FOOD SAFETY AND FOODBORNE PATHOGEN – A GLOBAL PERSPECTIVE ON THE DIVERSITY, COMBATING MULTIDRUG RESISTANCE AND MANAGEMENT

EDITED BY : Learn-Han Lee, Om V. Singh, Nurul-Syakima Ab Mutalib and Marta López PUBLISHED IN : Frontiers in Microbiology

#### 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 978-2-88966-239-5 DOI 10.3389/978-2-88966-239-5

### 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

## FOOD SAFETY AND FOODBORNE PATHOGEN – A GLOBAL PERSPECTIVE ON THE DIVERSITY, COMBATING MULTIDRUG RESISTANCE AND MANAGEMENT

Topic Editors:

Learn-Han Lee, Monash University Malaysia, Malaysia Om V. Singh, TSG Consulting, a Science Group Company, United States Nurul-Syakima Ab Mutalib, National University of Malaysia, Malaysia Marta López, Consejo Superior de Investigaciones Científicas (CSIC), Spain

A question raised by many individuals today – "How Safe is Our Food Consumed Today?" Food safety has become a hot topic and an important public issue due to the increasingly widespread nature of foodborne illnesses in both developed and developing countries. As food is biological in nature and supplies consumers with nutrients, it is also equally capable of supporting the growth of microorganisms from the environmental sources. A precise method of monitoring and detecting of foodborne pathogens including Salmonella sp., Vibrio sp., Listeria monocytogenes, Campylobacter and Norovirus is needed to prevent and control human foodborne infections. Clinical treatments of infection caused by foodborne pathogens are becoming tougher with the increase number of multidrug resistant pathogens in the environment. This situation creates a huge healthcare burden – e.g. prolonged treatment for infections, decrease in the efficacy of antibiotic, delay in treatment due to unavailability of new antibiotics, and increased number of deaths. As such, continuous investigation of the foodborne pathogens is needed to pave the way for a deeper understanding on the foodborne diseases and to improve disease prevention, management and treatments.

Citation: Lee, L.-H., Singh, O. V., Ab Mutalib, N.-S., López, M., eds. (2020). Food Safety and Foodborne Pathogen – A Global Perspective on the Diversity, Combating Multidrug Resistance and Management. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-239-5

# Table of Contents

*07 Breast Milk is a Potential Reservoir for Livestock-Associated*  Staphylococcus aureus *and Community-Associated* Staphylococcus aureus *in Shanghai, China*

Xiaoliang Li, Yun Zhou, Xianlin Zhan, Weichun Huang and Xing Wang

*16 Association Between* agr *Type, Virulence Factors, Biofilm Formation and Antibiotic Resistance of* Staphylococcus aureus *Isolates From Pork Production*

Yang Zhang, Dongyang Xu, Lei Shi, Rujian Cai, Chunling Li and He Yan


Julio Parra-Flores, Juan Aguirre, Vijay Juneja, Emily E. Jackson, Ariadnna Cruz-Córdova, Jesus Silva-Sanchez and Stephen Forsythe

*64 Discovery on Antibiotic Resistance Patterns of* Vibrio parahaemolyticus *in Selangor Reveals Carbapenemase Producing* Vibrio parahaemolyticus *in Marine and Freshwater Fish*

Learn-Han Lee, Nurul-Syakima Ab Mutalib, Jodi Woan-Fei Law, Sunny Hei Wong and Vengadesh Letchumanan

*77 Prevalence, Serotyping, Molecular Typing, and Antimicrobial Resistance of* Salmonella *Isolated From Conventional and Organic Retail Ground Poultry*

Ahmed H. Gad, Usama H. Abo-Shama, Katherine K. Harclerode and Mohamed K. Fakhr


Xiaojing Tian, Qianqian Yu, Donghao Yao, Lele Shao, Zhihong Liang, Fei Jia, Xingmin Li, Teng Hui and Ruitong Dai


Zongbao Liu, Uli Klümper, Lei Shi, Lei Ye and Meng Li


Shi Wu, Jiahui Huang, Feng Zhang, Qingping Wu, Jumei Zhang, Rui Pang, Haiyan Zeng, Xiaojuan Yang, Moutong Chen, Juan Wang, Jingsha Dai, Liang Xue, Tao Lei and Xianhu Wei

*182 Survival and Environmental Stress Resistance of* Cronobacter sakazakii *Exposed to Vacuum or Air Packaging and Stored at Different Temperatures*

Yichen Bai, Haibo Yu, Du Guo, Shengyi Fei and Chao Shi

*189 Probiotic Properties of* Enterococcus *Isolated From Artisanal Dairy Products*

Yousef Nami, Reza Vaseghi Bakhshayesh, Hossein Mohammadzadeh Jalaly, Hajie Lotfi, Solat Eslami and Mohammad Amin Hejazi

*202 Occurrence and Characterization of Methicillin Resistant* Staphylococcus aureus *in Processed Raw Foods and Ready-to-Eat Foods in an Urban Setting of a Developing Country*

Mohammad Aminul Islam, Sahana Parveen, Mahdia Rahman, Mohsina Huq, Ashikun Nabi, Zahed Uddin Mahmood Khan, Niyaz Ahmed and Jaap A. Wagenaar

*209 Multi-Laboratory Validation of a Loop-Mediated Isothermal Amplification Method for Screening* Salmonella *in Animal Food* Beilei Ge, Kelly J. Domesle, Qianru Yang, Thomas S. Hammack,

Shizhen S. Wang, Xiaohong Deng, Lijun Hu, Guodong Zhang, Yuan Hu, Xiaokuang Lai, Kyson X. Chou, Jan Ryan Dollete, Kirsten A. Hirneisen, Sammie P. La, Richelle S. Richter, Diyo R. Rai, Azadeh A. Yousefvand, Paul K. Park, Cindy H. Wu, Tameji Eames, David Kiang, Ju Sheng, Dancia Wu, Lori Hahn, Lisa Ledger, Cynthia Logie, Qiu You, Durda Slavic, Hugh Cai, Sherry L. Ayers, Shenia R. Young and Ruiqing Pamboukian

*222 A Rapid Method for Detection of* Salmonella *in Milk Based on Extraction of mRNA Using Magnetic Capture Probes and RT-qPCR*

Yalong Bai, Yan Cui, Yujuan Suo, Chunlei Shi, Dapeng Wang and Xianming Shi


Francesca Fanelli, Angela Di Pinto, Anna Mottola, Giuseppina Mule, Daniele Chieffi, Federico Baruzzi, Giuseppina Tantillo and Vincenzina Fusco

*260 Attenuation of Multiple* Vibrio parahaemolyticus *Virulence Factors by Citral*

Yi Sun, Du Guo, Zi Hua, Huihui Sun, Zhanwen Zheng, Xiaodong Xia and Chao Shi


Xuchu Wang, Silpak Biswas, Narayan Paudyal, Hang Pan, Xiaoliang Li, Weihuan Fang and Min Yue

*299 Corrigendum: Antibiotic Resistance in* Salmonella *Typhimurium Isolates Recovered From the Food Chain Through National Antimicrobial Resistance Monitoring System Between 1996 and 2016* Xuchu Wang, Silpak Biswas, Narayan Paudyal, Hang Pan, Xiaoliang Li,

Weihuan Fang and Min Yue

*300 Isolation and Characterization of Clinical* Listeria monocytogenes *in Beijing, China, 2014–2016*

Xiaoai Zhang, Yanlin Niu, Yuzhu Liu, Zheng Lu, Di Wang, Xia Cui, Qian Chen and Xiaochen Ma

*311 A Novel Mathematical Model for Studying Antimicrobial Interactions Against* Campylobacter jejuni

Mohammed J. Hakeem, Khalid A. Asseri, Luyao Ma, Keng C. Chou, Michael E. Konkel and Xiaonan Lu

*322 The Effect of Previous Life Cycle Phase on the Growth Kinetics, Morphology, and Antibiotic Resistance of* Salmonella *Typhimurium DT104 in Brain Heart Infusion and Ground Chicken Extract*

Jabari L. Hawkins, Joseph Uknalis, Tom P. Oscar, Jurgen G. Schwarz, Bob Vimini and Salina Parveen

*333 Building of Pressure-Assisted Ultra-High Temperature System and Its Inactivation of Bacterial Spores*

Dong Liang, Liang Zhang, Xu Wang, Pan Wang, Xiaojun Liao, Xiaomeng Wu, Fang Chen and Xiaosong Hu

*344 Enhanced Efficacy of Peroxyacetic Acid Against* Listeria monocytogenes *on Fresh Apples at Elevated Temperature*

Xiaoye Shen, Lina Sheng, Hui Gao, Ines Hanrahan, Trevor V. Suslow and Mei-Jun Zhu

*353 Embracing Diversity: Differences in Virulence Mechanisms, Disease Severity, and Host Adaptations Contribute to the Success of Nontyphoidal*  Salmonella *as a Foodborne Pathogen*

Rachel A. Cheng, Colleen R. Eade and Martin Wiedmann


Silin Tang, Renato H. Orsi, Hao Luo, Chongtao Ge, Guangtao Zhang, Robert C. Baker, Abigail Stevenson and Martin Wiedmann

*405 Composition and Dynamics of Bacterial Communities in a Full-Scale Mineral Water Treatment Plant*

Lei Wei, Qingping Wu, Jumei Zhang, Weipeng Guo, Qihui Gu, Huiqing Wu, Juan Wang, Tao Lei, Moutong Chen, Musheng Wu and Aimei Li

*417 Prevalence, Genotypic Characteristics and Antibiotic Resistance of*  Listeria monocytogenes *From Retail Foods in Bulk in Zhejiang Province, China*

Yunyi Zhang, Shilei Dong, Honghu Chen, Jiancai Chen, Junyan Zhang, Zhen Zhang, Yong Yang, Ziyan Xu, Li Zhan and Lingling Mei

*431 Epidemiological and Molecular Investigations on* Salmonella *Responsible for Gastrointestinal Infections in the Southwest of Shanghai From 1998 to 2017*

Xulin Qi, Pei Li, Xiaogang Xu, Yiqun Yuan, Shurui Bu and Dongfang Lin

# Breast Milk Is a Potential Reservoir for Livestock-Associated *Staphylococcus aureus* and Community-Associated *Staphylococcus aureus* in Shanghai, China

#### Xiaoliang Li 1†, Yun Zhou2†, Xianlin Zhan3†, Weichun Huang<sup>1</sup> and Xing Wang<sup>1</sup> \*

#### *Edited by:*

*Learn-Han Lee, Monash University Malaysia, Malaysia*

#### *Reviewed by:*

*Giuseppe Spano, University of Foggia, Italy Atte Von Wright, University of Eastern Finland, Finland David Christopher Coleman, Trinity College Dublin, The University of Dublin, Ireland*

> *\*Correspondence: Xing Wang*

*wx\_5166@163.com*

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

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 18 October 2017 Accepted: 18 December 2017 Published: 11 January 2018*

#### *Citation:*

*Li X, Zhou Y, Zhan X, Huang W and Wang X (2018) Breast Milk Is a Potential Reservoir for Livestock-Associated Staphylococcus aureus and Community-Associated Staphylococcus aureus in Shanghai, China. Front. Microbiol. 8:2639. doi: 10.3389/fmicb.2017.02639*

*<sup>1</sup> Department of Laboratory Medicine, Shanghai Children's Medical Center, Shanghai Jiaotong University School of Medicine, Shanghai, China, <sup>2</sup> Department of Intensive Care Unit, Huashan Hospital, Fudan University, Shanghai, China, <sup>3</sup> Department of Laboratory Medicine, The 455th Hospital of Chinese People's Liberation Army, Shanghai, China*

Breast milk is the first choice in feeding newborn infants and provides multiple benefits for their growth and development. *Staphylococcus aureus* usually exists in breast milk and is considered one of the most important causative infective agents. To be effective in preventing and controlling *S. aureus* infections among infants, the aim of this study was to determine the occurrence and molecular characteristics of *S. aureus* isolated from 1102 samples of breast milk between 2015 and 2016 in Shanghai, China. Out of 71 *S. aureus* strains isolated, 15 (21.1%, 15/71) were MRSA and all the strains were characterized by *spa* typing, Multi-Locus Sequence Typing, SCC*mec* typing, antibiotic resistance testing and virulence-associated genes. A total of 18 distinct sequence types (STs) and 36 *spa* types were identified within the 71 isolates, among which the most frequently represented was ST398 (19.7%, 14/71), followed by ST7 (18.3%, 13/71), ST59 (16.9%, 12/71). The three predominant STs accounted for more than one half of all *S. aureus* isolates. The most prevalent *spa* types were *t*091 (12.7%, 9/71), followed by *t*571 (8.5%, 6/71), *t*189 (7.0%, 5/71), *t*034 (5.6%, 4/71), *t*437 (5.6%, 4/71), and *t*701 (4.2%, 3/71). All MRSA isolates belonged to SCC*mec* IV and V, accounting for 66.7 and 33.3% respectively. Notably, 23 (32.4%) *S. aureus* strains were multidrug resistance (MDR), including 4 (5.6%) MRSA and 19 (26.8%) MSSA strains, and MDR isolates were mostly resistant to penicillin, erythromycin and clindamycin. All isolates exhibited simultaneous carriage of at least 5 of 33 possible virulence genes and the most prevalent genes detected were *icaA* (100%), *clfA* (100%), *hla* (100%), *sdrC* (94.4%), *hlg*2 (88.7%), *lukE* (57.8%). 39 (54.9%, 39/71) isolates, including 9 (12.7%) of MRSA isolates, harbored ≥10 tested virulence genes evaluated in this study. The *pvl* gene was detected in 8 strains, which represented 5 different STs, with ST59 being the most one. Overall, our findings showed that *S. aureus* strains isolated from breast milk were mainly MSSA (78.9%, 56/71) and exhibited high genetic diversity in Shanghai area of China. Breast milk was a reservoir for LA-SA (ST398) and CA-SA (ST59), which was likely a vehicle for transmission of multidrug-resistant *S. aureus* and MRSA lineages. This is a potential public health risk and highlights the need for good hygiene practices to reduce the risk of infant infections.

Keywords: breast milk, livestock-associated *Staphylococcus aureus*, community-associated *Staphylococcus aureus,* prevalence, antibiotic resistance

### INTRODUCTION

Breast milk is recognized as the best food for newborn infants, which contains all the nutrients that are essential to the children in the first 6 months and favors the development of the immune system (Albesharat et al., 2011). However, breast milk is not sterile and represents a complex ecosystem with a considerable diversity of bacteria instead. It is well known to be colonized by benefical flora with a majority of bifidobacteria, promoting development of infant's healthy gut microbiota. Not surprisingly, it may contains potentially pathogenic bacteria species (Barbosa-Cesnik et al., 2003). In addition, the collection, storage and transport of breast milk may introduce pathogenic contamination, increasing the risk of infection to these vulnerable premature infants. In fact, breast milk has been reported to act as a repository of bacteria for vertical transmission from mother to infant. Staphylococcus aureus is the most frequently isolated pathogenic bacteria in breast milk (Barbosa-Cesnik et al., 2003) and could cause a wide variety of infections including pneumonia, sepsis, skin lesion and food poisoning among infants.

S. aureus is a common colonizer of skin and mucous membranes in human and animals, and 30–50% of healthy adults are colonized with it during their lifetime. S aureus infection occurs following breaks in skin or mucosal barriers, ranging from mild skin and soft-tissue infections to severe systemic infections such as sepsis and necrotizing pneumonia (Lowy, 1998). S. aureus has been recognized as a major cause of hospital-associated (HA) infections worldwide firstly, thereafter it transferred into the communities and became an important causative agent of community-associated (CA) infections (Mediavilla et al., 2012; Li et al., 2016). Recently, S. aureus has been identified as an emerging pathogen in livestock, companion animals and humans in contact with livestock, which called livestock-associated S. aureus (LA-SA) (Fitzgerald, 2012). Besides, S. aureus strains have been reported in animal-source food such as meat, fish, milk and dairy products (Wang et al., 2014), suggesting these foods may serve as reservoirs and sources of community-associated S. aureus (CA-SA). So far, it has become a particular public threat to human and animal health. There are some differences between HA-MRSA, CA-MRSA, and LA-MRSA in molecular characteristics (Chuang and Huang, 2013; Chen and Huang, 2014). HA-MRSA isolates typically harbor relatively large SCCmec elements (types I-III), and are resistant to multiple antibiotics, including β -lactams. CA-MRSA isolates usually carry smaller SCCmec elements (types IV-V) and are only resistant to β-lactam antimicrobials and possess different exotoxin gene profile. Most LA-MRSA strains are host-specific and contain variable mobile genetic element (MGEs). In China, ST239 and ST5 are predominant HA-MRSA clones (Xiao et al., 2013), ST59 is the most prevalent CA-MRSA clone (Qiao et al., 2013; Chen and Huang, 2014), while ST9, ST97, and ST398 are the common LA-MRSA clones (Cui et al., 2009; Wang et al., 2015). The spreading of epidemic clones among the hospital, the community and the livestock environment makes the distinction among CA-MRSA, HA-MRSA, and LA-MRSA become blurred.

It has been reported that breastfeeding was associated with severe neonatal disease, including infantile pneumonia, neonatal sepsis and food poisoning (Le Thomas et al., 2001; Kayiran et al., 2014). However, there has been no recommendation to examine breast milk routinely for pathogenic bacteria when a mother feeds her own baby. So far, fewer data are available regarding the prevalence of S. aureus and MRSA in breast milk. The aim of this study was to determine the prevalence, antibiotic resistance, and molecular characteristics of S. aureus and MRSA isolated from breast milk samples between 2015 and 2016 in Shanghai. Such information could provide guidance for further clinical and epidemiologic studies, rational usage of antimicrobial agents.

