosomal protein S10 has been widely reported to be associated with Tet-class resistance and the impact of FnBP and RecB protein on Omad susceptibility need to be further verified.

## Relationship Between Candidate Genes Overexpression and Omad Susceptibility

The impacts of following candidate genes on Omad susceptibility in Omad-sensitive S. aureus isolates was conducted: USA300HOU\_RS03535, USA300HOU\_RS01625 (encodes a branched-chain amino acid transport system II carrier protein), USA300HOU\_RS00550 (encodes a Na/Pi cotransporter family protein), tet(K), NI36\_12460/NI36\_12465 (fnbp), and NI36\_00170 (recB). The former three were candidately implicated in Omad heteroresistance in our qRT-PCR experiments. Meanwhile, tet(K) was found frequently among S. aureus isolates with an Omad MIC of 1 mg/L, and the proteins

TABLE 3 | Transcriptional expression levels of eight DEGs between S221-O and S221 analyzed by RNA-Seq and qRT-PCR.


encoded by fnbp and recB have been hypothesized to participate in antimicrobial resistance evolution.

The overexpression plasmids pRS00550, pRS01625, pRS03535, pTet(K), pRecB, pRecB-M, pFnBP, and pFnBP-M, where -M suffix indicates a mutated variant, were transformed into clinical isolates with low expression of the target gene (**Supplementary Tables S9, S10**). Stable overexpression of the candidate genes was confirmed by qRT-PCR (**Supplementary Figure S5**). The influence of the overexpression of these six genes on Omad susceptibility was reported in **Table 5**. Briefly, although RS00550, RS01625, and RS03535 did not elevate Omad or Tig MICs in the absence of antibiotic pressure, PAP experiments showed that RS00550 or RS01625 overexpression could lead to Omad heteroresistance compared with negative findings in controls (**Table 5**). RS00550 and RS01625 homology analysis results are reported in **Supplementary Tables S11, S12**.

Overexpression of tet(K) did not elevate Omad or Tig MICs and had no apparent contribution to heteroresistance development in S. aureus. The overexpression of fnbp (NI36\_12 460) and its mutated type led to MIC creep (**Table 5**), supporting the possibility that FnBPs may participate in Omad resistance development. Homology analysis results for fnbp are shown in **Supplementary Table S13**. Overexpression of recB and its mutants did not impact Omad susceptibility in S. aureus.

## DISCUSSION

The presently observed low Omad MICs in this study support the supposition that Omad should be considered a prospective preferential choice for S. aureus infection treatment. Omad MICs ≥ 0.5 mg/L were more frequent among MSSA than MRSA in this study, and moreover, Tet-specific resistance genes, particularly tet(K) and tet(L), were found to be more common among MSSA than MRSA isolates, indicating that Omad may have greater efficacy against MRSA than MSSA. Our Omad MICs were higher than previously reported, perhaps due to regional variation and environmental factors (Villano et al., 2016; Pfaller et al., 2017a). We also obtained higher MICs for the newgeneration Tet-class drug eravacycline in isolates from China than had been reported for isolates from the United States and Europe, suggesting that Tet-class drug MIC dynamics should monitored across global regions with particular attention to MIC creep in China (Zhang et al., 2018; Zheng et al., 2018).

Major mechanisms of Tet resistance in both Gram-positive and -negative microorganisms have been linked to ribosomal protection proteins and efflux pumps, most of which can be overcome with new generation tetracyclines, including Tig and Omad (Jones et al., 2006; Noel et al., 2012; Draper et al., 2014; Macone et al., 2014; Honeyman et al., 2015; Grossman, 2016; Heidrich et al., 2016; Villano et al., 2016; Pfaller et al., 2017b; Argudin et al., 2018; Zhang et al., 2018). In this study, we obtained low Omad MICs for S. aureus, even among isolates harboring a ribosomal protection protein, namely Tet(M), or an efflux pump factor, namely Tet(K) and Tet(L), uncovering an apparent advantage of using Omad to overcome Tet-specific resistance mechanisms, particularly those mediated by Tet(M), Tet(K), and fmicb-10-02546 November 6, 2019 Time: 17:34 # 9

TABLE 4 | Non-synonymous mutations of candidate proteins correlated with Omad resistance found between MS4 and MS4O8.

fmicb-10-02546 November 6, 2019 Time: 17:34 # 10


TABLE 5 | Omad and Tig MICs in S. aureus derivatives with overexpression of candidate genes and their impact on PAPs.


FBP-M has T672S and/or I665V mutations. RecB-M has R10R, I23V, I23N, H24N, H24Q, V29L, and/or V35M mutation. Positive PAP results are highlighted with shading.

Tet(L). Notwithstanding, Tig MICs were shown recently to be increased by a high transcriptional level of tet(M) and tet(L) in Enterococcus faecium (Fiedler et al., 2016). Because we observed a higher frequency of Tet-specific genes in MSSA than in MRSA from China and three isolates with MICs ≥ 1 mg/L harbored tet(K) in this study, we hypothesize that tet(K) overexpression may elevate Omad MICs, as was found in E. coli (Linkevicius et al., 2016). Our data demonstrate that tet(K) overexpression does not impact S. aureus susceptibility to Omad in vitro, and thus indicate that Omad can overcome the Tet(K)-mediated resistance in S. aureus.

Prior epidemiological data have revealed ST239 and ST59 to be predominant MRSA STs internationally, with MSSA ST predominance being more variable across regions (Atshan et al., 2013; Leon-Sampedro et al., 2016; Recker et al., 2017). S. aureus clonality of drug-resistance and virulence factors has been reported (Atshan et al., 2013; Leon-Sampedro et al., 2016; Recker et al., 2017). The present examination of the relationship of ST clonality with Omad susceptibility revealed a far higher frequency of the top-three MRSA STs with MICs ≥ 0.5 mg/L than of the top-three such MSSA STs. Although a definite relationship of ST clustering with Omad MICs has not been established, it is noteworthy that ST59-MSSA were much more likely to have Omad MICs ≥ 0.5 mg/L than were ST59-MRSA.

Heteroresistance frequency is an important harbinger of lastresort antibiotic resistance risk (Zhang et al., 2018; Zheng et al., 2018). The present findings of heteroresistance in 16.98% of MSSA and 3.75% of MRSA with Omad MICs ≥ 0.5 mg/L suggest a need to be alert to selection resistance under Omad pressure for S. aureus, especially in strains from China. Moreover, we observed relatively high Tig MICs for Omad heteroresistancederived clones (2–8 mg/L), indicating a potential risk of Omad-Tig cross-resistance. Mutations affecting 30S ribosomal subunits, which have been reported to participate in Tet or Tig resistance, were not found in our heteroresistance-derived clones or clinical isolates with Omad MICs ≥ 1 mg/L, indicating that 30S ribosomal subunit mutations cannot explain Omad MIC creep and heteroresistance occurrence (Nguyen et al., 2014; Grossman, 2016; Argudin et al., 2018).

The progression of reduced Tig susceptibility in S. aureus has been linked with mepR/mepA encoded efflux pumps (Lupien et al., 2015). However, the possible role of efflux pumps and cell envelopes in Omad heteroresistance in S. aureus is unclear. Multiple reports have shown that protonophore efflux pump inhibitors (e.g., CCCP and PaβN) can be used to evaluate interactions between antibiotics and cell envelope components in bacteria (Klionsky et al., 2016; Bai et al., 2018; Yao et al., 2018). Here, we found that the efflux pump inhibitor PAβN and the cell envelope component inhibitor CCCP reduced Omad MICs of heteroresistance-derived S. aureus clones to as low as ≤0.03 mg/L and 0.12–1 mg/L, respectively, pointing to involvement of efflux pumps or cell envelopes in the progression of Omad heteroresistance in S. aureus (Klionsky et al., 2016; Bai et al., 2018). Omad MICs of isolates with Omad MICs ≥ 1 mg/L could also be reduced by CCCP and PAβN, supporting the notion that efflux pumps or cell envelopes may play an important role in reducing susceptibility to Omad. Our findings showing that overexpression of RS00550 or RS01625 can lead to Omad heteroresistance occurrence, despite having no impact on Omad MICs in the absence of Omad pressure, indicate that expression of genes can facilitate the formation of Omad resistance under antibiotic exposure. Our phylogenic analysis showed that both RS00550 and RS01625 encode efflux pump family proteins, supporting our hypothesis that efflux pump or membrane proteins contribute to Omad heteroresistance.

Crystallographic studies of the Thermus thermophilus 30S ribosomal subunit revealed at least one high-occupancy Tetbinding site and five other minor binding sites in 16S rRNA (Draper et al., 2014; Honeyman et al., 2015; Heidrich et al., 2016). Crystallographic studies with Tet, Tig, and Omad showed that, although they produced slightly different patterns of RNA cleavage and dimethylsulfate modification, all three antibiotics fmicb-10-02546 November 6, 2019 Time: 17:34 # 11

associated with the same binding site, albeit in somewhat different orientations. In several bacterial species, Tig and Omad have been shown to exhibit higher binding affinities and greater antitranslational potencies than Tet or minocycline, and 16SrRNA mutations have been shown to affect Tet binding sites in the 30S ribosomal subunit, which may confer Tet/Tig resistance (Nguyen et al., 2014; Grossman, 2016; Zheng et al., 2018). Consistent with previous reports, we found that greater numbers of 16S rRNA copies with genetic mutations were associated with higher levels Omad/Tig resistance in S. aureus isolates with Omad-resistance induced under Omad pressure (Nguyen et al., 2014; Grossman, 2016). This finding implicates the participation of 16SrRNA mutations in the development of the Omad resistance. Additionally, our finding of frequent 30S ribosomal protein S10 mutations in Omad-derived resistant isolates indicates that such mutations may be an important factor in Omad resistance evolution. It will be important to examine the unknown mechanism(s) underlying MIC elevation during Omad resistance evolution in S. aureus. In this study, NGS was performed to identify candidate genes that may be involved in Omad resistance development and FnBP was identified as a novel membrane molecule that may contribute to Omad MIC elevation. Mechanistically, we hypothesized that the overexpression of FnBP could alter the penetration potential of cell membranes.

## CONCLUSION

Omad exhibited excellent in vitro activity against clinical S. aureus isolates from China and might represent a preferential choice for the treatment of S. aureus infections. However, we must be alert to the potential risk of Omad heteroresistance in S. aureus, especially in strains with MICs ≥ 0.5 mg/L. Compared with MRSA, MSSA had relatively low MICs with a more facile tendency for the occurrence of Omad heteroresistance. Omad heteroresistance in both MSSA and MRSA could be reversed by CCCP and PaβN, indicating involvement of efflux pumps in Omad heteroresistance development in S. aureus. Furthermore, both RS01625 and RS00550, which encode efflux pump family proteins (a branched-chain amino acid transport system II carrier protein and an Na/Pi cotransporter family protein, respectively),

## REFERENCES


were found to contribute to Omad heteroresistance. FnBP emerged as a novel molecule to be associated with Omad resistance and Omad MIC elevation in S. aureus. The present data contribute to understanding potential resistance mechanisms that may impact clinical applications of Omad.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## AUTHOR CONTRIBUTIONS

QD and ZY initiated and designed the project. BB, ZP, and ZL performed the molecular biological experiments with bacteria. GX, XS, JZ, and ZC performed the PCRs. FZ, PL, and GX performed the bacterial culturing and MIC testing. All authors participated in the data analysis. QD, ZY, and BB wrote the manuscript incorporating comments from all authors.

## FUNDING

This work was supported by grants from the National Natural Science Foundation of China (Nos. 81170370 and 81601797); Science, Technology, and Innovation Commission of Shenzhen Municipality of basic research funds (No. JCYJ20170412143551332); Shenzhen Health and Family Planning Commission (SZXJ2018027 and SZXJ2017032); Sanming Project of Medicine in Shenzhen; the Shenzhen Nanshan District Scientific Research Program of the People's Republic of China (No. 2018010, 2018011, and 2018021); and provincial medical funds of Guangdong (No. A2018163).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02546/full#supplementary-material

aureus (MRSA) infection: an in-depth review of newer antibiotics active against an enduring pathogen. J. Glob. Antimicrob. Resist. 10, 295–303. doi: 10.1016/j. jgar.2017.05.019


fmicb-10-02546 November 6, 2019 Time: 17:34 # 12

omadacycline. Antimicrob. Agents Chemother. 58, 1279–1283. doi: 10.1128/ AAC.01066-13


**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 Bai, Lin, Pu, Xu, Zhang, Chen, Sun, Zheng, Li, Deng and Yu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# Increasing Frequencies of Antibiotic Resistant Non-typhoidal Salmonella Infections in Michigan and Risk Factors for Disease

Sanjana Mukherjee<sup>1</sup> , Chase M. Anderson<sup>1</sup> , Rebekah E. Mosci <sup>1</sup> , Duane W. Newton<sup>2</sup> , Paul Lephart <sup>2</sup> , Hossein Salimnia<sup>3</sup> , Walid Khalife<sup>4</sup> , James. T. Rudrik <sup>5</sup> and Shannon D. Manning<sup>1</sup> \*

<sup>1</sup> Department of Microbiology and Molecular Genetics, Michigan State University, East Lansing, MI, United States, <sup>2</sup> Clinical Microbiology Laboratory, University of Michigan, Ann Arbor, MI, United States, <sup>3</sup> Microbiology Division, Detroit Medical Center University Laboratories, Wayne State University, Detroit, MI, United States, <sup>4</sup> Microbiology, Immunology & Molecular Laboratories, Sparrow Hospital, Lansing, MI, United States, <sup>5</sup> Bureau of Laboratories, Michigan Department of Health and Human Services, Lansing, MI, United States

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### Reviewed by:

Angelo Berchieri Junior, São Paulo State University, Brazil Rafael Antonio Casarin Penha Filho, São Paulo State University, Brazil

> \*Correspondence: Shannon D. Manning mannin71@msu.edu

#### Specialty section:

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Medicine

Received: 31 July 2019 Accepted: 17 October 2019 Published: 08 November 2019

#### Citation:

Mukherjee S, Anderson CM, Mosci RE, Newton DW, Lephart P, Salimnia H, Khalife W, Rudrik JT and Manning SD (2019) Increasing Frequencies of Antibiotic Resistant Non-typhoidal Salmonella Infections in Michigan and Risk Factors for Disease. Front. Med. 6:250. doi: 10.3389/fmed.2019.00250 Non-typhoidal Salmonella (NTS) are important enteric pathogens causing over 1 million foodborne illnesses in the U.S. annually. The widespread emergence of antibiotic resistance in NTS isolates has limited the availability of antibiotics that can be used for therapy. Since Michigan is not part of the FoodNet surveillance system, few studies have quantified antibiotic resistance frequencies and identified risk factors for NTS infections in the state. We obtained 198 clinical NTS isolates via active surveillance at four Michigan hospitals from 2011 to 2014 for classification of serovars and susceptibility to 24 antibiotics using broth microdilution. The 198 isolates belonged to 35 different serovars with Enteritidis (36.9%) predominating followed by Typhimurium (19.5%) and Newport (9.7%), though the proportion of each varied by year, residence, and season. The number of Enteritidis and Typhimurium cases was higher in the summer, while Enteritidis cases were significantly more common among urban vs. rural residents. A total of 30 (15.2%) NTS isolates were resistant to ≥1 antibiotic and 15 (7.5%) were resistant to ≥3 antimicrobial classes; a significantly greater proportion of Typhimurium isolates were resistant compared to Enteritidis isolates and an increasing trend in the frequency of tetracycline resistance and multidrug resistance was observed over the 4-year period. Resistant infections were associated with longer hospital stays as the mean stay was 5.9 days for patients with resistant isolates relative to 4.0 days for patients infected with susceptible isolates. Multinomial logistic regression indicated that infection with serovars other than Enteritidis [Odds ratio (OR): 3.8, 95% confidence interval (CI): 1.23–11.82] as well as infection during the fall (OR: 3.0; 95% CI: 1.22–7.60) were independently associated with resistance. Together, these findings demonstrate the importance of surveillance, monitoring resistance frequencies, and identifying risk factors that can aid in the development of new prevention strategies.

Keywords: non-typhoidal Salmonella, antimicrobial resistance, epidemiology, risk factors, Michigan

## INTRODUCTION

The gram-negative pathogen, Salmonella enterica, is an important public health concern resulting in 93.8 million cases of foodborne infections globally each year (1). In 2015, infection with non-typhoidal S. enterica (NTS) serovars was a leading cause of death, with 90,300 deaths reported (2). Infections with NTS can cause nausea, vomiting, abdominal pain, myalgia (muscle pain) and arthralgia (joint pain), while hepatomegaly (liver enlargement), and splenomegaly (spleen enlargement) can develop in some cases (3). Systemic infections can also occur in immunocompromised patients (4). Because of these complications, NTS infections were estimated to contribute to 70 disability-adjusted life years (DALY) lost/100,000 persons worldwide in 2010 (5). In the U.S., the number of NTS infections was estimated to be 1.2 million per year with 23,000 hospitalizations and 450 deaths (6). Salmonella infections were also reported to have the highest mean cost of illness among all foodborne infections (7). Geographical differences in serovar prevalence have been documented as well (8). In Europe and Asia, for instance, S. Enteritidis was the leading cause of clinical infections in 2002, whereas S. Typhimurium was the highest in North America followed by S. Enteritidis, S. Newport and S. Heidelberg. The Typhimurium and Enteritidis serovars, however, have been reported to cause the greatest number of enterocolitis and bacteremia cases (3).

NTS has been frequently isolated from commercially raised chickens and other poultry, and contact with cattle, pigs, horses, and other domestic animals are important risk factors for NTS infections (9). In addition, 74,000 Salmonella infections have been attributed to reptile and amphibian exposures in the U.S. each year (10); contact with reptiles and cats was associated with salmonellosis in a prior Michigan study (11). Other studies have identified serovar-specific risk factors as well. A study in the Netherlands, for example, found consumption of raw eggs and products containing raw eggs to be linked to Salmonella Enteritidis infections, while exposure to raw meat and playing in a sandbox were risk factors for S. Typhimurium infections (12). Prior history of antibiotic use, living on a livestock farm, and international travel were also identified as risk factors for S. Typhimurium infections in Canada (13). NTS isolates have been recovered from environmental sources including water and soil and can often survive in these environments for extended periods of time (14, 15). Taken together, these studies indicate the importance of the environment as a reservoir for Salmonella.

Antibiotic-resistant NTS infections have also emerged and are increasing in frequency in the U.S. resulting in high hospitalization rates and ∼\$365,000,000 in medical costs every year (16). Fluoroquinolones, third generation cephalosporins, penicillins, macrolides, and trimethoprim-sulfamethoxazole are commonly prescribed for the treatment of salmonellosis, particularly in patients with immunocompromising conditions, young children and the elderly (17). Importantly, antibiotic resistant Salmonella infections have been linked to more severe disease outcomes including bloodstream infections as well as hospitalization (18) and multidrug resistance has emerged. Resistance to ampicillin, chloramphenicol, streptomycin, sulfonamides and tetracycline, previously defined as ACSSuT isolates, for instance, has been reported in multiple NTS serovars in different geographical regions (19–22). In addition to ACSSuT, resistance to ampicillin, streptomycin, sulfonamides, and tetracycline (ASSuT) has been observed in S. Typhimurium from humans, food animals and retail meats in the U.S. (23). The emergence of these multi-drug phenotypes in Salmonella are of great concern since alternative antibiotics that can be used to treat Salmonella infections are limited.

Emergence of widespread resistance in Salmonella is attributed to the overuse of antibiotics and limits the effectiveness of these antibiotics for the treatment of human infections. Michigan is not included in the CDC FoodNet surveillance network, which monitors the incidence of foodborne illnesses and collects case information from 10 states in the U.S., covering 15% of the U.S. population. Consequently, we sought to examine the distribution of NTS serovars causing disease in Michigan and evaluate risk factors for infection using an active surveillance system at four hospitals. We also calculated the frequency of antibiotic resistance in NTS isolates and identified factors associated with resistance and NTS infections. This study highlights the importance of enhanced surveillance for resistant pathogens to ensure that the most appropriate drug targets are used, and to identify risk factors for infection and patients with an increased risk of more debilitating conditions.

## MATERIALS AND METHODS

## Strain Source and Collection

From 2011 to 2014, 198 NTS isolates were collected as part of the Enterics Research Investigational Network (ERIN), which was set up in collaboration with the Michigan Department of Health and Human Services (MDHHS) and four major hospitals in southern Michigan. Serovar classification was conducted at the MDHHS using traditional serotyping protocols as recommended by the Association of Public Health Laboratories (24); serovar data was extracted from the Michigan Disease Surveillance System (MDSS) for each case. To ensure that the ERIN surveillance network was representative of enteric infections occurring in Michigan, the frequency of ERIN cases was shown to be similar to those identified throughout the state of Michigan during the same time period (25). Isolates were stored in Luria-Bertani broth with 10% glycerol at −80◦C until further testing.

## Antimicrobial Susceptibility Profiling

For NTS, susceptibilities to 24 antibiotics were determined by broth microdilution using Sensititre GN4F Trek plates (Trek Diagnostic Systems, Cleveland, OH, USA) according to the manufacturer's instructions. Nine antibiotic classes were tested and included: aminoglycosides (amikacin, gentamicin, tobramycin); β-lactam antibiotics [penicillins (ampicillin, piperacillin); β-lactam/β-lactamase inhibitor combinations (ampicillin/sulbactam 2:1 ratio, piperacillin/ tazobactam constant 4, ticarcillin/clavulanic acid constant 2); cephalosporins (cefazolin, ceftazidime, ceftriaxone, cefipime)]; carbapenems (imipenem, doripenem, ertapenem, meropenem); tetracyclines (tetracycline, minocycline); fluoroquinolones (ciprofloxacin, levofloxacin); glycylcyclines (tigecycline); nitrofurans (nitrofurantoin); monobactams (aztreonam); and anti-folates (trimethoprim/sulfamethoxazole). The minimum inhibitory concentration (MIC) was determined by identifying the lowest concentration of antibiotic that prevented visible bacterial growth. Escherichia coli ATCC 25922, which is susceptible to all antibiotics evaluated, was used as the quality control strain. The results of the susceptibility tests were interpreted as resistant or susceptible in accordance with published guidelines (26) and isolates were classified as multidrug resistant if they were resistant to three or more antimicrobial classes.

## Data Analysis

Epidemiological, demographic, and clinical data were obtained from the MDSS and managed using Microsoft Access and Excel. Season was classified as spring (March, April, and May), summer (June, July, and August), fall (September, October, and November) and winter (December, January, and February) based on the sample collection date; for those cases with a missing collection date, the stool arrival date and/or onset dates were used. Counties in Michigan were classified as urban or rural based on the classification scheme devised by the National Center for Health Statistics; only 10 Michigan counties were designated as urban (27). Based on the published rates of antibiotic prescription use in adults and children in Michigan (28), counties were classified as having high or low prescribing rates. High rates were defined as those counties where hospital service areas had >30% higher use rates relative to the state average. A dichotomous variable was created for length of hospital stay by defining a long hospital stay as one that was greater than 4 days.

SAS version 9.4 (SAS Institute, Cary, NC, USA) and Epi InfoTM version 7.0 were used for all statistical analyses. χ 2 test and Fisher's exact test were used for dichotomous variables to identify significant associations between the dependent and independent variables; a p-value ≤ 0.05 was considered significant. A univariate analysis was first conducted, and those variables found to have strong associations with the outcome (pvalue ≤ 0.20) were included in the multivariate analysis unless the variable contained >15 missing values. Multivariate analysis using forward logistic regression was performed to build a model containing significant variables (p-value ≤ 0.05) along with potentially confounding factors such as age and sex. The Mantel-Haenszel χ 2 test was used to test for trends and the student's t-test was used for testing statistical significance between means. Finally, the χ 2 test for equality of proportions was used for comparing sample proportions.

## RESULTS

## Characteristics of Cases With Non-typhoidal Salmonella (NTS) Infections

A total of 198 NTS cases were identified and roughly half were male (53.1%; n = 104) and between 18 and 52 years of age (n = 81; 40.9%). Forty-four (22.2%) cases were younger than 10 years of age and 14 were younger than 2 years. More cases self-reported as Caucasians (n = 125; 74.0%) than as African Americans (n = 33; 19.5%) or other races (n = 11; 6.5%); race was not known for 29 cases. Diarrhea (n = 172, 97.2%), abdominal pain (n = 135, 80.4%), and fever (n = 106, 69.3%) were commonly reported symptoms, though 70 (41.9%) patients also reported bloody diarrhea. Sixty-five patients (34.6%) were hospitalized between 1 day and 17 days; the average duration of hospitalization was 4.4 days.

No significant difference in the proportion of cases was observed by year (p-value = 0.075), though a higher number was reported in the summer and fall (n = 139; 70.2%) compared to the winter and spring months (p-value < 0.0001). Differences were also observed among the four participating hospitals, which represented 10.6, 23.2, 29.3, and 36.9% of the 198 cases. When stratified by county, 56.2% (n = 108) of the cases lived in rural counties and 43.7% (n = 84) resided in urban counties; the residence was not known for one case. A subset of five cases resided in other states (Colorado, Georgia, Illinois, Ohio, and South Dakota), though each patient developed symptoms and were diagnosed with salmonellosis while visiting Michigan. Of the 166 cases with travel information available, a significantly higher proportion of patients had not traveled in the past month (n = 103, 62.0%) when compared to those who reported travel (n = 63; 37.9%) (p-value = 0.0019).

Patients reporting contact with animals (n = 95; 61.3%) were also more likely to be affected when compared to those reporting no animal contact (n = 60; 38.7%) (p-value = 0.0049). Among these 95 cases, 12 (12.6%) had contact with reptiles such as turtles and lizards, and nine (9.5%) reported contact with livestock including cattle, horses, goats, or pigs. Contact with domestic animals (e.g., cats, dogs, and small mammals) was also reported in 83 of the 95 (87.4%) cases.

Because more cases resided in rural counties, we conducted a case-case analysis between the rural and urban cases to further detect differences in disease frequencies (**Table S1**); the five cases from other states were excluded from this analysis. Notably, any animal contact was more common in patients living in rural vs. urban counties (OR: 1.8; 95% CI: 0.95–3.57), however, the association was not statistically significant. Contact with "other" animals including small mammals, fish and/or amphibians was also more common in rural cases (Fisher's exact p-value = 0.0095), though only one case from the urban counties reported "other" contact. No significant difference was observed by season or for well water consumption in rural (n = 18, 19.8%) vs. urban (n = 7, 11.9%) cases; however, pork consumption was significantly lower (OR: 0.1; 95% CI: 0.02–1.12) and peanut butter consumption was significantly higher (OR: 2.1, 95% CI: 1.06–4.35) in rural vs. urban cases.

## Risk Factors for More Severe NTS Infections

To identify predictors of hospitalization, a marker for more severe infections, we used hospitalization as the dependent variable among the 165 cases with data available. In the univariate analysis, patients self-reporting nausea (OR: 2.0; 95% CI: 1.05–3.91) and vomiting (OR: 1.9; 95% CI: 0.99–3.64) were significantly more likely to be hospitalized as were patients over the age of 59 years (n = 16; 51.6%). Patients from urban counties were also more likely to be hospitalized (n = 34; 41.9%) than cases from rural counties (n = 29; 27.9%). No associations were identified with specific serovars.

Multivariate logistic regression using forward selection identified urban residence (OR: 2.1; 95% CI: 1.06–4.04) and vomiting (OR: 1.9; 95% CI: 0.99–3.78) to be the strongest predictors of hospitalization while controlling for age and sex. Because vomiting and nausea were associated with each other (p-value < 0.0001), only vomiting was included in the model. Age over 59 years met the p > 0.25 significance level for entry into the model but despite the positive association (OR: 2.0; 95% CI: 0.87–4.78), it was not statistically significant.

## Distribution of Salmonella enterica Serovars in Michigan Cases

The 198 NTS isolates were classified into 35 different S. enterica serovars; the serovar could not be determined for three isolates (**Table S2**). Among the 195 typed isolates, the predominant serovar was Enteritidis (n = 72; 36.9%) followed by Typhimurium (n = 38; 19.5%), Newport (n = 19; 9.7%), Hartford (n = 6; 3.1%), Saintpaul (n = 5; 2.6%), and Heidelberg (n = 4; 2.1%). The remaining 51 isolates represented 28 different serovars with fewer than three isolates per type. Nine isolates were classified as I 4, [5], 12:i:- /I 4, 5,12:i- (n = 3), I 4, 12:b- (n = 3), I 4, 12:i:- (n = 3), and III 50:Kz (n = 1), which likely represent variants of known serovars.

A significant difference in the proportion of Enteritidis cases was observed by hospital (p-value = 0.02). The highest proportion was observed at two sites with 24 (33.3%) and 25 (34.7%) Enteritidis cases, though 12 (16.7%) and 11 (15.3%) additional cases were identified at the remaining two sites. The proportion of Typhimurium cases also differed at the four hospitals (p-value=0.003) with frequencies of 44.7% (n = 17), 34.2% (n = 13), 13.2% (n = 5), and 7.9% (n = 3) per site. S. Heidelberg was recovered from only one case at each hospital, but significant differences were observed in the proportion of Newport cases across hospitals (p-value = 0.009) as well as cases with all other serovars combined (p-value = 0.0003).

Differences were also observed in the proportion of specific serovars by year. In 2011 and 2012, for instance, 21.1% (n = 12) of the 57 cases and 35.2% (n = 18) of the 51 cases, respectively, were classified as Enteritidis, whereas higher frequencies were observed in 2013 (n = 20 of 34; 58.8%) and 2014 (n = 22 of 53; 41.5%). The proportion of Typhimurium cases was similar in years 2011, 2012, and 2014 (average = 16.2%), though a greater proportion of cases (n = 12; 35.3%) were recovered in 2013. Of the 19 Newport cases, the proportion of Newport cases decreased in 2013 (n = 2; 10.5%) and 2014 (n = 3; 15.8%) relative to 2011(n = 7; 36.8%) and 2012 (n = 7; 36.8%), however, the difference between each period was not significant (p-value = 0.22). Only four Heidelberg cases were identified and three of these cases were from 2014. Among all other serovars reported, it is notable that the highest frequencies were recovered in 2011 (n = 28; 49.1%) followed by 2014 (n = 17; 32.1%) and 2012 (n = 17; 33.3%). A more diverse serovar population was reported in 2011 as well given that infections were caused by 26 different serovars, while 15 and 17 serovars were reported in 2012 and 2014, respectively. Only Enteritidis, Typhimurium and Newport were recovered in 2013.

Because the number of cases differed by season with the highest proportion of cases occurring in summer than in other seasons (p-value < 0.0001), we also examined seasonal variation by serovar (**Figure 1**). The number of Enteritidis cases was significantly higher in the summer months (n = 34; 47.2%) when compared to the fall (n = 11; 15.3%), winter (n = 10; 13.9%), and spring (n = 17; 23.6%) months (p-value = 0.0001). Similar trends were observed for Typhimurium in the summer (n = 16, 42.1%) vs. fall (n = 10; 26.3%), winter (n = 6; 15.8%), and spring (n = 6; 15.8%) months. The majority (n = 14; 73.7%) of the 19 Newport cases were reported in the summer as were most cases infected with other serovars (n = 31; 46.9%). During the summer months, the greatest proportion of cases were caused by Enteritidis (n = 34, 35.8%), Typhimurium (n = 16, 16.8%), and Newport (n = 14, 14.7%).

## Risk Factors for Infection With Specific Salmonella Serovars

Several serovar-specific factors were identified. In the univariate analysis, significantly more Enteritidis cases were reported in urban (n = 40; 57.9%) vs. rural counties (n = 31; 37.3%) (pvalue = 0.011). Enteritidis was also associated with consumption of bottled water at home (n = 11; 64.7%) relative to municipal (n = 30; 37.0%) (p-value = 0.03) or well water (n = 10, 52.6%) (p-value = 0.21) as well as pork consumption (OR: 7.2; 95% CI: 0.85–61.5). The small sample sizes and missing data records for these variables, however, may have skewed the associations and prevented their inclusion into the multinomial regression analysis.

By contrast, Typhimurium infections were more common in the 95 patients reporting animal contact (n = 24; 25.3%) than those without animal contact (n = 7; 12.1%) (p-value = 0.078). Contact with livestock (Fisher's exact p-value = 0.0002) and other animals such as amphibians, small mammals, and fish (OR: 3.8; 95% CI: 1.41–9.97) were more common in Typhimurium cases relative to cases infected with other serovars.

A multinomial logit model using forward selection was used containing the following three outcomes: (1) Enteritidis infection; (2) Typhimurium infection; and (3) infection by all other serovars. When controlling for age and sex, residence was found to be a significant predictor of Enteritidis but not Typhimurium infections. Specifically, cases living in rural areas were significantly less likely to have infections caused by Enteritidis (OR: 0.4; 95% CI: 0.23–0.85; p-value = 0.01) but not Typhimurium (OR: 1.2; 95% CI: 0.52–2.71; pvalue = 0.69); only the former association was significant.

Because of missing data and too few cases per outcome, however, no additional variables could be examined in the multivariate analysis.

## Antibiotic Resistance Profiles in 198 Non-typhoidal Salmonella (NTS) Isolates

Resistance to at least one antibiotic was observed among 30 of the 198 (15.1%) NTS isolates (**Figure 2**). Resistance to the β-lactam, ampicillin (11.6%), and tetracycline (11.1%) predominated followed by resistance to trimethoprim-sulfamethoxazole (2.5%), gentamicin (0.5%), and other β-lactams including cephalosporins like cefazolin (2.0%), ceftazidime (2.0%), and ceftriaxone (1.0%). No resistance was observed to 13 of the 24 antibiotics tested. Overall, 19 distinct antibiotic resistance patterns were observed among the 30 resistant NTS isolates (**Table S3**). Multidrug resistance (MDR) to ≥ 3 antimicrobial classes was observed in 15 (7.5%) isolates while four (2.0%) isolates were resistant to ≥ 4 antimicrobial classes; nine (4.5%) isolates were resistant to only one antimicrobial class. When stratified by serovar (**Figure 3**), only 10 of the 35 NTS serovars contained isolates that were resistant to at least one antibiotic. Among these 10 serovars, Enteritidis was significantly less likely to comprise resistant isolates than Typhimurium (Fisher's exact p-value = 0.022) as well as the other eight serovars combined (Fisher's exact p-value < 0.0001). In all, four S. Enteritidis (n = 72; 5.6%) and eight S. Typhimurium (n = 38; 21.0%) isolates were resistant to at least one antibiotic.

Important differences in resistance frequencies were also identified by year of isolation (**Figure 4**). Significant increases in resistance to tetracycline and cephalosporins, for example, were observed between 2011 and 2014 as were increases in multidrug resistance (p-value <0.05). No significant difference in the frequency of resistance to trimethoprim-sulfamethoxazole or gentamicin were noted.

A comparison between all NTS isolates from Michigan to those tested by the National Antimicrobial Resistance Monitoring System (NARMS) (29) during the same time period also revealed slight differences in resistance frequencies by antibiotic (**Figure 5A**). None of these differences, however, were significant. For Enteritidis, tetracycline resistance was lower in Michigan (n = 1, 1.4%) than NARMS (n = 48, 3.0%) isolates (**Figure 5B**), while resistance to ampicillin and tetracycline was higher in Typhimurium isolates from NARMS than Michigan (**Figure 5C**). Resistance to trimethoprimsulfamethoxazole was also higher in the Michigan Typhimurium isolates (n = 2, 5.3%) compared to the NARMS isolates (n = 21, 1.7%), yet none of the frequency differences

FIGURE 4 | Trends in antibiotic resistance among clinical non-typhoidal Salmonella (NTS) isolates from Michigan over time. Mantel-Haenszel chi-square was used to identify trends over time and calculate p-values. AMP, Ampicillin; TET, Tetracycline; SXT, Trimethoprim/sulfamethoxazole; CEPH, Cephalosporin; GEN, Gentamicin; MDR, Multidrug resistance (resistance to ≥3 antimicrobial classes).

observed among Enteritidis and Typhimurium isolates were statistically significant.

## Epidemiological Associations With Antibiotic Resistant NTS Infections

To identify factors associated with resistant NTS infections, we conducted univariate and multivariate analyses using resistance to at least one (≥ 1) antibiotic as the dependent variable. The univariate analysis demonstrated that the odds of resistance was significantly higher in Typhimurium isolates (OR: 4.5; 95% CI: 1.27-16.22) and other NTS serovars (OR: 4.6; 95% CI: 1.47-14.20) compared to Enteritidis isolates (**Table 1**). Higher resistance frequencies were also observed in counties with low antibiotic prescribing rates (n = 26, 16.9%) compared to those with high rates (n = 4, 10.3%), yet this difference was not statistically significant. Furthermore, frequencies of antibiotic resistant infections were higher in urban (n = 14, 16.7%) than rural (n = 16, 14.8%) areas and were lowest in the summer (n = 10, 10.3%) compared to winter, spring and fall (n = 20, 19.8%) (OR: 2.1; 95% CI: 0.95-4.86).

Notable differences in resistance frequencies were observed among the 59 patients who were hospitalized for one or more days. Among these 59 cases, the mean hospital stay was 5.9 days for the 39 patients hospitalized with resistant NTS isolates compared to 4.0 days for the 48 patients infected with pansusceptible isolates (**Table S4**). Six of the 23 (26.1%) patients with hospital stays of ≥5 days had resistant infections compared to five of the 36 (13.9%) patients with stays between 1 and 4 days (Student's t-test p-value <0.05). Cases with tetracycline resistant and susceptible infections had mean hospital stays of 6.0 days and 4.2 days, respectively (Student's t-test p-value = 0.068), whereas the mean hospital stay was 6.2 days for patients infected with ampicillin-resistant NTS compared to 4.0 days for patients with ampicillin-susceptible infections (Student's t-test p-value < 0.05). No association was observed between hospitalization and infection with either Enteritidis (n = 21, 30.9%) or Typhimurium (n = 11, 31.4%) when compared to all other serovars (n = 30, 36.6%).

Multivariate analysis using forward regression was performed to identify predictors of resistant infections. Because the frequency of resistance to ≥1 antibiotic was similar in isolates belonging to Typhimurium and the other NTS serovars except Enteritidis, they were grouped together for the multivariate analysis. After adjusting for sex and age, the model indicated that compared to Enteritidis, the remaining serovars were 3.8 times more likely to be resistant to at least one antibiotic (**Table 1**). Moreover, resistant NTS infections were significantly more likely to occur in fall (OR: 3.1; 95% CI: 1.22–7.60) than in the other TABLE 1 | Univariate and multivariate analysis to identify factors associated with antibiotic resistance in 198 clinical non-typhoidal Salmonella in Michigan, 2011–2014.


(Continued)

#### TABLE 1 | Continued


\*Depending on the variable examined, the number of isolates do not add up to the total (n = 198) because of missing data.

e Wald 95% confidence intervals (CI) for odds ratio (OR).

‡ p-value was calculated by Chi square test and Fisher's exact test was used for variables < 5 in at least one cells; ORs were not calculated for variables with <5 per cell and the Mantel-Haenszel Chi square test was used for serogroup.

£Logistic regression was performed using forward selection while controlling for variables that yielded significant (p-value ≤ 0.05) and strong (p-value ≤ 0.20) associations with in the univariate analysis and with a sufficient sample. Nineteen records were omitted from the analysis when hospitalization was added to the model, however, the associations between resistance and the other four variables remained the same with and without this variable. Hosmer and Lemeshow Goodness of Fit tests were >0.05 for both models with and without hospitalization. All variables were tested for collinearity.

three seasons. A strong association (OR: 2.7) was also observed between resistance and hospitalization stays longer than 5 days, however, it was not statistically significant while controlling for the other four variables.

## DISCUSSION

This study, which was conducted between 2011 and 2014 using data and isolates from an active surveillance system at four large, metropolitan hospital systems, indicated that most (64.1%) of the 198 cases were over the age of 19 and 16.2% were over 60 years of age. These data differ from those reported in a prior Michigan study of Enteritidis infections occurring between 1995 and 2001, which found children <4 years of age to have the greatest risk of infection (30). Although age was not significantly associated with Enteritidis infections or infections caused by any other serovars, these data suggest that the demographics of salmonellosis cases may have changed over time. Additional studies in different Michigan hospitals are therefore required.

Significantly more NTS infections occurred in Michigan during the summer and fall months consistent with a CDC report showing that most Salmonella infections occur between June and October in the U.S. (31). An increased frequency of infections caused by S. Enteritidis specifically (30) and other enteric pathogens has also been reported during the summer. Shiga toxin-producing E. coli (STEC) infections, for example, were highest in Michigan during the summer (37.3%) and fall (29.5%) over a 12-year period (2001–2012) (32). Such seasonal variation has been explained by recreational activities, inadequate cooking and suboptimal food storage temperatures during outdoor picnics and barbeques (33) as well as visiting petting zoos and farms (34). Variation in the distribution of serovars by season was also observed and is consistent with data from a study of 690,479 Salmonella infections reported to the CDC. Specifically, this study found that the number of different Salmonella serotypes causing human infection, or the "serotype richness," was greatest in the winter even though the greatest number of infections occurred during the summer (35).

Although salmonellosis frequencies were not significantly higher in patients residing in rural vs. urban areas (p-value = 0.083), urban residence was found to be a predictor of infection with S. Enteritidis using logistic regression. Intriguingly, urban residence was also associated with hospitalization due to any NTS infection, a finding that may be related to proximity of the health care facilities included in the study. The association between residence location and infection with specific serovars is supported by data from prior studies showing a lower prevalence of Enteritidis in the farm environment relative to other serovars. A Canadian study of urban and rural streams, for example, detected low frequencies of Enteritidis in the rural streams (36), while Enteritidis was rarely recovered from livestock and poultry in Alberta (37) or in the environment of commercial poultry farms in California (38). It is therefore possible that Salmonella serovars other than Enteritidis may be more widespread in the environment. An association was also observed between Typhimurium infections and history of animal contact, which has been reported in many prior studies. Indeed, one study indicated that living on a livestock farm was an independent risk factor for acquiring S. Typhimurium DT104 infections (13), while another study recovered indistinguishable S. Typhimurium DT104 isolates from a child and animals living on the same farm (39). The identification of animal contact as a risk factor for NTS infections in Michigan may be important as it highlights the need for additional studies to investigate this epidemiological association and develop more targeted prevention strategies. Indeed, we were only able to evaluate whether or not any animal contact was reported, which fails to adequately describe the level or duration of contact or even the well-being of the animals.

Importantly, a wide range of antibiotic resistance profiles in the 198 NTS isolates recovered from patients at four Michigan hospitals was observed. High frequencies of resistance to ampicillin and tetracycline were identified with an increasing trend in tetracycline resistance over the 4-year period; tetracycline resistance also contributed to the high frequency of MDR in the NTS isolates examined. It is notable that an increase in tetracycline resistance was not observed by NARMS for the same time period (p-value = 0.66) and hence, it is possible that Michigan harbors a unique population of resistant NTS. NARMS also reported an increased frequency of MDR in 2015 (12%) relative to 2008 (9.5%) (29). While tetracycline is not widely recommended for clinical use, antibiotics such as amoxicillin, penicillin, and sulfamethoxazole-trimethoprim are commonly prescribed to adults and children in Michigan (28). The high prescription rates of these antibiotics, among others, is likely to have an impact on resistance frequencies in the state. Unfortunately, actual antibiotic usage data was not available for NTS cases in our study as these data are not collected by the MDHHS during case interviews. We also observed an increasing trend in the frequency of MDR in NTS over the 2011–2014 time period, however, the difference was not significant by year and could be due to our focused surveillance of only four hospitals. It is possible that increases in MDR frequencies over time could be linked to international travel or food imports as suggested previously (40, 41); however, travel was not associated with MDR infections in our study and import risk was not evaluated.

Although tetracyclines are not widely used in human medicine, they are among the most commonly used antibiotics in livestock and poultry worldwide (42). The FDA has estimated that 3,535,701 (kg)<sup>2</sup> of tetracyclines are used in foodproducing animals in the U.S. each year representing 64% of all antibiotics used (43). While the use of tetracyclines in these animals decreased by 40% in 2016–2017 relative to the period between 2009 and 2017 (43), antibiotic residues and resistance determinants can persist in the environment (44) and animal reservoir, be transmitted to humans, and contribute to MDR in both gram-positive and -negative bacteria (45). Indeed, the emergence and spread of resistance between food animals and people has been documented. One historical study, for instance, found identical resistance patterns in E. coli isolates from livestock and farming families (46), while others have documented spread through food products and water (47–49). Since we have only examined resistance in clinical NTS isolates, additional studies are needed to quantify resistance frequencies in isolates from other sources to identify those strain types, virulence gene profiles, and resistance phenotypes that pose the highest risk of transmission to and infection in humans. A recent study, for example, documented different antibiotic resistance profiles among isolates from different sources and suggested that multiple sources in the food-chain are responsible for resistant Salmonella infections in humans (23).

As such, it is important to consider that the genetic and phenotypic diversity of NTS isolates in different geographical locations may play a role in human infections as specific NTS lineages circulating in Michigan may be more likely to carry resistance determinants. Support for this possibility comes from our finding that NTS serovars had varying resistance profiles. Notably, multivariate logistic regression identified serovar Enteritidis to be significantly less likely to be resistant to at least one antibiotic compared to all other NTS serovars. Serovar-dependent differences in resistance have been observed in NTS isolates in different geographic locations including Spain (50) and Brazil (51). The reason for serovar-specific differences is not clear, however, one study found that ciprofloxacinresistant S. Typhimurium isolates were more competitive during growth in vitro with increasing concentrations of ciprofloxacin than the ciprofloxacin-resistant S. Enteritidis isolates (52). Other studies have shown that certain NTS serovars such as Kentucky, Typhimurium, and Heidelberg, were more likely to have MDR (53–55), whereas some serovars (e.g., Enteritidis, Montevideo, Infantis, and Mbandaka) were more commonly pansusceptible or were resistant to fewer antibiotics (56, 57). Such differences could be due to the specificity of or ability to take up certain plasmids carrying resistance genes as some plasmids may be more commonly acquired across serovars and other bacterial species (58). Serovars such as Kentucky and Heidelberg have also been shown to have a mutation in the methyl mismatch repair (MMR) system, which may contribute to genetic heterogeneity (59). This genome plasticity has been offered as an explanation for higher frequencies of antibiotic resistance within these strain backgrounds (60). Differences in fitness between serovars, genetic plasticity and dissimilar resistance mechanisms could partly explain the varying frequencies of resistance in NTS serovars worldwide. Although serogrouping has been useful for differentiating NTS isolates for surveillance studies (61), future studies are needed to characterize the isolates using whole genome sequencing to identify bacterial features commonly associated with specific resistance phenotypes and carriage of specific resistance determinants. Having these data could help identify sources of resistant infections and identify whether some strains are epidemiologically linked.

Our study also identified season to be a predictor of resistant NTS infections, with fewer resistant infections occurring in the summer and more in the fall as is consistent with seasonal variation observed in prior studies. A study of fluoroquinolone resistance in Campylobacter, for example, observed higher frequencies in the winter and spring compared to the summer. This difference was attributed to higher consumption of poultry products containing resistant bacteria in the winter and more frequent exposure to susceptible Campylobacter through other sources in the summer (62). Our prior studies have also observed higher resistance frequencies in C. jejuni (63) and STEC (64) recovered from Michigan patients in the winter and/or spring and could be due, in part, to seasonal variation in antibiotic prescription rates (65, 66). Because prescribing rates may vary by the type of infection and geographic location given climate and other factors, additional studies are required to understand the most important factors that impact antibiotic resistance development and trends in Michigan.

While we did not identify antibiotic resistant NTS infections to be a significant predictor of hospitalization, we did observe a strong association between resistance in NTS and a longer duration (≥5 days) of hospitalization. This finding is not surprising given that resistant infections take longer to clear than susceptible infections during antibiotic treatment and is consistent with other studies. Importantly, studies in other pathogens have identified associations between resistance and mortality, hospitalization, hospital stay duration, and the need for surgery (18, 64, 67, 68).

Because NTS pathogens are a leading cause of enteric infections worldwide and have been shown to frequently resist clinically important antibiotics, continuous surveillance, and monitoring is critical. NTS isolation and routine testing for resistance is imperative in order to help medical personnel and public health officials determine and modify the course of treatment for NTS infections. Furthermore, identifying risk factors for NTS and resistant NTS infections may help in the development of disease management policies and antibiotic use standards aimed at curbing the spread of NTS and resistance determinants.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## ETHICS STATEMENT

The studies involving human participants were reviewed and approved by Institutional Review Boards at Michigan State University (MSU; Lansing, MI, USA; IRB #10-736SM), the MDHHS (842-PHALAB), and the four participating hospitals. Written informed consent from the participants' legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.

## AUTHOR CONTRIBUTIONS

SM performed the experiments with CA and RM organized samples and extracted the epidemiological data. SM and SDM managed the data, conducted the analyses, and drafted the manuscript. SM, SDM, JR, DN, HS, PL, and WK designed the study and organized sample collection at each site. All authors contributed and approved the manuscript content.

## FUNDING

This work was supported by the National Institutes of Health Enterics Research Investigational Network Cooperative Research Center (U19AI090872 to SDM), the Michigan State University (MSU) Foundation, and the United States Department of Agriculture (MICL02475). The Department of Microbiology and Molecular Genetics at MSU, the Ronald and Sharon Rogowski Fellowship and the MSU College of Natural Science provided student support to SM. A portion of these findings were included in SM dissertation from MSU (69).

## ACKNOWLEDGMENTS

We thank Ben Hutton and Jason Wholehan at the MDHHS and the laboratory staff at each participating hospital for help with specimen processing and culture as well as James Collins and Tiffany Henderson at the MDHHS Bureau of Epidemiology for help with MDSS.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed. 2019.00250/full#supplementary-material

## REFERENCES


bloodstream infections and hospitalizations. J Infect Dis. (2005) 191:554–61. doi: 10.1086/427263


those from animal and food sources. J Clin Microbiol. (2011) 49:1130–3. doi: 10.1128/JCM.01931-10


**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 Mukherjee, Anderson, Mosci, Newton, Lephart, Salimnia, Khalife, Rudrik and Manning. 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.

# Constructing and Characterizing Bacteriophage Libraries for Phage Therapy of Human Infections

Shelley B. Gibson<sup>1</sup> , Sabrina I. Green<sup>1</sup> , Carmen Gu Liu<sup>1</sup> , Keiko C. Salazar<sup>1</sup> , Justin R. Clark<sup>1</sup> , Austen L. Terwilliger<sup>1</sup> , Heidi B. Kaplan<sup>2</sup> , Anthony W. Maresso<sup>1</sup> \*, Barbara W. Trautner1,3,4 \* and Robert F. Ramig<sup>1</sup> \*

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Paul Hyman, Ashland University, United States Pieter-Jan Ceyssens, Sciensano (Belgium), Belgium

#### \*Correspondence:

Anthony W. Maresso maresso@bcm.edu Barbara W. Trautner trautner@bcm.edu Robert F. Ramig rramig@bcm.edu

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 25 May 2019 Accepted: 21 October 2019 Published: 12 November 2019

#### Citation:

Gibson SB, Green SI, Liu CG, Salazar KC, Clark JR, Terwilliger AL, Kaplan HB, Maresso AW, Trautner BW and Ramig RF (2019) Constructing and Characterizing Bacteriophage Libraries for Phage Therapy of Human Infections. Front. Microbiol. 10:2537. doi: 10.3389/fmicb.2019.02537 <sup>1</sup> Department of Molecular Virology and Microbiology, Baylor College of Medicine, Houston, TX, United States, <sup>2</sup> Department of Microbiology and Molecular Genetics, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, United States, <sup>3</sup> Center for Innovations in Quality, Effectiveness and Safety, Michael E. DeBakey VA Medical Center, Houston, TX, United States, <sup>4</sup> Department of Medicine, Baylor College of Medicine, Houston, TX, United States

## Phage therapy requires libraries of well-characterized phages. Here we describe the generation of phage libraries for three target species: Escherichia coli, Pseudomonas aeruginosa, and Enterobacter cloacae. The basic phage characteristics on the isolation host, sequence analysis, growth properties, and host range and virulence on a number of contemporary clinical isolates are presented. This information is required before phages can be added to a phage library for potential human use or sharing between laboratories for use in compassionate use protocols in humans under eIND (emergency investigational new drug). Clinical scenarios in which these phages can potentially be used are discussed. The phages presented here are currently being characterized in animal models and are available for eINDs.

Keywords: phage libraries, phage therapy, host range, phage characteristics, killing spectrum, human infection

## INTRODUCTION

The crisis in clinical care imposed by the increasing resistance of bacterial infections to antibiotics threatens to return clinical practice to the pre-antibiotic era (Boucher et al., 2009; Centers for Disease Control [CDC], 2013; Bassetti et al., 2017; World Health Organization [WHO], 2017). The situation is particularly acute for infections caused by Gram-negative pathogens for which few new antibiotics are in the pipeline. The bacterial "mutagenic tetrasect" (mutation, transformation, transduction, and conjugation) is responsible for the rapid evolution of bacteria and suggests that bacteria are so flexible in their ability to adapt that production of antibiotics by pharmaceutical companies, will never be able to keep up with the evolution of resistance against that drug. Fortunately, a natural alternative to conventional chemical antibiotics (Ghosh et al., 2018; Stearns, 2019) exists in the form of bacteriophages (phages). Thus, phages can evolve to efficiently target specific bacteria and have been used to treat complex drug-resistant bacterial infections in a procedure termed phage therapy (Ghosh et al., 2018).

Although phage therapy has great potential as a treatment for antibiotic resistant infections, it is not without problems. Phage have a similar mutation rate to bacteria; the organisms reproduce

so rapidly that the high numbers lead to very large mutant populations on which selection can operate to enrich selected phenotypes. Host range expansion by evolution and selection has been achieved (Burrowes, 2011; Mapes et al., 2016; Burrowes et al., 2019) and is extremely rapid (unpublished data). In addition, phage ecologists have estimated the total number of phages on Earth to be greater than 10<sup>31</sup> (Suttle, 2005). This suggests that the environment is a plentiful resource for new phages; indeed, environmental phages to drug-resistant and pandemic Escherichia coli that have excellent efficacy in animal models of infection have been discovered, characterized, and tested in as short a time period as 14 days (Green et al., 2017).

Phages were discovered over a century ago, and phage therapy has a long history (Debarbieux et al., 2018; Gelman et al., 2018). However, the advent of chemical antibiotics led to virtual abandonment of phage therapy in most of the world, whereas in the countries of Eastern Europe phage therapy was continuously pursued (Chanishvili, 2016; Myelnikov, 2018). As the development of antibiotic resistance has grown, the interest in phage has been rekindled (Kutter et al., 2015). Studies with compassionate use investigational new drug (IND) have generated considerable excitement for use of phage therapy in human subjects (Wright et al., 2009; Schooley et al., 2017; Chan et al., 2018; Aslam et al., 2019). Recently the application of phage therapy to human infections was reviewed (El Haddad et al., 2018), and the majority of the studies analyzed showed efficacy (87%) and safety (67%), however only a few of the studies examined the development of phage-resistant bacteria during therapy. Bacteria become resistant to phage infection (phageresistant) (Labrie et al., 2010), through mutational changes just as they become antibiotic-resistant upon treatment with antibiotics. The problem of phage-resistance is often overcome by [i] the use of single broad host range phages, [ii] phages for which development of phage-resistance carries high bacterial fitness costs (Chan et al., 2016; unpublished data), or [iii] mixtures (cocktails) of phages (generally recognizing different bacterial surface receptors). Others have argued that phage resistance is not a problem in phage therapy because phage resistant bacteria often have fitness defects and new environmental phages active on the resistant host can be isolated (Ormala and Jalasvuori, 2013). Indeed it is likely that phages capable of infecting a resistant host can be isolated from the environment or evolved in the laboratory (Mapes et al., 2016, unpublished data). However, these operations are time consuming and best avoided. In addition, not all phage resistant hosts were found to have fitness defects, so that they could persist in the patient (Wei et al., 2010, our unpublished data).

The development of phages for use against human infections has been described as following two pipelines (Pirnay et al., 2011). The "prêt à porter" (ready-to-wear) is a method, in which a medicinal product of a single broad host range phage is developed and undergoes safety and efficacy testing. This is time consuming and costly but yields products that can be licensed by regulatory agencies. In contrast, in the "surmesure" (custom made) method many phages are isolated and characterized and can be combined as appropriate to treat the infection. This method is flexible, inexpensive, and can rapidly respond to infections with phage- or antibiotic- resistant bacteria. However, sur-mesure approaches cannot currently be licensed, but therapeutic use of phage produced through this approach is under active discussion (Debarbieux et al., 2016; Pirnay et al., 2018). We have chosen the latter approach in which [i] libraries of well-characterized phages are generated and stored, [ii] as the clinical laboratory is assessing the antibiotic-resistance of the bacterial isolate (∼48 h), it is also tested for sensitivity to phages from the library (<48 h), [iii] phages to which the clinical isolate is sensitive are selected for mono-phage-therapy or used to construct cocktails containing several individual phages. Two therapeutic options are available: the patient can be treated with the phage alone, or phage plus antibiotic.

Here, we describe the preparation of well-characterized libraries of E. coli-specific, Pseudomonas aeruginosa-specific, and Enterobacter cloacae-specific phages for use in a sur-mesure approach to phage therapy, which will ultimately result in a therapeutic that is personalized to the specific infection of the individual patient. We provide information on the phages including: descriptions of their sources, their efficacy against a panel of clinical strains, some basic infection characteristics (burst size and absorption rates), and their DNA sequence and annotation to determine if they harbor lysogenic, antibiotic resistance or toxin genes and their morphologic description; providing the means to select high quality phage(s) for use in therapy. Also described are the clinical scenarios for which these phage libraries have been developed, as their proposed use shaped the development of the libraries.

## MATERIALS AND METHODS

## Bacterial Strains

The laboratory strains used were E. coli (MG1655) and P. aeruginosa (PAO1 and BWT111). A collection of 13 E. coli ST131 strains (see **Supplementary Figures S1–S3**) was obtained from Dr. Jim Johnson (University of Minnesota). Two strains of E. cloacae were isolated from a phage therapy candidate with an infected hip prosthesis. One isolate was from a wound swab and other from the fluid exudate of the wound (collected at different times). De-identified clinical isolates of E. coli, P. aeruginosa, and E. cloacae and their antibiotic susceptibility data were obtained from the clinical microbiology laboratory at the Houston Veterans Administration Hospital or Baylor St. Luke's Hospital. Collection of de-identified clinical isolates was approved by the Baylor College of Medicine Institutional Review Board (IRB). An isolated colony of each clinical isolate was streaked on LB agar and grown overnight. A single colony from the LB plate was grown overnight in LB medium, diluted 1:10 into LB medium containing 15% glycerol, and frozen at −80◦C. In cases where clinical isolates appeared to be mixed, the desired species was isolated from differential plates.

## Phages

Four P. aeruginosa-specific phages φKMV, φPA2, φPaer4, and φE2005-24-39 (hereafter called φE2005) were previously described (Mapes et al., 2016). These were the only Pseudomonas

wild type phages used in this work. All E. coli-specific and E. cloacae phages used here were isolated from environmental samples (see **Figures 1**, **4A,B**, **7** for source species) by plaque assay. Fecal samples were suspended to ∼50% (w/v) in PBS, shaken, and centrifuged to remove debris. The supernatant was filtered through a 0.22 micron filter, and 0.1 ml was plated with 0.8% LB top agar containing 100 µl of an overnight culture of the desired isolation strain. After overnight incubation, well-isolated


FIGURE 1 | Summary of characterization of Escherichia coli phages. The characteristics, DNA sequences, growth properties and a summary of phage killing spectra are presented for each phage.

plaques were picked into 1.0 ml phage storage buffer (Mapes et al., 2016), allowed to sit overnight for phage diffusion at 4◦C, and 0.5 ml of suspended phage was used to prepare plate stocks using the isolation strain as host. Plate stocks were harvested and stored at 4◦C.

## Host Range Determination/Efficiency of Killing (Virulence)

To determine phage host range a spot titration protocol was used that allowed us to determine both host range and relative phage killing (EOP). Five microliters of serial 10-fold dilutions of a CsCl purified phage stock (1010–10<sup>12</sup> pfu/ml) were spotted on freshly seeded lawns of control, isolation, or clinical strains. The host range and titer were determined by formation of individual plaques within the area of the spot at terminal dilution. This avoided false positives by determining host range at dilutions where phenomena such as lysis from without (Abedon, 2011) or complementation between defective phages would not be expected. Phage virulence was determined as the efficiency of plating (EOP) (Mirzaei and Nilsson, 2015). EOP was calculated by dividing the titer of the phage at the terminal dilution on the test strain by the titer of the same phage on its isolation strain. On this basis, phages were classified as highly virulent (0.1 < EOP > 1.00), moderately virulent (0.001 < EOP < 0.099), avirulent but active (EOP < 0.001), or avirulent (no plaques detected – see **Figures 2**, **5**, **8**).

## Host Range Expansion (HRE)

Pseudomonas aeruginosa-specific phages were subjected to the HRE protocol as described (Burrowes, 2011; Mapes et al., 2016; Burrowes et al., 2019). Briefly, in a 96 well plate, different host strains were placed in each of the eight rows at a dilution of 1:1000 of overnight culture. Serial 10-fold dilutions of phage (a single phage or a phage mixture) were placed in the 12 columns and the plate was incubated (37◦C) with shaking (225 RPM) for 18 h. Following incubation, for each bacterial strain (row) the supernatant from the well with complete lysis and the adjacent well with higher phage dilution (partial lysis) were combined, into a single tube with the corresponding complete and partial lysis wells of the other bacterial strains. The pooled lysate was treated with CHCl<sup>3</sup> and filtered through a 0.22 micron filter. The filtered lysate was the yield of the 1st cycle of HRE. This filtered lysate was serially diluted 10-fold, and the experiment was repeated using the pooled lysate as the phage and the same bacterial strains as host for cycle 2. The HRE was repeated up to 30 cycles, and yielded a mixture of phages that had replicated on at least one of the host bacterial strains. The heterogeneous mixture of phages in the lysate from any cycle of HRE can be assayed for plaque formers on a host refractory to the parental phage(s), plaques purified, and phage stocks with expanded host range produced (Mapes et al., 2016). The HRE protocol was also successfully applied to E. coli-specific phages (data not shown).

The HRE protocol exposes the lysate (including mutants) that arose during a cycle to new bacteria (unevolved) of the strains used in the previous cycle. Some of the mutations may allow phages to infect and replicate on bacterial strains that were previously refractory to phage, thus amplifying the mutant that contained the host range expanding mutation.

## DNA Sequencing and Annotation of Phage Genomes

CsCl purified phages were submitted to the Center for Metagenomics and Microbiome Research at Baylor College of Medicine for DNA extraction, sequencing and assembly as described previously (Green et al., 2017). Briefly, DNA samples were constructed into Illumina paired-end libraries. The libraries had an average final size of 660 bp (including adapter and barcode sequences) and were pooled in equimolar amounts to achieve a final concentration of 10 nM. The library templates were prepared for sequencing on the Illumina MiSeq. After sequencing, the.bcl files were processed through Illumina's analysis software (CASAVA), which demultiplexes pooled samples and generates sequence reads and base-call confidence values (qualities). The average raw yield per sample was 802 Mbp. For analysis, the adapter sequences were removed, and the sequence was then assembled using SPAdes v3.5.0 (Bankevich et al., 2012) on careful mode, retaining only contigs longer than 1,000 bp and with an average coverage of 1000x or greater. This generated 1–2 contigs per sample, with an average of 74% of the original reads mapping with 100% identity to the final contigs. Genomes were analyzed using both PATRIC's comprehensive genome analysis service (Wattam et al., 2017) and EDGE Bioinformatic software (Li et al., 2017). Gene calling and genome annotation was performed using PROKKA (version 1.13) (Lo and Chain, 2014), RAST (Zerbino and Birney, 2008; Bankevich et al., 2012), GLIMMER3 (version 3.02) (Peng et al., 2010), and GeneMarkS (version 4.28) (Peng et al., 2012). **Figures 1**, **4A,B**, **7** show ORF predictions from the RAST pipeline and tRNA predictions from ARAGORN (version 1.2.36) (Seemann, 2014). ORFs were searched for virulence and antibiotic resistance genes by using BLAST (Aziz et al., 2008) to compare assembled genomes against the Virulence Factor Database (VFDB) (McNair et al., 2018), the PATRIC virulence factor database (Delcher et al., 2007), the Antibiotic Resistance Gene Database (ARDB) (Besemer, 2001) and the Comprehensive Antibiotic Resistance Database (CARD) (Laslett, 2004). ShortBRED (version 0.9.4M) (Johnson et al., 2008) was used for targeted searches of ORFs for genes in the VFDB, CARD, and the Resfam antibiotic resistance gene database (Chen et al., 2016). Genus was inferred from closest sequenced relatives identified by using BWA-Mem (version 0.7.9) (Mao et al., 2015) to aligning contigs to NCBI's RefSeq database and by ORF homology using PHAge Search Tool Enhanced Release (PHASTER) (Liu and Pop, 2009). Phage lifestyles were determined by classifying the genomes using PHACTs (McArthur et al., 2013), using PHASTER to predict integrases and attachment sites, and by parsing all versions of the annotated genome for "integrase." No virulence genes of known toxicity (viral or bacterial) or genes involved in lysogeny were detected. Thus it appears the phages examined here are devoid of any known lysogenic or toxic elements that would preclude their use in phage therapy.


FIGURE 2 | Representative data for antibiotic sensitivity and phage killing (EOP) of clinical isolates of E. coli. Shown are the properties of the E. coli clinical isolates on the left, including: source, date of isolation and antibiotic sensitivity data (VITEK2). On the right are shown the killing spectra of the phages on the individual E. coli clinical isolates. The keys to antibiotic sensitivity and phage killing (EOP) are shown at the bottom of the figure.

## Phage Growth Parameters

The percentage of phage adsorbed in 10 min and the adsorption constant were determined for each phage on its isolation strain. Burst size and latent period (one-step growth curves) were also determined for each phage on its isolation strain (Kropinski, 2009, 2018).

## RESULTS

## Escherichia coli Phages

## E. coli Phage Isolation

Our primary target for E. coli phage isolation was extraintestinal pathogenic E. coli (ExPEC) of the pandemic sequence type 131 (ST131). ExPEC are commonly associated with bacteremia and urinary tract infections, and the ST131 lineage is characterized by multi-drug resistance and its high frequency of isolation over the past 10 years. Interestingly, a rapid screen of common laboratory E. coli-specific phages (T2, T4, T6, T7, λ vir) revealed that none of them formed plaques on the ST131 strains tested. As a result we began to isolate phages from the environment, concentrating on avian and canine species which are known reservoirs for E. coli ST131 (Johnson et al., 2001, 2017). Our first phage isolates were from mixed goose/duck feces collected at a local park (φHP3), canine feces from a dog park (φEC1) and chicken feces from a rehabilitation farm (φCF2). We subsequently isolated phages from raw sewage collected at a local sewage treatment plant (φES series; see **Figure 1**). All phages were plaque-purified three times prior to use.

## E. coli Phage Characteristics

fmicb-10-02537 November 9, 2019 Time: 14:7 # 6

The phage characteristics are summarized in **Figure 1**. The phages varied in plaque size, but plaques tended to be small and clear. Some produced halos around the plaques, suggesting enzymatic activity on the surrounding cells. Regardless of the small plaque sizes, reasonable titer plate stocks were obtained, and all phages could be CsCl purified to about 10<sup>11</sup> pfu/ml. Adsorption curves revealed that the phages ranged widely in adsorption (16–98%). For all the E. coli phages, one-step growth curves revealed the latent period was in the range of 22–40 min and burst sizes ranged from 9.6 to 80.7 pfu/cell. Sequence analysis revealed genome sizes ranging from (53,242 to 170,254 bp) with variable G + C content. The number of encoded ORFs and tRNAs identified was also variable. Notably, none of the sequences contained features that would preclude their use in phage therapy, such as genes to establish and maintain lysogeny, produce toxins or virulence factors, or confer antibiotic resistance (Merabishvili et al., 2018; Hyman, 2019). All of the E. coli phages had myovirus morphology except for φES17 which was a podovirus with an elongated head. Phage ES17 also contained, at marginal statistical significance, an integrase gene when examined with PHACTs and PHASTER software. When colonies were isolated in the presence of excess φES17, no phages were isolated following growth of those colonies in the presence of mitomycin C (data not shown). We concluded that φES17 is a lytic phage, lacking an integrase.

## Host Range of E. coli Phages

The host ranges of the phages was determined as described in Section "Materials and Methods" by spot testing serial 10-fold phage dilutions on the isolation strain and on other laboratory and clinical isolates. The virulence of the phage was determined by comparing the titer on a test strain with the titer on the isolation strain (EOP = titer test/titer isolation). Phages with EOP > 0.1 were considered highly virulent and most useful. Phages with EOP in the range of 0.001–0.099, were considered moderately virulent and may be useful if high titers can be produced. **Figure 2** shows the host range and virulence results for eight E. coli phages on a representative selection of E. coli clinical isolates. A complete determination of host range and virulence on (1) characterized ST131 strains, (2) a collection of paired clinical isolates from two sites (urine and blood) collected from the same patient on the same day, and (3) a set of clinical isolates collected between November 2015 and December 2018 is shown (**Supplementary Figures S1–S3**).

At least one of the E. coli phage isolates was able to kill each member of the ST131 collection, except for E. coli strain JJ1886 (**Supplementary Figures S1–S3**). For individual phages, 8–62% of ST131 strains were killed at EOP > 0.1 and 31–69% were killed at EOPs in the range of 0.001–0.099 (**Supplementary Figures S1–S3**). A cocktail of as few as two E. coli phages (φHP3 and φES17) was capable of killing 12/13 (92%) of ST131 strains (**Supplementary Figures S1–S3**).

Among the 76 E. coli clinical isolates (24 paired blood and urine isolates from the same patient on the same day; 40 single clinical isolates, mostly from patients with UTI; and 12 clinical isolates from urine of catheterized spinal cord injured [SCI] patients) the eight E. coli phages killed from 4 to 57% at an EOP > 0.1. If EOPs between 0.001 and 0.099 (moderately virulent) were included very little increase in the number of clinical isolates killed was observed, except for φHP3 where the number killed was increased by 50% (**Figure 1**). For the 24 paired blood and urine isolates, a cocktail of as few as three of the E. coli phages (φCF2, φES12, and φES17) could be assembled that killed them all at EOP > 0.1. Among the 40 clinical isolates primarily of urinary origin, cocktails (φHP3, φES17, and φES19) capable of killing 35/40 (88%) of the isolates at EOP > 0.1 could be made. Among 12 isolates originating from SCI patients, a cocktail of four phages (φHP3, φEC1, φES12, and φES17) could be made that killed 9/12 (75%) E. coli strains at EOP > 0.1. Among all 76 of the E. coli clinical isolates, we noted no correlation between killing at high efficiency and date of isolation (November 2015– December 2018) or antibiotic-sensitivity phenotype. A summary of the antibiotic sensitivity and phage killing of the 89 total E. coli isolates examined is shown in **Figure 3**. Although none of the individual phages killed more than 50–55% of the 76 bacterial strains at high efficiency, a three phage cocktail increased the high efficiency killing to nearly 90% (**Figure 3**).

FIGURE 3 | Antibiotic Sensitivities and Phage Killing Phenotypes of E. coli Clinical Isolates (N = 89). (A) Antibiotic sensitvities using cut off values used in the microbiology lab at the Houston VA Hospital. (B) The phage killing phenotypes were based on EOP. Strong killers, EOP > 0.1; Moderate Killers, 0.099 > EOP > 0.001; Weak Killers, EOP < 0.00099 but positive; None, no growth. Strong and moderate killers have EOP high enough to be useful in phage therapy. The phage cocktail consisted of equal titers of phages: φHP-3, φES-12, and φES17. Pip/Tazo, piperacillin/tazobactam; TMP/SMX, trimethoprim/sulfamethoxazole.

## Pseudomonas aeruginosa Phages Origin of P. aeruginosa Phages

fmicb-10-02537 November 9, 2019 Time: 14:7 # 7

Four P. aeruginosa-specific phages previously isolated and used by other laboratories were used here: φKMV (Lavigne et al., 2003; Chibeu et al., 2009), φPA2 (ATCC 14203-B1; McVay et al., 2007), φPaer4 (Fu et al., 2010), and φE2005 (Liao et al., 2012). All four phages as a mixture, or single phages, were used in the host range expansion protocol (HRE) as described (Mapes et al., 2016; see section "Materials and Methods"). The HRE protocol generates phage mutants able to infect and replicate on bacterial strains that were previously resistant to the phage (i.e., broadening their host range). The four parental phages subjected to HRE as a mixture lysed 38% of 16 bacterial strains (development strains) used. After 30 cycles of HRE 75% of the 16 development strains were lysed by the phage mixture. When 10 strains different from those used in the HRE process (test strains) were tested 100% of them were lysed by the phage mixture resulting from 30 HRE cycles (Mapes et al., 2016). During this directed evolution process all phages present are mixed, so that at any cycle the lysate is a heterogeneous mixture of phages. Individual plaques were picked after 20 and 30 cycles of HRE and plaque purified after plating on the desired host. For example, P. aeruginosa strain DS38 one of the development strains, was not lysed by the parental phage mixture. The heterogeneous phage mixture from HRE cycle 30 formed plaques on strain DS38 indicating that it contained phages with the host range expanded to DS38. Purified phage clones were generated on strain DS38 from 108 plaques picked from the 30 cycle lysate containing the heterogeneous mixture phages. Among the 108 phage clones that all lysed DS38, there were 30 different killing spectra when they were tested against the 16 development and 10 test strains (Mapes et al., 2016). Similarly, φKMV was subjected to five cycles of HRE, and was found have expanded host range phages in the lysate of cycle 5.

## P. aeruginosa Phage Characteristics

The characteristics of parental and host range expanded phages are shown in **Figures 4A,B**. The characteristics of the phages were variable and similar to those seen for E. coli phages (**Figure 1**). Importantly, DNA sequencing revealed that each of the phages derived from the HRE of the four phage mixture represented only mutants of one of the parental phages. This result indicated that recombination between parental phages did not contribute to expansion of host range in the progeny examined. Thus, the morphology of the HRE-derived clones was not determined but assumed to be like that of the parental phage (**Figures 4A,B**). Sequencing also revealed that none of the phages contained genes that would be detrimental to their use in phage therapy. The growth properties of the HRE-derived phages were also similar to the parental phages.

## Host Range of P. aeruginosa Phages

Host range and virulence of P. aeruginosa phages on representative P. aeruginosa clinical isolates is shown in **Figure 5**. Characterization of the host range and virulence on the complete set of P. aeruginosa clinical isolates tested is shown in the supplementary information (**Supplementary Figures S4, S5**). Compared to the parental phages, the HRE-derived phages had expanded host range when tested against the development strains. They lysed, 19–69% of those strains compared to 12–31% for the parental phages. Likewise the HRE-derived phages had expanded host range when tested against the test strains; they lysed 20–90% of test strains compared to 1–10% for the parental phages (**Supplementary Figures S4, S5**). The parental and HRE-derived phages were then tested against a collection of 64 clinical isolates, which were mostly isolated from patient urine samples that were obtained between November 2015 and August 2017, and displayed a spectrum of antibiotic resistances. Examination of the killing activity (**Supplementary Figures S4, S5**) revealed the HRE-derived phages had expanded host range relative to the parental phages, although many of the phages were expanded at EOP < 0.001, an EOP too low to be useful. It is possible that additional rounds of HRE on some the clinical isolates could generate phage with an EOP in a useful range (EOP > 0.001). Greater numbers of HRE cycles led to greater expansion of host range in the isolated phage clones (**Figures 5**, **6** and **Supplementary Figures S4, S5**), both at highly efficient killing (EOP > 0.1) and at useful levels of killing (EOP > 0.001). This was observed for both HRE using four parental phages (compare 20 and 30 cycles) and for 5 cycles of HRE using a single parental phage φKMV (**Figure 5** and **Supplementary Figures S4, S5**).

**Figure 6** summarizes the antibiotic sensitivity and phage killing phenotypes of the phages on the 64 total P. aeruginosa clinical isolates examined. The increase in useful killing with cycles of the HRE protocol and with mixing of cocktails is shown in **Figure 6B**.

The lack of recombination in HRE-derived phages observed here, contrasts with the contribution of recombination reported by others (Burrowes et al., 2019). In retrospect, this finding is not surprising, since the four phages used were distant phylogenetically, making homology-driven recombination unlikely. The HRE-derived phage sequences contained mutations spread randomly across the genome, but all of them had mutations in the tail fiber gene as would be expected if the host range expansion was based on tail fiber-bacterial receptor interactions. In addition, the sequence analysis revealed no genes that would preclude the use of the HRE-derived phages in phage therapy. Thus, the HRE-derived P. aeruginosa phages are classified as variants of the parental phage to which they corresponded (**Figures 4A,B**).

# Enterobacter cloacae Phages

## E. cloacae Phage Isolation

Enterobacter cloacae phages were isolated from raw sewage collected on two different days by plaquing on a phage therapy candidate's isolates (**Figure 7**).

## E. cloacae Phage Characteristics

Sequence analysis of the phages revealed that the four phages were similar and T4-like (**Figure 7**). The growth properties of the E. cloacae phages were somewhat variable but had parameters within expected values (**Figure 7**). The E. cloacae phages contained no genes that would preclude their use in phage therapy (**Figure 7**).

## E. cloacae Phage Host Range and Virulence

The antibiotic sensitivity and phage killing phenotypes of four phages on E. cloacae clinical isolates are shown in **Figure 8**. The phages were strong killers, especially for E. cloacae isolates from LVAD infections where the original source of the infection may have been the skin. Only one phage strongly killed E. cloacae isolates from UTI. More isolates from various sites must be examined to determine if site of origin of the bacteria affects the efficacy of phage killing. **Figure 9**, a summary of antibiotic sensitivity and phage killing of the isolates examined, shows > 70% of strains killed by all individual phages and only a small gain in killing by a cocktail composed of two phages when compared to the best single phage.


FIGURE 4 | Continued


FIGURE 4 | (A,B) Summary of characterization of Pseudomonas aeruginosa phages. The characteristics, DNA sequences, growth properties and a summary of phage killing spectra are presented for each phage.

## DISCUSSION

This study presents a "sur mesure" approach to phage therapy (Pirnay et al., 2011). Here phage libraries were constructed, characterized, and prepared for use in preclinical or clinical situations. Specifically, we plan to concurrently test clinical isolates against phages from the appropriate library to identify phages for mono- or cocktail-based therapy while they are being characterized in the clinical microbiology laboratory. It will then be the physician's choice to treat the patient with phage alone, antimicrobials alone or the combination of the two. A broader interpretation of the "sur mesure" approach is to develop a phage library using the full spectrum of bacterial strains available in the clinical microbiology laboratory of a specific medical facility, so that the shelf-ready phage strains or cocktails can reasonably be expected to cover multidrug-resistant organisms that cause infections in patients in that facility. These libraries can be tested in the clinical laboratory of the hospital,


or the hospital can send isolates for phage susceptibility testing to basic science laboratories that agree to participate, such as those associated with educational institutions (Center for Phage Technology, Texas A&M University, United States; The Tailored Antimicrobials and Innovative Laboratories for Phage Research [tialφr], Baylor College of Medicine, Houston, TX, United States), government laboratories (The Biological Defense Research Directorate of the Naval Medical Research Center, United States; The Eliava Institute, Tbilisi, Georgia; The Phage Therapy Unit, The Hirzfeld Institute, Poland; Center for Innovative Phage Applications and Therapeutics [IPATH], UCSD, San Diego, CA, United States), or industry partners (AmpliPhi Biosciences; Adaptive Phage Therapeutics). For example, we are prospectively collecting bacterial strains from the urine of all hospitalized patients with SCI at our Veterans Affairs Hospital, so that we can create phage libraries to treat any bacterial pathogens causing urosepsis among those SCI patients. Similarly, at the Baylor-St. Luke's hospital we are collecting all bacterial strains causing infections of left-ventricular assist device (LVAD) drivelines. In both situations our goal is to create a phage library that is able to treat infections caused by the most antibiotic resistant bacteria in that specific clinical setting. While creating phage libraries for many species seems like an attainable goal, it is likely to be more difficult for some species. Phages against Staphylococcus aureus are infrequently isolated from environmental samples (Mattila et al., 2015; Latz et al., 2016) and identification of phages active against Clostridium difficile required induction of lysogens (Hargreaves et al., 2015). Thus, construction of large libraries of phage will depend on the target bacterial species.

Despite the near certainly that phage-resistant bacteria will emerge during therapy, we envision multiple clinical scenarios in which even a single well-timed dose of phage, in addition to standard antibiotics, may be life saving. For example, rapid initiation of effective antimicrobial therapy is essential to

preventing clinical deterioration in sepsis (Kalil et al., 2017). Many patients are at high risk for sepsis caused by antibiotic resistant organisms, by virtue of prior healthcare exposures and/or known colonization with multidrug-resistant organisms. When such high-risk patients present with sepsis, a few early doses of a broad-spectrum phage cocktail used empirically, together with empiric antibiotics could act as a safety net, ensuring adequate coverage of the causative organisms, until the microbiology lab can identify the organism and determine its antibiotic sensitivities. In this scenario a "sur mesure" phage cocktail developed against the full panel of multidrug-resistant organisms isolated in the clinical microbiology laboratory of that specific institution would be used for initial treatment together with empiric antibiotics. Another example of a clinical scenario in which a single dose of phage might be very useful would be to temporarily sterilize a patient's urine prior to an invasive urologic procedure. Phage cocktails mixed specifically for the organisms found in standard pre-procedure urine cultures at a given institution would offer a more targeted approach than our current approach, which involves wiping out the bladder and much of the bowel flora with broad spectrum antibiotics.

In contrast, treatment of biofilm infections, such as those that cause life-threatening LVAD infections, would likely require a longer course of phage therapy, in part because of the longer clinical time frame given the chronicity of LVAD infections. New phage cocktails could be mixed to address phage-resistant bacterial pathogens that might develop during the course of therapy. Alternatively, treating these biofilm infections with phage and antibiotics simultaneously may allow for synergy, particularly if the phage is able to restore antibiotic susceptibility in the infecting pathogen (Comeau et al., 2007; Ryan et al., 2012; Chaudhry et al., 2017). This approach of re-mixing sur-mesure phage cocktails and using them together with an antibiotic to which the infecting organism is resistant was successful in treating a patient with disseminated Acinetobacter infection (Schooley et al., 2017).

To achieve these clinical goals, we have demonstrated that unmanipulated phages isolated from the environment on E. coli ST131, are capable of lysing as many as 58% of a collection of 76 E. coli clinical isolates. Cocktails of as few as three of the individual phages (φHP3, φES12, and φES17) were capable of lysing 92% of the 76 clinical isolates. These results indicate that environmental samples provide good reservoirs of phages capable of being used against E. coli, and that highly effective cocktails can be generated from them. In all cases the cocktails tested were highly effective against clinical isolates, killing at EOP > 0.1. The highly effective killing of the phages and the high titers obtained in the CsCl-purified preparations indicates that these phages should be useful in clinical therapy where a concentrated dose could be administered without fear of generating a septic response due to the presence of contaminating endotoxin (**Figures 1**, **4A,B**, **7**). Similar broad coverage was found for E. cloacae phages. In addition, we demonstrated that laboratory isolates of P. aeruginosa-specific phages can evolve to expand their host ranges to P. aeruginosa clinical isolates. A mixture of four parental phages subjected to 20 or 30 cycles of host range expansion was capable of killing 2–44% of the 64 clinical isolates tested, whereas the uncycled parental phages could lyse only 2–23% of the clinical isolates. However, cocktails containing as few as three individual HRE-derived phages were capable of killing 52 of the 64 clinical isolates tested (81%) (**Figure 6**). Additional cycles of HRE using the 12 clinical isolates not killed by any of the phages at useable EOP (>0.001), seems likely to further expand the host range among those isolates.

In addition to phage isolation for E. coli-, P. aeruginosa-, and E. cloacae-specific phage collections, a number of parameters were characterized that can be useful in choosing phages for phage therapy, making new phage isolates, and general work with the phages (Abedon, 2017). Our "phage master lists" (**Figures 1**, **4A,B**, **7**) contain information on the source of the phages, morphology, growth properties, DNA sequence, and host range much like a Physician's Desk Reference provides useful parameters for chemical antibiotics. The DNA sequence analyses and morphologies of the phages are important to establish the relationship of the individual phage to other phages in the databases (Weber-Dabrowska et al., 2016;


FIGURE 7 | Summary of characterization of Enterobacter cloacae phages. The characteristics, DNA sequences, growth properties and a summary of phage killing spectra are presented for each phage.

Casey et al., 2018). In addition, DNA sequence analysis provides important information on the properties of the phage genome, ensuring that phages can be used as therapeutic agents because they do not encode genes to establish and maintain lysogeny, toxins, virulence factors, or antibiotic resistance. The data on adsorption constant, adsorption rate, latent period, and burst size all represent parameters that can affect the success of phage therapy (Weber-Dabrowska et al., 2016). Finally, in our determinations of phage host range we examined EOP, a parameter that allows one to determine the relative killing power of a phage on a test strain compared to its killing power on the isolation strain. EOP has been shown to be an excellent method for estimating phage virulence on a given bacterial strain. Simple spot tests of high titer phage were found to overestimate


FIGURE 8 | Representative data for antibiotic sensitivity and phage killing (EOP) of clinical isolates of E. cloacae. Shown are the properties of the E. cloacae clinical isolates on the left, including: source, date of isolation and antibiotic sensitivity data (VITEK2). On the right are shown the killing spectra of the phages on the individual E. cloacae clinical isolates. The keys to antibiotic sensitivity and phage killing (EOP) are shown at the bottom of the figure.

both the virulence and the host range of a phage (Mirzaei and Nilsson, 2015). Indeed, we have shown that phage virulence and bacterial susceptibility to the phage determined in vitro allowed us to predict the outcome of therapy in vivo (Green et al., 2017). While bacterial receptors for the phages were not identified here, that information is important for the rational mixing of phage cocktails. We are in the process of identifying receptors for phages in our libraries, and have identified the receptor for φHP3 as lipopolysaccharide in the E. coli JJ2528 host (unpublished data). Having all these parameters at hand aids in the selection of a phage for monotherapy, or a mixture of phages for cocktail therapy. Cesium chloride purified stocks of all phages described here exist and their endotoxin content has been reduced below clinically permissible levels, so that they can quickly be put to use. Our E. coli, P. aeruginosa, and E. cloacae phage libraries are now ready for rigorous in vivo studies in animal models of urinary tract infections and LVAD infections, as well as available for compassionate use protocols in humans.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## ETHICS STATEMENT

The collection of de-identified clinical isolates was approved by the Baylor College of Medicine Institutional Review Board (IRB).

## AUTHOR CONTRIBUTIONS

AM, RR, HK, and BT conceived the experiments and guided their performance. SBG, SIG, CL, JC, AT, and KS performed the experiments and analyzed the results. BT arranged for collection

of clinical isolates and corresponding antibiotic sensitivity data. RR wrote the manuscript with editing from BT, HK, and AM. All authors contributed to manuscript revision, read and approved the submitted version.

## FUNDING

This work was supported by the following funding sources: 1R21AI121545 (RR), The Mike Hogg Fund (AM, RR, BT), The MacDonald General Research Fund (BT, AM, RR), Baylor College of Medicine seed funds (AM), and Veterans Administration RR&D I01 RX002595 (BT). The research reported here was supported in part by the U.S. Department of Veterans Affairs, Veterans Health Administration, Health Services Research and Development Service at the Center for Innovations in Quality, Effectiveness and Safety, Michael E. DeBakey VA Medical Center, Houston, TX, United States (CIN 13-413). The views expressed in this article are those

## REFERENCES


of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States Government.

## ACKNOWLEDGMENTS

We thank Jim Johnson for providing well-characterized E. coli ST131 isolates from his collection. Laura Dillon and Dr. Faisal Cheema assisted in collecting clinical isolates from the hospital. Dr. Jason Kaelber and Kara Schoenemann performed electron microscopy.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02537/full#supplementary-material




**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 Gibson, Green, Liu, Salazar, Clark, Terwilliger, Kaplan, Maresso, Trautner and Ramig. 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 Lytic Bacteriophages Infecting Multidrug-Resistant Shiga Toxigenic Atypical *Escherichia coli* O177 Strains Isolated From Cattle Feces

Peter Kotsoana Montso1,2, Victor Mlambo<sup>3</sup> and Collins Njie Ateba1,2 \*

<sup>1</sup> Bacteriophage Therapy and Phage Bio-Control Laboratory, Department of Microbiology, Faculty of Natural and Agricultural Sciences, North-West University, Mmabatho, South Africa, <sup>2</sup> Food Security and Safety Niche Area, North-West University, Mmabatho, South Africa, <sup>3</sup> Faculty of Agriculture and Natural Sciences, School of Agricultural Sciences, University of Mpumalanga, Mbombela, South Africa

#### *Edited by:*

Leonardo Neves de Andrade, University of São Paulo, Brazil

#### *Reviewed by:*

Robert Czajkowski, University of Gdansk, Poland Kim Stanford, Alberta Ministry of Agriculture and Forestry, Canada Josefina Leon-Felix, Centro de Investigación en Alimentación y Desarrollo (CIAD), Mexico

> *\*Correspondence:* Collins Njie Ateba collins.ateba@nwu.ac.za

#### *Specialty section:*

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

*Received:* 10 July 2019 *Accepted:* 07 November 2019 *Published:* 26 November 2019

#### *Citation:*

Montso PK, Mlambo V and Ateba CN (2019) Characterization of Lytic Bacteriophages Infecting Multidrug-Resistant Shiga Toxigenic Atypical Escherichia coli O177 Strains Isolated From Cattle Feces. Front. Public Health 7:355. doi: 10.3389/fpubh.2019.00355 The increasing incidence of antibiotic resistance and emergence of virulent bacterial pathogens, coupled with a lack of new effective antibiotics, has reignited interest in the use of lytic bacteriophage therapy. The aim of this study was to characterize lytic Escherichia coli O177-specific bacteriophages isolated from cattle feces to determine their potential application as biocontrol agents. A total of 31 lytic E. coli O177-specific bacteriophages were isolated. A large proportion (71%) of these phage isolates produced large plaques while 29% produced small plaques on 0.3% soft agar. Based on different plaque morphologies and clarity and size of plaques, eight phages were selected for further analyses. Spot test and efficiency of plating (EOP) analyses were performed to determine the host range for selected phages. Phage morphotype and growth were analyzed using transmission electron microscopy and the one-step growth curve method. Phages were also assessed for thermal and pH stability. The spot test revealed that all selected phages were capable of infecting different environmental E. coli strains. However, none of the phages infected American Type Culture Collection (ATCC) and environmental Salmonella strains. Furthermore, EOP analysis (range: 0.1–1.0) showed that phages were capable of infecting a wide range of E. coli isolates. Selected phage isolates had a similar morphotype (an icosahedral head and a contractile tail) and were classified under the order Caudovirales, Myoviridae family. The icosahedral heads ranged from 81.2 to 110.77 nm, while the contractile tails ranged from 115.55 to 132.57 nm in size. The phages were found to be still active after 60 min of incubation at 37 and 40◦C. Incremental levels of pH induced a quadratic response on stability of all phages. The pH optima for all eight phages ranged between 7.6 and 8.0, while at pH 3.0 all phages were inactive. Phage latent period ranged between 15 and 25 min while burst size ranged from 91 to 522 virion particles [plaque-forming unit (PFU)] per infected cell. These results demonstrate that lytic E. coli O177-specific bacteriophages isolated from cattle feces are highly stable and have the capacity to infect different E. coli strains, traits that make them potential biocontrol agents.

Keywords: atypical enterophagenic *E. coli* O177, bacteriophages, bacteriophage therapy, biocontrol, biological properties, multi-drug resistance, shiga-toxigenic *E. coli*

## INTRODUCTION

Bacteriophages (phages) are self-replicating viruses, which are capable of infecting and lysing their specific host bacteria (1). They are ubiquitous organisms on Earth estimated to number at 1030-10<sup>32</sup> (2). Phages are relatively safe, nontoxic, and harmless to animals, plants, and humans (3, 4). They are found in various environments related to their host such as in food, soil, sewage water, feces, and farm environments (2). Several bacterial species such as Campylobacter, Escherichia coli, Listeria, Salmonella, Pseudomonas, and Vibrio species are used as hosts to isolate their specific bacteriophages (5–7). Because of their host specificity and nontoxicity, lytic phages are considered to be an alternative solution to combat antimicrobial-resistant pathogens. Outbreaks of listeriosis and widespread occurrence of multidrug resistance in E. coli, Salmonella, and Staphylococcus species have been reported in South Africa (8–11). However, there has been no attempt to use bacteriophages to control antibiotic-resistant pathogens, in either hospital settings or food industry.

Antibiotic resistance in foodborne pathogens, particularly E. coli species, remains a public health concern. Antibiotic-resistant pathogens do not only increase economic and social costs but are also responsible for severe infections in humans (12). In 2014, foodborne infections caused an estimated 600 million illnesses and 420,000 deaths across the globe (13). In addition, 978 listeriosis cases were reported in South Africa from 2017 to 2018, resulting in 183 deaths (11). The leading pathogenic bacteria of concern are E. coli O157, Campylobacter, Listeria, and Salmonella species (14). Moreover, recent reports revealed that non-O157 strains, particularly O26, O45, O103, O111, O121, and O145, exhibit multidrug resistance and are among the leading causes of foodborne infection (15).

In view of the above, several interventions, such as physical, chemical, and biological methods, have been devised and implemented at all levels of the food chain to combat foodborne infection and the spread of antibiotic-resistant pathogens (16). However, these conventional methods have significant drawbacks such as corrosion of food processing plants, environmental pollution, change of food matrices, development of antibiotic resistance, and toxic effects of chemical residues (17). In addition, application of chemical agents coupled with a lack of effective enforcement regulations in food may hamper international trade and thus affect the economy of the exporting country (16, 18). Therefore, lack of new antibiotics and inefficacy of conventional strategies to combat multidrug-resistant bacterial pathogens necessitate the search for alternative control strategies such as the use of bacteriophages. Given their biological properties as explained above, lytic phages can be applied at all levels of the food chain, including preharvest application. Preharvest intervention has the advantage of preventing the transmission of foodborne pathogens from food-producing animals to human.

Considering the virulence and antibiotic resistance profiles of the E. coli O177 strain, coupled with the lack of new antibiotics and limitations of conventional strategies to mitigate antibiotic resistance, there is a need to expand the search for novel bacteriophages for biocontrol application. Therefore, the current study was designed to isolate and characterize lytic E. coli O177-specific bacteriophages as potential biocontrol agents. Stability and viability of the phages were determined under temperature and pH ranges that would be obtained in a live ruminant to assess their stability for preharvest use in these animals.

## MATERIALS AND METHODS

## Bacterial Strain

Multidrug-resistant and virulent atypical enteropathogenic E. coli O177 strain was used to isolate E. coli O177-specific bacteriophages. The atypical enteropathogenic E. coli O177 isolates were obtained from cattle feces and confirmed through PCR analysis. The isolates were further screened for the presence of virulence and antimicrobial gene determinants. Prior to phage isolation, 40 E. coli O177 isolates stored at −80◦C were resuscitated on MacConkey agar and incubated at 37◦C for 24 h. A single colony from each sample was transferred into 15-ml nutrient broth in 50-ml falcon tubes. The samples were incubated in a shaker (160 rpm) at 37◦C for 3 h until the growth reached an optical density (OD) of 0.4–0.5 (600 nm).

## Enrichment and Isolation Purification of *E. coli* O177-Specific Bacteriophages

Escherichia coli O177-specific bacteriophages were isolated from cattle feces using E. coli O177 environmental strain following the enrichment method (19, 20) with some modifications. Twenty fecal samples were collected from two commercial feedlots and two dairy farms. Samples were collected directly from the rectum using arm-length rectal gloves, placed in a cooler box containing ice packs, and transported to the laboratory. Five grams of each fecal sample was dissolved in 20 ml of lambda diluent and vortexed to obtain a homogeneous mixture. The mixture was centrifuged at 10,000 × g for 10 min using Hi Centrifuge SR (model: Z300, Germany) to sediment fecal matter and other impurities. An aliquot of 10 ml from the supernatant was extracted and filter-sterilized using a 0.22-µm pore-size syringe filter (GVS Filter Technology, USA) to obtain crude phage filtrates. For enrichment, 5 ml of each filtrate was added to 100 µl of exponential-phase (OD<sup>600</sup> = 0.4–0.5) culture of each of the 40 E. coli O177 host strains in 10 ml of double-strength tryptic soya broth (TSB) supplemented with 2 mM of calcium chloride (CaCl2). The samples were incubated in a shaking incubator (80 rpm) at 37◦C for 24 h. After incubation, the samples were centrifuged at 10,000 × g for 10 min using Hi Centrifuge SR (Model: Z300, Germany) to remove bacterial cells and sample debris. The supernatant was filter-sterilized with a 0.22-µm poresize Acrodisc syringe filter (GVS Filter Technology, USA) to obtain crude phage filtrates.

Subsequently, a spot test was performed to determine the presence of phages as previously described (19). Briefly, 100 µl of exponential-phase (OD<sup>600</sup> = 0.4–0.5) culture of the bacterial host was mixed with 3 ml of soft agar (0.3% w/v agar) held at 50◦C, then poured onto modified nutrient agar (MNA) plates so as to create a bacterial lawn, and allowed to solidify for 15 min. Ten microliters of each crude phage filtrate was spotted on bacterial lawn, and the plates were incubated at 37◦C for 24 h. After incubation, the plates were observed for the presence of clear zones or plaques at inoculated points. Plaques were picked using a sterile pipette tip and suspended in 1 ml of lambda diluent [10 mM of Tris Cl (pH 7.5), 8 mM of MgSO4·7H2O] in 2-µl Eppendorf tubes. The tubes were left at room temperature to allow phages to diffuse into the solution. The tubes were then centrifuged at 11,000 × g for 10 min, and the supernatant was filter-sterilized with a 0.22-µm pore-size Acrodisc syringe filter (GVS Filter Technology, Germany).

## Bacteriophage Purification and Propagation

Bacteriophages were purified from single plaque isolates using the soft agar overlay method (21, 22). Plaque assay was performed, and the plates were incubated at 37◦C for 24 h. After incubation, single plaques from each plate were picked based on their sizes and clarity using a sterile pipette tip and were resuspended in 1 ml of lambda diluent in 2-µl Eppendorf tubes. The tubes were left at 4◦C for 24 h to allow phage to diffuse into the buffer. The tubes were then centrifuged at 10,000 × g for 10 min, and the supernatant was filtersterilized using a 0.22-µm pore-size Acrodisc syringe filter (GVS Filter Technology, Germany). The purification process was repeated three consecutive times until homogeneous plaques were obtained for each phage isolate. Purified phages were propagated using E. coli O177 host bacteria. One hundred microliters of pure phage stocks was mixed with 100 µl of exponential-phase (OD<sup>600</sup> = 0.4–0.5) culture of corresponding host(s) in a 50-ml falcon tube containing sterile 10-ml doublestrength TSB supplemented with 2 mM of CaCl2. The mixture was incubated in a shaking incubator (150 rpm) at 37◦C for 24 h. After incubation, the samples were centrifuged at 8,000 × g for 10 min at 4◦C, and the supernatant was filter-sterilized using a 0.22-µm pore-size Acrodisc syringe filter (GVS Filter Technology, Germany). Ten-fold serial dilutions were prepared, phage titers were determined using plaque assay, and the titers were expressed as PFU per milliliter. The stock phages were stored at 4◦C for further analysis.

## Characterization of Selected *E. coli* O177-Specific Bacteriophages Host Range and Cross Infectivity of the Phage Isolates

The host range of eight selected phage isolates was evaluated against 50 bacterial hosts [13 E. coli O177, 12 E. coli O157, 12 E. coli O26, and 10 Salmonella species (environmental strains), 1 Pseudomonas aeruginosa (ATCC 27853), 1 Salmonella enterica (ATCC 12325), and 1 Salmonella typhimurium (ATCC 14028)], and all environmental species were isolated from cattle feces. Phage isolates were selected based on different plaque morphologies and clarity of the plaques and sizes. The spot test technique was performed to determine lytic spectrum activity of each phage isolates as previously described (21). The bacterial lawns of all the selected bacterial hosts were prepared on MNA plates. Ten microliters of phage stock (107–10<sup>9</sup> PFU/ml) was spotted on bacterial lawn and allowed to air-dry under laminar airflow for 10 min. The plates were incubated at 37◦C for 24 h. After incubation, the plates were observed for the presence of plaques at the point of application, and the phage lytic profiles were classified into three categories according to their clarity: clear, turbid, and no lysis (23). The test was performed in triplicates for each phage isolate.

## Efficiency of Plating of Phages

Efficiency of plating (EOP) was performed to determine lytic efficiency of the phage in comparison with their suitable host bacteria as previously described (24), with modification. Fifteen bacterial strains (five E. coli O177, five E. coli O26, and five E. coli O157) were selected based on their sensitivity against the phages. Ten-fold serial dilutions of phages were prepared to obtain single plaques. An aliquot of 100 µl of each phage (1 × 10<sup>4</sup> PFU/ml) was mixed with 100 µl of exponential-phase (OD<sup>600</sup> = 0.4–0.5) culture of each bacterium in 50-ml sterile falcon tubes and left for 10 min at room temperature to allow the phage to attach to the host. Then, 3 ml of soft agar (0.3% w/v) was added to the tube, and the mixture was poured onto MNA plates. Three independent assays were performed for each phage isolate. After solidifying, the plates were incubated at 37◦C for 24 h. After incubation, the number of plaques per plate was counted. The EOP was calculated as the ratio between the average number of plaques on target host bacteria (PFUs) and average number of plaques of reference host bacteria (PFUs). The EOP was classified as high (EOP ≥ 0.5), moderate (EOP > 0.01 < 0.5), and low (EOP ≤ 0.1) based on the reproducible infection on the targeted bacteria (25). The following formula was used to calculate EOP values:

Relative EOP = average number of plaques on targeted host bacteria (PFUs) average number of plaques on reference host bacteria (PFUs)

## Transmission Electron Microscopy Analysis

Eight phage isolates were subjected to transmission electron microscopy (TEM), and phage morphotype was determined using negative staining techniques as previously described (26), with some modifications. Briefly, phages were propagated to obtain high titer (108–10<sup>11</sup> PFU/ml). Ten milliliters of each phage (108–10<sup>11</sup> PFU/ml) was concentrated in 50-ml falcon tubes by adding 10% (w/v) PEG, and the mixture was incubated at 4◦C overnight to allow precipitation of the phage particles. After incubation, phage particles were sedimented by centrifugation at 11,000 × g for 10 min at 4◦C. The supernatant was discarded, and the pellet was washed three times with 0.1 M ammonium acetate (pH 7.0). The pellet was resuspended in 200 µl of ammonium acetate. Ten microliters of concentrated phage solution was deposited on 200-mesh copper grids with carboncoated formvar films. The phage particles were allowed to adsorb for 2 min, and excess liquid was drained off with a sterile filter paper. The grid was allowed to air-dry. A drop of 1% (w/v) ammonium molybdate (aqueous, pH 6.5) was added to negatively stain the phage particles and allowed to air-dry for 10–15 min. The grid containing the specimen (phage particles) was then loaded into a transmission electron microscope (model: FEI Tecnai; TEM, Czech Republic) and operated at 120 kV to scan and view phage images with a magnification range of 20,000–100,000. Micrographs were taken with a Gatan bottommounted camera using Digital Micrograph software at 80 kV and a magnification range of 20,000–250,000. The images were taken, and morphology characteristics were used to classify phage isolates as previously described (27).

## Effect of Different Temperatures and pH on the Stability and Viability of Phages

Phage stability and viability were evaluated across different temperatures (37 and 40◦C) over a 60-min period in a temperature-controlled incubator. The concentration of the host bacteria and phage titers was standardized before starting the experiment. One hundred microliters of exponential-phase culture [10<sup>5</sup> colony-forming unit (CFU)/ml] and 100 µl of phages (10<sup>5</sup> PFU/ml) was added to 10 ml of double-strength TSB supplemented with 2 mM of CaCl2. The tubes were incubated in a preset shaking incubator at 37 and 40◦C for 60 min, and samples were taken at 10, 30, and 60 min of incubation and assessed for viability and concentration using double-layer agar (22). Plaque assay was performed in triplicates for each sample, and the results were expressed as PFU per milliliter. For pH, phages were exposed to different pH levels (3.0, 4.5, 6.3, 7.0, 8.5, and 10.0) in a 48-h incubation period. Ten milliliters of sterile double-strength TSB (amended with 2 mM of CaCl2) was distributed into 50-ml falcon tubes to prepare different pH solutions. The pH was adjusted using hydrochloric acid (HCL, 6M) or sodium hydroxide (NaOH, 6M). One hundred microliters of each bacterial host (10<sup>5</sup> CFU/ml) and their corresponding phage (10<sup>5</sup> PFU/ml) isolates were added to 10 ml. The tubes were incubated in a preset shaking incubator (80 rpm) at 37◦C for 48 h. Samples were taken at 24 and 48 h of incubation, and the phage titer for each sample was determined using the standard plaque assay. Plaque assay was performed in triplicates for each sample, and the results were expressed as PFU per milliliter.

## One-Step Growth Curve

A one-step growth curve experiment was performed to determine the latent period and burst size of the selected phages as previously described (21), with some modifications. Briefly, 5 ml of exponential-phase culture of each host was centrifuged at 8,000 × g for 5 min at 4◦C. The pallet was resuspended in 10 ml of double-strength TSB supplemented with 2 mM of CaCl<sup>2</sup> to obtain an OD of 0.4–0.5 (600 nm). The bacterial concentration was adjusted using sterile double-strength TSB to obtain 1 × 10<sup>8</sup> CFU/ml. Each phage (10<sup>8</sup> PFU/ml) was added to its respective host bacterial suspension to achieve multiplicity of infection (MOI) 1.0. The mixture was left at room temperature for 10 min to allow phages to adsorb to the host bacteria. After 10 min, 1.5 µl of the mixture was transferred into 2-µl Eppendorf tubes and centrifuged at 11,000 × g for 10 min to remove unadsorbed phage particles. The pellet was resuspended in 100 µl of TSB supplemented with 10 mM of magnesium sulfate (mTSB) and transferred into 9.9 ml of prewarmed mTSB. The samples were incubated in a shaking incubator (160 rpm) at 37◦C for 1 h. Two hundred microliters was drawn from each sample at 5-min intervals for 60 min. Plaque assay was performed in triplicates for each samples to determine phage titer. The data generated were used to determine the latent period, burst time, and phage relative burst size per infected cell. The burst size was calculated as the ratio of the final count of released phage progeny to the initial count of infected bacterial host cell during the latent period using the following formula as previously described (28):

Relative burst size = final titer (PFU) − initial titer (PFUs) initial titer (PFUs) (PFUs)

The relative burst size at different time points was plotted against time to determine the latent period and burst size of each phage isolate.

## Statistical Analysis

The viability and stability of phages were tested at different temperatures and pH levels. The data were converted to log<sup>10</sup> PFU per milliliter and analyzed using SAS (2010). The effect of temperature, time, and phage type on viability and stability of phages was analyzed using the general linear model (GLM) procedure of SAS (2010) for a 2 (temperature) × 3 (time) × 8 (phages) factorial treatment arrangement according to the following model:

$$\begin{aligned} \label{eq:1} Y\_{ijkl} &= \mu \, + \, T\_i \, + \, \mathbb{S}\_j \, + \, V\_k \, + \, (T \times \mathbb{S})\_{ij} \, + \, (T \times \, V)\_{ik} \, \\ &+ \, (\mathbb{S} \times \, V)\_{jk} \, + \, (T \times \mathbb{S} \times \, V)\_{ijk} \, + \, E\_{ijkl} \end{aligned}$$

where Yijkl is the observation of the dependent variable ijkl; µ is the fixed effect of population mean for the variable; T<sup>i</sup> is the effect of temperature; S<sup>j</sup> is the effect of time; V<sup>k</sup> is the effect of phages; (T × S)ij is the effect of interaction between temperature at level i and time at level j; (T × V)ik is the effect of interaction between temperature at level i and phage at level k; (S × V)jk is the effect of interaction between time at level j and phage at level k; (T × S × V)ijk is the effect of interaction between temperature at level i, time at level j, and phage at level k; and Eijkl is the random error associated with observation ijkl.

Phage viability and stability data in response to incremental levels of pH were evaluated for linear and quadratic effects using polynomial contrasts. Response surface regression analysis (Proc RSREG; SAS 2010) was applied to describe the responses to pH according to the following quadratic model: y = a + bx + cx<sup>2</sup> , where y is the response variables, b and c are the coefficients of the quadratic equation, a is the intercept, x is the pH level, and – b/2c is the x value for maximum response. For all statistical tests, significance was declared at p ≤ 0.05.

## RESULTS

## Isolation, Purification, and Propagation of Bacteriophages

Thirty-one lytic E. coli O177-specific bacteriophages were isolated from cattle feces. Phages were able to infect 15% of the E. coli O177 isolates used for isolation. Phage isolates were designated as ECPV, according to the genus of the host bacteria,

FIGURE 2 | Representative image depicting spot test results of phages on different Escherichia coli strains.

followed by the notation of phage virus and a numeric number as identity. Phage isolates revealed different plaque morphologies in terms of sizes, ranging from small to large (1–2 mm, respectively) plaques (**Figure 1**). A large proportion (71%) of the phage isolates revealed large plaques, while a small proportion (29%) showed small plaques on their preferred hosts. All the phages revealed clear (complete lysis) plaques, and no turbid plaques were observed. Phage titer after propagation ranged from 6.2 × 10<sup>5</sup> to 3.1 × 10<sup>13</sup> PFU/ml. Phage EC3A2PV had the lowest titer, while phage EC198B1PV had the highest titer compared to other phage isolates.

## Host Range of Phages and EOP Analysis Against Different *E. coli* Strains

A spot test was performed to determine the host range of eight selected lytic phage isolates against 50 bacterial hosts comprising different bacterial species. The results indicated that the phages were capable of infecting E. coli species (E. coli O177, E. coli 0157, and E. coli O26 environmental strains) tested (**Figure 2**). All phage isolates produced clear plaques against all E. coli O177 and 83–100% of E. coli O26 strains (**Table 1**). Three phages (EC10C3PV, EC11B2PV, and EC12A1PV) were able to infect E. coli O157 (75–83%; **Table 2**). None of the phages could infect ATCC strains and environmental Salmonella species. The EOP analysis was performed on 15 (five E. coli O177, five E. coli O26, and five E. coli O157) isolates that were susceptible to phages on the spot test. Although spot test results revealed clear plaques on E. coli O177, EOP results exhibited various lytic patterns of the phages. Even though EOP analysis revealed high (EOP ≥ 0.5) productive infection on E. coli O177, moderate infections were observed (**Table 2**). Four phages revealed high EOP values (0.5– 0.8) on E. coli O177 isolates. On the other hand, EOP analysis exhibited moderate and low productive infections on E. coli O26 and E. coli O157 isolates (EOP values range from 0.0 to 0.4 and 0.0 to 0.3, respectively).

## Morphological Characterization of Phages Based on TEM

Eight selected phage isolates were subjected to TEM analysis to determine their morphotype. Transmission electron micrograph images of the phages and structural dimensions are shown in **Figure 3** and **Table 3**, respectively. Phage isolates were classified as per the International Committee on Taxonomy of Virus (ICTV) classification based on the three-dimensional structure observed. All phage isolates showed similar morphotype on TEM analysis. Structurally, the phages had an icosahedral head and a neck attached to a long contractile tail, with tail fibers, and they were classified under the order Caudovirales, Myoviridae family. The phage icosahedral heads ranged from 81.2 ± 6 to 95.6 ± 3 nm while the contractile tails ranged from 118.1 ± 0.3 to 135 ± 2 nm. Phage EC10C2PV had the longest icosahedral head with a diameter of 95.6 ± 3 nm and the longest contractile tail of 135 ± 2 nm with fibers. Phage EC10C3PV had the smallest icosahedral head with a diameter of 81.2 ± 6 nm and the shortest contractile tail of 118.1 ± 0.3 nm with fibers.

## Phage Stability and Viability Against Different Temperatures

The results showed a significant (p < 0.001) time × temperature interaction effect on the stability and viability of the phages.

#### TABLE 1 | Host range infection of the phages.


<sup>a</sup>,b,c,dATCC 27853, ATCC 12325, ATCC 14028, and environmental strains, respectively.

TABLE 2 | Efficacy of plating of phages against different Escherichia coli serotypes.


ID, identity.

<sup>a</sup>Reference host.

FIGURE 3 | Transmission electron micrograph images of representative phage isolates negatively stained with 1% ammonium molybdate. Both phages (A: EC11B2PV; B: EC118BPV) belong to the myoviridae family and are showing icosahedral capsid and long contractile tail with tail fibers. The bars indicate scale (100 nm).

Incubation of phages from 10 to 60 min resulted in significant phage growth at 37◦C (**Figure 4**). The growth from 10 to 30 min ranged from 8.55 to 8.75 log<sup>10</sup> PFU/ml (at 37◦C). Phage EC3A1PV revealed the lowest growth rate, while phage EC10C3PV exhibited the fastest growth rate from 10 to 60 min. Phage growth at 40◦C when incubated for 10 to 60 min is TABLE 3 | Phage dimensions based on transmission electron microscopy (TEM) analysis.


depicted in **Figure 5**. From 10 to 30 min, seven phages revealed significant growth rates (ranging from 8.71 to 9.13 log<sup>10</sup> PFU/ml). A decrease in phage EC12A1PV's growth rate was observed (from 10 to 60 min) while two phages (EC10C2PV and EC10C3PV) exhibited a decrease in growth rate from 30 to 60 min of incubation period.

An exposure of individual phages to 37◦C for 10–30 min showed a significant increase in growth rate with time (0.4– 2.3% growth rate at 37◦C; **Figures 6**, **7**). The average growth

FIGURE 6 | Survival and stability of individual phages when exposed to different temperatures for 10 min. The error bars represent the standard deviation. PFU, plaque-forming unit.

rate of the phages increased from 0.2 to 0.17 log<sup>10</sup> PFU/ml at 37◦C (10–30 min of incubation period). Phages revealed various responses when incubated at 40◦C. A decline in growth rate in other phages was observed. When incubated for 10 min at 40◦C, EC10C3PV's growth rate declined the most by 4.8%, while phage EC3A1PV's growth rate was the least affected, showing only a 0.2% decline. After 30 min of incubation, a decline in growth rate was observed in phages EC366BPV (0.2%), EC366VPV (0.2%), and EC10C3PV (2.4%) (**Figure 7**). When incubated for 60 min at 40◦C, phages EC10C2PV (0.28 log<sup>10</sup> PFU/ml), EC10C3PV (0.39

log<sup>10</sup> PFU/ml), and EC12A1PV (0.37 log<sup>10</sup> PFU/ml) exhibited the greatest decline in growth rates when compared to their respective rates for 60 min at 37◦C (**Figure 8**).

## Phage Stability and Viability Against Different pH Levels

Response surface regression analysis revealed quadratic effects (p < 0.0001) of pH on phage stability when incubated for 24 and 48 h (**Figures 9A–H**, **10A1–H1**, respectively). The pH optima for all the phages ranged from 7.6 to 8.0 with the R 2 values ranging from 0.90 to 1.0 when incubated for 24 h (**Figures 9A**–**H**). Three phages showed maximum stability at pH 8.0 determined from the following quadratic equations: y = −13.9 (±2.93) + 6.4 (±0.98)x – 0.4 (±0.07)x 2 (EC10C2PV), y = −13.9 (±3.04) + 6.4 (±1.01)x – 0.4 (±0.08)x 2 (EC366BPV), and y = −13.9 (±2.97) + 6.4 (±0.99)x – 0.4 (±0.08)x 2 (EC366VPV). Phages EC10C3PV and EC11B2PV exhibited maximum stability at pH 7.6, which was determined from the following quadratic equations: y = −13.4 (±2.87) + 6.1 (±0.95)x – 0.4 (±0.07)x 2 and y = −13.5 (±2.94) + 6.1 (±0.98)x – 0.4 (±0.07)x 2 , respectively. When incubated for 48 h, pH optima for phage stability ranged from 7.9 to 8.0 with R 2 values ranging from 0.90 to 1.0 (**Figures 10A1–H1**). Seven phages exhibited maximum stability at a higher (8.0) optimum pH while only one phage showed maximum stability at a lower (7.9) pH determined from the quadratic equation y = −13.8 (±3.10) + 6.3 (±1.03)x – 0.4 (±0.08)x 2 .

## One-Step Growth Curve Bacteriophages

A one-step growth curve analysis for the eight phage isolates was performed to determine the latent period and relative burst size per infected bacterial cell. Data generated were analyzed and used to construct triphasic curves (**Figures 5**–**11A–H**). The latent period for all the phages ranged from 15 to 25 min (average = 20 ± 3.8 min). Phages EC11B2PV and EC3A1PV had the longest latent period (25 min), while phages EC118BPV and EC366VPV had the shortest (15 min) latent period. The latent period for the other four phages was 20 min. In terms of burst size, phages EC10C3PV and EC12A1PV had the largest burst size per infected cell (522 and 367 PFUs, respectively), while phage EC366VPV had the smallest burst size (91 PFUs) per infected cell.

## DISCUSSION

The emergence of antibiotic resistance in foodborne pathogens has revitalized interest in the possible exploitation of lytic bacteriophages as an alternative biocontrol strategy. Because of their ability to lyse multidrug-resistant pathogens, lytic bacteriophages are considered as a natural and green technology for food preservation and safety (29). The isolation, identification, and full characterization of the bacterial host is a prerequisite for the successful isolation of suitable lytic bacteriophages intended for biocontrol of antimicrobial foodborne pathogens (30). Furthermore, reliable, reproducible, and efficient methods need to be employed for selection of suitable phage candidates for biocontrol application (31). In this study, 31 lytic E. coli O177-specific bacteriophage isolates were successfully isolated from cattle feces using multidrug-resistant atypical enteropathogenic E. coli O177 as a host. Since cattle are the main reservoirs of the atypical enteropathogenic E. coli O177 strain, this supports the idea that phages are present in every ecosystem where their hosts exist (4). The phages exhibited clear and discrete plaques with different sizes. The plaque size ranged between small and large (1–2 mm, respectively) sizes while phage titers ranged from 6.2 × 10<sup>5</sup> to 3.1 × 10<sup>13</sup> PFU/ml. Interestingly, a large proportion (71%) of phage isolates produced large and clear plaques on their preferred hosts. These characteristics were similar to those reported for E. coli O157-, Listeria-, Pseudomonas-, Salmonella-, and Vibrio-specific phages (23, 32, 33). From a biocontrol point of view, strictly lytic phages with high titers are considered as ideal candidates for biocontrol application (4, 34).

Host range is one of the most important criteria when selecting phages intended for biocontrol of antimicrobial foodborne pathogens (35). Eight phages were selected to determine phage host range. The selection criteria were based on the lytic profiles, plaque clarity, and size of the phages. A spot test revealed that the phages were capable of infecting different E. coli strains from two different categories [environmental atypical enteropathogenic E. coli O177 and Shiga toxin-producing E. coli (E. coli O26 and E. coli O157)]. Clear plaques were predominantly observed on E. coli O177 and E. coli O26 serotypes. Interestingly, three phages exhibited clear plaques on E. coli O177, E. coli O26, and E. coli O157 strains, suggesting that these phages were polyvalent, infecting strains from two different categories. Despite this, no phage could infect all the ATCC strains and environmental Salmonella species tested in this study. This could be attributed to the fact that ATCC strains and Salmonella species lack specific receptors for phage attachment. Based on EOP analysis, phages revealed high efficiency (EOP ≥ 0.5) on the E. coli O177 strain. Despite the fact that all the phages revealed clear plaques on E. coli O26 and O157 strains on the spot test, only three phages exhibited medium to low EOP (<0.5) on E. coli O26 and O157 serotypes. This suggested that phages were highly specific to the E. coli O177 strain. Moreover, host specificity is regarded as a desirable characteristic for potent phage application, particularly in live animals to ensure that they have little or no impact on the beneficial gut microflora (2, 6). Furthermore, infectivity variation might be due to the

non-specific binding receptors on the host cell wall or the presence of phage-resistant strains (6).

A negative staining procedure was used for TEM analysis. Based on TEM results, all eight phage isolates revealed a similar morphotype. Structurally, the phages had an icosahedral head and a neck attached to a long contractile tail with tail fibers connected to the baseplate. The icosahedral head of the phages ranged from 81.2 ± 6 to 95.6 ± 3 nm in size while the contractile tails ranged from 118.1 ± 0.3 to 135 ± 2 nm. Based on these characteristics, phage isolates were classified under the order Caudovirales and Myoviridae family (27). Moreover, these characteristics were similar to those of T1-7-like E. coli phages (4, 36). Given the fact that the Myoviridae family contains doublestranded DNA phages (4), all eight phages were presumptively classified under linear double-stranded DNA phages. The tail fibers contain proteins, which help the phage to recognize their specific receptors on the bacterial cell wall and thus restrict the phage from binding to non-specific bacterial cell (37). This explains the host specificity of the phages isolated in this study.

External factors such as pH and temperature may influence the stability and infectivity of the phages (7). These factors may fluctuate, particularly in live animals because of diet and/or ambient temperature. In view of this, phages intended for biocontrol application, particularly in live ruminants, must be tested against an appropriate range of pH and temperature. For this reason, the effect of exposure to different temperatures (37 and 40◦C) for different times on infectivity and stability of eight phages was evaluated. Given that complete bacterial lysis by phage takes 20–40 min (33), phage growth at different temperatures was monitored after 10, 30, and 60 min. Furthermore, the incubation temperatures were selected because

standard deviation.

the temperature in the digestive system of the ruminant ranges from 37 to 40◦C. The ability of phages to survive at these temperatures suggests that they can be applied in live animals as biocontrol agents. When incubated at 37◦C, phages revealed a significant growth rate at each time point. Phages EC10C3PV, EC11B2PV, and EC366VPV revealed the fastest growth rates; phage EC3A1PV showed the slowest growth rate from 10 to 60 min. These results are similar to those reported in previous studies (28).

Phages revealed variable growth patterns when exposed to 40◦C. Generally, phages showed a decline in growth rate at 40◦C when compared to their growth rate at 37◦C. When exposed to 40◦C for 10–30 min, seven phages exhibited a significant growth rate. However, one phage revealed a decline in growth rate after 30 min while two phages, EC10C2PV and EC10C3PV, only exhibited a decline in growth from 30 to 60 min at 40◦C. This demonstrates that these three phages were less stable at high temperature and, therefore, their application in live animals is limited because the rumen temperature is 39◦C. Despite this, the other five phages were fairly stable at 40◦C, suggesting they may be suitable for biocontrol application in live animals.

The response surface regression analysis revealed a significant relationship between pH and phage stability. The optimal pH for phages at different incubation times ranged from pH 7.6 to 8.0 (24 h) and pH 7.9 to 8.0 (48 h). When incubated for 24 and 48 h, all phages exhibited similar growth trends and survival over a wide range of pH (4.5–10.0). Despite this, all the phages were sensitive to low pH (3.0) with no activity being observed after 24 and 48 h of incubation at this pH. This is consistent with previous studies, which reported that exposure of phages to pH 3.0 and below significantly reduced the viability and stability of phages (38, 39). Although the optimum pH for all the phages is 7.6–8.0, phages revealed good stability even at lower pHs of 6.3 and 7.0, which encompasses rumen pH values (6.5–6.9). And this indicates their potential suitability for use in preharvest intervention strategies that may be designed for application in ruminants.

Phage latent period and burst size are important parameters to consider when selecting phages for biocontrol purposes (5, 31). Phages with a short latent period and large burst size are more effective in inactivating bacteria and are thus considered to be suitable for biocontrol application (35). One-step growth curves revealed that the eight phages have different patterns of growth, suggesting that they have distinct genotypes. They displayed outstanding characteristics such as short latent periods and large bust sizes, which make them attractive for the control of the E. coli O177 strain. The latent period of phages ranged from 15 to 25 min, while the burst size ranged from 91 to 522 PFU per host cell. In addition, the average latent period for all the phages was 20 ± 3.8 min while the burst size was 260 ± 144 PFU per host cell. These results were consistent with those reported previously (35). Two phages, EC118BPV and EC366VPV, had the shortest latent period (15 min) while EC11B2PV and EC3A1PV had the longest latent period (25 min). Phages EC10C3PV, EC118BPV, and EC12A1PV had the largest burst size per infected cell (522 and 367 PFU per host cell, respectively). Interestingly, these three phages also showed broad host range in the spot test. This demonstrates that these phages are better suited for biocontrol application (40).

## REFERENCES


In conclusion, lytic bacteriophages infecting the E. coli O177 environmental strain were successfully isolated in this study. Furthermore, phages were capable of infecting three E. colistrains from two different categories, atypical enteropathogenic E. coli (E. coli O177) and Shiga toxin-producing E. coli (E. coli O26 and E. coli O157). Despite this, no phage could infect ATCC strains and environmental Salmonella species tested. Considering strong lytic activity, broad spectrum, and stability at different temperatures and pH levels, phages isolated in this study are considered as potential candidates for in vivo control of the E. coli O177 strain. However, further studies using appropriate in vitro and in vivo models are required to evaluate the efficacy of E. coli O177-specific phages in reducing E. coli O177 in live animals and meat products. Moreover, whole-genome sequence analysis is also required to determine the presence of undesirable genes in these phages.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## ETHICS STATEMENT

The statement, Ethical clearance was obtained from the Faculty of Natural and Agricultural Sciences Ethics committee, North West University prior to the commencement of the study. An ethics number NWU-01223-19-S9 was assigned to the study.

## AUTHOR CONTRIBUTIONS

VM and CA conceived and designed the experiments, contributed reagents, material, and analysis tools. PM performed the experiments. PM, VM, and CA wrote the paper and data analysis.

## FUNDING

This research was supported financially by the National Research Foundation (Grant Number: 112543) and the North-West University Postgraduate Bursary.

## ACKNOWLEDGMENTS

The authors wish to thank Dr. Anine Jordaan for her technical support during TEM analysis for bacteriophages and Mr. B. J. Morapedi for his assistance in sample collection.


Escherichia coli, Klebsiella pneumoniae and Enterobacter species. Front Microbiol. (2019) 10:574. doi: 10.3389/fmicb.2019.00574


**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 Montso, Mlambo and Ateba. 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.

# Is Meropenem as a Monotherapy Truly Incompetent for Meropenem-Nonsusceptible Bacterial Strains? A Pharmacokinetic/Pharmacodynamic Modeling With Monte Carlo Simulation

## Xiangqing Song\*, Yi Wu, Lizhi Cao, Dunwu Yao and Minghui Long

Department of Pharmacy, Hunan Cancer Hospital, The Affiliated Cancer Hospital of Xiangya School of Medicine, Central South University, Changsha, China

Edited by: Hemda Garelick, Middlesex University, United Kingdom

## Reviewed by:

Jozsef Soki, University of Szeged, Hungary César de la Fuente, University of Pennsylvania, United States

> \*Correspondence: Xiangqing Song sxqmaster@163.com

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 04 July 2019 Accepted: 14 November 2019 Published: 29 November 2019

#### Citation:

Song X, Wu Y, Cao L, Yao D and Long M (2019) Is Meropenem as a Monotherapy Truly Incompetent for Meropenem-Nonsusceptible Bacterial Strains? A Pharmacokinetic/Pharmacodynamic Modeling With Monte Carlo Simulation. Front. Microbiol. 10:2777. doi: 10.3389/fmicb.2019.02777 Infections due to meropenem-nonsusceptible bacterial strains (MNBSs) with meropenem minimum inhibitory concentrations (MICs) ≥ 16 mg/L have become an urgent problem. Currently, the optimal treatment strategy for these cases remains uncertain due to some limitations of currently available mono- and combination therapy regimens. Meropenem monotherapy using a high dose of 2 g every 8 h (q 8 h) and a 3-h traditional simple prolonged-infusion (TSPI) has proven to be helpful for the treatment of infections due to MNBSs with MICs of 4–8 mg/L but is limited for cases with higher MICs of ≥16 mg/L. This study demonstrated that optimized two-step-administration therapy (OTAT, i.e., a new administration model of i.v. bolus plus prolonged infusion) for meropenem, even in monotherapy, can resolve this problem and was thus an important approach of suppressing such highly resistant bacterial isolates. Herein, a pharmacokinetic (PK)/pharmacodynamic (PD) modeling with Monte Carlo simulation was performed to calculate the probabilities of target attainment (PTAs) and the cumulative fractions of response (CFRs) provided by dosage regimens and 39 OTAT regimens in five dosing models targeting eight highly resistant bacterial species with meropenem MICs ≥ 16 mg/L, including Acinetobacter baumannii, Acinetobacter spp., Enterococcus faecalis, Enterococcus faecium, Pseudomonas aeruginosa, Staphylococcus epidermidis, Staphylococcus haemolyticus, and Stenotrophomonas maltophilia, were designed and evaluated. The data indicated that meropenem monotherapy administered at a high dose of 2 g q 8 h and as an OTAT achieved a PTA of ≥90% for isolates with an MIC of up to 128 mg/L and a CFR of ≥90% for all of the targeted pathogen populations when 50% fT > MIC (50% of the dosing interval during which free drug concentrations remain above the MIC) is chosen as the PD target, with Enterococcus faecalis being the sole exception. Even though 50% fT > 5 × MIC is chosen as the PD target, the aforementioned dosage regimen still reached a PTA of ≥90% for isolates with an MIC of up to 32 mg/L and a CFR of ≥90% for Acinetobacter spp., Pseudomonas aeruginosa, and Klebsiella pneumoniae populations. In conclusion, meropenem monotherapy displays potential competency for infections due to such highly resistant bacterial isolates provided that it is administered as a reasonable OTAT but not as the currently widely recommended TSPI.

Keywords: meropenem, monotherapy, meropenem-nonsusceptible bacteria, meropenem-resistant bacteria, pharmacokinetic/pharmacodynamics, Monte Carlo simulation

## INTRODUCTION

The increasing emergence of meropenem-nonsusceptible bacterial strains (MNBSs), such as Enterobacteriaceae, Pseudomonas aeruginosa, and Acinetobacter spp., which are defined as any isolate displaying minimum inhibitory concentration (MIC) of 2, 4, and 4 mg/L with meropenem, imipenem, and/or doripenem, respectively [Clinical and Laboratory Standards Institute (CLSI), 2019], has become a serious global health concern (Nordmann, 2014; Iovleva and Doi, 2017; Bulens et al., 2018; Moghnieh et al., 2018; Huang et al., 2019). The resulting infections, which have increasingly been identified not only in hospitals (Snitkin et al., 2012; Onori et al., 2015) but also in the community (Kelly et al., 2017; Salomão et al., 2017), result in excessive morbidity, mortality, and costs (Lemos et al., 2014; Bartsch et al., 2017) and severely limit treatment options.

Some traditional agents, such as polymyxins, tigecycline, fosfomycin, and aminoglycosides, etc. and currently relatively novel ones, such as ceftazidime-avibactam, meropenemvaborbactam, and imipenem/cilastatin-relebactam, etc. in monotherapy exhibit good potency for MNBSs; however, these agents are unfortunately limited by either significant shortcomings for the traditional agents (e.g., nephrotoxicity for polymyxins and aminoglycosides, increased mortality for tigecycline, and insufficient blood concentration due to the oral dosage form and dosage for fosfomycin) or geographical availability restrictions or unlisting for the novel ones (Satlin and Walsh, 2017; Karaiskos et al., 2019). Likewise, meropenemcontaining combination therapy (MCCT) with synergism, which is currently being widely studied and recommended for MNBSs, mostly for meropenem-nonsusceptible Klebsiella pneumoniae, is also limited for the following reasons: (i) it has not been satisfactorily investigated in large-scale randomized clinical trials, despite the existence of sporadic controlled trials and in vitro studies on this form of therapy (Liu et al., 2016; Oliva et al., 2017; Satlin and Walsh, 2017; Paul et al., 2018), and (ii) the extrapolation of MCCT based on meropenem-nonsusceptible K. pneumoniae to other MNBSs with different resistance mechanisms lacks further investigation and verification; and importantly, (iii) its efficacy in certain situations remains controversial relative to monotherapy since some of them do not truly improve the clinical outcomes (Del Bono et al., 2017; Paul et al., 2018). Consequently, the choice of MCCT or monotherapy for the treatment of infections due to different MNBSs remains a matter of debate, and the optimal treatment for MNBSs remains uncertain.

Meropenem, at recommended optimal dosing conditions, is often included in treatment regimens for infections with a MNBS. Indeed, both theoretical and clinical studies have shown that an optimal regimen using a high dose of 2 g every 8 h (q 8 h) and a 3-h prolonged infusion for meropenem as monotherapy improves its efficacy against MNBSs with MICs of 4–8 mg/L (Jaruratanasirikul et al., 2015; Tumbarello et al., 2015). However, the majority of these isolates often have meropenem MICs ≥16 mg/L (Tumbarello et al., 2015; Gomez-Simmonds et al., 2016; Cojutti et al., 2018), limiting the utility of this approach. However, to the best of our knowledge, this outcome occurs because traditional simple prolonged-infusion (TSPI) leads to both a decrease in the peak concentration and a delay in the peak time, resulting in an apparent incompetency for meropenem against the isolates with MICs ≥16 mg/L. For this reason, reoptimization of the method for administering meropenem in monotherapy may continue to show promise regarding the successful management of these problematic isolates, prompting us to design a theoretically optimized two-step-administration therapy (OTAT, see OTAT design in the part of Materials and Methods) for these problematic isolates. It is believed that this reoptimization will not only be crucial for maximizing microbiological outcomes but also be particularly important when better treatment options for these pathogens strains are absent.

Given that few attempts have been made to determine the optimal treatment for meropenem as a monotherapy against the increasing number of MNBSs with MICs ≥ 16 mg/L, this study focused on only these strains. An attempt was made to design OTAT regimens for meropenem to determine whether its use as a monotherapy can achieve an acceptable pharmacokinetic (PK)/pharmacodynamic (PD) exposure to illustrate whether meropenem monotherapy is incompetent for use with these

**Abbreviations:** MNBSs, meropenem-nonsusceptible bacterial strains; MIC, minimum inhibitory concentration; MCCT, meropenem-containing combination therapy; TSPI, traditional simple prolonged-infusion; OTAT, optimized two-stepadministration therapy; PK, pharmacokinetic; PD, pharmacodynamic; EUCAST, European Committee on Antimicrobial Susceptibility Testing; PTAs, probabilities of target attainment; CFRs, cumulative fractions of response; IVB, i.v. bolus; AUC, area under the concentration-time curve; ELF, epithelial lining fluid; f T > MIC, the time that the unbound (free) drug concentrations remain above the minimum inhibitory concentration; %f T > MIC, percentages of the dosing interval during which unbound (free) drug concentrations remain above the MIC; LRTIs, lower respiratory tract infections; f(Cmin)/MIC, ratio of the trough concentration of unbound (free) drug to the minimum inhibitory concentration; CLcr, creatinine clearance; KPC, Klebsiella pneumoniae carbapenemase.

strains, with the intent of defining optimal dosage regimens for meropenem against such strains if possible.

## MATERIALS AND METHODS

## Study Design

Meropenem-specific serum PK parameters were obtained from published literature and microbiological susceptibility data for the targeted pathogens were obtained from the European Committee on Antimicrobial Susceptibility Testing (EUCAST); together with dosing parameters, such as the dose and infusion time, these data were incorporated into a PK/PD model. Monte Carlo simulation was used to calculate the probabilities of target attainment (PTAs) at different MICs and the cumulative fractions of response (CFRs) for the targeted bacteria population with a pooled MIC distribution provided by each dosage regimen against 8 targeted bacterial species with doubling MICs between 16 and 512 mg/L from the EUCAST database for a given PK/PD target. A PTA or CFR of ≥90% and the causal dosage regimens were considered optimal.

## OTAT Design

Considering that the i.v. bolus (IVB) administration mode can rapidly and maximally achieve the loading drug concentration for a given agent and that prolonged infusion can continuously maintain the efficacious drug exposure, using these two techniques in combination is speculated to be optimal and competent for coverage of highly resistant bacterial isolates for a given pathogen. Therefore, OTAT represents such an administration mode in this study in which a loading dose of the total amount of the tested drug is first administered via a rapid IVB, immediately followed by the remainder of the experiment via prolonged infusion after the first dose, as illustrated in **Figure 1**.

## Meropenem Dosage Regimens

Meropenem dosage regimens, including the injection method, were chosen based on the licensed and studied regimens used for infected adults with normal renal function, including 0.5 g every 8 h (q 8 h), 1 g q 8 h, and 2 g q 8 h. In addition, 0.75 g every 6 h (q 6 h) and 1.5 g q 6 h were also investigated as the modified dosage regimens of 1 g q 8 h and 2 g q 8 h, respectively. Since meropenem in solutions is stable for only ∼6 h at room temperature (Kuti et al., 2004), the remainder of the experiment can only be administered by prolonged infusion up to 6 h, and 4 to 6-h prolonged infusion for the remainder were thus simulated. In this study, the following 39 dosage regimens presented in **Table 1** were investigated.

## Meropenem Tissue PK Profiles and PD Model Associated With Clinical Response

Meropenem tissue penetration to the infection site is critical for obtaining a good clinical outcome in patients with different infection sites. Previous studies indicated that the mean meropenem tissue penetration (i.e., the mean tissue/concomitant serum concentration) after injection ranged from ∼0.2 to 1 for different tissues as follows: ∼0.5 for liver, lung, skin, uterus, ovaries, rectum, prostate, thyroid, trachea, and lymph nodes with the exception of a very small concentration in the brain and cerebrospinal fluid (Harrison et al., 1989), 0.95 for peritoneum (Hextall et al., 1991), 0.2 for bronchial secretions (Bergogne-Bérézin et al., 1994), 1.1 for skin exudate (Wise et al., 1990), and 0.85-0.87 for blister fluid (Mouton and Michel, 1991). However, Byl et al. (1999) reported it to be 0.17–0.43 for lung, 0.20– 0.55 for bronchial mucosa, and 0.18–0.26 for pleural tissues. Regarding meropenem tissue penetration, especially for lung, a recent study conducted by Lodise et al. (2011) reported that the ratio of the area under the concentration-time curve (AUC) in epithelial lining fluid (ELF) to the AUC in plasma

FIGURE 1 | Concentration-time profiles via TSPI and OTAT. IVB, i.v. bolus; PI, prolonged infusion; OTAT, optimized two-step-administration therapy; TSPI, traditional simple prolonged-infusion; MIC, minimum inhibitory concentration; f, fraction of unbound drug; fT > n × MIC, the time that the unbound (free) drug concentrations remain above the MIC by n-fold.



(AUCELF/AUCplasma ratio) for meropenem varied substantially between patients with ventilator-associated pneumonia, with the 10th and 90th percentile ELF exposures that were 3.7–178% of the plasma AUC values and with the mean and median AUC exposures that were 81.6 and 25.42% of the plasma values. Overall, these observations imply that logically, mean 1– 2 times meropenem plasma concentrations or exposures would achieve the desired tissue concentration or exposure for most tissues if they increase proportionally, with the exception of ∼5 times for lung, bronchial and pleural tissues. Theoretically, it would be more accurate to predict and establish a drug regimen based on the relationship between the tissue drug concentration and tissue exposure targets. However, to the best of our knowledge, penetration of meropenem in infected tissues to achieve the exposure targets has not been studied often, and drug concentrations in extracellular compartments are difficult to determine; thus, correlations between the PK/PD index in the tissue and antimicrobial effects are less well-understood (Nightingale and Mur, 2007). Therefore, serum PK parameters based on the plasma drug concentrations are most commonly used as surrogates for establishing and estimating the PK/PD indices in some studies (Kuti et al., 2003; Ikawa et al., 2011), and so it is with the present study.

Regarding the correlations between the PK/PD index and clinical response to meropenem therapy, Roberts et al. (2014) reported that 50 and 100% of the ratios of f T > MIC to a dosing interval are independent factors that influence the clinical outcome of patients receiving meropenem or other β-lactams and that a higher PK/PD index is associated with a higher likelihood of a positive clinical outcome. Likewise, Zhou et al. (2011) also found that f T > MIC is an independent influencing factor for predicting clinical success and that the cutoff value using f T > MIC based on serum concentrations in elderly patients with lower respiratory tract infections (LRTIs) is 76%. However, Li et al. (2007) studied the indices of clinical PD for LRTIs through meropenem serum concentrations in 101 patients and considered that the minimum concentration of drug in serum f(Cmin)/MIC > 5 rather than %f T > MIC is the only significant predictor of clinical response since 100% of f T > MIC is achieved in the majority of LRTI patients. However, there is no consensus regarding which strategy (%f T > MIC vs. f(Cmin)/MIC > 5) is better. In the present study, %f T > MIC is therefore used as the PD target associated with the clinical response of meropenem therapy.

Generally, 40–50% of f T > MIC for meropenem based on the serum concentration is usually used for predicting clinical and microbiological outcomes and for optimizing dosage regimens in most current meropenem PK/PD studies (Burgess et al., 2007; Watanabe et al., 2007; Ikawa et al., 2011; Kondo et al., 2014). However, given the profiles of meropenem tissue penetration described previously, we consider that (i) this exposure target in plasma may be underestimated when 40–50% of f T > MIC for meropenem is required in infected tissues; and (ii) it is reasonably speculated that a meropenem drug concentration of 1–2 × MIC in plasma for most types of infection but 5 × MIC in plasma for pulmonary, bronchial and pleural infection is sufficient to achieve the pathogen MIC without considering the influence of inflammation on meropenem tissue penetration. Based on all the abovementioned considerations, especially the profiles of meropenem tissue penetration, the targets of 50% f T > MIC (mainly for bacterial peritonitis or intraabdominal infections, bloodstream infections, skin and soft tissue infections, or urinary tract infections), 50% f T > 2 × MIC (mainly for bacterial hepatitis, metritis, oophoritis, proctitis, or prostatitis, etc.), and 50% f T > 5 × MIC (mainly for LRTIs, such as pneumonia, bronchitis, or pleural infections) based on the serum concentration were applied as the optimal PK/PD index in terms of obtaining adequate meropenem exposures regarding its bacterial killing and clinical efficacy for various types of infection or infected sites in the present study.

The %f T > MIC in OTAT was calculated using the following one-compartment intravenous infusion equation, as modified from a previously reported equation (Li et al., 2006):

• Meropenem was administered via TSPI;

$$f\text{T} > \text{MIC} = T\_{\text{inf}} - \frac{V\_d}{CL} \times \ln\left(\frac{R\_0/CL}{R\_0/CL - \text{MIC}}\right) + \frac{V\_d}{CL} \times 1$$

$$\ln\left(\frac{R\_0/CL - R\_0/CL \times e^{-CLT\_{\text{inf}}/V\_d}}{MIC}\right)$$

Then

$$f\text{T} > n \times \text{MIC} = T\_{\text{inf}} - \frac{V\_d}{\text{CL}} \times \text{Ln} \left(\frac{R\_0 / \text{CL}}{R\_0 / \text{CL} - n \cdot \text{MIC}}\right) + 1$$

$$\frac{V\_d}{\text{CL}} \times \text{Ln} \left(\frac{R\_0 / \text{CL} - R\_0 / \text{CL} \times e^{-\text{CL}T\_{\text{inf}} / V\_d}}{n \cdot \text{MIC}}\right)$$

• Meropenem was administered via OTAT, and prolonged infusion was started immediately following the completion of IVB (see **Figure 1**);

$$\begin{aligned} f\text{T} &> n \times \text{MIC} = T\_{\text{inf}} + \frac{V\_d}{\text{CL}} \times \\ \ln\left(\frac{f \cdot \text{Dose}\_{\text{bol}} \cdot \text{CL} /\_{V\_d} + f \cdot \text{Dose}\_{\text{inf}} \cdot \left(1 - e^{-\text{CL} \cdot T\_{\text{inf}} / V\_d}\right) /\_{T\_{\text{inf}}}}{\text{CL} \cdot n \cdot \text{MIC}}\right) \end{aligned}$$

and %f T > n × MIC = f T > n × MIC × 100/DI.

where f is the fraction of unbound drug, f T is the time that the drug is in unbound (free) form, MIC is the minimum inhibitory concentration, f T > MIC is the time that the unbound (free) drug concentrations remain above MIC, f T > n × MIC is the time that the unbound (free) drug concentrations remain above the MIC by n-fold, n is an integer set as 1, 2, or 5 in the present study, %f T > n × MIC is the percentage of the dosing interval during which unbound (free) drug concentrations remain above the MIC by n-fold, Tinf (h) is the infusion time, R<sup>0</sup> (mg/h) is the zero-order infusion rate calculated as whole-dose × f/Tinf in TSPI, Dosebol (mg) is the dose administered via IVB in OTAT, Doseinf (mg) is the dose administered via prolonged infusion in OTAT, CL (L/h) is the plasma clearance rate of the experiment, V<sup>d</sup> (L) is the volume of distribution of the experiment at steady state, e is the exponent, Ln is the natural logarithm, and DI (h) is the dosing interval.

## Meropenem Population PK Parameters

Serum population PK parameters for meropenem were obtained from previously published studies documenting adult patients (preferably those describing infection studies when available) with normal renal function (i.e., creatinine clearance (CLcr) ≥ 50 ml/min) or healthy volunteers (when the desired data from infected populations were unavailable). Meropenem population PK parameters were determined by the PK model established by Li et al. (2006) as follows: CL(L/h) = 14.6 × (CLcr/83)0.62 × (AGE/35)(−0.34) , Vc(L) = 10.8 × (WT/70)0.99, and Vp(L) = 12.6 where CLcr (ml/min) is the creatinine clearance of the patient calculated according to the Cockcroft-Gault equation based on the patient's ideal body weight (Cockcroft and Gault, 1976), WT (kg) is the ideal body weight of the patient, V<sup>c</sup> (L) is the central volume of distribution, and V<sup>p</sup> (L) is the peripheral volume of distribution. The data in this study were chosen for our analysis because compared with other studies (Gonçalves-Pereira and Póvoa, 2011), this study had a relatively large number of patients (N = 79), and all of the patients had various types of infections, including 52 patients with intra-abdominal infections, 21 patients with ventilator-associated pneumonia, and six patients with community-acquired pneumonia, and were treated with meropenem; therefore, the PK data obtained for meropenem are relatively representative. The meropenem population PK parameter estimates at steady state based on the demographic characteristics of the subjects in this study are summarized as follows: CL 14.97 ± 4.13 L/h and V<sup>d</sup> (V<sup>c</sup> + Vp) 23.86 ± 2.46 L. The ranges of the unbound fraction (f) were calculated from the protein binding data, and estimates of f for meropenem (0.85–0.98) were obtained from the package insert of the product (USP PACKAGE INSERT., 1996) and from the pharmacokinetic studies (Kuti et al., 2005), if measured.

## Microbiological Susceptibility Data

The microbiological susceptibility data, including the targeted bacterial species, the number of isolates with meropenem MICs ≥ 16 mg/L and the corresponding MIC frequency distributions (in **Table 2**), were derived from the EUCAST database [European Committee on Antimicrobial Susceptibility Testing (EUCAST), 2019]. This database was chosen for our analysis because it provided the most current and comprehensive collection of MIC data for the antibiotics and organisms modeled in the current study. Bacterial species with isolates having meropenem MICs ≥ 16 mg/L (≥100 strains) include mainly Acinetobacter baumannii (763 strains), Acinetobacter spp. (1,209 strains), Enterococcus faecalis (1,317 strains), Enterococcus faecium (1,554 strains), P. aeruginosa (4,841 strains), Staphylococcus epidermidis (175 strains), Staphylococcus haemolyticus (103 strains), and Stenotrophomonas maltophilia (3,935 strains). These bacterial species will therefore be modeled as targets for investigating meropenem exposures in monotherapy against MNBSs with MICs ≥ 16 mg/L.

## Monte Carlo Simulation

A 5,000-subject simulation was performed by Crystal Ball software (version 7.2.2; Decisioneering, Inc., Denver, CO, USA) to calculate the probability of achieving the requisite PK/PD exposure (i.e., 50% f T > MIC, 50% f T > 2 × MIC, and 50% f T > 5 × MIC) for each dosage regimen, referred to as the PTA against the isolates at a specific MIC and the CFR against a population of an organism with a pooled MIC distribution. Prior to the simulations, PK parameters were


TABLE 2 | No. of targeted bacterial isolates and corresponding MIC frequency distributions at MICs ≥16 mg/L collected from the EUCAST database.

AB, Acinetobacter baumannii; AS, Acinetobacter spp.; EFS, Enterococcus faecalis; EFM, Enterococcus faecium; PA, Pseudomonas aeruginosa; SE, Staphylococcus epidermidis; SHA, Staphylococcus haemolyticus; SM, Stenotrophomonas maltophilia.

assumed to follow log-normal distributions and the fraction unbound f followed a uniform distribution, whereby the probability was equal within the specified range. The PTA was determined by calculating the fraction of subjects who attained the target at a specific MIC and was determined for MICs between 16 and 512 mg/L. A regimen with a PTA of ≥90% against the isolates at this MIC was considered optimal. The overall expectation value for the PTA (i.e., CFR) is related to PD target attainment in that it expresses the probability of a given dosage regimen achieving the desired exposures against an entire population of pathogens. The CFR percentages for each organism were calculated by multiplying the PTA at each MIC by the percentage of isolates of each of the modeled organisms actually found at that MIC. A regimen with a CFR of ≥90% against an organism population of was considered optimal.

## RESULTS

## Probability of Target Attainment

PTA vs. MIC profiles for simulations of different dosage regimens are presented in **Figure 2**. A PTA of ≥90% was considered satisfactory. For attainment of the classical PD target, i.e., 50% f T > MIC, the dosage regimens of 0.5 g q 8 h, 1 g q 8 h, 2 g q 8 h, 0.75 g q 6 h, and 1.5 g q 6 h would be insufficient for the treatment of bacteria with MICs ≥ 16 mg/L if administered via 4- to 6-h TSPI, regardless of the MIC values and PD targets. However, they would be adequate if administered as a reasonable OTAT. Specifically, the dosage regimen of 0.5 g [e.g., 0.25 g (5 min IVB) + 0.25 g (5–6 h)] q 8 h for the isolates with MICs of 16 mg/L, 1 g [e.g., 0.5 g (5-min IVB) + 0.5 g (5 h)] q 8 h for the isolates with MICs of 32 mg/L, 0.75 g [e.g., 0.5 g (5 min IVB) + 0.25 g (6 h)] q 6 h for the isolates with MICs of 64 mg/L, 2 g [e.g., 1.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h and 1.5 g [e.g., 1 g (5-min IVB) + 0.5 g (6 h)] q 6 h for the isolates with MICs of 128 mg/L produced sufficient PK/PD exposures when 50% f T > MIC was used as the PD target. However, the power of these dosage regimens was weakened by nearly half when 50% f T > 2 × MIC was chosen as the PD target and substantially reduced when 50% f T > 5 × MIC was chosen as the PD target. Interestingly, the dosage regimen of 2 g [preferred 1.5 g (5-min IVB) + 0.5 g (5 h)] q 8 h still showed good antibacterial properties for the isolates with MICs of up to 32 mg/L even though the highest PD target of 50% f T > 5 × MIC was used. Of note, as a modification of the dosage regimen of 2 g q 8 h, the dosage regimen of 1.5 g q 6 h produced a PTA of ≥90% for the isolates with MICs of 16 mg/L when 50% f T > 5 × MIC was used as the PD target when it was administered as an OTAT [e.g., 1.5 g [1 g (5-min IVB) + 0.5 g (5 h)] q 6 h], suggesting a potentially useful dosage regimen for these isolates. **Table 3** summarizes the coverage of various dosage regimens for the pathogen isolates with meropenem MICs ≥ 16 mg/L in different types of infection at the condition of achieving ≥90% PTA.

## Cumulative Fraction of Response

CFR vs. various targeted pathogen populations for simulations of different dosage regimens are displayed in **Figure 3**. A CFR of ≥90% was considered optimal. Obviously, only regimens with OTAT achieved a CFR of ≥90% for the targeted pathogen population, regardless of the PD targets and dosing models. Based on currently pooled MIC distributions, when 50% f T > MIC was chosen as the PD target, the dosage regimen of 0.5 g [preferred 0.25 g (5-min IVB) + 0.25 g (6 h)] q 8 h yielded a CFR of ≥90% for only the Acinetobacter spp., S. epidermidis, and S. maltophilia populations; however, the dosage regimens of 1 g [preferred 0.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h, 2 g [e.g., 0.5 g (5-min IVB) + 1.5 g (6 h)] q 8 h, 0.75 g [preferred 0.5 g (5-min IVB) + 0.25 g (6 h)] q 6 h, and 1.5 g [e.g., 0.5 g (5 min IVB) + 1 g (5 h)] q 6 h for all of the targeted pathogen populations achieved the requisite CFR, with E. faecalis being the sole exception. When using a higher PD target of 50% f T > 2 × MIC, the majority of the simulated dosage regimens had decreased coverage of the targeted pathogens population at the condition of achieving ≥90% CFR. However, the dosage regimens of 2 g [preferred 1.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h and 1.5 g [preferred 0.5 g (5-min IVB) + 1 g (6 h) or 1 g (5 min IVB) + 0.5 g (6 h)] q 6 h still covered the A. baumannii, Acinetobacter spp., E. faecium, P. aeruginosa, S. epidermidis, S. haemolyticus, and S. maltophilia populations at this condition.

TABLE 3 | Summary of the coverage of various dosage regimens for the pathogen isolates with MICs of ≥16 mg/L at the condition of achieving ≥90% PTA and/or the targeted pathogen population with pooled MIC distributions between 16 and 512 mg/L at the condition of achieving ≥90% CFR in different types of infection.


AB, Acinetobacter baumannii; AS, Acinetobacter spp.; EFS, Enterococcus faecalis; EFM, Enterococcus faecium; PA, Pseudomonas aeruginosa; SE, Staphylococcus epidermidis; SHA, Staphylococcus haemolyticus; SM, Stenotrophomonas maltophilia.

NA, not applicable.

P16, P32, P64, P<sup>128</sup> signify the pathogen isolates with MICs of 16, 32, 64, and 128 mg/L, respectively.

Regimens with P<sup>x</sup> (x = 16, 32, 64 or 128) + (y) (y = AB, AS, EFM, PA, SE, SHA, SM, or a combination) signify that they are competent for the treatment of infections caused by the pathogen isolates actually found at that MIC if the exact MIC values are available and/or for the treatment of infections caused by the targeted pathogen population identified with only bacterial species if the exact MIC values are unavailable.

FIGURE 3 | CFRs of achieving 50% fT > MIC, 50% fT > 2×MIC, and 50% fT > 5 × MIC for meropenem with various dosage regimens simulated for the targeted bacteria populations with pooled MIC distributions between 16 and 512 mg/L. AB, Acinetobacter baumannii; AS, Acinetobacter spp.; EFS, Enterococcus faecalis; EFM, Enterococcus faecium; PA, Pseudomonas aeruginosa; SE, Staphylococcus epidermidis; SHA, Staphylococcus haemolyticus; SM, Stenotrophomonas maltophilia; 50% fT > MIC, 50% of the dosing interval during which free drug concentrations remain above the MIC; 50% fT > 2 × MIC, 50% of the dosing interval during which free drug concentrations remain above the MIC by two-fold; 50% fT > 5 × MIC, 50% of the dosing interval during which free drug concentrations remain above the MIC by five-fold.

However, when using an aggressive PD target of 50% f T > 5 × MIC, these simulated dosage regimens had a drastically decreased coverages of the targeted pathogen populations at the condition of achieving ≥90% CFR. Surprisingly, however, the dosage regimen of 2 g [1.5 g (5-min IVB) + 0.5 g (5 h)] q 8 h still reached ≥90% CFR for the Acinetobacter spp., P. aeruginosa, S. epidermidis, S. haemolyticus, and S. maltophilia populations. **Table 3** summarizes the coverage of various dosage regimens for the targeted bacterial population with pooled MIC distributions between 16 and 512 mg/L in different types of infection at the condition of achieving ≥90% CFR.

## DISCUSSION

To the best of our knowledge, this study is the first to analyze meropenem as a monotherapy aimed at MNBSs with MICs ≥ 16 mg/L. In this study, we considered the tissue penetration profiles of meropenem for infections at different sites when establishing the PK/PD model associated with clinical response. We believe that this study is worthwhile because the PK/PD outcomes generated from our data could help clinicians treat these isolates more effectively through optimization of dosage regimens, especially when better treatment options are absent. Notably, our results support that meropenem as a monotherapy is still competent for isolates with MICs ≥ 16 mg/L provided that the drug is administered as a reasonable OTAT but not as the currently widely recommended TSPI.

## TSPI vs. OTAT for Meropenem Against Highly Resistant Bacterial Isolates

Currently, the optimal meropenem dosage is undergoing ardent evaluation to develop new strategies to overcome increasing meropenem resistance and maximally preserve the effectiveness of this drug. TSPI for meropenem has often been the preferred optimal mode. Indeed, TSPI for meropenem against central nervous system infections due to P. aeruginosa (Capitano et al., 2004) or S. marcescens (Nicasio et al., 2007) and ventilator-associated pneumonia due to gram-negative bacilli and for treating neutropenic patients with fever have resulted in successful clinical responses when compared with those resulting from intermittent infusion (Lorente et al., 2006; Fehér et al., 2014). However, the benefits of TSPI for meropenem reported by these studies were found using meropenem-susceptible bacterial strains. However, for infections due to MNBSs, especially those with MICs ≥ 16 mg/L, the benefit of this approach for improving clinical efficacy is unknown, and clinical data are limited.

Indeed, this approach is unfavorable for highly resistant bacterial isolates because it reduces the initial bactericidal effects due to both the decrease in meropenem peak concentration and the delay in its peak time, as demonstrated by Eguchi et al. (2010). It is therefore difficult for meropenem to achieve the MIC of highly resistant bacterial isolates, even at a high dose. This reduction in initial bactericidal effects may also be the reason why some studies found that meropenem, even at 3 g/day (e.g., 1 g q 8 h over a 3-h infusion or 0.5 g every 4 h (q 4 h) over a 4 h infusion), displayed poor PK/PD exposures for isolates with MICs ≥ 16 mg/L but showed a relatively good effect for isolates with MICs ≤ 8 mg/L (Vourli et al., 2016; Zhao et al., 2017), thus inferring that TSPI for meropenem may be beneficial for isolates with low MICs. Likewise, our data confirmed this inference because using TSPI for meropenem, even at a high dose of 2 g q 8 h and using 50% f T>MIC as the PD target, did not yield a PTA ≥ 90% for isolates with MICs of 16 mg/L. Thus, this conventional approach of using TSPI to optimize PD exposure is rendered futile with the emergence of higher MICs (Avery and Nicolau, 2018). For the data obtained herein, OTAT is preferable because it provides a higher meropenem exposure relative to that of TSPI regardless of the MIC values, dosing models and PD targets. Additionally, meropenem administered via a loading dose of 0.5 g over a 30-min infusion followed immediately by 0.5 g q 4 h over a 4-h infusion [i.e., an OTAT regimen of 0.5 g (0.5 h) + 0.5 g (4 h)] reportedly achieved better outcomes than that administered intermittently against bacteria of intermediate susceptibility (Zhao et al., 2017), thus demonstrating the superiority of OTAT for meropenem against MNBSs.

## Competence of Meropenem in Monotherapy Against Highly Resistant Bacterial Isolates

Currently, clinical experience with meropenem monotherapy for MNBSs with MICs ≥ 16 mg/L is indeed limited because from a clinical point of view, determination of meropenem MICs ≥ 8 mg/L for the identified stains by susceptibility tests causes the vast majority of clinicians to switch to other better options. However, previous studies indicated that meropenem using a high dose and TSPI (e.g., 2 g q 8 h over a 2- to 3-h infusion) in monotherapy can provide some therapeutic benefit and therefore be considered to treat infections due to MNBSs with MICs ≤ 4 or even ≤8 mg/L based on the therapeutic efficacy of K. pneumoniae carbapenemase (KPC)-producing K. pneumoniae (Daikos and Markogiannakis, 2011; Tzouvelekis et al., 2012; Hsu and Tamma, 2014; Tumbarello et al., 2015). Inconsistent with these findings, on a theoretical basis, our data indicated that even for the isolated K. pneumoniae strains with meropenem MICs of up to 32 mg/L and utilization of an aggressive PD target of 50% f T > 5 × MIC, meropenem in monotherapy, even at the same daily dose used in the abovementioned studies, can still produce desired PK/PD exposures provided that it is administered using the dosage regimen of 2 g [1.5 g (5-min IVB) + 0.5 g (5 h)] q 8 h, as shown in **Figure 2**.

Regarding the treatment of infections due to MNBSs with MICs ≥ 16 mg/L, most reports currently focus on MCCT for meropenem-nonsusceptible K. pneumoniae. Joint guidelines prepared by the Working Party of the British Society for Antimicrobial Chemotherapy, the Healthcare Infection Society and the British Infection Association (Hawkey et al., 2018) considered that MCCT including a high dose and continuous infusion of meropenem would be appropriate for K. pneumoniae with MICs > 8 and <64 mg/L. In addition, some studies have indicated that MCCT may grant a survival benefit relative to that of meropenem in monotherapy when the MIC of K. pneumoniae is <16 mg/L (Tumbarello et al., 2012; Daikos et al., 2014). Moreover, even for strains with MICs ≥ 16 mg/L, MCCT including a high dose of 13.2 g/day and continuous infusion of meropenem with levels optimized by therapeutic drug monitoring were helpful in obtaining a favorable clinical outcome for infections due to KPC-producing K. pneumoniae with MICs of 16–64 mg/L (Pea et al., 2017). Another study also demonstrated the benefits of MCCT using a high dose of 6 g daily and 3-h extended infusion for meropenem for an infection due to meropenem-nonsusceptible K. pneumoniae strains with MICs ≥16 mg/L (Giannella et al., 2018).

Understandably, these reports imply that MCCT is preferable for treating meropenem-nonsusceptible K. pneumoniae infections arising from strains with MICs > 8 or even ≥16 mg/L. Likewise, some retrospective, prospective observational cohort and multicentric studies have confirmed the superiority of MCCT over monotherapy for the treatment of infections due to KPC-producing K. pneumoniae, especially for bloodstream infections (Zarkotou et al., 2011; Qureshi et al., 2012; Tumbarello et al., 2012; Daikos et al., 2014). In contrast, another most recently retrospective study (Kuti et al., 2019), in which PD exposures of meropenem as an adjunctive treatment in plazomicin-based combination therapy (i.e., plazomicin plus meropenem) were evaluated to investigate its synergy in combination against meropenem-nonsusceptible K. pneumoniae with MICs ≥ 64 mg/L, showed PD exposures of 0% for meropenem at the target of ≥40% f T > MIC when meropenem was administered as 2 g q 8 h over a 3-h infusion and therefore concluded that plazomicin monotherapy was sufficient and optimization of meropenem therapy was not required for the combination to achieve microbiological response and clinical efficacy against serious meropenem-nonsusceptible K. pneumoniae infections, including bloodstream infections, hospital acquired pneumonia or ventilator-associated pneumonia.

It should also be noted that because most conclusions on the superiority of MCCT reached from the abovementioned reports were derived based on meropenem-nonsusceptible K. pneumoniae, the extrapolation of MCCT using a high dose and prolonged infusion for meropenem to other MNBSs with different resistance mechanisms requires further investigation. Coincidentally, the recent Amsterdam Investigator-Initiated Absorb Strategy All-Comers trial (Paul et al., 2018), in which a randomized, controlled, superiority trial was conducted at six hospitals to investigate the superiority of MCCT (i.e., colistin plus meropenem) vs. colistin alone, showed no differences in the outcomes of patients treated with MCCT or colistin monotherapy for infections (including bloodstream infections, ventilator-associated pneumonia and/or hospital acquired pneumonia, and urinary tract infections) due to carbapenem-resistant Acinetobacter spp., Enterobacteriaceae, and P. aeruginosa in which 97% of the isolates had meropenem MICs > 8 mg/L. This result supported that colistin monotherapy is equipotent to MCCT for such infections, and it is therefore unnecessary to use MCCT to treat infections due to these MNBSs. Overall, the choice of MCCT or monotherapy for the treatment of infections due to MNBSs remains a matter of debate. Also, it is important to note that co-administration of different antibiotics may lead to important concomitant adverse effects, including Clostridium difficile infection, selection of further resistances, or nephrotoxicity (Petrosillo et al., 2013).

Inconsistent with these reports, our data supported that even meropenem monotherapy, rather than plazomicin monotherapy, colistin monotherapy, or MCCT, is sufficient for the treatment of such infections due to these pathogens (i.e., the abovementioned Acinetobacter spp., P. aeruginosa, and Enterobacteriaceae, including meropenem-nonsusceptible K. pneumoniae) and for situations in which meropenem exposures are not reduced relative to plasma exposures, such as bloodstream infections, skin and soft tissue infections, and urinary tract infections. Meropenem monotherapy, in which it must be administered at a high dose of 2 g q 8 h or 1.5 g q 6 h and as an OTAT, can achieve optimal PK/PD exposures for MNBSs with MICs of up to 128 mg/L and for pathogen populations with a pooled MIC distribution, as demonstrated by achieving a PTA of ≥90% at an MIC of 128 mg/L and a CFR of ≥90% for Acinetobacter spp., P. aeruginosa, and Enterobacteriaceae populations when using the dosage regimen of 2 g [1.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h or 1.5 g [1 g (5-min IVB) + 0.5 g (6 h)] q 6 h and using 50% f T > MIC as the PD target for such types of infection.

However, meropenem monotherapy did not display acceptable PK/PD exposures for the isolates with MICs ≥ 256 mg/L in the present study regardless of the dosing models and PD targets. Interestingly, another PK/PD study (Del Bono et al., 2017) in which meropenem using a high dose of 2 g q 8 h and 3-h TSPI in MCCT (i.e., meropenem plus tigecycline or gentamicin or colistin, or meropenem plus tigecycline plus gentamicin or colistin) was used to treat bloodstream infections due to KPC-producing K. pneumoniae with actual meropenem MICs ≥ 256 mg/L also showed that meropenem did not achieve the PD target of T > 40% 1 × MIC in these isolates based on the measured meropenem levels despite the MCCT used. Moreover, at this dosage condition, meropenem could have attained PTAs of only 68 and 32% at that PD target in isolates with hypothetical MICs of 16 and 32 mg/L, respectively, despite the MCCT used. Thus, no synergisms were concluded between meropenem and the co-administered agents in this study. This conclusion was confirmed by the study conducted by Kuti et al. (2019), in which meropenem administered as 2 g q 8 h over a 3-h infusion had no synergy on the co-administered plazomicin when used for the treatment of infections due to meropenem-nonsusceptible K. pneumoniae with meropenem MICs ≥64 mg/L. One reason why unsatisfactory meropenem exposures were observed in these studies may be that TSPI for meropenem reduced its AUC exposures above the MIC, especially for values >16 mg/L, due to both the decrease in its peak concentration and the delay in its peak time. Based on these observations and ours, when administered as a reasonable OTAT and used for bloodstream infections, meropenem monotherapy displayed equally discontented exposures against highly resistant bacterial isolates with MICs ≥ 256 mg/L but showed a superior effect against relatively lowly resistant bacterial isolates with MICs ≤ 128 mg/L when compared with that of MCCT including a high dose and TSPI of meropenem. This finding suggests a possible usefulness of meropenem monotherapy for the treatment of such infections caused by isolates with MICs of up to 128 mg/L.

High MICs often result in insufficient antimicrobial tissue exposures, especially when the drug is used in inappropriate dosage regimens and used for infected sites with poor drug penetration. Regarding this issue, our data provide very useful directions on whether meropenem at the optimal dosage regimens can be used as a monotherapy for the treatment of infections due to isolates with high MICs at different sites based on the exact MICs if available. Given the profiles of meropenem tissue penetration, for infections due to isolates with MICs ≥ 16 mg/L occurring in the liver, skin, uterus, ovaries, rectum, prostate, trachea, etc., meropenem monotherapy with at least a 3 g daily dose yields good outcomes, e.g., the dosage regimen of 1 g [preferred 0.5 g (5-min IVB) + 0.5 g (5–6 h)] q 8 h for isolates with MICs of 16 mg/L, the dosage regimen of 2 g [preferred 1.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h, 0.75 g [preferred 0.5 g (5-min IVB) + 0.25 g (6 h)] q 6 h or 1.5 g [preferred 1 g (5-min IVB) + 0.5 g (6 h)] q 6 h for isolates with MICs of up to 64 mg/L, and is thus sufficient. Satisfactorily, for infections occurring in the peritoneum, the dosage regimen of 0.5 g [preferred 0.25 g (5-min IVB) + 0.25 g (5–6 h)] q 8 h would be sufficient for isolates with MICs of 16 mg/L, and the dosage regimen of 1 g [preferred 0.5 g (5-min IVB) + 0.5 g (5–6 h)] q 8 h, 0.75 g [preferred 0.5 g (5-min IVB) + 0.25 g (6 h)] q 6 h, 2 g [preferred 1.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h or 1.5 g [preferred 1 g (5-min IVB) + 0.5 g (6 h)] q 6 h for isolates with MICs of up to 32, 64, and 128 mg/L would be adequate, respectively. Surprisingly, even for infections located in the lung, bronchus, and pleura, meropenem with the preferred regimen of 2 g [1.5 g (5-min IVB) + 0.5 g (5 h)] q 8 h still proved effective for isolates with MICs of up to 32 mg/L, suggesting that for LRTIs and pleural infections, meropenem monotherapy can still perform good bactericidal action on isolates with MICs of up to 32 mg/L.

However, the exact MIC values, especially those >16 mg/L, are often unavailable because automated systems such as VITEK-2, which indicate MIC values as high as >16 mg/L, are unable to determine the precise MIC. Regarding this problem, our data provided the CFR for the targeted bacterial population and summarized the treatment options (**Table 3**). As a potential monotherapy, meropenem with an aggressive dosage regimen of 2 g [preferred 1.5 g (5-min IVB) + 0.5 g (6 h)] q 8 h or 1.5 g [preferred 1 g (5-min IVB) + 0.5 g (6 h)] q 6 h may be the best choice and is worth trying in its empiric therapy, especially for critically ill patients, because even at a PD target of 50% f T > 2 × MIC, these dosage regimens could produce a CFR of ≥90% for all of the tested bacterial populations with MICs ≥16 mg/L, including A. baumannii, Acinetobacter spp., E. faecium, P. aeruginosa, S. epidermidis, S. haemolyticus, and S. maltophilia, with the sole exception of E. faecalis. This result suggests that in the empiric therapy of meropenem, these dosage regimens would be competent for the vast majority of infections (e.g., bloodstream infections, intraabdominal infections, skin, and soft tissue infections, and urinary tract infections) due to these pathogens based on the profiles of meropenem tissue penetration. However, for LRTIs and pleural infections, meropenem monotherapy with the most promising dosage regimen of 2 g [1.5 g (5-min IVB) + 0.5 g (5 h)] q 8 h would be sufficient for infections due to only Acinetobacter spp., P. aeruginosa, S. epidermidis, S. haemolyticus, and S. maltophilia. **Table 4** summarizes some preferred dosage regimens in the empiric therapy of meropenem for different types of infection based on our analysis.

Of note, these optimal dosage regimens sufficient for MNBSs are also adequate for meropenem-susceptible bacterial strains since the regimens established at higher MICs generate higher PTAs and CFRs at low MICs. In addition, we need to monitor renal function regularly and adjust the meropenem dose as required, particularly for the high dose regimens, as the majority of the patients infected with meropenemnonsusceptible K. pneumoniae are critically ill and have altered renal function (Daikos and Markogiannakis, 2011). Although meropenem monotherapy displays satisfactory PK/PD exposures against highly resistant bacterial isolates, especially when it is administered at a high dose of 2 g q 8 h in OTAT, the accompanying safety issues are also worthy of our attention, especially when a high dose is used. However, although the most frequent adverse events associated with meropenem use, such as diarrhea, rash, nausea, and vomiting, thrombocytosis, eosinophilia and changes in hepatic biochemistry and the possible episodes of seizures are reported, meropenem exhibits an acceptable safety profile with good central nervous system and gastrointestinal tolerability, even at a high dose of up to 6 g per day (2 g q 8 h), and shows also a favorable safety profile in a number of special patient populations, including elderly, renally impaired, pediatric, and neutropenic patients, patients with cystic fibrosis and those with meningitis (Norrby, 1995; Norrby et al., 1995; Norrby and Gildon, 1999; Linden, 2007). Therefore, the dosage regimens recommended herein for highly resistant bacterial strains should be safe and worthwhile to try.

TABLE 4 | Summary of preferred treatment option recommendations in the empiric therapy of meropenem for different types of infection based on our analysis.


## Study Limitations

The present study has some limitations. First, our data did not include meropenem concentrations in plasma or infected sites. Second, the results of simulation analysis were not validated by evaluating clinical outcomes, limiting the generalization of the conclusions. Third, the susceptibility data used in the modeling for our predictions of target attainment were obtained from the EUCAST database. As such, our findings should be interpreted and extrapolated while paying attention to the different or changing susceptibility profiles in one's own hospital. Fourth, a one-compartment model was used to calculate the meropenem PD exposures, whereas other studies have suggested that its in vivo pharmacokinetic disposition is best fitted using a twocompartment model. However, studies on its PK have been published using both one- and two-compartment models as well as non-compartmental analysis (Christensson et al., 1992; Leroy et al., 1992a,b).

Despite these limitations, our data are believable and instructive for prescribers because it considered the profiles of meropenem tissue penetration, used the more representative PK parameters, and integrated the most current and comprehensive MIC data. Importantly, even with f(Cmin)/MIC > 5 recommended by Li et al. (2007) as a PD target associated with clinical response, our data derived from 50% f T > 5 × MIC also confirmed that meropenem monotherapy is competent for infections due to isolates with MICs of up to 32 mg/L, and importantly in the absence of alternative treatments, our data provide very useful directions on how to choose an optimal dosage regimen for meropenem monotherapy for the treatment of infections due to isolates with high MICs at different sites. To the best of our knowledge, this is the first PD analysis in which meropenem was used as a monotherapy aimed at MNBSs with MICs ≥ 16 mg/L and in which the profiles of meropenem tissue penetration were taken into account for setting the PD targets and calculating the PTAs and CFRs. In addition, we plan to perform a clinical trial to confirm our findings and to validate the optimal dosage by using the established equation for %f T > MIC and the Monte Carlo simulation.

## REFERENCES


## CONCLUSIONS

When faced with the daily challenge of infections due to MNBSs, we should try to reduce the gap between the available medical evidence for using meropenem against such infections and the dearth of alternative therapeutic options, some of which have not been sufficiently explored and/or whose efficacy in certain situations remains doubtful. Whether we can continue to use meropenem in the presence of MNBSs with high MICs remains controversial. The data analyses presented herein support the opinion that meropenem monotherapy can still be considered for use against MNBSs provided that (i) the MIC for the infecting pathogen isolates is ≤32 mg/L and (ii) a reasonable OTAT is used to drive the PK/PD profiles to acceptable exposures. However, in the absence of control trials, the continued appraisal of meropenem for use as a monotherapy, along with the optimal dosage regimens in clinical experience, will provide further important information on its utility against MNBSs.

## DATA AVAILABILITY STATEMENT

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

## AUTHOR CONTRIBUTIONS

XS performed the model simulation and wrote the manuscript. YW, LC, DY, and ML conceptualized and supervised the manuscript. All authors approved the final version of the manuscript.

## ACKNOWLEDGMENTS

The authors are grateful to Profs. Lu and Shen for their directions on meropenem anti-infection therapy, and we thank all members of our hospital library for access to their information resources.

modelling with Monte Carlo simulation to the development of susceptibility breakpoints for Neisseria meningitidis. Clin. Microbiol. Infect. 13, 33–39. doi: 10.1111/j.1469-0691.2006.01617.x


bile and dosing considerations for biliary tract infections based on sitespecific pharmacodynamic target attainment. Antimicrob. Agents Chemother. 55, 5609–5615. doi: 10.1128/AAC.00497-11


with ventilator-associated pneumonia. Antimicrob. Agents Chemother. 55, 1606–1610. doi: 10.1128/AAC.01330-10


**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 Song, Wu, Cao, Yao and Long. 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.

# Multiple Klebsiella pneumoniae KPC Clones Contribute to an Extended Hospital Outbreak

Carolina Ferrari<sup>1</sup> , Marta Corbella1,2, Stefano Gaiarsa<sup>1</sup> , Francesco Comandatore3,4 , Erika Scaltriti<sup>5</sup> , Claudio Bandi3,6, Patrizia Cambieri<sup>1</sup> , Piero Marone<sup>1</sup> and Davide Sassera<sup>7</sup> \*

<sup>1</sup> Microbiology and Virology Unit, Fondazione IRCCS Policlinico San Matteo, Pavia, Italy, <sup>2</sup> Biometric and Medical Statistics Unit, Fondazione IRCCS Policlinico San Matteo, Pavia, Italy, <sup>3</sup> Pediatric Research Center Romeo ed Enrica Invernizzi, University of Milan, Milan, Italy, <sup>4</sup> Department of Biomedical and Clinical Sciences "L. Sacco", University of Milan, Milan, Italy, <sup>5</sup> Risk Analysis and Genomic Epidemiology Unit, Istituto Zooprofilattico Sperimentale della Lombardia e dell'Emilia Romagna (IZSLER), Brescia, Italy, <sup>6</sup> Department of Biosciences, University of Milan, Milan, Italy, <sup>7</sup> Department of Biology and Biotechnology "L. Spallanzani", University of Pavia, Pavia, Italy

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

## Reviewed by:

Costas C. Papagiannitsis, University of Thessaly, Greece Philippe Glaser, Institut Pasteur, France Marco Maria D'Andrea, University of Rome Tor Vergata, Italy

> \*Correspondence: Davide Sassera davide.sassera@unipv.it

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 12 June 2019 Accepted: 13 November 2019 Published: 29 November 2019

#### Citation:

Ferrari C, Corbella M, Gaiarsa S, Comandatore F, Scaltriti E, Bandi C, Cambieri P, Marone P and Sassera D (2019) Multiple Klebsiella pneumoniae KPC Clones Contribute to an Extended Hospital Outbreak. Front. Microbiol. 10:2767. doi: 10.3389/fmicb.2019.02767 The circulation of carbapenem-resistant Klebsiella pneumoniae (CRKP) is a significant problem worldwide. In this work we characterize the isolates and reconstruct the spread of a multi-clone epidemic event that occurred in an Intensive Care Unit in a hospital in Northern Italy. The event took place from August 2015 to May 2016 and involved 23 patients. Twelve of these patients were colonized by CRKP at the gastrointestinal level, while the other 11 were infected in various body districts. We retrospectively collected data on the inpatients and characterized a subset of the CRKP isolates using antibiotic resistance profiling and whole genome sequencing. A SNP-based phylogenetic approach was used to depict the evolutionary context of the obtained genomes, showing that 26 of the 32 isolates belong to three genome clusters, while the remaining six were classified as sporadic. The first genome cluster was composed of multi-resistant isolates of sequence type (ST) 512. Among those, two were resistant to colistin, one of which indicating the insurgence of resistance during an infection. One patient hospitalized in this period was colonized by two strains of CRKP, both carrying the blaKPC gene (variant KPC-3). The analysis of the genome contig containing the blaKPC locus indicates that the gene was not transmitted between the two isolates. The second infection cluster comprised four other genomes of ST512, while the third one (ST258) colonized 12 patients, causing five clinical infections and resulting in seven deaths. This cluster presented the highest level of antibiotic resistance, including colistin resistance in all 17 analyzed isolates. The three outbreaking clones did not present more virulence genes than the sporadic isolates and had different patterns of antibiotic resistance, however, were clearly distinct from the sporadic ones in terms of infection status, being the only ones causing overt infections.

Keywords: KPC, genomic epidemiology, MDR, Klebsiella pneumoniae, nosocomial outbreak, colistin resistance

## INTRODUCTION

fmicb-10-02767 November 29, 2019 Time: 16:45 # 2

Klebsiella pneumoniae (Kp) is ubiquitous in the environment, part of the normal intestinal microbiota in humans and capable of colonizing the skin and nasopharynx of healthy individuals (Podschun and Ullmann, 1998; Broberg et al., 2014). Kp can persist on abiotic surfaces of different origin through the synthesis of biofilm, which can also make bacteria resistant to the action of antimicrobial agents (Di Martino et al., 2003). In immunocompromised or debilitated hospitalized patients with severe underlying diseases, Kp causes urinary tract, respiratory tract and bloodstream infections (Podschun and Ullmann, 1998) as well as other less frequent diseases, including osteomyelitis, arthritis (Ghorashi et al., 2011), and meningitis (Ko et al., 2002; Tumbarello et al., 2006; Nordmann et al., 2009). Kp is responsible for roughly 12% of Gram-negative infections in hospital intensive care units (ICUs) in Europe (European Centre for Disease Prevention and Control [ECDC], 2016). Kp invasive infections are associated with high rates of morbidity and mortality due to the high prevalence of resistance to most available antimicrobial agents (Patel et al., 2008; Borer et al., 2009). This is an emerging concern in clinical care resulting in an increase of mortality rates and costs.

The most commonly used class of antibiotics against nosocomial infections is β-lactams, which includes penicillin derivatives, cephalosporins, monobactams and the most recently developed carbapenems. Frequent use and abuse of these drugs, combined with the transmissibility of resistance determinants mediated by mobile elements (plasmids, transposons, and other integrative conjugative elements), has contributed to the spread of resistance to β-lactams by Kp (Mathers et al., 2015; Navon-Venezia et al., 2017). In the last 20 years, the emergence of isolates resistant to carbapenems has limited the efficacy of this last line treatment option hampering the use of this whole class of antibiotics, with few alternatives (Mathers et al., 2015; Navon-Venezia et al., 2017). One of the most common mechanism of resistance to carbapenems in Kp is the K. pneumoniae carbapenemase (KPC). This is an Ambler molecular class A serine enzyme that is able to hydrolyze a broad variety of β-lactams. KPC carbapenemases are plasmid-encoded, they have been originally associated with the Kp clonal group 258 (CG258) (Samuelsen et al., 2009; Breurec et al., 2013) but are not limited to it, as a number of occurrences of strains of other sequence types (ST) carrying the gene have been reported (Giakkoupi et al., 2010; Qi et al., 2010; Tzouvelekis et al., 2013; Markovska et al., 2015; Oteo et al., 2016; Villa et al., 2016; Wei et al., 2016; Aires et al., 2017). KPC-carrying Kp strains have recently spread worldwide, with some countries, including Italy, being heavily affected. In Italy, in 2017, 33.9% of the Kp nosocomial infections were caused by KPC strains (Sabbatucci et al., 2017). Most of these strains belong to the CG258, which in Italy has been shown to have been imported on four occasions, giving rise to four Italian-subclades (Gaiarsa et al., 2015). The spread of carbapenem-resistant strains has led to the resurgence of the use of colistin, previously abandoned due to its nephrotoxicity and neurotoxicity (Javan et al., 2015; Velkov et al., 2018). This has however resulted in the emergence of colistin-resistant strains, both due to the insurgence of chromosomal mutations (Cannatelli et al., 2014; Olaitan et al., 2014) and to the acquisition of plasmid-encoded resistance genes (Di Pilato et al., 2016; Liu et al., 2016).

Whole genome sequencing (WGS) of bacterial isolates is increasingly used for epidemiological investigations (Sabat et al., 2013). The use of genomics in clinical settings as a routine tool could in the future be an important aid to the microbiologist to accurately identify and characterize outbreaks at early stages, and to identify transmission routes. However, only from the combination of typing data with clinical and demographic data the correct interpretation of the origin and evolution of an outbreak can be obtained (Van Belkum et al., 2007).

The present retrospective study analyzes a nosocomial outbreak, lasting 10 months (August 2015 – May 2016), caused by K. pneumoniae KPC, in an ICU and the period immediately before and after, for a total of 12 months. Thanks to the combination of genomic, microbiological and clinical data it was possible to reconstruct the epidemic event and to characterize the multiple unrelated strains that spread in mostly temporally non-overlapping periods.

## THE EVENT

At the Fondazione IRCCS Policlinico San Matteo in Pavia, a 900 bed hospital in Northern Italy, cardio-respiratory patients are admitted to a specific Cardiorespiratory ICU where preoperative assessment, anesthetic treatment and intensive post-operative treatment of the patients undergoing cardiac and thoracic surgery are performed. This ICU has eight beds and is managed by 37 staff members. All ICU patients are subjected to surveillance through rectal swab at admission and once a week during the stay in the ward, to monitor for carbapenem-resistant Enterobacteriaceae colonization. In case of positivity, additional contact precautions are applied, which are interrupted only after three consecutive negative surveillance samples and are resumed in case of a single new positive screening swab.

Retrospectively, an increase in the number of carbapenemresistant Klebsiella pneumoniae (CRKP) colonizations and infections in the ICU was observed in the period from August 2015 to May 2016. The first infection was reported on 15th August 2015. After this event, the number of infected and colonized patients gradually increased and peaked twice, in December 2015 and March 2016. When the first increase in cases was detected (December 2015), additional surveillance measures were undertaken: increase in the training of health personnel and in the use of disposable devices, more accurate daily cleaning, and increased passive surveillance. Environmental screenings were performed in the ward after the second peak in cases in March 2016, all resulting negative. No cohorting or spatial isolation of individual patients were applied. The situation returned to the norm, with no clinical infections, in June 2016 (**Figure 1**). The period from June 2015 to May 2016 was thus analyzed to understand the characteristics of the observed prolonged outbreak. A total of 426 patients were hospitalized in the ICU during that period and the procedure for rectal swab screening

for Carbapenem-Resistant Enterobacteriaceae identified 23/426 CRKP-colonized patients (average age: 61.1 years; range: 27– 80 years). Out of these 23 patients, 18 were negative at the admission time into the ICU. Two of the five patients (numbers 6 and 22) that were positive at admission at the ICU were previously hospitalized in other wards (respectively another ICU and Cardiac Surgery) were they became colonized. The remaining three patients (numbers 2, 4, and 7) were positive at the first swab, suggesting positivity before hospitalization. During the event 11/23 (47.8%) patients died, 2/23 (8.7%) were discharged for rehabilitation and 10/23 (43.5%) were transferred to other hospitalization facilities or hospital wards. Eleven of these 23 (47.8%) patients developed overt infections, with CRKP isolated from blood, respiratory tract, urine and wounds. Seven of the 11 infected patients died while at the ICU. The demographic and hospitalization characteristics of the 23 patients are reported in **Table 1**. During the same period, a total of eight other patients resulted infected by non-carbapenem resistant Kp, thus 58% of total Kp infections were CRKP.

## MATERIALS AND METHODS

## Ethics Statement

The study was designed and conducted in accordance with the Helsinki declaration and approved by the Ethics Committee of Fondazione IRCCS Policlinico San Matteo in Pavia, Italy.