### MATERIALS AND METHODS

### Sample Collection and Bacterial Isolation

From January 2015 to December 2016, a total of 1102 breast milk samples were collected from pediatric patients' mothers in a university hospital in Shanghai (Shanghai Children's Medical Center, affiliated with Shanghai Jiao Tong University). For milk collection, the breast of these mothers were cleaned with water and dried. Cotton swabs with 70% ethanol were used to disinfect the surfaces of the breast. The first few streams of milk were dropped. The collected milk was kept in a cooler with ice and transported to the laboratory within 2 h. The milk samples were cultured on 5% blood plate and inoculated at 37◦C for 24 h. S. aureus isolates were confirmed by classic microbiological methods: Gram stain and catalase and coagulase activity on rabbit plasma. They were further identified by biochemical characterization using the Api-Staph test (bioMérieux, Lyon, France). All S. aureus isolates recovered from breast milk samples were each from a separate mother. These isolates were processed in Class II Biological Safety Cabinets. All strains were stored at −70◦C and grown overnight on sheep blood agar plates at 37◦C.

This study was approved by the Ethics Committee of Shanghai Children's Medical Center, and all isolates were collected with informed consents prior to sample collection.

### Antimicrobial Susceptibility Testing

The antibiotic susceptibility profiles of all S. aureus isolates in the current study were performed using the bioMe'rieux VITEK2 system following manufacturer's instructions. Results were interpreted according to the recommendations and definitions of the Clinical and Laboratory Standards Institute (CLSI, 2015). The following 17 drugs were tested: cefoxitin (FOX), linezolid (LZD), ciprofloxacin (CIP), clindamycin (DA), erythromycin (E), trimethoprim-sulfamethoxazole (SXT), moxifloxacin (MOF), nitrofurantoin (FD), vancomycin (V), tetracycline (TET), penicillin (P), rifampicin (RF), levofloxacin (LVX), ampicillin (AMP), gentamicin (GM), quinupristin/dalfopristin (Q/D), and tigecycline (TGC). S. aureus ATCC 29213 was used as a quality control.

### MLST Analysis

All S. aureus isolates were performed according to the protocol of Enright (Enright and Spratt, 1999) on the S. aureus MLST website (http://saureus.mlst.net) to detect the following seven housekeeping genes (Aanensen and Spratt, 2005): carbamate kinase (arcC), shikimate dehydrogenase (aroE), glycerol kinase (glp), guanylate kinase (gmk), phosphate acetyltransferase (pta), triosephosphate isomerase (tpi), and acetyl coenzyme A acetyltransferase (yqiL). PCR amplicons of seven S. aureus housekeeping genes were obtained from chromosomal DNA. The sequences of the PCR products were compared with the existing alleles available from the MLST website, and alleles and ST were assigned by submitting the sequences. Clustering of related STs, which were defined as clonal complexes (CCs), was determined using eBURST (based on related STs).

### SCC*mec* Typing

Staphylococcal cassette chromosome mec (SCCmec) typing was carried out discriminating the mec complex and the cassette chromosome recombinase(ccr) genes complex as described elsewhere (Kondo et al., 2007), which was based on a set of multiplex PCRs (M-PCRs) with 14 primers. SCCmec types I– V were assigned according to the combination of the ccr type and mec class. MRSA isolates that could not be assigned to any expected type were defined as nontypable (NT).

### *Spa* Typing

In S. aureus, the polymorphic X region of staphylococcal protein A (spa) gene consists of a variable number of 24 bp repeat units (Shopsin et al., 1999) that allow isolates to be distinguished from one another. The spa typing was based on variations of the repeat units. Amplification and sequencing of the X region were performed as described previously by Koreen et al. (2004). The spa typing was assigned by submitting the data to the S. aureus spa type database (http://www.ridom.de/spa-server/).

### Detection of Virulence Genes

All S. aureus isolates were subjected to a multiplex PCR assay for the detection of 33 staphylococcal virulence genes: the staphylococcal enterotoxin genes (sea, seb, sec, sed, see, seg, seh, sei, sej, sel, sem, sen, seo, sep, seq, sek), the toxic shock syndrome toxin (tsst), the arginine catabolic mobile gene(arcA), the exfoliative toxin genes (eta, etb), the leukocidin (lukF/S-PV, lukE, lukM) (Lina et al., 1999), the bacteriocin (bsaA), the hemolysin gene (hla, hlb, hlg, hlg2), and the adhesin genes (clfA, icaA, sdrC, sdrD, and sdrE) as previously described (Arvidson and Tegmark, 2001; Peacock et al., 2002; Bubeck Wardenburg et al., 2007).

### Statistical Analysis

Statistical analyses were performed using Stata software (version 10.1/SE, Stata Corp, College Station, TX, USA). We used the χ 2 and Fisher's exact tests, as appropriate for analysis of categorical data. Statistical significance was set at P ≤ 0.05.

### RESULTS

### Prevalence of *S. aureus* and MRSA in Breast Milk

Overall 1102 breast milk samples, collected from a university hospital in Shanghai between 2015 and 2016, were subjected to bacteriological analysis. Seventy-one (6.4%, 71/1102) strains of S. aureus isolated from 71 breast milk, 15 (21.1%, 15/71) were MRSA. PCR assay for mecA and disk diffusion test with oxacillin confirmed methicillin resistance of 15 isolates.

### MLST, SCC*mec,* and *spa* Typing

The evolutionary and genetic diversity of S. aureus isolates within breast milk was analyzed by MLST (**Table 1**). There were 18 distinct STs identified within the 71 isolates, among which the most frequently represented was ST398 (19.7%, 14/71), followed by ST7 (18.3%, 13/71), ST59 (16.9%, 12/71). These three predominant STs accounted for more than one half of all S. aureus isolates. Other STs represented included ST188 (7.0%, 5/71) and ST6 (7.0%, 5/71) with five isolates, ST1 (4.2%, 3/71) and ST5 (4.2%, 3/71) with three isolates, ST15 (2.8%, 2/71), ST20 (2.8%, 2/71), ST88 (2.8%, 2/71), ST615 (2.8%, 2/71), and ST630 (2.8%, 2/71) with two isolates, and 6STs (ST8, ST12, ST22, ST25, ST508, and ST1290) with one isolate. Eight isolates harboring pvl were distributed among 5 different STs, including ST59 (4 isolates) as well as ST188, ST1, ST615, ST22 (1 isolate each).

Thirty-six spa types were observed among the 71 isolates. The most prevalent spa types were t091 (12.7%, 9/71), followed by t571 (8.5%, 6/71), t189 (7.0%, 5/71), t034 (5.6%, 4/71), t437 (5.6%, 4/71), t701 (4.2%, 3/71). Each of the remaining spa types was represented in less than three isolates.

The eBURST analysis was performed on all the S. aureus isolates by using all STs available in the MLST database was shown. This methodology revealed that the strains clustered into 9 CCs (CC398, CC7, CC59, CC1, CC5, CC72, CC15, CC20, CC78) and 4 singletons (**Figure 1**). The largest cluster was CC398 with 14 isolates, followed by CC7 with 13 isolates, CC59 with 12 isolates, CC1 with 9 isolates, CC5 with 8 isolates, CC72 with 8 isolates, CC15 with 2 isolates, CC20 with 2 isolates and CC78 with 2 isolates.

SCCmec typing was performed on 15 MRSA isolates. Among them, only two types (type IV and V) were found. Two thirds of them were type IV (66.7%, 10/15), and one third were type V (33.3%, 5/15).



There was a strong association observed between specific ST and spa types. The ST398 genotype was associated mainly with spa t571 (6/14) and spa t034 (4/14), less frequently with three types: t2582, t6606, and t7160. The ST7 genotype was mainly linked with t091 (9/13), less frequently with t796, t1685, and t14204. The ST59 genotype was associated primarily with spa t172 (5/12) and t437 (4/12). All the ST188 genotype was associated with spa t189.

### Antimicrobial Susceptibility Testing

The antimicrobial resistance profiles of 71 S. aureus isolates according to MLST were listed in **Table 2**. All the strains were susceptible to vancomycin, linezolid, nitrofurantoin, rifampicin, tigecycline and quinupristin/dalfopristin. Resistance to penicillin was observed in the majority (84.5%), followed by erythromycin (35.2%), clindamycin (29.6%), tetracycline (22.5%), cefoxitin (21.1%), ampicillin (21.1%) and trimethoprim-sulfamethoxazole (16.9%). The resistance rates to other antibiotics tested were less than 6%, including 5.6% to gentamicin, 4.2% to levofloxacin, 4.2% to ciprofloxacin, and 2.8% to moxifloxacin.

Among the 71 S. aureus isolates, 23 (32.4%) strains were resistant to ≥3 antibiotics, including 4 (5.6%) MRSA and 19 (26.8%) MSSA strains. In the MSSA strains, eight (11.2%) strains were resistant to 3 antibiotics and mostly resistant to penicillin, erythromycin and clindamycin(Supplementary Table 1), seven (9.9%) strains showed resistance to 4 antibiotics, and four (5.6%) strains were resistant to ≥5 antibiotics, however, only four MRSA strains were found to be resistant to at least three antibiotics.

### Virulence Gene Profiles

The distribution of 33 putative virulence genes varied among the 71 S. aureus strains according to STs (**Table 3**). All of these virulence genes except lukM and etb genes were identified within multiple isolates, and all isolates exhibited simultaneous carriage of at least 5 virulence genes. Thirty-nine (54.9%, 39/71) isolates harbored ≥10 tested virulence genes, among which were 2 isolate with 20 genes, 1 isolates with 19 genes, 1 isolates with 18 genes, 3 isolates with 17 genes, 2 isolates with 16 genes, 1 isolates with 15 genes, 7 isolates with 14 genes, 6 isolates with 13 genes, 1 isolates with 12 genes, 7 isolates with 11 genes and 8 isolates with 10 genes. Compared with MSSA isolates, the carriage rates for arcA and seq genes in MRSA isolates were significantly higher, while those of sdrD and lukE were significantly lower. The pvl gene was detected in 8 strains, which represented 5 different STs, with ST59 being the most common.

Adhesion genes were present in most of the S. aureus isolates; 100% carried the icaA and clfA genes, 94.4% harbored sdrC, 67.6% carried sdrE and 49.3% carried sdrD.

The most prevalent toxin genes detected were hla (100%), hlg2 (88.7%), lukE (57.8%), hlb (43.7%). The carriage rates for tsst (11.3%) and eta (7.0%) in breast milk isolates were low.

The carriage of staphylococcal enterotoxin genes was a strong association with MLST profiles.

Thirteen classical enterotoxin genes (sea, seb, sec, sed, see, seg, seh, sei, sem, sen, seo, seq, sek) were detected within these strains (**Table 4**). Overall, each enterotoxin gene was found in multiple S. aureus isolates, ranging from 5.6 to 31.0%. No enterotoxin gene

was found in ST1290 and ST630 isolates. The see-sep genes were present in the ST7 strains, whereas, the sed-sej genes were present in ST5 and ST615 strains. All ST5, ST20, ST22, ST25, ST26, ST508, and ST615 strains harbored seg-sei-sem-sen-seo genes, but ST59 isolates mainly carried seb-sek-seq genes.

### Molecular Characteristics of the Prevalent Clone ST398

In this study, ST398 (19.7%, 14/71) was found to be the most prevalent clone, including 5 MRSA and 9 MSSA isolates, which was associated primarily with spa t571 (6/14) and spa t034 (4/14), less frequently with three types: t2582, t6606, and t7160. Among 14 ST398 stains, all were susceptible to vancomycin, linezolid, nitrofurantoin, ciprofloxacin, moxifloxacin, levofloxacin, rifampicin, gentamicin, tigecycline and quinupristin/dalfopristin. The highest levels of resistance were observed for penicillin (100%), cefoxitin (35.7%), and ampicillin (35.7%). The resistance rates to other antibiotics tested were 21.4% to clindamycin,21.4% to erythromycin,7.1% to trimethoprim-sulfamethoxazole and 7.1% to tetracycline. In addition, there were no significant differences in antibiotic sensitivities between ST398 and non-ST398 isolates (Supplementary Table 2).

All ST398 isolates exhibited icaA, clfA, sdrC, hla, hlb, and hlg genes, however, the frequency of carriage for hlg2, sdrD, lukE, seb, and sek was significantly lower than that for non-ST398 isolates (Supplementary Table 3). In addition, there were no significant differences on the positive rate of pvl between ST398 and non-ST398 strains.

### DISCUSSION

Breast milk is considered to be the best source of nutrients for infant growth and development in the world. However, breast milk isn't always sterile and may contain pathogenic bacteria that could cause infections especially in premature infants. S. aureus is a common colonizer of skin and mucous membranes in human and infection by S. aureus is often occur following breaks in skin or mucosal barriers. S. aureus is one of the most frequently isolated pathogenic bacteria in breast milk (Barbosa-Cesnik et al., 2003) and could cause a wide variety of infections including pneumonia, sepsis, skin lesion and food poisoning among infants. Given these dangerous consequences, it is urgent to understand the prevalence, molecular characteristics and virulence profiles of S. aureus isolates from breast milk in order to implement right measures to control infection and transmission.

The detection rate of S. aureus in breast milk varies substantially worldwide, ranging from 2.5 to 100% in different countries. In Brazil, studies on the frequency of S. aureus in breast milk have shown differences between 2.5 and 34%. In the present study, 71 (6.4%, 71/1102) S. aureus strains were isolated from 1102 breast milk samples and 15 (1.4%, 15/1102) have identified as MRSA. This indicates that S. aureus is an important pathogenic bacterium in breast milk now and suggests the urgent need for active surveillance of S. aureus and MRSA infection and transmission in mothers and infants.

In the current study, ST398 (19.7%, 14/71), ST7 (18.3%, 13/71), and ST59 (16.9%, 12/71) were the three predominant STs, accounting for 54.9% of all S. aureus isolates. Surprisingly, ST398 was found to be the most frequently represented ST in breast milk. ST398 is a typical livestock-associated type (Graveland et al., 2011; Qiao et al., 2014), which first observed among pigs and pig farmers in Netherlands in 2003, then found in Austria, Germany and Denmark (Fluit, 2012). Afterward, it became the overwhelmingly dominant lineage in Europe and North America. Previous studies showed that patients carrying this type were usually in contact with animal reservoirs of these MRSA. Recently, ST398 clones were found in different samples



**12**



TABLE 4 | Frequencies of staphylococcal enterotoxin genes among the molecular types of 71 *S. aureus* isolates recovered from breast milk.


of patients in China, including sputum, blood, pus and secretion (Zhao et al., 2012; He et al., 2013; Song et al., 2017). Moreover, breast milk also became the source of ST398 in our study and favored the transmission between mothers and infants. There was no evidence shown that all the mothers had ever been exposed to livestock because of the lack of adequate information. It is very difficult to speculate on the origins of these isolates because of the absence of epidemiological data linking these to animals. However, livestock-associated S. aureus usually harbored an intact beta-toxin gene (hlb) and no lysogenic prophages encoding the immune evasion complex genes (sea, sep, sak, scn, and chp genes) (van Wamel et al., 2006). Among ST398 isolates in this study, they all harbored an intact hlb gene and didn't carry sea and sep genes. This was powerful evidence that these strains were of animal origin. Two strains lacked all the immune evasion complex genes, and others harbored one, two or three of sak, scn and chp genes. The sak, scn and chp genes are usually encoded by hlb-disrupting bacteriophages, this suggested the other isolates may harbor prophages integrated somewhere else besides the hlb gene. ST7, found in a total of 13 MSSA isolates, was the second common ST in the present study. ST7 has also been reported to be one of the most dominant MSSA genotypes in invasive CA-SA infection in Chinese children (Qiao et al., 2014). Another study in our group showed ST7 also was one of the common genotypes causing bovine mastitis in Shanghai between 2014 and 2015 (Li et al., 2017). ST7, which was considered as a pandemic clone, have arisen in communities and spread across the country. In addition, it is well known ST59 is the most predominant CA-MRSA clone in the Asia-Pacific region, including Taiwan and Hong Kong (Chuang and Huang, 2013). In China, previous studies revealed that ST59-MRSA-IV was the major lineage accounting for up to two-thirds of isolates (Geng et al., 2010). CC59 was also reported to be the most common clonal complex among the patients with SSTIs (Yu et al., 2015). Similar to these findings, ST59 were also found to be one of the dominant types in our study and still spread widely in the communities. From these results, our study provided evidence for the existence of two different lineages of S. aureus in breast milk in China: LA-SA and CA-SA.

The invasive potential of S. aureus largely depends on the carriage of a battery of virulence factors associated with adhesion, acquisition of nutrients, tissue penetration, evasion of host defenses and toxin-mediated responses (Dinges et al., 2000; Bubeck Wardenburg et al., 2007; Diep and Otto, 2008). Consistent with other findings, the prevalence of icaA, clfA, sdrC among S. aureus isolated from breast milk in our study were high (100, 100, and 94.4%), supporting the statement that adherence of S. aureus to host cells was the crucial initial step for bacterial pathogenicity. The distribution of some virulence genes, especially enterotoxin genes, was correlated with the different S. aureus lineages. All ST398 isolates harbored hlg gene, but the frequency of carriage for hlg2, sdrD, lukE, seb, and sek was significantly lower than that for non-ST398 isolates (P < 0.05). Compared with other ST isolates, ST7 isolates harbored less hlb

### REFERENCES


but more see-sep. None of ST59 isolates carried hlg, sdrD, and lukE, but contained seb-sek-seq genes, which was significantly higher than that among non-ST59 isolates. In addition, All ST5, ST20, ST22, ST25, ST26, ST508, and ST615 strains harbored seg-sei-sem-sen-seo genes. These findings implied distinctive virulence genes have different roles in the pathogenicity of S. aureus lineages.

In conclusion, our findings showed that breast milk was a reservoir for LA-SA (ST398) and CA-SA(ST59) and was likely a vehicle for transmission of multidrug-resistant S. aureus and MRSA lineages. This is a serious public health risk and highlights the need to implement good hygiene practices. Additional studies are required to assess the source of contamination of breast milk samples and the risk of infection to babies.

### AUTHOR CONTRIBUTIONS

XW: designed the studies and obtained funding; XW, XL, YZ, XZ, and WH: performed the experiments and/or analyzed the data; XW and XL: wrote the manuscript.

### FUNDING

This study was supported by the National Natural Science Foundation of China (grant 81301392) and the Training Program for Outstanding Young Teachers in Higher Education Institutions (ZZjdyx13132), the Training Program for Clinical Medical Young Talents in Shanghai (HYWJ201605), Visiting Scholar Research Program and SCMC-EPT Program to XW.

### ACKNOWLEDGMENTS

The authors would like to thank all the mothers who contributed their specimens for this study. We thank the microbiologists and technical staff of Shanghai Children's Medical Center for collecting the bacterial isolates and laboratory testing.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2017.02639/full#supplementary-material


swine and workers in China. J. Antimicrob. Chemother. 64, 680–683. doi: 10.1093/jac/dkp275


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

Copyright © 2018 Li, Zhou, Zhan, Huang and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Association Between agr Type, Virulence Factors, Biofilm Formation and Antibiotic Resistance of Staphylococcus aureus Isolates From Pork Production

Yang Zhang<sup>1</sup>† , Dongyang Xu<sup>2</sup>† , Lei Shi<sup>3</sup> , Rujian Cai<sup>4</sup> , Chunling Li<sup>4</sup> \* and He Yan1,3 \*

<sup>1</sup> School of Food Science and Engineering, South China University of Technology, Guangzhou, China, <sup>2</sup> Institute of Genomics, Huaqiao University, Xiamen, China, <sup>3</sup> State Key Laboratory of Food Safely Technology for Meat Products, Xiamen, China, <sup>4</sup> Institute of Animal Health, Guangdong Academy of Agricultural Sciences, Guangzhou, China

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Maria José Saavedra, Universidade de Trás-os-Montes e Alto Douro, Portugal Florence Dubois-Brissonnet, AgroParisTech-Institut des Sciences et Industries du Vivant et de l'Environnement, France

#### \*Correspondence:

Chunling Li lclclare@163.com He Yan yanhe@scut.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 08 March 2018 Accepted: 25 July 2018 Published: 20 August 2018

#### Citation:

Zhang Y, Xu D, Shi L, Cai R, Li C and Yan H (2018) Association Between agr Type, Virulence Factors, Biofilm Formation and Antibiotic Resistance of Staphylococcus aureus Isolates From Pork Production. Front. Microbiol. 9:1876. doi: 10.3389/fmicb.2018.01876 Livestock-associated Staphylococcus aureus colonization and/or infections exist in pigs and people in frequent contact with pigs. In this study, a total of 130 S. aureus isolates obtained from different stages of pork production were subjected to antimicrobial susceptibility, biofilm formation, as well as PCR screening to identify virulence genes, and the accessory gene regulator alleles (agr). Among all 130 S. aureus isolates, 109 (83.8%, 109/130) isolates were positive for agr. All swine farms isolates belonged to agr IV, whereas S. aureus isolated from slaughterhouse and retail indicated diverse agr types. All isolates exhibited biofilm formation ability, and raw meat isolates (belonging to agr I) exhibited a greater ability to form strong biofilms than swine farms isolates (belonging to agr IV). agr-positive isolates were associated with more virulence genes than agr-negative isolates. Most biofilm-producing isolates were positive for microbial surface component recognizing adhesive matrix molecule (MSCRAMM), capsule type and ica group genes. The results illustrate a significant association between the prevalence rate of MSCRAMM, capsule type and ica group genes among isolates producing weak, moderate and strong biofilms. The high prevalence of resistance to ciprofloxacin, gentamicin, tetracycline, clarithromycin, clindamycin, and trimethoprimsulfamethoxazole were mainly observed in moderate and weak biofilm producers. Our findings indicate that S. aureus isolates from pork production displayed diverse molecular ecology.

Keywords: Staphylococcus aureus, agr typing, biofilm formation, virulence gene, antibiotic resistance, pork production

### INTRODUCTION

Staphylococcus aureus is an important zoonotic pathogen that is responsible for a variety of infectious diseases characterized by septicemia and sepsis (Crombe et al., 2013; Song et al., 2015). China is one of the world's largest pork producers with more than 470 million pigs, accounting for ∼50% of the total numbers in the world (Krishnasamy et al., 2015). Consecutively, several reports suggested transmission between pigs and humans causing livestock-associated S. aureus

**16**

(LA-SA) colonization in 23–45% of pig-farmers (Voss et al., 2005; Huijsdens et al., 2006; Smith et al., 2008) and 4.6% of pigcare veterinarians (Wulf et al., 2006). The "One Health" concept recognizes that human health or livestock or wildlife health are interconnected and bound to the animal-human-ecosystems in which they (co)exist. Occupational exposure to swine has been associated with increased Staphylococcus aureus carriage, and increased risk of colonization and infections of different hosts (Witte et al., 2007; Graveland et al., 2011; Price et al., 2012; Kock et al., 2013; Ge et al., 2017; Davis et al., 2018). The risk of zoonotic transmission to humans demands our deep understanding of S. aureus contamination and ecology in the swine production.

The rapid development of resistance to multiple antimicrobial agents increases the difficulty treating S. aureus infections and biofilm production facilitate this organism to survive in the presence of antibiotics (Dhanawade et al., 2010; Bhattacharya et al., 2018). Several studies have demonstrated that low doses of certain antibiotics could induce biofilm formation, indicating that biofilm regulation might be involved in the global response to external stresses, including antibiotics (Hoffman et al., 2005; Kaplan, 2011). Previous studies regarding quantitative correlation between biofilm formation and antibiotics resistance have yielded different results. For example, Neopane et al. (2018) concluded that the biofilm-positive strains have a higher tendency to exhibit multidrug resistance and methicillin resistance compared to biofilm-negative strains, while Eyoh et al. (2014) indicated that there was no significant difference in the percentage of multi-drug-resistance (MDR) among biofilm producers and non-biofilm formers for both medical and nonmedical personnel.

Staphylococcus aureus produces a wide variety of protein toxins, such as exfoliative toxins, Panton-Valentine leukocidin, hemolysins, enterotoxins, and toxic shock syndrome toxin. Among the large array of S. aureus virulence factors the MSCRAMMs (microbial surface component recognizing adhesive matrix molecules) includes different adhesins, which are essential for initial stages of infection (Magro et al., 2017). MSCRAMMs, which includes fibronectin binding proteins (FnbA and FnbB), fibrinogen binding proteins (ClfA, ClfB and Efb), capsule proteins (Capsule type 5 and 8) and collagen binding proteins (Cna), can bind to a variety of mammalian extracellular proteins and abiotic surfaces (Donlan, 2002). Furthermore, the formation of a highly organized multicellular biofilm is related to the polysaccharide intercellular adhesin (PIA) production, which is controlled by the ica operon (Cramton et al., 1999). Therefore, the numbers and combinations of toxin genes may contribute to the pathogenicity of S. aureus.

While previous studies have documented the prevalence of S. aureus isolates in bovine mastitis (Fluit, 2012; Snel et al., 2015; Artursson et al., 2016; Kot et al., 2016; Magro et al., 2017), there is a lack of data regarding the prevalence and characterization of S. aureus in pork production. A thorough understanding of the correlation between the observed polymorphism in genotype and virulence, and the diversity in production practices is important for targeted mitigation. In this study, an extensive study was conducted involving systematic sampling of three commercial swine farms, a contracted slaughterhouse for the designated farms, and a retail market in Xiamen, China to profile S. aureus isolates along the production, processing and retail chain. The data enabled tracking of the spread of S. aureus from pork production and a better understanding of the evolution of S. aureus.

### MATERIALS AND METHODS

### Bacterial Strains and Antibiotic Susceptibility

From September – December 2014, three commercial swine farms with > 5000 pigs, one large slaughterhouse and several terminal markets were selected from Xiamen City, People's Republic of China, and 501 samples were collected from these places for S. aureus isolation. Pigs were born and raised in these three commercial swine farms with distance for more than 25 km from each other and then were sent to the slaughterhouse. These three swine farms and the slaughterhouse were vertically integrated pork processing plant, meaning pigs originated from these three swine farms contracted to sell hogs exclusively to the slaughterhouse. However, terminal samples from the markets did not totally originate from the slaughterhouse tested in the present study.

Briefly, a total of 501 non-duplicate samples were collected from the pork industry, including three commercial swine farms (sty door and soil, n = 71; nasal swabs, n = 97), one slaughterhouse (pork, n = 173), and terminal markets (pork, n = 160). Isolation and identification of S. aureus were performed according to China's National Technical Standard GB4789.10- 2010 and the special gene nuc was targeted by PCR for identifying S. aureus (Brakstad et al., 1992). Contamination with S. aureus was detected in 26.0% (130/501) of the total samples, and the prevalence of S. aureus was highest in the slaughterhouse (35.8%, 62/173) followed by the market (24.4%, 39/160) and the farm (17.3%, 29/168).

These isolates were assessed for antimicrobial susceptibility by the Kirby-Bauer disk diffusion method described by the Clinical and Laboratory Standards Institute (CLSI, 2012). The antibiotic disks used (Hangzhou Microbial Reagent Co., Ltd., Hangzhou) included ciprofloxacin (5 µg), penicillin (10 µg), gentamicin (10 µg), tetracycline (30 µg), clarithromycin (15 µg), clindamycin (2 µg), chloramphenicol (30 µg), sulfamethoxazoletrimethoprim (25 µg), nitrofurantoin (30 µg), rifampin (5 µg), cephalothin (30 µg), minocycline (30 µg), cefoxitin (30 µg) and oxacillin (1 µg).

### agr Genotyping

Bacterial genomic DNA template was extracted from the isolates by a commercial DNA extraction kit (Biomed, Beijing, China). The agr types (I–IV) were determined by a multiplex PCR assay as described by Gilot et al. (2002). In brief, multiplex PCR was performed with the following primers: Pan (5<sup>0</sup> -ATG CAC ATG GTG CAC ATG C-3<sup>0</sup> ), agr1 (5<sup>0</sup> -GTC ACA AGT ACT ATA AGC TGC GAT-3<sup>0</sup> ), agr2 (5<sup>0</sup> -TAT TAC TAA TTG AAA AGT GGC CAT AGC-3<sup>0</sup> ), agr3 (5<sup>0</sup> -GTA ATG TAA TAG CTT GTA TAA TAA TAC CCA G-3<sup>0</sup> ) and agr4 (5<sup>0</sup> -CGA TAA TGC


(Continued)


#### TABLE 1 | Continued

fmicb-09-01876 August 17, 2018 Time: 15:12 # 4

CGT AAT ACC CG-3<sup>0</sup> ). These primers yield a PCR product of 441, 575, 323, or 659 bp corresponding to agr group I, II, III, and IV, respectively. Each assay contained 2 µL of prepared DNA template, 2.5 µL of 10× Easy Taq Buffer [Takara Biomedical Technology (Beijing) Co., Ltd, China], 1 µL of 10 mM deoxynucleotide triphosphate [Takara Biomedical Technology (Beijing) Co., Ltd, China], 1 µL of upstream and downstream primers (10 µM), and 0.125 µL of DNA polymerase (5 U/µL) [Takara Biomedical Technology (Beijing) Co., Ltd, China], and the final system volume was adjusted to 25 µL with sterile ultrapure water. The PCR conditions were as follows: 1 cycle at 94◦C for 5 min; 26 cycles at 94◦C for 30 s, 55◦C for 30 s, and 72◦C for 1 min; and finally 1 cycle at 72◦C for 10 min. All PCR products were analyzed by electrophoresis on a 1.5% (w/v) agarose gel.

### Identification of Virulence Determinants

The nucleotide sequences of all PCR primers used in this study and their respective amplified products and specific Tm (◦C) are listed in **Table 1**. All the oligonucleotide primers were synthesized by Sangon Biotech (Shanghai, China). Each assay contained 1 µL of prepared DNA template, 2.5 µL of 10× Easy Taq Buffer [Takara Biomedical Technology (Beijing) Co., Ltd, China], 1 µL of 10 mM deoxynucleotide triphosphate [Takara Biomedical Technology (Beijing) Co., Ltd, China], 1 µL of upstream and downstream primers (10 µM), and 0.125 µL of DNA polymerase (5 U/µL) [Takara Biomedical Technology (Beijing) Co., Ltd, China], and the final system volume was adjusted to 25 µL with sterile ultrapure water. The PCR conditions were as follows: an initial denaturation at 95◦C for 5 min; 30 cycles of 95◦C for 30 s, specific Tm for 30 s, and 72◦C for 40–90 s depending on the PCR product length; and a final extension at 72◦C for 10 min. Sequencing of the extracted PCR product was performed by Beijing Genomics Institute (Shenzhen, China) and the data were analyzed with the GenBank database using the BLAST algorithm at the National Center for Biotechnology Information web site<sup>1</sup> .

### Biofilm Formation

Quantification of biofilm formation was performed by spectrophotometry in microplates (Nest Biotechnology Co., Ltd. Wuxi, China) using crystal violet staining as previously described (Pereyra et al., 2016). Briefly, 20 µL of bacterial log phase culture was added to 200 µL of fresh 1% glucose BHI in 96-well flat-bottom microtiter plates. S. aureus ATCC25923 (biofilmforming) and S. epidermidis ATCC12228 (not biofilm-forming) were used as positive and negative controls, respectively. BHI without bacteria served as the blank. The plates were incubated at 37◦C for 24, 48, and 72 h under aerobic conditions. After each sampling time, wells were washed three times with 300 µL of sterile phosphate-buffered saline (PBS; pH 7.2) and drained by inversion. Subsequently, 200 µL of methanol was added to each well and the plates were dried for 15 min. The adherent cells were stained with 150 µL of 0.1% crystal violet solution for 15 min and then washed twice with sterile water. Bound crystal violet was dissolved by treatment with 150 µL of 95% ethanol for 10 min, and OD<sup>570</sup> was measured for the stained bacteria and control wells. The experiment was performed in triplicate. An OD<sup>570</sup> value of 0.3 was taken as the cutoff point to differentiate between biofilm producers and non-biofilm-producer strains [cut-off value (ODc) = average OD of negative control + 3× standard deviation (SD) of negative control] (Pereyra et al., 2016). The quantitative classification of biofilm production based on ODc and average OD values was carried out, resulting in four categories of strains: strong biofilm producers (OD > 4 × ODc), moderate biofilm producers (4 × ODc > OD > 2 × ODc), weak

<sup>1</sup>www.ncbi.nlm.nih.gov



<sup>a</sup>The number in parentheses represents the percentage of isolates in the corresponding genotype.

biofilm producers (2 × ODc > OD > ODc), and no biofilm producers (OD < ODc) (Pereyra et al., 2016).

### Growth Rate Analysis

The growth of 12 strong, 12 moderate and 12 weak biofilm formers were measured according to Qi et al. (2016). Briefly, isolates were cultured in BHI agar for 18–24 h and adjusted to 0.5 McFarland units with 0.85% NaCl medium, and diluted 1: 20 in BHI medium. The cultures were incubated for 24 h at 37◦C with shaking at 200 rpm and the bacterial growth was monitored by measuring the OD<sup>600</sup> values of the culture. All experiments include three independent replicates.

### Statistical Analysis

Statistical analysis was performed with SPSS v.22.0 (SPSS Inc., Chicago, IL, United States). Differences groups were compared using the chi-squared test and a p-value of <0.05 was deemed to be significant. Spearman's rank correlation test was used for comparison of biofilm formation ability and multi-drugresistance (MDR).

### RESULTS

### agr Genotyping

By multiplex PCR, the agr types were successfully identified in 109 isolates, and 21 isolates were non-typeable for agr locus. As shown in **Table 2**, the agr I was most prevalent (39.2%; 51/130), followed by agr IV (32.3%; 42/130), agr II (9.2%; 12/130) and agr III (3.1%; 4/130). All swine farms isolates belonged to agr IV, whereas S. aureus isolated from slaughterhouse and retail indicated diverse agr types.

### Prevalence and Distribution of Virulence Genes

As illustrated in **Figure 1**, nearly all isolates harbored the hla (95.4%), hlb (100%) and hld (98.5%) genes, encoding alpha-, beta-, and delta-hemolysins respectively. No isolate harbored bap, pvl, or tsst. It was found that the bbp, cna and cap8 genes were detected only in isolates obtained from slaughterhouse and terminal markets. As shown in **Table 3**, the most frequent numbers of toxin genes per isolate were 11∼14 in all S. aureus isolates (**Table 3**). Notably, one isolate harbored 16 toxin genes and 5 isolates harbored 15 toxin genes, which were obtained from slaughterhouse (**Table 3**).

The average toxin gene number was also examined based on agr genotyping, and a higher average number of toxin genes was found in the agr-positive isolates compared to agr-negative isolates. The agr-positive isolates were associated with a high average number of toxin genes (averaging 13.2 for agr II, 12.6 for agr I, 12.6 for agr IV and 12.0 for agr III), whereas the agrnegative isolates were associated with a lower average number of toxin genes (averaging 9.9) (**Figure 2**). The distribution of virulence genes differed among the isolates according to the agr genotyping. Among the MSCRAMMs genes, the prevalence of 3 genes was significantly different between the agr-positive and agr-negative isolates: clfA (p < 0.01), clfB (p < 0.01) and fnbA (p < 0.05). The capsule multiple type (carriage of both capsule type 5 and 8) (p < 0.01) and icaC gene (p < 0.01) were positively associated with agr-positive isolates (**Figure 3**).