## Bacterial Strain Identification and Susceptibility Testing

During the entirety of the event, all surveillance rectal swabs were sent to the Microbiology and Virology Unit and were directly plated on chromID CARBA Agar (BioMérieux, Marcy-l'Étoile, France) to screen for the presence of carbapenemase-producing Enterobacteriaceae. In the same period, all clinical specimens (blood samples, bronchial aspirates, bronchoalveolar wash, urine samples, and wound swabs) were all cultured in parallel on all the following media: Columbia Blood Agar with 5% sheep blood, chocolate agar, selective media or on Schaedler agar and 5% sheep blood (BioMérieux SA, Marcy-l'Étoile, France) anaerobically and incubated at 37◦C overnight. Colonies suspected to be Kp based on morphology were identified with the MALDI Biotyper 3.1 system based on Matrix-Assisted Laser Desorption Ionization time-of-flight (MALDI-TOF) (Bruker Daltonics, Bremen, Germany).

We selected a subset of the Kp isolates from surveillance and clinical samples based on sample type, collection date and resistance profile, to perform additional characterization. Antibiotic susceptibility testing and minimum inhibitory concentrations (MICs) determinations were performed using the BD Phoenix 100 automated system (Becton, Dickinson and Company, Franklin Lakes, NJ, United States) and interpreted following the clinical breakpoints of the version 6.0 of the European Committee on Antimicrobial Susceptibility Testing [EUCAST] (2016)<sup>1</sup> . Tigecycline and carbapenems MICs were confirmed by E-test strips (BioMérieux, Marcy-l'Étoile, France). As recommended by European Committee on Antimicrobial Susceptibility Testing [EUCAST] (2016)<sup>2</sup> and the European Centre for Disease Prevention and Control [ECDC] (2016)<sup>3</sup> , colistin MICs were confirmed performing the broth microdiluition (MIC-Strip Colistin (MERLIN Diagnostika GmbH, Germany).

## DNA Extraction and Genome Sequencing

Total DNA extraction was performed for the selected isolates using the QIAamp DNA mini kit (Qiagen, Italy) according to manufacturer's instructions. DNA was sequenced using the Illumina MiSeq platform (Illumina Inc., San Diego, CA, United States), with paired-end runs of 2 × 250 bp, after Nextera XT library preparation.

## Genomic Analyses

Sequencing reads were quality checked using FastQC<sup>4</sup> and trimmed using the Trimmomatic software (Bolger et al., 2014). SPAdes-3.10.1 (Bankevich et al., 2012) was then used to assemble the pair-end reads using the accurate setting. Multilocus sequence typing (MLST) profiles and virulence/resistance gene variants were determined in silico using the Kleborate tool<sup>5</sup> . Presence of the mcr gene was tested using ResFinder Version 3.1.0 (database updated to February 20, 2019) (Zankari et al., 2012). Plasmid content was characterized using PlasmidFinder Version 2.0.1 (database updated to November 20, 2018) (Carattoli et al., 2014), while the contigs containing the blaKPC gene were

<sup>5</sup>https://github.com/katholt/Kleborate

<sup>1</sup>http://www.eucast.org/fileadmin/src/media/PDFs/EUCAST\_files/Breakpoint\_ tables/v\_6.0\_Breakpoint\_table.xls

<sup>2</sup>http://www.eucast.org/fileadmin/src/media/PDFs/EUCAST\_files/General\_ documents/Recommendations\_for\_MIC\_determination\_of\_colistin\_March\_ 2016.pdf

<sup>3</sup>https://ecdc.europa.eu/sites/portal/files/media/en/publications/Publications/ enterobacteriaceae-risk-assessment-diseases-caused-by-antimicrobial-resistantmicroorganisms-europe-june-2016.pdf

<sup>4</sup>www.bioinformatics.babraham.ac.uk/projects/fastqc/


TABLE 1 | Table showing the hospitalization characteristics of all patients admitted to the ICU in the examined period that resulted positive to carbapenem-resistant Klebsiella pneumoniae at least once.

compared using the BLASTn and the Mauve software (Darling et al., 2010) and subjected to Maximum Likelihood phylogeny using RAxML (Stamatakis, 2014) with the GAMMA substitution model, considering the Ascertainment bias and applying the Lewis correction (Lewis, 2001). The tree topology reliability was tested using 100 bootstrap replicates.

To estimate genomic variability, the obtained genome sequences were added to a selected dataset of K. pneumoniae genomes extracted from the PATRIC database (Wattam et al., 2016). PATRIC genomes were selected, using an in-house script, to be the closest in genomic distance to our strains. In detail, each genome was compared to all PATRIC database genomes using Mash (Ondov et al., 2016) and the 50 best hits were selected. All the obtained best hits lists were merged removing duplication, to obtain the final genomic dataset. CoreSNPs were extracted from the resulting dataset following a published method (Gaiarsa et al., 2015). Briefly, the Mauve software (Darling et al., 2010) was used to align all novel genomes and the similar PATRIC genomes to a well-characterized complete genome reference [NZ\_CP006923 (DeLeo et al., 2014)]. Individual alignments were merged using a Python script to obtain a multi-alignment file, allowing to extract coreSNPs (defined as variations of a single nucleotide flanked on each side by two nucleotides conserved in all the genomes analyzed).

The distribution of the coreSNPs distances among strains was visualized using the software R Version 3.2.3 (R Core Team, 2013) and the coreSNP cut-off threshold was determined manually by looking at the graph and finding that there were two groups of genome pairs, those with less than 15 SNPs and those with more than 35 SNPs (**Figure 2**). The distribution allowed us to safely infer that two genomes could be considered as part of the same transmission cluster if their distance in number of coreSNPs was lower than the 15 SNPs threshold value. The coreSNPs alignment was used to perform a phylogenetic analysis using the software RAxML (Stamatakis, 2014) with the GAMMA substitution model, considering the Ascertainment bias and applying the Lewis correction (Lewis, 2001). The tree topology reliability was tested using 100 bootstrap replicates.

The genomes of the epidemic clusters were investigated also in order to detect possible unique genomic characteristics. A SNP based phylogeny was generated for each cluster, using the same approach as the global phylogeny, and adding to each cluster the genome not part of the cluster that resulted the closest in the global phylogeny. The phylogenies and the genomic alignments of each cluster and their outgroup were used as inputs for a recombination analysis with the software ClonalFrameML (Didelot and Wilson, 2015). Furthermore, the presence and abundance of Insertion Sequences (IS) was tested using the software ISSeeker (Adams et al., 2016) against all IS retrieved from the Issaga database (Varani et al., 2011) using "Klebsiella" as organism keyword (on date July 17, 2019).

## RESULTS

An outbreak of CRKP occurred in the cardio-respiratory ICU from August 2015 to May 2016, involving 23 patients (12 colonized, 11 infected). In order to characterize the event, we retrieved all the CRKP strains isolated during the event and in the 2 months before. We obtained a total of 144 samples, 98/144 (68.1%) surveillance samples and 46/144 (31.9%) clinical samples (16 blood cultures, 12 bronchial aspirates, 1 bronchoalveolar wash, 8 urines, and 9 wound swabs). A total of 32/144 K. pneumoniae isolates, from 22 rectal swabs and 10 clinical specimens, were selected based on sample types and collection date, including at least one per each of the 23 positive patients, for antibiotic resistance characterization and genome sequencing (**Table 2**). The results of the antibiotic susceptibility tests are reported in **Supplementary Table S1** and show that all the isolates are resistant to at least one carbapenem: 32/32 to Ertapenem, 31/32 to Meropenem, 29/32 to Imipenem. Moreover, 31 isolates are resistant to aminoglycosides (25 to Amikacin, 22 to Gentamicin, 28 to Tobramycin). Twenty isolates are resistant to Fosfomycin (MICs > 64 mg/L), 20 to Colistin (MICs ranging from 4 to >64 mg/L) and three are resistant to Tigecycline (MICs > 2 mg/L). Only one strain from a surveillance sample (genome ID: 1880), resulted to be resistant to all five classes of antibiotics.

Whole-genome sequences were obtained for the 32 selected strains using Illumina technology and assembled into draft genomes after read polishing. Genomes resulted to be of good quality (average N50: 191,110; average number for contigs above 500 bp = 116; average genome size (5,671,622 bp, see **Supplementary Table S2** for complete genome characteristics). Genomes have been submitted to EMBL and are accessible under accession PRJEB32609 (ERP115310). Genomes were characterized by calculating the MLST, detecting antibiotic resistance determinants and virulence factors (see **Supplementary Tables S3**, **S4** for complete results). The most prevalent ST was ST258 (n = 19, 59.4%), followed by ST512 (n = 11, 34.4%). One isolate was found to belong to ST45 (3.1%). One new ST was found, a single-locus variant (-SLV) of ST940 (genome 1998 from patient 5, now registered as ST3985). Among the 5 patients with more than one isolate sequenced, one resulted to be colonized by isolates of different STs (patients 5).

Results of the antibiotic resistance factors analysis (**Supplementary Table S3**) were grouped in 12 drug classes according to the ARG-ANNOT database (Gupta et al., 2014), with β-lactamases divided into the six Lahey classes (Bush and Jacoby, 2010). All 32 analyzed isolates carried at least one ESBL gene (3.1% CTX-M-15, 3.1% SHV-1, 87.5% SHV-11, 6.3% SHV-12) and/or inhibitor-resistant β-lactamase genes (87.5% TEM-54 and 3.1% TEM-122). All the isolates were confirmed to be resistant to carbapenems (9.4% KPC-2 and 90.6% KPC-3). Kleborate identified colistin resistance as truncation or loss of mgrB gene in 20 isolates (62.5%), if less than 90% of the reference gene was covered by sequencing reads. Manual analysis allowed to detect the presence of an Insertion Sequence (IS5-like) interrupting mgrB in two resistant isolates (1880 and 1753). The interruption due to this mobile element has been previously reported (Cannatelli et al., 2014). All the other resistant isolates, belonging to cluster 3, exhibited a frameshift mutation in mgrB due to a 10 nucleotide insertion in position 109 of the gene. Presence of mcr genes was investigated with ResFinder, but they were never detected. Colistin antibiogram and genomic data resulted to be coherent for all clones, with two exceptions, clone 1760 indicated as resistant in Kleborate but resulting sensitive at the microdilution test and clone 1870 resistant to the phenotypic test but not to the genomic analysis.

Results of the virulence determinants presence show that only 4/32 strains (12.5%) contained the ybt locus, which encodes the biosynthesis pathway of the siderophore yersiniabactin (Lam et al., 2018). None of the strains were found to harbor other siderophores (aerobactin and salmochelin were screened). All the four ybt positive strains were isolated from surveillance rectal swabs of different patients and were sporadic cases. Two of them (genomes 1758 and 1760) showed ybt allele 13, carried by integrative conjugative element: ICEKp 2. The other two isolates (genomes 1826 and 1998) showed respectively the variants: ybt 10 (carried by ICEKp 4) and ybt 4 (plasmid-borne allele). The association between the mobile genetic elements found and the allelic variants of ybt gene is in agreement with previous observations (Lam et al., 2018). The biosynthetic pathway for the production of the toxin colibactin, as well as the virulence factor rmpA/rmpA2 (responsible for the expression of the hypermucose-viscous phenotype) were absent from all 32 strains. Finally, using the Kleborate tool we identified three distinct K-loci. The most common K-locus was KL107 (n = 28, 90.3%), followed by KL106 (n = 2, 6.5%) and KL24 (n = 1, 3.2%). As expected from the literature (Shu et al., 2009), they were found to be associated with wzi alleles 154, 29, and 101,



respectively. K-locus was not typable for one isolate (1998) (see **Supplementary Table S4** for full results).

The distribution of the coreSNPs distances among the 32 sequenced genomes was calculated and plotted to determine a threshold cut-off indicating epidemiological relatedness. A clear threshold was detected at 16 coreSNPs (**Figure 2**), allowing to determine the presence of three clades of genomes. Within each clade, all genomes have reciprocal coreSNP distances lower than the calculated threshold. A global coreSNPs maximumlikelihood phylogeny was performed, including the 32 genomes investigated in this work and 172 related genomes extracted from the PATRIC database, in order to contextualize our strains within the surrounding Kp diversity. The genomes with coreSNPs distance below the threshold clustered in three monophyletic clades (**Figure 3** for the cladogram, see also **Supplementary Figure S1** that shows subtrees of the three clusters) and were thus considered part of three separate outbreak clusters (green, red, and violet clusters in **Figures 3**, **4**). Five of the other six genomes do not cluster with other genomes of the outbreak and are thus considered sporadic cases. The last remaining genome (2205) is the phylogenetic sister group of cluster 2, but presents a relatively high number of coreSNPs with the other genomes of the cluster (average 42 coreSNPs), and was thus considered as a separate, sporadic, case.

Comparative genomic analyses were performed within the three clades and to compare them to their nearest neighbor, showing very limited genomic variation. No recombination was detected at the origin of any of the three epidemic clades (**Supplementary Figure S2**). IS content of isolates of each of the three outbreaking clusters was compared using ISfinder. The IS content resulted to be stable within each cluster, while each cluster appears to have more ISs than the evolutionary closest sporadic isolates. Specifically, isolates of Cluster 1 have IS of families ISL3 and IS1 that are absent in the sporadic sister group. Isolates of cluster 2 present IS66 sequences, absent in the evolutionary closest sporadic isolate. Isolates of cluster 2 are also richer in ISL3 and IS6 sequences than their neighbor and those of cluster 3 are richer in IS5 and IS6. It must be noted however that the accuracy, especially quantitatively, of such an analysis on draft genomes is limited, due to the difficulties in assemblying IS sequences (**Supplementary Figure S3**).

FIGURE 3 | Phylogenetic tree based on coreSNPs from 32 Klebsiella pneumoniae KPC strains isolated from the Intensive Care Unit and 172 related genomes retrieved from the PATRIC database. Sporadic strains are highlighted in blue; in green, red, and violet are the genomes belonging to three monophyletic clusters (distance in SNPs < 16).

FIGURE 4 | Timeline of the presence of carbapenem-resistant Klebsiella pneumoniae positive patients in the ICU in the period from June 2015 to May 2016. The periods of hospitalization are colored in gray. The other colors identify the sequenced samples based on genomic characterization. The colors green, red, and violet correspond to the three clusters identified based on SNPs distance, are reported sporadic cases in blue. Samples obtained from clinical infections (not surveillance swabs) are indicated by stars. The crosses indicate patients death. Black circles indicate the first positive CRKP rectal swab for each patient, negative at the admission. Black triangles indicate the first positive CRKP rectal swab for patients already CRKP-colonized at the admission time into the hospital. Black diamonds indicated the patients that became positive in other hospital wards.

## DISCUSSION

fmicb-10-02767 November 29, 2019 Time: 16:45 # 8

Here we present the results of the integrated characterization of the CRKP strains isolated during a 1-year period (10 months outbreak plus 2 months before) in the cardio-respiratory ICU of the San Matteo hospital in Pavia, Northern Italy. Retrospectively, starting in August 2015, we observed a strong increase in CRKP isolated from routine surveillance rectal swabs, taken from each ICU patient every week indicating a high number of colonized patients. We also observed an increase in CRKP isolations from clinical samples, indicating overt infections. This trend continued until June 2016, with a total of 23 colonized patients, among which 11 infected, prompting the investigation presented here. Antibiotic resistance characterization coupled with genome sequencing of 32 isolates from these 23 patients (**Table 2**), collected in the outbreak period and in the 2 months before, allowed the characterization of the extended outbreak. Most strains were found to belong to CG258, one of the most widespread KPC-bearing group in Italy and worldwide (Gaiarsa et al., 2015; Onori et al., 2015).

Performing a SNPs distance analysis (**Figure 1**) we found that twenty-six of the 32 isolates resulted to belong to three genome clusters (coreSNPs < 16) that appeared in the ICU, infected multiple patients, and then disappeared, with one single strain lasting for most of the event (**Figure 4**, see also **Supplementary Figure S1** for phylogenetic reconstructions of the three clusters). The remaining six genomes resulted to be unrelated and were consequently classified as sporadic isolates, as they all showed more than 35 coreSNPs with all the other analyzed genomes. This result is coherent with a recent multicenter study that proposed to set a threshold for SNPs distance in Kp outbreak at 21 (David et al., 2019). Interestingly, four of the six sporadic isolates represent the strains characterized in the 2 months before the outbreak (June– August 2015, see **Figure 4**), indicating that multiple strains were present in the ICU, but they were only responsible for colonization, did not cause clinical infections, nor they spread to multiple patients. The six sporadic isolates are characterized by variable levels of antibiotic resistance (number of resistance classes determined with Kleborate ranging from 3 to 11). Two of them are resistant to colistin (genomes ID: 1753, 1760). In terms of virulence, the only isolates that present yersiniabactin are four of the six sporadic isolates, all isolated from colonized, not infected, patients. Most sporadic isolates belong to the CG258 (n = 4), one belongs to ST45, already reported as a KPC-bearing strain in Italy (Cristina et al., 2016), and the final one to a novel ST, variant of ST940, now classified as ST3985. These sporadic isolates could be epidemiologically important, acting as a "reservoir" for the spread of the mobile element carrying blaKPC.

The strain responsible for the first cluster of isolations, belonging to ST512, was isolated from patient 5 from a screening sample in August 2015 and then from clinical samples in September. The patient remained colonized until his death in March. The isolates of this cluster are characterized by the resistance to 10 antibiotic classes, remaining susceptible to colistin, with two exceptions. The first one (genome ID: 1880) is the only strain of the cluster, belonging to patient 6, which present a correspondence between phenotypic (MIC > 64 mg/L) and genotypic (loss/truncation of mgrB gene) colistin resistance. The other colistin-resistant strain of the cluster (genome ID: 1870) was the third of four in the temporal series isolated from patient 5 and resulted phenotypically colistin-resistant (MIC = 4 mg/L), even though no colistin-resistance mutation/gene was detected in the genome. In order to investigate the presence of other mutations potentially causing colistin resistance, the genomes of the isolates belonging to patient 5 were manually examined, detecting 2 SNPs unique to the colistin-resistant isolate. These SNPs are co-localized in an intergenic region upstream of the pca operon, potentially responsible for the β-ketoadipate pathway. Understanding whether these SNPs could be related to colistin resistance would require additional investigations.

One additional CRKP was isolated from patient 5 on December 1st. This isolate was unrelated to those of cluster 1, belonging to ST3985 (a novel SLV of ST940). In order to evaluate whether this strain acquired the KPC plasmid from the ST512 strain co-present in patient 5, we compared the contigs harboring the KPC gene, performing a phylogenetic analysis including contigs from isolates of cluster 1, from the ST3985, and from another sporadic isolate (1753), as control. We found that the KPC-harboring contig from the ST3985 was divergent from all identical contigs retrieved from the ST512 isolates belonging to cluster 1, but also from the sporadic control, which resulted sister group of the isolates of the cluster (**Supplementary Figure S4**).

Four isolates were grouped in cluster 2, belonging again to ST512. This clone was short-lived but was transmitted from patient 7 to three other patients in the course of 6 weeks and caused one clinical infection. The isolates of this cluster are resistant to most antibiotics, retaining susceptibility to colistin and tigecycline. Two of four isolates resulted phenotypically sensitive to aminoglycosides, while genome analysis are only in partial agreement, indicating that all four strains carry the aph3- Ia gene, which should confer resistance to this class of antibiotics.

The largest of the clusters is represented by 17 strains, isolated from 12 patients during 5 months, causing five clinical infections and resulting in three deaths. The isolates of this cluster are those characterized by the highest level of resistance, retaining susceptibility only to tigecycline. Genomic analysis agrees with this result indicating the presence of resistance genes against 11 antibiotic classes, one more than the two other clusters. Indeed, all the isolates of this cluster resulted resistant to colistin with phenotypic tests, and the analysis of the genome content using Kleborate highlighted a possible resistance-causing truncation in the mgrB gene in all 17 strains. Manual analysis allowed to detect a frameshift mutation in mgrB due to a ten nucleotide insertion in position 109 of the gene.

## CONCLUSION

Over the course of 1 year, nine different strains of Carbapenem resistant K. pneumoniae were isolated in the ICU from 23 patients, 12 only colonized, 11 infected. Three strains colonized multiple patients, were the cause of all the seven clinical

infections reported in the ICU in this period and were responsible for the outbreak that lasted for 10 months. The three outbreaking strains did not present more virulence genes than the six sporadic isolates, four of which were actually the only ones exhibiting yersiniabactin. In terms of antibiotic resistance, both outbreaking and sporadic strains were phenotypically resistant to most classes, a result confirmed by strong repertoires of resistance genes. A clear correlation between antibiotic resistance profiles and number of colonizations or infections is thus not clearly evident. Our results highlight the importance of the overall environmental context, possibly more than the intrinsic characteristics of a strain, in determining the spread of different CRKP isolates.

## DATA AVAILABILITY STATEMENT

The genomes generated for this study can be found in the EMBL EBI repository, under the PRJEB32609 entry.

## ETHICS STATEMENT

Neither ethics committee approval, nor informed consent were required as all collected data were fully anonymized, there was no contact with patients and/or their families and no interventions or changes to treatment and management were made, in accordance with institutional guidelines.

## AUTHOR CONTRIBUTIONS

CF, CB, PM, and DS designed the project. MC and PC performed the microbiological analyses. ES performed the genome sequencing. CF, SG, and FC performed the bioinformatic analyses. CF, MC, and DS performed the epidemiological analyses. CF, MC, SG, and DS drafted the manuscript. CB, PM, and DS finalized the manuscript. All authors read and approved the final manuscript.

## REFERENCES


## FUNDING

This research was supported by the Italian Ministry of Education, University and Research (MIUR): Dipartimenti di Eccellenza Program (2018–2022) – Department of Biology and Biotechnology "L. Spallanzani", University of Pavia to DS.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02767/full#supplementary-material

FIGURE S1 | Recombination analysis of Cluster 1 (A), Cluster 2 (B) and Cluster 3 (C). The analysis was performed using the software ClonalFrameML and including the evolutionary closest sporadic genome as outgroup.

FIGURE S2 | Insertion sequences (IS) content of the three epidemic clusters. The evolutionary closest sporadic genome was added as a reference, last in order in the figure for each cluster.

FIGURE S3 | Phylogenetic trees of the three genomic clusters [Cluster 1 in panel (A), Cluster 2 in panel (B) and Cluster 3 in panel (C)], based on coreSNPs from the global alignment described in Figure 1, determined using RAxML, with 100 bootstrap, showing branch lengths and patients numbers. For each cluster, the sister group genome, as determined by the tree in Figure 1 was used as outgroup.

FIGURE S4 | Phylogenetic comparison of the KPC-bearing contigs of the 32 isolates. The contig from isolate 1998, the ST3985 genome, is clearly distinct from the contigs from the ST512 strains belonging to Cluster 1 (1897, 1935, 1955, 1961).

TABLE S1 | Antimicrobial susceptibility profiles determined using the BD Phoenix 100 System on the 32 investigated Klebsiella pneumoniae strains.

TABLE S2 | Characteristics of the sequenced genomes.

TABLE S3 | Predicted antimicrobial susceptibility profiles determined using the Kleborate tool on 32 Klebsiella pneumoniae KPC isolates.

TABLE S4 | Factors of the 32 Klebsiella pneumoniae KPC isolates.



pneumoniae sequence type 258 as a typical opportunistic pathogen. Antimicrob. Agents Chemother. 57, 5144–5146. doi: 10.1128/AAC.01052-13


**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 Ferrari, Corbella, Gaiarsa, Comandatore, Scaltriti, Bandi, Cambieri, Marone and Sassera. 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.

# Over-Expression of ISAba1-Linked Intrinsic and Exogenously Acquired OXA Type Carbapenem-Hydrolyzing-Class D-ß-Lactamase-Encoding Genes Is Key Mechanism Underlying Carbapenem Resistance in Acinetobacter baumannii

#### Edited by:

Leonardo Neves de Andrade, University of São Paulo, Brazil

#### Reviewed by:

Andres Felipe Opazo-Capurro, Universidad de Concepción, Chile Jitendraa Vashistt, Jaypee University of Information Technology, India

#### \*Correspondence:

Sheng Chen shechen@cityu.edu.hk

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 24 July 2019 Accepted: 19 November 2019 Published: 04 December 2019

#### Citation:

Wong MH-y, Chan BK-w, Chan EW-c and Chen S (2019) Over-Expression of ISAba1-Linked Intrinsic and Exogenously Acquired OXA Type Carbapenem-Hydrolyzing-Class D-ß-Lactamase-Encoding Genes Is Key Mechanism Underlying Carbapenem Resistance in Acinetobacter baumannii. Front. Microbiol. 10:2809. doi: 10.3389/fmicb.2019.02809 Marcus Ho-yin Wong<sup>1</sup>† , Bill Kwan-wai Chan<sup>1</sup>† , Edward Wai-chi Chan<sup>1</sup> and Sheng Chen1,2 \*

<sup>1</sup> State Key Laboratory of Chemical Biology and Drug Discovery, The Hong Kong Polytechnic University, Kowloon, Hong Kong, <sup>2</sup> Department of Infectious Diseases and Public Health, Jockey Club College of Veterinary Medicine and Life Sciences, City University of Hong Kong, Kowloon, Hong Kong

Acinetobacter baumannii is an important clinical pathogen which often causes fatal infections among seriously ill patients. Treatment options for managing infections caused by this organism have become limited as a result of emergence of carbapenem resistant strains. In the current study, whole genome sequencing, gene expression studies and enzyme kinetics analyses were performed to investigate the underlying carbapenem resistance mechanisms in fourteen clinical A. baumannii strains isolated from two hospitals, one each in Hong Kong and Henan Province, People's Republic of China. A large majority of the A. baumannii strains (11/14) were found to belong to the International Clone II (IC-II), among which six were ST208. Twelve of these strains were carbapenem resistant and found to either harbor blaOXA−23/blaOXA−72, or exhibit overexpression of the blaOXA−<sup>51</sup> gene upon ISAba1 insertion. Enzymatic assay confirmed that the OXA variants, including those of blaOXA−51, exhibited strong carbapenemdegrading activities. In terms of other intrinsic mechanisms, a weak correlation was observed between reduced production of outer membrane porin CarO/expression resistance-nodulation-division (RND) efflux AdeB and phenotypic resistance. This finding implied that over-production of carbapenem-hydrolyzing-class D-ß-lactamases (CHDLs), including the intrinsic blaOXA−<sup>51</sup> gene and the acquired blaOXA−<sup>23</sup> and blaOXA−<sup>24</sup> elements, is the key mechanism of carbapenem resistance in A. baumannii. This view is confirmed by testing the effect of NaCl, a known blaOXA inhibitor, which was found to cause reduction in carbapenem MIC by twofolds to eightfolds, suggesting that inhibiting OXA type carbapenemases represents the most effective strategy to control phenotypic carbapenem resistance in A. baumannii.

Keywords: Acinetobacter baumannii, carbapenem resistance, OXA-23, OXA-51, mechanisms

## INTRODUCTION

fmicb-10-02809 December 2, 2019 Time: 13:48 # 2

Acinetobacter baumannii is an important Gram-negative pathogen that often causes serious hospital infections, especially among immunocompromised patients in intensive care units (ICUs) (Bergogne-Berezin and Towner, 1996). The increasing mortality due to A. baumannii infections is of major concern as this pathogen exhibits the potential to evolve into carbapenem resistant variants through acquiring antibiotic resistanceencoding mobile genetic elements, which is often exacerbated by the intrinsic low membrane permeability of this organism. These features render A. baumannii one of the bacterial pathogens that exhibits the highest resistance rate in clinical settings (Peleg et al., 2008). In 2013, the United States Center for Disease Control and Prevention estimated that as many as 11,500 A. baumannii infections occurred annually, among which 63% were multidrug resistant, resulting in 500 deaths (Queenan et al., 2012). Likewise, A. baumannii is responsible for more than 1/5 of all clinical Gram-negative bacterial infections in Hong Kong and other Asia-Pacific regions, with a high portion being multidrug resistant (Liu et al., 2012). Recently, the World Health Organization has listed carbapenem-resistant A. baumannii to be "Priority 1: Critical" in its "Global Priority List of Antibiotic-Resistant Bacteria to Guide Research, Discovery and Development of New Antibiotics," further highlighting the worsen situation caused by this pathogen (World Health Organisation, 2017).

Carbapenem resistance in A. baumannii has been attributed to intrinsic cellular mechanisms, including loss of outer membrane porins (OMP) and over-expression of efflux pumps, which could result in alteration of cytoplasmic antimicrobial drug concentration and hence its bactericidal effect (Magnet et al., 2001; Siroy et al., 2005). Several OMPs, including CarO, HMP-AB and OmpW, were found to be involved in transportation of β-lactams across cytoplasmic membrane of this bacterial pathogen (Gribun et al., 2003; Siroy et al., 2006). While OMPs are responsible for the uptake of antibiotics, the multi-drug efflux systems are believed to be involved in removal of drugs by pumping them out of the cell. In particular, the resistancenodulation-division (RND) type efflux pumps, have long been hypothesized to play a role in rendering resistance toward various antibiotics. In A. baumannii, the most extensively studied RND efflux system is the adeABC gene product, which exhibits substrate specificity toward various β-lactams, including fluoroquinolones, aminoglycosides, tetracyclines and chloramphenicol (Higgins et al., 2004). Nevertheless, evidence confirming a direct linkage between carbapenem susceptibility and the presence/absence of these porin proteins and efflux systems in A. baumannii is currently not available.

Enzymatic mechanisms have been regarded as the key factors that mediate development of carbapenem resistance in Gram negative bacteria, including A. baumannii. Instead of blaIMP, blaVIM and blaNDM which are commonly identified in other bacterial pathogens, the carbapenem-hydrolyzing-class-D β-lactamases (CHDLs) are regarded as key determinants underlying the emergence of carbapenem-resistant A. baumannii (Poirel and Nordmann, 2006). CHDLs denote the OXA-type β-lactamases which exhibit carbapenem hydrolyzing activity. There are various types of blaOXA genes which are known to be harbored by A. baumannii, including blaOXA−51, blaOXA−23, blaOXA−24/40, blaOXA−58, blaOXA−143, and blaOXA−<sup>235</sup> (Higgins et al., 2013; Evans and Amyes, 2014). A considerable number of studies have been conducted on enzymes encoded by these resistance genes due to their uniqueness in A. baumannii, particularly the blaOXA−51-like β-lactamases, the genetic determinant of which is inherent in A. baumannii chromosome and can be readily overexpressed as a result of promoter activation by insertion sequences such as ISAba1 (Turton et al., 2006). Apart from this chromosomal resistance gene, plasmidborne blaOXA−23-like, blaOXA−24/40-like, and blaOXA−58-like elements are also frequently identified in resistant isolates. Among them, blaOXA−<sup>23</sup> is the most prevalent CHDL-encoding element in carbapenem-resistant A. baumannii worldwide (Mugnier et al., 2010). A previous study in China reported that 96.5% of carbapenem-resistant A. baumannii isolates carried blaOXA−23-like elements, and that over 96% of those isolates belonged to the Clonal Complex CC92 (Ji et al., 2014). In Hong Kong, the majority of carbapenem-resistant A. baumannii were blaOXA−23-like carrier that belonged to sequence type ST26 (Ho et al., 2010). Similarly, dissemination of blaOXA−23-bearing A. baumannii was also observed in other Asian countries, including Taiwan, Japan, and Korea (Peleg et al., 2008).

A comprehensive study was performed in 2013 to investigate the interplay between intrinsic and extrinsic mechanisms in mediating development of antimicrobial resistance in A. baumannii, including efflux systems, membrane porins, and production of CHDL β-lactamases (Rumbo et al., 2013). The study concluded that the blaOXA−<sup>24</sup> and blaOXA−<sup>58</sup> are the major determinants of carbapenem resistance in this organism, and that blaOXA−<sup>51</sup> and porin proteins were not involved in antimicrobial resistance of blaOXA−<sup>24</sup> and blaOXA−58-carrying isolates (Rumbo et al., 2013). Nevertheless, the exact role of intrinsic mechanisms in mediating carbapenem resistant phenotypes in A. baumannii strains carrying blaOXA−<sup>23</sup> and blaOXA−<sup>51</sup> remained unclear. In this work, we showed that intrinsic resistance mechanisms including RND-mediated efflux and reduced expression of porin proteins did not play a major role in mediating onset of carbapenem resistance in A. baumannii. Instead, the phenotype was mainly conferred by CHDL encoded by the blaOXA−23, blaOXA−72, and blaOXA−51-like genes. In addition, we identified several variants of blaOXA−51, which exhibited carbapenemhydrolyzing activities that resemble those encoded by blaOXA−<sup>23</sup> in terms of substrate profile, as over-expression of these blaOXA−<sup>51</sup> variants in A. baumannii upon insertional activation by ISAba1 conferred the host strain a carbapenem resistant phenotype identical to blaOXA−23-carrying strains.

## MATERIALS AND METHODS

## Bacterial Isolates

A total of 14 representative Acinetobacter baumannii clinical strains were first included in the genome sequencing, gene expression study, and western blot analysis as described below. The strains were isolated from patients of two hospitals, one each

in Hong Kong and Henan Province, People's Republic of China, during the period between 2000 and 2013. These strains exhibited various carbapenem resistance phenotypes and genotypes. The genetic identity of these isolates was confirmed by the Vitek II bacterial identification system prior to further analysis. The ethic approval for this study was covered by human subject ethic approval, 2018-039, approved by the Second Affiliated Hospital of Zhejiang University, Zhejiang, China. An addition 453 clinical carbapenem-resistant A. baumannii strains isolated from four different regions of China, were included in the screening of blaOXA genes in the latter part of the study. The experimental flow is illustrated by **Supplementary Figure S1**.

## Antimicrobial Susceptibility Test

Minimal inhibitory concentration (MIC) of carbapenems was tested for the 14 Acinetobacter baumannii strains by using the microdilution method and interpreted according to CLSI guidelines (Clinical and Laboratory Standards Institute, 2016). Briefly, bacterial strains were grown on MH agar. Bacterial cell suspensions at a concentration equivalent to a 0.5 McFarland Standard were prepared and inoculated into microplate wells containing different concentrations of carbapenem. The final volume in each well was 150 µL. The effect of the efflux pump inhibitor Phenylalanine-arginine β-naphthylamide (PAβN) and CHDL inhibitor Sodium chloride on the susceptibility of the test strains to carbapenem was determined by adding the compounds to specific wells of the microtiter plate to produce a concentration of 30 µg/ml and 200 mM, respectively. The MIC test was repeated twice to ensure the accuracy of the result.

## Whole Genome Sequencing and ST Typing of Acinetobacter baumannii

Whole genomic DNA was extracted from the strains using the InvitrogenTM PureLinkTM Genomic DNA Mini kit, followed by sequencing with an Illumina <sup>R</sup> NextSeq 550 system. Raw reads generated in this study and the Illumina reads of 11 strains obtained from the NCBI database were trimmed or filtered to remove low-quality sequences and adaptors. Genome data were annotated with the RAST tool (Overbeek et al., 2014) and Prokka (Seemann, 2014). Scaffolds obtained were analyzed by Geneious 9.7. A. baumannii type strain ATCC19606 was used as reference throughout the analysis. ST profiles were determined according to the A. baumannii MLST (Oxford) database using sequences extracted from whole genome sequencing. Discrimination of International Clone II was based on analysis of the blaOXA−<sup>51</sup> gene as described previously (Matsui et al., 2013).

## qRT-PCR Analysis on Gene Expression

Overnight culture of A. baumannii isolate was inoculated into fresh LB broth until OD 0.6 was reached. Total RNA was extracted using the QIAGEN RNeasy <sup>R</sup> Mini Kit. The extracted RNA was further treated with the Invitrogen Turbo DNAfreeTM Kit to remove any DNA contaminants. 1 µg of purified mRNA samples were reverse-transcribed to cDNA by using the InvitrogenTM SuperScriptTM III First-Strand Synthesis SuperMix kit for qRT-PCR. Real-time PCR was performed on a Roche LightCycler <sup>R</sup> 480 System and PowerSYBRTM Green PCR Master Mix was used as the reaction medium. The parameters of PCR were as followed: the reaction mixture was first incubated at 95◦C for 10 min for complete denaturation of template and activation of DNA polymerase; followed by 40 cycles of denaturation at 95◦C for 10 s, annealing and polymerization at 60◦C for 1 min. The melting temperature of PCR product was measured after completion of PCR to ensure the absent of non-specific PCR product. Expression level of endogenous genes was normalized with gyrB and A. baumannii ATCC19606 was used as control for comparison. The gene expression study was performed in triplicate. Primers used in this study are listed in **Supplementary Table S1**.

## Western Blot of blaOXA−51-Like Proteins

Cell lysates were solubilized by boiling with SDS running buffer for 10 min and subsequently separated by SDS-PAGE with 12% separating gel. Proteins were transferred to a PVDF membrane followed by blocking by skimmed milk for 1 h and incubated with mouse anti-OXA-51 antibody at 4◦C overnight. The goat anti-mouse antibody was used as the secondary antibody. The signal was generated by HRP substrate and detected by chemiluminator. The broad range anti-GADPH (ABCAM) was used as loading control.

## Further Screening of Carbapenem-Resistant Acinetobacter baumannii Clinical Isolates

An addition 453 clinical carbapenem-resistant A. baumannii strains, isolated from four different regions of China, were further screened for the presence of blaOXA−23, presence of ISAba1 in the promoter region of blaOXA−23, blaOXA−<sup>51</sup> and presence of ISAba1 in the promoter region of blaOXA−51. The primers sequences are shown in **Supplementary Table S1**. To screen for the presence of ISAba1 in the promoter region of blaOXA−<sup>23</sup> or blaOXA−51, PCR assay was performed using primer ISAba1-F was used together with reverse primer of blaOXA−<sup>23</sup> or blaOXA−51. Strains that did not contain blaOXA−<sup>23</sup> with insertion of ISAba1 in the promoter region were selected for Western-blot analysis as mentioned above to check for the over-expression of OXA-51 in these strains.

## RESULTS AND DISCUSSION

## Antimicrobial Susceptibility and Sequence Analysis of Acinetobacter baumannii Strains

Fourteen clinical A. baumannii strains were first tested for their susceptibility toward carbapenems. Resistance toward imipenem and meropenem was observable in 12 of the 14 strains, with MIC ranging from 8 to ≥64 µg/ml (**Table 1**). The fourteen strains were then subjected to whole-genome sequencing. Based on the sequences obtained, about half of the strains were found to belong to ST208 (6 out of 14), whereas three strains (AB2, AB5, and MH5) belonged to ST940. For the remaining five strains they belonged to different ST types. Only the three ST940 strains


TABLE 1 | Geneticand phenotypic characteristics of 14Acinetobacter baumannii clinical strains tested in this study.

Presence of endogenous genes and other resistance determinants were investigated by analysis of whole genome sequences.@MIC performed in the presence of NaCl.#Chromosome-based. HK, Hong Kong; HN, Henan province, China. IMP, imipenem; MER, meropenem; COL, colistin; PAβN, Phenylalanine-arginineβ-naphthylamide.

(32)

were found to belong to International Clone II (IC II) based on analysis on the intrinsic blaOXA−<sup>51</sup> gene harbored by these strains (**Table 1** and **Supplementary Figure S6**). The findings corroborated with those in previous studies which reported the dissemination of A. baumannii Clonal Complex 92 strains within the Asia Pacific region, including Taiwan, South Korea and China (Kim et al., 2013). It has previously been demonstrated that the high genome plasticity of A. baumannii may result in sporadic loss of endogenous genes, altering the physiological status and hence antibiotic susceptibilities of the organism (Roca et al., 2012). In view of the possibility that resistance formation is due to loss of specific physiological functions, the relationship between presence of various intrinsic determinants of carbapenem sensitivity was determined; the determinants tested included the outer membrane porin-encoding genes carO and ompA, as well as the efflux gene adeABC and its regulator adeRS. It was found that all the test strains carried the intact carO and ompA genes without insertion of IS elements; this finding was consistent with that of a previous report (Rumbo et al., 2013). However, three strains (AB1, AB3, and MH5) were found to lack adeRS, with the latter two also lacking adeB. The intrinsic blaOXA−<sup>51</sup> gene was detectable in all test strains, among them, the three ST940 strains were found to carry the variant gene blaOXA−99, whereas the blaOXA−66, blaOXA−79, blaOXA−<sup>82</sup> and blaOXA−<sup>83</sup> variants were identified in the other strains. Interestingly, the ISAba1 element was found to be inserted into the promoter region of blaOXA−<sup>51</sup> in five strains, all of which belonged to IC-II. Apart from carriage of various intrinsic CHDL genes, half of the strains were found to have acquired the blaOXA−<sup>23</sup> gene, two of which were located in the chromosome. One strain was also found to harbor the blaOXA−<sup>72</sup> element (a variant of blaOXA−24) (**Table 1**). It has been described that alteration of amino acids of the adeRS gene product could result in change of expression level of adeB (Wen et al., 2017). Sequences of carO, adeB and adeRS were then analyzed and aligned in attempt to establish their relationship with carbapenem susceptibility. However, it was shown that the genetic differences detectable in these genes were linked closely to their genetic types (IC-II and non IC-II) rather than carbapenem sensitivity (**Supplementary Figures S2**–**S5**). Other known carbapenemase genes such as blaNDM−<sup>1</sup> were not detectable in the test strains. The combinational analysis of carbapenem resistance genotypes and their genetic characteristics showed that strains exhibiting carbapenem resistance were either carrying an additional CHDL gene (blaOXA−<sup>23</sup> and blaOXA−24) or having an ISAba1 linked upstream region in their blaOXA−<sup>51</sup> element. Consistently, two carbapenem-susceptible isolates carried neither blaOXA−<sup>23</sup> or blaOXA−<sup>24</sup> nor insertion of ISAba1 in the intrinsic blaOXA−<sup>51</sup> (**Tables 1**, **2**).

## Efflux Pumps Did Not Mediate Carbapenem Resistance in Acinetobacter baumannii

The effect of RND-efflux systems on carbapenem susceptibility in A. baumannii clinical strains was evaluated by supplementing Phenylalanine-arginine β-naphthylamide (PAβN) at a concentration of 30 µg/ml in determination of MIC toward carbapenem. As shown in **Table 1**, we found that addition of PAβN exhibited minimal effect on altering carbapenem MIC except for two strains (**Table 1**). This finding demonstrated that RND efflux pumps only play a minimal role in mediating carbapenem resistance in A. baumannii strains of hospital origins and that the resistance phenotype is mainly conferred by the blaOXA enzymes.

TABLE 2 | Expression level of genes related to carbapenem resistance in A. baumannii strains.


Expression levels were normalized with ATCC19606 for carO, adeB, adeR and blaOXA-51. Strains lacking the corresponding genes were depicted by "n.a." Expression of adeB was not detected in AB2 and AB5.


TABLE 3 | Surveillance of mechanisms of carbapenem resistance in clinical carbapenem-resistant A. baumannii isolates from different regions of China.

MER, meropenem; IMP, imipenem; ERA, ertapenem. #OXA-51 expression level was determined by Western-blot using OXA-51 specific antibody developed in our lab.

To further investigate the role of various host determinants in mediating changes in carbapenem susceptibility, the expression level of various putative resistance genes, including the RND efflux pump gene adeB, its regulator adeR, the outer membrane porin gene carO, as well as the blaOXA−<sup>23</sup> and blaOXA−51 like β-lactamase-encoding elements, were further analyzed by RT-qPCR. The expression level of each gene was normalized by the gene encoding the gyrase protein subunit B (gyrB) of the A. baumannii type strain ATCC19606. The results are shown in **Table 2** and **Supplementary Figures S7–S11**. For carO, varied expression level was observed among the test strains, with most isolates exhibiting a level similar to or lower than that of ATCC19606. For the two susceptible strains, AB10 and MH1, carO expression was lower than that of ATCC19606 and most of the other carbapenemresistant counterparts. The result is in line with a previous study, in which a decreased expression of carO has been described in carbapenem susceptible isolates recovered from Brazil (Fonseca et al., 2013). It should be noted that although strains AB7, AB8, and MH3 exhibited elevated carO expression by an amplitude >twofolds, their contribution to carbapenem resistance was hard to be addressed due to the presence of other major resistance mechanism, presence of OXA types of carbapenemases (**Tables 1**, **2**).

An unexpected phenomenon regarding the adeB expression level of the test strains was observed. It was proposed that elevated expression of adeABC might contribute to reduced carbapenem susceptibility, possibly due to enhanced extrusion of intracellular antibiotics through efflux activity (Kim et al., 2013). Nevertheless, except for the two test strains AB3 and MH5, which lacked adeB, all other test strains exhibited a very low transcript level of this efflux gene, with a range of 0.07 to 0.31 compared to ATCC19606. Transcript of adeB was not detected in strains AB2 and AB5. Of note, adeB expression in the two carbapenem sensitive strains AB10 and MH1 was similar to most of their resistant counterparts. In order to evaluate whether decreased adeABC expression was due to effect of transcription regulation, expression level of adeR was also examined in an attempt to determine the degree of correlation between the two. Contrary to the data on adeB, similar or higher transcription level of adeR was observed among the majority of the test isolates, ranging from 0.91 to 3.63 folds when compared to ATCC19606 (**Table 2**). Correlation between expression of adeB and adeR, as well as adeR and carbapenem susceptibility, therefore could not be established. Taken together, the results suggest that the adeABC gene plays a negligible role in mediating changes in carbapenem susceptibility, particularly in strains which harbor additional CHDL genes such as blaOXA−<sup>23</sup> and blaOXA−24. Indeed, the results of genotypic analysis corroborated with those of the expression study in that a lack of adeABC efflux system was observed in several resistant strains.

## Overexpression of OXA-23 and Intrinsic OXA-51 Through ISAba1 Insertion Was the Major Mechanisms of Carbapenem Resistance

Antimicrobial susceptibility test revealed that all blaOXA−<sup>23</sup> carriers were resistant to imipenem and meropenem but

exhibited varied MIC values, presumably due to the fact that these strains carried various variants of blaOXA−<sup>23</sup> and blaOXA−51. Expression level of blaOXA−<sup>51</sup> and blaOXA−<sup>23</sup> was examined in an attempt to determine the degree of contribution of carbapenemases encoded by these genes to reduction in carbapenem susceptibility of the A. baumannii strains tested in this study. All blaOXA−<sup>23</sup> genes were found to carry insertion in the promoter region by ISAba1. Consistently, expression level of the blaOXA−<sup>23</sup> gene was found to be up-regulated 2.24 to 7.64 folds among the strains. Discrepancies between the effects of plasmidborne and chromosome-based elements were not observed (**Tables 1**, **2**). Among the 6 blaOXA−<sup>23</sup> carriers which lacked an over-expressed blaOXA−51, MIC values of imipenem and meropenem ranged from 16 to ≥64 µg/ml and 8 to ≥64 µg/ml, respectively. In particular, the higher transcript level of blaOXA−<sup>23</sup> did not necessarily confer higher resistance level toward carbapenems. Since identical blaOXA−<sup>23</sup> was harbored by all the strains, the possibility of varied activity toward carbapenem in individual variants can be ruled out. The lack of association between blaOXA−<sup>23</sup> expression and susceptibility might then be attributed to physiological status of individual A. baumannii strains. Result of blaOXA−<sup>51</sup> expression revealed consistency between ISAba1 insertion and transcript level of this endogenous carbapenemase. For the five test strains which carried the IS element, blaOXA−<sup>51</sup> expression ranged from 25.09 to 64.96 folds compared to ATCC19606. To ensure that the overexpression of blaOXA−<sup>51</sup> was consistent with their protein production, we took advantage of the monoclonal antibody specific to OXA-51 variants that produced by our lab to examine the production level of these OXA-51 variants by western blotting (**Figure 1**). It was shown that protein production correlated well with the qRT-PCR data, and that the five strains apparently over-produced the OXA-51 enzymes. The result confirmed that the ISAba1 insertion sequence is a key factor that promotes blaOXA−<sup>51</sup> expression. To summarize, although the physiological status of individual strain may affect carbapenem susceptibility to some extent, resistance was observed only among A. baumannii strains which carried either ISAba1 blaOXA−<sup>51</sup> or additional CHDL-encoding genes such as blaOXA<sup>23</sup> and blaOXA−72.

To confirm that overexpression of OXA-23 and intrinsic OXA-51 is the major mechanisms of carbapenem resistance in clinical A. baumannii, a total of 453 carbapenem-resistant clinical A. baumannii isolated from four different regions of China were screened for the presence of these two mechanisms (**Table 3**). Our results showed that 404/453 strains carried over-expressed OXA-23 with ISAba1 in the promoter region of blaOXA−23, while the rest of 49 strains that did not carry blaOXA−<sup>23</sup> exhibited an insertion of ISAba1 in the promoter region of blaOXA−<sup>51</sup> and over-expression of OXA-51, which was confirmed by Western-blot using specific antibody to OXA-51 (**Table 3**). These data suggested that mechanisms of carbapenem resistance in clinical A. baumannii were mediated by either overproduction of OXA-23 or OXA-51 through insertion of ISAba1 in their promoter region.

Correlation analysis of the enzymatic activity of OXA types carbapnemase and meropenem MIC of strains producing these enzymes indicated that: (1) A. baumannii strains that over-expressed OXA-23/OXA-24 regardless of the status of over-expression of OXA-51 variants generally displayed higher MIC of meropenem (≥16 µg/ml); (2) A. baumannii strains that over-expressed OXA-51 variants only displayed relatively low MIC of meropenem, with OXA-83 (32 µg/ml) being higher than OXA-79/OXA-83 (8 µg/ml); and (3) Western-blot experiments indicated that OXA-51 variants exhibited base line expression without the insertion of ISAba1, which is probably true for extrinsic OXAs. OXA variants with lower activity like OXA51 variants are needed to be overexpressed to counter carbapeneme resistance phenotype, while variants with high catalytic activity such as OXA-72 may not need to be overexpressed, its basic line expression might be enough to mediate high carbapenem resistance in A. baumannii in the case of MH2 in this study.

## CONCLUSION

Acinetobacter baumannii is an important pathogen that may cause fatal infections in nosocomial settings. It was thought that the interplay of multiple resistance mechanisms, including over-expression of endogenous efflux systems, suppression of expression of porin proteins, reduced membrane permeability and carriage of CHDL-encoding elements could lead to carbapenem resistance in this bacterial pathogen. In the current study, through systematic characterization of a set of clinical A. baumannii strains with different OXA-23 and MIC profiles, we demonstrated that the RND efflux system and membrane porin proteins were not the key factors that conferred carbapenem resistance in the A. baumannii strains tested. Instead the OXA-23 and OXA-24 enzymes (OXA-72 variant) were found to be the key elements underlying carbapenem resistance in A. baumannii strains as a result of insertion of the ISAba1 element, which provides a promoter for over-expression of the blaOXA−<sup>23</sup> and blaOXA−<sup>23</sup> genes (Nigro and Hall, 2016). In A. baumannii strains lacking these genes, carbapenem resistance was mainly due to the overexpression of blaOXA−<sup>51</sup> variants, again through insertion of the promoter-bearing ISAba1 elements. Findings in this study emphasized the role of blaOXA type enzymes in mediating carbapenem resistance in this major clinical pathogen. These data provided essential information for the design of new strategies to treat clinical infections caused by carbapenemresistant A. baumannii strains.

## DATA AVAILABILITY STATEMENT

The whole genome shotgun sequences of isolates included in this study have been deposited in GenBank with the following accession numbers: AB1, SAMN09667770; AB2, SAMN09667771; AB3, SAMN09667772; AB4, SAMN09667773; AB5, SAMN09667774; AB7, SAMN09667775; AB8, SAMN-09667776; AB10, SAMN09667777; MH1, SAMN09667778; MH2, SAMN09667779; MH3, SAMN09667780; MH5, SAMN09667781; MH6, SAMN09667782; and MH7, SAMN09667783.

## AUTHOR CONTRIBUTIONS

fmicb-10-02809 December 2, 2019 Time: 13:48 # 8

MW and BC designed and performed the research, and drafted the manuscript. EC edited the manuscript. SC supervised the study and edited the manuscript.

## REFERENCES


## FUNDING

This work was supported by the Health and Medical Research Fund of the Food and Health Bureau, Hong Kong (15141322) to SC.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02809/full#supplementary-material


FEMS Microbiol. Lett. 258, 72–77. doi: 10.1111/j.1574-6968.2006. 00195.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 © 2019 Wong, Chan, Chan and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Edited by: Raffaele Zarrilli, University of Naples Federico II, Italy

#### Reviewed by:

Mohamamd Emaneini, Tehran University of Medical Sciences, Iran Morovat Taherikalani, Lorestan University of Medical Sciences, Iran

#### \*Correspondence:

Durg Vijai Singh durg\_singh@yahoo.co.in; singhdv@ils.res.in; dvsingh@cusb.ac.in

†These authors have contributed equally to this work

#### ‡Present address:

Sasmita Panda, Department of Pathology and Microbiology, University of Nebraska Medical Center, Omaha, NE, United States

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 08 August 2019 Accepted: 12 November 2019 Published: 04 December 2019

#### Citation:

Aggarwal S, Jena S, Panda S, Sharma S, Dhawan B, Nath G, Singh NP, Nayak KC and Singh DV (2019) Antibiotic Susceptibility, Virulence Pattern, and Typing of Staphylococcus aureus Strains Isolated From Variety of Infections in India. Front. Microbiol. 10:2763. doi: 10.3389/fmicb.2019.02763

# Antibiotic Susceptibility, Virulence Pattern, and Typing of Staphylococcus aureus Strains Isolated From Variety of Infections in India

Shifu Aggarwal<sup>1</sup>† , Smrutiti Jena<sup>1</sup>† , Sasmita Panda<sup>1</sup>†‡, Savitri Sharma<sup>2</sup> , Benu Dhawan<sup>3</sup> , Gopal Nath<sup>4</sup> , N. P. Singh<sup>5</sup> , Kinshuk Chandra Nayak<sup>6</sup> and Durg Vijai Singh1,7 \*

1 Infectious Disease Biology, Institute of Life Sciences, Bhubaneswar, India, <sup>2</sup> Jhaveri Microbiology Centre, LV Prasad Eye Institute, Brien Holden Eye Research Centre, Kallam Anji Reddy Campus, Hyderabad, India, <sup>3</sup> Department of Microbiology, All India Institute of Medical Sciences, New Delhi, India, <sup>4</sup> Department of Microbiology, Institute of Medical Sciences, Banaras Hindu University, Varanasi, India, <sup>5</sup> Department of Microbiology, Faculty of Medical Sciences, University of Delhi, New Delhi, India, <sup>6</sup> Institute of Life Sciences, Bhubaneswar, India, <sup>7</sup> Department of Biotechnology, Central University of South Bihar, Gaya, India

Staphylococcus aureus is one of the major causes of nosocomial infections. This organism produces powerful toxins and cause superficial lesions, systemic infections, and several toxemic syndromes. A total of 109 S. aureus strains isolated from a variety of infections like ocular diseases, wound infection, and sputum were included in the study. Minimum inhibitory concentration (MIC) was determined against 8 antimicrobials. PCR determined the presence of 16S rRNA, nuc, mecA, czrC, qacA/B, pvl, and toxin genes in S. aureus isolates. Pulse-field gel electrophoresis (PFGE), multi-locus sequence typing (MLST), SCCmec, spa-, and agr-typing and serotyping determined the diversity among them. All isolates of S. aureus were resistant to two or more than two antibiotics and generated 32 resistance patterns. These isolates were positive for 16S rRNA and S. aureus-specific nuc gene, but showed variable results for mecA, czrC, and qacA/B and pvl genes. Of the 32 methicillin-resistant S. aureus (MRSA), 13 strains carried SCCmec type V, seven type IV, two type III, and nine carried unreported type UT6. Of the 109 strains, 98.2% were positive for hlg, 94.5% for hla, 86.2% for sei, 73.3% for efb, 70.6% for cna, 30.2% for sea, and 12.8% for sec genes. Serotypes VII and VI were prevalent among S. aureus strains. PFGE analysis grouped the 109 strains into 77 clusters. MLST classified the strains into 33 sequence types (ST) and eight clonal complexes (CCs) of which 12 were singletons, and two belong to new allelic profiles. Isolates showed 46 spa-types that included two new spa-types designated as t14911 and t14912. MRSA and methicillin-susceptible S. aureus (MSSA) isolates were diverse in terms of antibiotic resistance pattern, toxin genotypes, SCCmec types, serotypes and PFGE, MLST, and spa-types. However, few isolates from eye infection and wound

infection belong to CC239, ST239, and spa-type t037/t657. The study thus suggests that S. aureus strains are multidrug resistant, virulent, and diverse irrespective of sources and place of isolation. These findings necessitate the continuous surveillance of multidrug-resistant and virulent S. aureus and monitoring of the transmission of infection.

Keywords: antibiotic susceptibility, virulence, MLST, spa-typing, PFGE, biofilm, Staphylococcus aureus

## INTRODUCTION

Staphylococcus aureus commensal to human skin and mucous membranes could cause nosocomial (Lindsay and Holden, 2004) and systemic infections (Jarraud et al., 2002). The isolation of methicillin-resistant S. aureus (MRSA) from ocular infections varies from 3 to 30% in a hospital in India and other countries (Shanmuganathan et al., 2005; Freidlin et al., 2007). MRSA strains belonging to ST5, ST72, and ST88 and isolated from severe eye infections in India were resistant to all antibiotics except tetracycline, chloramphenicol, and cefazolin (Nadig et al., 2012). Godebo et al. (2013) showed that 94.5% of S. aureus isolated from wound infection were resistant to penicillin, 91.8% to ampicillin, and 76.7% to oxacillin.

Several studies have shown the presence of toxin genes among MRSA. The presence of the sea gene in MRSA varies from country to country (Mehrotra et al., 2000; Kim et al., 2006; Wang et al., 2013). However, hla gene was present in all isolates (Shukla et al., 2010). MRSA isolated from conjunctivitis in Nigeria belonging to ST88 and SCCmec type IV were positive for pvl gene (Ghebremedhin et al., 2009). However, pvl gene positive methicillin-susceptible S. aureus (MSSA) strains belonged to ST30 (D'Souza et al., 2010). S. aureus carrying the pvl gene and belonging to ST239, ST5, and ST88 was reported from a teaching hospital in China (Liu et al., 2009). MSSA belonging to ST121 and spa-type 287 isolated from community-acquired pneumonia in young patients carried the virulence genes (cna and bbp) and pvl (Baranovich et al., 2010). The role of virulence genes in S. aureus pathogenesis may vary from one infection type to another type of infections. Dhawan et al. (2015) reported the isolation of SCCmec type IV and V clones of MRSA in an Indian hospital. Several other workers also showed a decrease in SCCmec III MRSA isolation but increased SCCmecIV and V MRSA isolation (Hsu et al., 2005; D'Souza et al., 2010). Multidrug-resistant isolates belonging to ST239 and SCCmec type III were slowly replaced by multidrugsusceptible ST22 (SCCmec type IV) and ST772 (SCCmec type V) in hospitals (D'Souza et al., 2010).

Several molecular biology techniques like multi-locus sequence typing (MLST), pulse-field gel electrophoresis (PFGE), SCCmec typing, and spa-typing have been used to study epidemiology and clonal diversity of S. aureus (Maslow et al., 1993; Norazah et al., 2001; Ghaznavi-Rad et al., 2011). However, not a single technique alone could discriminate the bacteria because of differences in the degree of typeability, reproducibility, and discriminatory power (Tenover et al., 1994). Overall analysis of different typing techniques can provide information on diversity of the isolates that can be useful for outbreak investigations. In India, S. aureus is rated as one of the major pathogen causing a variety of infections and showing resistance to several antibiotics; however, not much information is available on their antibiotic susceptibility, virulence profile, and genomic diversity. In this study, our aim was to determine the antibiotic susceptibility pattern, virulence profiles, and genomic diversity among MRSA and MSSA isolated from patients with a variety of infections, including ocular diseases and collected from different parts of India from 2007 to 2015. Genetic, serotype, and phenotypic data were used to determine whether isolates from a variety of infections had similar characteristics.

## MATERIALS AND METHODS

## Bacterial Strains

A total of 109 S. aureus strains isolated from patients visited/admitted to hospitals with infections in different part of India between July 2007 and November 2015 were included in the study. These isolates were from LV Prasad Eye Institute, Bhubaneswar (n = 54), comprised of microbial keratitis (n = 18), eyelid abscess (n = 8), endophthalmitis (n = 5), Steven Johnson syndrome with bacterial keratitis (n = 9), suture-related infections (n = 3), and other ocular infection (n = 5); LV Prasad Eye Institute, Hyderabad (n = 10) comprised of cornea scrapping (n = 5), pus from eye (n = 4), and suture-related infections (n = 1); Institute of Medical Sciences, Banaras Hindu University, Varanasi (n = 21) comprised of wound infection (n = 16) and unknown sources (n = 5); All India Institute of Medical Sciences, New Delhi (wound infection n = 10); and University College of Medical Sciences, Delhi (wound infection n = 9). Also, five isolates were from the conjunctiva of the asymptomatic healthy volunteers LV Prasad Eye Institute, Bhubaneswar. We conducted the study following the guidelines mentioned in the Declaration of Helsinki. We identified all the 109 isolates by using biochemical tests including Gram staining, catalase production, fermentation of glucose and mannitol, and ID32 STAPH strips using ATBTM NEW v.1.0.0 software on an ATBTM reader (bioMerieux, France) (Panda et al., 2014). The amplification of the S. aureus nuc gene confirmed the identity of isolates (Hirotaki et al., 2011). We used S. aureus ATCC 25293 and S. aureus ATCC 29213 as quality control strains for antibiotic susceptibility testing, and S. aureus ATCC 25923 and ATCC 43300 as a reference for serotyping, PFGE, MLST, and spa-typing.

## Coagulase Gene Typing

Coagulation-inhibition test with coagulase type I–VIII-specific antisera (staphylococcal coagulase antiserum kit; Denka Seiken, Inc., Tokyo, Japan) was conducted to determine the coagulase type of S. aureus following the manufacturer's instructions (Goh et al., 1992). Briefly, a single colony for each test

strain was suspended in BHI broth (Becton Dickinson Co.) and incubated at 37◦C for overnight. Then centrifuged the culture and 0.1 ml of the supernatant used as test antigen. Distributed an aliquot (0.1 ml) of the test antigen into ten tubes followed by addition of 0.1-ml aliquots of anticoagulase types I–VIII sera to first eight tubes, except 9th and 10th tubes which were used as positive and negative controls and incubated at 37◦C for 1 h. After that, 0.2 ml of diluted rabbit plasma was added to each tube and incubated at 37◦C for 1 h. Visual inspection judged the coagulation of plasma after 2, 4, 24, and 48 h and accordingly, strains were typed based on results obtained with staphylocoagulase reaction showing coagulation inhibition.

## Minimum Inhibitory Concentration (MIC) Determination

Minimum inhibitory concentrations (MICs) of oxacillin, chloramphenicol, vancomycin, tetracycline, gentamicin, erythromycin, clindamycin, and trimethoprim were determined by broth microdilution methodology as recommended by the CLSI breakpoints. The 96-well plates were incubated at 37◦C and were read for turbidity after 24 h.

## Polymerase Chain Reaction (PCR) Assays

The presence of genes encoding for methicillin resistance (mecA), the nuclease (nuc), Panton-Valentine leukocidin (pvl), cadmium resistance (czrC), and quaternary ammonium resistance (qacA/B) was determined by hexaplex PCR (Panda et al., 2014). PCR identified the presence of msrA, ermA, ermC (erythromycin resistance), tetK (tetracycline resistance) genes (Duran et al., 2012). Also, PCR determined the presence of gene encoding for resistance to aminoglycosides [aac (6')/aph (2), aph (3'-III)] by the method described earlier (Schmitz et al., 1999). The presence of catpC221, catpC223, and catpC194 (chloramphenicol resistance) was determined by PCR as described by Argudín et al. (2011). The mphC (clindamycin resistance) gene was detected by PCR method described earlier (Panda et al., 2016).

## SCCmec Typing

Two PCRs, MPCR1 and MPCR2 were used to detect the presence of mec complex, ccr complex, and SCCmec type among S. aureus (Kondo et al., 2007).

## Virulence Gene Profile and Accessory Gene Regulator (Agr) Typing

PCR determined the presence of Staphylococcal enterotoxin (SE) genes encoding for seA, seC, and seI (Monday and Bohach, 1999; Jarraud et al., 2002). Also, the presence of hemolysin genes, hlA and hlG, was determined by PCR (Mitchell et al., 2010; Paniagua-Contreras et al., 2012). PCR was used to detect the presence of collagen adhesion (cna) and extracellular fibrinogen binding protein (efb) among S. aureus strains (Zecconi et al., 2006). The presence of intracellular adhesion genes (icaA, icaD) was determined by PCR as described by Arciola et al. (2001). PCR amplification was carried out to determine the presence of agr alleles using group-specific primers as described by Gilot et al. (2002).

## Pulsed-Field Gel Electrophoresis (PFGE)

Pulsed-field gel electrophoresis of S. aureus genomic DNA digested with SmaI (NEB) was carried out by the protocol described for S. aureus by Centre for Disease Control and Prevention. The dendrogram of similarity showing the clustering of the isolates according to banding patterns was generated with Bionumerics software, version 7.1 (Applied Maths, Belgium) using the Dice index and the un-weighted pair group method with arithmetic average (UPGMA) with 0.5% optimization and 1% position tolerance. Isolates showing similarity coefficient of up to 80% were considered belonging to similar pulsotype (Van Belkum et al., 2007).

## Multi-Locus Sequence Typing (MLST)

The internal fragments of seven housekeeping genes, viz., arcC, gmk, aroE, glpF, pta, tpi, and yqil were amplified by PCR method described earlier (Enright and Spratt, 1999). The amplified products were purified (ExoSAP; Affymetrix, Cleveland, OH, United States) and both strands sequenced using an ABI sequencer model 3500 (Life Technologies, Marsiling, Singapore) at the sequencing facility of the Institute of Life Sciences (Bhubaneswar, India). The nucleotide sequences were aligned using Mega 5.2 software. After manually comparing with reported alleles, STs were assigned accordingly. Sequencing was performed in biological duplicates to confirm the presence of novel alleles.

The advanced cluster analysis was performed to define the clonal complexes (CCs) by using Bionumerics software, version 7.1 (Applied Maths, Belgium). A minimum spanning tree (MST) was constructed using the MLST data and partitions were created to form clusters. The similarity in at least six alleles grouped isolates of S. aureus in one CC. The central ST of each separation was used to designate a CC.

## Spa-Typing

PCR amplified the polymorphic X region of Staphylococcus protein A (spa) gene following the conditions mentioned earlier (Nelson et al., 2007). Amplified products were purified, and both strands were sequenced using an ABI sequencer model 3500 (Life Technologies, Marsiling, Singapore) at the sequencing facility of the Institute of Life Sciences (Bhubaneswar, India). The nucleotide sequences were aligned using Mega 5.2 software. Repeat succession in the polymorphic X-region assigned the spa-types, and accordingly the MST was generated using Bionumerics 7 software (Applied Maths, Belgium) using gap creation cost 250%, gap extension cost 50%, duplicate production cost 25%, duplicate expansion cost 25%, and maximum duplication three repeats.

## Statistical Analysis

fmicb-10-02763 December 2, 2019 Time: 13:50 # 4

We performed principal coordinates analysis (PCoA) and discriminant analysis (DA) using PAST program v2.17 for the antibiotic resistance genes and virulence genes in MRSA and MSSA isolates with regard to sources of isolation (Hammer et al., 2001). We carried out the DA using default values to confirm the hypothesis of whether MRSA and MSSA isolates are different.

## RESULTS

## Hexaplex PCR

All the isolates of S. aureus were positive for 16S rRNA and S. aureus-specific nuc genes. Hexaplex PCR discriminates between MSSA and MRSA isolates. Thirty-one of 109 (29.4%) methicillin-resistant strains were positive for the mecA gene, and 77 (70.6%) methicillin sensitive isolates were negative for the mecA gene. One of the methicillin-resistant strains of S. aureus was negative for the mecA gene. Among 109 isolates, 43 (39.4%) isolates comprising 23 of the 77 (29.9%) MSSA and 20 of the 31 (64.5%) MRSA isolates were positive for pvl gene. Of the 31 MRSA isolates, two (6.5%) strains were positive for the czrC gene and four (12.9%) isolates were positive for qacA/B gene, and remaining isolates were negative for both czrC and qacA/B genes (data not shown).

## Coagulase Serotyping

Serotyping classified S. aureus isolates into I–VIII serotypes by using coagulase typing scheme. Twelve of the 109 (11%) strains belong to serotype I, 11 (10%) to serotype II, nine (8%) to serotype III, 14 (12.8%) to serotype IV, 12 (11%) to serotype V, 19 (17.4%) to serotype VI, 20 (18.3%) to serotype VII, and 12 (11%) to serotype VIII, respectively. Nine of 31 (29%) MRSA belong to serotype VI and 17 of 78 (21.8%) and MSSA isolates belong to serotype VII (**Table 1**). Nine of the 24 (37.5%) isolates from wound infection belong to serotype VI and 16 of 64 (25%) isolates from eye infection belonged to serotype VII.

## Antibiotic Resistance Genes

One hundred two of the 109 S. aureus isolates were multidrug resistant showing resistance to two or more antibiotics. All the strains were susceptible to vancomycin when tested by broth microdilution assay. Thirty-one isolates of S. aureus were resistant to oxacillin and carried the mecA gene; however, one isolate of S. aureus resistant to oxacillin was negative by PCR for the mecA gene. The remaining 77 isolates were sensitive to oxacillin and negative by PCR for the mecA gene (**Table 1**).

Ninety-five isolates of S. aureus resistant to chloramphenicol carried cat: pC221 gene; however, 86 isolates carried cat: pC223 and 37 isolates carried cat: pC194 gene, respectively. Twenty isolates carried all the three genes tested; however, 83 isolates were positive for cat: pC221 and cat: pC223 and 37 isolates for cat: pC221 and cat: pC194 genes, respectively (**Table 1**). One of the isolates sensitive to chloramphenicol was negative by PCR for all three genes. In contrast, 15 strains of S. aureus susceptible to chloramphenicol were positive for cat: pC221 and 14 for cat: pC223 genes, respectively.

Twenty-nine isolates were phenotypically resistant to tetracycline of which 29 isolates were positive for tetK, 25 for tetL, and 28 for tetM genes. Twenty-five isolates carried all the three genes tested; however, three strains carried tetK and tetM genes and one isolate tetL and tetM genes. In contrast, 76 isolates sensitive to tetracycline were positive for the tetM gene, 66 for tetL, and 29 for tetK genes. Among them, 27 isolates carried all the three genes, six had tetK and tetM, and 39 strains had tetL and tetM genes, respectively. One isolate sensitive to tetracycline was negative by PCR for all three genes tested (**Table 1**).

A total of 54 isolates were resistant to gentamicin of which 45 isolates were positive for aac(6')/aph(2') and aph (3'-III) genes and nine isolates for aph (3'-III) gene only. In contrast, 43 gentamycin sensitive isolates showed positive results for aac(6')/aph(2') and aph (3'-III), seven isolates for aac(6')/aph(2'), and two isolates for aph (3'-III) genes. However, 56 isolates sensitive to gentamicin were negative by PCR for aac(6')/aph(2') and aph (3'-III) genes (**Table 1**).

Of the 91 isolates of S. aureus showing resistance to macrolides carried erythromycin resistance genes. Twenty-eight isolates carried all the erythromycin resistance genes, namely, msrA, ermA, and ermC. Fifty-one isolates were positive for two genes, of which 30 isolates carried msrA and ermC genes, and 21 strains had ermA and ermC genes. Besides, 12 isolates were positive for a single gene of which five isolates carried the ermC gene, and seven isolates had msrA gene. In contrast, two of the 10 erythromycin sensitive isolates carried msrA and ermC genes, four strains possess msrA and ermC genes, and three isolates had the ermC gene. Of the 64 isolates carrying the mphC gene, 22 isolates were phenotypically resistant to clindamycin (**Table 1**). None of the 17 strains showing sensitivity to erythromycin carried any of the erythromycin resistance genes. One of the resistant isolate not carrying any of the erythromycin resistant genes is likely to be mediated by an as-yet-unknown mechanism.

Similarly, 74 isolates were resistant to trimethoprim of which 45 isolates were positive for dfrA, dfrB, and dfrG genes, 27 strains for dfrB and dfrG genes, and one isolate each for dfrB and dfrG genes, respectively. In contrast, 34 isolates sensitive to trimethoprim were also positive for dfrA, dfrB, and dfrG genes; however, one strain was positive for the dfrG gene (**Table 1**).

## D-Test and Macrolide Resistance

Ninety of 109 (89.9%) S. aureus isolates that exhibited erythromycin resistance were evaluated for MLSB resistance phenotype, namely, iMLSB, cMLSB and MSB. Seventy eight of 90 (79.5%) isolates were erythromycin-resistant but clindamycin susceptible were tested for D-test. We found 14 isolates (10 MRSA and four MSSA) showed iMLSB phenotype, and 12 (two MRSA and 10 MSSA) had MSB phenotype. Seven erythromycinresistant isolates comprising six MRSA and one MSSA had cMLSB phenotype. The remaining 45 isolates (14 MRSA and 31 MSSA) did not show any MLSB phenotypes.

Among MRSA and MSSA showing cMLSB resistance phenotype, three of six MRSA isolates possessed the ermA and ermC genes and one each possessed ermC gene, msrA, ermC, mphC genes, and ermC and mphC genes. One MSSA isolate was positive for msrA, ermA, and ermC genes. On the hand, one

Typing of

Staphylococcus aureus

Strains

fmicb-10-02763 December 2, 2019 Time: 13:50 # 5

TABLE 1 | Antibioticresistance patterns and presence of antibiotic resistance genes inStaphylococcus aureusisolates from different parts of India.

#### Phenotypic antibiotic resistancepattern

#### Number of isolates showing presence of gene(s) encoding for


MRSA: methicillin-resistant Staphylococcus aureus; MSSA: methicillin-susceptible Staphylococcus aureus, OX: oxacillin, GEN: gentamicin, ERY: erythromycin, TET: tetracycline; CL: clindamycin; CHL: chloramphenicol; TMP: trimethoprim. Isolates showing phenotypic resistance to given antibiotic(s) are shown in bold. Number in brackets indicate phenotypic sensitive isolates.

fmicb-10-02763 December 2, 2019 Time: 13:50 # 6


TABLE 2 |Resultof D-test obtained with MRSA and MSSA isolates showing presence of erythromycin resistance genes and its correlation with MLSB phenotypes amongStaphylococcus aureus.

S: sensitive; R: resistance;ER: erythromycin; CL: clindamycin.



of the two MRSA isolates showing MSB phenotype had msrA, ermC genes and other strain had msrA, ermC, and mphC genes (**Table 2**). Of the 12 MSSA, six isolates contained msrA and ermC genes, one isolate each contained ermC and ermA genes, respectively, two strains had mphC gene. The remaining isolates did not carry any of the genes tested. Of the 10 MRSA, six isolates with iMLSB phenotype had ermC gene. One isolate each carried msrA, ermA, and ermC genes, ermA, ermC genes, msrA, ermC, and mphC genes, respectively. The remaining one isolate did not possess any of the resistance genes. Of the four MSSA isolates that showed iMLSB phenotype, three strains were positive for msrA, ermC genes, and one isolate was positive for ermC gene (**Table 2**).

Of the 109 S. aureus isolates tested for the presence of MLSB resistance genes, 102 isolates carried one or more erm genes. Three strains carried all the erythromycin resistance genes, namely, msrA, ermA, and ermC. Fifty-one isolates were positive for two genes, of which 46 isolates carried msrA and ermC genes, and five had ermA and ermC genes. Besides, 37 isolates were positive for a single gene of which 34 isolates carried the ermC gene, two isolates had ermA gene, and three isolates had the msrA gene (**Table 2**). In contrast, four of the 13 erythromycinsensitive isolates carried msrA and ermC genes. One strain each had the ermC gene and msrA gene. The remaining isolates did not carry any resistance genes. Twelve of the 21 mphC genepositive isolates showed phenotypic resistant to clindamycin. The remaining nine isolates were sensitive to clindamycin (**Table 2**). Eight erythromycin-resistant strains did not carry any of the erythromycin-resistant genes is likely to be mediated by an asyet-unknown mechanism.

## SCCmec Typing

The presence of the mec complex and ccr complex classified S. aureus strains into different SCCmec types. Thirty-one MRSA isolates showed four known SCCmec types of which 13 (40.6%) belong to type V, nine (28.1) belong to type UT6, seven (21.9%) belong to type IV, and two (6.3%) belong to type III (**Table 3**). One isolate showing phenotypic resistance to methicillin but negative for mecA gene carried C1 type of ccr complex but lack mec complex. Of the 32 methicillin-sensitive isolates lacking the mec complex, 14 isolates carried ccrA1B1, one strain possesses ccrA4B4, and 17 isolates had ccrA3B3 and ccrA4B4 type of ccr complex, respectively (**Table 3**).

## Toxin Gene Profiles

Of the 109 isolates, 34 (31.2%) isolates harbored sea gene, 14 (12.8%) isolates sec gene, 93 (85.3%) isolates sei gene, 76 (69.7%) cna gene, 101 (92.6%) isolates hla gene, 107 (98%) isolates hlg gene, and 84 (77%) isolates carried efb gene, respectively. All the isolates, except one isolate, was positive for the hlg, and carried multiple virulence genes (**Table 4**).

Ninety-one isolates comprising 26 MRSA and 65 MSSA were positive for both icaA and icaD genes, but five strains containing three MRSA and two MSSA were negative for both icaA and icaD genes. Two of the three MRSA isolates were positive for icaA gene, and another strain was positive for icaD gene. Similarly, nine of the 10 MSSA isolates were positive for icaD gene and one isolate for icaA gene, respectively (**Table 4**).

Also, a total of 25 toxin genes combinations was obtained with 109 strains belonging to 77 PFGE patterns, 32 sequence types (STs), 46 spa-types, and five agr-types. Twenty-three isolates belonging to five MRSA and 18 MSSA showed a toxin pattern comprising sei-cna-hla-hlg-efb genes. On the other hand, five MRSA and two MSSA showed another virulence pattern composed of sea-sec-sei-cna-hla-hlg-efb genes. The remaining isolates showed 23 different virulence gene patterns (**Table 4**).

## Agr-Typing

Of 109 S. aureus strains, 40 (36.7%) isolates belong to agr-I, 31 (28.4%) isolates to agr-III, 18 (16.5%) to agr-II, and seven (6.4%) belong to agr-IV; however, 13 (11.9%) isolates were not typeable by the method employed (**Table 4**). Of the 32 MRSA isolates, 20 (62.5%) belong to agr-I, five (15.6%) to agr-II, three (9.4%) to agr-III, and remaining isolates were untypeable. On the other hand, 28 of 77 (36.4%) MSSA isolates belong to agr-III, 20 (25.9%) to agr-I, 13 (16.9%) to agr-II, seven (9%) to agr-IV, and nine (11.7%) isolates were untypeable. There was a good correlation between virulence patterns and specific molecular types (**Table 4**). The sea-sei-cna-hla-hlg-efb was the dominant virulence pattern shown by MRSA belonged to SCCmec type UT6, and agr type I, followed by sei-hla-hlg-efb and sei-cna-hla-hlg-efb pattern showed

TABLE 4 | Source, clonal complex, sequence-, spa-, SSCmec-, and agr-types and virulence profiles of S. aureus isolated from different parts of India.


(Continued)

#### TABLE 4 | Continued

fmicb-10-02763 December 2, 2019 Time: 13:50 # 9


(Continued)

#### TABLE 4 | Continued

fmicb-10-02763 December 2, 2019 Time: 13:50 # 10


MRSA: methicillin-resistant Staphylococcus aureus; MSSA: methicillin-sensitive Staphylococcus aureus; CC: clonal complex; ST: sequence type; SSC: Staphylococcal cassette chromosome; agr: accessory gene regulator; Unk: unknown; UT: untypeable; NT: non-typeable; <sup>∗</sup> Isolates with ccrAIB1 complex but lack mec complex; pvl: Panton-valentine leucocidin; icaA: intracellular adhesion gene A; icaD: intracellular adhesion gene D; sea: staphylococcal enterotoxin A; sec: staphylococcal enterotoxin C; sei: staphylococcal enterotoxin I; cna: collagen adhesion; hlyA: hemolysin A; hlyG: hemolysin G; efb: extracellular fibronectin binding protein.

by MSSA isolates belonged to agr type I and III, respectively (**Supplementary Table S1**).

## Spa-Typing

Analysis of the aligned sequence of the polymorphic X region of spa gene using the spa-typing plug-in tool of Bionumerics 7 software showed 46 spa-types (**Figure 1**). MST analysis classified the strains into six major clusters, seven minor clusters, and 30 singletons. We designated cluster as a minor cluster that contained less than five but more than two strains. Of the 109 S. aureus isolates, 11 (10%) isolates belong to spa-type t442, 10 (9%) to t037, nine (8.2%) to t2663, eight (7.3%) to t657, six (5.5%) to t127, five (4.5%) isolates each to t345 and t3841, and four isolates each belong to t021, t091, and t1309. In addition, four (3.6%) isolates belong to t1309, three (2.7%) isolates belong to t948, and two (1.8%) each belong to t148, t386, t012, t159, and

t272, respectively. Moreover, one isolate each of 30 strains belong to single spa-types, namely, t15579, t8179, t14912, t010, t852, t005, t310, t309, t1328, t302, t1149, t10579, t007, t14911, t2952, t693, t2526, t8078, t5562, t448, t4104, t177, t098, t084, t2505, t3204, t1839, t064, t12406, and t4285 (**Figure 1**). Whereas 10 of 32 (31.2%) MRSA isolates belong to t037, 11 of 77 (14.3%) MSSA isolates belong to t442. S. aureus strain ATCC 25923 showed spa-type t948 along with three test isolates. We found two novel spa-types, namely, t14911 and t14912 among S. aureus strains after submission of nucleotide sequences to the Ridom spa server. Spa-type t14912 showed a close association with major spa-type t442, but 14911 spa-type was diverse and unrelated. One of the isolates was not assigned any spa-type (**Figure 1**).

## Multi-Locus Sequence Typing (MLST)

Multi-locus sequence typing of 109 S. aureus isolates showed 32 STs, eight CCs, and 12 singletons (**Figure 2**). The major ST comprised of ST1 (12.8%), ST5 (11.9%), ST772 (11%) followed by ST239 (9.2%), ST672 (8.3%), and ST2233 (8.3%). Also, we found two new allelic profiles designated as unknown not reported earlier among S. aureus strains (**Supplementary Table S2**). Of the eight CCs, CC5 contained 14 isolates, CC22 had eight isolates, CC30 had six isolates, CC121 had five isolates, CC239 had 11 isolates, CC772 had 26 isolates, CC813 had two isolates, and CC2884 contained four isolates, respectively. Of the major CCs, CC30 contained five STs, namely, ST30, ST503, ST714, ST938, and ST1482, CC121 contained four STs, namely, ST120, ST121, ST1964, ST2160, and CC772 had three STs, namely, ST772, ST1, and new unknown ST (**Figure 2**). Seven of the 32 (21.8%) of MRSA strains belong to ST239, spa-type t037, and SCCmec type UT6. However, 14 (18.2%) of MSSA strains possessing ST1 belong to different spa-types, namely, t127, t948, t177, t693, t098, and t386, of which few strains carry ccr complex but devoid of mec complex (**Table 4**). However, few isolates from eye infection and wound infection belong to CC239, ST239, and spa-type t037/t657. Reference strain of S. aureus ATCC25923 belonged to ST30 and CC30.

## Pulsed-Field Gel Electrophoresis

SmaI-digested genomic DNA of S. aureus yielded bands classifying the 109 strains into 77 pulsotypes that includes two identical pairs (12 and 19A), three major clusters (1, 3, and 19), 17 minor clusters (14, 15, 17, 19, 20, 22, 24, 25, 28, 32, 57, 58, 63, 67, 69, 71, and 73), and 56 singletons. Four isolates were untypeable by the method employed. We found a total of 24 PFGE patterns among 32 MRSA isolates, of which one isolate was untypeable. Similarly, 77 MSSA isolates showed 53 PFGE patterns, of which three MSSA isolates were untypeable (**Figure 3**). MSSA isolates belonging to the major pulsotype 19 contained seven subtypes 19A, 19B, 19C, 19D, 19E, 19F, and 19G. These isolates were mostly from ocular infection and belong to ST1, agr type III, except one subtype 19G which belongs to ST6 and agr type I. S. aureus strain ATCC 25923 showed pulsotype 14. A dendrogram was generated using Bionumerics 7 software and percentage similarity with a cut-off of 80% and dice coefficients.

## Statistical Analysis

Principal coordinates analysis segregates MRSA and MSSA isolates, except for few isolates with 25.75% of explained variance for antibiotic resistance genes (**Figure 4A**) and 26% for virulence genes (**Figure 5A**). We used axis one for the highest percentage

of representation. DA graph showed that MRSA isolates grouped within more positive values, whereas MSSA isolates grouped within negative values for both antibiotic resistance genes and virulence genes (**Figures 4B**, **5B**). Predominant biomarkers were determined by calculating the coefficient of discriminant function and considered when the value was equal to 0.5 or >0.5. For antibiotic resistance genes, MRSA isolates are discriminating in the biomarker of resistance to ermA (0.8407), mphC (2.0167), tetK (2.3495), tetL (2.0604), and dfrA (1.3116), whereas the MSSA isolates were discriminating in resistance to aac(6')/aph(2) (−0.351), aph3 (−2.7179), ermC (−0.8473), tetM (−0.522), cat:pC221 (−2.421), cat:pC223 (−6.601), dfrB (−0.443), and dfrG (−0.603). For virulence genes, MRSA isolates are discriminating in the biomarker of resistance to icaA (0.67169), seA (0.68593), seC (2.3245), cnA (0.90744), and hlA (0.54797). On the other hand, MSSA isolates were discriminating in resistance to icaD (−2.1945), seI (−0.58795), and hlG (−1.4999). PCoA and discriminant function of antibiotic resistance and virulence genes of S. aureus isolates with source and place of isolation was heterologous and complex (data not shown).

## DISCUSSION

We used hexaplex PCR for detection of MRSA and MSSA isolates along with the presence of mecA, pvl, czrC, and qacA/B genes. We found a good correlation between oxacillin resistance and the presence of the mecA gene. However, one isolate showing resistance to oxacillin and lack mecA gene indicate the occurrence of different mechanism of methicillin resistance. Twenty of 31 MRSA and 23 of 77 MSSA isolates were positive for pvl gene indicating the prevalence of pvl gene among MRSA strains from the wound and eye infections. This finding is in contrast to those who did not find such correlation among clinical isolates (Shittu et al., 2011); therefore, it cannot be used as a reliable marker for MRSA. The presence of czrC and qacA/B genes among the number of MRSA isolates indicates their possible association with the mecA gene; however, further investigation is required to authenticate these findings.

Coagulase gene typing has been used to characterize S. aureus strains. Hwang and Kim (2007) showed the presence of coagulase


serotype II among 54.4% MRSA and serotype VII among 30.9% MSSA. In contrast, we found serotype VII was present among 22% of MSSA isolates and serotype VI in 28.1% of MRSA isolates. These observations thus suggest that there is a difference in the presence of serotypes with regard to MRSA and MSSA.

Genetic determinants study among S. aureus showed a good correlation between resistance to aminoglycosides, chloramphenicol, clindamycin, erythromycin, trimethoprim, and tetracycline, and the presence of corresponding resistance genes. In this study, we found 85.3% strains showing resistance to chloramphenicol carried the pC221 gene; however, some of these strains also carried either pC223 or pC194 or both genes. Although one of 109 strains sensitive to chloramphenicol did not carry any of these genes, 13.8% strains showing sensitivity to chloramphenicol carried either pC221 or pC223 genes. These observations thus suggest that chloramphenicol sensitive strains carrying antibiotic resistance genes can develop resistance against this drug on exposure.

The aminoglycoside-modifying enzyme, encoded by aac (6')-aph(2") gene, is responsible for resistance against aminoglycosides (Vanhoof et al., 1994). Besides, two other genes encoding for aph(3.III) and ant(4, IV) are accountable for aminoglycoside resistance, but their frequency is less compared to aac(6')-aph(2") among staphylococci (Busch-SØRensen et al., 1996). In this study, we found 41.3% S. aureus possesses both aac(6')-aph(2") and aph (3, III) genes and 8.3% contained aph (3, III) gene and showed phenotypic resistance to gentamycin. These findings thus suggest that there are strains which harbor aminoglycoside resistance genes other than aac(6')-aph(2") and few strains had aph (3, III) only. At least 47.7% strains of S. aureus that were sensitive to aminoglycosides contain either aph (3, III) or aac(6')-aph(2") or both; however, three strains susceptible to gentamycin lack resistance genes. These findings are in contrast to those workers who reported that all aminoglycoside-resistant strains carried aac(6')-aph(2") (Price et al., 1981; Lovering et al., 1988; Dornbusch et al., 1990; Vanhoof et al., 1994; Martineau et al., 2000). The presence of aminoglycoside resistance gene among gentamycin sensitive isolates of S. aureus indicates that there is likely hood development of aminoglycoside resistance among S. aureus upon exposure to these drugs.

Similarly, 83.4% strains of S. aureus resistant to erythromycin harbored any of the four genes, namely, ermA, ermB, ermC, and msrA; however, an strain sensitive to erythromycin did not

isolates and positive values to MRSA isolates.

carry any of the genes. Previously, it was reported that the ermA gene is dominant among erythromycin resistance genes in S. aureus (Kaur and Chate, 2015). In contrast, we found the presence of the ermC gene in 83.4% strains compared to 49% of ermA gene. Kaur and Chate (2015) reported that majority of MRSA strains showed constitutive MLSB (cMLSB) resistance; however, two isolates had inducible MLSB (iMLSB) phenotype. In this study, 64.5% MRSA and 37.1% MSSA strains belong to iMLSB phenotype; however, 35.4% of MRSA and 43.5% of MSSA strains belong to cMLSB phenotype. This difference could be due to less number of MRSA isolates used in the study, and MSSA isolates were multidrug resistant. Seventeen strains showing sensitivity to erythromycin harbored one of the resistance genes, and one of the strains resistant to erythromycin did not possess any of the resistance genes to indicate that these strains are likely to develop resistance and mediated by an unknown mechanism.

About trimethoprim resistance, 67.8% strains harbored any of the three genes, namely, dfrA, dfrB, and dfrG. The remaining strains showing sensitivity to trimethoprim also carried all or one of the three genes. In this study, 73 of 74 trimethoprim resistance strains possess dfrG and dfrB genes; 45 strains carried the dfrA gene. These findings are in contrast to those who reported the presence of the dfrG gene in 92% strains, dfrA in 7% strains, and one strain carried a dfrB among trimethoprim resistance strains in a travel-associated skin and soft tissue infection study in Europe (Nurjadi et al., 2015).

Like other antibiotic resistance, 26.6% phenotypic resistance strains carried one or all the three tetracycline resistance genes, namely, tetK, tetL, and tetM. One of the strains sensitive to tetracycline was devoid of carrying any genes. However, the majority (69.7%) strain showing sensitivity to tetracycline carried one or all three resistance genes indicating that these isolates could develop resistance after exposure to an antibiotic. From this study, it is clear that erythromycin and gentamicin were least active; however, vancomycin and clindamycin were the most effective drugs. These results corroborate the finding of Pai et al. (2010), who also reported that vancomycin and clindamycin are the most effective drugs.

SCCmec type V was predominant type among MRSA strains followed by SCCmec type UT6, IV, and III, respectively. This finding is similar to Nadig et al. (2012), who also reported the prevalence of SCCmec type V among isolates from eye infections. To our knowledge, we are the first to inform of the presence of SCCmec type UT6 among S. aureus from India. The combination of SCCmec IV, V, and pvl gene was reported as the genetic markers for a community-associated MRSA (Bhutia et al., 2015). Similarly, our study showed the presence of SCCmec V (40.6%), IV (21.9%), and pvl (64.5%); therefore it can be used as a marker for hospital-associated infections. However, new UT6 SCCmec type is emerging in India. Many untypeable strains carried ccr complex but no mec complex. This observation thus suggests the ability of such strains to acquire mec complex and became a known or unknown SCCmec type.

A total of 25 unique toxin combination was found among S. aureus strains, of which at least one toxin gene was present in a given strain. Sotto et al. (2008) reported the presence of sei and sea genes in S. aureus isolated from diabetic foot ulcer. Similarly, we found the presence of sei and sea genes in both MRSA and MSSA strains. Although we noted the high percentage of hlg (98%) and hla (92.6%) among in S. aureus comprising both MSSA and MRSA, other workers reported the presence of these genes in mupirocin resistant in MRSA isolates in China (Liu et al., 2012). Moreover, the distribution of virulence genes with regards to source and place of isolation was complex. Gowrishankar et al. (2016) reported the isolation of 84% MRSA strains carrying icaADBC genes from patients with pharyngitis. Also, in this

values to MRSA isolates.

study, 81.3% MRSA and 84.4% MSSA carrying icaA/icaD genes were isolated from the eye and wound infections (**Supplementary Table S3**). Absence of icaA/icaD genes in S. aureus strains was similar to those of the previous report (Agarwal and Jain, 2013).

Several molecular genotyping tools are used to trace the origin of the strain, and distribution of CC with regard to methicillinresistant, methicillin-sensitive, sources and place of isolation. We determined the population structure of S. aureus isolated from ocular and wound infections from different parts of India using MLST, agr-typing, spa-typing, and PFGE.

Multi-locus sequence typing analysis showed the presence of six major ST(s) comprising ST1, ST5, ST772, ST239, ST672, and ST2233, respectively. While ST239-MRSA-UT6 was the typical type among MRSA isolates from wound infection, ST772-MRSA-V were from eye infections (Nadig et al., 2012). Similarly, ST772- SCCmec-V were reported slowly replacing multidrug resistant ST239-SCCmec-III in Asian studies (D'Souza et al., 2010). This finding is in contrast to Suzuki et al. (2012), who reported the presence of ST5 and ST764 among MRSA strains from the infected eye and healthy conjunctiva sacs. Also, Mohammadi et al. (2014) showed emergence of SCCmec-III with variable antimicrobial resistance profiles in Iran. We found ST772-MRSA-V with spa-type t345 and t657 belonging to dominant CC772 among wound infection isolates. Besides, we reported two new spa-types among S. aureus strains from India.

There were eight CCs, namely, CC30, CC121, CC772, CC813, CC239, CC28841, CC22, and CC5 present among S. aureus represents different PFGE clusters. CC30 and CC121 comprising different STs were almost equally distributed among MRSA and MSSA isolates. Whereas CC772-ST772 was dominant among MRSA, CC772-ST1 was prevalent among MSSA isolates. Similarly, CC239-ST239 and CC22-ST22 were prevalent among MRSA isolates and CC5-ST5, CC813-ST813, and CC28841- ST28841 were more commonly found in MSSA isolates. The prevalent CC among Varanasi isolates (mostly wound infection) were CC772 followed by CC239 besides the presence of CC30, and CC121. However, isolates from wound infection from Delhi showed varied results. Whereas AIIMS isolates showed CC CC30, CC22, and CC813, UCMS isolates showed the presence of CC239 and CC22 CCs. Interestingly isolates from Hyderabad (eye infection) had CC239 but isolates from Bhubaneswar (eye infection) showed the presence of CC772, CC5, CC2884, and CC30.

Mobasherizadeh et al. (2019) reported the prevalence of CC5 and CC30 and other CCs among MRSA isolates isolated from nasal carriage in Iranian hospitals. Similarly, CC8, CC121, CC1, CC45, and CC5 were reported in MRSA isolates from Malaysia (Ghasemzadeh-Moghaddam et al., 2011). These observations indicate the existence of different CCs in India and Asian countries. MLST and spa-typing was better than PFGE and toxin genotyping a finding unusual from those who reported a good correlation between various typing schemes. Overall, there was diversity in genotypes, antimicrobial resistance, and virulence determinants among MRSA and MSSA strains.

From this study, it is clear that S. aureus strains sensitive to antibiotics but carried antibiotic resistance genes could develop resistance upon exposure to antibiotic(s), and vancomycin and clindamycin were the most effective drugs. ST239-SCCmec UT6/t035 were dominant clones among S. aureus. There was diversity in genotypes, antimicrobial resistance, and virulence determinants among MRSA and MSSA strains, therefore suggests continuous surveillance of multidrug-resistant strains circulating in the community/hospitals in India, to take adequate measures to control the infection.

## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## ETHICS STATEMENT

fmicb-10-02763 December 2, 2019 Time: 13:50 # 16

The studies involving human participants were reviewed and approved by the Institutional Review Board (IRB) of LV Prasad Eye Institute (LEC/08/110/2009) and by the Institute Ethics Sub-Committee (IESC) of All India Institute of Medical Sciences, New Delhi (IESC/T-34/2013), and the data were analyzed anonymously and reported. The patients/participants provided their written informed consent to participate in this study.

## AUTHOR CONTRIBUTIONS

SA, SJ, SS, and DS conceived the experiments. SA, SJ, and SP conducted the experiments. SA, SJ, SP, SS, BD, GN, NS, and DS analyzed the results. KN performed statistical analysis. SA, SJ, and DS wrote the manuscript. All authors reviewed and approved the manuscript.

## REFERENCES


## FUNDING

This study was supported by the Department of Science and Technology (Grant No. SR/SO/HS-117 to DS). This study, in part, was also supported by the fund contributed by the Department of Biotechnology, New Delhi, to the Institute of Life Sciences, Bhubaneswar. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

## ACKNOWLEDGMENTS

SA and SJ are grateful to the Institute of Life Sciences, Bhubaneswar and Department of Science and Technology, New Delhi, respectively for providing Senior Research Fellowship.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02763/full#supplementary-material




**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 Aggarwal, Jena, Panda, Sharma, Dhawan, Nath, Singh, Nayak and Singh. 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.

# Baicalin Inhibits Biofilm Formation and the Quorum-Sensing System by Regulating the MsrA Drug Efflux Pump in Staphylococcus saprophyticus

Jinli Wang1,2, Haihong Jiao<sup>3</sup> , Jinwu Meng1,2, Mingyu Qiao1,2, Hongxu Du1,2, Miao He1,2 , Ke Ming1,2, Jiaguo Liu1,2 \*, Deyun Wang1,2 and Yi Wu1,2

<sup>1</sup> MOE Joint International Research Laboratory of Animal Health and Food Safety, College of Veterinary Medicine, Nanjing Agricultural University, Nanjing, China, <sup>2</sup> Institute of Traditional Chinese Veterinary Medicine, College of Veterinary Medicine, Nanjing Agricultural University, Nanjing, China, <sup>3</sup> Key Laboratory of Tarim Animal Husbandry Science and Technology of Xinjiang Production & Construction Corps, College of Animal Science, Tarim University, Alar, China

## Edited by:

Raffaele Zarrilli, University of Naples Federico II, Italy

#### Reviewed by:

Adline Princy Solomon, SASTRA University, India Mara Di Giulio, Universitá degli Studi "G. d'Annunzio" Chieti e Pescara, Italy Iqbal Ahmad, Aligarh Muslim University, India

> \*Correspondence: Jiaguo Liu liujiaguo@njau.edu.cn

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 23 September 2019 Accepted: 18 November 2019 Published: 10 December 2019

#### Citation:

Wang J, Jiao H, Meng J, Qiao M, Du H, He M, Ming K, Liu J, Wang D and Wu Y (2019) Baicalin Inhibits Biofilm Formation and the Quorum-Sensing System by Regulating the MsrA Drug Efflux Pump in Staphylococcus saprophyticus. Front. Microbiol. 10:2800. doi: 10.3389/fmicb.2019.02800 Staphylococcus saprophyticus (S. saprophyticus) is one of the main pathogens that cause serious infection due to its acquisition of antibiotic resistance. The efflux pump decreases antibiotic abundance, and biofilm compromises the penetration of antibiotics. It has been reported that baicalin is a potential agent to inhibit efflux pumps, biofilm formation, and quorum-sensing systems. The purpose of this study was to investigate whether baicalin can inhibit S. saprophyticus biofilm formation and the quorum-sensing system by inhibiting the MsrA efflux pump. First, the mechanism of baicalin inhibiting efflux was investigated by the ethidium bromide (EtBr) efflux assay, measurement of ATP content, and pyruvate kinase (PK) activities. These results revealed that baicalin significantly reduced the efflux of EtBr, the ATP content, and the activity of PK. Moreover, its role in biofilm formation and the agr system was studied by crystal violet staining, confocal laser scanning microscopy, scanning electron microscopy, and realtime polymerase chain reaction. These results showed that baicalin decreased biofilm formation, inhibited bacterial aggregation, and downregulated mRNA transcription levels of the quorum-sensing system regulators agrA, agrC, RNAIII, and sarA. Correlation analysis indicated that there was a strong positive correlation between the efflux pump and biofilm formation and the agr system. We demonstrate for the first time that baicalin inhibits biofilm formation and the agr quorum-sensing system by inhibiting the efflux pump in S. saprophyticus. Therefore, baicalin is a potential therapeutic agent for S. saprophyticus biofilm-associated infections.

Keywords: baicalin, efflux pump, biofilm, quorum sensing (QS), Staphylococcus saprophyticus

## INTRODUCTION

Staphylococcus saprophyticus, a member of the opportunistic coagulase-negative Staphylococci (CoNS), is a common pathogen of acute uncomplicated urinary tract infection, francolin ophthalmia, and bovine mastitis (Mahato et al., 2017; Mlaga et al., 2017; Wang J. et al., 2019). It has also been reported that S. saprophyticus is a common gastrointestinal flora in pigs and cows and

may thus be transferred to humans by eating these respective foods. The use of antibiotics to treat and prevent bacterial infections has made an unprecedented impact on improving human health. However, multidrug-resistant bacteria have seriously threatened people's health in recent decades. Wang et al. discovered that over 90% of S. saprophyticus isolated from ready-to-eat food displayed multiantibiotic resistance (Wang Y. et al., 2019). Bacteria are resistant to antibiotics due to target mutations, multidrug efflux pumps, drug-inactivating enzymes, biofilm formation, etc. Recently, S. saprophyticus has been shown to form biofilms in umbilical catheters (Martins et al., 2019).

Bacterial biofilms are adherent complex communities of bacteria encased within extracellular polymeric substances (EPSs), and they are considered intrinsically resistant to antibiotic treatment and host defenses (Thurlow et al., 2011; Paharik and Horswill, 2016). It has been estimated that nearly 60% of nosocomial infections in the human body are the result of biofilm formation on medical devices such as indwelling catheters and prostheses (Hou et al., 2012). The MIC of antibiotics toward bacteria biofilm is 1000-fold higher than that of planktonic counterparts (Costerton et al., 1987). Bacterial biofilms may represent an important barrier to the therapy of bacterial infections due to their lower sensitivity to antibiotic treatment. Therefore, the development of biofilm inhibitors based on a completely novel concept is urgently needed.

Recently, several studies have provided evidence to show that genetic inactivation and chemical inhibition of efflux pumps resulted in transcriptional inhibition of biofilm matrix components and a short biofilm formation (Baugh et al., 2014; Van Acker and Coenye, 2016; Sabatini et al., 2017). Biofilm formation is regulated by the quorum-sensing (QS) system. QS is a potential target for the therapy of bacterial biofilm infections. Staphylococcus uses a canonical Gram-positive twocomponent QS system encoded by the accessory gene regulator (agr) locus. Autoinducing peptide (AIP) is the signaling molecule of the agr system. When the cell density increases, the secretion of AIP is upregulated. In most Gram-positive QS bacteria, AIP is processed and exported by ABC transporters (Bassler, 1999). Our previous study indicated that azithromycin-resistant S. saprophyticus (ARSS) was resistant to macrolide antibiotics (Wang J. et al., 2019). The ermA, ermB, ermC, mphC, and msrA genes are the main resistance genes of macrolide antibiotics in S. saprophyticus, and the msrA gene is most prevalent in hospitals (Le Bouter et al., 2011). The MsrA efflux pump encoded by the msrA gene belongs to the ATP-binding cassette (ABC) transporters. We hypothesized that MsrA efflux pump inhibitors influence the agr system and biofilm formation.

Several natural plant products act as efflux pump inhibitors (Stavri et al., 2007). Baicalin (**Figure 1**; Moore et al., 2016), a type of flavonoid derived from the roots of Scutellaria baicalensis Georgi, exerts many biological activities and pharmacological effects, including remarkable antibiofilm, antibacterial (Sass et al., 2019), antiviral (Chen et al., 2018), and immune-enhancing ability (Ge et al., 2012). Our previous study has shown that baicalin possessed synergistic anti-ARSS with azithromycin (Azm; Wang J. et al., 2019). Chen et al. elucidated that a sub-inhibitory concentration of baicalein can prevent biofilm

formation by inhibiting the agr system of S. aureus (Chen et al., 2016). Thus, we hypothesized that baicalin has the potential to be an effective therapeutic strategy against S. saprophyticus biofilm formation and the agr system by inhibiting efflux pumps.

Considering these aspects, this study evaluated whether baicalin could effectively inhibit the efflux pump, biofilm formation, and QS system in ARSS. We also tried to explain the relationship between the efflux pump and biofilm formation and the QS system. To our knowledge, there are no papers regarding the relationship between efflux and biofilm formation and the QS system in S. saprophyticus.

## MATERIALS AND METHODS

## Bacterial Strains and Culture Conditions

The azithromycin-resistant S. saprophyticus (ARSS) strain used in this experiment was isolated in 2016 from francolins suffering from ophthalmia in a francolin farm in Jiangsu Province, China. ARSS was coagulase negative and resistant to azithromycin with an MIC value of 1000 mg/L (Wang J. et al., 2019). S. saprophyticus ATCC 15305 was purchased from the China Center of Industrial Culture Collection (CICC). Strains were routinely cultured on brain heart infusion broth (BHI; Haibo, Qingdao, China) or nutrient broth (NB, Haibo, Qingdao, China) and incubated at 37◦C.

## Determination of Growth Kinetics

For the growth kinetics assay, overnight ARSS cultures were prepared and cocultured with sub-inhibitory concentrations of baicalin or verapamil (an efflux pump inhibitor as a positive control compound). Normal saline containing up to 2% NaHCO<sup>3</sup> was used to prepare a 4000 mg/L baicalin solution. We have previously demonstrated that the MIC value of ARSS to baicalin was 500 mg/L (Wang J. et al., 2019). The sub-inhibitory concentrations of baicalin were used in this experiment, and the final concentrations of baicalin were 250, 125, 62.5, and 31.25 mg/L. The final concentration of verapamil was 250 mg/L. At the same time, the S. saprophyticus control group (SS group, not containing drug) was set. The cell densities based on the optical density at 600 nm in BHI were measured every 2 h from 0 h to 24 h.

## Efflux of Ethidium Bromide

fmicb-10-02800 December 7, 2019 Time: 17:11 # 3

The efflux of ethidium bromide (EtBr) by ARSS was implemented as previously described with some modifications (Smith and Blair, 2014): bacterial strains were grown to an OD600 nm of 0.6. The cells were pelleted by centrifuging at 4000 rpm for 10 min at room temperature, and the precipitation was washed twice with the same volume of phosphate buffer saline (PBS). Then, the OD600 nm of the bacterial suspension was adjusted to 0.3 with PBS containing 1 mM MgCl2. EtBr was added at a final concentration of 7.81 mg/L (1/2 MIC). Cultures were incubated at 37◦C with shaking for 60 min. After centrifugation for 10 min at 4000 rpm, the supernatant was discarded, and the pellet was resuspended in PBS with 1 mM MgCl<sup>2</sup> and 5% glucose to energize the cells. Aliquots of 0.05 ml were put into each well of black, clear-bottom 96-well microplates that contained 0.05 ml baicalin or verapamil at sub-inhibitory concentrations prepared at 2-fold serial dilutions in PBS with 1 mM MgCl<sup>2</sup> and 5% glucose. Verapamil at 250 mg/L was used as a known efflux pump inhibitor, while the S. saprophyticus control (SS) group contained an equal volume of PBS with 1 mM MgCl<sup>2</sup> and 5% glucose. The fluorescence was measured over 60 min at excitation and emission wavelengths of 530 nm and 590 nm, respectively, in a Tecan Infinite 200 Pro (Switzerland).

## Measurement of the ATP Content and Pyruvate Kinase Activity

Overnight cultures were used to inoculate 4 ml of NB containing baicalin at concentrations of 250, 125, 62.5, and 31.25 mg/L and incubated at 37◦C with shaking (180 rpm). The samples from each group were removed at 24 h. Then, the ATP content and pyruvate kinase (PK) activity were determined using an ATP assay kit (Beyotime, China) and a Pyruvate Kinase Assay Kit (Solarbio, Beijing, China) according to the manufacturer's instructions.

## Measuring Gene Transcription Levels via RT-PCR

Overnight cultures were used to inoculate 4 ml of NB containing baicalin at concentrations of 250, 125, 62.5, and 31.25 mg/L and incubated at 37◦C with shaking (180 rpm). The samples used to detect the transcription level of the msrA gene from each group were removed at 24 h in the efflux mechanism experiment. Overnight cultures of ARSS were diluted into fresh BHI. Then, 0.5 ml of diluted cultures was applied to each well of sterile 96-well flat-bottom tissue culture plates that contained an equal volume of sub-inhibitory concentrations of baicalin (final concentrations of 250, 125, 62.5, and 31.25 mg/L) or verapamil (final concentration of 250 mg/L). The S. saprophyticus control (SS) group contained an equal volume of fresh BHI. The plates were incubated at 37◦C without shaking. The samples from each group were removed at 24 h and 48 h postinoculation for RNA extraction. These samples were used to detect the transcription levels of msrA, agrA, agrC, RNAIII, and sarA in biofilms. Then, total RNA was extracted by using a Bacteria RNA kit (Vazyme, Nanjing, China) as recommended in the manufacturer's instructions. The value of A260/A280 was confirmed to be between 1.8 and 2.1. The reverse transcription assay took place in a PCR instrument (2720 Thermal Cycler PCR instrument, Applied Biosystems, America) by using a HiScript II 1st Strand cDNA Synthesis Kit (Vazyme, Nanjing, China). Reverse transcription was carried out at 50◦C for 15 min and 85◦C for 5 s. Real-time PCR was reacted in a PCR instrument (StepOnePlusTM Real Time PCR instrument, Applied Biosystems, United States) using ChamQTM SYBR <sup>R</sup> qPCR Master Mix according to the manufacturer's instructions. The cycling parameters were as follows: holding stage of 95◦C for 3 min; 40 cycles at the cycling stage of 95◦C for 10 s and 60◦C for 60 s; one melt curve stage of 95◦C for 15 s, then 60◦C for 60 s, and 95◦C for 15 s. The 16S rRNA of ARSS was chosen as the housekeeping control gene. The sequences of primers used in this experiment are listed in **Table 1**.

## Semiquantitative Determination of Biofilm Formation

Semiquantitative biofilm assays were conducted as described previously with some modifications (Liu et al., 2017). Briefly, overnight cultures of ARSS or ATCC 15305 strains were diluted into fresh BHI. Then, 0.1 ml of diluted cultures was applied to each well of sterile 96-well plates that contained an equal volume of sub-inhibitory concentrations of baicalin (final concentrations of 250, 125, 62.5, and 31.25 mg/L) or verapamil (final concentration of 250 mg/L). The S. saprophyticus control (SS) group and the ATCC group contained equal volumes of fresh BHI. The negative control group contained only an equal volume of fresh BHI. Then, the plates were incubated at 37◦C for 24 and 48 h without shaking. Culture supernatants were gently removed, and wells were washed with PBS twice to remove the floating cells, followed by fixation with 2.5% glutaraldehyde for 1.5 h and finally air-dried. The adherent bacteria in the wells were stained for 20 min with 1% (wt/vol) crystal violet and then rinsed thoroughly with PBS until the negative control wells (without biofilms) appeared colorless. To quantify biofilm formation, 0.2 ml of 33% glacial acetic acid was added to the wells of plates that were stained with crystal violet. Biofilm formation was measured with a ThermoTM MultiskanTM FC



enzyme-labeled instrument at 570 nm. The results are presented as the values in the experimental groups minus the values in the negative control group.

## CLSM Protocol Studies for Biofilm Inhibition

Biofilms were stained with the fluorescent LIVE/DEAD BacLightTM bacterial viability kit L7012 (Molecular Probes, Invitrogen) according to the manufacturer's instructions. Bacteria with damaged or intact cell membranes stain fluorescent green when SYTO 9 was used alone. Briefly, fluorescent green (SYTO 9) was used according to the product information manual supplied by the manufacturer. Afterward, 1 ml of overnight cultures that were diluted by BHI was used to grow biofilms on cover slides in 6-well microtiter plates containing 1 ml of baicalin at concentrations of 62.5, 125, 250, and 500 mg/L, verapamil (VP group) at a concentration of 500 mg/L or an equal volume of BHI (SS group) for 24 or 48 h at 37◦C without shaking. Then, the supernatant was removed, and the adherent organisms were stained with dye at room temperature in the dark for 15 min. Finally, these carriers were rinsed three times with 0.85% NaCl and detected under CLSM (Nikon A1).

## Scanning Electron Microscope Studies for Biofilm Morphology

Biofilm morphology was observed using an S3400N scanning electron microscope (SEM; Hitachi, Japan). In this experiment, 0.5 ml of overnight cultures that were diluted with fresh BHI were used to grow biofilms on round cover slides in 24-well microtiter plates containing 0.5 ml of baicalin at a concentration of 500 mg/L or an equal volume of BHI (SS group) for 48 h at 37◦C without shaking. Each biofilm slice was washed with PBS, fixed in 2.5% glutaraldehyde overnight at 4◦C, and then rinsed thoroughly three times with fresh PBS (pH 7.4). The slices were passed through an ethanol gradient (e.g., 50, 70, 80, and 90%) for 15 min, passed through 100% ethanol (three times for 10 min) for dehydration, dried, and then coated with gold (Fujimura et al., 2008).

## Baicalin and Azithromycin Combined Efficacy in vitro on Mature Biofilm

In order to compare the combined effects of baicalin and Azm against ARSS biofilms, 48 h biofilms were prepared in 24-well plates. Briefly, overnight cultures of ARSS were diluted into fresh BHI. Then, 1 ml of diluted cultures was applied to each well of sterile 96-well flat-bottom tissue culture plates. The plates were incubated at 37◦C for 48 h without shaking. Then, the culture supernatants were gently removed. The plates were gently washed twice with PBS. Baicalin and verapamil in the presence or absence of Azm were added to each well in BHI. The total volume per well was 1 ml. The final concentrations of verapamil and Azm were 250 and 7.8 mg/L, respectively. The final concentrations of baicalin were 250, 125, 62.5, and 31.25 mg/L. Finally, the plates with biofilms were cultured for 24 h at 37◦C. After the treatment was completed, planktonic bacteria were discarded by washing with PBS, and clumps were disrupted by sonicating. The bacterial CFU counts in biofilm were performed by plating serial dilutions on NB agar. This experiment was conducted three times in parallel.

## Correlation Analysis

Correlations among the relative expression of the efflux gene and the ability of biofilm formation and the relative expression of agr system-associated genes were determined using Pearson's correlation coefficient.

## Statistical Analysis

Relative gene expression data were analyzed by the 2−11CT method. Duncan's Multiple Range Test was used to determine the differences among groups by the SPSS Software Package version 20.0 (IBM, Armonk, NY, United States). The results were expressed as the mean ± standard deviation (SD). Differences were considered statistically significant at p < 0.05.

## RESULTS

## Sub-Inhibitory Concentrations of Baicalin and Verapamil Did Not Influence Bacterial Growth in vitro

Our previous investigation indicated that the MICs of baicalin and azithromycin against ARSS were 500 and 1000 mg/L, respectively (Wang J. et al., 2019). The results of the growth kinetics assay showed that the growth kinetics of the ARSS under sub-inhibitory concentrations of baicalin and verapamil did not exhibit any difference from the control group when the bacteria were cultured in BHI, indicating that in a nutrient-rich environment, the sub-inhibitory concentrations of baicalin and verapamil may not affect S. saprophyticus growth (**Figure 2**).

## Baicalin Inhibited EtBr Efflux

The inhibitory effect of increasing concentrations of baicalin and the known efflux pump inhibitor verapamil (VP) on EtBr

efflux is shown in **Figure 3**. A concentration-dependent effect was observed for baicalin. In the presence of verapamil or 250, 125, or 62.5 mg/L baicalin, the rate of efflux reduction was significantly higher than that of the SS group, as presented in **Figure 3A** (p < 0.05). Importantly, the rates of efflux reduction were not significantly different from that of the VP group when 250 or 125 mg/L baicalin was added. MsrA is an ATP-dependent efflux pump, suggesting that ATP is essential for MsrA efflux function (Hu et al., 2015). PK is associated with ATP production. Thus, the ATP content and PK activity were detected. Baicalin markedly decreased the content of ATP and the activity of PK (**Figures 3B,C**) (p < 0.05). The MsrA efflux pump was encoded by the msrA gene. Baicalin significantly reduced the relative expression of the msrA gene at 250, 125, and 62.5 mg/L, as presented in **Figure 3D** (p < 0.05). However, baicalin at 31.25 mg/L did not exert beneficial effects.

## Biofilm Formation Was Impeded by Baicalin and Verapamil

To examine whether baicalin or verapamil inhibited ARSS biofilm formation, the ability of biofilm formation was investigated. These results reflected that the A<sup>570</sup> values of ARSS biofilm in the SS group were significantly higher than those of the ATCC group for 24 and 48 h. We found that baicalin reduced biomass in a dose-dependent manner in spite of whether the cultures were grown for 24 or 48 h, while 31.25 mg/L baicalin had almost no effect at 24 h in ARSS (**Figures 4A,B**). Additionally, verapamil significantly decreased biofilm formation compared with that of the SS group (not containing drug). Interestingly, the A<sup>570</sup> values were not different when verapamil or 250 mg/L baicalin was added for 24 or 48 h. Moreover, fluorescence microscopy revealed similar results (**Figure 5**). From **Figures 5**, **6**, we find that many bacteria accumulated together and were enrolled in large amounts of extracellular matrix in the SS group. However, when baicalin was added, only a few bacteria adhered to the glass side and did not form mature biofilms.

## Combined Anti-ARSS Biofilm Efficacy of Baicalin Plus Azm

To determine whether baicalin was effective at disrupting ARSS biofilms combined with Azm, 48-h biofilms were treated with

significantly (p < 0.05).

baicalin and verapamil in the presence or absence of Azm. The results showed that Azm alone did not eradicate ARSS in the formed biofilms. However, combination treatment with verapamil + Azm or baicalin + Azm markedly reduced the counts of bacteria on plates in a concentration-dependent manner (p < 0.05), while 31.25 mg/L baicalin had almost no effect. Furthermore, there was no significant difference between the 250 and 125 mg/L baicalin + Azm groups and the verapamil + Azm group (**Figure 7**).

## Baicalin Inhibited the Relative Expression Level of the msrA Efflux Gene in Biofilm

We measured the expression of the msrA gene in ARSS biofilm bacteria at 24 and 48 h by RT-PCR to detect the relationship between efflux and biofilm formation. As shown in **Figure 8**, the relative expression of the msrA gene was significantly decreased in a dose-dependent manner when baicalin was added compared with relative expression in the SS group (p < 0.05). However, the transcript level of the msrA gene was not significantly affected by 31.25 mg/L baicalin at 24 h.

## In the ARSS Background, Baicalin and Verapamil Lowered the Activity of the agr System

In Staphylococcus, the two-component QS system agr is an important contributor to the establishment of biofilm and infection by this bacterium (Boles and Horswill, 2008). In order to detect the transcription levels of specific RNA in different samples, the levels of agr system-associated gene transcripts were measured using RT-PCR. After the bacterial biofilms were treated with either 250, 125, or 62.5 mg/L baicalin or 250 mg/L verapamil for 24 and 48 h, the transcription levels of agrA, agrC, sarA, and RNAIII genes significantly decreased in a dose-independent manner (p < 0.05). Treating the biofilms with 31.25 mg/L baicalin (24 or 48 h) did not change agr system-associated gene expression levels. More importantly, no significant difference in agr systemassociated gene expression levels resulted from treatment with 250 mg/L baicalin and verapamil (**Figure 9**).

## Pearson's Correlation Coefficients Between the Relative Expression of the Efflux Gene and Biofilm Formation and the Transcript Levels of agr System-Associated Genes

The correlation coefficients among the measured indices of msrA efflux gene relative expression and biofilm formation and transcript levels of agr-associated genes are presented in **Table 2**. As shown in **Table 2**, the relative expression of the msrA efflux gene was positively correlated with the A<sup>570</sup> values of biofilm (p < 0.01) and the relative expression of agr system-associated genes (p < 0.01). In particular, the relative expression of agrA and agrC was positively correlated with the relative expression of sarA and RNAIII.

## DISCUSSION

Efflux pumps, which are membrane-bound proteinaceous transporters, are in charge of transporting xenobiotics or chemotherapeutic agents that are otherwise harmful for bacterial survival. Many researchers have reported that multidrugresistant (MDR) efflux pumps play a prominent role in the biology of bacteria and have roles in drug resistance, cell division, pathogenicity, and, as recently described, the formation of biofilms (Kvist et al., 2008; Buckley et al., 2006). MsrA is an MDR efflux pump of the ABC family. It was established to expel various macrolide antibiotics in Staphylococcus. The MsrA efflux

FIGURE 5 | Influence of baicalin on biofilm formation assessed by fluorescence microscopy (200×). Static biofilms after exposure to baicalin or verapamil for 24 h (A–F) and 48 h (G–L) were stained with SYTO 9. ARSS within biofilms on glass carriers display green fluorescence. Control group (A,G); verapamil group (B,H); 250 mg/L baicalin group (C,I); 125 mg/L baicalin group (D,J); 62.5 mg/L baicalin group (E,K); 31.25 mg/L baicalin group (F,L).

pump has been well established as a model system for studying efflux inhibition in Staphylococcus (Chan et al., 2013).

In the present study, baicalin was assessed for its MsrA efflux pump inhibitory activity in ARSS, which expresses the msrA efflux gene and is resistant to Azm. Baicalin could inhibit 33% of the EtBr extrusion from ARSS. Importantly, there was no significant difference in the rate of efflux reduction resulting from treatment with either the positive control compound verapamil

or 250 mg/L baicalin (**Figure 3A**). These results suggested that baicalin could inhibit the MsrA efflux pump and has the potential to be an efflux pump inhibitor. MsrA is an ATP-dependent efflux pump, indicating that the inhibitory activities of baicalin against the ARSS efflux pump were more correlated with the ATP-dependent process in bacteria (Hu et al., 2015). In this study, our results indicated that baicalin significantly decreased the content of ATP compared with the SS group. PK, a finalstage enzyme in glycolysis, converts phosphoenolpyruvate (PEP) to pyruvate and ATP (Zoraghi et al., 2010). PK is critical for bacterial survival. Chan et al. proved that diosmetin could statistically reverse the resistance of MRSA to erythromycin possibly by inhibiting the MsrA efflux pump in vitro and MRSAspecific PK selectively, and they also confirmed that baicalein could inhibit the activity of PK (Chan et al., 2013; Chan et al., 2011). Because the chemical structure of diosmetin is similar to baicalin, it is interesting to detect whether baicalin has potential inhibitory actions on ARSS PK. In this study, our results revealed that the enzymatic activity of ARSS PK was inhibited by baicalin. It is possible that interfering with ATP production by baicalin may stop the function of the MsrA efflux pump and contribute to the synergistic action of baicalin and Azm against ARSS. We also elucidated that baicalin could dosedependently inhibit the transcript level of the msrA gene. Taken together, these results indicated that baicalin interfered with ATP generation by inhibiting PK enzymatic activity and decreasing the relative expression of the msrA gene, ultimately inhibiting the MsrA efflux pump.

It has been reported that increasing resistance is associated with active efflux in bacterial biofilms (Van Acker and Coenye, 2016; Danquah et al., 2018). Baugh et al. demonstrated that numerous functional efflux pumps give rise to biofilm matrix expression in Salmonella Typhimurium and that the addition of a variety of efflux inhibitors inhibited biofilm formation (Baugh et al., 2012). Therefore, we hypothesized that the efflux inhibitor baicalin could be used as a biofilm formation inhibitor. In the present study, sub-inhibitory concentrations of baicalin or positive control verapamil did not affect S. saprophyticus growth (**Figure 2**) but reduced the values of A<sup>570</sup> measured by staining with crystal violet. ARSS was unable to form mature biofilms under baicalin or verapamil incubation conditions even after extended incubation times up to 48 h. These results indicated that baicalin and verapamil inhibited biofilm formation. Pearson's correlation coefficient analysis indicated that the relative expression of the efflux gene was positively correlated with biofilm formation (p < 0.01). Therefore, baicalin can inhibit biofilm formation by inhibiting the MsrA efflux pump in ARSS. Evidently, biofilms render the cells less accessible to the defense

FIGURE 8 | Influence of msrA gene relative expression in biofilm bacteria of baicalin (using 16S rRNA as the internal parametric gene). The relative expression of the msrA gene was determined in samples prepared from 24 h (A) and 48 h (B) ARSS biofilms in BHI using quantitative RT-PCR. Data are derived from three biological repeats. Bars in the same index without the same letters differ significantly (p < 0.05).

system of the organism and reduce antibiotic concentrations inside the target pathogen. The treatment of biofilm-related infections is a critical clinical problem in the current era. Because

TABLE 2 | Pearson's correlation coefficients between the relative expression of msrA and biofilm formation ability and the expression of agr system-associated genes.


∗∗p < 0.01, <sup>∗</sup>p < 0.05. 0.8–1.0, extreme correlation; 0.6–0.8, strong correlation; 0.4–0.6, moderate correlation; 0.2–0.4, weak correlation; 0.0–0.2, very weak correlation or no correlation.

of frequent reports on resistance to antimicrobials, the synergistic antibacterial method may be a better treatment strategy (Drugeon et al., 1999). The combined action of molecules was reported to eradicate the biofilms (Baugh et al., 2014). Overall, this evidence encouraged the investigation of the roles of baicalin in the prevention and eradication of biofilms. In this study, baicalin and Azm increased antibiotic permeability by disrupting the already-formed biofilms. Verapamil was used as a positive control compound to evaluate the combined effects of subinhibitory concentrations of baicalin with Azm on biofilms that were established for 48 h. S. saprophyticus could not be eradicated by Azm alone, and both baicalin + Azm and VP + Azm could significantly decrease the counts of bacteria compared with the counts for the SS group. Presumably, baicalin may be a potential agent for treating biofilm infections by either inhibiting biofilm formation or eradicating biofilms in combination with Azm.

In addition, multidrug efflux pumps often secrete metabolites involved in QS (Polkade et al., 2016). QS is a process of

bacterial cell–cell communication that allows bacteria to sense cell density and change bacterial gene expression patterns to alter bacteria group behaviors at high cell numbers (Rutherford and Bassler, 2012; Camilli and Bassler, 2006). This cross-talk between bacteria is believed to be essential for the establishment of bacterial biofilms and bacterial biofilm infection (Danquah et al., 2018). QS regulates the expression of genes encoding virulence factors involved in a range of toxins, adhesion molecules, and compounds that influence immune function. Chen et al. (2016) prophase studies demonstrated that baicalein interfered with the QS system and affected bacterial virulence. Therefore, the potential target for the treatment of bacterial biofilm infection is the QS system. Staphylococcus biofilm is regulated by a Grampositive two-component QS system encoded by the agr locus. An increase in cell density accounts for a prompt upsurge in the production, secretion, and detection of AIP. When AIP accumulates, it binds with agrC, which is a membrane-bound histidine kinase. Then, agrC autophosphorylates at a conserved histidine and transfers the phosphate group to an aspartate on the response regulator agrA (Lina et al., 1998). AgrA activates the divergently encoded P3 promoter, which controls the expression of RNAIII (Novick et al., 1993). Most of the effects of QS regulating virulence in Staphylococcus are achieved by direct and indirect regulation of RNAIII (Queck et al., 2008). Our results showed that 250, 125, and 62.5 mg/L baicalin and verapamil downregulated the transcript levels of agrA, agrC, RNAIII, and sarA in a dose-dependent manner compared with that of the SS group. In most Gram-positive QS bacteria, AIP was processed and exported by ABC transporters (Bassler, 1999). The MsrA efflux pump belongs to the ABC transporters. We found that there was a significant correlation between the relative expression of agr system-associated genes and the efflux gene (p < 0.01, **Table 2**). Therefore, we inferred that baicalin may inactivate the agr system by inhibiting the efflux of AIP. Recent findings (Ge et al., 2012; Drugeon et al., 1999) have demonstrated that RNAIII regulates biofilm formation and induces toxin production, such as plasma-coagulase, enterotoxin, and thermostable nuclease. Additionally, RNAIII is a transcriptional regulator of the Staphylococcal accessory regulator A (sarA) family (Lewis, 2007). Recent evidence has elucidated that SarA, as a central regulatory element, controls the production of staphylococcus virulence factors (Van Acker and Coenye, 2016). Therefore, we inferred that baicalin may regulate virulence by inhibiting the agr system.

## REFERENCES


## CONCLUSION

In conclusion, baicalin effectively inhibited the MsrA efflux pump, biofilm formation and the agr system in ARSS. In addition, there is a significant positive correlation between efflux and biofilm formation and the agr system. Therefore, we believe that S. saprophyticus biofilm-related infections could be treated by baicalin combined with Azm. To our knowledge, we first elucidated the positive relationship between efflux and biofilm formation in S. saprophyticus.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article.

## AUTHOR CONTRIBUTIONS

JW, HJ, and JL conceived and designed the experiments. JW, JM, MQ, HD, MH, and KM performed the experiments. JW and HJ analyzed the data. DW and YW contributed to the reagents, materials, and analysis tools. JW wrote the manuscript. All authors read and made additions to the manuscript during revision stages.

## FUNDING

This work was supported by the Fundamental Research Funds for the Central Universities (Y0201700441 and KYYJ201803), the National Natural Science Foundation of China (Grant Nos. 31572557 and 31772784), the Special Fund for Agro-Scientific Research in the Public Interest (201303040 and 201403051), and the Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD).

## ACKNOWLEDGMENTS

We thank all other staff at the Institute of Traditional Chinese Veterinary Medicine of Nanjing Agricultural University for their assistance with the experiments.


RN4220/pUL5054 and inhibition of MRSA pyruvate kinase. Phytomedicine 20, 611–614. doi: 10.1016/j.phymed.2013.02.007


**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 Wang, Jiao, Meng, Qiao, Du, He, Ming, Liu, Wang and Wu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# Increasing Prevalence of ESBL-Producing Multidrug Resistance Escherichia coli From Diseased Pets in Beijing, China From 2012 to 2017

#### Yanyun Chen<sup>1</sup>† , Zhihai Liu2,3† , Yaru Zhang1,4, Zhenbiao Zhang<sup>1</sup> , Lei Lei<sup>2</sup> and Zhaofei Xia<sup>1</sup> \*

#### Edited by:

Leonardo Neves de Andrade, University of São Paulo, Brazil

#### Reviewed by:

Tiago Casella, Faculdade de Medicina de São José do Rio Preto, Brazil Nilton Lincopan, University of São Paulo, Brazil Rafael Antonio Casarin Penha Filho, São Paulo State University, Brazil

\*Correspondence:

Zhaofei Xia zhaofeixiacau@126.com †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 23 August 2019 Accepted: 25 November 2019 Published: 10 December 2019

#### Citation:

Chen Y, Liu Z, Zhang Y, Zhang Z, Lei L and Xia Z (2019) Increasing Prevalence of ESBL-Producing Multidrug Resistance Escherichia coli From Diseased Pets in Beijing, China From 2012 to 2017. Front. Microbiol. 10:2852. doi: 10.3389/fmicb.2019.02852 <sup>1</sup> Department of Veterinary Internal Medicine, College of Veterinary Medicine, China Agricultural University, Beijing, China, <sup>2</sup> College of Chemistry and Pharmaceutical Sciences, Qingdao Agricultural University, Qingdao, China, <sup>3</sup> Beijing Advanced Innovation Center for Food Nutrition and Human Health, College of Veterinary Medicine, China Agricultural University, Beijing, China, <sup>4</sup> The New Hope Liuhe Co., Ltd., Qingdao, China

We investigated antimicrobial resistance trends and characteristics of ESBL-producing Escherichia coli isolates from pets and whether this correlates with antibiotic usage in the clinic. Clinical samples containing E. coli from diseased cats and dogs were screened for antibiotic sensitivity and associated genotypic features. We identified 127 E. coli isolates from 1886 samples from dogs (n = 1565) and cats (n = 321) with the majority from urinary tract infections (n = 108, 85%). High rates of resistance were observed for β-lactams and fluoroquinolones and resistance to > 3 antibiotic classes (MDR) increased from 67% in 2012 to 75% in 2017 (P < 0.0001). This was especially true for strains resistant to 6–9 antibiotics that increased from 26.67 to 60.71%. Increased rates in β-lactam use for clinical treatment accompanied these increasing resistance rates. Accordingly, the most frequently encountered subtypes were blaCTX-<sup>M</sup> (n = 44, 34.65%), blaCTX-M-<sup>65</sup> (n = 19) and blaCTX-M-<sup>15</sup> (n = 18) and qnrB (n = 119, 93.70%). The blaCTX-M-isolates possessed 36 unique pulsed field electrophoretic types (PFGEs) and 28 different sequence types (STs) in ST405 (7, 15.9%), ST131 (3, 6.8%), ST73, ST101, ST372, and ST827 (2, 4.5% each) were the most prevalent. This data demonstrated a high level of diversity for the blaCTX-M-positive E. coli isolates. Additionally, blaNDM-<sup>5</sup> was detected in three isolates (n = 3, 2.36%), comprised of two ST101 and one ST405 isolates, and mcr-1 was also observed in three colistin-resistant E. coli with three different STs (ST6316, ST405, and ST46). Our study demonstrates an increasing trend in MDR and ESBL-producing E. coli and this correlated with β-lactam antibiotic usage for treatment of these animals. This data indicates that there is significant risk for the spread of resistant bacteria from pets to humans and antibiotic use for pets should be more strictly regulated.

Keywords: multidrug resistance, antimicrobial drug usage, companion animals, ESBL, Escherichia coli

## INTRODUCTION

fmicb-10-02852 December 6, 2019 Time: 18:11 # 2

Antimicrobial resistance has become one of the most challenging problems for public health and results in 700,000 deaths annually (O'Neill, 2016). Antibiotic misuse has led to the spread of antibiotic resistance genes (ARGs) in humans, food animals, pets, songbirds, water, and soil and even agricultural plants, and this represents a significant threat to public health security (Carter et al., 2018; Hartantyo et al., 2018; Anderson et al., 2019; Chen et al., 2019; Gros et al., 2019; Sanchez et al., 2019; Vikesland et al., 2019). However, novel ARGs have emerged that encode resistance to carbapenems (blaNDM, blaIMP, blaVIM, and blaKPC) (Perez and Bonomo, 2019) colistin (mcr-1 to 9) and tigecycline (tetX3 and tetX4) (Carroll et al., 2019; He et al., 2019; Wang et al., 2019). The current limited development of novel drugs and substitutes makes the use of ARGs monitoring even more important to develop comprehensive and integrative measures for antimicrobial resistance.

Over the past two decades, there has been a significant number of infections caused by bacteria expressing extendedspectrum-β-lactamase (ESBL) and carbapenemases (Logan and Weinstein, 2017). In particular, ESBL isolates have been found in humans (Pitout and Laupland, 2008), animals, the environment (water and soil) (Runcharoen et al., 2017), meat and even vegetables (Yang et al., 2019). ESBL are becoming more common because this phenotype is being selected for by the use and exposure to β-lactams, especially the cephalosporins. This has generated a vicious cycle of drug resistance and decreased therapeutic effects. The increasing use of cephalosporin has been linked to Escherichia coli infections in pigs (Hammerum et al., 2014) and a high frequency of ESBL-producing E. coli was directly linked with a high consumption of third- or fourthcephalosporins (Andersen et al., 2015).

One area that has not been thoroughly investigated is ARG presence in companion animals. This group comes in intimate contact with humans and pet contact can lead to bacterial spread to humans (Lloyd, 2007). In particular, E. coli is a common pathogenic agent isolated from pets (Mathers et al., 2015) and is often present in dogs and cats with urinary tract infections (UTIs), pyometra and respiratory tract infections (Karkaba et al., 2019; Moyaert et al., 2019). Multidrug resistance (MDR) E. coli isolates have emerged in companion animals in the United States and Europe (Morrissey et al., 2016), but data for China is lacking.

Here, we explored the effects of antibiotic dose on resistance phenotypes in pet bacterial isolates over a 6-year period. The aim of this study was to investigate the dissemination of ESBLproducing multidrug-resistant pathogens in diseased pets and the correlation between resistance rates and consumption of β-lactam antibiotics.

## MATERIALS AND METHODS

## Samples and Identification of Bacterial Isolates

Animal samples were collected at the Veterinary Teaching Hospital of China Agricultural University (VTH-CAU), between January 2012 and June 2017. This study was approved by the China Agricultural University Animal Ethics Committee and the approval document (No. AW08104102-2) (see **Supplementary Information**). We collected 1886 samples from dogs (1565) and cats (321) and where some samples were gathered from different infections in the same pet. All samples consisted of urine (UTI) samples (1398, 74.1%) and samples from pyoderma (125, 6.7%), ear swabs (6.1%), effusions (115, 6.1%) and other specimens (132, 7.0%), which included pus from the uterus and soft tissue infections and several trachea lavage fluids.

Bacterial strains were recovered using blood agar and MacConkey agar plates that were incubated at 37◦C for 24 h and single pink colonies were collected from each isolation plate. Subsequently, the DNA of individual clones was extracted by Fast Pure Bacteria DNA Isolation Mini Kit (Vazyme Biotech, Nanjing, China) and used as templates for PCR. PCR amplification of the 16S rDNA gene were performed for all isolates as previously described (Brianna et al., 2013), and amplicons were sequenced to confirm bacterial genus using the BLAST algorithm<sup>1</sup> .

## Antimicrobial Susceptibility Testing

Antimicrobial susceptibility of isolates was performed using the broth microdilution method according to Clinical and Laboratory Standards Institute guidelines (CLSI, 2015a). The breakpoints for other antimicrobials used the CLSI (M100-S25 or Vet01-A4/Vet01-S2) and EUCAST (CLSI, 2015b; EUCAST, 2019), where the breakpoints (R) for tigecycline, orbifloxacin, enrofloxacin, and marbofloxacin were recommended as ≥0.5, ≥4, ≥4, and ≥8 µg/ml, respectively. The screening panel consisted of 17 antibiotics that included ampicillin, cefazolin, cefotaxime, ceftriaxone, meropenem, amoxicillinclavulanic acid, aztreonam, ciprofloxacin, enrofloxacin, marbofloxacin, orbifloxacin, chloramphenicol, amikacin, gentamicin, doxycycline, colistin, and tigecycline. Isolates with resistance to three or more categories of antimicrobial agents were classified as MDR (Magiorakos et al., 2012). E. coli ATCC 25922 was used as the quality control strain. Resistance was categorized according to Standardized International Terminology and our 17 test antibiotics were contained within nine categories (Magiorakos et al., 2012).

## Survey of Antimicrobial Drug Usage at the VTH-CAU

Antibiotic usage at VTH-CAU was recorded between January 2014 and September 2017 to correlate antibiotic usage and the resistance of E. coli isolates for each study animal.

## ARG Detection

Screening of E. coli isolates for ARG types was conducted using PCR for each isolate depending on the antibiotic resistance phenotype. The ARGs we examined were (1) carbapenemase genes blaNDM, blaIMP blaKPC, blaVIM, blaOXA, blaAIM, blaBIC, blaDIM, blaGIM, blaSIM, and blaSPM including blaNDM and blaCTX-<sup>M</sup> subtyping (Poirel et al., 2011), (2) β-lactamase genes

<sup>1</sup>https://www.ncbi.nlm.nih.gov/

blaSHV, blaTEM, and blaCTX-<sup>M</sup> (Casella et al., 2018), (3) plasmidmediated AmpC β-lactamase genes blaMOX, blaCMY, blaLAT, blaDHA, blaACC, blaMIR, blaACT, and blaFOX (Perez-Perez and Hanson, 2002), (4) colistin resistance genes mcr-1-8 (Rebelo et al., 2018; Wang X. et al., 2018; Yang et al., 2018; Carroll et al., 2019), and (5) plasmid-mediated quinolone resistance (PMQR) genes qnrA, qnrB and qnrS including whether the gyrA and parC genes in the quinolone resistance determining region (QRDR) were mutated (Komp et al., 2003; Kraychete et al., 2016; Onseedaeng and Ratthawongjirakul, 2016). PCR primers used for screening are shown in **Supplementary Table S1**. All PCR amplicons were sequenced to confirm gene identity.

## Transconjugation Assays and Whole Genome Sequencing

Conjugation assays were performed between clinical isolates and E. coli J53 to evaluate whether blaNDM and mcr-1 were mobilizable. Transconjugants were selected on MacConkey agar containing 100 mg/L sodium azide and 1 mg/L meropenem or colistin. Presumptive transconjugants were identified using PCR screening for blaNDM and mcr-1.

Transconjugant DNA was extracted and used for whole genome sequencing (WGS). A library of 250-bp paired-end was constructed by using a NEXT Ultra DNA Library Prep kit (New England Biolabs, Beverley, MA, United States) and sequenced using an Illumina HiSeq 2500 system at Bionova Biotech (Beijing, China). Raw data was de novo assembled using the SPAdes algorithm v.3.10.0. ARGs and plasmid incompatibility groups were analyzed using ResFinder v.3.2<sup>2</sup> , RASTtk v.2.0<sup>3</sup> and PlasmidFinder 2.1<sup>4</sup> , respectively.

## PFGE and MLST Typing of E. coli Strains

The clonal relatedness of blaCTX-<sup>M</sup> positive was determined using pulsed-field gel electrophoresis (PFGE) typing conducted as previously described (Tenover et al., 1995). PFGE patterns were visually inspected and gel images were analyzed using InfoQuest FP software (Biorad, Hercules, CA, United States). Group analysis of PFGE profiles was performed using the Dice coefficient and the unweighted pair group method with arithmetic means. Simultaneously, multilocus sequence typing (MLST) analysis was conducted using the following E. coli gene set: recA, adk, fumC, icd, mdh, purA, and gyrB. The results were interpreted using the MLST database<sup>5</sup> . Sequence types (STs) of 131 clades in E. coli isolates were also screened using multiplex conventional PCR assays as previously described (Matsumura et al., 2017).

## Statistical Analysis

Tests of statistical significance was determined using Fisher's exact test with Yates continuity correction in GraphPad Prism 6 (San Diego, CA, United States) and the level of significance was set at P < 0.05. All figures were designed by ggplot2<sup>6</sup> and GraphPad Prism 6.

## RESULTS

## Samples and E. coli Isolates

We isolated 127 E. coli strains that included 108 (85.0%) urine and 5 (3.9%) uterus, 4 (3.1%) abdominal fluid, 3 (2.4%) pyoderma, 2 (1.6%) soft tissue infectious sites, and 1 (0.8%) from synovial fluid. The E. coli isolation rate was 6.73% (127/1886) and ranged from 4.42 to 9.41% per year. Resistance to β-lactams was extremely high for these isolates and included resistance to ampicillin (77.9%), cefotaxime (58.3%) and ceftriaxone (58.3%) and cefazolin (65.35%) with the exception of amoxicillin + clavulanic acid (22.8%) (**Supplementary Tables S2, S3**).

The range of resistance rates was narrow for ampicillin (68.42**–**92.86%) compared with cefotaxime (36.67**–**72.73%), ceftriaxone and cefotaxime (36.67**–**72.73%), cefazolin (43.33**–** 86.36%), and amoxicillin + clavulanic acid (10.00**–**27.27%). These data demonstrated a significant increase in resistance rates from 2012 to 2017 and the cefotaxime, ceftriaxone, cefazolin, and amoxicillin + clavulanic acid rates more than doubled (**Supplementary Figure S1** and **Supplementary Table S3**). The aztreonam resistance rate showed the greatest variability and was maximal in 2014 (71.43%) and minimal in 2012 (13.33%). We also found high rates of resistance to the four fluoroquinolones we tested (52.76–57.48%) except for 2013 (36.84%) (**Supplementary Table S3**). The other resistance groups we examined displayed irregular trends or slight increases over the 2012**–**2017 study period. Notably, almost all isolates showed high susceptibility to the "last-resort" antibiotics colistin, meropenem, and tigecycline and only 5 (3.94%) of our isolates were resistant to colistin and 3 (2.4%) to meropenem (**Figure 1** and **Supplementary Table S3**).

The MDR values for our isolates from these pets were high with an overall MDR rate that increased 73.2% from 2012 to 2017 (n = 93). These MDR rates from 2012 to 2017 were 66.67, 68.42, 64.29, 85.71, 81.82, and 75.00% in 2017, respectively (P < 0.0001) (**Supplementary Table S4**). Specifically, MDR prevalence in 6–9 antimicrobial categories exhibited an obvious increase (**Figure 1** and **Supplementary Table S4**). The MDR of our 44 CTX-M-producing E. coli was greater than for the non-CTX-M-expressing isolates (**Supplementary Figures S5a,b**). Moreover, all blaCTX-M-positive E. coli were MDR strains possessing resistance to the penicillins [ampicillin (AMP)], non-extended spectrum cephalosporins [cefazoline (CZO)], extended-spectrum cephalosporins [cefotaxime (CTX) and ceftriaxone (CRO)]. The most common MDR pattern was resistance to quinolones (ciproxacin, enrofloxacin, orbifloxacin, and marbofloxacin) and the tetracyclines (doxycycline) and aminoglycosides (gentamycin) + β-lactams (AMP, CZO, CTX, and CRO). When this data was viewed solely by the number of antibiotics per MDR isolate, resistance rates to 11–17 antibiotics

<sup>2</sup>https://cge.cbs.dtu.dk

<sup>3</sup>http://rast.nmpdr.org/rast.cgi

<sup>4</sup>https://cge.cbs.dtu.dk/services/PlasmidFinder/

<sup>5</sup>http://enterobase.warwick.ac.uk/species/ecoli/allele\_st\_search

<sup>6</sup>https://ggplot2.tidyverse.org/

increased significantly over the study period from 13.3, 31.6, 35.7, 35.7, 50.0, and 53.6% from 2012 to 2017, respectively (**Supplementary Figure S2** and **Supplementary Table S5**).

We performed a correlation analysis between antibiotic usage and the rate of antibiotic resistance between 2014 and 2017. Amoxicillin-clavulanic acid (1.3 × 10<sup>7</sup> mg/yr), doxycycline (5.7 × 10<sup>6</sup> mg/yr), ampicillin (1.9 × 10<sup>6</sup> mg/yr), and enrofloxacin (8.1 × 10<sup>5</sup> mg/yr) were the most widely used antibiotics at the animal facility. With the widespread use of FQs in VTH-CAU, >50% of the clinical E. coli isolates showed resistance to FQs since 2012. Interestingly, even though fluoroquinolone usage had decreased dramatically, resistance rates to this antibiotic class remained high. In addition, the use of β-lactams, including amoxicillin-clavulanic acid, ampicillin, and ceftriaxone, had increased over the study period and was positively correlated with an increase in β-lactam resistance that was also related to dosage. Meropenem and imipenem were not used at the facility and we found no resistance from 2012∼2015 although three resistant E. coli were detected in 2016 (**Supplementary Figure S3**).

## ARG Prevalence and NDM Plasmid Characterization

In our study, we identified 10 ARG types and mutations in gyrA and parC. The qnrB gene was present in 119 (93.70%) of the isolates (**Supplementary Figure S4**). In the gyrA gene of 77 quinolone-resistant isolates, 38 carried the mutations S83L and D87N, 9 S83L and D87Y, 2 D87N, 3 D87G, and 14 S83L (**Supplementary Table S6**). In the parC gene we found the mutations S80I (22 isolates), S80I and E84G (14 isolates) and S80I and E84V (8 strains). The ESBL and pAmpC-containing isolates harbored blaCTX-<sup>M</sup> (n = 44, 34.65%), blaSHV (n = 21, 16.55%), blaOXA (n = 9, 7.09%), blaCMY (n = 12, 9.45%), blaFOX (n = 4, 3.15%), and blaNDM (n = 3, 2.36%). CTX-M alleles were assigned to two main clusters; CTX-M-14 and -15 including seven CTX-M genotypes (-14, -15, -64, -65, - 116, -127, and -174) where CTX-M-65 (43.18%, n = 19) and CTX-M-5 (40.91%, n = 18) predominated. All 44 blaCTX-Mpositive isolates were resistant to cefotaxime and ceftriaxone and were classified as MDR. When compared with blaCTXnegative isolates, the blaCTX-positive isolates showed significantly greater resistance to second and third-generation cephalosporins, the fluoroquinolones, aztreonam, doxycycline, gentamicin, and chloramphenicol (P < 0.05) (**Table 1** and **Supplementary Figure S5**). The only observed carbapenemase gene we identified was NDM-5 that was present in three isolates 16DU02, 16DF03, and 16XXI8 and all showed resistance to meropenem and all were collected from dogs in 2016. Of note, 16DU02 and 16DF03 were isolated from same dog but the samples were isolated from urine and an abdominal effusion, respectively, and they carried same genes blaNDM-5, blaTEM, blaCTX-M-65, and qnrB. The 16XXI8 isolate recovered from dog urine possessed blaNDM-5, blaTEM, blaCTX-M-15, blaOXA, and qnrB. Additionally, the colistin resistance gene mcr-1 was detected in three blaCTX-<sup>M</sup> positive E. coli while other mcr variants were undetected.

We further examined our mcr-1 and blaNDM isolates and tested whether these genes were present on mobile elements. In our conjugation tests, only blaNDM in three E. coli isolates


TABLE 1 | Minimum inhibitory concentration (MIC) of antimicrobial agents for clinical E. coli isolates from cats and dogs in Beijing, China, 2012–2017 (n = 127)‡ .

‡The 127 isolates consisted of 107 isolates from urinary tract, 5 from uterus, 4 from abdominal fluid, 1 from synovial fluid, and 10 from pyoderma and soft tissue. †P-values were determined by Fisher's exact test. <sup>∗</sup>P < 0.05; ∗∗P < 0.01.

were successfully transferred at frequencies of 4.86 × 10−8– 8.02 × 10−<sup>7</sup> . Two complete 46,161 bp blaNDM-5-harboring plasmids pP16NDM-502 (MN701974) and pP16NDM-503 (MN701975) were obtained from strains 16DU02 and 16DF03 transconjugants, respectively. The backbone sequences were assembled and contigs and gaps were identified by additional PCR and sequence analyses. The other NDM-1-carrying plasmid was unsuccessfully assembled because of fragmentary and short contigs (**Supplementary Figure S6**). The two completely assembled plasmids were all in the IncX3 replication group. In addition, blaNDM-<sup>5</sup> was contained within an insertion sequence (IS) cassette (1ISAba125-IS5-blaNDM-ble-trpF-dsbC-IS26). blaNDM-<sup>5</sup> and bleMBL were the only ARGs present in the two plasmids. Homology analysis revealed that pP16NDM-502 and pP16NDM-503 were ≥ 99% identical to the following IncX3 blaNDM plasmids: (i) pNDM\_MGR194 (KF220657) from a Klebisella pneumoniae human isolate in India, (ii) p1079-NDM (MG825384) from a chicken E. coli isolate in China. (iii) pL65-9 (CP034744) from E. coli in goose in China, (iv) pZHDC40 (KY041843) from E. coli human isolate in China, (v) pQDE2- NDM (MH917280) from K. pneumoniae human isolate in China, (vi) pCRCB-101\_1 (CP024820) from a Citrobacter freundii human isolate in Korea (vii) p128379-NDM (MF344560) from Enterobacter hormaechei from a human sample in China and (viii) pAD-19R (KX833071) from a chicken E. coli isolate in China (**Supplementary Figure S6**). We were unable to transfer the mcr-1 gene by conjugation in three separate tests and we could not assemble a complete plasmid sequence. However, no complete plasmid sequence of mcr-1 was successfully assembled by using three clinical isolates genomes. Although the analysis of incompatibility group in three clinical isolates genomes by analyzed PlasmidFinder-2.0 Serve revealed they possessed IncFIB, IncFIC, IncFII, and IncHI2 typical fragments, all mcr-1 genes weren't located on those fragments (data not showed).

## PFGE and MLST Typing

In our group of 127 E. coli isolates, most (n = 109, 85.8%) were successfully characterized by PFGE typing and included 38/44 blaCTX-M-positive isolates. These 38 isolates were obtained from cats (6) and dogs (32) and could be subdivided into 36 unique PFGE patterns. Interestingly, two isolates obtained from urine (15cu184) and perirenal effusion (15cu186) samples showed identical PFGE patterns and carried the same blaCTX-M-<sup>127</sup> subtype but were recovered from different animals (**Figure 2**).

We also found universality in MLST types and identified 28 different STs although 1 isolate failed to type. The most prevalent were ST405 (7, 15.9%), ST131 (3, 6.8%) ST73, ST101, ST372, and ST827 (2, 4.5% each). All ST405 strains were collected from dogs but in different years. Three ST131 isolates (UTI-27, UTI-166, and DU40) belonged to clade C1, in which UTI-27 and DU40 were assigned as C1-nM27, and UTI-166 as C1-M27. The blaCTX-M-<sup>65</sup> and blaCTX-M-<sup>15</sup> positive E. coli exhibited the greatest ST diversity and contained 13 and 14 STs, respectively. Additionally, three blaNDM-5-positive isolates belonged to ST405 (n = 1) and ST101 (n = 2) and three carryingmcr-1 E. coli were classified as three different STs: ST6316, ST405, and ST46 (**Figure 3**).

## DISCUSSION

In this study, we investigated the antibiotic resistance profiles and trends in MDR E. coli isolates collected from diseased dogs and cats. We identified 127 E. coli isolates and most were


associated with UTI (84.2%) accounting for 20.4% (108/529) of the confirmed bacterial UTI cases. In contrast, in the United States nearly 30% of UTI isolates from pets were E. coli (Ling et al., 2001; Hall et al., 2013). Previous studies in European countries based on 22,256 isolates from dogs and cats with UTI between 2008 and 2013 showed that E. coli was the most common pathogen in both dogs (59.50%) and cats (59.31%) (Marques et al., 2016). Isolation of E. coli from the respiratory tract is much less frequent and occurred in 10–15% of dog respiratory tract infections (Rheinwald et al., 2015; Morrissey et al., 2016).

Our 127 clinical E. coli isolates showed high prevalence rates of resistance to β-lactams (58.3–77.9%) and quinolones (52.8–57.4%). This pattern was quite different from the data of dog and cat E. coli isolates from United States that identified resistance to ampicillin at 40%, cephalexin 98% and doxycycline at 100% (Thungrat et al., 2015). An Australian study revealed that canine clinical E. coli isolates had low rates of resistance to quinolones (9.1–9.3%) and among 392 canine UTI isolates, 9.9–10.2% were resistant to third-generation cephalosporins (Saputra et al., 2017).

In the current study, the overall MDR frequency of 73.2% was higher than that observed in studies from the United States (52%) and Poland (66.8%) (Rzewuska et al., 2015; Thungrat et al., 2015). In contrast, a European multi-centre study on AMR of various bacteria isolated from companion animals with UTIs showed a much lower frequency of MDR among E. coli isolates (1.4–29.7%). The high MDR rates observed in the current study indicated that currently available antimicrobial treatment options for E. coli infections in companion animals are limited. We also identified an increase in E. coli MDR prevalence between 2012 and 2017 and increased rates of thirdgeneration cephalosporins and amoxicillin-clavulanic acid usage was accompanied by increased AMR and MDR rates. Based on these findings, we speculate that the rising incidence of multidrug-resistant E. coli might be associated with the heavy use of these antibiotics in companion animals at the hospital.

Among the 44 CTX-M-producing E. coli isolates in this study, blaCTX-M-<sup>65</sup> (43.2%) and blaCTX-M-<sup>15</sup> (40.9%) were the most blaCTX-<sup>M</sup> variants. Currently, more than 220 CTX-M-lactamases have been reported and clustered into five subgroups, containing

CTX-M-1, -2, -8, -9, and -25 depending on amino acid sequence homology (Peirano and Pitout, 2019). CTX-M-15 (CTX-M-1) is the most frequent CTX-M variant worldwide (Karim et al., 2001) especially in South-East Asia, China, South Korea, and Japan (Peirano and Pitout, 2019). Since 2000, blaCTX-M-<sup>15</sup> has emerged worldwide and is the most prevalent ESBL globally (Peirano and Pitout, 2019). The spread of blaCTX-M-<sup>15</sup> contributed to the dissemination of MDR bacterial isolates in animals but also in humans; a significant public health concern (Mugnaioli et al., 2006; Canton et al., 2012; Liu et al., 2016b). While the blaCTX-M-<sup>15</sup> variant was also common in canine E. coli isolates from both Shanxi, China (n = 40) and the United States (n = 50), blaCTX-M−<sup>123</sup> predominated in those studies (Liu et al., 2016a,b). The blaCTX-M-<sup>65</sup> was the most prevalent subtype we found and is frequently detected in Salmonella isolates. In a previous study, the blaCTX-M-<sup>65</sup> gene (n = 131) was identified among 153 ESBL-positive Salmonella isolates from poultry slaughterhouses (n = 121) and humans (n = 10) (Bai et al., 2016). Similarly, a study on chickens and pigs in China surveyed seven blaCTX-M-<sup>65</sup> among Salmonella isolates (Zhang et al., 2016). All these studies indicated more frequent occurrences in China and is supported by data from 2005 to 2017 in another study (Bevan et al., 2017). Additionally, blaCTX-M-<sup>55</sup> prevalence has increased in recent years in China among animal and human isolates, although not in our present study. Conversely, blaCTX-M-<sup>14</sup> has a remarkable reduction based on data for China (Bevan et al., 2017). We found only one blaCTX-M-<sup>14</sup> isolate while this gene is globally present (Bevan et al., 2017). Although other blaCTX-<sup>M</sup> variants, such as blaCTX-M-64, -174, -116, and -<sup>127</sup> rarely occur, their presence in diseased pets suggests that diversity and evolution of blaCTX-<sup>M</sup> had occurred in these companion animals.

All blaCTX-carrying isolates in the current study had MDR profiles and showed high resistance rates to β-lactams, quinolones, doxycycline, gentamycin, and chloramphenicol. This suggested that other resistance genes may be co-transferred with blaCTX, making it even more difficult to eliminate the spread of MDR. Most blaCTX-<sup>M</sup> positive strains in other studies were also resistant to quinolones due to topoisomerase modifications of qnr genes (Lahlaoui et al., 2014) as we found in the present study. Additionally, mutations in gyrA

and parC also can be responsible for quinolone resistance. Our results showed S83L and D87 alterations to N, Y, G generated in gyrA were commonly associated with resistance to fluoroquinolones and mostly generate high level resistance (Basu and Mukherjee, 2019).

Only three E. coli isolates containing blaNDM-<sup>5</sup> were identified in our study animals with UTI in late 2016. Five (3.94%) were colistin resistance and three carried mcr-1. Previous reports implied that the transmissible blaNDM and mcr-1-carrying plasmids play a major role in the dissemination of these genes (Wang et al., 2017). For example, mcr-1, initially named as mobile colistin resistance gene, was generally considered as mediating the rapid spread of bacterial colistin resistance worldwide due to its mobile plasmid association. In contrast, we found that mcr-1 could not be mobilized by conjugation. In a previous study, 14/23 mcr-1-positive isolates were successfully transferred and six plasmids were non-transferable (Zhou et al., 2017). The mcr-1 gene was located in the chromosome of E. coli from a goose isolate (Lu et al., 2019). Furthermore, ISApl1 transposon can mediate mcr-1 transfer from chromosome to plasmids and this is the reason for its current global distribution (Wang R. et al., 2018). So, mcr-1-carried in a non-transferable plasmid or chromosome may be spread by ISApl1 or others transposons. In our study, we could not assemble a complete plasmid sequence for the mcr-1 genomic data; and mcr-1 genes were not present on those plasmid fragments and may be chromosomal. There are nine variants of mcr and we identified only mcr-1 (Carroll et al., 2019). This gene is the most prevalent variant globally and has been detected in almost 40 countries/regions across five continents in both involving developed and non-developed countries. Moreover, the mcr-1 gene was found in more than 11 bacterial species and in diverse locations such as rivers, public beaches, well water, wastewater, hospital sewage, foods (vegetables and meats), animals (wild birds, housefly/blowfly, cattle, pigs, poultry, and companion animals) (Feng, 2018). Our study provided new evidence for the above claim and we identified three mcr-1 positive E. coli in diseased pets. Recently, carbapenem-resistant Enterobacteriaceae (CRE) have posed a threat to humans and animals because they exhibited resistance to most β-lactams including carbapenems, further compromising treatment of MDR infections (Gupta et al., 2011; Potter et al., 2016).

Carbapenem-resistant Enterobacteriaceae is mediated largely by the production of carbapenemase especially for NDM isolates. NDM-5, an NDM-1 variant, exhibited increased enzyme activity to carbapenems (Rogers et al., 2013) and its gene blaNDM-<sup>5</sup> has been reported worldwide (Khan et al., 2017) and is the most prevalent variant in China (Shen et al., 2018). In support of this, 84 (52%) NDM-5-producing E. coli were collected from 161 blaNDM carrying CRE in chickens (Wang et al., 2017). Our study revealed that pets have become a reservoir of NDM-5-producing E. coli. The blaNDM-<sup>5</sup> gene is also associated with different plasmids such as IncFIA/B, IncFII, IncN, and IncX3 (Sun et al., 2015; Tyson et al., 2019; Zhang et al., 2019) and IncX3 was the dominant type (Li et al., 2018; Zhang et al., 2019). Similarly, the blaNDM-5-harboring plasmid was assigned to the IncX3 type, and their transferability was confirmed in our study indicating a risk of blaNDM-<sup>5</sup> plasmid transfer between bacteria. However, further studies should be conducted to determine the origins of blaNDM and mcr-1 in the E. coli isolates from companion animals at the hospital. This will be helpful in designing measures to control the spread of MDR isolates.

All CTX-M-producing E. coli in our study displayed a diversity of PFGE patterns and STs demonstrating that blaCTX-<sup>M</sup> encoding ESBLs are present in diverse E. coli. Additionally, our diseased pet samples were from animals having no prior contact suggesting that clonal spread had a low frequency and it was not likely that clonal spread of blaCTX-M-positive E. coli occurred between the pets. But ESBL or other resistance genes can transfer between different E. coli by transferable genetic elements, as previously reported (Kim et al., 2019). However, two E. coli isolates (15cu184 and 15cu186) from same cat but different sample sources shared the same PFGE, ST, and blaCTX-M−<sup>127</sup> types, indicating that clonal spread may be occurring. It was reported that dissemination of Enterobacteriaceae can occur between pets and their owners by both horizontal transfer and clonal expansion (Yao et al., 2016).

In our study, ST405 E. coli (n = 7) was the most prevalent isolates carrying blaCTX-M. They presented different blaCTX-<sup>M</sup> subtype genes including blaCTX-M-<sup>15</sup> (n = 3), blaCTX-M-<sup>65</sup> (n = 3), and blaCTX-M−<sup>64</sup> (n = 1). In a clinical study from King Abdulaziz Medical City (KAMC) in Riyadh (Alghoribi et al., 2015), six ST405 E. coli harboring blaCTX-<sup>M</sup> were detected in UTI and ST405 (P ≤ 0.02) were significantly associated with ESBL production. ST405 extended-spectrum β-lactamase-producing E. coli (ESBL-EC) was also detected in animals, in which one ST405 ESBL-EC was screened in barbary macaques (Macaca sylvanus) in Algeria, notably, it held the mcr-1, blaTEM−1, and qnrB19 genes and blaCTX-M-<sup>15</sup> (Bachiri et al., 2017). We observed a similar pattern and a UTI25 (ST405 ESBL-EC) isolate harbored mcr-1, blaTEM, qnrB, and blaCTX-M−64. Additionally, ST405 ESBL-EC (16XXI8) carried blaNDM-5, blaOXA, blaTEM, and qnrB. These data indicated ST405 ESBL-EC has become the vector of multiple resistance genes, containing carbapenemand colistin-resistance genes. In a Sweden project, ST405 ESBL-EC was also found in UTI from a diseased cat (Bogaerts et al., 2015). We found seven ST405 ESBL-EC that were all recovered from dogs but in different years. Thus, companion animals already contain the hosts of ST405 ESBL-EC. In addition, except for ST405, we also identified two ST101 E. coli carrying blaNDM-<sup>5</sup> (16DU02 and 16DF03). They were from the same animal source and possessed the same ST, PFGE, plasmid type and blaNDM-5, suggesting a clonal origin. ST101 was reported to be strongly associated with the NDM genotype although most were NDM-1 rather than NDM-5. Recently, five blaNDM-5 positive E. coli were identified associated with ST101 and ST1196 (Ranjan et al., 2016; Aung et al., 2018). In contrast, only two ST101 NDM-5-producing E. coli were observed in another study and most were ST167 and ST410 but were clonally spread in a hospital (Sun et al., 2019). This was similar to our study and suggested that ST101 may be becoming the most important clone for dissemination of blaNDM-5, similar to ST167 and ST410.

The clone ST131 was observed in three E. coli isolates. This is the predominant E. coli lineage in extraintestinal pathogenic E. coli (ExPEC) isolates worldwide and are associated with global community and nosocomial dissemination (Marie-Hélène et al., 2014). Currently, ST131 is classified as three clades: A, B, and C (Peirano and Pitout, 2019). In our study, all ST131 belonged to clade C and ST131 is in clade C 80% of the time (Peirano and Pitout, 2019). Clade C evolved from clade B and further evolved into C1 and C2. Recently, a new C1 subclade C1- M27 was identified in animal and human (Matsumura et al., 2016), including companion animals (Melo et al., 2019), and we found one C1-M27 subclades that are rare in China, but prevalent in Europe (Ghosh et al., 2017; Merino et al., 2018). The other ST131 belonged to subclade C1-nM27. C1 is commonly associated with quinolone resistance, in agreement with our results. The ESBL-EC isolates were commonly associated with ST131, where 40–80% of ESBL ExPEC belong to ST131 E. coli (Marie-Hélène et al., 2014), and CTX-15 was the most prevalent ESBL enzyme in ST131 ESBL-EC (Alghoribi et al., 2015). Our study demonstrated that blaCTX-M-<sup>65</sup> was found in two ST131 ESBL-EC and one blaCTX-M-<sup>15</sup> isolate. Almost all ST131 isolates were resistant to fluoroquinolones, especially for ESBL CTX-M-15 isolates (Marie-Hélène et al., 2014). We found similar results that almost all CTX-M ESBL isolates (40/44) were resistant to quinolones. ST131 ESBL E. coli possessed more frequent resistance to amikacin than non-ST131 ESBL isolates, but showed more frequent susceptibility to gentamicin or trimoxazole. On the other hand, the non-ESBL isolates such as E. coli ST131 were more frequently resistant to fluoroquinolones than non-ST131 isolates. Therefore, the resistance to quinolones may be associated with ST131 rather than ESBL presence (Marie-Hélène et al., 2014). These findings indicated that quinolones resistance may have been the predecessor of ESBL enzymes. We found that ST131 and ST405 were correlated to ESBL production as previously found (Alghoribi et al., 2015). Other STs in our study occurred infrequently. Recently, a water sample study revealed that ESBL-producing E. coli isolates were present 15 different STs (ST10, ST46, ST48, ST58, ST69, ST101, ST117, ST131, ST141, ST288, ST359, ST399, ST405, ST617, and ST4530) (Said et al., 2016), We detected ST46, ST101, ST131, ST359, and ST405 in our study. Multiple ST types (ST46, ST1286, ST10, ST29, ST101, and ST354) have also been found in chickens and they carried mcr-1 and produced ESBLs (Wu et al., 2018). In our work, 28 diverse STs in ESBL-EC were found.

## CONCLUSION

We found a high prevalence of MDR E. coli isolates from diseased cats and dogs in Beijing and this rate has markedly

## REFERENCES

Alghoribi, M. F., Gibreel, T. M., Farnham, G., Al Johani, S. M., Balkhy, H. H., and Upton, M. (2015). Antibiotic-resistant ST38, ST131 and ST405 strains are the leading uropathogenic Escherichia coli clones in Riyadh, Saudi Arabia. J. Antimicrob. Chemother. 70, 2757–2762. doi: 10.1093/jac/dkv188

increased over the last 6 years. The widespread use of thirdgeneration cephalosporins and amoxicillin-clavulanic acid at the veterinary teaching hospital has likely contributed to the increasing frequency of β-lactam resistance in these isolates. These strains carried blaCTX, blaNDM-5, and mcr-1 and most possessed an MDR profile. Diversity analysis of the PFGE patterns and STs of these clinical E. coli isolates from different origins suggested that the dissemination of blaCTX have broad reservoirs of E. coli. Future studies should be undertaken to identify the MDR transmission mechanisms and establish national standards for the rational use of antibiotics in companion animals.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## ETHICS STATEMENT

This study was approved by the Agricultural University Animal Ethics Committee, China. The animals were given the best practice veterinary care and informed consent was granted by the owners.

## AUTHOR CONTRIBUTIONS

ZX was responsible for the study design. YC, ZL, YZ, and ZZ assisted in the data collection. ZL, YC, and LL interpreted the data. ZL, YC, and ZX completed the report writing. All authors revised, reviewed, and approved the final report.

## FUNDING

This work was supported by the Beijing Science and Technology Plan Project (Z171100001517008) and the National Natural Science Foundation of China (31422055).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02852/full#supplementary-material


extended-spectrum cephalosporinase-producing Escherichia coli in Danish pig herds. Vet. J. 204, 345–350. doi: 10.1016/j.tvjl.2015.03.014


on-farm livestock waste treatment plant. J. Hazard. Mater. 378:120716. doi: 10.1016/j.jhazmat.2019.05.109


fmicb-10-02852 December 6, 2019 Time: 18:11 # 10


with blaNDM-5 from a dog in the United States. Microbiol. Resour. Announc. 8:e00872-19. doi: 10.1128/MRA.00872-19


**Conflict of Interest:** YZ was employed by the company of Shandong New Hope Liuhe Group Ltd.

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 Chen, Liu, Zhang, Zhang, Lei and Xia. 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.

# Down-Regulation of Flagellar, Fimbriae, and Pili Proteins in Carbapenem-Resistant Klebsiella pneumoniae (NDM-4) Clinical Isolates: A Novel Linkage to Drug Resistance

#### Edited by:

Raffaele Zarrilli, University of Naples Federico II, Italy

## Reviewed by:

Gokhlesh Kumar, University of Veterinary Medicine Vienna, Austria Rita Berisio, Italian National Research Council (CNR), Italy

> \*Correspondence: Asad U. Khan asad.k@rediffmail.com

#### †Present address:

Divakar Sharma, Central Research Facility, Mass Spectrometry Laboratory, Kusuma School of Biological Sciences, Indian Institute of Technology, New Delhi, India

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 17 August 2019 Accepted: 27 November 2019 Published: 17 December 2019

#### Citation:

Sharma D, Garg A, Kumar M, Rashid F and Khan AU (2019) Down-Regulation of Flagellar, Fimbriae, and Pili Proteins in Carbapenem-Resistant Klebsiella pneumoniae (NDM-4) Clinical Isolates: A Novel Linkage to Drug Resistance. Front. Microbiol. 10:2865. doi: 10.3389/fmicb.2019.02865 Divakar Sharma<sup>1</sup>† , Anjali Garg<sup>2</sup> , Manish Kumar<sup>2</sup> , Faraz Rashid<sup>3</sup> and Asad U. Khan<sup>1</sup> \*

1 Interdisciplinary Biotechnology Unit, Aligarh Muslim University, Aligarh, India, <sup>2</sup> Department of Biophysics, University of Delhi, New Delhi, India, <sup>3</sup> SCIEX Pvt. Ltd., Gurgaon, India

The emergence and spread of carbapenem-resistant Klebsiella pneumoniae infections have worsened the current situation worldwide, in which totally drug-resistant strains (bad bugs) are becoming increasingly prominent. Bacterial biofilms enable bacteria to tolerate higher doses of antibiotics and other stresses, which may lead to the drug resistance. In the present study, we performed proteomics on the carbapenemresistant NDM-4-producing K. pneumoniae clinical isolate under meropenem stress. Liquid chromatography coupled with mass spectrometry (LC–MS/MS) analysis revealed that 69 proteins were down-regulated (≤0.42-fold change) under meropenem exposure. Within the identified down-regulated proteome (69 proteins), we found a group of 13 proteins involved in flagellar, fimbriae, and pili formation and their related functions. Further, systems biology approaches were employed to reveal their networking pathways. We suggest that these down-regulated proteins and their interactive partners cumulatively contribute to the emergence of a biofilm-like state and the survival of bacteria under drug pressure, which could reveal novel mechanisms or pathways involved in drug resistance. These down-regulated proteins and their pathways might be used as targets for the development of novel therapeutics against antimicrobial-resistant (AMR) infections.

Keywords: Klebsiella pneumoniae (NDM-4), proteomics, bioinformatics, pathway enrichment, biofilm, carbapenem resistance

## INTRODUCTION

Klebsiella pneumoniae is a gram-negative bacteria of the family Enterobacteriaceae. In clinical settings, the emergence and spread of drug-resistant K. pneumoniae are worsening the medical situation worldwide. Carbapenems have been considered the last line of defense in the treatment of drug-resistant infections (Paterson, 2000; Paterson and Bonomo, 2005). Interrupted use of

**Abbreviations:** CLSI, Clinical and Laboratory Standards Institute; ESBLs, extended spectrum beta-lactamases; LB, Luria– Bertani; MIC, minimum inhibitory concentration; STRING, Search Tool for the Retrieval of Interacting Genes/Proteins.

carbapenem during the course of treatment leads to the emergence of carbapenem-resistant infections. Carbapenemases are produced that cleave or hydrolyze the carbapenem drugs and contribute to carbapenem resistance. Carbapenemase overproduction and porin deficiency are the two major causes of carbapenem resistance (Ambler et al., 1991; Martínez-Martínez et al., 1999; Jacoby et al., 2004; Loli et al., 2006). Several explanations have been put forward to explain the mechanisms of carbapenem resistance, but our information is as yet incomplete or fragmentary.

Biofilm formation is among the mechanisms known to be responsible for microbial drug resistance. During biofilm formation, bacteria first become sessile and then colonize and grow up from surfaces. The biofilm protects the bacteria from various stresses like altered pH, osmolarity, and nutrient scarcity (Costerton and Lewandowski, 1995; Fux et al., 2005; McCarty et al., 2012) and blocks the entry of drugs to the bacterial communities (Costerton et al., 1999; Stewart and William Costerton, 2001; Sharma et al., 2019c). In the first step of biofilm formation, bacteria lose their motility and become sessile. We assume that decreased expression of proteins related to motility could lead to biofilm formation and thus might contribute to the development of drug resistance. Comparative proteomics addressing the whole-cell proteins of drug-resistant microbes with or without drug pressures have been reported previously (Lata et al., 2015; Khan et al., 2017; Sharma et al., 2018a; Qayyum et al., 2019; Sharma et al., 2019a). However, little information is available regarding the bacterial proteome related to biofilm, and, to the best of our knowledge, no data has yet been reported related to the proteome of drug-resistant microbes, especially carbapenem-resistant K. pneumonia, in relation to motilitymediated drug resistance.

In this study, we used comparative proteomics and systems biology-based approaches to investigate the correlation of the decreased expression of motility-related proteins (flagellar, fimbriae, and pili) with biofilm formation, which may lead to the development of drug resistance. Proteomics and systems biology approaches are both among the potential strategies for exploring biological problems such as the mechanisms of drug resistance. In the present study, we used liquid chromatography coupled with mass spectrometry (LC–MS/MS) to determine the expression of the motility-related proteome of a carbapenem-resistant K. pneumoniae (NDM-4) clinical isolate under meropenem stress. The results of this study could lead to the exploration of novel therapeutics targets against carbapenem resistance.

## MATERIALS AND METHODS

## Strain Selection and Drug Susceptibility Testing

An NDM-4-encoding carbapenem-resistant K. pneumoniae clinical isolate (AK-97) was selected for this study. This was reported in our earlier study, which showed its presence in the NICU of a northern Indian Hospital (Ahmad et al., 2018). Drug susceptibility testing (DST) against the drug meropenem was carried out via the micro-dilution method according to CLSI guidelines (Wayne, 2014).

## Culture Scaling, Drug Induction, and Protein Sample Preparation

A single colony of K. pneumoniae was inoculated in LB broth and kept at 37◦C at 220 rpm, and a sub-MIC (32 µg/ml) of meropenem was used for induction in a 200 ml culture flask. Bacteria were grown up to the exponential phase (OD<sup>600</sup> = 0.8), and cells were harvested by centrifugation at 8000 × g for 8 min at 4 ◦C. The cells were washed with normal saline and re-suspended in a lysis buffer [50 mM Tris–HCl containing 10 mM MgCl2, 0.1% sodium azide, 1 mM phenyl-methyl-sulfonyl-fluoride (PMSF), and 1 mM ethylene glycol tetra-acetic acid (EGTA); pH 7.4] at a concentration of 1 g wet weight per 5 ml. Cell lysis was performed by intermittent sonication with a sonicator with the power at 35% amplitude (Sonics & Materials Inc., Newtown, CT, United States) for 10 min at 4◦C. Further, the homogenate was centrifuged at 12,000 × g for 20 min at 4◦C, and the supernatant was precipitated overnight at −20◦C by adding cold acetone in excess (1:4) (Lata et al., 2015; Sharma and Bisht, 2016; Sharma et al., 2019a). The precipitated protein was collected by centrifugation (12,000 × g, 20 min), allowed to air dry, and then suspended in an appropriate volume of protein-dissolving buffer. The protein concentration was estimated using the Bradford (1976) assay. All of the experiments were replicated biologically and technically.

## Separation and Identification of the Proteome by nanoLC-TripleTOF 5600 MS

Equal concentrations of protein samples were trypsinized, and digested proteins were analyzed using a TripleTOF 5600 MS (AB Sciex, Foster City, CA, United States) equipped with an Eksigent MicroLC 200 system (Eksigent, Dublin, CA, United States) with an Eksigent C18 reverse-phase column (150 × 0.3 mm, 3 µm, 120 Å) (Sharma et al., 2019a). For protein identification, spectral libraries were generated using information-dependent acquisition (IDA) mode after injecting 2 gm of tryptic digest on the column using an Eksigent NanoLC-UltraTM 2D Plus system coupled with a SCIEX Triple TOF <sup>R</sup> 5600 system fitted with a NanoSpray III source. The samples were loaded on the trap (Eksigent Chrom XP 350 µm × 0.5 mm, 3 µm, 120 Å) and washed for 30 min at 3 µl/min. A 120 min gradient in multiple steps (ranging from 5 to 50% acetonitrile in water containing 0.1% formic acid) was set up to elute the peptides from the ChromXP 3-C18 (0.075 × 150 mm, 3 µm, 120 Å) analytical column. Technical replicates of the nanoLC-TripleTOF 5600 MS experiments were performed.

## Sequential Window Acquisition of all Theoretical Fragment Ion Spectra (SWATH) Analysis for Label-Free Quantification

For label-free quantification (SWATH analysis), data-dependent analysis (DDA) mode was applied for both samples to generate high-quality spectral ion libraries by operating the mass spectrometer with specific parameters (Sharma and Bisht, 2016).

In the SWATH acquisition method, the Q1 transmission window was set to 12 Da from the mass range 350–1250 Da. A total of 75 windows were acquired independently with an accumulation time of 62 ms, along with three technical replicates for each of the sets. The total cycle time was kept constant at <5 s. Protein PilotTM v. 5.0 was used to generate the spectral library. For labelfree quantification, peak extraction and spectral alignment were performed using PeakView <sup>R</sup> 2.2 Software with the parameters set as follows: number of peptides, 2; number of transitions, 5; peptide confidence, 95%; XIC width, 30 ppm; XIC extraction window, 3 min. The data were further processed in MarkerView software v. 1.3 (AB Sciex, Foster City, CA, United States) for statistical data interpretation. In MarkerView, the peak area under the curve (AUC) for the selected peptides was normalized by the internal standard protein (beta-galactosidase) spike during SWATH accumulation. The results were extracted as three output files containing the AUC of the ions, the summed intensity of peptides for protein, and the summed intensity of ions for the peptide. All SWATH acquisition data were processed using SWATH Acquisition MicroApp 2.0 in PeakView <sup>R</sup> Software.

## Data Analysis

Data were processed with Protein Pilot Software v. 5.0 (AB Sciex, Foster City, CA, United States) utilizing the Paragon and Progroup Algorithm. The analysis was done using the tools integrated into Protein Pilot at a 1% false discovery rate (FDR) with statistical significance. In brief, the UniProt database searched for the K. pneumoniae taxonomy, which was download from the database in July 2018. The download included total combined (reviewed and un-reviewed) entries of 409,060 proteins. We used cRAP analysis to identify the proteins commonly found in proteomics experiments (unavoidable contamination) of protein samples. E. coli betagalactosidase (BGAL\_ECOLI-[P00722]) was used as a molecular weight marker to calibrate the system for sample acquisition. The internal standard was used for the normalization of statistical parameters. We exported the label-free quantified data and imported them into MarkerView software V1.3 to obtain statistical data for further interpretation. Triplicate data for each sample were normalized using the internal protein (betagalactosidase) area, which was initially spiked in the samples. After normalization, principal component analysis (PCA) was performed to check the possible correlated variables within the group. We plotted a volcano curve to determine the statistical significant fold change versus p-value for the control and test. Proteins with a significant fold change < 0.42 were considered down-regulated proteins. Peak extraction and spectral alignment were performed using PeakView software (v. 2.2, AB Sciex, Foster City, CA, United States) with the following parameter settings: number of peptides per protein, 5; number of transitions per peptide, 6; selected peptide confidence, 1% FDR; XIC width, 30 ppm; XIC extraction window, 3 min.

## Gene Ontology Term Assignment and Analysis

Klebsiella pneumonia subsp. pneumoniae (strain ATCC 700721/MGH 78578) was used as a reference strain to carry out the functional studies. The proteins obtained from LC– MS/MS were firstly aligned to the reference strain proteome. Reference strain proteins that showed alignment identity of ≥50% over 80% of the sequence length of the reference strain protein were considered as homologs of the LC– MS/MS identified proteins. The Gene Ontology (GO) terms associated with the reference strain protein were used for the functional annotation. We used the slim version of GO terms, which were obtained from the Gene Ontology Consortium<sup>1</sup> (Camon et al., 2003).

<sup>1</sup>http://www.geneontology.org/

TABLE 1 | Details of the down-regulated proteome (flagella-, fimbriae-, and pili-related proteins) under meropenem stress in Klebsiella pneumonia clinical isolates (NDM-4).


TABLE 2 | Functional analysis of down-regulated genes associated with Klebsiella pneumonia subsp. pneumonia (strain ATCC 700721/MGH 78578).


(Continued)

#### TABLE 2 | Continued

fmicb-10-02865 December 16, 2019 Time: 16:17 # 5


TABLE 3 | List of Klebsiella pneumonia sp. proteins mapped on E. coli K12 substr.DH10B.


## Protein–Protein Interaction Network Integration

To find the interaction partner(s) of down-regulated proteins, protein–protein interaction (PPI) information were obtained from the STRING database v10.0<sup>2</sup> and Cytoscape (version 3.6.1) (Shannon et al., 2003; Sharma and Bisht, 2017a,b,c; Sharma et al., 2018b, 2019b; Sharma and Khan, 2018). The PPI information provided by the STRING database has been established by experimental studies or by genomic analyses like domain fusion, phylogenetic profiling high-throughput experiments, co-expression studies, and gene neighborhood analysis. In the present study, only interactions with a score of ≥0.4 were used.

## RESULTS

## Identification of Proteins by LC–MS/MS

In this study, we grew the carbapenem-resistant isolate at meropenem sub-MIC 32 mg/L. Further, we identified the down-regulated proteome of the same bacteria by LC–MS/MS using a SWATH workflow; 1156 proteins

<sup>2</sup>http://www.string-db.org/

were quantified at 1% FDR with statistical significance as per the log fold change vs. p-value. Among them, 69 proteins were down-regulated (<0.42 log fold change vs. p-value) and are tabulated in the **Supplementary Material** (**Supplementary Table S1**).

In the present work, our main focus was on motility-related proteins. After critical analysis of the 69 down-regulated proteins, we found 13 proteins (around 19%) that belonged to the flagella-, fimbriae-, and pili-related protein functional groups (**Table 1**). These proteins are flagellar motor switch protein FliG, flagellar hook protein FlgE, negative regulator of flagellin synthesis FlgM, putative fimbriae major subunit StbA, flagellar hook-associated protein 2, chaperone FimC protein, flagellar biosynthesis protein FlgN, flagellar basal body protein, conjugal transfer protein TraC, fimbrial subunit type 1, chemotaxis regulator-transmits chemoreceptor signals to flagellar motor components CheY, and flagellin, all of which are involved in motility and its supporting processes.

On the basis of the parameters described in the section "Materials and Methods," we were able to map 67 out of the 69 down-regulated proteins on the proteome of the ATCC 700721/MGH 78578 strain of K. pneumoniae (**Supplementary Table S2**). Our GO results for down-regulated genes also show that most down-regulated proteins were involved in

nitrogen and other small molecule metabolism, ion binding, and oxidoreductase activity (**Table 2**).

## Protein Network Analysis

To construct the PPI network, the down-regulated proteins with motility-related functions were annotated using E. coli K12 strain DH10B homologs, as tabulated in **Table 3**. Among them, few proteins showed no hits in E. coli, and the rest of the proteins interacted with other proteins to make an interactome; this was visualized through Cytoscape (version 3.6.1) (**Figure 1**). Interacting proteins were color-coded by their functions: biofilm formation, blue; chemotaxis proteins, yellow; flagellar assembly, green. Square red nodes indicate down-regulated proteins.

## DISCUSSION

The development of carbapenem-resistant K. pneumoniae has worsened the medical situation around the globe. The emergence of carbapenem resistance is usually due to the overexpression of carbapenemases and loss of porins. Recently, we reported a cluster of over-expressed proteins in carbapenemresistant K. pneumoniae under meropenem pressure. These could be responsible for the drug resistance and belong to various categories such as the protein translational machinery complex, DNA/RNA modifying enzymes or proteins, proteins involved in carbapenem cleavage, modification, and transport, and energy metabolism- and intermediary metabolism-related

proteins (Sharma et al., 2019a). Therefore, we suggested that they could be potential targets for the development of novel therapeutics against this resistance. Biofilm formation is also one of the mechanisms that leads to the development of drug resistance. In the present study, we have found a group flagellar, fimbriae, and pili proteins that are down-regulated under meropenem stress (sub-MIC). We hypothesize that the downregulation of these proteins under meropenem stress makes the bacteria sessile or non-motile, leading to the emergence of a biofilm-like state that could contribute to carbapenem resistance in K. pneumoniae (NDM-4). Earlier microarray analysis of K. pneumoniae also reported the down-regulation of genes related to nitrogen metabolism, porin genes, and some membrane-associated proteins in association with the antibiotic resistance phenomenon (Doménech-Sánchez et al., 2006). The significance of the aforementioned pathways in antibioticevading mechanisms is also highlighted in several other reports (Yeung et al., 2011; Piek et al., 2014).

## A Hub of Flagellar, Fimbriae, and Pili Proteins Could Regulate Resistance

A group of flagellar, fimbriae, and pili proteins involved in the formation of the flagellar machinery complex and the regulation of motility processes were found to be down-regulated in meropenem-induced carbapenem-resistant K. pneumoniae clinical strains. These proteins are flagellar motor switch protein FliG, flagellar hook protein FlgE, negative regulator of flagellin synthesis FlgM, putative fimbriae major subunit StbA, flagellar hook-associated protein 2, chaperone FimC protein, flagellar biosynthesis protein FlgN, flagellar basal body protein FlgF, conjugal transfer protein TraC, fimbrial subunit type 1, chemotaxis regulator-transmits chemoreceptor signals to flagellar motor components CheY, and flagellin.

The observed expression down-regulation of flagella-, fimbriae-, and pili-related proteins provides clues to a novel mechanism of drug resistance. Flagellin, flagellar biosynthesis protein FlgN, and FlgM, respectively, cumulatively maintain equilibrium in the biosynthesis of flagella. Flagellin protein is part of the structural component of flagella, and biosynthesis of flagella is favored by flagellar biosynthesis protein FlgN (Bennett et al., 2001). FlgM is a negative regulator that switches off flagellar transport. FlgM can be exported from the cell via a flagellum, and the transport occurs only after the completion of hook formation (Hughes et al., 1993). This unique regulatory mechanism further postpones flagellin synthesis in the cell. Cumulatively, the expression of these proteins is involved in the flagella formation, regulation, and motility of the bacteria. Therefore, we suggest that, under meropenem pressure, downregulation of these proteins might make the bacteria sessile, which is the first step in biofilm-formation. Therefore, we assume that down-regulation of these proteins could create a biofilm-like state, which ultimately leads to the drug resistance.

Flagella are composed of three different parts: a filament (helical and long), hook (a curved and short structure), and basal body (a complex structure with a central rod and a series of rings). Flagellar motor switch protein (FliG), flagellar hook protein FlgE, flagellar hook-associated protein 2, and flagellar basal body protein FlgF together make up part of the flagellar motor switch complex (FliG, FliN, and FliM), which is involved in bacterial motility, after receipt and transduction of the signal by chemotaxis (Djordjevic and Stock, 1998). This is a complex apparatus that senses the signal from the chemotaxis sensory signaling system and is transduced into motility. Chemotaxis response regulator CheY transmits chemoreceptor signals to flagellar motor components and is believed to be the "on" switch that directly induces tumbles in the swimming pattern (Robinson et al., 2000). Physical interactions of CheY and switch proteins have not been reported. Chemotactic stimuli change the association of the CheY signal protein with the distal FliMNCFliN C ring (Dyer et al., 2009; Sarkar et al., 2010). In the present study, down-regulation of the chemotaxis response regulator (CheY) subsequently down-regulates signal transmission to the flagellar motor components, which may act as an "off " switch and make the bacteria sessile or non-motile. Further, it might lead to a biofilm-like state and could contribute to the drug resistance.

Putative fimbriae major subunit StbA, Fimbrial subunit type 1, conjugal transfer protein TraC, and chaperone FimC protein are involved in pillus organization, fimbriae formation, and their associated assemblies (Johnson and Clegg, 2010). FimC protein acts as a periplasmic pilin chaperone that not only protects the bacteria under stress through chaperone-mediated folding but is also involved in pillus formation (Klemm, 1992). In the periplasm, the FimC chaperone binds to the major and minor structural components and protects them from degradation. Conjugal transfer protein TraC, encoded by a gene, traC, presents on plasmid and is involved in the conjugation process as well as pili formation (Winans and Walker, 1985; Kado, 1994), leading

FIGURE 2 | Proposed model based on the flagellar-, fimbriae-, and pili-related proteome suggested the mechanism of carbapenem-resistance in K. pneumoniae (NDM-4) clinical isolates.

to the transfer of drug-resistant plasmid to bacteria. These down-regulated proteins (flagellar, fimbriae, and pili proteins) form a hub of proteins in the PPI network, which indicates their important role in flagellar, fimbriae, and pili assemblies, signaling through chemotaxis proteins, and biofilm formation (**Figure 1**). In this study, the down-regulation of fimbriae-, pili-, and conjugative-related proteins leads to the creation of the biofilm-like state, which may contribute to drug resistance. On the basis of the flagella-, fimbriae-, and pili-related proteome, we propose a model (**Figure 2**) that suggests the potential path or mechanism of carbapenem resistance in K. pneumoniae (NDM-4) clinical isolates. Overall this group of down-regulated proteins and their interactive protein partners cumulatively make a hub that leads to the formation of a biofilm-like scenario and could contribute to meropenem resistance.

## CONCLUSION

In brief, the present study focused on the down-regulated proteome of carbapenem-resistant K. pneumoniae clinical isolate (NDM-4) under meropenem pressure through proteomics and systems biology approaches. A group of down-regulated proteins was identified that belongs to the flagellar, fimbriae, and pili proteins. Therefore, we suggest that these proteins and their interactive protein partners cumulatively lead to the bacteria becoming sessile, which further creates a biofilm-like state and could contribute to the survival of bacteria under meropenem pressure, which might reveal a novel mechanism of drug resistance. Further research on these motility-related protein targets and their pathways may lead to the development of novel therapeutics against the worsening situation of drug resistance.

## REFERENCES


## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## AUTHOR CONTRIBUTIONS

DS designed the concept, and experimented and wrote the manuscript. AG and MK carried out the systems biology analysis. FR provided support in the LC–MS experiments and analysis. AK designed and guided the study and finalized the manuscript. All authors approved the final manuscript.

## FUNDING

Science and Engineering Research Board (SERB) is gratefully acknowledged for providing fellowship and funds to DS (SERB-N-PDF/2016/001622) to work at IBU-AMU Aligarh. The authors also acknowledge SCIEX Pvt. Ltd., Gurgaon, India, for data acquisition and proteomics facility support.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02865/full#supplementary-material

clinical isolates of Klebsiella pneumoniae by DNA microarray analysis. Clin. Microbiol. Infect. 12, 936–940. doi: 10.1111/j.1469-0691..01470.x



interactomic approach. Front. Cell. Infect. Microbiol. 7:240. doi: 10.3389/fcimb. 2017.00240


**Conflict of Interest:** FR was employed by SCIEX Pvt. Ltd.

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 Sharma, Garg, Kumar, Rashid and Khan. 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.

# Effect of Dietary Copper on Intestinal Microbiota and Antimicrobial Resistance Profiles of Escherichia coli in Weaned Piglets

Yiming Zhang1,2, Jian Zhou<sup>2</sup> , Zhenglin Dong1,2, Guanya Li1,2, Jingjing Wang<sup>2</sup> , Yikun Li<sup>2</sup> , Dan Wan<sup>2</sup> \*, Huansheng Yang<sup>1</sup> and Yulong Yin1,2

<sup>1</sup> Hunan International Joint Laboratory of Animal Intestinal Ecology and Health, Laboratory of Animal Nutrition and Human Health, College of Life Sciences, Hunan Normal University, Changsha, China, <sup>2</sup> Hunan Province Key Laboratory of Animal Nutritional Physiology and Metabolic Process, Key Laboratory of Agro-ecological Processes in Subtropical Region, National Engineering Laboratory for Pollution Control and Waste Utilization in Livestock and Poultry Production, Institute of Subtropical Agriculture, Chinese Academy of Sciences, Changsha, China

#### Edited by:

Leonardo Neves de Andrade, University of São Paulo, Brazil

#### Reviewed by:

Yulan Liu, Wuhan Polytechnic University, China Evelyne Forano, INRA Centre Auvergne – Rhône-Alpes, France

> \*Correspondence: Dan Wan w.dan@isa.ac.cn

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 15 July 2019 Accepted: 19 November 2019 Published: 17 December 2019

#### Citation:

Zhang Y, Zhou J, Dong Z, Li G, Wang J, Li Y, Wan D, Yang H and Yin Y (2019) Effect of Dietary Copper on Intestinal Microbiota and Antimicrobial Resistance Profiles of Escherichia coli in Weaned Piglets. Front. Microbiol. 10:2808. doi: 10.3389/fmicb.2019.02808 Copper is an essential microelement for animals, and not only it has been used as a feed additive at pharmacological doses in swine production to improve growth performance, but it also has an effect on intestinal microbes by enhancing host bacterial resistance. However, there are few reports on the effects of pharmacological doses of copper on intestinal microorganisms and the antimicrobial resistance profiles of pathogenic bacteria, such as Escherichia coli, in pigs. Therefore, this study aimed to investigate the effects of pharmacological doses of copper on the microbial communities in the hindgut and the antimicrobial resistance profiles of E. coli in weaned piglets. Twenty-four healthy weaned piglets aged 21 ± 1 days and with an average weight of 7.27 ± 0.46 kg were randomly divided into four groups. The control group was fed a basal diet, while the treatment groups were fed a basal diet supplemented with 20, 100, or 200 mg copper/kg feed, in the form of CuSO4. Anal swabs were collected at 0, 21, and 42 days of the trial, and E. coli was isolated. Meanwhile, the contents of the ileum and cecum from the control and 200 mg copper/kg feed groups were collected at 21 and 42 days for microbial community analysis and E. coli isolation. All isolated E. coli strains were used for antimicrobial resistance profile analysis. A pharmacological dose of copper did not significantly change the diversity, but significantly affected the composition, of microbial communities in the ileum and cecum. Moreover, it affected the microbial metabolic functions of energy metabolism, protein metabolism, and amino acid biosynthesis. Specifically, copper treatment increased the richness of E. coli in the hindgut and the rates of E. coli resistance to chloramphenicol and ciprofloxacin. Moreover, the rate of E. coli resistance to multiple drugs increased in the ileum of pigs fed a pharmacological dose of copper. Thus, a pharmacological dose of copper affected the composition of the microbial community, increased the antimicrobial resistance rates of intestinal E. coli, and was most likely harmful to the health of piglets at the early stage after weaning.

Keywords: copper, piglet, microbial community, Escherichia coli, antimicrobial resistance

## INTRODUCTION

fmicb-10-02808 December 13, 2019 Time: 16:6 # 2

Copper (Cu) is an essential trace element for animals, with many biological functions, including iron metabolism, immunity, protection from oxidative stress, and improvement in the activity of digestive enzymes (Gipp et al., 1967; Harris, 1991; Amachawadi et al., 2010; Braune et al., 2010; Huang et al., 2015). Feeding weaned piglets a high Cu diet, usually at pharmacological doses (150–250 mg Cu/kg feed) as CuSO4, Cu-AA, or Cu hydroxychloride, may improve their average daily weight gain and feed conversion ratio and reduce the frequency of diarrhea (Smith et al., 1997; Cromwell et al., 1998; Hill et al., 2000; Perez et al., 2011; Bikker et al., 2016; Espinosa et al., 2017).

However, as the Cu requirement of piglets is 4–6 mg Cu/kg feed (Okonkwo et al., 1979), the excess Cu results in an increase in the Cu content of pig slurry and, consequently, an accumulation of Cu in soils, which poses a high environmental risk (Jondreville et al., 2002). Meanwhile, excess amounts of Cu in the gut can affect the intestinal microbial community and result in reduced counts of streptococci, ureolytic bacteria, and total anaerobes in the feces (Fuller et al., 1960; Kellogg et al., 1966; Varel et al., 1987). Recently, it has been found that feeding piglets with 300 mg Cu/kg feed affected the abundance of Clostridium genera and decreased the relative abundance of butyrate-producing bacteria, such as Acidaminococcus, Coprococcus, and Roseburia in fecal microbiota (Zhang et al., 2019). It also altered the gut microbiota and decreased the counts of Lactobacilli and Enterobacteriaceae in the cecum and ileum of weaned piglets (Ole et al., 2005; Mei et al., 2010). However, the effects on gut microbiota composition and function when pharmacological amounts of Cu are added to feed have not been thoroughly investigated.

In pigs, enteric bacteria have been shown to develop Cu resistance by transferable tcrB in enterococci (Hasman and Aarestrup, 2005) or by the multicopper oxidase, PcoA, and the metal sponge, PcoE, in enterobacteria (Djoko et al., 2008; Zimmermann et al., 2012). Resistance to Cu in bacteria, particularly in enterococci, is often associated with resistance to antimicrobial drugs, such as macrolides and glycopeptides (Hasman et al., 2006; Yazdankhah et al., 2014), while Cu resistance in Escherichia coli is negatively associated with both tetA and bla (CMY-2) (Agga et al., 2014). Nevertheless, the potential connection between Cu resistance and antimicrobial drug resistance, particularly in the enterobacterium, E. coli, has not been fully elucidated. Therefore, the objective of the present study was to investigate the effects of different doses of Cu, including a pharmacological dose, on hindgut microbiota diversity, composition, and metabolic function in weaned piglets after 3 and 6 weeks of supplementation. Based on the results of microbiota analysis, E. coli was isolated from feces at 0, 21, and 42 days of the feeding period, and the drug resistance profile of the isolates was assessed for the first time.

## MATERIALS AND METHODS

## Animals, Feeding, and Sampling

Animal experiments were approved by the Animal Protocol Review Committee of the Institute of Subtropical Agriculture, Chinese Academy of Sciences (Changsha, China), and all experiments were performed according to the institutional animal welfare requirements. Forty-eight healthy weaned piglets, aged 21 ± 1 days (8.26 ± 0.17 kg), were randomly assigned to four groups, with six barrows and six females in each group. The piglets were housed individually and given free access to drinking water and food. The control group was fed a basal diet (**Table 1**), and the treatment groups were fed a basal diet supplemented with 20, 100, or 200 mg Cu/kg feed, as CuSO4·5H2O, for 6 weeks. Anal swabs were collected from piglets at days 0, 21, and 42 of the experimental period, for E. coli isolation and antimicrobial drug resistance detection. At days 21 and 42, piglets (n = 6) in the control and 200 mg Cu group were anesthetized with Zoletil (15 mg/kg BW, i.m.) and euthanized. The ileal and cecal contents were collected under sterile conditions, using a 15-ml sterile conical centrifuge tube and submitted for 16S rRNA sequencing, E. coli isolation, and antimicrobial drug resistance evaluation.

## 16S rDNA Sequencing

The ileal and cecal contents from piglets in the control (without Cu supplementation) group and the group receiving a pharmacological dose of Cu (200 mg Cu/kg feed, n = 6 per treatment) were collected for microbiota analysis, as described previously (Zhou et al., 2018; Xiong et al., 2019). Briefly, total bacterial DNA was extracted using a QIAamp DNA Stool Mini Kit (Qiagen, Hilden, Germany), and the V3–V4 hypervariable region of the microbial 16S rRNA genes were sequenced using the primers, 341F: 5<sup>0</sup> -CCTAYGGGRBGCASCAG-3<sup>0</sup> and 806R: 5 0 -GGACTACNNGGGTATCTAAT-3<sup>0</sup> ; Illumina adaptors; and molecular barcodes. Libraries were prepared using a TruSeq DNA PCR-Free Sample Preparation Kit (Illumina Inc., San Diego, CA, United States) and were assessed using a Qubit 2.0 Fluorometer (Thermo Scientific, Madison, WI, United States) and a Bioanalyzer 2100 system (Agilent Technologies, Santa Clara, CA, United States). Libraries were sequenced on a HiSeq 2500 platform (Illumina Inc., San Diego, CA, United States). Raw 16S rDNA sequences were assembled using the QIIME (v1.9.0) and FLASH software packages. Operational taxonomic units (OTUs) were analyzed using UPARSE (v7.0.1001), and



<sup>a</sup>Feed analyses were performed according to the AOAC guidelines (2004). <sup>b</sup>The vitamin–mineral premix provided the following per kilogram of feed: 100 mg of antioxidant, 100 mg of preservatives, 100 mg of choline chloride, 6,000 IU of vitamin A, 3,000 IU of vitamin D3, 20 IU of vitamin E, 1.8 mg of vitamin K3, 2.0 mg of thiamine, 6.0 mg of riboflavin, 4.0 mg of pyridoxine, 0.02 mg of vitamin B12, 26.0 mg of niacin, 18.0 mg of pantothenic acid, 3.2 mg of folic acid, 0.4 mg of biotin, 100 mg of Fe as FeSO4· 7H2O, 100 mg of Zn as ZnSO4·H2O, 50 mg of Mn as MnSO4·H2O, 1.2 mg of I as KI, and 0.30 mg of Se as Na2SeO3.

high-quality sequences were aligned against the SILVA reference database 1 and clustered into OTUs at a 97% similarity level using the UCLUST algorithm 2. Each OTU was assigned to a taxonomic level using the Ribosomal Database Project Classifier program v2.203. The assembled sequences were then submitted to the NCBI Sequence Read Archive (No. PRJNA551517) for open access.

## E. coli Isolation and Identification

Samples were collected with sterile swabs, and the swab sample was immediately inserted into a 15-ml sterile centrifuge tube containing 5 ml of phosphate-buffered saline. Then, 10 µl of each sample was swabbed onto MacConkey agar and incubated at 37◦C for 18–24 h. Three red- or pink-colored colonies were picked per sample and inoculated on eosin methylene blue agar plates at 37◦C for 18–24 h to obtain pure cultures. A single colony with a black metallic luster was picked and cultured in Luria–Bertani (LB) liquid medium at 37◦C for 8– 12 h. Bacterial smears were from the LB liquid culture, and Gram staining and microscopic examination were performed as described (Feyzioglu et al., 2014; Usaini et al., 2015). LB liquid cultures containing Gram-negative bacilli were swabbed onto eosin methylene blue agar plates and incubated at 37◦C for 18– 24 h. A single colony with a black metallic luster was picked, inoculated on a nutrient agar plate, and cultured at 37◦C for 12– 18 h. Single colonies from the nutrient agar plate were inoculated in biochemical identification tubes (Hangzhou Microbial Reagent Co., Ltd., Hangzhou, China) for lactose fermentation, and indole and citrate utilization tests, according to the manufacturer's instructions. Isolates with a Lac+/Ind+/Cit− phenotype were identified as E. coli. Finally, the 16S rRNA gene of isolated E. coli was amplified by PCR. DNA extraction and PCR amplification were performed as previously described (Blyton et al., 2013). Three oligonucleotide primers, 16E1, 16E2, and 16E3, were designed and used in the PCR assays for the identification of E. coli isolates, as previously described (Tsen et al., 1998). The sequences and locations of these primers are shown in **Supplementary Table S1**. PCR products were analyzed by gel electrophoresis through a 2% agarose gel, as previously described (Tsen et al., 1998) and sequenced by Sangon Biotech Co., Ltd. (Shanghai, China). Homology analysis between the amplified sequences and the E. coli 16S rRNA gene sequences published in GenBank was performed to confirming that the isolates were E. coli. Repetitive element PCR with both ERIC and (CGG)4 primers was also used to discriminate clonal isolates from the same samples (Adamus-Bialek et al., 2009). Isolates producing the same banding pattern were considered to be the same strain.

## Drug-Resistant Phenotype Identification

After PCR validation, the E. coli suspensions were subjected to antimicrobial susceptibility testing on Müller–Hinton agar using the Kirby–Bauer disk diffusion method, in accordance with the Clinical and Laboratory Standards Institute guidelines, using E. coli ATCC 25922 as an internal quality control. The antimicrobial disks used for the test included ampicillin (10 µg/disk), amoxicillin/clavulanic acid (20/10 µg/disk), ceftriaxone (30 µg/disk), meropenem (10 µg/disk), aztreonam (30 µg/disk), gentamicin (10 µg/disk), trimethoprim/sulfamethoxazole (1.25/23.75 µg/disk), amikacin (30 µg/disk), and ciprofloxacin (CIP) (5 µg/disk) (Oxoid Ltd., Hampshire, United Kingdom).

## Statistical Analysis

All statistical analyses were performed using SPSS 20.0 software (SPSS Inc., Chicago, IL, United States). α and β diversity were analyzed using QIIME (v1.7.0). Data for microbial metabolic functions were analyzed by Student's t-test only, due to minor changes in the microbiome data. The differences in isolation rates and antimicrobial resistance of E. coli among the groups were analyzed by Chi-square test. A P < 0.05 was considered statistically significant.

## RESULTS

## Bodyweight and Growth Performance

The average bodyweights of the weaned piglets at 21 ± 1 days in the control and Cu-supplemented groups were 8.13 ± 0.22 and 8.40 ± 0.26 kg, respectively. There were no significant differences in average daily gain, average daily feed intake, or feed conversion ratio between the control and Cu-supplemented groups in early (1–3 weeks) and late postweaning stages (4–6 weeks, **Table 2**).

## Microbiota Diversity in Intestinal Contents

fmicb-10-02808 December 13, 2019 Time: 16:6 # 4

There were no significant differences between treatments for the indices of α diversity, including observed\_species, Shannon, Simpson, Chao1, ACE, and PD\_whole\_tree indices, of ileal (**Supplementary Table S2**) and cecal microbiota (**Supplementary Table S3**). In addition, the weighted principal

TABLE 2 | The effect of Cu levels on the growth performance of piglets during the first 3 weeks and the second 3 weeks of feeding.


<sup>a</sup>ADG, average daily gain; ADFI, average daily feed intake; FCR, feed conversion ratio.

coordinate analysis plots of ileal (**Figure 1A**) and cecal microbiota (**Figure 1C**) verified that there were few differences in microbial communities between Cu treatments.

## Microbiota Composition of Intestinal Contents

Based on an unweighted pair-group method with arithmetic mean analysis, Firmicutes, Proteobacteria, Tenericutes, and Bacteroidetes were found to be the dominant bacteria in the ileal microbiome (**Figure 1B**), while Firmicutes, Bacteroidetes, Cyanobacteria, Proteobacteria, and Spirochaetes were the dominant bacteria in the cecal microbiome (**Figure 1D**). Furthermore, the linear discriminant analysis effect size method was used to determine the variation in microbial composition between groups. Minimal changes were observed after short-term feeding (3 weeks), with differences in abundance of the genera Sarcina, Pasteurella, Veillonella, Streptococcus, Mycobacterium, Lactococcus, and Sphaerochaeta in the ileal microbiome (**Figure 2A**) and in the family, Acetobacteraceae, and the genera, Veillonella, Pseudoflavonifractor, Defluviitaleaceae, Oribacterium, Ruminococcaceae (UCG008, 014), Lachnospiraceae (UCG003), and Phascolarctobacterium in the cecal microbiome (**Figure 2B**). Meanwhile, after a longer time of feeding (6 weeks) with Cu supplementation at a pharmacological dose, marked differences

were found in the abundance of the family, Clostridiaceae, the phylum Firmicutes, the genus, Sarcina, Lactobacillus paralimentarius, and swine manure bacterium in the ileal microbiome (**Figure 2C**), and the genera, Lactobacillus and Sporolactobacillus and E. coli in the cecal microbiome (**Figure 2D**). A heat map of the 40 most different genera showed the taxonomic distributions among groups after 3 and 6 weeks of treatment (**Supplementary Figure S1**). Next, the impact of Cu treatment on aerobic, anaerobic, mobile element-containing, facultatively anaerobic, biofilm-forming, Gram-negative, Grampositive, potentially pathogenic, and stress-tolerant bacteria were predicted via BugBase (Zhou et al., 2019). No significant differences were found to occur as a result of Cu treatment (**Supplementary Figure S3**).

## Predicted Metabolic Functions in the Gut Microbiota

Subsequently, to further investigate the changes in the function of the gut microbiota, the metagenome was generated, based on 16S rDNA sequencing results, using Picrust. The Kyoto Encyclopedia of Genes and Genomes annotation results were then submitted for principal component analysis. Microbial metabolism-related pathways at Kyoto Encyclopedia of Genes and Genomes level 3 were specifically filtered. No obvious segregation of ileal and cecal microbiomes was found in the third week (**Supplementary Figures S2A,B**). Furthermore, a t-test showed that Cu supplementation (200 mg Cu/kg feed) increased the abundance of microbiota related to the transcriptional machinery and tuberculosis in the ileal microbiome, while an increase in the abundance of microbiota related to methane metabolism was observed in the cecal microbiome at the third week (**Supplementary Figures S2C,D**). An obvious segregation was found in the ileal microbiome in the sixth week (**Figure 3A**), but there was still no obvious segregation in the cecal microbiome at this time point (**Figure 3B**). Cu supplementation decreased the abundance of microbiota related to energy metabolism, lysine biosynthesis, amino acid metabolism (alanine, aspartate, glutamate, histidine, glycine, serine, and threonine), butanoate metabolism, carbon-fixation pathways in prokaryotes, terpenoid backbone biosynthesis, nitrogen metabolism, benzoate degradation, and the metabolism of cofactors and vitamins in the ileal microbiome (**Figure 3C**). Meanwhile, Cu supplementation increased the abundance of microbiota related to amino acid biosynthesis (valine, leucine, isoleucine, and lysine), lipid biosynthesis protein, amino acid metabolism (glycine, serine, and threonine), and C5-branched dibasic acid metabolism and decreased the abundance of microbiota related to peptidases in the cecal microbiome (**Figure 3D**).

## Antimicrobial Resistance Profiles of E. coli Isolates From Intestinal Contents

Based on microbial composition analysis, we further determined the antimicrobial resistance profiles of E. coil isolated from the ileal and cecal contents. Isolates that were resistant to three

or more classes of antimicrobials were defined as multidrugresistant strains. A total of 119 E. coli isolates from ileal (n = 50) and cecal contents (n = 69) were included in disk diffusion testing, with 20 ileal isolates and 34 cecal isolates from control pigs and 30 ileal isolates and 35 cecal isolates from pigs fed a pharmacological dose of Cu (**Supplementary Table S4**).

Of the 119 isolates, 10 (8.40%) were not resistant to any of the antimicrobial agents tested, 9 (7.56%) were resistant to one antimicrobial agent, and 21 (17.65%) were resistant to two antimicrobial agents. Furthermore, 79 (66.39%) were resistant to at least three antimicrobial agents, and 47 (39.50%) of these were isolated from pigs fed a pharmacological dose of Cu. In particular, seven isolates (5.88%) were resistant to more than six classes of antimicrobials, and six (5.04%) of these were isolated from pigs fed a pharmacological dose of Cu (**Table 3**). The antimicrobial resistance rates of E. coli in the ileum and cecum are summarized in **Table 4**. Resistance to CIP in ileal isolates and resistance to chloramphenicol (C) in cecal isolates were found to significantly respond to Cu dose.

On the basis of these results, anal swabs were collected from four groups of pigs and used for E. coli isolation to investigate the development of resistance in response to Cu dose and feeding period. No significant differences were found in the rate of E. coli response to Cu dosage or feeding period (**Supplementary Table S5**). However, a dose-dependent effect on resistance to C and CIP was found at day 42 (P < 0.05, **Table 5**).

## DISCUSSION

For many years, a pharmacological dose of Cu (150–250 mg/kg feed) has been added to animal diets to improve growth performance. However, as the Cu requirement of pigs is 4– 6 mg/kg feed, the growth-promoting effect of dietary Cu is mostly attributed to the regulation of the composition and metabolism of the intestinal microbiota (Varel et al., 1987). Meanwhile, there are increasing concerns regarding the potential adverse effects of Cu accumulation on the intestinal microbiota, i.e., microbial drug resistance (Hasman and Aarestrup, 2005; Hasman et al., 2006; Djoko et al., 2008; Amachawadi et al., 2011; Zimmermann et al., 2012; Agga et al., 2014; Yazdankhah et al., 2014). The effects of Cu on the counts of coliforms and microbial drug resistance have previously been reported (Djoko et al., 2008;

TABLE 3 | Number of antimicrobial multiresistant E. coli isolates in ileum and cecum.


Zimmermann et al., 2012; Zhang et al., 2019). Thus, the present research focused on microbial diversity, composition, and function, as well as the potential risks of E. coli resistance in the intestine of piglets after pharmacological doses of Cu.

α and β diversity are important features of microbial communities. In our study, α diversity indices, including observed\_species, Shannon, Simpson, Chao1, ACE, and PD\_whole\_tree indices were not affected by dietary Cu. Our results were consistent with those of a previous study of high dietary Cu levels (300 mg/kg) in suckling piglets (Zhang et al., 2019) but in contrast to previous studies reporting that high dietary Cu levels (100–200 mg/kg, as CuSO4) significantly affect microbial species in the ileac, cecal, and colonic chyme of piglets (Varel et al., 1987; Hojberg et al., 2005; Namkung et al., 2006; Mei et al., 2010). A possible reason for these differences is that the method used to investigate the microbial



<sup>1</sup>The differences in antibiotic resistance of E. coli among groups were analyzed by Chi-square test. A P < 0.05 was considered statistically significant. <sup>a</sup>AMP, ampicillin; AMC, amoxicillin/clavulanic acid; CRO, ceftriaxone; MEM, meropenem; ATM, aztreonam; CN, gentamicin; AK, amikacin; CIP, ciprofloxacin; C, chloramphenicol; SXT, trimethoprim/sulfamethoxazole.


TABLE 5 | Antimicrobial resistance rates (95% CI) of Escherichia coli from anal swabs<sup>1</sup> .

<sup>1</sup>The differences in antibiotic resistance of E. coli among groups were analyzed by Chi-square test. a,bMeans in a row without a common superscript letter differ significantly for the same period. <sup>s</sup>CIP, ciprofloxacin; C, chloramphenicol.

flora is inconsistent among the different studies. As most bacterial species in the gut are yet to be cultured, 16S rRNA gene sequencing is recommended for the analysis of gut microbial communities. In the present study, unweighted pair-group method with arithmetic mean analysis showed that Firmicutes, Bacteroidetes, and Proteobacteria were the most abundant phyla in the ileum and cecum, which is similar to previous findings in piglets (Xiong et al., 2019; Zhang et al., 2019). Linear discriminant analysis effect size analysis also showed that the differences included, but were not limited to, the phylum Firmicutes, the family, Clostridiaceae, and the genera, Streptococcus, Lactococcus, and Ruminococcaceae, which were found to be affected in previous studies in which Cu was supplemented at high levels (200–300 mg/kg) as CuSO<sup>4</sup> (Fuller et al., 1960; Kellogg et al., 1966; Varel et al., 1987; Hojberg et al., 2005; Xia et al., 2005; Namkung et al., 2006; Mei et al., 2010; Wang et al., 2012; Song et al., 2013; Zhang et al., 2019). In addition, consistent with previous studies, the addition of high concentrations of Cu to the diet had no significant effect on the populations of aerobic, anaerobic, mobile element-containing, facultatively anaerobic, biofilm-forming, Gram-negative, Gram-positive, potentially pathogenic, or stress-tolerant bacteria (Varel et al., 1987; Hojberg et al., 2005; Mei et al., 2010; Zhang et al., 2019). Moreover, in accordance with the results of Zhang et al. (2019), carbohydrate metabolism by ileal microbes was significantly enhanced by high levels of dietary Cu.

Based on Picrust predictions, our study suggested that Cu supplementation affected the abundance of microbiota related to protein metabolism, amino acid synthesis and metabolism, and carbohydrate metabolism in the intestine. In accordance with this result, multiomics analysis showed that dietary Cu levels affected protein and carbohydrate metabolites and amino acid metabolism. The abundance of the genera, Lachnospiraceae and Ruminococcaceae, was positively correlated with energy metabolism pathways, and the abundance of the genus, Streptococcus, was negatively correlated with amino acid metabolism pathways (Zhang et al., 2019). Thus, the abundance of these species was also altered by Cu supplementation, resulting in effects on energy metabolism and amino acid biosynthesis and metabolism in piglets. In addition, we found that Cu supplementation affected the abundance of cecal microbiota related to

lipid metabolism, suggesting that Cu may affect lipid metabolism in the cecum by regulating the abundance of the cecal microbiota.

Specially, our results showed that the abundance of E. coli in ileal and cecal contents tended to increase with high levels of dietary Cu at 3 weeks, and its abundance significantly increased in the cecum after 6 weeks of a pharmacological dose of Cu. However, there was no significant difference in the rate of isolation of E. coli from anal swabs. This discrepancy may be due to the high abundance of E. coli in rectum. These results were of great interest, since E. coli is the main diarrhea-causing opportunistic pathogen in piglets and also has high a rate of drug resistance. Therefore, E. coli was isolated at the beginning, middle, and end of the trial. As E. coli is present at a high abundance in the intestine and feces, three clones were isolated from all samples for drug resistance analysis.

The Cu resistance system in E. coli mainly consists of the cut and pco systems (Brown et al., 1992, 1995). Other Cu resistance systems, such as the CusCFBA efflux system (Franke et al., 2003), have also been reported. The absorption, efflux, and distribution of Cu are controlled by these systems to balance the content of Cu inside and outside the cell and avoid excessive Cu in the cell (Harris, 1991; Bull and Cox, 1994). Under conditions of high Cu concentration, intestinal Cu-resistant E. coli were selected, and thus, the resistance of E. coli to Cu increased, resulting in an improvement in the adaptability to high levels of Cu and an increase in the abundance of E. coli. However, some previous studies have indicated that Cu inhibits the growth of E. coli in in vitro culture experiments (Ding et al., 2010; Reyes-Jara et al., 2016). This difference may be explained by the fact that the E. coli strain used for in vitro culturing did not possess a Cu resistance system, while the strains of E. coli in the intestinal tract varied, with some being resistant to Cu.

Copper promotes the selection of antibiotic resistance in E. coli (Page, 2003). Hölzel et al. (2012) found that high levels of Cu in liquid pig manure promoted the resistance of E. coli to β-lactam antibiotics. Our study showed that the addition of high levels of Cu in the diet also contributed to the resistance of E. coli to CIP and C in both the intestine and feces. The mechanism responsible for this increased antibiotic

resistance of E. coli after exposure to high levels of Cu was not clear. Previous studies have shown that the Curesistance gene (tcrB), the macrolides-resistance gene (ERM), and the glycopeptide-resistance gene (vanA) of Enterococcus faecium are located on the same plasmid. E. faecium isolates with tcrB have been shown to be resistant to macrolides (erythromycin) and glycopeptides (vancomycin) (Hasman and Aarestrup, 2002; Amachawadi et al., 2010, 2011, 2013). Therefore, Cu resistance may play a synergistic role in the development of antibiotic resistance, through coresistance, cross-resistance, or coregulatory resistance.

In addition, E. coli isolates in the present study were found to be highly multidrug resistant. A previous study has shown that >88% of E. coli isolates from pig feces are multidrug resistant, which is in agreement with the results of our study (Agga et al., 2014). In our study, of the 79 isolates (66.39%) that were resistant to at least three antimicrobial agents, 47 (39.50%) were from piglets treated with a pharmacological dose of Cu, and only 32 isolates (26.89%) were from piglets in the low-Cu group. In particular, six isolates from the group treated with a pharmacological dose of Cu were found to be resistant to six or more classes of antimicrobials, whereas only one isolate resistant to six or more classes of antimicrobials was found in the control group. Our results indicated that a high level Cu supplementation in the diet contributed to an increase in the rate of multidrug resistance of E. coli. The mechanism by which Cu promotes multiple drug resistance in E. coli remains unclear and should be the focus of future studies.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the https://www.ncbi.nlm.nih.gov/sra/PRJNA551517.

## REFERENCES


## ETHICS STATEMENT

The animal study was reviewed and approved by the Animal Protocol Review Committee of the Institute of Subtropical Agriculture, Chinese Academy of Sciences.

## AUTHOR CONTRIBUTIONS

YZ, DW, HY, and YY designed the study. YZ, JZ, ZD, GL, JW, and YL performed the animal feeding trial and sample analysis. YZ and DW wrote and revised the article.

## FUNDING

This work was supported by the National Key R&D Program of China (2016YFD0501201), the National Natural Science Foundation of China (31702127), the Young Elite Scientists Sponsorship Program by CAST (2018QNRC001), and the Hunan Province Key Laboratory of Animal Nutritional Physiology and Metabolic Process (2018TP1031).

## ACKNOWLEDGMENTS

The authors thank Beijing Novogene Bioinformatics Technology Co., Ltd. for microbiota analysis by 16S rRNA sequencing.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02808/full#supplementary-material



mucosal anti-stress capacity in weaned rats. Br. J. Nutr. 120, 241–249. doi: 10.1017/S0007114518001101


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

Copyright © 2019 Zhang, Zhou, Dong, Li, Wang, Li, Wan, Yang and Yin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(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.

# Clinical Impact of Antibiotics for the Treatment of Pseudomonas aeruginosa Biofilm Infections

Elodie Olivares1,2 \*, Stéphanie Badel-Berchoux<sup>3</sup> , Christian Provot2,3, Gilles Prévost<sup>1</sup> , Thierry Bernardi2,3 and François Jehl<sup>1</sup>

<sup>1</sup> University of Strasbourg, CHRU Strasbourg, Fédération de Médecine Translationnelle de Strasbourg, EA7290, Institut de Bactériologie, Strasbourg, France, <sup>2</sup> BioFilm Pharma SAS, Saint-Beauzire, France, <sup>3</sup> BioFilm Control SAS, Saint-Beauzire, France

Bacterial biofilms are highly recalcitrant to antibiotic therapies due to multiple tolerance mechanisms. The involvement of Pseudomonas aeruginosa in a wide range of biofilmrelated infections often leads to treatment failures. Indeed, few current antimicrobial molecules are still effective on tolerant sessile cells. In contrast, studies increasingly showed that conventional antibiotics can, at low concentrations, induce a phenotype change in bacteria and consequently, the biofilm formation. Understanding the clinical effects of antimicrobials on biofilm establishment is essential to avoid the use of inappropriate treatments in the case of biofilm infections. This article reviews the current knowledge about bacterial growth within a biofilm and the preventive or inducer impact of standard antimicrobials on its formation by P. aeruginosa. The effect of antibiotics used to treat biofilms of other bacterial species, as Staphylococcus aureus or Escherichia coli, was also briefly mentioned. Finally, it describes two in vitro devices which could potentially be used as antibiotic susceptibility testing for adherent bacteria.

Keywords: biofilms, antibiotic tolerance, biofilm-related infections, Pseudomonas aeruginosa, clinical laboratory technique, MBEC assay, antibiofilmogram

## INTRODUCTION

Bacterial biofilm was defined for the first time in 1978 as a structured community of microorganisms adhering to a surface and producing an extracellular matrix of polysaccharides (Costerton et al., 1978). It represents a particular behavior of bacteria triggered by the proximity of a surface and involving complex signaling networks, including quorum sensing (QS). Its discovery was attributed to the microscope inventor, Antoni Van Leeuwenhoek who observed bacteria clusters on dental plaque in 1684. He wrote in a report for the Royal Society of London: "The number of these animalcules in the scurf of a man's teeth are so many that I believe they exceed the number of men in a kingdom" (Biofilms: The Hypertextbook, 2011).

Nowadays, it is well-recognized that biofilms play an ecological role and have a significant impact in medicine by the development of healthcare-associated infections. The National Institutes of Health (NIH) estimated that bacterial biofilms are involved in 65% of microbial diseases and in more than 80% of chronic infections (Jamal et al., 2018). Sessile cells can colonize indwelling medical devices as any type of catheters, contact lenses, heart valves, and protheses. Their presence on retrieved infected implants is easily detectable by laboratory methods. Indeed, bacterial colony outgrowths can be revealed by culturing techniques but can also be directly visualized by microscopy methodologies (Dibartola et al., 2017). Biofilm formation is equally involved in nondevice-associated infections as periodontitis, osteomyelitis, and chronic infections (Srivastava and Bhargava, 2016). Pseudomonas aeruginosa biofilms are particularly deadly in cystic fibrosis (CF)

### Edited by:

Fabian Cieplik, University Medical Center Regensburg, Germany

#### Reviewed by:

Rodolfo García-Contreras, National Autonomous University of Mexico, Mexico Monique L. Van Hoek, George Mason University, United States

\*Correspondence:

Elodie Olivares elodie.olivares@biofilmpharma.com

## Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 29 July 2019 Accepted: 02 December 2019 Published: 09 January 2020

#### Citation:

Olivares E, Badel-Berchoux S, Provot C, Prévost G, Bernardi T and Jehl F (2020) Clinical Impact of Antibiotics for the Treatment of Pseudomonas aeruginosa Biofilm Infections. Front. Microbiol. 10:2894. doi: 10.3389/fmicb.2019.02894

patients. They also have a relevant impact on clinical outcomes of patients with chronic wounds (Mulcahy et al., 2014). Relevant animal models are now available to study the involvement of P. aeruginosa sessile cellsin vivo infections. Diabetic wounds were mimicked in mice by Watters et al. (2013) and a porcine model allowed replicating the development of bacterial infections in CF lungs (Pezzulo et al., 2012).

A specific feature of sessile cells is their inherent tolerance to antimicrobials. Despite this basic knowledge, classical antibiotic susceptibility testing, providing theminimalinhibitory concentration (MIC) of molecules, is performed on non-adherent bacteria. Results collected according to antibiogram methods cannot predict the therapeutic success of the corresponding antibiotic therapies against biofilms. Furthermore, it is now well-recognized that low doses of antibiotics, encountered during continuous and fluctuating treatments, can stimulate biofilm establishment and are partly responsible for biofilm-specific antimicrobial tolerance.

Currently, no guidelines exist to help clinicians treat this kind of infections, although they are involved in the majority of untreatable clinical cases. Therefore, it appears urgent to develop a susceptibility test specific to biofilm or to validate a new-existing method for a routine use in diagnostic labs.

This review summarizes the basic knowledge about the growth of bacteria within a biofilm and the main steps of its formation. The tolerance features of sessile microorganisms to antimicrobial molecules were also detailed as well as the beneficial or deleterious effects of antibiotics for biofilm treatment. Available diagnostic tools for the selection of appropriate therapies against adherent bacteria are discussed herein.

## THE BACTERIAL BIOFILM

## A Community Way of Life

The growth of bacteria within biofilms is a natural process. The entirety of microorganisms could be sessile and live attached to a surface. This community mode is different from the planktonic growth, in which bacteria are isolated and mobile in the environment. The sessile cells differ from the planktonic ones by their morphology, physiology, and gene expression. The ability to adhere and grow on a surface as a biofilm is a survival strategy allowing the colonization of the environment by microorganisms. Bacteria continuously switch from a planktonic phenotype to a sessile one. This state variation is strategic for the cell as it allows a rapid adaptation to environmental conditions (Lebeaux and Ghigo, 2012).

The use of microscope can highlight a specific mushroom-like structure, especially for P. aeruginosa biofilms. They are mainly composed of microorganism clusters, delimited by aqueous channels. These latter separate bacterial microcolonies and allow the flow of oxygen and nutriments in the deepest areas of the biofilm as well as the elimination of degradation products. Nevertheless, it appears hard to generalize the composition, structure and features of biofilms owing to the wide range of environments and bacterial species. External factors, as medium composition and/or genetic properties of bacteria, contribute to the perpetual structure variation of the sessile population.

The key step of biofilm development is the synthesis of the extracellular matrix. It incorporates all the elements apart the bacterial cells. By forming up to 90% of its total organic matter, the matrix is the main structural component of the bacterial biofilm. It is highly hydrated and mainly composed of exopolysaccharides, proteins, nucleic acids, and minerals (Limoli et al., 2015). Its composition depends on the bacterial species and growth conditions. It allows strengthening of the biofilm structure while keeping a high flexibility. It also plays a protective role as it enhances the tolerance of bacteria to antimicrobials by creating a physical barrier that limits their diffusion to other environmental factors (UV, pH, and osmotic pressure variations, desiccation, etc.).

During the early development of the bacterial structure, it has been highlighted that extracellular DNA (eDNA) is essential for the adhesion of microorganisms and for their intercellular cohesion (Whitchurch et al., 2002). Quantitatively, in the biofilm matrix of P. aeruginosa, eDNA is six times more abundant than proteins and eighteen times more abundant than carbohydrates. Its origin was confirmed as being genomic. Nucleic acids can arise either from the lysis of a part of sessile cells or from an active secretion by living bacteria through merging membrane vesicles (Okshevsky and Meyer, 2015).

## Development of a Mixed Environment

Bacterial biofilm can be formed in a few hours. Its general development consists of five main steps (Dufour et al., 2010). Mobile free-floating bacteria detect an available conditioned surface through environmental signals as pH variation, oxygen and nutriment concentrations, temperature, and osmolarity, etc. They are transported by physical forces or bacterial appendages (i.e., flagella). The flagellum is as much required in the surface arrival as in the biofilm formation initiation since mutants defective in its synthesis are not able to adhere (O'Toole and Kolter, 1998). In the same way, a complete but inactive flagellum does not allow the establishment of the biofilm (Vallet et al., 2001).

The increasing proximity of the support, which is conditioned by the fluids and flows to where it is exposed, allows the initial adhesion of bacterial cells by physicochemical and electrostatic interactions. At this stage, the adhesion is reversible (**Figure 1**). Besides the environmental influence, this attachment is strongly influenced by the nature of the surface itself. A rough and/or hydrophobic surface boosts the adhesion of microorganisms, contrary to a smooth and/or hydrophilic support. Furthermore, the organic molecules, which are present at the surface, also condition the cell attachment.

Following this first step, which can occur few seconds after the initial contact with the surface, a second stage of adhesion happens, allowing the strengthening of the bacteria-surface bonds by the implication of bacterial compounds, such as type IV pili or more generally, surface adhesins. The surface binding, becoming irreversible, enables subsequently the multiplication of adherent bacteria, and the formation of microcolonies.

As for the flagella, P. aeruginosa mutants defective in the production of type IV pili adhere to a surface by the formation of a cell monolayer but are not able to gather in microcolonies. This data confirms that the microcolony formation is a process

requiring bacterial mobility and not only a clonal growth from a bacterial cell (O'Toole and Kolter, 1998). In general, bacterial structures involved in the mobility of microorganisms are needed for the initial step of biofilm formation. They allow the approach and the colonization of the surface. The use of DNA microarrays on biofilms formed by the PAO1 strain, on sterilized granite pebbles in a continuous-culture model, showed that genes required for the synthesis of bacterial surface structures are repressed as soon as biofilm formation is initiated (Whiteley et al., 2001). They are no longer necessary for biofilm development and move on to compounds allowing its structuring and differentiation. It was shown that QS is involved in the structural steps of the biofilm. Analysis of P. aeruginosa wild type and lasI mutant highlighted an architectural difference in both biofilms (thinner and less heterogeneous for the mutant one) (Davies et al., 1998).

The first maturation step of biofilm development is defined by the production of the extracellular matrix. It allows a mechanical cohesion between bacterial cells and favors the switch from a "free life" to a "static life." Its composition fluctuates in space and time and determines the spatial configuration of the biofilm.

As for the gene inactivation of extracellular structures, studies on P. aeruginosa showed that bacteria have a "sense of touch," namely the ability to detect the presence of a surface and to combine a specific gene expression. During the first stages of adhesion, the transcription of genes involved in the alginate synthesis is activated to organize the matrix production after the formation of microcolonies (Davies and Geesey, 1995). More recent works showed that this gene regulation is directly dependent on the cyclic di-GMP (c-di-GMP), a central messenger present in the cytoplasm of bacterial cells, which controls the transition between the planktonic life and biofilm establishment and whose intracellular concentration is affected by environmental stimuli (Donné and Dewilde, 2015).

If growth conditions are optimal, a second phase of biofilm maturation occurs, defined by a growth in thickness. Therefore, the mature biofilm shows a complex 3D-structure. It can acquire a typical "mushroom-like" shape formed by bacterial columns on the basement of cells and in which bacteria are mobile. The whole biofilm is surrounded by an extracellular matrix. Through this structure, channels remain and allow the transfer of oxygen and nutriments required for the growth of sessile bacteria. Gradients of oxygen and pH also set up from the top to the bottom of the biofilm. These variations of concentrations within the biofilm lead to metabolic activity and growth differences of bacteria even in a monomicrobial community, their activity being increased at the surface and reduced at its center. These physicochemical differences lead to a physiological heterogeneity of microorganisms and generate the formation of environmental microniches constituted by bacterial subpopulations that are genetically identical but physiologically distinct (concerning the tolerance against antimicrobials for instance).

The final step of the biofilm formation cycle is its destructuring. Biofilm dispersion can be initiated by various factors as mechanical disruptions (abrasion), enzymatic degradations (enzyme secretion determined by QS) or even a lack in nutriments or an overpopulation (McDougald et al., 2011). Fractions of bacteria are removed from the community and are spread in the environment. Newly mobile and adherent individual cells will be able to explore and colonize new surfaces by the establishment of a new biofilm. A new cycle of adhesion/maturation can get back.

Biofilms must be considered as an elaborate and dynamic organization which constantly evolve to get used to its environment. The passing through the sessile state to the planktonic one plays a considerable role in the transmission of bacteria from environmental reserves to the host and also in the transmission between hosts and in the infection propagation in the individual (biofilm metastasis).

## BACTERIA UNDER SHELTER

fmicb-10-02894 December 23, 2019 Time: 16:4 # 4

The main advantage of the sessile way of life is the modification of the adherent bacteria in regard to their susceptibility to mechanisms of immune defenses and antimicrobials. Indeed, a single planktonic cell is vulnerable to the action of antibodies or phagocytes and is fairly sensitive to antibiotics. Conversely, bacteria that are embedded in a biofilm structure can be tolerant to the host immune system, antimicrobials, and biocide molecules (Høiby et al., 2011). Indeed, we usually talk about biofilm tolerance to antibiotics rather than resistance.

The resistance can be defined as the ability of a microorganism to grow when an antimicrobial compound is present in the environment. Resistance mechanisms are heritable and avoid the antibiotic interaction with its target. On the contrary, the term tolerance must be used for bacteria, which are able to survive in high concentrations of antimicrobials, but with a suspended growth. This feature, specific to the sessile bacterial life, is reversible, phenotypical, and non-inherited. A bacterial cell from a biofilm, which is resuspended in liquid medium, will get back an in vitro susceptibility to antimicrobials.

## Inefficiency of Immune System

The size of the bacterial biofilm is the first brake to the phagocytosis process. Even in immunocompetent individuals, components of the immune system are seldom effective against biofilm infections. During the innate immune response, macrophages and neutrophils are rapidly activated following the direct contact with bacteria (i.e., through the O-antigen of the bacterial lipopolysaccharide (LPS) or the alginate for P. aeruginosa). Then, the immediate immune response triggers an important accumulation of neutrophils around the biofilm structure associated with an oxygen exhaustion, which is due to an active stimulation of the oxidative metabolism, as the molecular oxygen is reduced in superoxide (Watters et al., 2016).

The phagocytic cells penetrate with difficulty the extracellular matrix. They are slowed down and become more vulnerable to the inactivation by bacterial enzymes. Besides, the extended lysis of neutrophils leads to the overflow of harmful compounds in the medium, which are responsible for consecutive tissue damages. The host immune response is the main cause of the healthy tissue degradation surrounding the bacterial infection.

Concerning the memory of the immune system response, it has been reported that specific antibodies against bacterial compounds as the elastase, the LPS or the flagellum are secreted in CF patients. These data show that antigenic determinants were neutralized during chronic lung infection. Unfortunately, it has been demonstrated that these antibodies contribute to the formation of immune complexes, which are precipitated into the parenchyma and lead to complement activation and opsonization of neutrophils, namely in an indirect way, to the nearby tissue degradation (Jensen et al., 2010).

## General Tolerance to Antimicrobials

Biofilm tolerance to external aggressions, notably to antibiotic treatments, is one of its exceptional features. It is well-known that the MIC of an antimicrobial, which is effective on sessile bacteria, is 10 to 1,000 times more concentrated that the one which would be active on their planktonic version (Schurek et al., 2012). This decreased antimicrobial susceptibility can have several causes. It can be inherent to the own organization of the biofilm (structure and functioning) but can also be acquired by transmission of resistance factors.

Given its complex architecture, the biofilm itself creates a protective environment for bacterial cells. It can be seen as an "innate" tolerance. As for compounds of the host immune system, the extracellular matrix forms a mechanical barrier limiting antibiotic diffusion within the biofilm and their access to microorganisms. The electrostatic charges or some matrix components bind antimicrobial molecules and trap them. The general high viscosity of the polymeric matrix can also prevent the antibiotics from reaching their effective concentrations in the deeper layers of the bacterial community. Consequently, bacteria in the outer layers of the biofilm die following an antimicrobial treatment, while those in the deeper layers have time to react (Paraje, 2011). This invasion delay should be enough to allow a progressive physiological adaptation of bacteria exposed to antimicrobials (expression of resistance genes, secretion of inactivating enzymes. . .). For instance, it has been demonstrated that alginate and eDNA in the biofilm matrix of P. aeruginosa could link aminoglycosides and play a role in the sessile bacterial tolerance to tobramycin (Hentzer et al., 2001). Similarly, an extracytoplasmic process of antibiotic sequestration by periplasmic glucans was highlighted. The locus ndvB was identified as being required for the production of cyclic glucans (Mah et al., 2003). The authors showed that polymers physically bound antimicrobial compounds in the periplasm, leading to the diffusion slowdown of antibiotics into the cells and preventing them from reaching their action sites. Nevertheless, this global diffusion barrier, specific of the biofilm matrix and the sessile cells, appears to be strain- and antibiotic-dependent. By itself, it cannot explain the radical tolerance of biofilms against antimicrobial agents.

In view of their own biofilm organization, bacterial metabolism plays an important part in antibiotic tolerance. The concentration gradients of metabolites, oxygen, and nutriments within the mature biofilm create bacterial niches that are less metabolically active. For example, in P. aeruginosa microcolonies, the oxygen is consumed faster at the surface than it diffuses into the deeper layers of the biofilm. Its graduated diffusion leads to the formation of hypoxic areas in the bacterial community (Serra and Hengge, 2014). Some of microorganisms could get back to a stationary phase in lowering their growth and multiplication rates as an induced stress response. This reduced metabolism of sessile cells is partly responsible for the tolerance associated with the biofilm, as the action mode of the majority of antimicrobials targets metabolic processes in growing

bacteria (replication, transcription, translation, or cell wall synthesis). Lots of works have validated the advantageous efficacy of antimicrobials on active bacteria, which are located in external areas of the biofilm. However, parallel studies showed that other types of molecules, such as SDS, EDTA, or chlorhexidine could conversely act on bacterial cells in stationary phase of growth, located in the internal niches (Ciofu et al., 2015).

The activation of efflux pumps by bacteria embedded in the extracellular matrix can also contribute to the inefficiency of antimicrobials in actively discharging them outside the biofilm structure before they can reach their target. These membrane transporters can be specific of a class of antibiotics or responsible for multidrug resistance. In Gram-negative bacteria as P. aeruginosa, efflux pumps are usually composed of a pump located in the inner membrane, an outer membrane factor, and a periplasmic fusion protein. The association of cell impermeability with the expression of the efflux system MexAB-OprM leads partly to the inherent resistance of the bacillus to antibiotics. The expression of some of them was demonstrated as being specific to the biofilm mode (Zhang and Mah, 2008).

The bacterial density and the spatial proximity of microorganisms within a mature biofilm promote the gene exchange and the resistance plasmid transmission. The horizontal gene transfer could be 1,000-fold more important in a bacterial community than between planktonic cells. Due to the starved local environment within the biofilm, bacteria are also subjected to random mutations and genetic rearrangements. This generation of bacterial variants, favored by natural selection, leads to a chromosomal resistance (Poole, 2012). The mutation frequency can be stimulated by environmental factors, as the presence of reactive oxygen species from the lung inflammatory response. These reagents, in damaging DNA, cause mutations in bacteria and lead to the diversity of bacterial phenotypes in the biofilm (Rodríguez-Rojas et al., 2012). Finally, by combination of several of these mechanisms, sessile bacteria rapidly became multiresistant.

It also demonstrated the existence of a "persister" bacterial population which could constitute a reserve allowing the infection relaunch after elimination of peripheral planktonic and sessile cells (Stewart, 2002).

## The "Persister Cells" Enigma

Persisters are regular cells exhibiting a specific non-growing phenotype, combined with an excessive tolerance to antibiotic concentrations. Their existence was firstly described in the 1940s (Bigger, 1944). The transcription downregulation of genes involved in motility and energy production was highlighted for these isolated bacteria. Consequently, as they are in a dormant state, antimicrobials are able to bind to their target molecules but not to impair their initial function (Lewis, 2005).

The presence of persisters can be easily detected in bacterial cultures by a process of biphasic death, further to an exposition to bactericidal antibiotic concentrations. Firstly, a lethal dose of antimicrobials will rapidly eradicate the sensitive bacterial population. A much slower second phase of death follows, reflecting the poor killing of persister cells. Finally, the end of antimicrobial treatment will allow the renewal of the bacterial community by regeneration of persister survivors (Conlon et al., 2015).

Each bacterium shows the capacity to be differentiated in a persister cell, but few of them are really observed during the early exponential phase of growth. Indeed, the genuine persister population is formed during the mid-exponential phase and finally, they reach up to 1% of the overall population at the stationary phase. This phenotypic conversion can be induced by environmental stimuli or stresses, as antibiotic exposure, which are predictive of immediate threats for cells, or they may be preexisting in the bacterial population (Harms et al., 2016; Fisher et al., 2017). It is assumed that stochastic modifications in genes can lead to the phenotypic switch along with the over-expression of specific toxin-antitoxin (TA) module proteins. Typically, the toxin portions are neutralized by their antitoxins, but under cellular stresses, proteases must be over-expressed and degrade the antitoxin proteins. In that case, the toxin modules are free to exert their toxic action on bacteria. The expression of many other compounds implicated in the persister phenotype can be induced by environmental stimuli. The signaling pathway of the SOS response and the alarmone ppGpp, two stringent responses to stress, also appears to be associated with the persistence of bacteria (Del Pozo, 2018).

## EFFECTIVENESS OF CONVENTIONAL ANTIBIOTICS

Despite the intensive tolerance of the biofilm to antimicrobials, certain conventional antibiotics still demonstrate activity against bacterial cells growing in the biofilm state.

In a recent study, Otani et al. (2018) showed that sub-MICs of ceftazidime reduce biofilm volume, inhibit twitching motility, and repress gene expression involved in bacterial adhesion and matrix production of P. aeruginosa PAO1. Roudashti et al. (2017) had previously noticed this effect of cephalosporin on motility and biofilm formation for the same strain.

Similarly, other common antimicrobials were described as being effective on biofilm behavior of P. aeruginosa. Subinhibitory doses of piperacillin/tazobactam altered the pathogenic potential of various clinical and laboratory strains of P. aeruginosa in reducing bacterial adhesion, in decreasing biofilm formation, swimming, and twitching motility and conversely in increasing the susceptibility of cells to oxidative stress (Fonseca, 2004). Indeed, one early step of biofilm formation which can be targeted for the prevention of chronic infection is bacterial adhesion to a surface. The process of twitching motility contributes to this part of virulence of sessile microorganisms. Wozniak and Keyser (2004) noticed that clarithromycin substantially inhibited cell translocation of P. aeruginosa through its type IV pilus as well as altered its biofilm architecture at sub-MIC levels. Another control strategy against bacterial biofilm is QS disruption. Azithromycin, ceftazidime, and ciprofloxacin showed, at subinhibitory concentrations, QS-inhibitory activities in bacteria (Skindersoe et al., 2008). This beneficial effect of the macrolide was also emphasized in an experimental urinary tract infection model. Antibiotic

concentrations below the MIC could inhibit the production of QS molecules, leading to the complete clearance of P. aeruginosa from the mouse kidneys (Bala et al., 2011). Azithromycin was also described as being able to prevent PAO1 biofilm formation in a flow cell biofilm model (Gillis and Iglewski, 2004).

Continuous treatment of colistin (25 µg/ml) turned out to be effective against the non-dividing central part of a P. aeruginosa biofilm growing in a flow chamber for 4 days. Associated with ciprofloxacin (60 µg/ml), which can kill metabolically active cells in the surface layers, this combination therapy showed a clinical efficacy for the early eradication treatment of bacteria in CF patients (Høiby et al., 2010). In a more recent study, the association of minimal biofilm inhibitory concentrations (MBICs) of fosfomycin with tobramycin (≥1024 µg/ml and from 8 to 32 µg/ml, respectively) has been demonstrated to be synergistic against CF isolates in in vitro models (Díez-Aguilar et al., 2018). Overall, the aminoglycosides usually prescribed in CF (amikacin and tobramycin) showed a preventive action on the early adhesion of clinical P. aeruginosa strains at various concentrations (sub-MICs, MICs, and so-called PK/PD doses) (Olivares et al., 2017).

The efficacy of antibiotic lock solution (meropenem, levofloxacin, and colistin) on P. aeruginosa clinical and reference strains was also confirmed. The antibiotic lock technique (ALT), using antimicrobial molecules, prevented bacterial regrowth in an in vitro antibiotic lock model. The efficacy of ALT to eliminate P. aeruginosa biofilms should be improved when the three antibiotics were used in combination with clarithromycin (Ozbek and Mataraci-Kara, 2016).

Finally, all the cited publications attest that some conventional molecules can still be active on P. aeruginosa in the context of chronic infections, in preventing its growth within the biofilm. Nevertheless, a lot of these studies were carried out on the reference PAO1 strain. To confirm the clinical effectiveness of antimicrobial treatments, antibiotic susceptibility testing must be performed on clinical isolates, and clinical trials must be planned.

Concerning the positive effect of classic antimicrobial therapies on other sessile pathogens, MICs and minimal bactericidal concentrations (MBCs) of rifampicin have demonstrated an activity against biofilms of Staphylococcus epidermidis and Staphylococcus aureus isolates associated with device infections, especially when it is used in association with other molecules as fusidic acid, vancomycin or ciprofloxacin in an in vitro biofilm model (Saginur et al., 2006). More recently, adherent staphylococci involved in skin and soft tissue infections were described as being more or less susceptible to multiples of tedizolid MICs and other comparator agents (vancomycin, linezolid, and daptomycin) (Delpech et al., 2018). The use of daptomycin-lock therapy (50 mg/ml) also showed a therapeutic advantage for the 24 h-treatment of a long-term catheter-related bloodstream infections by coagulase-negative S. epidermidis in a rabbit model (Basas et al., 2018). Similarly, subinhibitory concentrations of fluoroquinolones were able to reduce the number of sessile cells to prevent the adhesion of the corresponding S. epidermidis strains and to alter biofilm morphology (Szczuka et al., 2017).

This overall review of publications, dealing with the anti-biofilm property of conventional antimicrobials, can be completed with studies using Escherichia coli, another bacterial specie well-characterized for its capacity to form biofilm structures. In a recent article, Klinger-Strobel et al. (2017) noticed that colistin concentrations from 4 to 16 mg/l could reduce the amount of adherent E. coli bacteria and exert a matrix-reducing effect on biofilms in formation. Similarly, Butini et al. (2018) investigated the anti-biofilm property of gentamicin-eluting bone graft substitute against bacterial species involved in bone and implant-associated infections. Calcium sulfate bone graft substitutes served as local antibiotic delivery carrier, and gentamicin is one of the most used molecules for the treatment of bone-related infections. Therefore, they demonstrated that 12 µg/ml of released gentamicin were able to prevent E. coli adhesion and 23 µg/ml of the molecule could eliminate a 24 h-old biofilm. These data are promising as the applied concentrations are achievable for local treatment in bone and soft tissues. In another recent publication, a mupirocin spray was formulated and tested against E. coli strains, in a context of wound and surgical site infections. Inhibition and disruption of formed biofilms were achieved with a single and a sub-actual dose of the antibiotic spray (1 mg per spray or 1 mg per 50 mg of ointment), compared to the commercialized mupirocin ointment (Bakkiyaraj et al., 2017). Results showed that both formulations had an anti-biofilm action on E. coli sessile cultures in tissue culture plates but microscope studies provided complementary evidences that it remained more individual adherent cells with ointment formulation than for treatment with the antibiotic spray. Authors concluded that this efficacy on biofilm prevention and disruption was comparable with that of the ointment. Nevertheless, the spray use seemed beneficial as it included an easy application: while the ointment was removed from the application site upon washing, the spray formulation significantly resisted removal after a single wash.

Finally, for the treatment of recurrent urinary tract infections (UTIs) associated with the presence of biofilm, the use of amikacin, ciprofloxacin, and third-generation cephalosporins could be recommended. Indeed, various concentrations of these molecules, selected according the bioavailability of antibiotics in human urine, showed the ability to significantly reduce biofilm biomass in a study of 116 E. coli strains of UTIs (González et al., 2017).

Even if it seems that antimicrobials can be effective on prematurely adherent bacteria, a small percentage of persister cells develop a high tolerance to antibiotics and are typically involved in infection relapses. Moreover, the biofilm must be considered as a single compartment with its own pharmacokinetic parameters. This will influence local antibiotic concentration and its metabolization in the biofilm (Cao et al., 2015). All these combined factors lead to the recommended use of high doses of antimicrobials for long periods, which cannot always be practicable in patients because of severe systemic side effects.

Biofilm tolerance to antimicrobials is complex and above all multifactorial. A now long list of studies demonstrate that low doses of antimicrobials at the infection site might increase the selection of mutagenesis and the risk of biofilm formation initiation induction (Ciofu et al., 2017).

## INDUCTION OF BIOFILM FORMATION BY ANTIMICROBIALS

At high concentrations, antibiotics appear to be perfect bacterial killers. Their original function in the environment is fighting and inhibiting the growth of competitors (Linares et al., 2006). The production of antimicrobials by bacteria themselves allows the killing of predators. This application of the classical Darwinian principle supports the idea that they allow the competing colonization of soil by microorganisms. However, antibiotic molecules constitute only a small part of organic compounds produced by bacteria. Consequently, it can be assumed that they can affect the general modulation metabolic function in bacteria, as other signaling analogous (rhamnolipids, peptides, and QS signals, etc.). Supporting this assumption, phylogenetic analyses revealed that antimicrobial resistant genes were present in bacterial genomes millions of years before the modern use of antibiotics. A similar example concerns the metagenomic study of Alaskan soils, which demonstrated the existence of an ancient and varied collection of β-lactamase genes, whereas this antimicrobial family is not detected in the environment (Aminov, 2009).

In activating specific gene transcription, antimicrobial compounds seem to act as signaling molecules, regulating the homeostasis of bacterial communities. As sessile cells are significantly less sensitive to antimicrobials, biofilm formation would be a strategic evolution of bacterial populations to counteract non-lethal doses of antibiotics produced by soil microorganisms. This implies that antimicrobials can also be beneficial for the survival of susceptible planktonic cells in nature. Therefore, they can permit a more efficient colonization of heterogeneous environments. Especially at subinhibitory levels, antibiotics modulate bacterial virulence, stress response, motility, and biofilm formation (Song et al., 2016).

The first report describing the ability of low doses of antibiotics to interfere with bacterial functions was made by Gardner (1940). In the presence of subinhibitory concentrations of penicillin, numerous gram-negative, and positive pathogens formed elongated filaments.

Since then, similar studies introducing the structure modification of bacterial cells by antibiotics were published. P. aeruginosa showed the most morphological changes as bacteria reacted to meropenem and biapenem in forming a "bulge" midway along them (Horii et al., 1998). The authors also described a relationship between the induction of a new morphology and the amount of endotoxin released by bacteria. The modification of the bacterial size and shape was explained by the fact that antibiotic molecules inhibit the Penicillin-Binding-Proteins 2 and 3 (PBP-2 and PBP-3). They are involved in the assembly of the bacteria cell wall by the catalysis of the terminal stages of the peptidoglycan network. These proteins are the primary target of β-lactam antibiotics in Gram-negative bacteria. The inactivation of PBP-1 leads to bacteria lysis, whereas the inhibition of PBP-2 and PBP-3 is associated with either spherical cells or filamentous bacteria. These effects of antimicrobials onto the bacterial morphology were also observed in experimental infections in mice (Yokochi et al., 2000). More precisely, the authors highlighted a relationship between the shape of bacteria and their susceptibility to phagocytosis. Induced round bacteria were phagocytosed by peritoneal cells, whereas long filaments were not. Finally, it appeared that the morphological reorganization of bacteria is a reversible process as when antibiotics were no longer present, the induced spherical or filamentous population converts back to the normal bacillary form (Monahan et al., 2014). This transition is a strategy to survive antibiotic exposure as the biofilm formation.

Furthermore, exposure of P. aeruginosa strains to sub-MICs of ciprofloxacin leads to the selection of pre-existing mutants with high level resistance (Jørgensen et al., 2013). The analysis of the strains selected by subinhibitory doses of the fluoroquinolone showed phenotypic changes in bacteria, as decreased protease activity and swimming motility, down-regulation of the type III secretion system and higher levels of quorum-sensing signals (Wassermann et al., 2016). In some cases, this implies that antibiotic therapies may expose patients to detrimental sideeffects by accelerating pathogen adaptation and raising the risk of antimicrobial tolerance and spread in the commensal bacterial flora if bacteria are exposed to low antibiotic concentrations.

The first study demonstrating the real "inducer" property of antimicrobials on biofilm formation at subinhibitory concentrations dates back to 1988 (Schadow et al., 1988). The adherence of coagulase-negative staphylococci was increased to 65% after rifampicin treatment. Since then, numerous works focusing on the effect of low antibiotic doses on biofilm formation were published.

Although several studies have shown that aminoglycoside antibiotics act as antagonists of biofilm formation in vitro (see previous section of this paper), opposite data were collected and highlighted the ability of the same molecules to significantly induce the sessile growth of a variety of bacterial species. Hoffman et al. (2005) described the effect of subinhibitory doses of aminoglycosides in P. aeruginosa and E. coli biofilms. They showed that these antibiotics stimulated the expression of the arr gene, which encodes a phosphodiesterase whose substrate is the c-di-GMP. Reinforcing this idea, induction of biofilm formation after an exposure to sub-MICs of aminoglycosides was detected among half of a P. aeruginosa clinical isolate collection (Elliott et al., 2010).

Imipenem, a carbapenem molecule, was also able to substantially influence the expression of 34 genes in the common reference PAO1 strain. When subinhibitory concentrations of this antibiotic were applied to bacterial cultures, the alginate gene cluster, the main component of the biofilm matrix of P. aeruginosa was more than 10-fold induced (Bagge et al., 2004). The corresponding polysaccharide amount in biofilms was quantified by the authors and they found that the alginate level in the matrix of stimulated-biofilms was 20-fold higher than the one of the non-exposed controls.

Induction phenomenon of biofilm formation by antimicrobials was also described for other bacterial species. The use of β-lactam antibiotics at sub-MIC concentrations leads to the induction of the bacterial adhesion by a community-associated Methicillin-Resistant S. aureus (MRSA) strain and clinical isolates (Ng et al., 2014). All of the antimicrobial molecules tested

in this study exhibited a biphasic dose-response curve. Authors also described an inversely proportional relationship between the biofilm amount and the susceptibility of bacteria to methicillin: the more sensitive the strain to antibiotic, the lower is the concentration required to induced biofilm. Biofilm formation of a clinical isolate of Enterococcus faecalis, which commonly underlies prosthetic valve endocarditis and multiple device infections, was also significantly increased by low concentrations of ampicillin, ceftriaxone, oxacillin, and fosfomycin. This enhancement of biofilm establishment appeared to be specific of molecules inhibiting cell wall synthesis (Yu et al., 2017). Additionally, a collection of ninety-six clinical isolates of S. epidermidis, which originate from various samples as wounds, catheters, sputum, etc., was recovered by He et al. (2016). The authors described that 27% of erythromycin-resistant strains exhibited biofilm induction by 0.25 MIC of the molecule. The induction intensity ranged from 1.11-fold to more than twofold (He et al., 2016).

A more complete review article written by Kaplan (2011) gathers studies showing that subinhibitory concentrations of antimicrobial molecules can act as agonists of bacterial biofilms in vitro.

Biological responses of bacterial species are strain- and dosedependent. Some molecules can promote the biofilm formation at high levels and conversely be antagonists and repress its establishment at lower doses. Nonetheless, this dose-response relationship cannot be generalized as it can be inverted, depending on the considered antibiotic, which is used in the treatment for a specific laboratory or clinical strain. The discovery of this ecological function of antibiotics is essential as, in a clinical context, the induction of biofilm formation by low concentrations of antibiotics would contribute to the failure of antimicrobial therapies in case of biofilm-related infections. It can be speculated that this is a common phenomenon as microorganisms are usually under fluctuating doses of antibiotics during a chemotherapy.

## Clinical Tools Available for the Diagnosis and Treatment of Biofilm Infections

Traditionally, clinical microbiology laboratories have focused on the culture of isolated bacterial strains and provide their susceptibility to antibiotics in defining the breakpoints and the PK/PD parameters under planktonic growth conditions. The corresponding antibiotic therapies, based on non-adherent microorganisms, are often associated with treatment failures and/or recurrence of the infection. No guidelines are offered to clinicians to successfully treat biofilm infections, which can result to false-negative data if the samples do not significantly represent the main infection.

Besides, there is still no available standardized tool to detect easily the presence of sessile cells in a clinical sample and allow determination of their specific antibiotic susceptibility. As biofilm bacteria are inherently more tolerant to antimicrobials, the establishment of the corresponding breakpoints to predict therapeutic success is needed. New methods monitoring the effect/response of biofilm cells to antibiotic therapy must

FIGURE 2 | Regular MBEC with 96-well plate. Photography adapted from Parker et al. (2014).

be designed. Currently, two technologies were developed but not yet standardized for a rapid routine use in hospital laboratories: the Calgary device (also called MBEC assay) and the Antibiofilmogram <sup>R</sup> .

The Calgary biofilm device consists of a two-part reaction vessel (Ceri et al., 1999). A lid, composed of 96 pegs, forms the first and top component. The bottom component of the device is a standard 96-well plate, in which the pegs are designed to sit into each well. The microplate contains the medium allowing the growth of bacteria, which is set up in 96 equivalent biofilms at each peg site (**Figure 2**). Biofilm susceptibility testing is performed in transferring the lid with bacterial biofilms to a standard 96-well plate containing serial dilutions of antibiotics. The minimal biofilm eradication concentration (MBEC) is defined as the minimal concentration of antibiotic that prevents visible growth in the recovery medium used to collect sessile cells.

Moskowitz et al. (2004) evaluated the in vitro activity of twelve antimicrobials on a large number of CF sessile clinical isolates of P. aeruginosa using a modified Calgary device protocol. The MBICs of the antimicrobial agents were much higher than the corresponding conventional MICs. In a context of biofilm infections, this suggests that the use of broth microdilution susceptibility testing or other standard methods to guide therapy may not contribute to improve clinical outcomes. Quite the reverse, devices simulating biofilm growth conditions might guide therapy more effectively. To verify this hypothesis, clinical trials were already conducted. Unfortunately, no evidence that biofilm susceptibility testing performed with the Calgary device was more efficient than conventional techniques in terms of clinical outcomes was provided (Waters and Ratjen, 2017). Yau et al. (2015) introduced the first randomized controlled trial evaluating the utility of biofilm antimicrobial susceptibility testing in the treatment of CF pulmonary exacerbations. They concluded that the choice of antibiotic therapies based on biofilm behavior of bacteria did not improve clinical outcomes and did not decrease pulmonary bacterial loads. They explained the lack of Calgary method efficacy by an oversimplification

of the sessile growth conditions. Biofilm formation on a lid composed of 96 plastic pegs could not recreate the environment in vivo, in which sessile cells grow and express specific properties. Moreover, the determination of antimicrobial susceptibility of a selected isolate could underestimate the microbial diversity response to antibiotics.

The Antibiofilmogram <sup>R</sup> (ATBFG) method was specifically designed to investigate early steps of biofilm development, by rapidly screening antibiotic effective against sessile bacteria. Its functioning principle is based on the potential immobilization of magnetic microbeads by bacteria forming a biofilm in the well bottom of a microplate (Chavant et al., 2007).

Briefly, a given bacterial culture is mixed with the microbead suspension and loaded in the plate (**Figure 3**). Afterward, it is incubated and submitted to a magnetic field at a desired time point, without staining and/or washing stages. The formation of a brown spot in the bottom of wells reveals the free migration of beads during plate magnetization and so, the remaining-free state of bacteria. Conversely, the absence of visible spots reflects the bead blockage by a pre-forming biofilm (Olivares et al., 2016). The main advantage of this methodology is its capacity to collect data within a couple of hours, allowing comparison of antibiotic susceptibility of sessile bacteria to the results of classical antibiograms. But as for the Calgary method, the ATBFG is an in vitro assay, which does not provide information about structure or thickness of the mature biofilms.

Results of ATBFG performed on 29 clinical strains of S. aureus isolated from bone and joint infections (BIJs) were also published (Tasse et al., 2016). On the basis of antibiotic breakpoint values, the authors defined effective antimicrobial molecules against adhesion of the majority of S. aureus strains (rifampicin, linezolid, clindamycin, and fusidic acid), others inefficient against bacterial adherence (fosfomycin, ofloxacin, daptomycin, and gentamicin) and some of them whose efficacy was strain-dependent (cloxacillin, vancomycin, and teicoplanin). Data validity was confirmed by in vivo assays (catheterrelated infections in the mouse). Results showed that serum concentrations of cloxacillin, corresponding to the MBICs determined by ATBFG (either 2 or 4 µg/ml), allowed reduction to 3 log the bacterial biomasses colonizing the catheters for three clinical strains, whereas the simple MICs of the antibiotic were inefficient on biofilm formation.

In a CF context, the use of ATBFG on clinical P. aeruginosa strains also showed the capacity of two aminoglycosides (amikacin and tobramycin) to prevent bacterial adhesion at concentrations close to the MICs (Olivares et al., 2017). Only in vitro assays were performed in this study but an intermethod reproducibility was conducted through Crystal Violet staining and a tissue culture system, which validated the inhibitory effect of the antimicrobials on the early adhesion of P. aeruginosa isolates.

## CONCLUDING REMARKS

As demonstrated in previous sections, the treatment of bacterial infections with chemically distinct antibiotics can lead to a variety of responses from sessile bacteria. Despite the increased tolerance of microorganisms toward antimicrobials, some molecules are always effective against newly adherent bacteria. In clinical

practice, it is recommended that when possible, as for the diabetic foot for instance, to resort to topical administration to provide high local concentrations to the infection site without systemic side-effects.

Disappointingly, numerous studies have also described that low doses of antibiotics can significantly induce biofilm formation in vitro for a variety of bacterial species. The plausibility of this phenomenon in vivo must be considered as bacterial pathogens are exposed to sub-MIC concentrations of antimicrobials during a clinical therapy with fluctuating dosing regimens. More researches on antibiotic-induced biofilm formation are required to elucidate the involved mechanisms. Clinical trials that verify the relevance of this process in patients and the potential relationship with therapy failure will also be highly helpful. The prospect of complementary assays evaluating the susceptibility of free and sessile cells to antibiotics would also allow the optimization of the general use of antimicrobials in the treatment of biofilm-related infections.

## REFERENCES


## AUTHOR CONTRIBUTIONS

EO wrote the manuscript. All authors read and approved the submitted version.

## FUNDING

This study was received funding from the BioFilm Pharma SAS. The funder was not involved in the study design, collection, analysis, interpretation of data, and the writing of this article or the decision to submit it for publication. This study was supported by the 15th Fonds Unique Interministériel (FUI).

## ACKNOWLEDGMENTS

We thank BioFilm Pharma SAS for the financial support.



exposed to subinhibitory concentrations of antimicrobial agents. J. Infect. Dis. 157, 71–77. doi: 10.1093/infdis/157.1.71


**Conflict of Interest:** EO, CP, and TB were employed by the company BioFilm Pharma SAS, and SB-B, CP, and TB were employed by the company BioFilm Control SAS.

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 © 2020 Olivares, Badel-Berchoux, Provot, Prévost, Bernardi and Jehl. 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.

# Colistin Resistance Gene mcr-1 Mediates Cell Permeability and Resistance to Hydrophobic Antibiotics

Baiyuan Li<sup>1</sup> , Fang Yin<sup>2</sup> , Xuanyu Zhao3,4, Yunxue Guo<sup>3</sup> , Weiquan Wang3,4, Pengxia Wang<sup>3</sup> , Honghui Zhu<sup>5</sup> , Yeshi Yin<sup>1</sup> and Xiaoxue Wang3,4 \*

<sup>1</sup> Key Laboratory of Comprehensive Utilization of Advantage Plants Resources in Hunan South, College of Chemistry and Bioengineering, Hunan University of Science and Engineering, Yongzhou, China, <sup>2</sup> Department of Breast and Thyroid Surgery, The Fifth Affiliated Hospital, Sun Yat-sen University, Zhuhai, China, <sup>3</sup> Key Laboratory of Tropical Marine Bio-resources and Ecology, Guangdong Key Laboratory of Marine Materia Medica, RNAM Center for Marine Microbiology, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou, China, <sup>4</sup> University of the Chinese Academy of Sciences, Beijing, China, <sup>5</sup> State Key Laboratory of Applied Microbiology Southern China, Guangdong Provincial Key Laboratory of Microbial Culture Collection and Application, Guangdong Open Laboratory of Applied Microbiology, Guangdong Microbial Culture Collection Center, Guangdong Institute of Microbiology, Guangzhou, China

Edited by:

Raffaele Zarrilli, University of Naples Federico II, Italy

## Reviewed by:

Ximin Zeng, The University of Tennessee, Knoxville, United States Jian-Hua Liu, South China Agricultural University, China

> \*Correspondence: Xiaoxue Wang xxwang@scsio.ac.cn

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 19 September 2019 Accepted: 16 December 2019 Published: 10 January 2020

#### Citation:

Li B, Yin F, Zhao X, Guo Y, Wang W, Wang P, Zhu H, Yin Y and Wang X (2020) Colistin Resistance Gene mcr-1 Mediates Cell Permeability and Resistance to Hydrophobic Antibiotics. Front. Microbiol. 10:3015. doi: 10.3389/fmicb.2019.03015 Colistin is considered the last-resort antibiotic used to treat multidrug resistant bacteriarelated infections. However, the discovery of the plasmid-mediated colistin resistance gene, mcr-1, threatens the clinical utility of colistin antibiotics. In this study, the physiological function of MCR-1, which encodes an LPS-modifying enzyme, was investigated in E. coli K-12. Specifically, the impact of mcr-1 on membrane permeability and antibiotic resistance of E. coli was assessed by constructing an mcr-1 deletion mutant and by a complementation study. The removal of the mcr-1 gene from plasmid pHNSHP45 not only led to reduced resistance to colistin but also resulted in a significant change in the membrane permeability of E. coli. Unexpectedly, the removal of the mcr-1 gene increased cell viability under high osmotic stress conditions (e.g., 7.0% NaCl) and led to increased resistance to hydrophobic antibiotics. Increased expression of mcr-1 also resulted in decreased growth rate and changed the cellular morphology of E. coli. Collectively, our results revealed that the spread of mcr-1-carrying plasmids alters other physiological functions in addition to conferring colistin resistance.

Keywords: mcr-1, colistin resistance, permeability, hydrophobic antibiotics, plasmid

## INTRODUCTION

Colistin is one of the primary classes of antibiotics with activity against most gram-negative bacteria and is considered the last resort antibiotic for the treatment of infections caused by carbapenemresistant Enterobacteriaceae. In 2015, a plasmid-encoded colistin resistance gene named mcr-1 was described in Enterobacteriaceae isolated from humans and livestock in China (Liu et al., 2016).

Since then, polymyxins have rightfully drawn renewed attention to colistin resistance, and plasmid-mediated colistin resistance by mcr-1 has been reported worldwide in livestock, food and humans (Poirel et al., 2017; Li et al., 2018). The mcr-1 gene confers colistin resistance by encoding a phosphoethanolamine transferase that catalyzes the addition of a phosphoethanolamine moiety to lipid A in the bacterial outer membrane (OM) (Gao et al., 2016; Hinchliffe et al., 2017), which may modify the structure of lipid A and then decrease the growth rate, cell viability, and competitive ability and shape cytoplasmic structures (Yang et al., 2017).

The OM of gram-negative bacteria plays a crucial role in protecting cells against an adverse environment and exchanging material (Costerton et al., 1974). To work effectively, antibiotics must pass across the OM barricade to reach the inhibitory concentration inside the bacterial cell (Vergalli et al., 2017). Bacterial OMs with low permeability have been identified as robust barriers that prevent many antibiotics from reaching their intracellular targets (Nikaido, 2003). Antibiotics usually traverse the OM by one of two mechanisms: the lipid-mediated pathway responsible for macrolides and hydrophobic antibiotics, such as aminoglycosides (gentamycin, kanamycin), and general diffusion porins for hydrophilic antibiotics such as β-lactams (Benz, 1988; Nikaido, 2003). The lipid and protein compositions of the OM have a major impact on the susceptibility of the microorganism to antibiotics, and drug resistance involving modifications of these macromolecules is common (Benz, 1988). For instance, several studies showed that alternations in the hydrophobic properties of the membrane or null mutations in porins create resistance to β-lactam antibiotics (Miller, 2016; Ghai and Ghai, 2018). MCR-1 is a membrane-bound enzyme consisting of five hydrophobic transmembrane helixes and a soluble form located in the periplasmic space (Liu et al., 2016). A recent study reported that mutants with a highlevel colistin resistance are more susceptible to most antibiotics compared with their respective parental strains (Yang et al., 2017). However, whether other antibiotic resistances could be affected by expression of mcr-1 remain unclear. Since mcr-1 encodes a phosphoethanolamine transferase that modifies the structure of LPS of the OM, it raises the possibility that MCR-1 may affect the susceptibility of bacteria to hydrophobic antibiotics by changing the membrane permeability. Therefore, we analyzed the impact of mcr-1 expression on the membrane permeability of E. coli by constructing an mcr-1 deletion mutant strain and by constructing a vector to overexpress mcr-1.

## MATERIALS AND METHODS

## Bacterial Strains, Plasmids, and Growth Conditions

The bacterial strains and plasmids used in this study are listed in **Table 1**, and the sequences of primers used in this study are listed in **Supplementary Table S1**. The E. coli strains were grown in Luria–Bertani (LB) broth or on LB agar plates (with 10 g NaCl per liter) at 37◦C, except for E. coli carrying pKD46 or pCP20, which were grown at 30◦C. Antibiotics and other chemicals were used TABLE 1 | Strains and plasmids used in this study.


at the following final concentrations: chloramphenicol, 30 µg/ml; polymyxin B, 2 µg/ml; ampicillin, 100 µg/ml; isopropyl-β-dthiogalactopyranoside (IPTG), 0.5 mM; and arabinose, 1 mM. A total of 1 mM L-arabinose or 0.5 mM IPTG (Sigma) was used to induce Para or Plac, respectively.

## Construction of the Deletion Mutant

The coding region of the mcr-1 gene was deleted from E. coli K12 BW25113 carrying plasmid pHNSHP45 (Liu et al., 2016) following a one-step inactivation method (Datsenko and Wanner, 2000). The primers used in this study are shown in **Supplementary Table S1**. PCR products containing chloramphenicol resistance cassettes flanked by 39 bp of homology to the 5<sup>0</sup> and 3<sup>0</sup> termini of mcr-1 were electroporated into competent cells of parent strains carrying pKD46. To construct the mcr-1 deletion mutant, PCR products that included 37-nt homology extensions and 20-nt priming sequences for the chloramphenicol resistance gene cat, bordered by FLP recombination target (FRT) sites, were amplified from plasmid pKD3 (cat) using primers pKD46-mcrF/pKD46-mcrR. Removal of the mcr-1 gene from pHNSHP45 in E. coli was verified by PCR and DNA sequencing using the primer pair MCR-LF/MCR-LR. The FRT-flanked chloramphenicol cassette was removed after transformation with pCP20 as described previously (Datsenko and Wanner, 2000). pCP20 is a plasmid that carries ampicillin and chloramphenicol resistance genes and exhibits temperaturesensitive replication and thermal induction of FLP synthesis (Cherepanov and Wackernagel, 1995). CmR mutants were transformed with pCP20, and ampicillin-resistant transformants were selected at 30◦C, propagated non-selectively at 42◦C and then tested for loss of all antibiotic resistances.

## Plasmid Constructs

fmicb-10-03015 December 30, 2019 Time: 16:53 # 3

The pCA24N vector was used to express target genes in E. coli. The coding region of mcr-1 was PCR-amplified from genomic DNA of E. coli carrying pHNSHP45 using the primer pair pCA24N-mcrF/pCA24N-mcrR. PCR products were digested with SalI and XbaI and inserted into the corresponding sites of pCA24N. The correct constructs were verified by PCR with primer pair pCA24N-F/pCA24N-R and DNA sequencing. The same procedures were performed to fuse gfp before the stop codon of the mcr-1 gene, as well as for the construction of the plasmid pCA24N-mcr-1-gfp. To generate pCA24N-mcr-1 gfp, the coding region of mcr-1 without its stop codon was amplified with primers Mcr-1F(salI) and Mcr-1R(xbaI), and the coding region of gfp was amplified using pCA24N-gfp (Guo et al., 2017) as the template with primers Gfp-F(xbaI) and Gfp-R(EcoRI). The mcr-1 fragment was digested with SalI and XbaI, the gfp fragment was digested XbaI and RcoRI, and inserted into the corresponding sites of pCA24N. pCA24N-based expression vectors were transferred into the E. coli BW25113 host.

## Antibiotics Susceptibility Testing

The antibiotics tested include ampicillin, polymyxin B (PB), ceftazidime, ciprofloxacin, gentamycin (GEN), kanamycin, chloramphenicol, tetracycline, rifampicin, nalidixic acid, and spectinomycin. The antibiotic resistance level was described by the minimum inhibitory concentrations (MICs) determined using a custom-made 96-well MIC panel (Flentie et al., 2019). The results were interpreted according to the criteria of the Clinical and Laboratory Standards Institute (CLSI) (Clinical and Laboratory Standards Institute, 2017).

## Microscopy Exam

To evaluate cell membrane integrity, the membrane-specific redfluorescent dye FM4-64 (Thermo Fisher Scientific, Rockford, IL, United States) was used. Overnight cultures were diluted in 50 ml of fresh medium and cultured to an OD600 nm of 0.5, and cells were harvested by centrifugation (6000 × g, 2 min), washed and resuspended in phosphate buffered saline (PBS). Cells were then treated with 0.85% NaCl or 7.0% NaCl for 30 min followed by staining with 4 µg/ml FM 4-64 for 15 min in the dark at ambient temperature. Bacterial cells were imaged using a Zeiss Axiovert fluorescence microscope (Carl Zeiss Inc., Thornwood, NY, United States).

## Protein Localization

For localization of MCR-1 by GFP fusion, overnight cultures of BW25113 carrying pCA24N-mcr-1-gfp were inoculated into LB broth supplemented with chloramphenicol (30 µg/ml) to an OD600 nm of 0.1 and 0.5 mM IPTG was added to induce MCR-1-GFP expression for 2 h before imaging. Cells were washed with PBS and imaged with fluorescence microscopy (Zeiss Axiophot) using an oil immersion objective (100×).

## RESULTS AND DISCUSSION

## The Expression of mcr-1 Decreases Resistance to Hydrophobic Antibiotics

The plasmid pHNSHP45 was extracted from E. coli K12 C600/pHNSHP45 using the EZNA <sup>R</sup> plasmid mini kit I (Omega) and then electroporated into the E. coli K12 BW25113 host, which is susceptible to polymyxin B (MIC = 0.5 µg/mL). Positive clones were selected with 2 µg/mL polymyxin B and confirmed by PCR with the primer pair CLR5-F/R (**Supplementary Table S1**). To explore the impact of mcr-1, we deleted the mcr-1 gene with a one-step inactivated method using a λ-red recombinase expression plasmid pKD46 as described by Datsenko and Wanner (2000). The replacement of the mcr-1 gene with the cat cassette amplified from pKD3 (strain BW25113/pHNSHP45 1mcr-1:cat) and the elimination of the FRT-flanked cat cassette by using the FLP recombinase expression plasmid pCP20 to obtain strain BW25113/pHNSHP45 1mcr-1 were confirmed by PCR and DNA sequencing (**Figure 1**). As expected, the deletion of mcr-1 resulted in decreased resistance to polymyxin B (MIC = 0.5 µg/mL) (**Figure 1B**). The MIC value of PB for the mcr-1 deletion mutant was 0.5 µg/ml compared to an 8 µg/ml for the wild type (**Table 2**).

The asymmetric lipopolysaccharide (LPS)-phospholipid bilayer of the OM provides a formidable permeability barrier for both hydrophilic and hydrophobic antibiotics (Nikaido, 2003; Zgurskaya et al., 2015). Previous studies have found that small hydrophilic drugs use the pore-forming porins to cross the OM, while hydrophobic drugs diffuse across the LPS-phospholipid bilayer (Vaara, 1992; Nikaido, 2003). To explore whether MCR-1 expression demonstrates differences in different families of antibiotics, we performed MIC tests of the 1mcr-1 and wild-type strains using a 96-well MIC panel test assay that contained 11 antibiotics. As expected, the 1mcr-1 strain showed increased resistance to gentamicin, kanamycin and rifampicin. However, no effect was found for resistance to ampicillin, nalidixic acid, spectinomycin, or ciprofloxacin (**Table 2**). Interestingly, gentamicin, kanamycin, and rifampicin are hydrophobic antibiotics, while ampicillin, nalidixic acid, spectinomycin, and ciprofloxacin are hydrophilic antibiotics. Thus, the deletion of mcr-1 also increases resistance to hydrophilic antibiotics. mcr-1 confers colistin resistance through the addition of cationic phosphoethanolamine (pEtN) to phosphate groups on the lipid A component of LPS, which reduces the net anionic charge of the cell surface (Jeannot et al., 2017). LPS modification in Gram-negative bacteria plays a significant role in resistance to antimicrobial factors (Gunn, 2001). Thus, we proposed that the expression of mcr-1 resulted in the modification of LPS and disrupt the organization of the LPS-phospholipid bilayer, change its permeability, and therefore decrease the resistance to hydrophobic antibiotics, but the hydrophilic antibiotics traverse the OM through porin channels.

primer pair MCR-LF/LR. M, DNA marker DS5000. 1, the wild-type strain (E. coli K12 BW25113/pHNSHP45); 2, the deletion mutant 1mcr-1:cat (E. coli K12 BW25113/pHNSHP45 1mcr-1:cat); 3, the deletion mutant 1mcr-1 (E. coli K12 BW25113/pHNSHP45 1mcr-1). (B) The polymyxin B (PB) susceptibility test of the wild-type strain and the 1mcr-1 strain.

## The Expression of mcr-1 Decreases Survival During High Salt Stress

Acquired antibiotic resistance by horizontal gene transfer tends to be related to a fitness cost for bacterial hosts (Andersson and Levin, 1999; Andersson and Hughes, 2010; Vogwill and MacLean, 2015). The osmotic stresses of E. coli strains with and without mcr-1 were tested. The growth of the wild type strain was severely hindered in the presence of 7.0% NaCl in LB broth (total NaCl concentration was 7.0%) when compared with the 1mcr-1 strain, while no growth defect was observed for the two strains in regular LB broth containing 1.0% NaCl (**Figure 2A**). In addition, the growth of E. coli with or without mcr-1 in LB

TABLE 2 | Antimicrobial susceptibility of E. coli BW25113, the wild-type strain and the mcr-1 deletion mutant.


MIC values (µg/mL) of 11 antibiotics are listed. Three independent cultures of each strain were tested.

broth with different NaCl concentration was tested and the cell density was measured by optical density at 600 nm. The results showed that the cell density of the E. coli was decreased with the increase of NaCl concentration, and E. coli without mcr-1 was able to tolerance to higher NaCl concentration than E. coli with mcr-1 (**Supplementary Table S2**). Furthermore, we examined cell integrity using the red membrane dye FM 4-64 (Life Technologies, United States), which specifically stains the cell membrane. Exponentially growing cells (OD600 nm∼ 0.5) were collected, washed and incubated with 0.85% NaCl or 7.0% NaCl for 30 min. Cells were then collected and imaged by fluorescence microscopy (Zeiss Axiophot) using an oil immersion objective (100×). The cell membranes appeared intact in the presence of low concentrations of NaCl, while membrane integrity of wild type strain was only severely reduced in the presence of 7% NaCl (The membrane defects of wild type strain was about 79.2% compared with 1mcr-1 strain is 19.4%) (**Figure 2B**). Consistent with the above results, the 1mcr-1 strain demonstrated 10 fold higher viability than the wild-type cells in the presence of 7% NaCl (**Figure 2C**). Taken together, these results showed that the expression of the mcr-1 gene decreases tolerance to high salt stress. We proposed that the mechanism of the expression of the mcr-1 gene decreases cell fitness under high salt conditions may be due to expression of the mcr-1 gene affects the membrane permeability.

## The Expression of mcr-1 Affects the Growth and Cell Morphology of E. coli

After prolonged incubation on LB agar plates for 24 h, the edge of the colonies formed by the 1mcr-1 strain appeared more wrinkle compared with the wild-type strain (**Figure 2D**). Furthermore, the mcr-1 coding region from pHNSHP45 was cloned into plasmid pCA24N to construct pCA24N-mcr-1. To

30 min of incubation with LB broth containing 0.85% NaCl or 7.0% NaCl. (C) Cell viability of exponentially growing cells (OD600 nm∼0.5) after 30 min of incubation with LB broth containing 0.85% NaCl or 7.0% NaCl. Experiments were conducted with three independent cultures, and only representative images are shown in panels (B,C). (D) Colony morphology of E. coli with or without mcr-1 on LB agar plates.

further check the cell morphology of E.coli, the complementation experiments were carried out. As expected, the edge of the colonies formed by the 1mcr-1/pCA24N strain appeared more wrinkle compared with the 1mcr-1/pCA24N-mcr-1 (**Figure 2D**). Scanning electron microscopy (SEM) was then employed to study the cell morphology of E. coli K-12 BW25113/pCA24N-mcr-1 with the addition of 0.5 mM IPTG. As shown in **Figure 3A**, cells overexpressing mcr-1 via pCA24N-mcr-1 had rougher cell envelopes than the cells carrying empty pCA24N. In addition, overexpression of mcr-1 significantly reduced the cell growth rate (**Figure 3B**), which is consistent with a recent report indicating that increased expression of mcr-1 results in decreased growth rate and cell viability (Yang et al., 2017). To further check the localization of MCR-1 in E. coli, the green fluorescence protein gene gfp was fused to the C-terminus of the mcr-1 gene to express the fused protein MCR-1-GFP. As expected, MCR-1- GFP was located around the cellular membrane (**Figure 3C**), which further confirmed that MCR-1 is a membrane protein. As a control, the green fluorescence protein GFP was produced using pCA24N-gfp and the protein GFP was localized to the cytoplasm (**Supplementary Figure S1**). To further determine if the MCR-1- GFP fusion protein confers polymyxin resistance, we performed MIC tests of the BW25113/pCA24N and BW25113/pCA24Nmcr-1-gfp strains using a 96-well MIC panel test assay. The MIC value of PB for the BW25113/pCA24N-mcr-1-gfp was 8.0 µg/ml, but BW25113/pCA24N was 0.5 µg/ml, suggesting that GFP fused to the MCR-1 does not affect the polymyxin resistance of MCR-1. Thus, MCR-1 is localized at the membrane, and the expression of mcr-1 affects cellular morphology and colony morphology.

CONCLUSION

In this study, we demonstrated that MCR-1 is a membrane protein that localizes to the cellular membrane. Furthermore, MCR-1 increases the loss of the cell membrane integrity and decreases the MICs of gentamicin, kanamycin and rifampicin. Evolving colistin resistance by acquiring mcr-1 therefore challenges bacterial populations with an evolutionary tradeoff: input of mcr-1 protects the host against colistin but changes the membrane permeability and reduces resistance to hydrophobic antibiotics. This trade-off may further explain the balance between mcr-1 expression and bacterial survival. Several studies have documented a link between antibiotic use and the development of antibiotic resistance (Goossens et al., 2005; Bergman et al., 2009). Our results provide a further possibility of mcr-1 gene transfer, which was affected not only by colistin use but also by environmental cues, such as osmotic pressure conditions.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## AUTHOR CONTRIBUTIONS

XW and BL conceptualized and designed the project. BL, XW, FY, XZ, YG, WW, HZ, PW, and YY did the investigation, and data curation and analysis. BL and XW did the supervision and visualization. BL and XW wrote, reviewed, and edited the original draft.

## FUNDING

This work was supported by the National Science Foundation of China (31290233, 41230962, 41406189, and 31470141), the Chinese Academy of Sciences (XDA11030402), the China Postdoctoral Science Foundation (2017M622649), and the Science and Technology Innovation Leading Talent Program of Guangdong Province (2015TX01N036). XW is the recipient of the 1000-Youth Elite Program award (a program for the recruitment of global experts to China).

## REFERENCES

fmicb-10-03015 December 30, 2019 Time: 16:53 # 7


## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.03015/full#supplementary-material

crystal structures of the catalytic domain of MCR-1. Sci. Rep. 7:39392. doi: 10.1038/srep39392


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

Copyright © 2020 Li, Yin, Zhao, Guo, Wang, Wang, Zhu, Yin and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# CRISPR-Cas System in Antibiotic Resistance Plasmids in Klebsiella pneumoniae

#### Muhammad Kamruzzaman<sup>1</sup> \* and Jonathan R. Iredell1,2 \*

<sup>1</sup> Centre for Infectious Diseases and Microbiology, The Westmead Institute for Medical Research, The University of Sydney, Westmead, NSW, Australia, <sup>2</sup> Westmead Hospital, Westmead, NSW, Australia

CRISPR-Cas (clustered regularly interspersed short palindromic repeats-CRISPR-

#### Edited by:

Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil

## Reviewed by:

Ørjan Samuelsen, University Hospital of North Norway, Norway Edze Westra, University of Exeter, United Kingdom

#### \*Correspondence:

Muhammad Kamruzzaman muhammad.kamruzzaman@ sydney.edu.au Jonathan R. Iredell jonathan.iredell@sydney.edu.au

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 02 October 2019 Accepted: 06 December 2019 Published: 10 January 2020

#### Citation:

Kamruzzaman M and Iredell JR (2020) CRISPR-Cas System in Antibiotic Resistance Plasmids in Klebsiella pneumoniae. Front. Microbiol. 10:2934. doi: 10.3389/fmicb.2019.02934 associated protein) is a microbial adaptive immune system involved in defense against different types of mobile genetic elements. CRISPR-Cas systems are usually found in bacterial and archaeal chromosomes but have also been reported in bacteriophage genomes and in a few mega-plasmids. Klebsiella pneumoniae is an important member of the Enterobacteriaceae with which they share a huge pool of antibiotic resistance genes, mostly via plasmids. CRISPR-Cas systems have been identified in K. pneumoniae chromosomes, but relatively little is known of CRISPR-Cas in the plasmids resident in this species. In this study, we searched for CRISPR-Cas system in 699 complete plasmid sequences (>50-kb) and 217 complete chromosomal sequences of K. pneumoniae from GenBank and analyzed the CRISPR-Cas systems and CRISPR spacers found in plasmids and chromosomes. We found a putative CRISPR-Cas system in the 44 plasmids from Klebsiella species and GenBank search also identified the identical system in three plasmids from other Enterobacteriaceae, with CRISPR spacers targeting different plasmid and chromosome sequences. 45 of 47 plasmids with putative type IV CRISPR had IncFIB replicon and 36 of them had an additional IncHI1B replicon. All plasmids except two are very large (>200 kb) and half of them carried multiple antibiotic resistance genes including blaCTX−M, blaNDM, blaOXA. To our knowledge, this is the first report of multi drug resistance plasmids from Enterobacteriaceae with their own CRISPR-Cas system and it is possible that the plasmid type IV CRISPR may depend on the chromosomal type I-E CRISPRs for their competence. Both chromosomal and plasmid CRISPRs target a large variety of plasmids from this species, further suggesting key roles in the epidemiology of large plasmids.

#### Keywords: Klebsiella pneumoniae, Enterobacteriaceae, plasmid, CRISPR, antibiotic resistance

## INTRODUCTION

Acquisition of genetic material including virulence, fitness and antibiotic resistance genes by horizontal gene transfer (HGT) is an essential process in bacterial adaptation to different environments (Frost et al., 2005). In addition bacteria have acquired an adaptive immune system, clustered regularly interspaced short palindromic repeats and their associated Cas proteins

(CRISPR-Cas), which helps to limit the acquisition of genetic materials and defend against invasive bacteriophages and plasmids (Garneau et al., 2010; Barrangou, 2015; Samson et al., 2015).

A typical CRISPR-Cas locus is comprised of a CRISPR array, Cas genes and a leader sequence. A CRISPR array is comprised of nearly identical short (21 to 47 nucleotides) direct repeats, separated by unique DNA fragments (spacers) acquired from foreign DNA [mobile genetic elements (MGEs)]. The leader sequence is usually a (∼100–500 bp) AT rich region believed to serve as a promoter for the transcription of the CRISPR array (Marraffini, 2015). The CRISPR-Cas defense mechanism can be considered as three steps. In an initial adaptation step foreign DNA fragments (protospacers) from infecting bacteriophages and plasmids are incorporated into the CRISPR array as new spacers. These spacers provide the sequence specific memory for a targeted defense against subsequent invasions by the same bacteriophage or plasmid. The CRISPR array transcript is then processed to matured CRISPR RNAs (crRNAs). After expression of the array, mature crRNAs, aided by Cas proteins, identify specific targets and cleave the nucleic acid strands of corresponding viruses or plasmids (van der Oost et al., 2009; Garneau et al., 2010; Makarova et al., 2011b, 2015; Barrangou, 2015).

CRISPR-Cas systems show a great deal of diversity in their Cas protein composition, structure of effector proteins complex, genetic organization and localization in the genome, mechanism of adaptation, crRNA processing and interference. Based on the effector complexes CRISPR-Cas systems can be divided into two classes and six types (Class 1, including types I, III, IV and class 2, including types II, V and VI), those can be subdivided into at least 34 sub-types (Makarova et al., 2011b, 2015, 2018; Koonin et al., 2017; Hille et al., 2018), Class 1 CRISPR-Cas system provides interference by using multi-Cas effector protein complex whereas Class 2 uses single effector protein for interference (Hille et al., 2018). Generally there are signature genes for each type of CRISPR-Cas system and those include cas3 for type I, cas9 for type II, cas10 for type III, csf1 (large subunit, cas8-like) for type IV, cas12 for type V, and cas13 for type VI (Makarova et al., 2018). Types I and II CRISPR-Cas systems provide immunity against DNA (Brouns et al., 2008; Gasiunas et al., 2012) whereas type III systems may target DNA or RNA (Tamulaitis et al., 2014). Types I–III are well-studied and are generally found in chromosomes of bacteria and archaea, whereas types IV, V, and VI are three putative new types. Type IV systems are usually localized on plasmids or other MGEs and lack apparent adaptation modules (cas1 and cas2) and type V was identified in archaeal chromosome only (Makarova et al., 2011a, 2015). Type VI is another new type identified recently carrying HEPN-domain containing effector protein Cas13, which, unlike, other class II effector cleaves single stranded RNA (ssRNA) (Hille et al., 2018). HEPN RNase is a toxin domain of bacterial toxinantitoxin module and suggests that type VI includes dedicated RNA-targeting CRISPR-Cas system (Makarova et al., 2011b, 2018; Koonin et al., 2017).

The classification, functions and mechanism of actions of all CRISPR-Cas systems are well-characterized except type IV. A recent study demonstrated that the function of type IV system in the maturation of crRNAs and in the subsequent formation of a Cascade-like crRNA-guided effector complex (Ozcan et al., 2019). Type IV system can be classified as two sub-types, type IV-A and IV-B, based on the presence of DinG family helicase and type IV specific effector protein Csf5. Type IV-A encodes a DinG helicase (Csf4) and an effector protein Csf5 and whereas type IV-B lacks these proteins (Makarova et al., 2015, 2018; Koonin et al., 2017; Hille et al., 2018; Pinilla-Redondo et al., 2019) and usually, the type IV-A system carries CRISPR-array.

Klebsiella pneumoniae, a member of the bacterial family Enterobacteriaceae, is a common opportunistic hospital associated pathogen, accounting for about one third of total Gram-negative infections (Navon-Venezia et al., 2017). It causes a variety of infections including urinary tract infections, pneumonia, cystitis, wound infections, and life-threatening sepsis (Podschun and Ullmann, 1998). Occurrence of transmissible antibiotic resistance in this organism is a major problem worldwide. K. pneumoniae have a huge pool of antibiotic resistance genes that they share among other Enterobacteriaceae, mostly via self-transferrable plasmids (Navon-Venezia et al., 2017). Almost all modern antibiotic resistance (to carbapenems, cephalosporins, aminoglycosides, now even colistin) in these organisms is encoded on large (40–200 kb) low-copy (1–6 per cell) conjugative plasmids (Carattoli, 2009; Navon-Venezia et al., 2017). Plasmid-borne antibiotic resistance is acquired very quickly and, once acquired, could become fixed in the bacterial accessory genome by 'addiction systems' that poison cells from which the antibiotic resistance plasmid is lost (Hayes, 2003).

Several studies have identified CRISPR-Cas systems in K. pneumoniae chromosomes as I-E and I-E<sup>∗</sup> types (Ostria-Hernandez et al., 2015; Shen et al., 2017; Li et al., 2018) but little is known about CRISPR-Cas systems of plasmids in K. pneumoniae and other Enterobacteriaceae (Enas Newire et al., 2019). CRISPR-Cas systems are associated with relative antibiotic susceptibility in Streptococcus pyogenes and E. coli and the chromosomal CRISPR-Cas system is known to interfere with acquisition of antibioticresistant plasmids in E. coli (Zheng et al., 2014; Aydin et al., 2017). In this study, we examined 699 complete plasmid sequences from K. pneumoniae and 217 K. pneumoniae chromosomal sequences from the GenBank for the presence of CRISPR-Cas system and further analyzed the identified CRISPR-Cas systems and their spacers.

## MATERIALS AND METHODS

## Extraction of Complete Nucleotide Sequence of Plasmids and Chromosomes for K. pneumoniae From the GenBank

Klebsiella pneumoniae chromosome and plasmid sequences available in the GenBank database<sup>1</sup> were downloaded and

<sup>1</sup>https://www.ncbi.nlm.nih.gov/genome/?term=Klebsiella+pneumoniae

subjected to CRISPR analysis. For complete K. pneumoniae chromosomal sequences, after opening the database link, we selected "genome assembly and annotation report" and chose complete sequences and then extracted all the complete nucleotide sequences individually and saved as FASTA format sequence file. For complete plasmid sequences, in the same link we selected "plasmid annotation report" and downloaded plasmid sequences > 50 kb and saved as separate FASTA files for individual plasmid sequences.

## Identification and Characterization of CRISPR-Cas in Plasmid and Chromosomal Sequences

CRISPR was identified with CRISPRFinder<sup>2</sup> (Grissa et al., 2007) software. This algorithm locates direct repeat sequences of 23– 55 bp separated by variable sequences of a size no greater than 2.5 times or no less than 0.6 times the length of the repeated sequences (25–60 bp). When the algorithm detects at least three repeating regions that are exactly the same (in sequence and size), which are separated by variable sequences, it is considered a "confirmed CRISPR." If the algorithm locates two repeats separated by a variable sequence, it establishes the status of a "questionable CRISPR." For the present study we only considered those indicated by the program as "confirmed CRISPRs." In addition, with this platform, we searched for cas genes in regions adjacent to CRISPR sequences. Fasta formatted complete nucleotide sequence of each individual plasmid or chromosome was uploaded in the CRISPRFinder and run the program by using a default setting parameters and outcomes provided the possible CRISPR-array (CRISPR repeats and spacers). Spacers sequences were collected from CRISPRFinder outputs and saved to use for further analysis. The CRISPR region identified by CRISPRFinder was then detected on the plasmid or chromosomal sequences and nearly 10 kb upstream and downstream regions were analyzed for putative cas genes. The CRISPR-array neighboring genes and their respective protein sequences were analyzed by BLASTn and BLASTp searches for the GenBank identity. For nucleotide sequence analysis, megablast was performed by using following parameters: (i) expectation threshold (e-values) less than or equal to 0.01 and a score greater than 40, (ii) maximum target sequence was set at 1000, (iii) automatically adjusted parameters for short input sequences, (iv) different match/mismatch scores were selected to identify highly conserved to low conserved sequences. BLASTp for protein sequences were performed against nonredundant protein sequence database and against reference proteins sequence database with expectation values (e-value) less than or equal to 0.01 were considered significant as well as a coverage percentage of more than or equal to 80%. The identified CRISPR-array and cas genes were further verified by using CRISPRone software<sup>3</sup> (Zhang and Ye, 2017). The individual fasta formatted nucleotide sequence of plasmid and chromosome was run through CRISPRone software by using default settings.

## Search for Similar CRISPR-Cas System in GenBank Data

The cas genes identified in the putative type IV CRISPR-Cas systems in the plasmids of K. pneumoniae were used to fish similar type of CRISPR-Cas system in the GenBank data. Both the cas genes nucleotide sequences and amino acid sequences were used separately for BLASTn and BLASTp search in the GenBank data with the parameters mentioned earlier. The additional plasmid sequences identified with identical cas genes or Cas proteins were downloaded and analyzed for CRISPR-array and cas genes orientation by CRISPRFinder and CRISPRone software.

## Analysis of CRISPR Spacers and Identification of Spacers Protospacers Match

Spacers from respected plasmid CRISPR-Cas system were extracted from CRISPRFinder outputs and made a fasta formatted sequence file for all spacer pool by BioEdit software<sup>4</sup> . Each of the spacers sequence, their reverse complement sequence and both 3<sup>0</sup> and 5<sup>0</sup> truncated version were then searched against the spacer pool and identified all the unique spacers found in the plasmid CRISPRs and then plotted their distribution. Each of the unique spacer was then analyzed for their identity (match with protospacers) to GenBank sequences by nucleotide blast search (BLASTn) with parameters described earlier.

## Identification of Chromosomal CRISPR Type

Two different types of cas1 and cas3 alleles were found in K. pneumoniae genomes and CRISPR-Cas systems were further divided into types I-E or I-E<sup>∗</sup> (Li et al., 2018) on the basis of the cas1 and cas3 alleles and their localization in the chromosome.

## Plasmid Characterization

The presence of antibiotic resistance genes in sequenced plasmids were identified by ResFinder 3.2<sup>5</sup> and plasmid replicon types by PlasmidFinder 2.1<sup>6</sup> (Carattoli et al., 2014).

## RESULTS

## CRISPR-Cas System in K. pneumoniae Plasmids

A total of 699 complete plasmid sequences of > 50-kb in size found in K. pneumoniae were extracted from the GenBank database. CRISPR-arrays were identified in 5% (37 of 699; **Table 1** and **Supplementary Table S1**) of the plasmids. The identified CRIPSR-arrays had direct repeats of 23–30 bp separated by a variable number (0–22) of spacer sequences of 25–57 bp and most of them are 30–33 bp long (**Supplementary Table S2**). Immediately upstream of the CRISPR-array an ∼130 bp conserved AT rich region was present, which may act as a

<sup>2</sup>https://crispr.i2bc.paris-saclay.fr/Server/

<sup>3</sup>http://omics.informatics.indiana.edu/CRISPRone/

<sup>4</sup>https://bioedit.software.informer.com/7.2/

<sup>5</sup>https://cge.cbs.dtu.dk/services/ResFinder/

<sup>6</sup>https://cge.cbs.dtu.dk/services/PlasmidFinder/

Kamruzzaman and Iredell

fmicb-10-02934 December 23, 2019 Time: 16:44 # 4

#### TABLE 1 |Characteristicsof CRISPR-Cas positive plasmids.


(Continued)

Plasmid Mediated CRISPR-Cas

# TABLE

 1 | Continued

#### Sl no. Plasmid Speciesa Accession no. CRISPR-array position DR-length (bp) No. of spacers Plasmid size (kb) Replicon type<sup>b</sup> Antibiotic resistance genes identified (annotation by ResFinderc) 16 p44-1 Kp CP025462.1 42604–43120 29 8 261 IncFIB None 17 TVGHCRE225 plasmid unnamed1 Kp CP023723.1 10881–11763 29 14 297 IncHI1B, IncFIB None 18 pOXA1\_020030 Kp CP028791.1 30205–30843 29 10 288 IncFIB blaOXA−1, aac(60)-Ib-cr, sul1 19 825795-1 plasmid unnamed1 Kp CP017986.1 148524–149585 30 17 244 IncHI1B, IncFIB None 20 pKp\_Goe\_304-1 Kp CP018720.1 24563–25679 24 18 246 IncHI1B, IncFIB None 21 pKp\_Goe\_021-1 Kp CP018714.1 7697–8748 30 17 246 IncHI1B, IncFIB None 22 pKp\_Goe\_026-1 Kp CP018708.1 54661–55777 24 18 246 IncHI1B, IncFIB None 23 pKp\_Goe\_024-1 Kp CP018702.1 5006–6122 24 18 246 IncHI1B, IncFIB None 24 pKp\_Goe\_832-1 Kp CP018696.1 102143–103259 24 18 246 IncHI1B, IncFIB None 25 pKp\_Goe\_473-1 Kp CP018687.1 116666–117727 30 17 246 IncHI1B, IncFIB None 26 pKp\_Goe\_579-1 Kp CP018313.1 94773–95834 30 17 246 IncHI1B, IncFIB None 27 AR\_0363 plasmid unnamed4 Kp CP027156.1 127062–128305 29 20 186 IncHI1B, IncFIB aac(60)-Ib, aac(60)-Ib-cr 28 pKPN-edaa Kp CP026398.1 7401–8525 29 18 249 IncHI1B, IncFIB blaNDM−1, blaOXA−1, blaDHA−1, aph(300)-Ib, aph(6)-Id, aac(60)-Ib-cr, qnrB4, sul1 29 pKPN-bbef Kp CP026172.1 227283–228407 29 18 244 IncHI1B, IncFIB None 30 pKpvST101\_5 Kp CP031372.1 144828–145262 25 10 210 IncHI1B, IncFIB None 31 pKpn23412-362 Kp CP011314.1 141865–142618 23 12 362 IncHI1B, IncFIB blaOXA−1, blaCTX−M−15, blaTEM−1B, blaOXA−1, aac(3)-IIa, aph(300)-Ib, aph(6)-Id, aac(60)-Ib-cr, sul2, dfr1 32 p1502320-3 Kp CP031580.1 7608–8781 29 19 87 ND None 33 pKP3301 DNA Kp AP018748.1 3677–4373 29 11 296 IncHI1B, IncFIB blaNDM−1, blaCTX−M−15, blaTEM−1A, blaOXA−1, aph(30)-VI, armA, aadA2, aac(60)-Ib, aac(60)-Ib-cr 34 SKGH01 plasmid unnamed 1 Kp CP015501.1 7545–7875 29 15 281 IncHI1B, IncFIB None 35 Plasmid\_A\_Kpneumoniae\_MS6671 Kp LN824134.1 16764–17823 29 17 280 IncHI1B, IncFIB None 36 pKJNM8C2.1 Kp CP030858.1 10639–11155 30 8 304 IncHI1B, IncFIB blaNDM−1, blaCTX−M−15, blaOXA−1, armA, aadA2, aac(60)-Ib-cr, qnrB1, sul1, dfrA12

fmicb-10-02934 December 23, 2019 Time: 16:44 # 5

Plasmid Mediated CRISPR-Cas

Plasmid Mediated CRISPR-Cas

fmicb-10-02934 December 23, 2019 Time: 16:44 # 6

# Frontiers in Microbiology | www.frontiersin.org



aKp, K. pneumoniae; Pg, Pluralibacter gergoviae; Ec, E. coli; Ko, Klebsiella oxytoca; Ro, Raoultella ornithinolytica; Kv, Klebsiella variicola.<sup>b</sup>ND, not detected.caac(60)-Ib also commonly annotated as aacA4; aph(300)-Ib, aph(6)-Id also commonly annotated as strA, strB.

**223**

leader sequence of this CRISPR. We also identified csf2, csf3, DinG helicase (csf4), cas6 (csf5), csx3, and cas10 homologs upstream and a reverse transcriptase (RTase) or maturase gene downstream of the CRISPR-array (**Figure 1**). Two genes of unknown function were also present in the Cas genes locus, but we could not identify the adaptation genescas1 orcas2, or evident homologs, in these plasmids. The structure of the CRISPR-Cas array and organization of cas genes identified here is very close to that of the type IV CRISPR-Cas system previously identified in the mega-plasmid of Aromatoleum aromaticum EbN1, an aromatic-degrading betaproteobacteria found in freshwater and soil habitats (Ozcan et al., 2019), even though a large subunit (Csf1) that acts as the signature protein for this type (Makarova et al., 2015) is absent from the system identified here. The Cas genes and their orientations in identified CRISPR-Cas systems are very similar among the plasmids except for the presence of Insertion Sequences (ISs) insertion events in some. ISs were also identified between the CRISPR-array and RTase gene in a few plasmids.

## CRISPR-Cas System in Other Enterobacteriaceae Plasmids

A BLAST search identified identical CRISPR-Cas systems in 10 other plasmids in GenBank, in addition to the 37 plasmids identified in K. pneumoniae (**Table 1**). Five of these 10 plasmids were from K. pneumoniae and 1 each from E. coli, K. oxytoca, Pluralibacter gergoviae, K. variicola, and Raoultella ornithinolytica. One of the plasmids from K. pneumoniae has a cas locus without any CRISPR-array identified. No match was found in chromosomal sequences on GenBank.

## Characterization of CRISPR-Positive Plasmids

A total of 47 type IV CRISPR-positive plasmids were analyzed. All but two are very large (>200 kb) and the largest is 430 kb (**Table 1**). Almost all had an IncFIB replicon identified by PlasmidFinder except for two in which no replicon match was identified. Most plasmids (36/47) also have an IncHI1B replicon (**Table 1**). Interestingly, although most of the plasmids (44/47) were found in Klebsiella species, only one plasmid has the characteristic Klebsiella type IncFIB<sup>K</sup> replicon (Garcia-Fernandez et al., 2012). The %GC content of almost all CRISPR-negative plasmids is > 50% and lower (∼44–46%) in CRISPR-positive plasmids (**Supplementary Table S1**). The only CRISPR-positive plasmid with an IncFIB<sup>K</sup> replicon had a GC content of ∼52%, similar to other CRISPR-negative plasmids of K. pneumoniae.

ResFinder identified multiple antibiotic resistance genes including blaCTX−M, blaNDM, blaOXA, armA and qnr genes, associated with resistance to β-lactam, carbapenem, aminoglycoside and quinolone antibiotics, in almost half the plasmids (24/47; **Table 1**).

## Analysis of Spacers From Plasmid CRISPRs

A total of 623 spacers from 46 CRISPR-positive plasmids were analyzed by BLASTn search, identifying 67 unique spacer sequences that made up the total pool of 623, including some repetition or reversed orientation of the same sequence, along with loss or gain of a few nucleotides (**Figure 2** and **Supplementary Table S2**). Spacer numbers varies without any relationship with plasmid size or the presence or absence of antibiotic resistance genes (**Table 1**). Five unique spacers (SP11, 36, 43, 55, 57) were found specific for plasmid sequences other than the match with CRISPR-array region and one of them (SP11) appears to match with the traL gene of 457 different CRISPR-negative plasmid sequences and another one (SP43) hits the traN gene of 260 different CRISPR-negative plasmid sequences in GenBank, mostly (>98%) from K. pneumoniae (**Table 2** and **Supplementary Table S2**). Both spacers are found in the CRISPR-array of 17 of the 46 CRISPR-positive plasmids we examined (**Table 2**). Spacers SP36, SP55, and SP57 have identity with plasmid transposases, transcriptional regulator and traH gene, respectively. Genes, traL, traN, and traH play an important role in plasmid transfer and are highly conserved among plasmids. Five unique spacers (SP8, 20, 42, 62, 63) were also found to match K. pneumoniae chromosomes but not those of their current host bacteria (**Table 2** and **Supplementary Table S2**). Two of them occurred in 17 of the 46 plasmids, targeting DUF1367 family protein and hypothetical protein genes for 111 and 139 K. pneumoniae chromosomes, respectively. Another spacer, that occurred in one plasmid only, appears to recognize a hypothetical protein gene from 3 of the K. pneumoniae chromosomes (**Table 2**). Many spacers (18/67) did not have any match in GenBank (**Supplementary Table S2**).

## Analysis of CRISPR-Cas System in K. pneumoniae Chromosomes

A total of 217 K. pneumoniae complete chromosomal sequences were extracted from GenBank (June 2019) and we identified that 81 of these (37%) carried CRISPR-Cas system on the chromosome. Of these 81, 45 were I-E and 36 were I-E<sup>∗</sup> type (**Supplementary Table S3**), consistent with previous reports (Ostria-Hernandez et al., 2015; Shen et al., 2017; Li et al., 2018).

CP018720.1), (∗∗∗CP018687.1, CP018714.1, CP017986.1), (#CP022612.1, CP012754.1, CP006799.1), (##MG845201.1, MG845200.1) and (###CP030858.1, CP008933.1) and marked with asterisk or hashtag were the representative from each group showed in the figure. The unique spacer sequences and their match with protospacers will be found in Supplementary Table S2.

## Relationship Between the Presence of Chromosomal CRISPR and Plasmid in K. pneumoniae Bacteria

We also gathered information about the presence of plasmids in those 217 K. pneumoniae isolates from GenBank. Most bacteria (185 of 217, 85%) carried plasmids, from 1 to 10 in number (**Figure 3A** and **Supplementary Table S3**) and 37% of these 217 had the putative CRISPR-Cas system. We found that the occurrence of chromosomal CRISPR is more in plasmid-free than plasmid-carrying strains (43 vs. 35%) (**Figure 3B**), a relationship that has been noted before (Li et al., 2018). We found that bacteria with chromosomal type I-E CRISPR had more plasmids (from 1 to 7 in number, most with 4–5 plasmids) whereas bacteria with chromosomal type I-E<sup>∗</sup> CRISPR had less plasmids (from 0 to 4 in number, mostly only 1 or 2 plasmids or none) (**Supplementary Table S3**).

## Analysis of Spacers From K. pneumoniae Chromosomal CRISPRs

A total of 2,464 spacers were extracted from chromosomal CRISPR-positive strains. A BLASTn search with these spacers

TABLE 2 | Spacers from plasmid mediated CRISPR specific to plasmids and K. pneumoniae chromosomes.


hp, hypothetical protein.


matched K. pneumoniae chromosomal sequences as well as different mobile elements including bacteriophages and plasmids. We identified 18 unique spacers matched sequences on plasmids (**Table 3**). Interestingly, 5 of these 18 spacers were from plasmid conjugative transfer region genes (traH, traG, traT, traN, traF) in several hundred different plasmid sequences in GenBank. Spacers matching traH and traG gene sequences from 413 and 401 different plasmids respectively were identified. One spacer matched the plasmid segregation gene parM, one matched the ubiquitous toxin antitoxin system gene hok-sok (**Table 3**), one matched the SAM-methyl transferase gene and another matched DNA sequence in a hypothetical gene located immediate upstream of SAM-methyltransferase on the plasmid. Three spacers matched different regions in the DUF3560 domaincontaining protein gene, which was found in turn on ∼1,000 plasmids in GenBank. One spacer matched ydeA, two matched a hypothetical protein and other three in intergenic regions of plasmid sequences; one spacer matched an intergenic region with identity to 524 different plasmids (**Table 3**) which, in a few cases, was present in multiple times in a single plasmid sequence.

## DISCUSSION

In this study we have identified a putative type IV CRISPR-Cas system in the plasmids of K. pneumoniae. To our knowledge, this is the first comprehensive report of type IV CRISPR-Cas system in plasmids of K. pneumoniae, specifically in antibiotic resistance plasmids.

Type IV CRISPR-Cas was previously identified in a megaplasmid of Aromatoleum aromaticum species, an aromaticdegrading β-proteobacteria found in freshwater and soils. The CRISPR-positive plasmids we describe are also very large (200– 430 kb; **Table 1**). Like the previously reported type IV CRISPR-Cas system (Makarova et al., 2015; Enas Newire et al., 2019; Pinilla-Redondo et al., 2019) homologs of cas region genes csf2, csf3, DinG helicase (csf4), cas6 (csf5) were present, the organization of the cas genes was very similar and the adaptation genes cas1 and cas3 were absent, all typical of the previously reported type IV CRISPR-Cas. The large subunit csf1 gene thought to be the signature gene for type IV system was absent in the CRISPR-Cas system reported here, but two additional cas

#### TABLE 3 | Plasmid specific spacers in K. pneumoniae chromosomal CRISPR.


<sup>a</sup>hp for hypothetical protein.

genes (csx3 and cas10 homologs) and two genes of unknown functions were identified in the cas gene locus, any of which may compensate for the absent csf1. The Cas1 protein of type IV CRISPR-Cas is a Zn-finger containing protein with a weak similarity to Zn-finger sequences of Cas10 and it has been suggested that Csf1 could be a highly divergent, inactivated and N-terminally truncated Cas10-like polymerase derivative (Makarova et al., 2011a). The presence of DinG helicase (csf4), only previously reported in type IV-A CRISPR-Cas (Koonin et al., 2017; Makarova et al., 2018; Pinilla-Redondo et al., 2019), further supports the designation of these plasmid systems as type IV-A CRISPR-Cas.

A preliminary study identified putative type IV CRISPR-Cas system in the IncH1B/IncFIB plasmids of Enterobacteriaceae (Enas Newire et al., 2019). In the present study most (45 of 47) of these very large CRISPR-carrying plasmids had an IncFIB replicon identified and most had an additional IncHI1B replicon. We identified only one of the IncF replicons that are thought to be typical (i.e., IncFIBK) of K. pneumoniae (Garcia-Fernandez et al., 2012; Villa and Carattoli, 2020). This suggests that CRISPR-positive plasmids might have originated from some other species in Enterobacteriaceae and transferred into K. pneumoniae. Analysis of %GC content shows that almost all CRISPR-positive plasmids have lower %GC (44–46%) than CRISPR-negative plasmids (∼50% or greater); we identified only one IncFIB<sup>k</sup> plasmid with CRISPR (pKPM501) and this had a 'normal' %GC of > 51%.

Plasmids have their own genetic modules that they can utilize to exist stably in certain bacterial host by competing with other plasmids. Plasmid incompatibility is one of the such mechanism by which two plasmids with similar or related replication genes cannot co-exist in the same cell. By interfering with host replication system only one plasmid of similar type can be efficiently replicate and segregate to daughter cell and others lost form the system (Novick, 1987; Austin and Nordstrom, 1990). Acquiring antibiotic resistance genes also give plasmids the advantage to maintain over sensitive plasmids at antibiotic selection pressure (Carattoli, 2013). Plasmid mediated toxin-antitoxin (TA) module also provide another alternative to displace incompatible plasmid by toxin mediated killing of plasmid free cells (Hayes, 2003; Yamaguchi et al., 2011). For example, if a cell carries two incompatible plasmids and one plasmid encodes a TA system, then after segregation of these incompatible plasmids, only plasmid carrying TA system will be maintained into daughter cells and cells carrying the other plasmid are eliminated from the population. Similar to those systems, it was suggested that plasmid mediated type IV CRISPR-Cas system may involve in the competition between plasmids by acquiring spacers specifically targeting different plasmids (Pinilla-Redondo et al., 2019). Chromosomal CRISPR are known to acquire spacers against different MGEs (Samson et al., 2015) and many plasmid-borne CRISPR spacers we found were also directed against other plasmids, including three unique spacers targeting 100% identical sequences (the common and highly conserved traN, traH, and traL of conjugative plasmids) in more than 700 other plasmids in GenBank. Large potentially expensive plasmids such as these CRISPR-positive plasmids may need this competitive edge and may reduce the overall plasmid burden in their host bacteria. Plasmid CRISPR spacers

targeting heterologous K. pneumoniae chromosomes may also have a role in determining the epidemiology of plasmids in this species.

Acquisition of a new plasmid produces burden to the host by reducing growth rate and lessened competitiveness of plasmidbearing hosts under conditions that do not select for plasmid genes (San Millan and MacLean, 2017). Although this fitness-cost can be mitigated over time through compensatory evolutions, however, the initial cost associated with plasmid carriage is one of the main barrier in the acquisition, maintenance and transfer of new plasmids (San Millan and MacLean, 2017; Dionisio et al., 2019). Multiple plasmids impose more fitness-cost related to single plasmid. Acquisition of plasmid mediated CRISPR spacers targeting other plasmids and host chromosome may provide advantage in the formation of plasmid co-integrate with other plasmids or integrated into the host chromosome by homologous recombination that might facilitate the stability and compatibility of the plasmids. CRISPR-Cas defense system not only identified in plasmids but also distributed in other MGEs including bacteriophages, T7-transposable elements and integrative conjugative elements (ICEs) (Faure et al., 2019; Koonin et al., 2019). The recruitment of CRISPR-Cas defense system by different MGEs may contribute to the evolution of both MGEs and defense systems.

Several previous studies identified and analyzed chromosomal CRISPR-Cas systems in K. pneumoniae by analyzing 52 (Ostria-Hernandez et al., 2015), 68 (Shen et al., 2017) and 97 (Li et al., 2018) complete and draft genome sequences. Here, we analyzed 217 complete K. pneumoniae chromosomes available in GenBank for the distribution of CRISPR-Cas systems, their types, acquired spacers and relationship between presence and absence of CRISPR and plasmids. Consistent with previous studies, we found type I-E and type I-E<sup>∗</sup> CRISPRs distributed in K. pneumoniae chromosomes. We found that chromosomal CRISPR-negative strains had more plasmids (**Figure 3** and **Supplementary Table S3**) and that K. pneumoniae with type I-E<sup>∗</sup> chromosomal CRISPR appeared to have less plasmids than those with type I-E.

Spacer sequences from chromosomal CRISPR matched different MGEs including plasmids. A total of 18 unique spacers were acquired from plasmids and many from conjugative transfer region genes, plasmid partition (parM) and stability genes (hok-sok). Acquired plasmid-specific spacers in K. pneumoniae chromosomal CRISPR may provide immunity against plasmids and, it has been suggested, promote or select for mobilization of important plasmid-borne antibiotic resistance genes such as blaCTX−<sup>M</sup> and blaKPC onto the chromosome (Huang et al., 2017). Similar phenomena have been directly observed for Streptococcus thermophilus CRISPR-Cas systems (Garneau et al., 2010).

Type IV CRISPR-Cas systems on plasmids lack genes for target cleavage enzymes (cas3 or cas10) (Makarova et al., 2015) but we have identified a putative cas10-like gene in these plasmid CRISPR-Cas system in Enterobacteriaceae. They also lack key adaptation modules (cas1 and cas2) but RNA processing and effector complex formation has been experimentally demonstrated for these systems in Aromatoleum aromaticum, in which a chromosomal type I-C CRISPR is also present (Ozcan et al., 2019). Importantly, we also noted that type IV CRISPR-Cas system-positive plasmids were found only in bacteria with chromosomal type I-E or I-E<sup>∗</sup> CRISPR-Cas, suggesting cross-talk between plasmid and chromosomal CRISPR which may compensate for the lack of adaptation and target cleavage functions encoded from plasmid mediated CRISPR.

Chromosomal CRISPR-Cas systems clearly protect some bacteria from horizontally acquired mobile elements (Palmer and Gilmore, 2010; Price et al., 2016). Multi-drug resistant Enterococcus lacking CRISPR-Cas (Palmer and Gilmore, 2010) more readily acquire new genes and adapt to new antibiotics (Price et al., 2016). Vibrio cholerae that acquired phage-inducible chromosomal islands (PICI) as a defense against bacteriophages (Novick et al., 2010; Seed, 2015) now must contend with bacteriophages that have acquired CRISPR-Cas with spacers directed against chromosomal PICI to inactivate that very defense system (Naser et al., 2017). We describe here a novel type IV CRISPR-Cas that is evidently circulating in Enterobacteriaceae plasmids, predominantly within K. pneumoniae, and appears to have a complementary relationship with chromosomal Type I-E/I-E<sup>∗</sup> CRISPRs. Plasmid CRISPR-Cas directed against other plasmids (and some K. pneumoniae chromosomes) provide another level of incompatibility in plasmid communities. Both plasmid and chromosomal CRISPR-Cas are evidently important determinants of the epidemiology of large antibiotic resistance plasmids in K. pneumoniae.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## AUTHOR CONTRIBUTIONS

MK conceived and designed the study, and generated the data. MK and JI analyzed the data and wrote the manuscript.

## FUNDING

This work was funded by a grant G1145914, from the National Health and Medical Research Council (NHMRC), Australia.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2019. 02934/full#supplementary-material

## REFERENCES



**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 Kamruzzaman and Iredell. 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.

# Linezolid and Rifampicin Combination to Combat cfr-Positive Multidrug-Resistant MRSA in Murine Models of Bacteremia and Skin and Skin Structure Infection

Yu-Feng Zhou1,2† , Liang Li<sup>3</sup>† , Meng-Ting Tao1,2, Jian Sun1,2, Xiao-Ping Liao1,2 , Ya-Hong Liu1,2,4 \* and Yan Q. Xiong3,5

<sup>1</sup> National Risk Assessment Laboratory for Antimicrobial Resistance of Animal Original Bacteria, College of Veterinary Medicine, South China Agricultural University, Guangzhou, China, <sup>2</sup> Guangdong Provincial Key Laboratory of Veterinary Pharmaceutics Development and Safety Evaluation, South China Agricultural University, Guangzhou, China, <sup>3</sup> The Lundquist Institute for Biomedical Innovation at Harbor-UCLA Medical Center, Torrance, CA, United States, <sup>4</sup> Jiangsu Co-Innovation Center for the Prevention and Control of Important Animal Infectious Diseases and Zoonoses, Yangzhou University, Yangzhou, China, <sup>5</sup> David Geffen School of Medicine at UCLA, Los Angeles, CA, United States

#### Edited by:

Leonardo Neves de Andrade, University of São Paulo, Brazil

### Reviewed by:

Sandeep Sharma, Lovely Professional University, India Xiang-Dang Du, Henan Agricultural University, China

#### \*Correspondence:

Ya-Hong Liu lyh@scau.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 09 September 2019 Accepted: 19 December 2019 Published: 14 January 2020

#### Citation:

Zhou Y-F, Li L, Tao M-T, Sun J, Liao X-P, Liu Y-H and Xiong YQ (2020) Linezolid and Rifampicin Combination to Combat cfr-Positive Multidrug-Resistant MRSA in Murine Models of Bacteremia and Skin and Skin Structure Infection. Front. Microbiol. 10:3080. doi: 10.3389/fmicb.2019.03080 Linezolid resistance mediated by the cfr gene in MRSA represents a global concern. We investigated relevant phenotype differences between cfr-positive and -negative MRSA that contribute to pathogenesis, and the efficacy of linezolid-based combination therapies in murine models of bacteremia and skin and skin structure infection (SSSI). As a group, cfr-positive MRSA exhibited significantly reduced susceptibilities to the host defense peptides tPMPs, human neutrophil peptide-1 (hNP-1), and cathelicidin LL-37 (P < 0.01). In addition, increased binding to fibronectin (FN) and endothelial cells paralleled robust biofilm formation in cfr-positive vs. -negative MRSA. In vitro phenotypes of cfr-positive MRSA translated into poor outcomes of linezolid monotherapy in vivo in murine bacteremia and SSSI models. Importantly, rifampicin showed synergistic activity as a combinatorial partner with linezolid, and the EC<sup>50</sup> of linezolid decreased 6-fold in the presence of rifampicin. Furthermore, this combination therapy displayed efficacy against cfr-positive MRSA at clinically relevant doses. Altogether, these data suggest that the use of linezolid in combination with rifampicin poses a viable therapeutic alternative for bacteremia and SSSI caused by cfr-positive multidrug resistant MRSA.

Keywords: MRSA, cfr, phenotype, biofilm, bacteremia, skin and skin structure infection, combination therapy

# INTRODUCTION

MRSA is particularly challenging due to its inherent pathogenicity and multidrug resistant phenotypes contributing to a variety of infectious diseases, ranging from skin and skin structure infection (SSSI) to bacteremia (Tong et al., 2015; Wang et al., 2019). An increased global incidence of MRSA infections associated with high mortality has been observed over the past decades (Bassetti et al., 2014; Hassoun et al., 2017). For example, in the United States, S. aureus is most often contracted as a nosocomial infection leading to more than 80,000 illnesses and 11,000 deaths yearly

**231**

(Lepak and Andes, 2016). Therefore, new alternative strategies for the treatment of such infections are urgently needed.

Linezolid has become an important drug for treating nosocomial infections due to MRSA, including those with reduced vancomycin susceptibility (e.g., VISA) (Dryden, 2011). However, linezolid resistance due to acquisition of the cfr (chloramphenicol and florfenicol resistance) gene has compromised MRSA treatment options (Long et al., 2006). The cfr gene encodes a 23S rRNA methyltransferase that confers combined resistance to phenicols, lincosamides, oxazolidinones, pleuromutilins, and streptogramin A (PhLOPS<sup>A</sup> phenotype) (Long et al., 2006; Witte and Cuny, 2011). In addition, there is only a low fitness cost to the host for cfr carriage and this facilitates its spread (LaMarre et al., 2011). Infections due to cfr-positive MRSA are increasing and pose a serious threat to the clinical success of oxazolidinone antibiotics (Witte and Cuny, 2011). Although the level of resistance to linezolid conferred by cfr is moderate, the ability of cfr to enhance bacterial survival in the presence of linezolid has been shown in vivo in a murine pneumonia model (Zhou et al., 2018). Linezolidresistant MRSA strains carrying cfr were also associated with prolonged use of linezolid in patients (Endimiani et al., 2011). These data suggest that in addition to the cfr-mediated linezolid resistance, cfr-positive MRSA may possess phenotypes associated with pathogenesis that contribute to poor in vivo treatment outcomes.

In this study, we profiled relevant phenotype differences between cfr-positive and -negative MRSA that contribute to bacteremia and SSSI. We examined whether these MRSA were susceptible to host defense cationic peptides (HDP) and assayed their biofilm forming abilities and binding to fibronectin (FN) and endothelial cells. In addition, we correlated in vitro phenotypes to linezolid resistance in vivo in murine SSSI and bacteremia models to characterize cfr-positive and -negative MRSA.

## MATERIALS AND METHODS

## Bacterial Strains and Background Information

Ten well-characterized MRSA strains were used in this study (Li et al., 2018; Zhou et al., 2018). Human clinical MRSA strains (161402, 161400, 161494, and 161813) were kindly provided by the Third Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) that obtained from hospitalized patients with pulmonary infections. MRSA strains of animal origin (N50, 6Y2C, HYP6, N4-2, HYXC4, and 2B3) were collected from blood and abscess cultures of sick animals at the Animal Diagnostic Laboratory of South China Agricultural University. All strains were identified by MALDI-TOF MS system (Ostergaard et al., 2015). Four MRSA strains of human origin were typed as ST 764 and spa-type t1081, and the remaining six strains were typed as ST 398 and ST 9 (Li et al., 2018; Zhou et al., 2018). The ST 764 MRSA has emerged as a novel hybrid variant of the ST 5 HA-MRSA lineage with the characteristics of CA-MRSA in Asia, causing invasive infections (necrotizing fasciitis and bacteremia) in both hospital and community settings (Takano et al., 2013). The ST 398 MRSA has been reported in China and Europe that was responsible for zoonotic infections in patients with pneumonia and SSSIs (Stegger et al., 2010; van der Mee-Marquet et al., 2011).

## Linezolid-Based Combination Susceptibility Testing and Time-Kill Curves

The MICs of linezolid and other ten antibiotics (oxacillin, cefotaxime, amikacin, azithromycin, tetracycline, vancomycin, clindamycin, retapamulin, ciprofloxacin, and rifampicin) against clinical MRSA isolates were conducted by the broth microdilution method as recommended (CLSI, 2015). S. aureus ATCC 29213 served as the quality control strain. Fold reduction in MIC was determined by dividing the MIC of the antibiotic alone by its MIC in the presence of 0.5 mg/L linezolid. Three biological replicates were done for each combination and the means of fold reduction were used for generating heat maps. In vitro interactions between linezolid and rifampicin were evaluated by the checkerboard method, and a fractional inhibitory concentration index (FICI) of ≤0.5 was deemed synergistic (Zhou et al., 2018).

In vitro time-kill curves were performed to compare the activity of linezolid and rifampicin alone and in combination against two representative cfr-positive and -negative strain sets. In brief, a starting inoculum of ∼10<sup>6</sup> cfu/mL logarithmic phase MRSA cells was used to expose to linezolid (16 mg/L) with or without rifampicin (0.5 mg/L). The drug concentrations were chosen to mimic the free serum steady-state peak concentrations (fCmax) at the usual clinical doses in human (i.e., 600 mg for linezolid, 300 mg for rifampicin) (Andes et al., 2002; Sirgel et al., 2005; Chik et al., 2010; Dryden, 2011). MRSA densities were determined by the serial viable counts collected over 24 h incubation and expressed as log<sup>10</sup> cfu/mL. Synergistic effect was defined as the combination caused ≥2 log<sup>10</sup> cfu/mL reduction vs. the single drug.

## In vitro Concentration-Effect Relationship

Concentration-effect curves were used to evaluate linezolid potency against cfr-positive and -negative MRSA. Briefly, an overnight culture of MRSA cell was washed, adjusted to 0.5 McFarland units and diluted in cation-adjusted Mueller Hinton broth to a final density of 10<sup>6</sup> cfu/mL. The testing procedure consisted of two groups, and each group included tubes with two-fold increasing concentrations of linezolid from 0.015 to 32 mg/L, in the presence and absence of 0.5 × MIC rifampicin. After 16 h of incubation at 37◦C, absorbance of each tube was measured at OD600nm to quantify bacterial growth and normalized with the no drug control. The relationship between linezolid concentrations and antibacterial potency was calculated using the Hill sigmoid Emax equation: E = Emax + (E<sup>0</sup> – Emax)/1 + 10∧[(log EC<sup>50</sup> – C) × Hill slope] using GraphPad Prism 8 software (Zhou et al., 2017).

## In vitro HDP Susceptibility

fmicb-10-03080 December 31, 2019 Time: 13:30 # 3

The tPMPs were prepared from thrombin-stimulated platelets isolated from fresh rabbit blood and their bioactivity was quantified using Bacillus subtilis ATCC 6633 as previously described (Yeaman et al., 1992). Human neutrophil peptide-1 (hNP-1) and cathelicidin LL-37 were purchased from Peptides International (Louisville, KY, United States) and Eurogentec (Fremont, CA, United States), respectively. In vitro HDP susceptibilities were assessed by adding tPMP (2 mg/L equivalent) to 10<sup>3</sup> cfu/mL MRSA cells and hNP-1 (5 mg/L) or LL-37 (20 mg/L) to 10<sup>5</sup> cfu/mL MRSA cells (Xiong et al., 2009; Seidl et al., 2011a). The HDP concentrations were selected to cover the peptide concentrations that did not rapidly kill MRSA cells over 2 h of incubation based on previous studies (Seidl et al., 2011a). Results were expressed as the percentage of the initial inoculum that survived exposure to HDPs.

## Adherence to Fibronectin and Endothelial Cells

Six-well tissue culture plates were coated using 50 mg/L purified human FN (Sigma Chemical, St. Louis, MO, United States) overnight at 4◦C, and then treated with 3% bovine serum albumin for 3 h to prevent non-specific adhesion (Xiong et al., 2009). The human microvascular endothelial cell line (HMEC-1) was cultured as previously described (Seidl et al., 2012). Logarithmic-phase MRSA cells were added to FN-coated plates (5 × 10<sup>3</sup> cfu/mL) and endothelial cell monolayer-coated plates (5 × 10<sup>5</sup> cfu/mL; MOI = 1:1), and then incubated for 1 h at 37◦C under static conditions. For FN binding assay, unbound bacteria were removed by washing the plates with PBS, and melted tryptic soy agar (TSA; 2 mL) was added into each well and allowed to solidify. For endothelial cell binding assay, unbound bacteria were removed by washing the plates with Hanks balanced salt solution (HBSS) and permeabilized using 1.0% Triton X-100 (Seidl et al., 2012), after which bacterial numbers per well were determined by serial dilutions and plating on TSA. Adherence was expressed as the percentage of the initial inoculum bound.

## Biofilm Formation, Extracellular Polysaccharide (EPS) and DNA Determinations

The ability of MRSA to form biofilm was determined as described previously (Seidl et al., 2011a). Briefly, overnight cultured MRSA at 0.5 McFarland units (∼10<sup>8</sup> cfu/mL) was diluted 1:100 into brain heart infusion (BHI) broth supplemented with 0.5% glucose. 200 µL of the suspension was transferred into 96 well plates and incubated for 18 h at 37◦C. After incubation, the plates were washed with PBS, air dried and stained with 0.1% safranin. The adhering dye was dissolved in 30% acetic acid, and absorption was measured at OD490nm to quantify biofilm formation.

The water-soluble and -insoluble EPS synthesized by the biofilms was examined using the anthrone-sulfuric method (Chen et al., 2016). Briefly, 24 h biofilms were rinsed, removed and dispersed by sonication at 20 kHz for 5 s. The suspension was centrifuged at 6000 × g for 10 min at 4◦C, and the supernatant was collected for water-soluble EPS determinations. The pellets were resuspended in PBS, washed and air-dried to ensure all the water-soluble EPS was discarded. The dry weight of each biofilm was measured to adjust biomass differences between cfr-positive and -negative MRSA. The water-insoluble EPS was extracted using 1.0 M NaOH under agitation for 2 h at 37◦C and quantified using an anthrone-sulfuric acid colorimetric assay (Chen et al., 2016).

Release of extracellular DNA (eDNA) was determined from 18 h MRSA biofilm using a microplate fluorescence assay with Hoechst dye 33258 (Leggate et al., 2006). Protocols for extraction and purification of eDNA from MRSA biofilms were described in detail elsewhere (Rice et al., 2007). The eDNA was quantified using an EnSight fluorescence plate reader at Ex350/Em<sup>460</sup> (PerkinElmer, Waltham, MA, United States). Purified salmon sperm DNA was used to generate a standard curve. To account for differences in biomass, the average OD490nm of each unwashed biofilm was determined to calculate the amount of eDNA per relative biomass.

## Hemolytic Activity and Nuclease Production

Hemolytic activity was evaluated by spotting 2 µL of MRSA suspension (∼10<sup>8</sup> cfu/mL) onto 5% sheep blood agar plates and incubated at 37◦C for 24 h (Seidl et al., 2011b). The diameters of the zones of clearance (cm) indicating hemolytic activity were measured. Nuclease production was assessed by spotting 15 µL of filtered culture supernatants of the strains into wells cut into DNase test agar (Beenken et al., 2010). Plates were incubated overnight at 37◦C. Nuclease activity was then assessed by overlaying the agar with 1 M HCl to precipitate undigested DNA and define the zone diameters (cm) of clearance (Beenken et al., 2010).

## In vivo Murine Bacteremia and SSSI Models

Six-week-old, pathogen-free female ICR mice (25–27 g from Guangdong Medical Lab Animal Center, Guangzhou, China) were used in this study. All animal experimental procedures were approved by the South China Agricultural University (SCAU) Institutional Ethics Committee (2017B075 and 2017018) and performed in accordance with the SCAU Institutional Laboratory Animal Care and Use guidelines. For bacteremia model, mice were infected via the tail vein with a 0.5 mL bacterial suspension delivering ∼105.5−6.<sup>0</sup> cfu/mouse (Thakker et al., 1998). For SSSI model, 0.1 mL of bacterial suspension consisting of ∼107.<sup>0</sup> cfu was inoculated subcutaneously into the flanks of mice (Tseng et al., 2011). Four representative MRSA strains were selected for in vivo studies based on their in vitro phenotypes and MLST types that included cfr-positive and -negative MRSA.

To assess the therapeutic efficacy of linezolid and rifampicin alone and in combination, mice were randomized at 24 h (bacteremia model) and 48 h (SSSI model) post-infection to receive: (i) no therapy (control); (ii) linezolid at 100 mg/kg, orally twice daily; (iii) rifampicin at 5 mg/kg, orally twice daily; or (iv) a combination of linezolid and rifampicin. The linezolid and

rifampicin doses were selected to mimic the pharmacokinetic profiles of recommended human clinical doses (i.e., 600 mg and 300 mg, orally twice daily for linezolid and rifampicin, respectively) (Chik et al., 2010; Zhou et al., 2018). Treatments lasted for 3 and 5 days for the bacteremia and SSSI models, respectively. Groups of five or six mice were included at each dose regimen. Control and antibiotic-treated mice were sacrificed either at the beginning of treatment (untreated controls) or 12 h after the last antibiotic dose, respectively. At sacrifice, the target tissues (blood, spleen and kidney for bacteremia model, and skin abscess for SSSI model) were removed and quantitatively cultured. Bacterial densities in infected tissues were calculated as the mean log<sup>10</sup> cfu/g. of tissue and log<sup>10</sup> cfu/mL of blood (± SD). In addition, the mean areas of superficial skin lesions were quantitated for statistical comparisons in the SSSI model.

## Statistical Analysis

In vitro studies were performed with three biological replicates in triplicate. Two-tailed Student's t-test was used to compare relevant phenotype differences between cfr-positive and -negative MRSA groups. Mann-Whitney non-parametric test was used to analyze MRSA densities in target tissue among different groups.

## RESULTS

## Linezolid-Based Combination Potentiated Activity Against cfr-Positive MRSA

As expected, all the study MRSA isolates were resistant to oxacillin with MICs ranging from 8 to 128 mg/L. The MICs of linezolid were markedly higher in MRSA isolates harboring the cfr gene (1–8 mg/L) than in those lacking the cfr (0.5– 2 mg/L; **Table 1**). Fold reductions in MICs of amikacin and vancomycin were observed for part of cfr-positive MRSA isolates when combined with the sub-MIC levels of linezolid at 0.5 mg/L. However, the broad-spectrum antibiotics including cefotaxime and ciprofloxacin displayed a limited MIC reduction. Notably, in the presence of linezolid, rifampicin achieved the highest therapeutic potential as a combinatorial partner with a greater than 8-fold reduction in MIC against 8/10 MRSA isolates, and this was independent of cfr expression (**Figure 1A**). The combination of linezolid and rifampicin resulted in synergistic activity against 5/6 cfr-positive MRSA isolates and 2/4 cfrnegative MRSA isolates, with FICIs ranging from 0.375 to 0.5 (**Table 1**).

Control cultures increased ∼3-log<sup>10</sup> cfu/mL for both cfrpositive and -negative MRSA over a 24 h of incubation. Rifampicin alone had the similar bacterial growths vs. their control groups (**Supplementary Figure S1**). Of note, despite having the same MICs, linezolid alone at 16 mg/L resulted in greater bacterial killing for cfr-negative strain HYXC4 (1.70 log<sup>10</sup> cfu/mL) vs. the cfr-positive strain 6Y2C (0.78-log<sup>10</sup> cfu/mL; **Supplementary Figure S1**). Importantly, the combination of linezolid (16 mg/L) and rifampicin (0.5 mg/L) showed a synergistic bactericidal effect compared to each drug alone regardless of the presence ofcfr gene (**Supplementary Figure S1**).

For cfr-negative MRSA group, the 50% maximal killing effect occurred at an average linezolid concentration of 0.71 mg/L (**Figure 1B**) and this decreased 3.4-fold to 0.21 mg/L in the presence of sub-MIC levels of rifampicin (**Supplementary Table S1**; paired t-test, P < 0.01). Expression of the cfr gene increased linezolid concentrations required to achieve 50% maximal effect to 2.01 mg/L and were significantly higher than the cfr-negative test group (**Table 2**; P < 0.05). However, when combined with rifampicin, the concentration of linezolid required to achieve 50% maximal effect was only 0.34 mg/L for cfr-positive MRSA group (**Figure 1C**). In fact, this level was comparable to the concentration required to potentiate rifampicin for cfr-negative


<sup>a</sup>OXA, oxacillin; LZD, linezolid; RIF, rifampicin. <sup>b</sup> Interpreted as synergy (FICI ≤ 0.5), no interaction (0.5 < FICI ≤ 4) or antagonism (FICI > 4).

group (**Table 2**). Although the concentration of 0.34 mg/L linezolid was insufficient to inhibit growth of MRSA carrying the cfr gene, its combination with rifampicin provided a promising alternative to overcome MRSA infections irrespective of cfr expression.

## In vitro HDPs Susceptibility and Adherence to FN and Endothelial Cells

As a group, the cfr-positive MRSA exhibited significantly higher survival rates after exposure to 2 mg/L tPMP (73.8%) or 5 mg/L hNP-1 (78.5%) compared with the cfr-negative strain group (43.1 and 57.7%, respectively; P < 0.01). Similarly, a markedly reduced LL-37 killing was observed in cfr-positive vs. cfr-negative MRSA group (P < 0.005; **Figure 2A**). The cfr-positive MRSA demonstrated significantly higher binding rates to FN (14.8%) compared with the cfr-negative MRSA (4.02%; P < 0.005), despite a relatively low adherence to FN observed with the cfr-positive strain N4-2. Consistent with FN binding profiles, cfr-positive MRSA strain group bound substantially better to human endothelial cells than cfr-negative group (11.30 vs. 3.87%, P < 0.005; **Figure 2B**).

## Biofilm Formation, EPS and eDNA Determinations and Nuclease Productions

We compared cfr-positive and -negative MRSA strain groups with respect to in vitro biofilm capacity and composition. Overall, the cfr-positive MRSA group had a greater ability to form biofilms

TABLE 2 | Calculated EC<sup>50</sup> and Hill slope (N) values representing the antimicrobial potency of linezolid alone or with 0.5 × MIC rifampicin against cfr-positive and -negative MRSA strains.<sup>a</sup>


<sup>a</sup>EC50, the linezolid concentration required to achieve 50% of maximal effect (Emax); N, the Hill coefficient that described the slope of the dose-response curve. <sup>b</sup>P < 0.05 for EC<sup>50</sup> of linezolid alone in cfr-positive vs. -negative strains. <sup>c</sup>P < 0.01 for EC<sup>50</sup> of linezolid and rifampicin in combination vs. linezolid alone.

compared with the cfr-negative group (OD490nm 1.51 vs. 0.33, P < 0.01; **Figure 2C**). The production of the water-soluble EPS ranged from 161 to 593 µg/g, with the average production in cfrpositive group being considerably higher than that in cfr-negative group (487 µg/g vs. 202 µg/g, P < 0.01). A similar pattern was observed for the water-insoluble EPS between cfr-positive and -negative strain groups (P < 0.01; **Figure 2D**). Of note, the 24 h old MRSA biofilms showed increased production of the watersoluble EPS compared to water-insoluble EPS in both groups. In addition, the average amount of eDNA present in the cfrpositive MRSA biofilms (10.6 ± 1.37 ng) was 2.1-fold greater than that present in the cfr-negative MRSA biofilms (4.96 ± 0.88 ng), a statistically significant difference (P < 0.005; **Figure 2E**). These results were further corroborated by a significantly lower level of nuclease production in the cfr-positive MRSA group (P < 0.005; **Figure 2F**). In addition, except for strain 161494, the cfr-positive MRSA strain group possessed more α-hemolysin activity, compared with weak or non-detectable α-hemolysin production in cfr-negative MRSA group (**Figure 2F**).

## In vivo Responsiveness to Linezolid and Rifampicin Alone and in Combination

In the murine SSSI model, MRSA densities in skin abscesses and the areas of skin lesions in mice infected with cfr-negative MRSA isolates 161400 and HYXC4 were significantly reduced after 5 days of linezolid monotherapy as compared to their untreated controls (P < 0.005, **Figure 3**). In contrast, for cfr-positive MRSA isolates 161402 and 6Y2C, the mice did not respond to linezolid monotherapy and residual abscess-tissue MRSA densities and the areas of skin lesions similar to those in their respective controls (**Figure 3**). Of note, combination therapy with linezolid and rifampicin resulted in 1.5–3.6 log<sup>10</sup> cfu/abscess reductions in MRSA densities vs. linezolid monotherapy against both cfr-positive and -negative MRSA infections (**Figures 3A,C**). This result occurred despite the lower MRSA densities in skin abscesses observed in cfr-negative isolates (P < 0.05; **Figure 3**).

In the bacteremia model, we found the similar results. Linezolid monotherapy resulted in uniform and highly significant reductions of MRSA densities in all the target tissues of mice infected with cfr-negative MRSA isolates 161400 and HYXC4. For instance, a ≥2.0 log<sup>10</sup> cfu/mL reductions in blood density and ≥1.0 log<sup>10</sup> cfu/g reductions in spleen and kidney densities were observed with linezolid monotherapy as compared to their respective control mice infected with cfr-negative MRSA (P < 0.01). However, bacteremia caused by cfr-positive MRSA isolates showed the opposite results: (i) a ≥1.0 log<sup>10</sup> cfu/g increases in both spleen and kidney (161402, **Figures 4A,B**) or (ii) no response to linezolid monotherapy, with similar residual target-tissue MRSA densities compared with untreated controls (6Y2C; **Figures 4D–F**). Importantly, the combination of linezolid and rifampicin showed efficacies higher

than each monotherapy in mice infected with both cfr-positive and -negative MRSA (P < 0.001; **Figure 4**).

## DISCUSSION

S. aureus is a common opportunistic pathogen that causes a wide range of infections affecting skin and soft tissues as well as invasive infections that include bacteremia, endocarditis and pneumonia (Bassetti et al., 2014; Hassoun et al., 2017). In particular, bacteremia caused by MRSA is often associated with a high mortality rate, even with appropriate antibiotic treatments (Seidl et al., 2011a). Linezolid has been an important drug for therapy of MRSA infection. However, the emergence and rapid spread of horizontally transferable cfr determinant in MRSA has become a substantial concern (Witte and Cuny, 2011). The effective therapeutic alternatives were limited, especially for the vancomycin-intermediate and linezolidresistant MRSA carrying cfr gene (Barber et al., 2016). Fortunately, unlike the chromosomally encoded resistance mechanisms,cfr has been shown to confer the low-level resistance to linezolid (Locke et al., 2014). Therefore, the combination therapy is an appealing option for retaining the clinical utility of linezolid.

Linezolid is a valuable alternative to glycopeptide antibiotics (e.g., vancomycin) and an oral formulation allows a rapid intravenous to oral switch (Dryden, 2011; Bassetti et al., 2014). As described in previous clinical studies, linezolid achieved significantly greater efficacy (e.g., higher cure rates) and earlier discharges from hospital than vancomycin treatment in patients with MRSA-complicated SSSI (Sharpe et al., 2005). This clinical outcome may be ascribed to a good penetration into skin and soft tissues with almost 100% oral bioavailability (Bassetti et al., 2014). Furthermore, linezolid treatments for MRSA bacteremia had roughly equivalent clinical and microbiological outcomes compared with vancomycin (Shorr et al., 2005). In our previous pneumonia model, addition of rifampicin to linezolid significantly decreased fAUC/MIC targets in both plasma and lung epithelial lining fluid (Zhou et al., 2018). Here, we extended the effectiveness of linezolid and rifampicin combination to the clinically relevant murine models of SSSI and bacteremia due to cfr-positive and -negative MRSA. More importantly, rifampicin decreases S. aureus FN binding that is a further advantage of the combination therapy (Rasigade et al., 2011).

This is supported by other studies showing successful clinical outcomes with oral linezolid and rifampicin combination therapy in the management of recurrent and persistent MRSA bacteremia (Schwalm et al., 2004).

Indeed, we observed significant phenotype differences between cfr-positive and -negative MRSA that contribute to bacteremia and SSSI. The capability of S. aureus to circumvent clearance mediated by locally secreted HDPs is an important factor for its pathogenicity (Seidl et al., 2011a). In the current study, we demonstrated that cfr-positive MRSA isolates tended to be more resistant to key innate HDPs from neutrophils (hNP-1), platelets (tPMPs), and epithelial cells (LL-37), compared with the cfr-negative MRSA isolates. These findings suggested that cfr-positive MRSA might be more capable to survive in the bloodstream early in the course of skin infections. In particular, survival rates of >40% after 2 h of exposure to HDPs were positively correlated with the severity of endovascular infections and reduced responsiveness to antimicrobial therapy (Yeaman et al., 1992; Seidl et al., 2011a). Similarly, we observed a dramatic relationship between the reduced HDP killing in vitro and decreased efficacy of linezolid-based therapy in the murine bacteremia model.

Interestingly, cfr-positive MRSA group exhibited greater biofilm formation and higher EPS and eDNA productions compared with cfr-negative strain group. Biofilms enhance bacterial resistance to HDPs and antibiotics due to poor penetration past this barrier and this was the case for cfr-positive MRSA (Donlan and Costerton, 2002). Cathelicidin LL-37 can protect against MRSA-induced skin infections but cfr-positive MRSA were also able to resist the adverse effects of LL-37 exposure (Haisma et al., 2014). The alterations may contribute to the poor outcomes of linezolid therapy in mice infected with cfr-positive MRSA isolates in the SSSI model.

Invasive S. aureus must attach to extracellular matrix ligands or surface proteins on host cells to enable adhesion and internalization (Xiong et al., 2009). Therefore, the ability to bind FN is necessary for inducing S. aureus infections (e.g., bacteremia) (Seidl et al., 2011a). Consistent with this, increased FN binding ofcfr-positive MRSA correlated with worse outcomes of linezolid-based mono- and combination therapies in the murine bacteremia model. This observation is also in line with previous reports that the development of a hyper-adhesive FN binding phenotype contributed to persistent MRSA bacteremia and infective endocarditis (Xiong et al., 2009, 2015). In addition, we found that cfr-positive MRSA isolates adhered better than cfr-negative isolates to human endothelial cells, and this may facilitate MRSA colonization and provide an advantage in the pathogenesis of invasive MRSA infections.

The exact mechanisms how the presence of cfr correlates with the phenotypic characteristics remain to be fully elucidated. In the recent past, it was assumed that the development of antibiotic resistance was linked to virulence and fitness costs (Beceiro et al., 2013). However, the acquisition of cfr in S. aureus has been associated with low fitness cost that potentially facilitates the

growth rates, invasiveness and transmission capacity (LaMarre et al., 2011; Beceiro et al., 2013). In addition, in most of our study MRSA strains, cfr co-expressed with the erm gene (erythromycin resistance) that has a significant correlation with biofilm formations (Beceiro et al., 2013; Li et al., 2018). Previous studies in Enterococcus species exhibited that strong biofilm formation was more prevalent among linezolidresistant compared with -sensitive isolates (Osman et al., 2020). S. aureus is a highly adaptable bacterium capable of dynamic changes in its virulence and resistance phenotypes when exposure to the host defenses or antibiotics (Abdelhady et al., 2015). Thus, additional unknown mechanisms likely contribute to the adaptive response phenotypes in vitro and linezolid-associated outcomes in vivo. Studies including comparative genomic and transcriptomic analysis are in progress in our laboratories to further determine other possible mechanisms.

Our investigation has several limitations. For example, we assessed a relatively small number of cfr-positive and -negative MRSA strains despite the different clonal types that raise the possibility of strain-dependent bias. In addition, some strains were isolated from animals and we only examined four representative MRSA strain sets in our animal models. Future studies should examine the phenotypic characteristics and usefulness of this combination in a larger population of strains from patients and in the clinical settings. Moreover, based on current findings, we do not know whether the relationship between the phenotypic profiles and the outcome of linezolid-based treatment is "cfr specific" or other unknown mechanisms. Although this is beyond the scope of the present study, future mechanism-based studies is warranted to better understand the precise factors responsible for this potential correlation.

Of note, the previous study reported that bacteriostaticbactericidal antibiotic combinations could result in attenuation of bactericidal activity (Lobritz et al., 2015). However, our results showed the synergistic bactericidal effect for linezolid and rifampicin combination. This is supported by the previous observation that the combination of linezolid and rifampicin resulted in 3.1-log<sup>10</sup> cfu/mL killing in vitro against staphylococcal biofilm (El Haj et al., 2018). Similarly, linezolid used in combination with rifampicin was more effective than their monotherapies, reducing the planktonic MRSA cells by >3.0 log<sup>10</sup> cfu/mL in the cage fluids of foreign-body infections (Baldoni et al., 2009). In light of the divergent effects that observed between our results and previous study, future investigation is warranted to better understand the precise mechanism of this combination.

In summary, our results indicated that increased FN and endothelial cell adhesion, reduced susceptibility to HDPs and robust biofilm formation all contributed to linezolid treatment outcomes we found in vivo. Combination therapy with linezolid and rifampicin significantly enhanced therapeutic efficacy against experimental bacteremia and SSSI due to cfr-positive and -negative MRSA isolates.

## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## ETHICS STATEMENT

This study was carried out in accordance with the recommendations of ethical guidelines of South China Agricultural University (SCAU). The protocol of in vivo studies and isolation procedures for animal-origin strains were approved by the SCAU Institutional Animal Ethics Committee (approval no. 2017B075). Individual written informed consent for the use of samples was obtained from all animal owners. Humanorigin strains were kindly provided by the Third Affiliated Hospital of Sun Yat-sen University (Guangzhou, China), and isolation procedures were carried out in accordance with relevant guidelines with written informed consent from all subjects. The isolation and use protocols of human-origin strains were reviewed and approved by the Human Research Protection Office of SCAU Institutional Ethics Committee (approval no. 2017018).

## AUTHOR CONTRIBUTIONS

Y-HL, YX, and Y-FZ designed the study. Y-FZ, LL, and M-TT carried out the experiments. Y-FZ, JS, and X-PL analyzed the data. Y-FZ wrote the manuscript. LL and YX revised the manuscript. All authors read and approved the final manuscript.

## FUNDING

This study was supported by the National Key Research and Development Program of China (2016YFD0501300), the Program of Changjiang Scholars and Innovative Research Team in University of Ministry of Education of China (IRT\_17R39), and the Foundation for Innovation and Strengthening School Project of Guangdong, China (2016KCXTD010).

## ACKNOWLEDGMENTS

The authors thank the Third Affiliated Hospital of Sun Yat-sen University for providing the clinical S. aureus isolates.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.03080/full#supplementary-material

## REFERENCES

fmicb-10-03080 December 31, 2019 Time: 13:30 # 10



persistent methicillin-resistant Staphylococcus aureus bacteremia in vitro and in an experimental endocarditis model. J. Infect. Dis. 199, 201–208. doi: 10.1086/ 595738


**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 Zhou, Li, Tao, Sun, Liao, Liu and Xiong. 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.

# Topical Therapeutic Efficacy of Ebselen Against Multidrug-Resistant Staphylococcus aureus LT-1 Targeting Thioredoxin Reductase

Chuanjiang Dong<sup>1</sup>† , Jingxuan Zhou<sup>2</sup>† , Peng Wang1,2† , Tao Li<sup>1</sup> , Ying Zhao<sup>3</sup> , Xiaoyuan Ren<sup>4</sup> , Jun Lu<sup>3</sup> , Jun Wang2,5 \*, Arne Holmgren<sup>4</sup> and Lili Zou<sup>1</sup> \*

#### Edited by:

Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil

#### Reviewed by:

Jianhua Wang, Chinese Academy of Agricultural Sciences, China David Leitsch, Medical University of Vienna, Austria Luigi Messori, University of Florence, Italy

#### \*Correspondence:

Jun Wang wangjfox@gmail.com Lili Zou zoulili@ctgu.edu.cn; zoulili@mail3.sysu.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 19 July 2019 Accepted: 16 December 2019 Published: 15 January 2020

#### Citation:

Dong C, Zhou J, Wang P, Li T, Zhao Y, Ren X, Lu J, Wang J, Holmgren A and Zou L (2020) Topical Therapeutic Efficacy of Ebselen Against Multidrug-Resistant Staphylococcus aureus LT-1 Targeting Thioredoxin Reductase. Front. Microbiol. 10:3016. doi: 10.3389/fmicb.2019.03016 <sup>1</sup> The First College of Clinical Medical Science, China Three Gorges University, Yichang, China, <sup>2</sup> The Institute of Infection and Inflammation, Medical College, China Three Gorges University, Yichang, China, <sup>3</sup> Key Laboratory of Luminescent and Real-Time Analytical Chemistry (Southwest University), Ministry of Education, College of Pharmaceutical Sciences, Southwest University, Chongqing, China, <sup>4</sup> Division of Biochemistry, Department of Medical Biochemistry and Biophysics, Karolinska Institutet, Stockholm, Sweden, <sup>5</sup> Translational Neuroscience & Neural Regeneration and Repair Institute/Institute of Cell Therapy, The People's Hospital of China Three Gorges University, Yichang, China

As a thiol-dependent enzyme, thioredoxin reductase (TrxR) is a promising antibacterial drug target. Ebselen, an organo-selenium with well-characterized toxicology and pharmacology, was recently reported to have potent antibacterial activity against Staphylococcus aureus. In this paper, we demonstrated that ebselen has strong bactericidal activity against multidrug-resistant (MDR) S. aureus based on taking TrxR as a major target and disruption of the redox microenvironment. Further, the topical therapeutic efficacy of ebselen for staphylococcal skin infections was assessed in a rat model. Treatment with ebselen significantly reduced the bacterial load and the expression of pro-inflammatory cytokines tumor necrosis factor-α (TNF-α), interleukin-6 (IL-6) and interleukin-1 beta (IL-1β) in S. aureus skin lesions; further, wound healing and pathological changes were obvious improved in ebselen-treated rats compare to controls. Finally, ebselen was found to sensitize S. aureus to curcumin, which may be due to their synergistic effects in inhibiting bacterial TrxR. Altogether, ebselen is an effective topical antibacterial agent in animal model of MDR S. aureus LT-1 skin infection. This may lay the foundation for further analysis and development of ebselen as an antibacterial agent for topical treatment of MDR staphylococcal infections.

Keywords: ebselen, thioredoxin reductase, topical treatment, Staphylococcus aureus, curcumin

## INTRODUCTION

Novel antimicrobials and new bacterial cell targets are urgently needed to overcome the everincreasing worldwide antimicrobial resistance (AMR), which is now widely accepted as a global health threat with leading causes of morbidity and mortality (Brown and Wright, 2016; Baptista et al., 2018). In May 2018, the World Health Organization (WHO), the Food and Agriculture Organization of the United Nations (FAO), and the World Organization for Animal Health (OIE) signed a Memorandum of Understanding (MoU) to support governments, health care workers, and stakeholders to engage and collaborate in the fight against AMR (Balkhy et al., 2018). Limiting the

emergence of AMR is critical to preserve our ability to treat bacterial and microbial infections not only in humans, but also in animal, plant, food, and environment circumstances (Bhatia, 2018).

Staphylococcus aureus is a highly adaptable, "Janus-faced" Gram-positive pathogen, for which humans are the only known reservoir (Kobayashi et al., 2015; Balasubramanian et al., 2017; Dweba et al., 2018). S. aureus is present in approximately 30% of the human population, and its presence has been linked to skin rashes, wound infections, pleuropulmonary, bacteremia, infective endocarditis, and device-related infections (Coates et al., 2014; Tong et al., 2015). In the pre-antibiotic era, the case fatality rate (CFR) for S. aureus was ∼80%; it has since decreased and plateaued at 15∼50% over the past several decades since the introduction of penicillin (van Hal et al., 2012). However, the adaptive evolution of S. aureus during the modern antibiotic era has enabled its acquisition of antibiotic resistance, thus increasing disease burden worldwide (Pantosti et al., 2007; McGuinness et al., 2017).

Ebselen or 2-phenyl-1,2 benzisoselenazol-3(2H)-one, an organo-selenium compound, is a clinical trial drug with well-characterized toxicology and pharmacology (Zou et al., 2017; **Figure 1**). Recent studies have shown that ebselen possesses bactericidal activity against Gram-positive, including multidrug-resistant (MDR) clinical isolates of S. aureus. Bacterial thioredoxin reductase (TrxR) was assessed to be a critical target, and the ebselen inhibition is through a reaction with the active site dithiol of TrxR to act as a competitive inhibitor (Lu et al., 2013; Thangamani et al., 2015; Zou et al., 2018). Bacterial TrxR transfers an electron from NADPH to Trx allowing it to regulate functions of various critical cellular proteins, including ribonucleotide reductase (RNR), methionine-S-sulfoxide reductase (Msr), thiol peroxidase (Tpx), bacterioferritin comigratory protein (Bcp) (Holmgren, 1985; Lu and Holmgren, 2012, 2014a; Leitsch, 2017). Bcp is critical in antioxidant defense and DNA synthesis and repair (Lu and Holmgren, 2012, 2014b). However, animal experiments testing antibacterial activity of ebselen are rare. Thus, the aim of this study was to evaluate the bactericidal activity of ebselen against MDR S. aureus in vitro, and to assess its topical therapeutic efficacy, anti-inflammatory properties, and potential clinical applications in staphylococcal skin infection rat model.

## RESULTS

## Antibacterial Activity of Ebselen Targeting S. aureus TrxR

Staphylococcus aureus LT-1 was isolated from patients with cutaneous infections in The first clinical hospital of Yichang (China), and identified as an MDR strain (**Tables 1**, **2** and **Supplementary Figure S1**). LT-1 cells with logarithmic growth were treated with different concentrations of ebselen for 16 h. The antibacterial effect of ebselen on the growth of S. aureus was investigated in microplates by a spectrophotometer, which estimated cell number. As shown in **Figure 2A**, ebselen inhibited S. aureus growth with a minimal inhibition concentration (MIC) of 2.2 µg/ml (8 µM). Meanwhile, the positive control gentamycin inhibited S. aureus growth with a MIC of 0.85 µg/ml (1.08 µM). Further, the propidium iodide (PI) nuclear staining which represents the bacterial membrane permeability was performed after treatment with 22 µg/ml ebselen. PI stains the nucleic acids inside dead cells, or those with damaged membranes. In agreement with the inhibitory effect on bacterial growth curve, when LT-1 cells were treated with ebselen, there was a significant increase in PI positive cells (P < 0.001, **Figure 2B**).

The effect of ebselen on the morphology of S. aureus was detected by transmission electron microscopy (**Figures 2C–H**). The morphology of S. aureus changed significantly when treated with ebselen compared to the control. Untreated S. aureus cells have a smooth surface and a complete cell membrane and cell wall (**Figures 2C,D**). After 20 min treatment with 22 µg/ml ebselen, the S. aureus cell membrane and cell wall were ruptured, cytoplasmic material flowed out, and the cells eventually died (**Figures 2E,F**). In contrast, gentamycin-treated S. aureus cells showed no obvious morphological changes compared to untreated cells (**Figures 2G,H**).

The inhibitory effect of ebselen on S. aureus TrxR activity was detected by dithiobis nitrobenzoic acid (DTNB) assay. Interestingly, ebselen treatment significantly inhibited bacterial TrxR activity when compared with untreated cells (P < 0.001, **Figure 2I**). In addition, the mean fluorescent intensity (MFI) of reactive oxygen species (ROS) was detected by Flow cytometry, and showed that the ROS production level in ebselen-treated S. aureus cells was significantly upregulated compared to the control (P < 0.01, **Figure 2J**).

Altogether, these results shown that ebselen is an effective antibacterial compound that inhibits S. aureus TrxR in vitro, and that ROS production is one of the key virulent factors for its bactericidal activity.

## The Therapeutic Efficacy of Ebselen in Rat Model of Staphylococcal Skin Infections

Thirty rats were randomly divided into three groups: control group, ebselen-treated group, and gentamycin-treated group. Thick scalpel cuts were made into the dermis of the skin on the posterior upper back and neck of rats and inoculated with S. aureus LT-1 cells. Two days post-infection, the bacteria from the three groups showed no obvious difference in colony forming units (average of 3.8 ± 1.0 (Log<sup>10</sup> 3 ) CFU/ml) (P < 0.05). Rats were then treated topically with 25 mg/kg ebselen, 5 mg/kg gentamycin or PBS once per day for 5 days, respectively. The results are highlighted in **Figure 3A**. After 5 days posttreatment, the rats receiving ebselen had a significantly reduced mean bacterial count compared with the group receiving PBS (P < 0.05). The ebselen-treated group had the highest reduction (43 ± 26.4 CFU/ml), followed by gentamycin (57 ± 17.3 CFU/ml) and control (403 ± 130.1 CFU/ml). There was no obvious difference between ebselen and gentamycin in their antibacterial activities (P > 0.05).



AMY, amygdalin; PIPLC, phosphatidyl phosphatase C; dXYL, D-xylose; ADH1, arginine double hydrolase 1; BGAL, β-D-galactosidase; AGLU, α-glucosidase; APPA, alanine- phenylalanine-proline aromaminase; CDEX, cyclodextrin; AspA, L-aspartate arylamine; BGAR, β-galactopyranosidase; AMAN, α-Mannosidase; PHOS, phosphatase; LeuA, leucine aromaminase; ProA, proline aromaminase; BGURr, β-glucuronidase; AGAL, α-galactosidase; PyrA, pyroglutaminase; BGUR, β-Dglucuronidase; AlaA, alanine aromatase; TyrA, tyrosine aromaminase; dSOR, D-sorbitol; URE, urease; POLYB, polycolistin B tolerance; dGAL, D-galactose; dRIB, D-ribose; lLATk, lactate produces alkali; LAC, lactose; NAG, N-acetyl-D-glucosamine dMAL D-Maltose; BACI, bacillus peptide tolerance; NOVO, novomycin tolerance; NC6.5, 6.5% NaCl growth; dMAN, D-mannitol; dMNE, D-mannose; MBdG, methyl-B-D- glucopyranoside; PUL, pullulan; dRAF, D-raffinose; O129R, O/129 tolerant; SAL, salicin; SAC, saccharose; dTRE, D-trehalose; ADH2s, arginine double hydrolase 2; OPTO, optoxin tolerance.



Wound healing was also evaluated according to the formula as follows: F = (A−B)/A<sup>∗</sup> 100%, where F: wound healing rate, A: original wound area, and B: wound area post-treatment. As the results showed in **Figure 3B**, ebselen-treated rats had a wound healing rate that was significantly higher than the control (P < 0.01). The group treated with ebselen had the highest healing rate in wound (44.43 ± 5.41), followed by gentamycin (42.43 ± 5.54), and PBS (33.56 ± 6.12). There was no difference in wound healing in rats treated with ebselen and gentamycin (P > 0.05).

## Topical Effect of Ebselen on Inflammatory Cytokines of Staphylococcal Skin Infection

To study the immune-modulatory activity of ebselen in a topical application against S. aureus LT-1 skin infection, forty rats were randomly divided into 4 groups: A to D. Scalpel cuts were made in the dermis of the skin on the posterior upper back and neck of the rats. Rats from group A served as a control group and were not infected or treated. Rats from groups B to D were inoculated with the S. aureus strain LT-1 and further topically treated with 25 mg/kg ebselen, 5 mg/kg gentamycin, or PBS once per day for 5 days, respectively. An ELISA assay was used to measure the presence of pro-inflammatory cytokines, including tumor necrosis factor-α (TNF-α), interleukin-6 (IL-6), and interleukin-1 beta (IL-1β). As shown in **Figure 4**, ebselen significantly reduced the expression of all three tested proinflammatory cytokines compared to rats treated with PBS (P < 0.05). For TNF-α, the group treated with ebselen had the highest reduction in expression (44.83 ± 3.59 pg/ml), followed by gentamycin (47.01 ± 4.19 pg/ml), and PBS (51.90 ± 6.98 pg/ml) (**Figure 4A**), while there was no difference between ebselen and gentamycin (P > 0.05). For IL-1β, the group treated with ebselen had the highest reduction in expression (64.83 ± 8.91 pg/ml), followed by gentamycin (75.61 ± 7.37 pg/ml), and PBS (102.02 ± 11.22 pg/ml) (**Figure 4B**), while there was difference between ebselen and gentamycin (P < 0.05). Finally, for IL-6, the group treated with ebselen had the highest reduction in expression (38.03 ± 5.85 pg/ml), followed by gentamycin (45.21 ± 8.7 pg/ml), and PBS (71.92 ± 10.0 pg/ml) (**Figure 4C**), while there was difference between ebselen and gentamycin (P < 0.05). Overall, ebselen treatment had the greatest effect in reducing the expression of pro-inflammatory cytokines, and its anti-inflammatory activity was considerably higher than gentamycin.

FIGURE 3 | Therapeutic efficacy of ebselen in treating staphylococcal skin lesions. Rats were treated with 25 mg/Kg ebselen, 5 mg/Kg gentamycin (positive control) and PBS once daily for 5 days, respectively. (A) The bacterial load was calculated by counting the colonies; (B) The wound healing was evaluated according to the formula as follows: F = (A-B)/A∗100%, where F: wound healing rate, A: original wound area, B: wound area post-treatment. (∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; student's t-test. Data are presented as means ±SD of three independent experiments).

Further, hematoxylin and eosin (H&E) and Masson staining were performed using the new formative tissues. Staining revealed that PBS-treated rats had reduced number of fibroblasts and microvascular, and increased number of inflammatory cells and erythrocytes. Meanwhile, the ebselen or gentamycintreated rats had dense granulation tissues and microvascular, and less scattered lymphocytes (**Figures 5A–H**). Anti-CD64 antibody was further used to detect activated mature neutrophil by Immunohistochemistry (IHC). Rats treated with ebselen had less infiltrative neutrophils than those treated with PBS (**Figures 5I–L**). Moreover, fibroblasts were detected by S100a4 antibody, also termed FSP-1 (fibroblast-specific protein-1). The results showed an increased number of fibroblasts in ebselen or gentamycin-treated rats compared to the PBS-treated group (**Figures 5M–P**).

Altogether, these results demonstrated that ebselen may influence the repair of damaged skin, and its efficacy to do so is considerably higher than that of gentamycin.

## Synergistic Activity of Ebselen With Curcumin Against S. aureus in vitro

As an important constituent of turmeric, curcumin, has various biological activities due to its antioxidant mechanism. Previous studies have shown that curcumin has antibacterial activity against various bacteria, including S. aureus (Song et al., 2012), and that it operates through blocking the assembly dynamics of filamentous temperature-sensitive protein Z (FtsZ) in the Z ring at the site of division in bacterial cells (Rai et al., 2008; Kaur et al., 2010), affecting biofilm initiation (Rudrappa and Bais, 2008). It may also have bacterial membrane lysing

properties (Tyagi et al., 2015), yet whether it targets bacterial TrxR remains unknown.

The synergistic inhibition effect of bacterial growth by ebselen and curcumin was tested on S. aureus ATCC25923. The growth of S. aureus was substantially inhibited by 5 µM ebselen and 10 µM curcumin in 96 well-plate 4 h post-incubation (P < 0.01, **Figure 6A**). Furthermore, as showed in **Figure 5B**, adding 5 µM ebselen to 10 µM curcumin-treated S. aureus could also significantly inhibit cells growth (P < 0.05, **Figure 6B**). In addition, the Bliss model of synergism against S. aureus was used, and the degree of synergy of 5 µM ebselen and 10 µM curcumin is 0.89 2 h post-incubation (**Supplementary Figure S2**).

To detect whether treatment with 5 µM ebselen and 80 µM curcumin affected bacterial membrane permeability, PI nuclear staining was performed. Consistent with the inhibitory effect on bacterial growth, treatment with ebselen and curcumin increased the number of PI positive S. aureus cells compared to that of ebselen or curcumin alone (P < 0.01, **Figure 6C**).

## The Inhibition Activity of Bacterial TrxR by Ebselen and Curcumin

Thioredoxin reductase activity in S. aureus was assessed following treatment with 5 µM ebselen and 80 µM curcumin. Although curcumin alone could inhibit TrxR activity, the combination with ebselen exhibited a significantly stronger inhibitory effect (P < 0.01) (**Figures 7A,B**).