### Quantification of Biofilm Biomass and Growth Rate Analysis

Biofilm formation was analyzed, and all the isolates were able to form biofilm. The biomass of biofilms formed by most isolates increased continuously during incubation for 72 h at 37◦C (**Table 4**). Biofilm strong producers are mainly in slaughterhouse and biofilm biomass increase with time. No significant difference in the growth rates of the strong, moderate and weak biofilm formers was observed, indicating that the difference in biofilm formation was not due to the growth rate.

### Correlation Between Virulence Genes and Antibiotic Resistance in Biofilm Producing S. aureus

The relationship between prevalence of biofilm-associated genes and biofilm formation ability (incubation for 24 h at 37◦C) of S. aureus isolates was further analyzed (**Figures 4**, **5**). Considering the studied gene status, 19 different gene patterns were observed (**Table 5**). The most prevalent gene pattern was clfA-clfB-ebpS-eno-fib-cap5-icaA-icaC-icaD which was identified in 13 (10.0%) of 130 isolates. However, there was only one strong biofilm producer, nine moderate biofilm producers and three weak biofilm producers in this genes pattern. Conversely,

more significantly different in isolates from different food sources (p < 0.01).

Number of the toxin gene per isolate (n) Number of S. aureus isolates<sup>a</sup> Total number of isolates (130)<sup>b</sup> Swine farms (29) Slaughter house (62) Terminal markets (39) 16 0 (0) 1 (1.6%) 0 (0) 1 (0.8%) 15 0 (0) 5 (8.1%) 0 (0) 5 (3.8%) 14 1 (3.4%) 13 (21.0%) 5 (12.8%) 19 (14.6%) 13 14 (48.3%) 9 (14.5%) 10 (25.6%) 33 (25.4%) 12 13 (44.8%) 18 (29.0%) 14 (35.9%) 45 (34.6%) 11 1 (3.4%) 5 (8.1%) 6 (15.4%) 12 (9.2%) 10 0 (0) 3 (4.8%) 3 (7.7%) 6 (4.6%) 9 0 (0) 2 (3.2%) 0 (0) 2 (1.5%) 8 0 (0) 2 (3.2%) 0 (0) 2 (1.5%) 7 0 (0) 2 (3.2%) 0 (0) 2 (1.5%) 6 0 (0) 2 (3.2%) 1 (2.6%) 3 (2.3%)

TABLE 3 | The toxin genes number of S. aureus isolates from different stages of pork production.

<sup>a</sup>The number in parentheses represents the percentage of isolates with the corresponding number of toxin genes for all S. aureus isolates of the same part in pork production. <sup>b</sup>The number in parentheses represents the percentage of isolates with the corresponding number of toxin genes for all S. aureus isolates.

among the genes patterns of clfA-clfB-ebpS-eno-fib-fnbB-cap5 cap8-icaA-icaC-icaD (3.8%,5/130), clfB-eno-fib-fnbB-cap5-cap8 icaA-icaD (1.5%, 2/130), clfB-bbp-eno-fib-cap5-cap8-icaA-icaCicaD (1.5%, 2/130), clfA-clfB-eno-fib-fnbB-cap5-cap8-icaA-icaD (1.5%, 2/130), and clfB-eno-fib-cap5-cap8-icaA-icaC-icaD (1.5%, 2/130), all isolates showed strong biofilm formation ability (**Table 5**). A comparison between the strong, moderate, and weak biofilm producers in the isolates showed a significant difference in the prevalence of virulence genes among these isolates.

To determine whether biofilm formation was correlated with resistance to any particular antibiotic(s), we compared the biofilm forming capacities (incubation for 24 h at 37◦C) among isolates with different resistance profiles for the 14 antibiotics (**Table 6**). Resistance to ciprofloxacin, gentamicin,

tetracycline, clarithromycin, clindamycin and trimethoprimsulfamethoxazole were significantly higher in moderate biofilm producers and weak biofilm producers than in strong biofilm producers (**Table 6**). Notably, resistance to nitrofurantoin was only found in strong biofilm producers (7.1%, 4/56) and moderate biofilm producers (1.8%, 1/56) (**Table 6**). Resistance to penicillin, cefoxitin and chloramphenicol showed no significant difference among strong biofilm producers, moderate biofilm producers and weak biofilm producers (**Table 6**). Regarding multidrug resistance, no significant association to strong, moderate or weak biofilm producers was observed (**Table 7**).

### DISCUSSION

The agr (accessory gene regulator) system is a peptide quorumsensing system present in all the Staphylococci and a dominant regulator of pathogenesis and biofilm development in S. aureus (Boles and Horswill, 2008; Paharik and Horswill, 2016). All the swine farms isolates were agr type IV, whereas the slaughterhouse and terminal markets isolates indicated diverse agr types. In addition, isolates belonging to agr-positive group had a higher number of toxin genes than those belonging to agr-negative group (p < 0.05), suggesting that agr profiles may be associated with the virulence potential of S. aureus, which is consistent with a previous finding (Cheung et al., 2011). Raw meat isolates (belonging to agr I) exhibited a great ability to form strong biofilms than swine farms isolates (belonging to agr IV). Previous studies have shown that biofilm formation in S. aureus isolated

#### TABLE 4 | Biofilm phenotype of 130 S. aureus isolates at different time points.


<sup>a</sup>The number in parentheses represents the percentage of isolates with the corresponding number of biofilm phenotype for all S. aureus isolates of the same part in pork production. <sup>b</sup>Biofilm-forming ability was measured after 24, 48, and 72 h at 37◦C in terms of biofilm biomass by crystal violet staining. The results are presented by optical density (OD) determination of three independent repeats and compared to ATCC 25923 (biofilm-positive) and ATCC12228 (biofilm-negative).

from bovine mastitis with agr I is higher than those with other agr types (Bardiau et al., 2013; Bardiau et al., 2014; Khoramrooz et al., 2016).

The prevalence of virulence genes involved in biofilm formation and staphylococcal toxin genes were investigated. Most biofilm-producing isolates were positive for MSCRAMM, capsule type and ica group genes. The data show a significant association between the prevalence rate of MSCRAMM, capsule type and ica group genes among isolates producing weak, moderate and strong biofilms. Approximately 92.3% (120/130) of all isolates harbored icaA and icaD genes simultaneously, which were similar to those from previous studies (Szweda et al., 2012; Pereyra et al., 2016). Moreover, although both pvl and tst genes were not detected in the tested isolates, hemolysins and enterotoxin-producing genes (data not shown) were found. This suggests that these isolates exhibit pathogenic potential.

In the present study, all S. aureus isolates were biofilm producers. Biofilm formation is influenced by numerous factors, such as sugar content and concentration (glucose versus lactose), proteolytic enzymes and biofilm-associated genes, etc. (Coelho et al., 2008). In this study, biofilm production was higher for raw meat isolates compared to swine farms isolates. There was a difference in the prevalence of several genes involved in adhesion and biofilm production between raw meat and

FIGURE 5 | Diagram showing the antibiotic resistance pheno- and agr types, virulence genes profiles and biofilm phenotype of S. aureus isolated from different stages of pork production. The diagram was established on the basis of the presence and absence of selected determinants. For antibiotic resistance phenotype, black indicates resistance, gray indicates intermediate, and white indicates susceptible. CIP, ciprofloxacin; PEN, Penicillin; GEM, gentamicin; TET, tetracycline; CLR, clarithromycin; CHL, chloramphenicol; SXT, trimethoprim-sulfamethoxazole; NIT, nitrofurantoin; RIF, rifampicin; CLI, clindamycin; CEF, cephalothin; MIN, minocycline; OXA, oxacillin; FOX, cefoxitin; For virulence genes profiles, black indicates presence and white indicates absence. For biofilm phenotype, black indicates strong, gray indicates moderate, and white indicates weak.

TABLE 5 | The prevalence of biofilm related genes pattern and their associations with biofilm production in 130 S. aureus from different stages of pork production.


<sup>a</sup>Biofilm phenotype was measured after 24 h at 37◦C.

swine farms isolates. However, further studies are required to quantify the expression of relevant genes. Moreover, biofilm biomass increased proportionally as biofilms aged, which is accordance with previous findings (Akinbobola et al., 2017). High variability in biofilm biomass was found among isolates throughout the time course of biofilm formation (24 – 72 h), which is in accordance with previous findings (Marino et al., 2011; Va<sup>0</sup> zquez-Sa0nchez et al., 2014). Moreover, our study demonstrated the potential association between antibiotic resistance and biofilm-forming ability of S. aureus. Apart from resistance to penicillin, the high prevalence of resistance to ciprofloxacin, gentamicin, tetracycline, clarithromycin, clindamycin and trimethoprim-sulfamethoxazole were mainly observed in moderate and weak biofilm producers. Together,


#### TABLE 6 | Biofilm formation and antibiotic resistance pattern of 130 S. aureus isolates from different stages of pork production.

<sup>a</sup>The number in parentheses represents the corresponding number of biofilm phenotype for S. aureus isolates of antibiotic resistance. Biofilm phenotype was measured after 24 h at 37◦C, and the number of strong biofilm producers, moderate biofilm producers and weak biofilm producers were 56, 56 and 18, respectively.

TABLE 7 | Occurrence of multidrug resistant pattern and their associations with biofilm phenotype in 130 S. aureus from different stages of pork production.


<sup>a</sup>Biofilm phenotype was measured after 24 h at 37◦C.

Qi et al. (2016) reported that for Acinetobacter baumannii, there was a statistically negative correlation between antibiotic resistance and biofilm forming capacity, suggesting that biofilmforming strains are less dependent on antibiotic resistance than no biofilm-forming strains for survival. Previous studies have demonstrated that biofilm resistance to antimicrobials is multifaceted, including reduced penetration of the agent into biofilms due to the presence of extracellular matrix, biofilm heterogeneity and biofilm-specific phenotypes such as expression of efflux pump and persister cells (Stewart and Costerton, 2001; Akinbobola et al., 2017). Moreover biofilm resistance is known to vary from one microorganism to another (Mah and O'Toole, 2001). Thus our further study will focus on the enhancement in resistance of our Staphylococcus aureus after biofilm formation.

In summary, our study revealed agr type diversity, virulence potential, antibiotic multiresistance and high biofilm formation ability of S. aureus isolated from pork production. All swine farms isolates belonged to agr IV, whereas S. aureus isolated from slaughterhouse and retail indicated diverse agr types. Raw meat isolates (belonging to agr I) exhibited a great ability to form strong biofilms than swine farms isolates (belonging to agr IV). Most biofilm-producing isolates were positive for MSCRAMM, capsule type and ica group genes. The results illustrate a significant association between the prevalence rate of MSCRAMM, capsule type and ica group genes among isolates producing weak, moderate and strong biofilms. Clarifying these mechanisms could provide novel insights that would prevention against S. aureus biofilm-related infections.

### AUTHOR CONTRIBUTIONS

fmicb-09-01876 August 17, 2018 Time: 15:12 # 11

HY and LS participated in the design of this study. RC, DX, LS, and CL provided assistance for concepts, design, literature search, data acquisition, and manuscript preparation. YZ collected important background information, carried out the study, and performed the statistical analysis. HY and YZ drafted the manuscript. HY and DX performed the manuscript review. All the authors have read and approved the content of the manuscript.

### REFERENCES


### FUNDING

This work was supported by the National Key Basic Research Program (2016YFD0500606), the Science and Technology Planning Project of Guangdong Province, China (2014A020214001 and 2016A020219001), the Fundamental Research Funds for the Central Universities, SCUT (D2170320), and the Central Universities constructs the world first-class university (discipline) and Characteristic Development Guidance Special Fund (K5174960).



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

Copyright © 2018 Zhang, Xu, Shi, Cai, Li and Yan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Effect of Co-infection of Food-Borne Pathogenic Bacteria on the Progression of Campylobacter jejuni Infection in Mice

Gang Wang1,2, Yufeng He1,2, Xing Jin1,2, Yonghua Zhou<sup>3</sup> \*, Xiaohua Chen<sup>4</sup> , Jianxin Zhao1,2,5,6, Hao Zhang1,2,6,7 and Wei Chen1,2,7,8 \*

<sup>1</sup> State Key Laboratory of Food Science and Technology, Jiangnan University, Wuxi, China, <sup>2</sup> School of Food Science and Technology, Jiangnan University, Wuxi, China, <sup>3</sup> Key Laboratory of National Health and Family Planning Commission on Parasitic Disease Control and Prevention, Jiangsu Provincial Key Laboratory on Parasite and Vector Control Technology, Jiangsu Institute of Parasitic Diseases, Wuxi, China, <sup>4</sup> College of Life Sciences and Environment, Hengyang Normal University, Hengyang, China, <sup>5</sup> International Joint Research Laboratory for Probiotics, Jiangnan University, Wuxi, China, 6 Institute of Food Biotechnology, Jiangnan University, Yangzhou, China, <sup>7</sup> National Engineering Research Center for Functional Food, Jiangnan University, Wuxi, China, <sup>8</sup> Beijing Innovation Centre of Food Nutrition and Human Health, Beijing Technology and Business University, Beijing, China

#### Edited by:

Om V. Singh, Johns Hopkins University, United States

### Reviewed by:

Tu Anh Ngoc Huynh, University of Wisconsin–Madison, United States Ben Pascoe, University of Bath, United Kingdom

#### \*Correspondence:

Yonghua Zhou zhouyonghua@jipd.com Wei Chen chenwei66@jiangnan.edu.cn

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 29 March 2018 Accepted: 06 August 2018 Published: 22 August 2018

#### Citation:

Wang G, He Y, Jin X, Zhou Y, Chen X, Zhao J, Zhang H and Chen W (2018) The Effect of Co-infection of Food-Borne Pathogenic Bacteria on the Progression of Campylobacter jejuni Infection in Mice. Front. Microbiol. 9:1977. doi: 10.3389/fmicb.2018.01977 Campylobacter is a well-known food-borne pathogen that causes human gastroenteritis. Food products that contain Campylobacter may also be contaminated by other pathogens, however, whether this multiple contamination leads to more severe infection remains unclear. In this study, mice were gavaged with Campylobacter jejuni and other food-borne pathogenic bacteria to mimic a multiple infection. It was demonstrated that the C. jejuni load was elevated when the mice were co-infected with C. jejuni and Salmonella typhimurium, and the campylobacteriosis that followed was also enhanced, with features of decreased body weight, heavier bloody stools and more pronounced inflammatory changes to the colon. In addition, infection with C. jejuni was also promoted by co-infection with entero-invasive Escherichia coli but unaffected over time. In contrast to S. typhimurium and entero-invasive E. coli, co-infection by Listeria monocytogenes showed little effect on C. jejuni infection and even hindered its progress. In addition, the intestinal microecology was also affected by co-infection of C. jejuni with other pathogens, with an increased relative abundance of unclassified Enterobacteriaceae, decreased levels of butyric acid and changes in the abundance of several genera of gut microbe, which suggests that some food-borne pathogenic bacteria might affect the progression of C. jejuni infection in mice by influencing the composition of the gut microbiota and the resulting changes in SCFA levels. Collectively, our findings suggest that co-infection of Campylobacter with other pathogenic bacteria can impact on the progression of infection by C. jejuni in mice, which may also have implication for the etiology of Campylobacter on human health.

Keywords: Campylobacter, food-borne pathogenic bacteria, co-infection, campylobacteriosis, gut microbiota, butyric acid

### INTRODUCTION

fmicb-09-01977 August 21, 2018 Time: 8:17 # 2

Campylobacter is a food-borne pathogen and the leading cause of human gastroenteritis around the world. The incidence of campylobacteriosis has been estimated to be 4.4 per 1000 people, with 1.3 million cases in the United States (Scallan et al., 2011), 5.8 per 1000 person–years in the Netherlands (Havelaar et al., 2012), and 1.2 per 1000 person–years in China (Huang et al., 2018). Campylobacter infection usually results in symptoms such as bloody diarrhea, abdominal pain and fever, and the course of the disease is self-limited in most cases. However, an increased risk of irritable bowel syndrome and inflammatory bowel disease was observed amongst those infected with Campylobacter (Marshall et al., 2006). Moreover, some peripheral neuropathies such as Guillain–Barré syndrome and Miller Fisher syndrome are also long-term consequences of Campylobacter infection (Poropatich et al., 2010; Sejvar et al., 2011).

Campylobacter is most frequently found in poultry, but it can also be found in other food items such as pork, beef, and raw milk. The prevalence of Campylobacter contamination in retail poultry and by-products exceeds 50% around the world, varying from 0 to 100% (Sahin et al., 2015). In many developed countries, the rate of Campylobacter contamination usually exceeds 60%, but it can be much lower in developing countries such as China and Brazil (Silva et al., 2017; Zhu et al., 2017). Notably, when food items are contaminated with Campylobacter, they may also be contaminated with other foodborne pathogens. Many reports have found food samples that harbor more than one species of pathogen. In Ireland, 10 of 25 raw chicken samples were contaminated with multiple pathogens (Gorman et al., 2002). Another study also described the presence of more than two pathogens in ready-to-eat meat products (Gibbons et al., 2006). In addition, Campylobacter strains from different sources can also be detected in the same poultry house (Hiett et al., 2002). Due to the conventional intestinal microbiota, Campylobacter cannot effectively colonize mice sufficiently to exert obvious clinical symptoms (Dorrell and Wren, 2007). Acute ileitis induced by Toxoplasma gondii can abrogate the colonization resistance of mice, after which high loads of Campylobacter can be achieved (Haag et al., 2012a). However, it remains unclear whether other co-infection of pathogenic bacteria with Campylobacter can exacerbate the campylobacteriosis. To date, few studies have focused directly on campylobacteriosis caused by Campylobacter and other pathogenic bacteria.