Since the Trx system is the major mediator in redox balance, ROS production was assessed following treatment with 5 µM ebselen and 80 µM curcumin as detected with H2DCF-DA by flow cytometry. When S. aureus cells were treated with curcumin alone, ROS production was increased (P < 0.05); however, the effect of ebselen and curcumin in combination on ROS production was significantly higher (P < 0.001) (**Figure 7C**).

Altogether, the above results demonstrated that targeting S. aureus TrxR is one of the antibacterial mechanisms of curcumin, and ebselen enhances its efficacy.

## DISCUSSION

The WHO stated that 1 in 10 patients suffer an infection while receiving medical care and that more than half of surgical site infections may be antibiotic-resistant (Odell, 2010). Meanwhile, skin infections caused by MDR S. aureus have been defined as a major public health threat worldwide (McCaig et al., 2006). Therefore, new and effective treatment strategies are warranted (Antonelou et al., 2011).

As a major thiol-dependent enzyme system exists in both mammalian and most Gram-positive bacteria, Trx system transfers electrons from NADPH to corresponding substrates via TrxR, which is critical for DNA synthesis, defense against oxidative stress, repair of oxidized proteins, and post-translational modifications (Lillig and Holmgren, 2007; Tejman-Yarden et al., 2013; Leitsch, 2017; Zou et al., 2017, 2018). Bacterial TrxR has notable differences in components, enzyme structure, and reaction mechanisms compared with corresponding mammalian host (Holmgren, 1989; Sandalova et al., 2001; Lu et al., 2013), which guaranteed it act as an appropriate target by specific antibiotics. We and other groups previously found that the selenazol drug ebselen was a lead compound for this antibiotic principle (Winterbourn and Hampton, 2008; Lu and Holmgren, 2014a; Thangamani et al., 2015). Briefly, mammalian TrxR is a large (55 kDa/subunit) selenium-containing enzyme, and ebselen react with it to form Se-Se bond to be a substrate for better redox homeostasis. Meanwhile, bacterial TrxR is a small enzyme (35 KDa/subunit) free from selenium, and ebselen react with it to form Se-Se

bond to act as a competitive inhibitor that could upregulate ROS production (Lu et al., 2013; Lu and Holmgren, 2014a; Zou et al., 2017).

In recent studies, we and other groups have shown that ebselen has potent antimicrobial activity against most Gram-positive bacteria including MDR S. aureus (Lu et al., 2013; Thangamani et al., 2015). As a well-known organoselenium compound, ebselen is under phase II clinical trials for stroke, hearing loss, atherosclerosis, and other conditions, which guaranteed its safety and tolerance in human being (Pratta et al., 1998; Ishii et al., 2000; Zhang et al., 2002; Johnsen-Soriano et al., 2007). In this study, the bactericidal activity of ebselen was confirmed by visible spectrophotometer, transmission electron microscopy, DTNB assay, and Flow cytometry in vitro. These results verified the potent bactericidal activity of ebselen against MDR S. aureus, and demonstrated one of its ability to inhibit bacterial TrxR and disrupt the intracellular redox microenvironment.

The bacterial Trx system is important for reducing various cellular antioxidants, such as RNR, Msr, Tpx, Bcp. Thus, inhibition of TrxR is highly related to the excessive production

of ROS, which may disrupt many physiological processes and cause cellular damages or even cell death (Zou et al., 2017). In order to detect the intracellular ROS level, ebselen-treated cells were stained with H2DCFH-DA. FACS results confirmed that ebselen inhibited the electrons transfer from TrxR to its corresponding oxidized substrates, which greatly influences ROS removal. This result demonstrated that ROS elevation is a major player in determining the bacterial fates, which is consistent with previous reports showing that ROS produced by some clinically used antibiotics contribute to their bactericidal efficacy (Kohanski et al., 2007; Morones-Ramirez et al., 2013; Zou et al., 2018).

Further, the rat model with staphylococcal skin wound infection of S. aureus LT-1 strain was used to determine the therapeutic efficacy of ebselen on bacterial burden and infection-induced inflammation, in vitro. As this model represented a staphylococcus infection of open skin wounds, it provided the opportunity to evaluate the efficacy of topical ebselen treatment. Ebselen showed high therapeutic efficacy against MDR S. aureus LT-1 caused skin infection, including substantially reducing the bacterial burden by day 5 (10-fold), increasing fibroblasts and microvascular, and reducing scattered lymphocytes. Moreover, since increased host inflammatory cytokines are mediators for clinical severity of staphylococcal skin infections, the expression of infection-induced inflammation cytokines was also evaluated (Montgomery et al., 2013; Sharma-Kuinkel et al., 2013). The results showed that TNF-α, IL-1β, and IL-6 all contributed to host defense during a S. aureus skin wound infection, whereas IL-1β was more critical during a deeper intradermal S. aureus infection. A similar study demonstrated that during the intradermal infection, the inducible IL-1β response of the bone-marrow-derived recruited cells of the abscess was a more critical determinant for host defense (Miller et al., 2007). Our results showed that ebselen dramatically decreased the infection-induced inflammation that provide a favorable outcome in wound healing, which might be highly related to its recognized immunemodulatory, anti-inflammatory, and anti-oxidant activities (Schewe, 1995; Ren et al., 2017, 2018). Together, the results demonstrated that ebselen has a higher skin repairing efficacy than gentamycin, which may make it a better antimicrobial than gentamycin, especially because it is predicted to have less severe side effects.

Moreover, we have investigated whether ebselen has the potential to act synergistically with the traditional polyphenol, curcumin, against S. aureus. The results showed that TrxR is one of the antibacterial targets of curcumin, and ebselen sensitized its ability.

Together, these results suggested that ebselen could serve as an alternative topical agent to gentamycin to treat Staphylococcal skin infections. The sensitivity of bacteria to ebselen is dependent on the antioxidant systems equipped in bacteria. Notably, ebselen increased the sensitivity of S. aureus to curcumin, which is a commonly used polyphenol antimicrobials component. Importantly, the combination of ebselen with other antimicrobials may help overcome the MDR in bacteria.

## MATERIALS AND METHODS

## Rat and Bacterial Strains

Sprague Dawley rats (male, 180–200 g) were purchased from China Three Gorges University, and approval from the Medical Animal Care & Welfare Committee of China Three Gorges University was obtained prior to using the animals for research. All rats were kept in individually cages under a constant darklight cycle in a conventional SPF animal house and were free access to food and water.

Staphylococcus aureus LT-1 was isolated from the skin lesion of a clinical patient in the First Affiliated Hospital of Three Gorges University in Hubei Province, China, with an approval for research from the Ethics Committee of the First Affiliated Hospital of Three Gorges University and written informed consent of the patient. S. aureus LT-1 strain was thoroughly identified by biochemical, 16S rRNA PCR (5<sup>0</sup> -GATAACCTACCTATAAGACT-3<sup>0</sup> and 5<sup>0</sup> - TCCATCTATAAGTGACAG-3<sup>0</sup> as primers), and MALDI-TOF MS (bioMrieux VITEK MS), and stored in our laboratory (**Tables 1**, **2** and **Supplementary Figure S2**). S. aureus ATCC25923 was stored in our lab.

## Regents

Bacteria cells were cultured in Luria Bertani (LB) medium (EMD millipore). 2-Phenyl-1, 2-benzisoselenazol-3(2H)-one (ebselen) (Daiichi), protease inhibitor cocktails (Roche), Escherichia coli DHB4 Trx was from IMCO Corp. (Stockholm, Sweden)<sup>1</sup> , Rabbit anti-sheep IgG-HRP (Santa cruz), Curcumin (Selleck), all the other reagents were from Sigma-Aldrich.

## The Inhibition of S. aureus LT-1 Growth by Ebselen

The inhibition of S. aureus LT-1 growth by ebselen was measured by visible spectrophotometer. S. aureus LT-1 cells were grown (37◦C, 220 rpm) till an OD 600 nm of 0.4 and diluted 100 times to be treated with different concentrations of ebselen (gentamycin was used as positive control) for 16 h at 37◦C in 96 wells plate, and the absorbance values at 600 nm were measured.

In addition, nuclear staining reagent propidium iodide (PI) was used to detect the inhibition efficiency of S. aureus LT-1 by ebselen. LT-1 cells were grown (37◦C, 220 rpm) till an OD 600 nm of 0.4 and treated with 22 µg/ml ebselen for 40 min at 37◦C, and washed three times with PBS and the cells were collected by centrifugation (6,000 rpm, 5 min). Nuclei were

<sup>1</sup>http://www.imcocorp.se

stained by 5 µg/ml PI for 30 min at 37◦C, and were washed three times and re-suspended in PBS after the incubation, and the fluorescence measured by flow cytometry (BECKMAN COULTER, CytoFLEX).

## Effect of Ebselen on Bacterial Morphology

Staphylococcus aureus LT-1 was grown till an OD 600 nm of 0.4 in LB medium, and treated with 22 µg/ml ebselen, 64 µg/ml gentamycin and PBS for 20 min, respectively. Cells were obtained by centrifuging (4◦C, 13000 rpm, 15 min), and fixed with 2.5% glutaraldehyde. The morphology and structure of S. aureus cells were observed under transmission electron microscopy (Hitachi H-7500, Japan).

## The Inhibition of S. aureus LT-1 TrxR Activity by Ebselen

Staphylococcus aureus LT-1 cells were cultured till an OD 600 nm of 0.4 and incubated with 22 µg/ml ebselen for 40 min. LT-1 cells were obtained by centrifugation (4◦C, 5000 rpm, 5 min), which were washed three times and re-suspended in 50 mM Tris–EDTA buffer (pH 7.4), and the protein inhibitor cocktail was added to decrease the protease activity. Finally, the cells were disrupted with before sonication (240 W, 5 min). The cell supernatants were obtained by centrifugation (4◦C, 12000 rpm, 10 min), and the protein concentration was measured by Bradford assay and used for TrxR activity assay.

The determination of TrxR activity was performed in 96 wells plate. 25 µg obtained cell lysate was incubated with 2 mM EDTA, 200 µM NADPH at 37◦C for 5 min, and added 5 µM E. coli Trx, 2 mM DTNB. The absorbance at 412 nm was detected with a VERSA micro-wells plate reader for 5 min, and the slope was used to represent TrxR activity. The activity of the untreated group was considered as 100%.

## Determination of ROS Production in Ebselen Treated S. aureus LT-1

Staphylococcus aureus LT-1 was grown till an OD 600 nm of 0.4 in LB medium, and incubated with 22 µg/ml ebselen for 40 min. The LT-1 cells were obtained by centrifugation (4◦C, 5000 rpm, 5 min). The pellets were washed three times with PBS and stained with 10 µM H2DCFH-DA for 30 min at 37◦C. After the incubation, cells were washed three times and re-suspended in PBS, and the ROS production was quantified by flow cytometry (BECKMAN COULTER, CytoFLEX).

## Rat Model of S. aureus LT-1 Skin Wound Infection

All experiments were performed in accordance with the relevant guidelines and regulations. Healthy male SD rats (body weight 180–200 g) were used for this study. Animals were housed adaptively for 1 week in SPF room and were free access to food and water. For bacterial load and wound healing tests, thirty rats were randomly divided into three groups: control group, ebselen-treated group, and gentamycin-treated group. Thick scalpel cuts were made into the dermis of

the skin on the posterior upper back and neck of rats and inoculated with 200 µl 3.26 × 10<sup>9</sup> CFU/ml S. aureus LT-1 cells. Rats were then treated intradermal administered with 80 µl 25 mg/kg ebselen, 5 mg/kg gentamycin or PBS once per day for 5 days, respectively. Twenty-four hours after the last treatment, the wound was excised for bacteria load on S. aureus after homogenization. The wound healing was calculated by formula as follows: F = (A−B)/A<sup>∗</sup> 100%, where F: wound healing rate, A: original wound area, and B: wound area post-treatment.

For inflammatory cytokines, pathological and IHC detection, forty rats were randomly divided into four groups: A to D, and the skin cuts were constructed as described above. Further, rats from group A served as a control group and were not infected or treated. Rats of groups B–D were inoculated with 200 µl of S. aureus strain LT-1 (3.26 × 10<sup>9</sup> ). Group B was intradermal administered 80 µl PBS; Group C and group D were administered 80 µl 25 mg/kg ebselen and 5 mg/kg gentamycin, respectively. All groups were treated once a day for 5 days. Blood serum was used to detect the cytokines level by ELISA. TNF-α, IL-6, and IL-1β ELISA Kits (Uscn Life Science, Inc.) were used for the quantification of cytokines. The experiment was carried out as the manufacture instructions.

## Histological Analysis

Rats from four groups were treated as described above, and were euthanized and lesional 8-mm punch biopsy (Acuderm) tissues specimens were bisected and fixed in formalin (10%) and embedded in paraffin. Paraffin sections (4 µm thick) were cut and stained with hematoxylin and eosin and Masson stain.

Further, anti-CD64 antibody and anti-S100a4 antibody were used to present neutrophils and fibroblasts, respectively, by IHC.

## Synergistic Antibacterial Activity of Ebselen Combined With Curcumin

Staphylococcus aureus ATCC25923 cells were incubated till an OD600 nm of 0.4 and diluted 100 times in LB medium. In 96 wells plate, the diluted bacteria were incubated with serial concentrations of ebselen and curcumin for 16 h at 37◦C and read the OD600nm values by VERSA microplate reader. In addition, PI staining was used to identify whether cells were synergistically killed by 5 µM ebselen and 80 µM curcumin as described above.

Further, the TrxR activity and ROS production were performed by DTNB and flow cytometry, respectively, to detect the synergistically inhibition effect of 5 µM ebselen and 80 µM curcumin on S. aureus as described above.

## The Synergist Degree Measurement

The synergist degree was measured by Bliss independent model which described as previous study and the synergist degree S was calculated by the following formula (Hegreness et al., 2008): S = (fX0/f00)(f0Y/f00)-(fXY/f00), fXY refers to the growth rate of bacteria under the two drugs in combination treatment, the concentration of one drug is X and the other is Y; fX0, f0Y refers to the growth rate of bacteria just one drug treatment, the concentration of one drug is X and the other is Y; f<sup>00</sup> refers to the growth rate of bacteria without drug; S refers to the degree of synergy. The value of the combined degree S is between −1 and 1, and the value is closer to −1, indicating that the antagonism effect, and the value is closer to 1, indicating that the synergy efficient.

## Statistical Analyses

Statistical analyses were assessed by Graph Pad Prism 6.0 (Graph Pad Software, La Jolla, CA, United States). Statistical analysis was assessed between two groups using the Student's two–tailed t-test; among multiple groups comparisons, one-way ANOVA analysis was performed. P < 0.05 was considered to indicate a statistically significant difference. P-values of < 0.05 were considered as significant (∗∗P < 0.01; ∗∗∗P < 0.001).

## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## ETHICS STATEMENT

The animal study was reviewed and approved by the Medical Animal Care & Welfare Committee of China Three Gorges University.

## AUTHOR CONTRIBUTIONS

CD, JZ, PW, TL, and YZ performed the experiments. JL and AH conceived the project. XR, JW, and LZ drafted the manuscript and analyzed the results.

## FUNDING

We are grateful for the support of the Swedish Research Council Medicine Grant (13x-3529), the National Natural Science Foundation of China (81903105), and Yichang Medical Treatment and Public Health Foundation (A19-301-50).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2019. 03016/full#supplementary-material

FIGURE S1 | 16 S rRNA Identification of Staphylococcus aureus LT-1. S. aureus LT-1 cells was identified by 16 S rRNA PCR using primers as following: 5 0 -GATAACCTACCTATAAGACT-3<sup>0</sup> and 5<sup>0</sup> -TCCATCTATAAGTGACAG-3<sup>0</sup> , and the predictive product is 115 bp. M, marker; 1–4, PCR products.

FIGURE S2 | Synergistic activity of ebselen with curcumin. The Bliss Model for Synergy confirms a synergistic effect, between 5 µM enselen and 10 µM curcumin. Degree of synergy was quantified after 2 h of treatment with ebselen in combination with curcumin.

## REFERENCES

fmicb-10-03016 January 2, 2020 Time: 14:34 # 11



**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 Dong, Zhou, Wang, Li, Zhao, Ren, Lu, Wang, Holmgren and Zou. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# New Tetramic Acids Comprising of Decalin and Pyridones From Chaetomium olivaceum SD-80A With Antimicrobial Activity

Xinzhu Wang<sup>1</sup> , Liya Zhao<sup>2</sup> , Chao Liu<sup>3</sup> , Jun Qi<sup>2</sup> , Peipei Zhao<sup>2</sup> , Zhaoming Liu<sup>4</sup> , Chunlei Li<sup>2</sup> , Yingying Hu<sup>1</sup> , Xin Yin<sup>2</sup> , Xin Liu<sup>2</sup> , Zhixin Liao<sup>1</sup> \*, Lixin Zhang2,5 \* and Xuekui Xia<sup>2</sup> \*

<sup>1</sup> Department of Pharmaceutical Engineering, School of Chemistry and Chemical Engineering, Southeast University, Nanjing, China, <sup>2</sup> Shandong Provincial Key Laboratory for Biosensor, Biology Institute, Qilu University of Technology (Shandong Academy of Sciences), Jinan, China, <sup>3</sup> Institute of Agro-Food Science and Technology, Shandong Academy of Agricultural Sciences, Jinan, China, <sup>4</sup> Guangdong Institute of Microbiology, Guangdong Academy of Sciences, Guangzhou, China, <sup>5</sup> State Key Laboratory of Bioreactor Engineering, East China University of Science and Technology, Shanghai, China

#### Edited by:

Raffaele Zarrilli, University of Naples Federico II, Italy

#### Reviewed by:

Guojun Wang, Florida Atlantic University, United States Annalisa Guaragna, University of Naples Federico II, Italy

#### \*Correspondence:

Zhixin Liao zxliao@seu.edu.cn; 101010762@seu.edu.cn Lixin Zhang lxzhang@ecust.edu.cn Xuekui Xia xiaxk@sdas.org

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 31 August 2019 Accepted: 09 December 2019 Published: 15 January 2020

#### Citation:

Wang X, Zhao L, Liu C, Qi J, Zhao P, Liu Z, Li C, Hu Y, Yin X, Liu X, Liao Z, Zhang L and Xia X (2020) New Tetramic Acids Comprising of Decalin and Pyridones From Chaetomium olivaceum SD-80A With Antimicrobial Activity. Front. Microbiol. 10:2958. doi: 10.3389/fmicb.2019.02958 Cycloaddition reactions such as intramolecular Diels–Alder (IMDA) are extremely important in constructing multicyclic scaffolds with diverse bioactivities. Using MycB as a biomarker, three new polyketides – Chaetolivacines A (1), B (3), and C (4) – with one known compound Myceliothermophin E (2) comprising of decalin and 4-hydroxy-2-pyridones were obtained from the culture of Chaetomium olivaceum SD-80A under the guidance of gene mining. The structures of these compounds were established using detailed 1D, 2D NMR, and high-resolution electron spray ionization mass spectroscopy (HRESIMS) analysis. The relative and absolute configurations of the compounds 1, 3, and 4 were elucidated by NOESY and ECD. The biosynthesis pathways of these compounds were proposed, which involves in three key genes ChaA [polyketide synthase-non-ribosomal peptide synthetases (PKS-NRPS)], ChaB, and ChaC. Compounds 1–4 were tested for their antimicrobial activities, and compounds 2 and 3 showed moderate bioactivity against Staphylococcus aureus (SA) and methicillinresistant S. aureus (MRSA) with MIC values of 15.8 and 27.1 µM. The results showed that configuration of C-21 in 3 and 4 is important for anti-SA and anti-MRSA activities. This study reveals the significant potential of the genus Chaetomium in producing new PKS-NRPS, therefore increasing the speed in the mining for new sources of antimicrobial agents.

Keywords: Chaetomium olivaceum, tetramic acids, isolation, anti-MRSA, biosynthesis pathway

## INTRODUCTION

The increasing resistance to drugs of bacteria like methicillin-resistant Staphylococcus aureus (MRSA) has become a major threat to public health (Coast et al., 1996; Stanton, 2013). The first strain of MRSA appeared in Cairo in 1961. Since then, the specific strain has spread to become a worldwide problem (Chapman et al., 2005). MRSA has increased in prevalence during the past decade due to the steady growth of elderly and immunocompromised patients and the

**254**

emergence of multidrug-resistant (MDR) bacterial strains. Because MRSA is one of the most common and problematic bacteria associated with increasing antimicrobial resistance, continuous efforts to discover compounds, develop alternative therapies, and create faster diagnostics methods are required (Kurosu et al., 2013). MRSA continues to be associated with significant morbidity and mortality rates. Vancomycin was the "gold standard" for treatment of serious MRSA infections; however, the emergence of less-susceptible strains, poor clinical outcomes, and increased nephrotoxicity associated with highdose therapy have challenged vancomycin's role as a first-line therapy. Linezolid is recommended for PO or IV treatments of skin and skin structure infections (SSSIs) and pneumonia caused by MRSA. Daptomycin (IV) should be considered for patients with MRSA bacteremia and right-sided endocarditis as well as for cases of complicated SSSIs; however, daptomycin should not be used to treat MRSA-related pneumonia. Tigecycline and telavancin are alternative (IV) treatments for SSSIs caused by MRSA; however, safety concerns have limited use of these agents (Rodvold and McConeghy, 2013). Recently, more and more attention has been devoted to fungi, which is an important resource to produce lead compounds for drugresistant bacteria (Ghisalberti, 2003). Chaetomium sp. is a large fungal group (Royles, 1995) known for its secondary metabolites with physiological activity and is considered a potential biological control group (Hazuda et al., 1999; Kirk et al., 2001). A variety of secondary metabolites can be produced by the Chaetomium sp., such as decalinepolyketides containing a tetramic acid (2,4-pyrrolidine-2,4-dione) ring; it is one of the most important classes of metabolites (Yang et al., 2006, Yang S.W. et al., 2007). The tetramic acid (2,4-pyrrolidinedione) ring system has been reported since the early 20th century (Osterhage et al., 2000) when the first simple derivative was prepared. Compounds containing this specific structural unit exhibit a diverse range of biological activities (Soytong et al., 2001). For instance, codinaeopsin (Kontnik and Clardy, 2008; Ramanathan et al., 2013) is active against Plasmodium falciparum, the causative agent for the most lethal form of malaria. Equisetin (Vesonder et al., 1979; Wheeler et al., 1999) suppresses germination or inhibits the growth of various monocotyledonous and dicotyledonous seeds. Additionally, it inhibits the growth of young seedlings and causes necrotic lesions on the roots, cotyledons, and coleoptiles of tested plant seedlings. Trichosetin (Marfori et al., 2002a, 2003b) is produced by the dual culture of Trichoderma harzianum and Catharanthus roseus callus and shows remarkable antimicrobial activity against the Gram-positive bacteria MRSA and Bacillus subtilis. UCS1025A (Nakai et al., 2000) exhibited antimicrobial and antiproliferative activities.

Many important compounds are produced by polyketide synthase-non-ribosomal peptide synthetase (PKS-NRPS) hybrid clusters, such as fusarin C (Song et al., 2004), equisetin (Kato et al., 2015), and decalinepolyketides (Han et al., 2017). This type of compounds exhibit a wide variety of biological activities (Boettger and Hertweck, 2012; Chooi and Tang, 2012). In recent years, with the development of whole genome sequencing technology, an increasing number of genomes have been released. The separation of natural products, which were biosynthesized by the regulation of particular gene clusters, has become more efficient under the guidance of genome mining. According to recent studies, collecting and mining "Diels–Alderases" (DAases) from fungi may lead to the discovery of new decalin-containing natural products (Li et al., 2016). Based on the above literature, the trans-decalin tetramic acids structure is widely present in bioactive natural products isolated from fungi. In this study, under the guidance of biosynthesis gene cluster analysis of C. olivaceum SD-80A, four derived decalin-containing natural products were obtained. Apart from myceliothermophin E (**2**) (Yang Y.L. et al., 2007; Shionozaki et al., 2012), the compounds **1**, **3**, and **4** are new. Their activity against drug-resistant bacteria was tested for compounds **1**–**4**. Compounds **2** and **3** showed moderate bioactivity against MRSA (MIC = 15.8 and 21.7 µM, respectively). In conclusion, we report the detailed isolation, structural elucidation, biological activities, and possible biosynthesis pathways of these compounds. This is also the first time that these compounds have been found to exhibit anti-MRSA activity.

## MATERIALS AND METHODS

## Genome Mining of Biosynthetic Gene Clusters

The genome C. olivaceum SD-80A was firstly deciphered using de novo sequencing technology. As reported, DAases have many important biological functions (Zheng et al., 2016). The diversity of DAases have been found to simulate the discovery of certain structural features compounds (Li et al., 2016). In this work, MycB (NCBI Accession No. AEO57198), a kind of DAases used as a signature biosynthetic marker from Myceliophthora thermophila was used as probe for mining the homologous protein based on the C. olivaceum SD-80A genome. Gene cluster analysis was implemented using anti-SMASH 3.0 (Weber et al., 2015), and gene function was predicted by NCBI combined with reported references (Doroghazi et al., 2014). Amino acid sequence alignment was performed by BLAST<sup>1</sup> .

## Producing Fungus and Fermentation

The fungus C. olivaceum SD-80A was isolated from a fece of Boselaphus tragocamelus collected in New Delhi, India. The strain was grown on a PDA medium for 7 days in the dark at

<sup>1</sup>https://blast.ncbi.nlm.nih.gov/Blast.cgi

TABLE 1 | The features of the Chaetomium olivaceum SD-80A genome.


FIGURE 1 | Comparison of Cha cluster and homologous gene cluster myc containing a PKS-NRPS hybrid, a DAse and a trans-ER. PKS-NRPS, polyketide synthase–non-ribosomal peptide synthetase; ER, enoylreductase; DAse, Diels–Alderases; MFS, major facilitator superfamily; α/β hydrotase, Alpha/beta hydrolase family.

28◦C, then transferred to five 500 mL Erlenmeyer flasks with a PDB medium for 5 days at 28◦C, with 120 rpm as the seed medium. After the seed medium grew well, the culture was evenly transplanted to 30 1 L Erlenmeyer flasks on an ultra-clean bench,

each containing 50 mL of water and 50 g of rice. Fermentation was conducted under stationary conditions for 30 days at 28◦C. After the fermentation, we procured the fungus of C. olivaceum SD-80A.



<sup>a</sup>Compounds 1 and 2 were measured at <sup>1</sup>H NMR (400 MHz), and <sup>13</sup>C NMR (100 MHz) with CDCl<sup>3</sup> as solvent.

## Solvent Extraction and Purification of Secondary Metabolites

The fungus C. olivaceum SD-80A was extracted with ethyl acetate (5 × 2 L, 24 h each time) at 25◦C to create a dark brown residue (12.0 g) after removal of the solvent in vacuo. The ethyl acetate extract (12.0 g) was subjected to a silica gel column using petroleum ether-EtOAc (100:0– 0:100, v/v) as an eluent to create seven fractions (Fr. A– G). Fr. A (0.6 g) was separated with semi-preparative highperformance liquid chromatography (HPLC) using CH3OH-H2O (60:40–100:0, v/v) as the mobile phase to create Fr. A.1 (306.7 mg). Fr. A.1 was further followed by semipreparative HPLC by using CH3OH-H2O (80:20–100:0, v/v) as the mobile phase to create **1** (22.5 mg) and **2** (48.6 mg). Fr. C (2.8 g) was subjected to a silica gel column using petroleum ether-EtOAc (15:85–0:100, v/v) to generate Fr. C.1- C.6. Fr. C.3 was separated by semi-preparative HPLC eluted by CH3OH-H2O (80:20–100:0, v/v) to yield compounds **3** (8.1 mg, Rt = 15.0 min) and **4** (9.7 mg, Rt = 17.6 min) (**Supplementary Figure S25**).

## Antimicrobial Activity Assay of Fractions

All antimicrobial assays mentioned in the article were performed using Gram-positive bacteria pathogens S. aureus (SA, ATCC 6538) and MRSA (clinical strain from Chaoyang Hospital, Beijing, China) to which vancomycin was used as a positive control, and Gram-negative bacteria Pseudomonas aeruginosa strain 14 (PA14, ATCC 15692) to which ciprofloxacin was used as a positive control. For each organism, a loopful of glycerol stock was streaked on an LB-agar plate, which was incubated overnight at 37◦C. The Candida albicans strains (type G5, 17#, ATCC 10231; type SC5314, ATCC MYA-2876) were grown on Sabouraud dextrose agar and incubated overnight at 28◦C; Amphotericin B was used as a positive control.

The fresh (about 3 days old) single colony bacteria were removed from the YPD medium plate with an inoculating ring and mixed in 1 mL RPMI1640 medium. The concentration of bacterial solution was then properly adjusted. In a 96-well plate, an 80 µL bacterial solution of the above concentration was added to each well; the first column was added with antibiotics as the positive control, and the last column was added with DMSO as the negative control. The compound was then added to be tested in the middle. During the operation, the concentration of a compound in the intermediate part was successively half diluted by the pipette. In this study, all isolated compounds (**1**–**4**) were screened for their antibacterial activity and antibiotic resistance bacteria activity toward a panel of bacteria using the MIC method.

## Experimental Apparatus and Materials

The 1D and 2D NMR spectra were recorded in CDCl<sup>3</sup> on a BRUKER ASCEND 400 AVANCE III HD spectrometer with tetramethylsilane as the internal standard. All NMR assignments were based on the <sup>1</sup>H-1H COSY, HSQC, HMBC, and NOESY spectroscopic data. High-resolution electron spray ionization mass spectroscopy (HRESIMS) data were acquired on an Agilent 6520 Q-TOF mass spectrometer (Agilent Technologies, United States). ECD spectra were recorded on a JASCO J-815 spectropolarimeter using a Nicolet Magna-IR 750 spectrophotometer. Sephadex LH-20 (Pharmacia, Sweden), MCI gel (CHP20P, 75-150 µm, Mitsubishi Chemical Industries Ltd.) and YMC ODS-AQ (S-50 µm, 12 nm, YMC Co., Ltd., Japan). Thin-layer chromatography was performed on a precoated GF254 plate (Qingdao Marine Chemical Co., Ltd., China). Spots were detected under UV light followed by heating after spraying with a 10% vanillin/H2SO<sup>4</sup> solution. Semi-preparative HPLC was carried out on an Agilent 1260 Infinity instrument equipped with a DAD detector and a Waters Sunfire Prep C-18 column (10 × 250 mm, 5 µm).

## Genome Analysis and Functional Assays

To get a genome of Chaetomium olivaceum SD-80A, de novo whole genome sequencing approach was used and a 500 bp library was sequenced on Illumina HiSeq 2500 platform. In this work, the total size of predicted assembly is 34782639 bp. The resulting 3.47 G of data were assembled to 10658 scaffolds using the SOAP de novo (v2.04) assembler (Luo et al., 2012). The

average G + C content is 55.57% (**Table 1**). The amino acid sequence of MycB (XP\_003662443.1) was used as a probe to blast the genome of C. olivaceum SD-80A. Four homologous gene sequences that were separately located at scaffold 1389, scaffold 1605, scaffold 647, and scaffold 1433 were mined with different levels of identities of 68, 32, 26, and 28%, respectively. Herein, the scaffold 1389 has a higher identity and the E-value is 0. Therefore, it is a likely candidate. Furthermore, the identity between the homologous protein encoded by the gene at scaffold 1389 and MycB was up to 68%. The corresponding gene was designated as ChaB. Combined with the gene cluster analysis results of the C. olivaceum SD-80A genome, ChaB and a PKS-NRPS gene located at scaffold 1389 were adjacent (**Figure 1**). This PKS-NRPS gene showed an identity of 68% with the myceliothermophin gene (MycA, XP\_003662442) from M. thermophila ATCC 42464 was named as ChaA. Through further sequence analysis, the MycC (XP\_003662444) from M. thermophila ATCC 42464 has a 55% identity sequence homology to ChaC, another adjacent gene of ChaB. The presence of these three genes in the C. olivaceum SD-80A genome suggests that the strain may produce decalin-containing compounds during secondary metabolism.

## Structural Analysis and Identification of Compounds

To verify our inferences, secondary metabolites of C. olivaceum SD-80A were accumulated through fermentation. Four decalincontaining compounds were obtained as follows (**Figure 2**): Compound **1** was obtained as yellow amorphous powder, and its molecular formula was designated as C29H35NO<sup>2</sup> according to the ion peaks at m/z 430.2266 [M + H]<sup>+</sup> under its HRESIMS (**Supplementary Figure S1**), which indicates that it contains 12 degrees of unsaturation. The formula was supported by the <sup>1</sup>H NMR and <sup>13</sup>C NMR data. The <sup>1</sup>H NMR spectrum (**Table 2** and **Supplementary Figure S2**) exhibited one active hydrogen at δ<sup>H</sup> 8.65 (22-NH), five CH<sup>3</sup> at δ<sup>H</sup> 0.90 (H-11), 0.86 (H-12), 1.48 (H-13), 1.43 (H-16), and 1.46 (H-17), four

CH at δ<sup>H</sup> 5.37 (H-1), 5.08 (H-15), 7.61 (H-20), and 6.34 (H-24), five benzene ring CH at δ<sup>H</sup> 7.50 (H-26, H-30), 7.45 (H-27, H-29), and 7.35 (H-28). The <sup>13</sup>C NMR data (**Table 2** and **Supplementary Figure S3**) exhibited 29 signals consisting of five methyls, three methylenes, 13 methines, and eight quaternary carbon atoms, including two carbonyl atoms at δ<sup>C</sup> 169.8 and 196.6. Furthermore, δ<sup>C</sup> 134.8 (C-19), 143.9 (C-20), 134.2 (C-21), and 169.8 (C-23) with the N atom could be elucidated to a pyridine ring.

The <sup>1</sup>H-1H COSY (**Supplementary Figure S4**) correlations of H-3, H-4, H-5, H-6, H-7, H-8, and H-9/12, correlations of H-15, H-16, correlations of H-26, H-27, H-28, H-29, and H-30 indicate three parts in blue color of **1** (**Figure 3**). The HMBC (HSQC and



<sup>a</sup>Compounds 3 and 4 were measured at <sup>1</sup>H NMR (400 MHz) and <sup>13</sup>C NMR (100 MHz) with CDCl<sup>3</sup> as solvent. Chaetolivacine A (1), yellow amorphous powder, [α] 20 D = + 18.6 (c 0.1, MeOH); HRESIMS m/z 430.2266 [M + H]+(calcd for C29H36NO2, 430.2263). Myceliothermophin E (2), white amorphous powder, HRESIMS m/z 396.2897 [M + H]<sup>+</sup> (calcd for C26H38NO2, 396.2903) (Yang Y.L. et al., 2007). Chaetolivacine B (3), yellow amorphous powder, [α] 20 D = + 28.5 (c 0.1, MeOH); HRESIMS m/z 462.3036 [M + H]<sup>+</sup> (calcd for C30H40NO3, 462.3033). Chaetolivacine C (4), yellow amorphous powder, [α] 20 D = −18.8 (c 0.1, MeOH); HRESIMS m/z 462.2999 [M + H]<sup>+</sup> (calcd for C30H40NO3, 462.3003).

HMBC see **Supplementary Figures S5, S6**) correlations from H-1 to C-3, C-5, C-9, C-10, and C-13; from H-4 to C-3, C-5, and C-10; from 11-Me to C-1, C-5, and C-9, from 12-Me to C-7 and C-9; and from 13-Me to C-1 and C-3 revealed the presence of the decalin ring moiety. HMBC correlations from H-15 to C-16 and C-3 displayed a butene group linked to C-3. Furthermore, HMBC correlations of H-4, H-5, and H-20 to C-18 supported a tetramic acid moiety and the decalin ring joined with C-18. HMBC correlations of H-24 to C-29, C-21, C-26, and C-30 and correlations of H-26, H-20, and H-30 to C-24 indicate that the benzene connected to C-24.

A NOESY experiment (**Supplementary Figure S7**) combined with a coupling constant resulted in two 1,3-diaxial correlations. J = 12.0 Hz of H-4 and H-5 indicated the two protons are in an axial position. Additionally, H-4 was related to H-11 and H-5 was related to H-9α, which indicates that **1** has a trans-decalin ring (H-4/H-11 and H-5/H-9α). The other NOESY correlations from H-11 to H-6β, from H-6β to H-8, suggest that the orientation of CH3-12 is equatorial. In addition, it can also be elucidated from the NOESY correlation that H-3, H-4, H-8, and H-11 are β configurations, and H-5 is an α configuration. Thus, the relative configurations of **1** is determined as 3S ∗ 4R ∗ 5R ∗ 8R ∗ 10S ∗ (**Figure 4**).

To establish the absolute configuration, ECD calculations were carried out to compare with the experiment plot. At first, the search for conformers was carried out by Spartan'14 with a Molecular Merck Force Field (MMFF). As a result, two conformers with an energy <10 kcal/mol were subjected to optimization to a minimum energy of b3lyp/6-31 + g(d, p), which consequently gave a Boltzmann distribution of 87.7 and 12.3%, respectively, based on their free energies (δG). Secondly, the theoretical ECD plot in methanol was predicted at a b3lyp/6- 311 + g(d, p) level and weighted with Boltzmann distribution. Solvent effects were taken into consideration using the selfconsistent reaction field (SCRF) method with a polarizable continuum model (PCM). The calculated spectrum **1** showed an excellent fit with the experimental plot, suggesting that the absolute configuration should be 3S, 4R, 5R, 8R, and 10S. Finally, the overall characterization of compound **1** was elucidated and named as Chaetolivacine A. The experimental method is based on the experience of predecessors (Yang Y.L. et al., 2007).

The molecular formula of **2** was confirmed as C26H37NO<sup>2</sup> through HRESIMS (**Supplementary Figure S8**), which has nine degrees of unsaturation. The <sup>1</sup>H and <sup>13</sup>C NMR data (**Table 2** and **Supplementary Figures S9, S10**) were very similar to that of **1**, except that the monosubstituted benzene ring connected to C-24 in **1** was replaced by the isopropyl in **2**. Comparing the <sup>1</sup>H NMR and <sup>13</sup>C NMR data (**Table 2**) with that reported data, compound **2** was determined as myceliothermophin E (Yang Y.L. et al., 2007).

Compound **3** was isolated as yellow amorphous powders. The molecular formula of **3** was designated as C30H39NO<sup>3</sup> according to the ion peaks at m/z 462.3036[M + H]<sup>+</sup> in its HRESIMS (**Supplementary Figure S11**). The <sup>1</sup>H NMR and <sup>13</sup>C NMR data (**Table 3** and **Supplementary Figures S12, S13**) of **3** are similar to that of **1**, except that the signal of one double bond at C-21/C-24 in **1** (δC21 134.2, δC24 119.5, δH24 6.34, s) was reduced to C-21 (δ<sup>C</sup> 89.8), C-24 (δ<sup>C</sup> 44.2), H-24 (δH24 3.12, d, J = 14.4 Hz, 3.02, d, J = 13.6 Hz), and a methoxy group OCH3-24 (δ<sup>C</sup> 49.8) was connected to C-21. HMBC correlations from OCH3-24 to C-20, C-21 confirmed the elucidation. Then, the structure of **3** was elucidated and named as Chaetolivacine B.

Compound **4** was also obtained as yellow amorphous powders. The molecular formula of **4** was designated as C30H39NO<sup>3</sup> according to the ion peaks at m/z 462.2999 [M + H]<sup>+</sup> in its HRESIMS (**Supplementary Figure S18**). The <sup>1</sup>H and <sup>13</sup>C NMR data of **4** (**Supplementary Figures S19, S20**) are highly similar to that of **3**, except for δ<sup>H</sup> 7.40 (H-20) of **3** and δ<sup>H</sup> 7.33 (H-20) of **4**. Compound **3**'s Rt = 15.0 min is different from **4**'s Rt = 17.6 min in HPLC (**Supplementary Figure S25**). The assignments of <sup>1</sup>H and <sup>13</sup>C signals of **4** are summarized in **Table 3**.

Attentive analyses of HSQC, <sup>1</sup>H–1H COSY, and HMBC spectra of **3** (**Supplementary Figures S14–S16**) and **4** (**Supplementary Figures S21–S23**) revealed that they have the same planar structures, which are proposed to be diastereomers. Spectral data indicated that **3** and **4** are a pair of isomers with different configurations. The observations also support the above deduction. However, based on the different chemical shifts of H-3, H-4, and the deoxy-tetramic acid moieties proposed, the configuration of C-21 in **3** and **4** is reversed (Yang Y.L. et al., 2007). In our situation, the NOESY correlation between H-15 and H-24 in **3** (**Supplementary Figure S17**) in the absence of the correlation in **4** (**Supplementary Figure S24**) suggests the relative configuration of C-21. Thus, the relative configurations of **3** and **4** were determined as 3S ∗ 4R ∗ 5R ∗ 8R ∗ 10S ∗ 21S ∗ and 3S ∗ 4R ∗ 5R ∗ 8R ∗ 10S ∗ 21R ∗ , respectively (**Figure 5**).

The absolute configuration of **3** and **4** was further elucidated by comparing the experimental ECD spectra with those calculated at the b3lyp/6-311 + g (d, p) and b3lyp/6-31 + g (d, p) levels. The results suggest that they were epimers at C-21 while both of them perform the absolute configuration in deoxy-tetramic acid moieties at 3S, 4R, 5R, 8R, 10S (Yang Y.L. et al., 2007).

## Antimicrobial Activity of Compounds

The isolated compounds **1**–**4** were evaluated for their antimicrobial properties, specifically their abilities against drug-resistant bacteria. Their antimicrobial activity against SA, MRSA, G5, SC5314, 17#, and PA14 were evaluated. As shown in the results, compounds **1**–**4** exhibited weak inhibitive activity against SC5314, 17#, and G5 as well as PA14; however, compounds **2** and **3** showed inhibitive activity against SA and MRSA (**Table 4**). Compound **3** showed moderate anti-SA bioactivity with an MIC of 10.8 µM. Compounds **2** and **3** showed potent inhibitory effects against MRSA with MIC values of 15.8 and 27.1 µM. For the isomers 21R-**3** and 21S-**4**, the MIC value of inhibition of SA and MRSA increased from 10.8 and 27.1 µM, respectively, to over 100 µM, implying that configurations of C-21 played a critical role in the biological activity toward SA and MRSA.

## Postulated Biogenetic Pathways

It has been reported that some secondary metabolites such as oteromycin (Singh et al., 1995; Uchiro et al., 2013), ZG-1494a (West et al., 1996), UCS1025 A and B (Agatsuma et al., 2002; Lambert and Danishefsky, 2005), trichosetin (Marfori et al., 2002), and talaroconvolutins A–D (Suzuki et al., 2000) have a subtle biogenetic pathway in the deoxy-tetramic acid ring system. Although the pathways of the newly isolated compounds **1**, **3**, and **4** remain unclear, a reasonable biological pathway could be assumed. The three genes ChaA, ChaB, and ChaC may play an important role in the synthesis of compounds **1**–**4** (Li et al., 2016). The proteins encoded by ChaA and ChaC functioned with


Vancomycin (McCormick et al., 1995) was positive control of SA and MRSA, MIC = 0.84 µM. Ciprofloxacin as positive control of PA, MIC = 0.19 µM; amphotericin B, positive control for SC5314, 17#, and G5, MIC = 1.69 µM.

aminoacyl polyketide aldehyde **5** from eight malonyl-CoAs and four S-adenosylmethionines (SAM). Following ChaA-catalyzed constructions and the release of aminoacyl polyketide aldehyde **5**, then Knoevenagel condensation yielded the intermediate ketone **6**; as a substrate, **6** could lead to the formation of the endo product **7** by ChaB, and **7** will be used as a substrate to get **9** through enolization. Finally, **9** would be oxidized to form **1**. Another possible biosynthesis pathway of **1** is the creation of **6** by tautomerization, oxidation, and cyclization (**Figure 6**). Compounds **3** and **4** could be formed by the hydroxyl attached to C-21 from different directions of the double bond planar and that are then methylated. The pathway involves leucine in **2** or phenylalanine in **1**, **3**, and **4** that participates in a polyketide reaction.

Drug-resistant bacteria are microorganisms that produce tolerance to corresponding antibiotics after a long period of antibiotic selection. MRSA has been recognized as one of the major pathogens that cause nosocomial infections (Imamura et al., 2001). Chaetomium, which belongs to the family of Chaetomiaceae, contains >100 species derived from marine and terrestrial environments. In previous investigations, Chaetomium producing >200 bioactive metabolites has been a valuable source for mining anti-MRSA compounds (Zhang et al., 2012). Our study clearly shows that C. olivaceum hosts rich genes, including I PKS, NRPS, PKS-NRPS, terpene, and other kinds. However, trans-decalin ring systems were found in Chaetomium for the first time. In this study, compounds **1**–**4** containing naphthalene type tetramic acid groups produced by C. olivaceum SD-80A were obtained. The terminal of the tetramic acid group in compounds **1**, **3**, and **4** were identical benzene rings. Compounds **3** and **4** were identified as isomers with different configurations of C-21. These compounds were tested against drug-resistant bacteria and revealed moderate activity, and the configuration of C-21 is important for anti-MRSA activity in these kind of compounds. As reported before, the configuration of C-21 in the two sets of diastereomers of myceliothermophins A–E also play a critical role in the biological toxicities toward cancer cells (Yang Y.L. et al., 2007). The results revealed the significant potential of the genus Chaetomium in producing new compounds with PKS-NRPS, which will speed the mining for new sources of antimicrobial agents.

Genome mining has recently been used for the discovery of new secondary metabolites. Cycloaddition reactions such as IMDA reactions are found to occur widely among fungal natural product pathways of PKS and NRPS (Minami and Oikawa, 2016), like lovastatin (Auclair et al., 2000) and cytochalasin (Scherlach et al., 2010). Therefore, using DAases (MycB) as biosynthesis marker to focus C. olivaceum SD-80A efficiently from our fungi library, then decalin-containing natural products **1**–**4** were obtained, and these compounds involved three proteins encoded by key genes ChaA responding to the synthesis of the acetate-derived portion, including the decalin. Concurrently, the NRPS adds the amino acid ChaB, which responds for a Diels–Alder reaction, and ChaC, an auxiliary enoyl reductase (ER) is created. The predicted biosynthesis pathway provided a theoretical basis for the future biosynthesis of these compounds using a heterologous expression. For compounds **1**–**4**, leucine in **2** or phenylalanine in **1**, **3**, and **4** involved in the biosynthesis pathway **2** with isopropyl showed stronger bioactivity than **1** with phenyl against MRSA with an MIC of 15.8 µM and over 100, respectively. Therefore, the different groups from amino acid as a precursor may play an important role in the antimicrobial activity of this kind of compound. The experiments for more novel bioactive compounds of this class using different amino acid as a precursor are ongoing.

## CONCLUSION

In this study, using genome mining and ChaB as a probe, compounds **1**–**4** were isolated as polyketones containing naphthalene-type tetramic acid groups produced by C. olivaceum SD-80A. This is the first report of this kind in regard to compounds generated from Chaetomium. The terminal of the tetramic acid group in compounds **1**, **3**, and **4** were identical benzene rings, and compounds **3** and **4** were identified as isomers with different configurations of C-21. These compounds were tested against drug-resistant bacteria and revealed moderate activity, and configurations of C-21 are important for anti-MRSA activity in this kind of compound. The biosynthetic pathway was proposed, which provides a theoretical basis for the future biosynthesis of these compounds in C. olivaceum SD-80A. This also presents an interesting potential application in food, feed, as well as pharmaceutical industries. The experiments for more novel bioactive compounds of this class using different amino acid as precursor are ongoing.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## AUTHOR CONTRIBUTIONS

XX, ZXL, and LXZ conceived, designed, and coordinated the study, edited the manuscript, and critically revised the manuscript. XW contributed to the extraction and purification of the compounds, data acquisition, analysis, and interpretation, and manuscript drafting. LYZ did the bioinformatics, genome analysis, and helped in the manuscript writing. ZML did the chemical calculation. JQ and XL helped with the experimental assays and the manuscript revision. PZ, CLL, and XY helped with the antimicrobial experiments and verifying the identity of all the strains. CL and YH contributed to the manuscript preparation.

## FUNDING

We gratefully acknowledge the financial support of the 2016 Scientific Research Innovation Plan for Academic

Degree Postgraduates in Jiangsu Universities (KYLX16\_0264, Southeast University), the National Natural Science Foundation of China (Nos. 31320103911, 31430002, and 31770024), the Qinghai Key R&D and Transformation Project (Qinghai Science and Technology Department) (2017-NK-C25), the Open Project of Qinghai Key Laboratory of Qinghai–Tibet Plateau Biological Resources, Grant No. 2017-ZJ-Y10, the Priority Academic Program Development of Jiangsu Higher Education Institutions (No. 1107047002), the Natural Science Foundation from Shandong Province (Nos. ZR2017MH025 and ZR2017ZB0206), and the Taishan Scholar Project from Shandong Province to LXZ.

## REFERENCES


## ACKNOWLEDGMENTS

We thank Zhang's Lab for their excellent technical assistance. We gratefully thank the Shandong Analysis and Test Center for NMR core facility (BRUKER).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02958/full#supplementary-material



**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 Wang, Zhao, Liu, Qi, Zhao, Liu, Li, Hu, Yin, Liu, Liao, Zhang and Xia. 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.

# COL<sup>R</sup> Acinetobacter baumannii sRNA Signatures: Computational Comparative Identification and Biological Targets

Viviana Cafiso<sup>1</sup> \*, Stefano Stracquadanio<sup>1</sup> , Flavia Lo Verde<sup>1</sup> , Veronica Dovere<sup>2</sup> , Alessandra Zega<sup>1</sup> , Giuseppe Pigola<sup>3</sup> , Jesús Aranda<sup>4</sup> and Stefania Stefani<sup>1</sup>

<sup>1</sup> Department of Biomedical and Biotechnological Sciences, University of Catania, Catania, Italy, <sup>2</sup> Department of Translational Research and New Technology in Medicine and Surgery, Azienda Ospedaliero Universitaria Pisana, University of Pisa, Pisa, Italy, <sup>3</sup> Department of Clinical and Experimental Medicine, University of Catania, Catania, Italy, <sup>4</sup> Departament de Genètica i Microbiologia, Facultat de Biociènces, Universitat Autònoma de Barcelona, Barcelona, Spain

#### Edited by:

Benjamin Andrew Evans, University of East Anglia, United Kingdom

#### Reviewed by:

Younes Smani, Institute of Biomedicine of Seville (IBIS), Spain Margarita Poza, Biomedical Research Institute, Spain

> \*Correspondence: Viviana Cafiso v.cafiso@unict.it

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 29 October 2019 Accepted: 19 December 2019 Published: 17 January 2020

#### Citation:

Cafiso V, Stracquadanio S, Lo Verde F, Dovere V, Zega A, Pigola G, Aranda J and Stefani S (2020) COL<sup>R</sup> Acinetobacter baumannii sRNA Signatures: Computational Comparative Identification and Biological Targets. Front. Microbiol. 10:3075. doi: 10.3389/fmicb.2019.03075 Multidrug-Resistant (MDR) and Extensively Drug Resistant (XDR) Acinetobacter baumannii (Ab) represent a serious cause of healthcare-associated infections worldwide. Currently, the available treatment options are very restricted and colistin-based therapies are last-line treatments of these infections, even though colistin resistant (COLR) Ab have rarely been isolated yet. In bacteria, small non-coding RNAs (sRNAs) have been implicated in regulatory pathways of different biological functions, however, no knowledge exists about the sRNA role on the biological adaptation in COL<sup>R</sup> Ab. Our study investigated two Italian XDR isogenic colistin-susceptible/resistant (COLS/R) Ab strain-pairs to discover new sRNA signatures. Comparative sRNA transcriptome (sRNAome) analyses were carried out by Illumina RNA-seq using both a Tru-Seq and a Short Insert library, whilst Ab ATCC 17978 and ACICU Reference Genome assembly, mapping, annotation and statistically significant differential expression (qvalue ≤ 0.01) of the raw reads were performed by the Rockhopper tool. A computational filtering, sorting only similarly statistically significant differentially expressed (DE) sRNAs mapping on the same gene in both COL<sup>R</sup> Ab isolates was conducted. COL<sup>R</sup> vs. COL<sup>S</sup> sRNAome, analyzed integrating the DE sRNAs obtained from the two different libraries, revealed some statistically significant DE sRNAs in COL<sup>R</sup> Ab. In detail, we found: (i) two different under-expressed cis-acting sRNAs (AbsRNA<sup>1</sup> and AbsRNA2) mapping in antisense orientation the 16S rRNA gene A1S\_r01, (ii) one under-expressed cis-acting sRNA (AbsRNA3) targeting the A1S\_2505 gene (hypothetical protein), (iii) one underexpressed microRNA-size small RNA fragment (AbsRNA4) and its pre-microAbsRNA<sup>4</sup> targeting the A1S\_0501 gene (hypothetical protein), (iv) as well as an over-expressed microRNA-size small RNA fragment (AbsRNA5) and its pre-microAbsRNA<sup>5</sup> targeting the A1S\_3097 gene (signal peptide). Custom TaqMan <sup>R</sup> probe-based real-time qPCRs validated the expression pattern of the selected sRNA candidates shown by RNA-seq. Furthermore, analysis on sRNA 1A1S\_r01, 1A1S\_2505 as well as the over-expressed A1S\_3097 mutants revealed no effects on colistin resistance. Our study, for the first time, found the sRNAome signatures of clinical COL<sup>R</sup> Ab with a computational prediction of their targets related to protein synthesis, host-microbe interaction and other different biological functions, including biofilm production, cell-cycle control, virulence, and antibiotic-resistance.

Keywords: XDR Acinetobacter baumannii, colistin resistance, small RNAs, Illumina RNA-seq, bioinformatics

## INTRODUCTION

fmicb-10-03075 January 8, 2020 Time: 18:45 # 2

The Multidrug-Resistant (MDR) Gram-negative pathogens included recently in the WHO black list (Tacconelli et al., 2018), i.e., Acinetobacter, Pseudomonas, and various Enterobacterales (including Klebsiella, E. coli, Serratia, and Proteus) represent a serious health problem worldwide. In particular, focusing on the Acinetobacter baumannii infections, the therapeutic options for their treatment are very limited and mainly draw on colistinbased therapies (Hancock and Chapple, 1999; Zavascki et al., 2007; Vila and Pachón, 2012). Therefore, the increasing use of colistin (COL) and the spread of colistin resistant (COL<sup>R</sup> ) can consequently determine an increase in polymyxin resistance onset (Ko et al., 2007; Gales et al., 2011).

Small non-coding RNAs (sRNAs) [∼30–300 nucleotides (nt) length] have been recognized as a major class of regulatory molecules in bacteria (Barquist and Vogel, 2015). They unroll a regulatory function affecting gene expression – by base pairing to their related target mRNA – modulating transcription, translation, mRNA stability, DNA maintenance, and silencing. Functionally, sRNAs are involved in the regulation of a wide range of physiological responses, reacting to environmental signals, such as pH or temperature shifts (Wassarman, 2002). They can help the modulation of changes in cellular metabolism to optimize use of available nutrients and improve the probability for survival, as well as contributing to virulence (Chabelskaya et al., 2010; Gottesman and Storz, 2011; Sayed et al., 2012; Álvarez-Fraga et al., 2017; Kröger et al., 2018).

sRNAs act in trans or cis depending on the transcription start position where the sRNA is transcribed with respect to their regulated gene. Trans-encoded sRNAs are transcribed at a genetic locus separated from the gene that they regulate and they often work via an imperfect base pairing with their target mRNAs. Conversely, cis-acting sRNAs are transcribed from the same genetic locus that they regulate even thought in an antisense orientation to the target gene. Since transcribed proximally to their target, cis-antisense RNAs share a perfect match with their targets allowing duplex formation and a stringent regulation (Thomason and Storz, 2010; Chang et al., 2015; Georg and Hess, 2018).

A third class of small RNA, the microRNA-size small RNA fragments (∼15–26 nt) have been identified in a few different species of bacteria by next generation sequencing (NGS). These molecules could originate from a cut of a longer precursor (presRNA) in analogy with maturation of the eukaryotic microRNAs (Bloch et al., 2017).

Few mechanisms have been proposed to the underlying colistin resistance in Ab (Li et al., 2006; Adams et al., 2009; Falagas et al., 2010; Moffatt et al., 2010; Arroyo et al., 2011; Cai et al., 2012; Hood et al., 2013; Parra-Millán et al., 2018; Cafiso et al., 2019), however, no investigations have been conducted so far on the role that sRNAs exert on the biological adaptations of colistin resistance acquisition. Few studies have analyzed the sRNA contribution to the A. baumannii biology and antimicrobial resistance. Weiss et al. (2016) offer a detailed view of the sRNA content of Ab and provide new insights into the evolution and role of these regulatory molecules. In detail, using RNA-seq, 78 Ab sRNAs were identified in the AB5075 background, grouped in six classes of similar sRNAs, with one particularly abundant and homologous to regulatory C4 antisense RNAs found in bacteriophages. Sharma et al. (2014) found three new sRNAs, namely AbsR11, AbsR25, and AbsR28, hypothesizing an sRNA involvement in the regulation of antibiotic resistance in bacteria, specifically in cryptic A. baumannii.

Our study aimed to gain new insight on the sRNA signature and biological target of clinical COL<sup>R</sup> Ab by high-throughput technology RNA-seq.

Our data define the distinctive signatures of COL<sup>R</sup> Ab sRNAs revealing computational predicted targets involved in the protein synthesis machinery, host-microbe interactions, pathways involved in biofilm production, cell cycle control, virulence, and antibiotic-resistance.

## MATERIALS AND METHODS

## Bacterial Strains

Two Italian Extensively Drug Resistant (XDR) isogenic colistin-susceptible/resistant (COLS/<sup>R</sup> ) clinical Ab strainpairs (1-S/R, 2-S/R) were investigated. Ab strain-pair source, antimicrobial susceptibility, molecular characterization, genomic epidemiology, and phylogeny were previously characterized (Cafiso et al., 2019).

## RNA-Seq

To optimize data, RNA-seq was carried out using the Illumina Mi-seq Standard pipeline on two biological replicates consisting of two different libraries of each strain i.e., a Single-end library with 50 bp reads (Short-Insert library) and a Paired-end Read library with 150 bp reads (Tru-Seq library).

For RNA extraction, a single colony of each Ab strain was grown in 10 ml of Cation-adjusted Mueller-Hinton broth (Ca-MHB) (Becton Dickinson) and incubated at 37◦C overnight. The overnight cultures were then diluted 1:50 in 50 ml of Ca-MHB in a sterile 250 ml flask and incubated with shaking at 250 rpm at 37◦C. Bacterial pellets were harvested at mid-log growth phase (OD<sup>600</sup> 0.6 ∼1.5 × 10<sup>7</sup> CFU/ml, 18 h) according to the strain growth-curves (**Supplementary Figure S1**).

RNA extraction was performed using the NucleoSpin RNA kit (Macherey-Nagel, Düren, Germany) following the manufacturer's protocol with minor modifications according to the previously published protocols (Cafiso et al., 2019).

The total RNA quality was checked by the 2200 TapeStation RNA Screen Tape device (Agilent, Santa Clara, CA, United States) and the RNA concentrations were determined using an ND-1000 spectrophotometer (NanoDrop, Wilmington, DE, United States). RNA Integrity Number (RIN) values, ranging from 1 to 10 with 10 being the highest quality, was determined by the Agilent TapeStation 2200 system. Only RNAs with preserved 16S and 23S peaks and with RIN values >8 were used for library construction. The RIN values >8 indicated intact and high quality RNA samples usable for downstream applications as previously published (Fleige and Pfaffl, 2006).

## Library Preparation and Sequencing

The Tru-Seq library (TS) was an A Paired-end library with reads of 150 bp and average insert size of 350/400 bp. After sequencing, raw reads were processed using FastQC v0.11.2 to check data quality and, then, reads were trimmed by Trimmomatic v.0.33.2 to remove the adapters for Paired-end reads. A minimum base quality of 15 over a 4-bases slidingwindow was required. Only trimmed reads with a length above 36 nucleotides were included in the downstream analysis (Cafiso et al., 2019).

The Short-Insert library (SI) was processed with an A Singleend stranded library with reads of 50 bp. After sequencing, raw reads were processed using FastQC v0.11.2 to evaluate data quality. Reads were then trimmed using Trimmomatic v.0.33.2 to remove sequencing adapters for Single-end reads, requiring a minimum base quality of 15 (Phred scale) and a minimum read length of 15 nucleotides. Only trimmed reads were included in the downstream analysis (Cafiso et al., 2019).

## Tru-Seq and Short-Insert Library Analysis

TS and SI RNA-seq reads were aligned on Ab ATCC 17978 (CP000521.1) and on Ab ACICU (CP000863.1) Reference Genomes, assembled and quantified using Rockhopper v.2.03 opportunely developed to investigate the bacterial gene structures and transcriptomes as well as validated to identify novel sRNAs (McClure et al., 2013; Tjaden, 2015; Cafiso et al., 2019).

To obtain the data, analyses were carried out with default parameters and verbose output. Rockhopper normalizes read counts for each sample using the upper quartile gene expression level. Starting from the p-values calculated according to the Anders and Huber approach, differentially expressed genes (DEGs) were selected as statistically significant by computing q-values ≤ 0.01 based on the Benjamini–Hochberg correction with a false discovery rate of <1%. In addition, Rockhopper is a tool using biological replicates when available and surrogates when biological replicates for two different conditions are unavailable, considering the two conditions under investigation surrogate replicates for each other (McClure et al., 2013).

## Comparative sRNA Prediction

A double computational filtering was carried out on the library analysis data outputs, first for the DE sRNAs in the COL<sup>R</sup> strains versus their COL<sup>S</sup> counterparts and, thus, to identify only those mapping on the same gene in both COL<sup>R</sup> Ab isolates with a statistically significant q-value ≤ 0.01.

## Determination of Small RNA Functional Categories

Functional categories of small RNA target genes were investigated by different bioinformatic tools including BLAST, PANTHER (Protein ANalysis THrough Evolutionary Relationships) Classification System, Gene Ontology (GO) Consortium, ExPASy, and KEGG.

## RNA-Seq Data Accession Number

RNA-seq data of the two different libraries have been deposited in the NCBI GEO database under study accession no. GSE109951.

## I-Tasser ab initio Structure Modeling

For the I-Tasser (Iterative Threading asseMBLY Refinement) analysis (Zhang, 2009; Roy et al., 2010, 2012; Yang and Zhang, 2015), the first N-terminal 120 AA of the Ab ATCC 17978 AbsRNA<sup>2</sup> target (A1S\_2505) was selected as a target to obtain more significant predictions, as the C-terminal part of the A1S\_2505 protein was excluded because of its low coverage with threading templates identified from the PDB-library: MTYQYHDESIVTELPEDTVFVFGSNMAGQHGSGAARVASQ HFGAVEGVGRGWAGQSFAIPTLNEHIQQMPLSQIEHYVEDF KVYAKNHPKMKYFVTALGCGIAGYKVSEIAPLFKGIHHN.

For the analysis of the microAbsRNA<sup>4</sup> target (A1S\_0501) the whole FASTA sequence of the protein was used as a template.

## Validation of sRNA-Seq Expression

AbsRNA<sup>3</sup> and microAbsRNA<sup>5</sup> were selected to be validated since they are representative of under-expression and over-expression, respectively. Particularly, their RNA-seq expression levels were validated by Custom TaqMan <sup>R</sup> Small RNA Assays as follow: a single colony of each strain was grown in 10 ml of Ca-MHB (Becton Dickinson) and incubated at 37◦C overnight. The overnight cultures were diluted 1:50 in 50 ml of Ca-MHB in a sterile 250 ml flask and incubated with shaking at 250 rpm under normal atmospheric conditions at 37◦C. Bacterial pellets were harvested in mid-log growth phase, lysed by adding lysozyme (10 mg/ml) and incubated for 1 h at 37◦C. Small RNA extraction was performed by using the mirVanaTM miRNA Isolation Kit (Ambion, Austin, TX, United States) according to the manufacture's protocol following the enrichment procedure for small RNAs. Extracted sRNAs were quantified using Eppendorf BioPhotometer D30 to assess their quality and to properly dilute them to the amount suggested by Custom TaqMan <sup>R</sup> Small RNA Assays protocol (Applied BiosystemsTM), then a stem-loop qRT-PCR was performed, one of the most commonly used real-time PCR approaches to quantify sRNAs. The quantification assay was divided into two steps: i) RNA was reverse-transcribed into cDNA using a stem-loop specific primer, and ii) the quantified RT


TABLE 1 | Oligonucleotides used in this work.

fmicb-10-03075 January 8, 2020 Time: 18:45 # 4

product was consequently used as a template for real-time qPCR with TaqMan <sup>R</sup> Fast Advanced Master Mix (Applied Biosystems). The stem-loop specific primer used for RT and the specific TaqMan <sup>R</sup> probes and primers used for real-time qPCR were provided by Custom TaqMan <sup>R</sup> Small RNA Assay Design Tool on the Applied Biosystems websiteTM. All real-time qPCRs were performed in triplicate, using Agilent AriaMx Real-Time PCR System, with three different biological replicates using one of the thermal profiles suggested by the Custom TaqMan <sup>R</sup> Small RNA Assays protocol, which provided a first enzymatic activation step of 95◦C for 10 min, followed by 40 cycles of 95◦C for 15 s and 60◦C for 60 s (Salone and Rederstorff, 2015). The expression levels of AbsRNA<sup>3</sup> and microAbsRNA<sup>5</sup> are shown as the increment/decrement fold-change (FC) in COL<sup>R</sup> (1-R, 2-R) vs. COL<sup>S</sup> strains (1-S, 2-S) in RNA-seq and real-time qPCR.

## Statistics

AbsRNA<sup>3</sup> and microAbsRNA<sup>5</sup> expression levels found by Custom TaqMan <sup>R</sup> probe-based real-time qPCRs were expressed as means ± standard deviations and analyzed by the oneway analysis of variance (ANOVA) using the on-line Free Statistics Calculators-DanielSoper (Soper, 2019) considering a p-value ≤ 0.01 as statistically significant.

## Ab ATCC 17978 Mutant Construction

Due to the difficult to manipulate the clinical A. baumannii strain-pairs related to the lack of antibiotic markers for the mutants selection, A. baumannii ATCC 17978 strain was used to generate 1A1S\_r01 and 1A1S\_2505 mutants as well as to overexpress the A1S\_3097 gene.

Gene inactivation was carried out as previously described (Aranda et al., 2010). Briefly, an internal fragment of the target gene was PCR-amplified from the A. baumannii ATCC 17978 genome, using the appropriated primers (listed in **Table 1**). The internal fragment was cloned into the pCR-BluntII-TOPO plasmid (Invitrogen), introduced by electroporation in Escherichia coli DH5α (Clontech) and selected in kanamycin-containing LB plates. Purified plasmids were then introduced by electroporation in A. baumannii ATCC 17978 strain and selected on kanamycin-containing plates. Recombinant clones were confirmed by sequencing (Macrogen) of the PCR products obtained by using the appropriate primers (**Table 1**). For overexpression, the A1S\_3097 gene was cloned, using the indicated primers (**Table 1**), into XbaI-NcoI sites of the pET-RA vector (Aranda et al., 2010). The recombinant plasmid was introduced in E. coli DH5α and once correct construction was verified by both PCR and sequencing (Macrogen), in ATCC 17978. Finally, A. baumannii transformants overexpressing the A1S\_3097 gene were selected on rifampicin- and kanamycin-containing plates and confirmed by PCR with the pETRAFW and pETRARV primers (**Table 1**).

## RESULTS

## Comparative Transcriptome Analysis

To define the characterizing sRNA traits of the two COL<sup>R</sup> vs. COL<sup>S</sup> clinical Ab strains, a comparative analysis of the DE sRNAs was conducted by a computational double cross-filtering.

RNA-sequencing generated 1,307,792 – 1,175,327 – 1,173,332 – 1,270,020 total reads in 1-S, 1-R, 2-S, and 2-R, respectively, with 96, 96, 72, and 93% mapped reads on Ab ATCC 17978; as well as 97, 97, 76, and 95% reads aligned on Ab ACICU for the TS library, whilst 2,353,045 – 2,041,858 – 1,804,167 – 1,819,349 total reads in 1-S, 1-R, 2-S, and 2-R, respectively, with 57, 56, 53, and 56% mapped reads on Ab ATCC 17978 and 59, 58, 54, and 57% on Ab ACICU for the SI library. RNome structures were previously published (Cafiso et al., 2019).

As shown in **Tables 2**, **3** and **Supplementary Table S1**, the comparative statistically significant filtering-analysis of the sRNAome sorting for DE sRNAs of the COL<sup>R</sup> vs. COL<sup>S</sup> Ab parental strains returned two different underexpressed cis-acting sRNAs (AbsRNA<sup>1</sup> and AbsRNA2) mapping in antisense orientation the 16S rRNA gene A1S\_r01, one under-expressed cis-acting sRNA (AbsRNA3) targeting the A1S\_


2505/ACICU\_02783 gene (hypothetical protein), one underexpressed microRNA-size small RNA fragment (AbsRNA4) and its pre-microAbsRNA<sup>4</sup> targeting the A1S\_0501 gene (hypothetical protein), as well as an over-expressed microRNAsize small RNA fragment (AbsRNA5) and its pre-microAbsRNA<sup>5</sup> targeting the A1S\_3097 gene coding a signal peptide involved in the cytokinin biosynthesis. In detail, the two different cisacting sRNAs – mapping on the Ab ATCC 17978 Reference Genome – targeted two different positions (**Table 2**) of the A1S\_r01 gene with a size of 35 bp in strain-pair 1 (AbsRNA1) and 39 bp in strain-pair 2 (AbsRNA2), no mapped regions were found on the Ab ACICU Reference Genome. The 70–75 nt cis-acting AbsRNA<sup>3</sup> mapped in an analog position both on Ab ATCC 17978 and Ab ACICU Reference Genomes targeting the A1S\_2505/ACICU\_02783 gene in both the Ab strain-pairs. The 107 nt pre-microAbsRNA<sup>4</sup> covering the A1S\_0501 mapping on Ab ATCC 17978 Reference Genome and its smaller fragment of 21 bp (microAbsRNA4) were found in Ab strain-pair 2 and in Ab strain-pair 1, respectively. Similarly, a 228 pre-microAbsRNA<sup>5</sup> targeting the A1S\_3097 gene in Ab strain-pair 2 and its inner fragment of 20 nt in Ab strain-pair 1 were found. Furthermore, both Ab strain-pairs presented the same aforementioned sRNAs with similar expression profiles (over- or under-expression), though the q-value did not allow, in some cases, to return the same fragments in both strain-pairs, considering the two different genomic annotations. Moreover, the only sRNA with a statistically significant expression in Ab ACICU Reference Genome was ACICU\_02783 (AbsRNA3) in both strain-pairs (**Supplementary Table S1**).

The sRNA nucleotide positions (transcription start and stop) reported by Rockhopper tool were shown in **Table 2**. In addition, none of these sRNAs targeted the 5<sup>0</sup> or 3<sup>0</sup> untranslated regions (UTR) of their target genes.

## I-Tasser

To predict the putative role of AbsRNA2, an A1S\_2505/ACICU\_02783 conserved domain (CD) BLAST search and I-Tasser ab initio protein structure prediction were computationally investigated. The three-dimensional structure of a protein can be very informative and useful to understand functional characteristics of proteins with unknown functions. This is because the structure of a protein provides the precise molecular details that often facilitate experimental characterization of an expected function. In a case in which there is no expected function, the structure of a protein can be used to facilitate its functional predictions by using the structure as a search template for better-characterized proteins that share regions of structural similarity (Kemege et al., 2011). The CD-BLAST search provided a match with the PHA00684 super family (cl10259) domain related to a protein of unknown function. On the contrary, analyzing the concordances of the highest significant prediction of the I-Tasser TM-align structural alignment and the COACH Predicted biological function, we resolved the structure and biological function as similar to the Orphan Macrodomain Protein (human C6orf130) with O-Acyl-ADP-ribose deacylase activity. In particular, the closest structural similarity of


 base per million mapped reads; •q-value ≤ 0.01 according to the Rockhopper guidelines were considered statistically significant.

the targeted A1S\_2505 was the PDB-Hit 2lgrA (TM-score 0.925) matching the human protein C6orf130, previously published as an Orphan Macrodomain Protein (human C6orf130) with an O-Acyl-ADP-ribose deacylase activity, which catalyzes the deacylation of O-acetyl-ADP-ribose, O-propionyl-ADP-ribose, and O-butyryl-ADP-ribose to produce ADP-ribose (ADP-r) with acetate, propionate, and butyrate, respectively. Due to the structural similarity, we can speculate that A1S\_2505/ACICU\_02783 could have a function similar to O-Acyl-ADP-ribose deacylase. This structural prediction was also supported by the COACH Predicted biological function, 2l8rA, defining the human protein C6orf130 in complex with ADP-ribose (C-score 0.59), matching the O-acetyl-ADPribose deacetylase receptor binding the ADP-ribose in the AA residues – G19 D20 L21 F22 H32 C33 I34 S35 R39 A42 I44 A45 L47 A87 P118 R119 I120 G121 C122 G123 L124 D125 Y150 L152- representing the binding sites. This ligand binding site (BL0101984) showed GO Molecular Functions of purine nucleoside binding (GO:0001883), hydrolase activity (GO:0016787) and deacetylase activity (GO:0019213) as well as the GO Biological Process of the purine nucleoside metabolic process (GO:0042278). Regarding the computational prediction of the putative role of AbsRNA3, the A1S\_0501 hypothetical protein referred to an integral membrane component (GO:0016021), whilst the CD BLAST search provided a CD of the cytoskeletal protein RodZ containing Xre-like HTH and DUF4115 domains related to the cellcycle control, cell division, and chromosome partitioning (cl34261 Superfamily). This result was also supported by the I-Tasser predictions. By LOMETS, the A1S\_0501 protein showed homology with the highest Norm Z-score (1.71) and 0.33 coverage with 2wus hit referred as a bacterial structural protein actin MreB that can be complexed with the cell shape protein RodZ. As regards the A1S\_0501 GO and the consensus prediction of GO term, obtained from I-Tasser, the A1S\_0501 protein showed a molecular function of DNA polymerase activity (GO:0034061) (GO-score 0.38) and DNA binding (GO:0003677) (GO-score 0.34) and the biological process of nucleic acid metabolism (GO:0090304) (GO-score 0.38).

## Validation of sRNA Expression Profile

Custom TaqMan <sup>R</sup> probe-based real-time qPCRs, dedicated for the analysis of bacterial sRNAs, validated and confirmed the RNA-seq expression profiles of two selected sRNA candidates: AbsRNA<sup>3</sup> and microAbsRNA<sup>5</sup> in both Ab strain-pairs, as shown in **Figure 1**. In detail, AbsRNA<sup>3</sup> had a statistically significant under-expression (p-value ≤ 0.01), whilst microAbsRNA<sup>5</sup> showed a statistically significant overexpression (p-value ≤ 0.01) in both COL<sup>R</sup> Ab strains compared with their COL<sup>S</sup> parents.

## sRNA Target Mutants

No COL MIC changes with respect to the wild-type Ab ATCC 17978 (COL-MIC 1 mg/L) were observed in the 1A1S\_r01 Ab ATCC 17978 and 1A1S\_2505 Ab ATCC 17978 mutants as well as in the WT + A1S\_3097 (**Table 4**).

fmicb-10-03075 January 8, 2020 Time: 18:45 # 6



†q-value ≤ 0.01 according to the Rockhopper guidelines were considered statistically significant.

## DISCUSSION

The comparative (sRNAome) integrated to bioinformatics, computational cross-double filtering and experimental validations of COL<sup>R</sup> versus COL<sup>S</sup> Ab strain-pair revealed distinctive small RNA signatures in COL<sup>R</sup> Ab.

Small non-coding RNAs have been identified so far as crucial regulatory elements in bacteria showing high structural diversity and molecular action mechanisms. The most intensively studied prokaryotic sRNA regulators are cis-acting sRNAs or trans-encoded sRNAs, however, the more recent discovery of microRNAs in prokaryotes represents a challenging field of investigation regarding bacterial regulatory mechanisms.

Our clinical COL<sup>R</sup> Ab were distinguished by 3 cis-acting sRNAs and 2 microRNA-size small RNA fragments involved in different biological networks. These distinctive features span a different area of bacterial biology involving protein synthesis apparatus, host-microbe interactions, biofilm production, cellcycle control, virulence and antibiotic resistance that emerge as sRNA targets. Notably, they do not seem apparently related to colistin resistance mechanism as demonstrated by our preliminary data on 1A1S\_r01, 1A1S\_2505 and high expressed A1S\_3097 Ab ATCC 17978 mutants showing no COL MIC variations of mutants compared to the WT Ab ATCC 17978 – reflecting, however, the wide biological adaptations that the coexistence of colistin resistance and the XDR profile implies.

Regarding the novel AbsRNA1, AbsRNA<sup>2</sup> and AbsRNA3, we can speculatively assume that via a cis-antisense regulation mechanism they could post-transcriptionally regulate the target translation. On the contrary, no regulation appears at the transcriptional level as demonstrated by the lack of statistically significant differential expression in these AbsRNA targets according to previously published transcriptomic data (Cafiso et al., 2019). On top, in COL<sup>R</sup> strains, AbsRNA<sup>1</sup> and AbsRNA<sup>2</sup> under-expression could indicate that they are characterized by sRNAs modulating the protein synthesis machinery and the amount of active ribosomes in agreement with other previous findings (Fernández-Reyes et al., 2009). The occurrence of the under-expressed AbsRNA<sup>3</sup> – targeting the O-acetyl-ADPribose – provided evidence that COL<sup>R</sup> Ab are distinguished by an AbsRNA<sup>3</sup> involved in the RNase III inhibition previously associated with different biological functions, including biofilm production, virulence, and antibiotic resistance in Gram-negative bacteria (Chen et al., 2011; Kim et al., 2013; Song et al., 2014). In fact, the O-acetyl-ADP-ribose is a substrate for several related macrodomain proteins such as the human MacroD1, human MacroD2, Escherichia coli YmdB. In E. coli, YmdB is an RNase III inhibitor that modulates many different functions including biofilm formation (Kim et al., 2013), and E. coli adaptive resistance to aminoglycosides via an antibiotic stress-induced sequential modulation of the endoribonucleolytic activity of RNase III and RNase G (Song et al., 2014).

Regarding the microRNA-size small RNA fragments (microAbsRNA<sup>4</sup> and its pre-microAbsRNA<sup>4</sup> as well as the microAbsRNA<sup>5</sup> and its pre-microAbsRNA5), our data suggest that these sRNAs exist in a premature form (pre-microAbsRNA) that act as precursors of their mature form, microAbsRNAs, likely obtained cutting the premature form. Moreover, the underexpressed microAbsRNA<sup>4</sup> and its pre-microAbsRNA<sup>4</sup> targeting a gene coding a structural membrane protein with a CD similar to cytoskeletal protein RodZ related to the cell-cycle control,

cell division and chromosome partitioning, could speculatively address its regulation of these functions. In fact, the GO-term prediction highlighted two possible molecular functions, a DNA polymerase activity and a DNA-binding function. Furthermore, the microAbsRNA<sup>4</sup> target (A1S\_0501) was previously listed as a transcript that decreased significantly upon exposure to NaCl in A. baumannii, but no relationship with the mechanisms of antimicrobial tolerance in response to monovalent cations was previously found (Hood et al., 2010).

For the over-expressed microAbsRNA<sup>5</sup> and its premicroAbsRNA5, targeting a signal peptide involved in the cytokinin biosynthetic process, we have to keep in mind that A. baumannii is a versatile pathogen that can adhere and invade numerous cell types displaying varying degrees of susceptibility to invasion, stimulating the proinflammatory immune response (Choi et al., 2008). The stimuli and signaling pathways implicated in cell death are not yet established; however, they involve imbalanced calcium homeostasis, pro-inflammatory cytokines, and oxidative stress, all traits related to strain virulence (Smani et al., 2011; Mortensen and Skaar, 2012). MicroAbsRNA<sup>5</sup> can regulate host-microbe A. baumannii interactions that shape the pathogenesis of Ab infection mediated by the host immune response.

In this study, we experimentally and computationally discovered five statistically significant DE sRNAs characterizing COL<sup>R</sup> Ab strains speculatively implicated, as cis-antisense sRNAs, in the regulation of protein synthesis machinery via under-expressed AbsRNA<sup>1</sup> and AbsRNA<sup>2</sup> and in different biological functions, including biofilm production, virulence and aminoglycoside-resistance via an under-expressed AbsRNA3. Likewise, we found two microAbsRNAs that may be involved in cell-cycle control via an under-expressed microAbsRNA<sup>4</sup> and in the host-microbe interaction via an over-expressed microAbsRNA5.

Colistin resistance onset in A. baumannii entails dissimilar biological adaptations – not exclusively related to colistinresistance – supporting the extremely complex and dynamic nature of this life-threatening microorganism and the urgent need to elucidate the role of small RNAs, whose only the tip of the iceberg is known.

## REFERENCES


This work offers a model for the identification of sRNA signatures and the prediction of their targets in A. baumannii. Although we do not have clear information on their functions yet, our bioinformatic analysis may provide indications regarding the cellular roles of these new sRNAs.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the NCBI GEO Database: GSE109951.

## AUTHOR CONTRIBUTIONS

VC and SSte conceived and designed the study. VC, SStr, FL, VD, and AZ performed the genomics, transcriptomics, real time qPCR, and bioinformatics. GP contributed to the bioinformatics analysis. JA carried out the mutant construction. All authors analyzed the data and contributed and approved the manuscript.

## FUNDING

This study was supported by a research grant PRIN 2017SFBFER from MIUR, Italy. This study was also supported by grant BIO2016-77011-R from the Ministerio de Economía y Competitividad. JA is a Serra Húnter Fellow, Generalitat de Catalunya, Barcelona, Spain.

## ACKNOWLEDGMENTS

We wish to thank the Scientific Bureau of the University of Catania (Italy) for language support.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.03075/full#supplementary-material

phosphoethanolamine modification of lipid A. Antimicrob. Agents Chemother. 55, 3743–3751. doi: 10.1128/AAC.00256-11


immune-evasion molecule. PLoS Pathog. 6:e1000927. doi: 10.1371/journal.ppat. 1000927


**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 Cafiso, Stracquadanio, Lo Verde, Dovere, Zega, Pigola, Aranda and Stefani. 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.

# Reducing the Consumption of Antibiotics: Would That Be Enough to Slow Down the Dissemination of Resistances in the Downstream Environment?

Christophe Merlin\*

Université de Lorraine, CNRS, LCPME, Nancy, France

Keywords: antibiotic resistance dissemination, antibiotic consumption, collateral antibiotic effect, resistance gene transfer, downstream environment

## CONTROLLING THE CONSUMPTION OF ANTIBIOTICS

#### Edited by:

Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil

#### Reviewed by:

Diamantis Plachouras, European Centre for Disease Prevention and Control (ECDC), Sweden Martina Barchitta, University of Catania, Italy

\*Correspondence:

Christophe Merlin christophe.merlin@univ-lorraine.fr

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 17 October 2019 Accepted: 09 January 2020 Published: 28 January 2020

#### Citation:

Merlin C (2020) Reducing the Consumption of Antibiotics: Would That Be Enough to Slow Down the Dissemination of Resistances in the Downstream Environment? Front. Microbiol. 11:33. doi: 10.3389/fmicb.2020.00033 The emergence and dissemination of antibiotic resistance is now understood as an unavoidable aspect of bacterial evolution following the consumption of antibiotics (Courvalin, 2005). This dramatic phenomenon is well illustrated by the relationship existing between the occurrence of resistances and the consumption of antibiotics (Furuya and Lowy, 2006; Davies, 2007; Davies and Davies, 2010). Mechanistically speaking, the increasing occurrence of antibiotic resistant bacteria (ARB) has been widely attributed to the selection of resistant variants that pre-exist in susceptible communities (Andersson and Hughes, 2014). Such resistant bacteria are supposedly outcompeting the rest of the microbial communities in a context where antibiotics are administrated at relatively high levels, which means that local concentrations are well-over the Minimum Inhibitory Concentrations (MICs). Despite the fact that the increasing occurrence of antibiotic resistances among bacteria has been recognized decades ago as resulting from antimicrobial drug consumption, only recently has the seriousness of the situation been considered by international, national and local health organizations/agencies. This awareness led to series of reports and recommendations intending to educate and improve practices of health professionals and consumers, in order to preserve the effectiveness of our therapeutic potential (MSS, 2011; World Health Organization, 2015; O'Neil Report, 2016; EUR-Lex, 2018). Considering the correlation between antibiotic consumption and occurrence of resistances in bacteria (Davies, 2007), most recommendations proposed to take action in the public health and veterinary/farming domains by limiting the inappropriate exposure of bacteria to antibiotics in order to slow down a natural evolution toward resistance and its spread in the downstream environment in a One Health context (World Health Organization, 2015, 2017). Limiting the inappropriate exposure of bacteria to antibiotics implicitly means (i) reducing the need for antibiotics, which can be achieved with infection control measures that would limit the epidemic spread of resistant bacteria, and (ii) a better usage of antibiotics so as to reduce our overall antibiotic consumption when unnecessary.

Even if there is a great disparity between countries regarding the consumption of antibiotics (European Centre for Disease Prevention and Control (ECDC), 2017), change in practice remains difficult to implement when public health is concerned. In any case, taking action to reduce antibiotic resistance requires a coordinated and multi-sectorial approach combining political commitment, resources, specific governance mechanisms, and practical managements, as recently reported by World Health Organization (2018). In its 2018 reports, the ECDC indicated that the overall consumption of antibiotics in the EU did not significantly change in the community and the hospital sectors, while a few decreasing and increasing trends were observed for some countries over the 2013–2017 period. Changes in consumption were probably more visible in veterinary medicine. In a report covering the 2011–2016 period on veterinary antibiotics sale, the European Surveillance of Veterinary Antimicrobial Consumption related an overall decrease of 20% aggregated for 25 countries (European Surveillance of Veterinary Antimicrobial Consumption (ESVAC), 2018). This was tentatively explained by the implementation of policies and measures aiming at reducing the misuse of antibiotics. Even if the studied period is too short yet to draw robust conclusion, the first effects of such responsible-use campaigns start to be visible. In France for instance, an unprecedented national plan to reduce the antibiotic consumption in the animal sector has been initiated (Ecoantibio, 2017). This led to a drastic 39% reduction of antibiotic prescription in veterinary medicine in 6 years, all animals considered. The reduction was even stronger for critical antibiotics such as fluoroquinolones (81% reduction) and last generation cephalosporin (75% reduction). According to the French surveillance network of antimicrobial resistance in pathogenic bacteria of animal origin, these measures were followed by a net diminution of pathogenic ARBs (RESAPATH, 2016). As reported by the French National Public Health Agency (Santé Publique France, 2018), using data also presented by the European Food Safety Agency, the proportion of resistant E. coli for C3G went down from 16% to <2% in poultry between 2010 and 2017, which was dramatically increasing before 2010 (Bourély et al., 2018; European Food Safety Agency (EFSA), 2018; Santé Publique France, 2018). Although more results are necessary to comfort these results, they tend to demonstrate that a better use leading to a reduced consumption of antibiotics can rapidly result in a sensible decrease of the relative occurrence of ARBs. If several other reports are rather encouraging to pursue in that direction (Seppala et al., 1997; Aarestrup et al., 2001; Dutil et al., 2010), the relationship between occurrence of resistance and antibiotic consumption does not always follow this trend. Indeed, even if it is not the vast majority of the reported cases, stopping or increasing the consumption of a given antibiotic does not always result in the concomitant decrease or increase of the corresponding resistances, and this may vary according to the studied environment, the public/animal concerned, and the antibiotic and bacteria considered. For instance, Lai et al. (2011) reported a negative correlation between a decreasing consumption of cefotaxime and the rate of cefotaxime resistant-Escherichia coli pathogens isolated in a Taiwanese university hospital. Similar trends were reported for the consumption of ceftriaxone and ceftriaxone-resistant E. coli and Klebsiella spp. in a Turkish hospital setting (Altunsoy et al., 2011). Negative correlations between antibiotic consumptions and development of resistances can also work the other way around, and may depend on the bacterial species considered. In a Korean study covering six university hospitals, Kim et al. (2018) observed contrasted results following an increased consumption of fluoroquinolones, where the resistance rate for ciprofloxacin in E. coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa, either increased, remained stable or decreased, respectively over an 8-years period. Surprisingly, the same authors also found negative correlation between decreasing consumptions of aminoglycosides and the resistance rate for third generation cephalosporins and ciprofloxacin, thus disconnecting a given drug consumption from its effect on the corresponding antibiotic resistance, at least for a few documented cases. To go further, it is worth noting that carbapenem-resistant P. aeruginosa could be isolated from animals that have not been previously treated with carbapenems. In this case, the resistance to carbapenem was attributed to an efflux pump dysregulation (rather than a carbapenemase) resulting from mutations possibly selected by disinfectants and other antibiotics in veterinary practices (Haenni et al., 2017), thus showing that resistant phenotypes can emerge independently from the presence of the corresponding antibiotics. On the other hand, the identification of antibiotic resistance genes in metagenomes from 30,000-years old sediments reminds us that resistance phenotypes and their corresponding genes probably existed before the so-called antibiotic era (D'Costa et al., 2011). Taken together, these observations clearly indicate that the emergence and the dissemination of antibiotic resistances in bacteria cannot solely be explained by a simple selective process occurring during antibiotics therapy, even if the latter is an important driving parameter in many instances.

## THE ANTIBIOTIC COLLATERAL EFFECTS

Tackling the spread of antibiotic resistance will surely require a better usage of antibiotics in order to slow down the emergence of resistant variants associated to antibiotic therapies. But, considering the indispensability of antibiotics in modern medicine, antibiotic resistances will continue to be released in anthropogenically-impacted environments where ARBs can persist, accumulate, transfer their resistant genes (ARGs) to indigenous microbes, and finally re-enter the food chain and contaminate human and animal guts for a new round of selection (Davies and Davies, 2010). It should be noted here that the environment has been described as a reservoir of ARGs in several occasions (Berendonk et al., 2015). Considering that the dissemination of antibiotic resistances lies on the acquisition of resistance but also implies a transmission, and therefore a contact, between people, or with wastewater, or manure, or animals, tackling the dissemination of ARB and ARGs will surely require controlling both the usage of antibiotics but also the route of transmission, especially at the environmental level. With that respect, Collignon et al. (2018) recently proposed that the transmission of ARB and ARGs was probably the dominant contributor to consider for controlling antibiotic resistance, which implies to act at other levels than the antibiotic consumption as well.

The global reduction of consumption is not the sole important measure implemented by national and international organizations for better usage antibiotics. The classification of antimicrobial agents as critically important molecules for human health (WHO classification list), the restriction of their availability/delivery, and the confinement of particular antibiotic usages to human medicine are important measures aiming at preventing the emergence of particular resistances in the animal husbandry sector and their dissemination in the human health sector (EUR-Lex, 2018; OIE: World Organisation for Animal Health, 2018; World Health Organization, 2019). Nevertheless, confining the usage of a given antibiotic is likely to be of limited impact if collateral effects were to be observed between antibiotics of different nature on the emergence and the dissemination of unrelated ARGs. Lately, Scornec et al. (2017) demonstrated that Tn916, a mobile genetic element involved in the dissemination of an ARG for tetracycline, could exhibit a 1,000-fold increase of its transfer frequency when exposed to sub-inhibitory concentrations of tetracyclines, but also macrolides, lincosamides, and streptogramins. This means that not only sub-inhibitory concentrations of an antibiotic could stimulate the dissemination of its corresponding resistant gene, but that collateral stimulation by other antibiotics is also possible. This tends to rule out the effectiveness, at least partially, of any measure that would be based on confining the usage of cross-reacting antibiotics. On another ground, the use of trace metal elements such as zinc oxide or copper sulfate, is frequently used as an antibiotic alternative to promote growth of livestock and poultry. Consequently to such practices, several authors reported a concomitant increase of ARB and ARGs that are likely to result from coselective processes, as ARGs and metal resistance genes can collocate on the same genetic entities (mobile genetic elements) (Hasman et al., 2006; Yin et al., 2017; van Alen et al., 2018).

## CONCLUSION

Collateral effects of antibiotics at sub-inhibitory concentrations and trace metal elements clearly highlight the fact that the

## REFERENCES


antibiotic resistance risk should not be associated with the sole antibiotic therapy practices, and should rather be considered as a multifactorial problem where co-selection and stimulation of horizontal gene transfer also fully applies. Further in depth epidemiological studies should allow determining the extent of such collateral effects outside the context of a Petri dish, and may explain why some antibiotic resistances escape any reduction of occurrence while reducing the corresponding antibiotic consumption. On the other hand determining exhaustively which antibiotic molecules exhibit collateral effects, and at which concentrations, would be an additional step toward antibiotic risk assessment, whether for therapeutic practices or for the effect of antibiotics once diluted in the downstream environments.

## AUTHOR CONTRIBUTIONS

CM wrote this opinion paper.

## FUNDING

The author wishes to thank the French National Agency for Research (ANR) (Project ReguloMobile, No. ANR-13-ADAP-0009) and the French Agency for Food, Environmental and Occupational Health & Safety (Project AquaResist, PNREST Anses, 2018/1/052) for research funding.

## ACKNOWLEDGMENTS

This article was based upon work carried out in the framework of the COST Action ES1403: New and emerging challenges and opportunities in wastewater reuse (NEREUS), supported by COST (European Cooperation in Science and Technology). The author also wishes to thank Sandrine Baron for fruitful discussion during the preparation of the manuscript.


préservé! Available online at: https://www.santepubliquefrance.fr/maladies-ettraumatismes/infections-associees-aux-soins-et-resistance-aux-antibiotiques/ resistance-aux-antibiotiques/documents/rapport-synthese/consommationd-antibiotiques-et-resistance-aux-antibiotiques-en-france-une-infectionevitee-c-est-un-antibiotique-preserve (accessed October 15, 2019).


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

Copyright © 2020 Merlin. 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.

# Evaluation of Machine Learning Models for Predicting Antimicrobial Resistance of Actinobacillus pleuropneumoniae From Whole Genome Sequences

Zhichang Liu1,2,3,4, Dun Deng1,2,3,4, Huijie Lu1,2,3,4, Jian Sun<sup>5</sup> , Luchao Lv<sup>5</sup> , Shuhong Li1,2,3,4 , Guanghui Peng1,2,3,4, Xianyong Ma1,2,3,4, Jiazhou Li1,2,3,4, Zhenming Li1,2,3,4, Ting Rong1,2,3,4 \* and Gang Wang1,2,3,4 \*

#### Edited by:

Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil

#### Reviewed by:

Denise Mara Soares Bazzolli, Universidade Federal de Viçosa, Brazil Quézia Moura, Federal University of Grande Dourados, Brazil

#### \*Correspondence:

Ting Rong rongting@gdaas.cn Gang Wang wanggang@gdaas.cn

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 28 May 2019 Accepted: 10 January 2020 Published: 06 February 2020

#### Citation:

Liu Z, Deng D, Lu H, Sun J, Lv L, Li S, Peng G, Ma X, Li J, Li Z, Rong T and Wang G (2020) Evaluation of Machine Learning Models for Predicting Antimicrobial Resistance of Actinobacillus pleuropneumoniae From Whole Genome Sequences. Front. Microbiol. 11:48. doi: 10.3389/fmicb.2020.00048 1 Institute of Animal Science, Guangdong Academy of Agricultural Sciences, Guangzhou, China, <sup>2</sup> State Key Laboratory of Livestock and Poultry Breeding, Guangzhou, China, <sup>3</sup> Key Laboratory of Animal Nutrition and Feed Science of Ministry of Agriculture (South China), Guangzhou, China, <sup>4</sup> Guangdong Engineering Technology Research Center of Animal Meat Quality and Safety Control and Evaluation, Guangzhou, China, <sup>5</sup> National Veterinary Microbiological Drug Resistance Risk Assessment Laboratory, College of Veterinary Medicine, South China Agricultural University, Guangzhou, China

Antimicrobial resistance (AMR) is becoming a huge problem in countries all over the world, and new approaches to identifying strains resistant or susceptible to certain antibiotics are essential in fighting against antibiotic-resistant pathogens. Genotypebased machine learning methods showed great promise as a diagnostic tool, due to the increasing availability of genomic datasets and AST phenotypes. In this article, Support Vector Machine (SVM) and Set Covering Machine (SCM) models were used to learn and predict the resistance of the five drugs (Tetracycline, Ampicillin, Sulfisoxazole, Trimethoprim, and Enrofloxacin). The SVM model used the number of co-occurring k-mers between the genome of the isolates and the reference genes to learn and predict the phenotypes of the bacteria to a specific antimicrobial, while the SCM model uses a greedy approach to construct conjunction or disjunction of Boolean functions to find the most concise set of k-mers that allows for accurate prediction of the phenotype. Fivefold cross-validation was performed on the training set of the SVM and SCM model to select the best hyperparameter values to avoid model overfitting. The training accuracy (mean cross-validation score) and the testing accuracy of SVM and SCM models of five drugs were above 90% regardless of the resistant mechanism of which were acquired resistant or point mutation in the chromosome. The results of correlation between the phenotype and the model predictions of the five drugs indicated that both SVM and SCM models could significantly classify the resistant isolates from the sensitive isolates of the bacteria (p < 0.01), and would be used as potential tools in antimicrobial resistance surveillance and clinical diagnosis in veterinary medicine.

Keywords: machine learning, Support Vector Machine, Set Covering Machine, antimicrobial resistance, Actinobacillus pleuropneumoniae, genomics

## INTRODUCTION

fmicb-11-00048 February 5, 2020 Time: 15:34 # 2

Antimicrobial resistance (AMR) in bacteria from humans and food-producing animals is becoming an urgent threat to the control of bacterial infections. Identification of strains resistant or susceptible to certain antibiotics is essential in fighting against antibiotic-resistant pathogens. Typically, the determination of antimicrobial susceptibility is done either by disk diffusion or minimum inhibitory concentration (MIC) assays. Identification of resistance-specific markers by PCR or microarray hybridization not only corroborates phenotypic results but is also useful for epidemiological purposes, as there are often multiple different genes that can confer resistance to a given antimicrobial agent (Bossé et al., 2017). With the increasing throughput and decreasing cost of DNA sequencing, whole genome sequencing (WGS) may be an alternative for routine surveillance of resistance profiles and for identification of emerging resistances (Mahé and Tournoud, 2018).

Actinobacillus pleuropneumoniae causes porcine pleuropneumonia, which is present in almost all the countries of the world. Pleuropneumonia can affect all ages of pigs and may result in great economic losses in pig production particularly as it causes serious respiratory distress and death. A. pleuropneumoniae is divided into 15 serotypes based on the antigenic properties of capsular polysaccharides and cell wall lipopolysaccharides. None of the serotype provides a cross-immune response for another serotype and therefore restricts the application of vaccine (Kim et al., 2016). A. pleuropneumoniae can be killed by using effective antimicrobials. However, resistant mutants increased gradually due to the misuse of antimicrobials (Zhang et al., 2018). Knowledge of resistance profiles for A. pleuropneumoniae is required to inform treatment decisions.

The presence or absence of specific resistance genes must be associated with resistance (and susceptibility) to particular antibiotics, and then the resistance profiles for all genes in a particular isolate must be added together to provide the predicted susceptibility profile for that organism. The routine studies make genotype-to-phenotype predictions based on identifying the AMR genes in the draft genomes via web servers like ResFinder (Zankari et al., 2012), the Comprehensive Antibiotic Resistance Database (CARD) (McArthur et al., 2013), and Resfams (Gibson et al., 2015).

With the help of computational tools, reference-based or reference-free machine-learning algorithms have been used increasingly to build models that correlate genomic variations with phenotypes. In supervised learning, each example consists of an input and an expected outcome. The goal of the algorithm is to learn a model that accurately maps any input to the correct outcome.

In this study, we propose to apply the Support Vector Machine (SVM) and Set Covering Machine (SCM) algorithm to accurately predict their phenotype against five antimicrobial agents (Tetracycline, Ampicillin, Sulfisoxazole, Trimethoprim, and Enrofloxacin) from the whole genomes of 96 isolates of A. pleuropneumoniae.

## MATERIALS AND METHODS

## Data

The WGS reads and binary resistance phenotypes of 5 antimicrobial agents (tetracycline, ampicillin, sulfisoxazole, trimethoprim, and enrofloxacin) of 96 isolated strains of A. pleuropneumoniae data were obtained from Bossé et al. (2017). The WGS reads were downloaded from the European Nucleotide Archive (Study: PRJEB2343<sup>1</sup> ) and the phenotypes of the isolates against the antimicrobial agents were downloaded from the **Supplementary Material** of the same study<sup>2</sup> . Acquired resistance genes of the antimicrobial agents were downloaded from ResFinder Database as reference genes<sup>3</sup> .

For enrofloxacin, even though resistance might be mediated by the acquired qnr genes, resistance to fluoroquinolones in the A. pleuropneumoniae is most often mediated by mutations in the target genes gyrA, parC, and parE (Wang et al., 2010; Pesesky et al., 2016; Zhang et al., 2018). Therefore, gene sequences of the quinolone resistance determining regions (QRDR) of gyrA (residues 68–106), parC (residues 68–106), and parE (residue 425–478) of all the isolates were translated into amino acid and aligned with the same regions of the reference gyrA (GenBank accession number ABN73394), parC (GenBank accession number ABN73680) and parE (GenBank accession number ABN74341), respectively. All the DNA sequences of QRDR with no mutation in amino acid were appended into a FASTA file as reference genes (see **Supplementary Material**).

The WGS reads were further assembled using Velvet 1.2.08 (Zerbino and Birney, 2008). The contigs of the strains along with the AMR genes downloaded from ResFinder Database and the gene sequences of QRDR for recognition of enrofloxacin point mutation were subsequently split into k-mers (sequence of k nucleotides) of length 31 using the Ray Surveyor tool (Déraspe et al., 2017).

## Reference-Based SVM Model

With the input of resistance genes of interest as reference genes, the matrix of the co-occurring k-mers in the genome of the strains and the reference genes were simultaneously built by the Ray Surveyor tool during the splitting process. Support Vector Machine (SVM; radial basis function kernel) used the number of co-occurring k-mers of the strain and the reference genes of the specific antimicrobial to learn and predict the phenotypes of each isolate. The SVM was implemented in the Python sklearn package<sup>4</sup> .

The dataset was randomly divided into three subsets of equal size by the ID of the strains, and two subsets were used for training while the other one was used for testing. The training and testing process repeated three times so that every subset of the strains could be used to evaluate the performance of the model.

<sup>1</sup>https://www.ebi.ac.uk/ena/data/view/PRJEB2343

<sup>2</sup>https://www.frontiersin.org/article/10.3389/fmicb.2017.00311/full# supplementary-material

<sup>3</sup>https://bitbucket.org/genomicepidemiology/resfinder\_db/downloads/ <sup>4</sup>http://scikit-learn.org/

## Reference-Free SCM Model

fmicb-11-00048 February 5, 2020 Time: 15:34 # 3

Unlike the SVM model which included k-mers of reference genes in the dataset, the SCM used to learn sparse and interpretable models of phenotypes by reference-free k-mers comparisons are performed implemented in Kover, an open-source software implemented in the Python and C programming languages<sup>5</sup> . Kover automates the machine learning analysis (e.g., dataset splitting, model selection, and model evaluation) without making assumptions about the underlying genetic mechanisms. The k-mers and phenotypic data of all the strains were used and packaged into a Kover dataset, and then split the dataset into a training set (2/3 of the Kover data) and a testing set (1/3 of the Kover data) according to the same ID of the datasets of the SVM model. The training set was used to learn a model containing combination rules of both conjunction (logical-AND) and disjunction (logical-OR) at most 5 rules, the testing dataset was used for testing the accuracy of the model.

## Model Selection and Performance Evaluation

In order to minimize the waste of the training dataset and avoid overfitting, five-fold cross-validation was performed on the training set of the SVM and SCM model to select the best hyperparameter values. The best hyperparameter values selected from the five-folds cross-validation were then averaged and chosen to evaluate the performance of the model.

The performances of the SVM and SCM model were evaluated in terms of sensitivity, specificity, accuracy, and precision. They were defined as: sensitivity = TP/(TP + FN), specificity = TN/(TN + FP), accuracy = (TP + TN)/(TP + FP + TN + FN), and precision = TP/ (TP + FP). Where TP was the number of resistant strains predicted to be resistant, TN was the number of sensitive strains predicted to be sensitive, FP was the number of sensitive strains predicted to be resistant, and FN was the number of resistant strains predicted to be sensitive.

## RESULTS

A total of 96 clinical A. pleuropneumoniae isolates were included in the study, with 58, 19, 46, 16, 6 of the isolates resistant to Tetracycline, Ampicillin, Sulfisoxazole, Trimethoprim, and Enrofloxacin, respectively. There were 8 isolates were resistant to four kinds of antimicrobials (Tetracycline, Ampicillin, Sulfisoxazole, and Trimethoprim); 17 isolates were resistant to 3 kinds of antimicrobials, 10 of them were resistant to Tetracycline, Ampicillin, and Sulfisoxazole, 7 of them were resistant to Tetracycline, Ampicillin, and Trimethoprim, respective; 22 isolates were resistant to 2 kinds of antimicrobials, 20 of them were resistant to Tetracycline and Sulfisoxazole, one of them was resistant to Tetracycline and Ampicillin, one of them was resistant to Sulfisoxazole and Trimethoprim, respectively; 18 isolates were resistant to single antimicrobial, 12 and 6 of them were resistant to Tetracycline and Enrofloxacin, respectively; and 31 were sensitive to all kinds of the five antimicrobials (**Figure 1**).

The gyrA QRDR DNA fragments of all the 90 Enrofloxacin sensitive isolates were the same as that region of the reference gyrA, while part of the isolates contained the same QRDR DNA fragments as the reference parC or parE genes. And including those fragments, there were 5 and 2 DNA fragments in the 90 Enrofloxacin sensitive isolates that code the same amino acid as reference parC and parE, respectively.

A total of 4,299,871 distinct k-mers of length 31 were obtained from the 96 genomes of A. pleuropneumoniae. By comparing the k-mers of the genes downloaded from ResFinder, a range of 509∼607 k-mers of tet(B) gene were found in the genome of 50 strains, 540∼613 k-mers of tet(H) gene in 5 strains, 454∼463 k-mers of blaROB-1 gene in 19 strains, 299∼402 k-mers of sul2 gene in 46 strains, and 172∼236 k-mers of dfrA14 gene together with 13 k-mers of dfrA30 gene in 16 strains of the bacteria, respectively. For enrofloxacin, 53 and 84 k-mers of gyrA QRDR in the genomes of 7 and 89 isolates, 84 and 126 k-mers of parC and parE QRDR in 96 isolates, respectively (**Figure 2**). No k-mer of qnr genes were found in the genomes of the isolates.

The training accuracy (mean cross-validation score) and the testing accuracy of SVM and SCM models of five drugs were above 90% (**Figure 3**), indicating that both of the SVM and SCM models were not overfitted. Average and standard deviation of the sensitivity, specificity, accuracy, and precision measured on the 3 randomly partition testing sets representing the whole 96 unduplicated strains of the bacteria were provided in **Table 1**. The accuracies of Ampicillin, Sulfisoxazole, Trimethoprim, and Enrofloxacin were 1.00 ± 0.00, indicating that no false positive and no false-negative strain of bacteria were predicted by both of the models.

Even though 3 and 8 out of 58 phenotype resistant strains were predicted to be sensitive for tetracycline from SVM and SCM model, respectively, the sensitivity and accuracy of both of the models were still high enough for prediction. Of the 3 false-negative isolates (MIDG3342, MIDG3352, and MIDG3356) predicted by SVM and SCM, no acquired tetracycline-resistant genes were found in the genome of those isolates. Of the other 5 false negative isolates predicted by SCM, all of them were found to carry the tetH gene and predicted to be true positive by SVM.

Correlations between the phenotype and the model predictions of 3 subsets of testing datasets represented 96 unduplicated strains of A. pleuropneumoniae isolates were shown in **Table 2**. The results indicated that both SVM and SCM models could significantly classify the resistant isolates from the sensitive isolates of the bacteria (p < 0.01).

## DISCUSSION

Support Vector Machine (SVM) has been applied to several biological problems such as prediction of protein-protein interactions, homology detection, gene expression analysis, drug discovery, and drug resistance analysis (Cui et al., 2012; Li et al., 2016; Kouchaki et al., 2018). To our knowledge, it's the first time to use the SVM model to predict the drug resistance

<sup>5</sup>https://github.com/aldro61/kover/

FIGURE 1 | The phenotype of 96 isolates. (A) Bar plot of phenotype availability for the different drugs. (B) Venn diagram quantifying the number of instances of co-occurrence of resistance between drugs.

based on the counts of co-occurring k-mers between the genome and the reference resistance genes. Using the reference gene fragments of QRDR built by the authors could cover the deficit that no point mutation databases provide the reference genes specific for A. pleuropneumoniae. And this new method could be used for phenotype prediction of the other genes or bacteria with point mutation.

Unlike the KmerResistance, which uses the "winner takes all strategy"(Clausen et al., 2016), the exact number of co-occurring k-mers between the genome and the reference resistance genes were counted. The supervised machine learning itself can learn from the situation where k-mers not being able to match due to the mismatch, indel, non-perfect assembly, or genomic rearrangements in the query genome from the training dataset

FIGURE 3 | Bars with red color show the mean accuracy for the tuned model with five-fold cross-validation on the training dataset. Bars with blue color are the accuracy of the tuned model on the test dataset. The error bars are standard deviations. (A) SVM model, (B) SCM model.

TABLE 1 | Prediction metrics on test datasets using the best performing SVM and SCM models.


TABLE 2 | Correlation of phenotype and model predictions of SVM and SCM models.


R resistant, S sensitive. Correlation between phenotype and model predictions were calculated using Fisher's exact test in python.

and predict the correct answer while the same situation happened in the test dataset.

The Set Covering Problem is a classical question in combinatorics, computer science, operations research, and complexity theory. As of now, one of the most relevant applications of SCP is given by crew scheduling problems in railway and mass-transit transportation companies, where a given set of trips has to be covered by a minimum-cost set of pairings (Caprara et al., 2000). In this study, The SCM algorithm uses a greedy approach to construct conjunction (logical-AND) or disjunction (logical-OR) of Boolean functions to find the most concise set of genomic features (k-mers) that allows for accurate prediction of the phenotype. A conjunction model assigns the positive class to a genome if all the rules output true, whereas a disjunction model does the same if at least one rule outputs true. The method was validated by generating models that predict the antibiotic resistance of C. difficile, M. tuberculosis, P. aeruginosa, and S. pneumoniae for 17 antibiotics (Drouin et al., 2016). The obtained models, implemented in Kover, were proven to be accurate, faithful to the biological pathways targeted by the antibiotics, and they provide insight into the process of resistance acquisition.

The numbers of isolates predicted to be sensitive or resistant by SVM was exactly the same as the result predicted by mapping the reference resistant genes against the assembly of the WGS data (Bossé et al., 2017). This indicated that the SVM model is excellent in classifying the phenotype of the bacteria. In general, reference-based SVM model should be equally successful whether

they are applied to a small or large set of pathogens since the accuracy of the prediction rely mainly on whether there were reference resistant genes in the reference databases like ResFinder or other built-in databases.

Until now, no point mutation was reported in the amino acid of GyrB QRDR of A. pleuropneumoniae. Of the amino acid of parE QRDR (residues 440–479) of 96 isolates in this study, 29 sensitive isolates had substitutions of D479E in the parE protein. The other 61 sensitive isolates and the 6 resistant isolates did not have substitutions of D479E comparing with the amino acid of reference parE gene. The finding indicated that mutation of D479E in the parE gene might not be related to the resistant of the bacteria against Enrofloxacin. So, DNA sequences of QRDRs of gyrA (residues 68–106), parC (residues 68–106), and parE (residue 425–478) of sensitive isolates were chosen and appended to a FASTA file as reference genes for the SVM model to learn and predict the phenotypes of the bacteria against Enrofloxacin.

The SCM model, regardless of the resistant mechanism of which were acquired AMR genes or point mutation in the chromosome, by comparing the difference of the k-mers between the resistant strains and the sensitive strains, finds the most concise set of equivalent k-mers that allows for accurate prediction of the phenotype.

Any approach that uses machine learning models requires adequate input data to form a "training set" to train the machine learning model and a "testing set" to assess the performance of the model (Macesic et al., 2017). Among the five antimicrobial agents, the resistant background of A. pleuropneumoniae against tetracycline is more complicated than the others. There were 58 phenotype resistant strains, with 50 isolates carrying tet(B), 5 isolates carrying tet(H), and 3 isolates did not have any tetracycline resistance genes detected. And up to now, we still could not able to collect the whole genome of the A. pleuropneumoniae with tet(H) genes publicly, therefore, after randomly split the limited data into training set or testing set, the SCM model did not have enough sample to learn from the training dataset and therefore lead to a relative lower accuracy while predicting the testing set of the model.

Both models have advantages and shortcuts. The referencebased SVM model performs well at classifying resistance from sensitive isolates regardless of the sample size of the training set since the counts of co-occurring k-mers between the genome and the reference resistance genes of the resistant isolates are significantly different from that of the sensitive isolates (**Figure 2**). But this method relies mainly on the database

## REFERENCES


and therefore cannot be used for predictions where resistance mechanisms have yet to be identified. The SCM model should need enough proportion of true phenotype data against false phenotype data as input to form a "training set" to train the model, but it provides a unique approach for deciphering, de novo, new biological mechanisms without the need for prior information (Drouin et al., 2016).

Even though both of the models can use raw reads to learn and predict the phenotype of the bacteria, it is recommended to use the assembled contigs as input data, since the genome assembly can increase the quality of the k-mer representation, reduces the number of unique k-mers and thus makes the process of splitting the genome into k-mers and building the matrix encoding the presence or absence of all k-mers by Ray Surveyor tool much faster (Drouin et al., 2016; Mahé and Tournoud, 2018).

## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## AUTHOR CONTRIBUTIONS

ZCL, TR, and GW designed the study and machine learning algorithm. DD, HL, JS, and LL analyzed the data and evaluated the biological relevance of the models. SL, GP, XM, JL, and ZML acquired the data and prepared it for analysis. ZCL wrote the manuscript.

## FUNDING

This work was financially supported by the Guangdong Modern Agro-industry Technology Research System (2018LM1080 and 2018LM2153), Science and Technology Program of Guangzhou (201707010217), and the Science and Technology Program of Guangdong Province (2014A070713026 and 2014A010107032).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00048/full#supplementary-material

whole genome data. J. Antimicrob. Chemother. 71, 2484–2488. doi: 10.1093/jac/ dkw184


reference-free genome comparisons. BMC Genom. 17:754. doi: 10.1186/s12864- 016-2889-6


predicting antimicrobial resistance profiles in gram-negative bacilli from whole genome sequence data. Front. Microbiol. 7:1887. doi: 10.3389/fmicb.2016.01887


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

fmicb-11-00003 February 7, 2020 Time: 15:18 # 1

# Ectopic Expression of Rv0023 Mediates Isoniazid/Ethionamide Tolerance via Altering NADH/NAD<sup>+</sup> Levels in Mycobacterium smegmatis

Shailesh Kumar Gupta1,2† , Rajendra Kumar Angara<sup>1</sup>† , Suhail Yousuf<sup>1</sup>† , Chilakala Gangi Reddy1,3† and Akash Ranjan<sup>1</sup> \*

<sup>1</sup> Computational and Functional Genomics Group, Centre for DNA Fingerprinting and Diagnostics, Hyderabad, India, <sup>2</sup> Graduate Studies, Manipal Academy of Higher Education, Manipal, India, <sup>3</sup> Regional Centre for Biotechnology, Faridabad, India

Tuberculosis (TB) caused by Mycobacterium tuberculosis (Mtb) accounts for nearly

#### Edited by:

Mattias Collin, Lund University, Sweden

#### Reviewed by:

Yu-Min Chuang, Yale University, United States Diana Machado, NOVA University Lisbon, Portugal

#### \*Correspondence:

Akash Ranjan akash@cdfd.org.in; dr.akash.ranjan@gmail.com †These authors have contributed

#### Specialty section:

equally to this work

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 19 July 2019 Accepted: 03 January 2020 Published: 07 February 2020

#### Citation:

Gupta SK, Angara RK, Yousuf S, Reddy CG and Ranjan A (2020) Ectopic Expression of Rv0023 Mediates Isoniazid/Ethionamide Tolerance via Altering NADH/NAD<sup>+</sup> Levels in Mycobacterium smegmatis. Front. Microbiol. 11:3. doi: 10.3389/fmicb.2020.00003 1.2 million deaths per annum worldwide. Due to the emergence of multidrug-resistant (MDR) Mtb strains, TB, a curable and avertable disease, remains one of the leading causes of morbidity and mortality. Isoniazid (INH) is a first-line anti-TB drug while ethionamide (ETH) is used as a second-line anti-TB drug. INH and ETH resistance develop through a network of genes involved in various biosynthetic pathways. In this study, we identified Rv0023, an Mtb protein belonging to the xenobiotic response element (XRE) family of transcription regulators, which has a role in generating higher tolerance toward INH and ETH in Mycobacterium smegmatis (Msmeg). Overexpression of Rv0023 in Msmeg leads to the development of INH- and ETH-tolerant strains. The strains expressing Rv0023 have a higher ratio of NADH/NAD+, and this physiological event is known to play a crucial role in the development of INH/ETH co-resistance in Msmeg. Gene expression analysis of some target genes revealed reduction in the expression of the ndh gene, but no direct interaction was observed between Rv0023 and the ndh promoter region. Rv0023 is divergently expressed to Rv0022c (whiB5) and we observed a direct interaction between the recombinant Rv0023 protein with the upstream region of Rv0022c, confirmed using reporter constructs of Msmeg. However, we found no indication that this interaction might play a role in the development of INH/ETH drug tolerance.

Keywords: XRE family of protein, whiB5, isoniazid resistance, ethionamide resistance, transcription regulation

## INTRODUCTION

Tuberculosis (TB) remains a major cause of death worldwide and the leading cause by a single infectious agent (World Health Organisation, 2018). Even though the disease can be cured and managed by several multidrug regimens, the emergence of multidrug-resistant (MDR) TB is proving to be a major challenge for complete eradication of the disease. Worldwide, MDR TB constitutes 3.5% of new TB cases and 18% of previously treated cases (World Health Organisation, 2018). To overcome these challenges and to better counter resistance in Mycobacterium tuberculosis (Mtb), understanding the mechanisms and deciphering the pathways majorly responsible for generating resistance are greatly required.

fmicb-11-00003 February 7, 2020 Time: 15:18 # 2

Isoniazid (INH), in combination with rifampicin (RIF), ethambutol (EMB), and pyrazinamide (PZA), forms the firstline therapy for TB. Cycloserine, ethionamide (ETH), and amikacin/capreomycin are used as second-line drugs<sup>1</sup> . INH was first used as an anti-TB drug in 1952 and shortly the first INHresistant Mtb (INHr) clinical isolates were reported (Bernstein et al., 1952; Fox, 1952; Middlebrook and Cohn, 1953). INH and ETH are prodrugs that are converted into their active forms by proteins encoded by katG (catalase peroxidase KatG) and ethA (monooxygenase EthA), respectively (Johnsson and Schultz, 1994; Johnsson et al., 1997; Baulard et al., 2000; DeBarber et al., 2000; Vannelli et al., 2002; Fraaije et al., 2004). Although katG and inhA (encoding an NADPH-dependent enoyl-ACP reductase) are the main genes involved in INH resistance, clinical isolates with mutations in the ndh gene have been identified in INH-resistant Mtb strains (Lee et al., 2001). While in the study by Lee et al. (2001), INH resistance mutations were observed only in the ndh gene, in other studies, ndh mutations occur simultaneously with mutations in other genes (Hazbón et al., 2006; Cardoso et al., 2007). The role of the ndh gene mutations in INH and ETH co-resistance in Mycobacterium smegmatis (Msmeg) and Mycobacterium bovis (Mbovis) has been shown (Vilcheze et al., 2005), while the role of ndh in conferring resistance in Mtb is yet to be determined. The gene ndh encodes the type II NADH dehydrogenase and its ortholog in Escherichia coli (E. coli) has been characterized. In E. coli, it exists in a monomeric state bound to the membrane, where it oxidizes NADH, reduces quinone, and catalyzes the transfer of electrons from reduced flavin to quinone (Jaworowski et al., 1981; Matsushita et al., 1987; Yagi, 1993; Gennis and Stewart, 1996; Kerscher, 2000). In Msmeg, mutations in ndh lead to an increase in NADH cellular concentration and inhibition of INH-NAD and ETH-NAD adducts formation (Vilcheze et al., 2005).

Xenobiotic response element (XRE) family of transcription factors are one of the most frequently occurring families of regulators in bacteria. Among the well-studied members of the XRE family are the lambda and Cro repressors, from lambda bacteriophage, and the prophage repressor Xre from Bacillus subtilis. The XRE family of regulators share a conserved N-terminal helix-turn-helix (HTH) DNA binding domain, while the C-terminal regulatory region is highly variable. The XRE family of regulators control diverse metabolic functions; e.g., SinR regulates developmental process in B. subtilis (Gaur et al., 1991), ClgR regulates Streptomyces growth and controls Clp proteolytic complex (Bellier and Mazodier, 2004), PuuR regulates putrescine utilization pathway in E. coli K-12 (Nemoto et al., 2012), and BzdR is involved in the anaerobic catabolism of benzoate in the denitrifying Azoarcus sp. strain CIB (Barragán et al., 2005). There are seven members of the XRE family of transcription regulators in Mtb: Rv0023, Rv0465c, Rv0474, Rv1129, Rv2017, Rv2021, and EspR (Rv3849). Except for EspR, which positively regulates the ESX-1 protein secretion system, the principal virulence determinant of Mtb (Raghavan et al., 2008), the remaining XRE transcription regulators in Mtb are uncharacterized.

Rv0023 is a regulator from the XRE family of transcriptional regulators, known to induce 488 genes and repress 404 genes (Rustad et al., 2014). Rv0023 regulon is enriched for the regulation of NAD reductases (Rustad et al., 2014). Rv0023 is transcribed in an operon together with Rv0024, a gene that codes for an NLPC/p60 family protein and is transcribed divergently from whiB5, belonging to the WhiB family of transcriptional regulators. The whiB5 gene product is a positive regulator of transcription and contributes to Mtb virulence and reactivation (Casonato et al., 2012). At present, very little is known about Rv0023 functions and its effect on Mtb physiology.

Here, we studied the effects of Rv0023 overexpression in Msmeg. The results show that ectopic expression of Rv0023 confers enhanced INH and ETH tolerance in Msmeg. Rv0023 ectopic expression downregulates the expression of the ndh gene and increases NADH/NAD<sup>+</sup> levels, which are known to be mediators of INH and ETH resistance in Msmeg (Miesel et al., 1998; Vilcheze et al., 2005). We further studied the regulation of the whiB5-Rv0023 locus and identified Rv0023 as a negative regulator of whiB5. We have characterized its binding site and identified the promoters of Rv0023 and whiB5.

## MATERIALS AND METHODS

## Bacterial Strains and Growth Condition

A complete list of strains used in this study is mentioned in **Table 1**. Cloning and plasmid propagation were done using E. coli strain DH5α. For protein expression, E. coli BL21 (DE3) was used. Both strains were grown in Luria Bertani (LB) medium at 37◦C. Mbovis BCG Pasteur 1173P2, Msmeg mc<sup>2</sup> 155, and the recombinant strains were grown in Middlebrook 7H9 (Himedia) broth supplemented with 10% OADC (oleic albumin dextrose catalase) (Himedia), 0.2% glycerol, and 0.05% Tween80 (20% stock) or on 7H10 agar without Tween80 at 37◦C. Kanamycin (50 µg/ml) (Sigma-Aldrich), ampicillin (100 µg/ml) (Sigma-Aldrich), and hygromycin (50 µg/ml) (Invitrogen) were used as and when required.

## Plasmids and DNA Manipulation

Cloning, genomic, and plasmid DNA isolations were done as per standard molecular biology procedures (Sambrook et al., 1989). The plasmids and primers used in this study are listed in **Tables 2**, **3**, respectively. Overlapping extension PCR was performed to generate site-directed mutations for promoter and critical residues studies. Sequences of all clones generated were confirmed by Sanger sequencing.

## Protein Expression and Purification

pET23a-0023 plasmid containing Rv0023 ORF was used to transform E. coli BL21 (DE3) strain. Cells were grown in LB broth at 37◦C till mid-log phase, and then 1 mM IPTG (isopropyl 1-thio-β-D-galactopyranoside) was added to induce the expression of protein. Cells were grown for another 4 h at 37◦C, after which cells were harvested and purified using Ni-NTA affinity chromatography as described earlier (Yousuf et al., 2018). The purity of recombinant protein was analyzed

<sup>1</sup>https://www.cdc.gov/tb/education/corecurr/

Frontiers in Microbiology | www.frontiersin.org

#### TABLE 1 | List of strains used in this study.

fmicb-11-00003 February 7, 2020 Time: 15:18 # 3


#### TABLE 2 | List of plasmids used in this study.


#### TABLE 3 | List of primers used for cloning.


Restriction site in the primer was underlined. The point mutations incorporated in SDM primers were highlighted in bold.

by 12% SDS-PAGE and protein concentration was measured by Bradford assay.

## β-Galactosidase Assay

fmicb-11-00003 February 7, 2020 Time: 15:18 # 4

Mycobacterium smegmatis mc<sup>2</sup> 155 strain was transformed with various constructs as required and grown in 7H9 media. The cultures were grown to mid-log phase and β-galactosidase activity was measured in Miller Units (MU) and all experiments were done in triplicate (Miller, 1972).

## Electrophoretic Mobility Shift Assay

To verify the interaction between Rv0023 and the upstream region of whiB5, electrophoretic mobility shift assay (EMSA) was performed. The 500-bp upstream region and the 50-bp downstream region of whiB5 were PCR amplified and labeled with γ-<sup>32</sup>P ATP (3000 Ci mmol−<sup>1</sup> ) using T4 polynucleotide kinase as per manufacturer's instructions (New England Biolabs). Labeled DNA was purified using nucleotide purification kit (Qiagen). Purified labeled DNA was then incubated with increasing concentrations of recombinant Rv0023 protein in an EMSA reaction buffer (25 mM HEPES, pH 7.9, 0.1 mM EDTA, 10 mM MgCl2, 20 mM KCl, and 5% glycerol) with 50 ng/µl poly(dI–dC) as a non-specific DNA competitor. The protein– DNA complex was incubated for 45 min and was resolved on a 6% non-denaturing polyacrylamide gel in 0.5 × tris-borate (TBE) buffer. The gel was run for 3–4 h to allow sufficient resolution of protein–DNA complex.

Electrophoretic mobility shift assay with commercially synthesized overlapping oligonucleotide and motifs was performed to identify the exact binding region of Rv0023.

## Primer Extension

To map the (+1) transcription start site (TSS), primer extension was performed as described earlier and in Cold Spring Harbor protocols (Carey et al., 2013; Angara et al., 2018). As the ORFs and intergenic region of whiB5-Rv0023 locus are 100% conserved between Mtb and Mbovis, we used Mbovis RNA for primer extension studies. RNA was isolated with the Qiagen RNeasy kit as per manufacturer's instructions. SuperScript III reverse transcriptase (Invitrogen) was used to transcribe cDNA from total RNA (10–15 µg) using 5<sup>0</sup> -end labeled γ-<sup>32</sup><sup>P</sup> ATP primers (**Table 4**). Primer extension products were resolved on 6% polyacrylamide/8 M urea gel in TBE buffer. The size of primer extension product was determined by generating a dideoxy sequencing ladder using pUC19 plasmid as template and M13 universal primer (USB; 70140 KT).

## Stress Assay

To check the sensitivity of Msmeg toward various stress conditions, MsmegWT, MsmegpVV16, and MsmegpVV0023


cells were grown in complete 7H9 media (supplemented with 10% OADC, 0.2% glycerol, and 0.05% Tween80) at 37◦C overnight. Then, the cells were centrifuged and washed with PBS. The cells were then suspended in 7H9 media and the cell concentration was adjusted to 0.02 OD at 600 nm. The sensitivity of bacterial cultures toward 0.1% SDS (Sigma Aldrich) and 5 mM hydrogen peroxide (Merck) was measured at 37◦C for 6 h. Sensitivity to 250 µg/ml lysozyme (Sigma Aldrich) was measured at 37◦C for 24 h. Susceptibility of Msmeg toward the antibiotics (Sigma Aldrich) INH (10 µg/ml, MIC of 5 µg/ml), RIF (10 µg/ml, MIC of 1 µg/ml), and ETH (2.5 µg/ml, MIC of 10 µg/ml) was determined at 37◦C for 24 h. For determining CFU numbers for each stress condition and antibiotics, the cells were serially diluted (10-fold) and plated onto 7H10 agar plates.

Further, the strains MsmegWT, MsmegpVV16, and MsmegpVV0023 were serially diluted (10-fold) and spotted on the 7H10 agar plates containing increasing concentrations of INH (0, 5, 10, and 15 µg/ml) and ETH (0, 2.5, and 5 µg/ml). The plates were incubated at 37◦C for 48 h.

## qRT-PCR

Total RNA from MsmegpVV16 (vector control) and MsmegpVV0023 (Rv0023 protein overexpressed) cultures were isolated using Qiagen RNeasy kit. SuperScript III reverse transcriptase (Invitrogen) was used to generate cDNA using random hexamers. Real-time PCR was carried on the BioRad CFX96 system using gene-specific primers (**Table 5**) and EvaGreen qPCR mastermix (Applied Biological Materials Inc.) as per standard protocol. The fold change in expression relative to MsmegpVV16 (vector control) was calculated after normalizing to sigA. The 2−11CT method was used to calculate relative changes in gene expression (Livak and Schmittgen, 2001).

## NADH/NAD<sup>+</sup> Cellular Concentration

To measure the cellular concentration of NADH and NAD+, MsmegpVV16 and MsmegpVV0023 cells were grown to an OD<sup>600</sup> of 0.8–1.2. Cells were collected (1 ml) by centrifugation (11,000 rpm for 2 min). The supernatant was removed and 300 µl of 0.2 M HCl (for NAD<sup>+</sup> extraction) or 0.2 M NaOH (for NADH extraction) was added to the cells. The cells were resuspended and incubated at 50◦C for 10 min after which extracts were cooled to 0◦C. The bacterial suspensions were neutralized by adding 0.1 M HCl (for NADH extraction) or

TABLE 5 | List of primers used in real-time PCR.


fmicb-11-00003 February 7, 2020 Time: 15:18 # 5

0.1 M NaOH (for NAD<sup>+</sup> extraction) dropwise while vortexing at high speed. Centrifugation was done to remove the cell debris and supernatant was transferred to a new tube and used immediately. NADH and NAD<sup>+</sup> concentrations were obtained by spectrophotometrically measuring the rate of MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl tetrazolium bromide, 4.2 mM] (Amresco), reduction by yeast alcohol dehydrogenase II (Sigma Aldrich) in the presence of PES (phenazine ethosulfate, 16.6 mM) (Sigma Aldrich) at 570 nm (Leonardo et al., 1996; San et al., 2002; Vilcheze et al., 2005). The concentration of nucleotide (NADH/NAD+) is proportional to the rate of MTT reduction.

## Bioinformatics

Rv0023 protein sequence (UniProt P9WMI3) was used as a query in the SynTax web server<sup>2</sup> for synteny analysis within the Mycobacteriaceae family (Oberto, 2013); a conserved domain database search was performed to identify HTH\_3 and XRE domains (Marchler-Bauer et al., 2015). Orthologous sequences of the whiB5–Rv0023 intergenic region from different mycobacterial species were retrieved from KEGG genome database and aligned using CLUSTAL OMEGA (Sievers et al., 2011).

<sup>2</sup>http://archaea.u-psud.fr/SyntTax

plates containing increasing concentrations of ETH.

serially diluted and spotted on plates containing increasing concentrations of INH. (C) MsmegpVV16 and MsmegpVV0023 cells were serially diluted and spotted on

## Statistical Analysis

fmicb-11-00003 February 7, 2020 Time: 15:18 # 6

Data were presented as mean ± SD. Student's t-test and oneway ANOVA were used to determine the statistical significance between groups and values with p < 0.05 were considered to be significant. GraphPad Prism software version 5.02 was used for the statistical analysis.

## RESULTS

## Overexpression of Rv0023 Confers Increased INH and ETH Tolerance in Msmeg

Rv0023 is a non-essential gene in Mtb, yet it regulates a high number of genes in the genome (Sassetti et al., 2003; Rustad et al., 2014). Msmeg genome does not harbor the ortholog of Rv0023 and hence serves as a good model to study the function of Rv0023. To ascertain the role of Rv0023, we constructed an overexpressing strain of Rv0023 in Msmeg, MsmegpVV0023. The wild-type MsmegWT, vector control MsmegpVV16, and MsmegpVV0023 strains were subjected to acid fast staining to rule out the effect of vector insertion on cellular integrity. The cells were found to be intact and acid fast (**Supplementary Figure 1**). To further study the role of Rv0023 in cellular physiology, MsmegWT, MsmegpVV16, and MsmegpVV0023 were subjected to different stress conditions as depicted in **Figure 1A**, and tolerance was measured in terms of CFU. The CFU was counted for all conditions and we observed that overexpression of Rv0023 leads to increased tolerance with regard to INH and ETH, and no effect was observed with any other stress conditions. Tolerance to INH and ETH was not seen either on the wild type or in vector control strains (**Figure 1A**).

The role of Rv0023 in generating tolerant strains was further confirmed via spotting the wild type, vector control, and Rv0023 overexpressed strains in the presence of increasing concentrations of INH and ETH. MsmegWT, MsmegpVV16, and MsmegpVV0023 cells were serially diluted and spotted on plates containing 5, 10, and 15 µg/ml of INH and 2.5 and 5 µg/ml of ETH. Even at higher concentrations, MsmegpVV0023 showed tolerance toward INH and ETH (**Figures 1B,C** and **Supplementary Figure 2**). The results indicate that Rv0023 expression specifically contributes toward the higher tolerance of INH and ETH in Msmeg.

## Rv0023 Expression in Msmeg Alters the NADH/NAD<sup>+</sup> Levels

Isoniazid and ethionamide are both prodrugs and they are activated by the protein catalase–peroxidase KatG and the NADPH-specific flavin adenine dinucleotide-containing monooxygenase, EthA, respectively. After activation, they react with NAD<sup>+</sup> to form INH-NAD and ETH-NAD adduct. This species then inhibits the enoyl ACP reductase InhA, which leads to the inhibition of mycolic acid biosynthesis and eventual mycobacterial cell death (**Figure 2A**). The major common genes to both pathways are the ndh (maintains NADH/NAD<sup>+</sup> ratio) and inhA. Msmeg orthologs of these and other important genes

TABLE 6 | Cellular concentration of NADH and NAD+.

fmicb-11-00003 February 7, 2020 Time: 15:18 # 7


involved in INH and ETH resistance were subjected to qRT-PCR quantification. Total RNA was isolated from MsmegpVV16 and MsmegpVV0023 and the expression levels compared between the two strains. We found that the expression level of ndh (MSMEG\_3621) in MsmegpVV0023 was approximately twofold lower than that of MsmegpVV16. No changes in the expression levels were observed in inhA (MSMEG\_3151), katG (MSMEG\_6384), and ethA (MSMEG\_6440) between both strains (**Figure 2B**). As ndh encodes NdhII, which oxidizes NADH to NAD+, we reasoned that the ratio of NADH/NAD<sup>+</sup> might be altered in MsmegpVV0023 strain. So, we measured the cellular ratio of NADH and NAD<sup>+</sup> in MsmegpVV16 and MsmegpVV0023 strains and found that the ratio of NADH/NAD<sup>+</sup> is increased in MsmegpVV0023 (**Table 6**). These results indicate that Rv0023 expression alters both the transcript levels of ndh gene and NADH/NAD<sup>+</sup> levels. To find out whether these effects are directly mediated by Rv0023 or through some indirect means, we studied the interaction of recombinant Rv0023 with the upstream region of ndh (MSMEG\_3261) gene. However, under the experimental conditions mentioned, we did not observe any such positive interaction (**Figure 2C**).

## Rv0023 Negatively Regulates whiB5

Rv0023 is in operon with Rv0024 and is transcribed divergently from whiB5 (**Figure 3A**). Rv0023 and whiB5 are absent from the non-pathogenic, fast-growing Msmeg but are present in Mtb. So, we tried to find the co-occurrence of these two genes in other mycobacterial species. Synteny analysis of Rv0023 and whiB5 was performed for important species of the Mycobacterium genus using the "SynTax" web server (**Figure 3B**). It was observed that whiB5 is in synteny with Rv0023 and that both are present only in pathogenic species of mycobacteria (**Supplementary Table 1**).

The syntenic relationship of the two regulators prompted us to probe the regulation of the whiB5-Rv0023 locus. To identify the regulatory elements at this locus, upstream regions of whiB5 and Rv0023 were cloned in the promoter-less vector pEJ414 and named pEJwhiB5 and pEJ0023, respectively. The coding region of Rv0023 was cloned in the pVV16 vector to obtain pVV0023. Msmeg was transformed with pEJwhiB5 or pEJ0023. The strains were again transformed with pVV023

fmicb-11-00003 February 7, 2020 Time: 15:18 # 8

to obtain MsmegpEJwhiB5-pVV0023 and MsmegpEJ0023 pVV0023 double transformants. Promoter activities were measured in double transformants (MsmegpEJwhiB5-pVV0023, MsmegpEJwhiB5-pVV16, MsmegpEJ0023-pVV0023, and MsmegpEJ0023-pVV16). Overexpression of Rv0023 caused approximately fivefold decrease in whiB5 promoter activity (**Figure 3C**), whereas no significant change in the promoter activity of Rv0023 was noted (**Figure 3D**). Rv0494, a FadR transcriptional regulator, was used as a negative control, which did not affect the promoter activities of any of these genes. These data suggest that Rv0023 negatively regulates the expression of whiB5, but it is not auto-regulatory.

## Rv0023 Binds to whiB5 in vitro

As we have observed that Rv0023 expression represses the promoter activity of whiB5 in Msmeg, it prompted us to further test whether Rv0023 interacts with whiB5 promoter or not. Five hundred base pairs upstream and 50 bp downstream of whiB5 gene containing the promoter were radiolabeled and EMSA was performed with the recombinant Rv0023.

(8 pmol) protein.

#### TABLE 7 | List of overlapping fragments used in Figure 4.

fmicb-11-00003 February 7, 2020 Time: 15:18 # 9


The radiolabeled upstream region of whiB5 interacted with the purified recombinant Rv0023 in a dose-dependent manner. No interaction of Rv0023 was observed with non-specific DNA (**Figure 4A**).

Both EMSA and β-galactosidase assays confirmed that Rv0023 interacts with whiB5 upstream and negatively regulates whiB5 expression. In order to find the exact binding site of Rv0023, the 550 bp upstream of whiB5 was divided into two fragments: UF1 of 340 bp and UF2 of 250 bp (**Figure 4B**). We observed that only the UF2 fragment binds to Rv0023 (**Supplementary Figure 3**). The UF2 fragment was further divided into six overlapping fragments, F1–F6 (**Figure 4B** and **Table 7**). Purified Rv0023 binds to fragments F1 and F2, whereas no binding was observed with other fragments (**Figure 4C**). So, we looked for a binding site at the overlapping region of F1 and F2. We generated new oligonucleotides (OL1–OL5) covering the F1–F2 overlapping region (**Figure 5A** and **Table 8**). EMSA with these radiolabeled fragments showed that Rv0023 binds to OL1 and OL3 fragments whereas no binding was observed with other fragments (**Figure 5B**). Close examination of these fragments revealed the presence of an imperfect palindrome present in both sequences, TATAcgtTTTT in OL1 and TATAcgcTTTT in OL3 (**Figure 5C**). To confirm the binding sites identified, we mutated TATAcgtTTTT to TGTGcgtTGTG in OL1 and TATAcgcTTTT to TGTGcgcTGTG in OL3 to obtain OL1M and OL3M, respectively (**Table 8**). Binding studies have shown that Rv0023 binds to the native OL1 and OL3 fragments, but not to the mutant fragments, OL1M and OL3M (**Figure 5C**). We conclude that there are two binding sites (TATAcgtTTTT and TATAcgcTTTT) of Rv0023 in the upstream region of whiB5 comprising an 11-bp imperfect palindromic sequence with a 3-bp spacer region.

## Analysis of Rv0023 Binding Site

The importance of each residue within the Rv0023 binding site was determined by generating 39 bp probes (M0–M12)


fmicb-11-00003 February 7, 2020 Time: 15:18 # 10


containing specific point mutations on each half of the imperfect palindrome and in the spacer region (**Figure 6A**). The probes were radiolabeled and were used in binding studies with recombinant Rv0023. The binding was observed with probes M0 (unaltered), M1, M4, and M5, and significant loss of binding was observed with M2, M3, M6, M7, and M8 fragments, suggesting the importance of these residues in Rv0023 binding. Changing the residues in the spacer region (M9–M10) did not affect Rv0023 binding; however, changing the length of spacer region by one base lead to significant loss of Rv0023 binding (M11–M12) (**Figure 6B**). As mentioned earlier, the organization of Rv0023 and whiB5 is conserved across pathogenic mycobacterial species. Therefore, we looked into the conservation of identified Rv0023 binding sites across the conserved genomes. For this, we aligned the upstream region of whiB5 orthologs from different pathogenic mycobacterial species strains and found that the two binding sites of Rv0023 are conserved across the analyzed mycobacterial species (**Figure 6C**). These results highlight the important residues for Rv0023 binding and its conservation across mycobacterial species.

## Rv0023 Binding Site Overlaps With the whiB5 Promoter Region

In order to understand the mechanism of Rv0023-mediated repression of whiB5, the TSSs of whiB5 and Rv0023 were mapped. The putative −10 region of whiB5 was identified earlier by RACE (Casonato et al., 2012) (**Figure 7A**). To test the functionality of the putative −10 region, we cloned the 250 bp upstream of whiB5 into lacZ reporter vector (pEJwhiB5WT). The promoter activity was measured using the

FIGURE 6 | Identification of important residues in the imperfect palindrome for Rv0023 binding. (A) Representation of probes (∼40 bp each) containing native binding site M0 and probe with mutation at specific positions in the binding site M1–M8 and probe with mutation in spacer regions M9–M12. (B) EMSA with radiolabeled probes [native (M0), mutated (M1–M9), and altered (M9–M12)] with constant concentration of purified Rv0023 protein. Fragments with (+) and without (–) Rv0023 (8 pmol) protein. (C) Multiple sequence alignment of whiB5 upstream from selected members of mycobacterial species. The two binding sites of Rv0023 are highlighted in rectangular boxes.

fmicb-11-00003 February 7, 2020 Time: 15:18 # 11

β-galactosidase assay in Msmeg. Mutations in the putative −10 region of whiB5 (atacgctt to gcacgcgg) abolished the promoter activity, confirming the −10 region of whiB5 (**Figure 7B**). To identify the promoter region of Rv0023, primer extension was performed. We identified the TSS at 228 bp upstream of Rv0023 start codon (**Figure 7C**). A −10-like hexamer sequence TCATAG was identified upstream of TSS. To validate the functionality of the −10 region of Rv0023, 400 bp upstream of Rv0023 was cloned into a lacZ reporter vector (pEJ0023WT). β-Galactosidase assay with WT promoter and mutated promoter (pEJ0023MUT) indicated significant decrease in the promoter activity of the mutant strain compared to WT (**Figure 7D**). Mapping the promoters of whiB5 and Rv0023 with Rv0023 binding site revealed that Rv0023 binding site overlaps with the promoter of whiB5 but not of the Rv0023 promoter region (**Figure 7E**).

## DISCUSSION

fmicb-11-00003 February 7, 2020 Time: 15:18 # 12

In this study, Rv0023, a member of the XRE family of transcriptional regulators, has been characterized, and its role in INH and ETH drug tolerance has been explored. The first part of our study concerns the role of Rv0023 in the physiology of mycobacteria. For our study, we used Msmeg as a surrogate model to study the effects of various stress conditions on wild type and Rv0023-expressing strains. It was observed that overexpression of Rv0023 confers higher tolerance toward INH and ETH in Msmeg. It was seen that the ectopic expression of Rv0023 alters the NADH/NAD<sup>+</sup> ratio, thereby increasing drug tolerance. In the second part of the manuscript, we identified Rv0023 as a transcriptional regulator and studied the regulatory effect at whiB5-Rv0023 locus and observed that Rv0023 negatively regulates whiB5 expression but is not auto-regulatory. Rv0023 regulates whiB5 expression by binding to specific sequences in the upstream region of whiB5. Our data indicate that there are two binding sites for Rv0023 upstream of whiB5 and one of the sites overlaps with the whiB5 promoter, thereby possibly occluding the binding of RNA polymerase. The binding site of Rv0023 is conserved across mycobacterial species as confirmed by multiple sequence alignment. Apart from this, we have also characterized the binding site of Rv0023 and found important bases for binding. While analyzing the spacer region, it was found that the length of the spacer region is important for proper binding.

Isoniazid, isonicotinic acid hydrazide, is a synthetic drug and the occurrence of INH-resistant strains is significantly more frequent than other drug-resistant Mtb clinical strains (Nachega and Chaisson, 2003). ETH, 2-ethylthioisonicotinamide, is a structural analog of INH and was first synthesized in 1956 (Grumbach et al., 1956). Many genes have been found to be associated with INH and ETH resistance in clinical isolates, but the role of transcription factors regulating INH and ETH co-resistance is poorly understood. Rv0023, a transcriptional regulator, conferring higher tolerance toward INH and ETH in Msmeg, provided us a chance to explore the role of transcription factors in INH and ETH co-resistance in Mtb. To elucidate the mechanism by which Rv0023 confers drug tolerance, we investigated two main genes known to be involved in INH and ETH co-resistance, inhA (enzyme involved in the synthesis of cell wall mycolic acid) (Banerjee et al., 1994) and ndh (NADH dehydrogenase maintains the NADH/NAD + ratio) (Miesel et al., 1998). We checked their expression levels in the vector control and Rv0023-expressing Msmeg strains and differential expression was only observed in the ndh gene. The ndh gene oxidizes NADH to NAD+, which maintains the ratio of NADH/NAD<sup>+</sup> in mycobacterial cells (Jaworowski et al., 1981). Loss of ndh function leads to higher levels of NADH in the cytosol. Previous studies have shown that the higher NADH/NAD<sup>+</sup> levels interfere with activation of INH/ETH drugs by competitive inhibition with the formation of INH-NAD and ETH-NAD and thus confer resistance to INH/ETH in Msmeg and BCG (Miesel et al., 1998; Vilcheze et al., 2005). So, we further measured the NADH/NAD<sup>+</sup> ratio in vector control and Rv0023-expressing Msmeg strains. The NADH/NAD<sup>+</sup> ratio was found to be increased in the Rv0023-overexpressed Msmeg strain, suggesting that negative regulation of ndh gene by Rv0023 is probably the mechanism by which Rv0023 confers INH and ETH tolerance in Msmeg. Our results were supported by previous studies where it was shown that Rv0023 regulon is enriched for NAD reductases (Rustad et al., 2014). Rv0023 did not bind to the upstream region of ndh, indicating that the regulation occurs possibly through indirect means, which may involve multiple intermediate gene products. Further studies are required to completely decipher this mechanism.

Synteny analysis has shown that whiB5-Rv0023 locus is present mainly in pathogenic species of mycobacteria. The gene whiB5 belongs to the WhiB family of transcriptional regulators, which are exclusive to actinomycetes, such as Mycobacterium and Streptomyces spp. (Soliveri et al., 2000). Earlier studies have shown that whiB5 is a global transcriptional regulator, regulating 58 genes of diverse functions, including sigM and genes encoding for Type VII secretion systems such as esx-2 and esx-4. It was shown that whiB5 has a role in Mtb virulence and reactivation (Casonato et al., 2012). As the whole locus of whiB5-Rv0023- Rv0024 is absent in Msmeg, it is difficult to gauge the true effect of Rv0023 in the physiology of Mtb. Further investigation in Mtb may reveal important aspects of Rv0023 regulation and its effects. Hence, taking into account the earlier studies and our current data, we surmise that Rv0023 plays a significant role in modulating genes associated with NADH/NAD<sup>+</sup> levels, which further influences mycobacterial physiology.

In conclusion, our work shows, for the first time, that Rv0023 has a role in conferring INH and ETH tolerance in Msmeg. Understanding the role of transcriptional factors in the development of drug resistance will open new avenues in the field of drug discovery and may provide important insights into Mtb physiology.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## AUTHOR CONTRIBUTIONS

AR, SG, RA, and SY conceived the hypothesis and rationale of the study, analyzed the results, and wrote the manuscript. SG performed all the experiments. CR contributed in cloning of constructs and recombinant protein purification.

## FUNDING

SG was supported by a Centre for DNA Fingerprinting and Diagnostics fellowship. RA was supported earlier by a Department of Biotechnology fellowship. SY was supported earlier by a Council of Scientific and Industrial Research fellowship at the Computational and Functional Genomics Group. CR was supported by a Centre for DNA Fingerprinting and Diagnostics fellowship.

## ACKNOWLEDGMENTS

fmicb-11-00003 February 7, 2020 Time: 15:18 # 13

We thank Dr. Roger Buxton (MRC, London, United Kingdom) for pEJ414 plasmid. Research in the AR laboratory was supported

## REFERENCES


by the Centre for DNA Fingerprinting and Diagnostics and Department of Biotechnology, Government of India.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00003/full#supplementary-material

mutations and evolution of multidrug-resistant Mycobacterium tuberculosis. Antimicrob. Agents Chemother. 50, 2640–2649. doi: 10.1128/aac.00112-06


fmicb-11-00003 February 7, 2020 Time: 15:18 # 14

DNA damage in Mycobacterium tuberculosis. Microbiology 147, 3271–3279. doi: 10.1099/00221287-147-12-3271


monooxygenase. J. Biol. Chem. 277, 12824–12829. doi: 10.1074/jbc. m110751200


**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 Gupta, Angara, Yousuf, Reddy and Ranjan. 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.

# Radezolid Is More Effective Than Linezolid Against Planktonic Cells and Inhibits Enterococcus faecalis Biofilm Formation

Jinxin Zheng1,2† , Zhong Chen1,2† , Zhiwei Lin<sup>2</sup>† , Xiang Sun1,2† , Bing Bai<sup>2</sup> , Guangjian Xu1,2 , Junwen Chen<sup>2</sup> , Zhijian Yu<sup>2</sup> \* and Di Qu<sup>1</sup> \*

<sup>1</sup> Key Laboratory of Medical Molecular Virology of Ministries of Education and Health, School of Basic Medical Science and Institutes of Biomedical Sciences, Shanghai Medical College of Fudan University, Shanghai, China, <sup>2</sup> Department of Infectious Diseases and the Key Laboratory of Endogenous Infection, Shenzhen Nanshan People's Hospital, The 6th Affiliated Hospital of Shenzhen University Health Science Center, Shenzhen, China

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### Reviewed by:

Lara Mendes Almeida, Federal University of Alagoas, Brazil Ali Al-Ahmad, Freiburg University Medical Center, Germany Daria Van Tyne, University of Pittsburgh, United States

#### \*Correspondence:

Zhijian Yu yuzhijiansmu@163.com Di Qu dqu@fudan.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 20 August 2019 Accepted: 28 January 2020 Published: 14 February 2020

#### Citation:

Zheng J, Chen Z, Lin Z, Sun X, Bai B, Xu G, Chen J, Yu Z and Qu D (2020) Radezolid Is More Effective Than Linezolid Against Planktonic Cells and Inhibits Enterococcus faecalis Biofilm Formation. Front. Microbiol. 11:196. doi: 10.3389/fmicb.2020.00196 The aim of this study was to compare the effects of radezolid and linezolid on planktonic and biofilm cells of Enterococcus faecalis. A total of 302 E. faecalis clinical isolates were collected, and the minimum inhibitory concentrations (MICs) of radezolid and linezolid were determined by the agar dilution method. Changes in the transcriptome of a highlevel, in vitro-induced linezolid-resistant isolate were assessed by RNA sequencing and RT-qPCR, and the roles of efflux pump-related genes were confirmed by overexpression analysis. Biofilm biomass was evaluated by crystal violet staining and the adherent cells in the biofilms were quantified according to CFU numbers. The MIC50/MIC<sup>90</sup> values of radezolid (0.25/0.50 mg/L) against the 302 E. faecalis clinical isolates were eightfold lower than those of linezolid (2/4 mg/L). The radezolid MICs against the highlevel linezolid-resistant isolates (linezolid MICs ≥ 64 mg/L) increased to ≥ 4 mg/L with mutations in the four copies of the V domain of the 23S rRNA gene. The mRNA expression level of OG1RF\_12220 (mdlB2, multidrug ABC superfamily ATP-binding cassette transporter) increased in the high-level linezolid-resistant isolates, and radezolid and linezolid MICs against the linezolid-sensitive isolate increased with overexpression of OG1RF\_12220. Radezolid (at 1/4 or 1/8× the MIC) inhibited E. faecalis biofilm formation to a greater extent than linezolid, which was primarily achieved through the inhibition of ahrC, esp, relA, and relQ transcription in E. faecalis. In conclusion, radezolid is more effective than linezolid against planktonic E. faecalis cells and inhibits biofilm formation by this bacterium.

Keywords: Enterococcus faecalis, radezolid, linezolid, resistance, biofilm, efflux pump

## INTRODUCTION

Enterococcus faecalis is a prominent example of a human pathogen that rapidly evolves and becomes refractory to a wide range of antimicrobials. In addition to the intrinsic and acquired resistance to many individual antimicrobials, the spread of multidrug-resistant (MDR) enterococci, especially those resistant to vancomycin (VRE), has further narrowed the choices for anti-infective

therapy (Ahmed and Baptiste, 2018). Linezolid (LZD), an important member of the oxazolidinone class of antibiotics, has proven to be highly effective against most gram-positive bacteria and is recommended as the first-line choice for the remedial treatment of VRE and other MDR enterococci infections (Whang et al., 2013). However, widespread LZD application has led to the rapid, global emergence of LZD-resistant clinical isolates, including Staphylococcus aureus, Staphylococcus


TABLE 1 | The distribution of radezolid and linezolid minimum inhibitory concentrations (MICs) in 302 Enterococcus faecalis clinical isolates.

RZD, radezolid; LZD, linezolid; S, sensitive; I, intermediate; R, resistant.

TABLE 2 | High-level linezolid resistance leads to the decreased sensitivity of Enterococcus faecalis to radezolid.


RZD, radezolid; LZD, linezolid; MIC, minimum inhibitory concentration.

epidermidis, E. faecalis, E. faecium, Mycobacterium tuberculosis, and Mycobacterium abscessus (Balandin et al., 2016; Zimenkov et al., 2017; Chen et al., 2018; Silva et al., 2019; Ye et al., 2019). The consequent renewed interest in the optimization of oxazolidinones led to the development of new antimicrobials such as radezolid (RZD, RX-1741) (Lemaire et al., 2010b), which showed greater potency than LZD against a broad range of grampositive bacteria, including VRE (Lemaire et al., 2010a; Wu et al., 2018, 2019). However, whether RZD is also effective against linezolid-resistant E. faecalis isolates remains unclear.

Numerous studies have demonstrated that LZD resistance is associated with mutations in domain V of the 23S rRNA gene and L3 and L4 ribosomal proteins, as well as with the acquisition of the cfr, cfr(B), or optrA genes (Sadowy, 2018). Recently, the ABC-F subfamily ATP-binding cassette protein PoxtA was also found to play a role in the decreased susceptibility of S. aureus and E. faecalis to oxazolidinones (Antonelli et al., 2018; Elghaieb et al., 2019; Hasman et al., 2019; Lei et al., 2019). Nevertheless, the extent to which RZD exerts enhanced antibacterial activity against E. faecalis when compared with LZD is still not known. Additionally, LZD has been reported to have good inhibitory effects on E. faecalis biofilms (Holmberg et al., 2012); however, it is also unclear whether RZD shows greater efficacy than LZD against E. faecalis biofilms. To address these questions, in this study, we compared the antibacterial effects of RZD and LZD against biofilm and planktonic cells of E. faecalis.

## MATERIALS AND METHODS

## Bacterial Strains and Antimicrobials

A total of 302 non-duplicate E. faecalis isolates were collected from different inpatients at Shenzhen Nanshan People's Hospital (Grade A, level III Hospital, 1500 beds), Shenzhen University, China, between January 1, 2011, and December 31, 2016. These E. faecalis isolates were obtained from urine (135 isolates), blood (37 isolates), pus or secretions (86 isolates), bile (25 isolates), and other clinical sources (19 isolates). Based on a previous study, the dominant multilocus sequence types (MLSTs) of these isolates were ST16 and ST179 (Zheng et al., 2017). The isolates were identified by the Phoenix 100 automated microbiology system (BD, Franklin Lakes, NJ, United States), following which two subcultured generations of all the 302 isolates were re-identified with matrix-assisted laser desorption ionization time-of-flight mass spectrometry (IVD MALDI Biotyper, Bruker, Bremen, Germany). E. faecalis strains ATCC29212 and OG1RF (ATCC47077) were used as reference strains.

Chloramphenicol (catalog no. HY-B0239), linezolid (catalog no. HY-10394), and radezolid (catalog no. HY-14800) were purchased from MedChemExpress (MCE, Shanghai, China).

## Antimicrobial Susceptibility Test and Detection of LZD Resistance Genes

The susceptibilities of E. faecalis isolates to clinically relevant antimicrobials were tested by the Phoenix 100 automated microbiology system. The minimum inhibitory concentrations (MICs) of LZD and RZD were determined by the agar dilution method according to Clinical and Laboratory Standards Institute (CLSI) guidelines. While no CLSI interpretive criteria existed for RZD against enterococci, the MICs of RZD for the quality control strain ATCC29212 were observed to range from 0.06 to 0.5 mg/L. Therefore, to analyze the distribution of MICs for RZD in these clinical isolates, the MICs were categorized into the following four levels: ≤ 0.125, 0.25, 0.5, and ≥ 1 mg/L. The four copies of the V domain of the 23S rRNA gene, as well as the rplC and rplD genes, were amplified by PCR and sequenced. The cfr, cfr(B), optrA, and poxtA genes were also amplified by PCR. The primers used for PCR are listed in **Supplementary Table S1**.

## In vitro Induction of High-Level LZD-Resistant Isolates and Efflux Inhibition Assay

The ATCC29212-0 and OG1RF (ATCC47077)-0 strains (LZDsensitive, linezolid MIC: 2 mg/L) were serially subcultured in Mueller–Hinton Broth (MHB) containing LZD. The initial inducing concentration of LZD was 0.5× the MIC, which was then successively increased to 1, 2, 4, 8, 16, 32, 64, and 128× the MIC. Strains were cultured at each concentration for 3– 5 passages before their exposure to the next concentration. Isolates from the final passage of each concentration were identified by matrix-assisted laser desorption ionization time-offlight mass spectrometry (IVD MALDI Biotyper, Bruker, Bremen, Germany), and the MICs of RZD and LZD were determined by the agar dilution method according to CLSI guidelines.

The efflux pump activities in LZD-resistant isolates were detected using the efflux pump inhibitor Phe-Argβ-naphthylamide (PAβN, Sigma, Shanghai, China). MICs for RZD and LZD were determined in the presence or absence of

TABLE 3 | Radezolid and linezolid minimum inhibitory concentrations (MICs) decreased in the presence of PAβN.


RZD, radezolid; LZD, linezolid; PAβN, Phe-Arg-β-naphthylamide (20 mg/L).

PAβN (20 mg/L) (Kothary et al., 2013). This assay was performed at least in triplicate.

## RNA Isolation and Sequencing

fmicb-11-00196 February 12, 2020 Time: 18:1 # 4

Total RNA isolation and RNA sequencing (RNA-seq) of the E. faecalis OG1RF wild-type isolate (OG1RF-0, LZDand RZD-sensitive) and the high-level LZD-resistant isolate OG1RF-55 (linezolid MIC: 256 mg/L; radezolid MIC: 8 mg/L) were performed as previously described (Wang et al., 2017). Briefly, planktonic E. faecalis cells were homogenized using 0.1-mm zirconia–silica beads in a mini-BeadBeater and the total RNA in the supernatant was purified using an RNeasy Mini Kit (Qiagen, Hilden, Germany). RNA-seq was performed according to the Illumina RNA sequencing sample preparation guide. Total RNA samples were treated with RNase-free DNase I (TaKaRa Biotechnology, Dalian, China). cDNA libraries were prepared using an RNA-seq sample preparation kit (Illumina, San Diego, CA, United States), and sequencing was performed with an Illumina HiSeq 2500 sequencer for 50 cycles. All these procedures were performed according to the manufacturers' protocols. Raw sequencing data were processed using the data collection software provided by Illumina. RNA-seq was performed in three independent experiments.

## RNA-Seq Data Analysis and RT-qPCR

Raw sequencing reads were preprocessed by filtering out rRNA reads, sequencing adapters, short fragment reads, and other low-quality reads. The remaining reads were mapped to the E. faecalis OG1RF reference genome (CP002621.1) at the National Center for Biotechnology Information (NCBI) website using Bowtie2 software (version 2.0.5) based on the local alignment algorithm. The alignments reported using Bowtie2 software were further processed with BED Tools software to determine transcript expression levels and their differential expression between each two of the three samples. Differential expression of all the transcripts was quantified using DEGseq software (version 2.16.1), and then the fold-change values were presented. For validation of the RNA-seq results, RT-qPCR was performed using the SYBR Premix Ex Taq II Kit (TaKaRa Biotechnology, Dalian, China) on a Mastercycler ep realplex system (Eppendorf, Hamburg, Germany) as previously described (Zheng et al., 2017). The primers used for RT-qPCR are listed in **Supplementary Table S2**. RT-qPCR was performed in triplicate at least three times.

## Overexpression of Efflux Pump-Related Genes in E. faecalis

The OG1RF\_12220 gene (multidrug ABC superfamily ATPbinding cassette transporter, mdlB2) and three ABC superfamily ATP-binding cassette transporter genes (OG1RF\_10126, OG1RF\_10665, and OG1RF\_10495) were amplified by PCR. The amplicons were purified and digested with endonucleases, and then cloned into the pIB166 plasmid for gene overexpression. Correct cloning was verified by PCR

TABLE 4 | Differential RNA levels of efflux pump-related genes in a high-level linezolid-resistant isolate.


The Enterococcus faecalis OG1RF wild-type isolate (OG1RF-0, linezolid- and radezolid-sensitive) and the high-level linezolid-resistant isolate (OG1RF-55, linezolid MIC: 256 mg/L; radezolid MIC: 8 mg/L) were inoculated and grown to the logarithmic phase (4 h), total RNA was isolated and sequenced by Illumina HiSeq 2500 sequencer. Data are the means of the results from three independent experiments.

and sequencing. Verified loaded plasmids were introduced into the E. faecalis OG1RF strain. All strains, plasmids, and primers used for overexpression analysis are listed in **Supplementary Tables S3, S4**.

## Biofilm Biomass Assay and Adherent Cell Detection

The biofilm biomass of 13 E. faecalis clinical isolates (16C1, 16C35, 16C51, 16C102, 16C106, 16C124, 16C138, 16C152, 16C166, 16C201, 16C289, 16C350, and 16C353) was detected by crystal violet staining as previously described (Zheng et al., 2019). For analysis of the eradication potential of RZD or LZD against established E. faecalis biofilms, E. faecalis isolates were inoculated into 96-well polystyrene microtiter plates with TSBG (tryptic soy broth with 0.25% glucose) for formation of mature biofilms. After 24 h of static incubation, the supernatants were discarded and plates were washed with 0.9% saline to remove unattached cells, following which fresh TSBG containing RZD or LZD was added. After 48 h of static incubation, with the medium replaced daily,

FIGURE 1 | Overexpression of OG1RF\_12220, OG1RF\_10126, OG1RF\_10665, and OG1RF\_10495 in the Enterococcus faecalis OG1RF strain. The RNA levels of the four genes were determined by RT-qPCR. The OG1RF wild-type isolate was used as the reference strain (mRNA level = 1.0). The OG1RF isolate containing the empty pIB166 vector (OG1RF-pIB166) was used as control.

the remaining biofilm biomass was determined by crystal violet staining. The numbers of adherent cells remaining in biofilms formed in 24-well polystyrene microtiter plates were determined

TABLE 5 | Radezolid and linezolid minimum inhibitory concentrations (MICs) increased with OG1RF\_12220 overexpression in the Enterococcus faecalis OG1RF strain.


by counting the number of colony-forming units (CFUs), as previously described (Zheng et al., 2019). To investigate whether RZD or LZD could inhibit E. faecalis biofilm formation, E. faecalis isolates were inoculated into 96-well polystyrene microtiter plates with TSBG containing RZD or LZD (at sub-MICs). After 24 h of static incubation, biofilm biomass was determined by crystal violet staining. Each assay was performed in triplicate at least three times.

## RT-qPCR to Determine the RNA Levels of E. faecalis Biofilm Formation-Related Genes

The RNA levels of 16 biofilm formation-related genes of the above 13 E. faecalis clinical isolates were determined by RTqPCR based on published reports (Tendolkar et al., 2005;

Nallapareddy et al., 2006; Guiton et al., 2009; Chavez de Paz et al., 2012; Soares et al., 2014; Dale et al., 2015; Frank et al., 2015; Akbari Aghdam et al., 2017; Zheng et al., 2018). The E. faecalis clinical isolates were inoculated into 100 mm × 20 mm nonpyrogenic polystyrene cell culture dishes with TSBG containing RZD or LZD (at 1/4× the MIC). After 6, 12, or 24 h of static incubation, total RNA was extracted from planktonic and biofilm E. faecalis cells for RT-qPCR. The primers used for RT-qPCR are listed in **Supplementary Table S2**. Each assay was performed in triplicate at least three times.

## Statistical Analysis

fmicb-11-00196 February 12, 2020 Time: 18:1 # 7

The data were analyzed using the Student's t-test. P-values < 0.05 were considered significant. All data were analyzed in SPSS version 16.0 (SPSS, Inc., Chicago, IL, United States).

## RESULTS

## RZD Was More Effective Than LZD Against E. faecalis Clinical Isolates

The distribution of MICs for RZD and LZD and their relationship with the susceptibilities for some other conventional antimicrobials are shown in **Table 1**. The data show that the MIC50/MIC<sup>90</sup> of RZD was eightfold lower than that of LZD, suggesting that the in vitro activity of RZD against the 302 E. faecalis clinical isolates was substantially higher than that of LZD. Moreover, the MICs of RZD against the 52 LZDnon-susceptible E. faecalis clinical isolates were also lower (**Supplementary Table S5**). Twenty-one LZD-non-susceptible clinical isolates presented mutations in domain V of the 23S rRNA gene, but only four isolates contained the optrA gene. As

FIGURE 3 | Planktonic growth of E. faecalis 16C106 and 16C350 isolates with radezolid (RZD) or linezolid (LZD) treatment. The growth of planktonic E. faecalis 16C106 cells treated with LZD (A) or RZD (B) and that of planktonic 16C350 cells treated with LZD (C) or RZD (D) was determined by measurement of the optical density at 600 nm (OD600). Data represent the average of three independent experiments (mean ± SD).

shown in **Supplementary Table S5**, only two of the four optrAcarrying isolates (16C112 and 16C154) harbored mutations in the V domain of the 23S rRNA gene. No mutations were detected in ribosomes L3 and L4, or in the cfr, cfr(B), poxtA genes in these isolates (data not shown).

## Effects of RZD on High-Level LZD-Resistant E. faecalis Isolates

To evaluate the effect of RZD on the high-level LZD-resistant E. faecalis isolates and explore the extent to which RZD overcame LZD-related resistance mechanisms, high-level resistance to LZD was induced in the E. faecalis ATCC29212 and OG1RF strains. **Table 2** shows the primary mutation loci in the 23S rRNA gene in the high-level LZD-resistant G2576U isolates. Mutations in ribosomes L3 and L4, and in the cfr, cfr(B), optrA, and poxtA genes were not detected in these isolates (data not shown). The

## MICs for RZD increased sharply to ≥ 4 mg/L if the isolates presented with LZD-induced mutations in the four copies of domain V of the 23S rRNA gene (linezolid MICs ≥ 64 mg/L).

## Efflux Pumps Are Involved in E. faecalis Resistance to RZD and LZD

The MICs for LZD or RZD showed a two- to eight-fold decrease in the presence of PAβN (**Table 3**). Thus, to explore the role of efflux pumps in RZD and LZD resistance, we assessed the changes occurring in the transcriptome of the high-level LZD-resistant OG1RF isolate (OG1RF-55, linezolid MIC: 256 mg/L; radezolid MIC: 8 mg/L) by RNA-seq (**Supplementary Figure S1** and **Supplementary Table S6**). Subsequently, we evaluated the RNA expression levels of efflux pump-related genes by RT-qPCR, and found that the transcriptional levels of four ABC superfamily ATP-binding cassette transporter

genes (OG1RF\_12220, OG1RF\_10126, OG1RF\_10665, and OG1RF\_10495) were increased in the high-level LZD-resistant OG1RF-55 isolate (**Table 4**). Finally, to confirm that the four genes were involved in the resistance to RZD and LZD, we overexpressed these genes in the LZD- and RZD-sensitive OG1RF isolate (**Figure 1** and **Supplementary Tables S3**, **S4**). As indicated in **Table 5**, the MICs for LZD and RZD increased only with overexpression of OG1RF\_12220 (mdlB2).

## RZD Inhibited E. faecalis Biofilm Formation to a Greater Extent Than LZD

Thirteen E. faecalis clinical isolates (biofilm-positive) were selected to compare the differential effects of RZD and LZD on E. faecalis biofilms (**Supplementary Table S7**). First, we compared the eradicating potential of RZD and LZD (at 8× their MICs) on established E. faecalis biofilms, and found no difference between them (**Figure 2**). Both drugs also elicited similar effects on adherent cells of established biofilms. We also compared the sub-MICs at which RZD and LZD inhibited E. faecalis biofilm formation. Based on a previous study (Mlynek et al., 2016) and our preliminary results, 1/4, 1/8, 1/16, and 1/32× MICs were used in this study. The 1/2× MIC was not used because the planktonic growth of E. faecalis was markedly affected under this concentration of both RZD and LZD (**Figure 3**). As shown in **Figure 4**, the RZD at 1/4 or 1/8× its MIC efficiently inhibited E. faecalis biofilm formation and to a greater extent than LZD. This trend was also observed in several of the 13 clinical isolates at MICs of 1/16 or 1/32× (**Figure 5**).

## RZD Treatment Reduced the RNA Levels of Biofilm Formation-Related Genes

Isolates 16C106 and 16C350 were selected for RT-qPCR to evaluate the RNA expression levels of 16 E. faecalis biofilm formation-related genes at different stages of biofilm formation.

The RNA levels of ahrC, cylA, esp, relA, and relQ markedly decreased when the isolates were treated with RZD or LZD at 1/4× their MICs for 6 h (**Table 6**). The RNA levels of these 16 genes in the other 11 E. faecalis clinical isolates were also determined, and, as indicated in **Table 7**, the transcriptional levels of ahrC, esp, relA, and relQ showed a significant decrease, especially in isolates treated with RZD.

## DISCUSSION

Radezolid, a novel biaryl analog of LZD, exhibits excellent activity against gram-positive bacteria, including methicillin-resistant Staphylococcus aureus (MRSA) and LZD-resistant staphylococci (Lemaire et al., 2010a; Locke et al., 2010). The results of this study further indicate that RZD also exerts stronger effects than LZD against several E. faecalis clinical isolates. One study demonstrated that, in LZD-resistant S. aureus isolates (linezolid MICs ranging from 8 to 32 mg/L), the MICs of RZD against these isolates were two- to eight-fold lower (from 1 to 4 mg/L) than those of LZD (Locke et al., 2010). In this study, we also focused on LZD-non-susceptible E. faecalis clinical isolates (linezolid MICs ranging from 4 to 32 mg/L); however, we found that the MICs of RZD for these isolates (from 0.25 to 1 mg/L) were 8- to 32-fold lower than those of LZD. We also showed that when high-level resistance to LZD was induced in E. faecalis (linezolid MICs ≥ 64 mg/L), the MICs of RZD increased from 1 to 8 mg/L. This result indicated that cross-resistance to RZD and LZD was emerging in these isolates, and that this was mainly due to mutations in the four copies of domain V of the 23S rRNA gene.

The oxazolidinone class of antimicrobials, which includes LZD, exhibited excellent effects against gram-positive bacteria, but showed poor activity against gram-negative bacteria such as Escherichia coli, primarily due to the enhanced activity of AcrB, a RND-type efflux pump (Schumacher et al., 2007; Schuster et al., 2019). However, whether efflux pumps have a role in the resistance of gram-positive bacteria to LZD or RZD remains uncertain. In the present study, we found that the MICs of LZD and RZD decreased in the presence of PAβN, an efflux pump inhibitor, indicating that efflux pumps are involved in E. faecalis resistance to LZD and RZD. We also found that the enhanced activity of OG1RF-12220 (mdlB2) led to increases in the MICs of both LZD and RZD. The expression of the efflux pump gene mdlB was reported to be substantially increased among fluoroquinolone-resistant isolates of Salmonella enterica (Chen et al., 2007). Both mdlB and mdlB2 belong to the efflux transport system of the ATP-binding cassette (ABC) superfamily, which plays an important role in antimicrobial resistance (Iannelli et al., 2018). We found that mdlB2, but not mdlB, was involved in the resistance to LZD and RZD. However, this result requires further experimental confirmation, especially in high-level LZD-resistant E. faecalis clinical isolates.

Several studies have indicated that LZD affects the biofilms of E. faecalis, both when administered alone or in combination with rifampicin or gentamicin (Bayston et al., 2012; Holmberg et al., 2012; Luther et al., 2014). In this study, we found that RZD had a significantly greater effect than LZD on planktonic E. faecalis cells. We further investigated whether this was also true for E. faecalis biofilm formation, but we did


TABLE 6 | Changes in the RNA expression levels of biofilm formation-related genes of isolates 16C106 and 16C350 with radezolid or linezolid treatment.

Radezolid and linezolid were used at 1/4× their minimum inhibitory concentrations (MICs). RNA levels were detected by RT-qPCR, with untreated isolate as the reference strain (mRNA level = 1.0). −, not detected. <sup>a</sup>16C106: Radezolid (6 h) vs. Linezolid (6 h), P < 0.01; <sup>b</sup>16C106: Radezolid (6 h) vs. Linezolid (6 h), P < 0.05; <sup>c</sup>16C350: Radezolid (6 h) vs. Linezolid (6 h), P < 0.01; <sup>d</sup>16C350: Radezolid (6 h) vs. Linezolid (6 h), P < 0.05.

TABLE 7 | Changes in RNA expression levels of biofilm formation-related genes in 11 clinical isolates with radezolid or linezolid treatment.


RZD, radezolid; LZD, linezolid. Radezolid and linezolid were used at 1/4× their minimum inhibitory concentrations (MICs). The RNA levels were detected by RT-qPCR, with untreated isolate as the reference strain (mRNA level = 1.0). −, not detected. <sup>a</sup>ahrC: RZD vs. LZD, P < 0.01; <sup>b</sup>ahrC: RZD vs. LZD, P < 0.05; <sup>c</sup>cylA: RZD vs. LZD, P < 0.05; <sup>d</sup>esp: RZD vs. LZD, P < 0.05; <sup>e</sup>esp: RZD vs. LZD, P < 0.01; <sup>f</sup> relA: RZD vs. LZD, P < 0.05; <sup>g</sup> relA: RZD vs. LZD, P < 0.01; <sup>h</sup> relQ: RZD vs. LZD, P < 0.05; <sup>i</sup> relQ: RZD vs. LZD, P < 0.01.

not find any difference between RZD and LZD in eradicating established biofilms or adherent cells in the biofilms of E. faecalis. Interestingly, we found that RZD (at 1/4 or 1/8× the MIC) strongly inhibited E. faecalis biofilm formation, and more effectively than LZD. This result was similar to that of a previous study, in which Wu et al. (2014) found that another oxazolidinone, FYL-67, could more strongly inhibit S. aureus biofilm formation than LZD.

In the present study, we also explored the reasons for the RZD-mediated reduction in E. faecalis biofilm formation, and found that the mRNA levels of ahrC, esp, relA, and relQ were significantly decreased with RZD treatment. Several studies have demonstrated that the Esp virulence factor, which has been found to support E. faecalis cell adherence, colonization, and persistence in the urinary tract, also plays an important role in E. faecalis biofilm formation (Toledo-Arana et al., 2001; Tendolkar et al., 2004; Zheng et al., 2017). The ahrC gene, which encodes a transcriptional regulator of the ArgR family, was found to be critical for E. faecalis attachment to polystyrene in vitro, and porcine heart valve surfaces ex vivo, during the early stages of biofilm formation (Frank et al., 2013). The hydrolase RelA and the small alarmone synthetase RelQ, encoded by the relA and relQ genes, respectively, have also been found to control (p)ppGpp metabolism and sustain biofilm formation in E. faecalis (Chavez de Paz et al., 2012). Our study indicated that the transcriptional levels of ahrC, esp, relA, and relQ were greatly decreased when E. faecalis clinical isolates were treated with RZD for 6 h, resulting in a significant reduction in biofilm formation by E. faecalis.

## CONCLUSION

fmicb-11-00196 February 12, 2020 Time: 18:1 # 12

We showed that radezolid was more effective than linezolid against planktonic E. faecalis cells. Additionally, this study is the first to report that the mdlB2 gene is important for E. faecalis resistance to both RZD and LZD. We also found that RZD was more effective than LZD at inhibiting E. faecalis biofilm formation, which was mainly achieved through inhibition of the transcription of ahrC, esp, relA and relQ in E. faecalis.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the Sequence Read Archive (SRA) database under accession number PRJNA505107 (https://www.ncbi.nlm.nih.gov/bioproject/ PRJNA505107).

## ETHICS STATEMENT

All procedures involving human participants were performed in accordance with the ethical standards of Shenzhen University School of Medicine and with the 1964 Helsinki Declaration and its later amendments, and this study was approved by the Ethics Committee of the Shenzhen University School of Medicine. For this type of study, formal consent is not required.

## AUTHOR CONTRIBUTIONS

JZ designed the study, performed gene manipulation, analyzed and interpreted the RNA-seq data, performed the biofilm assay, and drafted the manuscript. ZC performed gene manipulation, MIC detection, mRNA extraction, and RNA-seq and RT-qPCR data analysis. ZL conducted the LZD in vitro induction, MIC detection, and analyses of RNA-seq and overexpression data. XS performed MIC detection, gene manipulation, RT-qPCR, biofilm assay, and bacteria counting in the biofilm assay. BB, GX, and JC performed MIC detection, LZD in vitro induction, RT-qPCR, biofilm assay, and bacteria counting in the biofilm assay. ZY and DQ designed the study, analyzed the data, and critically revised the manuscript for important intellectual content.

## REFERENCES


## FUNDING

This work was supported by grants from the National Science and Technology Major Projects for "Major New Drugs Innovation and Development" (grant number 2019ZX09721001); the Sanming Project of Medicine in Shenzhen (grant number SMGC201705029); Science, Technology and Innovation Commission of Shenzhen Municipality key funds (JCYJ20170412143551332 and JCYJ20180508162403996); and basic research funds (JCYJ20180302144721183 and JCYJ20180302144431923).

## ACKNOWLEDGMENTS

The authors thank Prof. Jingren Zhang (Center for Infectious Disease Research, School of Medicine, Tsinghua University, Beijing, China) for generously providing the pIB166 plasmid.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00196/full#supplementary-material

FIGURE S1 | The volcano plot of all the genes measured by RNA-seq.

TABLE S1 | PCR primers used for the detection of linezolid resistance genes.

TABLE S2 | RT-qPCR primers used for the detection of the RNA levels of the efflux pump-related genes and the biofilm formation-related genes.

TABLE S3 | Strains and plasmids used for the overexpression of OG1RF\_12220, OG1RF\_10126, OG1RF\_10665, and OG1RF\_10495 in E. faecalis OG1RF strain.

TABLE S4 | PCR primers used for the overexpression of OG1RF\_12220, OG1RF\_10126, OG1RF\_10665, and OG1RF\_10495 in E. faecalis OG1RF strain.

TABLE S5 | Radezolid against 52 linezolid-nonsusceptible E. faecalis clinical isolates.

TABLE S6 | Differential RNA levels of the genes in a high-level linezolid-resistant isolate.

TABLE S7 | Radezolid and linezolid minimum inhibitory concentrations (MICs)in 13 biofilm positive E. faecalis clinical isolates.

an intensive care unit. Infect. Dis. 48, 343–349. doi: 10.3109/23744235.2015. 1122225


antimicrobials. Antimicrob. Agents Chemother. 51, 535–542. doi: 10.1128/aac. 00600-06


aureus USA300 Biofilms. Antimicrob. Agents Chemother. 60, 2639–2651. doi: 10.1128/aac.02070-15


planktonic and adherent linezolid-resistant isolates of Enterococcus faecalis. J. Med. Microbiol. 68, 493–502. doi: 10.1099/jmm.0.000945


**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 Zheng, Chen, Lin, Sun, Bai, Xu, Chen, Yu and Qu. 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.

# Analysis of Identification Method for Bacterial Species and Antibiotic Resistance Genes Using Optical Data From DNA Oligomers

Ryan L. Wood<sup>1</sup> , Tanner Jensen<sup>2</sup> , Cindi Wadsworth<sup>2</sup> , Mark Clement<sup>2</sup> , Prashant Nagpal<sup>3</sup> and William G. Pitt<sup>1</sup> \*

<sup>1</sup> Chemical Engineering, Brigham Young University, Provo, UT, United States, <sup>2</sup> Computer Science, Brigham Young University, Provo, UT, United States, <sup>3</sup> Chemical and Biological Engineering, University of Colorado Boulder, Boulder, CO, United States

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Nabil Karah, Umeå University, Sweden Allison Hsiang, Ludwig Maximilian University of Munich, Germany Jason Sahl, Northern Arizona University, United States

> \*Correspondence: William G. Pitt pitt@byu.edu

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 09 September 2019 Accepted: 03 February 2020 Published: 20 February 2020

#### Citation:

Wood RL, Jensen T, Wadsworth C, Clement M, Nagpal P and Pitt WG (2020) Analysis of Identification Method for Bacterial Species and Antibiotic Resistance Genes Using Optical Data From DNA Oligomers. Front. Microbiol. 11:257. doi: 10.3389/fmicb.2020.00257 Bacterial antibiotic resistance is becoming a significant health threat, and rapid identification of antibiotic-resistant bacteria is essential to save lives and reduce the spread of antibiotic resistance. This paper analyzes the ability of machine learning algorithms (MLAs) to process data from a novel spectroscopic diagnostic device to identify antibiotic-resistant genes and bacterial species by comparison to available bacterial DNA sequences. Simulation results show that the algorithms attain from 92% accuracy (for genes) up to 99% accuracy (for species). This novel approach identifies genes and species by optically reading the percentage of A, C, G, T bases in 1000s of short 10-base DNA oligomers instead of relying on conventional DNA sequencing in which the sequence of bases in long oligomers provides genetic information. The identification algorithms are robust in the presence of simulated random genetic mutations and simulated random experimental errors. Thus, these algorithms can be used to identify bacterial species, to reveal antibiotic resistance genes, and to perform other genomic analyses. Some MLAs evaluated here are shown to be better than others at accurate gene identification and avoidance of false negative identification of antibiotic resistance.

Keywords: antibiotic resistance, machine learning, DNA sequencing, Raman spectroscopy, biomedical diagnostic

## INTRODUCTION

Novel DNA sequencing technologies have proliferated over the past two decades. Continual improvements in "next-generation sequencing" (NGS) and "third-generation sequencing" (TGS) have increased the fidelity and rate of sequencing, but it still takes hours or days to obtain complete sequences (van Dijk et al., 2018). Sequencing plays an essential role in biological classification, cell biology, forensic analysis, and gene manipulation for medical and research purposes. Furthermore, there are some diagnostic applications in which very rapid identification of a particular gene or genetic species becomes essential, while identification of all genes is not necessary. For example, in patients with septic shock from bacterial infections, identification of antibiotic-resistance genes is essential because the mortality rate increases 7.6% per hour of delay in administering correct

antibiotics (Kumar et al., 2006). Unfortunately, it takes more than 24 h to grow up the bacteria recovered from the blood of an infected patient, identify the species, and then determine to which antibiotics the organism is resistant, leading to a very high mortality rate for such infections (Kumar et al., 2009). Carbapenem resistance is one of the most concerning antibiotic resistances, as infections with carbapenem-resistant bacteria have a 48% mortality rate (Patel et al., 2008) caused in part by the good reluctance of physicians to initially prescribe carbapenem antibiotics without verifying resistance because of the severe side effects of carbapenems. The attending clinician wants to know the bacterial species and needs to know any resistance to antibiotics, with confidence that the diagnostic technique has a very low error rate of false negatives. As genome and plasmid sequencing can identify the species and previously identified resistance genes, it would be tremendously useful to perform bacterial sequencing in an hour or less. However, current and proposed NGS and TGS techniques still require much more time.

Herein we present a novel approach that is useful when the diagnostic objective is to rapidly identify the species of bacteria or the presence of an antibiotic resistance gene in the bacteria. Our approach employs a genomic analysis technique that has some data compression and data loss, but compensates by very rapid analysis of very short reads of DNA–sufficiently short length and suitably fast analysis that the species and resistance genes can be identified in about an hour. Such a process has been proposed and demonstrated for the identification of resistance genes in bacteria associated with bloodstream infections (Sagar et al., 2018; Korshoj and Nagpal, 2019). This technique employs a block optical sequencing (BOS) method using surface-enhanced Raman spectroscopy (SERS) to obtain a spectrum of short DNA oligomers of length k, called k-mers (Sagar et al., 2018; Korshoj and Nagpal, 2019). Because the Raman spectrum of each A, T, G and C base is known, the overall ATGC content of a single k-mer can be calculated by mathematical analysis of the k-mer spectrum. Sequence information is lost, but the base content– called block optical content (BOC)–is preserved. For example, the 10 bp DNA segment ATATGGCCTT would become a BOC datum of A2T4G2C2. For very rapid analysis, this BOC technique can be multiplexed by creating an array of 1000s of pyramidal peaks on a silicon wafer whose entire peak field can be imaged by a sensitive CCD camera. Using band-pass filters at discrete spectral windows, optical intensity at specific wavelengths can be obtained simultaneously from all peaks. Finally the optical spectra are processed to obtain the ATGC content of the DNA on each pyramid peak. The size of the tips is such that 10 bases, but not any longer length, fit within the SERS electromagnetic field "hot-spot" (Sagar et al., 2018). It is estimated that using a 1,000 × 1,000 array of SERS pyramids on a silicon wafer, 1,000,000 reads of DNA 10-mers can be done in about 100 s, using high-throughput Raman spectroscopy using quantum dot optical filters (Bao and Bawendi, 2015) and digital processing of the resulting spectra.

Such a technique is ideally suited for genomic identification of bacteria, as a typical bacterial genome is about 5,000,000 bp, or 10 Mbase of single-stranded DNA (ssDNA). Clipping this genome into 10-mer lengths would provide enough ssDNA from a single bacterium to cover the SERS pyramids on a 1,000 × 1,000 array. Bloodstream infections contain very low counts of bacteria, often on the order of 10 colony forming units (CFU) per mL of blood. Thus a 10 mL sample of blood would provide 100-fold more DNA than needed to place a 10-mer on each pyramid. Bacteria (and their DNA) can be collected from blood in minutes (Pitt et al., 2016; Alizadeh et al., 2017; Pitt et al., 2019), and the SERS analysis can commence immediately, followed by computations for identification of species and antibiotic resistance.

Compared to species identification, analysis of resistance genes on plasmids is more challenging since those genes are usually contained within 500 to 1,000 bases while an average plasmid is roughly 200,000 bases in length. Thus, there is a much smaller signal to background ratio, making detection of resistance genes more difficult. Nevertheless, we show herein that our technique can still identify specific genes.

While such technology seems promising, a working device with many 1000s of pyramid tips is still in development, thus leaving some experimental questions unanswered for now. While we wait, however, many of the theoretical questions can be answered, the largest of which is whether the loss of sequence data (the BOC reads give only content, not sequence) will make it difficult to uniquely identify a bacterial species or state unequivocally whether a known resistance gene is present. Another theoretical question is whether random mutations in the bacterial genomes will compromise correct identification, or whether random noise from the experimental optical measurements will reduce accuracy.

The main goal of this study was to determine whether data produced in 10-mer blocks could be used in a diagnostic device, meaning that the data could correctly identify species and antibiotic resistance genes with a realistic number of pyramid tips (≤1,000,000 tips). To answer this question, this study addressed four main objectives: (1) to determine how many BOC reads are needed for species identification; (2) to determine how many BOC reads are needed for single gene detections (such as an antibiotic resistance gene of around 800 bp); (3) to analyze how accuracy is affected by noise from the detecting instrument and from random gene mutations; (4) to identify which learning algorithms are best at accurately identifying the species and genes.

The present study answers these and other questions to show that only 10<sup>4</sup> BOC reads, even in the presence of mutations or experimental noise, are sufficient to identify bacterial species. The presence of resistance genes can be identified with an accuracy of 80% up to 95% using 10<sup>5</sup> BOC reads. Both of these results are well below the proposed 10<sup>6</sup> BOC reads, even when large error is present, showing that the algorithm could be used with high accuracy in a diagnostic device.

## MATERIALS AND METHODS

**Figure 1** provides an overview of the process for simulating the experimental data reads, converting these into a spectrum, and testing the machine learning algorithms (MLAs). A brief outline follows: a sequence is obtained (see section "DNA Sequences"); the sequence is broken into every possible 10-mer

for both strands of DNA (see section "Generating Sequence-Specific FBC Spectrum"); the 10-mers are binned according to the percent A, T, G, and C, resulting in the sequencespecific fractional base content (FBC) spectrum (see section "Generating Sequence-Specific FBC Spectrum"); experimental noise is simulated by introducing random errors into the sequence-specific FBC spectrum resulting in the simulated experimental FBC spectrum (see section "Simulating Gene Mutations and Experimental Errors"); the spectrum from a purely random sequence (bias) is subtracted from the simulated experimental FBC spectrum, producing a deviation spectrum (see section "Bias FBC Spectrum"); the FBC deviation spectra from many DNA samples are then analyzed by both principal component analysis (PCA) (see section "Principal Component Analysis") and the MLAs (see section "Testing the MLAs"); the MLAs are trained and cross-validated on one set of FBC deviation spectra and then tested against another set of never-before-seen FBC deviation spectra (see section "Testing the MLAs").

## DNA Sequences

While awaiting experimental data from the SERS instrument, simulated SERS BOC data were generated to determine the feasibility of the device in identifying bacterial species and antibiotic resistance genes. Reference genomes for 12 bacterial species, 728 plasmids containing 4 different types of carbapenem antibiotic-resistant genes, and 600 control plasmids not containing any carbapenem resistance genes were collected from the National Center of Biotechnology Information's reference sequence (NCBI RefSeq; see **Supplementary Material** for NCBI reference IDs for each genome and plasmid) (O'Leary et al., 2016). The DNA sequences were separated into genomic and plasmid DNA (gDNA and

pDNA, respectively) and studied separately using PCA and several MLAs.

Of the gDNA sequences, 10 of the 12 species are common organisms producing bloodstream infections (Bacteroides fragilis, Campylobacter jejuni, Enterococcus hirae, Escherichia coli, Escherichia fergusonii, Klebsiella pneumoniae, Salmonella enterica, Staphylococcus aureus, Streptococcus pneumoniae, and Streptococcus pyogenes) and were used in both training and testing (different genomic sequences were used for training and testing); and two additional species (Klebsiella aerogenes and Mycobacterium tuberculosis) were only used in testing (for taxonomy testing). Median GC% contents for the chosen species are included in **Supplementary Table 4**. One NCBI RefSeq genome was used for each species in the training set and one NCBI RefSeq genome was used for each species in the testing set. Only one was used because 1000 genomes are generated from each genome using the original genome FBC spectrum as the probability distribution for creating new genomes (see sections "Generating Sequence-Specific FBC Spectrum" and "Simulating Gene Mutations and Experimental Errors"). This means that no 2 simulated genomes are identical, even in the absence of error, and that the MLAs see the original genome and 999 variations for both the training and testing sets for each species, resulting in a similar analysis to one made with multiple NCBI RefSeq genomes per species.

For the pDNA, all sequenced plasmids from the NCBI RefSeq for the four carbapenem resistance plasmids [imipenemase 4 (IMP-4), Klebsiella pneumoniae carbapenemase 2 (KPC-2), New Delhi metallo-beta-lactamase 1 (NDM-1), and Verona integronencoded metallobeta-lactamase 1 (VIM-1)] were used in either training or testing (see **Table 1** for specific number used in training and testing). No variant plasmid data were used (a variant being KPC-4 or VIM-2), and all control plasmids were checked to make sure they did not contain the four carbapenem resistance plasmids or any variants. Two different tests were performed on the pDNA: the first test investigated whether the MLAs are able to identify the particular type of resistance (out of four types) or identify that none of these are present; the second test grouped the carbapenem resistance plasmids together and investigated whether the MLAs are able to identify the presence (or not) of any carbapenem resistance.



<sup>a</sup>Plasmids were tested against individual resistant types and not resistant (five groups), or as a resistant vs. non-resistant grouping (two groups).

## Generating Sequence-Specific FBC Spectrum

The physical optical instrument reads the BOC of each DNA k-mer bound to r number of SERS pyramids on a silicon chip in the instrument (Sagar et al., 2018). Since the size (i.e., k bases) of each read is known, these base fractions are converted into specific integer counts of nucleotides for each k-mer. BOC reads are written in the form AwTxGyC<sup>z</sup> where 0 ≤ w,x,y,z ≤ k, and w + x + y + z = k. For traditional sequencing, there are 4 <sup>k</sup> possible reads for a single k-mer; however, there are only (k + 3)!/(k!3!) BOC reads corresponding to the different ways of assigning the variables w, x, y, and z, given the previous constraints. The distribution of BOC reads, hereafter called the FBC spectrum, is defined as the probability distribution function of sampling any BOC read of a specific base composition (see **Figure 2**).

In order to create the FBC spectrum for each gDNA and pDNA sequences of interest, each sequence is decomposed into every possible 10-mer block using both complementary strands of DNA, since both strands will be present in a physical system and a 10-mer block from either strand has the same probability of adhering to the pyramid tip for the BOC reads. Therefore, in creating the FBC spectrum, the k-mers from both complementary strands of DNA are used. These blocks are then assigned to their corresponding bins to produce the FBC spectrum. Once all blocks are binned for the given sequence, each bin count is divided by the total 10-mer count of all 286 bins for that DNA sequence to get the sequence-specific probability distribution function, or FBC spectrum.

To simulate BOC reads on a multi-pyramid chip, the FBC spectrum for the selected plasmid or genome is randomly sampled 2.5 million times, using the sequence-specific probability distribution. This represents having more than one copy of the DNA sequence present in a physical experiment. From this 2.5 million-value array, the first r number of values are selected and distributed into bins to produce a simulated experimental FBC spectrum, where r represents the number of pyramid tips on the SERS-BOC device. This is done for both the training and testing sets.

## Simulating Gene Mutations and Experimental Errors

Due to mutations present in bacteria, any given bacteria species or plasmid will not have a perfectly identical FBC spectrum as that of the corresponding NCBI reference sequence. These natural gene mutations, which are on the order of 5 × 10−<sup>4</sup> to 5 × 10−<sup>9</sup> in bacteria (Denamur et al., 2002; Denamur and Matic, 2006), are expected to be overwhelmed by the experimental errors produced during experimental BOC reads and assignments. There are several sources for error in the optical sequencing reads. A few examples are under-digestion and/or over-digestion which results in non-uniform length k-mer sequences, k-mer sequences adhering to the pyramid tips such that not all bases can be read, portions of multiple k-mer sequences adhering to the same pyramid tip, and optical noise from the instrumentation (Sagar et al., 2018). While creating more realistic training and test

samples, a single error rate parameter was introduced to modify the BOC read that accounts for both the expected bacterial mutations and the instrument errors, producing FBC spectra with various levels of random error.

We define an error rate m (where 0 ≤ m ≤ 1) to be the fraction of bases in the reference sequence that are expected to contain an error. Assuming that the errors are randomly distributed throughout the reference sequence, the number of errors in a randomly selected 10-mer is the same as the number of errors in 10 randomly selected bases. The probability of selecting a 10 mer without any errors is therefore determined by a binomial distribution in which the number of trials is the same as the number of pyramid tips, r, and the probability of being errorless is one minus the error rate, 1-m. To simulate errors, a value of either [0,1] is sampled from the binomial distribution B(r,m). If a 0 is chosen, then a value is chosen from the sequence-specific FBC spectrum (the 2.5-million-value array described above in Section "Generating Sequence-Specific FBC Spectrum"). If a 1 is chosen, then a random value from the bias spectrum (see section "Bias FBC Spectrum" for details) is chosen. This is repeated until a list of values is created that is r in length; next, the list is distributed into bins to produce the "noisy" FBC bins. The resulting FBC 10-mer bin counts are divided by r to obtain the "noisy" FBC spectrum for the given plasmid or genome.

## Bias FBC Spectrum

We discovered that a key to enhancing the differences in the FBC spectra of various DNA sequences is to subtract from each FBC spectrum the spectrum of totally random ATGC, leaving a spectrum of deviations from randomness. The resulting spectrum is called the "deviation spectrum" for a particular sequence.

Because the k-mer size is given, a purely random spectrum (called the bias spectrum) can be generated by including every possible k-mer once. Since there are 4<sup>k</sup> possible sequential k-mers, and given that any BOC read (AwTxGyCz) has k!/(w!x!y!z!) ways of permuting the base counts to create nucleotide-specific k-mers, the bias spectrum can be calculated as:

$$
bar{\left(A\_{\le}T\_{x}G\_{\mathcal{Y}}C\_{z}\right)} = \frac{1}{4^{k}} \cdot \frac{k!}{\
u!\chi!\!\!/\mathcal{Y}!z!}$$

This bias spectrum is subtracted from the FBC spectrum of the particular DNA sequence to yield a unique-DNA-sequence FBC deviation spectrum. This unique-DNA-sequence deviation spectrum is the deviation from pure randomness and should oscillate around zero.

For our given k-mer size of 10, there are 286 bins in the FBC spectrum. The FBC spectrum for a sequence can be visualized by plotting the frequency of a 10-mer in the bins of the spectrum, as shown in **Figure 2**. **Figure 2A** shows the corresponding FBC spectrum for E. coli, K. pneumoniae, K. aerogenes, and the bias spectrum. **Figure 2B** shows the resulting deviation spectra for E. coli, K. pneumoniae, and K. aerogenes produced by subtracting the bias spectrum. In **Figure 2A**, the commonness of the sequence is indicated by the height of the peak, with taller peaks being more common, such as A3T2G3C<sup>2</sup> and A3T3G2C2. A few of these peaks are labeled for easy comparison with **Figure 2B**. As seen in **Figure 2B**, some FBC 10-mers appear more often than expected from a random sequence and some FBC 10-mers appear less than expected from a random sequence. While obvious that these bacteria do not have random sequences, it is useful and informative to observe that randomly breaking their DNA into all possible 10-mers does not produce a random FBC spectrum; in fact, unique features appear that suggest a species may be identified by its deviation spectrum.

To test the robustness of identification of species or genes by deviation spectra in the presence of real experimental noise and random mutations in the DNA sequence, random errors are introduced into the FBC spectra (see section "Simulating Gene Mutations and Experimental Errors" for details) and then the resulting noisy data are divided by r and the bias spectrum is subtracted to obtain a simulated FBC deviation spectrum (with noise) for the given plasmid or genome. For each gDNA and pDNA sequence (22 bacterial genomes and 1329 plasmids), 1000 simulated FBC deviation spectra were created from each specific sequence for both the training set and testing set.

## Principal Component Analysis

Principal component analysis is useful analysis is useful for reducing data dimensions data dimensions while retaining trends and patterns (Lever et al., 2017). This technique, which can reduce computational expense, is often used with biological data. For these reasons, PCA was used to investigate whether a simple data reduction analysis could easily identify the different species and antibiotic resistance genes using the noiseless and noisy FBC deviation spectra. As seen in **Figures 3**, **4**, visualizing the first two principal components allows for some of the data to be easily classified into the correct groups, while other data are unclassifiable. Adding a third principal component may help cluster the data, but information is not easily retrieved from three dimensional plots, especially with 1000s of data points. From this initial result, MLAs were subsequently examined to determine if supervised learning methods could classify the data.

## Testing the MLAs

After creating the FBC deviation spectra from noiseless and noisy DNA BOC reads, the data sets were split into a training set and a never-before-seen test set (both sets of FBC spectra are created through the process detailed by Sections "Generating Sequence-Specific FBC Spectrum," "Simulating Gene Mutations and Experimental Errors," and "Bias FBC Spectrum") and run through several MLAs to classify the bacterial species or the carbapenem resistance status of plasmids based on their FBC deviation spectra. The attributes used in the classification model were the values corresponding to the probability distribution from each FBC deviation spectrum. The training set was then randomly split into 900 and 100 deviation spectra for each DNA sequence for running 10-fold cross-validation. The MLAs were trained on the 900 deviation spectra from each DNA sequence and then tested against the remaining 100 deviation spectra from each DNA sequence of the training set and then tested against all of the never-before-seen sequences. For the gDNA, that meant that the MLAs were trained on 9,000 training deviation spectra (900 FBC spectra for each of the 10 training genome

sequences) and validated against the remaining 1,000 training deviation spectra from the training set and then tested against all the 12,000 testing deviation spectra (1,000 FBC spectra for each of the 12 testing genome sequences). The MLAs were trained and tested similarly for the pDNA. For the gDNA, the labels used for classifying were: B. fragilis, C. jejuni, E. hirae, E. coli, E. fergusonii, K. pneumoniae, S. enterica, S. aureus, S. pneumonia, and S. pyogenes. The other two species, K. aerogenes, and M. tuberculosis are used to investigate how the MLAs group unknown species. For the pDNA, the labels used for the individual classification tests were: KPC, NDM, VIM, IMP and No Resistance; and the labels used for the group classification tests were: Resistance and No Resistance.

For both the gDNA and pDNA, the FBC deviation spectra were created with the following parameters: k = 10; r = [10<sup>2</sup> ; 10<sup>3</sup> ; 10<sup>4</sup> ; 10<sup>5</sup> ; 10<sup>6</sup> ]; m = [0, 0.01, 0.05, 0.1, 0.25, 0.33, 0.5, 0.75, 0.9, 1]; and s = 1000; where k is the size of the k-mer, r is the number of pyramid tips for generating the sample FBC spectra, m is the fractional error rate, and s is the number of FBC deviation spectra created per DNA sequence (genome or plasmid). All 50 combinations of r and m were tested.

From the machine learning python package, Sci-kit learn (version 0.20.3) (Pedregosa et al., 2011), 11 different MLAs were tested from the following categories: linear MLAs, decision tree learning algorithms, Naïve Bayes learning algorithms, discriminant analyses, and a neural network. For this initial study, default parameters were used for all of these algorithms. Each of these classification models were chosen for their ability to fit data with positive and negative values and to fit data to the model using out-of-core fitting, except for the discriminant analyses (see **Supplementary Material** for further details). Presented in the results is the best algorithm from each category; the results of the other algorithms are in the **Supplementary Material**.

## Model Performance Testing and Statistical Analysis

The robustness of each classification model was studied by measuring the predictive accuracy as a function of the parameters of the simulated BOC data. For each optical sequencing read number (r) and each error rate (m), the performance of the model was quantified by the predictive accuracy and a confusion matrix, which keeps track of the true positives, true negatives, false positives, and false negatives for the sample (Powers, 2011). The accuracies and confusion matrix presented for each MLA are the average of 10 trials (n = 10) for the given r and m. Three different cutoff accuracies (95% for species, 90% for group plasmid, and 75% for individual plasmid) for the never-beforeseen sequences were chosen as criteria for assessing the effects of the different error rates and sequencing read numbers.

## Simulation

To generate large amounts of simulated experimental data on which to test the different MLAs, the data were produced using code written in Python 3.7 as described above. Reference genomes were downloaded directly from the NCBI RefSeq database. All code used for running the simulation is available at: https://github.com/rlwphd/DNAFingerprints. While these simulations were tested only on plasmids having a carbapenem antibiotic resistance gene, the simulation will work for any set of antibiotic resistance genes in which the full sequence of the gene-containing plasmid is known.

## RESULTS

## Principal Component Analysis

**Figures 3**, **4** visually display the first two principal components for the deviation spectra of the species and resistance genes. For species, the PCA data are visually distinct, even when significant noise is added to the BOC data. For resistance gene detection, the PCA could not produce a clear distinction. Details are discussed below.

For both bacterial species and antibiotic resistance gene identification, it is noted that the PCA produced in all cases contains an arch effect, which indicates that the principal components are not completely independent of each other and are thus not completely orthogonal to each other (Morton et al., 2017). Since the PCA assumes independence and orthogonality between the principal components, it is not completely reliable as a means of identification without first adjusting for the arch effect. However, PCA was used here to reveal the extent of differences that might be learned by the MLAs.

## Bacterial Species

For the bacterial species, we found that the PCA revealed significant differences between the various species, even with a 90% error rate for 10<sup>6</sup> reads. As a control, using a 100% error rate produced no differences (data not shown) because the PCA yields the same value for all species, regardless of the number of reads. We selected E. coli and E. fergusonii, two genetically very similar bacteria, as a stringent test for sensitivity and discrimination. Because of the similarity of E. coli and E. fergusonii, the graphical plot can only reveal the difference between these two species for error rates of 0–33% for 10<sup>5</sup> reads and 0–90% for 10<sup>6</sup> reads (see difference between **Figures 3A,B**). **Figure 3** contains an example of a non-overlapping (significantly different) PCA (A) and an overlapping (some difference to no difference) PCA (B). **Figure 3** shows the PCA of the FBC deviation spectra data for the bacteria species for 25% error at 10<sup>6</sup> reads (A) and at 10<sup>4</sup> reads (B) (see **Supplementary Material** for additional PCA figures). This figure shows that with enough BOC reads even in the presence of error, there exists significant differences between the species when using only the first two principal components. This indicates that the FBCs of each of these 12 species are distinct enough that the MLAs should be able to easily classify each species, potentially perfectly, even in the presence of noise, even differentiating very similar species such as E. coli and E. fergusonii.

## Antibiotic Resistance Genes

For the plasmids, the PCA revealed that there is a greater spread across the plasmid sequences with no clear differences using only two principal components for any number of

reads, even with no error. This is not surprising because the antibiotic resistant gene content could be affecting less than 1% of the FBC deviation spectrum (800 bases on a 200,000 base plasmid) for some of the samples. This indicates that the MLAs will need to detect the small resistance gene signal from a dominating background. In addition to a low signal competing with a strong background, there is a wide range of different DNA signatures in plasmids, increasing the complexity of the task. **Figure 4** highlights this varying range of DNA signatures; both control plasmids and plasmids containing resistance genes span the entire space of the PCA plot. Neither the individual resistance categorization (**Figures 4A,B**) nor the group resistance categorization (**Figures 4C,D**) provided any insights into clustering or separation. **Figure 4** shows the PCA of the FBC spectra data for the plasmids for 0% error at 10<sup>6</sup> reads (A&C) and at 10<sup>4</sup> reads (B&D) (see **Supplementary Material** for additional PCA figures).

## Classification Using MLAs

The outputs from 1000s of simulated SERS-BOC experiments were produced to examine a wide range of combinations of error rates and the number of reads from pyramid tips. The simulation

creates hypothetical BOC reads from different bacterial genomes or plasmids which are then converted to FBC deviation spectra and handed to the different MLAs for categorizing. Simulating the experiments allows us to show feasibility and test the robustness of the algorithm in the presence of noise and genetic variation, and to examine the device as a diagnostic tool for bacteria classification and resistance profiling.

The results detailed below showed that the best algorithms from each category are: Passive-Aggressive Classifier (PA, linear machine learning algorithm); Extra Trees Classifier (ET, decision tree algorithm); Gaussian Naïve Bayes (GNB, Naïve Bayes algorithm); Linear Discriminant Analysis (LDA, discriminant analysis algorithm); and the neural network (NN, whose default is 100 layers and the number of nodes determined by the number of input features). The results for each of these MLAs are presented below and the results for the other MLAs can be found in **Supplementary Tables 1**–**3**.

## Bacterial Species

The MLA analysis showed that we were able to accurately classify greater than 96% of the simulated unseen bacterial genome data sets (1000 samples per species) using the ET, LDA and NN MLAs

FIGURE 4 | Visualization of the individual and group plasmid principal component analysis. The plasmid FBC spectra cannot be distinguished using only two principal components (500 FBC spectra per plasmid). (A) Individual 10<sup>6</sup> BOC reads; (B) Individual 10<sup>4</sup> BOC reads; (C) Group 10<sup>6</sup> BOC reads; (D) Group 10<sup>4</sup> BOC reads. The error rate is 0% for (A–D). In (A–D), the darker squares represent the never-before-seen plasmids for the given resistance type with the lighter circles representing the plasmids trained on. The non-resistant plasmids are represented by a darker diamond for the never-before-seen plasmids with a lighter triangle for training plasmids.

employing as low as 10<sup>4</sup> BOC reads and up to 50% error. The PA MLA was the worst algorithm and never got better than 78% accuracy (see **Supplementary Material** for detailed table). As for the best algorithms, even at 90% error, the LDA and NN could accurately classify greater than 98% of the bacterial species at 10<sup>5</sup> BOC reads. This surprising result at very high noise levels is postulated to occur because random error added to the BOC generates random noise in the FBC spectrum, but the deviation spectrum has the bias spectrum subtracted out which removes random noise. Thus, the MLA models working on the FBC deviation spectra operate on data with a good signal-to-noise level, even up through 90% random error in the original BOC data.

The results suggest that a SERS-BOC device only needs a 100 × 100 pyramid array (∼10<sup>4</sup> reads per experiment) to accurately identify bacterial species. **Figure 5** shows the MLA accuracy for an error rate of 10% (A) and 25% (B) for all of the BOC reads. This figure shows that greater than 97% accuracy is maintained at 10<sup>4</sup> BOC reads for both 10 and 25% error. **Figure 5** also shows how the various MLAs hold up to error for 10<sup>4</sup> BOC reads (C) and 10<sup>3</sup> BOC reads (D).

The confusion matrix for the MLAs shows how the models mislabel the species they were trained on, and how they label the two species which are not defined in the model. **Figure 6** shows the species confusion matrix for the five MLAs for 10<sup>4</sup> BOC reads and 10% error. While most of the species in this study are not closely related (see **Supplementary Material** for details), the best three MLAs only had problems distinguishing between E. coli and E. fergusonii, which are genetically very similar (same genus). The ET had a sensitivity rate of 82% for E. coli and a specificity rate of 94% for E. fergusonii when comparing the two genomes. The LDA had a sensitivity rate of 92% for E. coli and a specificity rate of 100% for E. fergusonii when comparing the two genomes; meaning that the LDA could correctly identify E. fergusonii but not E. coli. The NN had a sensitivity rate of 93% for E. coli and a specificity rate of 100% for E. fergusonii when comparing the two genomes. All other genomes trained on had 100% identification for the ET, LDA, and NN MLAs. As for the classification of the two extra species (K. aerogenes and M. tuberculosis), all three MLAs identify M. tuberculosis as K. pneumoniae (unrelated) 100% of the time. K. aerogenes was identified as both K. pneumoniae (same genus) and S. enterica (same family) by all three MLAs. Other details are found in **Figure 6**.

## Antibiotic Resistance Genes

fmicb-11-00257 February 18, 2020 Time: 17:51 # 11

In the study of the individual resistance categorization (distinguishing four carbapenem resistance genes), the MLAs had difficulty achieving better than 75% accuracy. Only with 10<sup>5</sup> BOC reads (or more) did the ET algorithm attain better than 75% accuracy, and even this algorithm only achieved 80% accuracy at best. The LDA and ET algorithms had similar accuracy at 10<sup>4</sup> reads (see **Figure 7D**) but the LDA did not perform as well at 10<sup>5</sup> reads (**Figure 7C**). The GNB and PA MLAs were the worst but were still able to achieve 64% accuracy (see **Figure 7** and **Supplementary Material** for a detailed table). **Figure 7** shows the MLA accuracy for an error rate of 10% (A) and 25% (B) for all of the BOC reads for the individual scenarios. **Figures 7A,B** show that greater than 75% accuracy is maintained at 10<sup>5</sup> BOC reads for both 10 and 25% error. **Figure 7** also shows how the MLAs hold up to error for 10<sup>5</sup> BOC reads (C) and 10<sup>4</sup> BOC reads (D).

For the group resistance categorization, the MLA analysis was able to accurately classify 90% of the simulated unseen data set using the ET, LDA and NN MLAs down to 10<sup>4</sup> BOC reads and up to 50% error. The GNB MLA performed the worst of those examined but was still able to achieve 86% accuracy (see **Figure 8** and **Supplementary Material** for detailed table). **Figure 8** shows the MLA accuracy for an error rate of 10% (A) and 25% (B) for all of the BOC reads for the group scenario. **Figure 8** shows that greater than 90% accuracy is maintained at 10<sup>4</sup> BOC reads for both 10 and 25% error. **Figure 8** also shows how the MLAs hold up to error for 10<sup>5</sup> BOC reads (C) and 10<sup>4</sup> BOC reads (D).

The ET MLA model for the individual resistance categorization maintains 74% accuracy for up to 25% error for 10<sup>4</sup> BOC reads and maintains 80% accuracy for up to 25% error for 10<sup>6</sup> BOC reads. The individual results suggest that a SERS-BOC device would need to be at least a 100 × 100 pyramid array (10<sup>4</sup> reads per experiment), but a 1,000 × 1,000 pyramid array (10<sup>6</sup> reads per experiment) would be optimal for the best results to accurately identify antibiotic resistance genes on plasmids. For the group resistance categorization, the three best MLA models maintain good signal-to-noise levels up through 50% error. The group results suggest that a SERS-BOC device with a 100 × 100 pyramid array would be sufficient to accurately identify whether any carbapenem resistance gene was present in the presence of 50% experimental noise.

The confusion matrix for the MLAs shows how the models mislabel the plasmids, which allows us to calculate the falsenegative rate for the antibiotic-resistant plasmids as a group, as well as the sensitivity of each individual plasmid. The falsenegative rate metric is clinically important as it is incorrectly labeling resistant plasmids as not being resistant, which could result in the wrong (ineffective) antibiotics being given to the patient, possibly leading to death. The sensitivity metric shows which type of resistance gene is harder to identify. **Figure 9** shows the confusion matrix for the five MLAs for 10<sup>5</sup> BOC reads and 10% error for the plasmid sets for individual identification. **Figure 10** shows the confusion matrix for the five MLAs for 10<sup>5</sup> BOC reads and 10% error for the plasmid sets for group identification.

The sensitivity rate at 10<sup>5</sup> BOC reads and 10% error is the values on the diagonal on the confusion matrices shown in **Figures 9**, **10**. The clinically important false negative rates at 10<sup>5</sup> BOC reads with 10% error from the individual carbapenem resistance identification are 0.50% for LDA, 1.00% for ET, 1.25% for NN, 2.50% for PA, and 4.75% for GNB. The false-negative rates at 10<sup>5</sup> BOC reads with 10% error from the grouped carbapenem resistance identification are 2.6% for LDA, 4.5% for ET, 5.4% for NN, 5.9% for GNB, and 8.2% for PA. Thus, the LDA MLA appears best at avoiding false-negative errors when identifying resistance genes on plasmids, while the ET most accurately identifies the resistance genes.

## DISCUSSION

This research is the first of its kind and validates the SERS-BOC instrument (currently in development) as an excellent predictor of genetic signatures, even in the presence of genetic mutations and experimental noise. The use of FBC deviation spectra was able to achieve greater than 99% accuracy in classifying bacterial species using several types of MLAs.

We were able to detect individual antibiotic resistance genes on plasmids from their FBC deviation spectra with 80% accuracy and to detect the presence of a carbapenem resistance gene with 94% accuracy using MLAs. The implications of our research suggest that the use of MLA classifiers on fractional base composition data generated from a SERS-BOC type instrument (Sagar et al., 2018) has tremendous potential in accurately identifying both bacterial species and antibiotic resistance genes. With respect to bloodstream infection diagnosis, the creation of FBC models has the potential to help determine the bacterial species and the antibiotic-resistance profile associated with a bloodstream infection in a cost-efficient and time-efficient way, thus improving the outcomes for patients. Of most import, the false negative evaluations for carbapenem resistance genes were less than 3% using the LDA algorithm with 10<sup>5</sup> BOC reads.

We note with surprise the lack of spread within a species data cluster in PCA analysis even when introducing a 25% error rate. This indicates that the inclusion of error does not introduce enough variance to cause these species to have overlap in the principal component space until the number of BOC reads get low (10<sup>2</sup> and 10<sup>3</sup> reads). Because we are sampling at least a million different k-mers to generate the FBC spectrum, the distribution of any single training sample does not deviate significantly from the FBC of the non-mutated genome. The high number of reads available on a 1-million-pyramid SERS array would enable this high predictive ability, and may be more than necessary. Even with experimental error causing deviation from the reference FBC, the FBCs of different species are distinct

FIGURE 7 | Individual antibiotic-resistance model accuracy for different number of BOC reads and error rates. The accuracy of the five MLAs is shown with (A) 10% error, and (B) 25% error in identifying the never-before-seen plasmid sets for individual identification. The accuracy of the five MLAs is shown with (C) 10<sup>5</sup> BOC reads, and (D) 10<sup>4</sup> BOC reads in identifying the never-before-seen plasmid sets for individual identification. The red line represents a 75% accuracy threshold.

FIGURE 8 | Group antibiotic-resistance model accuracy with different number of BOC reads and error rates. The accuracy of the five MLAs is shown with (A) 10% error, and (B) 25% error in identifying the never-before-seen plasmid sets for group identification. The accuracy of the five MLAs is shown with (C) 10<sup>5</sup> BOC reads, and (D) 10<sup>4</sup> BOC reads in identifying the never-before-seen plasmid sets for group identification. The red line represents a 90% accuracy threshold.

enough that the simulated experimental noise had no negative effect on clustering in principal component space. The random error in BOC data from genomic mutation and experimental noise in the FBC spectra are subtracted out since we know what the random spectrum looks like. Thus, random errors reduce the signal levels but have little effect on noise levels.

The MLA analysis shows how robust the species model is at handling experimental noise in the FBC, allowing for substantial error rates while maintaining its predictive power. Again, one reason why the model can still accurately classify species even with high error rates is that the experimental noise mimics the random bias distribution. Since we subtract out this random bias before running the machine learning classifier, we are essentially "subtracting out" the effects of the mutation or noise, and what is left is a smaller size of the accurate FBC, which the MLA can still use to accurately classify the species of bacteria. For example, 10<sup>6</sup> BOC reads with 90% random errors still leaves 10<sup>5</sup> good reads once the randomness is subtracted.

This study challenged the algorithms with much more random error in the data than expected in actual experiments. The largest source of error is anticipated to be wrong assignment of the BOC values due to spectroscopic noise. Currently the accuracy of calling correct BOC is about 90% (Sagar et al., 2018), but accuracy is improving with time and experience and optical quality. Thus 25% error is probably a gross overestimate of actual experimental error, which we anticipate will be on the order of 10% or less. Random point mutations in bacteria are far less than 0.1% and have no bearing on the accuracy of species identification.

Plasmids come in all different sizes ranging from 2,000 to 200,000 bp long (and longer), and the carbapenem-resistant genes are about 800 bp. This means that the FBC that comes from the resistant genes has varying weight upon the FBC spectrum of the entire plasmid, which would explain why there is so much spread from one resistant plasmid to another in the PCA analysis. The ability for the MLAs to learn from that 800-bp region amongst all of the other data and noise shows great promise for this type of antibiotic resistance gene identification. Overall, the never-before-seen results show that the good performance of the models is not a result of overfitting; rather, the models are actually learning how to recognize the characteristic signatures in an FBC deviation spectrum when a specific target gene is included. Of course the application of the algorithm for antibiotic resistance

identification requires knowledge of an antibiotic resistance gene sequence, so this technique would not identify novel antibiotic resistance evolved by mutations.

In a real-world application, these models would need to be tuned to perform optimally, but the low sensitivity of the accuracy to error rates ranging from 0 to 90% proves the robustness of the models in dealing with noise. As previously stated, we postulate that this robustness stems from the subtraction of randomness from within the data set. A high number of BOC reads with high random error is similar to a model having less SERS-BOC readings, which our results show is still effective down to the 10<sup>4</sup> BOC reads. Additional sources of error in a real-world setting could include DNA contamination (human or bacterial), lack of good separation between plasmids and genomes (which would make the resistant genes on plasmids less distinguishable), and untested or new bacterial species causing the blood infections. We anticipate the continued development of the SERS-BOC instrument that will provide real BOC data from which FBC data can be created to further test our assumptions and postulates. Further application of this method could go beyond analyzing bacterial bloodstream infections into other clinical scenarios that need fast and reliable analysis of only a few key genes of known sequence.

There are known limitations to this study. The first and foremost is that all of the data had to be simulated because a large enough device to produce experimental data are still being developed. Another limitation is that only a single group of antibiotic resistance genes (carbapenem resistant genes) was tested. Therefore, the results cannot be broadcasted to other groups of resistance genes or to identifying other genes. Also, this technique can only identify antibiotic resistance through known DNA sequences. Another limitation is the limited number of bacterial species used. Because of the focus on testing the technology for clinical use in bloodstream infections, a select group of bacterial species was used for this initial study. Future studies will test the ability of this technique to identify between all species. This study also did not tune the MLAs or perform a costbenefit analysis to identify the factors influencing the accuracy of the MLAs. However, this study provides the initial groundwork for further exploration in using a SERS-BOC device to identify species and genes based on 10-mer DNA sequences.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## AUTHOR CONTRIBUTIONS

fmicb-11-00257 February 18, 2020 Time: 17:51 # 15

RW contributed to the concept, ran the simulations, and contributed to the writing. TJ and CW contributed to the concept and ran initial simulations. MC advised the simulations and contributed to the writing. PN contributed to

## REFERENCES


the concept and contributed to the writing. WP contributed to the concept, advised the simulations, and contributed to the writing.

## ACKNOWLEDGMENTS

RW acknowledges some graduate student stipend from NIH grant R01AI116989 and the Chemical Engineering Department at Brigham Young University.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00257/full#supplementary-material


**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 Wood, Jensen, Wadsworth, Clement, Nagpal and Pitt. 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.

# Essential Metabolic Routes as a Way to ESKAPE From Antibiotic Resistance

Angélica Luana C. Barra<sup>1</sup> , Lívia de Oliveira C. Dantas <sup>1</sup> , Luana Galvão Morão<sup>1</sup> , Raíssa F. Gutierrez <sup>1</sup> , Igor Polikarpov <sup>1</sup> , Carsten Wrenger <sup>2</sup> \* and Alessandro S. Nascimento<sup>1</sup> \*

<sup>1</sup> São Carlos Institute of Physics, University of São Paulo, São Carlos, Brazil, <sup>2</sup> Department of Parasitology, Institute of Biomedical Sciences, University of São Paulo, São Paulo, Brazil

Antibiotic resistance is a worldwide concern that requires a concerted action from physicians, patients, governmental agencies, and academia to prevent infections and the spread of resistance, track resistant bacteria, improve the use of current antibiotics, and develop new antibiotics. Despite the efforts spent so far, the current antibiotics in the market are restricted to only five general targets/pathways highlighting the need for basic research focusing on the discovery and evaluation of new potential targets. Here we interrogate two biosynthetic pathways as potentially druggable pathways in bacteria. The biosynthesis pathway for thiamine (vitamin B1), absent in humans, but found in many bacteria, including organisms in the group of the ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumanii, Pseudomonas aeruginosa, and Enterobacter sp.) and the biosynthesis pathway for pyridoxal 5′ -phosphate and its vitamers (vitamin B6), found in S. aureus. Using current genomic data, we discuss the possibilities of inhibition of enzymes in the pathway and review the current state of the art in the scientific literature.

### Edited by:

Filipa Grosso, University of Porto, Portugal

#### Reviewed by:

Roberto Contestabile, Sapienza University of Rome, Italy Andrea Mozzarelli, University of Parma, Italy

#### \*Correspondence:

Carsten Wrenger cwrenger@icb.usp.br Alessandro S. Nascimento asnascimento@ifsc.usp.br

#### Specialty section:

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

> Received: 28 October 2019 Accepted: 27 January 2020 Published: 28 February 2020

#### Citation:

Barra ALC, Dantas LOC, Morão LG, Gutierrez RF, Polikarpov I, Wrenger C and Nascimento AS (2020) Essential Metabolic Routes as a Way to ESKAPE From Antibiotic Resistance. Front. Public Health 8:26. doi: 10.3389/fpubh.2020.00026 Keywords: ESKAPE pathogens, thiamine, pyridoxal 5′ -phosphate, antibiotic resistance, vitamin biosynthesis

## INTRODUCTION

Antibiotic resistance is an urgent threat to human health and requires urgent actions from physicians, patients, industries, governmental agencies, and the academic community worldwide. According to the last document from the Centers for Disease Control and Prevention (CDC) regarding antibiotic resistance in the United States, from 2013, the number of people with serious infections caused by resistant bacteria reaches 2 million every year, with at least 23,000 deaths per year directly caused by these infections (1). The situation is similarly warning in Europe, where 25,100 deaths were reported from the European Center for Disease Prevention and Control in 2007 (2). Globally, 700,000 deaths are estimated every year as a consequence of antibiotic resistance (3). The same CDC document lists four general action lines to address antibiotic resistance: (i) preventing infections and the spread of resistance; (ii) tracking resistant bacteria; (iii) improving the use of current antibiotics; and (iv) developing new antibiotics (1).

Since bacteria have a short doubling time and efficient mechanisms for plasmid sharing, the development of antibiotic resistance is a very efficient defense mechanism. So, as previously said by Walsh and Wencewicz, the development of resistance is not a matter of if, but rather a matter of when (4), and despite the title of this paper (which is rather provocative), there is no way we can escape from it (5). On the contrary, the development/discovery of new antibiotics will tend to be a continuous goal to be achieved in drug discovery pipelines.

Interestingly, even after what has been named as the "golden age" of antibiotic development, the drugs currently in the market are restricted to only five molecular targets and/or pathways (4): (i) the peptidoglycan/cell wall biosynthesis, site of action of beta-lactam antibiotics, for example; (ii) the protein biosynthesis, where the ribosome is an important target; (iii) DNA replication and RNA transcription; (iv) the folate biosynthesis pathway, and (v) the disruption of the bacterial membrane. Given the relevance of antibiotic therapy and the emergence of antibiotic resistance, the available choices of current drugs are very narrow in the context of the mechanism of action restricted to only five molecular/pathway targets.

Antibiotic resistance, as a threat to human health, should be addressed in multiple and simultaneous ways. For the academia, an interesting way to address antibiotic resistance is the discovery and validation of new targets/pathways that could be specifically targeted by new antibiotic candidates. It is worthy of note that, according to Kelly and Davies (3), no new class of antibiotics was discovered and released for routine treatment since the 1980s, highlighting the outstanding role that the academia can have in the preliminary research for discovery and validation of druggable targets/pathways.

About 7% of the Escherichia coli genome, 303 genes, were shown to be composed of essential genes (5, 6). Under stress conditions caused by a limited medium, additional 119 genes show some condition-dependent essentiality, including several genes related to the metabolism and synthesis of essential molecules such as amino acids and nucleotides (5), and obviously, some of these targets/pathways can be interesting choices for the development of new antibiotic candidates. A promising approach to screen compounds for their activity in metabolic pathways was shown by Zlitni et al. (7). The authors searched for antibacterial compounds under poor nutrient media and found three potential compounds in this screening strategy (7). Worthy of note, the metabolic profile of existing antibiotics showed that the supplementation of vitamins B5, B6, B1, and B2 did not significantly reverse the antimicrobial effect of any of the 24 inhibitors assayed (7), suggesting that the current antibiotics do not explore these essential pathways.

In this context, the enzymatic routes for the biosynthesis of vitamins are very interesting pathways to be explored as potential targets for the discovery of new antibiotic candidates. Vitamin B1, for example, cannot be synthesized by humans, although several microorganisms can synthesize this vitamin, including pathogens. In the absence of thiamin (vitamin B1), the activity of several carbohydrate metabolism enzymes is impaired, including pyruvate dehydrogenase, which connects glycolysis and the citric acid cycle (8). Other thiamin-dependent enzymes are transketolase, α-ketoacid decarboxylase, α-ketoacid dehydrogenase, and acetolactate synthase (9). A very similar scenario was observed for pyridoxal 5′ -phosphate (vitamin B6) in the model Gram-positive organism Bacillus subtilis (10).

The pathways involved in the synthesis of vitamins, in particular, vitamin B1 (thiamin) and B6 (pyridoxal), seem to be of great relevance, since they are involved in central processes in the metabolism of carbohydrates and amino acids, and the corresponding enzymes are found in most bacteria, fungi, and plants but not in humans (8), favoring the development of specific drugs with minimal side effects due to the interaction with the host. However, a few questions still stand: (i) How feasible are the targets involved in the pathways for the biosynthesis of thiamin and pyridoxal for the microorganisms with a higher emergency in terms of resistance, or ESKAPE (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumanii, Pseudomonas aeruginosa, and Enterobacter sp.)? (ii) What enzymes are present, and what do we know about them?

Here we used the available genomic and proteomic data to address these questions focusing on the ESKAPE pathogens. Vitamins B1 and B6 were chosen for this analysis, since the biosynthetic pathways for these enzymes have already been validated as molecular targets for some human pathogens, such as thiamin for Plasmodium falciparum (11), or pyridoxal for P. falciparum (12) or Trypanosoma brucei (13), for example.

## METHODS

The analyses provided here are the result of the interrogation of the enzymes in the biosynthesis pathways for thiamin and pyridoxine phosphate or pyridoxal phosphate using the KEGG database (14, 15). For this purpose, the KEGG pathway for thiamin metabolism (map 00730) was listed for the ESKAPE pathogens using, whenever possible, commercial strains rather than specific or antibiotic-resistant strains. Briefly, for each pathogen (E. faecium, S. aureus, K. pneumoniae, A. baumanii, P. aeruginosa, and Enterobacter species), a KEGG search for the maps of thiamin and pyridoxal metabolism was done, looking for the existing genes in commercial strains for each of the ESKAPE organisms. For the sake of comparison, some additional resistant strains were listed, with no differences in the observed genes for the pathways under study. A similar search was carried out for vitamin B6 metabolism (map 00750) in KEGG, interrogating the existing enzymes for the ESKAPE pathogens in comparison with humans. The analysis was focused in the enzymes in the biosynthesis of B1 and B6 vitamins in an attempt to identify the most promising targets for future medicinal chemistry studies. Finally, a table listing the existing genes for each pathogen was compiled and is presented in the following sections.

Using the existing literature, we sought to identify whenever possible the cases of enzymes with functional redundancy with another enzyme for the same organism. The identified cases were discussed as less promising targets for medicinal chemistry campaigns. Additionally, the existing data for enzyme inhibition using natural or synthetic compounds was compiled to provide initial proof-of-concept clues about the druggability of the identified promising targets.

When necessary, the sequence of an enzyme of the pathway was used to search for homologs using BLAST (16) searches against the PDB (17) or the non-redundant database of proteins using BLAST default parameters, i.e., minimum expected threshold of 10, a word size of 6, and the BLOSUM62 substitution matrix.

## RESULTS AND DISCUSSION

## Thiamin Biosynthesis

The biosynthesis pathway for thiamin involves two branches, as shown in **Figure 1**. In summary, thiamin is synthesized from 4 amino-5-hydroxymethyl-2-methylpyrimidine (HMP) (superior branch in **Figure 1**) and 5-(2-hydroxyethyl)-4-methylthiazole (THZ, inferior branch). Both compounds are phosphorylated by ThiD and ThiM, respectively. Finally, the enzyme ThiE, central to the pathway, is responsible for synthesizing thiamine phosphate (TMP) by merging the two branches (18). Thiamin phosphate can be dephosphorylated to thiamine and then pyrophosphorylated to thiamin diphosphate (TPP) by TPK. Alternatively, thiamin can be degraded to HMP and THZ by TenA, a thiaminase II enzyme.

According to the KEGG pathway database, humans lack many of the enzymes of this pathway, but not all of them. In humans, a TPK (TPK1) enzyme is found, with UNIPROT ID Q9H3S4. Other than that, the remaining enzymes in the biosynthetic pathway for thiamin are missing in humans, making them very attractive for the design of chemical probes that could be used as proof-of-concept compounds.

The analysis of the thiamin pathway for E. faecium (ATCC 8459) in the KEGG database did not identify genes for ThiE, ThiM, and TenA. For this species, only ThiD and TPK were positively identified. However, a BLAST search for TenA homologous within the non-redundant database restricted to E. faecium (TAXID 1352) identified a single result for thiaminase II (GenBank: SAM74984.1), four results for ThiE (SAZ10134.1, WP\_086323306.1, WP\_010732763.1, and WP\_072538983.1), and two results for ThiM (SAZ10238.1, EJY48288.1). The small number of hits in the BLAST search may suggest some issues in the annotation.

Interestingly, when the Enterococcus faecalis thiamin pathway is compared with the pathway observed for E. faecium, many differences are found. The KEGG pathway for E. faecalis (ATCC 29212) shows that the entire pathway, as described in **Figure 1**, is found: ThiD, ThiM, ThiE, TPK, and TenA, in contrast with E. faecium. Additionally, there is no change in the repertoire of enzymes in the pathway for the Vancomycin-resistant strain V583 compared to the ATCC 29121 strain.

For S. aureus (NCTC 8325) and K. pneumoniae (subsp. pneumoniae ATCC 43816 KPPR1), the KEGG pathway indicates that all the enzymes in the pathway are observed, with no changes to an S. aureus-resistant strain such as COL (MRSA) compared to the NCTC 8325. For A. baumannii (ATCC 14978), all enzymes are observed but TPK. Instead, thiamin phosphate may be converted directly to thiamin diphosphate by a thiamin-monophosphate kinase, and then converted to thiamin triphosphate by an adenylate kinase. In the case of P. aeruginosa (NCGM 1900), some enzymes are missing: ThiD, TenA, TPK. Finally, Enterobacter sp. (638) has ThiD, ThiE, ThiM and misses TenA and TPK.

The overall panel of enzymes for the thiamin pathway for the ESKAPE pathogens is summarized in **Table 1**.

TenA was shown to play a dual role in thiamine synthesis and salvage (19): beyond its function to hydrolyze thiamin, TenA also deaminates aminopyrimidine to form 4-amino-5-hydroxymethyl-2-methylpyrimidine (HMP), with the latter activity being about 100 times faster than the former (19, 20). In some organisms, such as Bacillus subtilis, the deaminase activity is somewhat redundant with the ThiC (ThiA) activity, since both activities provide HMP or HMP phosphate, which can be further phosphorylated by ThiD (9).

In the other branch of the thiamin biosynthesis pathway, thiazole phosphate can be generated from thiazole alcohol incorporated from the medium and phosphorylated by ThiM, or it can be synthesized in a series of enzymatically catalyzed reactions using amino acids such as glycine or tyrosine as substrates (9). In this context, according to Begley and coworkers (9), lesions in the ThiM gene were observed to be prototrophic. In contrast, selective mutations of ThiD and ThiE led to the requirement of externally provided thiamin (9).

ThiE is central to the pathway since the enzyme merges the two branches to finally synthesize thiamin phosphate (**Figure 1**). It is important to highlight that many of the mutational analysis was carried out in E. coli, and as **Table 1** shows, there is significant variance among bacterial species.

Taking together, the analysis of the literature suggests that ThiD and ThiE may represent interesting targets for the development of chemical probes to further evaluate their function. TPK is expressed in humans, and the inhibition of this enzyme could lead to harmful effects on the human host. TenA inhibition could be overcome by the somewhat functional redundancy with thiC (thiA), and ThiM function can be dispensable since thiazole phosphate can be synthesized de novo.

ThiD and ThiE are found in most ESKAPE organisms. Exceptions are P. aeruginosa, which lacks ThiD, and E. faecium, which lacks ThiE, according to the KEGG data. For ThiE, in particular, it was reported that 4-amino-2-trifluoromethyl-5 hydromethylpyrimidine (CF3-HMP), an HMP analog, can be converted by ThiD to CF3-HMP pyrophosphate, which in turn inhibits ThiE (8, 21). This enzymatic inhibition culminated with the inhibition of E. coli growth (8), suggesting that this strategy might be promising for the development of new inhibitors.

In terms of structural biology, some crystal structures of ThiD are available by the time of writing of this paper, including the enzymes from Salmonella enterica [PDB ID 1JXH (22)], Clostridioides difficile (PDB ID 4JJP), B. subtilis [PDB ID 2I5B (23)], Thermus thermophilus (PDB ID 1UB0), A. baumannii (PDB ID 4YL5), Bacteroides thetaiotaomicron (PDB ID 3MBH), and the bifunctional enzyme from Saccharomyces cerevisiae (PDB ID 3RM5). Typically, ThiDs are folded as typical ribokinases, with a central eight-stranded β-sheet surrounded by eight αhelices, and structural studies suggest that some surface loops may have a structural change based on the presence of the nucleotide (23).

TABLE 1 | Panel of observed enzymes for the thiamine pathway in ESKAPE pathogens.


Interestingly, the S. aureus pyridoxal kinase enzyme (SaPdxK) has a dual role, phosphorylating pyridoxal and pyridoxine in the pyridoxal de novo biosynthesis pathway as well as HMP in the thiamin biosynthesis pathway, with a K<sup>M</sup> almost 20 times greater for HMP than for pyridoxal and kcat values three times faster for pyridoxal (24). So, SaPdxK has some redundancy with S. aureus ThiD (SaThiD), with less efficiency, though.

SaThiD was shown to be potentially inhibited by Rugulactone (Ru0), a natural product, as well as by its derivatives Ru1 and Ru2, with IC<sup>50</sup> values ranging from 14 to 32µM (25). In the absence of thiamine in the medium, the MIC observed for Ru1 was four times lower than in the presence of thiamine for Listeria monocytogenes (25), suggesting that the inhibitory effect of Rugulactone is due to ThiD inhibition, although Rugulactone also inhibits other kinases.

For ThiE, the crystal structures of a few enzymes are available, including the enzyme from Pyrococcus furiosus (PDB ID 1XI3), B. subtilis [PDB IDs 1G4T, 3O15, 3O16, 1G69, 1G4E, 1G67, 1G4P, 1YAD (26)], Mycobacterium tuberculosis (3O63), and for the bifunctional enzyme from Candida glabrata [PDB IDs 3NL2 and 3NM1 (27)]. No crystal structure of an ESKAPE pathogen, ThiE, is available to date. From the structural point of view, ThiE is typically folded as an α/β TIM barrel, with thiamine binding at the top of the barrel, as elegantly shown in the set of crystal structures determined for the B. subtilis enzyme (26).

At this point, an important question has to be addressed: can the blockade of thiamine de novo biosynthesis pathway be overcome by internalization of the vitamin? In principle, several microorganisms can import thiamine from the environment (28). So, how effective can the inhibition of ThiD and ThiE be particularly for pathogenic bacteria? This question can be addressed in several levels. On the first level, the essentiality of ThiD, for example, has been shown some pathogenic microorganisms (23), such as Streptococcus pneumonia (29), Haemophilus influenza (30), and M. tuberculosis (31). This is not the case for other bacteria. For B. subtilis, for example, ThiD gene was shown to be dispensable (23). Whether ThiD and ThiE are essential for the ESKAPE pathogens is still an open question to be addressed by further investigation. On a second level, Nodwell et al. showed that L. monocytogenes, which is known to uptake thiamine from the environment, is still sensible to Rugulactone and its derivatives Ru1 and Ru2 (25). When grown in a chemical defined media without thiamine, the inhibitory effects of Rugulactone are potentiated with a 4-fold reduction in the MIC (25). However, the inhibitory effect observed even in the presence of thiamine highlights that the blockade of the de novo biosynthesis pathway is still a promising strategy. As a third level, the natural compound bacimethrin, isolated from Bacillus megaterium and from Streptomyces albus is known to be toxic for bacteria with MIC at a low micromolar range (32, 33). The mechanism of action of bacimethrin is based on the formation of 2′ -methoxy-thiamine where bacimethrin is used by ThiD instead of HMP. 2′ -Methoxy-thiamine pyrophosphate inhibits E. coli growth at concentrations 15 times lower than bacimethrin (32). Interestingly, some Salmonella enterica thiD mutants were shown to be bacimethrin resistant (34). Together, these data suggest that a very promising approach can be devised by the design of suicide drugs, i.e., compounds that are recognized by the enzymes in the pathway but lead to final compounds that cannot be used as thiamine substituents. Bacimethrin is a good example of a suicide inhibitor, and although its effects can be reverted by increased thiamine concentrations, the vitamin concentration that pathogens are exposed to is usually defined within a narrow range.

## Pyridoxal Biosynthesis

Pyridoxal, pyridoxine, and pyridoxamine, together with their respective phosphate esters, compose the six vitamers for vitamin B6 and can be interconverted (35). There are two biosynthetic routes for pyridoxal 5′ -phosphate: a deoxyxylulose 5-phosphatedependent pathway found in some bacteria and a deoxyxylulose 5-phosphate-independent pathway found in all kingdoms (35, 36). This second and widespread pathway, curiously, depends on only two enzymes, as shown in **Figure 2**.

In this pathway, the enzyme Pdx2 converts L-glutamine into L-glutamate and an ammonium molecule. The latter is thought to

diffuse to the Pdx1 active site, which also uses glyceraldehyde 3 phosphate and D-ribose 5-phosphate to synthesize pyridoxal 5′ phosphate (35, 37). Interestingly, in order to properly synthesize pyridoxal 5′ -phosphate, a complex involving 12 Pdx1 molecules and 12 Pdx2 molecules has to be assembled (38–40), and this assembly is further stabilized by the interaction with Lglutamine (41).

Similar to what is observed to the thiamine biosynthesis pathway, humans lack the genes for Pdx1 and Pdx2, making the development of specific antibiotic candidates an attracting strategy. In the same direction, for some pathogenic organisms such as Helicobacter pylori (42), M. tuberculosis (43), and Streptococcus pneumoniae (44), the depletion in vitamin B6 resulted in reduced virulence, indicating that the biosynthesis of this vitamin may be a good strategy for the design of new antibiotic candidates.

Very interestingly and in contrast to what was observed for the thiamine pathways, from the ESKAPE pathogens, S. aureus is the only microorganism that has the ribose 5-phosphate-dependent pathway for the biosynthesis of pyridoxal 5′ -phosphate, as shown in **Table 2**. A single enzyme (Pdx1 or Pdx2 alone) is never observed for these organisms (**Table 2**).

A comparison between the pyridoxal/pyridoxine phosphate pathways for E. faecium and E. faecalis shows that both pathogens miss the Pdx enzymes. On the other hand, some other pathogenic organisms, including organisms with high antibiotic resistance rates, have already been identified as susceptible to modulations of the pyridoxal phosphate biosynthesis pathway. Some organisms that have been the focus of basic research include P. falciparum and Plasmodium vivax, the malaria pathogens (45–48).

In terms of the structural biology of the enzymes in the vitamin B6 biosynthesis pathway, the S. aureus Pdx1 enzyme (UNIPROT ID Q2G0Q1) is a close homolog of Bacillus subtilis Pdx1 [81% identity, PDB ID 2NV1 (40)], Geobacillus stearothermophilus Pdx1 [78% identity, PDB ID 1ZNN (49)], and T. thermophilus enzyme (67% identity, PDB ID 2ZBT). S. aureus Pdx2 has Geobacillus kaustophilus [PDB ID 4WXY (38)], G.s stearothermophilus (PDB ID 1Q7R), and B. subtillis [PDB IDs 2NV0 and 2NV2 (40)] enzymes as its close homologs, with 60, 60, and 58% identity in sequence similarity, respectively.

In terms of its structure, the Pdx1 enzyme has a typical (β/α)<sup>8</sup> barrel fold with a central eight-stranded β-barrel surrounded


TABLE 2 | Panel of observed enzymes for the pyridoxal 5′ -phosphate pathway in ESKAPE pathogens.

by eight α-helices (49). The most impressing feature in the Pdx1 enzyme structure is its quaternary arrangement, where six Pdx1 enzymes interact with each other to form a "donut-like" arrangement with about 100 Å in diameter. For Pdx2, a Rossman fold is observed with an α/β/α sandwich topology (40). Curiously, in the active PLP synthase complex, 12 Pdx1 molecules (a twolayer donut) interact with 12 Pdx2 enzymes. However, Pdx2 molecules do not interact with each other in this complex (40).

Using a structural homology model, Reeksting et al. identified some ribose 5′ -phosphate analogs with interesting in vitro inhibitory effects on the enzyme Pdx1 from P. falciparum (12). The compounds were identified in a structure-based computational screening campaign and were shown to have inhibitory effects in vivo, as well as the in vitro effect. The authors also showed that the in vivo effects could be at least partially suppressed in a mutant strain overexpressing Pdx1 and Pdx2, in a clear indication that the effect of the analogs was due to their inhibition of the pyridoxal 5′ -phosphate biosynthesis pathway (12).

The inhibition of the Pdx2 activity by the glutamine analog acivicin was also demonstrated to be a feasible strategy for the inhibition of the vitamin B6 biosynthesis pathway (50). Raschle et al. showed that when inhibited by acivicin, Pdx2 is incapable of interacting with Pdx1, thus disrupting the pyridoxal 5 ′ -phosphate synthase activity. Interestingly, acivicin is a covalent inhibitor that binds to a cysteine residue. So, it seems that multiple strategies may be available for the design of new binders, including the inhibition of Pdx2 (covalent and non-covalent), the inhibition of Pdx1, and possibly the inhibition of the assembly of the 24-mer complex with Pdx1/Pdx2. Another possibility to address the inhibition of PLP biosynthesis might be at the level of the Pdx1 transcription, which was shown to be regulated by PdxR (44). In principle, if PdxR action could be regulated by an exogenous chemical probe, the PLP synthase activity would be regulated as well. However, a proof-of-concept experimental validation is still necessary.

Finally, a salvage pathway for vitamin B6 is found in many bacteria, as well as in humans. In E. coli, it involves two enzymes: PdxK and PdxY. These kinases act by phosphorylating the vitamers pyridoxal, pyridoxine, and pyridoxamine into their phosphorylated forms (23, 35, 51). Interestingly, in some organisms, including pathogenic microorganisms such as S. aureus, the PdxK activity is carried out by the HMP kinase ThiD, showing some functional convergence between the vitamin B6 and B1 pathways (24). Not surprisingly, this enzyme has also been shown to be a validated druggable target for some pathogens, such as Trypanosoma brucei, for example (13), although specificity may be an important issue, since humans also have genes for PdxK known to be inhibited by drugs like theophylline, for example, with known neurotoxic effects (52).

Again, the same question asked for the thiamine biosynthesis pathway can be asked here: How can the microbial uptake of vitamin B6 overcome the blockade of the biosynthetic pathway? Some evidence suggested that the blockade might be effective. For example, in S. pneumoniae, the deletion of Pdx1 resulted in defective growth of the bacteria. The growth was shown to be restored under increased concentrations of vitamin B6 (44). A similar scenario was also observed for M. tuberculosis (43), suggesting that the uptake might not be enough to meet the pathogen's requirement of the vitamin.

In conclusion, the biosynthesis pathways for vitamin B1 (thiamine) and B6 (pyridoxal 5′ -phosphate) may be interesting molecular targets for the development of new chemical probes aiming to inhibit the synthesis of these essential cofactors in pathogenic organisms. Model compounds, mainly based on substrate analogs showed that inhibition of the vitamin B1 biosynthetic enzymes, ThiD and ThiE, as well as Pdx1, Pdx2, and PdxK in the pyridoxal 5′ -phosphate biosynthesis pathway have shown an initial proof of concept for the model, and it is up to the scientific community and/or researcher in the industry to explore these targets further.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding authors.

## AUTHOR CONTRIBUTIONS

CW, IP, and AN conceived the project. LM and RG analyzed the thiamine pathway, while AB and LD analyzed the pyridoxal pathway. All authors wrote and approved the manuscript.

## FUNDING

Financial support was provided by Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) through grants 2017/18173-0, 2015/26722-8, and 2015/13684-0 and by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), through grants 303165/2018-9 and 406936/2017-0.

## ACKNOWLEDGMENTS

The authors thank the funding agencies FAPESP and CNPq. We are also indebted to Maria Auxiliadora M.

## REFERENCES


Santos, Lívia Regina M. Margarido, Josimar Sartori, and João Fernando Possatto for their technical support. This manuscript has been released as a Pre-Print at BioRxiv (53).


**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 Barra, Dantas, Morão, Gutierrez, Polikarpov, Wrenger and Nascimento. 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.

# In vitro Activity of a New Fourth-Generation Cephalosporin, Cefoselis, Against Clinically Important Bacterial Pathogens in China

#### Jing-Wei Cheng1,2† , Jian-Rong Su<sup>2</sup>† , Meng Xiao1,3, Shu-Ying Yu1,3,4, Ge Zhang1,3 , Jing-Jia Zhang1,3, Yang Yang1,3, Si-Meng Duan1,3, Timothy Kudinha5,6, Qi-Wen Yang1,3 \* and Ying-Chun Xu1,3 \*

<sup>1</sup> Department of Clinical Laboratory, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing, China, <sup>2</sup> Center of Clinical Laboratory, Beijing Friendship Hospital, Capital Medical University, Beijing, China, <sup>3</sup> Beijing Key Laboratory for Mechanisms Research and Precision Diagnosis of Invasive Fungal Diseases (BZ0447), Beijing, China, <sup>4</sup> Graduate School, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing, China, <sup>5</sup> School of Biomedical Science, Charles Sturt University, Orange, NSW, Australia, <sup>6</sup> Centre for Infectious Diseases and Microbiology Laboratory Services, ICPMR – Pathology West, Westmead Hospital, The University of Sydney, Sydney, NSW, Australia

The objective of this study was to systematically evaluate the in vitro activity of cefoselis and other comparators against common bacterial pathogens collected from 18 hospitals across China. Minimum inhibitory concentrations (MICs) were determined by the broth microdilution method following Clinical and Laboratory Standards Institute (CLSI) guidelines. Cefoselis showed poor activity against extended-spectrum β-lactamase (ESBL)-producing Escherichia coli, Klebsiella pneumoniae, and Proteus mirabilis, with susceptibility rates of < 10% each, while the susceptibility rates of this antibiotic against non-ESBL-producing strains of these organisms were 100%, 94.3%, and 97.0%, respectively. Cefoselis exhibited susceptibility rates of 56.7–83.3% against other tested Enterobacteriaceae isolates. For Acinetobacter baumannii and Pseudomonas aeruginosa isolates, the susceptibility rates to cefoselis were 18.7% and 73.3%, respectively. All methicillin-resistant Staphylococcus aureus (MRSA) strains were resistant to cefoselis, while all methicillin-sensitive S. aureus (MSSA) strains were susceptible to this antibiotic. In conclusion, cefoselis showed good activity against non-ESBL-producing E. coli, K. pneumoniae, and P. mirabilis, MSSA, and was also potent against Enterobacteriaceae, P. aeruginosa, and Streptococcus.

Keywords: cefoselis, cefepime, antimicrobial resistance, in vitro activity, China

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### Reviewed by:

Norma Margarita de la Fuente Salcido, Universidad Autónoma de Coahuila, Mexico Helio S. Sader, JMI Laboratories, United States

#### \*Correspondence:

Qi-Wen Yang yangqiwen81@vip.163.com Ying-Chun Xu xycpumch@139.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 07 August 2019 Accepted: 24 January 2020 Published: 28 February 2020

#### Citation:

Cheng J-W, Su J-R, Xiao M, Yu S-Y, Zhang G, Zhang J-J, Yang Y, Duan S-M, Kudinha T, Yang Q-W and Xu Y-C (2020) In vitro Activity of a New Fourth-Generation Cephalosporin, Cefoselis, Against Clinically Important Bacterial Pathogens in China. Front. Microbiol. 11:180. doi: 10.3389/fmicb.2020.00180

## INTRODUCTION

fmicb-11-00180 February 26, 2020 Time: 18:8 # 2

Multidrug-resistant (MDR) pathogens, especially the ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter species), are the leading cause of nosocomial infections throughout the world, which is usually caused by excessive drug usage or prescription and inappropriate use of antimicrobials. Understanding the mechanisms, developing novel antimicrobial agents, and knowing the latest antimicrobial resistance patterns of bacterial pathogens are crucial to combat these public health challenges (Santajit and Indrawattana, 2016; Sheu et al., 2018; Gajdacs, 2019).

Cefoselis is a member of the fourth-generation cephalosporins which exhibit a wider antibacterial spectrum activity than the third-generation cephalosporins to both Gram-negative and Gram-positive bacteria (King et al., 1995). The wide antibacterial spectrum of cefoselis is attributed to the resistance to hydrolysis by the chromosomal β-lactamases and the rapid penetration through the bacterial cell wall (Giamarellos-Bourboulis et al., 2000). However, few reports have been published in China on the antimicrobial activity of cefoselis against common bacterial pathogens. The objective of this study was to better understand the in vitro activity of cefoselis against common Gram-positive and Gram-negative bacterial pathogens in China.

## MATERIALS AND METHODS

## Ethics

The protocol was approved by the Human Research Ethics Committee of Peking Union Medical College Hospital (no. S-K262). Peking Union Medical College Hospital did not require written informed consent from participants because this was an in vitro study on bacteria isolates without any private data of the human participants.

## Clinical Isolates

A total of 1188 bacterial isolates derived from 18 hospitals in China (January 2014–December 2016) were studied. The bacterial species distribution is shown in **Table 1**. The isolates from each of the participating hospitals were re-identified by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) (Bruker Daltonics GmbH, Bremen, Germany) at the Central Lab, Peking Union Medical College Hospital (Beijing), China.

## Antimicrobial Susceptibility Test Method

Minimum inhibitory concentrations (MICs) were determined by the broth microdilution method following Clinical and Laboratory Standards Institute (CLSI) guidelines. Thirty-two antimicrobial agents were tested against the isolates, among which 17 agents were against Gramnegative organisms, 24 agents against Staphylococcus spp., and 19 agents against Streptococcus spp. Cefoselis was obtained from Hansoh Pharma, and the other agents were

TABLE 1 | Distribution of bacterial species.


ESBL, extended-spectrum β-lactamase; MRSA, methicillin-resistant S. aureus; MSSA, methicillin-sensitive S. aureus; PSSP, penicillin-susceptible S. pneumoniae; PRSP, penicillin-resistant S. pneumoniae.

provided by AstraZeneca. Interpretation of the antimicrobial testing results was based on CLSI M100-S28 (CLSI, 2018). Escherichia coli ATCC 25922, P. aeruginosa ATCC 27853, K. pneumoniae ATCC 700603, S. aureus ATCC 29213, and Streptococcus pneumoniae ATCC 49619 were used as the quality control strains.

## Extended-Spectrum β-Lactamase Detection

Phenotypic identification of extended-spectrum β-lactamase (ESBL) production in E. coli, K. pneumoniae, and Proteus mirabilis, was carried out using CLSI-recommended methods. If the cefotaxime or ceftazidime MICs were ≥ 2 µg/ml, the MICs of cefotaxime + clavulanic acid (4 µg/ml) or ceftazidime + clavulanic acid (4 µg/ml) were comparatively determined. ESBL production was defined as an eightfold or greater decrease in MICs for cefotaxime or ceftazidime tested in combination with clavulanic acid compared to their MICs without clavulanic acid.

## RESULTS

## In vitro Activity of Antimicrobial Agents Against Enterobacteriaceae

Against ESBL-producing E. coli, K. pneumoniae, and P. mirabilis isolates, cefoselis, cefepime, cefotaxime, and ceftriaxone showed relatively low susceptibility rates, with drug resistance rates of > 87%. Against non-ESBL-producing strains, most



ESBL, extended-spectrum β-lactamase; R, resistant; S, sensitive.

TABLE 3 | In vitro activity of antimicrobial agents against Enterobacteriaceae strains.


R, resistant; S, sensitive.

antibiotics revealed good activity, of which cefoselis, cefepime, showed > 94% antimicrobial susceptibility rates. For Citrobacter freundii, Enterobacter aerogenes, Enterobacter cloacae, Serratia marcescens, and Proteus vulgaris isolates, the susceptibility rates for cefoselis ranged from 56.7% to 83.3%, which were slightly lower than that of cefepime, with susceptibility rates ranging from 70% to 100%. Meropenem and amikacin exhibited high activity against all the Enterobacteriaceae strains (**Tables 2**, **3**).

## In vitro Activity of Antimicrobial Agents Against Non-fermentative Gram-Negative Organisms

The most active agents against A. baumannii were tigecycline and minocycline, with susceptibility rates of 58.6% and 45.5%, respectively. The other analyzed agents were less effective, with susceptibility rates of < 30%. Furthermore, against P. aeruginosa

TABLE 4 | In vitro activity of antimicrobial agents against Acinetobacter baumannii and Pseudomonas aeruginosa strains.

Antimicrobial agents A. baumannii (198) P. aeruginosa (30) %R %S %R %S Piperacillin/tazobactam 75.3 23.2 13.3 80.0 Ceftazidime 76.8 20.7 20.0 73.3 Ceftriaxone 78.8 7.1 96.7 0 Cefotaxime 78.8 14.6 100.0 0 Cefoselis 80.8 18.7 26.7 73.3 Cefepime 75.8 19.7 20.0 73.3 Cefoxitin 96.5 2.5 96.7 0 Aztreonam 84.3 5.1 20 73.3 Ertapenem – – – – Imipenem 75.3 24.7 20.0 80.0 Meropenem 75.3 24.2 10.0 76.7 Amikacin 69.7 29.8 6.7 93.3 Ciprofloxacin 76.8 21.2 16.7 76.7 Levofloxacin 59.6 23.2 16.7 76.7 Minocycline 28.3 45.5 – –

R, resistant; S, sensitive.

isolates, amikacin exhibited the highest in vitro activity, with a susceptibility rate of 93.3%. The susceptibility rates for cefoselis, cefepime, and ceftazidime were all 73.3% each for this organism (**Table 4**).

Tetracycline 81.3 16.2 – – Tigecycline 22.7 58.6 96.7 3.3

## In vitro Activity of Antimicrobial Agents Against MRSA and MSSA

Against methicillin-resistant S. aureus (MRSA) strains, linezolid, vancomycin, and teicoplanin exhibited a susceptibility rate of 100% each, followed by tigecycline (97.9%), and trimethoprim–sulfamethoxazole (TMP-SMX) (94.8%). All strains were resistant to ceftazidime, ceftriaxone, cefoselis, and cefepime. Against 100 methicillin-sensitive S. aureus (MSSA) strains, most antibiotics showed good activity, except for ampicillin, tetracycline, and erythromycin. All strains were susceptible to ceftazidime, ceftriaxone, cefoselis, and cefepime (**Table 5**).

## In vitro Activity of Antimicrobial Agents Against Streptococcus Strains

For the penicillin-susceptible S. pneumoniae (PSSP), betahemolytic Streptococcus strains, and viridans group Streptococcus strains, cefoselis and cefepime both showed very high antimicrobial activities, with susceptibility rates of > 90%. Against 15 penicillin-resistant S. pneumoniae (PRSP) strains, the susceptibility rate of cefoselis was higher than that of cefepime (60.0% vs. 40.0%). Linezolid, vancomycin, and tigecycline exhibited 100% antimicrobial activity against all the Streptococcus strains (**Table 6**).


TABLE 5 | In vitro activity of antimicrobial agents against MRSA and MSSA strains.

MRSA, methicillin-resistant S. aureus; MSSA, methicillin-sensitive S. aureus; R, resistant; S, sensitive.

## Comparison of Cefoselis and Cefepime Against Common Clinical Pathogens

Cefoselis exhibited a slightly lower antimicrobial activity than cefepime against Enterobacteriaceae and non-fermentative Gram-negative organisms, but a little higher activity than cefepime against MRSA, MSSA, PSSP, beta-hemolytic Streptococcus, and viridans group Streptococcus strains. The cumulative percentage MIC distributions of cefoselis and cefepime against common clinical pathogens are shown in **Table 7**.

## DISCUSSION

The Enterobacteriaceae family is a major group of pathogens causing several community- and hospital-acquired infections, among which the ESBL rates in E. coli and K. pneumoniae in China have been reported as high as 60–70% and 30– 40%, respectively (Yang et al., 2010, 2013). According to our previous study, the genotype distribution of ESBLproducing strains among bacterial species was diverse, and blaCTX-M was the major ESBL gene, with occurrences in 99.5% of E. coli, 91.1% of K. pneumoniae, and 97.5% of P. mirabilis strains (Yang et al., 2015). In the present

TABLE 6 | In vitro activity of antimicrobial agents against Streptococcus strains.


PSSP, penicillin-susceptible S. pneumoniae; PRSP, penicillin-resistant S. pneumoniae; R, resistant; S, sensitive.

study, cefoselis and cefepime both showed poor activities against ESBL-producing E. coli, K. pneumoniae, and P. mirabilis, which may be attributed to the specific ESBL genes present in China, albeit further studies are needed for confirmation.

Against other Enterobacteriaceae strains, cefoselis exhibited a slightly lower antimicrobial activity than cefepime, but a higher activity than third-generation cephalosporins. The relatively high activities of fourth-generation cephalosporins against Enterobacteriaceae may be attributed to the low affinity for chromosome-mediated AmpC β-lactamases (D'Angelo et al., 2016), which are the common β-lactamases in Enterobacteriaceae isolates from China. A multicenter, double-blind, randomized clinical trial in China revealed equal clinical efficacy and safety of intravenous cefoselis and cefepime injection for the treatment of acute, moderate, and severe bacterial infections (Liu et al., 2014).

Acinetobacter baumannii was one of the bacteria considered to be of maximum resistance and is classified as a priority category according to the bacterial groups classified by priority categories of need for new antibiotics (Tacconelli and Magrini, 2017). In this study, A. baumannii exhibited low susceptibility to most of the tested antibiotics, with susceptibility rates ranging from 5.1 to 58.6%. Against P. aeruginosa, cefoselis and cefepime showed equal antimicrobial activities for this organism, with susceptibility rates of 73.3% each, which are slightly higher than those in a previous study in China; thus, these two antibiotics could be used to treat infections caused by P. aeruginosa, in combination with other antibiotics (Zhang et al., 2016).

Although the prevalence of MRSA in China showed a markedly decreasing trend from 69.0% in 2005 to 35.3% in 2017, as per the China Antimicrobial Surveillance Network (CHINET) program, MRSA remains a major pathogen responsible for nosocomial infections (Hu et al., 2018). No isolates were found to be resistant to vancomycin, linezolid, and teicoplanin in this study. Tigecycline and TMP-SMX also showed good activities, which were similar to previous studies (Zhao et al., 2012; Zhang et al., 2015). Vancomycin, linezolid, and TMP-SMX were recommended by the Infectious Diseases Society of America (IDSA) to treat MRSA infections (Liu et al., 2011).

Teicoplanin can be an effective alternative to vancomycin for treating patients infected by MRSA as the two therapies are similar in both efficacy and safety (Peng et al., 2013). Tigecycline was often recommended as a second- or thirdline agent for MRSA infections when alternative agents cannot be used (Rodvold and McConeghy, 2014). Ceftaroline fosamil was the first FDA-approved cephalosporin with any activity against MRSA, but the low susceptibility among MRSA isolates in China needs attention (Lodise and Low, 2012; Zhang et al., 2015).

The bacterial isolates were collected from 2014 to 2016, and the susceptibility has certainly changed in the last 5 years for most organisms. More recently collected strains should be involved in further studies. This is a limitation of the study. In conclusion, cefoselis exhibited good antimicrobial activity against non-ESBL[Frame1]-producing E. coli, K. pneumoniae, P. mirabilis, and MSSA and was also potent against other Enterobacteriaceae, P. aeruginosa, and Streptococcus.

TABLE 7 |Cumulativepercentage MIC distributions of cefoselis and cefepime against common clinical pathogens collected in China.


methicillin-resistant S. aureus; MSSA, methicillin-sensitive S. aureus; PSSP, penicillin-susceptible S. pneumoniae; PRSP, penicillin-resistant S. pneumoniae; MIC, minimum fmicb-11-00180 February 26, 2020 Time: 18:8 # 6

Cheng et al.

ESBL,

inhibitory concentration.

extended-spectrum

 aMIC90 values are in boldface.

β-lactamase;

 MRSA,

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## AUTHOR CONTRIBUTIONS

fmicb-11-00180 February 26, 2020 Time: 18:8 # 7

J-WC, J-RS, and TK wrote the manuscript. S-YY, GZ, J-JZ, YY, and S-MD collaborated the strains and performed the antimicrobial susceptibility tests. MX, Q-WY, and Y-CX designed and supervised the study.

## FUNDING

This work was supported by the National Key Research and Development Program of China (2018YFE0101800,

## REFERENCES


2018YFC1200100, and 2018YFC1200105), Chinese Academy of Medical Sciences (CAMS) Initiative for Innovative Medicine (Grant No. 2016-I2M-3–014), CAMS Innovation Fund for Medical Sciences (Grant No. 2016-I2M-1–014), Beijing Innovation Base Cultivation and Development Special Fund (Z171100002217068), Outstanding Talents Training Funding Project of Dongcheng District, Beijing (2017), Graduate Innovation Fund of Peking Union Medical College (Grant No. 2017-1002-1-21), and the National Natural Science Foundation of China (Grant No. 81101287). The funders had no role in the study design, data collection and analysis, decision to publish, or in the preparation of the manuscript.

## ACKNOWLEDGMENTS

We thank all the 18 hospitals for providing the study strains.

producing Enterobacteriaceae: current evidence and future prospects. Expert Rev. Anti Infect. Ther. 16, 205–218. doi: 10.1080/14787210.2018.1436966


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

# Metadata Analysis of mcr-1-Bearing Plasmids Inspired by the Sequencing Evidence for Horizontal Transfer of Antibiotic Resistance Genes Between Polluted River and Wild Birds

#### Edited by:

Raffaele Zarrilli, University of Naples Federico II, Italy

#### Reviewed by:

Ruichao Li, Yangzhou University, China Stefano Gaiarsa, San Matteo Hospital Foundation (IRCCS), Italy

#### \*Correspondence:

Kun Yang cookyoung@scu.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 15 September 2019 Accepted: 18 February 2020 Published: 10 March 2020

#### Citation:

Lin Y, Dong X, Wu J, Rao D, Zhang L, Faraj Y and Yang K (2020) Metadata Analysis of mcr-1-Bearing Plasmids Inspired by the Sequencing Evidence for Horizontal Transfer of Antibiotic Resistance Genes Between Polluted River and Wild Birds. Front. Microbiol. 11:352. doi: 10.3389/fmicb.2020.00352 Yufei Lin† , Xiaohong Dong† , Jiao Wu† , Dawei Rao, Lihua Zhang, Yousef Faraj and Kun Yang\*

Department of Pharmaceutical and Biological Engineering, School of Chemical Engineering, Sichuan University, Chengdu, China

We sequenced the whole genomes of three mcr-1-positive multidrug-resistant E. coli strains, which were previously isolated from the environment of egret habitat (polluted river) and egret feces. The results exhibit high correlation between antibiotic-resistant phenotype and genotype among the three strains. Most of the mobilized antibiotic resistance genes (ARGs) are distributed on plasmids in the forms of transposons or integrons. Multidrug-resistant (MDR) regions of high homology are detected on plasmids of different E. coli isolates. Therefore, horizontal transfer of resistance genes has facilitated the transmission of antibiotic resistance between the environmental and avian bacteria, and the transfer of ARGs have involved multiple embedded genetic levels (transposons, integrons, plasmids, and bacterial lineages). Inspired by this, systematic metadata analysis was performed for the available sequences of mcr-1-bearing plasmids. Among these plasmids, IncHI2 plasmids carry the most additional ARGs. The composition of these additional ARGs varies according to their geographical distribution. The phylogenetic reconstruction of IncI2 and IncX4 plasmids provides the evidence for their multiregional evolution. Phylogenetic analysis at the level of mobile genetic element (plasmid) provides important epidemiological information for the global dissemination of mcr-1 gene. Highly homologous mcr-1-bearing IncI2 plasmids have been isolated from different regions along the East Asian-Australasian Flyway, suggesting that migratory birds may mediate the intercontinental transportation of ARGs.

Keywords: antibiotic resistance, intercontinental dissemination, migratory birds, antibiotic resistance gene, horizontal gene transfer

## INTRODUCTION

fmicb-11-00352 March 6, 2020 Time: 17:19 # 2

Horizontal gene transfer (HGT) plays an important role in the global dissemination of antibiotic resistance, while the mobilization of ARGs is the first and also the most important evolutionary step for their horizontal transfer. The discovery of any new mobilized resistance gene always attracts great attention of the researchers. As evidenced by a number of studies, these mobilized resistance genes have been increasingly spreading all over the world (Stokes and Gillings, 2011). Typical cases include the discovery of blaKPC, blaNDM, and mcr-1 genes. The blaKPC gene was first identified in 1996, and now it can be detected in many regions of the world (Munoz-Price et al., 2013). Its mobilization is related to a 10 kb Tn3-based mobile transposon Tn4401 (Naas et al., 2008; Cuzon et al., 2011). The blaNDM gene was first identified in 2008 in a K. pneumoniae isolate recovered from a Swedish patient, who had previously been hospitalized in New Delhi, India (Yong et al., 2009). Thereafter, this resistance gene was detected in different regions of the world (Dortet et al., 2014). Its mobilization is associated with an ISAba125 composite transposon Tn125 (Poirel et al., 2012). The mcr-1 gene was first discovered in China (Liu et al., 2016), and the same ARG was then reported in various regions of the world (Wang et al., 2018). Its mobilization is mediated by an ISApl1 composite transposon Tn6330 (Snesrud et al., 2016; Li et al., 2017).

Although the effect of HGT on the dissemination of antibiotic resistance has been realized, the spread of specific resistant bacterial clones is also widely concerned (Munoz-Price et al., 2013; Matamoros et al., 2017). In early 1990s, it was found that the moving of transposons between carrier replicons (plasmids) resulted in the spread of ARGs between bacterial species (Liebert et al., 1999). Recently, it has been described as a nested Russian doll-like mobility of ARGs, which once again drew the attention of researchers (Sheppard et al., 2016; Wang et al., 2018; Hernando-Amado et al., 2019). These studies clearly explain the mechanism of the horizontal transfer of resistance genes at multiple genetic levels (transposons, integrons, plasmids, bacterial lineages and bacterial species) (Wang et al., 2018). In our previous study (Wu et al., 2018), we demonstrated that wild birds could transport antibiotic resistance from contaminated river to the surrounding environment, and the spread of antibiotic resistance was not mainly due to the transfer of resistant bacterial clones. Among those resistant E. coli isolates in the aforementioned work, no strong correlation was observed between strain genotypes (repetitive-element PCR genotyping) and their drugresistance patterns. Therefore, we concluded that the horizontal transfer of resistance genes was the main mechanism of resistance transmission in that process. Nonetheless, there was no direct sequencing evidence at the time. In this study, we sequenced the whole genomes of three mcr-1-positive multidrugresistant E. coli strains isolated in the previous work to pursue the evidence for horizontal transfer of resistance genes. Inspired by the sequencing results, we systematically analyzed mcr-1-bearing plasmid sequences downloaded from GenBank database. We tried to obtain epidemiological information of the mcr-1 gene by analyzing its global dissemination at multiple levels of mobile genetic elements (MGEs), especially at the plasmid level.

## MATERIALS AND METHODS

## mcr-1-Positive Strains

In our previous study, we found that the wild birds (egrets) could mediate the environmental transmission of antibiotic resistance in local area, from the polluted Jin River to birds' nightly habitat, Wangjianglou Park (Wu et al., 2018). Some of the E. coli strains isolated in that work exhibited colistin resistance (n = 6), among which three isolates were proved to be mcr-1-positive and multidrug-resistant. Two of the three strains were isolated from the river (Jin River) polluted by antibiotic-resistant bacteria and the third one was isolated from egret feces. Three E. coli isolates were respectively labeled as W5-6, W2-5, and BE2-5.

## Antibiotic Susceptibility Testing

The minimum inhibitory concentrations (MICs) of 10 antibiotics against the three E. coli isolates has previously been determined (Wu et al., 2018) via a modified broth micro-dilution method as per ISO 20776-1:2006 CLSI using 96-well microtiter plates (Clinical and Laboratory Standards Institute [CLSI], 2019). The only modification in our method is in the step of inoculum preparation. Typically, the E. coli isolates were first grown in LB-broth-loaded 96 well plate at 37◦C overnight to reach the stationary phase. The cell cultures were then 10<sup>3</sup> fold diluted and used as inoculum. When inoculating the testing 96-well plates containing antibiotics of different concentrations, a 48-pin replicator was used to improve the efficiency of experimental operation. The MIC endpoints were determined as the lowest concentration, at which there was no visible growth after 20 h of incubation at 37◦C. Duplicate tests for each antibiotic concentration were conducted. Positive and negative controls were conducted in antibiotic-free LB to ensure the growth of environmental E. coli isolates under lab conditions and sterility of the assay, respectively. Quality control of the procedure was conducted by using the susceptive E. coli standard strain ATCC 25922.

## Bacteria Culture and Whole Genomic DNA Extraction

The whole genome DNA was extracted from fresh bacterial cell mass recovered from 100 mL LB pure culture containing colistin of 4 µg/mL with the FastDNA <sup>R</sup> Spin Kit for Soil (MP Biomedicals, France). About 1.2 mL of high-quality DNA was obtained for each E. coli strain. A small amount of each DNA sample was used for the amplification and sequencing of 16S rRNA gene to ensure that the DNA sample was obtained from a pure culture. The presence of mcr-1 gene in the DNA sample was also verified via PCR. The primer pair for amplifying 16S rRNA gene was 27F (5<sup>0</sup> -AGAGTTTGATCCTGGCTCAG-3 0 ) and 1522R (5<sup>0</sup> -AAGGAGGTGATCCANCCRCA-3<sup>0</sup> ) and that for mcr-1 was CLR5-F (5<sup>0</sup> -CGGTCAGTCCGTTTGTTC-3<sup>0</sup> ) and CLR5-R (5<sup>0</sup> -CTTGGTCGGTCTGTAGGG-3<sup>0</sup> ) (Liu et al., 2016).

Thereafter, the remaining bacterial DNA samples were kept frozen (−40◦C) until sequencing.

## Whole Genome Sequencing for Three E. coli Isolates

The bacterial genome DNA was detected by agarose gel electrophoresis and quantified by Qubit. The genomes of the three isolates were sequenced using Single Molecule, Real-Time (SMRT) technology performed at Beijing Novogene Bioinformatics Technology Co., Ltd. The low quality reads were filtered by the SMRT 2.3.0 (Berlin et al., 2015), and the filtered reads were de novo assembled to generate contigs without gaps (Koren and Phillippy, 2015). All genome sequences (chromosome and plasmid sequences) of the three E. coli isolates were deposited into GenBank database under the BioProject PRJNA495707 with BioSample numbers SAMN10230266 to SAMN10230268 and accession numbers CP032986 to CP032995.

The strain types (STs) of the three E. coli isolates were determined from their assembled genomes using online MLST service provided by the Center for Genomic Epidemiology<sup>1</sup> according to Achtman's MLST scheme (Wirth et al., 2006; Larsen et al., 2012).

## Prediction of Open Reading Frames (ORFs) and Gene Functions

We used GeneMarks to predict ORFs for the three E. coli isolates (Besemer et al., 2001). We used seven databases to predict gene functions. They were GO (Gene Ontology) (Ashburner et al., 2000), KEGG (Kyoto Encyclopedia of Genes and Genomes) (Kanehisa et al., 2004, 2006), COG (Clusters of Orthologous Groups) (Tatusov et al., 2003), NR (Non-Redundant Protein Database) (Li et al., 2002), TCDB (Transporter Classification Database) (Saier et al., 2014), Swiss-Prot (Bairoch and Apweiler, 2000), and TrEMBL (Magrane and UniProt, 2011), respectively. A whole genome BLAST search (E-value less than 10−<sup>5</sup> , minimal alignment length percentage larger than 40%) was performed against above seven databases (Altschul et al., 1990). ARGs were annotated via BLAST searching the ORFs against the Comprehensive Antibiotic Resistance Database (CARD) (Liu and Pop, 2009; McArthur et al., 2013). We set the thresholds of e-value and "best identity" at 10−<sup>5</sup> and 80%, respectively.

## Epidemiological and Phylogenetic Analysis of the mcr-1-Bearing Plasmids

The discovery of the mobilized colistin resistance gene, mcr-1, is in the era of rapid development of next-generation sequencing technology. During only 4 years, hundreds of sequences of mcr-1-bearing plasmids have been uploaded to the database, which provides a good chance for metadata analysis of these plasmids. We retrieved the GenBank database with keywords "**mcr-1**" (or mcr1) and "**plasmid**" in June 2018. From the hit entries, the **circular** sequences were manually picked and recorded the metadata. Identification of the plasmid incompatibility group was performed for downloaded plasmid sequences via the CGE online

<sup>1</sup>https://cge.cbs.dtu.dk/services/MLST/

services PlasmidFinder v1.3<sup>2</sup> (Carattoli et al., 2014). ResFinder<sup>3</sup> was used to determine other acquired resistance genes on these plasmids (Zankari et al., 2012).

Most of the published mcr-1-bearing plasmids belong to three main incompatibility groups, IncHI2, IncI2, and IncX4 (Li et al., 2017; Matamoros et al., 2017). According to the respective characteristics of the three incompatibility plasmid groups, we performed a targeted analysis. The analysis routine is illustrated in **Supplementary Figure S1**.

## RESULTS AND DISCUSSION

## Genome Assembly and Annotation for Three E. coli Isolates

The original sequencing data obtain more than 50 × coverage of the whole genome for the three E. coli isolates. Chromosomes and plasmids of the three isolates are all assembled in circular contigs with no gaps. Three E. coli isolates are of different strain types (**Table 1**). The number, size and Inc-types of plasmids are also different among the three isolates (**Table 1**). E. coli W5-6 contains 3 plasmids. The other two isolates both carry 2 plasmids. The plasmid pMCR\_W5-6 gives the biggest size (241 kbp) among all plasmids. The numbers of genes predicted by GeneMarks are 4898, 4834, and 4576 for W5-6, W2-5, and BE2-5, respectively. The annotation of the three genomes with the COG database clusters these genes into 23 classes with annotation rate of 89.6, 90.7, and 93.8%, respectively (**Supplementary Figure S2**). Genomes of all the three E. coli strains, especially the strain W5-6, contain a considerable number of MGEs (Mobilome in **Supplementary Figure S2**).

## Antibiotic-Resistant Phenotype and Genotype of Three E. coli Isolates

All the three strains show multidrug resistance. The E. coli W5- 6 demonstrates the highest level of drug resistance among the three isolates. It shows resistance against 9 out of 10 tested antibiotics. The other two E. coli isolates of lower resistance levels are resistant to 7 antibiotics (**Table 2**).

Eighty-five putative ARGs (via BLAST against the CARD database) are annotated in the whole genome of E. coli W5- 6, 20 of which are on plasmids. The numbers of ARGs on chromosomes of E. coli W2-5 and BE2-5 are 65 and 66, respectively. The numbers of ARGs on plasmids of these two isolates are 3 and 11, respectively (**Table 3**). Fifty-two ARGs are common among all the three E. coli isolates. However, most of these shared ARGs (n = 44, 84.6%) are efflux pump genes, and they are distributed on their chromosomes, which implies that efflux pump genes are highly conserved in the same bacterial species. These efflux pump genes do not appear to be significantly associated with the antibiotic-resistant phenotype of the three strains. Although these efflux pump genes may be just involved in some detoxification process (Martinez et al., 2009), we still list them below (**Table 3**) for reference. The

<sup>2</sup>https://cge.cbs.dtu.dk/services/PlasmidFinder/

<sup>3</sup>https://cge.cbs.dtu.dk/services/ResFinder/

TABLE 1 | Genome assembly results for three E. coli isolates.


<sup>a</sup>The plasmids bearing mcr-1 gene. <sup>b</sup>The plasmids with MDR region of similar structure. <sup>c</sup>Strain type (ST) or incompatibility replicon type of plasmids (Inc). Multiple replicons on the same plasmid are indicated in bracket.

only non-efflux-pump common ARG with a definite resistant phenotype among the three isolates is the colistin resistance gene mcr-1. The mcr-1 gene is located on the plasmids of different Inc-types in the three E. coli strains. The plasmids pMCR\_W5-6, pMCR\_W2-5, and pMCR\_BE2-5 are of the IncHI2, IncI2, and IncP1-type, respectively (**Table 1**). The mcr-1 sequence on plasmid pMCR\_W2-5 lost the downstream ISApl1 (**Supplementary Figure S3**).

For Aminoglycosides, resistance genes encoding three types of antibiotic inactivation enzymes are detected among the three E. coli isolates, aminoglycoside acetyltransferases [aac(3)- IV], phosphotransferase [aph(3<sup>0</sup> )-Ia, aph(4)-Ia, aph(300)-Ib and aph(6)-Id] and nucleotidyltransferase (aadA1 and aadA2) genes. E. coli W5-6 possesses all the three kinds of resistance genes (**Table 2**), and exhibits the highest resistance level to aminoglycosides. It is resistant to three out of four tested aminoglycosides: kanamycin A, gentamicin and streptomycin (**Table 2**). E. coli W2-5 has the ARGs, which render resistance to streptomycin [aph(300)-Ib and aph(6)-Id], and exhibits resistance to this antibiotic. E. coli BE2-5 does not have these two ARGs, while it has aph(3<sup>0</sup> )-Ia, which enables it to resist against kanamycin A (McArthur et al., 2013). For tetracycline, we detected corresponding ARGs in the genomes of all the three E. coli isolates either on chromosome [tet(A), tet(B) and tet(D)] or on plasmid [tet(A)], which is in good agreement with their resistance to this antibiotic (**Tables 2**, **3**). For β-lactams, different β-lactamase genes are detected on the plasmids of E. coli W5-6 (blaCTX−M−<sup>14</sup> on pMCR\_W5-6 and blaTEM−<sup>1</sup> on p2\_W5-6) and W2-5 (blaCTX−M−<sup>55</sup> on p2\_W2-5), while no β-lactamase gene is detected for E. coli BE2-5, as it is susceptible to ceftriaxone (**Table 2**). For quinolones, genes encoding subunits of efflux pump complex conferring resistance to fluoroquinolone (oqxA and oqxB) (Kim et al., 2009) are detected on the plasmid (pMCR\_W5-6) of E. coli W5-6. Point mutations in gyrA and parC genes conferring resistance to fluoroquinolones are detected in E. coli W5-6 and W2-5 (Oram and Fisher, 1991; Tankovic et al., 1996). In E. coli BE2-5, qnrS2 gene is detected on its plasmid (p2\_BE2-5), which confers it weak resistance to ciprofloxacin and nalidixic acid (Gay et al., 2006; **Table 2**). Besides above ARGs conferring resistance to the tested antibiotics, resistance genes against sulfonamides (sul2 etc.), chloramphenicol (cmlA1), fosfomycin (fosA3), and trimethoprim (drfA12) are also detected among the three E. coli isolates. Point mutation in glpT gene conferring resistance to fosfomycin is detected in all three E. coli isolates (Takahata et al., 2010; **Table 3**). Overall, the antibioticresistant genotype is in good agreement with the resistant phenotype among the three E. coli isolates. The only exception is that no corresponding resistance gene is detected to explain the unexpected resistance of E. coli BE2-5 against ampicillin.

## Evidence for the Horizontal Transfer of ARGs Between Environmental and Avian E. coli

Almost all the ARGs detected in the avian E. coli (BE2-5) are found in two other environmental strains (W2-5 and W5-6, **Table 3**), which is consistent with our previous view that the antibiotic resistance (resistant bacteria or resistance genes) is mainly transferred from the polluted river (Jin River) to the wild bird (egret) (Wu et al., 2018). Most importantly, highly homologous MDR regions are detected on two plasmids of different Inc-types (pMCR\_W5-6 and p2\_BE2-5) in E. coli W5- 6 and BE2-5, respectively (**Figures 1A,B**). The sizes of the MDR regions are around 12,800 bp (**Figure 1D**). Although the size and overall structure of the two plasmids are different (**Figures 1A,B**), the composition of the ARGs and MGE sequences (gene sequences encoding transposase and integrase) and their arrangement in the MDR regions are almost the same (**Figure 1C** and **Supplementary Figure S4**). The only difference is that the MDR region on the plasmid pMCR\_W5- 6 missing an efflux pump gene mef(B) in comparison to the plasmid p2\_BE2-5. On the plasmid pMCR\_W5-6, there even remains a 33 nt 3<sup>0</sup> fragment of the mef(B) gene. This provides sequencing evidence for gene deletion and recombination during the evolution of the MDR region. The shared ARGs includes the genes conferring resistance to aminoglycosides [aadA1, aadA2 and aph(3<sup>0</sup> )-Ia], sulfonamides (sul3), trimethoprim (drfA12) and chloramphenicols (cmlA1). The shared MGE sequences, especially the IS26 sequences that are detected more than 5 copies on both plasmids (**Figures 1A,B** and **Supplementary Figure S4**),


should play an important role during the formation of the MDR regions, and provide conditions for the relocation of these ARGs. Since the highly homologous MDR regions are detected on the plasmids of different Inc-types and in the E. coli of different strain types, the evolution of the MDR regions must have involved the process of HGT. The identical drug-resistant transposon containing aph(3<sup>0</sup> )-Ia

gene is also found on the chromosome (W5-6Chr) and plasmids (pMCR\_W5-6 and p2\_BE2-5) of different bacteria. Similarly, the resistance gene [aph(3<sup>0</sup> )-Ia] is also flanked with transposase genes (tnpA IS26) on both sides (**Supplementary Figure S5**). The ARGs on other plasmids are illustrated in **Supplementary Figure S6**. Most of these ARGs also show close correlation with MGEs.

The spread of antibiotic resistance is closely related to the horizontal transfer of resistance genes, regardless in the clinical or in a natural environment, which is highly polluted with antibiotics or antibiotic resistant bacteria (Stokes and Gillings, 2011). In our case, Jin River (an urban river in Chengdu, Sichuan, China) is highly polluted with antibiotic resistant bacteria, and the avian inhabitants (egrets) there are also highly affected. The resistance genes conferring antibioticresistant phenotype in the three E. coli isolates are mainly distributed on plasmids or on chromosome, but in forms of transposons (**Supplementary Figure S5**). The resistance genes on plasmids are also closely related to MGE sequences (transposase and integrase genes on plasmids as shown in **Figure 1** and **Supplementary Figure S6**). From environmental (polluted river water) and avian (egret) E. coli of different strain types, we have detected almost identical MDR regions on the plasmids of different incompatibility types. ARGs, which are flanked by MGE sequences, present as transposons or integrons in the MDR regions. Identical drug-resistance transposon [IS26-aph(3<sup>0</sup> )-Ia-IS26] is also found on chromosome and plasmids of these host bacteria. These results indicate that HGT plays a crucial role in the environmental dissemination of antibiotic resistance, and the transfer of ARGs must involve multiple embedded genetic levels (transposons, integrons, plasmids, and bacterial lineages), which is so-called the nested "Russian doll" model of genetic mobility (Sheppard et al., 2016).

## Global Dissemination and Multiregional Evolution of mcr-1 Plasmids

The metadata of 228 circular mcr-1-bearing plasmids were collected, as shown in **Supplementary Table S1**. The geographical distribution of plasmids of different incompatibility groups is depicted in **Supplementary Figure S7**. It is worth noting that the map outlines only the types of plasmid replicons on different continents. Some plasmids carry two or more (up to five) replicons (**Supplementary Table S1**). Therefore, a plasmid may have been counted more than once. The result still reflects the local diversity of plasmid types. Asia contributes the most types of mcr-1-bearing plasmids (13 out of 16). The other two continents in the Northern Hemisphere, Europe and North America also exhibit high diversity of plasmid types, 9 and 6, respectively. On the contrary, the regions in the Southern Hemisphere show poor plasmid

fmicb-11-00352 March 6, 2020 Time: 17:19 # 5

TABLE 3 | CARD database annotated ARGs in three E. coli isolates.


<sup>a</sup>A represents the common efflux pump genes in all three E. coli isolates, which include acrA, acrB, acrE, acrR, acrS, baeR, baeS, cpxA, cpxR, crp, emrA, emrB, emrD, emrE, emrK, emrR, emrY, evgA, evgS, gadW, gadX, hns, kdpE leuO, marA, mdfA, mdtA, mdtE, mdtF<sup>b</sup> , mdtG, mdtH, mdtL, mdtM, mdtN, mdtO, mdtP, mexN, msbA, msrB, patA, rob, tolC, yojI. <sup>b</sup>There are two copies of mdtF on the chromosome of W5-6, three copies of mdtF on both chromosomes of W2-5 and BE2-5, two copies of mfd on the chromosome of W2-5, and two copies of tet(B) and tet(D) on the chromosome of W5-6. <sup>c</sup>Specific point mutations in each gene are listed in the bracket. The reference sequences are from the genome of E. coli K-12 MG1655 (Accession No. NC\_000913.3).

diversity (**Supplementary Figure S7**). Sampling bias may have led to this discrepancy. However, twenty-eight mcr-1 bearing plasmids recovered from South America give only two plasmid types, IncX4 and IncI2. Up to some extent it reflects that the Southern Hemisphere (at least South America) suffers fewer invasions of various mcr-1-bearing plasmids. Among these mcr-1-bearing plasmids, the numbers of IncI2, IncX4 and IncHI2 plasmids are ranked at the top 3. Their detection rates in various Enterobacteriaceae hosts on different continents are illustrated in **Figure 2A**. IncI2 plasmids exhibit the most diverse bacterial hosts (in 6 unique host species of Enterobacteriaceae) and the most extensive geographical distribution (in all continents).

The IncHI2-type plasmid is the most diverse plasmid and harbors a large MDR region (Li et al., 2017). Therefore, it is considered as a genetic element mediating the transmission of MDR genes. A wide range of resistance genes and MGEs can be found in a Mosaic MDR region of IncHI2-type plasmids (Li et al., 2017). We recorded the additional ARGs on each IncHI2 plasmid (**Supplementary Table S1**). Co-occurrence network analysis, showing the correlation between these ARGs and their geographic distribution, was performed in Matlab\_R2016a. The composition of ARGs in this MDR region varies according to their geographical distribution (**Figure 2B**). Such an MDR region is also discovered in our sequenced IncHI2-type plasmid, pMCR\_W5-6 (**Figure 1B**). Unlike other ARGs concentrated in such an MDR region, mcr-1 gene is located in another location on the plasmid away from the MDR region (**Supplementary Table S1** also lists the interval between mcr-1 and other ARGs). It has a unique transposon structure, ISApl1-mcr-1 orf-ISApl1 (Snesrud et al., 2016; Li et al., 2017). The insertion of mcr-1-bearing transposon in such type of plasmid should be a late independent event. The exact time of the insertion event is hard to be determined, but the backbone structure of the IncHI2-type plasmid at the time of mcr-1 insertion can be speculated. The ARGs with high occurrence frequency should be the original backbone structure of this type of plasmids (**Figure 2B**). Besides mcr-1, two other ARGs, floR and aadA2, are common all over the world on this type of plasmids. It has been reported that the resistance (floR) to florfenicol – a veterinary drug – is commonly associated with mcr-1 (Matamoros et al., 2017; Shen et al., 2018), which supports the animal origin of the mobilized mcr-1 gene (Matamoros et al., 2017; Wang et al., 2018). The mcr-1-bearing IncHI2 plasmids recovered from Asia (mainly from China) possess the most diverse ARGs (**Figure 2C**), which must be due to the extensive use of antibiotics in this area (Zhang et al., 2015). The three unique ARGs on European IncHI2 plasmids are aadA12, blaTEM1A and catA1. This may be attributed to regional evolution events in Europe.

The IncX4-type plasmids are the most conserved and the smallest (mostly around 33 kb) ones among all types of mcr-1-bearing plasmids (Li et al., 2017). The sequences of IncX4 type plasmids were adjusted from the same start point and

transposase genes, tnpA IS26 etc.). The recombination event (mef(B) gene) in the MDR region is indicated in blue (A,C, on plasmid p2\_PE2-5). The mismatches between sequences were indicated with asterisks in (D). A BRIG-Easyfig version of this figure is shown in supporting information as Supplementary Figure S4.

aligned with the multiple sequence alignment program MAFFT<sup>4</sup> (Katoh et al., 2017). The aligned sequences were used to construct a phylogenetic tree in MEGA7 (Kumar et al., 2016). The IncI2 type plasmids contain a site-specific recombination system, the shufflon. The shufflon generates variants of the PilV protein, a minor component of the thin pilus. The shufflon is one of the most difficult regions for de novo genome assembly, because of its structural diversity even in a single bacterial colony (Sekizuka et al., 2017). Therefore, the shufflon structure affects the alignment of the plasmid sequences, and thus affects the phylogenetic reconstruction. For this type of plasmid (n = 92), we excised the shufflon region from each plasmid sequence and used the remaining part for phylogenetic analysis as we did for IncX4-type plasmids. From the phylogenetic trees of the two types of plasmid (**Supplementary Figures S8, S9A**), we detected the genotypic clusters of a single geographical origin. Typically, the plasmid clusters of South America are

observed in both phylogenetic trees, which indicates relatively seldom exchange of mcr-1-bearing plasmids between South America and other geographical regions. Geographical barriers may result in such a regional evolution. Meanwhile, Asia shows the closest communication with other geographical regions, and the interconnection between Europe and North America is relatively high. In addition, according to the phylogenetic trees of the two types of plasmids, the intercontinental exchange of IncI2 plasmids seems more frequent than that of IncX4 plasmids. The diversity of bacterial hosts of IncI2 plasmids (**Figure 2A**) may have facilitated their transportation among different continents, which can also explain why IncI2 plasmids occupy the largest share of the mcr-1-bearing plasmids (41% of 229 mcr-1-bearing plasmids).

The shufflons extracted from IncI2 plasmids were carefully annotated the structure. Typical shufflon structures are illustrated in **Figure 3**. There is a pair of inverted sfx repeats (IRL: **GKGCCAATCCGGT**NBS**TGG** and IRR**: CCA**SVN**ACCGGATTGGCMC**) at both ends of each shufflon

<sup>4</sup>https://mafft.cbrc.jp/alignment/server/

ARGs on IncHI2 plasmids (B) and Venn diagram showing the numbers of shared and unique additional ARGs on IncHI2 plasmids at different regions (C). In (A), the size of the nodes (plasmid type, bacterial species or continent) is proportional to their detection rate (log-transformed), and the width of the edge between two nodes is proportional to the detection rate of one node (plasmid type or bacterial species) from another (bacterial species and/or continent). Blue nodes indicate bacterial species, green of different saturations the plasmid types and the nodes of other colors the continents. In (B), the size of the nodes (ARG or geographical region) indicates the frequency of their occurrence, and the width of the edges is proportional to the co-occurrence rate between two nodes (between two ARGs or between an ARG and its location). Blue nodes indicate additional ARGs, red the mcr-1 gene and the nodes of other colors the geographical regions. Ten most frequently occurred ARGs are shown as blue circles.

segment, and also a highly conserved inverted or direct repeat sequence (GGAGGCCA) between adjacent shufflon segments. The shufflon structure also exhibits regional disparity (**Supplementary Table S2**), which is in good agreement with the clustering mode of IncI2 plasmids in the phylogenetic tree (**Supplementary Figure S9B**). IncI2-type plasmids isolated from Asia contain almost all kinds of shufflon structures. The shufflon rearrangement is closely related to plasmid transmission to a broad range of the Enterobacteriaceae (Ishiwa and Komano, 2003, 2004). This can explain why the most diverse Enterobacteriaceae hosts bearing mcr-1-positive IncI2 plasmids are detected in Asia (**Figure 2A**). The IncI2 plasmids isolated from Europe and North America commonly contain the shufflon segment E (except those plasmids with only one shufflon segment), while, to date (June 2018), no IncI2 plasmid from South America has been detected with definite structure of shufflon segment E (**Supplementary Table S2**). The mcr-1-bearing IncI2 plasmids in Europe and North America may mainly be derived from a plasmid with shufflon segment E, while those in South America may have been originated from one without the segment. Geographical barriers retain this original mark.

Some shufflon sequences cannot be annotated as accurate segmental structure, which could be due to sequencing/assembly mistakes (**Supplementary Table S2**). We also found an unnoticed shufflon segment structure that had never been annotated in the database. It is common and highly conserved in several IncI2 type plasmids (**Supplementary Table S2**, KY565557, CM008278, CP028153, CP006264, CP007134, FR851304, and CP030766), which cannot be accidental events. Thus we assigned the segment as segment S (**Figure 3**, KY565557).

The mechanism of Nested Russian Doll-like genetic mobility must be common for the worldwide dissemination of various mobilized ARGs (Dortet et al., 2014; Sheppard et al., 2016; Wang et al., 2018). The blaNDM and blaKPC genes are mainly located on conjugative plasmids of several different incompatibility groups (Dortet et al., 2014; Sheppard et al., 2016), which is similar to the behavior of mcr-1 gene. ARGs may be flanked by different MGE sequences at their mobilization, and exhibit different characteristics in their relocation process. The transfer of mcr-1 is mediated by the insertion sequence ISApl1. The high activity of this MGE enables mcr-1 gene to jump flexibly between different plasmids and between different bacterial species (**Supplementary Table S1**). At bacterial genome level, no genotypic clustering by geographical origin and isolation source has been observed (Matamoros et al., 2017). While at the transposon level, scientists have predicted the time of the initial mobilization of mcr-1 (Wang et al., 2018). In this work, at the plasmid level (one important genetic level involved in HGT) and according to the analysis routine illustrated in **Supplementary Figure S1**, we have found that the global spread of mcr-1-bearing plasmids is accompanied by their multiregional evolution. Based on the intrinsic mechanism of HGT, we believe that the analysis of mobilized ARGs at multiple levels of MGEs (transposons, integrons, and plasmids) can give important epidemiological information about their dissemination.

## Possible Intercontinental Transportation of Resistance Plasmids

Since we have found sequencing evidence for the horizontal transfer of ARGs between the environmental and avian bacteria, it is possible that migratory birds mediate the intercontinental transportation of the resistance plasmid. Two typical cases were noted at the analysis of IncI2 plasmids. First, the IncI2 plasmid, pJIE3685-1 (KY795978), isolated from Australia is of almost the same size (∼60960 bp) with two plasmids isolated from China (Taiwan), p5CRE51-MCR-1 (CP021176 from a human E. coli), and pP111 (KY120365, from a porcine Salmonella enterica subsp.). The position and the segment composition of the shufflon structure is also the same in the three plasmids (**Supplementary Table S2**), which indicates that the shufflon of the three plasmids is equivalent. Alignment of the remaining parts of the three plasmids after shufflon excision shows high identity among them with mismatches of only several (∼7 nts out of 59,042 nts) nucleotides (**Figure 4A**). In addition, another IncI2 plasmid from Australia, pJIE2288-1 (KY795977), is also highly identical (∼8 nts mismatches out of 59,081 nts)

to the three plasmids from Asia (**Figure 4B**), pRYU2912C-1 (AP018412, Japan, from a E. coli), pSh113-m4 (KY363994, Shanghai, China, from a human Shigella sonnei) and pSh069 m6 (KY363995, Shanghai, China, from a human Shigella sonnei). Such close genetic relationships between these plasmids suggest that they do not seem to originate from a common evolutionary ancestor, but rather are duplicate offspring of the same plasmid. Two Australian plasmids are from E. coli strains that were isolated from two clinical patients at New South Wales. Neither patients had ever gone abroad or had taken colistin/polymyxin antimicrobial drugs during hospitalization (Ellem et al., 2017). We noticed that these highly homologous IncI2 plasmids were all found on the same flyway of migratory shorebirds, i.e., the East Asian-Australasian Flyway (**Supplementary Figure S7**). Many species of migratory shorebirds take extreme long-distance migration between eastern Russia and Oceania. During the migration, they stage once in Eastern Asia (typically the Yellow Sea region) for around 40 days, resting and refueling for the subsequent flight (Battley et al., 2012). There is continentalscale pollution of ARGs at the estuaries along the east coast of China, where the migratory shorebirds pass by Zhu et al. (2017). Migratory shorebirds can be an option for tracking the source of the resistance plasmids.

During the outbreak of avian influenza, migratory birds as potential global spreaders had attracted the attention of scientists (Liu et al., 2005; Normile, 2006; Olsen et al., 2006). However, according to a number of previously reported studies, the risk of migratory birds mediating intercontinental exchange of influenza virus, especially those highly pathogenic ones, is fairly low (Krauss et al., 2007; Langstaff et al., 2009). A sick bird can hardly accomplish the arduous task of long-distance (intercontinental) migration. Nevertheless, migratory shorebirds carrying drugresistant symbiotic bacteria can cross the **Wallace Line** and bring ARGs from Asia to Australia (**Supplementary Figure S7**). Since the discovery of the mobilized colistin resistance gene, several studies have reported the detection of mcr-1-positive E. coli in migratory birds (Liakopoulos et al., 2016; Mohsin et al., 2016; Ruzauskas and Vaskeviciute, 2016; Sellera et al., 2017). Salt tolerant mcr-1-positive E. coli strains have also been isolated from recreational waters of public urban beaches (Fernandes et al., 2017). Therefore, mcr-1-bearing plasmids may be transported, via avian migration, from Eastern Asia to Australia, where the usage of antibiotics is under strict control but is definitely affected by migratory shorebirds. In fact, mcr-1-positive E. coli has recently been isolated from wild bird (silver gull) in Australia (Mukerji et al., 2019). In view of limited data available for bacterial ARGs of migratory birds, it is difficult to discuss the global dissemination of ARGs in the context of avian ecology. Our analysis does not imply that avian migration is the main route for the global dissemination of antibiotic resistance. In today's highly globalized world, global population mobility and international trade, especially the trade of food animal, must be the main channels (Matamoros et al., 2017; Shen et al., 2018; Hernando-Amado et al., 2019). Even so, the large number of migratory birds and their fixed migration path will have a lasting impact on the receptor environment. In particular, different countries execute different strategies for using antibiotics. Antibiotics used for human in one country may be used as animal feed additives in another. If migratory birds transport antibiotic resistance from the latter environment to the former, the cause originated in one country may have a devastating consequence in another.

Nowadays, with the global dissemination of antibiotic resistance, no country can be immune from it. The application of antibiotics should follow the strategy of global unification.

## DATA AVAILABILITY STATEMENT

Publicly available datasets were analyzed in this study. This data can be found in GenBank database under the accession numbers CP032986 to CP032995.

## AUTHOR CONTRIBUTIONS

KY and JW designed the study. DR isolated the E. coli strains and evaluated their drug resistance. JW and YL prepared the DNA samples for sequencing and analyzed the sequencing data. XD and LZ performed the metadata analysis of the mcr-1-bearing

## REFERENCES


plasmids. YL, JW, and KY constructed the manuscript. YF edited the manuscript and put forward constructive suggestions on it. All authors reviewed, revised, and approved the final report.

## FUNDING

We gratefully acknowledge the financial support from the Initiating Research Fund for Talent Introduction of Sichuan University (YJ201355) and the Natural Science Foundation of China (21677104).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00352/full#supplementary-material


underlying dissemination of the colistin resistance determinant. J. Antimicrob. Chemother. 72, 393–401. doi: 10.1093/jac/dkw411


rearrangement regions, shufflons, in IncI2 plasmids. Sci. Rep. 7:928. doi: 10. 1038/s41598-017-01082-y


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

fmicb-11-00352 March 6, 2020 Time: 17:19 # 12

# Molecular Analysis of Selected Resistance Determinants in Diarrheal Fecal Samples Collected From Kolkata, India Reveals an Abundance of Resistance Genes and the Potential Role of the Microbiota in Its Dissemination

## Rituparna De\*, Asish Kumar Mukhopadhyay and Shanta Dutta

#### *Edited by:*

*Inge C. Gyssens, Radboud University Nijmegen Medical Centre, Netherlands*

#### *Reviewed by:*

*Veranja Chathurani Liyanapathirana, University of Peradeniya, Sri Lanka Ellen Stobberingh, National Institute Public Health, Netherlands*

> *\*Correspondence: Rituparna De rituparna26@gmail.com*

#### *Specialty section:*

*This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health*

> *Received: 29 June 2019 Accepted: 18 February 2020 Published: 11 March 2020*

#### *Citation:*

*De R, Mukhopadhyay AK and Dutta S (2020) Molecular Analysis of Selected Resistance Determinants in Diarrheal Fecal Samples Collected From Kolkata, India Reveals an Abundance of Resistance Genes and the Potential Role of the Microbiota in Its Dissemination. Front. Public Health 8:61. doi: 10.3389/fpubh.2020.00061* *Division of Bacteriology, National Institute of Cholera and Enteric Diseases, Kolkata, India*

Twenty-five diarrheal fecal samples from Kolkata were examined to determine the relative abundance of antimicrobial resistance genes (ARGs) against eight common classes of antibiotics with polymerase chain reaction (PCR) and Sanger sequencing. Relative abundance of an ARG was calculated as the percentage of fecal samples showing the presence of that particular ARG. The frequency of occurrence of resistance marker against each class of antibiotic was calculated as the percentage of fecal samples carrying at least one resistance marker for that particular class of antimicrobials. Antibiogram of *Vibrio cholerae* (*V. cholerae*) O1 strains isolated from four of these samples was obtained by disc diffusion method and was compared with the ARG profile of corresponding fecal samples from which the strains were isolated. A 464 bp amplicon of the V3-V4 region of bacterial 16S rDNA was obtained by PCR from 9 of these 25 samples using the primer pair *S-D-Bact-0341-b-S-17* and *S-D-Bact-0785-a-A-21* and sequenced to determine the major operational taxonomic unit (OTU). These 9 samples represented diarrhea due to diverse etiology and also unresolved etiology as determined by culture method. We conclude that the diarrheal intestinal microbiome has a common gene pool of ARGs against the major classes of antibiotics and may be serving as a reservoir of ARG dissemination. ARG profile of cholera stool showed that ARGs present in the gut of cholera patients may be transferred to the *V. cholerae* genome and pose a serious threat to the treatment of cholera by triggering resistance against potential drugs to which contemporary strains of *V. cholerae* were found to be sensitive in the present study. Fecal samples which were culture negative for diarrheal pathogens we tested also carried ARGs and OTU. Abundance of resistance markers against macrolides, tetracyclines, and aminoglycosides was the highest. Phylum Proteobacteria was the most abundant OTU suggesting proteobacterial blooms characteristic of disturbed gut microflora. Our study is the first comparative study of ARG profile of diarrheal samples with varying etiologic agent revealing the presence of ARGs against the most important classes of antibiotics in the gut of diarrheal patients by common, robust molecular methods, which are easily accessible by molecular epidemiological laboratories worldwide.

Keywords: antimicrobial resistance, antimicrobial resistance genes, microbiota, microbiome, metagenomics, diarrhea, pathogen, commensal

## INTRODUCTION

Antimicrobial resistance (AMR) among diarrheal pathogens has emerged as a critical threat to the clinical management of diarrheal cases. Oral rehydration therapy (ORT) is the primary treatment for diarrhea and antibiotic therapy is used as a supplementary treatment to reduce severity and morbidity. Pathogens have developed resistance to multiple antibiotics which were used for controlling these infections giving rise to multi-drug resistance (MDR) which is leading to higher number of deaths. These are extremely difficult to treat with known chemotherapeutic agents in diarrheal patients for whom primary treatment with oral rehydration solution (ORS) is insufficient.

Common enteric pathogens like Klebsiella pneumoniae and Eschericia coli have developed resistance against last resort antimicrobials like carbapenem and these are, in turn, serving as potential agents for transmission of ARGs into the environment and the community<sup>1</sup> (1). Resistome analysis to understand the antimicrobial resistance (AMR) profile in pathogens is urgently required in order to discern divisive methods to prevent the transmission and spread of genetic determinants of AMR. The members of the microbiota in the environment and in humans are the primary sources of ARGs. These serve as potential reservoirs for the persistence and transmission of ARGs. The indiscriminate use of antibiotics in farm and for clinical and veterinary practices has led to the emergence of antimicrobial resistance as a critical threat. In this study we have attempted to report about the profile of selected resistance determinants obtained from fecal samples using simple polymerase chain reaction (PCR) with specific primers to detect a spectrum of antimicrobial resistance determinants that are involved in diverse antimicrobial resistance mechanisms in bacteria and which are associated with the most common classes of antibiotics advocated for diarrheal treatment. These include tetracyclines, macrolides, amphenicol, aminoglycosides, carbapenem, trimethoprim, sulfamethoxazole, and quinolones. The antimicrobial genetic determinants selected to serve as markers for these classes of antimicrobial resistance include those encoding enzymes, efflux proteins, and proteins which are involved in diverse mechanisms like cell wall degradation of bacteria and inhibition of protein synthesis. The presence of these genetic determinants was further confirmed by sequencing these genes where standard positive control DNA was not available. The results revealed the existence of antimicrobial resistance determinants against major classes of antibiotics in fecal DNA samples. The study helped us to forebode the possibility of ARGs being transmitted in the near future from the microbiota into pathogens. Our study suggests that major members of the gut community are serving as reservoirs of ARGs as samples from which pathogens could not be isolated by conventional culture methods also presented an ARG profile. We also report the relative abundance of different ARGs in the diarrheal gut microbiota in Kolkata and the suburban areas. Our study is the first addressing the relative abundance of ARGs of different classes of antimicrobials with the help of common and economic laboratory tools in the gut microbiome of the local diarrheal patients in Kolkata and the suburban areas. The study would provide valuable understanding about the threat posed by the presence of ARGs in the gut of diarrheal patients in parts of the world where the economically backward population is under perpetual threat of diarrheal diseases due to lack of sanitation and for whom antimicrobial therapy is of utmost importance to reduce severity and mortality due to diarrhea along with ORS administration. Therefore, understanding the distribution of ARGs is important to reduce their transmission from their reservoir with interceptive methods.

## MATERIALS AND METHODS

## Sample Collection and Ethical Clearance

Twenty-five diarrheal stool samples were collected from the Bacteriology Division laboratory of National Institute of Cholera and Enteric Diseases (NICED), which routinely receives stool samples from the adjoining Infectious Diseases Hospital (IDH) and the B. C. Roy Hospital (BCH) for systematic screening and isolation of enteric pathogens from stool of diarrheal patients from Kolkata and the suburban areas. Thus, these 25 samples represented diverse etiology and the diarrheal population of Kolkata and the suburbs. These patients eliminated loose watery stool more than three times in a day and suffered mild to severe dehydration. Stool samples included in the study were from patients of age 2 months and above and were from male and female patients. Twenty-four of these samples were from patients who were admitted to IDH for 1–4 days for treatment of acute diarrhea and in these patients diarrhea lasted for 1–5 days. One sample KOL18B2-6 was collected at the outpatient ward of BCH from the patient who had symptoms of mild diarrhea. In this patient diarrhea lasted for 2 days. After collection, each stool sample was given a unique identity number for the study. All stool samples were collected and handled in a manner conforming to ethical rules and regulations of the local governing bodies and the institute (NICED). **Table 1** presents a description of the stool samples used for the study.

<sup>1</sup>https://www.who.int/news-room/fact-sheets/detail/antimicrobial-resistance

#### TABLE 1 | Stool samples used in the study, their description and AMR profile.


*Location*\**: K, Kolkata; S, Suburban area.*

*Age: y, years; m, months.*

*Sex: M, Male; F, Female.*

## Isolation of Genomic DNA From Standard Laboratory Strains

Standard laboratory strains N16961 (V. cholerae O1, El Tor), O395 V. cholerae O1, classical) and MO10 (V. cholerae O139) served as control for this study. The strains were obtained from the NICED strain repository and initially grown overnight at 37◦C on TCBS (thiosulphate- citrate- bile salts-sucrose) (BD DifcoTM, U.S.A) plates followed by sub-culturing on Luria agar (BD DifcoTM, U.S.A) plates containing 1% NaCl (sodium chloride) (Merck-Millipore, U.S.A). A loopful of culture was dissolved in TE (Tris-EDTA) buffer (pH 8) (Sigma Aldrich, USA) followed by extracting DNA using phenol-chloroform-isoamyl alcohol mixture in the ratio 25:24:1 and alkaline pH and finally eluting DNA in nuclease-free water.

## Isolation of Genomic DNA From Stool Samples

Microbial genomic DNA was isolated from stool samples using QIAampUCP Pathogen Mini Kit (cat.no. 50214, Qiagen Inc., MD, USA), using the protocol provided by the manufacturer, for isolation of ultra-clean DNA. Accordingly, stool samples were subject to mechanical lysis using glass beads in a mini vortex-mixer (3020 Spinix, Tarsons) followed by enzymatic lysis using buffer containing pre-mixed bacterial cell-wall degrading enzyme cocktail under highly denaturing conditions at elevated temperatures of 70◦C and protein removal and nuclease inactivation using a combination of Proteinase K (20 mg/ml) and buffer APL2 using spin columns (QIAamp UCP Mini Column, Qiagen Inc., MD, USA) wherein microbial nucleic acid is adsorbed on the silica membrane by centrifugation at 8,000 g for 1 min and washing with buffer containing ethanol followed by final elution using elution buffer containing Tris-EDTA (Ethylenediaminetetraacetic acid). The concentration of DNA from each sample was quantified using the NanoDrop Lite UV-Vis Spectrophotometer (Thermo Fisher Scientific, MA, USA) and diluted to obtain a uniform concentration of 25ng/ul for all samples using ultra-pure water obtained by purification with the Milli-Q <sup>R</sup> Integral Water Purification System (Merck-Millipore, USA) followed by autoclaving at 121◦C for 15 min at 15psi. The DNA was used as template for PCR and Sanger sequencing. In the method described above salt and pH conditions help in the complete removal of proteins and other contaminants, which can inhibit downstream enzymatic reactions.

## AMR Profiling

AMR profile of each sample was obtained by PCR using specific primer set consisting of forward and reverse primers to amplify the gene of interest involved in antimicrobial resistance mechanism (**Table 2**). PCR was performed in a final reaction volume of 25 µl containing 1x GoTaq <sup>R</sup> Green master mix (cat.no. M7123, Promega Corporation, WI, USA), 0.4uMupstream and downstream primers and 1 ng/µl DNA template using a 96-well thermalcycler (GeneAmp <sup>R</sup> PCR system 9700, Applied Biosystems, Thermo Fisher Scientific, MA, USA). A list of genes and their corresponding primer sequences, annealing temperatures and size of the amplified fragment has been presented in **Table 3**. The PCR products were run on 1– 1.5% agarose gel (depending on amplicon size) prepared using 1X TAE (Tris-acetic acid-EDTA) buffer followed by staining in ethidium bromide solution (10 mg/ml) and visualized in a UV-transilluminator (BioRad) and the results documented. For experimental control standard laboratory strains N16961, O395, and MO10 were used as template for the amplification of ARGs pre-documented as present or missing in these strains by previous reports archived in the public database NCBI. The amplicon size was determined using 100 bp and 1 kb DNA ladder (NEB, MA, USA and GeNetBio Corp., Korea).

## Sanger Sequencing for Confirmation of ARGs

PCR products obtained above and containing the amplicon of interest was purified using the Wizard <sup>R</sup> SV Gel and PCR Clean-Up System (cat. no. A9281, Promega Corporation, WI, USA) that can purify100 bp to 10 kb DNA fragments with Wizard <sup>R</sup> SV Minicolumns. DNA quantity and quality was adjusted to fulfill the requirements for Sanger sequencing. Purified PCR products with optimum absorbance ratios (A260/280) between 1.8

TABLE 2 | The eight major classes of antibiotics and Mobile Genetic Elements (MGEs) and their corresponding ARGs whose presence or mutations were determined in the study.


and 2.0 were selected to serve as template. Cycle sequencing was carried out with the BigDyeTM Terminator v3.1 cycle sequencing kit (cat. no. 4337455, Applied Biosystems, Thermo Fisher Scientific, MA, USA) using 10 µl reaction mixture containing BigDyeTMTerminator v3.1 Ready Reaction Mix, BigDyeTMTerminator v3.1 Sequencing Buffer, final concentration of 3.2 pMol of either forward or reverse primer and DNA template quantity of 1–5 ng depending on the size of the fragment being sequenced. Cycling conditions consisted of denaturation at 96◦C for 1 min and 25 cycles of denaturation, annealing, and extension at 96◦C for 10 s, 50◦C for 5 s, and 60◦C for 4 min, respectively. The cycle sequencing product was purified using 3M sodium acetate solution (pH 4.8) and absolute ethanol (Emsure, Merck, Germany) and washed with 70% ethanol and finally vacuum dried. The dried pellet was resuspended in 10 µl Hi-DiTM Formamide and loaded onto the Applied BiosystemsTM 48-capillary DNA Analyzer 3730 (Thermo Fisher Scientific, MA, USA) for sequencing and collection of data using the recommended signal to noise ratio. Accordingly, unambiguous clear curves with sharp peaks indicating clear base-calling and low background noise were considered. Curves with minimum signal 300 for all the bases and maximum noise 20 were set as the threshold for accepting raw data. Chromatograms considered for analysis had maximum signal in the range of G = 964– 4,176; A = 518–2,967; T = 537–2,435; C = 662–4,186. Noise was confined to baseline (signals 0–50) and did not interfere with data interpretation. Raw data was viewed and analyzed using SeqManTMII (version Windows 32 SeqMan 5.01) software (DNASTAR Inc., WI, USA) and assembled sequence data in the clear range was accepted and considered. This was subject to local alignment using Megablast program of Nucleotide BLAST (Basic Local Alignment Search Tool) on the NCBI (National Center for Biotechnology Information) server.

## Calculation of Relative Abundance and Frequency of ARGs

Relative abundance of ARGS present in fecal samples was calculated. For each ARG it was represented as the percentage of fecal samples found to be carrying that particular ARG. This can be represented by the formula:

TABLE 3 | Primers (5′ -3′ ) used in the study.


*(Continued)*

#### TABLE 3 | Continued


*(Continued)*


#### For ARG A

(Number of fecal samples positive by PCR for ARG A/ n)∗100

Where ARG A is any ARG which has been included in the study and PCR positive and n is the total number of fecal samples used in the study. Here, n is 25.

Frequency of occurrence of resistance marker was calculated by the formula:

For any particular class of antibiotic:

(Number of fecal isolates found positive for at least one ARG marker against that class / n)<sup>∗</sup> 100

Where, n is the total number of fecal samples used in the study. Here, n is 25.

On the basis of these two parameters described above we concluded which group of antibiotics has the highest abundance of resistance marker.

## 16S Conserved and Variable Region Amplification

PCR amplification for confirming the presence of bacterial DNA in the DNA isolated from fecal specimens was carried out for the conserved regions C1-C9 and variable regions V3-V4 of 16S rDNA using standard primer sets targeting these regions. PCR amplification was carried out in 25 µl final reaction volume using the same reagents described above. The cycling conditions for PCR using primer sets 130–139 and S-D-Bact-0341-b-S-17- S-D-Bact-0785-a-A-21 were described by Bag et al. (13) and Klindworth et al. (14), respectively.

## Sanger Sequencing of 16s rDNA Variable Region

For representative samples the 464 bp fragment obtained by PCR amplification of variable region V3-V4 of 16S rDNA with primer set S-D-Bact-0341-b-S-17- S-D-Bact-0785-a-A-21 was subject to Sanger sequencing. The PCR product was purified and subject to cycle sequencing reaction and finally sequenced using the 48-capillary DNA Analyzer 3730 (Thermo Fisher Scientific, MA, USA) following the same procedure described above. Sequence assembly and analysis was done using SeqManII software (DNASTAR Inc., WI, USA) and the assembled sequence data in the clear range was accepted and considered. With the help of the Megablast program of the local alignment tool Nucleotide BLAST on the NCBI server the sequenced data was matched with archived and annotated 16S rDNA sequences available in GenBank, the NIH genetic sequence database.

## Routine Isolation, Laboratory Culture, and Antibiotic Susceptibility Test of Diarrheal Pathogens

The enteric pathogen associated with diarrheal etiology in each fecal sample was isolated using enrichment technique. Accordingly, the samples were streaked onto selective and differential media plates for the detection of common diarrheal pathogens, Vibrio sp., E. coli, Salmonella sp., Shigella sp., Aeromonas sp., Campylobacter sp. in the Bacteriology Division Laboratory at NICED. Accordingly, bacterial culture plates TCBS, HEA (Hektoen enteric agar), XLD (Xylose Lysine Deoxycholate), MacConkey, Blood agar were used for each fecal specimen. Culture plates were incubated overnight at 37◦C (3– 5 days for Campylobacter sp.) and colonies from the plates were subject to biochemical test using Triple Sugar Iron Agar for identification of acid (lactose/sucrose fermenting and nonfermenting), gas, and hydrogen sulfide producing organisms. Single colonies from TCBS plates were subjected to oxidase and string test for the confirmation of V. cholerae.

Confirmed V. cholerae strains were further tested for antibiotic susceptibility by Kirby-Bauer Disc Diffusion method. The strains were grown on LA (Luria agar) plates overnight at 37◦C and a loopful of culture was taken from the plate and inoculated into Luria broth and OD of this culture suspension was adjusted to obtain OD of 0.5 McFarland. With the help of autoclaved

swab stick bacterial culture from this broth was spread onto Mueller-Hinton Agar (MHA) plates and allowed to dry in a biosafety cabinet. Antibiotic discs (BD BBLTM, NJ, U.S.A.) were placed on the surface of MHA plates with sterile forceps to test sensitivity toward common antibiotics, namely, ampicillin (10 µg), ceftriaxone (30 µg) chloramphenicol (30 µg), ciprofloxacin (5 µg), gentamicin (30 µg), ofloxacin (5 µg), imipenem (10 µg), azithromycin (15 µg), gentamicin (10 µg), norfloxacin (10 µg), nalidixic acid (30 µg), streptomycin (10 µg), SXT (trimethoprimsulfamethoxazole) (1.25, 23.75 µg), and tetracycline (30 µg). The plates were incubated overnight at 37◦C. The diameter of the clear zone (zone of inhibition) around each disc was measured, recorded and compared with the zone diameter interpretive chart provided by the manufacturer and which follows CLSI (Clinical and Laboratory Standards Institute) guidelines<sup>2</sup> . Accordingly, the strains were categorized as resistant, intermediate or susceptible.

## Sanger Sequencing for Detecting Single Nucleotide Polymorphisms (SNPs) in Genes Involved in Quinolone Sensitivity

Four primer sets gyrA\_seqFOR2-Rev2 gyrB\_seqFOR-Rev, parC\_seqFOR-Rev, parE\_seqFOR-Rev (**Table 2**) were designed to amplify regions of genes encoding Topoisomerase II subunits A and B (gyr A and gyrB) and Topoisomerase IV subunits A and B (parC and parE). Sequencing was carried out to study SNPs in quinolone resistance determining regions (QRDR) in fecal DNA from representative samples. Amplification of fragments of gyrA, gyrB, parC, and parE was carried out by PCR with primers gyrA\_seqFOR2-Rev2, gyrB\_seqFOR-Rev, parC\_seqFOR-Rev, parE\_seqFOR-Rev which were designed from the N16961 sequence (reference sequence ID. NC\_002505; Gene ID: 2615137) to amplify 771 bp fragment of gyrA, 662 bp fragment of gyrB, 729 bp fragment of parC and 640 bp fragment of pare which were purified using the Wizard <sup>R</sup> SV Gel and PCR Clean-Up System (cat. no. A9281, Promega Corporation,

<sup>2</sup>www.clsi.org

WI, USA) and used as template for cycle sequencing reaction with the BigDyeTM Terminator v3.1 cycle sequencing kit (cat. no. 4337455, Applied Biosystems, ThermoFisher Scientific, MA, USA) and were sequenced on the Applied BiosystemsTM 48-capillary DNA Analyzer 3730 (Thermo Fisher Scientific, MA, USA) in the same manner as already described in the preceding section (Sanger Sequencing for confirmation of ARGs). Sequence data was assembled with SeqMan II software (DNASTAR Inc., WI, USA) and the sequence identity was matched by global alignment with the corresponding genomic regions of standard V. cholerae O1 El Tor strain N16961 to map single nucleotide polymorphisms (SNPs).

## RESULT

## 16S Conserved Region PCR

A 1,500 bp amplicon of conserved region of 16S rDNA was obtained spanning the C1-C9 region of bacterial genomic DNA in fecal samples on PCR amplification with primer set 130–139. The results confirmed the presence of bacterial DNA (**Figure S1**).

## AMR Profile

PCR was performed targeting a group of 49 genes associated with 8 different classes of common antibiotics like carbapenem, tetracycline, quinolones, aminoglycosides, macrolide, trimethoprim, sulfamethoxazole, amphenicol, and components of mobilome like transposase, SXT integrase, and markers of classes 1, 2, and 4 integron. PCR results showed amplification of genes aadA1, aad2, strAB, aac (2), aac(6′ )-Ib-cr, tetABCDEM, SXT integrase, floR, dhfr1, dfrA1, dfrA12, dfrA15, cat1, sul2, aph, ereA, mefA, mphA, tnpA, int1, int2, int4. This revealed that the microbial genomic DNA bore ARG markers against aminoglycoside, tetracycline, macrolide, chloramphenicol, trimethoprim, sulfamethoxazole, and mobile genetic elements (MGEs). **Table 1** shows the results obtained on screening for ARGs and the different classes of antibiotics against which they are active. All of the 25 samples tested for the presence of ARGs, showed genetic determinants associated with resistance against at least two classes of these broad-spectrum antibiotics. Where standard control strains were not available to compare and confirm the results, PCR products were sequenced using Sanger Sequencing to determine and confirm the identity of the gene. Accordingly, genes tetABCDEM, aph, dhfr1, mphA, mefA, dfrA12, tnpA, aadA1, aad2, aac(3)/(aacC), aac(6′ )- Ib-cr, ereA, int1, int2 were sequenced and the assembled sequence on local alignment using Megablast program of Nucleotide BLAST confirmed the identity of these genes thereby confirming their presence in the fecal samples of diarrheal patients in Kolkata and suburban areas. Representative gel pictures showing the results of ARG profiling are available in **Figures S2–S4**. The assembled sequences and the results obtained on alignment of these sequences using BLAST are available as **Supplementary Material S5**.

The relative abundance of different ARGs (**Figure 1**) and the frequency of occurrence of resistance marker against each class of antibiotic (**Figure 2**) were calculated. It was found that 96% samples carried mphA and mefA genes encoding resistance against macrolide and hence these two genes were found to have highest occurrence followed by 92% occurrence of strA and strB encoding aminoglycoside resistance followed by tetracycline resistance due to the occurrence of 84% isolates with tetA gene, 80% isolates with tetB gene and 76% isolates with tetE gene. Other genes with high occurrence rate are cat1 for chloramphenicol resistance (40%), dfrA1 (62.5%), dfrA12 (33%)



*VC, V. cholera. Antibiogram for VC*\**.*

both encoding trimethoprim resistance, aad2 (64%), aadA1 (72%), aac(3) (72%), aac6(64%) all encoding resistance against aminoglycosides. Lower rate of occurrence was seen among tetC (4%), ereA (8%), tetD and tetE (12% each), floR (12%), dhfr1 and dfrA15 (4% each), and aph (16%).

In addition 3 isolates showed the presence of SXT integrase, 19 isolates showed the presence of tnpA encoding transposase and 21 isolates carried integron class1, 11 isolates carried integron class 2 and 8 isolates carried integron class 4 and 4 isolates were found to carry all the three classes of integrons showing the presence of MGEs and signatures associated with MGEs.

## Isolation of Diarrheal Pathogens by Conventional Culture and Antibiotic Sensitivity Test of *V. cholerae*

Pathogens responsible for causing diarrhea in the patients were isolated from the fecal samples by conventional culture method and confirmed using routine biochemical tests. **Table 1** shows pathogens isolated from each fecal sample. V. cholerae O1 strains were subject to antibiotic sensitivity test by Kirby-Bauer Disc Diffusion method. The antibiogram results were compared with results of PCR AMR profile of corresponding fecal sample from which the strain was isolated (**Table 4**). On comparison of ARG profile of feces with the antibiogram of its corresponding pathogen it was clear that fecal samples carried ARGs which were not always being expressed by pathogens. At the same time even the pathogen's antibiogram revealed resistance against those classes of antibiotics against which resistance markers were not detected by PCR. Fecal sample KOL18B3-1 carried the following genes tetABM, cat1, floR, ant, dfrA12, dfr1, aac(3), aac6′ , aadA1, strAB, mphA, mefA, tnpA, intI1, intI2, intI4, sxt but the antibiogram of the V. cholerae strain isolated from this fecal sample showed that the strain was not resistant against tetracycline and macrolides. Similarly, V. cholerae strains isolated from fecal samples KOL18B2-1 and KOL18B2-4 were sensitive to chloramphenicol but the samples showed the presence of gene cat1, which affords resistance against chloramphenicol.

KOL18B2-6, KOL18B2-7, KOL18B3-2, KOL18B3-4, KOL18B3-7, KOL18B3-9, KOL18B3-10, KOL18B3-16, KOL18B3-17 failed to produce conclusive results by culture method for the identification of any pathogen that we routinely test for diagnosis. However, from all of these samples, resistance determinants were detected successfully by PCR indicating the presence of bacterial DNA and ARGs (**Table 5**).

## 16S Variable Region Sequencing

Four hundred sixty-four bp amplicon from the V3-V4 region of 16S rDNA of representative samples was obtained by PCR and subject to Sanger sequencing. Accordingly DNA from nine samples namely, KOL18B2-1, KOL18B2-2, KOL18B3-1, KOL18B3-2, KOL18B3-3, KOL18B3-9, KOL18B3-10, KOL18B3- 11, KOL18B3-12 and genomic DNA from two laboratory strains of V. cholerae namely O395 and N16961 were used for the amplification of 464 bp amplicon which was subject to nucleotide sequencing on the ABI platform and the sequence data was subject to pairwise alignment using local alignment software Nucleotide BLAST on the NCBI server. From KOL18B3-2, KOL18B3-3, and KOL18B3-10 no pathogen had been isolated by culture but they carried ARGs as revealed by PCR. The assembled sequence data obtained from sequencing of DNA from all of the above fecal samples could be successfully aligned with known annotated 16S ribosomal DNA sequences archived in the GenBank public database (**Table 6**). The results revealed significant information by indicating the presence of phyla Bacteroidetes, Firmicutes, and Proteobacteria (classes βproteobacteria and γ -proteobacteria) in the fecal samples from Kolkata. The 16S variable region DNA sequencing revealed that a majority of fecal samples carried DNA from family Enterobacteriaceae as 16S rDNAamplicons from KOL18B2-1, KOL18B3-9, and KOL18B3-12 aligned 92–100% with 16S rDNA of Eschericia sp. and KOL18B3-2, matched 99% to uncultured Eschericia sp. KOL18B2-2 contained DNA that matched 89% with Bacteroidetes DNA. KOL18B3-3 and KOL18B3-10 matched up to 98% with DNA from Bifidobacteriaceae and KOL18B3- 11 carried DNA from Streptococcus sp. revealing the presence of Firmicutes in KOL18B3-3, KOL18B3-10, and KOL18B3-11. KOL18B3-1 carried DNA that matched with uncultured βproteobacterium and uncultured Firmicutes (Clostridium sp.). 5 out of 9 representatives and therefore >55% carried DNA from phylum Proteobacteria, 4 out of 9 samples and hence TABLE 5 | Fecal samples with resistome but no culturable pathogen.


TABLE 6 | 16S rDNA V3-V4 region sequence identity of OTU obtained using Megablast.


>44% carried DNA from phylum Firmicutes and 1 out of 9 fecal DNA samples and hence 11% revealed the presence of DNA from phylum Bacteroidetes. These results are available in **Figure S6**. The assembled sequences from O395 and N16961 used as experimental control matched 100% with V. cholerae 16S rDNA. The assembled sequences are available as **Supplementary Material S7**.

## Mutations in Quinolone Resistance Determining Regions (QRDR) of Topoisomerase II and IV Genes

Quinolone resistance develops due to mutations in the genes encoding topoisomerase II and IV. Twelve samples were subject to SNP analysis. From these fecal samples different diarrheal pathogens had been isolated by culture method. Accordingly, KOL18B2-1, KOL18B2-2, KOL18B2-4, KOL18B2-5, KOL18B3-1, KOL18B3-2, KOL18B3-3, KOL18B3- 5, KOL18B3-12, KOL18B3-13, KOL18B3-14, and KOL18B3- 15 were included for studying SNPs in the QRDRs. Clear amplification results showing presence of Topoisomerase II and IV genes of V. cholerae origin was obtained (for gel picture refer to **Figure S8**) by PCR using primers gyrA\_seqFOR2-Rev2, gyrB\_seqFOR-Rev, parC\_seqFOR-Rev, parE\_seqFOR-Rev which were designed from the N16961 sequence (reference sequence ID: NC\_002505.1) Samples used were KOL18B2-1 for gyrB, KOL18B2-2, and KOL18B3-1 for gyrAB, parCE. These were the samples from which V. cholerae was isolated as the diarrheal pathogen. Amplified fragments of gyrAB, parCE from KOL18B2- 2 and KOL18B3-1 were subject to SNP analysis by Sanger sequencing. Accordingly 771 bp fragment of gyrA, 662 bp fragment of gyrB, 729 bp fragment of parC and 640 bp fragment of parE were sequenced on the ABI platform. The assembled data obtained using the SeqMan II software was subject to pairwise alignment with the nucleotide sequence of the corresponding regions in the N16961 genome available in GenBank on NCBI (NC\_002505.1) The result of pairwise alignment revealed 100% pairwise match of nucleotide sequence between the assembled 586 bp gyrA fragment of KOL18B3-1 with that of N16961 (NC\_002505.1) corresponding to positions 1754436 to 1755021 at the locus VCO1597. The assembled fragment of gyrB subunit from KOL18B2-2 was of length 626 bp and it showed 99.84% identity to the corresponding 626 bp region of gene gyrB at the locus VC00005 positions 4668–5295 of the reference sequence V. cholerae strain N16961 (NC\_002505.1) chromosome 1 (accession no. CP028827.1). The alignment presented a gap showing a deletion at the 16th position of the aligned fragment which corresponded to position 5280 from the 3′ end in the reference sequence that has an "A." The assembled and edited fragments of parC and parE from the same sample were of length 322 and 594 bp. These showed 100% identity to 322 bp fragment of parC gene at locus VC00463 and to 594 bp fragment of parE gene at locus VC00462 in the N16961 sequence (accession no. CP028827.1) on pairwise alignment using BLAST. The assembled and edited 420 bp fragment of pare gene from KOL18B3-1 was aligned with reference sequence N16961 (accession no. CP028827.1) and also with parE sequence from KOL18B2-2. The pairwise alignment showed this region corresponded to positions 476083– 476502 in the reference sequence. This 420 bp fragment of parE from KOL18B3-1 showed 99.76% sequence identity to both. A single base substitution was observed at position corresponding to position 476151 of reference sequence where in "A" in the reference sequence had been replaced by "T" in the fecal DNA sequence. The same difference was noticed in KOL18B3-1 when its nucleotide sequence of the same genomic region of 420 bp was compared with that of KOL18B2-2 that lacked this SNP. The nucleotide change was not observed in KOL18B2-2and it bore 100% identity to reference sequence over this genomic region.

## DISCUSSION

AMR due to MDR has emerged as a critical obstacle to the treatment of enteric infections like diarrhea. Therefore, research directed toward leveraging this problem is an urgent need of the hour. The study described here has provided crucial insights into the gut microbiota composition and its resistome to understand transmission of ARGs between pathogens and microbiota. For a complete understanding of the microbiome, metagenomic analysis by next-generation sequencing (NGS) is the key that enables intensive analysis. However, it is an expensive technique and may not be within the affordability and outreach of a majority of laboratories, particularly in developing countries. In such situations, common molecular tools like PCR and Sanger sequencing, used decisively can produce far fetching outcome.

We have used these simple yet useful and robust methods to study the ARG composition in diarrheal fecal samples and to identify major members of the intestinal microbiota of diarrheal patients from Kolkata and the suburban areas. The QIAamp UCP Pathogen Mini Kit is a high quality kit for microbial DNA extraction and purification from small volumes of body fluids in only two hours. This kit provides high yield of DNA, requires no downstream purification and is compatible with Sanger Sequencing and deep sequencing techniques. In addition, since this kit is used for microbial DNA extraction from body fluids and urine, it is perfectly suitable for diarrheal stool which is liquid too. Fecal samples contain DNA from different members of the fecal microbiota and in such a polymicrobial community it is most likely that the most abundant member will be identified on the basis of Sanger sequencing of its 16S ribosomal DNA marker. This technique has been used for more than two decades for identification of bacterial taxa and also of unculturable bacteria for phylogenetic study (15–17). However, this is also one limitation of phylogenetic analysis by Sanger Sequencing of 16S rDNA. Since one of the major goals of the study was to identify the most abundant phylum present in the samples and to detect the presence of bacterial DNA in samples which were found to be devoid of any culturable bacterial pathogen which we tested using pure culture technique 16S rDNA Sanger Sequencing method was the method of choice as an alternative to NGS and was applied to achieve our goal. The variety of taxonomic entities identified in a sample depends largely on the primers selected for amplification of 16S rDNA. The efficiency of different primers has been scrutinized by several groups who have attempted to select the one with the best coverage of bacterial taxa (14–16). The primers we used in the study to examine OTUs have been reported to be the most suitable for bacterial identification up to the group and genus levels and in identifying the most abundant groups (14) thereby serving our purpose of identification of the most abundant phyla. Diarrhea like other enteric infectious diseases is accompanied with dysbiosis of the microbiome. Our study indicates that in majority of diarrheal patients proteobacteria is the dominant phylum followed by firmicutes suggesting a dysbiotic microbiota with proteobacterial "blooms" which is a characteristic of an abnormal microbiota (18). Detection of a spectrum of 49 ARGs was carried out to study resistance against major classes of antibiotics recommended for the treatment of diarrhea. Primers used for the study were selected or designed from whole genome sequences of different organisms archived in GenBank, NCBI for different classes of antibiotics based on information provided by previous published reports of other workers indicating the source or origin of these genes. ARGs involved in resistance against tetracyclines, macrolides, and aminoglycosides were found to be the major components of the resistome irrespective of the diarrheal pathogen isolated or the major operational taxonomic unit (OTU) present in any particular fecal isolate. The ARGs mphA and mefA encode phosphotransferase and efflux pump, respectively. The phosphotransferase is involved as a modifying enzyme which inactivates macrolides. These genes are active against erythromycin and azithromycin (8). Both these genes have been found to be abundantly present in gram negative bacteria according to published reports by other groups (8). Primers used to amplify these genes were designed from Enterobactericeae nucleotide sequences (8). These genes have been reported to be plasmid-borne and have been found to be transferred from E. coli to Shigella sp. (8). In our study we found from the antibiogram of V. cholerae strains that these were sensitive toward macrolides. These were isolated from samples KOL18B2-1, KOL18B2-2, KOL18B2-4, KOL18B3-1. However, in these samples these genes were found to be present. The study helped us to predict the future possibility that these ARGs present in the gene pool may at any point of time be transferred into organisms like Vibrio cholerae which come in contact with them and confer resistance upon these pathogens and contribute to the spread of resistance of macrolide drugs which are recommended last resort drugs for treatment of diarrheal diseases like cholera. mefC and mphG which were examined using primers designed from genomic sequence of Photobacterium damselae subsp. damselae strain 04Ya311 (accession no. AB571865) (9), representing ARGs from marine bacteria, were not detected in any fecal sample. Our results demonstrated that among tetracycline resistance determinants classes A and B were widespread among the clinical samples, which were examined. KOL18B2-3, KOL18B2-7, and KOL18B3-12 were found to carry class D tetracycline resistance gene and KOL18B2-3, KOL18B3- 3, and KOL18B3-12 were found to carry class E tetracycline resistance gene. Classes D and E tetracycline resistance gene have been reported to be widespread among environmental gram-negative bacteria from marine sediment and fish intestine (19, 20) and are not often isolated from clinical strains. Tetracycline resistance gene was first isolated from Shigella dysenteriae (21). Latter studies showed they were widespread among families of class γ-Proteobacteria like Enterobacteriaceae, Vibrionaceae, Aeromonadaceae, and Moraxellaceae. They have been reported in clinical and environmental isolates of these classes and have been isolated from Acinetobacter sp., Vibrio sp., and Aeromonas sp. in addition to E. coli and Salmonella sp. (11, 22). The tetracycline resistance determinants are found on MGEs like plasmids and transposons and have been found to be effectively transferred among bacterial species by conjugation. The resistance determinants selected for this study are involved in efflux of tetracyclines except class M which may have a function in protection of bacterial ribosomes against tetracycline action. Class M appeared in a number of samples used in this study and has been previously reported in a number of gram-positive and gram-negative genera (21). The primers used in this study for the detection of all these genes were designed from members of family Enterobacteriaceae like E. coli and S. Ordonez and were of plasmid or transposon origin (11).

The human intestine is an optimum environment for conjugal transfer of genetic components among bacterial members. ARGs present in feces and representing the intestinal resistome may be easily transferred to pathogen genome and contribute to AMR development against those drugs in the future. From the presence of components of mobilome (SXT integrase, three different classes of integrons and transposase) one can forebode that ARGs present in the fecal microbiome are prone to dissemination by HGT, which is a common phenomenon in bacteria.

In the context of cholera, which is a common diarrheal illness in poverty-stricken parts of the world, predominantly in the low and middle income countries, the presence of a gene pool rich in ARGs in cholera stool against the most potential classes of antimicrobials namely, macrolides, tetracyclines, and aminoglycosides which are the drugs of choice in the treatment of cholera is highly alarming. The strains of V. cholerae used for antibiogram analysis were found to be sensitive to these drugs. However, the presence of resistance markers against these antimicrobials in the corresponding fecal resistome and in the population on the whole can admonish of grave danger lurking in the near future when these ARGs can well find their way into the genome of V. cholerae and contribute to resistance. HGT is a common occurrence in V. cholerae, which occurs when the organism passes through the human gut and also in the environment (23). It is a bacteria with dual lifestyle and can be an important agent in the transmission of ARGs between the gut and the environment (24).

Another significant finding of our study is the coexistence of 16S rDNA signature of Bifidobacteria sp. and ARGs in KOL18B3- 3 and KOL18B3-10. Bifidobacteria sp. is an early intestinal colonizer and commensal which impacts the development of newborns in multiple beneficial ways and have been developed as probiotics for the prevention and treatment of diseases (25). Recent work by Bag et al. (26) and Duranti et al. (27) demonstrated the existence of ARGs in whole-genome sequence (WGS) of Bifidobacteria sp. Bag et al. showed the physical linkage of these ARGs against different classes of antimicrobials to MGEs in the WGS of a strain of Bifidobacterium longum indica that was isolated from feces of a healthy Indian adult (26). Duranti et al. found interspecies variation in breakpoint levels for antibiotics like streptomycin and tetracycline indicating the role of HGT in the evolution of AMR in Bifidobacteria sp. (27). Acquisition of ARGs by Bifidobacteria sp. has lead to its emergence as MDR and these in turn will serve as vehicles of transmission of ARGs.

This is the first study addressing the frequency and distribution of different classes of ARGs of diverse origin in the local intestinal milieu of diarrheal patients in Kolkata and the suburban areas. The structure and composition of microbiome is influenced by geography, ethnicity, and cultural practices. The study is distinct. It has been conducted using fecal samples from a discrete population. The study helped us to compare ARG profile between fecal samples from diarrheal patients afflicted with diarrhea due to different diarrheal pathogens. It helped us to conclude that fecal samples may share many ARGs in common, irrespective of the etiologic agent isolated from it. This is suggestive of a common gene pool acting as ARG reservoir and aiding in frequent lateral and vertical genetic exchange in pathogens and commensals. The study demonstrated that fecal samples from which routinely diagnosed pathogens were not isolated by conventional culture method also presented a resistome. These stool samples might have been harboring viral or eukaryotic diarrheal pathogens or any other enteric bacterial pathogen which was not tested for as it is beyond our scope. Further, it is difficult to obtain information from patients regarding self administration of antibiotics prior to their admission to the hospital. Self-administration of antibiotics is quite common in India due to over-the-counter sale of antibiotics and slack rules governing their purchase. The 16S rDNA sequence alignment from these fecal samples indicated that commensals and other members of the disturbed gut microflora or undiagnosed diarrheal agent may be contributing to the resistome. Although, metagenomics is the ultimate answer to the complex question about the structural and functional diversity of the microbiome, our findings obtained using common molecular methods is a highly significant pilot study indicating the presence and involvement of culturable and culture-resistant organisms in the development of AMR without the use of deep sequencing techniques and whole-genome sequence analysis. These findings are of immense importance for public health. It has enabled us to understand which resistant markers associated with different classes of antibiotics are present in the contemporary diarrheal microbiome and the relative abundance and frequency of the resistant markers will serve as parameters to understand and predict the potential threat that each of these classes of antibiotics will encounter leading to AMR crisis. Finally, our study will help in devising research strategies to intervene the transmission route of these genes to alleviate the problem of AMR.

## DATA AVAILABILITY STATEMENT

All relevant data is contained within the article. All datasets generated for this study are included in the article/**supplementary material**.

## ETHICS STATEMENT

The study was conducted with the approval of the funding bodies and fecal samples were collected with the approval of the Institutional Ethics Committee, NICED. Informed consent was obtained from the participants of the study. Information obtained from the study participants was kept confidential. Information was kept locked in office. Any personal information was never passed to the third party.

## REFERENCES


## AUTHOR CONTRIBUTIONS

RD planned, designed and executed experiments, analyzed the data, and wrote the paper. AM executed the experiments and analyzed the data. SD analyzed the data and wrote the paper.

## ACKNOWLEDGMENTS

The authors would like to thank the Department of Health Research and the Indian Council of Medical Research, Govt. of India for financial support for this research (Grant number R.12015/01/2018-HR). We would like to thank Dr. G. Balakrish Nair, for his inspiration and support for this work.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh. 2020.00061/full#supplementary-material


fish farming environment. Appl Environ Microbiol. (2001) 67:5675–82. doi: 10.1128/AEM.67.12.5675-5682.2001


**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 De, Mukhopadhyay and Dutta. 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.

# In vitro Activity of Apramycin Against Carbapenem-Resistant and Hypervirulent Klebsiella pneumoniae Isolates

Mingju Hao<sup>1</sup>† , Xiaohong Shi<sup>1</sup>† , Jingnan Lv<sup>2</sup>† , Siqiang Niu<sup>3</sup> , Shiqing Cheng<sup>4</sup> , Hong Du<sup>2</sup> , Fangyou Yu<sup>5</sup> , Yi-Wei Tang<sup>6</sup> , Barry N. Kreiswirth<sup>7</sup> , Haifang Zhang<sup>2</sup> \* and Liang Chen<sup>7</sup>

<sup>1</sup> Department of Laboratory Medicine, Shandong Provincial Qianfoshan Hospital, The First Hospital Affiliated With Shandong First Medical University, Jinan, China, <sup>2</sup> Department of Clinical Laboratory, The Second Affiliated Hospital of Soochow University, Suzhou, China, <sup>3</sup> Department of Laboratory Medicine, The First Affiliated Hospital of Chongqing Medical University, Chongqing, China, <sup>4</sup> Department of Laboratory Medicine, Shandong Provincial Hospital Affiliated to Shandong University, Jinan, China, <sup>5</sup> Department of Clinical Laboratory, Shanghai Pulmonary Hospital, Tongji University School of Medicine, Shanghai, China, <sup>6</sup> Department of Laboratory Medicine, Memorial Sloan Kettering Cancer Center; Department of Pathology and Laboratory Medicine, Weill Medical College of Cornell University, New York, NY, United States, <sup>7</sup> Hackensack-Meridian Health Center for Discovery and Innovation, Nutley, NJ, United States

#### Edited by:

Luciene Andrade Da Rocha Minarini, Federal University of São Paulo, Brazil

#### Reviewed by:

Qi Wang, Peking University People's Hospital, China Willames M. B. S. Martins, Federal University of São Paulo, Brazil Ana Paula D'Alincourt Carvalho-Assef, Oswaldo Cruz Foundation (Fiocruz), Brazil

#### \*Correspondence:

Haifang Zhang haifangzhang@sina.com †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 14 September 2019 Accepted: 27 February 2020 Published: 13 March 2020

#### Citation:

Hao M, Shi X, Lv J, Niu S, Cheng S, Du H, Yu F, Tang Y-W, Kreiswirth BN, Zhang H and Chen L (2020) In vitro Activity of Apramycin Against Carbapenem-Resistant and Hypervirulent Klebsiella pneumoniae Isolates. Front. Microbiol. 11:425. doi: 10.3389/fmicb.2020.00425 Objective: The emergence of carbapenem-resistant and hypervirulent Klebsiella pneumoniae (CR-hvKp) strains poses a significant public threat, and effective antimicrobial therapy is urgently needed. Recent studies indicated that apramycin is a potent antibiotic with good activity against a range of multi-drug resistant pathogens. In this study, we evaluated the in vitro activity of apramycin against clinical CR-hvKp along with carbapenem-resistant non-hvKp (CR-non-hvKp) isolates.

Methods: Broth microdilution method was used to evaluate the in vitro activities of apramycin, gentamicin, amikacin, imipenem, meropenem, doripenem, ertapenem and other comparator "last-resort" antimicrobial agents, including ceftazidime-avibactam, colistin and tigecycline, against eighty-four CR-hvKp and forty CR-non-hvKp isolates collected from three Chinese hospitals. Multilocus Sequence typing (MLST), molecular capsule typing (wzi sequencing) and antimicrobial resistance genes were examined by PCR and Sanger sequencing. Pulsed-field gel electrophoresis and next generation sequencing were conducted on selected isolates.

Results: Among the 84 CR-hvKp isolates, 97.6, 100, 97.6, and 100% were resistant to imipenem, meropenem, doripenem and ertapenem, respectively. Apramycin demonstrated an MIC50/MIC<sup>90</sup> of 4/8 µg/mL against the CR-hvKp isolates. In contrast, the MIC50/MIC<sup>90</sup> for amikacin and gentamicin were >64/>64 µg/mL. All CR-hvKp isolates were susceptible to ceftazidime-avibactam, colistin and tigecycline with the MIC50/MIC<sup>90</sup> values of 0.5/1, 0.25/0.5, 1/1, respectively. For CR-non-hvKp, The MIC50/<sup>90</sup> values for apramycin, gentamicin and amikacin were 2/8, >64/>64, and >64/>64 µg/mL, respectively. There were no statistical significance in the resistance rates of antimicrobial agents between CR-hvKp and CR-non-hvKp groups (p > 0.05). Genetic analysis revealed that all CR-hvKp isolates harbored blaKPC-2, and 94% (n = 79) belong to the ST11 high-risk clone. 93.6% (44/47) of amikacin or gentamicin resistant strains carried 16S rRNA methyltransferases gene rmtB.

Conclusion: Apramycin demonstrated potent in vitro activity against CR-hvKp isolates, including those were resistant to amikacin or gentamicin. Further studies are needed to evaluate the applicability of apramycin to be used as a therapeutic antibiotic against CR-hvKp infections.

Keywords: apramycin, susceptibility, carbapenem resistance, hypervirulent Klebsiella pneumoniae, aminoglycoside

## INTRODUCTION

fmicb-11-00425 March 13, 2020 Time: 17:22 # 2

Klebsiella pneumoniae is a clinically important pathogen, causing a wide range of diseases including pneumonia, urinary tract infections (UTIs), bloodstream infections, in hospital settings among neonates, elderly and immunocompromised individuals (Paczosa and Mecsas, 2016; Bengoechea and Sa Pessoa, 2019). Hypervirulent K. pneumoniae (hvKp) is an increasing recognized pathotype that is more virulent than classical K. pneumoniae (cKp), which showed the propensity to cause serious, lifethreatening infections in otherwise healthy individuals in the community (Shon et al., 2013; Gu et al., 2018).

Historically, hvKp isolates are susceptible to most antibiotics, however, multidrug-resistant hvKp, especially carbapenemresistant strains, are increasingly reported in the past few years (Cejas et al., 2014; Li et al., 2014; Siu et al., 2014). The emergence of carbapenem-resistant and hypervirulent K. pneumoniae (CRhvKp) was either due to the acquisition of a pLVPK-like virulence plasmid by classic carbapenem-resistant K. pneumoniae strains (Gu et al., 2018) or through acquisition of a carbapenemaseencoding plasmid by hvKp isolates (Shon et al., 2013). The rmpA, rmpA2, iutA, iucABCD and iroBCDN genes harbored by pLVPK-like virulence plasmid is responsible for the virulence phenotype (Yang et al., 2019). A recent study from China reported a fatal outbreak of ventilator associated pneumonia caused by ST11 CR-hvKp strains (Gu et al., 2018). More recently, one hypermucoviscos isolate with KPC-3 carbapenemase-encoding plasmid was firstly described in the United States, which also exhibited colistin heteroresistance (Wozniak et al., 2019). The rapid dissemination of CR-hvKp strains in different global regions underscores the urgent need of appropriate antibiotic therapy against these life-threating pathogens.

Aminoglycosides including amikacin, gentamicin and tobramycin, which belong to the 4,6-disubstituted deoxystreptamine (DOS) subclass, are among the few drugs that retain certain in vitro activity against CRE (Livermore et al., 2011). However, the acquisition of 16S rRNA methylases confer high level resistance to the 4,6-disubstituted DOS aminoglycosides, which imposes serious challenge for clinical management (Hu et al., 2017). Apramycin is of the 4-monosubstituted DOS subclass, markedly different in its chemical structure from other clinically used aminoglycoside antibiotics. Its unique structure makes apramycin molecules inherently resilient to almost all resistance determinants commonly found in multi-drug resistant strains (Juhas et al., 2019). Indeed, recent studies indicated that apramycin is a potent antibiotic with good activity against a range of clinical pathogens, including multidrugresistant Mycobacterium tuberculosis, carbapenem-resistant

Enterobacteriaceae, spectinomycin-resistant Neisseria gonorrhoeae and Staphylococcus aureus (Smith and Kirby, 2016; Hu et al., 2017; Truelson et al., 2018; Galani et al., 2019; Riedel et al., 2019). In order to evaluate whether apramycin can be served as an alternative antibiotic to treat the CR-hvKp infections, we examined the in vitro activity of apramycin against a collection of CR-hvKp isolates collected from three different hospitals in China. Its in vitro activity was compared with the results from other aminoglycosides (amikacin and gentamicin) and some "last resort" antibiotics, including colistin, tigecycline and ceftazidime-avibactam.

## MATERIALS AND METHODS

## Bacterial Isolates

Eighty-four unique (one isolate per patient) CR-hvKp isolates, including 65 isolates from our previous study (Yu et al., 2018a), were included. They were collected from three hospitals across different regions in China (Shandong, Jiangsu and Shanghai) between 2014 and 2019. These isolates were obtained from sputum (53.6%, 45/84), urine (14.3%, 12/84), blood (14.3%, 12/84), pus (4.8%, 4/84), ascites (3.6%, 3/84) and other sources (9.5%, 8/84). In this study, we defined an hvKp strain based upon demonstration of a positive string test (hypermucoviscosity) and co-harboring the genes rmpA (rmpA or rmpA2) and iutA as previously (Yu et al., 2018a). All 84 isolates were hypermucoviscous (string test >5 mm) and were positive for the rmpA/rmpA2 and iutA by PCR analysis (Yu et al., 2018b). In addition, 40 unique strains of carbapenem resistant non-hypervirulent K. pneumoniae (CRnon-hvKp) isolates, with negative hypermucoviscous phenotype and negative rmpA/rmpA2 and iutA PCR results, were collected from the same three hospitals, and were included to compare with the susceptibility results of CR-hvKp.

## Antibiotic Susceptibility Assay

The antimicrobial susceptibility was assessed via a minimum inhibitory concentration (MIC) broth microdilution method. Results were interpreted using Clinical Laboratory Standards Institute (CLSI) breakpoints (Clinical and Laboratory Standards Institute, 2019) with the exception of tigecycline and polymyxin B, which were interpreted using EUCAST breakpoints (The European Committee on Antimicrobial Susceptibility Testing, 2019). National Antimicrobial Resistance Monitoring System (NARMS) breakpoints for enteric bacteria were used for the interpretation of apramycin susceptibility, in which strains were classified as susceptible (MIC ≤ 8 µg/mL), intermediate (MIC = 16∼32 µg/mL), or resistant

(MIC ≥ 64 µg/mL) (National Antibiotic Resistance Monitoring System [NARMS], 2002). MICs were performed in duplicate on two separate days. Escherichia coli ATCC 25922 and Pseudomonas aeruginosa ATCC 27853 were used as the quality control strains in each experiment. K. pneumoniae ATCC 700603 was used as quality control strain for ceftazidime-avibactam which was tested against ceftazidime-avibactam and ceftazidime alone to confirm the activity of avibactam in the combination.

## Molecular Typing and Screening of Antibiotic Resistance Genes

All CR-Kp isolates including CR-hvKp and CR-non-hvKp were genotyped by multilocus Sequence Typing (MLST) and wzi sequencing (Brisse et al., 2013). Pulsed-field gel electrophoresis (PFGE) of Xbal-digested genomic DNA samples of 15 selected CR-hvKp isolates was performed with a CHEF MAPPER XA apparatus (Bio-Rad, United States). PFGE patterns were analyzed using GelJ software 2.0v (Heras et al., 2015).

A multiplex PCR was used to detect KPC, NDM, VIM, IMP, and OXA-48-like carbapenemase genes, followed by Sanger sequencing (Chen et al., 2011; Chavda et al., 2016; Evans et al., 2016). Common aminoglycoside modifying enzyme (AME) and RMT-encoding genes were determined by PCR and Sanger sequencing. AME coding genes, including aac(3<sup>0</sup> )-IIa, aac(3<sup>0</sup> )- IId, aac(3<sup>0</sup> )-IV, aac(6<sup>0</sup> )-IIa, aph(3<sup>0</sup> )-Ia, aph(3<sup>0</sup> )-Ib, ant(2<sup>0</sup> )-Ia and ant(3<sup>0</sup> )-Ia, and ARM/RMT methyltransferases coding genes, including armA, rmtA, rmtB, rmtC, rmtD and npmA (Wu et al., 2009), were examined. The primers used in this study were listed in **Supplementary Table S1**.

The apramycin resistant isolate, KpApr172, was subject to next generation sequencing using the Illumina HiSeq system (Illumina, San Diego, CA, United States). Genomic DNA was isolated using a Wizard <sup>R</sup> Genomic DNA Purification Kit (Promega, Madison, WI, United States). Sequencing reads were de novo assembled using Spades 3.12.0 (Bankevich et al., 2012). In silico multi-locus sequence typing was performed using MLST 2.0 (Larsen et al., 2012) while the acquired antimicrobial resistance genes were identified using ResFinder 3.0 (Zankari et al., 2012). The plasmid replicons in the sequenced isolates were identified using PlasmidFinder 2.0 (Carattoli et al., 2014). The whole-genome sequence of KpApr172 was deposited in the GenBank database under accession number WUJI00000000.

## Statistical Analysis

Chi-square and Fisher's exact analysis were employed to compare resistance rate of antimicrobial agents between CR-hvKp and CR-non-hvKp groups. Statistical analyses were performed using GraphPad Prism 6.0 (GraphPad Software, La Jolla, CA, United States). P value of <0.05 was considered to indicate statistically significant differences.

## RESULTS

## Susceptibility Testing

The in vitro activities of apramycin and other antimicrobials against 84 clinical isolates CR-hvKp and 40 CR-non-hvKp strains were summarized in **Table 1**. The CR-hvKp isolates showed high-level resistance to four tested carbapenems, with MIC50/<sup>90</sup> values 128/>256, 128/> 256, 64/128 and 128/>256 µg/mL for meropenem, imipenem, doripenem, and ertapenem, respectively, 97.6, 100, 97.6, and 100% of the isolates were resistant to imipenem, meropenem, doripenem and ertapenem. All isolates were resistant to ceftazidime and 97.6% (n = 82) isolates were resistant to aztreonam. Apramycin exhibited significantly better antimicrobial activity than the clinical standard-of-care aminoglycosides gentamicin and amikacin, with MIC50/<sup>90</sup> values 4/8 µg/mL). The MIC ranged from 1 to 16 µg/mL (**Figure 1**). By contrast, the MIC50/<sup>90</sup> values for gentamicin and amikacin were both >64/>64 µg/mL.

For CR-non-hvKp, 2.5, 55, and 57.5% of the isolates were resistant to apramycin, gentamicin and amikacin, respectively. The MIC50/<sup>90</sup> values for apramycin, gentamicin and amikacin were 2/8, >64/>64, and >64/> 4 µg/mL, respectively. One isolate was resistant to all three aminoglycosides including apramycin (MIC > 128 µg/mL). Five NDM producing CRnon-hvKp isolates were resistant to ceftazidime-avibactam (MIC > 128 µg/mL), but they remained susceptible to colistin and tigecycline. There were no statistical significance in the resistance rates of antimicrobial agents between CR-hvKp and CR-non-hvKp groups (p > 0.05).

## Molecular Typing and wzi Genes Sequencing

Multilocus sequence typing showed that most clinical CR-hvKp strains (94%, 79/84) belong to ST11, the most prevalent CR-Kp ST in China (Gu et al., 2018), while the remaining five isolates were from ST268 (n = 2), ST65 (n = 1), ST412 (n = 1), and ST595 (n = 1). Among the 79 ST11 isolates, 44 have the capsular polysaccharide wzi allele 64 (wzi64) (corresponding with capsular type KL64), and the other 35 isolates carry wzi209 (corresponding with capsular KL47). For CR-non-hvKp strains, 70% (28/40) of the isolates were identified as ST11, carrying wzi64 (n = 18) and wzi209 (n = 10), respectively. The other 12 CR-non-hvKp isolates were from ST2407 (n = 3), ST1308 (n = 3), ST25 (n = 2), ST307 (n = 1), ST1308 (n = 1), ST412 (n = 1), and ST48 (n = 1). To further explore the relatedness of individual CR-hvKp strains, we randomly selected 15 ST11 CR-hvKp strains from the three hospitals for PFGE analysis (**Supplementary Figure S1**). The results showed these isolates generally clustered by hospitals and capsular KL types, but displayed various PFGE profiles. Our results suggested that apramycin is actively against CR-hvKp and CR-non-hvKp isolates most prevalent STs from different geographical regions (Jiangsu, Shanghai, and Shandong) in China.

## Genotyping of Carbapenemase and Aminoglycoside Resistance Genes

The distribution of carbapenemase and aminoglycoside resistance genes were summarized in **Table 2**. All CR-hvKp isolates and 72.5% (29/40) of CR-non-hvKp isolates harbored blaKPC-2. Five CR-non-hvKp strains harbored blaNDM-1 (n = 4) and blaNDM-5 (n = 1). 93.6% (44/47) of the CR-hvKp and 95.6%



(22/23) of the CR-non-hvKp isolates that were resistant to amikacin or gentamicin carried the RMT gene rmtB.

Notably, one CR-non-hvKp isolate (KpApr172) was resistant to apramycin. PCR showed that it contains the aminoglycoside acetyltransferase 3-IV gene aac(3)-IV, which encodes resistance to apramycin, gentamicin, netilmicin and tobramycin. Further next generation sequencing revealed that KpApr172 was a strain ST11 and harbored the KL47 capsular type. Resistance gene mining indicated that KpApr172 harbored 17 antimicrobial resistance genes encoding resistance TABLE 2 | Prevalence of carbapenemase and aminoglycoside resistance genes in CR-hvKp and CR-non-hvKp isolates.


to β-lactams (blaKPC-2, blaNDM-5, blaCTX-M-3, and blaSHV-12), aminoglycosides [armA, aadA2, aac(3)-IId, aac(3)-IV and aph(4)-Ia], macrolid [msr(E) and mph(E)], bleomycin (ble), florfenicol (floR), quaternary ammonium compound (qacE11) and sulfonamide-trimethoprim (sul1, sul2 and dfrA12). The quinolone resistance-determining region (QRDR) genes gyrA, gyrB, parC and parE were further examined, and we observed mutations encoding amino acid substitutions at Ser83-Ile and Asp87-Gly within the QRDR regions of GyrA, and Ser80-Ile in ParC. Examination of the outer membrane protein OmpK35 and OmpK36 genes revealed a premature stop codon at AA63 in OmpK35, due to an adenine (A) deletion mutation at nt 82, and the glycine and aspartic acid insertion at AA134. In silico plasmid replicon typing identified five plasmid groups in KpApr172, belonging to incompatibility groups X3, R, FII and ColE-like (n = 2), respectively.

In order to further characterize the transferability of apramycin resistance gene aac(3)-IV, we attempted to transfer them via conjugation to a recipient E. coli J53 AzR strain (data not shown), however, we were not able to transfer the aac(3)- IV by conjugation, suggesting the gene may be harbored by a non-conjugative plasmid or on the chromosome. Examination of the aac(3)-IV-containing contig from the genome assemblies revealed that aac(3)-IV was located on a 13,397 bp segment, along with the resistance genes, floR, sul2 and aph(4)-Ia (**Figure 2**). BLASTn analysis showed that the sequences of this contig were identical (100% query coverage and 100% identities) to the corresponding regions in some plasmids, including pHNAH67 (KX246266, from E. coli of chicken), pCFSA122-1 (CP033224, from Salmonella enterica of pork), pSH16G4466 (MK477617,

from Salmonella enterica in human), pXGE1mcr (KY990887, from E. coli of cow feces), pHNSHP45-2 (KU341381, from E coli of pig feces), and chromosome of E. coli YSP8-1 (CP037910, from pig feces) from animal or human sources in China.

## DISCUSSION

Aminoglycosides resistance is mostly due to the chemical modification by aminoglycoside-modifying enzymes (AMEs), including acetyltransferases (AACs), phosphotransferases (APHs) and nucleotidyltransferases (ANTs) (Ramirez and Tolmasky, 2010). Some studies had shown AAC(6<sup>0</sup> )-Ib is one of the most prevalent and clinically relevant AMEs, which confers the resistance to amikacin and other aminoglycosides (Ramirez et al., 2013). A Greek study showed more than 70% of carbapenemases-producing Enterobacteriaceae strains carry aac(6<sup>0</sup> )-Ib gene (Galani et al., 2019). However, only 7.14% of CR-hvKp and 7.5% of CR-non-hvKp isolates were found to carry this gene in our study. By contrast, we found a high frequency of ARM/RMT methyltransferases gene in our study. ARM/RMT methyltransferases were found to confer high levels of resistance to most aminoglycosides other than apramycin (Galimand et al., 2003). ARM/RMT genes armA and rmt had been frequently described in clinical isolates throughout America, Europe and India (Fritsche et al., 2008). Our study showed 93.6% (44/47) CRhvKp and 95.6% (22/23) CR-non-hvKp amikacin or gentamycin resistant strains contained the 16S rRNA methylase gene rmtB. These results suggested that gentamycin and amikacin have limited activities against clinical CRKp isolates, including CR-hvKp, in Chinese hospitals.

In comparison, our results showed that all CR-hvKp isolates were susceptible to apramycin, whereas ∼ 50% of them were resistant to amikacin or gentamycin. Similarly, almost all CR-non-hvKp strains (39/40) were susceptible to apramycin irrespective of the presence of KPC-2 or NDM carbapenemase genes. In addition, our results were in accordance with previous studies that apramycin is not be inactivated by general AMEs and methylases (Livermore et al., 2011), making it an potential antibiotic against clinical amikacin or gentamycin resistant strains.

Nevertheless, apramycin resistance has also been described in clinical isolates worldwide (Threlfall et al., 1986; Chaslus-Dancla et al., 1989, 1991; Hunter et al., 1993; Johnson et al., 1995; Mathew et al., 2003; Zhang et al., 2009). Production of the acetylase aac(3<sup>0</sup> )-IV was the major mechanism underlying apramycin resistance in clinical isolates. The aac(3<sup>0</sup> )-IV gene encoding apramycin resistance was usually harbored by a mobile plasmid (Chaslus-Dancla et al., 1989; Zhang et al., 2009). In this study, we identified one aac(3<sup>0</sup> )-IV-harboring apramycin resistant isolate KpApr172 in CR-non-hvKp with a high MIC value of >128 µg/ml. Genomic analysis showed that KpApr172 belonged to the epidemic CRKp ST11 clone, while the aac(3<sup>0</sup> )- IV-harboring contigs showed highly identity to plasmid or chromosome sequences of isolates from animal sources. Since apramycin is only licensed for use in farm animals in China, we suspected that horizontal transfer of apramycin resistant genes from animals to human might be responsible for the dissemination of apramycin resistance (Yates et al., 2004). Therefore, aac(3<sup>0</sup> )-IV gene should be preferably screened for apramycin resistant isolates in clinical settings.

In this study no CR-hvKp isolates were found to be resistant to the "last resort" antimicrobial agents, including ceftazidimeavibactam, colistin and tigecycline, however, the treatment arsenal for CR-hvKp could be easily compromised by the spread of mobile resistant genes. Plasmid-mediated colistin and tigecycline resistance mechanisms, MCR-1 and Tet(x), have been described in China (Liu et al., 2016; He et al., 2019; Sun et al., 2019). In addition, the FDA approved cefetazidime-avibactam is ineffective against metallo-β-lactamase (e.g., NDM) producers. Furthermore, the original KPC-2 ST11 CR-hvKp outbreak study showed that despite being in vitro susceptible to tigecycline and colistin, long-term treatment with colistin or tigecycline (alone or in combination with several other antibiotics) was not able to eradicate KPC-2-producing ST11 CR-hvKp isolates, resulting in fatal outcomes (Gu et al., 2018). This convergence of increased resistance and hypervirulence of CR-hvKp underscores the need for the improvement of the antimicrobial treatment choices.

Our study showed apramycin had a potent in vitro activity to CR-hvKp irrespective of most commonly occurring carbapenemases and aminoglycoside modifying enzymes. In addition, recent studies had showed that apramycin was

associated with fewer ototoxic and nephrotoxic side effects in comparison with other human use aminoglycosides antibiotics (Perzynski et al., 1979; Matt et al., 2012; Ishikawa et al., 2019). These findings suggested apramycin and apramycin-derivative compounds could be a promising antibiotic for the treatment of CR-hvKp infections.

To our knowledge, this is the first study to examine the activities of apramycin against CR-hvKp clinical isolates. Our results indicate that apramycin remain highly active against CRhvKp. These data support a possible role for apramycin in the treatment of infections due to this hypervirulent and multidrugresistant strains. A further clinical evaluation of apramycin treatment efficacy is warranted.

## DATA AVAILABILITY STATEMENT

The whole-genome sequence of KpApr172 was deposited in the GenBank database under accession number WUJI00000000.

## AUTHOR CONTRIBUTIONS

MH, XS, and JL contributed to work, data analysis, and manuscript preparation. SN, SC, HD, and FY analyzed the data. Y-WT and BK prepared the manuscript. HZ and LC contributed to study design, data analysis, and manuscript preparation.

## REFERENCES


## FUNDING

This study was supported by the Jiangsu Overseas Visiting Scholar Program for University Prominent Young and Middleaged Teachers and Presidents, the Natural Science Foundation of Jiangsu Province (BK20181173), the grant from the Medicine and Health Science Technology Development Project of Shandong Province, China (2017WS007), Key Research and Development Project of Jiangsu Provincial Science and Technology Department (BE2017654), Gusu Health Youth Talent of Suzhou, Jiangsu Youth Medical Talents Program (QN-866 and 867), and the Cultivate Fund from Shandong Provincial Qianfoshan Hospital (No. QYPY2019NSFC0623). This work was in part supported by the grants from the National Institutes of Health (R01AI090155 to BK).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00425/full#supplementary-material

FIGURE S1 | Dendrogram of PFGE patterns of 15 CR-hvKp isolates from hospitals in Shanghai (a), Suzhou (b), and Jinan (c), respectively. PFGE typing revealed most of the CR-hvKp strains not belong to the same clone.

TABLE S1 | Primers used in this study.


**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 Hao, Shi, Lv, Niu, Cheng, Du, Yu, Tang, Kreiswirth, Zhang and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

fmicb-11-00425 March 13, 2020 Time: 17:22 # 7

# Gastrointestinal Carriage of Vancomycin-Resistant Enterococci and Carbapenem-Resistant Gram-Negative Bacteria in an Endemic Setting: Prevalence, Risk Factors, and Outcomes

Alexandra Vasilakopoulou<sup>1</sup> , Polyxeni Karakosta<sup>1</sup> , Sophia Vourli 1,2, Aikaterini Tarpatzi 1,2 , Paraskevi Varda<sup>2</sup> , Maria Kostoula<sup>2</sup> , Anastasia Antoniadou2,3 and Spyros Pournaras 1,2 \*

<sup>1</sup> Clinical Microbiology Laboratory, Medical School, Attikon University General Hospital, National and Kapodistrian University of Athens, Athens, Greece, <sup>2</sup> Infection Control Committee, Attikon University General Hospital, Athens, Greece, <sup>3</sup> 4th Department of Internal Medicine, Medical School, Attikon University General Hospital, National and Kapodistrian University of Athens, Athens, Greece

## Edited by:

Raffaele Zarrilli, University of Naples Federico II, Italy

## Reviewed by:

Martina Barchitta, University of Catania, Italy Stephan Göttig, University Hospital Frankfurt, Germany

#### \*Correspondence:

Spyros Pournaras spournaras@med.uoa.gr

#### Specialty section:

This article was submitted to Infectious Diseases–Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

Received: 31 October 2019 Accepted: 14 February 2020 Published: 18 March 2020

#### Citation:

Vasilakopoulou A, Karakosta P, Vourli S, Tarpatzi A, Varda P, Kostoula M, Antoniadou A and Pournaras S (2020) Gastrointestinal Carriage of Vancomycin-Resistant Enterococci and Carbapenem-Resistant Gram-Negative Bacteria in an Endemic Setting: Prevalence, Risk Factors, and Outcomes. Front. Public Health 8:55. doi: 10.3389/fpubh.2020.00055 Background: Gastrointestinal carriage of vancomycin-resistant enterococci (VRE) and carbapenem-resistant Gram-negative bacteria (CRGN) constitutes a major public health concern as it may be followed by clinical infection development or lead to intra-hospital dissemination. Detection of carriers and implementation of infection control measures are essential in every hospital. In this study we determined the point prevalence of VRE and CRGN in the fecal flora of the inpatients of a tertiary university hospital in Greece. We determined risk factors for carriage and examined the impact of carriage on hospital outcomes.