In this study, mice were co-infected with Campylobacter jejuni and other pathogenic bacteria [Salmonella typhimurium, enteroinvasive Escherichia coli (EIEC) and Listeria monocytogenes] to determine whether this multiple infection would lead to more severe campylobacteriosis in mice. Moreover, the changes in the microbiota and the level of shortchain fatty acids (SCFAs) in feces were also monitored to determine if multiple infections also affect the intestinal microenvironment which may explain the potential mechanism by which co-infection with food-borne pathogenic bacteria influences the progression of C. jejuni infection in mice.

### MATERIALS AND METHODS

### Bacterial Strains and Culture Conditions

Campylobacter jejuni NCTC 11168 (ATCC 700819), S. typhimurium SL1344, EIEC (ATCC 43893), and L. monocytogenes (ATCC 19114) were acquired from the Culture Collection of Food Microorganisms of Jiangnan University (Wuxi, China). Columbia blood agar base plates (Oxoid, United Kingdom) supplemented with 5% sterile sheep blood and C. jejuni selective supplement (Oxoid) were used to culture C. jejuni strains under microaerophilic conditions (5% O2, 10% CO2, 85% N2) for 48 h at 37◦C. S. typhimurium, EIEC and L. monocytogenes were cultured with brain-heart infusion broth (Haibo, China) for 24 h at 37◦C.

### Animals and Experimental Design

Three-week-old female C57BL/6 mice were obtained from Shanghai Laboratory Animal Center (Shanghai, China) and used in all animal experiments. Six mice were housed in each cage, with a 12 h light–dark cycle in a controlled environment (temperature, 22 ± 2 ◦C; humidity, 50 ± 5%). All experimental procedures (#JIPD2017029) were approved by the Animal Care and Use Committee at Jiangsu Institute of Parasitic Diseases. All experiments conformed to the China Ministry of Science and Technology Guide for the Care and Use of Laboratory Animals.

Mice were infected with pathogenic bacteria or parasites by gavage at a volume of 0.2 mL. C. jejuni concentration was adjusted to 2 × 10<sup>9</sup> colony-forming units (CFU)/mL in sterile phosphate-buffered saline solution (PBS), and S. typhimurium, EIEC, L. monocytogenes were used at doses of 1 × 10<sup>5</sup> CFU/mL in sterile PBS. Cysts of the T. gondii ME49 strain (from Jiangsu Institute of Parasitic Diseases Remington, Wuxi, China) were obtained from the homogenized brains of mice infected with 10 cysts for 2 months, and the mice were infected perorally with 100 T. gondii cysts. Sterile PBS was used as a naive control. The mice were divided into 11 groups; the treatment of each group is shown in **Table 1**. Each mouse's body weight was recorded every 3 days, and stool samples were collected during the experimental procedure. The mice were anesthetised before sacrifice with an injection of 100 mg/kg body weight of ketamine, and plasma and colonic tissues were collected for further analysis.

### Detection of C. jejuni in Feces

Stool samples were collected from all mice and transported to the laboratory on an ice pack. The feces were resuspended in sterile PBS and serially diluted. Diluted samples were spread on Columbia blood agar with C. jejuni selective supplement and incubated under microaerobic conditions at 37◦C for 48 h. After incubation, C. jejuni numbers were determined by CFU.

### Bloody Stool Assay

A bloody stool detection kit (Jiancheng, China) was used to assess the presence of blood in fecal samples, which acted as a clinical sign of C. jejuni-induced infection. The detection kit was used immediately after the fecal samples were collected, and the

level of bloody stool was divided into four grades according to a previous study (Siegmund et al., 2001).

### Determination of Colon Histopathology

The colonic samples were fixed in 4% neutral buffered paraformaldehyde and embedded in paraffin. Sections (5 µm) were stained with haematoxylin and eosin for light microscopic examination (magnification, ×100). The degree of inflammation and damage was evaluated using a histopathological score system (Paclik et al., 2008), modified as follows:

Score 0: Histological findings identical to those of normal mice.

Score 1: Loss of goblet cells begins.

Score 2: Single isolated cell infiltrates within mucosa; loss of goblet cells begins.

Score 3: Mild scattered to diffuse cell infiltrates within mucosa and submucosa; loss of goblet cells begins.

Score 4: Mild scattered to diffuse cell infiltrates within mucosa and submucosa; loss of goblet cells.

Score 5: Severe inflammation; loss of goblet cells; loss of crypts.

Score 6: Severe inflammation; loss of goblet cells; extensive ulceration.

This analysis was performed blind by a pathologist.

### Cytokine Assay

Blood samples were collected and centrifuged (1200 × g, 15 min) to obtain serum. Before cytokine assays, serum samples were treated using a Milliplex MAP Kit (Merck, Germany) according to the manufacturer's instructions. A Luminex MAGPIX system (Luminex, United States) was used to detect the levels of cytokines in the treated serum samples.

### DNA Extraction, PCR, and 16S rDNA Sequencing

The fecal samples were stored at −80◦C before detection, and microbial genome DNA was extracted using a FastDNA Spin Kit for Soil (MP Biomedical, United States) following


the manufacturer's instructions. The V3–V4 region of the 16S rRNA gene was amplified by PCR. The products were separated in 1.5% (w/v) agarose gel, purified with a QIAquick Gel Extraction Kit (Qiagen, Germany) and quantified with a Quant-iT PicoGreen dsDNA Assay Kit (Life Technologies, United States). A TruSeq DNA LT Sample Preparation Kit (Illumina, United States) was used to establish libraries that were sequenced for 500 + 7 cycles on Illumina MiSeq using a MiSeq Reagent Kit. QIIME pipeline was used to analyze the sequence data of 16S rRNA as described previously (Wang et al., 2017).

### Short-Chain Fatty Acid Assay

Fecal samples stored at −80◦C were steeped in saturated NaCl solution for 30 min using sterile tubes and homogenized. Sulfuric acid (10%; 20 µL) was added to acidify the solution. Diethyl ether (800 µL) was added with an injection syringe to extract SCFAs. The tubes were centrifuged at 14,000 rpm for 15 min, and the supernatants collected. Anhydrous sodium sulfate was used to eliminate the remaining water, and the treated supernatants were analyzed using gas chromatography-mass spectrometry (GC-MS). The parameters for GC-MS in our experiment were established with reference to Sun et al. (2015).

### Statistical Analysis

Statistical analyses were performed using GraphPad Prism 5, and the data are expressed as mean ± SD. SPSS 20.0 (SPSS Inc., United States) was used for significance analysis. Comparisons between groups were made with a two-tailed Student's t-test, and a two-sided p-value of less than 0.05 was considered to indicate statistical significance.

### RESULTS

### Co-infection of S. typhimurium Elevated the C. jejuni Load in Mice

As shown in **Figure 1**, feces from all infected groups (C. jejuni, T. gondii+ C. jejuni, S. typhimurium + C. jejuni, EIEC + C. Jejuni, and L. monocytogenes + C. jejuni) were positive for C. jejuni, whilst C. jejuni was not detected in the mice from the control group (naive, T. gondii, S. typhimurium, EIEC, and L. monocytogenes). T. gondii infection led to significantly higher C. jejuni loads on day 9 than C. jejuni infection alone (P < 0.05). In addition, in the T. gondii coinfection group, more culturable C. jejuni was detected at day 9 than at day 7, and the number of viable bacteria reached 10<sup>9</sup> CFU per gram of feces. S. typhimurium infection also aggravated C. jejuni colonization on day 7 in comparison with C. jejuni alone (P < 0.05), which was more effective than T. gondii, with the number of C. jejuni reaching 10<sup>8</sup> CFU per gram of feces. On day 9, although C. jejuni loads in a few mice co-infected with S. typhimurium decreased, the loads remained mostly stable in other mice, where C. jejuni loads could reach 10<sup>8</sup> CFU per gram of feces. In contrast, coinfection with EIEC or L. monocytogenes did not significantly enhance the C. jejuni loads on days 7 or 9 (P > 0.05). Compared

with the overall increase of C. jejuni loads by T. gondii, the effects on C. jejuni colonization by three food-borne pathogens were distinct and pathogen specific. More individuals co-infected with S. typhimurium maintained the high C. jejuni loads. A few individuals co-infected with EIEC still had a higher C. jejuni load. However, the C. jejuni in most individuals co-infected with L. monocytogenes cleared, indicating the potential negative impact of L. monocytogenes on C. jejuni colonization.

## S. typhimurium Decreased the Body Weight of Mice When Co-infected With C. jejuni

Severe pathogenic infection can cause intestinal inflammation in mice, followed by a decrease in body weight. **Figure 2** shows that no significant decrease occurred in the body weights of the mice in the C. jejuni infection group. Consistent with previous reports (Grainger et al., 2013), the mice in the T. gondii only group and those in the T. gondii+ C. jejuni group showed significant decreases in body weight compared with the naive control group (P < 0.05), indicating that T. gondii infection seriously damaged their health status. Infection with three food-borne pathogens alone caused little change in the mice's body weights. However, after co-infection with C. jejuni, the mice with S. typhimurium showed significant decreases in body weight compared with healthy mice; mice in the EIEC+ C. jejuni group only showed significant weight loss (P < 0.05) on days 5 and 7; mice in the L. monocytogenes and L. monocytogenes + C. jejuni groups only showed slight weight loss on day 5, but this effect completely reversed by day 9. It is noteworthy that, unlike the overall decrease in body weight in the T. gondii+ C. jejuni group, the decrease in mice body weight in the S. typhimurium + C. jejuni group showed obvious individual differences. Similar differences can also be seen in body weight gain (**Supplementary Figure S1** and **Supplementary Table S1**). On day 5, except for the mice in the C. jejuni group, all of the mice in the pathogen infection groups showed significantly reduced weight gain. The rate of weight gain recovered in all of the groups infected or co-infected with EIEC or L. monocytogenes. Although there were no significant differences between the weight gain of the S. typhimurium infected (and co-infected) groups and that of the naive group on day 9, obvious individual differences in the weight gain of the mice within this group were observed. The weight gain of each mouse corresponded to the C. jejuni load. In addition, several mice died in the

S. typhimurium + C. jejuni, and in the T. gondii+ C. jejuni groups.

### Co-infection With S. typhimurium and C. jejuni Promoted Bloody Stools in Mice

Mice infected with C. jejuni had mild blood-positive feces by day 7, and the symptoms were partly relieved by day 9 (**Figure 3**). In T. gondii+ C. jejuni and T. gondii only groups, up to 50% of infant mice presented severe blood-positive (Grade 2) feces samples by days 7 and 9. In general, bloody stools were significantly more frequent in all co-infected groups than in groups infected with a single pathogen (either C. jejuni or another pathogenic bacteria), except that the bloody stools in the L. monocytogenes + C. jejuni group were less frequent than those in the L. monocytogenes group by day 9. Severe blood-positive (Grade 2) fecal samples were also observed in the S. typhimurium co-infection group (days 7 and 9) and in the EIEC coinfection group (day 9). It is noteworthy that although the frequency of bloody stools in the S. typhimurium only group was similar to or even lower than that in the EIEC and L. monocytogenes groups, the bloody stools in the S. typhimurium + C. jejuni group were more serious than those in all of the other coinfected groups, whilst the frequency of bloody stool caused by S. typhimurium infection was reduced to a relatively low level between 7 and 9 day.

### C. jejuni Relieved the Damage to Colonic Tissue Caused by Co-infected Food-Borne Pathogenic Bacteria

After the mice were killed (day 11), colon samples were collected from each group for histological analysis. As shown in **Figure 4A**, the mice from the naive group displayed normal villi, crypts, and muscular layer with a large number of goblet cells. Mice infected with C. jejuni alone had fewer goblet cells, whilst the structure of villi, crypts, and muscular layer was normal (**Figure 4B**). Infection with T. gondii alone resulted in severe colonic pathological changes, including a loss of goblet cells, damage to the crypt and villi architecture and inflammatory cell infiltration (**Figure 4D**), indicating that T. gondii infection induced serious inflammation and tissue injury. However, the histopathological scores in **Figure 4K** indicate that co-infected C. jejuni partly relieved the damage to the colonic tissue. S. typhimurium infection alone caused the highest histopathological scores in the groups of food-borne pathogenic bacteria infection alone, whilst EIEC caused the lowest scores, similar to that of C. jejuni (**Figures 4F,H,J**). It is noteworthy that, like the T. gondii+ C. jejuni group (**Figure 4C**), co-infected C. jejuni partly relieved the damage to the colonic tissue caused by S. typhimurium and L. monocytogenes, whilst it had no effects on the damage caused by EIEC (**Figures 4E,G,I**).

## Effects of Co-infected Food-Borne Pathogenic Bacteria on Inflammation

The concentration of inflammatory factors in blood serum was also investigated. As shown in **Figure 5**, infection with C. jejuni did not induce any of the inflammatory cytokines. However, the concentrations of interferon (IFN) γ, tumour necrosis factor (TNF) α, interleukin (IL) 6, and IL-10 were significantly elevated due to infection with T. gondii alone, whilst the levels of IL-1α were down-regulated compared with other groups. In the T. gondii+ C. jejuni group, the concentrations of IFN-γ and IL-6 were higher than those in the T. gondii only group, whilst the concentrations of TNF-α, IL-10, and IL-1α were lower than those in the T. gondii group. As in the T. gondii+ C. jejuni group, the levels of IFN-γ, TNF-α, and IL-6 were also increased in some mice co-infected with C. jejuni in the EIEC + C. jejuni group, although no changes in any of the mice in the EIEC group. Moreover, IFN-γ, TNF-α, IL-10, and IL-6 in some mice in the S. typhimurium and S. typhimurium + C. jejuni groups also showed slight increases, although these were not significant. No significant variation could be detected in other cytokines, such as IL-1β, IL-2, IL-4, IL-12, and IL-17 (data not shown).

### Co-infection of Food-Borne Pathogenic Bacteria Alter the Composition of SCFAs in Feces

The contents of acetic acid, propionic acid and butyric acid in feces were analyzed by GC-MS to evaluate the metabolism of the intestinal microbiota. **Figure 6** shows that C. jejuni infection alone did not cause significant changes in the composition of SCFAs in mouse feces compared to that in the naive group. In contrast, infection by EIEC or L. monocytogenes alone resulted in distinct decreases in the levels of acetic acid, propionic acid and butyric acid in mouse feces (P < 0.05). Interestingly, co-infection with C. jejuni led to recovery of the SCFA level in the EIEC group but showed no effects on the SCFA level in the L. monocytogenes group, indicating the specific alteration of microbial metabolism. Infection with T. gondii or S. typhimurium alone only decreased the level of butyric acid (P < 0.05). Co-infection with C. jejuni showed no significant effects on the SCFAs level influenced by T. gondii or S. typhimurium except for a further decrease in the acetic acid level in the S. typhimurium + C. jejuni group (P < 0.05).

### Co-infection of Food-Borne Pathogenic Bacteria Alter the Diversity and Relative Abundance of the Gut Microbiota

The estimated richness (Chao-1) and Shannon index were used to evaluate the community diversity of each sample. **Supplementary Figure S2** shows that infection by C. jejuni alone seemed to cause no obvious differences in the diversity of gut microbiota. Infection by T. gondii, S. typhimurium, EIEC, or L. monocytogenes alone caused significant decreases in gut microbiota diversity. In addition, there was a tendency toward further reduction in the diversity of the gut microbiota in mice co-infected with C. jejuni, although individual differences existed between different mice. At phylum level (**Figure 7**), Firmicutes, Bacteroidetes, and Proteobacteria were dominant amongst the experimental groups. Tenericutes, Actinobacteria, TM7, Verrucomicrobia, and Cyanobacteria were also found in the feces of mice with a relative abundance of less than 1%. Mice co-infected with T. gondii and C. jejuni and those infected

with EIEC alone showed a decrease in the relative abundance of Firmicutes compared with the naive group (P < 0.05). Each experimental group had between 20 and 40% Bacteroidetes in feces; no statistical differences were seen (P > 0.05). Notably, infection by different pathogens induced evident changes in Proteobacteria. The relative abundance of Proteobacteria in the naive control group was less than 1% but increased significantly in the T. gondii + C. jejuni, the S. typhimurium + C. jejuni, the S. typhimurium, EIEC, and the L. monocytogenes groups. The abundances of Proteobacteria in the T. gondii + C. jejuni, the S. typhimurium + C. jejuni, the S. typhimurium and the EIEC groups were positively correlated with the corresponding pathogen loads in the intestine. Although L. monocytogenes is not a member of the Proteobacteria, it significantly increased the abundance of Proteobacteria in the gut. Interestingly, although EIEC and L. monocytogenes increased the abundance of Proteobacteria, the presence of C. jejuni significantly decreased the abundance of this phylum, indicating an acceleration in the

elimination of pathogenic bacteria from the host in the late period of co-infection.

The microbial composition was further analyzed at the genus level, to further explore the differences observed in the Proteobacteria phylum. **Figure 8A** shows all genera with a relative abundance of more than 1%. Unclassified genera within Enterobacteriaceae is the only genus in the genera listed in **Figure 8A** that belongs to Proteobacteria phylum. The relative abundance of unclassified Enterobacteriaceae in the experimental groups was then checked, and significant increases in unclassified Enterobacteriaceae were found in the T. gondii + C. jejuni, S. typhimurium + C. jejuni, the S. typhimurium, the EIEC, and the L. monocytogenes groups (P < 0.05), which exhibited the same tendency as Proteobacteria phylum (**Figure 8B**). This indicates that all four food-borne pathogenic bacteria can increase the abundance of unclassified Enterobacteriaceae. In addition to changes in the levels of unclassified Enterobacteriaceae, decreases in the abundance of unclassified Clostridiales and Lachnospiraceae corresponded to exposure of pathogenic bacteria, except no significant reduction

significantly from the naive group (P < 0.05).

in abundance was observed in the group infected with C. jejuni alone (**Supplementary Figure S3**). The abundance of these two genera also showed positive correlations with the diversity of the gut microbiota and the SCFA level. In addition, although low C. jejuni loads in the C. jejuni group caused no significant changes in the different indexes, the abundances of Bacteroides and Lactobacillus were significantly changed by infection with C. jejuni alone. The abundance of Turicibacter was also relatively high in the two co-infection groups with high C. jejuni loads. Moreover, although large individual differences were observed, the abundances of Dorea and unclassified S24-7 showed significant increases only in some of the mice in the EIEC + C. jejuni group, and were highly correlated with the recovery of SCFA level in this group.

## DISCUSSION

Although many efforts have been made to prevent C. jejuni infections, it remains the most common food-borne pathogen from a global perspective (Haagsma et al., 2013). Humans can become infected with C. jejuni via contaminated water and food and by direct contact (Domingues et al., 2012). However, these pathways are compatible for other food-borne pathogenic bacteria as well, and the shared pathways lead to cross contamination. Contamination by multiple pathogenic bacteria, including different strains of Campylobacter from different sources (Konkel et al., 2007; Atterby et al., 2018), may cause symptoms that are more complex and serious than those caused by infection by only one pathogen. This study was performed to investigate whether co-infection with other common pathogenic bacteria affects the symptoms caused by C. jejuni. The results of our study show that co-infection with S. typhimurium significantly increased the C. jejuni load and resulted in more severe symptoms of C. jejuni infection. EIEC promoted infection by C. jejuni to some extent, but the promotion effects subsided with time. Co-infection with L. monocytogenes had no effects on C. jejuni load and even showed some reduction. Moreover, during co-infection with pathogenic bacteria, variations in the abundance of gut microbes such as unclassified Enterobacteriaceae, Clostridiales, and Lachnospiraceae, and corresponding changes in the level of SCFAs in the gut were observed (**Supplementary Tables S2–S6**).

Campylobacter jejuni has the disadvantage of sporadic colonization and barely triggers disease-defining clinical manifestations in mice due to their well established robust intestinal microbiota. In this study, 3-week-old female C57BL/6 mice with reported susceptibility to C. jejuni were used (Field et al., 1981). However, C. jejuni colonization levels remained low and even decreased beyond the infection day, which is consistent with the colonization resistance reported previously (Heimesaat et al., 2013). In addition, no significant pathological symptoms were found with C. jejuni alone, which may be related to the low colonization by C. jejuni. We therefore used T. gondii infection model (Liesenfeld, 2002; Haag et al., 2012b), that make the host more susceptible to C. jejuni infection as a positive control to assess the impact of food-borne pathogens on C. jejuni infections.

Because infection with high doses of S. typhimurium, EIEC, and L. monocytogenes can result in loss of weight, enteritis, and death in a mouse model (Miller and Burns, 1970; Medici et al., 2005; Martins et al., 2013; Franca et al., 2015), these food-borne pathogens are expected to promote the infection of C. jejuni. However, our study showed that only co-infection with S. typhimurium can promote C. jejuni colonization, and induce weight loss and bloody stools, even with a low infectious dose (3 × 10<sup>4</sup> CFU/mouse) of S. typhimurium. Bloody stools and weight loss can be caused by C. jejuni alone to some extent (Haag et al., 2012a; Heimesaat et al., 2014). In the presence of S. typhimurium, C. jejuni was found to promote weight loss, bloody stools and even death of the mice. These results suggest that the presence of S. typhimurium can enhance C. jejuni numbers and exacerbate the corresponding symptoms. Although no significant difference in weight loss was found between the EIEC + C. jejuni group and the EIEC group, the bloody stools in the EIEC + C. jejuni group suggest that the presence of EIEC may promote C. jejuni infection to a limited extent. L. monocytogenes is a poor colonizer of the mouse intestine due to lack of recognition by mouse intestinal cell receptors (Lecuit et al., 1999) and germ free mice (Manohar et al., 2001) and mice pre-treated with streptomycin can increase the pathogen load (Takeuchi et al., 2006). However, even low-dose L. monocytogenes gavage still caused damage to the gut health of the mice. Unexpectedly, it seems that the symptoms in the L. monocytogenes + C. jejuni group were milder than those in the L. monocytogenes group. Taking the C. jejuni loads into consideration, it seems that the co-existence of these two pathogenic bacteria impedes each other's infection process. Corresponding with the above results, histological analysis show that the mice infected with C. jejuni alone displayed mild pathological changes in the colon, likely due to sporadic colonization of C. jejuni. However, different to the results of weight loss and bloody stool, except for the EIEC group, it seems that the co-infected C. jejuni instead reduces the

degree of intestinal tissue lesions. This finding may be related to the colonization area of the pathogens, and further investigation is needed.

Although the major site of T. gondii replication is different from that of the other four food-borne pathogenic bacteria (Dubey et al., 1997), T. gondii promoted infection with C. jejuni in mice. This indicates that the synergistic effect may not be confined to the site of infection but rather a systemic effect. T. gondii infection leads to Th1-type immunopathology in mice, which causes an elevated IFN-γ concentration (Heimesaat et al., 2007; Zhou and Wang, 2017). The effects of one pathogen on the host's immune system may be the cause of the host's susceptibility to or tolerance of other pathogens. IL-10−/− mice that show heavier pathogen loads and more severe clinical signs are usually used to establish models of C. jejuni infection (Bereswill et al., 2011; Sun et al., 2012). This indicates that changes in the host's immune status may have an important impact in C. jejuni infection. It has been reported that TNFα, IL-1, IL-4, and IL-10 are all elevated in mice after C. jejuni infection (Al-Banna et al., 2012). In this study, co-infection of T. gondii and C. jejuni further increased the levels of IFN-γ and IL-6 caused by T. gondii, much in the same way as the changes in cytokines induced by C. jejuni in gnotobiotic mice (Bereswill et al., 2011), which suggests that T. gondii infection promotes C. jejuni infection symptoms. In addition, from the level of cytokines, it seemed that EIEC does promote C. jejuni infection to a certain extent, whilst L. monocytogenes infection without any promotion on C. jejuni infection, which is also confirmed by the disease indicators described above. However, unexpectedly, although S. typhimurium can promote C. jejuni infection and its related symptoms in mice from pathogen loads and disease index, and S. typhimurium may activate local inflammatory processes in the colon with elevated levels of IFN-γ, TNF-α, and IL-6 (Castillo et al., 2011), S. typhimurium infection, whether alone or together with C. jejuni, did not result in statistical changes in cytokine levels in this study, despite slight increases in IFN-γ and IL-6 in a few mice. This indicates that infection promotion on C. jejuni by S. typhimurium differs somewhat from that by T. gondii.

Studies have shown that in addition to the immune system, the colonization of C. jejuni in the intestine is closely related to the gut microbiota (Bereswill et al., 2011). Even in the C57/6J mice that were proven to be less susceptible to C. jejuni colonization, a small number of individuals with high C. jejuni loads also showed significant differences in the composition of their caecal microbiota. C. jejuni colonization did not incite visible pathologic changes, but was associated with increased abundances of Coriobacteriaceae, Lachnospiraceae, and Ruminococcaceae (Lone et al., 2013). This suggests that the gut microbiota might play an important role in determining the extent of which C. jejuni can colonize the mice gut. An investigation in humans also indicated that low diversity of gut microbiota may result in C. jejuni infection (Kampmann et al., 2016). In addition, infection with S. typhimurium, EIEC and other pathogens can lead to a decline in the diversity of the gut microbiota and a decrease in the abundance of some intestinal microbes (Chassaing et al., 2014; Mon et al., 2015; Borton et al., 2017; Zhang et al., 2017; Du et al., 2018). In this study, because of the low load of C. jejuni, no significant change in the diversity of gut microbiota was found in the group infected with only C. jejuni. However, all other groups treated with pathogens showed significant decreases in the diversity of the gut microbiota. Co-infection with C. jejuni generally resulted in a tendency toward further reduction of the diversity of the gut microbiota, suggesting that infection with some food-borne pathogens may increase the colonization rate of C. jejuni by decreasing the diversity of gut microbiota.

Firmicutes and Bacteroidetes are the two dominant phyla in the murine intestinal microbiota, and the abundance of these phyla is related to host health (Hansen et al., 2010). In this study, the two groups whose Firmicutes/Bacteroidetes ratio decreased also showed an increase in the relative abundance of Proteobacteria, a phylum including a wide variety of pathogens, such as Escherichia, Salmonella, Vibrio, Helicobacter, Yersinia, Legionella, and many other notable genera (Madigan et al., 2018). In addition, although no significant decrease was seen in the level of Firmicutes, the S. typhimurium, S. typhimurium+ C. Jejuni, and L. monocytogenes groups also showed significant increases in Proteobacteria. This might reflect the effective colonization of this three pathogen in the mice. Unexpectedly, the increase in Proteobacteria abundance caused by infection with EIEC or L. monocytogenes disappeared due to the presence of C. jejuni. This was further confirmed by the variation in abundance of unclassified Enterobacteriaceae at the genus level. Infection with some pathogens has been reported to increase the level of Enterobacteriaceae in the gut (Corridoni et al., 2012; David et al., 2014). The lack of change in unclassified Enterobacteriaceae abundance in the T. gondii group but significant increase in the T. gondii+ C. jejuni group in this study suggests that the increase in unclassified Enterobacteriaceae is a manifestation of C. jejuni infection (Sakaridis et al., 2018). The increased abundance of unclassified Enterobacteriaceae may also predicts the possibility of an elevated rate of Salmonella colonization in association with C. jejuni infection. However, for EIEC and L. monocytogenes, there appears to be an antagonistic relationship that makes C. jejuni and these two pathogens unable to co-exist at high levels in the host's gut. This may be due to the different effects of these bacteria on the immune system, or differential effects on the composition of the gut microbiota.

Changes in the abundances of unclassified Clostridiales and Lachnospiraceae also showed similar patterns with respect to gut microbiota diversity. As the Clostridiales includes a portion of the intestinal bacteria producing butyric acid (Liu et al., 2010; Myszka et al., 2012; Cai et al., 2013), these decreases can also explain the general decline of intestinal butyric acid levels in infected mice. It has been reported that butyric acid has potent antiinflammatory effects and can efficiently maintain the integrity of the intestinal mucosa (Mishiro et al., 2013). Therefore, infection with these pathogenic bacteria may cause intestinal damage by affecting the abundance of SCFA-producing bacteria in the intestinal tract. The decrease in unclassified Lachnospiraceae observed in this study was also consistent with previous studies (Kampmann et al., 2016), but no correlations of Coprococcus and Dorea abundances were found in our study. Furthermore, similar to Dorea, unclassified S24-7 also showed a significant

increase in only some mice in the EIEC + C. jejuni group, consistent with the recovery of SCFAs in this group. Both of these genera have been reported to produce acetic, propionic and butyric acids (Taras et al., 2002; Ormerod et al., 2016; Meisel et al., 2017; Bishehsari et al., 2018). Therefore, the specific regulation of gut microbiota by EIEC and C. jejuni, resulting in the restoration of SCFA levels may also be a cause of infection remission. In addition, similar to the findings of a previous study (Borewicz et al., 2015), the levels of Turicibacter in this study also increased due to S. typhimurium or T. gondii infection, which correlates with the promotion of C. jejuni colonization by these two pathogens. It has been reported there was a significantly lower proportion of Turicibacter in Tnf-/- compared to WT mice both prior to and after colitis induction (Jones-Hall et al., 2015). According to the levels of TNF-α in this study, the changes in cytokine and Turicibacter abundance showed a certain degree of coincidence. Whether the increase in Turicibacter abundance and TNF-α levels contribute to colonization by C. jejuni still needs further study. Besides, the changes in the abundances of Parabacteroides and Lactobacillusshowed similar tendency in this study. Taking the effects on the host by these genera into account (Sanchez et al., 2010; Clarke et al., 2011), these changes appear to be a self-protection by the host to alleviate further infections and injuries. However, this protection seems to be limited because the abundances of these genera decrease in seriously coinfected mice. Therefore, changes in the gut microbiota caused by pathogens would further affect aspects of the gut environment such as metabolites, nutrients, and immune factors, influencing the infection progression of subsequent pathogens.

### CONCLUSION

This study demonstrates that different food-borne pathogenic bacteria co-infected with C. jejuni exert different effects on the progress of C. jejuni infection in mice. Co-infection with S. typhimurium can significantly increase the C. jejuni burden in mice and lead to more severe campylobacteriosis. Moreover, co-infection with EIEC promotes infection by C. jejuni to some extent, but this promotion disappears over time. In contrast, co-infection with L. monocytogenes has little effect on C. jejuni infection and even hinders its progress. In addition, an increase

### REFERENCES


in the relative abundance of Enterobacteriaceae and a decreased level of butyric acid were also observed during co-infection of C. jejuni with other pathogenic bacteria. Changes in the abundance of some intestinal microbes may be directly related to the progression of C. jejuni infection and they might thus be used as indicators of C. jejuni infection. Given the possibility of co-infection, it is clear from this study that some foodborne pathogenic bacteria might play an important role in the progression of C. jejuni infection.

### AUTHOR CONTRIBUTIONS

GW, YZ, and WC conceived and designed the experiments. GW, YH, XJ, and XC performed the experiments. JZ and HZ analyzed the data. YZ, JZ, HZ, and WC contributed reagents, materials, and the analysis tools. GW and YH wrote the paper. All authors contributed to manuscript revision, read, and approved the submitted version.

### FUNDING

This work was supported by the National Natural Science Foundation of China (No. 31671839), the National Natural Science Foundation of China Key Program (No. 31530056), the National Natural Science Foundation of China (Nos. 31601444 and 31301407), the Fundamental Research Funds for the Central Universities (JUSRP51501), a project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions, the national first-class discipline program of Food Science and Technology (JUFSTR20180102), the Program of Collaborative Innovation Center of Food Safety and Quality Control in Jiangsu Province, and the project/research supported by Scientific Research Fund of Hunan Provincial Education Department (15B034).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.01977/full#supplementary-material

promotes SCFA production, and suppresses colon carcinogenesis. Genes 9:E102. doi: 10.3390/genes9020102


serovar Typhimurium infection in mice. BMC Microbiol. 11:177. doi: 10.1186/ 1471-2180-11-177


butyrate deficiency associated with increased susceptibility to colitis. ISME J. 11, 15–30. doi: 10.1038/ismej.2016.114


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

Copyright © 2018 Wang, He, Jin, Zhou, Chen, Zhao, 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.

# Phylogenetic Relatedness Among Plasmids Harbored by Campylobacter jejuni and Campylobacter coli Isolated From Retail Meats

### Daya Marasini, Anand B. Karki, Mark A. Buchheim and Mohamed K. Fakhr\*

Department of Biological Science, The University of Tulsa, Tulsa, OK, United States

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Santiago Castillo Ramírez, Universidad Nacional Autónoma de México, Mexico Andrea Zuccolo, Scuola Sant'Anna di Studi Avanzati, Italy Beatrix Stessl, Veterinärmedizinische Universität Wien, Austria

> \*Correspondence: Mohamed K. Fakhr mohamed-fakhr@utulsa.edu

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 25 May 2018 Accepted: 23 August 2018 Published: 12 September 2018

#### Citation:

Marasini D, Karki AB, Buchheim MA and Fakhr MK (2018) Phylogenetic Relatedness Among Plasmids Harbored by Campylobacter jejuni and Campylobacter coli Isolated From Retail Meats. Front. Microbiol. 9:2167. doi: 10.3389/fmicb.2018.02167 Campylobacter jejuni and Campylobacter coli are two of the major causes of foodborne illness. In this study, 29 plasmids isolated from 20 retail meat isolates of Campylobacter jejuni and Campylobacter coli were fully-sequenced individually or as a part of a whole genome sequencing approach. The fully-sequenced plasmids ranged in size from 3 to 119 kb. Molecular characterization of the sequenced plasmids was based on pangenomic analysis and types of genes present on these plasmids and similar ones from GenBank. The plasmids were categorized into four different groups. These groups include type-1 that consisted mainly of pTet plasmids with the tetO gene, type-2 plasmids commonly found in C. coli strains, type-3 which has pVir plasmids, and type-4 that consisted mainly of smaller plasmids. The type-2 plasmids were unique, common among C. coli strains, and carried several conjugative transfer genes. The type-2 plasmids were most similar to a plasmid from Helicobacter pullorum. Maximum parsimony analysis and NeighborNet analysis were used to assess the phylogenetic relatedness among the 29 plasmid sequences presented in this study in addition to the other 104 plasmid sequences of Campylobacter species available in GenBank to date. Results from MP analysis revealed multiple lineages among Campylobacter plasmids which was supported by NeighborNet analysis. Clustering of plasmids did not conform to species-specific clades which suggested an intra-species dissemination of plasmids among Campylobacter species. To our knowledge, this is the first extensive phylogenetic analysis of Campylobacter plasmids sequenced to date.