Materials/Methods: A point prevalence study of VRE/CRGN rectal carriage of inpatients was conducted on March 2018. Specimens were selectively cultured for VRE/CRGN, microorganisms were biochemically identified, submitted to antibiotic susceptibility testing, and tested for carbapenemase production. Data on potential risk factors and hospital outcomes were collected at the time of culture and until hospital discharge. Multivariable logistic and linear regression models were used, adjusting for confounders.

Results: Four hundred ninety-one patients were enrolled in the study. Of them, 64 (13.0%) were positive for VRE carriage, 40 (8.2%) for CRGN, and 10 patients (2.1%) for both VRE and CRGN. VRE carriage was independently associated with age over 65 years (adjusted OR: 2.4 [95%CI: 1.3, 4.5]) and length of stay (LOS) before rectal sampling (OR: 1.1 [95%CI: 1.0, 1.1]). Carriage of CRGN was associated with 11 days increase of LOS after rectal sampling (β-coef: 11.4 [95%CI: 1.6, 21.2]), with a 3.5-fold increased risk of acquiring a resistant pathogen after rectal swabbing (RR: 3.5 [95%CI 1.2, 9.9]) and with a 6-fold increased risk of mortality (RR: 6.1 [95%CI: 2.1, 17.9]), after adjusting for sex, age, and comorbidity index.

**377**

Conclusions: High prevalence rates were found for VRE and CRGN carriage among the inpatients of our hospital. Prolonged hospitalization and age were independent risk factors for VRE carriage, while CRGN carriage was associated with increased risk of acquiring a resistant pathogen, prolonged hospital stay, and increased mortality.

Keywords: vancomycin-resistant enterococci, carbapenem-resistance, carriage, risk-factors, mortality, length of stay

## INTRODUCTION

The wide dissemination of carbapenem-resistant Gram-negative bacteria (CRGN) and vancomycin-resistant enterococci (VRE) limits therapeutic alternatives and represents a global public health threat (1). The consequences of multidrug-resistant (MDR) bacterial infections include high morbidity and mortality and considerable economic loss (2). A recent study estimated the impact of infections caused by antimicrobial-resistant bacteria in countries within the EU and the European Economic Area for 2015. It was estimated that ∼670,000 infections with resistant bacteria were documented in EARS-Net data, with these infections accounting for ∼33,000 attributable deaths and 870,000 disability-adjusted life-years (DALYs). Notably, Greece and Italy contributed the highest burden among all participating countries and, for Greece, most of the infections were due to carbapenem- or colistin-resistant bacteria (3). The hospital environment seems to serve as the breeding grounds for MDR organisms (MDRO) (4). Asymptomatic rectal carriage of these organisms may precede infection and constitutes a reservoir for transmission that may remain unidentified in hospitals that do not implement active surveillance testing (5).

While Greece is considered one of the most common countries in Europe for antimicrobial resistance (6), only a limited number of studies have focused on MDRO rectal carriage and colonization to assess their prevalence, risk factors, and associated adverse outcomes (7–11). A previous study from our hospital has reported a prevalence of 14.3% of VRE carriage among hospitalized patients, identified invasive devices and duration of antimicrobial treatment as risk factors, and found that VRE carriage was not an independent predictor of mortality (12). Similar studies from other regions have reported VRE carriage rates ranging from 2 to 37% (13– 15), while prevalence rates for CRGN rectal carriage ranged from 5.3 to 52% (16–18). At the same time, there is an increased risk of carbapenem-resistant enterobacteriaceae (CRE) infection and mortality in patients who test positive for carriage of CRE (19, 20). To the best of our knowledge, there is no previous study from Greece focusing on both VRE and CRGN rectal carriage, exploring respective risk factors and adverse outcomes.

The primary objective of the present study was: (i) to determine the prevalence of rectal carriage of VRE and CRGN, (ii) to identify risk factors for VRE/CRGN rectal carriage, and (iii) to examine the impact of VRE/CRGN rectal carriage on hospital outcomes in inpatients of a University General Hospital in Greece.

## MATERIALS AND METHODS

## Subjects

The University Hospital "Attikon" in Athens is a modern tertiary care teaching hospital and is the largest in the West Attica region (2,000,000 population), with 750 beds in total and >71.000 admissions/year. The hospital attends to cases of high complexity in internal medicine and surgery.

The first part of the present project was a point prevalence study of VRE/CRGN rectal carriage of hospital patients that was conducted on March 22nd and 23rd, 2018. Adult patients hospitalized in all surgical and internal medicine departments were surveyed by obtaining rectal swab cultures. In total, 17 medical and surgical wards participated in the study: general internal medicine, cardiology, dermatology, neurology, respiratory medicine, obstetrics/gynecology, cardiothoracic surgery, neurosurgery, urology, otorhinolaryngology, orthopedics, vascular surgery, hematology, oncology, nephrology, gastroenterology, and general surgery. The special and intensive care units were not included in the study, as they were already on active surveillance for VRE/CRGN carriage. The psychiatric ward was also excluded because patients lacked the mental competency necessary to participate. The second part of the project was a cohort study that included all participants from the first part and followed them from the time of rectal swab culture until death or discharge from the hospital. Face-to-face completed questionnaires together with medical records and communication with physicians were used to obtain information on potential risk factors at the time of rectal swabbing, while all examined outcomes were extracted from medical records retrospectively. The study was approved by the institutional review board of the hospital (62,17/10/2017) and all patients provided informed consent after a complete description of the study.

## Culture, Identification, and Susceptibility Testing

A rectal swab was obtained from every consenting hospitalized patient. The swabs were transferred by using transport swabs in Amies transport medium (Biomedics, Madrid, Spain) and were transported to the microbiology laboratory for selective culture of VRE and CRGN. Bile-esculin agar with vancomycin (6 mg/L) and MacConkey agar with meropenem (1 mg/L) were used for selective cultivation. Microorganisms were biochemically identified by Phoenix automated microbiology system (BD Diagnostic Systems, Sparks, MD) and submitted to antibiotic susceptibility testing, according to EUCAST 2018 guidelines and breakpoints. The combination disk test was used for screening carbapenemase production using meropenem 10 µg disks (BIO-RAD, Marnes-la-Coquette, France) with or without inhibitors [phenyl boronic acid (PBA), ethylenediaminetetraacetic acid (EDTA)] (21). The guidelines of the European Committee on Antimicrobial Susceptibility Testing (EUCAST) were applied for detection of resistance mechanisms and specific resistances of clinical and/or epidemiological importance (22). A meropenem disk with PBA and EDTA was also included to detect double carbapenemase producers (KPC and VIM), which have been found in Greek hospitals since 2009 (21, 23). The immunochromatographic assay, NG-test CARBA 5 (NG Biotech, 35480 Guipry, France) that discriminates KPC, IMP, VIM, NDM, and OXA-48-like producers, was also used.

## Rectal Carriage of VRE/CRGN

All patients that tested positive for VRE and/or CRGN in the rectal swab culture were defined as carriers. Standard infection control measures to reduce transmission were used in these cases. The health personnel had to implement contact precautions (gloves, gowns) for all encounters with the carriers. The wards focused on thorough cleaning of the environment surrounding the positive patient, especially the patient care equipment. Whenever it was possible, the patient was isolated in a single room (24, 25).

## Risk Factors for VRE/CRGN Rectal Carriage

Following informed consent and after obtaining the rectal swab, every hospitalized patient completed a face-to-face questionnaire which was captured in a standardized form. These data, along with parameters retrieved from the patients' records, included information on age, gender, ward, length of stay (LOS) before rectal swabbing, transfer from another hospital, comorbidities, presence of indwelling medical devices, chronic immobilization, last year hospitalization or ICU admission, and specific therapies, such as immunosuppressive therapy, antineoplastic, or antimicrobial chemotherapy.

Comorbidities included: chronic kidney disease, diabetes, dermatologic lesions, hematological malignancy, solid organ malignancy, metastatic disease, neurologic disease, heart failure, coronary artery disease, chronic liver disease, cerebrovascular disease, peripheral vascular disease, and chronic obstructive pulmonary disease. A slightly modified Charlson comorbidity index (CCI) (26) that predicts the 10-years mortality for a patient having a range of 17 comorbid conditions was also calculated. Each condition is assigned with a score of 1, 2, 3, or 6, depending on the risk of death associated with this condition; the scores are then summed up and give a total score which predicts mortality. The clinical conditions and scores are as follows: One for each: myocardial infarct, heart failure, peripheral vascular diseases, dementia, cerebrovascular disease, chronic lung diseases, connective tissue diseases, ulcer, and mild chronic liver diseases. Two for each: hemiplegia, moderate or severe kidney diseases, diabetes with or without complications, tumor, leukemia, lymphoma. Three for each: moderate or severe liver disease. Six for each: metastatic solid tumor, AIDS. In our modified version of CCI (modified CCI), we used chronic immobilization instead of hemiplegia and we did not use data on mild chronic liver diseases, since they were not present in any patient.

Indwelling medical devices included central lines, urinary catheters, pacemakers, and other devices, such as external wound drains, enteral feeding tubes, endotracheal tubes, and tracheostomies.

Chronic immobilization is defined as loss of anatomical movement due to alteration of physiological function, which in daily practice is commonly defined as more than three-day-bed rest or inability to perform mobile activity on a bed, transfer, or ambulation (27, 28).

Last year, hospitalization or ICU admission included hospitalization in an acute care hospital (ward or ICU, respectively) for two or more days in the past 1 year.

Previous antibiotic therapy was defined as prescription of antibiotics for at least 2 days within the past 1 year. Immunosuppressive therapy included administration of steroids, cyclophosphamide, azathioprine, methotrexate, mycophenolate mofetil, and calcineurin inhibitors during the last year.

## Hospital Outcomes

Hospital outcomes were collected prospectively during hospitalization and included: (i) Mortality: death from any cause during hospitalization, (ii) Hospital LOS, (iii) LOS after rectal swabbing, and (iv) Isolation of a resistant pathogen [VRE, methicillin-resistant Staphylococcus aureus (MRSA), carbapenem-resistant Gram-negative bacteria and Gramnegative bacteria resistant to three or more of: beta-lactams, aminoglycosides, quinolones, co-trimoxazole]. from: (a) any clinical culture, and (b) blood, after rectal swabbing.

## Statistical Analysis

Statistical analysis was performed using the statistical package STATA, version 13 (StataCorp, College Station, TX). Univariate associations between background characteristics and VRE/CRGN rectal carriage were studied using Pearson's chi-square test for categorical variables (with Fisher's exact test for groups with <5 subjects expected in a cell) and nonparametric Kruskal–Wallis tests for continuous non-normally distributed variables (tested by the Shapiro–Wilk normality test). Since potential risk factors and VRE/CRGN rectal carriage were measured in a cross-sectional design, associations of potential risk factors with VRE/CRGN rectal carriage were estimated with univariate logistic regression models. Estimated associations were described as odds ratios (OR) with 95% confidence intervals. All variables with a p ≤ 0.050 in univariate analysis were included in a multiple regression model to examine independent risk factors for VRE/CRGN rectal carriage. In the prospective part of our study and in order to estimate the risk of VRE/CRGN rectal carriage on hospital outcomes, we used multivariable log-binomial or log-Poisson (if convergence failed) regression models to estimate relative risks (RRs) with 95% CIs for categorical outcomes (resistant pathogen in any culture and in blood after rectal swabbing, mortality) and multivariable linear regression models to estimate β coefficients with 95% CIs for continuous outcomes (hospital LOS, LOS after rectal swab). Modified Charlson Comorbidity Index, age, and sex were included as confounders a priori in all analyses. All association testing was conducted assuming a p ≤ 0.050 significance level.

## RESULTS

In total, 507 adult patients hospitalized in medical wards were contacted to participate and 491 provided rectal swabs were included in the analysis (participation rate: 96.6%) (**Figure 1**). Baseline characteristics of the study population according to rectal carriage of CRGN and VRE are presented in **Table 1**. Forty patients were identified as CRGN carriers (prevalence: 8.2%) and 64 as VRE carriers (prevalence: 13.0%). Of the carrier patients, 30% were hospitalized in single rooms. All VRE isolates were identified as VanA-phenotype E. faecium (high level resistance to vancomycin and teicoplanin). Molecular detection of van genes was not performed The respective prevalence for carbapenem-resistant (CR) enterobacteriaceae (CRE), CR-Acinetobacter spp., and CR-Pseudomonas spp. carriers was 5.9, 1.8, and 1.2%, respectively. Four different types of carbapenemases were detected in CRE colonized patients: KPC, 65.5% (n = 19); NDM, 24.1% (n = 7); VIM, 6.9% (n = 2); and OXA-48, 3.5% (n = 1). Patients colonized with VRE tended to be older (>65 years) compared to non-colonized patients. Regarding CRGN carriers, they were characterized by a higher modified CCI, were more likely to have a urine catheter, and to have been hospitalized in a medical ward/ICU or to have received antibiotics during the last year, compared with non-carriers. Moreover, both groups of CRGN and VRE carriers tended to have longer LOS before rectal swabbing. Mortality for non-carriers was 4.8% (n = 19), lower than that for CRGN (40%, p < 0.001) and VRE carriers (15.9%, p < 0.001). Hospital outcomes, such as hospital LOS and subsequent resistant pathogens isolated from blood, were also more frequent in both subgroups of CRGN and VRE carriers, compared to non-colonized patients (**Table 1**).

**Table 2** presents the results of the univariable analysis on the association between potential risk factors and CRGN and VRE carriage. The detection of VRE carriage was associated with age over 65 years (OR: 1.9 [95%CI: 1.1, 3.4]) and LOS before rectal swab (OR: 1.1 [95%CI: 1.0, 1.1]). In a multivariate logistic regression model, these variables were also independently associated with VRE carriage [adjusted OR: 2.4 [95%CI: 1.3, 4.5] and 1.1 [95%CI: 1.0, 1.1], respectively]. CRGN carriage was associated with LOS before rectal swab [OR: 1.1 (95%CI: 1.0, 1.1)], hospitalization [OR: 10.4 (95%CI: 1.3, 82.3)] and ICU admission during the past 1 year [OR: 9.3 (95%CI: 2.4, 36.0)], presence of a urinary catheter [OR: 3.5 (95%CI: 1.0, 12.4)] and pacemaker [OR: 6.0 (95%CI: 1.1, 32.6)], metastatic disease [OR: 4.9 (95%CI: 1.2, 20.4)], and high modified CCI [OR: 1.2 (95%CI: 1.0, 1.3)]. When these variables were included in the multivariable model, none of them were identified as an independently associated risk factor for CRGN carriage.

**Table 3** shows the multivariable analysis estimating the effect of CRGN and VRE rectal carriage on hospital outcomes, after

TABLE 1 | Comparison of baseline characteristics between MDRO carriers and non-carriers.


IQR, interquartile range; LOS, length of stay; ICU, intensive care unit; CCI, Charlson comorbidity index; CRGN, carbapenem-resistant Gram-negative bacteria; VRE, vancomycinresistant enterococci.

Bold indicates significant differences (p ≤ 0.050) of ANOVA for continuous variables and x<sup>2</sup> analysis for categorical variables.

Numbers may not correspond to the total due to missing numbers.

(n = 491).

adjusting for sex, age, and modified CCI. Carrying CRGN was associated with 38 days increased hospital LOS (β-coef: 38.0, [95%CI: 22.6, 53.4]) and 11 days increase in LOS after rectal swabbing (β-coef: 11.4, [95%CI: 1.6, 21.2]). More importantly, it is associated with a 3.5-fold increased risk for acquiring a resistant pathogen after rectal swabbing (RR: 3.5, [95%CI 1.2, 9.9]) and with a 6-fold increase risk for mortality (RR: 6.1, [95%CI: 2.1, 17.9]). We then separately examined different TABLE 2 | Associations between potential risk factors and carriage of CRGN and VRE.


OR, odds ratio; CI, confidence interval; LOS, length of stay; ICU, intensive care unit; CCI, Charlson comorbidity index; NA, not applicable. Bold indicates significant differences (p ≤ 0.050). \*Numbers do not correspond to the total in every risk factor, due to missing data.

#### TABLE 3 | Risk of CRGN and VRE carriage for hospital outcomes.


RR, relative risk; β-coef, beta coefficients; CI, confidence interval; LOS, length of stay; CR, carbapenem-resistant; VRE, vancomycin-resistant enterococci; NA, not applicable. Models were adjusted for sex, age and modified Charlson comorbidity index. The values in bold are statistically significant.

CRGN subgroups in order to identify the specific pathogen underlying the observed associations; sample size was marginal for firm conclusions in some cases, but CR-enterobacteriaceae carriage was predictive of all outcomes (**Supplementary Table 1**). Regarding VRE carriers, although a statistical significant risk was found for hospital LOS, resistant pathogen in blood after rectal swabbing, and mortality in the crude model (RR: 9.9, [95%CI: 4.8, 15.1], RR: 3.9, [95%CI: 1.3, 11.5] and RR: 3.3, [95%CI: 1.6, 6.8] respectively), statistical significance was in all circumstances not shown after adjusting for confounders.

## DISCUSSION

In the present study, we calculated simultaneously, for the first time, prevalence rates for VRE and CRGN carriage among inpatients of a Greek tertiary hospital and recognized prolonged hospitalization and age as independent risk factors for VRE carriage. We also showed that CRGN carriage is associated with increased risk of acquiring a resistant pathogen after rectal swabbing, prolonged hospital stays, and increased mortality.

The present study revealed a high prevalence of VRE and CRGN carriage among inpatients of our hospital (13.0 and 8.2% respectively). A previous study from the same hospital had determined a VRE carriage rate of 14.3% (29). Other studies have reported carriage rates that varied with geographic location and the general condition of patients (critically ill or not) (16, 17, 30– 33). Greece is regularly regarded as an environment with highselection pressure for the emergence of extensively drug-resistant Gram-negative bacilli in Europe, due to the over-consumption of antimicrobials both in the community and in the hospitals (34).

The multivariable analysis of potential risk factors showed that prolonged hospitalization and advanced age represent independent risk factors for VRE carriage. VRE can survive on environmental surfaces for a long time, and environmental contamination has been identified as a potential risk factor for VRE transmission to healthcare workers' hands and gloves and, subsequently, to patients (35, 36). Hand hygiene compliance rates in Greek hospitals have been reported to range from 22 to 43% (37–39). Since Greece is currently in the midst of a financial crisis, all Greek hospitals suffer from reductions in nursing and cleaning personnel, which may lead to compromises in infection control practices (40–42). All these factors might play a role in the further spread of VRE, especially in cases of prolonged hospitalization. In addition, long hospital stay may be associated with increased antibiotic consumption, which can also contribute to VRE selection and carriage (43). Moreover, the association of old age with VRE carriage may be attributed to several factors, such as alterations of the immune system, malnutrition, and social and economic factors (32, 44–46).

The univariate analysis revealed several risk factors for CRGN carriage: presence of indwelling devices (urinary catheter, pacemaker), CCI, metastatic disease, prolonged hospitalization before the fecal swab sampling, last year hospitalization, and ICU admission. Although none of these factors were independently associated with CRGN carriage in the multivariable logistic regression model, they all deserve to be considered. Indwelling devices are related to disease severity (47) and are recognized risk factors for healthcare-associated carriage and infection with MDR pathogens (48). Furthermore, serious underlying disease as a risk factor for CRGN carriage has been previously described (49) and can be explained by the patients' exposure to invasive procedures, as well as from impaired host defenses and extensive use of antibiotics. Almost invariably, these patients have longer hospitalization than patients with less severe illnesses (4). Also, previous hospitalization was identified as a risk factor for CRGN carriage. Our hospital is a referral center, where patients from all over the country are admitted for treatment. The majority of these patients have complex medical diseases, prolonged exposure to a healthcare setting, and an extensive use of antibiotics.

Gut carriage of CRGN was independently associated with increased hospital LOS, risk of acquiring a resistant pathogen after rectal swabbing, and mortality. Several studies have demonstrated that gut carriage and subsequent colonization by K. pneumoniae in hospitalized patients is associated with a greater risk of infection by the colonizing strains (50, 51). Little is known, though, about the mechanisms that promote progression from carriage to infection. Increased total LOS among CRE carriers has been previously described (19, 49), as well as the association with increased rate of CRE infections and high mortality (19, 20). Previous research has demonstrated that CRE infections are associated more often with sepsis and increased early mortality rate (52), particularly in vulnerable patients such as pediatric, geriatric, immunosuppressed, hospitalized, and chronically ill (53, 54).

Our study has some limitations. We might have underestimated the prevalence of OXA-48 producers that weakly hydrolyze carbapenems (MIC for meropenem lower than 1 mg/L) because of the selective media we used for the culture (55). Moreover, our study lacks enough power to establish a causal relationship between possible risk factors and MDRO carriage, since the study design was cross-sectional. We were also not able to gather detailed information regarding previous exposure to antibiotics such as carbapenems and vancomycin or to assess their impact on MDRO carriage. In addition, we didn't collect subsequent rectal samples in order to differentiate between transient carriage and colonization. More importantly, we did not collect the required data in order to define specific types of infections from resistant pathogens; instead, we collected data on isolation of a subsequent resistant pathogen from a clinical sample. Our analysis was limited to all-cause, rather than attributable, mortality. Thus, we are unable to determine whether the mortality rates of CRGN carriers were directly attributable to infection or were more likely to occur in patients with other fatal illnesses. Furthermore, although statistically significant conclusions were produced, confidence intervals were wide due to the relatively small sample size. Finally, although we tried to incorporate all known potential risk factors for carriage, infection, and mortality, we acknowledge that residual confounding from unmeasured covariates is still possible.

Nevertheless, the results of this study provide valuable information about the CRGN and VRE carriage burden in our hospital and can be used for improving our infection control strategy. After this study, the infection control team has been reinforced with more personnel. The health care employees of our hospital have been informed about the findings of the study and further educational activities on effective infection control practices have been provided. A new point prevalence study of MDRO carriage is scheduled for the first trimester of 2020 in order to compare results and assess the impact of the intensified infection control effort.

In conclusion, VRE and CRGN represent a serious public health problem and carrier patients represent a silent threat for hospitals. Efforts to limit the spread of MDRO need to be optimized.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## ETHICS STATEMENT

The studies involving human participants were reviewed and approved by Institutional Review Board, University General Hospital Attikon. The patients/participants provided their written informed consent to participate in this study.

## REFERENCES


## AUTHOR CONTRIBUTIONS

SP, AA, and SV contributed to the study design and reviewed the manuscript. AV and PK contributed to the data analysis and manuscript preparation. SV and AT contributed to the data analysis. PV and MK collected and reviewed patients' data and contributed to the data analysis.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh. 2020.00055/full#supplementary-material


bacteria in hospitalized patients. Clin Microbiol Infect. (2013) 20:1– 55. doi: 10.1111/1469-0691.12427


**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 Vasilakopoulou, Karakosta, Vourli, Tarpatzi, Varda, Kostoula, Antoniadou and Pournaras. 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.

# Inactivation of Glutamine Synthetase-Coding Gene glnA Increases Susceptibility to Quinolones Through Increasing Outer Membrane Protein F in Salmonella enterica Serovar Typhi

Ana R. Millanao1,2,3, Aracely Y. Mora<sup>1</sup> , Claudia P. Saavedra4,5, Nicolás A. Villagra6,7 , Guido C. Mora<sup>8</sup> and Alejandro A. Hidalgo<sup>1</sup> \*

<sup>1</sup> Escuela de Química y Farmacia, Facultad de Medicina, Universidad Andres Bello, Santiago, Chile, <sup>2</sup> Instituto de Farmacia, Facultad de Ciencias, Universidad Austral de Chile, Valdivia, Chile, <sup>3</sup> Facultad de Ciencias Químicas y Farmacéuticas, Universidad de Chile, Santiago, Chile, <sup>4</sup> Laboratorio de Microbiología Molecular, Departamento de Ciencias Biológicas, Facultad de Ciencias de la Vida, Universidad Andres Bello, Santiago, Chile, <sup>5</sup> Millennium Institute on Immunology and Immunotherapy, Departamento de Ciencias Biológicas, Facultad de Ciencias de la Vida, Universidad Andres Bello, Santiago, Chile, <sup>6</sup> Escuela de Tecnología Médica, Universidad Andres Bello, Santiago, Chile, <sup>7</sup> Departamento de Laboratorios Clínicos, Escuela de Medicina, Pontificia Universidad Católica de Chile, Santiago, Chile, <sup>8</sup> Instituto de Investigación Interdisciplinar en Ciencias Biomédicas SEK, Facultad de Ciencias de la Salud, Santiago, Universidad SEK, Santiago, Chile

Edited by: Kunihiko Nishino, Osaka University, Japan

## Reviewed by:

Ricardo Oropeza, National Autonomous University of Mexico, Mexico Xiangmin Lin, Fujian Agriculture and Forestry University, China

> \*Correspondence: Alejandro A. Hidalgo alejandro.hidalgo@unab.cl

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 30 July 2019 Accepted: 27 February 2020 Published: 20 March 2020

#### Citation:

Millanao AR, Mora AY, Saavedra CP, Villagra NA, Mora GC and Hidalgo AA (2020) Inactivation of Glutamine Synthetase-Coding Gene glnA Increases Susceptibility to Quinolones Through Increasing Outer Membrane Protein F in Salmonella enterica Serovar Typhi. Front. Microbiol. 11:428. doi: 10.3389/fmicb.2020.00428 Ciprofloxacin is the choice treatment for infections caused by Salmonella Typhi, however, reduced susceptibility to ciprofloxacin has been reported for this pathogen. Considering the decreased approbation of new antimicrobials and the crisis of resistance, one strategy to combat this problem is to find new targets that enhances the antimicrobial activity for approved antimicrobials. In search of mutants with increased susceptibility to ciprofloxacin; 3,216 EZ-Tn5 transposon mutants of S. Typhi were screened. S. Typhi zxx::EZ-Tn5 mutants susceptible to ciprofloxacin were confirmed by agar diffusion and MIC assays. The genes carrying EZ-Tn5 transposon insertions were sequenced. Null mutants of interrupted genes, as well as inducible genetic constructs, were produced using site-directed mutagenesis, to corroborate phenotypes. SDS-PAGE and Real-time PCR were used to evaluate the expression of proteins and genes, respectively. Five mutants with increased ciprofloxacin susceptibility were found in the screening. The first confirmed mutant was the glutamine synthetase-coding gene glnA. Analysis of outer membrane proteins revealed increased OmpF, a channel for the influx of ciprofloxacin and nalidixic acid, in the glnA mutant. Expression of ompF increased four times in the glnA null mutant compared to WT strain. To understand the relationship between the expression of glnA and ompF, a strain with the glnA gene under control of the tetracycline-inducible Ptet promoter was created, to modulate glnA expression. Induction of glnA decreased expression of ompF, at the same time that reduced susceptibility to ciprofloxacin. Expression of sRNA MicF, a negative regulator of OmpF was reduced to one-fourth in the glnA mutant, compared to WT strain. In addition, expression of glnL and glnG genes (encoding the two-component system NtrC/B that may positively regulate OmpF) were increased in the glnA mutant. Further

studies indicate that deletion of glnG decreases susceptibility to CIP, while deletion of micF gene increases susceptibility CIP. Our findings indicate that glnA inactivation promotes ompF expression, that translates into increased OmpF protein, facilitating the entry of ciprofloxacin, thus increasing susceptibility to ciprofloxacin through 2 possible mechanisms.

Keywords: glnA, ciprofloxacin, OmpF, S. Typhi, transposon

## INTRODUCTION

Salmonella Typhi is the etiological agent of typhoid fever, endemic in many developing countries worldwide. This disease is exclusive of humans with a mortality rate of 10%. The emergence of the extensively drug-resistant S. Typhi H58 strains in Pakistan, which are resistant to the first-line drugs (ampicillin, chloramphenicol, and cotrimoxazole), fluoroquinolones, and third-generation cephalosporin, is a real threat with potential to become typhoid fever untreatable (Cabello, 2018; Johnson et al., 2018; Klemm et al., 2018). Thus, typhoid fever requires alternative pharmacological treatment, including new antimicrobials and development of new combined therapies to revitalize existing antibiotics to prolong its useful life and slow down the emergence of resistance (Cottarel and Wierzbowski, 2007; Fajardo et al., 2008; Liu et al., 2010; Rodas et al., 2010; Sabbagh et al., 2012; González-Bello, 2017).

To find mutants with increased susceptibility to Ciprofloxacin (CIP), a screening was performed over 3,216 insertional mutants of S. Typhi STH2370 with the EZ-Tn5 TM transposon exposed to sub-MIC concentrations of CIP. We identified both: a small number of known and new genetic determinants of resistance to quinolones. One such mutant in the glnA gene that encodes for glutamine synthetase (GS) was further characterized. GS produces glutamine from glutamate and ammonia and has a crucial role in nitrogen metabolism. The internal concentration of glutamine is the main intracellular signal for regulating nitrogen availability in enteric bacteria (Zimmer et al., 2000; Switzer et al., 2018). Nitrogen is essential for the biosynthesis of macromolecules in bacteria; thus, the adaptive response to metabolic stress induced by starvation of nitrogen (as it is the case of glutamine auxotrophic bacteria) could affect bacterial physiology, including the susceptibility to antimicrobials which today is known to be modulated by metabolism (Maria-Neto et al., 2012; Peng et al., 2015; Vestergaard et al., 2017; Cui et al., 2019).

One example that relates nitrogen metabolism and susceptibility to antimicrobials is the observation of Escherichia coli strains resistant to magainin I (a cationic peptide) which overexpress GS (Maria-Neto et al., 2012). In Streptococcus pneumoniae it was demonstrated that glnA was repressed in the presence of penicillin, as well as, the inhibition of GS enhanced susceptibility to penicillin. Hence, glutamine conferred a protective role against penicillin when added to the culture medium (El Khoury et al., 2017). In the same line, methicillinresistant Staphylococcus aureus and methicillin-susceptible S. aureus with lower expression of GS decreased their level of methicillin resistance (Gustafson et al., 1994). It is proposed that in S. aureus GS participates in the production of constituents of the cell envelope, therefore maintaining the cell wall thickness and the level of crosslinking on peptidoglycan (Gustafson et al., 1994; Lima et al., 2013). In Mycobacterium tuberculosis, GS plays a crucial role in the cell wall biosynthesis; therefore, inhibition of GS in M. tuberculosis with the inhibitor L-methionine-SRsulfoximine increases susceptibility to isoniazid. Again, this study concludes that inhibition of GS activity affects the synthesis of peptidoglycan layers of the cell wall, changing susceptibility (Harth and Horwitz, 2003).

In this work, we describe that glnA mutants of S. Typhi increase their susceptibility to quinolones. Interestingly OmpF, a porin forming a homotrimer channel for the influx of CIP and nalidixic acid, was augmented in the S. Typhi glnA mutant. Further studies using a tetracycline-inducible system, revealed an inverse correlation between glnA and ompF expression. Our findings suggested that glnA inactivation increases the susceptibility of S. Typhi to CIP through increasing OmpF.

## MATERIALS AND METHODS

## Bacterial Strains and Growth Conditions

Salmonella Typhi SHT2370 an antibiotic-sensitive virulent clinical strain was obtained from the Infectious Diseases Hospital Lucio Cordova, Santiago, Chile (Valenzuela et al., 2014). Mutants used in this study were derived from S. Typhi STH2370 and are shown in **Table 1**. Bacteria were grown in LB medium (1% peptone, 0.5% yeast extract, 0.5% NaCl) at 37◦C with aeration. When required, the medium was supplemented with Kanamycin (KAN) 50 mg/L, Ampicillin (AMP) 100 mg/L, Chloramphenicol (CAM), CIP (at different concentrations) or Chlortetracycline (CTET at different concentrations). Media were solidified with agar (15 g/L) as required. Mutants in genes glnL and glnG were produced by P22 transducing null mutations with the KAN<sup>R</sup> cassette from S. Typhimurium.

## Transposon Insertional Mutagenesis and Screening

Mutagenesis with the commercial transposon EZ-Tn5 TM (Epicentre, CA, United States) was performed as indicated by the provider (protocols available at https://www.lucigen.com/docs/manuals/MA155E-EZ-Tn5-r6kg -Ori-KAN-2-Transposome-Kit.pdf at February 17, 2020) with minor modifications. In brief, S. Typhi STH2370 was grown to OD<sup>600</sup> 0.4, 5 mL were washed six times with sterile water, resuspended in 100 µL of sterile water and electroporated with



1 µL of EZ-Tn5 transposome at a concentration of 0.1 pmol/µL. Electroporated bacteria were recovered in 1 mL of LB broth, and incubated with aeration for 40 min before plating on LB-agar containing KAN 50 mg/L and overnight incubation. Mutants S. Typhi mutants STH2370 zxx::EZ-Tn5 were organized on a 6 × 8 pattern in 67 lots creating a collection of 3,216 mutants. The primary screening was performed by transferring colonies to LB plates with CIP 0.05 mg/L using a 48-pins microplate replicator. Mutants with increased susceptibility were selected by visual examination and confirmed by disk diffusion assay. The susceptibility of confirmed mutants was quantified by MICs assays before proceeding with cloning and sequencing of interrupted genes.

## Cloning and Sequencing

The mutants with at least two-fold-change in the MIC were subjected to cloning. Briefly, genomic DNA of mutants was extracted, digested with EcoRV and ligated with T4 DNA ligase. Ligated DNA was electroporated in E. coli DH5αλpir, recovered in 1 mL of LB broth, and incubated with aeration for 40 min at 37◦C before plating on LB-agar containing KAN 50 mg/L. KAN<sup>R</sup> colonies were grown, and plasmidial DNA was columnpurified to perform Sanger sequencing. Sequences were analyzed using the Basic Local Alignment Search Tool available at http: //www.ncbi.nlm.nih.gov.

## Disk Diffusion Test

The antibiotic susceptibility tests were performed by Kirby-Bauer disk diffusion method according to CLSI guidelines (CLSI, 2012a). Suspensions of 2 mL of microorganism were grown overnight in liquid culture using LB medium with aeration. The bacterial suspensions were diluted 1:1000 in 0.9% NaCl and 100 µL of this suspension were seeded on Mueller Hinton (MH) agar plates pH 7.3. CIP susceptibilities tests were performed with 5 µg CIP disks onto seeded plates incubated at 37◦C for 16– 18 h. Halos were recorded as the average of 2-crossed diameters in biological triplicate assays.

## Determination of MIC

MIC was determined using broth microdilution susceptibility testing according to CLSI guidelines (CLSI, 2012b). Briefly, 0.9% NaCl was used to prepare inoculum until turbidity McFarland 0.5, with an equivalent density to 1–2 × 10<sup>8</sup> CFU/mL. The inoculum was then diluted to reach a concentration of 5 × 10<sup>4</sup> CFU/well. Antibiotics were serially diluted in base-2 and each well-inoculated with 50 µL of diluted bacteria. The plates were incubated at 37◦C for 16 h and read at 600 nm. The MIC was determined as the concentration that allowed less than 50% growth density compared to the non-antibiotic control.

## Construction of S. Typhi Site-Directed Mutants

Site-directed mutants of S. Typhi STH2370 were constructed by allelic exchange facilitated by the Red/Swap Technique (Datsenko and Wanner, 2000). Briefly, PCR product, with a resistance cassette flanked by regions with homology to the target DNA sequences, was electroporated into strains carrying pKD46, recovered in 1 mL of LB broth and incubated for 1 h at 37◦C with aeration before plating onto plates with the corresponding antibiotic to select mutants. When needed the resistance cassette was eliminated to produce FRT scar mutants by introducing pCP20 which encodes the Flippase recombinase that promotes scission from FRT sequences flanking the resistance cassette. The primers used for these constructions are in **Supplementary Table 1**.

## Construction of Tetracycline-Dependent Mutants

Construction of the tetracycline-dependent (TD) mutants S. Typhi glnATD, S. Typhi ompFTD and S. Typhi STY4173TD, was carried out using a modification of the Red-Swap method for promoter mutation of target genes as described by Hidalgo et al. (2016). The primers used for these constructions are in **Supplementary Table 1**.

## Preparation of Outer Membrane Proteins (OMPs) and SDS-PAGE

The OMPs fraction of protein was obtained by a method previously described (Lobos and Mora, 1991). Two mL of bacteria were centrifuged at 10,000 g for 10 min. The pellet obtained was resuspended in 10 mM Tris Hydrochloride pH 8, sonicated for 100 s and centrifuged at 7,000 g for 5 min. The supernatant obtained was centrifuged at 13,000 g for 45 min, and the pellet was resuspended in 10 mM Tris Hydrochloride pH 8 with 2% Triton X-100. The suspension was incubated at 37◦C for 30 min with occasional stirring and centrifuged at 13,000 g for 45 min. The pellet containing OMPs was resuspended in 100 mM Tris

Hydrochloride pH 8, 2% SDS and quantified using BCA protein assay kit. Next, 50 µg of protein were heated at 98◦C for 5 min and resolved in an SDS-PAGE at 100 V. The gels were rinsed with water, pre-treated for 60 min with 40% ethanol and 10% acetic acid, washed twice for 10 min and stained in 0.1% acid Coomassie blue R250, 2% orthophosphoric acid, 10% ammonium sulfate, and 20% methanol overnight. Finally, gels were washed with 1% acetic until all Coomassie particles were removed. High resolution pictures of gels were open in the ImageJ software. Using the area selection tool, a horizontal line was traced to cut pick that were then analyzed by their density, which was compared to density of WT strain for the bands corresponding to OmpF (Gallo-Oller et al., 2018).

## RNA Isolation, Reverse Transcription, and Real-Time PCR (RT-qPCR)

For extraction of RNA, overnight cultures were diluted 100 times and cultured in fresh LB or LB + CTET as appropriate, until reaching OD<sup>600</sup> 0.5. RNA was extracted using the Trizol method according to the provider instructions. Briefly, 20 mL of bacterial cell was harvested, washed twice with ice-cold water and treated with lysozyme at RT before adding TrizolTM (Invitrogen). The samples were mixed before adding ice-cold chloroform, centrifugation and recovering the supernatant. RNA was precipitated with isopropanol, washed with 70% v/v ethanol and resuspended in nuclease-free water before treatment with DNase I. Reverse transcriptions of samples were performed with 1 µg of DNase-treated RNA using SuperScriptTM II Reverse Transcriptase (Invitrogen) with Random Hexamers (Invitrogen). Primers for the genes dnaN and lon were used to normalize the expression of target genes. Although dnaN and lon are both good expression controls, dnaN was preferred for assays that used tetracyclines to minimize distortion of results as dnaN remained unchanged compared to lon, in presence of CTet. Quantification was carried out in an EcoTM Real-Time PCR System using Brilliant II SYBRTM Master Mixes. The fold change was calculated as described by Pfaffl (Pfaffl, 2001). The primers used for these assays are listed in **Supplementary Table 1**.

## Statistical Analysis

Student's t-test was realized if samples obtained of different experiments have passed Shapiro–Wilk test for normality. The data are presented as mean ± SD of at least biological triplicates. Statistical analyses of results from RT-qPCR were calculated using the paired Student's t-test (GraphPad Prism version 8.0, San Diego, CA, United States).

## RESULTS

## Screening of S. Typhi zxx::EZ-Tn5 to Identify Mutants With Increased Susceptibility to Ciprofloxacin

We screened 3,216 **S**. Typhi STH2370 **zxx**::EZ-Tn**5** mutants on MH-agar plates with CIP 0.05 mg/L. A total of 43 susceptible mutants with poor or not growth were selected by visual analysis of the colonies. Confirmation of CIP susceptible mutants was performed by using disk diffusion assay. From this test, 10 mutants were confirmed, as their inhibition halos were 6 mm greater compared with WT control and were subjected to MIC assays. Eight mutants were twice as sensitive, one mutant was four-times more sensitive, and one mutant was eight-times more sensitive to CIP, compared with the WT counterpart. These mutants were set for cloning and further sequencing of the interrupted genes. All selected mutants that were sent to sequencing had one transposon insertion only. An extensive search was performed for information linking the mutated genes to antibiotics resistance. Five mutants, recC::EZ-Tn5, cysE::EZ-Tn5, aefA::EZ-Tn5, dacC::EZ-Tn5 and glnA::EZ-Tn5, resulted of especial interest after considering two criteria: (i) the level of susceptibility of the mutant and (ii) the relevance/novelty of the interrupted gene. **Supplementary Figure 1** shows the screening workflow to identify genes leading to susceptibility. Subsequently, a secondary screening was performed, over the five mutants of interest, to determine whether observed susceptibility was general or specific for one antimicrobial or a family of them. Diameters of inhibition haloes for the five insertional mutants selected with different antimicrobials were recorded (**Supplementary Table 2**). After discarding the best candidate (S. Typhi aefA::EZ-Tn5) whose increased susceptibility was product of polar effects produced by the transposon over the acrAB locus upstream of aefAgene, we began the characterization of mutantS. Typhi STH2370 glnA::EZ-Tn5 that showed increased susceptibility to NAL and CIP in the screening (**Supplementary Table 2**). While the other three mutants were not further characterized.

TABLE 2 | Susceptibility to ciprofloxacin of WT S. Typhi and null mutants in glnA.


TABLE 3 | Induction of glnA in glnATD mutant decreases the susceptibility to ciprofloxacin (CIP).


## S. Typhi glnA Mutants Are More Susceptible to Ciprofloxacin

To rule out polar effects that might modulate antimicrobial susceptibility, in the S. Typhi glnA::EZ-Tn5, we constructed mutant S. Typhi STH2370 glnA::kan and S. Typhi STH2370 glnA::FRT (**Supplementary Figure 2**). All mutants in glnA gene (glnA::EZ-Tn5, glnA::kan, and glnA::FRT) presented the same range of increased susceptibility to CIP with a diameter of inhibition haloes ∼11 mm larger and a MIC twice as sensitive compared to the WT strain (**Figure 1** and **Table 2**). The observation that S. Typhi glnA::FRT (as well as all null mutants in glnA gene) shows the same level of susceptibility, rule out any possible polar effect governing the observed phenotype, as FRT sequence is by far less invasive that resistance cassettes.

## Inactivation of glnA Gene Increases the Expression of OmpF

Because inactivating glnA gene increases susceptibility to CIP, we searched for information regarding determinants of susceptibility to this antibiotic. As outer membrane proteins may be implicated in antibacterial entry through the cell envelope (Nikaido, 2003), we decided to study the profiles of OMPs in the glnA::FRT. SDS-PAGE profile of OMPs, showed a threefold increase of OmpF in S. Typhi STH2370 glnA::FRT, compared with WT (**Figures 2A,B**). Same result was observed for all glnA null mutants in S. Typhi, the results for S. Typhi glnA::FRT and S. Typhi glnA::kan are shown in **Supplementary Figure 3**, compared to S. Typhi WT and a S. Typhi ompF mutant. Interestingly, deletion of ompF produces changes in the OMPs pattern, at the level of making OMPs unrecognizable by simple Coomassie staining. Previous observation reported changes on the amount of OMPs in an ompF null mutant of a clinical isolate of S. Typhi. However, in this mutant main porins, such as ompC and ompA did not change their migration pattern (Villarreal et al., 2014). It could be the case that the extend of changes produced after ompF is mutated differs between isolates of S. Typhi. Despite that, abundance of OmpF would be the responsible of increased susceptibility to CIP, as ompF is also augmented in the ompF inducible mutant, but not other porins, as shown in further experiments. Next, RTqPCR was performed to measure ompF expression in glnA::FRT mutant. As shown in **Figure 2C**, the expression of ompF gene was at least four-times higher in the glnA::FRT mutant compared with WT strain.

## Induction of glnA Inversely Correlates With Susceptibility to Ciprofloxacin

To confirm how changes in glnA expression cause changes in ompF expression and susceptibility to CIP, we modulated the expression of glnA by using a tetracycline-inducible system as previously described (Hidalgo et al., 2016). For that purpose, we successfully replaced the glnA gene promoter with

the tetRA cassette (**Supplementary Figure 4**). This mutant allows modulation of glnA expression by using tetracyclines, to study ompF expression and susceptibility to CIP as a function of glnA expression. Because both OmpF and TetA (the efflux pump for tetracyclines) participate in transport through membranes, we excised tetA gene, from the tetRA cassette, to reduce interferences, caused by the presence of TetA in the inner membrane. This last strain was named as the tetracycline-dependent glnA mutant (or simply glnATD), as shown in **Supplementary Figure 4**. Using the same strategy, we constructed the tetracycline-dependent ompF mutant (ompFTD), and the same inducible construct was produced for the unrelated gene STY4173 (4173TD) that encodes for a putative protein, as a control. Details regarding these mutants are shown in **Supplementary Figure 5**. Using the S. Typhi glnATD mutant, we observed that inhibition haloes in the presence of CIP decreased, as the concentration of CTET increases in the plates, to values comparable to the inhibition diameters observed for the WT strain (from 46.0 mm without induction to 38.1 mm with 0.05 µg/mL of CTET). In addition, in uninduced conditions, the glnATD mutant shows an inhibition diameter closer to glnA::FRT null mutant, 46.0 mm and 48.0 mm, respectively (**Table 3**).

In contrast, using the ompFTD mutant, the inhibition haloes in the presence of CIP increased as the concentration of CTET increases in the plates, to values close to the inhibition diameters observed for the WT strain (from 31.9 mm without induction to 41.0 mm in the plates with 0.05 µg/mL of CTET). Induction of STY4173TD expression did not cause significant changes in the inhibition halo to CIP (**Table 3**). The treatment with CTET did not produce significant changes in susceptibility of either WT or glnA::FRT strain to CIP.

## Expression of glnA Inversely Correlates With ompF Expression

In the view of the decreased susceptibility to CIP when glnA is induced with CTET, we decided to assess the expression of glnA and ompF transcripts in the tetracycline-dependent mutants by RT-qPCR. Expression of glnA was increased as CTET concentrations increased (**Figure 3A**). Next, we investigated the expression of ompF in the glnATD with and without

TABLE 4 | Susceptibility to ciprofloxacin of S. Typhi null mutants in genes glnL, glnG, micF, and ompF.


CTET induction. We found that in the glnATD mutant without induction, ompF is overexpressed close to five-times compared with WT strain (**Figure 3B**), the expression of ompF in the glnATD mutant without induction, is similar to the expression of ompF observed in S. Typhi glnA::FRT null mutant (**Figure 2C**). When glnA expression was induced with CTET we observed a significant decrease in ompF expression. Using the ompFTD mutant, we found that without induction, expression of ompF is dramatically reduced compared to WT strain. Induction of ompFTD mutant with 0.05 µg/mL of CTET significantly induced expression of ompF over 150-times compared to uninduced condition and over 1.5-times compared to the WT strain. Higher concentration (0.1 µg/mL) induced ompF only 50 times compared to uninduced condition and produces only half the expression of ompF observed in WT strain (**Figure 3C**). As a control, we measured the induction of ompF after inducing the unrelated gene STY4173 (**Figure 3D**). The presence of CTET induced expression of ompF two to threefold compared to uninduced condition and was slightly higher compared to WT strain in the presence of 0.1 µg/mL CTET. The results indicate that expression of glnA inversely correlates with expression of ompF, while increased expression of ompF may be responsible of increased susceptibility to CIP.

## Inactivation of glnA Reduces Expression of MicF

MicF is a small RNA that promotes degradation of ompF transcript and blocks translation by interacting with the Shine-Dalgarno sequence (Morillon, 2018). Because of this direct effect of MicF over OmpF, expression of MicF sRNA was studied in the glnA mutant and in the glnATD mutant with different levels of glnA expression. As observed in the **Figure 4A**, the S. Typhi glnA::FRT null mutant expresses around one-fourth of MicF, compared to the S. Typhi WT. To corroborate this result, MicF expression was studied in the S. Typhi glnATD mutant. Increased expression of glnA with 0.05 and 0.1 µg/mL CTET (as shown in **Figure 3A**) parallel with increased expression of MicF (**Figure 4B**). It is also observed that in uninduced conditions, MicF levels are slightly reduced compared with WT strain, probably MicF expression is sensitive to low levels of GlnA that transcribes even in the absence of induction with CTET. Conversely, deletion of micF gene slightly increased susceptibility to CIP in S. Typhi (**Table 4**). MicF is a well-described sRNA that negatively regulates transcript levels of ompF. Whether the induction of MicF by GlnA is direct or through other factors is unknown.

## The Deletion of glnA Activates Transcription of NtrBC Two-Component System

In E. coli previous findings indicate that ompF is under the control of NtrBC two-component system encoded by glnL and glnG, as observed in microarray assays (Zimmer et al., 2000). In enterobacteria, these genes are found downstream of glnA and are involved in the stress response of enteric bacteria to nitrogen-limited growth (Bhagirath et al., 2019) (**Supplementary Figure 2**). The deletion of glnA increases the

expression of glnL and glnG, 12- and 8-times respectively, compared to S. Typhi WT (**Figure 5**), indicating that glnA negatively regulates expression of glnL and glnG through maintaining steady concentrations of glutamine. Mutating glnA reveals a positive regulation of NtrBC on the expression of ompF. This regulation pathway may underly the mechanism of increased susceptibility to CIP in the glnA mutants. To study the phenotype associated to glnL and glnG, null mutations of these genes were transduced into S. Typhi to test for resistance to CIP. As shown in **Table 4**, deletion of glnL gene has no effect on resistance to CIP. When glnG, which encodes for a transcriptional regulator, is deleted in S. Typhi, the mutant shows moderately, yet significantly, resistant to CIP. It is plausible that while over expression of GlnG correlates with susceptibility, deletion of glnG produces resistance. However, the correlation of activation of NtrBC two-component system [encoded by glnL (NtrB) and glnG (NtrC) genes] need to be further validated.

## DISCUSSION

The threat of extensively drug-resistant S. Typhi isolates requires the development of new treatment strategies. One approach is the search for antibiotics adjuvants-compounds (Murima et al., 2014; Baquero and Martínez, 2017). In a non-exhaustive screening of S. Typhi STH2370, we found the mutant S. Typhi glnA::EZ-Tn5, with half the MIC to CIP. This change reveals a low-level resistance, mediated by an intrinsic mechanism of protection in a microorganism that is in principle susceptible to CIP. It is thought that the genes determining such intrinsic lowlevels of resistance (2- to 4-times the MIC) might be of clinical importance and they could be target for adjuvant therapies (Baquero and Martínez, 2017). GS plays a crucial role in nitrogen metabolism whose main function is to provide glutamine to the cell. In Enterobacteria, the reaction catalyzed by GS is the only biosynthetic pathway for glutamine synthesis; thus, glnA deletion results in glutamine auxotrophy (Ikeda et al., 1996; Shimizu, 2013). The glnA gene has not been previously related with changes in susceptibility to quinolones, in Gramnegative bacteria. Expression of the major porin OmpF, a wellcharacterized channel for CIP and NAL entry, was increased across all glnA null mutants tested and the inducible glnA mutant without induction.

Interestingly, the mechanism behind increased susceptibility differs from affecting integrity of the cell wall, as antimicrobials targeting the cell wall have little or no effect on the glnA null mutant (**Supplementary Table 2**). Our results contrast with previous reports regarding GS and resistance to antimicrobials (Harth and Horwitz, 2003; Lima et al., 2013; El Khoury et al., 2017). It was found an inverse correlation between expression of glnA and ompF and a direct correlation between expression of glnA and micF. Decreased expression of micF in the glnA null mutant might explain at least in part the increased expression of ompF and increased susceptibility to CIP in the glnA mutants.

The growth under nitrogen-limiting conditions activates the Nitrogen regulation stress response, under the control of NtrBC two-component system encoded by glnL (NtrB) and glnG (NtrC) genes. This two-component system is composed of NtrB, the cognate histidine kinase that phosphorylates NtrC to its active form to regulate expression of genes including glnA, as previously described (van Heeswijk et al., 2013). In this study, inactivation of glnA resulted in increased expression of glnL and glnG genes.

Importantly, it was found that expression of ompF is controlled by NtrC in E. coli (Zimmer et al., 2000). Therefore, this increase in the expression of glnL and glnG may explain the increased OmpF and increased susceptibility when glnA expression is suppressed or reduced as we illustrate in **Figure 6**. Indeed, mutation of glnG (encoding for NtrC transcriptional regulator) produces moderate resistance to CIP. This observation is consistent with increased glnLG expression in the glnA mutants, which may trigger ompF expression, finally allowing entry of CIP through OmpF porin.

Other mechanisms, modulating antimicrobial resistance were studied including regulators MarA, MarR, and RamA; effluxsystem components AcrB and TolC and porin OmpC. We found no differences in the expression of the genes encoding for such antimicrobial's resistance factors (**Supplementary Figure 6**). Thus, enhanced susceptibility to quinolones observed in the glnA mutants could be linked to the increase of OmpF and associated with nutritional stress due to the decrease of glutamine and lower expression of MicF.

## CONCLUSION

The experimental results show that the inactivation of glnA gene drives to increased susceptibility to CIP in S. Typhi, through increasing outer membrane protein OmpF. The finding that genes implicated in nitrogen metabolism, such as glnA, modulate susceptibility to antimicrobials can be useful in search of potential targets to combat antimicrobial resistance.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## REFERENCES


## AUTHOR CONTRIBUTIONS

ARM and AH participated in the conception of ideas, performed the experiments, analyzed and interpreted the data, and participated in the drafting of manuscript. AYM performed experiments, collected and analyzed the data, and revised the draft. CS, NV, and GM collected and analyzed the data and participated in drafting of manuscript.

## FUNDING

This work was supported by the Fondo Nacional de Ciencia y Tecnología (FONDECYT, Government of Chile) Grants 11150588 (AH), 1151393 (GM), and 1160315 (CS); UNAB Regular Grants DI-15-19/RG (AH) and DI-4-17/RG (NV); and ECOS-CONICYT grant C16B04 (AH). ARM has been a predoctoral fellow at Comisión Nacional de Ciencia y Tecnología grant 21120035 (CONICYT, Government of Chile).

## ACKNOWLEDGMENTS

We thank Dr. Lionello Bossi, Dr. Nara Figueroa-Bossi, and Dr. Felipe Cabello for valuable comments on the manuscript. We thank Mr. Víctor Ahumada for technical assistant. Part of the results included in this manuscript were presented during the XXXVIII Meeting of the Chilean Society for Microbiology 2016, Valdivia, Chile.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00428/full#supplementary-material


protective response in mice. Biochem. Biophys. Res. Commun. 470, 313–318. doi: 10.1016/j.bbrc.2016.01.058


causes loss of resistance to the antimicrobial peptide protamine. J. Antimicrob. Chemother. 65, 1624–1630. doi: 10.1093/jac/dkq197


**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 Millanao, Mora, Saavedra, Villagra, Mora and Hidalgo. 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.

# Confronting Tigecycline-Resistant Acinetobacter baumannii via Immunization Against Conserved Resistance Determinants

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Maria Bagattini, University of Naples Federico II, Italy Hua Zhou, Zhejiang University, China Weiwei Huang, Chinese Academy of Medical Sciences and Peking Union Medical College, China

#### \*Correspondence:

Te-Li Chen tecklayyy@gmail.com †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 20 August 2019 Accepted: 12 March 2020 Published: 31 March 2020

#### Citation:

Chiang M-H, Yang Y-S, Sun J-R, Wang Y-C, Kuo S-C, Lee Y-T, Chuang Y-P and Chen T-L (2020) Confronting Tigecycline-Resistant Acinetobacter baumannii via Immunization Against Conserved Resistance Determinants. Front. Microbiol. 11:536. doi: 10.3389/fmicb.2020.00536 Ming-Hsien Chiang1,2† , Ya-Sung Yang<sup>3</sup>† , Jun-Ren Sun<sup>4</sup> , Yung-Chih Wang<sup>3</sup> , Shu-Chen Kuo<sup>5</sup> , Yi-Tzu Lee6,7, Yi-Ping Chuang<sup>8</sup> and Te-Li Chen2,9 \*

<sup>1</sup> Department and Graduate Institute of Biology and Anatomy, National Defense Medical Center, Taipei, Taiwan, <sup>2</sup> Graduate Institute of Life Sciences, National Defense Medical Center, Taipei, Taiwan, <sup>3</sup> Division of Infectious Diseases and Tropical Medicine, Department of Internal Medicine, Tri-Service General Hospital, National Defense Medical Center, Taipei, Taiwan, 4 Institute of Preventive Medicine, National Defense Medical Center, Taipei, Taiwan, <sup>5</sup> National Institute of Infectious Diseases and Vaccinology, National Health Research Institutes, Zhunan, Taiwan, <sup>6</sup> School of Medicine, National Yang-Ming University, Taipei, Taiwan, <sup>7</sup> Department of Emergency Medicine, Taipei Veterans General Hospital, Taipei, Taiwan, <sup>8</sup> Department and Graduate Institute of Microbiology and Immunology, National Defense Medical Center, Taipei, Taiwan, <sup>9</sup> Institute of Clinical Medicine, National Yang-Ming University, Taipei, Taiwan

Antimicrobial-resistant (AMR) bacterial infections, including those caused by Acinetobacter baumannii, have emerged as a clinical crisis worldwide. Immunization with AMR determinants has been suggested as a novel approach to combat AMR bacteria, but has not been validated. The present study targeted tigecycline (TGC) resistance determinants in A. baumannii to test the feasibility of this approach. Using bioinformatic tools, four candidates, AdeA, AdeI, AdeK, and TolC, belonging to the resistance-nodulation-division (RND) efflux pump were identified as highly conserved and exposed antigens from 15 A. baumannii genomes. Antisera generated from recombinant proteins showed the capability to reserve Hoechst 33342, a substrate of the efflux pump, in bacterial cells. The rTolC antisera had the highest complement-dependent killing and opsonophagocytosis effect compared to the sera from phosphate-buffered saline immunized mice. Among the antisera, anti-rAdeKspecific antisera decreased the minimal inhibitory concentration of TGC in 26.7% of the tested isolates. Immunization with rAdeK significantly potentiated TGC efficacy in treating TGC-resistant A. baumannii pneumonia in the murine model. The bacterial load (7.5 × 10<sup>5</sup> vs. 3.8 × 10<sup>7</sup> , p < 0.01) and neutrophil infiltration in the peri-bronchial vasculature region of immunized mice was significantly lower compared to the PBSimmunized mice when TGC was administrated concomitantly. Collectively, these results suggest that active immunization against resistance determinants might be a feasible approach to combat multidrug-resistant pathogens in high risk population.

Keywords: immunization, resistant determinant, Acinetobacter, tigecycline, efflux pump

## INTRODUCTION

fmicb-11-00536 March 27, 2020 Time: 19:19 # 2

Antimicrobial-resistant (AMR) bacterial infections have emerged as a serious problem in clinical settings worldwide. By 2050, 10 million people may die annually from AMR infections (Jansen et al., 2018). Unfortunately, the launch of new antibiotics is rapidly compromised by the emergence of resistance (Alekshun and Levy, 2007). Therefore, the development of new strategies to combat AMR bacteria is urgently needed.

Recently, vaccine development has been increasingly advocated as a new solution to AMR bacteria (McConnell et al., 2011; Huang et al., 2014; Jansen et al., 2018), especially for those with high risk of acquired infection (Gagneux-Brunon et al., 2018). Vaccine candidates often target capsule polysaccharides and virulent factors that are responsible for disease pathogenesis (Rappuoli et al., 2019). Unfortunately, AMR bacteria-targeting vaccines have not been very successful until now due to the heterogeneity in the expression levels of vaccine antigens in different circulating strains of the target pathogen (Garcia-Quintanilla et al., 2016; Rappuoli et al., 2019). Moreover, the immunity induced by single or multiple antigens, which is usually obtained both in vitro or in vivo, might be ineffective in protecting the host because the pathobiology in human infection is more complicated and remains obscure (Perez and Bonomo, 2014).

Antibiotic resistance determinants could be considered as potential vaccine candidates (Ni et al., 2017). Although there are variations in these resistance determinants in clinical isolates, this vaccination strategy might still be effective in eradicating resistant strains by means of replacing these strains which have reduced fitness with susceptible strains when vaccination coverage is high (Joice and Lipsitch, 2013; Niewiadomska et al., 2019). This approach has an advantage when the resistance determinants are consistently present in vitro and in vivo and also possess less selection pressure on bacteria (Niewiadomska et al., 2019). However, this approach has not been validated yet.

The present study tested this idea in Acinetobacter baumannii. A. baumannii is considered one of the most problematic bacteria by the Infectious Disease Society of America (IDSA) (Boucher et al., 2009) because of the rapid evolution of multidrug and pan-drug resistant strains. Currently, only few strains of this bacteria are still susceptible to last-line antibiotics such as, carbapenem and tigecycline (TGC) (Sun et al., 2013; Ni et al., 2016). In order to maximize the coverage of immunization, the antibiotic resistance determinants used for the vaccine candidate should be the major resistance mechanism of the antibiotic of interest in particular bacterial species. These determinants should be universal in different strains, conserved in sequence homology, and accessible by the immune system. In this aspect, carbapenem resistance determinants are not suitable vaccine candidates, as its major resistance mechanism is production of different classes of carbapenemases (Nordmann and Poirel, 2019), which are very diverse in their protein sequences and structures. On the contrary, the major mechanism for TGC resistance is overexpression of efflux pumps (Sugawara and Nikaido, 2014). These are universal and are conserved in A. baumannii strains (Ardehali et al., 2019), and therefore, might be good vaccine candidates to test this immunization approach. The present study utilized bioinformatics tools to identify conserved and surface-exposed antigens of the chromosomalencoded resistome of A. baumannii. Several protein components of the resistance-nodulation-division (RND) efflux system were identified, which have been associated with TGC resistance in A. baumannii. We propose an immunization approach using TGC resistance determinants in a murine pneumonia model to combat multidrug-resistant A. baumannii.

## MATERIALS AND METHODS

## Bioinformatic Tools

Microbial Genome Database (MBGD) was used for comparative analysis of completely sequenced microbial genomes to identify core genes that are universal from 15 A. baumannii genomes (**Supplementary Table S1**; Uchiyama et al., 2014). PSORTb 3.0.2 (Yu et al., 2010), CELLO2GO (Yu et al., 2014), or SOSUI-GramN (Imai et al., 2008) were applied to predict the conserved residues and sub-cellular localization of these proteins. Comprehensive Antibiotic Resistance Database (CARD) was used to predict the resistome from raw genome sequence using Resistance Gene Identifier (RGI) software (Jia et al., 2017).

## Bacterial Strain Preparation

A. baumannii ATCC17978 reference strain was purchased from the American Type Culture Collection (ATCC). TGC-resistant clinical A. baumannii isolates were obtained from Tri-Service General Hospital in Taiwan (Sun et al., 2014). All isolates were identified using conventional biochemical and genomic methods as previously described (Sun et al., 2014).

## Construction and Purification of Antigens

Recombinant AdeA (A1S\_1751), AdeI (A1S\_2735), AdeK (A1S\_2737), and TolC (A1S\_0255) from ATCC17978 were amplified (the primers are listed in **Supplementary Table S2**) and cloned into a pET-29a expression vector (Vovagen, Darmstadt, Germany) with a 6× polyhistidine tag fused to the C-terminus of the recombinant protein. The resulting plasmids were expressed and purified as described in our previous study (Chiang et al., 2015). Purified proteins were digested with trypsin for subsequent liquid chromatography–mass spectrometry/mass spectrometry (LC-MS/MS) analysis and protein identification was conducted by Mission Biotech Co., Ltd., Taiwan. Sequence similarity comparison of adeA, adeI, adeK, and tolC from ATCC17978 and all isolates were analyzed using MEGA7 (Kumar et al., 2016).

## Mouse Immunogenicity Assessment and Pneumonia Models

All animal studies were approved by the National Defense Medical Center Institutional Animal Care and Use Committee (NDMC IACUC-17-206). Female C57BL/6 mice (6 weeks old) were bred in a barrier facility under specific pathogen-free conditions. C57BL/6 mice (n = 10/group) were subcutaneously

(sc.) immunized with 10 µg of individual recombinant antigens formulated with Complete Freund's Adjuvant/Incomplete Freund's Adjuvant (CFA/IFA) (Invivogen, Hong Kong), on days 0, 14, and 28. Blood samples were collected before the last immunization and tested against each immunogen. Immunoglobulin G (IgG) antibody titers were determined using antigen-specific enzyme-linked immunosorbent assays (ELISAs).

For conducting efficacy studies, immunized mice were challenged intra-tracheally (IT) on day 42 with a lethal dose [3 × 10<sup>7</sup> colony-forming units (CFUs)] of mid-log phase AB247 strain mixed with 10% porcine mucin (Sigma-Aldrich, MO, United States). The use of porcine mucin is to enhance the infectivity of A. baumannii (McConnell et al., 2011). TGC (10 mg/kg/d, q12h., sc.) treatment regimen was adopted from that used in a previous study (Pichardo et al., 2010). After 24 h therapy, the blood, lung, spleen, and kidney were homogenized and plated to evaluate for the CFUs. For histological analysis, the excised lungs were placed in vials containing 4% formaldehyde. The lungs were placed under vacuum overnight, paraffinembedded, and stained with hematoxylin and eosin (HE). Histological scores were assigned by independent pathologists by evaluating 3–5 fields, according to the following criteria (Noto et al., 2017): 0, no pathology; 1, minimal infiltrates of neutrophils in alveolar spaces; 2, low numbers of neutrophils in alveoli; 3, moderate numbers of neutrophils and hemorrhage in alveoli with occasional lobar involvement and focal necrosis of alveolar-wall neutrophils in bronchioles; 4, marked numbers of neutrophils, consolidation, and widespread alveolar necrosis.

## Flow Cytometry

ATCC17978 and clinical isolates from late-log-phase growth (OD<sup>600</sup> ≈ 1.8) in Luria-Bertani (LB) cultures were diluted in PBS containing 0.5% (w/v) bovine serum albumin (BSA) as a blocking buffer to an OD<sup>600</sup> of 0.03. Each specific antiserum was added at a 1:100 dilution with the bacterium. Unbound antibody was removed by washing twice, then fixed by incubating with 4% formaldehyde/PBS for 10 min on ice. A secondary antibody, goat anti-mouse IgG-PE (Invitrogen Corp., Carlsbad, CA, United States) at 0.1 mL per well, was added at a 1:100 dilution and incubated for 30 min. The bacteria were analyzed using a FACSCalibur flow cytometer (BD, Franklin Lakes, NJ, United States). Wash buffer, secondary antibody, or PBS Immunize serum were used as negative controls.

## Hoechst 33342 (H33342) Accumulation Assay

H33342 accumulation assays were carried out as described by Richmond et al. (2013). Strains were grown to an OD<sup>600</sup> of 0.4 and resuspended in PBS at room temperature, and the suspension was adjusted to an OD<sup>600</sup> of 0.1. Centrifugation steps were carried out at 2500 g. The wells of a black microtiter plate (Corning, Amsterdam, Netherlands) were inoculated with 176 µL of bacterial suspension and 20 µL of 10 µg/mL H33342. After 5 min equilibration, 4 µL of the efflux pump inhibitor, phenylalaninearginine-β-naphthylamide (PAβN), or specific antisera were added. The fluorescence intensity was recorded every 1 min for 60 min on a SPECTRAmax5 fluorometer (Molecular Devices, Sunnyvale, CA, United States) at excitation and emission wavelengths of 355 and 460 nm, respectively.

## Complement and Opsonophagocytosis Bactericidal Assays

ATCC17978 was freshly grown to a final bacterial cell concentration of 10<sup>6</sup> CFU/mL, and aliquoted into 96 well microtiter plates (10 µL, 10<sup>4</sup> cells/well). For complementary studies, 10 µL heat-inactivated or immune sera were mixed with 80 µL of undiluted human complement and added to the wells for 1 h at 37◦C. The samples were plated for bacterial enumeration. Bacteriolysis activity was defined as [1 − (CFU immune sera at 60 min/CFU of PBS-immunized antisera at 60 min)] × 100.

For the opsonophagocytic kill assay, RAW 264.7 macrophages were cultured in RPMI 1640 (Irvine Scientific, Santa Ana, CA, United States) with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin, and glutamine (Gemini BioProducts), and 50 µM β-mercaptoethanol (Sigma-Aldrich, St. Louis, MO, United States). RAW 264.7 cells were stimulated with 100 nM PMA (Sigma-Aldrich) for 3 days. RAW 264.7 macrophages (2 × 10<sup>5</sup> /well) and ATCC17978 (1 × 104CFU/well) were added into the wells along with the heat-inactivated or immune sera (5%). After a 1 h incubation with gentle shaking, the samples were serially diluted and plated. Serum killing rates were counted by comparing the number of reduced CFUs with those observed using PBS-immunized antisera.

## RNA Isolation and Quantitative Reverse Transcriptase – PCR

The expression level of adeA, adeI, adeK, and tolC were measured by quantitative real-time polymerase chain reaction (PCR) assays as described previously (Rosenfeld et al., 2012). The mRNA expression of rpoB from ATCC17978 was used for normalization with specific primers (**Supplementary Table S2**). The relative gene expression was expressed as fold-change calculated by the 11Ct method. Gene expression levels ≥ 2-fold compared to that for the reference strain, ATCC17978, were considered significant overexpression. Each experiment was performed in duplicates and at least twice independently.

## Minimal Inhibitory Concentration (MIC) Determination by Broth Microdilution Method

The MICs of TGC were determined by broth microdilution methods in Mueller Hilton broth and interpreted according to the Clinical and Laboratory Standards Institute guidelines (CLSI, 2017). Since MIC breakpoints are not established for TGC in Acinetobacter spp., we used the Food and Drug Administration breakpoints set for Enterobacteriaceae (Pillar et al., 2008). Each experiment was performed in triplicate.

## Statistical Analysis

Statistical analyses were performed using GraphPad Prism 7 software. All graphical values were represented as means ± standard error of the mean (SEM). Tests of statistical

FIGURE 1 | A rational strategy for potential antigen identification and verification from A. baumannii genomes. (A) The phylogenetic lineage of 15 completely sequenced A. baumannii strains analyzed in this study was constructed by the neighbor-joining method performed in MEGA7. Homology core genes were analyzed using the web-based MBGD tool. (B) Bioinformatics tools including PSORTb 3.0.2, CELLO2GO, and SOSUI-GramN were utilized to identify 462 non-redundant outer membrane or extracellular proteins from 2728 core genes. Four antimicrobial-resistant genes associated with efflux pumps were identified in the 462-protein datasets based on Comprehensive Antibiotic Resistance Database (CARD) analysis. OMP, outer membrane protein; EC, extracellular protein. (C) Purified proteins were analyzed by 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Lane M-standard protein markers, Lane 1-purified rAdeI protein, Lane 2-purified rTolC protein, Lane 3-purified rAdeK, Lane 4-purified rAdeA protein. (D) Western blot analysis of purified recombinant proteins using mouse anti-His antibody. The arrow indicates dimerization of rAdeA (confirmed by LC-MS/MS).

significance were performed using one-way analysis of variance and Kruskal–Wallis tests with post hoc analysis. Differences were considered significant for p < 0.05.

## RESULTS

## In silico Screening of Conserved Outer Membrane Efflux Pump Proteins and Generation of Recombinant Proteins

Fifteen A. baumannii genomes included 12 multidrug resistant strains were used for analysis. The antibiotic susceptibility profiles of these isolates are listed in **Supplementary Table S1**. These strains were grouped by multilocus sequence typing (MLST) (Laure et al., 2010) and three strains belonged to international clone I (IC I), eight to IC II, three to IC III, and the last strain was unclassified, indicating a wide coverage of A. baumannii strains (**Figure 1A**). A total of 2728 core genes were identified; among them, 462 non-redundant proteins were predicted to be outer membrane or extracellular proteins. CARD predicted that among the 462 conserved and surface-exposed proteins, four were associated with antibiotic resistance, including AdeA (A1S\_1751), AdeI (A1S\_2735), AdeK (A1S\_2737), and TolC (A1S\_0255) and were selected as vaccine candidates (**Figure 1B**). All the proteins belonged to the chromosomally encoded RND efflux pump family and had been associated with resistance to multiple antibiotics, including TGC (Sun et al., 2013). The four recombinant proteins were expressed and purified (**Figure 1C**). The proteins were identified and confirmed by immunoblotting (**Figure 1D**) and LC-MS/MS analysis (**Table 1**).

## Polyclonal Antibody Production and Functional Analysis

We then produced antigen-specific polyclonal antibodies from C57BL/6 mice using immunization (**Figure 2A**). All recombinant antigens (rTolC, rAdeK, rAdeI, rAdeA) formulated with CFA/IFA induced strong antigen-specific IgG antibody responses (IgG titers > 10<sup>5</sup> , **Figure 2B**) on day 42 after immunization. The results indicated that all four antigens are highly immunogenic. Antigen-specific antisera were used to verify the location of these proteins in the outer surface of bacteria by flow cytometry (**Figure 2C**). Data confirmed that all the antisera could bind on the surface of ATCC17978, with significantly higher intensities than that for the PBS control antisera (**Figure 2D**). In vitro complement-dependent and opsonophagocytosis bactericidal assays were used to assess the potential bacteria-killing activity of each antiserum. The results showed that 2–79% of ATCC17978 were inhibited by these antisera (**Figures 2E,F**). Among them, rTolC-specific antisera had the highest killing efficacy compared to the sera from PBS-immunized mice.

The accumulation of bis-benzamide H33342 dye provides a reliable method to evaluate the effect of agents that can block efflux pumps (Richmond et al., 2013). The fold-change of fluorescence intensity dramatically increased after adding PAβN and specific antisera, except for the PBS antisera control (**Figure 3A**, p < 0.001). Notably, fluorescence intensities were slightly decreased after 40 min of antisera treatment.

To examine the synergistic effect of antisera in the MICs range for TGCs in A. baumannii, 15 clinical TGC-R A. baumannii isolates were tested (**Table 2**). The TGC MICs of all isolates were > 4 mg/L. The sequence similarities of adeA, adeI, adeK, and tolC were determined (**Table 3**) and the similarities were very high among the isolates (98.2–100%), compared to the ATCC17978 sequence. The quantitative expression levels of adeA, adeI, adeK, and tolC were also determined. In general, all strains showed more than one pump overexpression phenotype. TGC MICs were significantly reduced after a combination with rAdeK antisera was used (**Figure 3B**), and five of the tested isolates (AB099, AB247, AB294, AB304, AB347) revealed more than fourfold reduction in TGC MICs (**Table 2**), and all five strains had adeA overexpression, and similarities in adeA sequences was 100% in 4 of the isolates. The similarity in adeA sequence of AB099 was 99.8%, with only a single amino-acid substitution (N295T). However, no significant differences were found among the responses of strains to anti-rAdeK antisera regarding their sequence homology (**Table 3**), gene expression level (**Table 2**) and binding ability of antisera to bacterial cells (**Figure 3C**). Furthermore, we examined the synergistic effect of anti-AdeK serum and four antibiotics including amikacin, meropenem, colistin and ampicillin/sulbactam, remained active against different portions of A. baumannii isolates. We found that only two isolates demonstrated a twofold reduction in the ampicillin-sulbactam MIC value from 64/32 to 32/16 µg/mL in the presence of anti-AdeK (**Supplementary Table S3**).

## TGC Activity Potentiation by rAdeK Immunization in a Mouse Pneumonia Model

According to the in vitro data, rAdeK antisera showed the highest capability to potentiate the effect of TGC against TGC-R A. baumannii, and also had an addition role to in

TABLE 1 | Confirmation of the purified recombinant proteins by liquid chromatography–mass spectrometry/mass spectrometry (LC-MS/MS).


individual antigen (rAdeI, rTolC, rAdeK, and rAdeA) or phosphate-buffered saline (PBS) formulated with Complete Freund's Adjuvant/Incomplete Freund's Adjuvant (CFA/IFA) on days 0, 14, and 28. Blood samples were collected before sacrifice and tested against each antigen. (B) Serum IgG antibody titers against each antigen as indicated by enzyme-linked immunosorbent assay. (C) Flow cytometry (FACS) analysis demonstrating the surface accessibility of the antigen-specific antisera. Cells from late-log-phase cultures were probed with a control non-specific antibody (shaded histograms) or individual antisera (unshaded histograms). Each histogram shows the fluorescence intensity distribution of >20,000 flow cytometry events. (D) Histogram representing quantitative analysis of the binding intensity from FACS studies. Bars indicate the means of at least three independent experiments ± SEM. \*\*\*p < 0.001. (E) Bacteria inhibition by each antigen-specific mouse antisera at 1:10 dilution, used in the presence of human complement and ATCC17978. (F) The complement-dependent opsonophagocytosis assays were performed with mouse antisera (1:10), ATCC17978 and RAW 264.7 macrophages. The results represent the percentage of bacterial survival in the assay after 1 h of incubation at 37◦C compared to that in phosphate-buffered saline (PBS)-Immunized antisera. Bars indicate the means of at least three independent experiments ± SEM. ns., non-significance. \*\*p < 0.01, Heat, complement replaced with a heat-inactivated complement.

tigecycline-resistant (TGC-R) A. baumannii clinical isolates (n = 15) compared to original or specific antisera [phosphate-buffered saline (PBS), rAdeA, rAdeI, rAdeK, and rTolC] co-administrated by broth microdilution method. The experiments were performed in triplicates. The p-value was determined by Kruskal–Wallis test with Dunn's multiple comparison analysis. \*\*p < 0.01. (C) Histogram representing quantitative analysis of the binding intensity between anti-AdeK antisera and isolates from the FACS study. The bars indicate the means of at least three independent experiments ± SEM. \*\*\*p < 0.001.

antibody mediated killing, so rAdeK was selected as a vaccine candidate. The protocol for murine pneumonia model is shown in **Figure 4A**. An intratracheal challenge with a clinical isolate of AB247 was performed. AB247 was selected because of its high TGC-resistant phenotype (MIC 16 mg/L) and its significant response in broth microdilution evaluation. The results showed that TGC administration in rAdeK-immunized mice significantly reduced the bacterial load in the lungs compared to that in mice without immunization (**Figure 4B**, p < 0.001). The bacterial load was also reduced in the kidney, spleen, and blood, but without statistical significance (**Figures 4C–E**). The histopathology of lung tissues showed that neutrophil infiltration in the peribronchial vasculature region was lower in rAdeK-immunization and TGC-treated mice (**Figure 5**) compared to that in other groups. More importantly, there was no evidence of morbidity or mortality among mice during the immunization course, which suggested that rAdeK is safe and suitable for immunization.

## DISCUSSION

AMR bacterial infection have been increasing dramatically in recent decades and account for about 80% of all severe bacterial infections (Du et al., 2018). Resistant phenotypes can arise from overexpression of intrinsic efflux activity to effectively respond to antibiotic or toxin related challenges (Du et al., 2018). Inhibition of antimicrobial determinants, such as efflux pump inhibitors (EPIs) is a feasible approach to preserve and improve the clinical performance of antituberculosis agents and overcome crucial drug-resistance challenges (Song and Wu, 2016). The use of EPIs could facilitate the revival of antibiotics and could be suitable for clinical applications. Unfortunately, no EPIs are yet available for clinical use (Abdali et al., 2017). We developed an alternative vaccine strategy targeting an antibiotic efflux pump. We tested this idea in TGC-resistant A. baumannii. We found that immunization with the RND efflux pump outer membrane protein, AdeK, interfered with the efflux activity of bacteria and, also induced antibodies with bactericidal effects (**Figure 6**). When co-administrated with TGC therapy, this strategy efficiently attenuated TGC-R A. baumannii pneumonia infection. A recent study showed that the combined treatment of anti-outer membrane vesicle serum and antibiotics could increase the intracellular aggregation of antibiotics by affecting porin function (Huang et al., 2019). The strategy significantly improved the antibiotic susceptibility of drug-resistant A. baumannii and strengthened the feasibility of combination therapy with vaccine and antibiotics. Accordingly, predominant resistance determinants with high homology, such as MecA, that confer resistance to methicillin in Staphylococcus aureus might also be good candidates for this approach.

Efflux-mediated TGC resistance has been extensively investigated in A. baumannii, especially in RND efflux pumps such as AdeABC, AdeFGH, AdeIJK, and AcrAB-TolC, in which overexpression leads to a TGC-resistant phenotype (Sugawara and Nikaido, 2014). Moreover, mutations in the adeR and adeS regulatory genes have been detected in TGC-R clinical isolates with efflux pump overexpression (Sun et al., 2014). Therefore,




<sup>a</sup>Bold font indicates a fourfold reduction in MIC value. Experiments were performed in triplicate and repeated three times, with similar results. <sup>b</sup>OE: overexpression, gene expression level ≥ twofold compared to that of the ATCC17978 reference strain was considered significant overexpression; +, Yes; −, No.


TABLE 3 | Comparisons of efflux pump protein sequence similarities to A. baumannii ATCC 17978.

inhibition of efflux activity is a promising approach to restore TGC susceptibility. RND efflux systems usually comprise three different components assembling into a functional complex, including outer membrane protein, middle periplasmic protein and inner membrane protein (Du et al., 2018). AdeA and AdeI belonged to the inner membrane protein of the system, whereas AdeK and TolC belonged to the outer membrane protein of the system. The results of kinetic accumulation study in the present study showed that all the antisera could potentially attenuate bisbenzimide H33342 efflux activity in ATCC strains, indicating H33342 is the substrate of these efflux systems. Decreased accumulation of H33342 after 40 min of antisera treatment (**Figure 3A**) might indicate the activation of the redundant efflux system to extrude H33342. Compared to other antisera, rAdeK antisera efficiently reduced TGC MIC levels in the clinical isolates; this result might indicate the major contribution of AdeIJK in TGC resistance in A. baumannii (Rosenfeld et al., 2012). It is interesting that when targeting the same RND system (AdeIJK), different components of the system could have different effects. For example, antisera against AdeK could

against A. baumannii AB247 [tigecycline (TGC) minimum inhibitory concentration of 16 µg/mL], twice a day by subcutaneous injections. Bacterial loads were determined in different organs after the mice were sacrificed and were compared among the four groups. (B–E) Bacterial load in the mice lungs (B), kidney (C), spleen (D), and blood (E) after 24 h TGC treatment compared among the four groups studied. The bars indicate the means ± the SEM. \*\*p < 0.01; \*\*\*p < 0.001.

potentiate TGC effects but antisera against AdeI could not. This result indicated that the outer membrane component might be a better vaccine candidate than inner membrane protein when combined with antibiotic use. As demonstrated in **Figure 3C**, antisera against rAdeK derived from ATCC17978 could crossreact with all fifteen clinical isolates by flow cytometry assay. The in vivo study also demonstrated a good response in AdeKvaccinated and TGC-treated mice, as they demonstrated less lung inflammation and reduced bacterial load in the lung. The bacteria load in other tissues was also lower in the AdeKvaccinated and TGC-treated mice, but the difference was not significant. This might be due to the lower bacterial load (about 10<sup>3</sup> CFU/g) in the no treatment or single treatment arms (rAdeK vaccinated or TGC-treated group), as A. baumannii is a bacterium with low virulence.

In addition to the potentiation of antibiotic treatment, the antisera had other roles as it could enhance the antibody mediated killing of bacteria. In the in vivo study, the effect of AdeK immunization might come from two aspects, one is antagonizing the efflux pump against antibiotic extrusion, another is through antibody mediated killing. It is unknown how the antisera of the efflux pump could reverse the antibiotic resistance. It is reported that antibodies may affect the function of specific antigens via conformation changes (Roguin and Retegui, 2003), or that this effect could be a result of blocking the channel of antibiotic extrusion. Some efflux pumps are associated with bacterial virulence and biofilm formation, which are responsible for host cell adhesion and invasion (Du et al., 2018). For example, TolC is a virulence factor associated with toxin translocation in E. coli (Lee et al., 2012). In our study, antisera derived from immunization with rTolC conferred significant complementdependent bactericidal and opsonophagocytic activity. It is worth to determine whether rTolC could be an ideal vaccine candidate in future studies.

The non-responding strains also had similar adeK sequence and overexpression level of the gene. Unknown mechanisms for TGC resistance may have been present in the non-responding strains. Li et al. (2016) identified > 50 possible drug efflux pumps that could contribute to multidrug resistance from over 1000 genomes of A. baumannii strains. In addition, the antibody might need to be optimized to more effectively block the most critical epitope of the efflux protein. Future research should focus on physical data in three-dimensional structures of efflux pumps and specific antibodies need to be elucidated to understand the structure-function relationships in these pumps (Roguin and Retegui, 2003). Although only 26.7% of strains had reduced TGC MIC after adding antisera, low coverage of the bacteria population (even 1–4%) by immunization with resistant determinants might still be effective in eradicating the resistant population (Niewiadomska et al., 2019).

We examined the effect of the anti-AdeK antisera on amikacin, meropenem, colistin, and ampicillin/sulbactam (**Supplementary Table S3**). These results showed no significant synergistic effect to reverse the resistance against these antibiotics. These was not unexpected, as the major resistant mechanisms of these antibiotics are not through efflux pump. Instead, the major resistance mechanism for amikacin, meropenem and ampicillin/sulbactam is the production of antibiotic modifying or degrading enzymes (Lee et al., 2017), whereas modification or loss of lipopolysaccharide confers colistin resistance in A. baumannii.

It is important to note that efflux pumps are conserved not only in A. baumannii but also in different bacterial genera. The AdeK protein has sequence homology to efflux proteins in other nosocomial "bad bugs" listed by the Infectious Diseases Society of America (Boucher et al., 2009; **Supplementary Table S4**), thus having the potential of broader coverage and application in the near future. A recent report also supported that AdeK and other 24 resistant determinants are predicted as vaccine candidates to strengthen antibiotic treatments (Ni et al., 2017).

## CONCLUSION

In conclusion, our results demonstrate that active immunization with antibiotic-resistant determinants may be a promising approach to combat multidrug-resistant pathogens in highrisk population.

## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## ETHICS STATEMENT

The animal study was reviewed and approved by National Defense Medical Center Institutional Animal Care and Use Committee (NDMC IACUC-17-206).

## AUTHOR CONTRIBUTIONS

M-HC, T-LC, and Y-PC contributed to the conception and design of the studies. M-HC, Y-SY, S-CK, and Y-PC contributed to the execution of the animal vaccination studies. M-HC, Y-TL, and J-RS carried out the laboratory in vitro assays. M-HC, Y-SY, Y-CW, S-CK, Y-TL, Y-PC, and T-LC were involved in the drafting, revision, and approval of the final version of the manuscript.

## FUNDING

This work was supported by the Ministry of Science and Technology (Grant Nos. MOST 105-2628-B-016-003-MY2, MOST 105-2314-B-016-039-MY3, MOST 107-2314-B-016-051- MY3, MOST 107-2320-B-016-003, MOST 108-2314-B-016-029, and MOST 109-2320-B-016-002-MY2), Tri-Service General Hospital (Grant Nos. TSGH-C107-098 and TSGH-C108-137). The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

## ACKNOWLEDGMENTS

This study was supported by the ACTION study group. The members of the ACTION study group include Yea-Yuan Chang (National Yang-Ming University Hospital, Yilan, Taiwan), Yuag-Meng Liu (Changhua Christian Hospital, Changhua, Taiwan), S-CK (National Institute of Infectious Diseases and Vaccinology, National Health Research Institute, Miaoli County, Taiwan), Chang-Pan Liu (MacKay Memorial Hospital, Taipei, Taiwan), T-LC (Graduate Institute of Life Sciences, National Defense Medical Center, Taipei, Taiwan), Y-TL (Taipei Veterans General Hospital, Taipei, Taiwan), and Y-SY (National Defense Medical Center, Taipei, Taiwan).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2020. 00536/full#supplementary-material

## REFERENCES

fmicb-11-00536 March 27, 2020 Time: 19:19 # 12


system of Escherichia coli. Antimicrob. Agents Chemother. 58, 7250–7257. doi: 10.1128/AAC.03728-14


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

# Prevalence, Characterization, and Drug Resistance of *Staphylococcus Aureus* in Feces From Pediatric Patients in Guangzhou, China

Xiaolan Ai 1†, Fei Gao1†, Shuwen Yao<sup>2</sup> , Bingshao Liang<sup>1</sup> , Jialiang Mai <sup>1</sup> , Zhile Xiong<sup>1</sup> , Xiantang Chen<sup>3</sup> , Zhuwei Liang<sup>1</sup> , Hongling Yang<sup>1</sup> , Zhiying Ou<sup>1</sup> , Sitang Gong<sup>1</sup> , Yan Long<sup>1</sup> \* and Zhenwen Zhou<sup>1</sup> \*

<sup>1</sup> Clinical Laboratory, Guangzhou Women and Children's Medical Center, Guangzhou Medical University, Guangzhou, China, <sup>2</sup> Clinical Laboratory, Guangzhou Children's Hospital, Guangzhou Medical University, Guangzhou, China, <sup>3</sup> Clinical Laboratory, Zengcheng Maternity and Children's Health Care Center, Guangzhou Medical University, Guangzhou, China

#### *Edited by:*

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### *Reviewed by:*

Rima Abdallah Moghnieh, Makassed General Hospital, Lebanon Honghu Chen, Zhejiang Center for Disease Control and Prevention, China

#### *\*Correspondence:*

Yan Long longyangmc@163.com Zhenwen Zhou zzw6248@126.com

†These authors have contributed equally to this work and share first authorship

#### *Specialty section:*

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Medicine

*Received:* 10 November 2019 *Accepted:* 23 March 2020 *Published:* 24 April 2020

#### *Citation:*

Ai X, Gao F, Yao S, Liang B, Mai J, Xiong Z, Chen X, Liang Z, Yang H, Ou Z, Gong S, Long Y and Zhou Z (2020) Prevalence, Characterization, and Drug Resistance of Staphylococcus Aureus in Feces From Pediatric Patients in Guangzhou, China. Front. Med. 7:127. doi: 10.3389/fmed.2020.00127 Background: Staphylococcus aureus (S. aureus) is a major pathogen of human infections. Its fecal carriage serves as a risk factor for nosocomial transmission and disease development. However, the rate of S. aureus fecal carriage among Chinese children has not yet been reported. Therefore, we sought to investigate the prevalence, characterization, and drug resistance of S. aureus isolated from pediatric patients' feces in Southern China.

Methods: Fecal samples (2059) from pediatric patients in three centers in Guangzhou were cultured. From which, 412 S. aureus isolates were identified via selective mediums and automated VITEK Mass Spectrometer analysis. Antibiotic susceptibility was determined and DNA sequencing of seven housekeeping genes were used for multilocus sequence typing analysis.

Results: The fecal carriage rates were 20.0% for S. aureus and 4.5% for methicillin-resistant S. aureus (MRSA). Moreover, S. aureus fecal carriage was positively correlated with outpatient status and gastroenteritis diagnosis. Moreover, age-related patterns were observed with respect to prevalence of S. aureus. Besides, a total of 76 sequence types (STs) were identified, including 25 newly assigned STs and 28 clonal complexes (CCs). ST188, ST6, and ST15 were the most prevalent methicillin-sensitive S. aureus (MSSA) clones, while ST59 and ST45 were the major MRSA clones. S. aureus isolates also exhibited high rates of penicillin (84.2%), erythromycin (38.8%), and clindamycin (35.9%) resistance. Specifically, all ST30 and ST338 isolates were resistant to erythromycin and clindamycin, 61% of ST7 were resistant to tetracycline, and 84% of ST45 exhibited resistance and intermediate resistance to rifampicin. Also, CC59 (ST338 and ST59) and CC45 exhibited different antibiotic resistance patterns.

Conclusion: These results demonstrate the colonization dynamics and molecular epidemiology of S. aureus in child feces in Southern China. Further, they suggest an urgency for strengthening the surveillance programs in China and provide important information for the prevention and treatment of S. aureus infection.

Keywords: *Staphylococcus aureus*, prevalence, characterization, drug resistance, child fecal carriage

## INTRODUCTION

Staphylococcus aureus (S. aureus) is a major pathogen of human infection that causes diseases ranging from minor skin infections to severe bacteremia, necrotizing pneumonia, and lifethreatening sepsis (1–3), and thus is a major global threat to human health. S. aureus can colonize multifarious body regions, including the anterior nares (4), skin (5), intestinal tract (6), oropharynx (7), and so on. Colonization is a crucial risk factor for the subsequent development of infections (8). Specifically, the importance of S. aureus fecal colonization was described as early as 1960 (9), in a study that demonstrated rectal carriage of S. aureus earlier than from the nose or throat. Subsequently, several studies have confirmed the clinical importance of S. aureus fecal carriage (10, 11). Additionally, S. aureus fecal carriage may contribute to environmental contamination (12), which can lead to nosocomial transmission and infection. Previous studies have reported fecal carriage of S. aureus in adults from Nigeria (13) and India (14), and a recent study investigated intestinal colonization by S. aureus and Clostridium difficile in healthy adult fecal samples from China (15), and a few studies have reported on S. aureus isolated from pediatric patients' feces samples in China.

Multilocus sequence typing (MLST) has become one of the most popular methods for evaluating S. aureus strains; however, only limited MLST studies of S. aureus from stool samples are available. Methicillin-sensitive (MSSA) strains ST30, ST398, and ST133 were detected from 100 healthy human fecal samples in Spain (16), while ST15, ST188, and ST59 were identified in six S. aureus isolates from stool specimens of diarrheal infants (17). Unfortunately, the diversity of molecular S. aureus types in these studies was limited due to the relatively small population size, which may have led to misinterpretation or inaccurate conclusions to be drawn regarding S. aureus colonization in fecal samples.

In addition to molecular characterization, antibiotic resistance, notably regarding the emergence and evolution of multi-drug-resistant (MDR) S. aureus, has become a major focal point in research across the world. Methicillin-resistant S. aureus (MRSA), which begins with resistance to methicillin or most β-lactam antibiotics and gradually develops co-resistance to vancomycin (18, 19), limits the use of alternative anti-infective drugs and threatens patient's health. Hence, drug resistance should be closely monitored to provide the basis for clinical antibacterial infection treatment, including exploring antibiotic resistance of S. aureus isolated from fecal samples.

Thus, the aims of this investigation were to evaluate the prevalence, molecular genotyping and antibiotic resistance of S. aureus isolated from pediatric patients' fecal samples in Southern China.

## MATERIALS AND METHODS

## Ethics

All patients were recruited voluntarily and provided informed consent from the participants or the guardians. The study was approved by the research ethics committee of the Guangzhou women and children's medical center (registration no. 2016081029).

## Bacterial Isolates and Data Collection

This study enrolled children from three medical centers in Southern China between August and November 2018, including Guangzhou Women and Children's Medical Center (Tianhe District, central Guangzhou), Guangzhou Children's Hospital (Yuexiu District, western Guangzhou), and Zengcheng Maternity and Children's Health Care Center (Zengcheng District, northern Guangzhou). A total of 2059 non-duplicate pediatric stool samples (1308 outpatients and 751 inpatients) were collected. Approximately 20 mg of stool sample was streaked onto a selective mannitol salt agar medium (Hope Bio-technology, Qingdao, China) within 4 h of sample collection, and incubated in a humidified atmosphere at 37◦C with 5% CO<sup>2</sup> for 24 h. Suspected S. aureus colonies from each sample were evaluated based on morphology and sub-cultured on Columbia Blood Agar Medium (Detgerm Microbiology Technology, Guangzhou, China) (20). All isolates were further identified for their species assignment by the automated VITEK MS (bioMérieux, Marcyl'Étoile, France). Identified S. aureus was further confirmed by detecting femB (21). We also collected a range of clinical information from the laboratory information system, including gender, age, types, diagnosis, and fecal occult blood test (FOBT). Accordingly, the patients were classified into six age groups: 0– 28 days, newborn; 28 days–3 months, young infant; 3 months–1 year, older infant; 1–3 years, child; 3–6 years, pre-school age; 6–18 years, school age and puberty (22, 23). Among these patients, the oldest was 17 years old, the youngest was 1 day, and the median age was 10 months and 23 days.

## Antibiotic Susceptibility Tests

Antibiotic susceptibility for 15 antibiotics (penicillin, oxacillin, gentamicin, rifampicin, levofloxacin, ciprofloxacin, trimethoprim/sulfamethoxazole, clindamycin, erythromycin, macrodantin, linezolid, vancomycin, quinoputin/dafutin, tetracycline, and tigecycline) was detected by VITEK 2 AST-GP67 cards (bioMérieux) using the automated VITEK2 compact system (bioMérieux). Antibiotic minimum inhibitory concentration (MIC) was determined according to the published guidelines (24). The quality control strain used in antibiotic susceptibility analysis was S. aureus ATCC 29213. MRSA was defined as an oxacillin-resistant isolate, and multidrug-resistant (MDR) isolates were identified as isolates with resistance to three or more non-β-lactam antibiotics (25).

## DNA Extraction

Total DNA was extracted from 1.0 ml of nutrient broth medium culture grown overnight. After centrifugation, the supernatant was discarded and the S. aureus isolates were resuspended in 200 µl of enzymatic lysis buffer (Sangon Biotech, Shanghai, China). Subsequently, S. aureus solution was incubated at 37◦C for 30 min with 3 µl of lysostaphin (Sigma-Aldrich, Shanghai, China), mixed with 200 µl of Buffer BD (Sangon Biotech), and 200 µl of 100% ethanol (Guangzhou Chemical Reagent Factory, Guangzhou, China), and then transferred to an absorbing column (Sangon Biotech). Next, the Ezup Column Bacterial Genomic DNA Extraction Kit (Sangon Biotech) was used in accordance with the manufacturer's instructions.

## PCR Detection of *femB* and *mecA*

The femB gene plays an important role in formatting the pentaglycine bridges that stabilize peptidoglycan chains in S. aureus, while the mecA gene is the most important cause of S. aureus resistance to oxacillin (26). Thus, to further confirm the presence S. aureus and MRSA, we detected the expression of femB and mecA by PCR. The primers used for femB and mecA genes were described previously (27), and extracted DNA was amplified using TaqTM (Takara, Tokyo, Japan) following the manufacturer's instructions. Following amplification and extension, the PCR amplicons were separated on 1% agarose gels stained with ethidium bromide and visualized under UV illumination (TEX-20 M, Life Technologies, Carlsbad, USA).

## MLST Typing

All isolates were analyzed by multilocus sequence typing (MLST) according to a previously published procedure (28). The PCR products were purified and sequenced by a commercial sequencing company (Beijing Genomics Institute, Shenzhen, China). DNA sequencing of seven housekeeping genes (arcC, aroE, glpF, gmk, pta, tpi, and yqiL) was used for MLST analysis. Sequence types (STs) were determined by searching the S. aureus MLST database (https://pubmlst.org/saureus/), which included new emerging MLST alleles and MLST types. Clonal complex (CC) analysis was conducted using the eBURST v.3 programme (https://www.mlst.net/eburst/) according to our previously described protocol (27). Based on STs, a UPGMA dendrogram was constructed with START2.

## Statistical Analysis

Statistical analyses were carried out with SPSS software 20 (SPSS Inc., Chicago, USA). The chi-square (χ²) test or Fisher's exact test were applied to dichotomous or categorical variables, which were described as frequencies and proportions. P < 0.05 was considered statistically significant.

## RESULTS

## Prevalence of *S. aureus* and MRSA Was Associated With Clinical Features

A total of 2059 fecal samples were collected from pediatric patients in three hospitals, from which 412 S. aureus and 93 MRSA isolates were identified. The overall colonization prevalence of S. aureus and MRSA were 20.0 and 4.5%, respectively. Accordingly, we classified the 2059 patients into two groups consisting of those that were S. aureus positive (n = 412) and negative (n = 1,647), to analyze the correlation of S. aureus fecal carriage with different clinical features. As shown in **Table 1**, the fecal carriage of S. aureus was not associated with patient gender (P = 0.149); however, it was positively correlated with outpatient status and gastroenteritis diagnosis (P < 0.01). We also observed a positive relationship between S. aureus and different age groups. The minimum S. aureus carriage rate was TABLE 1 | Correlation of fecal carriage of S. aureus in pediatric patients with different clinical features.


+: S. aureus-positive group; –: S. aureus-negative group. Clinicopathological features were assessed using the chi-square test, \*\*P < 0.01.

in newborn patients (7.3%), while the maximum rate was in infant patients (young infants, 26.6% and older infants 26.3%), after which the carriage rate gradually descended with increasing age of the patients (**Table 1**). In addition, we observed a positive correlation between FOBT results and S. aureus carriage (P < 0.01, **Table 1**).

Subsequently we divided the 412 positive S. aureus patients into two groups, MRSA (n = 93) and MSSA (n = 319), to explore the relationship of MRSA and clinical features. However, we found that MRSA was significantly correlated with inpatient status (P = 0.002), and not gender, age, or FOBT results (**Table 2**).

## Antibiotic Susceptibility of *S. aureus* and MRSA

The antibiotic susceptibility results for the 412 S. aureus isolates according to MLST are presented in **Table 3**. All 93 MRSA strains were resistant to cefoxitin screening and carried the mecA gene. The S. aureus strains exhibited highest rate of resistance to penicillin (PEN, 84.2%), followed by erythromycin (ERY, 38.8%), clindamycin (CLI, 35.9%), tetracycline (TE, 14.6%), and sulfamethoxazole-trimethoprim (SXT, 6.1%). The resistance rates of antibiotics were lower for gentamicin (GEN, 2.7%), levofloxacin (LEV, 1.9%), ciprofloxacin (CIP, 1.9%), and rifampicin (RIF, 0.7%); however, all isolates were susceptible to macrodantin, linezolid, vancomycin, dalfopristin/quinupristin (QDA), and tigecycline. Compared to the MSSA group, the MRSA group had significantly higher rates of resistance to PEN (P < 0.01), ERY (P < 0.01), CLI (P < 0.01), and TE (P = 0.03) and intermediate resistance to RIF (P < 0.01). Although 22.8% of all strains exhibited MDR, the MRSA group had a significantly higher rate (74.2%) compared to that of the MSSA group (7.8%) (P < 0.01).

## Molecular Characterization of *S. aureus*

According to the results of the MLST method, 76 unique STs were identified among 412 S. aureus isolates, including 25 novel STs. Based on eBURST analysis, the 76 STs were classified into 28 CCs, including 14 groups and 14 singletons (**Figures 1**, **2**). The three most abundant STs among all S. aureus isolates were ST188



MRSA, methicillin-resistant S. aureus; MSSA, methicillin-susceptible S. aureus. Clinicopathological features were assessed using the chi-square test, \*\*P < 0.01.

(12.9%), ST45 (12.1%), and ST59 (10.0%), comprising 35% of all isolates. Among the MRSA group, the three most abundant STs were ST59 (37.6%), ST45 (35.5%), and ST1 (5.4%), comprising 78.5% of all strains. Among MSSA group, ST188 (16.0%), ST6 (9.7%), and ST15 (8.5%) were the three prevalent STs. Further, the most common CCs among all strains were CC188, CC45, and CC59, representing 39.6% of all clones. Specifically, within the MRSA group, CC59 (41.9%), CC45 (35.5%), and CC1 (6.5%) were the three most abundant CCs, while in the MSSA group, the most common clone was CC188 (17.9%), followed by CC5 (11.6%) and CC6 (11.0%) (**Table 4**).

As shown in **Figure 2**, there were 25 newly assigned STs (ST5307 to ST5330, and ST5353) in this study, of which many were single-locus variants (SLVs). Among the 25 novel STs, 26 strains were identified, including 2 MRSA and 24 MSSA isolates; both MRSA isolates belonged to CC59, and the most abundant CC in the MSSA groups was CC188. Finally, we identified 18 novel SLVs in seven housekeeping genes, which were subsequently assigned as new alleles (**Table S1**).

## Association of Antibiotic Resistance With Specific *S. aureus* Sequence Types

When analyzing the correlation between antibiotic resistance profiles and unique sequence types (STs) in genotypes of the S. aureus isolates, 16 STs each had more than three strains that were selected in this study. Some specific STs were determined to be closely associated with certain antibiotic resistance patterns, while some exhibited high sensitivity. As shown in **Figure 2** and **Figure S1**, all ST338, ST25, ST30, and ST1281 isolates were resistant to PEN, while ST72, ST950, and ST965 showed high sensitivity to PEN. However, ST950 and ST965 showed high resistance to CLI and ERY, yet were sensitive to all other antibiotics, with an MDR rate of 0%, similar to the MDR rate

TABLE 3 | Antibiotic susceptibility of S. aureus and MRSA isolates from pediatric patients' feces.


SXT, trimethoprim/sulfamethoxazole; QDA, dalfopristin/quinupristin. R, Resistant; I, Intermediate; NA, not applicable. <sup>a</sup>Antibiotic resistance of MRSA vs. MSSA by chi-squared test (two-sided); <sup>b</sup>Rifampicin intermediate of MRSA vs. MSSA by chi-squared test (two-sided), \*\*P < 0.01, \*P < 0.05.

of ST1281 and ST15. Similarly, ST188, with the largest number of strain types and largest antibiotic resistance coverage, had a very low MDR rate (3.8%). All ST30 and ST338 isolates were resistant to ERY and CLI. ST59, which belongs to CC59 with ST338, also showed a high resistance rate (78%) to ERY and CLI. Additionally, ST59 and ST338 had a higher rate of resistance to TE, with the two highest MDR rates of 75.6 and 60.0%, respectively. In addition to the STs mentioned above, ST7 exhibited the highest resistance rate (61.1%) to TE, while ST45, another predominant ST in the MRSA isolate group, had the highest resistance and intermediate resistance rate (84.0%) to RIF (**Figure 2** and **Figure S1**).

## DISCUSSION

S. aureus fecal carriage may contribute to environmental contamination, facilitate nosocomial transmission, and promote human disease development. Moreover, fecal carriage of S. aureus among children is more likely to cause disease infection due to their immature and underdeveloped immune system (29). Since S. aureus fecal colonization has been identified as a risk factor for infection disease development (30), our findings may serve to advance the current understanding regarding S. aureus fecal colonization dynamics and prevention of S. aureus infection.

In this study, the S. aureus fecal carriage rate was determined to be 20.0% in pediatric patients from Guangzhou, while the MRSA carriage rate was 4.5%. These results agree with the reported prevalence of pooled estimates for S. aureus and MRSA fecal carriage rates (16.8–36.3%, 0.7–27.0%, respectively) (30). However, the prevalence of S. aureus in this study was higher than that reported from participants with nosocomial diarrhea in Germany (7%) (31) compared to healthy adults in China (3.51%) (15), but was lower than that reported in a previous Nigerian study (31.7%) (13). Moreover, in China, the prevalence of MRSA nasal colonization in children between 2005 and 2015 was 4.4% (32), similar to the 4.5% carriage rate detected in our study, but slightly higher than a previous study in American children with cancer (2.9%) (33). Although the prevalence of S. aureus and MRSA is dynamic due to differences in geographical regions, age, gender, and health status, future studies continue to be warranted to better characterize the cause of these differences.

In our study, the fecal carriage of S. aureus was positively correlated with outpatient status and gastroenteritis diagnosis. Pediatric gastroenteritis primarily manifests as abdominal pain, diarrhea, and vomiting, and these patients comprise the majority of outpatients. Moreover, S. aureus has been described as the most common global causative pathogen of food-borne illness, while studies have reported it to be associated with infantile diarrhea (17), corresponding with our observed positive correlation between S. aureus fecal carriage and gastroenteritis in children. In addition, our study found that fecal S. aureus colonization was lowest during the first 4 weeks of life, after which it increased rapidly during the first year, followed by a gradual decline until 17 years of age. This may be explained by the underdeveloped intestinal function and microbiota composition in newborn patients. With improved intestinal function and increased diversity of intestinal microbes, S. aureus colonization increased within the following year. Further, human milk oligosaccharides have been suggested to be a strong contributor to bacterial reproduction in the infant gut and to stimulate S. aureus growth (34), providing an important function for breastfeeding in early life. Subsequently, with the consumption of a comprehensive diet and enhanced immune function, the fecal

carriage of S. aureus decreases with increasing age. Additionally, within this study, S. aureus fecal carriage was higher in patients that tested FOBT positive, which may be explained by virulence factors, especially staphylococcal enterotoxins, produced by S. aureus, causing intestinal damage that leads to intestinal bleeding. However, this hypothesis requires further validation.

We also determined that fecal carriage of MRSA was positively associated with inpatients status as hospitalized patients are more likely to be infected with MRSA (35). Interestingly, 7 of the 68 hospitalized patients with S. aureus in their stool also had it within in their sputum or alveolar lavage fluid, demonstrating similar antibiotic susceptibility patterns, including 2 MRSA and 5 MSSA isolates (data not shown). This may suggest that the S. aureus isolated from stool sample was consistent with the source in the sputum or alveolar lavage fluid, and fecal carriage of S.

#Representing 16 STs that each had more than three strains. <sup>a</sup>Resistance and intermediary resistance of RIF. Different colors represent different clonal complexes.


TABLE 4 | Genotype ranking of S. aureus isolated from pediatric patients.

aureus may be associated with infection in other parts of the body.

Based on the MLST results, 412 S. aureus isolates were divided into 76 STs, with fewer MRSA isolates (14 STs) than MSSA isolates (69 STs), indicating that MRSA isolates were more genetically stable than MSSA isolates. The most commonly reported S. aureus isolates in China are diverse and include ST1 (36), ST6 (20), ST5 (37), and ST188 (38) according to different regions, ages, and resources. Similarly, many different MSSA isolates have been identified throughout China, including ST7 and ST188 (39). In our study, ST188, ST6, and ST15 were the most frequently observed STs in the MSSA group. ST188 was reported as a major cause of childhood infections in China, due to its high adhesion and biofilm formation ability (38). In addition, ST15 and ST188 were the two most prevalent clones isolated from infantile diarrhea fecal samples (17). Although the types of MSSA strains are diverse, the most abundant ones have not changed significantly among children in China.

MRSA strains demonstrated strong homology with the prevalent clones in this study, which were determined to be ST59 (37.6%) and ST45 (35.5%). This result was consistent with a previous study reported by Ding et al. in Chinese children (40). ST59, a predominant MRSA clone causing CA-MRSA infections among children (41), was not only predominant in Chinese cities, including Shanghai (39), Sichuan (42), and Taiwan (43), but also is considered to be a prevalent isolate throughout the Asia-Pacific region (44). It is worth noting that ST45, the second most prevalent MRSA strain, has an increased carriage rate in children, compared to 18.8% for the MRSA isolated in our previous study (27). ST45, known as a Berlin clone, was also a common isolate throughout European countries, and has now spread to Australia (45), Singapore (46), and China (41). Further, previous studies from Shanghai have shown that ST239 was the most frequent MRSA clone between 2005 and 2010 (47), which was replaced by the increasingly abundant ST5, ST59, and ST398 clones between 2008 and 2017 (48). However, ST239 was not identified in our study, and ST5 and ST398 only accounted for 2.2% of MRSA isolates. Alternatively, ST45 and ST59 were not only the two dominant clones in the MRSA group, but also the second and the third most abundant clones in all S. aureus isolates. ST45-MRSA is attributed to the acquisition of mecA by a MSSA clone in the community (49). Therefore, our results suggest that ST45 would alter the MRSA clone structure in this region of China, allowing for the development of additionally prominent clones in Chinese children, such as ST59.

The drug resistance of S. aureus has attracted great attention worldwide, especially in China, due to the abuse of antibiotics and the recent emergence of MDR bacteria. Consistent with a previous report (50), we show that S. aureus isolates exhibited a high rate of resistance to PEN, ERY, and CLI, which may be the result of the excessive use of PEN and macrolides (51). The current study demonstrated that S. aureus was more susceptible to vancomycin and linezolid, and displayed 100% sensitivity to vancomycin, linezolid, macrodantin, and QDA. Similarly, relatively low rates of resistance to CIP and LEV were observed in S. aureus isolates, which may be a result from the infrequent use of fluoroquinolones in pediatric patients due to their reported cartilage toxicity. In addition, S. aureus strains can differ in their antibiotic resistance patterns with specific STs. CC59 (ST59 and ST338) exhibited the highest MDR rate and a specific antibiotic resistance pattern (ERY-CLI-TE), while another common clone, CC45 (ST45), exhibited a different antibiotic resistance pattern (ERY-CLI-RIF). ST45 resistance or intermediate resistance to RIF is caused by rpoB mutations (52). However, strains assigned to the same cluster seemed to have similar resistance patterns, suggesting that further genotyping of the S. aureusstrains may assist in designing more effective clinical treatment regimens.

Certain limitations were noted within this study. Firstly, more detailed clinical information was difficult to obtain, especially for outpatients, which account for the majority of patients; therefore, the study was limited in its ability to analyze the effect of additional risk factors for S. aureus carriage or MRSA colonization, including premature birth, duration of hospital stay, mother carriage status, history of antibiotic intake, etc. Secondly, inpatient status is strongly correlated with the carriage of MRSA; however, due to the diversity of the hospital environment, we were unable to differentiate MRSA strains based on the various areas in which the inpatients were admitted to; the source of MRSA is worth exploring further. Lastly, we only applied a single method for S. aureus typing, which limited access to more specific and detailed prevalent molecular characterization of S. aureus.

In summary, the S. aureus carriage rate in pediatric feces in Southern China was as high as 20%. MSSA and MRSA exhibited significant differences in genotyping and antimicrobial susceptibility, with ST59 and ST45 emerging as two major MRSA clones and ST188 as the most prevalent MSSA clone. Antibiotic resistance patterns of S. aureus were also found to be closely associated with specific STs. These findings clarify the colonization dynamics and molecular epidemiology of S. aureus from child feces in Southern China and suggest an urgent need to strengthen the surveillance programs in this region, while also providing important information regarding the prevention and treatment of S. aureus infection.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the https:// pubmlst.org/bigsdb?db=pubmlst\_saureus\_seqdef&page= downloadProfiles&scheme\_id=1.

## ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the research ethics committee of the Guangzhou women and children's medical center (registration no. 2016081029). Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

## REFERENCES


# AUTHOR CONTRIBUTIONS

ZZ and YL initiated and designed the study. XA, FG, SY, BL, JM, ZX, XC, and ZL performed the experiments and/or analyzed the data. XA wrote the draft. HY, ZO, SG, ZZ, and YL revised the manuscript. All authors have approved the final version.

## FUNDING

This work was supported by grants from the Natural Science Foundation of Guangdong (Nos. 8451012001001570 and 9151012001000009), Guangdong Science and Technology Department (Nos. 2014A020212013 and 2016A020215013), Medical Health Science and Technology Foundation of Guangzhou (Nos. 20171A010267, and 20181A011039), Paediatric Institute Foundation of Guangzhou Women and Children's Medical Center (Nos. Pre-NSFC-2019-014 and IP-2019-022), and Guangzhou Science Technology and Innovation Commission (No. 201707010010).

## ACKNOWLEDGMENTS

We would like to thank all the patients who provided their specimens and clinical data for this study.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed. 2020.00127/full#supplementary-material

Figure S1 | Antibiotic resistance of S. aureus isolates from pediatric feces linked to sequence types, as illustrated by the tri-color scale.


Nigeria. J Health Popul Nutr. (2011) 29:446–53. doi: 10.3329/jhpn.v29i 5.8898


prevalence meta-analysis and review of influencing factors. PLoS ONE. (2016) 11:e0159728. doi: 10.1371/journal.pone.0159728


isolates from child patients of high-risk wards in Shenzhen, China. Jpn J Infect Dis. (2017) 70:479–84. doi: 10.7883/yoken.JJID.2016.328


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

# Resistance of Klebsiella pneumoniae Strains Carrying blaNDM-1 Gene and the Genetic Environment of blaNDM-1

Tianxin Xiang<sup>1</sup>† , Chuanhui Chen<sup>2</sup>† , Jiangxiong Wen<sup>3</sup> , Yang Liu<sup>1</sup> , Qi Zhang<sup>1</sup> , Na Cheng<sup>3</sup> , Xiaoping Wu<sup>3</sup> and Wei Zhang<sup>2</sup> \*

<sup>1</sup> Department of Hospital Infection Control, The First Affiliated Hospital of Nanchang University, Nanchang, China, <sup>2</sup> Department of Respiratory and Critical Care, The First Affiliated Hospital of Nanchang University, Nanchang, China, <sup>3</sup> Department of Infectious Disease, The First Affiliated Hospital of Nanchang University, Nanchang, China

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

## Reviewed by:

Zhi Ruan, Zhejiang University, China Andres Felipe Opazo-Capurro, University of Concepción, Chile

#### \*Correspondence:

Wei Zhang ndzhangwei2016@126.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 26 September 2019 Accepted: 25 March 2020 Published: 30 April 2020

#### Citation:

Xiang T, Chen C, Wen J, Liu Y, Zhang Q, Cheng N, Wu X and Zhang W (2020) Resistance of Klebsiella pneumoniae Strains Carrying blaNDM-1 Gene and the Genetic Environment of blaNDM-1. Front. Microbiol. 11:700. doi: 10.3389/fmicb.2020.00700 Objective: Regional dissemination is the major cause of the widespread prevalence of a plasmid-encoding NDM-1 enzyme. We investigated the drug resistance, joint efficiency, and gene environment of a Klebsiella pneumoniae strain carrying blaNDM-1 gene.

Materials and Methods: Carbapenem-non-susceptible strains were analyzed using the VITEK 2 Compact. Strains carrying blaNDM-1 were identified using polymerase chain reaction and sequencing. Antimicrobial susceptibility testing and plasmid conjugation experiments were then conducted. Strains carrying blaNDM-1 were subjected to Southern blot analysis. After the gene mapping of blaNDM-1, library construction, and sequencing, plasmids were subsequently spliced and genotyped using the software Glimmer 3.0, and then analyzed using Mauve software.

Results: Among 1735 carbapenem-non-susceptible strains, 54 strains of blaNDM-1 positive bacteria were identified, which consisted of 44 strains of K. pneumoniae, 8 strains of Acinetobacter baumannii and 2 strains of Escherichia coli. Strains carrying blaNDM-1 had a resistance rate of more than 50% in most antibiotics. Plasmid conjugation between strains carrying blaNDM-1 and E. coli strain J53 had a success rate of 50%. Southern blot analysis indicated that each strain had multiple plasmids containing blaNDM-1. Among the five plasmids containing blaNDM-1 in K. pneumoniae for sequencing, two plasmids with complete sequences were obtained. The findings were as follows: (i) The p11106 and p12 plasmids were highly similar to pNDM-BTR; (ii) the p11106 and p12 plasmids showed differences in the 20–30 kb region (orf00032–orf00043) from the other six plasmids; and (iii) blaNDM-1 was located at orf00037, while ble was found at orf00038. Two tnpA genes were located in the upstream region, and orf00052 (tnpA) in the 36 kb region was in the downstream sequence.

Conclusion: blaNDM-1-containing bacteria exhibit multidrug resistance, which rapidly spreads and is transferred through efficient plasmid conjugation; the multidrug

**419**

resistance of these bacteria may be determined by analyzing their drug-resistant plasmids. The presence of ble and tnpA genes suggests a possible hypothesis that blaNDM-1 originates from A. baumannii, which is retained in K. pneumoniae over a long period by transposition of mobile elements.

Keywords: carbapenem resistance, NDM-1, Klebsiella pneumoniae, plasmid, genetic characteristics

## INTRODUCTION

fmicb-11-00700 April 28, 2020 Time: 17:24 # 2

The clinical application of sulfa drugs can be traced back to the 1930s, which marked a new era of antimicrobial therapy. Once exposed to antibacterial drugs, bacteria spontaneously change their metabolic pathways or produce corresponding inactivating substances to resist antibiotics, exhibiting drug resistance. Notably, the abuse of antibiotics poses selective pressure of bacteria, conferring a survival advantage on drug-resistant bacteria. Consequently, numerous drug-resistant bacteria are spread in different pathogens.

New Delhi metallo-β-lactamase1 (NDM-1), also known as metallo-β-lactamase or metal-β-lactamase, was first isolated from a highly infectious and pathogenic multidrug-resistant strain of Klebsiella pneumoniae in 2009 (Yong et al., 2009). Cases of infection with the blaNDM-1 gene have subsequently been reported in more than 20 countries and regions worldwide, including the United Kingdom and India (Kumarasamy et al., 2010). The therapeutic efficacy of multiple antibiotic treatments for bacteria carrying blaNDM-1 is usually unsatisfactory. Therefore, bacterial strains carrying blaNDM-1 are also called superbugs. With a prolonged length of stay, bacteria-carrying blaNDM-1 have a higher probability of being isolated from stool samples. However, infection with blaNDM-1 cannot be determined from clinical symptoms and signs (Bush and Fisher, 2011). Superbugs pose a serious challenge to antibiotic therapy.

The blaNDM-1 gene is mainly distributed in plasmids and occasionally in the chromosomes of Escherichia coli, Pseudomonas aeruginosa, and Proteus mirabilis (Girlich et al., 2015; Rahman et al., 2015; Shen et al., 2017). In clinical practice, blaNDM-1 plasmids from different bacterial species isolated from the same patient typically have a similar structure, suggesting the significance of plasmids in the spread of blaNDM-1. Plasmids containing blaNDM-1 vary in size from 30–300 kb, and exist in different types, such as IncA/C, IncL/M, and IncR (Carattoli et al., 2015; Gamal et al., 2016). Specifically, IncA/C has a wide range of hosts and can exist in multiple strains, such as Enterobacteriaceae, Pseudomonas, Acinetobacter, and Vibrio cholerae, providing convenience for blaNDM-1 in different species of bacterial hosts (Wailan and Paterson, 2014).

The sequences and genetic modes of blaNDM-1 have been identified in previous studies. However, the blaNDM-1 gene environment is yet to be determined. In addition, studies mostly focus on blaNDM-1 in Acinetobacter (Bontron et al., 2016; Wang et al., 2017) and rarely on blaNDM-1 carried by K. pneumoniae. In the present study, 1735 carbapenem-nonsusceptible bacteria were collected from The First Affiliated Hospital of Nanchang University. These strains were sequenced, conjugated, and compared with the corresponding plasmids. This study aimed to analyze the differences in the blaNDM-1 gene environment and provide directions in clarifying the origin and propagation of the blaNDM-1 gene.

## MATERIALS AND METHODS

## Sample Collection and Identification

Approval was obtained from the Medical Ethics Committee of The First Affiliated Hospital of Nanchang University and informed consent was obtained from each subject. Carbapenemnon-susceptible bacteria were then collected from The First Affiliated Hospital of Nanchang University from January 2013 to December 2016. A total of 1735 carbapenem-nonsusceptible bacteria were isolated and then identified using VITEK 2 Compact (Pioneering Diagnostics, France) at the Microbiology Laboratory at The First Affiliated Hospital of Nanchang University.

## Polymerase Chain Reaction and Sequencing

Polymerase chain reaction (PCR) templates were prepared by boiling. Fresh bacteria were harvested and diluted in 500 µL ddH2O for 10 min in a boiling water bath. The supernatant was collected as PCR templates. Subsequently, a PCR system with a total volume of 50 µL was prepared, consisting of 25 µL of Taq Mix (Takara, Dalian), 2 µL of forward primer, 2 µL of reverse primer, 2 µL of template, and 19 µL of ddH2O. Subsequently, 29 cycles of PCR were conducted. The blaNDM-1 primer sequences were forward 5'-GGCGGAAGGCTCATCACGA-3' and reverse 5'-CGCAACACAGCCTGACTTTC-3'. The amplified product was 287 bp.

The PCR products were analyzed using electrophoresis with 1% agarose gel and 1 × TAE at 120 V for 25 min. Strains verified by sequencing to contain blaNDM-1 were used as markers. Electrophoresis results were obtained using an ultraviolet (UV) transilluminator. Positive PCR products were sequenced (Synbio Technology, Suzhou, China), and the sequencing results were compared using the software BLAST. Strains that were positive for PCR and matched the sequencing results were identified as those containing blaNDM-1.

## Antimicrobial Susceptibility Testing

Antimicrobial susceptibility testing was conducted using the zone of inhibition test, and the comprehensive drug resistance of each antibiotic was determined using the E-test. The antibiotics screened in this study included imipenem, meropenem, ertapenem, amikacin, amoxicillin/clavulanicacid, aztreonam, ceftriaxone, ceftazidime, cephalosporins, cefoxitin, cefazolin, ciprofloxacin, gentamicin, levofloxacin, piperacillin/tazobactam, trimethoprim/sulfamethoxazole, tetracycline, ticarcillin/clavulanicacid, and tobramycin.

## Plasmid Conjugation

fmicb-11-00700 April 28, 2020 Time: 17:24 # 3

The receptor strain was sodium azide-resistant E. coli strain J53. The donor and receptor strains were implanted in the Mueller-Hinton plate and cultured at 37◦C for 16–18 h. Strains in appropriate amounts were inoculated in a glass tube containing 5 mL of LB medium and then cultured at 37◦C for 16–18 h. Subsequently, 400 µL of the donor strain and 200 µL of the receptor strain were added to a glass tube containing 800 µL of LB broth medium and then cultured at 37◦C for 16–18 h. Meanwhile, the donor strain, screened in 180 µg/mL sodium azide, and the receptor strain, screened in 0.5 µg/mL of imipenem, were used as blank controls. Exactly 100 µL of the aforementioned mixture was added to the Mueller-Hinton plate and cultured at 37◦C for 16–18 h. Conjugation strains in good condition were ultimately identified using the VITEK 2 compact automatic microbial identification instrument.

## Southern Blot Analysis

Southern blot analysis was conducted using 1% agarose gel with 1 × TAE and run on 120 V electrophoresis for 40 min. The gel was incubated in 0.25 mol/L HCl for 15 min, 0.5 mol/L NaOH for 20 min twice, and 0.1 mol/L phosphate buffer for 15 min twice. Membrane transfer was conducted in a 20 × saline-sodium citrate buffer. The membrane was washed in a 2 × saline–sodium citrate buffer and then dried in a baking oven at 80◦C for 2 h. PCR products carrying blaNDM-1 were labeled using the DIG High Prime DNA Labeling and Detection Starter Kit I (Roche, United States). DIG-labeled DNA products were prepared and examined using Southern blot analysis, and images were obtained using a UV transilluminator.

## Plasmid Sequencing

Five qualified plasmids containing blaNDM-1 were used to construct a sequencing library. These 5 vectors were from 4 strains, which showed high resistance to the antibiotics and successfully conjugated with E. coli J53. The data are provided as **Supplementary Material** and **Supplementary Table S2**. Briefly, 1 µg of plasmid was placed in a Covaris tube, and the DNA was separated into 400 bp fragments using Covaris S2 (Covaris, United States). Small DNA fragments were generated for library construction using the NEXTflex DNA Sequencing Kit compatible with Biomek FXp (Bio Scientific, United States). The library fragments were subjected to pairedend sequencing (2 × 150 bp) on the HiSeq2500 Sequencing System (Illumina, United States).

Clean reads after pre-processing were assembled using Velvetver.1.2.03 software. Gene prediction and annotation analyses were performed using Glimmer 3.0 software. The p11106 and p12 plasmids were compared with the plasmid sequences without blaNDM-1 of the seven species. Similarities in the plasmid sequences were depicted using Mauve software. Gene functions in different regions were annotated and analyzed.

## RESULTS

## Screening and Identification of Strains Carrying blaNDM-1

Among the 1735 carbapenem-non-susceptible strains harvested in this experiment, 54 strains (3.1%) were blaNDM-1-positive. These strains consisted of 44 strains of K. pneumoniae, 8 strains of A. baumannii, and 2 strains of E. coli. The blaNDM-1 gene was not found in P. aeruginosa, Enterobacter cloacae, Bacillus, Maltophilia, or Pseudomonas cepacia. All sequencing results of the 54 strains were consistent with the NCBI database<sup>1</sup> . The positive rates of each strain are listed in **Table 1**.

These 54 multidrug-resistant strains were obtained from 43 patients at The First Affiliated Hospital of Nanchang University from January 2013 to December 2016. The patients were from different cities and provinces. Temporal and regional differences suggested that the same strains could not be causing the outbreak. Patient records are shown in **Supplementary Table S1**.

## Determination of Drug Resistance

The resistance rate of the strains carrying blaNDM-1 exceeded 50% in most antibiotics (**Figure 1**). The tested antibiotics exhibited nearly 100% resistance to imipenem and more than

<sup>1</sup>http://blast.ncbi.nlm.nih.gov/Blast.cgi

TABLE 1 | Positive rate of the blaNDM-1 gene in each strain.


90% resistance to meropenem and piperacillin. Meanwhile, the blaNDM-1-positive strains showed the lowest resistance (40%) to amikacin.

## Plasmid Conjugation

fmicb-11-00700 April 28, 2020 Time: 17:24 # 4

The plasmid conjugation experiment was conducted on all blaNDM-1-positive strains and J53. A total of 27 strains (21, K. pneumoniae strains; 5, A. baumannii strains; and 1, E. coli strain) successfully conjugated their plasmids containing the blaNDM-1 gene (**Table 2**). The success rate of plasmid conjugation was 50%.

## Location of the blaNDM-1 Gene

BlaNDM-1-positive strains that were successfully conjugated were subjected to Southern blot analysis to detect the location of

TABLE 2 | Plasmid conjugation of strains.


blaNDM-1. The results further demonstrated the conjugation of blaNDM-1 in E. coli strain J53 (**Figure 2**).

## Sequencing of Plasmids

Plasmid libraries were constructed from genomic DNA. Qualified plasmid DNAs were sequenced, and the results suggested excellent qualifications for library construction and sequencing (**Table 3**). After removal of DNA from host genes and other plasmid DNAs, approximately 20% of the reads were extracted from target plasmid DNAs with over 2000× coverage.

Subsequently, the clean reads of each plasmid were synthesized using Velvet. Large-fragment sequence assembly is presented in **Table 4**. Although coverage was sufficiently high (>2000×), the synthesis was not satisfactory, with the contigs between 33 and 161. Thus, artificial gap closing with the KU862632.1 strain was conducted using the Cytoscape platform, and two complete plasmid (p12 and p11106) sequences were generated. Synthesis of p243323, p32, and p7-1973 failed (**Table 4**).

The size of p12 was the same as that of p11106 (58757 bp). Cytosine was located on 50396 bp of p11106, thymine was located in p12, and the remaining regions were the same. In the following gene prediction and annotation analysis, p12 was used as an example. As

FIGURE 2 | Southern blot analysis for blaNDM−<sup>1</sup> localization. The conjugation of blaNDM−1-positive strain J53 was visualized. The 12 strains are denoted by red arrows. Lanes 12, 13, 17, 18, 20, 21, 22, and 23 are visualized bands that indicate Klebsiella pneumoniae strains carrying blaNDM−1. Lanes 6 and 10 indicate Acinetobacter baumannii strains carrying blaNDM−1. Lanes 1 and 2 indicate Escherichia coli strains carrying blaNDM−1. The lower part of this figure is mirrored with the top part.



TABLE 4 | Statistics of plasmid synthesis.


depicted in **Figure 3**, 46 genes were reverse transcribed, and 39 were forward transcribed without chain specificity (p = 0.627). We uploaded the sequencing data to the NCBI database under the SRA (Sequence Read Archive) accession number PRJNA596354.

## Comparative Analysis of the blaNDM-1 Gene Environment

Plasmid sequences from seven species were subjected to similarity analysis using Mauve software. The baseline characteristics of the selected plasmids are listed in **Table 5**. p11106 and p12 were compared with the aforementioned plasmids, and their similarity information is depicted in **Figure 4** and **Table 6**.

In addition to the plasmid pNDM-BTR, differences between p11106 and p12, and the remaining seven plasmids were

#### Xiang et al. blaNDM-1 in Klebsiella pneumoniae Strains

#### TABLE 5 | Plasmid information.

fmicb-11-00700 April 28, 2020 Time: 17:24 # 6


mainly enriched in the 20–30 kb region. Although the plasmid pMR3-OXA181 was not absent in this region, it exhibited a considerably low homology to p11106 and p12. The corresponding genes mapping this 10 kb region were orf00032–orf00043. Moreover, gene annotation showed that two genes (TnpA and a mobile element protein) with similar functions were located in the upstream sequences of this region. BlaNDM-1 was also located in these sequences (orf00037). IMP-4 and β-lactamase were absent in this region. The gene annotations of orf00032–orf00043 are presented in **Table 7**.

As shown in **Table 7**, (i) the p11106 and p12 plasmids were highly similar to pNDM-BTR; (ii) the p11106 and p12 plasmids showed differences in the 20–30 kb region (orf00032–orf00043) from the other six plasmids; and (iii) blaNDM-1 was located in the middle of the orf00032–orf00043 region, whereas ble was found nearby. Two tnpA genes were on the top, and orf00052 (tnpA) in the 36 kb region was in the downstream sequence.

## DISCUSSION

Among the 1735 carbapenem-non-susceptible strains, 54 (3.1%) of blaNDM-1-positive bacteria were identified, consisting of 44



in the 20–37 kb region (green region), with orf00032–orf00052 in p11106 and p12. Two TnpA transposases are located in this region (orf00032 mapped 20169–19597 bp, negative strand; orf00033 mapped 20160–20864 bp, positive strand), indicating that gene insertion is highly possible in this region of p11106 and p12. Similarly, additional 3.5 kb sequences present in the 3' UTR of p11106 and p12 were mapped orf00078–orf00082. This may also be explained by gene transposition or insertion.

TABLE 7 | Gene annotation of orf00032–orf00042.


strains (4.7%) of K. pneumoniae, 8 strains (1.5%) of A. baumannii, and 2 strains (2.1%) of E. coli. Our findings are consistent with previous studies showing that blaNDM-1 is mainly carried by K. pneumoniae, E. coli, A. baumannii, and E. cloacae (Deshpande et al., 2010; Rolain et al., 2010; Zarfel et al., 2011; Dongmei et al., 2017). Among the collected strains carrying blaNDM-1 in our hospital, K. pneumoniae (4.7%) had the highest positive rate, which was inconsistent with other reports that identified Acinetobacter and Klebsiella as the major blaNDM-1 positive bacteria (Chen et al., 2014; Zhang et al., 2014; Shengshu et al., 2015). Both Acinetobacter and Klebsiella are significant in the regional dissemination of blaNDM-1-positive bacteria. K. pneumoniae could be the vital preservation host of blaNDM-1.

Antimicrobial susceptibility testing indicated that NDM-1 resistant bacteria exhibited strong resistance to most antibiotics (except for amikacin), with a total resistance rate above 50%. The drug resistance of NDM-1 to aminoglycosides, βlactam antibiotics/β-lactamase inhibitors, and carbapenems are relatively high (Shengshu et al., 2015). NDM-1 has a drug resistance of up to 100% for dolipenem, ampicillin, furadantin, cefazolin, cefuroxime, cefotaxime, ceftriaxone sodium, ceftazidime, and cefoxitin. Its drug resistance rates to some carbapenems are as follows: meropenem, 97.30%; imipenem, 97.30%; and ertapenem, 92.00%. However, the drug resistance rates of NDM-1 to polymyxin B (37.93%), amikacin (32.65%), tigecycline (7.69%), and polymyxin E (2.33%) are significantly lower, compared with other antibiotics. NDM-1-resistant bacteria might be mostly multidrug-resistant, and plasmids containing blaNDM-1 can perform multiple drugresistant gene transfers (Barguigua et al., 2015; Sarkar et al., 2015; Villa et al., 2015; Liu et al., 2016).

Our experiment obtained 27 successfully conjugated J53 strains resistant to sodium azide, which is consistent with previous studies (Wailan et al., 2015; Kocsis et al., 2016). Conjugation with K. pneumoniae, A. baumannii, and E. coli as donor plasmids containing blaNDM-1 achieved a success rate of approximately 50%. Southern blot analysis confirmed that blaNDM-1 was mainly expressed in plasmids, and each drug-resistant strain carrying blaNDM-1 could contain multiple blaNDM-1-positive plasmids. The blaNDM-1-positive plasmids exhibited a relatively strong ability for conjugation transfer. Specifically, IncA/C showed adaptation to a wide range of hosts. It can be found in Enterobacteriaceae, Pseudomonas, Aeromonas, Vibrio cholera, and other bacterial genera, and can be transmitted within or between strains through conjugation (Wailan and Paterson, 2014; Carattoli et al., 2015; Gamal et al., 2016).

Sequencing of five plasmids obtained from the collected K. pneumoniae strains was highly qualified (coverage >2000×). Regardless, the synthesis of p243323, p32, and p7-1973 failed; only p11106 and p12 were successfully synthesized. This result could be explained by the complexity of the plasmid structure and interruption of the host genome and other plasmids during plasmid extraction. Among the sequencing reads, only 20–30% of the plasmids were identified to carry theblaNDM-1 gene, significantly increasing the difficulty of synthesis. The only difference between p11106 and p12 was the 50396 bp (orf00074 was mapped) region, where c.587T > C occurred in p11106 and thus resulted in p.196Ile > Thr. A total of 85 genes were contained in the p11106 sequence, and most of them had functions that could be identified using BLAST.

The structures of plasmids p11106 and p12 were similar, and their strains Kp.11 and Kp.12 were also similar in resistance (**Supplementary Table S2**). This finding suggests that resistant plasmids are a key factor in determining the drug resistance of strains.

The genetic environment of a certain gene helps reveal the origin and genetic characteristics of the gene. Poirel et al. (2011) uncovered the full-length or truncated ISAba125 sequences in the upstream region of blaNDM-1 and conserved bleomycin (ble) resistance genes in the downstream region. They hypothesized that ISAba125 sequences are carried by blaNDM-1 when it transfers from the original host, and the blaNDM-1 and ble genes come from the same original host strain (Poirel et al., 2011). In 2012, Toleman analyzed all available blaNDM-1-associated sequences and found that ISAba125 sequences are always present in the 100 bp upstream region of blaNDM-1, and demonstrated that blaNDM-1 is a chimera. The chimeric process occurs in A. baumannii and is mediated by ISAba125 (Jones et al., 2014, 2015; Toleman et al., 2015). Our sequencing results identified certain differences in the 20–30 kb region between p11106, p12, and the compared plasmids, except for pNDM-BTR. On the basis of the aforementioned findings, we propose the following: (a) p11106 and p12 are structurally similar to pNDM-BTR, indicating that p11106 and p12 may present characteristics of pNDM-BTR (McGann et al., 2015); (b) the differences in the 20–30 kb region between p11106, p12, and the compared plasmids support the theory of gene polymorphism (Potron et al., 2011; Mata et al., 2012; Datta et al., 2017); (c) the existence of ble indicates that blaNDM-1 may originate from A. baumannii; and (d) the presence of two TnpA in the upstream region of blaNDM-1 and orf00052 (TnpA) in the 36 kb downstream region suggests that blaNDM-1 is acquired from plasmid

transposition and long-term preservation (Campos et al., 2015; An et al., 2016; Khong et al., 2016).

This experiment screened drug-resistant strains carrying blaNDM-1 gene from carbapenem-non-susceptible bacteria collected from The First Affiliated Hospital of Nanchang University between January 2013 and December 2016. We assessed their drug resistance and plasmid conjugation transfer ability. Two complete plasmid sequences were obtained after sequencing analysis of five blaNDM-1-positive plasmids isolated from K. pneumoniae and compared with blaNDM-1-negative plasmids with known sequences. However, data on abundance were limited owing to the small sample size of drug-resistant strains as well as complexity and interferences during plasmid synthesis. In addition, the conclusion drawn from sequencing in this study is yet to be verified by molecular experiments. Gene deletion (IMP-4 and β-lactamase) occurred in p11106 and p12. The presence of a potential relationship between gene deletion and blaNDM-1 requires further study.

## CONCLUSION

In conclusion, NDM-1-resistant bacteria exhibit multidrug resistance and spread to a certain extent in the hospital via efficient plasmid conjugation transfer. K. pneumoniae may be an important intermediate retention host in the dissemination process. Nosocomial blaNDM-1-carrying bacteria are of concern.

Drug-resistant plasmids may be a key factor in determining drug resistance. The p11106 and p12 plasmids containing the blaNDM-1 gene isolated from K. pneumoniae exhibit the characteristics of the pNDM-BTR plasmid and have a polymorphic gene environment. The existence of ble in the surrounding environment of blaNDM-1 suggests that blaNDM-1 is derived from A. baumannii. Meanwhile, the presence of TnpA in both the upstream and downstream regions of blaNDM-1 indicates that blaNDM-1 may be acquired from gene transposition and persists over a long period in K. pneumoniae strains.

## REFERENCES


## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available from the corresponding author upon reasonable request.

## ETHICS STATEMENT

Written informed consent was obtained from all participants, and the purpose and procedures of the study were explained to them. Ethical approval for this study was obtained from the Institutional Review Board of The First Affiliated Hospital of Nanchang University.

## AUTHOR CONTRIBUTIONS

TX and CC designed the study, collected and analyzed the data, and drafted the manuscript. TX, CC, JW, YL, and QZ contributed to the performance of the experiment and data collection. NC and XW reviewed the study design. WZ contributed to the review of data analysis and data interpretation. All authors have approved the final version of the manuscript.

## FUNDING

This work was supported by the Fund Project of Jiangxi Health and Family Planning Commission (20181058) and the Natural Science Foundation of Jiangxi Province (20181BAB215033).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00700/full#supplementary-material

and India. Antimicrob. Agents Chemother. 59, 7387–7395. doi: 10.1128/aac. 01458-15



**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 Xiang, Chen, Wen, Liu, Zhang, Cheng, Wu and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# Prevalence, Genetic Diversity, and Temporary Shifts of Inducible Clindamycin Resistance Staphylococcus aureus Clones in Tehran, Iran: A Molecular–Epidemiological Analysis From 2013 to 2018

Mehdi Goudarzi<sup>1</sup> \*, Nobumichi Kobayashi<sup>2</sup> , Masoud Dadashi<sup>3</sup> , Roman Pant ˚ucek ˇ 4 , Mohammad Javad Nasiri<sup>1</sup> , Maryam Fazeli<sup>5</sup> , Ramin Pouriran<sup>6</sup> , Hossein Goudarzi<sup>1</sup> , Mirmohammad Miri<sup>7</sup> , Anahita Amirpour<sup>8</sup> and Sima Sadat Seyedjavadi<sup>9</sup>

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### Reviewed by:

Kiiyukia Matthews Ciira, Mount Kenya University, Kenya Artur J. Sabat, University Medical Center Groningen, Netherlands

#### \*Correspondence:

Mehdi Goudarzi gudarzim@yahoo.com; m.goudarzi@sbmu.ac.ir

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 22 August 2019 Accepted: 23 March 2020 Published: 30 April 2020

#### Citation:

Goudarzi M, Kobayashi N, Dadashi M, Pant ˚ucek R, Nasiri MJ, ˇ Fazeli M, Pouriran R, Goudarzi H, Miri M, Amirpour A and Seyedjavadi SS (2020) Prevalence, Genetic Diversity, and Temporary Shifts of Inducible Clindamycin Resistance Staphylococcus aureus Clones in Tehran, Iran: A Molecular–Epidemiological Analysis From 2013 to 2018. Front. Microbiol. 11:663. doi: 10.3389/fmicb.2020.00663 <sup>1</sup> Department of Microbiology, School of Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran, <sup>2</sup> Department of Hygiene, School of Medicine, Sapporo Medical University, Sapporo, Japan, <sup>3</sup> Department of Microbiology, School of Medicine, Alborz University of Medical Sciences, Karaj, Iran, <sup>4</sup> Department of Experimental Biology, Faculty of Science, Masaryk University, Brno, Czechia, <sup>5</sup> Department of Virology, Pasteur Institute of Iran, Tehran, Iran, <sup>6</sup> School of Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran, <sup>7</sup> Department of Critical Care and Anesthesiology, Imam Hossein Hospital, Shahid Beheshti University of Medical Sciences, Tehran, Iran, <sup>8</sup> Department of Internal Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran, <sup>9</sup> Department of Mycology, Pasteur Institute of Iran, Tehran, Iran

The prevalence of Staphylococcus aureus as an aggressive pathogen resistant to multiple antibiotics causing nosocomial and community-acquired infections is increasing with limited therapeutic options. Macrolide-lincosamide streptogramin B (MLSB) family of antibiotics represents an important alternative therapy for staphylococcal infections. This study was conducted over a period of five years from August 2013 to July 2018 to investigate the prevalence and molecular epidemiology in Iran of inducible resistance in S. aureus. In the current study, 126 inducible methicillin-resistant S. aureus (MRSA) (n = 106) and methicillin-sensitive S. aureus (MSSA) (n = 20) isolates were characterized by in vitro susceptibility analysis, resistance and virulence encoding gene distribution, phenotypic and genotypic analysis of biofilm formation, prophage typing, S. aureus protein A locus (spa) typing, staphylocoagulase (SC) typing, staphylococcal cassette chromosome mec (SCCmec) typing, and multilocus sequence typing. Of the 126 isolates, 76 (60.3%) were classified as hospital onset, and 50 (39.7%) were classified as community onset (CO). Biofilm formation was observed in 97 strains (77%). A total of 14 sequence types (STs), 26 spa types, 7 coagulase types, 9 prophage types, 3 agr types (no agr IV), and 9 clonal complexes (CCs) were identified in this study. The prevalence of the inducible MLSB (iMLSB) S. aureus increased from 7.5% (25/335) to 21.7% (38/175) during the study period. The iMLSB MRSA isolates were distributed in nine CCs, whereas the MSSA isolates were less diverse, which mainly belonged to CC22 (7.95%) and CC30 (7.95%). High-level mupirocin-resistant strains belonged to ST85-SCCmec IV/t008 (n = 4), ST5-SCCmec IV/t002 (n = 4), ST239-SCCmec III/t631 (n = 2), and ST8-SCCmec IV/t064 (n = 2) clones, whereas low-level mupirocin-resistant

strains belonged to ST15-SCCmec IV/t084 (n = 5), ST239-SCCmec III/t860 (n = 3), and ST22-SCCmec IV/t790 (n = 3) clones. All the fusidic acid–resistant iMLSB isolates were MRSA and belonged to ST15-SCCmec IV/t084 (n = 2), ST239-SCCmec III/t030 (n = 2), ST1-SCCmec V/t6811 (n = 1), ST80-SCCmec IV/t044 (n = 1), and ST59-SCCmec IV/t437 (n = 1). The CC22 that was predominant in 2013–2014 (36% of the isolates) had almost disappeared in 2017–2018, being replaced by the CC8, which represented 39.5% of the 2017–2018 isolates. This is the first description of temporal shifts of iMLSB S. aureus isolates in Iran that identifies predominant clones and treatment options for iMLSB S. aureus–related infections.

Keywords: methicillin-resistant S. aureus, methicillin-susceptible S. aureus, inducible resistance, staphylocoagulase, SCCmec, agr allotype, MLST

## INTRODUCTION

Staphylococcus aureus is one of the most common aggressive pathogen that causes many diseases in humans and animals such as skin and soft tissue infections, osteomyelitis, bacteremia, and endocarditis (Gordon and Lowy, 2008). The expression of virulence factors promoting adhesion needs nutrients, and evasion of host immunologic responses including cell surface components (collagen-binding protein, clumping factor, fibronectin-binding protein, and elastin-binding protein), secreted factors (staphylokinase toxic shock syndrome toxin-1), hemolysin, exfoliative toxins (ETA and ETB), staphylococcal enterotoxins (SEs), and lipase and Panton–Valentine leukocidin (PVL), besides the presence of antibiotic resistance genes, turns S. aureus into a very pathogenic microorganism (Gordon and Lowy, 2008; Gould et al., 2012).

In addition to the aforementioned, the biofilm-forming of S. aureus strains can play a key role in pathogenesis and resistance to antimicrobials (Luther et al., 2018). The rate of infections due to S. aureus, especially the antibiotic-resistant strains, has dramatically increased recently, which is becoming a serious problem all over the world (Gould et al., 2012). Emerging simultaneous resistance to multiple antibacterial agents underscores the necessity for therapeutic alternatives for the treatment of bacterium-related infections (Pantosti et al., 2007; Dadashi et al., 2018). Although the use of effective antibiotics such as vancomycin, linezolid, and quinupristin– dalfopristin is considered appropriate for therapy, widespread utilization of these antibiotics has made the current usage of these therapeutic options often unsuccessful (Pantosti et al., 2007; Gould et al., 2012; Wang et al., 2012; Dadashi et al., 2018). In recent years, the use of macrolide–lincosamide– streptogramin group B (MLSB) antibiotics has been favored, which is regarded as an alternative approach to treating such infections (Patel et al., 2006). Clindamycin, a member of MLSB family, serves as one such effective therapeutic alternative for treating S. aureus infections, because of its proven efficacy, safety, convenience of administration (parenteral and oral), and excellent pharmacokinetic properties. However, one important issue in clindamycin administration is the potential emergence of inducible clindamycin resistance, which may increase the risk of clinical failure (Chavez-Bueno et al., 2005; Adhikari et al., 2017).

Recent published data indicate that there has been a concurrent worldwide increase in the prevalence of inducible clindamycin resistance in different areas (Abimanyu et al., 2012; Wang et al., 2012). Evidence of epidemiological researches has revealed that iMLSB S. aureus strains are mostly genetically distinct from each other. One of the most prevalent iMLSB S. aureus strains in the United States belongs to sequence type 8 (ST8) strains, whereas in European countries, the majority of iMLSB S. aureus strains circulating was affiliated with the clonal complexes CC5 and CC8 (Ilczyszyn et al., 2016). Many studies across the world have focused on molecular epidemiology and analysis of iMLSB S. aureus strains isolated from the clinic (Lewis and Jorgensen, 2005; Patel et al., 2006; Adhikari et al., 2017). In our country, data on the prevalence or genetic diversity of inducible clindamycin resistance among S. aureus are very unknown, and there have been no published data on the molecular epidemiological characterization of the inducible clindamycin-resistant S. aureus strains in Iran. Here we provide molecular characterization of iMLSB S. aureus strains. With this aim, phenotypic and genotypic resistance patterns, presence of different classes of prophages, biofilm-forming ability, presence of the icaABCD, adhesion genes, and virulence factors were assessed. Then, multilocus sequence typing (MLST), staphylococcal cassette chromosome mec(SCCmec), agr,coa, and spa typing methods were used to characterize the genotype of the iMLSB S. aureus strains. To the best of our knowledge, this is the first report on the molecular characterization of the inducible clindamycin-resistant S. aureus strains from Iran.

## MATERIALS AND METHODS

## Study Population, Bacterial Isolation, and Inducible Clindamycin Resistance Screening

A total of 1,161 non-duplicated clinical S. aureus isolates were obtained from different clinical specimens including wound, blood, pus, urine, sputum, conjunctivitis, and body fluids from both genders and all age groups of patients, over a period of 5 years from August 2013 to July 2018. This cross-sectional study was conducted in four hospitals affiliated to Shahid Beheshti

University of Medical Sciences (Loghman, Shohada, Taleghani, Emam Hossein). The processing of all samples was done in 2 h. Among these isolates, 126 S. aureus isolates were identified with iMLSB phenotype based on the routine biochemical techniques such as Gram staining; colony morphology comprising shape, size, color, and hemolysis patterns; catalase test; tube coagulase test; growth on mannitol salt agar; and DNase. Polymerase chain reaction (PCR) assay targeting the S. aureus–specific nuc gene was applied to verify the isolates (Goudarzi et al., 2016b). Staphylococcus aureus strains were studied to identify inducible resistance phenotypes according to the Clinical and Laboratory Standard Institute (CLSI) D-zone test. The D-zone test was performed by placing 15 µg erythromycin and 2 µg clindamycin disks at a spaced 15 to 26 mm apart from center to center on the Mueller–Hinton agar (Merck, Darmstadt, Germany) plate inoculated with a 0.5 McFarland-equivalent bacterial suspension. The results also were read at 16 to 18 h incubation at 37◦C and ambient air, using transmitted and reflected light; inducible clindamycin resistance was verified if the clindamycin zone of inhibition adjacent to the erythromycin disk (D-shape) was flattened. A confirmatory microdilution broth test was done on all isolates for further confirmation. Briefly, any growth in the same well that contained 4 µg/mL erythromycin and 0.5 µg/mL clindamycin was set as a positive test and vice versa. Staphylococcus aureus ATCC 25923 was used to perform routine quality control of antibiotic disks. Confirmed isolates were kept into tryptic soy broth (TSB; Merck) with 20% glycerol at −70◦C for molecular testing.

Hospital-onset (HO) S. aureus was set if the positive culture of S. aureus was obtained on or after 96 h of admission to a hospital. Community-onset (CO) S. aureus was set if the culture was obtained prior to 4 days of hospitalization with one or more of the following criteria: (1) a history of hospitalization, surgery, dialysis, or residence in a long-term care facility in 12 months prior to culture date or (2) the presence of a central vascular catheter within 2 days before S. aureus culture. Invasive S. aureus infection was defined according to the Centers for Disease Control and Prevention defining isolation of S. aureus from typically sterile body sites such as the blood, bone, fluids (pericardial, joint/synovial, peritoneal, cerebrospinal, and pleural), internal body sites (brain, lymph node, pancreas, ovary, liver, spleen, heart, and kidney), or other normally sterile sites (Goudarzi et al., 2016a). The Ethics Committee of the Shahid Beheshti University of Medical Sciences in Tehran, Iran, certified the protocol of this project (IR.SBMU. MSP.REC.1396.412).

## Determining Resistance Pattern

Phenotypic methicillin resistance screening was performed by placing the cefoxitin disk (30 µg) on Mueller–Hinton agar (Merck), previously inoculated with a 0.5 McFarland-equivalent bacterial suspension. In vitro susceptibility of the isolates to kanamycin, ciprofloxacin, penicillin, quinupristin–dalfopristin, rifampin, tetracycline, linezolid, teicoplanin, amikacin, tobramycin, gentamicin, and trimethoprim–sulfamethoxazole (Mast; Merseyside, United Kingdom) was done according to the modified Kirby–Bauer disk diffusion method on Mueller– Hinton agar plates, as per CLSI recommendation. The European Committee for Antimicrobial Susceptibility Testing (EUCAST) breakpoint was used for interpreting the results of fusidic acid and ceftriaxone disks. As specified by the CLSI guidelines, the broth microdilution method was applied for determining the minimum inhibitory concentration (MIC) of vancomycin and mupirocin. Minimum inhibitory concentration values of 8 to 256 and ≥512 µg/mL were established to show low-level and high-level mupirocin resistance (LLMUPR, HLMUPR) of the strains, in the respective order. As specified by the CLSI guidelines, MIC breakpoints for vancomycin were set as follows: susceptible, ≤2 µg/mL; intermediate, 4–8 µg/mL; and resistant, ≥16 µg/mL. Fusidic acid MICs were specified by the broth microdilution method and interpreted based on EUCAST guidelines. Minimum inhibitory concentration breakpoints were set as follows: susceptible, ≤1 µg/mL; and resistant, >1 µg/mL. The S. aureus ATCC 25923 and ATCC 29213 strains were put to test as control strains. Sigma Chemical Co. (St. Louis, MO, United States) provided the study with the powders of antibiotics.

## DNA Extraction

To extract DNA, overnight cultures of S. aureus strains on 5% sheep blood agar (Merck) were used by applying the InstaGene Matrix kit (Bio-Rad, Hercules, CA, United States) following the manufacturer's instructions and adding lysostaphin (Sigma-Aldrich; St. Louis, United States) for bacterial lysis. Gel electrophoresis and NanoDrop 2000 spectrophotometer (Thermo, Wilmington, Delaware, United States) were applied, respectively, to test the quality and quantity of isolated S. aureus DNA (Goudarzi et al., 2016a,b). Seemingly, if the purity were appropriated, it would be applied as the template for PCR. Two hundred microliters of elution buffer [10 mM Tris–Cl, 0.5 mM EDTA (pH 9.0)] was applied to elute the extracted DNA, which was then kept at −20◦C until use.

## Detecting Antimicrobial Resistance Determinants and Virulence Factors

Polymerase chain reaction method was used for genotypic amplification of the mecA gene (Goudarzi et al., 2016a,b). Isolates exhibiting phenotypic resistance to particular antimicrobial agents were put to test for the presence of resistance genes [mecA, vanA, vanB, mupB, mupA, fusA, fusB, fusC, erm(A), erm(B), erm(C), tet(M), ant (4<sup>0</sup> )-Ia, aac (6<sup>0</sup> )-Ie/aph (200), aph (3<sup>0</sup> )-IIIa] (Castanheira et al., 2010a; Nezhad et al., 2017). Virulence factors involved SE genes (sea, seb, etb, sec, sed, see, seg, seh, sei, and sej) (Monday and Bohach, 1999), exfoliative toxin genes (eta and etb), Panton–Valentine leukotoxin genes lukS-PV and lukF-PV (pvl), and toxic shock syndrome toxin genes (tst) (Goudarzi et al., 2016a; Nezhad et al., 2017).

## Phenotypic Analysis of Biofilm Formation

Slime production assay or Congo red agar (CRA) method and microtiter plate (MtP) assay were recruited as in vitro methods for the detection of phenotypic biofilm formation. In CRA method, briefly, after preparation of CRA by adding 0.8 g of CR (Sigma-Aldrich; St. Louis, United States) to 1 L of brain heart infusion agar (Merck) and autoclaving it, filters used for

add saccharose (36 g) (Sigma-Aldrich; St. Louis, United States) to CRA. Bacteria were inoculated on CRA and incubated at 37◦C for 24 h and then overnight at room temperature. Biofilm formation was categorized in four levels based on colony color that strains appeared: (i) strong biofilm producer strains (very black colonies), (ii) moderate biofilm producer strains (black colonies), (iii) weak biofilm producer strains (gray colonies), and (iv) biofilm non-producer strains (red colonies) (Arciola et al., 2002; Yousefi et al., 2016).

The MtP assay as a quantitative method for biofilm detection was carried out as described previously. Concisely, an overnight culture of bacterial isolates in TSB (Merck) containing 1% glucose was diluted to 1:100 with fresh medium. Sterile MtP with flat-bottomed 96-well polystyrene was filled with 200 µL of the diluted culture and incubated at 37◦C for 24 h. After incubation, wells were washed three times with 200 µL of phosphate-buffered saline (pH 7.2) to remove planktonic bacteria. Afterward, wells were fixed by 99% methanol, dried at room temperature, and then stained with 0.1% safranin. Safranin dye bound to the adherent cells was dissolved with 1 mL of 95% ethanol per well. As a negative control, 200 µL of TSB–1% glucose was used. The optical density (OD) of each well was measured using an enzyme-linked immunosorbent assay reader at a wavelength of 490 nm. Optical density cutoff defined as average OD of negative control + 3 × standard deviation of the negative control. Biofilm formation of strains was analyzed according to the absorbance of the safranin-stained attached cells and interpreted as per the criteria described by Stepanovic et al. (2007) ´ . Accordingly, the degree of biofilm production was categorized into strong, moderate, weak, or without biofilm. For quality control, Staphylococcus epidermidis ATCC 35984 strain was used in each run.

## Genetic Analysis of Biofilm Formation

Polymerase chain reaction assays for the detection of icaABCD, cna, ebp, fnbB, fnbA, clfB, clfA, and bap genes were performed as described previously (Nemati et al., 2009; Yousefi et al., 2016). Detection of arginine catabolic mobile elements for definitive confirmation of USA300 was performed as previously described by Diep et al. (2008).

## Prophage Typing

iMLSB-positive S. aureus were characterized by prophage typing by using multiplex-PCR assay and oligonucleotide primers SGA, SGB, SGF (SGFa and SGFb), SGD, and SGL prophages as explained previously by Pantu˚cek et al. (2004) ˇ and Rahimi et al. (2012).

## Staphylococcus aureus Protein A Locus (spa) Typing

Staphylococcus aureus isolates underwent spa typing as recommended by Harmsen et al. (2003); all PCR products were Sanger sequenced in both directions. The sequences obtained were edited using the Chromas software (version 1.45; Technelysium, Tewantin, Australia). Ridom SpaServer database<sup>1</sup> was applied in order to assign the spa type.

## Staphylocoagulase Typing

Four sets of multiplex PCR reactions were used for assigning staphylocoagulase (SC) types (I–X) according to the procedure of Hirose et al. (2010). Set A contained primers for identifying SC types I, II, III, IVa, IVb, Va, and VI, whereas set B contained primers for identifying SC types VII, VIII, and X. Set 3 was used for identifying SC types IX and Vb. SC types IVa and IVb were distinguished using a set of four primers.

## SCCmec Typing

The multiplex PCR amplification was done with specific primers for SCCmec typing, as recommended by Boye et al. (2007). The prototype strains applied as control strains for typing were S. aureus ATCC 10442 (type I), S. aureus N315 (type II), S. aureus 85/2082 (type III), S. aureus MW2 (type IV), S. aureus WIS 173 (type V), and S. aureus HDE288 (type VI).

## agr Typing

Multiplex PCR was performed for agr-type detection using primer set comprising a common forward primer (pan-agr) and reverse primers (agr1, agr2, agr3, and agr4) specific to each agr group (Gilot et al., 2002).

## Multilocus Sequence Typing

All the 126 S. aureus isolates with iMLSB phenotype were further characterized by MLST as described by Enright et al. (2000) by sequencing an internal fragment of seven unlinked housekeeping genes to identify the following allelic profiles: phosphate acetyltransferase (pta), carbamate kinase (arcC), triosephosphate isomerase (tpi), shikimate dehydrogenase (aroE), guanylate kinase (gmk), acetyl-coenzyme A acetyltransferase (yqiL), and glycerol kinase (glp). The purification of PCR products was done by performing the Qiagen PCR purification kit. In addition, the two strands were sequenced on the ABI Prism 377 automated sequencer (Applied Biosystems, Perkin-Elmer Co., Foster City, CA, United States). Finally, STs were submitted to the online MLST website through the submission of DNA sequences<sup>2</sup> .

## Ethics Statement

The current study protocol was approved by the Ethics Committee of the Shahid Beheshti University of Medical Sciences in Tehran, Iran (IR.SBMU.MSP.REC.1396.700). Written informed consent was obtained from participants.

## RESULTS

## Isolation and Identification of S. aureus

A total of 126 S. aureus isolates with inducible resistance phenotype were identified, including 106 methicillin-resistant

<sup>1</sup>http://www.spaserver.ridom.de

<sup>2</sup>https://pubmlst.org/

S. aureus (MRSA) and 20 methicillin-sensitive S. aureus (MSSA) isolates representing 84.1 and 15.9% of isolates, respectively. Precisely, of 126 iMLSB S. aureus isolates, 41 were collected from female patients (32.6%), and the rest were collected from male patients (85, 67.4%). In the present study, 43 strains (34.1%) were isolated from wound, 32 (25.4%) from blood, 15 (11.9%) from body fluids, 14 (11.1%) from pus, 12 (9.5%) from urine, 7 (5.6%) from sputum, and 3 (2.4%) from conjunctiva. Of the 126 isolates, 76 (60.3%) were classified as HO, and 50 (39.7%) were classified as CO. According to the case notes, the rates of invasive and non-invasive S. aureus with iMLSB phenotype were found to be 37.3 and 62.7%, respectively. The patients' average age was 38 years. The patients were distributed in four age groups: 21 patients ≤20 years (16.7%), 60 patients between 21 and 45 years (47.6%), 30 patients between 46 and 65 years (23.8%), and 15 patients ≥65 years (11.9%). Regarding the occurrence of inducible resistance in S. aureus strains, data exhibited that most cases belonged to the age groups of 21 to 45 years between 2013 and 2017, whereas in 2018, more than half of the cases were found to be in the age group between 46 and 65 years. Patients with invasive infections and HO were older. Of 21 patients aged ≤20 years, 12 (57.1%) of 60 patients were between 21 and 45 years, 28 (46.7%) of 30 patients were between 46 and 65 years, and 9 (30%) had CO infections. However, among 15 patients aged ≥65 years, only 1 (6.7%) had CO infections. There were also trends toward an increasing incidence of HO infections in elderly patients.

## Antimicrobial Susceptibility

The highest and lowest rates of resistance in 126 S. aureus isolates tested were related to penicillin (91.3%) and fusidic acid (5.6%), respectively. The entire strains were susceptible to teicoplanin, linezolid, and vancomycin. Resistance to tested antibiotics was higher in MRSA than in MSSA. The rate of resistance to the tested antibiotics, with the exception of ciprofloxacin, was higher among HO S. aureus strains across CO S. aureus strains. The frequency of resistance rate among MRSA and MSSA strains to antimicrobial agents is presented in **Table 1**.

The entire S. aureus isolates were susceptible to vancomycin such that 15 isolates (11.9%) had MIC ≥ 0.5 µg/mL, 38 (30.2%) had MIC ≥ 1 µg/mL, and 73 (57.9%) exhibited MIC ≥ 2 µg/mL. The data of the microbroth dilution method illustrated that 23 isolates (18.3%) were mupirocin-resistant; of these, 11 (8.7%) and 12 (9.6%) were LLMUPR and HLMUPR, respectively. Of the 126 isolates of S. aureus with iMLSB phenotype, seven (5.6%) were fusidic acid-resistant. In total, 80.2% of isolates (101/126) were multidrug resistance (MDR), and the MDR rate for MRSA isolates (85.8%) was significantly higher than that for MSSA (50%). Information about simultaneous resistance patterns and distribution of clinical samples is presented in **Table 2**.

## agr and SCCmec Typing

By using agr typing method, all 126 S. aureus isolates with iMLSB phenotype could be typed agr I–III. Of these isolates tested, 77 (10 MSSA and 67 MRSA) were agr I, 20 (MRSA) were agr II, and 29 (10 MSSA and 19 MRSA) were agr III. Staphylococcal cassette chromosome mec typing for MRSA isolates indicated that type IV was the predominant SCCmec type (57.6%, 61/106), followed by SCCmec III (33%, 35/106), SCCmec V (7.5%, 8/106), and SCCmecII (1.9%, 2/106). In addition, no MRSA isolates harbored SCCmec type I.

## Determination of Adhesion and Biofilm Production Ability

Biofilm formation was observed in 97 strains (77%) using the CRA and MtP methods, whereas 29 strains (23%) were confirmed as non–biofilm producer strains. Of the total MRSA isolates, MtP detected 18 (17%) as weak, 23 (21.7%) as moderate, and 42 (39.6%) as strong biofilm producers. By CRA, 20 isolates (18.9%) were able to produce biofilm weakly, and 26 (24.5%) and 37 (34.9%) isolates presented moderate and potent biofilm formation, respectively. In MSSA isolates, MtP indicated weak biofilm production in five isolates (25), moderate biofilm production in three (15%), and strong biofilm production in six isolates (30%), whereas CRA detected four isolates (20%) as weak, five (25%) as moderate, and five (25%) as strong biofilm producers. The analysis of icaA-D genes among tested S. aureus strains showed that the most frequently detected gene was icaD (n = 92, 73%), followed by icaA (n = 90, 71.4%), icaB (n = 76, 60.3%), and icaC (n = 67, 53.2%). Overall, six different biofilm patterns were identified. **Figure 1** presents the distribution of icaABCD among tested strains. Regarding the presence of adhesion genes, the seven most frequent adhesionrelated genes were clfA (93.7%, 118/126) followed by clfB (87.3%, 110/126), fnbB (61.9%, 78/126), cna (57.9%, 73/126), fnbA (54%, 68/126), ebp (50%, 63/126), and bap (3.2%, 8/126), respectively. The adhesion profiles are shown in **Figure 1**. None of the MSSA strains carried the bap gene. In MSSA strains, three different adhesion patterns including clfA + clfB + fnbB (50%, 10/20), clfA + clfB + fnbB + fnbA + ebp + cna (25%, 5/20), and fnbB + fnbA + cna (25%, 5/20) were observed. Apparently, higher diversity among the adhesion patterns from MRSA (six patterns) than those from MSSA (three patterns) was noticed.

## Determination of Prophage Types and Toxin and Enterotoxin Encoding Genes

A total of 103 strains (81.7%) of the 126 inducible resistant S. aureus strains tested had one or more virulence genes. The most frequently attained virulence genes were pvl (23.8%, 30/126), followed by sea (19.8%, 25/126), tst (19%, 24/126), sec (11.1%, 14/126), sed (9.5%, 12/126), seg (4.8%, 6/126), seb (4%, 5/126), see (3.2%, 4/126), and eta (2.4%, 3/126), but she, sei, and sej encoding genes were detected in any of the total number of 126 isolates tested. In this analysis, six prophage types were grouped into seven separate prophage profiles. **Figure 1** provides a summary data of the number and frequency of prophage profiles among isolates tested.

The data for antibiotic resistance genes in S. aureus strains studied revealed that the most common gene was mecA in 106 strains (84.1%), followed by erm(A) in 43 strains (34.1%), erm(B) in 41 (32.5%), erm(C) in 361 (28.6%), ant (4<sup>0</sup> )-Ia in 35 (27.8%), tet(M) in 32 (25.4%), aac (6<sup>0</sup> )-Ie/aph (200) in 29 (23%), aph (3<sup>0</sup> )- IIIa in 17 (13.5%), mupA in 12 (9.5%), fusB in 6 (4.8%) strains,

TABLE 1 | Antimicrobial resistance pattern of MRSA and MSSA isolates with iMLSB phenotype.


TABLE 2 | Resistant pattern and distribution of samples in 126 S. aureus strains isolated from clinical sources.


<sup>a</sup>PG, penicillin; CRO, ceftriaxone; GM, gentamicin; K, kanamycin; AK, amikacin; TN, tobramycin; T, Tetracycline; CIP, ciprofloxacin; TS, trimethoprim–sulfamethoxazole; RI, rifampicin; MUP, mupirocin; FC, fusidic acid; SYN, quinupristin–dalfopristin. <sup>b</sup>B, blood; W, wound; P, pus; BF, body fluid; U, urine; S, sputum; C, conjunctiva.

and fusC in 1 strain. vanB, mupB, and fusA genes were not detected in none of experimented isolates.

## SC Typing

In the present study, all isolates could be typed by SC typing method. These isolates were distinguished into seven SC types (I–VI). The predominant SC type was III and included 56 isolates (44.4%), followed by II (28.6%, 36/126), IVa (9.5%, 12/126),

## VI (7.9%, 10/126), I (4%, 5/126), V (4%, 5/126), and IVb (1.6%, 2/126). A total of seven SC types and two types (II and IVa) were detected in MSSA isolates.

## MLST and spa Typing

All the isolates were typed by spa and MLST. A total of 26 spa types were attained in the entire 126 S. aureus. The isolates identified as invasive S. aureus were assigned to particular t790,

t002, t008, t064, t030, t037, and t631 spa types. For CO, the isolates were assigned to spa types t005, t1869, t019, t021, t10795, t6811, t002, t045, t008, t065, t030, and t037. Among MSSA isolates, the most common spa types were t021, t005, t1869, and t318 representing 35 (7/20), 25 (5/20), 25 (5/20), and 15% (3/20), and among MRSA isolates, the top four spa types were t037, t437, t030, and t008 representing 10.4 (11/106), 9.4 (10/106), 8.5 (9/106), and 7.5% (8/106) of isolates. Multilocus sequence typing identified 14 ST types in both MRSA and MSSA isolates, namely, ST22 (14.3%, 18/126), ST30 (11.9%, 15/126), ST1 (1.6%, 2/126), ST772 (4.8%, 6/126), ST8 (10.3%, 13/126), ST239 (23%, 29/126), ST585 (1.6%, 2/126), ST5 (5.63%, 7/126), ST225 (1.6%, 2/126), ST80 (9.5%, 12/126), ST59 (4.8%, 6/126), ST338 (3.2%, 4/126), ST15 (4%, 5/126), and ST45 (4%, 5/126), which were further grouped into nine CCs including CC1, 5, 8, 15, 22, 30, 45, 59, and 80. Among MSSA, CC22 and CC30 were the dominant clones, each clone with 50% (10/20) of the isolates being assigned to these genotypes, and all were recovered from non-invasive S. aureus and both CO (55%, 11/20) and HO (45%, 9/20) cases. CC8 (41.5%, 44/106), CC80 (11.3%, 124/106), and CC5 (8.5%, 9/106) were the top three CCs in MRSA isolates. **Table 3** represents the analysis of S. aureus clones obtained from patients based on invasive and non-invasive infections.

## Temporal Changes in Inducible Resistant S. aureus Genotypes

**Table 4** gives information about the percentage of inducible resistant S. aureus strains assigned to each CC for each year between 2013 and 2018. The prevalence of inducible resistance S. aureus isolates elevated in the experiment from 19.8% in 2013–2014 to 30.2% in 2017–2018. Apart from 2013 to 2014 in which CC22 genotype was the most prevalent CC among inducible resistance isolates (36%, 9/25), in the remaining years of study (2014–2018), the CC8 isolates were the most predominant genotype. The CC8, as a predominate genotype, was at its most prevalent in 2016–2017 (41.6%, 10/24). Its prevalence increased during the study period from 20% in 2013–2014 (5/25) to 39.5% (15/38) in 2017–2018. By contrast, the prevalence rate of CC22 decreased from 36% in 2013– 2014 to 0% in 2017–2018. The CC30, the third predominated CC among our clinical isolates, had its most prevalence in 2017–2018 (18.4%, 7/38). The prevalence of this type varied dramatically each year during 2013–2018; despite a decrease to 8% (2/25) in 2013–2014, a slight increase in the prevalence of this type was noted during 2014–2015 (9.5%, 2/21) and 2015– 2016 (16.7%, 3/18), followed by an overall decrease to 4.2% (1/24) in 2016–2017. The prevalence of this type increased to 18.4% (7/38) in 2017–2018.

The first CC45 isolates were found in 2015–2016. The prevalence of this genotype diminished during the study period from 22.2% (4/18) in 2015–2016 to 4.2% (1/24) in 2016–2017 apart from 2017–2018 when no CC45 was detected. The CC45 was the second predominant genotype in 2015–2016 (22.2%, 4/18). The CC15, absent during 2013–2016, was detected during 2016–2017 (8.3%, 2/24), but remained at relatively low prevalence (7.9%, 3/38) in 2017–2018.

TABLE 3 | Analysis of S. aureus clones obtained from patients based on invasive and non-invasive infections.


TABLE 4 | Distribution of characterized S. aureus isolates with iMLSB phenotype by year.


The CC59 isolates were not detected between 2013–2014 and 2015–2016. The CC59 isolates were detected for the first time in 2014–2015 (4.8%, 1/21), and increased to 8.3% (2/24) in 2016– 2017. This genotype was present at its greatest prevalence in 2017–2018, accounting for 18.4% (7/38) of inducible resistance strains. The prevalence of CC80 varied dramatically during the 5-year period, this genotype was not identified among any of the 2014–2015 and 2015–2016 tested strains. In 2013–2014, only 12% (3/25) of isolates exhibited this genotype. An increase in the prevalence of this genotype was noted between 2016 and 2017 to 16.7% (4/24), followed by a decline to 13.2% (5/38) in 2017–2018. The CC5 genotype was present in each year between 2013 and 2018. The highest prevalence of this genotype occurred in 2014– 2015 (14.3%, 3/21), followed by 11.1% (2/18) between 2015 and

2016. Overall, the prevalence of this genotype diminished during the study period from 8% (2/25) in 2013–2014 to 2.6% (1/38) in 2017–2018. The CC1 isolates were not identified between 2016 and 2018. It was the third predominant genotype in 2013–2014, accounting for 16% (4/25) of examined isolates, but it dropped to 9.5% (2/21) in 2014–2015 and then indicated an increase in 2015–2016 (11.1%, 2/18).

## Observations on Inducible Clindamycin-Resistant S. aureus Clones

The molecular characteristics of the isolates related to each genotype of inducible resistance MRSA and MSSA are shown in **Table 5**. Meanwhile, the details of each clonal lineage are explained below.

## CC1

In this study, most of the isolates identified as CC1/ST772- MRSA-V were represented by spa type t10795 (83.3%, 5/6). Among the CC1/ST772-MRSA-V isolates, resistance to aminoglycosides encoded by aac (6<sup>0</sup> )-Ie/aph (200) and aph (30 )-IIIa, and tetracycline encoded by tet(M) was common. Three isolates (60%) were able to produce biofilm strongly. The finding indicated that fnbB gene was present in all of the CC1 isolates. icaA and icaD were the most prevalent biofilm-related genes. More than half of the isolates (60%) carried pvl genes. The one remaining CC1/ST772-MRSA-V isolate was assigned to spa type t657 and coa type IVa. This isolate exhibited resistance to multiple antibiotics and harbored resistance genes mecA and erm(C); meanwhile, enterotoxin genes sea + seb carried simultaneously. The CC1/ST1-MRSA-V isolates exhibited spa type, t6811. One isolate carried fusc and exhibited resistance to fusidic acid at MIC 32 µg/mL. MDR pattern was detected among these isolates. Both isolates were identified to be biofilm producers albeit weakly. Toxin gene differs from those detected in CC1/ST772-MRSA-V found among the CC1/ST1-MRSA-V isolates, namely, sec. All the isolates exhibited erythromycin resistance encoded by erm(C).

## CC5

All CC5/ST5-MRSA-IV isolates were assigned to single spa type t002. This isolates carried the enterotoxin genes sea (42.9%, 3/7) and sed + seg (42.9%, 3/7) and the toxin tst (42.9%, 3/7). Among the CC5/ST5-MRSA-IV isolates, resistance to erythromycin encoded by erm(A), tetracycline encoded by tet(M), and mupirocin encoded by mupA was common. Five of HLMUPR strains, which harbored the mupA gene, belonged to this CC (41.7%, 5/12). Nearly 60% of all isolates that occurred in these clones showed moderate biofilm formation. The remaining 40% were not able to produce biofilm. The top three adhesion-related genes among CC5 isolates were clfA (100%, 9/9), clfB (77.8%, 7/9), ebp (75%, 6/9), and cna (75%, 6/9). Of nine CC5 isolates, icaA and icaD were found in seven isolates, whereas icaB and icaC were detected in six isolates. The other CC5 genotypes identified (CC5/ST225-MRSA-II) were assigned to spa type t045, which exhibited resistance to multiple antibiotics including aminoglycosides resistance encoded by ant (40 )-Ia (2) and aac (6<sup>0</sup> )-Ie/aph (200) genes. tst encoding gene was detected in all isolates. All CC5/ST225-MRSA-II showed strong biofilm formation. The predominant prophage profile was SGF + SGFa + SGFb.

## CC8

In this study, CC8 was represented by ST8-MRSA-IV (29.6%, 13/44), ST239-MRSA-III (65.9%, 29/44), and ST585-MRSA-III (4.5%, 4/44). spa types t008 (61.5%, 8/13) and t064 (38.5%, 5/13) predominated among the CC8/ST8-MRSA-IV isolates. Among 13 ST8-MRSA-IV strains, 8 isolates harbored the arcA and opp3AB genes and confirmed as USA300. The majority of CC/ST8-MRSA-IV (USA300), as a major global epidemic clone that has been noticed for its rapidity dissemination within the community and hospitals isolates, exhibited resistance to aminoglycosides encoded by aph (3<sup>0</sup> )-IIIa (61.5%, 8/13). Resistance to erythromycin encoded by erm(C) (38.5%, 5/13) and tetracycline encoded by tet(M) (38.5%, 5/13) was also detected in this genotype. Six strains belonging to CC/ST8-MRSA-IV clone demonstrated the HLMUPR phenotype and carried mupA gene. The findings showed that the majority of the isolates (77%, 10/13) was able to produce biofilm, although at different intensities. Nine isolates in this clone harbored the pvl genes (69.2%); the rest of four isolates carried the sed gene (30.8%).

The majority of CC8/ST239-MRSA-III isolates were assigned to spa type t037 (37.9%, 11/29), followed by t030 (31.1%, 9/29), t631 (13.8%, 4/29), t860 (10.3%, 3/29), and t388 (6.9%, 2/29). Resistance to multiple antibiotics simultaneously was identified in this genotype. Fusidic acid resistance encoded by fusc (6.9%, 2/29); tetracycline resistance encoded by tet(K) (41.4%, 12/29); aminoglycoside resistance encoded by ant (4<sup>0</sup> )-Ia (58.6%, 17/29), aac (6<sup>0</sup> )-Ie/aph (200) (37.9%, 11/29), and aph (3<sup>0</sup> )-IIIa (6.9%, 2/29); and mupirocin resistance encoded by mupA (6.9%, 2/29) were also noted. Almost more than half of the isolates carried sea (55.2%, 16/29), 48.3% (14/29) carried tst, and 17.2% (5/29) carried sec only three isolates harboring seg gene. All three ST239-SCCmec III/t860 isolates carried eta and demonstrated the LLMUPR phenotype. The CC8/ST239-MRSA-III isolates displayed high levels of biofilm production (86.2%, 25/29). Regarding the biofilm-related genes, the icaD (70.5%, 31/44) was common, followed by icaA (68.2%, 30/44), icaB (68.2%, 30/44), and icaC (61.4%, 27/44). The top three adhesion-related genes among CC8 isolates were clfA (100%, 44/44), clfB (86.4%, 38/44), and cna (86.4%, 38/44).

Two remaining CC8/ST585-MRSA-III isolates were assigned to spa type t713 and carried multiple resistance genes, that is, tet(M), erm(C), and enterotoxin genes see. Both of these strains indicated strong biofilm formation. The majority of the CC8 isolates was found to harbor SGF + SGFa + SGFb proghages.

## CC15

The five CC15/ST15 MRSA-IV isolates identified revealed the same spa, coa, agr, and prophage profile. Fewer than half of the isolates were resistant to fusidic acid encoded by fusB. All CC15/ST15 MRSA-IV isolates carried pvl genes. More than half of the isolates were biofilm producers. All CC15 isolates were positive for fnbB gene, and 80% of the isolates were found to carry all of the biofilm-related genes (icaA-D). One isolate was found

April 2020 | Volume 11 | Article 663

TABLE 5 |Phenotypicand genotypic typing data for inducible resistance MRSA (n = 106) and MSSA (n= 20) isolates studied in Tehran, Iran, between 2013 and 2018.


aPhenotypic profile are presentedinTable 2. <sup>b</sup>W, weak producer; M, moderate producer; S, strong producer, N, non-producer. cProphage profiles are presented inFigure 1A.

fmicb-11-00663 April 29, 2020 Time: 18:27 # 10

to carry none of the tested ica genes. All of the CC15 isolates possessed SGA + SGF + SGFa + SGFb + SGL prophage profile.

## CC22

Three spa types t852, t790, and t005 and two spa types t005, t1869 predominated among the CC22/ST22-MRSA-IV isolates (44.4%, 8/18) and CC22/ST22-MSSA (55.6%, 10/18) isolates. Among the MRSA isolates, resistance to erythromycin encoded by erm(C) and aminoglycosides encoded by aac (6<sup>0</sup> )-Ie/aph (200) were prevalent, whereas in MSSA isolates resistance to tetracycline encoded by tet(M) and that to aminoglycosides encoded by ant (40 )-Ia (3) were common. All the MRSA isolates were biofilm producers, whereas in MSSA isolates half of the isolates were able to produce biofilm. The results indicated that adhesion genes clfA and clfB were detected in all of the MRSA strains, and the most common ica genes were icaA and icaB. All the MSSA isolates carried clfA, clfB, and fnbB simultaneously, and icaA and icaD were the most prevalent biofilm-related genes. The isolates identified as CC22/ST22-MRSA-IV isolates showed a prevalence of 37.5% (3/8) for the pvl and tst genes. One isolate carried sea gene (12.5%, 1/8). The sec gene (30%, 3/10) was the only toxin gene detected among CC22/ST22-MSSA isolates. The carriage of pvl genes was more common among MRSA isolates than among MSSA isolates. The SGB + SGF + SGFa + SGFb prophage profile was detected in the majority of CC22 isolates.

## CC30

CC/ST30 isolates were found in both MRSA and MSSA isolates. Among MSRA isolates, the most predominant spa types were t605 and t019, representing 40% (2/5) and 60% (3/5) of isolates. Multiple resistance genes including erm(C), aac (6<sup>0</sup> )-Ie/aph (200), aph (3<sup>0</sup> )-IIIa, and tet(M) were identified among tested isolates. More than half of the CC/ST30-MRSA-IV isolates carried pvl, whereas isolates harboring sea were at a relatively low frequency. All CC/ST30-MRSA-IV isolates displayed agr type III and coa type II. Four isolates (80%) were able to produce biofilm, and the remaining 20% were non–biofilm producers. All the MRSA isolates carried biofilm (icaA, icaB, icaC, and icaD) and adhesion (clfA, clfB, and fnbB) related genes. In all the MSSA strains, fnbA, cna, and icaA genes were detected. Among MSSA isolates, spa types t021 and t318 represented 70% (7/10) and 30% (3/10) of isolates. Nearly 90% (9/10) of the CC/ST30- MSSA isolates could form biofilm at different intensities. Various prophage patterns with the majority of SGB (60%, 6/10) were identified among MSSA isolates. The majority of CC/ST30- MSSA isolates exhibited erythromycin resistance encoded by erm(B) (80%, 8/10), and erm(C) (50%, 5/10). The pvl and sed encoding genes were the only toxins produced by this genotype. The SGA + SGB + SGF + SGFa + SGFb + SGL was found as the most predominant prophage profile among CC/ST30-MRSA-IV isolates.

## CC45

All ST45-MRSA-IV isolates were assigned a spa type t038, agr I, and coa V. The isolates that exhibited resistance to multiple antibiotics were confirmed among these isolates. More than half of them (60%) were biofilm producers. All of these isolates carried icaB, clfA, clfB, and fnbB encoding genes simultaneously. All ST45-MRSA-IV isolates possessed erm(A) gene. The carriage of aac (6<sup>0</sup> )-Ie/aph (200) gene was found in 60% of isolates. The enterotoxin gene sec was present in 80% of isolates. Among prophage patterns, the SGF + SGFa + SGFb was common.

## CC59

Two CC5 MRSA isolates were detected as one with ST59 and another one with ST338. All CC59 isolates indicated a single spa type, t437, coa type VI, and SGB-SGF-SGFa-SGFb phage pattern. All CC59/ST59 isolates were discriminated into SCCmec types IV. Two-thirds of ST59-MRSA-IV isolates carried seb gene; however, no other toxin genes were identified. Three isolates accounting for half (50%) of the entire ST59-MRSA-IV isolates were non–biofilm producers, and only one isolate showed strong biofilm formation. Results also demonstrated that the icaA, icaD, clfA, and clfB genes were detected in all of the CC59 isolates. Almost one-fifth of these isolates were resistant to fusidic acid encoded by fusB gene. Four and two ST59-MRSA-IV isolates carried erm(C) (66.7%) and erm(B) (33.3%), respectively. All CC59/ST338 isolates were assigned into an SCCmec type III. Half of CC59/ST338-MRSA-III isolates harbored see encoding gene. Other toxin-encoding genes were not noticed. The predominant isolates carried erm(B) (75%, 3/4), and aminoglycoside resistance encoded by aph (3<sup>0</sup> )-IIIa was also common among these isolates (50%, 2/4). All of the CC59 isolates possessed SGB + SGF + SGFa + SGFb prophage profile.