Keywords: Campylobacter, plasmids, phylogenetic relatedness, retail meats, next generation sequencing

### INTRODUCTION

Foodborne bacterial illness caused by Campylobacter spp. in the United States ranks third after Salmonella spp. and Clostridium perfringens (Scallan et al., 2011). Most of the foodborne illnesses associated with Campylobacter spp. have been related to Campylobacter jejuni; whereas, the remaining have been attributed to Campylobacter coli (Acheson and Allos, 2001; Gillespie et al., 2002). Species from genus Campylobacter are known to have gained various types of antibiotic resistance, mostly tetracycline and aminoglycosides, followed by chloramphenicol (Taylor, 1986; Tenover et al., 1992). The majority of antibiotic resistance in bacteria is associated with plasmids. While several studies on plasmids of other foodborne pathogens like Escherichia coli and Salmonella spp. are available (Williams et al., 2013), only a few reports focused on plasmids of C. jejuni and C. coli. Various types of plasmids and their incompatibility groups were reported in other foodborne pathogens, but little is known about Campylobacter plasmids. The replicon typing and the RFLP analysis of the reference collection of ECOR, SARA, SARB and SARC (E. coli and Salmonella) plasmids showed unique RFLP patterns indicating variation among the plasmids of size greater than 30 kb (Williams et al., 2013). The IncX group of plasmids, which encode Type IV fimbriae in the Enterobactericeae, has also been expanded to four subtypes according to comparisons based on phylogenetic analysis (Johnson et al., 2012). Of the few plasmids studied in C. jejuni a majority (53%) had the tetracycline resistance gene, tetO (Schmidt-Ott et al., 2005). Approximately 29% of C. jejuni isolates obtained from bloody diarrhea samples contained plasmids that conferred tetracycline resistance (pTet) as well as virulence (pVir) plasmids (Schmidt-Ott et al., 2005). Next Generation Sequencing (NGS) technology has led to the characterization of a number of megaplasmids (up to 180.5 kb) of C. jejuni and C. coli isolated from various sources and bearing a spectrum of interesting genes such as the type VI secretion system(Gunther et al., 2016; Marasini, 2016; Marasini and Fakhr, 2016a,b,c, 2017a,b,c).

The tetO gene present in the most frequently encountered pTet plasmids was likely acting to maintain Campylobacter plasticity (Friis et al., 2007). Nucleotide sequence comparison of two tetracycline resistant plasmids of 45.2 and 44.7 kb in size showed the presence of the tetO gene, methylase and various homologous hypothetical genes present in both plasmids (Friis et al., 2007). These plasmids also contained various replication-associated and conjugation-associated genes that showed homology with a plasmid from Actinobacillus actinomycetemcomitans (Batchelor et al., 2004). The tetracycline


\*Plasmids sequenced in this study.

resistance gene tetO present in the C. jejuni plasmid of 45 kb in size showed a significant similarity to the tetM tetracycline resistance gene of Streptococcus spp., indicating the possible interchange of genetic information between these bacteria (Taylor, 1986). In most of the studies, tetO has been found to be located on plasmids; whereas, in other studies it was chromosomally located in both C. jejuni and C. coli (Pratt and Korolik, 2005; Marasini and Fakhr, 2016a,b,c, 2017a,b,c). The tetO determinant was found in the chromosome as a part of a transposon gene cassette in isolates of C. coli derived from turkey and swine (Pratt and Korolik, 2005). Some of the pTet plasmids are also known to contain the aminoglycoside phosphotransferase gene aphA-3 and aphA-7 kanamycin resistant determinants (Tenover et al., 1992; Crespo et al., 2016; Marasini and Fakhr, 2016b). Studies of C. jejuni and C. coli showed that the aphA-7 gene was also present in the smaller plasmids of 11.5 and 9.2 kb in size (Tenover et al., 1992). The C. jejuni strain 81-176 was found to contain a pVir plasmid encoding genes homologous to type IV secretion system found in Helicobacter pylori (Bacon et al., 2000). This pVir plasmid was thought to be associated with bloody diarrhea but the connection could not be confirmed (Louwen et al., 2006).

Besides these two major types of plasmids (pTet and pVir), various other cryptic plasmids have been identified and fully sequenced (Jesse et al., 2006; Miller et al., 2007). The plasmid pTIW96 from a wild bird isolate of C. jejuni was 3,860 bp in size with 5 ORFs. Two of the ORFs of this plasmid were similar to pCC2228-2 (Hiett et al., 2013) found in another C. coli plasmid. Sequence analysis of the two cryptic plasmids of an agricultural isolate of C. coli showed one of the plasmids contained an ORF with homology to a plasmid from C. upsaliensis (Jesse et al., 2006). To date, 127 Campylobacter plasmids have been completely sequenced and deposited in GenBank. Most of these are small plasmids that were isolated from clinical sources. The current study aimed to determine the DNA sequences and provide a molecular characterization of numerous plasmids from C. jejuni and C. coli strains isolated from retail meat sources. Phylogenetic relatedness among 29 different Campylobacter plasmids, ranging from 3 to 119 kb, and those available in the GenBank were also investigated to identify possible plasmid lineages and origins.

## MATERIALS AND METHODS

### Campylobacter Strains Used for Plasmid Isolation

A total of 29 plasmids from Campylobacter jejuni (19) and Campylobacter coli (10) were characterized (**Table 1**). However, 23 of these plasmid sequences were previously announced as part of whole genome sequences of Campylobacter strains (Marasini and Fakhr, 2016a,b,c, 2017a,b,c), and the remaining 6 plasmids were fully sequenced in this study and deposited in GenBank (**Table 1**). These plasmids were harbored by 20 Campylobacter isolates previously isolated from various retail meat samples in Tulsa Oklahoma (Noormohamed and Fakhr, 2012, 2013, 2014). The selection of the bacterial isolates for this plasmid study was based on the restriction pattern analysis and PFGE screening of megaplasmids detected in a previous study (Marasini and Fakhr, 2014).

### Plasmid Isolation and Sequencing

Whole genomic DNA and plasmid isolation from Campylobacter strains, sequencing in Illumina Miseq platform and sequence assembly process in CLC workbench version 7.5.1 have been TABLE 2 | List of all plasmids found in C. jejuni and C. coli strains from our laboratory and GenBank used for core genome and pangenome analysis.


(Continued)


#### TABLE 2 | Continued

The pCCDM105L, pCCDM18M, pCJDM67S, and pCJDM204S sequences were used as reference for blast analysis for type 1, type 2, type 3 and type 4 plasmids respectively. Type 1, type 2, type 3, and type 4 plasmids are highlighted in gray, blue, green, and red respectively.\*plasmids isolated and sequenced in our laboratory.

described previously (Marasini and Fakhr, 2016a,b,c, 2017a,b,c). Briefly, whole genome DNA isolation was carried out according to manufacturer's protocol with DNeasy Blood and Tissue kit (Qiagen Inc, Valencia, CA, United States) from cells grown micro-aerobically for 72 h in Mueller Hinton (MH) broth with 5% blood at 42◦C. The Qiagen plasmid midi kit (Qiagen Inc, Valencia, CA, United States) was used for plasmid isolation according to manufacturer's protocol. DNA quantification was done with a Qubit 2.0 fluorimeter using high sensitivity ds DNA assay kit (Life Technologies, CA, United States) and library preparation for sequencing was completed using a Nextera XT sample preparation kit (Illumina Inc, CA, United States) as per manufacturer's instructions. Sequencing was done on Illumina MiSeq platform using Illumina MiSeq V2 reagent kit 2×150 cycles (Illumina Inc, CA, United States). Sequence assembly was performed using CLC Genomics Workbench version 7.5.1. Plasmid sequences with several contigs were joined and made into a single contig using contig vs. contig alignment. Joints for the contigs were confirmed by PCR and Sanger sequencing.

Plasmid sequences have been deposited in Genbank (**Table 1**), and announced briefly as part of whole genome sequences (Marasini and Fakhr, 2016a,b,c, 2017a,b,c). Details of all plasmids isolated from our laboratory including their Genbank accession number, number of contigs, N50 and coverages are listed in **Table 1**. All plasmid sequences submitted to GenBank were annotated by the NCBI Prokaryotic Genome Annotation Pipeline. The RAST online tool (http://rast.nmpdr.org/rast.cgi) (Overbeek et al., 2014) was used to annotate all plasmids evaluated in this study. Circular plasmid renderings were constructed in CLC Genomics Workbench version 7.5.1.

### Phylogenetic and Genomic Analysis

Pangenomic analysis of all plasmids from our laboratory was used to group plasmids in this study according to presence of different genes. Core genome and pangenome analysis for each group of plasmids from our laboratory including similar plasmids from GenBank was carried out using the GView server (https:// server.gview.ca/). In addition to our 29 plasmid sequences, all available plasmid sequences of Campylobacter species were included to study the phylogenetic relatedness and possible transmission and origin of lineages of these plasmids. A total of 134 plasmid sequences of Campylobacter species (03/28/2018) from GenBank including one plasmid sequence of Helicobacter pullorum (plasmid 229336\_12) were aligned. In GView server, blast analysis (nucleotide) was carried out using GenBank files of plasmid sequences with e-value (<1e-10), alignment length cutoff value (100) and percent identity cutoff value (80). For phylogenetic analysis, sequence alignment was done using the online version of MAFFT version 7 (https://mafft.cbrc.jp/ alignment/server/) (Kuraku et al., 2013). Because the resulting alignment exhibited regions of non-overlap for various plasmids, a Maximum Parsimony (MP), character-based approach to phylogenetic analysis was used (i.e., neither distance-matrix methods nor nucleotide substitution models can be applied when extensive non-overlap exists). PAUP (Swofford, 2002) and MEGA 6 (Tamura et al., 2013) were used for phylogenetic construction by MP. Relative branch support was assessed using the bootstrap from 1,000 replicates. For comparison and validation of phylogenetic relatedness of plasmids inferred from MP tree, we also performed NeighborNet analysis with SplitsTree4 (Huson and Bryant, 2006).

### RESULTS

### Molecular Charcterization of the Sequenced Plasmids

A total of 29 plasmids were fully sequenced using the Illumina MiSeq desktop sequencer. A total of 19 plasmids from Campylobacter coli and 10 plasmids from C. jejuni were sequenced. The sizes of circular plasmids ranged from 3,002 to 119,543 bp. Based on pangenomic analysis and types of genes present, we categorized plasmids into four different groups (**Figure 1**). These groups include (1) type-1 plasmids (pTet plasmids) with tetO gene, (2) type-2 plasmids commonly found in C. coli strains, (3) type-3 plasmids (pVir plasmids) and (4) type-4 plasmids (plasmids < 6000 bp). All plasmids from our study are listed in **Table 1** and similar plasmids found in GenBank for each group after blast analysis are listed in **Table 2**.

FIGURE 2 | (A) Circular map of pTet (type-1) plasmid (pCCDM105L) showing the presence of various genes, (B) pangenome of pTet plasmids avialable in Genbank (incuding all pTet plasmids from our laboratory), (C) core genome for all pTet plasmids (red arrows in the outer circle indicate the core genome of all plasmid sequences used in the study), and (D) core genome among pTet plasmids isolated and sequenced from our laboratory.

### Type-1: pTet Plasmids

The most prevalent plasmid type in C. jejuni and C. coli strains was Type 1 (pTet). Of the 29 plasmids that were isolated and sequenced, 14 were pTet plasmids (**Figure 1**, **Table 2**). Plasmid pCCDM105L served as as an examplar for all pTet plasmids and also was used as reference for core genome and pangeome analysis (**Figures 2A–D**). Core genome analysis among pTet plasmids isolated from our laboratory showed various genes including genes for the Type IV secretion system (virB2, virB4, virB5, virB6, virB7, virB8, virB9, virB10, and virB11 genes) as core genome (**Table 3**, **Figure 2D**, **Supplementary Table 1**).

TABLE 3 | The common genes (with identified functions only) present in all of type-1 (pTet) plasmids from our laboratory isolates (details and percentage identity in Supplementary Table 1).


\*Only genes with identified functions are included, all hypothetical proteins are excluded in the list of core genome.

The core genome among pTet plasmids from our laboratory is summarized in **Table 3**. However, only the gene for TetO was found as core genome for all pTet plasmids of C. jejuni and C. coli from Genbank (including our 14 pTet plasmids) (**Figure 2C**, **Supplementary Table 2**). Pangenome analysis showed that most pTet plasmids share similar genomic composition and size, however, a few were determined to be megaplasmid due to the presence of extra DNA length that included some genes (**Figure 2B**, **Supplementary Table 3**). Extra Mu-like prophage genes are found to be inserted in the pcjDM plasmid (Marasini and Fakhr, 2016c), whereas, pCJDM67L, pCJDM202 (Marasini and Fakhr, 2016a) and pCC14983A-1 (from Genbank) harbor extra genes including several associated with the Type VI secretion system.

Few genes responsible for virulence and antibiotic resistance were found in different plasmids (**Supplementary Table 3**). A gene encoding virulence-associated protein 2 (VapD) was found in all pTet plasmids from our laboratory. The protein kinase gene was present in pccdm2, pccdm3, pCJDM210L and pCJDM204L. An aminoglycoside phosphotransferase gene was present in pccdm1, pCJDM, pCCDM183, and pCCDM224L. Histidine kinase and DNA-cytosine methyltransferase were present only in pCCDM224L. Kanamycin kinase, uridine phosphorylase, spectinomycin adenyl transferase, hygromycin B-phosphorylase, pyrrolidone–carboxylase peptidase, aminoglycoside adenyltransferase and streptothiricin acetyl transferase were present in pCCDM183 (Marasini and Fakhr, 2017c). The pCJDM plasmid harbors most of the multidrug resistance genes that are also present in pCCDM183 except uridine phosphorylase and spectinomycin adenyl transferase (Marasini and Fakhr, 2016c). All genes present in pTet plasmids and percentage similarity to other plasmids are listed in **Supplementary Table 3**.

### Type-2: Campylobacter coli Specific Plasmids

The type-2 plasmids are -the second-most prevalent group from our study (**Figure 1**, **Table 2**). These plasmids were found only in C. coli strains and were not found in any of the C. jejuni strains screened in our study. Type-2 plasmids range from 24 to 32 kb in size. Type-2 plasmids from our laboratory and similar plasmid sequences from GenBank are listed in **Table 2**. The plasmid sequence for pCCDM18M was used as reference for Blast, core genome and pangenomic analyses (**Figures 3A–C**, **Supplementary Tables 4**, **5**). Results from core genome analysis among type-2 plasmids and similar plasmids from GenBank are presented in **Table 4** (**Figure 3B**). A number of trb genes responsible for conjugative transfer were identified in these plasmids. A larger percentage of genes were conserved among the type 2 plasmids as compared to the pTet plasmids. In addition to these transfer genes, virD4, traI, gene for single-stranded DNA binding protein and traQ were common among all type-2 plasmids (**Figures 3B,C**, **Table 4**, **Supplementary Tables 4**, **5**). Few genomic differences were found among these plasmids (**Supplementary Table 5**). Meanwhile, few genes related to Type IV secretion system, virB1, putative antirepresser, phage Rha

proteins and mobile element protein were detected in several plasmids of this group (**Supplementary Table 5**).

## Type-3: pVir Type of Plasmid

There was only one plasmid of type-3 (pCJDM67S) among those sequenced for this project (**Table 2**, **Figure 4A**). This plasmid is similar to the pVir plasmid that was thought to be a virulence plasmid (Bacon et al., 2002). This plasmid also contains most of the hypothetical proteins observed in the pVir plasmid of Campylobacter jejuni 81-176 (Bacon et al., 2002) (**Supplementary Tables 6**, **7**). The pCJDM67S plasmid shares ssb, genes for DNA topoisomerase, VirB10, VirB9, DNA transformation competancy protein, VirB4, TraQ, and RepE as core genome similar to other pVir plasmids (**Table 5**, **Figures 4B,C**). Details of all genes present among all pVir plasmids (pangenome) used in this study are presented in **Supplementary Table 7**.

### Type-4: Small Plasmids

Seven small plasmids (<6 kb) were included in our study (**Table 2**, **Figure 1**). Except for pCJDM204S and pCJDM210S, which shared some homologous genes between them, remaining small plasmids did not share similar genetic composition. Most of these plasmids contain hypothetical protein-coding genes and replication initiater genes (**Figures 5A–D**). Published plasmid

TABLE 4 | Core genome for type-2 plasmids (including all plasmids from GenBank) presented in Table 2.


Details and percentage identity are available in Supplementary Table 4.\*Only genes with identified functions are included, all hypothetical proteins are excluded in the list of core genome.

sequences from our laboratory and some similar plasmids from GenBank share a replication initiation protein as core genome (**Figure 5D**, **Supplementary Table 8**). However, one of these sequences, pCCDM223S, only harbors hypothetical proteins (**Figure 5C**, **Supplementary Table 9**).

### Phylogenetic Analysis

The original alignment comprised 134 plasmid sequences. Seven duplicate sequences were excluded for the final round of phylogenetic analyses. Results from the MP analysis are shown

TABLE 5 | Core genome for all pVir plasmids of C. jejuni and C. coli strains used in this study (see Table 2).


\*Only genes with identified functions are included, all hypothetical proteins are excluded in the list of core genome.

in the **Figure 6**. The MP tree supported a distinctive clade of pTet plasmids (type-1) and pVir plasmids (type-3) from all Campylobacter species which consisted plasmids only from C. jejuni and C. coli strains. Type-2 plasmids from C. coli strains in our study and other similar plasmids from Genbank also form a separate clade in phylogenetic tree. Although, most type-2 plasmids are from C. coli strains, two plasmids from published sequences of C. jejuni are also included in this group. A single plasmid from our laboratory (pcjdm67) is allied in a monophyletic group of type-3 plasmids (**Figure 6**). Five small plasmids (<6 kb) from our study form part of a monophyletic group within a paraphyletic type-4 alliance (**Figure 6**). The pCCDM223S sequence forms a separate cluster with plasmid pCCON31 (from C. concisus) in the paraphyletic type-4 group (**Figure 6**). Numerous well-supported lineages (bootstrap values >95) are resolved by MP analysis of all plasmids from Campylobacter species. Not all plasmid sequences could be unambiguously categorized by pangenomic analysis (**Figure 6**). No species-specific clade was detected. Largely due to extensive regions of non-overlap between divergent plasmid sequences (i.e., missing data), relationships among pangenomic types and other major lineages are not resolved by these data. Results of the NeighborNet analysis revealed several major aggregates of plasmid sequences, all of which corresponded to robust branches on the MP tree (**Supplementary Figure S1**).