## CC80

The CC80 isolates exhibited different spa types including t044 (66.7%, 8/12), t131 (33.3%, 4/12). All the ST80-MRSA-IV isolates assigned agr III and coa II. One-fourth of isolates were categorized as non–biofilm producers. The results exhibited that the clfA and clfB genes were explored in the entire CC80 isolates. The gene encoding for icaD (91.7%, 11/12) was the most prevalent biofilm-related gene among isolates.

The carriage of pvl genes was confirmed in one-third of tested isolates. Resistances to aminoglycosides encoded by ant (40 )-Ia (41.7) and aac (6<sup>0</sup> )-Ie/aph (200) (33.3), erythromycin encoded by erm(A), and fusidic acid encoded by fusB were common. The most prevalent prophage pattern was SGA + SGB + SGF + SGFa + SGFb + SGL.

## DISCUSSION

This cross-sectional study provided several novel findings regarding the prevalence and genetic diversity of iMLSBpositive S. aureus strains isolated from clinical samples. It was observed that there was approximately a threefold increase in the prevalence of iMLSB S. aureus strains, from 7.5 to 21.7%, between 2013 and 2018. In contrast, Chavez-Bueno et al. (2005) suggested a significant decline in the trend for the prevalence of iMLSB phenotype from 93% in 1999 to 7% in 2002. Although our results markedly exhibited an increase in the prevalence of the inducible resistance among clinical S. aureus isolates in Iran, enhanced clinical and laboratory awareness of inducible resistance may

also be responsible for this rise. Molecular epidemiology and prevalence of inducible resistance among S. aureus strains change geographically and dynamically (Monecke et al., 2011; Boswihi et al., 2016). In this experiment, the prevalence of iMLSB among S. aureus was found to be 10.9% with 84.1% MRSA and 15.9% MSSA isolates exhibiting iMLSB. This prevalence was lower than the rates reported from Nepal (21%) (Adhikari et al., 2017) and Jordan (76.7%) (Aqel et al., 2017), whereas it was higher than Brazil (7.9%) (Bottega et al., 2014). However, variable rates of iMLSB among S. aureus isolates have been reported in previously published data from Iran ranging from 4.1 to 20.7% (Memariani et al., 2009; Khashei et al., 2018). Regarding the increasing incidence of inducible resistance, our study suggests the necessity to revise the prescription of macrolides, which could cause a decline in resistance patterns. The investigated population and the prevalence of different clones in across various regions of the world could likely explain the striking differences in inducible clindamycin resistance.

In the present experiment, 18.3% of isolates tested were found to be resistant to mupirocin. This prevalence is higher than that reported rate in Jordan (5%) (Aqel et al., 2012) and other reports in France (2.2%) (Desroches et al., 2013) and Greece (1.6%) (Petinaki et al., 2004). A research performed by Abbasi-Montazeri et al. (2013) in Iran notified that 6% of S. aureus strains obtained from burn patients were detected as mupirocinresistant. A high prevalence of resistance to mupirocin in this experiment indicates that there are unrestricted policies in the use of mupirocin for long periods and/or increasing trends of mupirocin prescription in our setup. Furthermore, we observed that 9.6 and 8.7% isolates had HLMUPR and LLMUPR phenotypes, respectively, which are higher than the rate stated in France (0.8%) (Desroches et al., 2013) and Korea (5%) (Yun et al., 2003). Different prevalence rates of HLMUPR S. aureus were reported from other researchers in Iran, 25% by Shahsavan et al. (2012) and 17% by Abbasi-Montazeri et al. (2013). We found that 9.5% of the examined isolates carried mupA gene, and all of them exhibited an HLMUPR phenotype. Researches from Spain (González-Domínguez et al., 2016) (27.2%) and Iran (Abbasi-Montazeri et al., 2013) (34%) revealed a higher rate of this gene among their S. aureus strains. Contrary to the study of Shahsavan et al. (2012), reporting mupA gene only in MRSA strains with cMLSB phenotype, the present data showed the existence of mupA gene in inducible resistance MRSA strains.

This study indicated that of 126 inducible clindamycinresistant S. aureus isolates, 5.6% were fusidic acid–resistant, which was similar to those in Canada (7%) and Australia (7%); nonetheless, it was far lower than the values previously reported in Greece (62.4%) and Ireland (19.9%) (Castanheira et al., 2010a,b). In a multicenter experiment performed from 2007 to 2011 in three referral hospitals in Tehran, Iran, of 726 tested S. aureus isolates, 3% of isolates were found to be resistant to fusidic acid (Deotale et al., 2010). In our collections, we found that fusB and fusC were present in six strains and one strain, respectively. A previously studied experiment from China noted a higher rate of fusB gene (10.5%); on the other hand, fusC and fusA genes were not present in any of tested isolates (Yu et al., 2015). These data highlighted that fusB is the predominant determinant responsible for resistance to fusidic acid among S. aureus in Iran. According to previously published data, different fusidic acid resistance rates were reported in both MRSA and MSSA strains. Notably, this study showed that fusidic acid resistance was only seen among MRSA isolates, which were in line with a report of China indicating that the prevalence of fusidic acid resistance among MRSA isolates was significantly higher than that among MSSA isolates (Yu et al., 2015).

According to the evidence, the ability to produce biofilm among S. aureus is diverse, with data ranging within 43 to 88% (Luther et al., 2018). Our results showed that, of the 126 isolates under investigation, 77% of strains could produce biofilm, whereas 23% were confirmed as non–biofilm producer strains. This finding is similar to those in Egypt (83.3%) (Gad et al., 2009), whereas based on a study of Iran biofilm production reported rate was 38.7% (a twofold decrease) (Mirzaee et al., 2014). In this study, the biofilm formability of MRSA isolates (78.3%) was higher than that of MSSA isolates (70%). Apparently, high capability of biofilm formation among MRSA strains was described by conducted studies from India (57.6%) (Mathur et al., 2006), China (66%) (Wang et al., 2010), and South Africa (37.8%) (Samie and Shivambu, 2011).

Data obtained from different studies suggest that the ability of biofilm formation in S. aureus strains is attributed to the expression of a wide range of virulence and adhesion factors of this bacterium (Nemati et al., 2009; Samie and Shivambu, 2011; Luther et al., 2018), but the correlation between the existence of specific virulence factor and biofilm production has been controversial. Researches has shown that ica ABCD and adhesion genes can affect the biofilm formation ability in S. aureus. In this study, the most frequent ica genes were icaD (73%), followed by icaA (71.4%), icaB (60.3%), and icaC (53.2%). According to an analysis by Mirzaee et al. (2014) on 31 clinical S. aureus isolates, icaD, icaA, and icaC genes with the exception of icaB were found in more than half of the isolates tested. In the study of Yousefi et al. (2016), of 39 isolates of S. aureus recovered from UTI patients, 69.2% were biofilm producers, and it was notable that all isolates carried icaA, fnbA, and clfA genes. In the current study, a high percentage of icaABCD genes made up 39.7% of the overall sample, which was close to a recent report in Iran (38.7%) (Mirzaee et al., 2014). The correlation between biofilm formation and the presence of adhesion genes is well established. In present survey, the most prevalent gene wasclfA (93.7%), followed by clfB (87.3), fnbB (61.9%), cna (57.9%), fnbA (54%), ebp (50%), and bap (3.2%), respectively. The attained data are consistent with a previous finding from Iran, which has displayed the role of the bap gene in biofilm production rarely. Our findings indicated that isolates with more adhesion and ica encoding genes were strictly associated with biofilm formation. Overall, our data confirmed the high ability of biofilm formation among inducible resistance S. aureus strains, which helps S. aureus to persist in infections. This scenario draws attention from clinicians to use treatment protocols in patients potentially infected with these bacteria.

Based on the results of coa typing, the top three coa types were III (44.4%), II (28.6%), and IVa (9.5%). This result was in contrast with the previous report by Hirose et al. (2010) in Japan, which indicated that coa types II, VII, and I accounted for 91.9%, 3.9%,

and 1.7% of isolates. We detected seven SC types (I–VI) among iMLSB S. aureus strains suggesting diverse genetic backgrounds of tested isolates in this region of Iran. In a research involving 157 S. aureus strains from clinical specimens, nine different patterns of coa gene were detected (Afrough et al., 2013). These findings were confirmed by results in Thailand (Janwithayanuchit et al., 2006) and Egypt (Omar et al., 2014) reported previously. Genetic variability in coa gene among studied isolates indicated that it could not be a predictor for specific iMLSB S. aureus strains.

In the present study, the findings revealed that seven different prophage profiles were identified in the isolates studied. A high diversity of prophage patterns among S. aureus isolates has been reported from the United States (Workman et al., 2006), Czech Republic (Pantu˚cek et al., 2004 ˇ ) and Iran (Rahimi et al., 2012). Our data exhibited that SGFa-SGFb (36.5%) and SGB-SGFa-SGFb (31.7%) were the major prophage profiles, which was in accordance with the study of Rahimi et al. (2012). SGFtype prophages are associated with the immune evasion cluster typical for S. aureus isolated from humans (Kahánková et al., 2010). SGA has been reported to be common among pvl-positive MRSA-IV isolates. This finding was consistent with previous research, revealing that all the pvl-positive isolates harbored SGA prophage type (Rahimi et al., 2012). Notably, the PVL was the most frequently encoded toxin in the tested isolates (23.8%). Previously published data from England and the United States (Holmes et al., 2005) indicated that the prevalence of pvl encoding genes among S. aureus was low (1.6%). Nevertheless, previous reports all over the world had shown a high prevalence of pvl among S. aureus strains (Monecke et al., 2011; Boswihi et al., 2016; Goudarzi et al., 2016a). In Iran, the pvl-positive rate ranged from 7.4 to 55.6% in S. aureus isolates. In our research, a relatively low but increasing prevalence of pvl-positive S. aureus strains was noted. According to our report, a recent study from Ireland showed an ascending trend of pvl among MRSA strains (from 0.2 to 8.8%), whereas this trend was diminishing for MSSA strains (20–2.5%) (Shore et al., 2014). Although this difference can, to some extent, reflect the origin of isolates and the type of sample, this could also be because of different pvl-encoding phages among S. aureus strains.

Considering the genes encoding enterotoxins, the data analysis of the current study exhibited that 54.8% of isolates carried one or more SE genes, with the sea (19.8%) and sec (11.1%) genes being the most commonly found. The attained data are consistent with the reports previously obtained from Iran and Turkey estimating enterotoxigenic S. aureus strains in 45 and 62.6% of isolates tested, respectively (Imanifooladi et al., 2007; Aydin et al., 2011). Based on earlier studies, the she, sei, and sej encoding genes were rarely present in S. aureus isolated from clinical samples, and similarly, these genes were not detected in the present survey either.

The present study displayed diversity in the numbers and the molecular types of iMLSB S. aureus clones identified in our health care settings. According to the current results, CC8 was the most prevalent clone in both CO and HO S. aureus isolates. The prevalence of CC8 was increased from 2013– 2014 (20%) to 2017–2018 (39.5%). These strains were found in association with pvl-encoding bacteriophage and HLMUPR, which carried mupA gene, which is similar to those reported in Iran (Azimian et al., 2014), Kuwait (Boswihi et al., 2016), and Ireland (Shore et al., 2014). According to evidence, phenotypic and genotypic resistance pattern in the ST8-IV isolates was found to be varied. In accordance with our research, high resistance to aminoglycosides and low resistance to tetracycline in ST8-MRSA isolates were described previously by researchers (Shore et al., 2014; Boswihi et al., 2016). The enterotoxin gene sed was the solo toxin gene detected among the ST8-IV isolates. Variability in the enterotoxin genes within ST8-IV isolates was reported by several investigators (Shore et al., 2014; Boswihi et al., 2016). The detection of the ST8-IV isolates in our study is of concern and highlights transmissibility and rapidity of its spread as an international epidemic MRSA clone in Iran.

ST239-MRSA-III, one of the most successful and persistent clones, was the most predominant genotype in CC8. Multiresistant ST239 clone was previously reported in South American, European, and some Asian countries, including Kuwait and Saudi Arabia (Shore et al., 2014; Boswihi et al., 2016). In this study, ST239 had five important spa types (t030, t037, t631, t860, and 388). The data in this experiment indicated that half of ST239-SCCmec III/t631 strains exhibited HLMUPR phenotype and harbored mupA gene. These findings are in parallel with previous reports from Iran (Goudarzi et al., 2016a) and those reported in India and Kuwait (Abimanyu et al., 2012; Boswihi et al., 2016). All ST239-SCCmec III/t860 strains exhibited the LLMUPR phenotype, which was similar to findings in other countries including China and Kuwait (Boswihi et al., 2016; Liang et al., 2018). According to the molecular typing results, ST239-SCCmec III/030 with 20.5% and ST239-SCCmec III/037 with 25% were recognized as the most common multiresistant ST239- MRSA-III type. These results were consistent with previous research performed by Li et al. (2018), who demonstrated multiresistant ST239-SCCmec III/030 as the most predominant genotype in China from 2005 to 2013, which decreased significantly in 2016. During a 12-year period in Norway, Fossum and Bukholm (2006) indicated ST239 as the most prevalent MRSA clones, which became endemic in hospital environments in Norway. However, molecular characteristics, especially antibiotic resistance gene and virulence factors, of our ST239-MRSA-III strains were similar to those of ST239 strains reported in Kuwait, China, and Saudi Arabia (Lewis and Jorgensen, 2005; Boswihi et al., 2016). The emergence of ST239-MRSA-III may be due to the import of this clone from neighboring countries.

Increasing prevalence rate of CC8 and disappearance of CC22 between 2013 and 2018 highlighted replacement and the changing clonal structure of MRSA in this region of Iran. As summarized in **Table 5**, pvl encoding genes were detected in two isolates of ST22-SCCmec IV/t852 and one isolate of ST22-SCCmec IV/t005. These pvl-positive genotypes were mainly distributed at some geographic area including England, Australia, Ireland, Kuwait, Iran, Germany, Saudi Arabia, and Nepal (Monecke et al., 2011; Shore et al., 2014; Boswihi et al., 2016). Of note, the majority of the ST22-SCCmec IV/t790 isolates was related to the tst gene, demonstrated LLMUPR phenotype, and could exhibit multiresistance. A recent report in Ireland

also displayed that most of pvl-positive ST22-MRSA-IV isolates were associated with resistance to gentamicin, trimethoprim, and ciprofloxacin and carried the erm(C), lnu(A), aacA-aphD, aadD, and mupA genes (Shore et al., 2014).

The ST22-MSSA isolates exhibited different antimicrobial resistance patterns including resistance to penicillin, gentamicin, tetracycline, kanamycin, rifampin, tobramycin, and amikacin. Our data are in concordance with a study conducted in Ireland, which reported the prevalence of low frequency of ST22-MSSA/t005 and ST22- MSSA/t1869 strains during the study period from 2002 to 2011. They indicated that ST22-MSSA was resistant to amikacin, ampicillin, fusidic acid, gentamycin, kanamycin, and tobramycin with multiple resistance genes blaZ, aacA-aphD, and dfrS1 detected among these isolates (Shore et al., 2014).

Our data showed that 40% of the CC/ST30 isolates carried pvl, which belonged to MRSA and MSSA strains. PVLpositive CC30/ST30-IV strains have been described in Iran (Goudarzi et al., 2016a,b), Kuwait (Boswihi et al., 2016), and Ireland (Shore et al., 2014). Importantly, the prevalence of CC30/ST30 increased from 8% in 2013 to 18.4% in 2018 in our country. In this connection, Boswihi et al. exhibited ST30 as the second dominant MRSA genotypes in Kuwait hospitals, which decreased from 30% in 2001–2003 to 22% in 2006. They also showed a low prevalence of ST30-IV-MRSA among tested isolates (2.9%) (Boswihi et al., 2016). Surprisingly, all MRSA isolates carried biofilm and adhesionrelated genes, and approximately 90% of the MSSA isolates were able to form biofilm at different intensities. In line with our study, Chamon et al. (2015) from Brazil showed biofilm production ability among ST30 strains. They also found bbp gene as a possible marker of this lineage. Congruent with the previous observations (Shore et al., 2014; Boswihi et al., 2016), all of our ST30 isolates belonged to agr type III, with different toxin and antimicrobial resistance patterns noted for this genotype.

As aforementioned, fewer than half of the CC/ST80-MRSA-IV isolates showed aminoglycoside resistance encoded by ant (4<sup>0</sup> )-Ia, aac (6<sup>0</sup> )-Ie/aph (200). Resistance to fusidic acid encoded by fusB was detected in one isolate. These results were different from a previous study conducted in Ireland, which exhibited all CC/ST80-MRSA-IV isolates were resistant to kanamycin and neomycin, encoded by aph (3<sup>0</sup> )-IIIa. In addition, our data revealed that resistances to tetracycline, fusidic acid, and erythromycin encoded by tet(K), fusB, and erm(C), respectively, were frequent among tested isolates (Shore et al., 2014). The variations could be a result of differences in the genetic backgrounds of the S. aureus strains. The prevalence of CC/ST80-MRSA-IV increased from 12% in 2013 to 13.2% in 2018. This increasing rate of ST80- IV raises the concern that this strain is becoming endemic in our hospitals.

Based on the evidence, ST80-IV is acknowledged as a toxigenic virulent isolate. In the present study, pvl-positive CC/ST80- SCCmec/t044 isolates were confirmed by approximately half of the isolates. ST80-MRSA-MRSA isolates harboring the pvl genes were previously reported from Kuwait, Malaysia, Singapore, and Ireland (Shore et al., 2014; Boswihi et al., 2016).

According to previously published data, CC/ST59 has limited geographical spread. In the present research CC/ST59 was present in 10 isolates, accounting for 7.9%. This clone was previously reported in Australia, Ireland, United Kingdom, Korea, Kuwait, and Taiwan (Monecke et al., 2011; Boswihi et al., 2016). In this study, the results revealed that our isolates carried resistance genes at a relatively low level, but erm(C) encoding resistance to erythromycin was noted in more than half of the isolates. Conversely, high frequencies of multiple resistance genes including erm(B), aph (3<sup>0</sup> )-IIIa, and tet(K) were identified among tested isolates of a study performed in Ireland (Shore et al., 2014). Notably, resistance to fusidic acid encoded by fusB was also detected in one isolate, which was similar to the previous report from Ireland (Shore et al., 2014). According to this experiment, 66.7% of the isolates were positive for seb (60.4%) gene. This is inconsistent with the research conducted in Ireland showing that seb gene could be a possible marker of this lineage. The observed frequency of CC59/ST338-SCCmecIII/t437 isolates was in accordance with previous reports, as this genotype is infrequently isolated (Li et al., 2013).

Another clone found among inducible resistance S. aureus strains was ST5-SCCmec IV/t002 (7.1%). Recent studies have shown the presence of CC/ST5- SCCmecIV/t002 clones in Asian and European countries, such as Iran, Japan, Korea, the United Arab Emirates, Kuwait, Ireland, and Australia (Monecke et al., 2011; Boswihi et al., 2016). The present data indicated that the prevalence of CC/ST5- SCCmecIV/t002 diminished during the study from 8% in 2013–2014 to 2.6% in 2017–2018. Conversely, a recent cross-sectional study performed in New Zealand on 3,323 patients from 2005 to 2011 documented seven most frequent MRSA clones. This study also indicated an ST5-SCCmecIV clone, which rapidly displaced ST30-SCCmecIV as the dominant CA-MRSA clone (Williamson et al., 2013). It was also observed that nearly half of ST5-SCCmecIV/t002 isolates were confirmed as HLMUPR strains. The finding of CC/ST5- SCCmecIV/t002 isolates in our screening indicated that resistance was observed for erythromycin encoded by erm(A), tetracycline encoded by tet(M), and mupirocin encoded by mupA. In an experiment conducted in 2010 in China, Song et al. (2013) revealed a different result. They showed a high prevalence of ST5-SCCmecIV/t002 among their clinical S. aureus isolates.

As shown in **Table 5**, the CC1 isolates belonged to two STs (ST772 and ST1). spa types t10795 and t657 pertained to ST772, whereas spa type t6811 was identified in ST1 isolates. ST772- SCCmecV, which is known as Bengal Bay clone, emerged in Bangladesh and was reported in New Zealand, Nepal, Italy, the United Arab Emirates, Malaysia, the United Kingdom, Ireland, Saudi Arabia, India, Australia, and Kuwait (Monecke et al., 2011; Shore et al., 2014; Boswihi et al., 2016). This clone was found in 6.4% of the examined isolates. All isolates were multiresistant and carried aac (6<sup>0</sup> )-Ie/aph (200), aph (3<sup>0</sup> )-IIIa, tet(M), and

erm(B). The attained data are in accordance to results of the study of Shore et al. (2014), which reported MDR pattern and carriage of ant (4<sup>0</sup> )-Ia, aac (6<sup>0</sup> )-Ie/aph (200), and tet(M) genes among CC1/ST772-SCCmecV isolates. The study demonstrated similarities in genetic characteristics including susceptibility to antibiotics and toxin-encoding genes of our ST772-SCCmecV isolates recently reported in Kuwait (Boswihi et al., 2016). Of two CC1/ST1-SCCmecV/t6811 isolates, one isolate carried fusc and exhibited resistance to fusidic acid at MIC 32 µg/mL. Boswihi et al. (2016) previously reported ST1-SCCmecV/t6811 carrying fusC gene from Kuwait.

In contrast to our study, which reported a low prevalence of ST15 (4%), a recent multicenter study performed in 25 European countries documented a relatively high prevalence of ST15 and reported it as the second most frequent clone across most of the European countries (Grundmann et al., 2014). In a study of 568 S. aureus isolates in 11 European countries, researchers reported a low level of this type among the tested isolates (Rolo et al., 2012). Our data demonstrated that all the ST15 isolates were pvl-positive and displayed the LLMUPR phenotype. The analysis of our previous study indicated that the most common mupirocinresistant MRSA isolates belonged to ST15-SCCmec IV/t084 (Goudarzi et al., 2017), which is in line with the present data. Although ST15 is more frequently detected in MRSA, there are reports that indicate the high distribution of this type among MSSA strains (Monecke et al., 2011; Boswihi et al., 2016). It was notable that all ST15 isolates harbored SCCmec IV, agr type II, and coa type I and belonged to spa type t084. Different antimicrobial resistance patterns were noted among these isolates, which is in accordance with previous studies (Rolo et al., 2012; Grundmann et al., 2014; Goudarzi et al., 2017).

In contrast to a study conducted in Kuwait, which reported CC45 as one of the most epidemics MRSA isolates with different antimicrobial resistance patterns, in the current research, a low frequency of the CC45 (4%) was observed among examined isolates. In a study conducted in the South of Poland, 26.1% of S. aureus isolates were found to be related to CC45 (Ilczyszyn et al., 2016). This ST was at its most prevalent in 2015, when it accounted for 22.2% (4/18) of inducible resistance isolates, reduced to 4.2% (1/24) in 2016, and disappeared in 2017–2018.

The strengths of our research included examining the prevalence and temporal differences in iMLSB S. aureus strains. It was the first study on the molecular characterization of inducible clindamycin-resistant S. aureus strains from Iran. However, the present research had limitations. One limitation of our study was the modest sample size and the impossibility of using typing methods such as pulsed-field gel electrophoresis (PFGE). It was not possible either to correlate demographics data with circulating clones owing to a lack of data linking patient characteristics. Another important limitation of our study was clfA- and clfB-negative S. aureus strains, which were unusual and required further analysis using whole-genome sequencing and microarray system.

## CONCLUSION

This was the first report of monitoring the prevalence and characterization of S. aureus isolates with the inducible resistance phenotype in Iran. Our investigation supports a detailed epidemiological survey on the prevalence and temporal differences in iMLSB S. aureus strains. It was ultimately attained that there is a considerable increasing trend for CC8 versus a decreasing trend for CC22. However, we revealed a shift in the clonal composition of MRSA isolates over time with the emphasis on a progressive replacement of CC22 clone by CC8 clones between 2013 and 2018. Indeed, our research indicated that iMLSB S. aureus isolates with similar genetic backgrounds exhibited specific virulence gene profiles, antimicrobial resistance patterns, and biofilm patterns. Increase in S. aureus with inducible resistance phenotype harboring SCCmecIV during the 5-year period makes sense that there is a shift in the iMLSB population from our community to hospital. Therefore, we conclude that there is a need for ongoing and nationwide surveillance studies to further evaluate of S. aureus with inducible resistance phenotype and to prevent these strains from becoming endemic in the Iranian hospitals.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## ETHICS STATEMENT

The current study protocol was approved by the Ethics Committee of the Shahid Beheshti University of Medical Sciences in Tehran, Iran (IR.SBMU.MSP.REC.1396.700). Written informed consent was obtained from participants.

## AUTHOR CONTRIBUTIONS

MG and HG conceived, designed, and supervised the study. MD, RaP, MN, MF, MM, SS, and AA performed the experiments. NK, RoP, and MG analyzed and interpreted the data. MG, HG, NK, SS, and RoP drafted and written the manuscript. All authors approved the final version of manuscript.

## FUNDING

This study was supported financially by the grant (No. 12688) from Research Deputy of Shahid Beheshti University of Medical Sciences, Tehran, Iran.

## ACKNOWLEDGMENTS

We are indebted to Dr. Edet E. Udo (Kuwait University) and Dr. Agnes Marie Sá Figueiredo (Universidade Federal do Rio de Janeiro) for providing reference strains of the S. aureus.

## REFERENCES

fmicb-11-00663 April 29, 2020 Time: 18:27 # 16


hospital setting by using novel software for spa repeat determination and database management. J. Clin. Microbiol. 41, 5442–5448. doi: 10.1128/jcm.41. 12.5442-5448.2003


isolates associated with lower respiratory infection. Afr. J. Microbiol. Res. 4, 2566–2569.


Yun, H.-J., Lee, S. W., Yoon, G. M., Kim, S. Y., Choi, S., Lee, Y. S., et al. (2003). Prevalence and mechanisms of low-and high-level mupirocin resistance in staphylococci isolated from a Korean hospital. J Antimicrob. Chemother. 51, 619–623. doi: 10.1093/jac/dkg140

**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 Goudarzi, Kobayashi, Dadashi, Pantu˚ˇcek, Nasiri, Fazeli, Pouriran, Goudarzi, Miri, Amirpour and Seyedjavadi. 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.

# Exploring the Potential of CRISPR-Cas9 Under Challenging Conditions: Facing High-Copy Plasmids and Counteracting Beta-Lactam Resistance in Clinical Strains of Enterobacteriaceae

#### Edited by:

Rustam Aminov, University of Aberdeen, United Kingdom

#### Reviewed by:

Yahong Liu, South China Agricultural University, China Sheng Chen, Army Medical University, China

#### \*Correspondence:

Hans-Peter Horz jhorz@ukaachen.de Simone Gonçalves dos Santos simonesantoskey@ufmg.br Tiago Antônio de Oliveira Mendes tiagoaomendes@ufv.br

†These authors share last authorship

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 13 November 2019 Accepted: 16 March 2020 Published: 30 April 2020

#### Citation:

Tagliaferri TL, Guimarães NR, Pereira MPM, Vilela LFF, Horz H-P, Santos SG and Mendes TAO (2020) Exploring the Potential of CRISPR-Cas9 Under Challenging Conditions: Facing High-Copy Plasmids and Counteracting Beta-Lactam Resistance in Clinical Strains of Enterobacteriaceae. Front. Microbiol. 11:578. doi: 10.3389/fmicb.2020.00578 Thaysa Leite Tagliaferri1,2, Natália Rocha Guimarães<sup>1</sup> , Marcella de Paula Martins Pereira<sup>1</sup> , Liza Figueiredo Felicori Vilela<sup>3</sup> , Hans-Peter Horz<sup>2</sup> \* † , Simone Gonçalves dos Santos<sup>1</sup> \* † and Tiago Antônio de Oliveira Mendes<sup>4</sup> \* †

<sup>1</sup> Department of Microbiology, Institute of Biological Sciences, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil, 2 Institute of Medical Microbiology, RWTH Aachen University Hospital, Aachen, Germany, <sup>3</sup> Department of Biochemistry and Immunology, Institute of Biological Sciences, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil, <sup>4</sup> Department of Biochemistry and Molecular Biology, Universidade Federal de Viçosa, Viçosa, Brazil

The antimicrobial resistance (AMR) crisis urgently requires countermeasures for reducing the dissemination of plasmid-borne resistance genes. Of particular concern are opportunistic pathogens of Enterobacteriaceae. One innovative approach is the CRISPR-Cas9 system which has recently been used for plasmid curing in defined strains of Escherichia coli. Here we exploited this system further under challenging conditions: by targeting the blaTEM−<sup>1</sup> AMR gene located on a high-copy plasmid (i.e., 100–300 copies/cell) and by directly tackling blaTEM−1-positive clinical isolates. Upon CRISPR-Cas9 insertion into a model strain of E. coli harboring blaTEM−<sup>1</sup> on the plasmid pSB1A2, the plasmid number and, accordingly, the blaTEM−<sup>1</sup> gene expression decreased but did not become extinct in a subpopulation of CRISPR-Cas9 treated bacteria. Sequence alterations in blaTEM−<sup>1</sup> were observed, likely resulting in a dysfunction of the gene product. As a consequence, a full reversal to an antibiotic sensitive phenotype was achieved, despite plasmid maintenance. In a clinical isolate of E. coli, plasmid clearance and simultaneous re-sensitization to five beta-lactams was possible. Reusability of antibiotics could be confirmed by rescuing larvae of Galleria mellonella infected with CRISPR-Cas9-treated E. coli, as opposed to infection with the unmodified clinical isolate. The drug sensitivity levels could also be increased in a clinical isolate of Enterobacter hormaechei and to a lesser extent in Klebsiella variicola, both of which harbored additional resistance genes affecting beta-lactams. The data show that targeting drug resistance genes is encouraging even when facing high-copy plasmids. In clinical isolates, the simultaneous interference with multiple genes mediating overlapping drug resistance might be the clue for successful phenotype reversal.

Keywords: CRISPR-Cas9, antimicrobial resistance, blaTEM, re-sensitization, resistance reduction, plasmid maintenance

## INTRODUCTION

fmicb-11-00578 April 28, 2020 Time: 17:25 # 2

Antimicrobial resistant microorganisms have become a public health concern due to their impact on human morbidity and mortality in recent decades. The reality of ineffective drugs allied to the insufficient launching of new antimicrobials may result in about 10 million deaths caused by multidrug-resistant organism infections in 2050 (O'Neil, 2016). The costs involved in the development of a new drug amount to around 800 million dollars and it can take up to 12 years until its commercialization (Adams and Brantner, 2006; Van Norman, 2016). Of particular concern is antimicrobial resistance (AMR) evolving in and spreading across species from the Enterobacteriaceae family, which includes plasmid encoded extended-spectrum beta-lactamases and carbapenemases as the main mechanism to disrupt antimicrobials (Savard and Perl, 2012).

Plasmids are well known for facilitating bacterial adaptation by their vertical and horizontal transmission. More specifically, small (<10 kb) multicopy plasmids have an important role in counteracting antimicrobial stressors. Multiple copies of a resistance gene increase the probability of mutational adaptation, thereby rapidly generating allele variations (Hall and Harrison, 2016). In addition, the repeated exposure to beta-lactams and high levels of beta-lactam resistance have been shown to be associated with an increase of plasmid copy number, reaching values higher than 100 copies/cell (San Millan et al., 2015, 2016). Hence, the increase of plasmid copy numbers can lead to maximum levels of AMR (San Millan et al., 2015, 2016; Hall and Harrison, 2016; Schechter et al., 2018).

Considering the capacity of plasmid-based resistance dispersal and the global public health impact of resistant bacteria, studies are needed for developing new technologies that could impede this progression. In recent years, CRISPR-Cas9 [clustered regularly interspaced short palindromic repeats (CRISPR)-CRISPR-associated protein 9 (Cas9)] has been demonstrated as an effective tool to cleave double-stranded DNA with accuracy (Ceasar et al., 2016). First studies have already successfully employed the CRISPR-Cas technology for genetically targeting AMR in different bacteria (Bikard et al., 2014; Citorik et al., 2014; Yosef et al., 2015). For instance, Yosef et al. (2015), not only reverted AMR but also eliminated the transfer of plasmids encoding two different beta-lactamases among Escherichia coli strains. Bikard et al. (2014) and Citorik et al. (2014) achieved plasmid curing and observed a CRISPR-Cas9 mediated cytotoxicity after specifically editing AMR genes in Staphylococcus aureus and in E. coli, respectively.

Given the importance of plasmid copy number variation in a cell (San Millan et al., 2015) and the overall complexity of AMR mechanisms within clinical isolates (Blair et al., 2015) we here investigated the potential of the CRISPR-Cas9 system under two challenging conditions. First, the CRISPR-Cas9 system was used to specifically target the blaTEM−<sup>1</sup> gene, located on the small high-copy plasmid pSB1A2, with 100–300 copies/cell (Yang et al., 2013; Registry of Standard Biological Parts, 2019), using a model E. coli strain. The blaTEM gene codifies for one of the most frequently encountered beta-lactamase in Enterobacteriaceae (Lachmayr et al., 2009; Rawat and Nair, 2010). Subsequently, we investigated the re-sensitization effect when targeting the same gene in clinical isolates of E. coli and related Enterobacteriaceae species.

## MATERIALS AND METHODS

## Escherichia coli Model Strain and Plasmids Used

The E. coli strain BL21 (Studier and Moffatt, 1986; Kim et al., 2017) was used in this study to validate the CRISPR-Cas9 system when targeting a high-copy plasmid. Bacteria were cultivated in LB medium at 37◦C and supplemented with ampicillin, 100 µg/ml; chloramphenicol, 30 µg/ml; and kanamycin, 50 µg/ml, depending on the plasmids harbored by the strains (**Table 1**). Three plasmids were used in this study, pSB1C3, pSB1A2, and pSB1K3, which are available at the Registry of Standard Biological Parts<sup>1</sup> under the accession numbers BBa\_K1218011, BBa\_J04450, and BBa\_I20260, respectively.

The pSB1C3 plasmid contains a constitutively expressed Streptococcus pyogenes derived CRISPR-Cas9 system. The sequence of the sgRNA responsible for the specificity of Cas9 cleavage was designed by indoor scripts using Pearl language based on public available sequences of the 861 bp-long blaTEM−<sup>1</sup> gene and its variants blaTEM−1a−<sup>d</sup> (**Supplementary Figure S1A**), including members of the Enterobacteriaceae. The sliding window method was used to find all 20-nucleotide length sequences with NGG flanking the 3-prime end. The conserved region most closely located at the 5<sup>0</sup> -end of the gene was selected (AGATCAGTTGGGTGCACGAGTGG). After synthesis of the 5<sup>0</sup> -located sgRNA, it was phosphorylated using a polynucleotide kinase (Thermo Fischer Scientific, MA, United States) and inserted into the plasmid pSB1C3 containing the CRISPR-Cas9 locus.

## Clinical Isolates

Escherichia coli 189A, Enterobacter hormaechei 4962 and Klebsiella variicola 68AI were isolated from patients with bacteremia in two different hospitals of Belo Horizonte, Brazil (**Supplementary Table S2**). Clinical strain collection was approved by the Human Research Ethics Committee from Universidade Federal de Minas Gerais (Brazil) under the protocol number ETIC 614/08 and written informed consent was obtained from each participant. E. hormaechei belongs to the Enterobacter cloacae complex and K. variicola is a member of the Klebsiella pneumoniae complex, as they share some biochemical and phenotypical features with E. cloacae and K. pneumoniae, respectively (Hoffmann et al., 2005; Barrios-Camacho et al., 2019). Because of this similarity, the species are frequently misclassified and the worldwide

<sup>1</sup>http://parts.igem.org/


CmR, chloramphenicol resistance as selective marker; AmpR, ampicillin resistance as selective marker; KmR, kanamycin resistance as selective marker.

presence of these bacteria in human infections may be underestimated, although their relevance as clinical pathogens has been demonstrated (Long et al., 2017; Beyrouthy et al., 2018; Barrios-Camacho et al., 2019; Rodríguez-Medina et al., 2019). The selection criteria of the isolates were the absence of a natural chloramphenicol-resistant phenotype, since the antibiotic was used as a selective marker of the pSB1C3 plasmid; the presence of the blaTEM−<sup>1</sup> gene and the absence of carbapenemase genes.

## Plasmid Transformation

Chemically competent E. coli BL21 were prepared using 0.1M MgCl2–CaCl<sup>2</sup> and competent cells were then transformed with plasmids by the heat-shock method (Chan et al., 2013; Lim et al., 2015). The clinical isolates used in this study received the CRISPR-Cas9 plasmid pSB1C3 by electroporation (Gonzales et al., 2013).

## Polymerase Chain Reactions

All primers used for polymerase chain reactions (PCRs), as well as the reaction conditions are provided in **Supplementary Table S1**. Cas9 functionality and blaTEM−<sup>1</sup> expression were assessed by reverse transcription quantitative polymerase chain reaction (RT-qPCR), which were performed based on cDNA from reverse transcription of total RNA. The 16S rRNA served as endogenous transcript for an internal control. Total bacterial RNA was extracted using TRIzol reagent and DNA was removed using DNase (both from Thermo Fischer Scientific, MA, United States). Integrity of extracted RNA was evaluated by agarose gel electrophoresis. Reverse transcription was performed with 1 µg of total RNA according to the iScript cDNA Synthesis Kit protocol (Bio-Rad, CA, United States). RT-qPCR was performed based on the CFX 96 Real Time PCR Detection System using the SYBR Green Master Mix (both from Bio-Rad, CA, United States). The analyses were performed by the relative standard curve method using a serial dilution of total cDNA pool as previously described (Larionov et al., 2005).

For determination of total plasmid copy number, the blaTEM−<sup>1</sup> gene was selected as a plasmid target. The amplified products of the chromosomal 16S rRNA gene and the blaTEM−<sup>1</sup> gene were cloned into TOPO vector (Thermo Fischer Scientific, MA, United States) for subsequent construction of the standard curves and absolute quantification. Determination of bacterial cell numbers was based on 16S rRNA gene quantification under consideration of the average 16S copy numbers per cell in strains (Klappenbach et al., 2000; Vetrovský and Baldrian, 2013 ˇ ). Establishment of standards with defined amounts of DNA was performed as previously described (Vianna et al., 2008).

## Growth Curves

Overnight bacterial cultures were adjusted to an OD600 nm of 0.1 and re-grown in LB medium with and without ampicillin. The growth kinetics were recorded every hour for 24 h similarly described in Jansen et al. (2018). Each assay was performed in triplicate and the whole experiment, starting from the pSB1C3 transformation, was performed three times on different days, accounting for technical and biological replicates.

## Disk Diffusion Test and Minimum Inhibitory Concentration

The disk diffusion test was performed in duplicate according to the Change to Clinical and Laboratory Standards Institute and the European Committee on Antimicrobial Susceptibility Testing (CLSI) protocol (Clinical and Laboratory Standards Institute, 2019). In total, 12 antibiotics were selected, namely chloramphenicol (CHL), ampicillin (AMP), ampicillin/sulbactam (SAM), amoxicillin/clavulanic acid (AMC), cefazolin (CFZ), cefoxitin, cefuroxime (CXM), ceftriaxone (CRO), ceftazidime (CAZ), cefotaxime (CTX), cefepime (FEP), and aztreonam (ATM). Measurements of diameter of the inhibition zones were performed in duplicate. For the Galleria mellonella assay, the minimum inhibitory concentration (MIC) of the clinical isolate of E. coli 189A was determined using the microdilution protocol (Clinical and Laboratory Standards Institute and Weinstein, 2012).

## Fluorescence Measurement

fmicb-11-00578 April 28, 2020 Time: 17:25 # 4

Fluorescence microscopy was performed using the EVOS <sup>R</sup> FL microscope (Life Technology, CA, United States). Cells were analyzed at the bright field and with the green fluorescent protein (GFP) (ex:470 nm/em:524 nm) and red fluorescence protein (RFP) (ex:530 nm/em:593 nm) fixed filters. For an overall analysis of RFP and GFP intensity, fluorescence was quantified using the Cytation 5 Cell Imaging Multi-Mode Reader (BioTeck, VT, United States). Fluorescence experiments were performed in triplicate and the whole experiment, starting from the pSB1C3 transformation, was performed three times on different days, accounting for technical and biological replicates.

Fluorescence-activated cell sorting (FACS) (BD FACS Canto II, BD, NJ, United States) was employed to distinguish between RFP positives and negative cells upon CRISPR-Cas9 treatment. Each measurement analyzed 30,000 events at a low flow rate. The following settings were used: an SSC voltage of 473; a FSC voltage of 398; and a PerCP-Cy5-5-A of 445 V.

## Sequencing Analyses

Sanger sequencing of purified PCR products was performed by Myleus Biotechnology (Minas Gerais, Brazil) and Eurofins Genomics (Luxembourg, Luxembourg). Sanger sequence analyses were performed using the Phred/Phrap pipeline and the Degenerate Sequence Decode program (DSDecodeM) (Liu et al., 2015). Deep sequencing of the genomes was performed using the Miseq platform (Illumina, CA, United States). Reads were trimmed to Phred15 using the Trimmomatic program (Bolger et al., 2014). The de novo assembly of the contigs was performed via the St. Petersburg genome assembler (SPAdes) (Bankevich et al., 2012) and resistance genes were detected via ResFinder (Zankari et al., 2012). Genome sequences were submitted to GenBank (accession number: E. coli SAMN12872130, E. hormaechei SAMN12872875 and K. variicola SAMN10216245.

## Galleria mellonella Infection Model

Larvae of the great wax moth G. mellonella were selected according to their length (2–3 cm, instar stage), weight (150– 250 mg), and excluded in case of dark coloration or limited activity. They were subsequently placed in the dark without food supply for 24-h acclimation prior to the infection assay (Harding et al., 2013). Larvae were challenged with E. coli 189A treated or untreated with CRISPR-Cas9 (∼4.0 × 10<sup>6</sup> CFU/ml). After 1-h of infection, either ceftriaxone (16 mg/kg) or sterile distilled water was administered in the front larvae proleg with a 10 µl syringe (Hamilton, NV, United States). Each group contained ten larvae sorted randomly, and the experiment was performed in triplicate (n = 30). Larvae were incubated at 37◦C, monitored after 24, 48, and 72 h and death was determined by absence of movement and unresponsiveness to touch.

## Statistical Analysis

Statistical analyses were performed using the GraphPad Prism (version 6, GraphPad Software, CA, United States), the VassarStats (NY, United States), as well as the BD FACSDiva and FlowJo softwares.

## RESULTS

## Resistance Reversal in a Model Strain of E. coli

Initially, the blaTEM−<sup>1</sup> gene located on the small high-copy plasmid pSB1A2 was introduced into the E. coli strain BL21, which was otherwise devoid of AMR. Located on this plasmid was also the RFP gene, as a marker, to verify the integrity of pSB1A2 upon CRISPR-Cas9 insertion. A second plasmid, pSB1K3, which contained the GFP gene, but no CRISPR-Cas9 target was also introduced into BL21. This plasmid served as an independent and indirect measurement of plasmid stability. The two plasmids pSB1A2 and pSB1K3 were chosen because of their high-copy number (i.e., 100–300 copies/cell) (Yang et al., 2013; Registry of Standard Biological Parts, 2019). For the design of a proper sgRNA, the conserved region most closely located at the 5<sup>0</sup> -end of the blaTEM−<sup>1</sup> gene was selected (**Supplementary Figure S1A**) in order to maximize the likelihood of an early stop codon generated during an eventual bacterial DNA repair mechanism (Chayot et al., 2010). Basic features of the three plasmids used in this study and associated strains are given in **Table 1**.

The presence and expression of Cas9 could be confirmed in CRISPR<sup>+</sup> and TEM+/CRISPR+, but not in BL21<sup>−</sup> or TEM<sup>+</sup> (**Figure 1A**), as expected. Growth curve analyses in the presence or absence of ampicillin confirmed antibiotic sensitivity of BL21<sup>−</sup> and CRISPR<sup>+</sup> (**Figures 1B,C**, respectively). In contrast, TEM<sup>+</sup> grew in the presence of ampicillin demonstrating the effectiveness of the introduced beta-lactamase gene (**Figure 1D**). The targeted re-sensitization to ampicillin was achieved in TEM+/CRISPR<sup>+</sup> (**Figure 1E**). Disk diffusion tests confirmed resistance of TEM<sup>+</sup> and sensitivity of TEM+/CRISPR<sup>+</sup> strains (**Supplementary Figure S1B**).

We observed a 100-fold lower RFP signal in TEM+/CRISPR+, compared to TEM<sup>+</sup> (p < 0.001) indicating a strong but not entire reduction of pSB1A2 (**Figures 2A,C**) upon CRISPR-Cas9 insertion. FACS results confirmed that only 0.005% of cells were RFP positive (**Figures 2D**, plot D.1 and D.3), which represents an around 150-fold reduction compared to the TEM<sup>+</sup> cells (**Figure 2D**, plot D.2). A control experiment with the pSB1C3 vector lacking the sgRNA led to a high percentage of RFP-positive cells (85.2%), similar to TEM<sup>+</sup> cells, confirming that the plasmid reduction was due to the presence of the sgRNA. This result combined with quantitative PCR analysis (qPCR) of the blaTEM gene indicated, that on average, the 0.005% persistent RFP positive TEM+/CRISPR<sup>+</sup> cells carried around 48 pSB1A2 copies/cell. In contrast, TEM<sup>+</sup> cells harbored around 100 copies/cell of pSB1A2. Since the GFP signals did not decrease (**Figures 2A,B**), pSB1A2 reduction was the result of CRISPR-Cas9 activity rather than any unspecific plasmid clearance by the cell.

In keeping with the RFP and plasmid reduction, the expression of the blaTEM−<sup>1</sup> gene was significantly reduced (by more than 2.5-fold, p < 0.05) in TEM+/CRISPR+, again confirming reduction but not entire plasmid loss (**Figure 2E**). Since those persistent gene expression levels did not confer ampicillin resistance (**Figure 1E** and **Supplementary Figure S1B**), it is likely

replicates ± standard deviation. Stars indicate statistical support for differences between strains with or without the CRISPR-Cas9 system, \*\*\*p < 0.001.

that gene defects occurred. In fact, deletions could be seen in the sequence data of the blaTEM−<sup>1</sup> resistance gene (**Supplementary Figure S2A**). This means, that any possible transmission of the remaining plasmid, vertically or horizontally, would not necessarily lead to the spread of blaTEM−1-based AMR, because of the non-functional resistance gene.

However, cases where the CRISPR-Cas9 was able to completely eradicate the high-copy plasmid pSB1A2 were also achieved, as evident by FACS analysis (**Figures 2D**, plot D.4) and confirmed by qPCR.

## Resistance Reversal in the Clinical Isolate of E. coli 189A

We next tested a blaTEM−<sup>1</sup> positive clinical isolate of E. coli 189A. Principally, considering that blaTEM variants potentially confer resistance to many beta-lactam antibiotics (Rawat and Nair, 2010), targeting this gene might restore the usability of several antibiotics. In fact, after CRISPR-Cas9 insertion into the clinical isolate of E. coli 189A, a re-sensitization was observed to AMP, CFZ, CXM, CRO, and CTX (**Figures 3A,B** and **Supplementary Figure S2B**), along with an introduced resistance against chloramphenicol co-mediated via pSB1C3. Based on qPCR, the blaTEM−<sup>1</sup> harboring plasmid was found to occur in low-copy numbers, as on average less than a single copy per bacterial cell was detected, indicating that some bacteria might be already plasmid free. After the CRISPR-Cas9 insertion, the blaTEM gene was not detectable anymore, indicating complete eradication of the plasmid (**Figure 3C**). The achieved plasmid clearance also resulted in the extinction of other resistance genes conferring resistance to beta-lactams and other classes of antibiotics. Genome sequencing identified the sul2 (sulfonamide), aph(300)-Ib and aph(6)-Id (both confer resistance

areas. The plot (D.3) represents the colony with higher RFP percentage (0.032%), while (D.4) demonstrates the colony with no RFP signal (0%). When analyzing all retrieved colonies, 0.005% of RFP positive cells were present upon CRISPR-Cas9 insertion. A control experiment with the pSB1C3 vector lacking the sgRNA led to a high percentage of RFP-positive cells (85.2%), similar to TEM<sup>+</sup> cells, confirming the sgRNA-dependant plasmid reduction. (E) Relative expression of the blaTEM−<sup>1</sup> gene confirming plasmid presence and gene functioning in the cells. Bars represent average and standard deviation of two replicate assays. The results were normalized with the 16S rRNA housekeeping gene. \*p < 0.05; \*\*\*p < 0.001.

replicates ± standard deviation. A.U., arbitrary units. (D) Histograms of negative (D.1) and the positive controls (D.2) of FACS analysis used to demarcate the gate

to aminoglycoside), as well as blaCTX−M−<sup>9</sup> (beta-lactam) genes before, but not after the CRISPR-Cas9 insertion.

To further investigate the meaningfulness of the resensitization approach, the reusability of CRO as one representative antibiotic was verified by infecting larvae of the great wax moth G. mellonella either with the CRISPR-Cas9 treated E. coli 189A (E. coli 189ACRISPR+) or with the E. coli 189A wild type (E. coli 189AWT).

Upon administration of CRO, 70% of larvae infected with E. coli 189ACRISPR<sup>+</sup> survived the time period of 72 h, as opposed

to only 30% of larvae infected with E. coli 189AWT (log-rank test, p < 0.01) (**Figure 3D**). No significant difference was seen between the CRO-treated larvae challenged with E. coli 189ACRISPR<sup>+</sup> and larvae used as controls (i.e., mock-infection with saline or administration of antibiotic only, or no treatment at all. Logrank test, p > 0.05). Conversely, it made no difference whether E. coli 189AWT infected larvae were treated with antibiotic or with water (log-rank test, p > 0.3) (**Figure 3D**). The outcome of this experiment indicates that antibiotic-treated larvae had a 5.44 times higher chance of survival, when challenged with E. coli 189ACRISPR<sup>+</sup> rather than with E. coli 189AWT (p < 0.01).

## Resistance Reduction in Clinical Isolates of E. hormaechei 4962 and K. variicola 68AI

A more complex situation was encountered when targeting the blaTEM−<sup>1</sup> gene of E. hormaechei 4962 and K. variicola 68AI. Upon CRISPR-Cas9 insertion, significant increases in the inhibition zones were observed for ATM, CTX, and CRO in the case of E. hormaechei (**Figures 4A,B** and **Supplementary Figure S2B**). However, the CLSI-defined threshold levels for an intermediate sensitivity was only reached for ATM. In some disk diffusion assays, a few colonies grew within the inhibition zones which were, however, not further analyzed (**Figure 4A**). For K. variicola, only non-significant increases in the inhibition zones were observed for CAZ and FEP (**Figures 4C,D** and **Supplementary Figure S2B**).

Further analysis revealed the continued presence of the blaTEM−<sup>1</sup> gene in E. hormaechei 4962 and K. variicola 68AI, in contrast to what was observed with the clinical isolate of E. coli 189A, which could be re-sensitized for several antibiotics. qPCRbased detection of the blaTEM−<sup>1</sup> gene showed that a substantial plasmid reduction had occurred in E. hormaechei 4962, as opposed to only a negligible fraction of plasmid reduction in K. variicola 68AI (**Figure 3C**). Since both E. hormaechei 4962

and K. variicola 68AI were found to harbor the blaTEM−<sup>1</sup> gene on a low-copy number plasmid (**Figure 3C**), we expected to achieve plasmid clearance. However, damages on the CRISPR-Cas9 loci after insertion were observed in both strains, affecting the expected CRISPR-Cas9-based outcome (**Supplementary Figure S3**). These findings explain the lower efficiency of resensitization when compared to the clinical strain of E. coli 189A.

Importantly, other genes conferring resistance to beta-lactams could be identified by genome sequencing in both strains. In E. hormaechei 4962, the intrinsic AmpC-type resistance gene was present (blaACT−7), as well as the beta-lactamases blaCTX−M−<sup>9</sup> and blaOXA−9. The same blaCTX−M−<sup>9</sup> was also detected in K. variicola 68AI, along with the blaLEN16 and blaLEN19 genes conferring resistance to beta-lactams. Since these resistance genes were not targeted by the sgRNA, failure to achieve fully reverted phenotypes was likely due to their presence in the bacteria, in addition to the damages in the CRISPR-Cas9 plasmid region.

## DISCUSSION

In this study, we investigated the potential of the CRISPR-Cas9 system to counteract antibiotic resistance mediated by the blaTEM−<sup>1</sup> gene harbored either in the high-copy plasmid pSB1A2 or when present in clinical isolates. In a model strain of E. coli having received the blaTEM−<sup>1</sup> gene on a high-copy plasmid (i.e., 100–300 copies/cell) the phenotype could be clearly reversed. However, an entire elimination of the plasmid and the target gene did not occur in some retrieved colonies. This outcome is likely linked with the initial high overall abundance of the vector, since CRISPR-Cas9 is able to completely clear a resistance gene plasmid, when the copy number per cell ranges between 50 and 70 (Citorik et al., 2014). Those remaining plasmid-positive cells, could be considered "persister cells" in an analogy to bacteria persisting antibiotic treatment (Van den Bergh et al., 2017), given their low fraction within the whole population, and their endurance in the presence of CRISPR-Cas9 concomitant with an otherwise re-sensitized phenotype (Wood et al., 2013; Van den Bergh et al., 2017). However, even facing high-copy plasmids, CRISPR-Cas9 demonstrated the potential to completely clear the vector from some bacterial colonies, indicating its promising application under this challenging condition. Optimization of the system may be required for a complete clearance of high-copy plasmids carrying resistance genes in the whole population.

The reason why a bacterial subpopulation maintains the plasmid while being attacked by CRISPR-Cas9 remains unknown. However, laboratorial and pathogenic E. coli strains possess an end-joining repair mechanism to bridge broken DNA ends (Chayot et al., 2010). Named as alternative end-joining (A-EJ), this restoration attempt typically leads to microdeletions in the sgRNA region (affecting regions of 1–10 bp), similar to those observed in this study (**Supplementary Figure S2A**; Chayot et al., 2010; Chen et al., 2019). Although the deletions lead to a non-functional blaTEM−<sup>1</sup> gene, the possible existence of

CRISPR-Cas9-persisters in the context of clinical isolates, could still result in therapeutic failure, because other intact resistance genes might still reside on the plasmid. Therefore, the detection and knowledge of CRISPR-Cas9-persisters may be important for devising strategies to minimize their potential clinical impact.

The deleterious effects on plasmids and bacterial cells of CRISPR-Cas9 targeting other resistance genes have been previously reported. Bikard et al. (2014) interfered with the kanamycin and the methicillin resistant genes aph-3 and mecA, in S. aureus (MRSA), while Citorik et al. (2014) re-sensitized E. coli that possessed the blaNDM−<sup>1</sup> and blaSHV−<sup>18</sup> genes. In both studies, a cytotoxicity of the CRISPR-Cas9 system was apparent with episomal and chromosomal targets. In the first situation, bacteria were killed because plasmid clearance led to the loss of the toxin-antitoxin genes. As the antitoxin is less stable than the toxin, its faster degradation results in cell toxicity as a consequence (Citorik et al., 2014). When the target was a chromosomal gene, the lethality was due to irreparable chromosomal damages (Bikard et al., 2014; Citorik et al., 2014). Conversely, no cytotoxicity was observed when CRISPR-Cas9 eliminated a resistance plasmid in bacteria lacking a toxinantitoxin system (Bikard et al., 2014; Citorik et al., 2014). Plasmid clearance was also achieved in another model strain of E. coli in which the blaNDM−<sup>1</sup> and blaCTX−M−<sup>15</sup> resistance genes were targeted (Yosef et al., 2015). The authors went one step further and simultaneously conferred resistance against the lytic phage T7, thereby providing the re-sensitized bacteria a selective advantage in the presence of the phage. Using this approach, antibiotic treatment along with administration of T7 should robustly oppress the re-emergence of resistant variants.

With the successful reversal of drug resistance in E. coli model strains, the consecutive step would be to test clinical isolates. In fact, in our study, plasmid clearance was achieved in the clinical strain of E. coli 189A when tackling the blaTEM−<sup>1</sup> gene present in a low-copy plasmid. Importantly, CRISPR-Cas9 activity led to a simultaneous re-sensitization to several beta-lactam antibiotics including CRO. This is particularly encouraging, given that third generation cephalosporins are "critically important antimicrobials," which need to be preserved as last resource treatments (Collignon et al., 2016). Re-sensitization effectiveness could be verified in vivo by rescuing E. coli 189A infected larvae with CRO, confirming the benefit of CRISPR-Cas9 for controlling drug resistant clinical isolates.

Both pSB1C3 vectors harbored by the clinical isolates of K. variicola 68AI and E. hormaechei 4962 demonstrated deletions in the CRISPR-Cas9 locus, and subsequent maintenance of the blaTEM−1-carrying low-copy plasmid. This is similar to what has been described by Bikard et al. (2014), who found that survival of S. aureus was due to the lack of the Cas9 region after CRISPR-Cas9 delivery, impeding DNA cleavage. Apart from this, plasmid maintenance with the presence of at least one other resistance gene not covered by our CRISPR-Cas9 system might explain why the resensitization in E. hormaechei 4962 and K. variicola 68AI was only partially successful. Given that clinical strains frequently harbor different resistance genes on plasmids (Bennett, 2009; Blair et al., 2015), this signifies that multiple sgRNA for potentially targeting different AMR genes need to be employed, for re-gaining full susceptibility to traditional antibiotics.

Even though the resistance reversal was not completely achieved to the sensitive category in E. hormaechei 4962, the intermediate sensitivity may still imply therapeutic success.

After all, according to the latest guidelines of the CLSI (2019) and The European Committee on Antimicrobial Susceptibility Testing [EUCAST] (2019), intermediate levels are to be interpreted as persisting clinical efficacy under elevated drug exposure.

Clearly, impediment of re-sensitization can also stem from overexpression of efflux pumps, or from reduction of permeability. Those intrinsic resistance mechanisms contribute to increased levels of resistance against cephalosporins and are typically found in clinical strains of Enterobacteriaceae (Blair et al., 2015). Nonetheless, overexpression of efflux pumps or alterations in porins in the clinical isolate of E. coli, if having occurred, did not prevent phenotype reversal. Furthermore, despite the versatile challenges imposed by clinical isolates, the interference with the blaTEM−<sup>1</sup> gene led after all to minor but clear resistance reductions in the other two clinical isolates.

Our findings support multiple aspects of the previous studies and complement the possible outcomes of the CRISPR-Cas9-based interference with plasmid-borne resistance genes, summarized in **Figure 5**.

When translating the use of the CRISPR-Cas9 system into clinical settings, phages might be an elegant approach for targeted delivery of the system into the bacterial pathogens. Successful application in mice and larvae using bio-engineered phages (Bikard et al., 2014; Citorik et al., 2014) pave the way for a possible future utilization in humans. Alternatively, phage-mediated CRISPR-Cas9 systems could be dispersed on surfaces to serve as preventive disinfection and cleansing procedures along with elaborate selection mechanisms that ensure stability of the CRISPR-Cas systems in bacterial cells, as demonstrated previously (Yosef et al., 2015). Importantly, cases in which CRISPR-Cas9 maintenance in the bacterial cell is intended, resistance genes commonly used as plasmid selective markers should be removed from the CRISPR-Cas9 vector in order to prevent further AMR spread. The possibility to restore sensitivity to traditional antibiotics might be superior to searching for new antimicrobials and potentially could deaccelerate the crisis that we are currently facing with AMR carrying bacteria.

## REFERENCES


## DATA AVAILABILITY STATEMENT

The genomic dataset generated for this study can be found in the Genbank (accession number: E. coli SAMN12872130, E. hormaechei SAMN12872875, and K. variicola SAMN10216245).

## ETHICS STATEMENT

Clinical strain collection was approved by the Human Research Ethics Committee from Universidade Federal de Minas Gerais (Brazil) under the protocol number ETIC 614/08 and written informed consent was obtained from each participant.

## AUTHOR CONTRIBUTIONS

TT, TM, SS, and H-PH designed the study. TT, NG, and MP performed the research. H-PH, LV, TM, and SS contributed with new reagents. TT, TM, and H-PH analyzed the data. TT, H-PH, SS, and TM wrote the manuscript.

## FUNDING

This work was supported by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq); Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES); Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG); and Fundação Arthur Bernardes and Bill and Melinda Gates Foundation (grant number OPP1193112). TT was supported by CNPq Scholarship - Brazil (grant number 203319/2016-6).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00578/full#supplementary-material

involving carbapenemase-producing strains. Emerg. Infect. Dis. 24, 1505–1515. doi: 10.3201/eid2408.180151



**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 Tagliaferri, Guimarães, Pereira, Vilela, Horz, dos Santos and Mendes. 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 KPC-Producing Serratia marcescens in an Intensive Care Unit of a Brazilian Tertiary Hospital

Roumayne L. Ferreira<sup>1</sup>† , Graziela S. Rezende<sup>1</sup>† , Marcelo Silva Folhas Damas<sup>1</sup>† , Mariana Oliveira-Silva<sup>2</sup> , André Pitondo-Silva<sup>2</sup> , Márcia C. A. Brito<sup>3</sup> , Eduardo Leonardecz<sup>1</sup> , Fabiana R. de Góes<sup>4</sup> , Emeline Boni Campanini<sup>1</sup> , Iran Malavazi<sup>1</sup> , Anderson F. da Cunha<sup>1</sup> and Maria-Cristina da Silva Pranchevicius<sup>1</sup> \*

#### Edited by:

Filipa Grosso, University of Porto, Portugal

#### Reviewed by:

Martina Barchitta, University of Catania, Italy Daria Van Tyne, University of Pittsburgh, United States

#### \*Correspondence:

Maria-Cristina da Silva Pranchevicius mcspranc@gmail.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 28 October 2019 Accepted: 21 April 2020 Published: 20 May 2020

#### Citation:

Ferreira RL, Rezende GS, Damas MSF, Oliveira-Silva M, Pitondo-Silva A, Brito MCA, Leonardecz E, Góes FR, Campanini EB, Malavazi I, da Cunha AF and Pranchevicius M-CS (2020) Characterization of KPC-Producing Serratia marcescens in an Intensive Care Unit of a Brazilian Tertiary Hospital. Front. Microbiol. 11:956. doi: 10.3389/fmicb.2020.00956 <sup>1</sup> Departamento de Genética e Evolução, Universidade Federal de São Carlos, São Carlos, Brazil, <sup>2</sup> Programas de Pós-graduação em Odontologia e Tecnologia Ambiental, Universidade de Ribeirão Preto, Ribeirão Preto, Brazil, <sup>3</sup> Laboratório Central de Saúde Pública do Tocantins, Palmas, Brazil, <sup>4</sup> Instituto de Ciências Matemáticas e de Computação, Universidade de São Paulo, São Carlos, Brazil

Serratia marcescens has emerged as an important opportunistic pathogen responsible for nosocomial and severe infections. Here, we determined phenotypic and molecular characteristics of 54 S. marcescens isolates obtained from patient samples from intensive-care-unit (ICU) and neonatal intensive-care-unit (NIUC) of a Brazilian tertiary hospital. All isolates were resistant to beta-lactam group antibiotics, and 92.6% (50/54) were not susceptible to tigecycline. Furthermore, 96.3% showed intrinsic resistance to polymyxin E (colistin), a last-resort antibiotic for the treatment of infections caused by MDR (multidrug-resistant) Gram-negative bacteria. In contrast, high susceptibility to other antibiotics such as fluoroquinolones (81.5%), and to aminoglycosides (as gentamicin 81.5%, and amikacin 85.2%) was found. Of all isolates, 24.1% were classified as MDR. The presence of resistance and virulence genes were examined by PCR and sequencing. All isolates carried KPC-carbapenemase (blaKPC) and extended spectrum beta-lactamase blaTEM genes, 14.8% carried blaOXA−1, and 16.7% carried blaCTX−M−1group genes, suggesting that bacterial resistance to β-lactam antibiotics found may be associated with these genes. The genes SdeB/HasF and SdeY/HasF that are associated with efflux pump mediated drug extrusion to fluoroquinolones and tigecycline, respectively, were found in 88.9%. The aac(6<sup>0</sup> )-Ib-cr variant gene that can simultaneously induce resistance to aminoglycoside and fluoroquinolone was present in 24.1% of the isolates. Notably, the virulence genes to (i) pore-forming toxin (ShlA); (ii) phospholipase with hemolytic and cytolytic activities (PhlA); (iii) flagellar transcriptional regulator (FlhD); and (iv) positive regulator of prodigiosin and serratamolide production (PigP) were present in 98.2%. The genetic relationship among the isolates determined by ERIC-PCR demonstrated that the vast majority of isolates were grouped in a single cluster with 86.4% genetic similarity. In addition, many isolates showed 100% genetic similarity to each other, suggesting that the S. marcescens that circulate in this ICU are closely related. Our results suggest that the antimicrobial resistance to many drugs currently used to treat ICU and NIUC patients, associated with the high frequency of resistance and virulence genes is a worrisome phenomenon. Our findings emphasize the importance of active surveillance plans for infection control and to prevent dissemination of these strains.

Keywords: Serratia marcescens, intensive care units, KPC, virulence and resistance genes, ERIC-PCR

## INTRODUCTION

fmicb-11-00956 May 19, 2020 Time: 16:32 # 2

Serratia marcescens is a Gram-negative bacillus that naturally resides in the soil and water and produces a red pigment at room temperature. Although previously considered non-pathogenic, this species has emerged as a prominent opportunistic pathogen found in nosocomial outbreaks in neonatal intensive care Units (NICUs), intensive care units (ICUs) and other hospital units over the last few decades (Enciso-Moreno et al., 2004; Moradigaravand et al., 2016; Ghaith et al., 2018).

The true occurrence of S. marcescens is still underestimated (Zingg et al., 2017). In NICUs, studies have showen that infected newborns are a potential source of S. marcescens (Cristina et al., 2019), although there is a constant increase of S. marcescens bacteremia across all age groups (Vetter et al., 2016; Phan et al., 2018). S. marcescens increasingly adapts to hospital environments (Yoon et al., 2005; Gastmeier, 2014). It accounts for 15% of all isolates from nosocomial infections (Raymond and Aujard, 2000). Although it is difficult to identify the source of S. marcescens during outbreaks, it is the third most frequent pathogen identified (Gastmeier et al., 2007), and more than one clone can be usually identified (David et al., 2006; Montagnani et al., 2015; Dawczynski et al., 2016).

Serratia marcescens associated with hospital outbreaks or epidemic events are commonly resistant to several antibiotics (Moradigaravand et al., 2016; Cristina et al., 2019). In fact, one important feature of S. marcescens is its resistance to narrow-spectrum penicillins and cephalosporins; nitrofurantoin; tetracycline; macrolides; cefuroxime; cephamycins; fluoroquinolone, and colistin (Stock et al., 2003; Liou et al., 2014; Moradigaravand et al., 2016; Sandner-Miranda et al., 2018). The resistance to some of these molecules may be intrinsic to this specie and is explained by either the presence of resistance genes on the chromosome or by the acquisition of such genes via horizontal transfer. It is noteworthy that the latter mechanism is considered the most important event that leads to multiple antibiotic resistance (von Wintersdorff et al., 2016; Sandner-Miranda et al., 2018).

Extended-spectrum β-lactamases (ESBLs) are a group of bacterial enzymes that can be rapidly transferred via plasmid exchange (Rawat and Nair, 2010) causing resistance to a broad range of β-lactams (Naas et al., 2008). Carbapenemases are the most versatile family of β-lactamases able to hydrolyze carbapenems and many other β-lactams (Jeon et al., 2015) including penicillins, cephalosporins, and monobactams (Anderson et al., 2007; Marschall et al., 2009). In general, bacteria carrying the blaKPC and/or ESBLs genes usually harbor other resistance genes associated with several classes of antimicrobials (Tzouvelekis et al., 2012; Cao et al., 2014; Ribeiro et al., 2016). Since S. marcescens has been acquiring a range of ESBLs and commonly exhibit co-resistance to many other classes of antibiotics, the infections caused by these multidrug-resistant (MDR) isolates impair therapy and limit treatment options (Yu et al., 1998; Mostatabi et al., 2013; Herra and Falkiner, 2018).

In this study, we investigated the phenotypic characteristics regarding antimicrobial resistance and the genotypic traits of S. marcescens isolated from a tertiary care hospital's ICUs including the search for resistance and virulence genes as well as the genetic relationship among the isolates. Our report describes MDR profile and KPC-producing S. marcescens isolates and highlight the importance of monitoring S. marcescens infection and the need of constant surveillance to support continuous and effective measures to prevent the spread of these strains.

## MATERIALS AND METHODS

## Study Design and Bacterial Isolates

From February 2014 to June 2015, a total of 54 S. marcescens were isolated of clinical specimens collected from 45 patients admitted to intensive-care-unit (ICU) and neonatal intensive care unit (NICU) of a tertiary care government hospital in Palmas, Tocantins, Brazil. Since 2013, there has been an increase in detection of S. marcescens isolates from hospital inpatients, and in 2015, the hospital reported an apparent S. marcescens outbreak that occurred from July to August 2015. Appropriate intervention measures were established, such as reviewing the infection control policies, hand antisepsis practices and determination of trends of isolation of S. marcescens over time. To trace the source of the infection, bacteria were isolated from various samples obtained from clinical indications of infections during the patients' ICU stay.

As part of the control measures, surveillance cultures were obtained from tracheal aspirate and rectal swabs from all ICU patients, on admission (within the first 24 h) and during the stay (once a week). Blood, wound, catheter tip, drain, sputum, urine, rectal swab, and tracheal aspirate samples were primarily sent to the hospital's laboratory, processed and cultured by standard microbiological techniques. The blood samples were inoculated first in blood culture bottles (Hemoprov-NewProv, Brazil). All clinical samples, including blood culture bottles giving positive signals were cultured onto MacConkey agar (Probac, Brazil), blood agar (Probac, Brazil), and chocolate blood agar (Probac, Brazil). Plates were incubated at 37◦C for up to 48 h. S. marcescens were identified by Gram staining, cultural characteristics in MacConkey agar (Probac, Brazil), blood agar (Probac, Brazil), and biochemical tests (Bactray I, II,III; Laborclin, Brazil). The antimicrobial susceptibility profile

was determined by Kirby-Bauer disk diffusion method. All S. marcescens isolates and the microbiological reports prepared at the hospital were sent to the Central Laboratory of Public Health of Tocantins (LACEN-TO) for further phenotypic validations.

## Bacterial Identification and Antimicrobial Susceptibility Test

Once samples were received at LACEN, bacterial identification and antimicrobial susceptibility tests were performed by the Vitek 2 system (Biomerieux, France), according to Clinical and Laboratory Standards Institute guidelines (CLSI, 2019). All 54 S. marcescens isolates were screened for susceptibility against 16 antimicrobial agents: ampicillin (AMP), ampicillin/sulbactam (SAM), piperacillin/tazobactam (TZP), cefuroxime (CXM), cefoxitin (FOX), ceftazidime (CAZ), ceftriaxone (CRO), cefepime (FEP), ertapenem (ETP), imipenem (IPM), meropenem (MEM), amikacin (AMK), gentamicin (GEN), ciprofloxacin (CIP), tigecycline (TGC), and colistin (CST). Broth microdilution method was performed to determine tigecycline and colistin minimum inhibitory concentration (MICs) and results were interpreted based on the European Committee on Antimicrobial Susceptibility Testing (EUCAST, 2018) criteria, available at https://www.eucast.org/fileadmin/src/media/PDFs/EUCAST\_

files/Breakpoint\_tables/v\_8.1\_Breakpoint\_Tables.pdf. All isolates were tested for carbapenemase production by Modified Hodge test, synergy test and ethylenediaminetetraacetic acid (EDTA) test under the CLSI guidelines (CLSI, 2019) as described elsewhere (Miriagou et al., 2010; Nordmann et al., 2011; Okoche et al., 2015; Ferreira et al., 2019). Multidrug-resistance S. marcescens isolates were classified by non-susceptibility to at least one agent of three or more antimicrobial categories (Magiorakos et al., 2012). S. marcescens is intrinsically resistant to AMP, SAM, CXM, FOX, and CST; therefore, these antibiotics were not included in the MDR classification (Magiorakos et al., 2012).

## Genomic DNA Extraction

Isolates of S. marcescens were subcultured on Brain Heart Infusion (BHI) agar (Oxoid, United Kingdom) and incubated for 24 h at 37◦C. All samples were submitted to genomic DNA extraction using the Wizard Genomic DNA Purification Kit (Promega, Madison, WI, United States), according to manufacture's instructions.

## Detection of Antibiotic-Resistance

Polymerase chain reaction (PCR) was performed for detection of β-lactamase genes (blaTEM, blaSHV variants, blaOXA−1, <sup>4</sup>and <sup>30</sup>, blaCTX−M−1group), carbapenemase genes (blaKPC, blaIMP, blaVIM, blaNDM, blaOXA−48) (Ferreira et al., 2019), plasmid mediated quinolone resistance (PMQR) gene (aac(6')-Ib-cr) (Wong et al., 2014; Mitra et al., 2019), resistance-nodulationdivision (RND) efflux pumps (SdeB, SdeY), and outer membrane gene (HasF, a TolC homolog) involved in energy-dependent efflux of antimicrobial agents (Kumar and Worobec, 2005b). The genes were amplified using specific primers designed to follow the conditions described in the references from **Table 1**. All primers were synthesized by Exxtend (Brazil). Amplicons were analyzed by gel electrophoresis in 1.0% agarose and visualized under ultraviolet (UV) light.

## Virulence Gene Detection

The presence of four virulence genes were assessed by PCR: genes PigP, a positive regulator of prodigiosin and serratamolide production; FlhD, a flagellar transcriptional regulator; ShlA, a pore-forming toxin with hemolytic activity; PhlA, a phospholipase A with hemolytic activity. The primers sequences amplicon sizes and annealing temperatures are listed in **Table 1**. Amplicons were analyzed by gel electrophoresis in 1.0% agarose and visualized under ultraviolet (UV) light.

## Sequence Analysis of Antibiotic-Resistance Markers and Virulence Genes

One amplicon of each studied gene was randomly selected for confirmation of identity by DNA sequencing using an automated sequencer (ABI 3500xL Genetic Analyzer; Applied Biosystems, Foster City, CA, United States). After amplification, we extracted the PCR products from agarose gels using the Illustra GFX PCR DNA (GE Healthcare), which were purified using the Gel Band Purification Kit (GE Healthcare), both according to manufacturer's instructions. Obtained sequences were edited with Bioedit v7.0.5 (Hall, 1999), compared with the nr database using the Blastn tool<sup>1</sup> and submitted to the GenBank database. Genes and their respective accession numbers: blaCTX – MK576103; blaKPC – MK576104; blaOXA – MK576105; blaTEM – MK576106; SdeB – MN583232; SdeY – MN583233; HasF – MN583234; aac(6<sup>0</sup> )-Ib-cr – MN583235; FlhD – MN583236; PigP – MN583237; ShlA – MN583238; PhlA – MN583239). Access to genetic heritage was approved by the National System for the Management of Genetic Heritage (SisGen n◦ AFF27ED).

## Enterobacterial Repetitive Intergenic Consensus Polymerase Chain Reaction

Enterobacterial repetitive intergenic consensus PCR (ERIC-PCR) analysis was performed to evaluate the genetic similarity among the 54 S. marcescens isolates using the primers and conditions previously described by Versalovic et al. (1994). PCR reactions were performed using the enzyme TaKaRa Ex Taq DNA Polymerase (Takara Bio, Kusatsu, Japan). The BioNumerics program version 5.1 (AppliedMaths, Keistraat, Belgium) was used to construct the unweighted pair group mean method (UPGMA) similarity dendrogram with Dice's similarity coefficient, following Ferreira et al. (2019).

## Statistical Analyses

In the analysis of contingency tables, we used Fisher's exact test and/or Barnard's exact test. Maximum likelihood did not present superior efficiency in relation to the previous methods (data not show). It was used logistic regression model with two predictor variables x<sup>1</sup> and x2. Statistical software R was used in all data analysis.

<sup>1</sup>https://blast.ncbi.nlm.nih.gov/

#### TABLE 1 | Sequences of primes used for detection of resistance markers.


## Ethical Considerations

In our study, we did not use/collect human genetic material and biological samples. Strains were part of the collection of the Central Laboratory of Public Health, (LACEN-TO), a health-care facility that is a reference in diagnosis in the state of Tocantins, Brazil. It was a retrospective study, and epidemiological data were obtained from a database or similar, which will be kept confidential in accordance with the with the terms of Resolution 466/12 of the National Health Council. These epidemiological data were also provided by LACEN-TO. Informed consent was not required according to resolution 466/12 concerning research involving humans of the National Health Council (Conselho Nacional de Saúde/Ministério da Saúde, Brasília, Brazil, 2012). The study was approved by the Committee of Ethics in Human Research of the Federal University of São Carlos (no. 1.088.936). Permission to conduct the study was also obtained from the Health Department of the State of Tocantins (Secretaria de Saúde do Estado do Tocantins – SESAU) and LACEN/TO.

## RESULTS

## Serratia marcescens Isolates

A total of 54 S. marcescens strains were isolated from 39 ICU and 6 NICU patients' samples at a tertiary hospital located in city of Palmas, Tocantins state. In six patients, 5 from ICU and 1 from NCIU, S. marcescens was isolated in more than one infection site. The prevalence of S. marcescens strains by age group was the following: 0–1 day (12.96%; n = 7), 18–59 years (38.89%, n = 21), 60 years or more (48.15%, n = 26). The median age of patients was 57.0 years (range, 0–93 years). S. marcescens strains were more frequently found in male (68.5%, n = 37)

than in female (31.5%, n = 17) patients (**Figure 1A**). Fortythree samples (79%) were from tracheal aspirate (33%, n = 18), rectal swab, (22%, n = 12), and blood (24%, n = 13) cultures, while 11 (21%) came from wound (9%, n = 5), catheter tip (4%, n = 2), surgical drain (4%, n = 2), sputum (2%, n = 1), and urine (2%, n = 1) cultures (**Figure 1B**). Antibiotic resistance profiles of S. marcescens isolated from the abovementioned different sites showed that all strains were resistant to β-lactams antibiotics. In addition, colistin (CST) and tigecycline (TGC) non-susceptibility pattern of S. marcescens per site of isolation was statistically significant (p < 0.01) in several organs (tracheal aspirate, blood, rectal swab, and wound) when compared with amikacin (AMK), gentamicin (GEN), and ciprofloxacin (CIP) antibiotics (**Figure 1C**).

## Antimicrobial Resistance Profile and Genetic Markers for Antibiotic-Resistance and Virulence Patterns

Serratia marcescens strains showed high-levels of resistance to all β-lactams (100%, n = 54) (TZP, CAZ, CRO, FEP, ETP, IPM, MEM), including high-levels of intrinsic resistance to β-lactams (AMP, SAM, CXM, FOX) (100%, n = 54) and colistin (CST) (96.3%, n = 52). Resistance to tigecycline (TGC) S. marcescens was found in nearly all isolates (92.6%; n = 50). However, for the antibiotics classes fluoroquinolones (CIP) (81.5%, n = 44) and aminoglycosides such as gentamicin (GEN) (81.5%, n = 44), amikacin (AMK) (85.2%, n = 46) (**Figure 2A**), high susceptibility profile was detected. In contrast, MDR was observed in 24.1% (n = 13) of the isolates, and the most common MDR profile was related to β-lactams-glycylcyclineaminoglycosides-quinolone (14.8%, n = 8), followed by βlactams-glicylcycline-quinolone (5.6%, n = 3), and (β-lactamsglicylcycline-aminoglycosides 3.7%, n = 2).

All 54 tested isolates harbored KPC-carbapenemase (blaKPC) and ESBL (blaTEM) genes. The ESBL-encoding genes blaOXA−<sup>1</sup> was detected in 14.8% (8/54), and the blaCTX−M−1group in 16.7% (9/54). However, the blaSHV variants, blaIMP, blaOXA−48, blaNDM, blaVIM, and mcr-1 genes were not detected. The aac(6 0 )- Ib-cr variant gene that can induce simultaneous resistance against aminoglycoside and fluoroquinolone was found in 13 (24.1%) strains. The RND pump efflux encoding genes SdeY and SdeB were identified in all strains while the outer membrane component gene (HasF) was present in 48 (88.9%). Thus, the coexistence of SdeY/HasF genes and SdeB/HasF was observed in 49 (88.9%) strains (**Figure 2B**). Finally, with the exception of one strain (Sm40), the virulence-associated genes PigP, FlhD, ShlA, and PhlA were regularly distributed among S. marcescens strains, which were detected in 98.2% of all strains (**Figure 2B**).

## Resistance Phenotype-Genotype Correlation and Genetic Markers for Virulence Factors

The correlation between the results of phenotypic and genotypic detection and the presence of virulence genes is shown in **Figure 3**.

All isolates carried blaKPC and conferred resistance to all beta-lactam, including carbapenem antibiotics. Furthermore, all detectable bla genes in blaCTX−M−<sup>1</sup> blaOXA−1, and blaTEM group presented ESBL phenotype. Of the 13 isolates with aac(6')- Ib-cr gene, 9 (69.2%) were non-susceptible to gentamicin, 7 (53.9%) to amikacin, and 8 (61.5%) to ciprofloxacin. Among the 49 (88.9%) HasF-positive isolates, 44 (81.5%) were nonsusceptible to tigecycline.

## ERIC-PCR

The ERIC-PCR results indicated that the majority of the isolates presented a rate of genetic similarity above 85% (**Figure 4**). Almost all strains (96.3%) were grouped into a large cluster named B cluster, sharing 86.4% of genetic similarity. In addition, the B cluster was separated into two sub-clusters named B1, with 21 isolates, and B2, with 31 isolates, sharing a genetic similarity of 96.1% and 100%, respectively. Although the cluster B1 presented two subgroups with 4 and 17 isolates, they showed 100% genetic similarity in each one. Interestingly, two strains (Sm38 and Sm40) were grouped separately within the A cluster and presented 71.4% of genetic similarity (**Figure 4**).

## DISCUSSION

Serratia marcescens is a prominent opportunistic pathogen that frequently causes infections in intensive care, surgical and dialysis units (Krishnan et al., 1991; Martineau et al., 2018). In Brazil, there are only few studies on S. marcescens (Ribeiro et al., 2013; Silva et al., 2015). Therefore, we here describe the presence of MDR S. marcescens isolates producing KPCcarbapenemase (blaKPC) and extended spectrum beta-lactamase (blaTEM, blaCTX−M−1group e blaOXA−1, <sup>4</sup>and <sup>30</sup>) in the state of Tocantins, Brazil. Tocantins, located southeast of the Northern Region, is the newest state of Brazil and shares borders with six states presenting intensive migration flow.

Serratia marcescens were isolated mainly from male patients with 60 or more years of age, similarly to previous studies that demonstrated advanced age male patients as presenting a higher risk of contracting S. marcescens infections (Ulu-Kilic et al., 2013; Kim et al., 2015; O'Horo et al., 2017). Samples with higher amounts of S. marcescens were those from tracheal aspirate, followed by blood, rectal swab, and wounds. Our findings corroborate studies by Kim et al. (2015) and Liou et al. (2014) that reported the respiratory tract as the main route of infection for S. marcescens. Other studies have also reported S. marcescens in other sites as bloodstream (Seeyave et al., 2006) and wounds (Us et al., 2017), demonstrating the versatility of these strains in colonizing the host and affecting a wide variety of physiological system.

In addition to the intrinsic resistance to the antibiotics AMP, SAM, CXM, FOX, and CST, we found multidrug-resistant (MDR) S. marcescens isolates to beta-lactam, glycylcycline, and/or aminoglycoside and quinolone group antibiotics. This is in line with other studies that have also reported MDR S. marcescens mainly to beta-lactam, aminoglycoside, and quinolone antibiotics groups (Stock et al., 2003), in hospital

environment (Merkier et al., 2013), and particularly in critically ill patients and neonatal intensive care units (Maragakis et al., 2008). We also observed a significant resistance to colistin and tigecycline in several colonization sites, as shown by a previous study (Silva et al., 2017).

All the S. marcescens strains tested here were resistant to β-lactams, including carbapenems antibiotics. Resistance to carbapenems used to be uncommon among Serratia species (Stock et al., 2003), but many resistant strains have now emerged throughout the world (Lin et al., 2016). Although few studies have related resistance to carbapenems in S. marcescens in Brazil (Milisavljevic et al., 2004; da Costa Guimarães et al., 2013; Ribeiro et al., 2013), Silva et al. (2015) obtained similar results. They have isolated S. marcescens resistant to imipenem, meropenem and ertapenem in samples from different infection sites of ICU patients in another Brazilian locality. Both results are troubling. Infections caused by carbapenem-resistant bacteria often do not respond to conventional treatment (Okoche et al., 2015), as produced carbapenemases hydrolyze not only carbapenems but also penicillins, cephalosporins and monobactams (Queenan and Bush, 2007). The most common carbapenem resistance of S. marcescens in Brazil is due to the production of carbapenamases, especially the KPC-2 type (da Costa Guimarães et al., 2013; Ribeiro et al., 2013), that is encoded by the gene blaKPC−2.

Enterobacteriaceae, such as S. marcescens, have the genes blaTEM−<sup>1</sup> and blaSHV−1; these genes express classical class A betalactamases, encoded by plasmid that hydrolyze first generation penicillins and cephaloporins (Bush, 2010). We found gene blaTEM in all isolates while gene blaSHVvariants was not detected. It is noteworthy that even though S. marcescens also carries the gene blaCTX−<sup>M</sup> (Yu et al., 2003; Kim et al., 2005; Tenover et al., 2013), few of our strains had the gene. Some strains also carried the genes blaOXA−1, <sup>4</sup> and <sup>30</sup>, that have been reported in few studies in Brazil or in other countries, either alone or associated with extended spectrum beta-lactamases genes (ESBL) (blaTEM, blaSHV e blaCTX−M) in S. marcescens strains. Although there are discrepancies in frequency rate and in genotyping of ESBL-producing S. marcescens (Cheng et al., 2006), the observed beta-lactam antibiotic resistance may have also been caused by the genes blaTEM, blaCTX−M, and

FIGURE 2 | (A) Percentage of clinical isolates not susceptible to 16 antibiotics tested. AMP (ampicillin), SAM (ampicillin-sulbactam), TZP (piperacillin-tazobactam), CXM (cefuroximeaxetil), FOX (cefoxitin), CAZ (ceftazidime), CRO (ceftriaxone), FEP (cefepime), ETP (ertapenem), IPM (imipenem), MEM (meropenem), AMK (amikacin), GEN (gentamicin), CIP (ciprofloxacin), TGC (tigecycline), CST (colistin); \* intrinsic resistance to antibiotics. (B) Percentage of genetic markers for resistance and virulence genes. blaKPC, blaTEM, blaCTX−M1, aac(6<sup>0</sup> )-Ib-cr: resistance genes; SdeB, SdeY, HasF: efflux pump and outer membrane component genes; PigP, FlhD, ShlA, PhlA: virulence genes.

blaOXA, since the production of broad-spectrum beta-lactamases enzymes (TEM-1, TEM-2, SHV-1, OXA-1) generate resistant to ampicillin, ticarcillin, piperacillin, piperacillin/tazobactam and cephalosporin antibiotics, and the enzymes CTX-M have hydrolytic activity against cefotaxime (Levy and Marshall, 2004; Sugumar et al., 2014).

FIGURE 3 | Phenotyping and genotyping of Serratia marcescens isolates. Sm represents Serratia marcescens and numbers represent identifications of strains. \*Sm is classified as multidrug resistant (MDR) strains. AMP (ampicillin), SAM (ampicillin-sulbactam), TZP (piperacillin-tazobactam), CXM (cefuroximeaxetil), FOX (cefoxitin), CAZ (ceftazidime), CRO (ceftriaxone), FEP (cefepime), ETP (ertapenem), IPM (imipenem), MEM (meropenem), AMK (amikacin), GEN (gentamicin), CIP (ciprofloxacin). Serratia marcescens are intrinsically resistant to TGC (tigecycline) and CST (colistin). AMP, SAM, CXM, FOX, and CST antibiotics were not included in the MDR classification. The blaIMP, blaOXA−48, blaNDM, blaVIM, blaSHV variants and mcr-1 genes were not detected. Blue box correlates with AMP, SAM, TZP, CXM, FOX, CAZ, CRO, FEP, ETP, IPM, MEM. Green box correlate with AMK, GEN, CIP. Multicolored box (HasF) correlates with CIP and TGC. Brown box shows number of S. marcescens caring virulence genes.

In our study, most strains of S. marcescens were only sensitive to aminoglycosides (gentamicin and amikacin) and fluoroquinolone (ciprofloxacin). Aminoglycosides are the oldest antibiotics that have been used less frequently in the last years, thus possibly preserving activity against some resistant bacteria that cause difficult to cure infections (Falagas et al., 2008; Gad et al., 2011). The observed low resistance to ciprofloxacin (18.18%) is in agreement with the results obtained by Sheng et al. (2002) who observed 20–30% resistance to quinolone in S. marcescens isolates. However, it is important to consider that S. marcescens is highly adaptable, so rates of resistance to fluoroquinolones diverge considerably among institutions (Young et al., 1980; Mahlen, 2011; Sader et al., 2014), including within Brazilian ones.

Resistance to fluoroquinolones may be caused by alterations in the target enzymes DNA gyrase and topoisomerase IV, and by acquisition of the transferable plasmid-mediated quinolone resistance (PMQR) determinants qnr, qepA, aac(6')-Ib-cr, and oqxAB (Veldman et al., 2011; Poirel et al., 2012; Moradigaravand et al., 2016). The gene aac(6<sup>0</sup> )-Ib-cr, a variant gene of the aminoglycoside acetyltransferase, was also present in most of strains that presented fluoroquinolone and/or aminoglycoside resistance. This finding is consistent with others studies that have shown that aac(6<sup>0</sup> )-Ib-cr may be associated with antibacterial resistance against fluoroquinolone and aminoglycoside (Kim et al., 2009, 2011) antibiotics.

Three RND-type efflux have been reported in S. marcescens, namely SdeAB (Kumar and Worobec, 2005a), SdeCDE (Kumar and Worobec, 2005a; Begic and Worobec, 2008), and SdeXY (Chen et al., 2003). SdeAB and SdeXY interact with HasF (an outer membrane component, TolC homolog gene) contributing to resistance against a wide variety of antimicrobial agents

# multidrug resistance (MDR) pattern.

(Begic and Worobec, 2008; Hornsey et al., 2010). Although we did not analyze the genes SdeA e SdeX, the genes SdeB and SdeX were present in all strains and the gene HasF was also found in most strains. Drug extrusion by efflux pumps as SdeAB-HasF comprises one of the main mechanisms for fluoroquinolones antibiotic resistance (Dalvi and Worobec, 2012). Additionally, Hornsey et al. (2010) has shown the intrinsic activity of the SdeXY-HasF efflux pump is responsible for the lower susceptibility to ciprofloxacin. Thus, in addition to the presence of the genes aac(6<sup>0</sup> )-Ib-cr [associated with plasmid-mediated quinolone resistance (PMQR)], the genes SdeB, SdeY, and HasF that encode RND-type efflux pump may have contributed to the observed ciprofloxacin resistance.

Our data shows high prevalence of tigecycline-resistant S. marcescens strains and SdeY and HasF genes. The reduced sensitivity of S. marcescens to tigecycline may be related to the

up-regulation of the SdeXY-HasF efflux pump (Hornsey et al., 2010). Although the gene SdeX was not analyzed, our findings strongly suggest that these genes may be responsible for the high tigecycline resistance.

Many bacteria produce virulence factors as hydrolytic enzymes and toxins that enable host invasion, bacterial proliferation and inhibit host defense mechanisms, sometimes resulting in host death (Aggarwal et al., 2017). S. marcescens strains were also evaluated for the presence of toxin genes ShlA and phlA, and all but the Sm40 strain carried these genes. Our findings are in agreement with other studies that reported the presence these genes in S. marcescens (strain SEN) (Aggarwal et al., 2017). FlhDC has been proposed as a regulator controlling flagellum biogenesis, biofilm formation, cell septation and expression of virulence factors during swarming (Givskov et al., 1995; Fraser and Hughes, 1999; Chilcott and Hughes, 2000; Lin et al., 2010). In our study, the presence of FlhC was not analyzed but the gene FlhD was present in almost all isolates. S. marcescens produces biosurfactant serrawettin and red pigment prodigiosin used in surfaces colonization (Hejazi and Falkiner, 1997). PigP is a positive regulator of prodigiosin production that regulates swarming and hemolysis through serratamolide production (Fineran et al., 2005; Shanks et al., 2013). In our study, PigP gene was found in almost all isolates (98.15%). Overall, our results suggest that the combination of these virulence genes could have contributed to the pathogenicity of S. marcescens strains.

The dendrogram based on ERIC-PCR fingerprint analysis demonstrated that the vast majority of isolates are closely related, sharing a genetic similarity of 86.4% (except for two strains). In addition, many isolates showed 100% similarity to each other. A study conducted by Ferreira-Firmo et al. (2019) evaluated nine S. marcescens from different clinical sources and three hospitals in Northeast Brazil showed a greater genetic diversity among the studied strains. Lin et al. (2016) studied 83 carbapenem-resistant S. marcescens isolates recovered from Zhejiang Provincial 501 People's Hospital, China, from which they found 63 blaKPC-2 positive strains sharing nine different profiles. Our results demonstrate the predominance of few genetic profiles grouped together, with similarity above 85%, indicating that, although bacteria have been isolated from different patients and devices, the circulating S. marcescens in this hospital is highly genetically related.

Our study has some limitations worth noting. There was a high number of S. marcescens isolated from rectal swabs and tracheal aspirate, and both cultures are recommended for surveillance in ICU and NICU patients. However, microbiological reports sent to LACEN were not clear regarding how many of these samples were analyzed for both clinical and surveillance purposes. This study was further limited by the duration of the research, which was relatively short (2014–2015), and by conventional phenotypic and genotypic techniques, that have their own particular strengths and limitations in detecting MDR strains. Nonetheless, we intend to extend the analysis period of S. marcescens isolates, and whole genome sequencing to type relevant MDR strains must be performed. We expect that this study broadens our understanding on epidemiology, antibiotic resistance, and putative virulence factors of these strains, and provides relevant information for the prevention and management of S. marcescens infections.

In conclusion, S. marcescens represents a problem for public health (Kurz et al., 2003) and the resistance pattern exhibited by clinical isolates along with the transmission to other clones show the importance of researching factors associated with the increase in frequency and/or emergence of infections caused by S. marcescens in Brazilian hospitals. The occurrence of antibiotic-resistant bacteria considerably varies according to country, region and susceptible population, and the mitigation of this problem in ICUs is especially associated with actions to control the spreading of such drugresistant bacteria (Ko et al., 2002; Hou et al., 2015). Thus, it is crucial to eliminate sources of resistance development and associated reservoirs as well as to overtake standardized sanitation and enforce mandatory notification to gather important data for continuous risk assessment evaluation and effective decision making to control these species in hospital environments.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the GenBank database. Genes and their respective accession numbers: blaCTX – MK576103; blaKPC – MK576104; blaOXA – MK576105; blaTEM – MK576106; SdeB – MN583232; SdeY – MN583233; HasF – MN583234; aac(6<sup>0</sup> )-Ib-cr – MN583235; FlhD – MN583236; PigP – MN583237; ShlA – MN583238; PhlA – MN583239). Access to genetic heritage was approved by the National System for the Management of Genetic Heritage (SisGen n ◦ AFF27ED).

## ETHICS STATEMENT

The studies involving human participants were reviewed and approved by the Committee of Ethics in Human Research of the Federal University of São Carlos (no. 1.088.936). In this work, all Serratia marcescens and the anonymous archival data related patient age, gender, and sample type were obtained from the Central Laboratory of Public Health of Tocantins (LACEN/TO, data's owner). Permission to conduct the present study was obtained from the Health Department of the State of Tocantins (Secretaria da Saúde do Estado do Tocantins – SESAU) and LACEN/TO. Patient consent was not required, since the data presented in this study do not relate to any specific person or persons. Written informed consent from the participants' legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.

## AUTHOR CONTRIBUTIONS

fmicb-11-00956 May 19, 2020 Time: 16:32 # 11

RF, GR, MD, MO-S, MB, and M-CP performed the experiments. FG and EL aided with statistical analysis. EC aided with the sequencing analysis and the sequence submission to the NCBI platform. M-CP conceived and supervised the study. M-CP, AC, IM, and AP-S wrote the manuscript and analyzed the data. MO-S performed ERIC-PCR.

## FUNDING

This work was supported by the Fundação de Amparo a Pesquisa do Estado de São Paulo (FAPESP grants 2020/05699- 6 to M-CP and FAPESP grant 2013/22581-5 to AP-S) and Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, grant 2013/485873 to M-CP). This study was partially financed by the Fundação de Amparo á Pesquisa do Estado de São Paulo-Brazil (FAPESP) as a fellowship to MD (FAPESP fellowship 2018/24213-7), and by

## REFERENCES


the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brazil (CAPES) – Finance Code 001 as a fellowship to GR, FG, and EC.

## ACKNOWLEDGMENTS

The authors thank the Laboratório Central de Saúde Pública do Tocantins, Palmas, TO, Brazil (LACEN-TO) who kindly provided the S. marcescens strains, and to Secretaria de Saúde do Estado do Tocantins (SESAU-TO) for facilitating the development of project.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.00956/full#supplementary-material

ß-lactamases in Enterobacteriaceae. J. Antimicrob. Chemother. 65, 490–495. doi: 10.1093/jac/dkp498


components in the control of prodigiosin and carbapenem antibiotic production. Mol. Microbiol. 56, 1495–1517. doi: 10.1111/j.1365-2958.2005.04 660.x


marcescens clones within the United Kingdom and Ireland. Genome Res. 26, 1101–1109. doi: 10.1101/gr.205245.116.Freely



Zingg, W., Soulake, I., Baud, D., Huttner, B., Pfister, R., Renzi, G., et al. (2017). Management and investigation of a Serratia marcescens outbreak in a neonatal unit in Switzerland - the role of hand hygiene and whole genome sequencing. Antimicrob. Resist. Infect. Control 6:125. doi: 10.1186/s13756-017- 0285-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 Ferreira, Rezende, Damas, Oliveira-Silva, Pitondo-Silva, Brito, Leonardecz, Góes, Campanini, Malavazi, da Cunha and Pranchevicius. 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.

# Prediction of Antimicrobial Resistance in Gram-Negative Bacteria From Whole-Genome Sequencing Data

Pieter-Jan Van Camp1,2, David B. Haslam3,4 and Aleksey Porollo2,4,5 \*

<sup>1</sup> Department of Biomedical Informatics, University of Cincinnati, Cincinnati, OH, United States, <sup>2</sup> Division of Biomedical Informatics, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, United States, <sup>3</sup> Division of Infectious Diseases, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, United States, <sup>4</sup> Department of Pediatrics, University of Cincinnati, Cincinnati, OH, United States, <sup>5</sup> Center for Autoimmune Genomics and Etiology, Cincinnati Children's Hospital Medical Center, Cincinnati, OH, United States

Background: Early detection of antimicrobial resistance in pathogens and prescription of more effective antibiotics is a fast-emerging need in clinical practice. High-throughput sequencing technology, such as whole genome sequencing (WGS), may have the capacity to rapidly guide the clinical decision-making process. The prediction of antimicrobial resistance in Gram-negative bacteria, often the cause of serious systemic infections, is more challenging as genotype-to-phenotype (drug resistance) relationship is more complex than for most Gram-positive organisms.

#### Edited by:

Leonardo Neves de Andrade, University of São Paulo, Brazil

#### Reviewed by:

Brock Aaron Arivett, Middle Tennessee State University, United States Zhi Ruan, Zhejiang University, China

## \*Correspondence:

Aleksey Porollo Alexey.Porollo@cchmc.org

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 30 January 2020 Accepted: 24 April 2020 Published: 25 May 2020

#### Citation:

Van Camp P-J, Haslam DB and Porollo A (2020) Prediction of Antimicrobial Resistance in Gram-Negative Bacteria From Whole-Genome Sequencing Data. Front. Microbiol. 11:1013. doi: 10.3389/fmicb.2020.01013 Methods and Findings: We have used NCBI BioSample database to train and crossvalidate eight XGBoost-based machine learning models to predict drug resistance to cefepime, cefotaxime, ceftriaxone, ciprofloxacin, gentamicin, levofloxacin, meropenem, and tobramycin tested in Acinetobacter baumannii, Escherichia coli, Enterobacter cloacae, Klebsiella aerogenes, and Klebsiella pneumoniae. The input is the WGS data in terms of the coverage of known antibiotic resistance genes by shotgun sequencing reads. Models demonstrate high performance and robustness to class imbalanced datasets.

Conclusion: Whole genome sequencing enables the prediction of antimicrobial resistance in Gram-negative bacteria. We present a tool that provides an in silico antibiogram for eight drugs. Predictions are accompanied with a reliability index that may further facilitate the decision making process. The demo version of the tool with pre-processed samples is available at https://vancampn.shinyapps.io/wgs2amr/. The stand-alone version of the predictor is available at https://github.com/pieterjanvc/ wgs2amr/.

Keywords: antimicrobial resistance, antibiotic resistance, whole-genome sequencing, machine learning, prediction, genotype-phenotype relationship

**Abbreviations:** AB, antibiotic; ABR, antibiotic resistance; ARG, antibiotic resistance gene; ARGC, antibiotic resistance gene cluster; MCC, Matthew's correlation coefficient; MIC, minimum inhibitory concentration; WGS, whole genome sequencing.

## INTRODUCTION

fmicb-11-01013 May 22, 2020 Time: 16:51 # 2

Since the discovery and widespread use of antibiotics (AB) in the early 20th century, resistance to those same AB has generally developed rapidly; often even within the first years of introduction (Marston et al., 2016). As a consequence, many bacteria have developed antibiotic resistance (ABR) to most of the major classes of AB, often seen in the Gramnegatives (Centers for Disease Control and Prevention, 2018). Effective treatment of these infections requires knowledge of the organism's susceptibility to the various AB, currently obtained by culturing bacteria in the clinical laboratory and subsequent testing for commonly used AB. Depending on the pathogen, this process may require 72 h or more. Drug susceptibility is usually reported to the clinician as either resistant or susceptible (sometimes intermediate is also used) with cut-offs based on the minimum inhibitory concentration (MIC) of an AB needed to halt growth or kill the pathogen in the lab. In serious systemic infections, early treatment with an effective antibiotic is paramount as unexpected resistance may lead to treatment failure, while fear of inadequate therapy may drive overly broad antibiotic use which contributes to extensively resistant and potentially untreatable bacteria (Marston et al., 2016).

With decreasing cost and increasing speed of high-throughput sequencing technology such as whole genome sequencing (WGS), in-depth analysis of pathogens is increasingly used in clinical decision-making. Studies already showed the potential of these techniques in ABR prediction in single, Gram-positive pathogens like Staphylococcus aureus and Mycobacterium tuberculosis. An example is the Mykrobe predictor that maps DNA sequencing data to a reference genome and a set of plasmid genes conferring ABR (Bradley et al., 2015). The model also accounts for polymorphism in select loci when predicting drug resistance. Data analysis and prediction is rapid, enabling it as a practical tool for clinical care during the decision-making process. PhyResSE is another tool that follows a similar strategy but may process data in minutes to a few days, attributing such time extension to more careful variant calling (Feuerriegel et al., 2015). Both tools report high accuracy of ABR prediction. However, their application to other pathogens like Gram-negative bacteria has not been described.

The prediction of ABR in Gram-negative bacteria, often the cause of serious systemic infections, is more challenging as the source of drug resistance is more complicated. For example, Gram-negative pathogens may possess one or more βlactamases with similar amino acid sequences but various activity against β-lactam-based AB (Livermore, 1998). They are also more likely to develop mutations that result in lower membrane permeability, or increase the expression of a variety of genes for excreting xenobiotics (efflux pumps) and for inactivating β-lactam-targeting drugs (Marston et al., 2016). There have been a few clinical predictors reported for assessing the risk of infection with resistant Gram-negative bacteria but they solely rely on data from the electronic health record and just predict the likelihood of infection with a resistant strain rather than individual drug resistance (Martin et al., 2013; Vasudevan et al., 2014). Thus far, only a handful of studies were published on the prediction of Gram-negative resistance using WGS data. These models were built for individual species, such as Neisseria gonorrhoeae or Klebsiella pneumonia, and were based on small sample sizes resulting in poor predictive accuracy (Eyre et al., 2017; Nguyen et al., 2018). A more extensive study recently published by Drouin et al. (2019) utilizes 107 machinelearning-based models, trained on WGS data, to predict ABR in 12 bacterial species, including six Gram-negatives, against a variety of AB with generally high accuracy. These models were trained without incorporating prior ABR knowledge as they were based on nucleotide k-mers from sequenced genomes of these pathogens, thus utilizing information from non-coding regions and polymorphism. Furthermore, their methodology produced small, human-interpretable decision trees, when the k-mers are mapped back to the respective genomes to make decisions interpretable (provided that k-mers belong to annotated genomic regions). However, the study was performed on heavily imbalanced datasets, while reporting only two-class accuracy, which could overestimate the real performance. Finally, isolates with intermediate resistance were excluded from the final predictors making the performance estimates on such samples uncertain. For more details on the bioinformatics approaches to the AMR analysis and prediction, the reader can refer to a recent review (Van Camp et al., 2020).

In this work, we present a machine learning-based method for the fast estimation of ABR in 5 Gram-negative species for 8 AB. The models were trained on WGS data and laboratory confirmed drug susceptibility. All isolates were assigned to either of two classes (susceptible or resistant) and subsequently used in the training and validation of binary predictors. Each model was evaluated using multiple performance measures. The presented workflow could inform early clinical decision-making on the choice of AB therapy (within a day) while waiting for the final antibiogram (typically 2–4 days) thereby decreasing the time to start effective therapy. A web-based demo application to show the potential clinical implementation is publicly available at https://vancampn.shinyapps.io/wgs2amr/. The stand-alone tool can be freely downloaded from https://github.com/pieterjanvc/ wgs2amr/.

## MATERIALS AND METHODS

## Pathogens and Antibiotics of Interest

This study focused on five common nosocomial Gram-negative organisms that can cause sepsis (Vincent, 2003; Couto et al., 2007): Acinetobacter baumannii, Escherichia coli, Enterobacter cloacae, Klebsiella aerogenes, and Klebsiella pneumoniae. These specific Gram-negative pathogens were chosen because, in addition to clinical relevance, they are the most represented in terms of DNA-sequenced and AMR annotated samples available at NCBI. For machine learning, the larger the dataset to train on, the more accurate and generalized model can be achieved. E. cloacae is the major representative of Enterobacter species and therefore part of ESKAPE pathogens. Their antibiotic susceptibility was evaluated for cefepime, cefotaxime, ceftriaxone, ciprofloxacin, gentamicin, levofloxacin, meropenem, and tobramycin. This panel of AB covers those most commonly used to treat Gram-negative bacterial infections. Consequently, these AB are most frequently tested for susceptibility against bacterial isolates.

## Public Data Collection

fmicb-11-01013 May 22, 2020 Time: 16:51 # 3

Meta-data for 6564 bacterial samples (isolates) were retrieved from the NCBI BioSample database using the "antibiogram" keyword filter. Of these, 4933 samples had the required information, such as bacterium name, antibiogram, and sequencing data accession number. For this work, all "intermediate" ABR levels were converted to "resistant" to project data to a binary classification problem (i.e., resistant versus susceptible). The list was subsequently refined to only include the bacteria and AB of interest (see section "Pathogens and antibiotics of interest") resulting in 2516 samples. Given the resistance to AB was highly imbalanced in the data (mostly skewed toward resistant phenotype), the samples were randomly chosen so that the number of susceptible, and resistant isolates for each antibiotic was as equal as possible in order to balance the input for machine learning models. This resulted in a final total of 946 samples (**Supplementary Table S1**). Of these, 3% of samples available for each species (total n = 31) were set aside to create a demo dataset to showcase the online application (see section "Preliminary pipeline implementation and evaluation" for details). The remaining 915 samples were used to build and evaluate eight XGBoost-based models, where available data for each antibiotic were randomly split in 70% training and 30% testing subsets. The overall flow of data collection is summarized in **Figure 1**. The counts of samples per species include: A. baumannii – 256; E. cloacae – 67; E. coli – 330; K. aerogenes – 51; and K. pneumoniae – 211. Of note, we did not stratify samples by different bacterial species during the model training as we intended our models to be species independent. **Table 1** shows the distribution of the 915 samples through the AB of interest.

Whole genome sequencing data for all samples were retrieved from the NCBI Sequence Read Archive (SRA) using the SRA toolkit (SRA Toolkit Development Team, 2019). In case multiple runs (SRR) of a sample (SRS) were available (e.g., sample was run through different sequencers or with different settings), the file TABLE 1 | Summary of the 915 samples used to build and evaluate antimicrobial resistance prediction models.


\*Note that not all samples were tested for every antibiotic, thus the counts per AB do not add up to 915.

with the smallest size was selected. Smaller file size may frequently be attributed to lower sequencing depth and/or shorter shotgun reads. We reasoned that, if a model is able to make predictions on smaller sequence files, it will most likely be applicable to larger files with better gene coverage. List of all samples used can be found in **Supplementary Table S1**.

## In-House Dataset of Samples

Gram-negative bacterial pathogens were collected from blood or urine of patients hospitalized at Cincinnati Children's Hospital Medical Center (CCHMC), 19 samples total. Organisms were identified to the species level and antimicrobial susceptibility testing was performed using the VITEK <sup>R</sup> 2 machine (Biomerieux) in the Diagnostic Infectious Diseases Testing Laboratory at CCHMC. Samples represent A. baumannii (n = 1), E. coli (n = 11), K. aerogenes (n = 2), and K. pneumoniae (n = 3). No in-house samples are available with E. cloacae. Two available samples of Klebsiella oxytoca are included. DNA was extracted from overnight liquid broth cultures using the QIAamp PowerFecal DNA Kit (Qiagen Inc, Germantown, MD, United States). Sequencing libraries were generated using the Nextera XT kit (Illumina Corporation, San Diego, CA, United States). Pooled libraries were sequenced on a NextSeq 500 (Illumina Corporation, San Diego, CA, United States) in the Microbial Genomics and Metagenomics Laboratory at

CCHMC using paired 150 bp reads to a depth of approximately 5 million reads per sample. Sample collection was approved by the Institutional Review Board (IRB) at CCHMC (IRB approval # 2016–9424: Molecular Epidemiology of Bacterial Infections). The in-house samples are available at the NCBI BioSample database (BioProject ID: PRJNA587095), where detailed metadata can be found (see **Supplementary Table S1** for sample IDs). Of note, the antimicrobial susceptibility testing with VITEK takes at least 72 h and generally requires a pure isolate, whereas sequencing preparation followed by the WGS data analysis can be completed under 48 h and does need to not rely on a pure colony (Scaggs Huang et al., 2019).

## Analysis of WGS Data

The sequencing files in the FASTQ format were aligned to 4579 antibiotic resistance genes (ARG) found in the NCBI Bacterial Antimicrobial Resistance Reference Gene Database (NCBI Accession: PRJNA313047, version as of September 26, 2018) using DIAMOND (Buchfink et al., 2015), a greatly improved version of BLASTx (Altschul et al., 1990) with regards to speed, and on par sensitivity. Given the same ARG can be present in multiple species (e.g., plasmid DNA), no filtering was performed based on original bacterium in which the ARG was sequenced. The default settings of DIAMOND were used to map shotgun reads to ARGs.

Antibiotic resistance genes coverage (C) within a sample was quantified as follows. First, reads aligned by DIAMOND were discarded if the alignment to the gene covered <90% of the read's length or yielded <90% sequence identity. Second, a gene was discarded if all aligned reads covered <90% of its length (i.e., protein sequence). The cutoff was chosen as being suboptimal after probing 70, 80, 90, and 100% coverage (**Supplementary Table S2**). If a gene was retained in the hit list, the number of alignments (n) was adjusted for gene length (i.e., the number of amino acids, L), and sequencing depth (i.e., total number of reads in the file divided by 10<sup>7</sup> , D). Such scaled data (Eq. 1) are better suitable for the machine learning.

$$\mathcal{C} = \frac{n}{L \times D} \tag{1}$$

## Clustering Similar Antibiotic Resistance Genes

Many of the ARG in the NCBI database have a high sequence similarity (e.g., polymorphism in strains or sequences derived from closely related species) that cannot be discriminated by the alignment techniques used in this study. Therefore, antibiotic resistance gene clusters (ARGC) were created using the cluster\_fast module of the USEARCH algorithm (Edgar, 2010) to group genes with ≥90% sequence identity together, naming them after the most representative gene as defined by USEARCH (i.e., cluster centroid). This threshold for grouping was chosen as a suboptimal compromise between balancing the number of genes per cluster, performance of the model, and the biological relevance of the genes grouped together (**Supplementary Table S3**). For cutoffs lower 85%, genes from different ABR classes started to group together, hence such cutoffs were excluded from consideration. The clustering resulted in the reduction of the potential input space for the machine learning models from 4579 ARG to 1027 ARGC, with 410 clusters (40%) consisting of just a single gene. To represent the coverage of each ARGC, the average coverage (Eq. 1) of all ARG detected in this ARGC was taken. Finally, of the 1027 ARGC, only 152 were found in our data and thereby used for subsequent machine learning.

## Building and Evaluating Machine Learning Models

Regression models are less stable on datasets where the input space is large, sparse, and the features are correlated (e.g., in the context of this work, drug resistance may be exerted by multiple ARGC; Farrar and Glauber, 1967; Devika et al., 2016). Using penalized regression (e.g., LASSO and Ridge regression) to reduce both the input space and select most important features can help increase performance of the model, but it still operates on the premise that input features are uncorrelated. In correlated datasets, feature selection will be distorted in this process resulting in less reliable model interpretation. Decision trees, on the other hand, inherently perform better in such cases as correlation does not influence the feature selection process (Piramuthu, 2008). In random forest models, hundreds to thousands of these trees are built, each with different subset of the input space, resulting in a more robust reporting of important features. Neural networks (NN), especially their currently popular application to deep learning, and require much larger training data-sets (tens of thousands to millions of input vectors/samples) than currently available for antimicrobial resistance (hundreds of samples) in order to demonstrate benefits of deep learning. Moreover, the resulting NN-based models represent a black box that would be difficult to dissect in order to see the decision making rules and factors influencing the decision (Van Camp et al., 2020).

XGBoost is an extreme gradient booster for decision trees that is capable of handling correlated inputs. It has innate support for sparse datasets (in our case, only a handful of ARGC are present in each sample) and can extract important features to provide additional insights in the decision-making process (Chen and Guestrin, 2016). Although XGBoost supports multiclass classification (i.e., model can choose between more than two classes), our samples can have resistance to multiple AB at the same time (i.e., multi-labeling classification), which is not supported and thus a separate, independent binomial model (resistance versus susceptible) was created for each antibiotic of interest (8 models total).

The input for each model was the list of ARGC and their presence (C > 0) or absence (C = 0) in each sample (Eq. 1). The output was binomial with label resistant (= 1) or susceptible (= 0) to the antibiotic of interest. The XGBoost models were trained with a learning rate of 0.1, maximum tree depth of 2, training subsampling of 0.8, and column subsampling of 0.8. All other parameters were kept default. The algorithm was run for a maximum of 300 iterations, but early stopping was done when no improvement was seen in 50 consecutive iterations using 10-fold cross-validation.

Despite the efforts to balance the number of samples with susceptible and resistant phenotypes per drug the final distributions on individual AB remained unbalanced (**Table 1**) because each sample was not tested for all drugs of interest. For two most imbalanced drugs, cefotaxime and ceftriaxone, an under-sampling was performed to reach 3:1 ratio and prevent the model overfitting toward the over-represented class. For heavily class-imbalanced data, standard performance measures like sensitivity or 2-class accuracy could overestimate the true performance. The Matthew's correlation coefficient (MCC) was therefore used as a more stringent performance statistic (Eq. 2). MCC measures binary classification in unbalanced datasets with range from -1 (inverse prediction) through 0 (random prediction) to 1 (perfect prediction).

$$\text{MCC} = \frac{TP^\*TN - FP^\*FN}{\sqrt{(TP + FP)(TP + FN)(TN + FP)(TN + FN)}} \quad \text{(2)}$$

where TP, TN, FP, and FN are true positive, true negative, false positive, and false negative instances, respectively.

For complete performance evaluation, two-class accuracy (Acc, Eq. 3), sensitivity (recall, R), precision (P), specificity (Sp), areas under ROC (AUC), and precision-recall (PR-AUC) curves are also provided.

$$Acc = \frac{TP + TN}{TP + FP + TN + FN} \tag{3}$$

where Acc is a two-class accuracy; TP, TN, FP, and FN are the same as in Eq. 2.

## Reliability Index

To provide an additional assessment how certain the prediction is by a given model, we introduce a reliability index (RI). The RI is based on the observation that in classification models values closer to extremes (0 or 1) are more likely to yield a correct prediction compared to values hovering around 0.5. Using adjusted model output (AMO, Eq. 4), we computed a misclassification rate (MR, Eq. 5) for every AMO in the test subset of each model, defined as the percentage of incorrect predictions in test cases with AMO equal or higher than a given cutoff (Eq. 5). A regression model was fit to this MR distribution for each drug and then was used to calculate the MR for new model outputs (**Supplementary Figure S1**). The RI is the inverse of the MR and simply defined as 1 – MR.

$$AMO\left(\mathbf{x}\right) = \begin{cases} \begin{array}{c} \mathbf{x} \\ 1 - \mathbf{x} \end{array}, \mathbf{x} \ge \mathbf{0.5} \end{array} \right\} \tag{4}$$

$$MR\left(\mathbf{c}\right) = \frac{\mathbf{f}\mathbf{p} + \mathbf{f}\mathbf{n}}{\mathbf{t}\mathbf{p} + \mathbf{f}\mathbf{p} + \mathbf{t}\mathbf{n} + \mathbf{f}\mathbf{n}}\tag{5}$$

where tp, fp, tn, and fp are the number of true positive, false positive, true negative, and false negative instances, respectively, predicted with AMO ≥ c.

## Feature Importance

XGBoost, being a random forest-based algorithm, can provide important features from the model once it has been built in order to evaluate the individual feature impact in the decision-making process. In our case, XGBoost lists the most important ARGC for each model. By design, random-forest-based algorithms ignore strongly correlated features while using only one in the model, as adding redundant features will not provide extra discrimination capabilities (see section "Building and evaluating machine learning models"). From a biology standpoint, however, it is interesting to know all the ARGC that occur in high frequency. Thus, when the most important features are extracted from the models, we reviewed the correlated features ARGC as well.

In consideration that organism(s)/strain(s) composition in the sample should not be known a priori, whereas de novo genome assembly may be inefficient and inaccurate, no genome assembly from WGS data is conducted in this work. Therefore, the prediction model is agnostic to the source of the detected antimicrobial gene, as to whether it is inherent to an organism or acquired via mobile genetic element (plasmid). Hence, no weighting scheme for plasmid-derived genes was considered for the model.

## Preliminary Pipeline Implementation and Evaluation

The long term goal of this project is to build a platform where prediction models like the ones presented here can be used in research or clinical practice to quickly estimate a bacterium's antibiogram from WGS data with sufficient accuracy, in order to inform early decisions about the correct AB use while awaiting the final antibiogram.

For illustration of developed models, an R-Shiny web-based application was developed where pre-processed samples from two datasets unseen in training (31 public, demo dataset, and 19 in-house samples, sections "Public data collection" and "Inhouse dataset of samples") can be individually submitted to the prediction models and subsequently compared to their actual ABR status. Furthermore, the application allows the user to explore the performance of the current ABR models in more detail, and review the important genes used in the decisionmaking process.

## RESULTS

## Data Collection and Pre-processing

After the processing and filtering (sections "Analysis of WGS data" and "Clustering similar antibiotic resistance genes"), of the 4579 ARG in the NCBI database 2605 (57%) were detected in at least one of the 946 samples. When clustered, only 152 ARGC (15%) were detected in the whole dataset. The median number of ARGC present in any sample is 10 resulting in a very sparse dataset. **Table 2** lists the most frequently found ARGC per species.

To review correlation between the 152 ARGC, a hierarchical clustering using Ward's algorithm (Ward, 1963) was performed (**Figure 2A**). An example of the strong correlation between several specific ARGC can be seen in **Figure 2B**. Even though some of the clusters have similar names, they represent different subtypes of the same gene (i.e., different enough in sequence to be placed in separate clusters, section "Clustering



\*The occurrence refers to the percentage of samples, stratified by species, where a given ARGC is found in the WGS data.

similar antibiotic resistance genes"). The full table with all ARGC pairwise correlation values can be found in supplements (**Supplementary Table S4**).

## XGBoost Model Training and Testing

Since sparse input in machine learning models can bias performance (both under- or overestimating) depending on the split in training and testing data (Wu et al., 2013), we trained 51 independent models for each antibiotic (polling isolates of all species together). Each model was trained and validated with a different split in order to see the performance distribution (refer to **Supplementary Figure S2** for the overall flow of model training). **Figure 3** shows the distribution of AUC based on the testing subset for each individual split. The model with the median performance over all 51 splits for each antibiotic (represented by the thick line in the boxplots) is assumed to be the closest to real-life performance (i.e., the least biased) and was chosen as the final model. The performance of these final models are detailed in **Table 3** and **Figure 4** (see **Supplementary Table S5** for the performance of all other models).

## Feature Importance

Regardless of fluctuations in model performance based on split in training and testing, the important features ARGC extracted from the models appeared to be largely the same per antibiotic prediction. **Table 4** shows the 5 most important ARGC (on average over the 51 models) for each ABR prediction model. The full table can be found in the supplements (**Supplementary Table S6**).

As mentioned in section "Feature importance", once a feature is chosen for the use in the decision-making, random forest-based methods often ignore other highly correlated features as they do not contribute to class discrimination. However, in the context of this study, when unused ARGC may provide additional biological


insights, we list all ARGC, chosen by the models and those correlated, in supplemental materials (**Supplementary Table S4**).

## Comparison With Other Algorithms

XGBoost models were compared with those based on LASSO and Ridge regression, which also can deal with data sparsity and have built-in feature selection (**Supplementary Table S7**). In 6–7 (depending on the metric to compare) out of 8 AB, XGBoost models appear to be slightly better than linear regression models. The real advantage of XGBoost over the linear regression algorithms is the robustness in the important feature selection. **Table 5** in conjunction with **Supplementary Table S6** demonstrate that XGBoost yields the best consistency in selecting top informative features compared to LASSO and Ridge regression models.

## Reliability Indexes

**Figure 5** shows distributions of the RI for each final model based on the corresponding testing set. While there is no unified cutoff for RI across all models, there is a clear trend that correct predictions tend to have a higher RI.

## Practical Implementation of Models

Demo dataset (section "Preliminary pipeline implementation and evaluation") is used to illustrate how new unseen samples could be run through the pipeline of preprocessing and subsequent prediction by the 8 models (**Figure 6**). For a given sample, each antibiotic is assigned a binary resistance prediction with a confidence (reliability index). For demonstration purposes, current implementation retrieves meta-data information for a given sample, such as species name (a header of the table), and known drug resistance status (last colored column).

**Figure 6** shows how prediction results on novel samples could be presented to the clinician in the format of an antibiogram. The online application provides a more intuitive way to explore the results and use of this pipeline. The predicted antibiograms from all extra samples (the Demo set based on public samples and the In-house dataset) can be explored in detail, and an additional tab (not shown) provides more information about the models and the important features. The summary of all predictions for Demo and In-house datasets can be found in **Figure 7**. The web-based demonstration and stand-alone versions of the application can be found at https://vancampn.shinyapps.io/wgs2amr/ and https: //github.com/pieterjanvc/wgs2amr/, respectively.

## DISCUSSION

fmicb-11-01013 May 22, 2020 Time: 16:51 # 7

This work provides a framework wherein bacterial samples can be tested quickly to obtain a preliminary antibiogram to guide initial antibiotic selection for treatment. Culturing bacteria and getting a full antibiogram can take 2–4 days, whereas WGS and the computational pipeline presented here only takes around a day. At present, sequencing is taking up the majority of time but is likely to decrease significantly with the improvements in sequencing technologies.

Presenting the results as an early antibiogram estimate (**Figure 6**) instead of just individual predictions, provides clinicians with a clear and intuitive way to inform the choice of the otherwise empiric initial AB. The general resistance pattern of the whole antibiogram can be informative in itself, even if there may still be errors in individual predictions. The latter is further aided by the addition of the RI that indicates how certain the models are on individual predictions. While there is not a single clear cut for the RI, **Figures 5**, **7** suggest that the majority of samples assessed with a high RI appear to have correct predictions. All of this helps early, informed AB choice that can decrease the time to start effective AB therapy while limiting the use of empiric broad-spectrum AB and slowing the development of new resistant strains. The predictions could become especially helpful in settings where resources do not permit the use of a full microbiology lab (e.g., in developing countries). Upcoming technologies, like the Nanopore MinION, (Z) (2019), will allow clinicians in the near future to sequence pathogens with smaller portable devices. This, coupled with analytical pipelines like the one presented here, could provide valuable information on pathogen resistance that would otherwise not be available. Regardless, the results will require clinical judgment.

All pathogens have several ARGC that are found in nearly every sample, regardless of its resistance status to the tested AB (e.g., class C extended-spectrum beta-lactamase EC-18 cluster was detected in 99% of E. coli isolates, **Table 2**). This underlines that Gram-negatives have no easy one-to-one genotype-phenotype relationship for some ARG as their presence does not equal resistance per se. A well-studied example of this is the AmpC gene, which is expressed in many species or strains, even those fully susceptible to AB (Bajaj et al., 2016). The complex relationships between the ARG and phenotype dictated the application of more complex machine learning algorithms, such as random forest (the basis of XGBoost). An additional advantage of XGBoost is that it provides a glimpse into its decision-making process by reporting the list of features ARGC most often used when building the model as a proxy for key decisions. The downside of this simplification is that caution is warranted when interpreting these features. Finally, given that the datasets are sparse, models are prone to having to rely on different features depending on the split in training and testing. The higher the consistency in selecting important features across different independent models, the more robust the final model is anticipated to be (**Table 5**).

An intuitive example is the class A extended-spectrum betalactamase cluster as an important feature in the cephalosporin prediction models (**Table 4**). Other ARGCs seem less relevant

FIGURE 3 | Performance of XGBoost models based on 51 different splits of the data. Boxplots represent distribution of AUC for the corresponding testing subsets, with thick lines indicating the median performance.

at first glance but might make more sense when interpreted in a broader context. AAC(6<sup>0</sup> )-Ib family aminoglycoside 6<sup>0</sup> - N-acetyltransferase is an aminoglycoside resistance gene, but apart from being important in the tobramycin model (an aminoglycoside), it is also found to be important in the models predicting cephalosporin and levofloxacin resistance. This is where a more careful interpretation of important features is warranted as the non-linearity of XGBoost models is starting to provide less intuitive connections. For example, this gene is found in a variety of Gram-negative species and is known to be present in many different plasmids, genomic islands and integrons that carry other genetic resistance genes. It could be that this gene is a proxy for other, more relevant genes or even other genetic factors in the genome associated with resistance

TABLE 4 | Top 5 most important features for each antibiotic model.


\*Gain is the relative importance of the ARGC in a prediction model, reported here as mean with standard deviation computed over 51 independent models for the same AB.

TABLE 5 | Counts of unique features found among top 5 across 51 independent models for each AB.


\*The lower number signifies the more consistent feature selection across data resampling.

(Wilson et al., 2016; Lehtinen et al., 2017). Furthermore, this gene is prominent in Klebsiella species which could be used during decision-making to take advantage of innate differences in resistance between species. The fact that our prediction models are species independent could make them more powerful when focusing on resistance patterns in contaminated, mixed, and metagenomic samples. The latter is part of the future goals of this work. To show that the presence of ARG only could easily predict species, a model with the same input, but trained on predicting species instead of ABR, was built and had a near perfect accuracy (**Supplementary Table S8**).

The most striking discordance between an antibiotic and its model's important ARGC features was observed for levofloxacin. None of the most important ARGC picked up by the model are directly related to quinolone resistance. This is likely because levofloxacin resistance is largely based on mutations or small variations (e.g., gyrase gene; Chen and Lo, 2003). Given our models do not incorporate such information (section "Clustering similar antibiotic resistance genes"), all important ARGC in the levofloxacin resistance model appear to be proxies for these mutations. This illustrates both the strength and limitations of models like XGBoost. It can make accurate predictions (**Table 3**) on correlated and complex data by using non-linear logic, but the interpretation of such models can be obscure and could limit biological understanding of the underlying processes.

As any other previous work in this early-stage field of predicting ABR based on the WGS data, our study has several limitations. One of the main challenges was the sparsity of the model input, which may result in biased performance depending on the split in training and testing data. Even after clustering highly similar ARG in ARGC (hence no account for polymorphism), we still ended up with some ARGC only seen once in the whole dataset (median presence of 10 out of 152 clusters per sample). The sparsity is likely because some genes are rare, or the dataset is not fully representative of all evaluated resistances (limitation of using publicly available data). If we would only have used the genes originally sequenced in the bacteria of interest, we might have had less sparsity, but would likely be underestimating the presence of resistance as many species have developed similar resistance genes or exchanged them in processes like horizontal gene transfer. The


FIGURE 6 | Example of the predicted antibiogram. An illustration of the output from the R shiny app using the sample SAMN07450853 from the Demo set. Other samples from the Demo and In-house datasets can be accessed through the app. The "reference" column compares the predicted resistance to the one confirmed in the clinical laboratory. This would normally not be present at the time the models do their prediction for de novo samples. Color coding used: green – correct, red – incorrect, and blue – unknown.

second reason performance suffered in some cases is the class imbalance between available susceptible and resistant samples, e.g., cefotaxime only has 50 susceptible samples (**Table 1**), and also the lowest performance (**Table 3**). By creating many independent models for each antibiotic and selecting the one with median performance, we ensured that the final model would be the closest estimate of the real-life performance (section "XGBoost model training and testing"). This technique is not to

be confused with model cross-validation, where different splits of data are used to enhance one final model, and additional validation data is needed to estimate the performance.

Other limitations are that the models cannot predict the level of resistance (i.e., as regression) as they were trained on a binary data (resistant versus susceptible). Using the MIC values as input could help in this case, but the data is not always available or may be inconsistent. Also, the current predictor accounts for presence or absence of certain known drug resistance genes and hence cannot detect the presence of previously unseen genes conferring new drug resistance. In other words, the prediction of a sample being resistant has higher confidence than being susceptible. Furthermore, due to the nature of sequencing data (DNA-seq), the models cannot incorporate resistance originating from the over-expression of genes targeted

by inhibitors. RNA-seq potentially could mitigate this problem but presents additional challenges. (Meta-)transcriptomics data, unfortunately, and remains mostly within research realm. Complexity of reference database, inference of organisms/strains, and their relative abundance data (Cox et al., 2017), dynamic gene expression profile upon different drug treatments, and overall complexity of data preparation/generation impede the application of meta-transcriptomics data for real-time predictions (Van Camp et al., 2020).

Finally, the models use sequencing data derived from isolates, but it remains to be seen how they would perform on contaminated or mixed (e.g., metagenomic) samples. All limitations will be further addressed in future studies. Nevertheless, this study has shown a great potential of sequencing data as a basis for the prediction of antimicrobial resistance in Gram-negative bacteria. We additionally focused on the more practical implementation of resistance prediction models by presenting the end-users (medical practitioners) with an easy to use and interpret interface where novel predictions on different AB are shown together as in a traditional antibiogram but with an additional RI to further assist during the decision making process.

To summarize, this work demonstrates that whole-genome sequencing coupled with modern machine-learning methods has great potential to deliver early estimations of the antibiogram for Gram-negative bacteria. The generated models, being trained solely on the presence-absence of the clusters of ARG, demonstrate promising performance and robustness to heavily class-imbalanced data. RI are introduced to provide further assessment of predictions and may be used in subsequent machine learning models to improve accuracy further. By presenting the results in the form of an antibiogram, we provide an intuitive way for the clinician to interpret predictions and guide the initial empiric antibiotic choice before the laboratory

## REFERENCES


results are available. This may help in shortening the time to start effective AB treatment.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the NCBI BioSample database (BioProject ID: PRJNA587095).

## AUTHOR CONTRIBUTIONS

P-JV did the data collection and curation, model training and validation, developed the R shiny app, and stand-alone analytical pipeline. DH and AP conceived of the study, evaluated results, and supervised the project. DH provided clinical samples for the In-house dataset. AP provided computational resources. All authors participated in writing the manuscript.

## FUNDING

This work was supported by the Centers for Disease Control (Contract Number OADS BAA 2016-N-17812) and funding from the Academic Research Committee at Cincinnati Children's Hospital Medical Center.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2020.01013/full#supplementary-material



**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 Van Camp, Haslam and Porollo. 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.

# Hospital-Associated Multidrug-Resistant MRSA Lineages Are Trophic to the Ocular Surface and Cause Severe Microbial Keratitis

Paulo J. M. Bispo1,2, Lawson Ung1,2, James Chodosh1,2 and Michael S. Gilmore1,2,3 \*

*<sup>1</sup> Department of Ophthalmology, Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, United States, 2 Infectious Disease Institute, Harvard Medical School, Boston, MA, United States, <sup>3</sup> Department of Microbiology and Immunobiology, Harvard Medical School, Boston, MA, United States*

Methicillin-resistant *Staphylococcus aureus* (MRSA) is a common cause of severe and difficult to treat ocular infection. In this study, the population structure of 68 ocular MRSA isolates collected at Massachusetts Eye and Ear between January 2014 and June 2016 was assessed. By using a combination of multilocus sequence typing (MLST) analysis, SCC*mec* typing and detection of the panton-valentine leukocidin (PVL) gene, we found that the population structure of ocular MRSA is composed of lineages with community and hospital origins. As determined by eBURST analysis of MLST data, the ocular MRSA population consisted of 14 different sequence types (STs) that grouped within two predominant clonal complexes: CC8 (47.0%) and CC5 (41.2%). Most CC8 strains were ST8, harbored type IV SCC*mec* and were positive for the PVL-toxin (93.7%). The CC5 group was divided between strains carrying SCC*mec* type II (71.4%) and SCC*mec* type IV (28.6%). Remaining isolates grouped in 6 different clonal complexes with 3 isolates in CC6 and the other clonal complexes being represented by a single isolate. Interestingly, major MRSA CC5 and CC8 lineages were isolated from discrete ocular niches. Orbital and preseptal abscess/cellulitis were predominantly caused by CC8-SCC*mec* IV PVL-positive strains. In contrast, infections of the cornea, conjunctiva and lacrimal system were associated with the MDR CC5 lineage, particularly as causes of severe infectious keratitis. This niche specialization of MRSA is consistent with a model where CC8-SCC*mec* IV PVL-positive strains are better adapted to cause infections of the keratinized and soft adnexal eye tissues, whereas MDR CC5 appear to have greater ability in overcoming innate defense mechanisms of the wet epithelium of the ocular surface.

Keywords: MRSA, Ocular infection, Molecular Epidemiology, Tissue tropism, biogeography of infections

## INTRODUCTION

Antimicrobial resistance in human infections has reached alarming levels and has become one of the major public health threats of the twenty first Century (1). Methicillin-resistant Staphylococcus aureus (MRSA) remains a leading cause of antibiotic-resistant infections at many anatomical sites (2, 3). MRSA initially were confined to the hospital environment, but in the mid 1990s began to

### Edited by:

*Filipa Grosso, University of Porto, Portugal*

#### Reviewed by:

*Tingtao Chen, Nanchang University, China Edet E. Udo, Kuwait University, Kuwait*

\*Correspondence: *Michael S. Gilmore michael\_gilmore@meei.harvard.edu*

#### Specialty section:

*This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health*

> Received: *01 November 2019* Accepted: *05 May 2020* Published: *03 June 2020*

#### Citation:

*Bispo PJM, Ung L, Chodosh J and Gilmore MS (2020) Hospital-Associated Multidrug-Resistant MRSA Lineages Are Trophic to the Ocular Surface and Cause Severe Microbial Keratitis. Front. Public Health 8:204. doi: 10.3389/fpubh.2020.00204* proliferate in the community (4), and are now leading causes of antibiotic-resistant infections in both settings (5). In US, strains within lineages that constitute clonal complex 5 (CC), notably the USA100 clone, are most commonly hospital-associated (5). USA300, a representative of CC8, has emerged as the most prevalent CA-MRSA clone in the US (5, 6). USA300 has also invaded the hospital setting where it is now a common cause of MRSA infections in American hospitals (7).

Ocular infections caused by MRSA have become increasingly common in the last two decades (8–11). These infections have been associated with serious ocular damage and permanent vision loss (12, 13), including bilateral blindness (14). Despite the growing importance of MRSA in ophthalmology, little is known about the population structure of MRSA causing the most common eye infections, or the microbial and host features that dictate this structure. The eye has extensive defenses for protection of vital structures from constant environmental exposure. These include mechanical barriers (e.g., lids, lashes), a polarized wet epithelium, a secreted tear film containing immunoglobulins and various other antimicrobial factors, mucins (secreted MUC5AC and shed epithelial cell surface-associated transmembrane mucins MUC1, MUC4, and MUC16), and cells of the innate immune system (15–17).

This unique environment of the ocular wet mucosa and its components are expected to act as selective forces that can shape the spatial distribution of microorganisms colonizing and infecting this ocular niche. The study of these ecological and geographical forces, as classically applied in ecology to study the biogeography of life in the natural world can now be combined with refined genetic and genomic epidemiology data to advance our understanding of community structures and distribution of microbes in different body sites (18, 19). We previously reported the genomic characteristics of a divergent cluster of unencapsulated Streptococcus pneumoniae strains that are uniquely tropic and adapted to the conjunctiva (20). These strains carry a set of genes that are absent or substantially different from those encoded within the genomes of encapsulated respiratory strains, which appear to be important for the pathogenesis of epidemic conjunctivitis. We have demonstrated that a unified model of microbial biogeography that incorporates classic ecological principles to explain community assemblage and dynamics can be applied to the understanding of this radical bifurcation in phylogeny and niche subspecialization of the unencapsulated S. pneumoniae conjunctivitis cluster (21). Because MRSA now rank among leading causes of a variety of ocular infections, to gain insight into particular features of importance in the pathogenesis of infection, it was of interest to determine the microscale biogeography of MRSA eye infections, whether dominant genetic lineages were associated with all sites of infection, or if there was evidence of a tissue tropism that would drive a specific population structure. We report that the population structure of ocular MRSA strains isolated at Massachusetts Eye and Ear (MEE) is dominated by the two major clonal complexes that cause infections at other body sites, but exhibit a distinct distribution in the types of infection they cause.

## METHODS

## Bacterial Strains

Protocols for obtaining bacterial isolates collected for infection diagnosis were approved by the MEE Institutional Review Board (IRB). Since this study only included discarded bacterial isolates that were frozen in our pathogen repository, written informed consent was waived by the MEE IRB. In total, 68 consecutive MRSA isolates recovered from January 2014 to June 2016 were analyzed for this study. For patients from whom multiple isolates from the same eye were obtained for infection diagnosis within a period of 6 months, only the first isolate was included. Specimens were obtained by the attending ophthalmologist or resident physician following institutional guidelines and submitted to the clinical laboratory for processing. Suspected S. aureus colonies were routinely identified using a combination of phenotypic methods including detection of coagulase and protein A by latex agglutination, followed by confirmation of species and antimicrobial susceptibility testing using the MicroScan Walkaway 40 Plus System (Beckman Coulter, Brea, CA). Isolates were stored at −80◦C in MicrobankTM cryopreservative tubes (ProLab Diagnostics). Frozen isolates were cultured twice on blood agar before further testing.

## Clinical Data Collection and Statistical Analyses

Demographic data and risk factors for MRSA infection were collected using the IRB-approved Research Electronic Data Capture (REDCap) tool, hosted by MEE and Harvard Medical School (22). General demographic data included age, sex and ethnicity. Ocular comorbidities, including any ophthalmic surgical history, ocular surface disease, eyelid disease, lacrimal system dysfunction, atopy, contact lens use and trauma were collected (see **Table 2** legend for full definitions). Patient systemic comorbidities and previous healthcare exposures were also captured. To identify possible healthcare exposures which may potentiate selective pressures for antibiotic-resistant infection, we identified these following groups in our data: patients residing in nursing homes and/or residential facilities; those requiring chronic ambulatory care such as renal replacement therapy (dialysis) and hospital-based infusions; and patients who had either inpatient hospital admission and/or day admission for eye surgery within the preceding 3 prior to developing an MRSA infection. For patients with MRSA keratitis, we recorded the presenting features of the ulcers according to an institutionwide clinical algorithm which mandates the collection of corneal cultures for lesions meeting any of the following criteria: ≥1+ cells in the anterior chamber; ≥2 mm infiltrate and/or the presence of ≥2 satellite lesions; or infiltrate located ≤3 mm from the corneal center (23). Simple 2 by 2 tests of proportion (Fischer's exact test) were used to compare CC5 and CC8 groups according to collected categorical variables, while age was compared using the non-parametric Wilcoxon rank-sum test.

## Antimicrobial Susceptibility Testing

In vitro susceptibility to ciprofloxacin (Fluka), ofloxacin, levofloxacin (TCI America), moxifloxacin, and besifloxacin (Sigma-Aldrich) was performed by broth microdilution methods according to the Clinical and Laboratory Standards Institute (CLSI) (24). Quality control was performed by testing the S. aureus ATCC 29213 control strain. The interpretative criteria for each antimicrobial agent tested were those published by CLSI (25).

## DNA Extraction

DNA extraction was performed using Chelex 100 molecular biology resin (Bio-Rad) as previously described (26). Purified genomic DNA was diluted 1:10, and was assessed for purity and DNA concentration using a Synergy 2 Multi-Mode Plate Reader and Take3 software system (BioTek).

## SCCmec Typing

PCR-based genotyping of the chromosomal cassette recombinase (ccr) and mec complexes comprising the SCCmec was determined by a combination of multiplex PCR designed to classify the mec complex and ccr complex using a previously published protocol (27). For each multiplex PCR assay, reference MRSA strains for SCCmec types II (USA100) and IV (USA800), provided by the Network of Antimicrobial Resistance in Staphylococcus aureus (NARSA) were included. SCCmec was considered nontypeable if mec and/or ccr complex gave no amplification results, if the isolate carried more than one ccr or mec complex, or if there was a mec/ccr complex combination not previously described.

## PVL Detection

The presence or absence of the Panton-Valentine Leukocidin (PVL) toxin gene was determined by PCR amplification of the LukS-PV-lukF-PV genes as previously described (28). Reference MRSA strains (provided by NARSA) USA300 and USA100 served as positive and negative controls, respectively.

## MLST

Multilocus sequence typing (MLST) was performed for all MRSA isolates using a scheme based on the sequencing of internal fragments of seven S. aureus housekeeping genes (arcC, aroE, glpF, gmk, pta, tpi, and yqiL). The PCR products were purified (QIAquick PCR purification kit; Qiagen), and both strands were sequenced by Genewiz Incorporated (South Plainfield, NJ). The sequences obtained were edited using Geneious R8 and sequence types (STs) were assigned using the S. aureus MLST database (https://pubmlst.org/saureus/). Clonal complexes (CC) were determined using the go eBURST algorithm (http://www. phyloviz.net/goeburst/).

## Statistics

Descriptive statistics were calculated using SPSS software (version 25, IBM, Armonk, New York), and proportions were compared by χ <sup>2</sup>or Fisher exact test, as appropriate. A P value of <0.05 was considered statistically significant.

## RESULTS

A total of 75 MRSA were identified from 281 S. aureus recovered from ocular sites at MEE from January 2014 to June 2016 (overall rate of 26.7%). The proportion of ocular MRSA isolates did not change considerably in 2014 (25.9%) compared to 2015 (22.3%), but was substantially higher during the sampling period of 2016 (37.7%). Of those, 7 MRSA were obtained from second cultures of the same patient eye, and were excluded from further study. The remaining 68 non-duplicate MRSA isolates were then analyzed. Sites of infection from which MRSA were isolated included orbital and preseptal abscess/cellulitis (n = 27), keratitis (n = 14), conjunctivitis (n = 9), lacrimal system infection (n = 8), eyelid margin infections (n = 4), endophthalmitis (n = 2), and miscellaneous (n = 4).

## Two Major Clonal Complexes Dominate the Ocular MRSA Population

Despite the clinical importance of MRSA, much remains to be learned about the pathogenesis of infection at different anatomical sites on and around the eye. Because the tissues of the eye and adnexa differ widely in host defenses (e.g., wet epithelium vs. keratinized epithelium and soft tissues), it was of interest to know whether some MRSA lineages were enriched in pathogenic features that select for one MRSA lineage over another. By using a combination of multilocus sequence typing analysis, SCCmec typing and detection of the PVLtoxin encoding gene, we found that the population structure of ocular MRSA is diverse, but dominated by the CC5 and CC8 lineages associated with infection at other anatomical sites (4, 5). As determined by eBURST analysis of MLST data (**Figure 1**), 14 different sequence types (STs) were identified, with most belonging to clonal clusters CC8 (47.0%, n = 32) and CC5 (41.2%, n = 28). The clonal cluster CC6 encompassed 3 strains (4.4%), and 5 strains represented single sequence types. Most CC8 strains (93.7%, n = 30) were ST8, harbored a SCCmec type IV and were positive for the PVL-toxin, common features of the USA300 strain. The CC5 group could be divided into those carrying a SCCmec type II (71.4%, n = 20), which includes isolates with the characteristics of the USA100 clone, and (28.6%, n = 8) SCCmec type IV, typical of the USA800 clone (**Table 1**).

Age at presentation ranged from 2 to 102 years (median 53.08) with CC5-infected patients being significantly older (median age, 68.05 vs. 35.9, p < 0.001) (**Table 2**). CC5-infected patients were more frequently subjected to eye surgery (66.7 vs. 16.7%, p <0.001), especially cataract surgery with implantation of intraocular lenses (44.4 vs. 10%, p = 0.006). Healthcare exposure was more common among patients infected with CC5 strains, including higher rates of topical antibiotic use at presentation (55.6 vs. 10.0%, p < 0.001) and prior to presentation (48.5 vs. 16.7%, p = 0.02). There was also a higher proportion of patients known to require non-acute clinical care and/or residents of aged care facilities in the CC5 group (22.2 vs. 3.3%, p = 0.05).

## Distribution of MRSA Lineages Across Different Ocular Niches

Although S. aureus causes a wide range of human infections, patterns of association of distinct genotypes have been noted with

TABLE 1 | SCC*mec* typing and PVL detection according to the clonal complex.

represent singleton STs in this particular population structure.


specific types of infection (29–32). The most well documented example is community-acquired skin and soft-tissue infection (SSTI) in the US, where a large majority of cases are caused by the CC8/USA300 lineage (29, 32). In agreement, ocular SSTIs including mainly orbital and preseptal abscess/cellulitis were found in this study to be caused mainly by CC8 SCCmec IV PVLpositive strains, characteristics of the USA300 clone (p < 0.001; **Table 3**). Interestingly, infections of the wet epithelial tissues of the ocular surface were substantially enriched in the CC5 lineage, which was particularly pronounced for infectious keratitis cases (p < 0.001; **Table 3**). Keratitis patients frequently presented with potentially sight-threatening corneal ulcers (85.7%) according to the 1,2,3 rule (23) for categorization of the severity of bacterial keratitis (**Table 5**).

TABLE 2 | Demographic and clinicopathologic data for patients with culturepositive MRSA ocular infections at MEE, 2014–2016 (*n* = 58).<sup>a</sup>


*<sup>a</sup>Out of 68 isolates included, 66 patient records were reviewed. One patient had two separate MRSA episodes (caused by different clones) and for one patient the record was not available.*

\**includes corneal degenerative disease, bullous keratopathy, exposure keratopathy, dry eye syndrome, atopy, Stevens-Johnson syndrome and toxic epidermal necrolysis.*

\*\**including blepharitis, trichiasis, floppy eyelid syndrome, ectropion and entropion, and lagophthalmos.*

\*\*\**including dacryocystitis and dacryoadenitis.*

*‡ including day-only procedures and overnight stays.*

*‡‡includes long-term care residents, nursing home care, chronic ambulatory care.*

*‡‡‡includes professions in close working contact with patients (e.g., nursing and allied health).*

*Bold values are represent statistically significant.*

## CC5 Strains Preferentially Associated With Infection of the Wet Epithelial Ocular Surface Are Multidrug Resistant

To compare rates of antibiotic resistance among ocular MRSA of various lineages and from the different ocular sites, we tested the sensitivities of all isolates to a panel of clinically relevant antibiotics using an automated microbiology system (MicroScan WalkAway). Overall, moderate to high rates of resistance to erythromycin (88.2%), levofloxacin (54.4%), and clindamycin (42.6%) was found among this MRSA collection. Resistance to gentamycin (2.9%) and tetracycline (1.5%) was rare (**Figure 2B**). Stratified analysis by clonal complex showed that CC5 strains are significantly more likely to be multidrug resistant (resistance to ≥3 non-beta-lactam antimicrobial classes) than CC8 strains (78.8 vs. 6.1%; p < 0.0001; **Figure 2A**). Ocular CC5-SCCmecII strains, which include isolates resembling the hospital-adapted clone USA100, were all resistant to erythromycin, clindamycin and levofloxacin (**Figure 2B**). CC5- SCCmec IV strains, including isolates with the characteristics of the USA800 clone, were frequently resistant to erythromycin (62.5%), levofloxacin (62.5%), and clindamycin (25.0%). While CC5 strains were frequently resistant to ≥3 non-beta-lactam antibiotics, CC8 isolates were usually resistant to only one antibiotic class in addition to beta-lactams, most often to the macrolide erythromycin (**Figures 2A,B**).

Because topical fluoroquinolone is widely used for empirical treatment of ocular infection, we assessed the in vitro susceptibility of the main ocular MRSA clonal complexes for the most commonly used fluoroquinolones (**Table 4**). The minimum inhibitory concentrations (MICs) for these agents were determined by reference broth microdilution (24, 25). CC5 isolates were in general highly resistant to the older fluoroquinolones ciprofloxacin, ofloxacin and levofloxacin. All canonical CC5-SCCmec II strains were resistant to oxifloxacin, levofloxacin, ofloxacin and ciprofloxacin with MIC<sup>90</sup> values >256µg/mL. Among CC5-SCCmec IV strains, the MIC<sup>90</sup> values ranged from 64µg/mL for moxifloxacin to >256µg/mL for levofloxacin, ofloxacin and ciprofloxacin (% of non-susceptible = 62.5% for the 3 drugs). Remarkably, many CC5 isolates were also highly resistant to the newer 8-methoxyfluoroquinolone moxifloxacin (MIC<sup>90</sup> 64µg/mL, 62.5% non-susceptible for CC5- SCCmec IV; MIC<sup>90</sup> 32µg/mL, 100% non-susceptible for CC5- SCCmec II).

## DISCUSSION

The population structure of S. aureus isolates from human infections is highly diverse (33). However, only a small proportion of these lineages have become successful MRSA clones that are now widely disseminated in both community and hospital settings (5). These epidemic lineages cause a variety of human infections, with some showing strong tropisms for specific body niches (29–32).

The eye has developed unique mechanisms to protect its delicate and exceptionally important structures against constant external disturbance (15–17). Because of this, we postulated that the uniqueness of this environment could be a driving force shaping the population structure of ocular MRSA infections, in a similar manner as reported for Streptococcus pneumoniae (20). To begin testing this hypothesis, we selected a collection of consecutive and non-duplicate MRSA strains prospectively isolated from a variety of eye infections, representing two major

TABLE 3 | Distribution of the main ocular MRSA clones across different infections.


*Bold values are represent statistically significant.*

complexes. (A) Frequency (%) of CC5 and CC8 strains resistant to additional non-beta-lactam antibiotic classes. (B) Resistance rates for common antibiotics representing 5 different classes. Statistical significance was determined using Fisher's exact test.

and distinct localizations: (i) the ocular surface and (ii) the ocular adnexal soft tissues. We found that the ocular MRSA population was dominated by two major clonal complexes that are also common causes of MRSA infections in other body sites: CC8 (47.0%) and CC5 (41.2%). The distribution of lineages within these clonal complexes followed a pattern of enrichment that was split into the main ocular niches tested, with preseptal and orbital abscess/cellulitis being predominantly associated to CC8, while ocular surface infections were frequently caused by CC5 strains, with a significant enrichment in keratitis (**Table 3**).

TABLE 4 | *In vitro* susceptibility profile for topically used fluorquinolones among ocular MRSA isolates according to the clonal group.


*a includes intermediate resistant isolates.*


\**The "1, 2, 3-Rule," originally conceived by Vital et al. (23), was formally implemented across MEE on July 1, 2015 as a means of identifying patients at greatest risk of developing sightthreatening complications. It is composed of three simple features on clinical examination, and the presence of any of these warrants the collection of corneal cultures and treatment with topical fortified antibiotics. These rules are:* ≥*1*+ *anterior chamber cells;* ≥*2 mm infiltrate and/or* ≥*2 satellite lesions; edge of lesion within 3mm of the corneal center.*

The basis for the discernable tropism of CC5 strains for the ocular surface is unknown, but viewed in light of the findings of others, we can develop a testable model. Strains grouped within CC5, notably the USA100 and USA800, are well established as major causes of healthcare-related infections in the US (5). CC5 strains are often replete with acquired antibiotic resistances (34, 35), and were implicated in the first 12 cases of frank vancomycin resistant S. aureus stemming from independent acquisitions of the vanA gene from Enterococcus spp. (36). Despite their hospital association, CC5-MRSA strains also occur in the community, potentially bridged by long-term care facilities. The nasal reservoir of MRSA of long-term care facility residents mirrors the molecular epidemiology of US hospitals, with CC5-related strains being predominant (37). Data from national surveillance programs show that CC5 lineages USA100 and USA800 are now widely disseminated in the community, even among noninstitutionalized individuals with no known risk factors for nasal colonization with these hospitaladapted clones (38). Although CC5 strains are frequently found among carriers in the community, they are only occasionally associated with community-acquired infection (32). In our population, despite the community origins of the patients in which ocular infections occurred, CC5 strains predominated as causes of infection of the wet epithelial surface of the eye. Keratitis in our series was predominantly caused by strains resembling the USA100 lineage, and mostly presented as severe and potentially sigh-threatening infections as determined the "1, 2, 3-Rule" (23). The severity of CC5-caused keratitis points toward the possible existence of virulence factors that could be particularly associated with corneal damage. In addition to carrying a variety of antimicrobial resistance genes, CC5 strains also posses a constellation of virulence genes (35). Of particular interest is the enterotoxin gene cluster (egc), which represents a unique group of enterotoxins with superantigen activities, and seems to be particularly enriched among CC5 strains (35), while being completely absent in CC8 strains (36, 39). Epidemiological observations have found an association of egc-encoded enterotoxins in the development of corneal ulcer in patients with atopic keratoconjunctivitis (40). Whether this locus is associated with exacerbation of the ocular surface inflammatory response and aggravation of the corneal damage is yet to be fully elucidated.

Empirical use of topical broad-spectrum antibiotics remains the first-line treatment for bacterial keratitis (41) and is commonly initiated with a topical newer fluoroquinolone ophthalmic solution (e.g., moxifloxacin 0.5%) (42). CC5 strains display generally higher fluoroquinolone MICs (nonsusceptibility rate for moxifloxacin of 89.3%, MIC<sup>90</sup> 64µg/mL) (**Table 4**). However, CC8-SCCmec IV strains associated with ocular infections at other sites were also often resistant to the commonly used fourth-generation fluoroquinolone, moxifloxacin (non-susceptibility rate using systemic breakpoints of 35.5%, MIC<sup>90</sup> 2µg/mL). The influence of degrees of resistance on population structure is unclear. In light of its pharmacokinetics on the ocular surface, neither CC5 nor CC8 would be predicted to respond to topical moxifloxacin therapy. Based on pharmacokinetic data for topical moxifloxacin in the cornea of pigmented rabbits (43), we calculated PK/PD (AUC0−24h/MIC<sup>90</sup> ratio) indices for the main MRSA lineages examined in our study (**Figure S1**). Although the index was higher for CC8 strains compared to CC5, all the indices were far below the PK/PD target that predicts clinical efficacy as determined for systemic infections (44) and also for keratitis treated with fluoroquinolone (45). In addition, a review of the medical records of 14 patients with MRSA keratitis included in this study showed that only 4/14 had prior exposure to a fluoroquinolone (Vigamox, n = 2; erythromycin plus ofloxacin, n = 1; and Vigamox plus tobramycin, n = 1), with no information available for 3 patients (**Table 5**). Together, the data on prior topical antibiotic use and the calculated PK/PD indices for the ocular MRSA lineages discounts the role of direct fluoroquinolone selection as the driver of the CC5 predominant population structure associated with keratitis.

Our results are consistent with a report of 30 cases of S. aureus keratitis in Japan, that also found CC5 strains to be the predominant cause of MRSA-infected ulcers (46). CC8 isolates were more frequently isolated from the healthy conjunctival sac (46). Together, these findings suggest an enhanced ability of CC5 to endure selective pressures and colonize the cornea and adjacent tissues. Lactoferrin, one the most abundant tear proteins with antimicrobial activity is inhibitory for S. aureus, including MRSA, but this activity varies among clinical isolates from different sites of isolation (47, 48). In one study, clinical S. aureus isolates causing bloodstream infections (50%) were frequently resistant to lactoferrin concentrations ≥ 20µM, while isolates from conjunctivitis (33%) and SSTI (13%) were less often resistant (47). Since these infections are caused by distinct MRSA lineages, with CC5 strains being commonly associated to bloodstream infections (49), variations across different genotypes may contribute to increased resistance to this tear antimicrobial compound, especially among CC5. Similarly, the antimicrobial activity of phospholipase A2 (PLA2), an enzyme that has been found to be the major tear molecule with bactericidal activity against staphylococci, varies against methicillin-susceptible and -resistant isolates, with MRSA being more resistant to its action following short incubations (50), pointing toward a variance of this enzyme in its ability to kill S. aureus according to the genetic background. Further, previous reports have demonstrated that S. aureus strains may be differently equipped to bind to and invade ocular tissues in vitro and in a rabbit model of conjunctivitis and keratitis (51, 52), suggesting the existence of specific sets of adhesive surface proteins that enhance S. aureus ability to bind and invade the ocular tissues in a strain-specific manner.

A recent study of S. aureus keratitis in South Florida (53) found both CC5 (40%) and CC8 (37.3%) MRSA among isolates, clearly showing that although CC5 strains are most common, CC8 strains in other circumstances are also capable of causing keratitis. Among our patients, 12 out of 14 (85.7%), were classified as potentially sight-threatening MRSA keratitis cases according to an institution-wide grading system based on the "1, 2, 3" rule (23). For these patients, corneal cultures were unequivocally positive at diagnostic levels for the pathogen. We do not yet know whether the differences between studies stem from levels of disease severity, or other factors such as association with contact lens wear. Larger studies with well-defined criteria for enrollment will be important for determining conditions that favor infection by various pathogenic, antibiotic resistant lineages of S. aureus.

In contrast to the wet epithelial surface of the eye, CC8 MRSA strains, typified by the USA300 lineage, predominate as causes of infections of the keratinized epithelium and eye soft tissues. Lineage enrichment has been also reported in patients with bloodstream infection with haematogenous complications (30) infective endocarditis (31) and respiratory tract infections (54). Among our patients, CC8 strains were the predominant causes of preseptal and orbital abscess/cellulitis (22 out of 27 cases), consistent with their known tropism for infecting keratinized epithelium and soft tissues (29, 32). Among the remaining abscess/cellulitis cases, only one was caused by a CC5 isolate and 3 by other lineages (**Table 3**). In the early 2000s, the USA300 clone appeared in outbreaks of community-acquired SSTIs in otherwise healthy people (55, 56). Because of its ability to rapidly spread through person-to-person contact and readily compete with commensal skin flora (56), the USA300 lineage quickly predominated as the main cause of SSTIs (29, 32), and a significant cause of severe invasive infections (6). USA300 is generally more virulent, causing infections of greater severity and associated with worse outcome compared to other MRSA clones (57, 58). The aggressiveness of USA300 type strains is thought to be associated to the carriage of a variety of virulence factors, including the pore-forming toxin PVL, which is predominantly found in CC8 strains (59), and has been linked to the ability of this strain to cause necrotizing infections (60).

Orbital and preseptal abscesses are SSTIs commonly caused by Staphylococcus aureus (61, 62), and MRSA rates appear to be rising (63). Severe cases have been reported, including orbital and periorbital necrotizing fasciitis (64–66), necrotizing conjunctivitis with orbital invasion (12) and orbital cellulitis with bilateral involvement that progressed to blindness (14). A prospective study (2012–2015) of children and adolescents presenting with staphylococcal periorbital and orbital cellulitis in Houston found that most of the S. aureus isolates in their population were methicillin-resistant (67%) and were genetically related to the USA300 clone (78%) (67). Similarly, in a series of 11 patients presenting with culture-positive MRSA infections of the eye and orbit in San Francisco, most (82%) were caused by USA300 (13). Preseptal and orbital abscess/cellulitis were among the most common manifestations in these cases, some presenting with extensive necrosis of the eyelid and orbital tissues.

Collectively, these results are consistent with various MRSA lineages being enriched for properties that enhance their ability to resist defenses and competitive pressures, and colonize and infect either the wet epithelial surface or the ocular adnexal tissues. We propose that many of the features that have allowed CC5-type MRSA to adapt and be transmitted in the hospital environment, and readily acquire antibiotic resistances, endows them with properties that enhance their ability to resist robust ocular surface defenses and infect the cornea. In a comparative analysis of the genomes of the CC5 strains that had acquired vancomycin resistance from enterococci, we identified a constellation of traits with the S. aureus pathogenicity island as likely involved with their ability to persist in a mixed infection and acquire resistances by horizontal gene transfer (36). These same traits may well enhance their survival despite the defenses of the wet epithelial surface. In contrast, CC8-SCCmec IV PVL-positive strains are already well known to have features that enhance their ability to colonize the skin and compete effectively for that niche with coagulase negative strains, including the ACME locus (68). We believe these features account for their association with infection of the keratinized surfaces of the ocular adnexa. Direct testing of isogenic mutants will be required to identify the major contributors to pathogenesis of various anatomical sites of the eye by MRSA, and to develop new approaches for mitigating that common threat. It is important to note that these findings represent the population structure of a relatively small ocular MRSA population isolated at the Massachusetts Eye and Ear, and validation of our results using a larger bacterial collection and isolates from other locations would be warranted.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## AUTHOR CONTRIBUTIONS

PB, LU, JC, and MG contributed conception and design of the study and wrote sections of the manuscript. PB and LU organized the database and performed the statistical analysis. PB wrote the first draft of the manuscript. All authors contributed to manuscript revision, read and approved the submitted version.

## FUNDING

This work was supported in part by NEI grant EY024285, and the Harvard-wide Program on Antibiotic Resistance NIAID grant AI083214. Funding agencies had no role in study design, data analysis, decision to publish or preparation of the manuscript.

## ACKNOWLEDGMENTS

The authors thank Rick Body, James Cadorette and other medical technologists from the Clinical Microbiology Laboratory at MEE for their support in creating a microbial repository of strains isolated from ocular and otolaryngology infections.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh. 2020.00204/full#supplementary-material

Figure S1 | PK/PD (AUC0−24h/MIC<sup>90</sup> ratio) indices for the main MRSA clones isolated in our study. AUC0−24h data was derived from pharmacokinetic studies of topical moxifloxacin in the cornea of pigmented rabbits (43). Dashed line indicates the PK/PD target that predicts clinical efficacy in keratitis patients treated with topical fluoroquinolone (45).

## REFERENCES


Approved Standard-Ninth Edition. CLSI document M07-A9. Wayne, PA: Clinical and Laboratory Standard Institutes. (2012).


among professional football players. N Engl J Med. (2005) 352:468–75. doi: 10.1056/NEJMoa042859


**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 Bispo, Ung, Chodosh and Gilmore. 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.

# Amoxicillin Increased Functional Pathway Genes and Beta-Lactam Resistance Genes by Pathogens Bloomed in Intestinal Microbiota Using a Simulator of the Human Intestinal Microbial Ecosystem

#### Edited by:

Leonardo Neves de Andrade, University of São Paulo, Brazil

#### Reviewed by:

Rosa Del Campo, Ramón y Cajal Institute for Health Research, Spain Pallavi Singh, Northern Illinois University, United States

## \*Correspondence:

Daqing Mao maodq@nankai.edu.cn Yi Luo luoy@nankai.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 22 October 2019 Accepted: 12 May 2020 Published: 04 June 2020

#### Citation:

Liu L, Wang Q, Lin H, Das R, Wang S, Qi H, Yang J, Xue Y, Mao D and Luo Y (2020) Amoxicillin Increased Functional Pathway Genes and Beta-Lactam Resistance Genes by Pathogens Bloomed in Intestinal Microbiota Using a Simulator of the Human Intestinal Microbial Ecosystem. Front. Microbiol. 11:1213. doi: 10.3389/fmicb.2020.01213 Lei Liu<sup>1</sup>† , Qing Wang1,2† , Huai Lin<sup>1</sup> , Ranjit Das<sup>1</sup> , Siyi Wang<sup>1</sup> , Hongmei Qi<sup>1</sup> , Jing Yang<sup>1</sup> , Yingang Xue<sup>3</sup> , Daqing Mao<sup>4</sup> \* and Yi Luo<sup>1</sup> \*

<sup>1</sup> College of Environmental Science and Engineering, Ministry of Education Key Laboratory of Pollution Processes and Environmental Criteria, Nankai University, Tianjin, China, <sup>2</sup> Hebei Key Laboratory of Air Pollution Cause and Impact (preparatory), College of Energy and Environmental Engineering, Hebei University of Engineering, Handan, China, <sup>3</sup> Key Laboratory of Environmental Protection of Water Environment Biological Monitoring of Jiangsu Province, Changzhou Environmental Monitoring Center, Changzhou, China, <sup>4</sup> School of Medicine, Nankai University, Tianjin, China

Antibiotics are frequently used to treat bacterial infections; however, they affect not only the target pathogen but also commensal gut bacteria. They may cause the dysbiosis of human intestinal microbiota and consequent metabolic alterations, as well as the spreading of antibiotic resistant bacteria and antibiotic resistance genes (ARGs). In vitro experiments by simulator of the human intestinal microbial ecosystem (SHIME) can clarify the direct effects of antibiotics on different regions of the human intestinal microbiota, allowing complex human microbiota to be stably maintained in the absence of host cells. However, there are very few articles added the antibiotics into this in vitro model to observe the effects of antibiotics on the human intestinal microbiota. To date, no studies have focused on the correlations between the bloomed pathogens caused by amoxicillin (AMX) exposure and increased functional pathway genes as well as ARGs. This study investigated the influence of 600 mg day−<sup>1</sup> AMX on human intestinal microbiota using SHIME. The impact of AMX on the composition and function of the human intestinal microbiota was revealed by 16S rRNA gene sequencing and high-throughput quantitative PCR. The results suggested that: (i) AMX treatment has tremendous influence on the overall taxonomic composition of the gut microbiota by increasing the relative abundance of Klebsiella [linear discriminant analysis (LDA) score = 5.26] and Bacteroides uniformis (LDA score = 4.75), as well as taxonomic diversity (Simpson, P = 0.067, T-test; Shannon, P = 0.061, T-test), and decreasing the members of Parabacteroides (LDA score = 4.18), Bifidobacterium (LDA score = 4.06), and Phascolarctobacterium (LDA score = 3.95); (ii) AMX exposure significantly enhanced the functional pathway genes and beta-lactam resistance genes, and the bloomed pathogens were strongly correlated with the metabolic and immune system diseases gene numbers (R = 0.98, P < 0.001) or bl2\_len and bl2be\_shv2 abundance (R = 0.94, P < 0.001); (iii) the changes caused by AMX were "SHIME-compartment" different with more significant alteration in ascending colon, and the effects were permanent, which could not be restored after 2-week AMX discontinuance. Overall results demonstrated negative side-effects of AMX, which should be considered for AMX prescription.

Keywords: amoxicillin, antibiotic resistance genes (ARGs), functional pathway genes, human intestinal microbiota, simulator of the human intestinal microbial ecosystem (SHIME)

## INTRODUCTION

fmicb-11-01213 June 2, 2020 Time: 22:39 # 2

Human intestinal microbiota co-exists in symbiosis with human beings and comprises with about 150 times more genes than the human genome, which makes intestinal microbiota become "another" genome of human beings (Qin et al., 2010). Moreover, human intestinal microbiota has been demonstrated to provide numerous important functions for the human health, including fermentation of indigestible dietary polysaccharides, synthesis of essential amino acids, and vitamins, modulation of the immune function and protection from the pathogens, as well as metabolism of the xenobiotic drugs (Chow et al., 2010; Yatsunenko et al., 2012; Cabreiro et al., 2013). Both human and veterinary antibiotics were detected in the collective gut of the Chinese population through our previous research (Wang et al., 2020). The antibiotics in our gut would kill or prevent the growth of commensal beneficial bacteria, which would cause the dysbiosis of intestinal microbiota, resulting multiple human diseases (Blaser, 2016; Ianiro et al., 2016; Flandroy et al., 2018; Kho and Lal, 2018). Antibiotics may also promote the spreading of antibiotic resistant bacteria (ARB) and antibiotic resistance genes (ARGs) in the human gut (Stecher et al., 2013; Barraud et al., 2018). One of the most commonly prescribed antibiotic in clinical and residential applications is amoxicillin (AMX), an inexpensive oral penicillin-type, beta-lactam antibiotic that kills a broad spectrum of bacteria by interfering with the synthesis of bacterial cell wall peptidoglycan layer (Zapata and Quagliarello, 2015; Zhang et al., 2015). The influence of AMX on human intestinal microbiota by in vivo research has been well studied and recorded, including human (Ladirat et al., 2014a; Pallav et al., 2014; Vrieze et al., 2014; Zaura et al., 2015; Oh et al., 2016) and human microbiota associated-animal models (Barc et al., 2008; Collignon et al., 2008; Collignon et al., 2010). Oh et al. (2016) noticed that AMX increased the abundance of Klebsiella in human intestinal microbiota. As a typical Gramnegative bacterial pathogen, K. pneumoniae is ubiquitous in the environment and symbioses in the human gut (Rock et al., 2014). However, the bloom of K. pneumoniae in human gut is often related to unhealthy status (Pena et al., 1997; Wiener-Well et al., 2010; Gorrie et al., 2017). Furthermore, no studies have suggested the bloom of K. pneumoniae caused by AMX exposure contributed to the increase of functional pathway genes and beta-lactam resistance genes.

Previous reports have revealed that changes of intestinal microbiota by AMX exposure are region-dependent and more significant effects were observed in the proximal colon than in the distal colon (Ladirat et al., 2014b; Marzorati et al., 2017). In vivo experiments detected the fecal samples that generally stand for the distal intestinal microbiota, which could not reveal the impacts of antibiotics on different gut regions. Moreover, in vitro experiment using simulator of the human intestinal microbial ecosystem (SHIME) can elucidate the direct effects of medicines on different regions of the human intestinal microbiota, which allows the complex human microbiota to be stably maintained in the absence of host cells (Van de Wiele et al., 2015). The SHIME model is known to be a useful tool for the in vitro study, which has already been used to identify the influence of bacteria and compounds on the colon microbiota, including probiotics such as Clostridium cluster XIVa and Bifidobacterium longum (Van den Abbeele et al., 2013; Truchado et al., 2015), prebiotics and prebiotics like compounds such as inulin, polyphenols and orange juice (Kemperman et al., 2013; Duque et al., 2016; Selak et al., 2016), and other toxic compounds such as Chlorpyrifos and Arsenic (Reygner et al., 2016; Yu et al., 2016). However, to the best of our knowledge, there are very few research that applied antibiotics into the SHIME model (Van den Abbeele et al., 2012; Bussche et al., 2015; Marzorati et al., 2017; Ichim et al., 2018; El Hage et al., 2019; Liu et al., 2020). These articles focused on the benefit of the mucosal environment, high-fiber diets, probiotic, and propionate-producing consortium in human intestinal microbiota. The effects of antibiotics, including AMX and other antibiotics mixture, vancomycin, and clindamycin, were limited on microbiota composition and metabolite. Also, there was little information available to demonstrate the link between the bloomed pathogens and functional pathway genes or ARGs. Hence, there is a need to study the influence of AMX on human intestinal microbiota in the SHIME model that integrates the entire gastrointestinal tract and maintains microbiome stability over an extended timeframe (Van de Wiele et al., 2015).

Since the reasonable dosage of AMX for the adult human study is about 750 to 1,500 mg day−<sup>1</sup> , and only half volume of the adult gut can be simulated in the SHIME model, here the direct effects of 600 mg day−<sup>1</sup> of AMX on the composition and function of the human fecal microbiota were followed by the previous studies (Pallav et al., 2014; Reijnders et al., 2016). Three reactors (representing the ascending, transverse, and descending colon, respectively) that inoculated with human intestinal microbiota were studied for the three groups: immediately before AMX administration during 0 to 21 days (a control group), AMX-exposure during 22 to 28 days (an AMX treated group), and after the AMX discontinuance during 29 to 42 days (a recovery

group). The 16S rRNA gene sequencing and high-throughput quantitative PCR (HT-qPCR) results revealed that AMX exposure caused a tremendous impact on the overall taxonomic composition of the gut microbiota, and increased functional pathway genes as well as ARGs. The changes were "SHIMEcompartment" different with more significant modulation in the ascending colon, which could not be restored after 2-week AMX discontinuance. Therefore, the results of our research demonstrated a severe impact and a negative side-effect of AMX related to health problems, which should be considered as a fundamental aspect of the cost-benefit equation for its prescription.

## MATERIALS AND METHODS

## The SHIME Model Experimental Setup and Sampling

The SHIME set up was formed by five double-jacketed reactors designated as the stomach, small intestine, ascending colon, transverse colon, and descending colon, respectively (**Supplementary Figure S1**). The last three reactors were inoculated with a mixture of fecal microbiota from a healthy adult volunteer, who did not suffer from gastrointestinal diseases or take antibiotics in the previous 6 months according to previous classic studies (Yu et al., 2016; Wang et al., 2018). And the differences between individuals may be alleviated by same culture condition (Van den Abbeele et al., 2013). The study was approved by the Biomedical Ethics Committees of Nankai University. The participant has given written informed consent to understand the study purpose, procedures, risks, benefits, and rights. The details of the SHIME system and the startup process are summarized in the **Supplementary Material**.

As shown in **Supplementary Figure S1**, during the first 2 weeks (0 to 14 days) of the experiment, control nutritional medium was added into the reactors to stabilize the microbial community. After this period, the SHIME was sequentially exposed to nutritious medium (15 to 21 days), and nutritious medium + 600 mg day−<sup>1</sup> AMX (22 to 28 days), each time system was maintained for 1-week. Then a nutritious medium was added and observed for another 2-weeks (29 to 42 days). Samples were collected and analyzed at six time points of 14, 21, 24, 28, 35, and 42 days from the ascending colon, transverse colon and descending colon, respectively. Each of the 18 samples is a mixture of three samples collected at specific time intervals in a day (Liu et al., 2020), and AMX was added after samples (C-A2, C-T2, and C-D2) had been collected in 22 days and discontinued after samples (AMX-A2, AMX-T2, and AMX-D2) had been collected in 29 days. Therefore, these samples were divided into three groups: before AMX administration during 0 to 21 days (control group), AMX exposure during 22 to 28 days (AMX treatment group) and after the AMX discontinuance during 29 to 42 days (recovery group), which were according to previous classic studies (Yu et al., 2016; Wang et al., 2018). Specifically, during each sampling time, three sterilized centrifuge tubes (50 ml) were used to collect the samples (10 ml) flow out from each colon vessel of the SHIME, respectively, and these samples were initially stored at −4 ◦C (Van de Wiele et al., 2010; Yu et al., 2016). After all the samples were collected in each sampling day, three samples from the same vessel were mixed into one sterilized centrifuge tube (50 ml), which was operated in a super-clean bench. Then the samples were centrifuged at 10,400 g for 10 min, and the separated supernatants and pellets were stored at −80◦C for further analyses.

## 16S rRNA Gene Sequencing and Analysis

Total DNA was extracted from the samples using the E.Z.N.A. stool DNA Kit (Omega, United States) according to the manufacturer's protocols. The V3–V4 region of the bacterial 16S rRNA gene was amplified by polymerase chain reaction (PCR) using primers 341F 5<sup>0</sup> -CCTAYGGGRBGCASCAG-3<sup>0</sup> and 806R 5 0 -GGACTACNNGGGTATCTAAT-3<sup>0</sup> (95◦C for 2 min, followed by 27 cycles at 95◦C for 30 s, 55◦C for 30 s, and 72◦C for 30 s, and a final extension at 72◦C for 5 min). Amplicons were pairedend sequenced (PE250) on Illumina Hiseq2500 platform. In the end, a total of 651,502 tags were obtained. The raw reads were deposited into the NCBI Sequence Read Archive (SRA) database under the accession number of SRR9330193–SRR9330210. The details of bacterial DNA extraction and PCR amplification of the 16S rRNA gene are described in the **Supplementary Material**.

Raw Illumina fastq files were de-multiplexed, quality-filtered, and analyzed using Quantitative Insights Into Microbial Ecology (QIIME) (Caporaso et al., 2010). The 16S rRNA gene sequences were classified taxonomically using the Ribosomal Database Project (RDP), and classifier 2.0.1 (Wang et al., 2007). The AMX exposure on alpha diversity was further measured by the taxon richness (Chao1 index), evenness (Simpson index), and diversity index (Shannon index) using all recommended samples. Besides, beta diversity of microbiota communities at baseline and after antibiotics were portrayed by nonmetric multidimensional scaling (NMDS) and principal coordinate analysis (PCoA) of weighted and unweighted UniFrac distances. Linear discriminant analysis effect size (LEfSe) analysis was performed to determine the bacterial taxa that significantly differed between the control and AMX exposure group using Galaxy application tool (Segata et al., 2011). Functional predictions of microbial community were performed to visualize the distribution of functional pathway genes in the three parts of the colon with different treatments using Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) (Langille et al., 2013). The accuracy of PICRUSt for the detection of these more challenging functional groups was good (minimum accuracy = 0.82), suggesting that their inference of gene abundance across various types of functions was reliable, and PICRUSt predictions had high agreement with metagenome sample abundances across all body sites (Spearman r = 0.82, P < 0.001). PICRUSt has been successfully manipulated in many previous research for predicting microbial function of human intestinal microbiota (McHardy et al., 2013; Bunyavanich et al., 2016; Labus et al., 2017; Puri et al., 2018). The details of taxonomical classification, LEfSe analysis, and functional predictions are described in the **Supplementary Material**.

## High-Throughput Quantitative PCR (HT-qPCR) and Analysis

High-throughput quantitative PCR reactions were performed to visualize the variation of ARGs during the treatment using Wafergen SmartChip Real-time PCR system, conducted by Anhui MicroAnaly Gene Technologies Co., Ltd. (Anhui, China). A total of 108 primer sets were used (Excel S1), including 102 primer sets to target the almost all major classes of ARGs found in the Chinese human gut microbiota (Hu et al., 2013), five mobile genetic elements (MGEs), and one 16S rRNA gene. The results were analyzed with SmartChip qPCR Software by excluding the wells with multiple melting peaks or amplification efficiency beyond the range (90–110%). Then data were screened with the conditions that a threshold cycle (CT) must be <31 and positive samples should have three replicates simultaneously. The details of HT-qPCR analysis are described in the **Supplementary Material**.

## Data Analysis

All the results were expressed as mean values and standard deviations. The statistical analysis was performed with SPSS 17.0 software (SPSS Inc., Chicago, IL, United States). The T-test was conducted to compare the differences between the groups, and all the statistical tests were two-tailed. The statistical significance was set at three different levels (∗P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001). Spearman test, Mantel test, and Procrustes test for correlation analysis between the microbiota and the functional pathway genes or ARGs were performed in R with the vegan package. Correlations between the pairs of variables were considered to be significant at R > 0.6, and P values were <0.05. The Gephi (V 0.9.1) software was used to visualize the bipartite network graphs using the Force Atlas algorithm.

## RESULTS

## AMX Exposure Increased Microbiota Diversity

The effects of 600 mg day−<sup>1</sup> AMX treatment on the gut bacterial community were revealed by the 16S rRNA gene sequencing of fecal samples collected from three different vessels designated as ascending, transverse, and descending colon. The vessels were set up for three groups: before AMX administration during 0 to 21 days (control group), AMX exposure during 22 to 28 days (AMX treatment group) and after the AMX discontinuance during 29 to 42 days (recovery group). The alpha diversity of the fecal microbiota was assessed in each group. The taxon richness (Chao1 index), evenness (Simpson index), and diversity index (Shannon index) of the three groups are shown in **Figure 1**. As compared with the control group, a substantially rising in the evenness (Simpson, P = 0.067, T-test) and diversity (Shannon, P = 0.061, T-test) was observed in the AMX treatment groups; however, no change occurred in the microbiota richness (Chao1, P = 0.564, T-test). Moreover, data displayed that an increasing of microbiota evenness and diversity caused by the AMX treatments

was continued after two weeks of AMX discontinuance (Simpson, P = 0.044, T-test; Shannon, P = 0.028, T-test).

In addition, the beta diversity of the microbial communities and weighted UniFrac distance was also affected by AMX treatment. As shown in **Figure 2** and **Supplementary Figure S2**, the samples collected after the AMX treatment were differed from the control group, because the two groups clustered far away from each other in both UniFrac NMDS and PCoA analyses. The average weighted UniFrac distance between the AMX treatment

(B) PCoA, and unweighted (C) and weighted (D) NMDS of UniFrac distances of samples in three groups (control, AMX, and recovery).

group and control group (AMX vs control) was significantly higher (P < 0.001, T-test) than that within the control group (control vs control). Similarly, the beta diversity results showed that the gut microbial composition remained comparable after 2-week of AMX discontinuance because these samples were clustered together with the AMX treatment groups observed in both UniFrac NMDS and PCoA analyses. The average weighted UniFrac distance between the recovery group and the control group (recovery vs control) was also significantly higher (P < 0.001, T-test) than the control group (control vs control).

## AMX Exposure Changed Microbiota Community Composition

The bacterial community compositions and their shifts at the phylum and genus levels after AMX exposure are shown in **Figure 3** and **Supplementary Figure S3**. Based on the 16S rRNA gene analysis, the taxonomic assignment was related to most four dominant phyla such as Bacteroidetes, Firmicutes, Proteobacteria, and Synergistetes, which accounted for 96.5 to 97.9% of the total community (**Figure 3A**). However, a noticeable increase in abundance of Bacteroidetes (from 16.3 to 19.9%) and Synergistetes (from 8.6 to 13.2%), and decrease of Proteobacteria (from 59.6 to 50.7%) were seen after AMX treatment. The shifted phenomena (Bacteroidete from 13.3 to 23.2%; Proteobacteria from 68.8 to 57.8%) were more prominent in the ascending colon than the transverse and descending colons (**Supplementary Figure S3A**). A significant alteration in the communities was also observed at the phylum level (Bacteroidetes from 19.9 to 30.5%; Proteobacteria from 50.7 to 39.4%) after 2-week of AMX discontinued.

At the genus level, the Bacteroides, Klebsiella, Megasphaera, and Pseudomonas were predominant (**Figure 3B**). The antibiotictreated subjects were shown to be substantially overgrown by Bacteroides (from 9.8 to 15.0%), Klebsiella (from 1.5 to 18.1%) and Pyramidobacter (from 0.8 to 10.0%), and declined in the percentage of Cloacibacillus (from 6.2 to 2.5%) and Parabacteroides (from 4.5 to 1.7%). Similarly, the changes of Bacteroides (from 8.4 to 20.5%) and Klebsiella (from 1.6 to 32.9%) were more evident in the ascending colon (**Supplementary Figure S3B**). As shown in **Figure 3B**, during the recovery period, it observed that the gut microbiota was not fully recuperated. Only, Bacteroides was inclined by abundance from 15.0 to 23.4%; however, all others were less retrieved (Klebsiella from 18.1% to 10.9%; Pyramidobacte from 10.0 to 8.5%; Cloacibacillus from 2.5 to 4.3%; Parabacteroides from 1.7 to 3.7%).

The LEfSe showed that the AMX exposure caused an obvious decrease in several taxa, including the members of Parabacteroides [linear discriminant analysis (LDA) score = 4.18], Bifidobacterium (LDA score = 4.06), and Phascolarctobacterium (LDA score = 3.95) (**Figure 4A**). The variation was accompanied by significant increases in the relative abundances of Klebsiella (LDA score = 5.26), and Bacteroides uniformis (LDA score = 4.75). However, after 2-week of AMX discontinuance, the decrease of Bifidobacterium (LDA score = 3.88), and increase of Klebsiella (LDA score = 4.67) and Bacteroides uniformis

(LDA score = 4.65) were still recognizable as compare with the control group (**Figure 4B**).

## AMX Exposure Increased Functional Pathway Genes

The hierarchy cluster heatmap analysis using metagenomic 16S rRNA gene sequencing was predicted by PICRUSt. Results showed that functional pathway genes, which included cellular processes, environmental information processing, genetic information processing, human diseases, metabolism, and

organismal systems were more abundant in the antibiotic exposure group than in the control group that represented by the ascending colon (**Figure 5**). For instance, as compared to the AMX-free sample C-A2, collected from the ascending colon before AMX administration, the gene numbers of human diseases related-pathways were 2.4–3.8 times enriched in the AMX exposed sample AMX-A2 that obtained after AMX treatment for 7 days. Similarly, the numbers of genes respect to functional pathways, for examples, membrane transport, transcription, xenobiotics biodegradation, metabolism, and excretory system were nearly 4.6–5.3 folds enriched in the AMX-A2 sample than C-A2. Moreover, these pathways were still maintained at a higher level (approximately 2.2–2.5 times enhanced than the control group) after the recovery phase.

To investigate whether the OTUs correlated with the functional pathway genes, the Mantel test and Procrustes analysis were used. The results showed that OTUs from the ascending, transverse, and descending colons were moderately correlated with the numbers of functional pathway genes (Mantel test, R = 0.743, P < 0.001). The Procrustes analysis indicated that OTUs obtained from the 16S rRNA gene data and the functional genes could be clustered according to the type of sample, which further exhibited a goodness-of-fit test (R = 0.669, P < 0.001, and 9,999 permutations) based on the Bray– Curtis dissimilarity metrics (**Supplementary Figure S4A**). As shown in **Figure 5**, the results demonstrated the co-occurrence patterns of significantly shifted microbial taxa and functional pathway genes during AMX-treatment. It can be seen that the abundances of most imminent bacteria after AMX exposure such as Bacteroides uniformis and Klebsiella were positively associated with the gene numbers of functional pathways; however, the suppressed bacteria like Phascolarctobacterium and Parabacteroides were negatively associated (**Figure 5**). Specifically, the correlation coefficients of Klebsiella with the gene numbers of the digestive system, immune system, and metabolic diseases were about 0.98 (P < 0.001) and the correlation coefficients of Phascolarctobacterium abundance with gene numbers of signaling molecules and interaction, transport and catabolism, endocrine system, cell growth, and death were about −0.77 (P < 0.01).

## AMX Exposure Increased the Abundance of Beta-Lactam and Tetracycline Resistance Genes

The relative abundances of ARGs such as beta-lactam and tetracycline resistance genes were substantially higher in the AMX exposure group as compared to the control group, while the multidrug-resistant ARGs and transposase were lower in the AMX exposure group (**Figure 6**). Notably, the relative log abundance of bl2\_len, bl2b\_tem1, and bl2be\_ctxm (aminoglycoside) were about 1.6–2.4 log units higher after AMX

treatment (AMX-A2) than in control (C-A2). Similarly, the tetb, tetc, and tetr (tetracycline) were about 1.4–2.5 log units higher than the control group. Besides, the reductions of relative log abundance in the ARGs such as mdte and tolc (multidrug) were about 1.2 log units after AMX treatment (AMX-A2) than control (C-A2), and Tn22 (transposase) was 3.0 log units. Similarly, these ARGs were unable to return at the baseline level following the recovery procedure with about 0.6 to 3.2 log units change of relative abundance compared with the control group.

To investigate whether the OTUs correlated with the resistome composition, Mantel test and Procrustes analysis were also performed to correlate the OTUs with the resistome using ascending, transverse and descending colons' samples. Our results showed that OTUs were strongly correlated to the ARG profiles (Mantel test, R = 0.895, P < 0.001). The Procrustes analysis demonstrated that the bacterial OTUs and the ARGs in HT-qPCR data could be clustered by the type of sample and exhibited a goodness-of-fit-test (R = 0.941, P < 0.001, and 9,999 permutations) by Bray-Curtis dissimilarity metrics (**Supplementary Figure S4B**). The network analyses of co-occurrence patterns between the significantly changed microbial taxa after AMX treatment and ARG subtypes were shown in **Figure 6**. Interestingly, a very similar pattern of results was observed between the microbial taxa and functional pathway genes. The abundance of bacteria that significantly decreased after AMX treatment was positively associated with multidrug ARGs, while the abundance of increased bacteria was positively associated with beta-lactam resistance genes. For example, the moderate correlation coefficient of Bifidobacterium longum with acra abundance was 0.70 (P < 0.05), and for Phascolarctobacterium with acrb and mdte were 0.67 (P < 0.05). The strong correlations were found in the abundance of Klebsiella with the abundances of bl2\_len, bl2b\_tem1, and bl2be\_shv2 (R = 0.9, P < 0.001), and in Bacteroides uniformis with bl2\_len and bl2be\_shv2 (R = 0.8, P < 0.01).

## DISCUSSION

## AMX Treatment Has a Tremendous Influence on the Overall Taxonomic Composition and Biodiversity

The SHIME model was stably operated in this study because the predominant phyla of Bacteriodetes, Proteobacteria, Synergistetes, and Firmicutes in the gut microbiome was previously demonstrated Yu et al. (2016). Firmicutes and Bacteroidetes are usually dominate in the microbiota of a healthy subject, however in the control microbiota of SHIME, Proteobacteria is majoritarian. For in vivo studies, highest percentage of Proteobacteria had also been observed in fecal samples from healthy human and animals (Gao et al., 2015; Li et al., 2015). Our previous in vitro study also discovered this phenomenon (Liu et al., 2020).

It has been reported that in vivo experiments of AMX caused by tiny effects on human gut microbiota composition (Vrieze et al., 2014; Reijnders et al., 2016). However, a definite shift in the phylum level in our research may attribute to the absence of disturbances from neurohumoral regulation, the individual differences, dietary habits, and physiological status using in vitro SHIME (Karl et al., 2018). The decrease of Proteobacteria shown in our study may also attribute to AMX is more effective for sensitive Gram-negative bacteria that belonging to Proteobacteria by interfering with the synthesis of bacterial cell wall peptidoglycan layer (Zapata and Quagliarello, 2015). And Bacteroides uniformis belonging to Bacteroidetes may be resistant to AMX to make them survive, because of the strong correlations were found in their abundance with bl2\_len and bl2be\_shv2 (R = 0.8, P < 0.01). Some studies even showed the opposite phenomena like increasing of Proteobacteria and decreasing of Bacteroidetes, which might be caused by the combined effects with other antibiotics

such as clarithromycin, fosfomycin, and metronidazole (Oh et al., 2016; Ishikawa et al., 2017). However, at the genus level, the blooming of intestinal microbiota such as Bacteroides uniformis and Klebsiella and decrease of Bifidobacterium occurred following AMX administration. These data were further supported by earlier reports (Ladirat et al., 2014a; Oh et al., 2016; Marzorati et al., 2017), while an increase of Parabacteroides was found in another study (Kabbani et al., 2017).

Amoxicillin treatment affected a vast number of sensitive intestinal bacterial species that directly engrossed their corresponding functions. As a result, the functionally redundant resistant intestinal bacterial species were initially present at low levels and became more abundant. Thus, to compensate this loss of function to maintain community function, and species with closer evolutionary relationship usually have more similar functions (Moya and Ferrer, 2016). For example, Van den Abbeele et al. (2012) have found that during antibiotic treatment, the abundance of antibiotic-resistant Pediococcus acidilactici was enlarged, while the amount of sensitive Lactobacillus mucosae was declined. Similarly, in this study, a reduction of sensitive Parabacteroides might be replaced by resistant Bacteroides species. Although the metabolic function of the intestinal microbiota could recover quickly, the unrestored composition of the intestinal microbiota was supported by increasing ARB, which may pose a significant therapeutic challenge and that needs more attention.

As demonstrated by several studies that administration of antibiotics is significantly associated with decrease in microbial community diversity and richness, our results that AMX exposure increased the microbial diversity seems counterintuitive (Dethlefsen et al., 2008). The study by Vrieze et al. (2014) discovered a trend toward increased taxonomic diversity after AMX exposure, which was in accordance with our study. However, in some other studies no significant effect on microbiome diversity was found, which were opposed to our study (Pallav et al., 2014; Zaura et al., 2015; Reijnders et al., 2016). For this phenomenon, the possible reason was that AMX resistant intestinal bacterial species could bloom and compensate for or even surpassed the loss of AMX sensitive species. Intestinal microbiota with higher biodiversity was usually more resistant to the perturbation and colonization by pathogens, which might help in maintaining the intestinal homeostasis (Li et al., 2016). However, the improved biodiversity by AMX exposure that discovered in our study is mainly caused by increased abundance of AMX resistant intestinal bacterial species, which included opportunistic pathogen of the human intestine. This phenomenon may not be considered healthy for people because the increasing of opportunistic pathogens may be related to some human diseases that may cause the transmission and diffusion of ARGs.

## Pathogen Contribute to the Increased Human Disease Pathway Genes and Beta-Lactam Resistance Genes

Some research articles have not mentioned the significant alteration of metabolic functions of the gut microbiota after AMX treatment (Vrieze et al., 2014; Reijnders et al., 2016), while the disruption of the metabolic activity of microbiota (increased in succinate and monosaccharide and oligosaccharide levels) in the fecal samples were found elsewhere (Ladirat et al., 2014a). Moreover, this study found that Klebsiella was positively associated with the gene numbers of functional pathways including cancer, metabolic and immune diseases, and other human diseases, which have not been reported yet in other research articles. K. pneumoniae is a typical Gramnegative bacterial pathogen, which frequently colonizes in the human gut and processes to infection diseases (Pena et al., 1997; Wiener-Well et al., 2010; Gorrie et al., 2017). The increase of the relative abundance of intestinal Klebsiella genus has been reported to associate with diverse human diseases such as pneumonia, inflammation, Crohn's disease, colitis, cystitis, liver abscess, and wound infections (Schneditz et al., 2014; Atarashi et al., 2017). All above findings suggested that Klebsiella might contribute to increase the pathway genes, including cancer, immune system diseases, infectious diseases, metabolic diseases, and neurodegenerative diseases.

It is also known from the literature that frequent use of antibiotics such as AMX may cause the transmission and proliferation of ARB and ARGs (Canton and Morosini, 2011; Zaura et al., 2015; Lenart-Boron et al., 2016). This study further demonstrated that the beta-lactamase resistance genes and the potential co-selected ARGs were increased after AMX exposure. The co-occurrence patterns between abundance of microbial taxa and that of ARG subtypes (**Figure 6**) showed that Bacteroides uniformis and Klebsiella might contributed to the increasing of ARGs including beta-lactam, tetracycline, and multidrug resistance. Therefore, treatment with AMX increases the number of resistance genes, even those not directly related to the antibiotic ingested. Bacteroides uniformis and Klebsiella are ubiquitous in the environment and are commensal in the human gut, notably the increased of their population, especially for Klebsiella is not usually considered as healthy. Moreover, resistance to beta-lactams via beta-lactamase production has mostly been described in Klebsiella spp. (Madhi et al., 2013; Marco et al., 2017). Bifidobacterium and Parabacteroides are most common or novel probiotics (Pinto-Sanchez et al., 2017; Wu et al., 2019), and the decreasing of these probiotics may cause dysbiosis of gut microbiota, which also leads to human health problem (Wischmeyer et al., 2016; George Kerry et al., 2018).

## AMX Caused "SHIME-Compartment" Different and Permanent Alterations

In this study, Klebsiella and Bacteroides uniformis were found to be more significantly enriched in the ascending colon than other two colonic-regions, which was in keeping with the previous reports of the alteration in intestinal microbiota by exposure of antibiotics mixture and other toxic compounds like chlorpyrifos and arsenic (Reygner et al., 2016; Yu et al., 2016; Marzorati et al., 2017). These studies revealed the "SHIME-compartment" specific effects that could be due to the inconsistent biodegradation of multiple compounds, gut microbiome community, and pH in different colon regions (Reygner et al., 2016; Yu et al., 2016; Marzorati et al., 2017). Moreover, a large number of functional pathways related genes was also shown to be increased in the ascending colon after AMX exposure, which further clarified that the primary effect observed at the level of the microbiota could also be identified at the genomic and metabolic levels (Garcia-Villalba et al., 2017; Wang et al., 2018). The intestinal microbiota is a key "organ" for the individual's health, and its response to AMX is started from the proximal colon and hence observed more obvious shifts there. This phenomenon suggested that more studies should focus on the proximal colon; however, it may be challenging to study by in vivo experiments that usually analysis the feces standing for the distal intestinal microbiota.

Furthermore, these AMX effects were still evident for at least 2 weeks after the AMX discontinuance, although the resilient tendency of microbial composition, functional pathway genes and ARGs were also observed. Pallav et al. (2014) revealed that AMX mediated opportunistic pathogen such as Escherichia/Shigella persisted up to 42 days after the interruption of antibiotic therapy. Similar phenomenon was also reported by Zwittink et al. (2018), whose work primarily discovered a 5 days treatment with combined of AMX and ceftazidime allowed Enterococcus to thrive and remain dominant up to 2 weeks, while the abundance of Bifidobacterium remained decreased till postnatal of 6 weeks after antibiotic treatment discontinuation. Therefore, our results demonstrated a serious negative side-effect of AMX, which could be persistent and different in the specific colon region and should be considered as an essential aspect of the risk assessment for AMX prescription.

## PERSPECTIVES

Typical antibiotics have been detected in the collective gut of the Chinese population in our previous research, which provided a reference for this study of the effects of AMX on human gut microbiota (Wang et al., 2020). This study provides the originality of being able to control the effect of antibiotics in the different parts of the colon, evaluating in situ the alterations of the microbiota. Although SHIME has been successfully manipulated according to previous classic studies (Van den Abbeele et al., 2013; Yu et al., 2016; Wang et al., 2018), there were still three limitations of this study. First, AMX could not be absorbed by SHIME on the small intestine nor interact with the upper microbiota, which make the burden reaching the colon in the SHIME is much higher comparing to practical exposure to human colon. Considering that about 77 to 93% AMX is absorbed by the gastrointestinal tract and their interaction with the upper microbiota, the putative exposure dosage of AMX in colon is about 10% of the original oral dosage (Spyker et al., 1977; Arancibia et al., 1980; Huttner et al., 2019). This model is better to include the interaction with the human cell to get a closer result to in vivo, which is also a very good suggestion. Therefore, we would like to use 60 mg day−<sup>1</sup> AMX (10% of this study) and include interaction with the human cell in our

future work, with which the results would be close to in vivo. Second, this study lack of biological replicates and appropriate controls, which makes it difficult to interpret the results in relation to the microbial community variations and changes observed. Therefore, we would collect biological replicates and run a parallel experiment with blank medium at T0 and other timelines in our future study to negate any background noise and make the research scientifically reasonable. Another limitation is that the composition of the microbiota from the donor has not been analyzed before introducing it into the simulator. The composition of the initial microbiota from the donor may affect the stabilized microbiota in SHIME. Therefore, we would analyze the composition of the microbiota from the donor before introducing it into the simulator and compare the stabilized microbiota in SHIME that using multiple donors either by mixing their fecal samples or adding individually to explore the effects of initial microbiota composition in our future work, which would make this study more interesting. Although our work has these limitations, our findings would be valuable for directing future work. The findings in this study suggested several numbers of opportunities for additional study. One avenue is to expand the analysis to incorporate multi-omics approaches of the metagenome, metatranscriptome, and metabolome. Functional and metabolite analysis by multi-omics approaches would refine the results of predicting microbial function. Including interaction with the human cell in the fermentation vessels or manipulate in vivo study is also a good avenue to confirm the actual AMX's effects on cancers, metabolic and immune diseases, and other human diseases that found in our in vitro experiments. On the other hand, the studies of virome and fungome may exert the substantial influences on the intestinal microbiome. Thus, an expanded analysis of other microorganisms will also be necessary. It is of interest to investigate the impacts of different kinds of antibiotics and a mixture of them on the gut microbiota. Moreover, research on the restoration effects of some prebiotics, probiotics, and synbiotics during and after antibiotic therapy is promising, which may help to discover several clinical strategies and restore side effects caused by antibiotics. As the increasing of opportunistic pathogens, functional genes and ARGs by antibiotics exposure may pose a significant therapeutic challenge, it should be more critical to take some efficient measures to reduce or even eliminate the effects caused by antibiotic treatment.

## CONCLUSION

Exposure to AMX had significantly altered the overall taxonomic composition of the gut microbiota with increasing taxonomic richness, functional pathway genes, and beta-lactam resistance genes. The changes were "SHIME-compartment" different and more substantial effect was observed in the ascending colon. The shifted human gut microbiota could not be restored after 2 weeks' of AMX discontinued. Importantly, most of the functional pathway genes and quantified beta-lactam resistance genes are positively associated with bacteria that increased after AMX exposure. Our results may open up new perspectives for the assessing of direct effects by antibiotics on the intestinal microbiota – a key "organ" in individual health. The results demonstrated the negative side-effects of AMX and should be considered for AMX prescription.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study can be found in the NCBI SRA database, SRR9330193–SRR9330210.

## ETHICS STATEMENT

The study was approved by the Biomedical Ethics Committees of Nankai University. The participant has given written informed consent to confirm they understand the study purpose, procedures, risks, benefits, and rights.

## AUTHOR CONTRIBUTIONS

LL carried out the laboratory work, analyzed the data, and wrote the manuscript. QW provided suggestions in the manuscript preparation and revised this manuscript. RD edited the language and improved the clarity of this manuscript. HL, SW, HQ, and JY carried out the laboratory work. YX provided funding support. DM and YL guided the laboratory work and revised this manuscript. All authors read and approved the final manuscript.

## FUNDING

This study was sponsored by the Key projects of the National Natural Science Foundation of China (41831287), China National Funds for Distinguished Young Scientists (41525013), National Natural Science Foundation of China (31870351, 41703088, 31670509, and 21607016), Key projects of Research and Development of Hebei Province (19273707D), and Key projects of Tianjin Natural Science Foundation (19JCZDJC40800).

## ACKNOWLEDGMENTS

The authors express the sincerest thanks to Professor Bing Wu at Nanjing University for the guidance of constructing SHIME model, and Guangzhou Gene Denovo Co., Ltd. and Anhui MicroAnaly Gene Technologies Co., Ltd. for genome sequencing and analysis.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2020. 01213/full#supplementary-material

## REFERENCES

fmicb-11-01213 June 2, 2020 Time: 22:39 # 12


intestinal tract. J. Agric. Food Chem. 65, 5480–5493. doi: 10.1021/acs.jafc. 7b02049



infants. Eur. J. Clin. Microbiol. Infect. Dis. 37, 475–483. doi: 10.1007/s10096- 018-3193-y

**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 Liu, Wang, Lin, Das, Wang, Qi, Yang, Xue, Mao and Luo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(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.

# *Glycosmis pentaphylla* (Rutaceae): A Natural Candidate for the Isolation of Potential Bioactive Arborine and Skimmianine Compounds for Controlling Multidrug-Resistant *Staphylococcus aureus*

#### Natarajan Murugan1†, Ramalingam Srinivasan1,2,3†, Athiappan Murugan<sup>4</sup> , Myunghee Kim<sup>2</sup> \* and Devarajan Natarajan<sup>1</sup> \*

<sup>1</sup> Natural Drug Research Laboratory, Department of Biotechnology, School of Biosciences, Periyar University, Salem, India, <sup>2</sup> Department of Food Science and Technology, Yeungnam University, Gyeongsan-si, South Korea, <sup>3</sup> Department of Biotechnology, K. S. Rangasamy College of Arts and Science, Namakkal, India, <sup>4</sup> Department of Microbiology, School of Biosciences, Periyar University, Salem, India

Several multidrug-resistant organisms have emerged, which increases the threat to public health around the world and a limited number of therapeutics were available to counteract these issues. Thus, researchers are trying to find out novel antimicrobials to overcome multidrug-resistant issues. The present study aimed to isolate antibacterial principles against the clinical isolates of multidrug-resistant (MDR) Staphylococcus aureus from the ethyl acetate extract of Gycosmis pentaphylla. The isolation and structural characterization of bioactive compounds were carried out using various chromatographic techniques (TLC, column, HPLC, and LC-MS) and spectral studies such as FT-IR, CHNS analysis, <sup>1</sup>H-NMR, and <sup>13</sup>C-NMR. The antimicrobial potential of isolated compounds was assessed according to the standard methods. The isolated compounds were identified as arborine and skimmianine, which exhibited a significant antibacterial effect with the lowest MIC and MBC values against MDR S. aureus and in vitro kinetic and protein leakage assays supported the antimicrobial activity. Significant morphological changes such as uneven cell surfaces and morphology, cell shrinkage, and cell membrane damages were observed in the MDR S. aureus upon the treatment of arborine and skimmianine. The present investigation concludes that the isolated arborine and skimmianine compounds from G. pentaphylla harbor a strong antibacterial activity against MDR S. aureus and may be used as alternative natural drugs in the treatment of MDR S. aureus.

Keywords: *Gycosmis pentaphylla*, antibacterial compounds, chromatography, spectral study, MDR *Staphylococcus aureus*, arborine, skinmmianine

#### *Edited by:*

Eliana De Gregorio, University of Naples Federico II, Italy

#### *Reviewed by:*

Ajay Kumar Sharma, Institute of Nuclear Medicine & Allied Sciences (DRDO), India A. Gnanamani, Central Leather Research Institute (CSIR), India Giovanni Di Bonaventura, Università degli Studi G. d'Annunzio Chieti e Pescara, Italy

#### *\*Correspondence:*

Myunghee Kim foodtech@ynu.ac.kr Natarajan Devarajan natarajpu@gmail.com

†These authors have contributed equally to this work

#### *Specialty section:*

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

> *Received:* 25 April 2019 *Accepted:* 21 April 2020 *Published:* 10 June 2020

#### *Citation:*

Murugan N, Srinivasan R, Murugan A, Kim M and Natarajan D (2020) Glycosmis pentaphylla (Rutaceae): A Natural Candidate for the Isolation of Potential Bioactive Arborine and Skimmianine Compounds for Controlling Multidrug-Resistant Staphylococcus aureus. Front. Public Health 8:176. doi: 10.3389/fpubh.2020.00176

## INTRODUCTION

Staphylococcus aureus is a Gram-positive bacterium that belongs to the family Staphylococaceae and is often found at a commensal on the skin, skin glands, and mucous membranes, particularly in the nose of a healthy individual (1). It is causing infections ranging from relatively mild skin and soft tissue infections to life-threatening sepsis, pneumonia, osteomyelitis, endocarditis, as well as toxin-mediated syndromes such as toxic shock syndrome and food poisoning (2). Multidrug resistant S. aureus (MRSA) is resistant to two or more antimicrobial agents like penicillin, oxacillin, ampicillin, and methicillin. In general, MRSA infections have been categorized into four groups based on their sources: healthcare-associated hospital-onset MRSA (HAHO-MRSA), community-associated MRSA (CA-MRSA), healthcare-associated MRSA with community-onset (HACO-MRSA), and livestock-associated MRSA (LA-MRSA) (3). Although MRSA infections mainly occur in hospitals, the human illness caused by community-associated MRSA (CA-MRSA) is increasing considerably (4).

In recent years, drug-resistant human pathogenic bacteria have been frequently reported (5). The drug-resistant Enterotoxigenic Escherichia coli pathogenic bacteria demand the development of new antimicrobial drugs with a novel mode of action, targeting either the cell membrane or intracellular targets (6). Increasing the bacterial resistance is prompting a resurgence in the research on the antimicrobial role of herbs available in nature as effective combinations against antibioticsresistant bacteria (7). Moreover, using natural products also help to diminish the toxicity of the drugs when they are used on humans (8).

Since ancient times, plants have been the primary sources of many therapeutic agents that possess several secondary metabolites with significant physiological effects. Recently, many active ingredients have been isolated from plants and those are used to develop synthetic analogs for the treatment of ailments. There are 300 plant species frequently used in 7,800 drug-manufacturing units around India, which consume more than 2,000 tons of herbs annually (9). Generally, bioactive principles are isolated from different parts of plants such as leaves, bark, roots, seeds, etc. based on folklore usage. According to WHO, medicinal plants would be the best source to obtain a variety of drugs (10), and ∼20% of the known plants were subjected to biological tests and a sustainable number of new antibiotics were introduced into the market (11). Concurrently, the characterizations of antimicrobial compounds from medicinal plants have been very challenging to the researchers (12). Hence, a systemic screening of plants for the identification of antimicrobial compounds to act against microbial pathogens needs an hour. Recently, many researchers have been focused on the isolation of potential bioactive compounds from medicinal plants to produce highquality and potential secondary metabolites responsible for the control of microorganisms (13, 14). With this background, the present study was aimed to isolate antibacterial principles from Glycosmis pentaphylla.

## MATERIALS AND METHODS

## Isolation and Identification of MDR *S. aureus*

About 500 specimens [wound (138), pus (122), blood (119), sputum (70), and urine (51)] were collected from patients suspected to have S. aureus infections from private and government hospitals in and around Salem and Namakkal Districts, Tamil Nadu, India. The present study has ethical clearance from the institutional ethical committee (reference no.: PU/IEC/HR/2014/008 dated: 31/06/2014), Periyar University, Salem, Tamil Nadu. The methods of isolation and identification of the selected MDR S. aureus strains were described in our previous published data (15, 16). The isolation and identification of MDR S. aureus were carried out by the standard methods such as colony morphology on differential media, microscopic observations, and biochemical tests, and confirmed by molecular analysis like 16s rRNA sequencing. The antibiotic-resistant potential of isolated S. aureus was identified using the standard antimicrobial susceptibility test and identification of antibioticresistant marker genes such as MecA, blaZ, Aph (III), etc. The 16s rRNA sequence of isolated MDR S. aureus (101, 270, 315, 319, and 410) strains used in the present study is deposited in GenBank and GenBank accession nos. are KU198419 (S. aureus 101), KX290715.1 (S. aureus 270), KX454514 (S. aureus 315), KX447584 (S. aureus 319), and KX447585 (S. aureus 410).

## Plant Material and Extraction

The healthy and young leaves of G. pentaphylla (**Figure 1**) were collected from Vellimalai village (Latitude 11◦ 47′ 55.6836′′ N, longitude 78◦ 41′ 58.0056′′ E), Villupuram district, Tamil Nadu, India. The taxonomic identification of collected plant material was done by the Botanical Survey of India (Southern Circle), Tamil Nadu Agricultural University, Coimbatore, Tamil Nadu, India (reference letter No BSI/SRC/5/26/2016/402). The voucher specimen of collected plant material was deposited (No. PU/BT/NDRL/2016/09) in the Natural Drug Research Laboratory, Department of Biotechnology, Periyar University, Salem, Tamil Nadu, India. The collected plant materials were washed twice in running tap water and shade dried at room temperature for 3 weeks. The air-dried plant leaves were pulverized, using an electric blender to make a fine powder. A total of 3 kg of powdered G. pentaphylla leaves was sequentially extracted with different organic solvents (1:10 solvent ratio) in an increasing polarity (hexane, chloroform, ethyl acetate, acetone, and methanol) using a Soxhlet apparatus until the efflux solvents become colorless. All extracts were filtered through filter paper (Whatman No. 1) and dried under vacuum at 40◦C. The dried crude solvent extracts were stored in a freezer at −4 ◦C for further study (17).

## Isolation of Antibacterial Principles

The preliminary antibacterial activity results of various crude extracts of G. pentaphylla (15) revealed that the crude ethyl acetate of G. pentaphylla possesses a significant high antibacterial

potential; thus, the ethyl acetate extract was selected for isolation of active principles. The selected ethyl acetate extract was subjected to different chromatography techniques and the structure of isolated bioactive compounds were identified using various spectral studies.

## Separation of Active Compounds by Chromatography

The ethyl acetate extracts (in both low and high concentration) were spotted on the TLC plate (TLC Silica gel F254, Merck, Germany, 10 × 10 cm<sup>2</sup> ) origin and kept in a TLC chamber containing the desired solvent system. The solvent traveled to the top of the TLC plate by capillary action until it reached the solvent front. The TLC plate was taken out of the solvent tank. The dried TLC plate was viewed by iodine vapor and visualized under UV light (low and high wavelength). The same method was followed to find the optimal solvent system with other TLC plates in various solvent systems like hexane:ethyl acetate, ethyl acetate:methanol, chloroform:ethyl acetate, and chloroform:methanol in 100:0 to 0:100 ratios by increasing 10% more polar solvents used in the solvent system until a good resolution was noticed (18). The selected bioactive extract of G. pentaphylla was subjected to column chromatography to obtain fractions by increasing polarity of eluents using different solvent systems (chosen from the TLC trials) of chloroform:methanol 100:0, 80:20, 70:30, 60:40, and 50:50 ratios by increasing 10– 20% more polar solvent used in the solvent system. The column (120 × 4 cm) was packed with a solution of silica gel (60– 120 mesh) with ethyl acetate using the wet slurry method. This involves preparing a solution of silica gel, with chloroform filled in the column up to one-third (about 40 cm) length. A significant amount of chloroform was added to the column and allowed to drain for silica gel setting in the column and the volume of the solvent collected was again poured back into the column.

The ethyl acetate extract (60 g) was absorbed into silica gel (100–200 mesh) by triturating in a mortar and left for 1 h to dry. The dried plant extract coated with silica gel was loaded on the prepared column by adding gently into the column filled with chloroform without any air bubbles and eluted with the desire solvent system. A total of 60 fractions was obtained and tested for their antibacterial potential against MDR S. aureus strains (clinical isolates) according to the method of Srinivasan et al. (19). The G. pentaphylla leaf ethyl acetate extract 9.7:0.3 chloroform:methanol ratio fraction (GPLCM-58F) showed significant antibacterial activity against S. aureus strains, which selected for further purification of bioactive compounds. The selected fraction was again monitored by TLC (precoated plate, 0.02 mm thick) to determine the optimal solvent system for further purification in column chromatography. Another small size column chromatography (60 × 2 cm) was employed for further purification of selected fractions as per the abovementioned procedures. The fractions were eluted with chloroform:methanol (9:1–9.8:0.2 ratio) solvent system by increasing 0.1% of the volume of the polar solvents used in the solvent system. All the collected column fractions were again examined for antimicrobial activity and purity by TLC (precoated plate, 0.02 mm thick) for the single spot and the R<sup>f</sup> values were calculated.

A total of two fractions (GPSC-55SF and GPSC-73SF) exhibited considerable antibacterial activity; these fractions again were subjected to preparative TLC purification (20). The preparative TLC plate was performed based on the noted R<sup>f</sup> values. The fractions were scraped with silica powder using the sterile scoop and the collected UV-active pure compounds were dissolved in ethyl acetate solvent, and it can allow stirring with a magnetic stirrer for 1 h. After the pure fraction of active compounds was separated in the solvent, it was collected and concentrated under vacuum conditions. The UV-active pure fractions of the compound (GP-1 and GP-2) with similar R<sup>f</sup> value was pooled together and subjected to the antimicrobial activity and spectral studies.

## Structural Identification of Isolated Compounds

## Liquid Chromatography–Mass Spectroscopy (LC-MS) Analysis

LC-MS analyses of the two isolated compounds were carried out according to the method of Natarajan et al. (21) using a Thermo/Finnigan Surveyor System consisting of a degasser, binary pump, autosampler, and column heater. The LC was coupled with the Ion Trap mass spectrometer (Thermofleet, LCQ-Fleet) supported with an ESI ion source. Data acquisition and mass spectrometric evaluation were carried out in a personal computer with Data Analysis software (Qual Browser; Thermo Electron, San Jose, CA). For the chromatographic separation, Acquity BEC 1.7-µm C18 column (2.1 × 50 mm) was used. The column was programmed to run 95% of 0.1% acetic acid in water and 5% acetonitrile for both compounds 1 and 2. The final, elution program was operated at the linear gradient level of acetonitrile from 100 to 5% for a minimum 2 min. The flow rate was 1 ml/min and the injection volume was 1 µl. The capillary voltage, column temperature, nebulizer pressure, and gas flow rate were set to 20 V, 300◦C, 40 psi, and 15 ml/min, respectively, for the entire MS analysis.

## Nuclear Magnetic Resonance (NMR) Spectroscopy Analysis

The structures of the isolated compounds were elucidated primarily by homonuclear (1D) NMR (22). <sup>1</sup>D NMR experiments included <sup>1</sup>H and <sup>13</sup>C NMR, which used to locate atom positions and fragment units. NMR spectra were recorded on a Bruker Avance 300-MHz and/or Bruker Avance 600-MHz spectrometers coupled with Topsin acquisition software. Samples were dissolved in 500–600 µl of suitable deuterated solvent. An NMR experiment on samples with very little mass was carried out using Shigemi NMR tubes (100–200 µl sample). Signals were recorded in chemical shifts (δ) and expressed in parts per million (ppm), with coupling constants (J) calculated in hertz (Hz).

## Fourier Transform Infrared Spectroscopy (FT-IR) Analysis

The UV-active pure compounds (GP-1 and GP-2) of fraction samples were used for FT-IR analysis. The dried compounds (each 3 mg) were encapsulated in 100 mg of KBr pellet, to prepare translucent sample discs. The compounds were loaded in an FTIR spectroscope (Shimadzu, Japan), measured from 400 to 4,000 cm−<sup>1</sup> and at a resolution of 4 cm−<sup>1</sup> .

## High-Performance Liquid Chromatography (HPLC) Analysis

The HPLC analysis of active compounds isolated from G. pentaphylla ethyl acetate extract was performed using the modified method of Hawry et al. (23). About 1 mg of concentrated sample was dissolved in 1 ml of chloroform:methanol (9.8:0.2) and 20 µl was injected to determine the purity of compounds.

## Antibacterial Activity of Isolated Compounds

The selected MDR S. aureus strain suspension culture was prepared by growing a single colony overnight in Luria-Bertani (LB) broth with a turbidity of 0.5 McFarland standards. The antibacterial activity of isolated compounds was evaluated against MDR S. aureus strains using the agar well diffusion method (24). The isolated compounds (10µg/ml) were added into the corresponding wells using a micropipette. Commercial antibiotic (amoxicillin) was used as a positive control, whereas DMSO served as a negative control. The plates were incubated at 37◦C for 24 h and the diameter of the growth inhibition zone was measured.

## MIC and MBC Test

Minimum inhibitory concentration (MIC) of the isolated compounds was carried out using the microdilution method (25), using LB broth (Hi-media, India), and inoculum was adjusted to 2.5 × 10<sup>5</sup> CFU/ml. In brief, 10 µl (2.5 × 10<sup>5</sup> CFU/ml) of each MDR S. aureus strains was added individually in 1 ml of LB broth. Different concentrations (0.5, 1, and 2 µg/ml) of the isolated compounds were mixed with test tubes containing the MDR S. aureus strains. After 24-h incubation, the MIC values were obtained by visual observation of bacterial growth. The minimum bactericidal concentration (MBC) value of the isolated compounds was evaluated using the method of Natarajan et al. (25). The MBC values were determined by sub-culturing (10 µl) the MIC dilutions into the sterile Müller Hinton agar plates and incubated at 37◦C for 24 h, and the results were observed.

## Mechanism of Action of Isolated Compounds Against MDR *S. aureus* Protein Leakage Assay

The impact of the isolated compounds on the MDR S. aureus cells was measured in the terms of leakage of intracellular protein materials. The MDR S. aureus were treated with the isolated compounds at 37◦C for 120 min; each cell suspension was centrifuged at 7,000 RPM. About 100 µl of each sample supernatant was mixed with 900 µl of Bradford reagent and then incubated for 10 min. The optical density was measured at 595 nm. Bovine serum albumin was used as a standard protein and the experiment was done in triplicate (26).

## In vitro Killing Kinetic Assay

The time-kill kinetic assays were performed in five test tubes (two sets) containing an initial inoculum of 1 × 10<sup>6</sup> CFU/ml in tryptic soy broth with isolated compounds according to the modified method of García et al. (27). Changes in the bacterial count during exposure of the isolated compounds to the bacteria were monitored in five test tubes. MDR S. aureus culture alone served as a control. The bacterial counts of the treated samples were determined in 3-h intervals up to 12-h periods (0, 3, 6, 9, and 12 h) of incubation. One hundred microliters of treated samples was diluted and 10 µl of each sample was spread on Baird–Parker agar plates and incubated at 37◦C overnight, and the control samples were incubated under the same conditions. The number of viable cells in each tube was estimated after counting bacterial colonies on plates and by multiplying the appropriate dilution factor (28). All the experiments were conducted in triplicate, and mean values were measured.

## SEM Analysis

The effect of isolated compounds on the morphology of MDR S. aureus strains was observed under scanning electron microscopy (SEM) after the bacterial cells were treated with the isolated compounds (arborine and skimmianine) for 6 h. After the treatment, 1 ml of each test bacterial strain was collected, centrifuged, and washed three times with phosphate buffer saline and incubated for 30 min at 4◦C and fixed with 2.5% glutaraldehyde. The MDR S. aureus cells were dehydrated in ethanol, freeze-dried under vacuum condition, coated with an ion sputtering apparatus, and observed through SEM (Hitachi S-3400N). All strains of MDR S. aureus cells that are not exposed to the isolated compounds and a standard reference strain of S. aureus (MTCC 96) [procured from Microbial Type Culture Collection and Gene Bank (MTCC), Institute of Microbial Technology, Chandigarh, India] cells were similarly processed and used as controls (29).

## RESULTS

## Isolation of Bioactive Compounds

Chloroform and methanol were used in various proportions as mobile phases. After elution, the purity of each fraction was tested by analytical TLC using chloroform:methanol in the ratio 9.7:0.3, which showed clear separation of fractions. The TLC containing nine major active compounds with different R<sup>f</sup> values and similar R<sup>f</sup> value fractions were scraped and pooled together (**Figure S1**) for further analysis. The ethyl acetate solvent extract of G. pentaphylla leaves (60 g) was applied to a silica gel column for isolation of the bioactive compounds. The column was eluted with a linear gradient solvent system consisting of chloroform:methanol, by increasing 10% polarity (100:0 to 0:100) of the solvents and the fractions were collected in a glass container. The similar R<sup>f</sup> value fractions were pooled together according to the TLC profile and serially numbered (GP1, GP2, GP3, etc.) and all those fractions were kept at room temperature to allow condensation. After solvent evaporation, all collected fractions were tested for their antibacterial activity against the MDR S. aureus strains.

## Spectroscopic Analysis

The structural characterization of the isolated compounds was identified using NMR, LC-MS mass spectrometry, and HPLC analysis. UV and IR spectroscopy along with the determination melting point was carried out as they required for both the physical and structural characterization of isolated compounds.

## HPLC Analysis

The purity of isolated compounds was checked by HPLC analysis. The isolated active compound shows a separation peak at a retention time of 3.212 and 3.434 min for active compounds 1 and 2 respectively. In both the solvent system used, the purity of the active compounds was indicated as a single sharp peak (**Figure S2**).

## Structural Elucidation of Bioactive Compounds

## Physical Properties of Isolated Compounds

Compound 1 (arborine) was a yellowish-green color and soluble in ethyl acetate, methyl chloride, and DMSO, and insoluble in water. The melting point of the compound was 160–161◦C and the yield was 120 mg. The R<sup>f</sup> value of compound 1 was 0.98 mm in analytical TLC using a chloroform:methanol 9.8:0.2 solvent system as the mobile phase, whereas compound 2 (skimmianine) appeared as yellow color powder, soluble in ethyl acetate, methyl chloride, and DMSO. The yield of the isolated compound was 150 mg and the melting point was 180◦C. The R<sup>f</sup> value of compound 2 was 0.5 mm in analytical TLC using a chloroform:methanol 9.8:0.2 solvent system as the mobile phase.


## FT-IR Spectrum Analysis of Compounds

The FT-IR spectrum of arborine shows a broad peak at 1,602 cm−<sup>1</sup> , which corresponds to the carbonyl group. The aromatic C–H stretching peak was observed at 2,925 cm−<sup>1</sup> while aliphatic C–H stretching frequency was found at 2,853 cm−<sup>1</sup> . The absorption band at 1,264 and 1,402 cm−<sup>1</sup> corresponds to the C=N stretch of amide and CH<sup>3</sup> bend, respectively. The aromatic C–H out of the plane bending vibration was detected at 765 cm−<sup>1</sup> (**Figure S3**, **Table 1**). The FT-IR analysis of skimmianine showed an absorption band at 1,616 cm−<sup>1</sup> due to the C=N group and characteristic phenyl ether C–O stretching vibration observed at 1,266 and 1,056 cm−<sup>1</sup> . The C–H stretching vibration corresponding to aromatic was observed at 3,116 cm−<sup>1</sup> while aliphatic C–H stretching vibration was identified at 2,937 and 2,839 cm−<sup>1</sup> . The characteristic methyl bend and aromatic C–H bend out of plane vibrations were seen at 1,390 cm−<sup>1</sup> and 739 cm−<sup>1</sup> , respectively. The stretching frequency of C–O in the furan ring was observed at 1,056 and 1,192 cm−<sup>1</sup> (**Figure S7**, **Table 1**).

## LC-MS and Elemental Analysis of Compounds

The mass spectra of arborine showed a molecular ion peak at m/z 251 (M+1)<sup>+</sup> corresponding to the molecular formula C16H14N2O. The molecular weight was exactly matching with expected structure arborine (**Figure S6**). The result of the CHNS/O analysis of arborine showed carbon 76.22%, hydrogen 5.48%, nitrogen 10.96%, and oxygen 6.39%, which is in agreement with theoretical values (**Table 2**). The mass spectra of skimmianine showed a molecular ion peak at m/z 260 (M+1)<sup>+</sup> corresponding to the molecular formula C14H13NO4. The molecular weight of the compound exactly matched the expected structure of skimmianine (**Figure S10**). The result of the CHNS/O analysis of skimmianine showed carbon 63.36%, hydrogen 4.98%, nitrogen 5.02%, and oxygen 24.96%, which is in agreement with theoretical values. The elemental analysis result was interpreted with molecular mass, which revealed the molecular formula of the GP-2 compound as C14H13NO<sup>4</sup> (**Table 2**).


TABLE 3 | NMR ( <sup>1</sup>H-NMR and <sup>13</sup>C-NMR) spectral data of arborine compound.


## <sup>1</sup>H-NMR and <sup>13</sup>C-NMR Analysis of Compounds

The <sup>1</sup>H-NMR spectrum of arborine revealed a total of 14 protons present in the compound. The characteristic N-methyl proton was observed at δ 3.63 ppm as a singlet. The benzylic protons present in the arborine was seen at δ 4.29 ppm as singlet (**Figure S4**). In the downfield aromatic region, eight aromatic protons were observed between δ 7.28 and δ 7.36 ppm as a multiplet. The quinazoline fused aromatic ring protons numbered 3, 1, and 4 were observed at δ 7.46 ppm (triplet), δ 7.71 ppm (doublet), and δ 8.37 ppm (doublet), respectively (**Figure S4**, **Table 3**).

In the <sup>13</sup>C-NMR of arborine, N-methyl and benzylic carbons are observed at δ 34.8 and δ 43.4 ppm, respectively. The highest TABLE 4 | NMR (1H-NMR and <sup>13</sup>C-NMR) spectral data of skimmianine compound.


carbon chemical shift δ 169.0 ppm was observed for amide carbonyl carbon. The chemical shift for imine carbon (C=N) was seen at δ 162.2 ppm. The aromatic carbon chemical shifts were observed between δ 114.4 and δ 141.4 ppm (**Figure S5**, **Table 3**). Three methoxy groups present in skimmianine showed chemical shift values between δ 4.03 and δ 4.49 ppm. All the methoxy protons in the upfield showed a clear singlet pattern. The two protons correspond to the furan ring observed at δ 7.02 and δ 7.56 ppm, respectively, and those protons show a doublet splitting pattern. The other aromatic protons are seen at δ 7.23 and δ 7.99 ppm, which also showed a doublet splitting pattern (**Figure S8**, **Table 4**). In the <sup>13</sup>C-NMR of skimmianine, the characteristic three methoxy carbons were seen at δ 56.7, δ 58.8, and δ 61.5 ppm. The highest carbon chemical shift δ 164.2 ppm was observed for the methoxy group-attached carbon, which is present in the pyridine ring part of skimmianine. The other methoxy-attached carbons are seen at δ 152.0 and δ 157.0 ppm. The characteristic imine carbon C=N was observed at δ 142.1 ppm. All aromatic magnetically non-equivalent carbons are observed between δ 101.9 and δ 142.8 ppm (**Figure S9**, **Table 4**).

## Antibacterial Activity of Isolated Compounds

Arborine and skimmianine compounds showed significant antibacterial activity against clinical isolated MDR S. aureus

TABLE 5 | Anti-bacterial activity of arborine and skimmianine compounds against MDR S. aureus.


TABLE 6 | MIC and MBC values of arborine and skimmianine compounds against MDR S. aureus.


strains reported that 25 mm to 28mm of growth inhibition zones (**Table 5**). The results of the antibacterial activity of arborine and skimmianine clearly showed a dose-dependent effect (**Table 6**). The antibacterial activity results of arborine indicated that MDR S. aureus 101 and 410 strain was highly susceptible to the isolated compounds with the maximum inhibition of growth zone (28 mm) and lower MIC (0.2µg/ml) and MBC (0.2µg/ml) values. Similarly, skimmianine compound exhibit the highest antibacterial activity against the MDR S. aureus 315 strain (28 mm) with the lowest MIC (0.2µg/ml) and MBC (0.2µg/ml) values. The antibacterial activity was noticed in low concentration with stronger inhibitory potential. The results indicated that arborine and skimmianine exhibit a strong growth inhibition effect on all tested pathogens. The overall results indicated that the arborine and skimmianine possess better antibacterial activity than commercial antibiotics. Even a low MIC and MBC concentration of isolated compounds expressed a remarkable amount of bactericidal activity against the tested MDR S. aureus strains.

## Bacterial Mechanism Study of Compounds Time Kill Kinetic Assay

The time-killing kinetic assay was used to analyze the posttreatment bacterial viability and to define the minimum time required to obtain the bactericidal effect. Both arborine and skimmianine compounds show similar time-killing kinetic patterns of bactericidal effect on the MDR S. aureus strains (**Figures 2A,B**). Bactericidal activity was gradually increased by the time up to 12 h exposure of the MDR S. aureus strains against arborine and skimmianine compounds at their respective MBC concentration for both the strains and the MDR S. aureus strains

were killed within this period. The arborine and skimmianine compounds expressed a time-dependent and prompt bactericidal potential against the tested MDR S. aureus strains, which leads to bacterial death at the early stationary phase, as shown in time-kill curves (**Figures 2A,B**) compared with the control cells.

compound on different MDR Staphylococcus aureus strains.

## Protein Leakage Assay

Arborine and skimmianine compounds are known to enhance protein leakage by increasing the membrane permeability in MDR S. aureus strains. To determine the impact of arborine and skimmianine compounds alone on protein leakage, the cells were treated with arborine for 75µg/ml, resulting in all strains exhibiting 54% protein leakage, and skimmianine compounds were treated for 100µg/ml and the leakage of proteins was 55%. When the cells were treated with arborine and skimmianine compounds, the amount of protein released from the cells was increased compared to control (commercial antibiotics with recommended doses) (**Figures 3A,B**).

## Morphological Changes of MDR S. aureus

In order to have a better understanding of the antibacterial mechanism, arborine and skimmianine compounds were further studied for their effects on bacterial cell surfaces. Using SEM analysis, the present study observed the morphological changes in all MDR S. aureus strains and the reference standard S. aureus MTCC-96 strain, upon treatment with the antibacterial effect of arborine and skimmianine compounds (**Figure 4**). In MDR S. aureus, the untreated bacterial cells (control strains) appeared as grape-shaped (cocci), and the surfaces of the cells were intact with no damage observed. The S. aureus cells treated with arborine and skimmianine compounds, was not able to maintain the cocci (grape) shaped characteristics. Moreover, uneven fragments were observed, which indicated the damages induced on bacterial cell membranes. Moreover, the treated S. aureus cell surfaces were uneven, the size of cells were reduced, and the cells seemed to be damaged. Through the SEM analysis, the present study confirmed that the MDR S. aureus membrane surfaces were damaged, upon the treatment with arborine and skimmianine compounds isolated from G. pentaphylla.

## DISCUSSION

Medicinal plants have been the major source for innumerable therapeutic agents, which are of great importance to the health of an individual as well as communities (30). G. pentaphylla ethyl acetate crude extract was separated into nine fractions by thin-layer chromatography. Only two fractions showed excellent antibacterial property. This fraction was composed of two compounds that were visualized in TLC under the UV chamber. This study clearly shows that the active fractions obtained from column chromatographic separation are found to be mixtures of compounds. The purity of the fractions was obtained by column chromatography. The column purified compounds were

FIGURE 4 | The effect of arborine and skimmianine on the morphology of different MDR Staphylococcus aureus strains observed under scanning electron microscopy. (A) Control (without any treatment), (B) arborine-treated, and (C) skimmianine-treated S. aureus cells (the numbers on the images represent the S. aureus strain numbers).

characterized by <sup>1</sup>H NMR, C <sup>13</sup> NMR, FT-IR, HPLC, and LC-MS. The compounds GP-1 and GP-2 exactly matched the structure of arborine and skimmianine compounds with the molecular formula C16H14N2O and C14H13NO4, respectively. The <sup>13</sup>C NMR also supports the presence of the N-methyl and benzylic group. The characteristic functional groups found at 1602 cm−<sup>1</sup> (C=O) and 1402 cm−<sup>1</sup> (C–N) in the FT-IR spectra confirmed the arborine structure. Finally, the LC-MS also complies with the mass of arborine (MW 250), which is observed as an M+1 peak (251).

Similarly, Chakravarti et al. (31) have isolated the alkaloid arborine compound from the leaves of Glycosmis arborea, and its structure was further confirmed as 2-benzyl-1-methylquinazol-4-one based on UV, FT-IR, and <sup>I</sup>HNMR analysis. The present findings correlated with the observations of the arborine compound isolated from the stem of G. pentaphylla, which was already reported by Chakravarti et al. (32), Govindachari et al. (33), and Muthukrishnana et al. (34). Ghani (35) reported that the G. pentaphylla leaves contain alkaloids arborine, arbornine, skimmianine, glycorine, glycophymine, glycophymoline, glycosmicine, and glycomide.

The structure of skimmianine was confirmed by <sup>1</sup>H NMR, <sup>13</sup>C NMR, FT-IR, and LC-MS. In <sup>13</sup>C NMR, three methoxy carbons were observed at δ 56.7, 58.8, and 61.5 ppm. The FT-IR values of about 1,616 cm−<sup>1</sup> (C=N) 1,266, 1,056 cm−<sup>1</sup> (C–O, ether), and 1,056, 1,192 cm−<sup>1</sup> (C-O, Furan) confirmed the structure of skimmianine. The LC-MS also confirmed the molecular weight of skimmianine, which was observed as M+1 peak (MW 260). The findings were in good agreement with earlier reports of Sreelekha (36) who isolated the skimmianine from Zanthoxylum rhetsa. Desai et al. (37) have isolated arborinine, cycleanine, isochondrodendrine, and skimmianine from the leaves of G. pentaphylla. Similarly, Sinhababu and Takur (38) have reported alkaloids arborine, arborinine, skimmianine, glycorine, glycosmicine, and an amide that was isolated from the flower heads of G. pentaphylla. On the other hand, Chakravarty et al. (39) isolated a skimmianine compound from G. arborea. Another study was done by Mester (40) and Greger et al. (41) isolated well-known alkaloids, arborine, and skimmianine compounds from G. parviflora.

The isolated arborine and skimmianine compounds show significant antibacterial activity against multidrug-resistant S. aureus isolates. Similar findings were reported by Bowen et al. (42), Chakravarti et al. (31), and Chakravarty et al. (39) who carried out the antibacterial activity of arborine and skimmianine compounds against both Gram-positive bacteria (S. aureus and MRSA) and Gram-negative bacteria (E. coli and S. typhimurium), which were inactive against S. aureus. Previously, Jeyachandran et al. (43) have reported that the plumbagin bioactive compound was isolated from the root extract of Plumbago zeylanica, exhibiting more toxic potential against S. aureus. The in vitro time-kill assay is one of the most commonly used experimental models to assess the antibacterial activity, efficiently characterizing the rate, extent, and timing of bacterial killing and regrowth (44). The time-kill test finds out the differences in the rate and extent of antibacterial activity over time, and it can also provide growth kinetics information (45). The overall result of this assay revealed a stronger bactericidal effect in arborine and skimmianine compounds against S. aureus than in a laboratory medium. The evaluation of protein leakage of the arborine and skimmianine showed a significant strong effect on the tested MDR S. aureus strains. A similar kind of protein leakage when treated with oil compounds from medicinal plants supports the findings of the present study (46).

The SEM result of the present study shows that bacterial cell membranes were significantly affected by the activity of arborine and skimmianine compounds. It is indicated that arborine and skimmianine bioactive compounds disturb the bacterial membrane and death to cells. Similar findings were reported by Campos et al. (47) who identified that the S. aureus and E. coli bacterial membranes and cytoplasm were highly affected due to the passive diffusion of plant metabolites into the cells that caused consequent cell disruption. These changes in bacteria cells may be due to the lysis of membrane and transformation caused by the damage on the permeability and integrity of membrane from arborine and skimmianine compounds. Consequently, the changes can lead to loss of inner cell materials (48, 49). The results of SEM were in good agreement with other findings of Paul et al. (50) and Sharma et al. (51) reported that other antimicrobials treated cells. Shen et al. (29) have reported that S. aureus damaged cell membrane affects the cell permeability, and outflow of cellular components that lead to cell death which supported the present study.

## CONCLUSION

The present study indicated that G. pentaphylla is a good source of arborine and skimmianine, which acts as potent antibacterial agents against MDR S. aureus clinical isolates. Besides, these compounds induced significant strong bactericidal effects on the MDR S. aureus such as intracellular molecular imbalance and cell membrane disturbances that caused cell death. Hence, this study recommends that the isolated compounds can be used as a template molecule for pharmaceutical drug design for the treatment of diseases caused by MDR S. aureus.

## REFERENCES


## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## AUTHOR CONTRIBUTIONS

NM and RS equally contributed to the experiments and wrote the manuscript. AM, MK, and DN designed the research work and carry out the corrections in the article.

## ACKNOWLEDGMENTS

The authors acknowledge the Department of Biotechnology, Periyar University, Salem, Tamil Nadu, India, for providing necessary laboratory facilities, Tamil Nadu State Council for Science and Technology for financial support under the S&T Project Scheme (TNSCST/S&T Projects/RJ/BS/2013-2014), and DST-FIST (SR/FIST/LSI-673/2016) for strengthening the instrumentation facility of the Biotechnology Department of Periyar University. We thank Dr. S. Gnanavel, Department of Chemistry, Government Engineering College, Salem, for his technical assistance in the isolation of compounds and Dr. G. Ramkumar, ICAR-Indian Institute of Horticultural Research, for his suggestions on the manuscript.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh. 2020.00176/full#supplementary-material


and Memecylon edule and their molecular docking studies (Ph.D Thesis). Periyar University, Salem, Tamil Nadu (2014). p. 20–43.


**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 Murugan, Srinivasan, Murugan, Kim and Natarajan. 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.

# Extended-Spectrum Beta-Lactamase-Producing *Escherichia coli* in Drinking Water Samples From a Forcibly Displaced, Densely Populated Community Setting in Bangladesh

Zahid Hayat Mahmud<sup>1</sup> \*, Mir Himayet Kabir <sup>1</sup> , Sobur Ali <sup>1</sup> , M. Moniruzzaman<sup>1</sup> , Khan Mohammad Imran<sup>1</sup> , Tanvir Noor Nafiz <sup>1</sup> , Md. Shafiqul Islam<sup>1</sup> , Arif Hussain<sup>1</sup> , Syed Adnan Ibna Hakim<sup>2</sup> , Martin Worth<sup>2</sup> , Dilruba Ahmed<sup>1</sup> , Dara Johnston<sup>2</sup> and Niyaz Ahmed<sup>1</sup>

#### *Edited by:*

Ilana L. B. C. Camargo, University of São Pãulo, Brazil

#### *Reviewed by:*

Rebecca Ashley Gladstone, University of Oslo, Norway Sebastian Guenther, University of Greifswald, Germany

> *\*Correspondence:* Zahid Hayat Mahmud zhmahmud@icddrb.org

#### *Specialty section:*

This article was submitted to Infectious Diseases - Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

> *Received:* 31 October 2019 *Accepted:* 14 May 2020 *Published:* 18 June 2020

#### *Citation:*

Mahmud ZH, Kabir MH, Ali S, Moniruzzaman M, Imran KM, Nafiz TN, Islam MS, Hussain A, Hakim SAI, Worth M, Ahmed D, Johnston D and Ahmed N (2020) Extended-Spectrum Beta-Lactamase-Producing Escherichia coli in Drinking Water Samples From a Forcibly Displaced, Densely Populated Community Setting in Bangladesh. Front. Public Health 8:228. doi: 10.3389/fpubh.2020.00228 1 International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh, <sup>2</sup> WASH Division, UNICEF Bangladesh, Dhaka, Bangladesh

## Introduction: Community-acquired infections due to extended-spectrum beta-lactamase (ESBL) producing Escherichia coli are rising worldwide, resulting in increased morbidity, mortality, and healthcare costs, especially where poor sanitation and inadequate hygienic practices are very common.

Objective: This study was conducted to investigate the prevalence and characterization of multidrug-resistant (MDR) and ESBL-producing E. coli in drinking water samples collected from Rohingya camps, Bangladesh.

Methods: A total of 384 E. coli isolates were analyzed in this study, of which 203 were from household or point-of-use (POU) water samples, and 181 were from source water samples. The isolates were tested for virulence genes, ESBL-producing genes, antimicrobial susceptibility by VITEK 2 assay, plasmid profiling, and conjugal transfer of AMR genes.

Results: Of the 384 E. coli isolates tested, 17% (66/384) were found to be ESBL producers. The abundance of ESBL-producers in source water contaminated with E. coli was observed to be 14% (27/181), whereas, 19% (39/203) ESBL producers was found in household POU water samples contaminated with E. coli. We detected 71% (47/66) ESBL-E. coli to be MDR. Among these 47 MDR isolates, 20 were resistant to three classes, and 27 were resistant to four different classes of antibiotics. Sixty-four percent (42/66) of the ESBL producing E. coli carried 1 to 7 plasmids ranging from 1 to 103 MDa. Only large plasmids with antibiotic resistance properties were found transferrable via conjugation. Moreover, around 7% (29/384) of E. coli isolates harbored at least one of 10 virulence factors belonging to different E. coli pathotypes. Conclusions: The findings of this study suggest that the drinking water samples analyzed herein could serve as an important source for exposure and dissemination of MDR, ESBL-producing and pathogenic E. coli lineages, which therewith pose a health risk to the displaced Rohingya people residing in the densely populated camps of Bangladesh.

Keywords: ESBL-producing *E. coli*, multidrug-resistant, drinking water, Rohingya camps, Bangladesh

## INTRODUCTION

Extended-spectrum beta-lactamase (ESBL)-producing E. coli have been recognized as a major multidrug-resistant bacteria implicated in serious hospital and community-acquired infections worldwide, especially in places where poor sanitation, and inadequate hygienic practices are very common (1–4). Infections caused by MDR-E. coli incur huge medical costs and limit treatment options (5–7).

Multidrug-resistant E. coli has been detected in different ecological niches in the community and environment (8, 9). For instance, ESBLs and New Delhi Metallo beta-lactamase 1 (NDM-1) producing E. coli were detected in drinking water and retail meat, respectively (10, 11). In Bangladesh, ESBL producing-E. coli were reported from drinking water as well as from river water samples (12, 13). Though E. coli has had a significant role in water microbiology as an indicator of fecal pollution, it is of greater public health concern when these E. coli isolates turn out to be multidrug-resistant pathogens (14). Detection of ESBL-producing E. coli in drinking water samples is important to recognize the risk of transmission of antimicrobial resistance (AMR) and gastrointestinal diseases. Transmission of ESBLencoding genes among bacteria is often plasmid-mediated (15), and aquatic environments provide ideal settings for horizontal transfer of AMR genes encoded on various forms of mobile genetic elements (16).

Though the majority of E. coli are typically innocuous, some E. coli variants are virulent and may inflict varying severity of enteric infections. Currently, there are six different E. coli pathotypes that have been documented to cause intestinal infections, they include, enterotoxigenic E. coli (ETEC), enteroinvasive E. coli (EIEC), enteropathogenic E. coli (EPEC), shiga toxin-producing E. coli (STEC), enteroaggregative E. coli (EAEC), and diffusely adhering E. coli (DAEC) (17). In Bangladesh, following rotavirus, the second most leading cause of diarrheal infections are caused by pathogenic E. coli (18). Several virulence genes such as st, lt (ETEC); bfp, eae (EPEC); aat, aai (EAEC) are associated with diarrheagenic E. coli pathotypes (19), which can be used to detect these pathotypes using PCR based gene amplification. Watery diarrhea is caused by the secretion of heat-labile (LT) and/or heat-stable (ST) enterotoxins from ETEC. Shiga toxin (Stx) expression is the unique feature of EHEC where systemic absorption of this toxin leads to possibly life-threatening complications. Multiple putative virulence factors expression for typical EAEC strains, containing the aggregative adherence fimbriae (AAF), dispersin, the dispersin translocator Aat, and the Aai type VI secretion system directs to adherence and triggering diarrhea. EPEC adhesion is associated with attaching and effacing adhesion and intestinal colonization, which also include bundleforming pili (BFP), EspA filaments and intimin (19, 20).

The contaminated drinking water was found to be responsible for a number of waterborne gastroenteritis outbreaks due to diarrheagenic E. coli (21–23). Therefore, it is pertinent to analyze the prevalence and properties of ESBL-E. coli from drinking waters in community settings, particularly, from human habitations that are projected to pose exceptionally high risks of waterborne diseases due to overcrowding, scarcity of safe drinking water, and unhygienic living conditions. In Bangladesh, the displaced Rohingya people are one such community with a population of ∼1.16 million who are living in 32 congested camps in a challenging hilly landscape of Cox's Bazar district (290,000 persons per square kilometer). This displacement, of a large population are facing compounding problems, particularly related to water, sanitation, hygiene and health care (24–27). Water from hand-pumped tube wells is the primary water supply for the people in Rohingya camps. Around 6057 water points and 50087 emergency latrines have been built (during the study in 2018). Moreover, in the absence of efficient community sanitation, insufficient sewage disposal, and treatment facilities, the risks of transmission of enteric pathogens become extremely high, and the community as a whole face serious public health concerns (28–30).

In our previous study, we analyzed source water (tubewell) samples, as well as POU drinking water samples, from Rohingya camps and found 10.5% source water and 34.7% POU water samples were contaminated with E. coli, which could cause waterborne diseases in the camps (26). An outbreak of ESBLproducing E. coli might create a medical emergency in a large congested habitation like Rohingya camps because of limited treatment options. The AMR surveillance, especially with regard to ESBL-producing E. coli has never been carried out in these camps. Therefore, this study aims to determine the prevalence of ESBL-producing, MDR, and virulent E. coli in drinking water samples. Furthermore, plasmid profiling and horizontal transfer of resistant gene analyses of isolated ESBL-producing E. coli will provide important insights in understanding the dissemination of resistance determinants.

## MATERIALS AND METHODS

## Bacterial Isolates

We employed drinking water samples from Rohingya camps collected in our previous study to culture E. coli isolates (26). From a total of 2512 E. coli contaminated water samples, 421 water samples were randomly selected for the present study. One random E. coli isolate was taken as a representative from each sample, which was further tested using the API-20E test kit (Biomerieux SA, Marcy-I'Etoile, France), and 384 API-20E confirmed E. coli were stored at −70◦C in 30% LB glycerolbroth for downstream analysis. Out of 384 E. coli isolates, 203 isolates were from the household water samples and 181 from source water samples. In brief, for each sample, 100 ml water was filtered through a 0.22µm membrane filter (Sartorius Stedim, Goettingen, Germany), the membrane filter paper was then firmly laid on the mTEC agar plate. Later, the culture plate was incubated for 2 h at 35 ± 0.5◦C, followed by further incubation for 22 ± 2 h at 44.5 ± 0.2◦C. After incubation, red to magenta-colored colonies, typical of E. coli colony was picked and subcultured on MacConkey agar plate and incubated at 35 ± 2 ◦C for 18 h to 24 h. After incubation, the characteristics of dark pink colonies typical of E. coli were obtained and confirmed using API-20E kit.

## Isolation and Confirmation of ESBL-Producing and Carbapenem-Resistant *E. coli* Using Chromagar

All 384 confirmed E. coli isolates were cultured on CHROMagar ESBL and CHROMagar KPC media at 37◦C for 18–24 h. The production of extended-spectrum beta-lactamases and carbapenemase was confirmed by observing the growth and characteristic colony morphology on respective culture media. Dark pink to reddish colonies on CHROMagar ESBL plate indicate ESBL producing E. coli whereas pink to reddish colonies on CHROMagar KPC media suggest carbapenemresistant E. coli.

## Confirmation of *E. coli* by VITEK 2

ESBL positive E. coli isolated from the ESBL CHROMagar plate were further confirmed by the VITEK 2 system (bioMérieux, Marcy I'Etoile, France) using VITEK 2 GN ID card. Enterobacter hormacchi (ATCC-700323) was used as a positive control for the identification in this system. For VITEK 2 assays, pure isolates were streaked on MacConkey agar plates and incubated at 35◦C overnight. One to three isolated colonies were selected from each MacConkey agar plate and suspended in saline for preparation of inoculum to obtain an absorbency of ∼0.5 McFarland Units before being subjected to VITEK 2 analysis.

## Detection of Diarrheagenic and ExPEC Associated Virulence Genes

Several virulence genes such as st, lt (ETEC); bfp, eae (EPEC); aat, aai (EAEC) are associated with diarrheagenic E. coli pathotypes (19) are used to detect the pathotypes using PCR based gene amplification. In the present study, PCR based screening of diarrheagenic virulence genes was carried out for all 384 E. coli isolates. Gene-specific primers entailing heat-labile (lt), heatstable (st), attaching and effacing (eae), anti-aggregation protein transporter (aat), bundle forming pilus (bfp) and aggR-activated island (aaiC) were used to detect the respective genes employing a multiplex PCR setup (31–34). The boiling lysis method was used to obtain the DNA template (35). A 3 µl template DNA was taken for a 25 µl PCR reaction containing 12.5 µl of 2X GoTaq G2 Green Master Mix (Promega, USA) with 0.44 µl of eae primer, the primers lt, st, aaiC, aat, bfp were taken in 0.4 µl volume each. The PCR was carried out at standard cycling conditions with an annealing temperature of 57◦C for 20 s. A separate PCR was performed for Shiga toxin genes (stx1 and stx2), which was described previously (36, 37). PCR for invasion plasmid antigen H (ipaH) and the invasion associated locus (ial) was performed according to the procedure described elsewhere (38–41). Primer details are tabulated in **Supplementary Table 1**.

To examine the presence of seven ExPEC associated virulence factors, we performed two multiplex PCRs on 55 non diarrheagenic ESBL-E. coli isolates (42). Among these, the first one was done to screen the presence of kpsMII (group II capsule), papA (pilus-associated protein A), sfaS (S-fmbrial adhesine), and focG (F1C fmbriae protein) genes; whereas the second multiplex was performed to detect hlyD (haemolysin D), afa (afmbrial adhesine), and iutA genes (aerobactin siderophore ferric receptor protein).

## Antibiotic Susceptibility by VITEK 2

Antibiotic susceptibility testing (AST) was performed using VITEK 2 system with VITEK 2 cards (AST-N280) for 19 antimicrobial agents according to the CLSI guidelines and manufacturer's recommendations; two additional antimicrobial agents (cefixime and ceftazidime) were also incorporated. E. coli ATCC 25922, susceptible to all drugs, was used for AST in each VITEK testing step as quality control. The 21 antibiotics tested included amikacin, amoxicillin/clavulanic acid, ampicillin, cefepime, cefixime, cefoperazone/sulbactam, ceftazidime, ceftriaxone, cefuroxime, cefuroxime axetil, ciprofloxacin, colistin, ertapenem, gentamicin, imipenem, meropenem, nalidixic acid, nitrofurantoin, piperacillin, tigecycline, and trimethoprim/sulfamethoxazole. Minimum inhibitory concentrations (MIC) were determined, and the isolates were classified into resistant, intermediate and susceptible as per CLSI criteria. The raw MIC data from the VITEK 2 assay are shown in **Supplementary Table 2**.

## Detection of ESBL, Quinolone, and Carbapenemase Resistance Genes by PCR

All CHROMagar confirmed ESBL-producing E. coli isolates were screened for molecular determinants of ESBL and carbapenem resistance comprising of blaSHV, blaTEM, blaCTX−M−<sup>15</sup> and all CTX-M-groups (blaCTX−M−1−group, blaCTX−M−2−group, blaCTX−M−8−group, blaCTX−M−9−group) including, blaOXA−1−group, blaOXA−47, and blaNDM−<sup>1</sup> were screened (34, 43–48). The gene blaCMY−<sup>2</sup> encoding for AmpC βlactamase was also screened by PCR as per a published protocol (12, 49). Besides, all 66 isolates were tested for the three qnr (quinolone) resistance genes; qnrA, qnrB, and qnrS according to the methods described by others (50–52).

# Plasmid Profiling and Conjugal Transfer

Plasmids from ESBL-E. coli isolates were extracted employing the Kado alkaline lysis method (53) and visualized after separation in low percent agarose gel (0.7%) electrophoresis. The size of extracted plasmids was determined comparing with size standard plasmids ran alongside. The following plasmids were used as size standards; Sa (23 MDa), RP4 (34 MDa), R1 (62 MDa), pDK9 (140 MDa), and E. coli V517 plasmids (1.4, 1.8, 2.0, 2.6, 3.4, 3.7, 4.8, and 35.8 MDa) (54). ESBL-producing E. coli were used as donors, and the sodium azide resistant strain E. coli-J53 was employed as a recipient for conjugation using broth mating assays at 30◦C for 19 ± 1 h. Transconjugants obtained were plated on MacConkey plates prepared with cefotaxime (20 µg/L) and sodium azide (100 mg/L), transconjugants were selected observing their growth and colony morphology. Transconjugants were analyzed for antibiotic susceptibility tests using VITEK 2 assay, plasmid profiling and presence of ESBL genes (53).

## Phylogenetic Analysis

As per the method described by Clermont and colleagues, the distribution of phylogenetic groups among ESBL- E. coli isolates was determined by performing multiplex PCRs after DNA extraction by boiling lysis (55).

## RESULTS

## ESBL-Positive but Carbapenem Sensitive *E. coli* Recovered From Drinking Water

To screen for ESBL-producing E. coli, 384 isolates were cultured on ESBL CHROMagar. The typical growth of pink colonies on the CHROMagar plate was considered as ESBL positive E. coli. Of the 384 isolates tested, 17.2% (n = 66) were found to be ESBL-producing E. coli (**Table 1**). About 15% (27/181), and 19% (39/203) ESBL producing E. coli originated from E. coli contaminated source water and POU water samples, respectively. Further, the ESBL producing E. coli was investigated for carbapenem resistance on CHROMagar KPC media, and none of the isolates was able to grow on CHROMagar KPC media. Therefore, we did not detect any carbapenem-resistance in our collection of ESBL-E. coli isolates.

## High Prevalence of MDR in ESBL-Producing *E. coli*

Antimicrobial susceptibility of 66 ESBL-E. coli isolates (**Supplementary Table 3**) were tested against 21 different antibiotics (**Supplementary Table 4**) using VITEK 2 assay. All 66 CHROMagar confirmed ESBL positive E. coli isolates demonstrated resistance against ampicillin, ceftriaxone, ceftazidime, cefixime, cefuroxime, and cefuroxime axetil (**Figure 1**). About 70% (46/66) of the isolates were found to be resistant to nalidixic acid, 37.9% (25/66) isolates were resistant to trimethoprim/sulfamethoxazole, whereas 22.7% (15/66) and 19.7% (13/66) isolates were resistant for cefepime and ciprofloxacin, respectively. However, no resistance was detected in any of the tested strains to the antibiotics used of aminoglycosides (amikacin and gentamicin), cefoperazone/sulbactam, glycylcycline, carbapenem, and polymyxins groups. It was found that 71% (47/66) of E. coli isolates were MDR that were resistant to at least three classes of antibiotics. Among the 47 MDR isolates, 20 were resistant to three different classes, and 27 were resistant to four different classes of antibiotics.

## blaCTX-M-1 Group Is the Predominant ESBL Gene Detected

The presence of molecular determinants of ESBLs was tested on all ESBL-producing E. coli isolates. Out of the 66 ESBL-E. coli isolates, 59% (39/66) isolates harbored both blaCTX−M−<sup>1</sup> group and blaCTX−M−<sup>15</sup> gene. However, 4.5% (3/66) isolates harbored either blaCTX−M−1group or blaCTX−M−<sup>15</sup> gene. The blaTEM β-lactamase gene was present in 35% (23/66) of isolates, and none of the isolates harbored other β-lactamase genes such as blaSHV, blaCTX−M−2−group, blaCTX−M−8−group, and blaCTX−M−9−group. The two genes, blaOXA−1−group and blaOXA−<sup>47</sup> were screened among 66 ESBL- E. coli isolates, but none of the isolates was found to be positive. In addition, the New Delhi metallo-β-lactamase gene, blaNDM−<sup>1</sup> as well as plasmid-mediated ampC-type β-lactamase gene the blaCMY−<sup>2</sup> was not present in any of the isolates. The quinolone resistance gene; qnrS, and qnrB were found in 34% (22/66) and 5% (3/66) of the ESBL-E. coli isolates, respectively (**Figure 2**, **Supplementary Table 5**).

## Distribution of *E. coli* Pathotypes

Screening for the presence of virulence factors demonstrated that 7% (29/384) of E. coli isolates were positive for at least one virulence factor out of the 10 E. coli pathotype-specific virulence genes tested. Ten isolates were positive for only aaiC gene and five isolates were positive for both aaiC and aat; whereas four isolates were positive for both bfp and eae genes. Heat labile (lt) gene was present in 7 isolates, whereas the heat stable (st) gene was found in 2 isolates and a single isolate was found positive for stx1. None of the isolates was positive for ipaH and iaa genes. Among the 29 pathogenic E. coli isolates detected; 52% (15/29) were EAEC, 31% (9/29) were ETEC, 14% (4/29) were EPEC and 4% (1/29) were EHEC (**Figure 3**).

When the ExPEC virulence genes were screened only three out of the seven virulence factors were detected that comprised of KpsMII, sfaS, and iutA genes, their prevalence rates were, 21.8% (12/55), 5.4% (3/55), and 16.4 % (9/55), respectively. Most of isolates (12/18) harboring ExPEC genes were affiliated to phylogenetic group D. However, out of the 55 isolates tested, only 6 isolates qualify as ExPEC as per the inclusion criteria (strains harboring at least two ExPEC associated virulence factors) 5 out of these 6 isolates were from phylogroup D.

The potential pathogenic (diarrheagenic and ExPEC) E. coli isolates showed high resistance rates 83% (39/47) to ampicillin, followed by 74% (35/47) to nalidixic acid, 68% (32/47) to cefuroxime, cefuroxime axetil, 61% (29/47) to cefixime, ceftazidime, and 51% (24/47) to sulfonamides. Out of 29 pathogenic E. coli 11 were found to be ESBL producing in this study (**Table 2**). Of note, all the pathogenic isolates detected were found to be susceptible to carbapenems, aminoglycosides (amikacin and gentamicin), and polymixin.

TABLE 1 | Antibiotic resistance pattern, presence of antibiotic resistance genes and plasmid patterns of ESBL-producing Escherichia coli isolated from water sample.


(Continued)

#### TABLE 1 | Continued


<sup>a</sup>Ak, Amikacin; Amc, Amoxicillin/Clavulanic Acid; Amp, Ampicillin; Fep, Cefepime; Scf, Cefoperazone/Sulbactam; Cro, Ceftriaxone; Cxm, Cefuroxime; Cfa, Cefuroxime Axetil; Cip, Ciprofloxacin; Cl, Colistin; Etp, Ertapenem; Cn, Gentamicin; Imp, Imipenem; Men, Meropenem; NA, Nalidixic Acid; Fd, Nitrofurantoin; Tzp, Piperacillin-Tazobactam; Tgc, Tigecycline; Sxt, Sulphamethoxazoletrimethoprim; Cfm, Cefixime; Caz, Ceftazidime.

## Phylogrouping

Among the ESBL-E. coli isolates, all phylogenetic groups were represented except for phylogroup F. The predominant phylogenetic group identified was B1 (23/66; 34.8%), followed by D (22/66; 33.3%), E (17/66; 25.7%), B2 and C (2/66; each 3%). Among the 47 multidrug resistant ESBL-E. coli (20/47) were of phylogroup D followed by A (13/47), B1 (12/47), C (2/47). Majority of isolates carrying diarrheagenic virulence genes were from B1(14/29; 48%) followed by A(6/29; 20.6%), B2, C, and unknown groups (2/29; each 7%) (**Figure 4**).

## Plasmid Analysis of ESBL-Producing *E. coli*

Plasmid profiling and conjugation analysis were performed to see whether the antibiotic-resistance genes of the 66 ESBL producing isolates were plasmid-mediated and can they be horizontally transferred. Plasmid number and size were determined using conventional lysis and agarose gel electrophoresis. About 63% (n = 42) isolates carried 1 to 7 plasmids ranging in size from ∼1 to 103 MDa (**Figure 5**), and the distribution of plasmids was heterogeneous (**Table 1**). Further, the plasmid containing isolates that showed the ESBL phenotype were tested for their ability to transfer the ESBL determinant by conjugation experiments. Nine isolates were able to transfer the cefotaxime resistance marker to a susceptible E. coli recipient with transfer rates ranging from 4.75 × 10−<sup>6</sup> to 1.19 × 10−<sup>4</sup> per donor cell (**Table 3**). Large plasmids (30–103 MDa) were transferred to the sodium azide resistant E. coli-J53 recipient. Whereas, the smaller plasmids (<30 MDa) were not seen to be transferred during conjugation. Among the nine donor isolates, two were able to transfer two plasmids each whereas seven isolates transferred single plasmids (**Table 3**).

## DISCUSSION

In our previous study, we investigated the occurrence of E. coli and fecal coliforms in source and household drinking water samples in Rohingya camps, wherein 10.5% tubewell water and

34.7% POU water samples were found to be contaminated with E. coli (26). In the current study, the ESBL-producing E. coli isolates from the contaminated drinking water samples of our previous study were characterized concerning antimicrobial susceptibility, dissemination of drug resistance, and pathogenic potential to comprehend the extent of public health threat due to the exposure of contaminated drinking water in Rohingya camps of Bangladesh.

ESBL-producing E. coli has been increasingly reported globally, it is not only restricted to clinical settings but also recovered from environmental niches like livestock, wildlife and particularly water (56–58). The pandemic spread of ESBL-producing Gram-negative bacteria is a serious health concern. Human habitation and the surrounding environment in Bangladesh are reportedly contaminated by antimicrobialresistant bacteria (59, 60). Reports emanating from developing countries like Bangladesh indicated a high prevalence of ESBLproducing E. coli in hospital and community drinking water samples (34, 61).

In the current study, we detected 17.2% (66/384) ESBLproducing E. coli, of which 71% (47/66) was multidrug-resistant. The high prevalence of multidrug-resistant E. coli among the ESBL-producing isolates implies that not only β-lactam antibiotics but resistance to other classes of antibiotics is being co-selected. In Bangladesh, cephalosporins and penicillins are the most commonly used antibiotics (62), which explain why all the 66 ESBL-producing isolates are found to be resistant to both the classes of antibiotics. Besides, the majority of ESBLproducing isolates were found to be resistant to the quinolone class of antibiotics (46 to nalidixic acid). This may reflect the overuse and misuse of antibiotics (63) as these drugs are often sold and distributed over the counter (64). The uncontrolled and unregulated use of antibiotics severely limits the therapeutic options as well as aids the rapid dissemination of resistance in such overpopulated Rohingya camps.

We found ESBL-producing E. coli isolates are 100% susceptible to the antibiotics tested of carbapenem, aminoglycoside (amikacin and gentamicin), glycylcycline, and polymyxin groups. This finding was similar to a study in Jordan, where all the E. coli isolates from drinking water were sensitive to carpapenem and glycylcycline (65). There might be several factors responsible for susceptibility, such as these drugs are rarely prescribed in Bangladesh (64) and are not readily available in the hard to reach hilly terrain like Rohingya camps.

In this study, most of the isolates were positive for blaCTX−M−<sup>1</sup> group and blaCTX−M−15, genes that concur the previous reports from Bangladesh (59, 60, 66). All blaCTX−M−<sup>1</sup>

group, blaCTX−M−<sup>15</sup> and blaTEM ESBL-positive E. coli isolates showed MDR phenotype ranging from 3 to 4 classes of antibiotics. Similar to our observation, a previous study from Bangladesh showed a high prevalence of CTX-M-15 in ESBLproducing E. coli cultured from drinking water samples from different households (34). The CTX-M group of beta-lactamases are a group of rapidly emerging ESBL genes globally, which has been predominantly detected in E. coli and Klebsiella spp. (56, 67–70). The NDM-1 producing bacteria have been reported in clinical isolates from Bangladesh (71), but in the current study, no NDM-1-producing E. coli was detected in the water samples.

The gene qnrB has been recognized in various enterobacterial species, such as E. coli and Klebsiella spp. (72–74). Plasmidmediated quinolone resistance is intervened by the genes (qnr) encoding proteins that protect DNA gyrase and topoisomerase IV against quinolone compounds (75). Among the nonclinical sources, qnr gene was reported in E. coli isolated from swine, livestock and poultry (76, 77). In the present study, 25 isolates harbored plasmid-mediated qnr genes comprising of 22 isolates positive for qnrS and 3 isolates for qnrB genes. Isolates harboring qnrS gene also demonstrated co-existence of blaCTX−M−<sup>15</sup> and blaCTX−M−<sup>1</sup> group gene. Additionally, they were resistant to a

#### TABLE 2 | ESBL-producing pathogenic E. coli.


range of 7–11 antibiotics, including ciprofloxacin and nalidixic acid. In contrast, all isolates containing qnrS gene also harbored the two ESBL genes, blaCTX−M−1group and blaCTX−M−15; they were resistant to 8 different antibiotics; most interestingly, they were susceptible to ciprofloxacin. Hence, the presence of qnrS gene alone may not be indicative of the isolate being resistant to fluoroquinolones as also been observed in a previous study (78).

Using PCR for virulence genes, 29 (7%) was found to be pathogenic out of 384 isolates from drinking water. The most prevalent pathotype was EAEC, accountable for 52% of the pathogenic isolates; followed by ETEC, EPEC, and EHEC responsible for 31, 14, and 4% of the pathogenic isolates, respectively. In addition to the diarrheagenic E. coli around 22% of E. coli isolates were at least positive for 1 ExPEC associated virulence genes, moreover 11% (6/55) of the isolates were dectected to be potential ExPEC strains. This indicates that the drinking water samples present potential risk of disease epidemic particularly, the diarreaheal disease, this assumes more importants as in this particular setting where the dirinking water is not treated before consumption. Though the reservoir for

FIGURE 5 | Agarose gel electrophoresis of plasmid DNA showing the patterns among the ESBL positive isolates. Lane-3: E. coli strain pDK9 (140, 105, 2.7, 2.1 MDa), Lane-6: V517 (23MDa), Lane-9:RP4 (36MDa), Lane-11: V517 (35.8, 3.4,3.7, 2,1.8,1.4), Lane-13: R1 (62MDa), Lane-1,2,4,5,7,8,10, and 12 are ESBL positive E. coli. The molecular weight of the markers is shown in the picture.



EAEC is still unclear, it is generally considered to be human (79– 81). The transmission of EAEC is often described as waterborne or foodborne; therefore, it is assumed to be transmitted by the fecal-oral route (82). The presence of ETEC in drinking water and environmental water has been reported previously in Bangladesh; viability after long term water incubation and capacity of biofilm formation might imply that the water is an important transmission route of ETEC (83–85). From the 29 pathogenic isolates, 11 were found to be ESBL positive and surprisingly, 10 of them were of EAEC pathotype. In recent studies of Iran and China, a high prevalence of ESBL in EAEC was also reported (86, 87). The alarming rate of ESBLproducing EAEC isolates recommends strict infection control policies to prevent additional spreading of the virulent and resistant EAEC strains. All phylogenetic groups were represented in the E. coli isolates indicating that they were not homogenous in their population structure instead they belonged to diverse phylogenetic backgrounds, mainly the phylogenetic groups that are associated with commensal (group B1) as well as pathogenic and antimicrobial resistant E. coli lineages (group D) were detected. The presence of ESBL producing E. coli among the phylogenetic group D strains represents major public health risks due to the spread of such strains via drinking water.

In this study, 64% of the isolates were observed to harbor plasmids ranging from 1 to 103 MDa and a negligible similarity of plasmids pattern among the isolates inferred their clonal diversity due to heterogenous people of diverse geographical origin. Conjugation experiments are important to understand the transfer potential of plasmids conferring extended-spectrumβ-lactamase resistance. It is reported that only plasmids above 35 MDa contain and transfer antibiotic resistance genes via conjugation (18, 88), and In line with other studies, we have also observed plasmid-mediated transfer of antibiotic resistance genes (18, 88). Conjugative plasmids, carrying cefotaxime resistance phenotype among different isolates, ranged from 42 to 103 MDa in size. These findings imply that horizontal gene transfer might worsen the existing antibiotic resistance scenario by speeding up the spread of antimicrobial resistance (AMR) within heterogeneous bacterial communities in environment (89).

Lack of proper sanitation and hygiene in a densely-populated area like Rohingya camps (26) might play a key role in the development and dissemination of AMR. Open defecation with poor personal hygiene, poor community sanitation and lack of controlled antibiotic usage has been reported to exacerbate the transmission of AMR infections (90). A previous study in four middle-income countries, Brazil, Indonesia, India, and Nigeria showed that improvement in water quality and sanitation alone could lead to reduction in antibiotic usage (90). The contamination of drinking water with ESBL-producing E. coli, as observed in this study could be due to poor sanitation and hygiene, including, open defecation, inappropriate fecal sludge management, etc. This study has shown that the environmental E. coli pose public health threat by being carriers of ESBL-genes. Moreover, these ESBL-producing E. coli were harboring virulence factors corresponding to major E. coli pathotypes. Limitations of this study include lack of genetic fingerprinting analysis of the antibiotic-resistant E. coli from drinking water, lack of exhaustive antimicrobial resistance gene, and extraintestinal pathogenic E. coli (ExPEC) virulence gene screening.

In conclusion, the findings of this study suggest that the drinking water samples analyzed herein could serve as an important source for exposure and dissemination of MDR, ESBLproducing and pathogenic E. coli variants, which also pose a

## REFERENCES

1. Aruna K, Mobashshera T. Prevalence of extended spectrum beta-lactamase production among uropathogens in south Mumbai and its antibiogram pattern. EXCLI J. (2012) 11:363–72.

health risk to the displaced Rohingya population residing in the densely populated camps in Cox's Bazar. Based on the results of this work we recommend that the policymakers should make considerable efforts in implementing strong infection control strategies by focusing on providing good quality water and ensuring water quality monitoring programs in the Rohingya camps.

## DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

## ETHICS STATEMENT

The Ethical Review Committee of International Centre for Diarrhoeal Disease Research, Bangladesh has approved the study.

## AUTHOR CONTRIBUTIONS

ZM, MI, SH, MW, KI, DJ, and NA planed and organized the study. ZM, MI, SA, MK, MM, TN, SH, MW, DJ, DA, and NA were involved to implement the study. ZM, SA, MK, KI, DA, and NA carried out laboratory work. ZM, MM, MI, SH, DA, and NA were involved to interpret the data. ZM, MK, SA, MM, MI, and NA had a major contribution in writing the manuscript. All authors contributed in proofreading the manuscript.

## FUNDING

This research study was funded by United Nations International Children's Emergency Fund (unicef), grant number BCO/PCA/2017/035-2018/001. International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b) acknowledges with gratitude the commitment of unicef to its research efforts. icddr,b is also grateful to the Government of Bangladesh, Canada, Sweden and the UK for providing core/unrestricted support. The authors are grateful to Frederic Geser of unicef Bangaldesh for his constructive input and cooperation during the study. The authors are also grateful to Department of Public Health Engineering (DPHE) Cox's Bazar and unicef Cox's Bazar, Bangladesh for their valuable guidance and cooperation during the study.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh. 2020.00228/full#supplementary-material

<sup>2.</sup> Hussain A, Shaik S, Ranjan A, Nandanwar N, Tiwari SK, Majid M, et al. Risk of transmission of antimicrobial resistant Escherichia coli from commercial broiler and free-range retail chicken in India. Front Microbiol. (2017) 8:2120. doi: 10.3389/fmicb.2017. 02120


clinical isolates of Escherichia coli. Vet Microbiol. (2008) 132:414–20. doi: 10.1016/j.vetmic.2008.05.009


an endemic area by real-time PCR. J Appl Microbiol. (2008) 104:1128–36. doi: 10.1111/j.1365-2672.2007.03628.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 Mahmud, Kabir, Ali, Moniruzzaman, Imran, Nafiz, Islam, Hussain, Hakim, Worth, Ahmed, Johnston and Ahmed. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Molecular Mechanisms and Epidemiology of Fosfomycin Resistance in Staphylococcus aureus Isolated From Patients at a Teaching Hospital in China

Wenya Xu<sup>1</sup> , Tao Chen<sup>1</sup> , Huihui Wang<sup>1</sup> , Weiliang Zeng<sup>2</sup> , Qing Wu<sup>1</sup> , Kaihang Yu<sup>2</sup> , Ye Xu<sup>1</sup> , Xiucai Zhang<sup>1</sup> and Tieli Zhou<sup>1</sup> \*

<sup>1</sup> Department of Clinical Laboratory, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China, <sup>2</sup> School of Laboratory Medicine and Life Science, Wenzhou Medical University, Wenzhou, China

## Edited by:

Eliana De Gregorio, University of Naples Federico II, Italy

#### Reviewed by:

Que Chi Truong-Bolduc, Massachusetts General Hospital and Harvard Medical School, United States Hong-Ning Wang, Sichuan University, China

> \*Correspondence: Tieli Zhou wyztli@163.com

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 28 November 2019 Accepted: 20 May 2020 Published: 26 June 2020

#### Citation:

Xu W, Chen T, Wang H, Zeng W, Wu Q, Yu K, Xu Y, Zhang X and Zhou T (2020) Molecular Mechanisms and Epidemiology of Fosfomycin Resistance in Staphylococcus aureus Isolated From Patients at a Teaching Hospital in China. Front. Microbiol. 11:1290. doi: 10.3389/fmicb.2020.01290 Staphylococcus aureus is a major cause of hospital- and community-acquired infections placing a significant burden on the healthcare system. With the widespread of multidrugresistant bacteria and the lack of effective antibacterial drugs, fosfomycin has gradually attracted attention as an "old drug." Thus, investigating the resistance mechanisms and epidemiology of fosfomycin-resistant S. aureus is an urgent requirement. In order to investigate the mechanisms of resistance, 11 fosfomycin-resistant S. aureus isolates were analyzed by PCR and sequencing. The genes, including fosA, fosB, fosC, fosD, fosX, and tet38, as well as mutations in murA, glpT, and uhpT were identified. Quantitative real-time PCR (qRT-PCR) was conducted to evaluate the expression of the target enzyme gene murA and the efflux pump gene tet38 under the selection pressure of fosfomycin. Furthermore, multilocus sequence typing (MLST) identified a novel sequence type (ST 5708) of S. aureus strains. However, none of the resistant strains carried fosA, fosB, fosC, fosD, and fosX genes in the current study, and 12 distinct mutations were detected in the uhpT (3), glpT (4), and murA (5) genes. qRT-PCR revealed an elevated expression of the tet38 gene when exposed to increasing concentration of fosfomycin among 8 fosfomycin-resistant S. aureus strains and reference strain ATCC 29213. MLST analysis categorized the 11 strains into 9 STs. Thus, the mutations in the uhpT, glpT, and murA genes might be the primary mechanisms underlying fosfomycin resistance, and the overexpression of efflux pump gene tet38 may play a major role in the fosfomycin resistance in these isolates.

Keywords: fosfomycin, Staphylococcus aureus, resistance mechanism, mutation, tet38

## INTRODUCTION

Staphylococcus aureus is a kind of facultative anaerobe pathogenic Gram-positive coccus with strong resistance and tolerance to harsh environments (Wang et al., 2020). At present, S. aureus has become a significant pathogen of nosocomial infections, such as deep-seated skin and soft tissue infections (SSTI), endocarditis, and other life-threatening severe infections (Mehraj et al., 2016). In

recent years, with the widespread use of antibiotics, the emergence of multidrug-resistant (MDR) S. aureus has become a major concern (Gatadi et al., 2019). In addition, the lack of effective clinical treatments against MDR S. aureus has rekindled the interest of clinicians in fosfomycin. It is an antimicrobial agent that was discovered in Streptomyces sp. It exhibits broad-spectrum activity against both Gram-positive and Gram-negative bacteria by inhibiting the peptidoglycan synthesis pathway, which is essential for the synthesis of the cell walls (Shorr et al., 2017). However, the number of fosfomycin-resistant S. aureus strains is increasing rapidly (Etienne et al., 1991).

Several mechanisms of fosfomycin resistance have been proposed, including the plasmid-encoded fosfomycin-modifying enzymes (FosA, FosB, FosC, FosD, and FosX) and the acquisition of chromosomal mutations (Nakaminami et al., 2008; Liu et al., 2017; Silver, 2017). Mutations in the MurA target enzyme and transporters (GlpT and UhpT) have been shown to be responsible for fosfomycin resistance (Michalopoulos et al., 2011). Additionally, the overexpression of target enzymes, MurA and Tet38 efflux pump, also contributes to fosfomycin resistance in S. aureus (Truong-Bolduc et al., 2018). Notably, there are no reports yet suggesting that fosfomycin can stimulate the expression of the efflux pump gene and mediate drug resistance. However, a few studies on S. aureus have described the drug sensitivity and resistance mechanism of fosfomycin in S. aureus, although they are not fully understood.

In the present study, we focus on the mutations of the target enzyme MurA, which catalyzes the initial step in the biosynthesis of peptidoglycan and transporters (GlpT and UhpT), as well as the overexpression of murA and tet38 efflux pump in 11 fosfomycin-resistant S. aureus. In addition, a strong correlation was established between fosfomycin resistance and efflux pump gene tet38 overexpression that has not been reported previously. The results of quantitative real-time PCR (qRT-PCR) indicated that the Tet38 efflux pump plays a vital role in fosfomycin resistance by pumping out the drug.

## MATERIALS AND METHODS

## Bacterial Strains

In 2018, a total of 200 S. aureus isolates were obtained from the First Affiliated Hospital of Wenzhou Medical University, a comprehensive teaching hospital in China. The bacteria were identified by matrix-assisted laser desorption/ionization time of flight mass spectrometry (MALDI-TOF MS; BioMérieux, Lyons, France). S. aureus ATCC 29213 (American Type Tissue Culture Collection, Manassas, VA, United States) was used as an endogenous control strain in antimicrobial susceptibility testing experiments. The study and consent procedure were approved by the Ethics Committee of the hospital.

## Antimicrobial Susceptibility Testing

The minimum inhibitory concentration (MIC) of fosfomycin for each clinical strain was determined using the agar dilution method, wherein the media were supplemented with glucose-6-phosphate (25 mg/L), according to the recommendations of the Clinical and Laboratory Standards Institute [CLSI], 2018 (Ushanov et al., 2020). The data were interpreted according to the European Committee on Antimicrobial Susceptibility Testing criteria (available at http://www.eucast.org/clinical\_breakpoints/) (susceptible, ≤32 mg/L; resistant, ≥64 mg/L), and the fosfomycin-resistant isolates were selected for further investigation. In addition, the MICs of fosfomycin-resistant S. aureus to other classes of antibiotics, including oxacillin, erythromycin, ciprofloxacin, levofloxacin, gentamicin, rifampicin, linezolid, vancomycin, and teicoplanin, were detected using the broth microdilution method.

## Detection of Fosfomycin-Resistant Genes

The DNA of fosfomycin-resistant and fosfomycin-susceptible S. aureus isolates was extracted using a Biospin Bacterial Genomic DNA Extraction Kit (Bioflux, Tokyo, Japan) and was utilized as the template for PCR amplification of the fosA, fosB, fosC, fosD, fosX, glpT, uhpT, murA, and tet38 genes; the primers are listed in **Table 1**. The PCR products were sequenced by Beijing Genomics Institute Technology Co., Ltd. (Shanghai, China), and the sequences were aligned by BLAST on the NCBI platform<sup>1</sup> . The PCR products of uhpT, glpT, and murA were sequenced to scan for mutations.

## Fosfomycin Treatment and Total RNA Isolation

Actively growing S. aureus specimens were treated with increasing concentrations of fosfomycin (1/8 MIC, 1/4 MIC, and 1/2 MIC) for 2 h, after which the cells were harvested, and total RNA was extracted using a Bacterial RNA Miniprep Kit (Biomiga, Shanghai, China) according to the manufacturer's instructions. Then, 1000 ng RNA was used as the template for reverse transcription using a RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific, Waltham, MA, United States) to obtain cDNA.

## Quantitative Real-Time PCR (qRT-PCR)

qPCR was performed on a CFX-96 TouchTM Real-Time PCR system (Bio-Rad, Hercules, CA, United States) using TB Green Premix Ex Taq II (Tli RNase H Plus) (2×) (Takara, Japan), specific primers (**Table 1**), and 100 ng cDNA as the template. Cycling conditions were as follows: 95◦C for 30 s, followed by 40 cycles of 95◦C for 5 s and 60◦C for 20 s. A melting curve was performed after each run (raising the temperature by 0.5◦C/s, from 65 to 95◦C). Each sample was run in triplicate, and the means of the Ct values were used for analysis. The relative expression levels of tet38 and murA genes were normalized to the gmk reference gene (Chen and Hooper, 2018). The quantification of the target genes was analyzed using the comparative threshold cycle 2−11Ct method. All experiments

<sup>1</sup>http://blast.ncbi.nlm.nih.gov/Blast.cgi



\*PCR product including surrounding sequences adjacent to the target gene. \*\*GenBank-EMBL-DDBL accession number.

were repeated in triplicate independently. The relative expression of the mRNA of the target gene was normalized to that of S. aureus ATCC 29213.

## Multilocus Sequence Typing (MLST)

Isolates were screened using a previously described method to detect the following seven housekeeping genes: carbamate kinase (arcC), shikimate dehydrogenase (aroE), glycerol kinase (glp), guanylate kinase (gmk), phosphate acetyltransferase (pta), triosephosphate isomerase (tpi), and acetyl coenzyme A acetyltransferase (yqiL) (Enright and Spratt, 1999). The sequences of the PCR products were compared with those available from the MLST website<sup>2</sup> for S. aureus. Also, the allelic number was determined for each sequence.

## Planktonic Growth Assay

The planktonic growth rates of 8 tet38-overexpressed S. aureus isolates were determined as described previously (Wijesinghe et al., 2019), with some modifications. Briefly, 8 tet38-overexpressed isolates and ATCC 29213 standard cell suspensions were prepared by adjusting the turbidity of suspension to 0.5 McFarland standard in sterile saline. Then, 200 µL of each suspension was inoculated in 20 mL sterile LB

<sup>2</sup>http://www.mlst.net

broth containing fosfomycin in 0, 1/8 MIC, 1/4 MIC, and 1/2 MIC, respectively, for growth at 37◦C and 180 rpm for 24 h. The growth rate of the planktonic bacteria was determined by measuring the optical density (OD) of the suspension in each well of the 96-well plate at 600 nm at 2-h intervals for 24 h using a microtiter plate reader (BioTek, United States). The growth curve was generated in triplicate for each experiment. ATCC 29213 served as the control strain.

## Statistical Analysis

The relative expression of murA and tet38 was compared using Student's t-test, and P-value < 0.05 was considered to be statistically significant.

## RESULTS

## Susceptibility to Fosfomycin and Other Types of Antibiotics

The susceptibility to fosfomycin of 200 S. aureus isolates was determined by the agar dilution method using glucose-6 phosphate (25 mg/L). The results showed that 5.5% (11/200) of the isolates were resistant to fosfomycin. Also, resistance to other antibiotics was determined (**Table 2**); 81.8% (9/11) of the



FOM, fosfomycin; OXA, oxacillin; ERY, erythromycin; CIP, ciprofloxacin; LVX, levofloxacin; GEN, gentamicin; RIF, rifampicin; LNZ, linezolid; VAN, vancomycin; TEC, teicoplanin. Superscript "R" indicates resistance.

TABLE 3 | Characteristics and amino acid substitutions in fosfomycin-resistant and fosfomycin-sensitive S. aureus.


+, carries the corresponding gene; –, does not carry the corresponding gene; None: nonsense mutation; TypeAuhpT: T1369 G; TypeIuhpT: G1364A; TypeIIuhpT: T1368G; TypeAglpT: C299T; TypeBglpT: G1064A; TypeIglpT: G583A; TypeIIglpT: T829G; TypeAmurA: G187A; TypeBmurA: G349T; TypeCmurA: G770A; TypeImurA: C371G; TypeIImurA: A873T.

isolates displayed resistance to erythromycin, while 72.7% (8/11) belonged to MDR S. aureus.

## Molecular Mechanisms of Fosfomycin-Resistant Isolates

Strains carrying the fosA, fosB, fosC, fosD, or fosX gene were not found in the current study (**Table 3**). Based on the classification method of Fu et al. (2015) we named the sense mutations as TypeA, TypeB, and TypeC according to the amino acid sequence, and the nonsense mutations were named as TypeI, TypeII, and TypeIII; the subscripts represent different genes (Fu et al., 2015). Three distinct mutations were detected in the uhpT gene of the 11 fosfomycin-resistant S. aureus isolates and the corresponding sensitive strains. Mutation TypeAuhpT, found in JP3212, resulted in an amino acid substitution at position 457 (Leu457Val) of UhpT. Conversely, the other two mutations (TypeI–IIuhpT), which resulted in distinct amino acid substitutions within the UhpT protein, were identified in fosfomycin-sensitive isolates, although one mutation

(TypeIIuhpT) was also found in fosfomycin-resistant S. aureus (**Figure 1** and **Table 3**).

Moreover, four different mutations were detected in the glpT gene (TypeA–BglpT and TypeI–IIglpT). Notably, TypeBglpT, found in the fosfomycin-resistant isolates JP3535, produced a premature stop codon within the glpT coding sequence at position 355 (Trp335Ter), thereby resulting in truncated proteins. In addition, TypeIIglpT was detected only in the fosfomycin-sensitive isolates (**Figure 1** and **Table 3**).

Of the 11 fosfomycin-resistant isolates, 6 contained one of the three different mutations (TypeA–CmurA) in the murA gene. TypeA–CmurA, which resulted in distinct amino acid substitutions within the MurA protein at positions 63 (Ala63Thr), 117 (Gly117Trp), and 257 (Gly257Asp), could only be found in fosfomycin-resistant isolates, and two mutations (TypeI–IImurA) could be found in both fosfomycin-resistant and fosfomycin-susceptible S. aureus (**Figure 1** and **Table 3**). Moreover, only one sense mutation was present in each fosfomycin-resistant S. aureus isolate.

## Expression Analysis of murA and tet38

qRT-PCR revealed significant differences in the expression of murA between resistant and susceptible groups of S. aureus as compared with S. aureus ATCC 29213 (P < 0.05) (**Table 4**). The data showed that the average expression level of murA gene decreased by 0.7-fold in fosfomycin-resistant and fosfomycinsusceptible S. aureus isolates. In addition, compared with the fosfomycin-susceptible S. aureus, the expression of murA in the resistance isolates was not significantly higher.

However, the results (**Table 4**) indicated that compared with that in ATCC29213 and susceptible isolates, the expression level of efflux pump gene tet38 in JP3212, JP3535, JP359, and JP3600 was elevated. Notably, the level of the tet38 gene in JP3212, JP3535, JP3592, and JP3600 was altered markedly (21. 60-, 2. 74-, 143. 36-, and 24.59-fold) as compared with that in ATCC29213 (**Table 4**).

## Exposure to Fosfomycin Resulted in Increased Expression of tet38 Efflux Pump Genes Among Some Resistant Isolates

The expression of tet38 in the presence of increasing amounts of fosfomycin with 0, 1/8 MIC, 1/4 MIC, and 1/2 MIC concentrations was determined by qRT-PCR. Notably, the expression of the tet38 gene in JP3187, JP3212, JP3244, JP3505 JP3535, JP3539, P3589, JP3592, and ATCC 29213 was upregulated with the increase in fosfomycin concentration as compared with the 0 MIC strains (**Figure 2** and **Table 4**). Also, 4.63-fold and 6.42-fold increases were noted in the expression of tet38 in JP3505 cells treated with 1/8 MIC (64 mg/L) and 1/4 MIC (128 mg/L) fosfomycin, respectively, as compared with that in cells that were not treated with fosfomycin (**Table 5**). A further 8.46-fold increase was observed in those treated with 1/2 MIC (256 mg/L) fosfomycin.



S1–S5 represent the 5 fosfomycin-susceptible S. aureus strains. ATCC 29213 served as the control strain. <sup>a</sup>The relative gene expression with more than 2-fold change compared with ATCC 29213 after fosfomycin induction is shown in bold.

## Molecular Typing

The 11 fosfomycin-resistant S. aureus specimens were categorized into 9 STs (**Table 2**): ST1 (n = 1), ST5 (n = 3), ST59 (n = 1), ST7 (n = 1), ST239 (n = 1), ST965 (n = 1), ST4539 (n = 1), ST4739 (n = 1), and a new ST that was found in the current study (ST 5708) (n = 1).

## Growth Rate

In order to gain quantitative insight into the fitness cost imposed by tet38-overexpressed isolates, the growth curves of 8 tet38-overexpressed S. aureus were recorded. We identified a fitness cost after fosfomycin induction. The growth of 8 tet38 overexpression strains was inhibited in LB at a subinhibitory concentration of fosfomycin (**Figure 3**).

## DISCUSSION

Due to the unique mechanisms of action, fosfomycin exhibits significant antimicrobial activity against a broad spectrum of pathogens, including S. aureus (Goto, 1977). A review described that the susceptibility of S. aureus to fosfomycin ranged from 33.2% to 100% in the nine available studies [odds ratio (OD) = 91.7%, 95% confidence interval (CI): 88.7–94.9%] (Vardakas et al., 2016). In the current study, the susceptibility rate of fosfomycin in S. aureus was 94.5% (189/200). However, the prevalence of fosfomycin resistance in clinical isolates of S. aureus has been reported with increasing frequency in many areas (Del Rio et al., 2014; Mihailescu et al., 2014; Shi et al., 2014).

The resistance mechanism of fosfomycin in Gram-negative bacteria has been widely reported; also, in a previous study, we reported the resistance of fosfomycin in ESBL-producing Escherichia coli (Bi et al., 2017). Fosfomycin enters the cell via two transporters, GlpT and UhpT, and mutations or insertions in glpT and/or uhpT genes result in the loss of function (Takahata et al., 2010). According to the study by Castaneda-Garcia et al. (2009), glpT inactivation played an essential role in the resistance to fosfomycin in Pseudomonas aeruginosa (Castaneda-Garcia et al., 2009). The murA gene is also closely related to fosfomycin resistance (Takahata et al., 2010; Couce et al., 2012). In addition, fosfomycin activity can be inhibited via the catalytic activity of FosA, FosB, and FosC, respectively (Garcia et al., 1995; Lee et al., 2012).

Among Gram-positive bacteria, the resistance mechanism of fosfomycin is rarely reported. In the current study, none of

TABLE 5 | Relative expression of efflux pump gene tet38 in fosfomycin-resistant S. aureus exposed to different concentrations of fosfomycin.


<sup>a</sup>The relative gene expression with more than 2-fold change compared with 0 MIC after fosfomycin induction is shown in bold.

the resistant strains carried the fosA, fosB, fosC, fosD, or fosX gene, indicating that these genes might not be the primary factors mediating the resistance of S. aureus against fosfomycin. Other studies have shown that the prevalence of fosfomycin resistance genes (fosA, fosB, and fosC) was not the predominant factor contributing to resistance in S. aureus (Xu et al., 2017). In addition, a total of 12 mutations were found in 11 strains of fosfomycin-resistant S. aureus by sequencing analysis. Of

these, 3 were detected in uhpT, and TypeAuhpT was carried only by fosfomycin-resistant strain JP3212, while TypeIuhpT and TypeIIuhpT were found in both fosfomycin-resistant and fosfomycin-sensitive strains, which likely did not contribute to fosfomycin resistance. Within the glpT gene of 11 drug-resistant strains, 2 mutations TypeA–BglpT were observed only in the drug-resistant strains. On the other hand, mutation TypeIglpT was widely detected in both drug-resistant and susceptible strains. Intriguingly, TypeBglpT, which generated a stop codon (TA1064G) at position 335 (**Figure 1**), was harbored in JP3535. Also, we found a mutation (TypeIIglpT) merely in the fosfomycinsensitive strains. Of the five murA gene mutations found in the drug-resistant strains, TypeImurA and TypeIImurA could also be found in susceptible strains, and the remaining three mutations (TypeA–CmurA) were found only in drug-resistant strains (TypeAmurA: JP3505; TypeBmurA: JP3589; TypeCmurA: JP3189, JP3235, JP3244; JP3592). In addition, TypeImurA was also found in fosfomycin-resistant strains (10/11). Among the above mutations, four mutation sites, TypeAglpT, TypeBglpT, TypeCmurA, and TypeIImurA, were consistent with those reported (Fu et al., 2015). We also found that the frequency of murA mutation in S. aureus was high, which might play a major role in mediating fosfomycin resistance, which needs an indepth investigation.

Although several studies have mentioned that the overexpression of the murA gene can greatly increase the MICs of fosfomycin, the difference in murA expression between fosfomycin-sensitive and fosfomycin-resistant S. aureus has not been reported (Garcia et al., 1995; Olesen et al., 2014). In the current study, the results of qRT-PCR revealed a significant difference between the fosfomycin-resistant and fosfomycinsusceptible S. aureus with respect to the expression of murA as compared with that of S. aureus ATCC 29213. However, statistical differences could not be detected between two types of strains (**Table 4**), indicating that the overexpression of target gene murA has no role in conferring fosfomycin resistance in the strains identified in this study. Interestingly, some resistant strains showed a downward trend in the expression of murA, suggesting that the role of the murA gene in fosfomycin needs to be studied further.

Recent studies have shown that the tet38 gene exerts a specific effect on fosfomycin resistance. According to the study by Truong-Bolduc, the overexpression of tet38 resulted in a fourfold increase in the MIC of fosfomycin compared with that of the parent strain (Truong-Bolduc et al., 2018). The results of the current study showed that the expression of the tet38 efflux pump gene in fosfomycin-resistant strains JP3212, JP3535, JP3592, and JP3600 was significantly higher than that in the control strain ATCC29213 and the susceptible strains (P = 0.007, P = 0.002, P < 0.001, P < 0.001, respectively). Furthermore, under the treatment of 0, 1/8 MIC, 1/4 MIC, and 1/2 MIC with fosfomycin, we found that the expression of efflux pump gene tet38 was upregulated in most resistant isolates, even in the reference strain ATCC 29213. Although nonsense mutation was detected in uhpT, glpT, and murA genes, the level of tet38 in JP3600 was high even without the drug, which might explain the resistance to fosfomycin. This phenomenon suggests that the tet38 efflux pump plays a critical role in mediating fosfomycin resistance. Reportedly, abscess and other factors can promote the expression of tet38 (Chen and Hooper, 2018), and the current study has shown that the stimulation of the drug also enhances the expression of the efflux pump, albeit the specific mechanism remains to be studied further. Moreover, the overexpression of tet38 can also lead to changes in the cost of bacterial fitness. Some studies have demonstrated that the global regulator MgrA functions as a direct regulator of tetR21, which is a TetR-like regulator of the tet38 efflux pump gene. TetR21 acts as a repressor of tet38 expression and may also regulate the expression of other bacterial resistance determinants (Truong-Bolduc et al., 2005, 2015). We speculated that the high expression of the tet38 gene in S. aureus might be related to the regulation of TetRlike regulator TetR21 and the global regulator MgrA. We will also continue to focus on these phenomena in other bacteria in future studies.

MLST analyses designated three fosfomycin-resistant S. aureus isolated to ST5. Combined with drug sensitivity, we found that the ST5 resistant strains were resistant to at least five antibiotics. Among 11 fosfomycin-resistant strains, 72.7% were MDR strains, and further follow-up treatment was essential. Wu et al. (2018) reported that ST5 and ST239 strains were usually resistant to fosfomycin and constituted the predominant HA-MRSA clones in China. The new sequence type found in the resistant strain has been submitted to the repository (see text footnote 2).

## CONCLUSION

A total of 11 fosfomycin-resistant strains were screened out from 200 S. aureus isolates, and the mechanism was explored. Our findings indicated that fosA, fosB, fosC, fosD, and fosX genes might not be the major resistant mechanism of S. aureus to fosfomycin. The mutations within the glpT, uhpT, and murA genes might play a critical role in conferring fosfomycin resistance. However, the role of overexpression of murA in fosfomycin resistance needs to be discussed further in S. aureus. Also, the phenomenon of overexpression in the tet38 gene under a subinhibitory concentration of fosfomycin needs to be investigated further.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

## AUTHOR CONTRIBUTIONS

WX conducted the experiments, analyzed the data, and wrote the manuscript. TC participated in experiments and writing. HW and WZ provided fosfomycin-resistant strains and participated in the analysis of results. KY and QW participated in the analysis of the results. TZ helped to design the study. YX and XZ designed the study and corrected the manuscript. All authors read and approved the manuscript.

## REFERENCES

fmicb-11-01290 June 24, 2020 Time: 17:40 # 9


## FUNDING

This study was supported by a research grant from the National Natural Science Foundation of China (no. 81171614).


Wu, D., Chen, Y., Sun, L., Qu, T., Wang, H., and Yu, Y. (2018). Prevalence of Fosfomycin Resistance in Methicillin-Resistant Staphylococcus aureus isolated from patients in a University Hospital in China from 2013 to 2015. Jpn. J. Infect. Dis. 71, 312–314. doi: 10.7883/yoken.jjid.2018.013

Xu, S., Fu, Z., Zhou, Y., Liu, Y., Xu, X., and Wang, M. (2017). Mutations of the Transporter Proteins GlpT and UhpT confer fosfomycin resistance in Staphylococcus aureus. Front. Microbiol. 8:914. doi: 10.3389/fmicb.2017.00914

Zhang, X., Bi, W., Chen, L., Zhang, Y., Fang, R., Cao, J., et al. (2020). Molecular mechanisms and epidemiology of fosfomycin resistance in enterococci isolated from patients at a teaching hospital in China, 2013-2016. J. Glob. Antimicrob. Resist. 20, 191–196. doi: 10.1016/j.jgar.2019.08.006

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

# Alternative Therapeutic Options to Antibiotics for the Treatment of Urinary Tract Infections

Paul Loubet<sup>1</sup> , Jérémy Ranfaing<sup>2</sup> , Aurélien Dinh<sup>3</sup> , Catherine Dunyach-Remy<sup>2</sup> , Louis Bernard4,5, Franck Bruyère4,6, Jean-Philippe Lavigne<sup>2</sup> \* and Albert Sotto<sup>1</sup>

<sup>1</sup> VBMI, INSERM U1047, Université de Montpellier, Service des Maladies Infectieuses et Tropicales, CHU Nîmes, Nîmes, France, <sup>2</sup> VBMI, INSERM U1047, Université de Montpellier, Service de Microbiologie et Hygiène Hospitalière, CHU Nîmes, Nîmes, France, <sup>3</sup> Service des Maladies Infectieuses, AP-HP Raymond-Poincaré, Garches, France, <sup>4</sup> PRES Centre Val de Loire, Université François Rabelais de Tours, Tours, France, <sup>5</sup> Service des Maladies Infectieuses, CHU Tours, Tours, France, <sup>6</sup> Service d'Urologie, CHU Tours, Tours, France

#### Edited by:

Ilana L. B. C. Camargo, University of São Paulo, Brazil

#### Reviewed by:

Nina Chanishvili, George Eliava Institute of Bacteriophage, Microbiology and Virology, Georgia Jason John Paxman, La Trobe University, Australia Saeid Bouzari, Pasteur Institute of Iran (PII), Iran

\*Correspondence:

Jean-Philippe Lavigne jean.philippe.lavigne@chu-nimes.fr

#### Specialty section:

This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology

Received: 27 November 2019 Accepted: 10 June 2020 Published: 03 July 2020

#### Citation:

Loubet P, Ranfaing J, Dinh A, Dunyach-Remy C, Bernard L, Bruyère F, Lavigne J-P and Sotto A (2020) Alternative Therapeutic Options to Antibiotics for the Treatment of Urinary Tract Infections. Front. Microbiol. 11:1509. doi: 10.3389/fmicb.2020.01509 Urinary tract infections (UTIs) mainly caused by Uropathogenic Escherichia coli (UPEC), are common bacterial infections. Many individuals suffer from chronically recurring UTIs, sometimes requiring long-term prophylactic antibiotic regimens. The global emergence of multi-drug resistant uropathogens in the last decade underlines the need for alternative non-antibiotic therapeutic and preventative strategies against UTIs. The research on non-antibiotic therapeutic options in UTIs has focused on the following phases of the pathogenesis: colonization, adherence of pathogens to uroepithelial cell receptors and invasion. In this review, we discuss vaccines, small compounds, nutraceuticals, immunomodulating agents, probiotics and bacteriophages, highlighting the challenges each of these approaches face. Most of these treatments show interesting but only preliminary results. Lactobacillus-containing products and cranberry products in conjunction with propolis have shown the most robust results to date and appear to be the most promising new alternative to currently used antibiotics. Larger efficacy clinical trials as well as studies on the interplay between non-antibiotic therapies, uropathogens and the host immune system are warranted.

Keywords: urinary tract infection, alternative therapeutics, vaccines, nutraceuticals, immunomodulating agents, probiotics, cranberry, bacteriophages

## INTRODUCTION

Urinary Tract Infections (UTIs) are frequent bacterial infections (Silverman et al., 2013) especially in women, estimated to affect more than one in every two women at least once in her lifetime (Foxman, 2014; McLellan and Hunstad, 2016). Uropathogenic Escherichia coli (UPEC) is the main pathogen isolated from patients with UTIs (>85%) (Flores-Meireles et al., 2015), while other Gram-negative rods (e.g., Proteus mirabilis, Klebsiella pneumoniae) and Gram-positive cocci (e.g., Staphylococcus saprophyticus, Enterococcus faecalis) are responsible for the remaining cases (Flores-Meireles et al., 2015). UTI recurrence is defined by the occurrence of more than two episodes in 6 months, or three in 12 months (Professionals, 2019). The annual incidence of UTI in women is estimated to be 30 per 1000 subjects (Laupland et al., 2007), with approximately

20–40% experiencing recurrence within 6–12 months (Ikähelmo et al., 1996; Foxman, 2014). The epidemiology of UTI changes significantly in the healthcare environment. Urethral catheterization is strongly associated with UTI, and the risk of infection increases with the length of catheterization (Suetens et al., 2018). Catheter-associated UTI (CAUTI) is the most common nosocomial infection (Suetens et al., 2018). CAUTI affects both sexes, with long-term urinary catheterization of both men and women almost invariably leading to detection of bacteria in the urine (bacteriuria). Long-term catheterization carries a daily risk of 3–7% for the development of symptomatic CAUTI (Saint et al., 2009).

The urinary tract is normally sterile, with the exception of the flora of the distal urethra which is diverse and reflects both the digestive flora, the cutaneous flora and genital flora (Lactobacilli in women). There are several physiological mechanisms to prevent the host from the development of an ascending infection. First, the urethra itself, which is an obstacle to the intravesical inoculation; second, the physicochemical characteristics of normal urine (osmolarity, pH, organic acid content) that makes growth of most of the bacteria colonizing the urethra difficult; third urination that eliminates most of the bacterial population; fourth the presence in the urine of glycoproteins and oligosaccharides acting as soluble receptors to capture bacteria and enhance their clearance. Finally, in case of bacterial colonization, three factors contribute to avoid the invasion of the mucous membrane (Sobel, 1997): (i) the presence of inhibitors of bacterial adhesion to the surface of urothelial cells (Tamm-Horsfall protein, mucopolysaccharides); (ii) the existence of a local bactericidal effect (independent of inflammatory response or immune response); (iii) a process of exfoliation of the infected urothelial cells. The occurrence of UTI implies either a flaw in these defense mechanisms or the development in the urethral flora of a virulent bacteria, termed uropathogenic. Only a minority of E. coli strains, are endowed with uropathogenicity by the production of one or more adhesins (fimbriae): (i) type 1 allowing low urinary tract colonization, (ii) type P inducing pyelonephritis by modification of ureteral peristalsis in binding to glomerulus and endothelial cells of vessel walls helping E. coli to cross the epithelial barrier to enter the bloodstream and causing hemagglutination of erythrocytes and by decreasing the renal filtrate flow due to the formation of dense bacterial communities within the tubular lumen (Roberts, 1991; Melican et al., 2011), and (iii) non-fimbrial adhesins such as UpaB that facilitate E. coli adherence to extracellular matrix proteins and colonization of the urinary tract (Paxman et al., 2019). An increased adherence of E. coli to uroepithelial cells is observed in patients with recurrent UTIs compared to healthy controls (Schaeffer et al., 1981). Moreover, it has been demonstrated that UPEC can invade and replicate within the bladder cells to form intracellular bacterial communities (Mulvey et al., 2001), which can be frequently found in urothelial cells in women with symptomatic UTIs (Rosen et al., 2007) and may act as a source of recurrence in women with same-strain recurrent UTIs (Beerepoot et al., 2012a). Finally, biofilm formation is a critical aspect of CAUTI (Soto et al., 2006; Beerepoot et al., 2012a). Mechanisms of recurrence in UTIs are not fully characterized. Besides pathogen virulence factors, an impaired mucosal immune response (with urinary IgA involved in the UPEC clearance from the bladder mucosa) of the urogenital tract may have a role in the host-pathogen process (Ingersoll and Albert, 2013; Abraham and Miao, 2015).

Long-term low dose antibiotic use is currently the keystone of the preventive treatment for UTI recurrence. Indeed, prophylactic antibiotics have been shown to decrease UTI recurrence by 85% compared to patients with placebo (relative risk (RR) 0.15, 95% confidence interval (95%CI) 0.08 to 0.28) (Albert et al., 2004). Moreover, with regard to urinary tract conditions such as neurogenic bladder, it has been suggested that weekly cycling of antibiotics could be the most optimal preventative strategy (Salomon et al., 2006; Dinh et al., 2019). Indeed, this original strategy seems effective with only a limited ecological effect on native gut microbiota according to long-term follow-up (Poirier et al., 2015). However, prolonged antibiotic use often results in the emergence of multidrug-resistant organisms (Beerepoot et al., 2012b) and increases the cost of care. Consequently, the development of new therapeutic options to prevent and treat UTIs, and most particularly recurrent UTIs, are of interest.

This review aims to describe all the existing non-antibiotic treatment options in UTI (**Table 1** and **Figure 1**).

## VACCINES

Vaccines have been studied to prevent recurrent UTI in the aim not to kill infectious pathogens but to protect the host against infection by priming the immune response to uropathogens. Different vaccine strategies have considered the use of both surface antigen or inactivated whole bacterium, from uropathogens, to generate protective antibodies as a preventive strategy for recurrent UTIs. An ideal vaccine will target factors critical for establishment of bladder colonization (O'Brien et al., 2016). In this way, vaccines containing O antigens (important virulence factors that are targets of both the innate and adaptive immune systems), fimbrial subunits (responsible for the attachment to host cells, the first step of UTI), α-hemolysin (a membrane-active protein exotoxin leading to serious tissue damage), siderophores and a variety of outer membrane siderophore receptors (allowing the sequestration of iron, the main source of bacterial growth) have been developed.

## Current Vaccine Solutions

There are currently four available vaccines with established results from randomized control trials (RCT): Uro−Vaxom <sup>R</sup> , Urovac <sup>R</sup> , ExPEC4V and Uromune <sup>R</sup> (Aziminia et al., 2019).

– Uro−Vaxom <sup>R</sup> , also known as OM−89, is comprised of bacterial extracts from 18 UPEC strains that mediated its effect by the ability of bacterial component pathogenassociated molecular patterns to non-specifically stimulate cells of the innate immune systems (Huber et al., 2000). This effect was shown in mouse models inducing an immunological defense response within the bladder.

TABLE 1 | Non-antibiotic therapeutic options for the treatment of urinary tract infections.


#### TABLE 1 | Continued

fmicb-11-01509 July 2, 2020 Time: 16:19 # 4


However, its use in human trials has shown conflicting results. Four placebo controlled studies (Tammen, 1990; Schulman et al., 1993; Magasi et al., 1994; Bauer et al., 2005) showed that taking one Uro-Vaxom <sup>R</sup> tablet daily for 3 months significantly reduced the number of UTIs in the treatment group, with a RR = 0.61 (95%CI 0.48–0.78) of developing a UTI in the treatment group during an observation ranging from 3 to 12 months. However, a recent multicenter double-blind control trial of 451 patients showed no significant difference in UTI rates between Uro-Vaxom <sup>R</sup> and the placebo (Wagenlehner et al., 2015).

– Urovac <sup>R</sup> is a mucosal vaccine in the form of a vaginal suppository containing 10 different strains of heatinactivated uropathogenic bacteria (six serotypes of E. coli strains, P. vulgaris, Morganella morganii, E. faecalis and

K. pneumoniae) (Yang and Foley, 2019). The aim of this vaccine preparation was to incorporate a broader range of commonly implicated uropathogens and thus provide broad protection. Overall, Urovac <sup>R</sup> has been shown to reduce the risk of recurrence (RR 0.75, 95% CI 0.63–0.89) (Uehling et al., 1997, 2003; Hopkins et al., 2007). This effect was more pronounced in the group that received a follow-up booster vaccination (Aziminia et al., 2019).


antibiotic prophylaxis (Lorenzo-Gómez et al., 2013, 2015). A prospective observational study showed that 59 out of 75 women (78%) which received 3 months of Uromune <sup>R</sup> treatment as a sub-lingual spray once a day, had no new UTIs during both the treatment and the one year of followup period (Yang and Foley, 2018). Another recent lager prospective study showed that 65% of 784 women had 0 or 1 UTI after 6 months of daily sub-lingual administration of Uromune <sup>R</sup> (Ramirez Sevilla et al., 2019).

This interesting result should be confirmed in a larger study with a control group. It has to be noted that the need for a 3-month daily administration raises questions on the immunogenicity and the expected compliance to this treatment. One international multicenter phase III RCT is currently underway.

A recent meta-analysis noted that the use of vaccines appeared to reduce UTI recurrence compared to placebo. However, the heterogeneity amongst studies renders interpretation and recommendation for routine clinical use difficult at present (Prattley et al., 2020). Further randomized controlled trials are warranted to assess the efficiency and the safety of the existing vaccines against UTIs.

## New Candidates

fmicb-11-01509 July 2, 2020 Time: 16:19 # 6

Several bacterial virulence factors involved in UTI are promising vaccine targets.

## Vaccines Targeting Adhesion

Bacterial adhesion to urothelium represents a crucial step in the pathogenesis of UTIs. It permits bacteria to resist mechanical elimination by the flow of urine and bladder and increases persistence of E. coli. One large family of adhesive organelles are pili assembled by the chaperone-usher pathway (CUP) pili. These pili are critical virulence factors in a wide range of pathogenic bacteria, including E. coli. CUP pili mediate adhesion to host and environmental surfaces, facilitate invasion of host tissues, and promote interaction of bacteria with each other to form biofilms (Sauer et al., 2004). The most studied of the CUP pili are the Type 1, P, and S pili, present in UPEC and allowing the colonization of mucosal surfaces. They promote both irreversible bacterial attachment and invasion of uroepithelial cell membrane within the bladder. Interactions mediated by these adhesins can stimulate a number of host responses that can directly influence the outcome of a UTI (Mulvey, 2002). They are thus promising vaccine candidates because of the importance of this adhesion. The potential exists to develop antibodies that block the adhesin-host cell receptor interaction and thus decrease bacterial colonization (Wizemann et al., 1999).

FimH is a major determinant of the adhesive subunit of Type 1 fimbriae, which has high tropism for urinary tract receptors. It binds to mono-mannose. Lack of this compound on renal epithelia has suggested a limited role in pyelonephritis. It has been targeted as a good candidate for a vaccine because of its critical role in cystitis pathogenesis (**Figure 1**). A FimCH vaccine protected cynomolgus monkeys from cystitis (Langermann et al., 2000). Another study on mice with an intranasal and intramuscular administration also showed a protection against cystitis, but the intranasal method produced a stronger immune response (Poggio et al., 2006). In a more recent study, a recombinant protein MrpH.FimH (consisting of a combination between two adhesins, FimH from UPEC and MprH from P. mirabilis) was injected with a transurethral instillation and the authors demonstrated a high immune response and a protection against E. coli and P. mirabilis (Habibi et al., 2016).

By comparison the Type P pilus is more involved in pyelonephritis due to the presence of globoseries glycosphingolipids (the receptor of this pilus) in kidney. This P fimbria helps to cross the epithelial barrier to enter the bloodstream and can cause hemagglutination of erythrocytes (Riegman et al., 1988). Approximatively 1,000 of subunits form a P fimbria. Among them the major constituent is the protein subunit PapA, and minor subunits are PapD, PapE, PapF, and PapG. So subunit vaccines based on this pilus have been developed to block renal colonization. Attempts on using the major subunit PapA as a vaccine, although initially favorable in mice (O'Hanley et al., 1985), failed due to the poor generation of protective antibodies most probably due to natural variation in the PapA pilus subunit (**Figure 1**; De Ree and Van den Bosch, 1987). A new vaccine based on purified PapDG protein demonstrated a good efficiency to prevent colonization of kidney on cynomolgus monkeys (Roberts et al., 2004).

## Vaccines Targeting Capsule

The main role of a capsule is to cover and protect E. coli from the host immune system. It provides protection against engulfment and complement-mediated bactericidal effect in the host (Johnson, 1991). Based on this implication in virulence strategy, capsules represent a promising vaccine target. In this way, conjugate UTI vaccines against UPEC capsule and lipopolysaccharide (LPS) components have shown protection in animal models after same-strain challenge. In early studies, intraperitoneal, subcutaneous or bladder injection of O-antigen from E. coli of different serotypes (O6, O8 and K13) protected rats and rhesus monkeys from pyelonephritis (Kaijser et al., 1983; Roberts et al., 1993; Kumar et al., 2005). A considerable challenge in formulating a vaccine targeting capsule or O-antigen, the most exposed component of LPS, is the great heterogeneity of serotypes among E. coli isolates. Indeed, six different O serotypes account for only 75% of UPEC isolates (Stenutz et al., 2006), making the formulation of a broadly protective conjugate vaccine impractical. Furthermore, some capsule serotypes, such as K1 and K5, are thought to evade the host immune response by molecular mimicry, potentially making them poor vaccine candidates. No studies on these vaccines have been conducted in humans yet.

## Vaccines Targeting Toxins

Several toxins secreted by UPEC play a consequential role as virulence factors in UTI. They have the ability to alter the host cell signaling cascade and modulate inflammatory responses. They also contribute to the stimulation of the host cell death and the ability to access deeper tissues. Many toxins have been reported including α-hemolysin (HlyA) or cytotoxic necrotizing factor 1 (CNF1). α-hemolysin is a pore-forming toxin, it can lyse erythrocytes, and induces the apoptosis of target host cells promoting the exfoliation of bladder epithelial cells (Johnson, 1991). CNF1 stimulates actin stress fiber formation and membrane ruffle formation resulting in the entry of E. coli into the cells. This protein interferes with polymorphonuclear phagocytosis and provokes apoptotic death of bladder epithelial cells (Asadi Karam et al., 2019). Following their importance in virulence, the toxins produced by UPEC have been used to develop vaccines to protect the bladder and kidney during a UTI. However, as they are secreted and thus removed from the bacteria, they do not make ideal vaccine candidates. Indeed, a purified α-hemolysin toxoid vaccine prevented renal injury, but not colonization, in mice after challenge with a hemolytic UPEC strain (**Figure 1**; O'Hanley et al., 1991). Of note, rather than being secreted as "naked" proteins, α-hemolysin and CNF1 are associated with outer membrane vesicles (OMVs), which bleb from the surface of Gram-negative bacteria during all stages of growth (Ellis and Kuehn, 2010). OMVs also contain adhesins, enzymes, and non-protein antigens like LPS. OMVs are intriguing vaccine candidates, and because they contain LPS

and other pro-inflammatory virulence factors, they should not require adjuvants to stimulate the immune system. However, no UPEC OMV vaccines have yet been tested.

## Vaccines Targeting Iron Metabolism

The acquisition of iron is crucial for bacteria life and E. coli uses iron for transporting and storing oxygen, DNA synthesis, electron transport and metabolism peroxides. However, the amount of iron availability is reduced in host. In response, E. coli produces siderophores, molecules that mediate iron uptake. Four siderophore systems have been identified such as yersiniabatin, aerobactin, enterobacterin, and salmochelin (Johnson, 1991).

Some studies have targeted siderophore, heme receptors and other functional molecules involved in iron acquisition. In one study, the authors tested six iron receptors and found that the antibodies against the yersiniabactin receptor, FyuA, conferred kidney protection in mice (Brumbaugh et al., 2013). In another mouse model, instead of targeting iron receptors, the same authors targeted molecules involved in iron metabolism. Of the six candidates, two conferred a protection to the bladder (IreA and IutA) and one protected the kidneys (Hma) after an intranasal immunization (Alteri et al., 2009).

## SMALL COMPOUNDS

In the current setting small compounds are low molecular weight molecules that are typically bacterial substrates or products or mimics thereof. They can act as inhibitors by binding active sites and substrate binding sites of proteins involved in pathogenicity and so impact bacterial infections.

## Small Compounds Targeting Adhesion

As previously noted, one of the critical mechanisms for the pathogenesis of the uropathogenic bacteria is its adhesion to uroepithelium (Beerepoot and Geerlings, 2016), due to fimbriae (specially the Type 1 and the P-fimbriae), playing a role in both cystitis and pyelonephritis (Beerepoot and Geerlings, 2016; Muenzner et al., 2016). The very conserved structure of the adhesive organelles makes them good candidates to develop antibacterial agents (Piatek et al., 2013). The small molecules targeting adhesion can be classified into two categories: those inhibiting the capacity of adhesion of the fimbriae, and those targeting fimbriae assembly.

## Pilicide

The main action of these molecules is to prevent the formation of UPEC pili by decreasing the levels of Type 1 and P piliation (Åberg and Almqvist, 2007). Pilicides are small molecules which have a ring-fused 2-pyridone backbone. Some pilicides act directly on pili assembly chaperones, through adhering to their hydrophobic substrate binding sites (Svensson et al., 2001; Pinkner et al., 2006). Others interfere with the transcription of pili genes and some cases genes involved in flagella biogenesis such as the pilicide ec240, the most potent inhibitor of Type 1 piliation and of type 1 pilus-dependent biofilm formation to date (Greene et al., 2014).

In vitro studies testing the potential of pilicides have shown promising results. These compounds decreased (i) the adhesion of UPEC strains on cells by a strong reduction (70–80%) of fimbriae density (Piatek et al., 2013), and (ii) the ability to form Type 1 pilus dependent biofilms (Pinkner et al., 2006). In a mouse study, the pilicide had a strong impact on adhesion and biofilm formation and also reduced the virulence in vivo. It is also interesting to note that it reduced the biofilm formation on abiotic surface (Cegelski et al., 2010).

To develop this compound into a therapeutic, further studies are needed to assess its pharmacokinetics and pharmacodynamics and to determine the concentration at which it accumulates in the bladder or other potential sites of infection.

## Small Compounds Targeting Urease

Urease, an enzyme which catalyzes the hydrolysis of urea, is crucial in the pathogenesis of several uropathogenic bacteria such as P. mirabilis, Klebsiella sp., Pseudomonas sp. and Staphylococcus sp. (Mobley and Hausinger, 1989). This enzyme leads to the alkalinization of the urine and the production of struvite and carbonate apatite that make up the major component of urinary stones (Burne and Chen, 2000). These conditions lead to the inflammation of the urogenital epithelia thus increasing the risk of catheter-associated biofilm formation that may contribute to pyelonephritis (Musher et al., 1975; Jacobsen et al., 2008), mainly due to both bacterial and host cysteine protease (Xu et al., 2017).

The most studied inhibitors of urease are hydroxamic acids (Amtul et al., 2002; **Figure 1**). These molecules have a high inhibitory activity against urease, by bonding to the two nickel ions in the urease active site (Benini et al., 2000). Initially, these molecules were used to treat UTIs by preventing urine alkalization (Griffith et al., 1978, 1988). However, because of the growing evidence of side effects such as mutagenic power, they were progressively phased out (Munakata et al., 1980; Bailie et al., 1986; Griffith et al., 1991).

Through similarly interacting with nickel ions in the urease active site, the phenyl phosphoramidates were found to have the highest inhibitory activity (Faraci et al., 1995). Studies testing these molecules in an in vitro model (Morris and Stickler, 1998) and in a rat model (Texier-Maugein et al., 1987) found promising results. Since then, no in vivo studies or clinical trials have been developed, probably due to the poor hydrolytic stability of these molecules which leads to a very short half-life (Pope et al., 1998).

Other molecules that possess inhibitory activity against bacterial urease have been developed, but they are not fully adapted to treat UTIs. One of them is the quinones, a class of active compounds with a high oxidizing potency (Zaborska et al., 2002). These molecules had not been evaluated in vivo models due to their cytotoxic and cancerogenic properties.

## Small Compounds Targeting Bacterial Capsule

Polysaccharide capsule biogenesis plays an important role in UPEC virulence. Like other pathogens, the capsule is used as a defense against opsonophagocytosis and complement-mediated killing (Roberts, 1995, 1996). The capsule is also involved in the formation of biofilm and the formation of intracellular bacterial communities (Llobet et al., 2008; Anderson et al., 2010). Until now, the human use of small-molecule inhibitors of UPEC capsule biogenesis was not available because of their antigenicity and poor bioavailability (Varki, 2008). Nevertheless, mouse studies identified two active agents (DU003 and DU01), that caused significant bacterial death (Goller et al., 2014; **Figure 1**).

## NUTRACEUTICALS

fmicb-11-01509 July 2, 2020 Time: 16:19 # 8

Nutraceuticals are pharmaceutical alternatives, consisting of all the foods or food products which provide medical benefits and can be delivered under medical form. They provide health benefits in addition to their basic nutritional value.

## Cranberry (Vaccinium macrocarpon)

Cranberry (Vaccinium macrocarpon Ait.) is a berry that grows in North America. In recent years, the use of cranberry has increased in the prophylactic approach of recurrent UTI (Howell et al., 2005). Although, its mechanism of action is unclear, there are several possible targets of cranberry (**Figure 1**).

The main efficacy is related to the antiadherence properties of cranberry (Ahuja et al., 1998; Liu et al., 2006) due to the A-type proanthocyanidin (PAC-A) that has been shown to be an important inhibitor of Type-I fimbriae E. coli adhesion to uroepithelial cells. Some in vitro and in vivo studies demonstrated the capacity of the cranberry to reduce the adhesion of bacteria to the cells (Ermel et al., 2012; Rafsanjany et al., 2015; Liu et al., 2019). Cranberry has also shown convincing results on motility and biofilm formation. Indeed, it has a negative impact on the swarming of Pseudomonas aeruginosa and P. mirabilis (Chan et al., 2013) and on the biofilm formation of E. faecalis, P. aeruginosa and E. coli (Ulrey et al., 2014; Rodríguez-Pérez et al., 2016; Wojnicz et al., 2016).

The use of cranberry has been associated with a decrease in the incidence of UTIs, although some conflicting results have been reported in the literature. Although cranberry products have been shown to significantly reduce the incidence of UTIs at 12 months (RR 0.65, 95% CI 0.46–0.90) compared with placebo/control in women with recurrent UTIs in a Cochrane review from 2008 (Jepson and Craig, 2008), an updated review concluded that cranberry products did not show any significant reduction in the occurrence of symptomatic UTI in the same population (Jepson et al., 2012).

A placebo controlled trial, published after the last review, showed that women randomized to cranberry juice had a nonsignificant reduction in numbers of P-fimbriated E. coli in urine and in the rate of symptomatic UTIs (Stapleton et al., 2012).

Because dosage, concentration and formulation of PAC-A are not well defined, the conflicting results may be explained by the difference in PAC-A concentration between cranberry formulations (juice, beverage, tablets) in the different studies making it difficult to choose one formulation over another (Anger et al., 2019).

It must be noted that many of the products containing cranberry used in studies are only for research purposes, limiting the prophylactic application of cranberries beyond research (Stapleton et al., 2012).

Finally, a recent publication showed that the cranberry proanthocyanidins had a variable effect on a collection of E. coli strains that could explain the discordant results observed in the clinical studies (Ranfaing et al., 2018a).

In order to enhance the effectiveness of cranberry, combinations of cranberry and other natural products with antimicrobial properties could be used, such as propolis. Propolis is a resinous material collected by bees from plants then mixed with wax and bee enzymes (AFSSA, 2007). Propolis has antimicrobial, anti-inflammatory, anti-tumoral, immunemodulatory and anti-oxidant activities (Boonsai et al., 2014). It has been used for several years to treat gastrointestinal disorders (food supplement) (Freitas et al., 2006), to promote oral health (mouthwash) (Pereira et al., 2011) and in dermatological care (creams and ointments) (Olczyk et al., 2014; Pasupuleti et al., 2017; Ranfaing et al., 2018b). In vitro studies showed that propolis potentiated the effect of cranberry proanthocyanidins on adhesion, motility (swarming and swimming), biofilm formation (early formation and fully-formed biofilm), iron metabolism and stress response of UPEC (Olczyk et al., 2014; Pasupuleti et al., 2017; Ranfaing et al., 2018a,b). Moreover, this association was active in all the E. coli strains studied, ruling out the variable effect observed with the cranberry used alone (Ranfaing et al., 2018a).

A recent RCT versus placebo in 85 women with recurrent UTIs showed a slight reduction in the number of cystitis events in the first 3 months in the propolis and cranberry group after adjustment on water consumption (0.7 vs. 1.3, p = 0.02), but no difference in the mean number of infections in women with at least one infection. Of note, the mean time to onset of the first cystitis episode was significantly longer in the propolis + cranberry group (70 vs. 43 days, p = 0.03) and tolerance to the treatment was similar in both groups (Bruyère et al., 2019).

## Hyaluronic Acid

The urinary bladder epithelium is composed of urothelial cells which carry specific sensors and properties as well as forming the first barrier to pathogens (Birder and de Groat, 2007). To maintain this capacity to fight infections, these cells produce sulfated polysaccharide glycosaminoglycan (GAG) which covers the epithelium and forms a non-specific anti-adherence factor. A major proportion of the GAG layer of the bladder is composed of hyaluronic acid (HA) and chondroitin sulfate (CS). Virulence factors (secreted by E. coli for example) damage the GAG layer to prepare its adhesion (Constantinides et al., 2004; **Figure 1**). One strategy for the management of UTI is based on the re-establishment of the GAG layer of the bladder epithelium with intravesical instillations of HA alone or in combination with CS. Various randomized and non-randomized studies have been performed.

A study investigated the impact of HA and CS on recurrent UTIs on 276 women (aged 18–75 years). The intravesical administration of HA and CS was given to 181 women and the standard treatment against recurrent UTIs was given to 95

women. There was a 49% reduction in the rate of recurrence (defined as one bacteriologically confirmed UTI in the year following the treatment initiation) in patients treated with HA + CS compared with standard care [adjusted OR 0.51 (95%CI 0.27–0.96)]. However, no significant difference was found when considering the number of recurrences or the median time to first recurrence (Ciani et al., 2016).

Another retrospective study in 157 women found similar results with a significant reduction in UTI recurrence and an increased time-to-recurrence between UTIs (Cicione et al., 2014).

It has to be noted that the administration protocol (different between participating centers) of weekly instillation for one month followed by monthly instillations may be a limiting factor for patients.

A synergy might exist between HA + CS and estrogen. This association has been explored in 145 postmenopausal women with mild-to-moderate urogenital atrophy and a history of recurrent UTI. Participants were divided into three groups: vaginal estrogen, oral HA, and oral HA + CS and vaginal estrogen. Oral treatments were effective in preventing recurrent UTI (number of patients with fewer than two infective episodes in the 6-month follow-up and fewer than three episodes in the 12-month follow-up), especially if administered with vaginal estrogen therapy. A slight effect on the HA alone but also a significant impact of the estrogen and the HA + CS on the recurrence of UTIs in postmenopausal women was observed (Torella et al., 2016).

Recently a meta-analysis suggested that HA ± CS decreased the rate of UTI recurrence and increased the time to recurrence (Goddard and Janssen, 2018). Moreover the authors noted the safety of HA therapy even if the intravesical instillation is more invasive than other administrations (usually per os). The combination therapy was more performant than the use of HA alone. It seems essential to perform a well-designed, randomized, controlled clinical trials with larger population.

## D-mannose

As seen above, CUP pili are important virulence factors and represent optimal targets for antivirulence compound development. E. coli binds to mannosylated host cells via their mannose-binding lectin domains of FimH.

Two developments have been proposed to prevent this interaction:


demonstrated a decrease in bladder colonization after an oral administration of mannosides and a prevention of acute and chronic UTI (Klein et al., 2010; Cusumano et al., 2011; **Figure 1**). In this way, an exogenous intake of D-mannose competitively blocks the interaction between the bacterial fimbriae and host cells (Cusumano et al., 2011; **Figure 1**). One RCT compared the recurrence of UTIs in a group of patients taking daily nitrofurantoin to a group treated with daily D-mannose powder. The risk of recurrence was similar between groups, whilst there was a reduction of the side effects in the D-mannose group (Kranjcec et al., 2014 ˇ ).

## Galabiose

Type P fimbriae adhere to the galabiose-like receptor via PapG (Larsson et al., 2003). As previously noted, this kind of fimbriae is essential for the pathogenesis of UTIs in helping the bacteria to reach the kidneys (Strömberg et al., 1990). Better understanding of the structure of the different variants of PapG (mostly PapG II and PapG III) might lead to drugs designed to target this adhesin (Sung et al., 2001; **Figure 1**). No in vivo studies on the impact of this sugar on UTIs have yet been performed. Evaluation on the most prevalent uropathogenic bacteria is essential to clearly evaluate the potential of galabiose to reduce UTI.

## Vitamin C

Vitamin C (ascorbic acid) is known to possess antioxidant and antimicrobial activities. As microbial infections cause reactive oxygen species (ROS) release by phagocytes, it is helpful in the limitation of infection through deactivation of microorganism killing. ROS may also cause damage to the host cells, therefore the level of ROS released by phagocytes should be reduced directly after infection (Liu et al., 2018). Vitamin C is an essential co-enzyme in the oxidative stress pathways, capable of ROS removal. Habash et al. (1999) suggested that vitamin C decreased the adhesion and microorganisms colonization of the biomaterials used in diagnostic/treatment procedures involving the urinary tract.

Moreover, two trials have studied the use of vitamin C to prevent UTIs. In a single-blind randomized study in thirteen spinal cord injury patients randomized to placebo or 500 mg ascorbic acid four times daily, there was no clinical benefit on urinary infection from the use of ascorbic acid (Castelló et al., 1996). In a second single-blind randomized trial in 110 pregnant women, participants taking a vitamin regimen with 100 mg ascorbic acid per day for 3 months showed a reduction in symptomatic UTIs incidence from 29.1 to 12.7% compared to participants following a vitamin regimen without ascorbic acid (Ochoa-Brust et al., 2007).

To date, there is no evidence of vitamin C action in the prevention of UTI.

## Chinese Herbal Medicine (CHM)

Chinese Herbal Medicine (CHM) is the ancient art of compiling complex herbal formulae usually comprising up to 15 herbs. CHM has been historically used to treat UTI. Some frequently used Chinese herbs are known to have significant diuretic,

antibiotic, immune enhancing, antipyretic, anti-inflammatory and pain relieving activities. Some of these herbs have shown an in vitro inhibitory activity against several uropathogens, especially against E. coli in decreasing its adherence to bladder epithelial cells (Tong et al., 2011). In an antibacterial test against mice, it was found that Sanjin tablets (composed by five kinds of CHM) have strong bacteriostasis activity (Hou and Wang, 2016). This product is used to treat acute uncomplicated lower UTI (Meng et al., 2015) and to reduce the symptoms of chronic UTI by reducing the secretory level of some urinary cytokines (Hou and Wang, 2016). A first meta-analysis of three RCTs including 282 women suggested that CHM significantly reduced recurrent UTI rates compared to antibiotics (RR 0.28, 95%CI [0.09 to 0.82]) (Flower et al., 2015). The second one has concluded that the current evidence is insufficient to support the efficacity and safety of Sanjin tablets for acute uncomplicated lower UTI (Pu et al., 2016). Only two of these RCTs reported adverse events (Gu et al., 2011; Zhao et al., 2011). Neither found any liver or renal impairment. Further studies are needed to definitively evaluate the potential of CHM on prevention/treatment of UTI.

## Other Phytochemicals

For centuries, plants have been used as alternative and traditional medicine around the world notably as therapies for infectious diseases (Banu and Kumar, 2009; Sharifi-Rad et al., 2016). Plants and their secondary metabolic derivatives are a major resource of antioxidants due to the presence of phenolic compounds including flavonoids, phenolic acids, or tannins. Different extracts from plants and spices have demonstrated antiinflammatory, antimicrobial and diuretic activities (Mickymaray and Al Aboody, 2019). They also exhibited anti-quorum sensing and anti-biofilm potentials (Issac Abraham et al., 2011; Mazarei et al., 2017). One of the main mechanism of action is the antiadhesive action due to the formation of H-bonds between the FimH protein ligand and the plant compounds (Jaiswal et al., 2018). Future clinical trials must be done after complete pharmacokinetic/pharmacodynamic analyses.

## IMMUNOMODULANT AGENTS

The innate immune system activation through the secretion of cytokines and the recruitment of macrophages and neutrophils rapidly occurs after the onset of UTI (Duell et al., 2012). Even in the absence of an effective antibiotic treatment, this immune reaction might be enough to counter the infection. In the case of the persistence of several bacterial strains in the bladder, persistent acute immune response and tissue inflammation can be observed. Moreover, multiple infections can lead to chronic inflammation that increases the risk of developing recurrent UTIs (Ferry et al., 2004; Hannan et al., 2010). There is an increase in the expression of cyclooxygenase (COX)-2 after an infection of the bladder epithelial cells by UPEC. Moreover, there is a correlation between the severity of the inflammation and the increase of COX-2 expression (Chen et al., 2011; **Figure 1**). COX-2 inhibition prevents urothelial transmigration by neutrophils and damage to the urothelial barrier and facilitates the innate responses (Hannan et al., 2014). A double-blind RCT on 79 women with uncomplicated UTI showed equivalent symptoms at Day 4, when taking ibuprofen (200 mg t.i.d) compared to ciprofloxacin (Bleidorn et al., 2010).

A recent 2 × 2 factorial placebo RCT evaluating ibuprofen in 382 women demonstrated a substantial reduction in antibiotic use in patients taking ibuprofen without differences in terms of symptom relief or speed of recovery (Moore et al., 2019).

Plant-based immunomodulants such as Green Tea Extract (GTE) have also shown promise (Bae et al., 2015). GTE contains an array of polyphenolic compounds, especially catechins. The biological properties of catechins are antioxidant, antiangiogenesis, antiproliferative activity, and antineoplastic (Cooper et al., 2005a,b). Some in vitro studies showed an antibacterial impact of GTE against UPEC (Hoshino et al., 1999; Arakawa et al., 2004; Reygaert and Jusufi, 2013) and in a rat model of cystitis, catechins significantly decreased inflammation and uroepithelium edema (Noormandi and Dabaghzadeh, 2015). These results are promising even though the mechanisms of action are still unclear.

## PROBIOTICS

A probiotic is a live microorganism that confers a health benefit.

## Vaginal Lactobacilli

Lactobacilli are frequently dominant microorganisms in vaginal flora (Mendling, 2016). Different observations can be done: (i) they have the ability to interfere with the adherence, growth and colonization of UPEC (Falagas et al., 2006); (ii) a change in the normal vaginal flora has been shown to facilitate the recurrence of UTIs (Cribby et al., 2008); and (iii) the use of commensal bacteria (such as Lactobacilli) reduces the proportion of uropathogens and thus restores bacterial homeostasis (Hardy et al., 2013). Although their exact mechanisms of action are still unknown, lactobacilli strains seem to have at least three different modes of action (Ng et al., 2018; **Figure 1**):


Probiotics can also modulate the immune system. In addition, bacterial strains secreting "immunomodulins" and cytokines are able to reduce infection by pathogenic bacteria (Galdeano and Perdigón, 2006; Kemgang et al., 2014). Lactobacillus species have these anti-inflammatory and immune-regulatory actions (Isolauri et al., 2001).

A non-inferiority randomized trial compared antibiotic prophylaxis and Lactobacillus prophylaxis in 252 postmenopausal women with recurrent UTIs who received 12 months of prophylaxis with trimethoprim-sulfamethoxazole (TMP-SMX), 480 mg once daily or oral capsules containing 10<sup>9</sup> CFU of Lactobacillus rhamnosus GR-1 and Lactobacillus reuteri RC-14 twice daily (Beerepoot et al., 2012b). The Lactobacillus treatment did not demonstrate non- inferiority, with an average number of symptomatic UTIs during the year of follow-up of 2.9 in the antibiotic group versus 3.3 in the lactobacilli group. The benefit of Lactobacilli was that it had no impact on antibiotic resistance compared to trimethoprimsulfamethoxazole.

A reduction in the recurrence rate of UTI with the use of lactobacilli products (pooled RR = 0.68, 95%CI 0.44 to 0.93, p < 0.001) (Ng et al., 2018) was underlined in a meta-analysis including 620 patients. In this study, two intravaginal suppositories (containing Lactobacillus crispatus CTV05, Lactobacillus rhamnosus GR1 and Lactobacillus reuteri RC14) had the highest efficacy (Ng et al., 2018).

A recent randomized, double-blind, placebo-controlled pilot study in 81 premenopausal women showed that the administration of Bio-Kult Pro-Cyan (a commercially available product containing probiotic strains (Lactobacillus acidophilus PXN 35, Lactobacillus plantarum PXN 47) and cranberry extract (36 mg/d PACs) twice-daily for 26 weeks, led to significantly lower number of recurrent UTIs compared to placebo (9.1 vs. 33.3%; P = 0.0053) (Koradia et al., 2019).

## Bacterial Interference: Escherichia coli Strain 83972

The intentional colonization of the bladder with a non-virulent strain, also called bacterial interference, has been studied among patients with neurogenic bladder. E. coli 83972 is a clinical strain, isolated from a woman with chronic urinary colonization and which has naturally lost its capacity to develop Type 1 and Type P fimbriae. This strain has been used for prophylactic purposes to deliberately colonized the bladders with this bacterium to prevent colonization/infection by pathogenic species.

In a mouse model of UTI, E. coli 83972 demonstrated a better fitness than a virulent strain of UPEC. In a poor environment, like the bladder, this difference in fitness is a crucial advantage for the competition between bacteria. The 83972 strain could reduce the impact of UTIs by a monopolization of resources and space (Roos et al., 2006).

Seven clinical studies are available: three are RCT, one of which is a crossover designed study; and four are prospective cohorts (Hull et al., 2000; Darouiche et al., 2001, 2005; Dashiff et al., 2011; Trautner et al., 2007; Prasad et al., 2009; Sundén et al., 2010). Sample sizes were small and varied from 12 to 44 patients. Clinical endpoints were the interval before first recurrence or the incidence of UTI during follow up.

Despite this heterogeneity, all studies demonstrated the ability of non-virulent strain to protect patients from UTI. One limit is the difficulty to achieve bladder colonization with the nonvirulent strain (only 38% of patients in one of the RCT) (Darouiche et al., 2011).

## Predatory Bacteria

Predatory bacteria are small, motile, deltaproteobacteria that are a predatory invader of other Gram-negative bacteria (**Figure 1**). They occupy an intraperiplasmic niche and kill, digest and lyse their host, the prey cell. Bdellovibrio and Micavibrio are the most studied predatory bacteria (Stolp and Starr, 1963). Bdellovibrio bacteriovorus uses its type IV pilus to adhere and penetrate the outer membrane (bdelopast) of its prey. Inside the bacteria, B. bacteriovorus modifies the membrane of its prey to allow its growth until it uses all the nutrients. This entire process takes only 2–3 h (Sockett, 2009). Several Gram-negative human pathogenic bacteria such as E. coli, Klebsiella spp. and Pseudomonas spp. can be targeted by these predatory bacteria (Dashiff et al., 2011). Furthermore, it has been showed in vitro that B. bacteriovorus significantly reduced the quantity of biofilm (Kadouri and O'Toole, 2005).

In vitro and mouse model studies have underlined the fact that the predatory bacteria have no negative impact on human cell lines (Gupta et al., 2016) nor on animals (Shatzkes et al., 2015). B. dellovibrioa is thus a potential therapeutic agent, but has not been yet applied in this way. No studies have investigated B. bacteriovorus to treat UTIs to date. However, these bacteria offer an exciting path for further research where in vivo studies should be the focus.

## BACTERIOPHAGES

Bacteriophages are viruses that parasitize a bacterium by infecting it and reproducing inside it and can act as bactericidal agents. Bacteriophage therapy is currently applied in different areas of medical research. Indeed, it has been recognized as an alternative treatment in localized infections such as otitis, infected burns and osteoarticular infections (Ferry et al., 2018; Jault et al., 2019). However, its use in UTI is scarce.

One research team has studied the combination of transurethral resection of prostate with bacteriophage therapy used instead of per operative antibiotics. They first demonstrated the in vitro lytic activity of commercial bacteriophage cocktails on 41 E. coli and 9 K. pneumoniae strains. The lytic activity of the bacteriophage cocktails varied between 66% and 93%. They also showed the potential of bacteriophage adaptation experiments to increase the lytic activity, leading to an increase from 66 to 93% for one of the cocktails (Sybesma et al., 2016).

In an animal model, Dufour et al. (2016) showed the ability of phage to treat an E. coli UTI. The administration of the same bacteriophage cocktail showed a significant reduction of the bacterial load in the E. coli kidney infection model as well as in the E. coli pneumonia model, but not in an E. coli sepsis model.

In vivo studies were performed with a commercial preparation called Pyo bacteriophage, composed of bacteriophage lines active against a broad spectrum of uropathogenic bacteria: S. aureus, E. coli, Streptococcus spp. (including Enterococcus spp.), P. aeruginosa, and Proteus spp. in nine patients planned for transurethral resection. Bacteria load decreased in two-thirds of the patients (6/9), without any associated adverse events (Ujmajuridze et al., 2018).

One case report described the treatment of a recurrent P. aeruginosa UTI associated with a bilateral ureteral stent. A phage cocktail containing phages with activity against Streptococcus pyogenes, S aureus, E. coli, P. aeruginosa, P. vulgaris, and P. mirabilis was administered. After 6 days of treatment, meropenem and colistin were started for 30 days. The results showed a 10-fold reduction of bacteria load in the urine after the first 5 days of phage treatment. The bacterial load was undetectable after 2 days of subsequent antibiotic treatment. Urine samples remained sterile until 1 year after the end of the antibiotic treatment.

Another case report showed the treatment of a recurrent UTI with an ESBL-producing K. pneumoniae in a renal transplant patient whose infection evolved into an epididymitis. The patient was definitively cured after an oral and intravesical bacteriophage treatment of 10 days following 6-week meropenem treatment (Kuipers et al., 2019).

One randomized, double-blind trial versus placebo in patients planned for transurethral resection of the prostate with UTI is ongoing. Patients will be randomized in a 1:1:1 ratio to receive 7 days of either: (i) bacteriophage (Pyo bacteriophage) solution, (ii) placebo solution, or (iii) antibiotic treatment (Leitner et al., 2017).

More human studies are warranted to further define the role of this treatment option in UTIs.

## CONCLUSION

Although prophylactic antibiotics remain the preferred preventive treatment in recurrent UTIs, the emergence of antimicrobial resistance worldwide has made the development of non-antibiotics strategies a priority. The better understanding of UTI mechanisms will help direct future research on the topic. Indeed, recent studies have revealed that infection with UPEC and a number of other Gram-negative uropathogens proceeds through dynamic intracellular and extracellular host niches during the course of acute and chronic infection.

Several targets such as uropathogenic adhesins, toxins, urease, iron metabolism and motility have been explored. Although

## REFERENCES


non-antibiotic prophylactic agents appear to be well tolerated and do not seem to increase the antimicrobial resistance of the commensal flora, most of therapeutic options displayed in this review are still preliminary.

Other approaches have also been evaluated in prevention of CAUTI. In case of prolonged utilization of catheter, some techniques have been developed to prevent bacterial growth and biofilm formation. These techniques, including the devices with antimicrobial coatings such as silver, peptides, enzymes or bacteriophages, provide minimal reduction in infection incidence and will need further assessment (Percival et al., 2015; Al-Qahtani et al., 2019).

While Lactobacillus-containing products appear to be the most promising new alternative to currently used antibiotics, cranberry products combined with propolis would need further investigations. Studies should now be designed to investigate the interaction between non-antibiotic therapies, uropathogens and the host immune system. Importantly, clinical trials must use standardized definitions of UTI (infection versus colonization based on urinary symptoms), treatment regimens and control groups as well as an assessment of the risk/benefit ratio especially on tolerance and antibiotic resistance and an economic evaluation of the reviewed therapeutics versus prolonged antibiotic treatments. Likewise, further research is needed for vaccines, which have shown potential in initial trials.

## AUTHOR CONTRIBUTIONS

PL, JR, J-PL, and AS wrote the manuscript. AD, CD-R, LB, and FB critically reviewed the manuscript. All authors contributed to the article and approved the submitted version.

## FUNDING

This work was supported by CHU Nîmes (Grant Thématiques phares).

## ACKNOWLEDGMENTS

PL, CD-R, AS and J-PL belong to the FHU INCh (Federation Hospitalo Universitaire Infections Chroniques, Aviesan). We thank the Nîmes University hospital for its structural, human and financial support through the award obtained by our team during the internal call for tenders Thématiques phares. We also thank Sarah Kabani for her editing assistance.

Ahuja, S., Kaack, B., and Roberts, J. (1998). Loss of fimbrial adhesion with the addition of Vaccinum macrocarpon to the growth medium of P-fimbriated Escherichia coli. J. Urol. 159, 559–562. doi: 10.1016/s0022-5347(01)63983-1


in hospitalized patients: a review on recent updates. J. Infect. Public Health 12, 760–766. doi: 10.1016/j.jiph.2019.09.009


model of urinary tract infection. Infect. Immun. 81, 3309–3316. doi: 10.1128/ iai.00470-13


Bdellovibrio bacteriovorus. J. Appl. Microbiol. 110, 431–444. doi: 10.1111/j. 1365-2672.2010.04900.x


immunity. Clin. Vaccine Immunol. 13, 219–226. doi: 10.1128/cvi.13.2.219-226. 2006




compounds from cranberry (Vaccinium macrocarpon) against Escherichia coli. Food Funct. 7, 1564–1573. doi: 10.1039/c5fo01441g


patients with incomplete bladder emptying. J. Urol. 184, 179–185. doi: 10.1016/ j.juro.2010.03.024



**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 Loubet, Ranfaing, Dinh, Dunyach-Remy, Bernard, Bruyère, Lavigne and Sotto. 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.

# Coexistence of the Oxazolidinone Resistance–Associated Genes *cfr* and *optrA* in *Enterococcus faecalis* From a Healthy Piglet in Brazil

Lara M. Almeida1,2 \*, Anthony Gaca<sup>3</sup> , Paulo M. Bispo<sup>3</sup> , François Lebreton<sup>3</sup> , Jose T. Saavedra<sup>3</sup> , Rafael A. Silva<sup>1</sup> , Irinaldo D. Basílio-Júnior <sup>1</sup> , Felipe M. Zorzi <sup>4</sup> , Pedro H. Filsner <sup>4</sup> , Andrea M. Moreno<sup>4</sup> and Michael S. Gilmore<sup>3</sup>

1 Institute of Pharmaceutical Sciences, Federal University of Alagoas, Maceió, Brazil, <sup>2</sup> Department of Clinical and Toxicological Analyses, Faculty of Pharmaceutical Sciences, University of São Paulo, São Paulo, Brazil, <sup>3</sup> Department of Ophthalmology and Department of Microbiology, Harvard Medical School, Boston, MA, United States, <sup>4</sup> School of Veterinary Medicine and Animal Science, University of São Paulo, São Paulo, Brazil

#### *Edited by:*

Filipa Grosso, University of Porto, Portugal

### *Reviewed by:*

Yeshi Yin, Hunan University of Science and Engineering, China Amira Awad Moawad, Friedrich Loeffler Institute, Germany

> *\*Correspondence:* Lara M. Almeida larameal@gmail.com

#### *Specialty section:*

This article was submitted to Infectious Diseases – Surveillance, Prevention and Treatment, a section of the journal Frontiers in Public Health

*Received:* 04 April 2020 *Accepted:* 11 August 2020 *Published:* 24 September 2020

#### *Citation:*

Almeida LM, Gaca A, Bispo PM, Lebreton F, Saavedra JT, Silva RA, Basílio-Júnior ID, Zorzi FM, Filsner PH, Moreno AM and Gilmore MS (2020) Coexistence of the Oxazolidinone Resistance–Associated Genes cfr and optrA in Enterococcus faecalis From a Healthy Piglet in Brazil. Front. Public Health 8:518. doi: 10.3389/fpubh.2020.00518 Oxazolidinones are one of the most important antimicrobials potentially active against glycopeptide- and β-lactam-resistant Gram-positive pathogens. Linezolid—the first oxazolidinone to be approved for clinical use in 2000 by the US Food and Drug Administration—and the newer molecule in the class, tedizolid, inhibit protein synthesis by suppressing the formation of the 70S ribosomal complex in bacteria. Over the past two decades, transferable oxazolidinone resistance genes, in particular cfr and optrA, have been identified in Firmicutes isolated from healthcare-related infections, livestock, and the environment. Our goals in this study were to investigate the genetic contexts and the transferability of the cfr and optrA genes and examine genomic features, such as antimicrobial resistance genes, plasmid incompatibility types, and CRISPR-Cas defenses of a linezolid-resistant Enterococcus faecalis isolated in feces from a healthy pig during an antimicrobial surveillance program for animal production in Brazil. The cfr gene was found to be integrated into a transposon-like structure of 7,759 nt flanked by IS1216E and capable of excising and circularizing, distinguishing it from known genetic contexts for cfr in Enterococcus spp., while optrA was inserted into an Inc18 broad host-range plasmid of >58 kb. Conjugal transfer of cfr and optrA was shown by filter mating. The coexistence of cfr and optrA in an E. faecalis isolated from a healthy nursery pig highlights the need for monitoring the use of antibiotics in the Brazilian swine production system for controlling spread and proliferation of antibiotic resistance.

Keywords: oxazolidinones, resistance, *Enterococcus faecalis*, *cfr* gene, *optrA* gene, livestock

## INTRODUCTION

Few drugs remain available for treating infections caused by antibiotic-resistant bacteria. Oxazolidinone antimicrobials, including linezolid and tedizolid, are among the few last-line therapies effective for multidrug-resistant (MDR) Gram-positive pathogens. Linezolid inhibits protein synthesis by targeting the peptidyl transferase center of the 50S subunit of bacterial ribosomes, blocking the binding of aminoacyl-tRNA to the A-site of the peptidyl transferase center (PTC), and also affecting the positioning of fMet-tRNA at the P-site, which prevents formation of the initiation complex (1– 4). Over the past two decades, however, mutations in domain V of the 50S ribosomal subunit of 23S rRNA or in the ribosomal proteins L3 and L4 (5, 6) and the transferable resistance genes cfr, optrA, and poxtA (7–9) have driven the spread of oxazolidinone resistance in Gram-positive bacteria in healthcare and animal agriculture settings.

The spread of the multiresistance gene cfr has raised concern since its first report in a bovine Staphylococcus sciuri isolate (10). The cfr gene initially reported occurred in Enterococcus spp. from healthcare-related infections in Thailand (11) and from livestock in China (12). So far, a BLASTn search of the GenBank database identifies 3 cfr homologs in enterococci. While cfr has been found in both human and animal isolates of Enterococcus faecalis (Thailand, China) (11–13), Enterococcus faecium (Italy, Ireland, U.S.) (14–16), Enterococcus casseliflavus, and Enterococcus thailandicus (China) (17), the cfr(B) variant has been detected only in clinical isolates of E. faecalis (Japan) (18) and E. faecium (U.S., Germany, Netherlands) (19, 20). The most recently described cfr(D) variant has only four entries so far recorded in NCBI's databases, all E. faecium (France, Ireland, Netherlands) (21).

A Cfr-mediated adenosine modification A2503 in the PTC of 23S rRNA, which confers resistance to the oxazolidinone, phenicol, lincosamide, pleuromutilin, streptogramin A, and 16 member-ring macrolide antimicrobials (22), was until 2015 the only known transferable oxazolidinone resistance mechanism. Since then, the ATP-binding cassette (ABC)-F protein OptrA (23) has also been identified as conferring resistance to oxazolidinones, including the newer molecule in the class, tedizolid (24). optrA was identified in both E. faecalis and E. faecium of human and animal origins (8, 25, 26), as well as in E. thailandicus and Enterococcus gallinarum isolated from hospitals in China (27). Elsewhere in Asia (18, 28), Europe (29, 30), and America (16, 31), optrA has been found in E. faecalis and E. faecium of both human and animal origins. In Africa, optrA-positive E. faecalis isolated from humans (32), urban wastewater (33), and food-producing animals (34) were also reported. The cfr and optrA genes can be either plasmid or chromosomally encoded, and the co-location of both in the same plasmid has already been described in a porcine Staphylococcus sciuri isolate in China (35) and in E. faecium and E. faecalis recovered from hospitalized patients and livestock from Europe and the US (14–16).

In this study, we investigated the genetic contexts and the transferability of the cfr and optrA genes from the linezolidresistant (LR) E. faecalis strain L9 (CP018004.1), which was isolated from a rectal swab collected from a healthy piglet in a surveillance study of antimicrobial susceptibility in Brazil's swine production system (36). Antimicrobial resistance genes, plasmid incompatibility types, epidemiology, and CRISPR-Cas (Clustered Regularly Interspaced Short Palindromic Repeat) defenses of LR E. faecalis L9 were also examined. Wholegenome sequencing (WGS) analysis revealed the presence of cfr associated with a transposable element capable of excision and formation of an intracellular circular intermediate flanked by IS1216E (CP041775.1), which is different from all previously known genetic contexts in Enterococcus spp. from human and animal sources. Further, the core araC-hp-optrA was found to be inserted into a conjugative Inc18 broad host-range plasmid of >58 kb (CP041776.1) in LR E. faecalis L9.

## MATERIALS AND METHODS

## Bacterial Isolation

LR E. faecalis L9 comes from a collection of 13 LR E. faecalis (linezolid MIC of 8 mg/L) that were screened from 245 MDR E. faecalis isolated from rectal swabs from healthy piglets (45 days old) in different states of Brazil (36). These 13 optrA-positive E. faecalis, epidemiologically unrelated (ST29, ST330, ST591, ST710, ST711), were recovered from different pigs found to be distributed in 6 out of the 7 states chosen for sample collection. Three LR E. faecalis isolated in the same state (DF) harbored both optrA and cfr (ST591 and ST29), but conjugal transfer of these resistance genes to an enterococcal recipient was achieved only using the ST29 E. faecalis strain L9 as donor in our previous filter mating assays. Therefore, here we investigated the mobile element types that enabled horizontal transfer of cfr and optrA.

## Whole-Genome Sequencing and Data Analysis

LR E. faecalis L9 was grown in brain heart infusion (BHI) broth at 37◦C (24 h). Genomic DNA was isolated using the QIAGEN DNeasy Blood & Tissue Kit, and quantified using Qubit dsDNA HS. Sequencing libraries were prepared with the Illumina Nextera XT DNA kit and sequenced on a MiSeq instrument (Illumina Inc., USA) at the Massachusetts Eye and Ear Infirmary (MEEI) Ocular Genomics Institute, as 250 nt paired-end reads. De novo assembling was performed using CLC Genomics Workbench 8.0.3. For genome annotations, both the RAST server (Rapid Annotation using Subsystem Technology) and the Prokaryotic Genome Annotation Pipeline (NCBI PGAP) were used. Genome data analysis was performed using BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi) and Center for Genomic Epidemiology (http://www.genomicepidemiology.org) online tools. ResFinder (https://cge.cbs.dtu.dk/services/ResFinder/) was used to identify acquired antimicrobial resistance genes, and PlasmidFinder (https://cge.cbs.dtu.dk/services/PlasmidFinder/) was used to determine plasmid incompatibility types. For detection of the oxazolidinone resistance determinants, LRE-Finder (https://cge.cbs.dtu.dk/services/LRE-finder/) was used as well. Multilocus sequence typing (MLST) loci were assigned by the MLST database (https://pubmlst.org/efaecalis/), and the presence of CRISPR-cas defenses was identified by CRISPRfinder (https://crisprcas.i2bc.paris-saclay.fr).

## Filter Mating Assay

Conjugation by filter mating as described previously by Jaworski and Clewell (37) was performed using LR E. faecalis L9 as donor, and the E. faecalisstrain OG1RF as recipient. Donor and recipient were grown overnight in BHI broth at 37◦C. One milliliter from donor culture plus 1 ml from recipient culture were inoculated in 3 ml of phosphate buffered saline (PBS) solution, filtered through a sterile 25-mm-diameter, 0.22-µm-pore-size membrane filter, and subsequently incubated on BHI agar at 37◦C for 24–48 h. PBS (5 ml) was used to wash the filters, and 500 µl of this solution was spread on BHI agar plates (100 × 15 mm Petri plates) containing 25µg/ml of fusidic acid, 25µg/ml of rifampicin to select for the OG1RF chromosomal markers, and 25 or 10µg/ml of chloramphenicol (CHL) to select for oxazolidinone and phenicol resistance genes; linezolid (LZD) (4µg/ml) instead of chloramphenicol was also tested to select for cfr and optrA. Conjugation efficiency (CFU/ml of transconjugants per CFU/ml of donors) was calculated as previously described (38). PCR using primer sets specific for optrA, cfr, poxtA, fexA, and cat genes (36) and Sanger sequencing were carried out to detect these resistances in OG1RF transconjugants. Minimum inhibitory concentrations (MIC's) of chloramphenicol, florfenicol, linezolid, and tedizolid were determined by broth microdilution testing according to the guidelines of the Clinical Laboratory Standards Institute (CLSI). E. faecalis ATCC 29212 was used as a control for antimicrobial susceptibility testing.

## RESULTS AND DISCUSSION

Enterococcus faecalis is a commensal bacterium of the gut microbiota of humans and various animal species and also a cause of infections in critically ill patients (39, 40). Besides being an important hospital pathogen, E. faecalis has emerged as a potential reservoir of oxazolidinone resistance genes in animal agriculture settings worldwide (8, 31, 34, 36, 41). It is of substantial concern that antibiotics used in foodproducing animals may be selecting for the proliferation of MDR E. faecalis lineages in which cfr and optrA coexist. Cfr rRNA methyltransferase confers resistance to six important antimicrobial classes that target the 50S ribosomal subunit (22), while the ATP-binding cassette (ABC)-F protein OptrA confers resistance to phenicol and oxazolidinone, including resistance to the new oxazolidinone tedizolid (23). The spread of cfr and optrA inter-species/genera has been driven by plasmids containing other important resistance determinants (14–16, 35). Therefore, oxazolidinone resistance can be co-selected by antimicrobials that have been largely used in swine production, such as phenicol, macrolide, lincosamide, and pleuromutilin.

## Antimicrobial Resistance Determinants of LR *E. faecalis* L9

ResFinder identified that LR E. faecalis L9 carries the lsa(A) gene, which is responsible for intrinsic LSAP resistance in E. faecalis, and acquired resistance genes for aminoglycoside (str), phenicol fex(A), phenicol and oxazolidinone (optrA), and tetracycline [tet(L), tet(M), tet(S)], in addition to the multiresistance gene cfr. LRE-Finder confirmed the presence ofcfr and optrA (CP041775.1 and CP041776.1, respectively), but the phenicol–oxazolidinone– tetracycline resistance gene poxtA was not found. 23S rRNA mutations were not detected in LR E. faecalis L9, nor were they identified in ribosomal protein genes rplC, rplD, and rplV (L3, L4, and L22, respectively).

## Genetic Context of *cfr* in the Porcine LR *E. faecalis* Isolate L9

A cfr-carrying DNA segment of 7,759 nt, pL9-A (CP041775.1), was found to be inserted into LR E. faecalis L9 (**Figure 1**). The cfr gene was flanked upstream by the Tn554-related 1tnpB gene. Further upstream of 1tnpB, a gene coding for RepUS18 was detected that was disrupted by the integration of an IS1216E. The repUS18 gene is often found in Inc18 broad host-range plasmids, which have been related to antimicrobial resistance gene transfer in enterococci. Downstream, the cfr gene was flanked by a recombinase rec gene, a gene coding for a hypothetical protein, and a plasmid recombination/mobilization pre/mob gene. In silico predictions indicated that the IS1216Eflanked segment pL9-A could excise and exist within the cell as a non-replicating circular intermediate in LR E. faecalis L9, which was confirmed by PCR and Sanger sequencing using the primers 5′AGGTTTAGAATAATCTCCCGA3′ and 5′GCTGACAACATATCTAATATCTCAA3′ .

pL9-A possesses 100% DNA identity over 93% of its length to a chromosomal DNA sequence from Proteus vulgaris PV-01 (JF969273) isolated from a pig nasal swab in China (42), and 99.98% and 99.83% DNA identity to the pSA8589 and p1128105 from Staphylococcus aureus 1900 (43) and S. aureus 1128105 (44) of human origin in the US (KC561137 and KJ866414, respectively), highlighting its very broad-range horizontal transfer capabilities. IS6 insertion sequence family elements, which have been commonly associated with antibiotic resistance genes, appear to be also involved in transposition events of the core cfr-rec-pre/mob in Gram-positive and Gramnegative bacteria.

The cfr-carrying mobile element pL9-A was distinctly different from all known genetic contexts of cfr in Enterococcus spp. from human and animal sources. The similarity of pL9- A to the cfr-carrying segments previously identified in bacteria of other genera indicates that it has most likely been acquired horizontally from other bacteria or, alternatively, could be intrinsic in some lineages of E. faecalis and then transferred to other bacteria. The IS1216E element appears to be involved in the acquisition and dispersal of pL9-A in LR E. faecalis L9. In Brazil, the cfr gene has been reported to date only in an ST398 MSSA strain of human origin (45) in a genetic context other than that observed in the porcine LR E. faecalis isolate L9.

## Plasmid-Borne *optrA*-Carrying Partial Sequence (pL9) in the Porcine LR *E. faecalis* Isolate L9

We recently reported that the core araC-hp-optrA of 3,453 nt in length, which was composed of genes coding for a hypothetical protein and an AraC family transcriptional regulator at the 5′ of optrA, was inserted upstream of an IS1216E element into a plasmid of >58 kb, which was not closed during de novo assembly of the high quality draft sequence (CP041776.1) (36). On the flank 5′ of the core araChp-optrA, LR E. faecalis L9 showed in silico a duplication of optrA, which was confirmed by PCR and Sanger sequencing using the primers 5′TTGAGTGAAATACCTGTGCG3′ and

repeats (IR) at the ends of an IS1216 that was inserted into repUS18. The 6,956-bp segment between the IRs shows high DNA identity to the corresponding stretches in pSA8589 (KC651137), p1128105 (KJ866414), and P. vulgaris PV-01 chromosomal sequence (JF969273). Alignment of these sequences revealed only a deletion of 7 bp in repUS18 from pL9-A and an insertion of 10 bp in the hypothetical protein on the flank 5′ of cfr in p1128105, which is represented by the slightly lighter shade of identity over this region.

5 ′TGATGGTAATATGGTGTTGGAA3′ . Further analysis of pL9 showed the presence of genes coding for the zeta–epsilon– delta (ω-ε-ζ) toxin–antitoxin (TA) system upstream of the duplication of optrA (**Figure 2**). The ω-ε-ζ TA module, a post-segregational killing system which acts at cell division eliminating progeny that fails to inherit plasmid copy, has

been found in various MDR Gram-positive bacteria, including the cfr-carrying conjugative plasmids pW9-2 from E. faecalis, and pW3 and p3-38 from E. thailandicus isolated from sewage in swine farm contexts in China (12) and pEF12- 0805 from E. faecium isolated from human blood in Italy (14). pL9 harbored a gene coding for the plasmid replication protein repUS1, which is found in Inc18 broad host-range plasmids. On the 5′ flank of repUS1, a 912-nt open reading frame (ORF) for the partitioning protein ParA which mediates plasmid segregation was found. These mechanisms ensure the maintenance of plasmids that exist in low-copy numbers in a bacterial population, such as Inc18 family plasmids. At the 3 ′ flank of repUS1, a 288-nt ORF for the replication control protein PrgN is present. Upstream to this region, an 1,494-nt ORF for ATP-binding cassette domain-containing protein came to our attention due to the very few entries so far recorded in NCBI's databases, as it only matches nucleotide sequences from 6 Lactococcus garvieae, which causes fatal hemorrhagic septicaemia in fish (South Korea and Japan), and 1 Lactococcus petauri isolated from human feces in China. This ORF codes for the ATP-binding cassette domain-containing protein (E. faecalis WP\_155282194.1), which has 81.74% DNA identity over 99% of its length to the Lactococcus ABC-F-type ribosomal protection protein (WP\_019291880.1).

pL9 was found to be inserted into a conjugative Inc18 plasmid of >58 kb. Inc18 broad host-range plasmids have been associated with a variety of antibiotic resistances in enterococci, including the high-level vanA glycopeptide resistance carried by Tn1546, which can be transferred to MRSA lineages (46). Inc18 plasmids can play a crucial role in the oxazolidinone resistance emergence, as they are widespread in enterococci, streptococci, and staphylococci in both clinical and environmental settings (47, 48). Moreover, most Inc18 plasmids carry locus coding for stabilization systems, such as the post-segregation killing (PSK) system (49), which has already been implicated in the persistence of the Tn1546-mediated vanA resistance in E. faecium (50). The presence of ORF's adjacent to optrA in pL9 that matched few or no DNA sequences available in GenBank indicates that further investigation is required to understand how new conjugative Inc18 plasmid mosaics are evolving and how that might favor the spread of oxazolidinone resistance in animal agriculture settings.

## Transferability of *cfr, optrA,* and Other Resistance Determinants in LR *E. faecalis* L9

Filter mating assays were carried out to determine the potential for conjugal transfer of cfr and optrA at different CHL and LZD concentrations (**Table 1**). optrA/fexA/tet(S)-carrying OG1RF-L9 transconjugants were selected at a frequency of 4 × 10−<sup>7</sup> transconjugant cells per donor cell using 25µg/ml CHL, but conjugation experiments failed to transfer cfr at 25µg/ml CHL. Decreasing CHL concentration from 25 to 10µg/ml, countless small colonies of optrA/cfr/fexA/tet(S) carrying OG1RF-L9 transconjugants could be selected. Linezolid could select only countless small colonies of optrA-positive OG1RF-L9 transconjugants; no cfr-positive OG1RF-L9 transconjugant was obtained, indicating that optrA is responsible for linezolid resistance, and cfr, for a lower-level chloramphenicol resistance phenotype in LR E. faecalis L9.

Tn558, which harbors the chloramphenicol/florfenicol efflux MFS transporter fexA gene and the tetracycline resistance ribosomal protection gene tet(S), was also transferred to OG1RF-L9 in a yet to be determined genetic context, as pL9 could not be closed during de novo assembly. A 204-nt fragment of a gene for a conjugal transfer protein 5′ of tet(S) that appears to be involved in its mobilization is identical to homologs occurring in Firmicutes as identified in a BLASTn search. A 210 nt ORF 3′ flank to Tn558 matched a gene coding for the replication-associated protein RepB identified in Listeria monocytogenes (KY613776.1 and KY613741.1) isolated from food in Canada, in Carnobacterium divergens (LT984411.1) from beef carpaccio in France, and in an optrA-carrying conjugative plasmid from the Enterococcaceae strain E508 (MK425645.1) in China. Another 606-nt ORF encoding a hypothetical protein at the extreme 3′ end of repB also matched ORF's from the L. monocytogenes, C. divergens, and Enterococcaceae isolates mentioned above, and the optrA-carrying Enterococcus avium isolate C674 (MH018573.1) from an asymptomatic healthy human in China. At 5′ , Tn558 is flanked by a 210-nt ORF for a hypothetical protein that only matches sequences from the optrA gene cluster from E. avium C674 and pStrcfr from Streptococcus suis S10 (KF129409.1). Further upstream, a 1,272-nt ORF for a Yfamily DNA polymerase is present, but no match was found for this nucleotide sequence.

## Bacterial Immunity of *cfr/optrA*-Carrying *E. faecalis* L9

Genome defenses for porcine LR E. faecalis L9 were investigated using CRISPRfinder. Clustered, regularly interspaced short palindromic repeat (CRISPR) loci provide an important defense against parasitic mobile element entry. MDR, hospital-adapted enterococcal lineages lack CRISPR defenses, which are thought to enhance the facility with which they acquire antibiotic resistances on mobile elements (39, 51). A CRISPR-related loci consisting of 9 spacers and direct repeat sequences of 36 bp was found in the L9 chromosome (1,708,044 to 1,708,673 bp), but genes coding for Cas proteins were not identified. A BLASTn search revealed that the L9 CRISPR-related loci possess 95.12% DNA identity over 100% of its length to the corresponding chromosomal DNA sequence from E. faecalis SGAir0397 (CP039434.1), which was recovered from air in Singapore. L9 CRISPR spacer sequences only matched to sequences from E. faecalis FDAARGOS\_324 (CP028285.1) isolated from a human eye in the US, and the cyanobacterium Geminocystis sp. isolate NIES-3708 (AP014815.1) from Japan, besides E. faecalis SGAir0397. LR E. faecalis L9 lacked the E. faecalis CRISPR1 locus (a CAS-TypeIIA cluster consisting of Csn2\_0\_IIA, Cas2\_0\_I-II-III, Cas1\_0\_II, Cas9\_1\_II) typically located between genes EF0672 and EF0673 (51).

TABLE 1 | Conjugation efficiency of cfr and optrA from E. faecalis L9 to E. faecalis OG1RF transconjugants.


<sup>a</sup>Graded levels of CHL and LZD were tested to select for oxazolidinone and phenicol resistance genes, in addition to 25 µg/ml fusid acid, and 25µg/ml rifampicin to select for the OG1RF chromosomal markers.

<sup>b</sup>Conjugation efficiency corresponds to the number of CFU transconjugants per CFU donors; ND (not determined): only small bacterial colonies (countless) were obtained, all optrA/fexA-positive.

<sup>c</sup>optrA could be efficiently transferred by conjugation at a frequency of 4 × 10−<sup>7</sup> per donor cell using 25µg/ml of CHL as previously reported (36), and also decreasing CHL concentration or using LZD instead of CHL.

To the best of our knowledge, this is the first report of the coexistence of optrA and cfr in a bacterial isolate in Brazil. The fact that LR E. faecalis L9 came from a pool of 13 LR E. faecalis collected from healthy piglets in swine herds distributed across 7 Brazilian States highlights the need for monitoring the use of antibiotics in the country's swine production system in order to preserve the few remaining last-line antibiotics to treat infections caused by MDR pathogens.

## DATA AVAILABILITY STATEMENT

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www. ncbi.nlm.nih.gov/genbank/, CP018004.1, https://www.ncbi.nlm. nih.gov/genbank/, CP041775.1, https://www.ncbi.nlm.nih.gov/ genbank/, CP041776.1.

## REFERENCES


## ETHICS STATEMENT

The study was approved by the Ethics Committee of Faculdade de Medicina Veterinária e Zootecnia- Universidade de São Paulo, under number CEUA N.8026060214.

## AUTHOR CONTRIBUTIONS

LA and MG planned the study and wrote the manuscript with suggestions from AG, PB, and FL. PF and AM provided a collection of 245 MDR E. faecalis strains isolated from swine. LA, JS, RS, and IB-J carried out the experimental work and genome sequencing. LA, FZ, AG, PB, and FL contributed to the bioinformatic analyses. All authors contributed to the article and approved the submitted version.

## FUNDING

This project has been funded by the Harvard-wide Program on Antibiotic Resistance NIH/NIAID AI083214, and FAPESP scholarship 2014/ 27267-0.

important gram-positive cocci in the United States from the LEADER surveillance program (2011 to 2015). Antimicrob Agents Chemother. (2017) 61:e00609–17. doi: 10.1128/AAC.00609-17


of Enterococcus faecalis. Antimicrob Agents Chemother. (2012) 56:3917–22. doi: 10.1128/AAC.00419-12


**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 Almeida, Gaca, Bispo, Lebreton, Saavedra, Silva, Basílio-Júnior, Zorzi, Filsner, Moreno and Gilmore. 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.