### DISCUSSION

The literature lacks much information about plasmids of C. jejuni and C. coli despite the fact that these two organisms are major causes of foodborne illness (Scallan et al., 2011). A total of 29 plasmids of different origins and from both C. jejuni and C. coli were fully sequenced for this investigation using the Illumina MiSeq technology. After complete analysis of the annotated genes by RAST (Overbeek et al., 2014), three major groups of plasmids (type 1, 2, and 3) and a few small plasmids less than 6 kb (type 4) were identified. Other foodborne pathogens, such as Salmonella spp. and E. coli, are also known to have variable plasmids as well as various types of incompatibility groups (Johnson et al., 2012; Williams et al., 2013). Thus, the diversity among plasmids from these pathogens is not unique, but confirms the assertion that plasmid diversity from even closely-related bacteria can be immense (Taylor et al., 1983; Tenover et al., 1985).

Only the gene for tetracycline resistance (tetO) is found to be in core genome among all pTet plasmids (**Figure 2C**). As noted previously, the tetO gene present in pTet plasmids (type-1) also shares sequence similarity with the tetM gene of Streptococcus spp., indicating the possibility of a genetic exchange between Gram-positive and Gram-negative bacteria (Taylor, 1986). Although the tetO gene found in pTet plasmids from Campylobacter is regarded as homologous to tetracycline resistance genes from other bacteria, the genetic composition of Campylobacter plasmids shows little similarity to plasmids of other bacteria. The presence of the tetO gene in both the chromosome and the plasmids of C. jejuni and C. coli indicates that the gene was either present in the chromosome and was later transferred with the integrated plasmids, or it might have reached the chromosome following acquisition of an integrated plasmid (Pratt and Korolik, 2005; Crespo et al., 2012). The high prevalence of pTet plasmids in Campylobacter strains from our study is similar to previous reports of clinical isolates in Germany (Schmidt-Ott et al., 2005). The presence of pTet (type-1) and pVir (type-3) in Campylobacter was also discussed in the previous investigation (Schmidt-Ott et al., 2005). We concur with Schmidt-Ott et al. (2005) that pVir plasmids are less prevalent than pTet plasmids in Campylobacter strains.

The pTet plasmids harbor important genes responsible for conjugation and virulence, exemplified by the Type IV secretion system (Bacon et al., 2000). The Type IV secretion system was reported in chromosomes of Helicobacter pylori (Fernandez-Gonzalez and Backert, 2014) and was conserved in the plasmids and various genomic islands in Campylobacter fetus (Graaf–van Bloois et al., 2016). Many hypothetical proteins of unknown function were observed in several previouslycharacterized plasmids (Batchelor et al., 2004) and also in plasmids sequenced for this investigation. Some of the genes responsible for conjugation are similar to genes from Actinobacillus actinomycetemcomitants (Batchelor et al., 2004), which might indicate transfer of conjugative genes between nonrelated microbes. Some of the pTet plasmids (pcjDM, pCCDM183, pccdm1, and pCCDM224L) were found to possess the aminoglycoside phosphotransferase genes. These pTet plasmids are all greater than 48 kb in size except pccdm1 which is only 44 kb in size. Some plasmids (i.e., pcjDM and pCCDM183) contained multidrug resistance genes such as aminoglycoside adenyl transferase, streptothiricin- and hygromycin-resistant genes. In addition, the pCCDM183 plasmid sequence also contained genes for kanamycin kinase and spectinomycin oadenyl transferase. The presence of aminoglycoside resistance genes in Campylobacter strains was also reported in previous studies (Tenover et al., 1992; Chen et al., 2013). Megaplasmids

of more than 80 kb are present in the pTet plasmids (type-1) group (Marasini and Fakhr, 2016a,c; Miller et al., 2016). In addition to the common genes among pTet plasmids (**Table 3**), the megaplasmids were shown to have an inserted segment of bacteriophage genes or other types of mobile genetic components (Gunther et al., 2016; Marasini and Fakhr, 2016c). Some of these megaplasmids also contained the complete Type VI secretion system with all 14 core genes present (Marasini and Fakhr, 2016a). The fact that 14/29 plasmids sequenced in our laboratory were pTet plasmids (**Table 2**) is not surprising since we previously reported that tetracycline resistance was prevalent among Campylobacter jejuni and Campylobacter coli strains isolated from various retail meats (Noormohamed and Fakhr, 2012, 2013, 2014). Functional analysis of the virulence and antimicrobial resistance genes present on these plasmids is worth investigating and may shed some light on the role of these genes in conferring the corresponding phenotypes.

The type-2 plasmids, which ranged in size from 24 kb to 32 kb, were similar to few others deposited in the GenBank. This plasmid group primarily consists of plasmids from C. coli strains. However, two plasmids from C. jejuni strains are also found to be allied in this group (**Figure 6**). A plasmid present in Helicobacter pullorum (i.e., plasmid 229336\_12) is a close ally of the type-2 plasmids (**Figure 6**). Most of the predicted genes in these plasmids are common to all plasmids in this group (type-2) except for a few hypothetical proteins. The core genome for type-2 plasmids included genes for conjugative transfer along with type IV secretion system genes such as virD4 and virB1(**Table 4**). Most of the conjugative transfer genes present in these plasmids were different from the Type IV conjugative transfer genes in other C. jejuni and C. coli plasmids (type-1 and type-3). The similarity between type-2 plasmids from C. jejuni and C. coli strains and the 229336\_12 plasmid (H. pullorum) might indicate a possible route of transmission and genetic interchangeability of these plasmids between species. The pCCDM67S plasmid from our study is similar to a pVir type plasmid, previously studied by Bacon et al. (2002). The pCCDM67S plasmid also contains orthologs of the Type IV secretion system found in Helicobacter pylori (Bacon et al., 2000).

In addition to types 1-3, we characterized other small cryptic plasmids (type-4) that share similar genomic composition and arrangements with previously-characterized plasmids of diverse sources (Jesse et al., 2006; Miller et al., 2007). Most of these plasmids contain the replication initiator protein and some unknown hypothetical proteins. Some of these plasmids also contain genes coding for Mob proteins.

The phylogenetic analysis clearly shows numerous, wellresolved lineages comprised of complete sequences for

FIGURE 6 | Maximum parsimony tree for all plasmid sequences of Campylobacter species (sequences from our laboratory are highlighted with colored circles). Categorization of plasmids from pangenomic analysis (Figure 1) are represented with shaded colors [type-1 (black), type-2 (blue), type-3 (green) and type-4 (red)] on the phylogenetic tree. Only boostrap values >70 are are mapped to the phylogenetic tree. Duplicate plasmid sequences were excluded from the analysis. The results of NeighborNet analysis for Campylobacter plasmids are illustrated in Supplementary Figure S1.

all Campylobacter plasmids reported to date (**Figure 6**, **Supplementary Figure S1**). The branching pattern of the phylogenetic trees supports our categorization of Campylobacter plasmids according to pangenomic analysis for type 1, 2, and 3 plasmids. In MP tree, the most prevalent pTet plasmids (type-1) were all grouped in one lineage. Similarly, type-2 and type-3 (pVir) plasmids formed distinctive clades with similar plasmids from Genbank. Type 1 (pTet), type-2, and type-3 (pVir) plasmid groups consist of plasmid sequences from C. jejuni/C. coli strains. The small (<6 kb) plasmid sequences from C. jejuni/C. coli strains cluster with plasmids from other Campylobacter species. One plasmid cluster including plasmid mp1 (CP014569.1) consists of plasmids from mostly C. fetus and non-jejuni/coli species of Campylobacter. This cluster is found near type-2 plasmids. Since we have very little knowledge of the distribution of these Campylobacter plasmids, we used Maximum Parsimony for the analysis of the plasmid sequences. In a previous study done by (Crespo et al., 2016), similar types of phylogenetic relationships were observed where the majority of plasmids with tetO genes were allied in one cluster and some smaller plasmids were allied in another cluster. Results of NeighborNet analysis indicate that plasmids found in Campylobacter species likely have a convoluted evolutionary history (**Supplementary Figure S1**). Nonethless, there is a 1:1 correspondence between the network (**Supplementary Figure S1**) and the tree from MP analysis (**Figure 6**) regarding major groupings of type 1, type 2 and type 3 plasmids (**Supplementary Figures S1B,D,F**). The root also shows correspondence between the two analyses. A portion of the type 4 group (**Figure 6**) forms a loose cluster in the network analysis (**Supplementary Figure S1E**). The type 4 group is a non-monophyletic assemblage in the MP analysis, too. This indicates that type 4 is either the most diverse plasmid type or may be comprised of additional, unrecognized types. We can

### REFERENCES


simply state that the well-resolved portions of the phylogeny provide a reasonable inference for relationships among plasmids.

Most of the isolated and sequenced plasmids are associated with isolates of C. jejuni and C. coli. This observation is consistent with the fact that most clinical cases of Campylobacter infection are associated with C. jejuni and C. coli strains (Acheson and Allos, 2001; Gillespie et al., 2002). The absence of any species-specific plasmid clades indicates intra-species dissemination of plasmids among Campylobacter species. Several divergent lineages are present in our analyses (**Figure 6**, **Supplementary Figure S1**) and these may be representatives of larger, but under-sampled plasmid lineages. Thus, the results of this study indicate that additional sampling will be needed to more fully understand the evolution and transmission of Campylobacter plasmids.

### AUTHOR CONTRIBUTIONS

MF conceived the research idea and design. DM and AK performed the experimental procedures. DM, AK, and MB performed the phylogenetic analysis. DM, AK, MB, and MF prepared the manuscript.

### ACKNOWLEDGMENTS

We would like to acknowledge financial support from the Research Office of The University of Tulsa (Tulsa, OK, USA) for awarding DM a student research grant.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.02167/full#supplementary-material

resistance in the turkey-derived Campylobacter jejuni strain 11601MD. Plasmid 86, 32–37. doi: 10.1016/j.plasmid.2016.06.001


including an 81.08-Kilobase plasmid. Genome Announc. 4, e00305–e00316. doi: 10.1128/genomeA.00305-16


from a turkey farm), harboring a novel gentamicin resistance mobile element. Genome Announc. 4, e01175–e01116. doi: 10.1128/genomeA.01175-16


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

Copyright © 2018 Marasini, Karki, Buchheim and Fakhr. 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.

# Virulence and Antibiotic Resistance Profiles of Cronobacter sakazakii and Enterobacter spp. Involved in the Diarrheic Hemorrhagic Outbreak in Mexico

Julio Parra-Flores<sup>1</sup> , Juan Aguirre<sup>2</sup> \*, Vijay Juneja<sup>3</sup> , Emily E. Jackson<sup>4</sup> , Ariadnna Cruz-Córdova<sup>5</sup> , Jesus Silva-Sanchez<sup>6</sup> and Stephen Forsythe<sup>7</sup>

<sup>1</sup> Departamento de Nutrición y Salud Pública, Facultad Ciencias de la Salud y de los Alimentos, Universidad del Bío-Bío, Chillán, Chile, <sup>2</sup> Departamento Agroindustria y Enología, Facultad de Ciencias Agronómicas, Universidad de Chile, Santiago, Chile, <sup>3</sup> Residue Chemistry and Predictive Microbiology Research Unit, Eastern Regional Research Center, Agricultural Research Service, United States Department of Agriculture (USDA), Wyndmoor, PA, United States, <sup>4</sup> Department of Biology, University of Nevada, Reno, Reno, NV, United States, <sup>5</sup> Laboratorio de Bacteriología Intestinal, Hospital Infantil de México, Federico Gómez, Mexico City, Mexico, <sup>6</sup> Grupo de Resistencia Bacteriana, Instituto Nacional de Salud Pública, Cuernavaca, Mexico, <sup>7</sup> Foodmicrobe.com, Nottingham, United Kingdom

### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Ondrej Holý, ˇ Palacký University Olomouc, Czechia Luis Augusto Nero, Universidade Federal de Viçosa, Brazil

> \*Correspondence: Juan Aguirre juan.aguirre@uchile.cl; juaguirr@vet.ucm.es

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 16 May 2018 Accepted: 29 August 2018 Published: 27 September 2018

#### Citation:

Parra-Flores J, Aguirre J, Juneja V, Jackson EE, Cruz-Córdova A, Silva-Sanchez J and Forsythe S (2018) Virulence and Antibiotic Resistance Profiles of Cronobacter sakazakii and Enterobacter spp. Involved in the Diarrheic Hemorrhagic Outbreak in Mexico. Front. Microbiol. 9:2206. doi: 10.3389/fmicb.2018.02206 Cronobacter spp. are bacterial pathogens that cause neonatal meningitis, septicemia, and necrotizing enterocolitis in infants with a lethality rate of 40–80%. Powdered infant formulas (PIF) have been implicated as the main vehicles of transmission. This pathogen can also cause infection through contaminated expressed breast milk, and it has been recovered from neonatal feeding tubes of neonates not fed reconstituted PIF and milk kitchen areas. This study analyzed antibiotic resistance profiles and the tissue virulence tests of Cronobacter sakazakii and Enterobacter spp. recovered from PIF, infant fecal matter's, and milk kitchen environment involved in a diarrheic hemorrhagic outbreak in 2011 in Mexico. The strains isolated from the outbreak had similar antibiotic resistance profiles and pathogenicity irrespective of isolation site, however, C. sakazakii strains isolated from PIF showed significantly higher invasive profiles than Enterobacter spp. (p = 0.001) and 83% were resistant to more than one antibiotic. The findings of this study can be used to complement existing information to better control Cronobacter and Enterobacter spp. contamination in PIF production, prevent its transmission, and improve infant food safety.

Keywords: Cronobacter sakazakii, Enterobacter hormaechei, powdered infant formula, virulence, antibiogram

## INTRODUCTION

Cronobacter infections are associated with adults and infants (Bowen and Braden, 2006; Holy and Forsythe, 2014; Alsonosi et al., 2015; Forsythe, 2018). Nevertheless, infections of premature neonates are of particular concern are due to their high lethality rate of between 40 and 80% (Joseph and Forsythe, 2012). The clinical manifestation of this pathogen in infants is mainly meningitis, septicemia, and necrotizing enterocolitis (Nazarowec-White and Farber, 1999; Van Acker et al., 2001; Baumbach et al., 2009; Hariri et al., 2013b) although diarrheal and urinary infections have also been observed (Friedemann, 2009).

From 2003 to 2009, 544 cases of Cronobacter spp. infection were identified in 6 states of the United States, especially among children <5 years of age (Patrick et al., 2014), indeed, Cronobacter is the genus that is the most commonly involved in cases of illness associated mainly with the consumption of contaminated powdered infant formula (PIF) rehydrated (Food Agriculture Organization of the United Nations [FAO] and World Health Organization [WHO], 2004, 2006) although, additional possible reservoirs from preparation utensils and the environment have been recognized (Friedemann, 2008; Siqueira-Santos et al., 2013; Holy and Forsythe, 2014) and contaminated expressed breast milk, where C. malonaticus strain was isolated from a breast abscess (Bowen et al., 2017). Additionally, C. sakazakii has been isolated from the enteral feeding tubes of neonates not fed reconstituted infant formula (Hurrell et al., 2009).

There are recommended biochemical methods to identify Cronobacter spp. (Api20E, ID32E, BIOLOG microarray, Vitek 2 System), but these can only be used for presumptive identification and they can have accuracy level as low as 43% (Cetinkaya et al., 2012; Joseph et al., 2013; Jackson and Forsythe, 2016). Several PCR primers have been proposed to identify members of the genus Cronobacter by amplifying specific sequences of variable and conserved regions of the 16S rRNA of the bacteria (Lehner et al., 2004; Hassan et al., 2007). Specific primers for the rpoB gene encoding the β region of the polymerase enzyme have been proposed for identifying Cronobacter species, but have not taken into account changes in the taxonomy of the species, giving false positive results with some Enterobacter species (Jackson et al., 2015; Jackson and Forsythe, 2016).

Baldwin et al. (2009) developed a 7-loci multilocus sequence typing (MLST) scheme for Cronobacter speciation and genotyping. The MLST scheme has an open access database<sup>1</sup> that contains >2,400 strains and >350 whole genomes along with corresponding metadata and updates according to changes in taxonomy. This approach has led to the recognition of clonal complexes (CC) within the Cronobacter genus. Of special significance is the recognition of the Cronobacter sakazakii CC4 pathovar which is strongly associated with neonatal meningitis cases (Joseph and Forsythe, 2011; Sonbol et al., 2013; Hariri et al., 2013a; Forsythe et al., 2014; Forsythe, 2018).

Jackson et al. (2015) provided the re-evaluation of a previous study done by Flores et al. (2011) of C. sakazakii outbreak caused by consuming contaminated reconstituted PIF in Mexico, which had used phenotyping and rpoB PCR probe method to identify the isolates, whereas Jackson et al. (2015) used DNA sequencing, and showed that the strains were E. hormaechei and Enterobacter spp. (undesignated species), demonstrating for the first time, the possible transmission of Enterobacter from PIF to infants (Jackson et al., 2015). This possible transmission suggests that these organisms may pose a risk to infants consuming rehydrated PIF (Jackson et al., 2015). In fact, this risk was estimated by Parra-Flores et al. (2016) in a risk based assessment under a probabilistic approach of reconstituted PIF contaminated with different inoculum size of Cronobacter, differing heat treatment to prepare the PIF and storage temperature.

Important aspects to be considered in the severity and prognosis of Cronobacter infection are the presence of antibioticresistance (Caubilla-Barron et al., 2007; Kilonzo-Nthenge et al., 2012; Xu et al., 2015), and virulence factors (Townsend et al., 2008). Such virulence factors can include iron acquisition and the invasiveness and adhesion in cell lines such as HEp-2 and CaCo-2 (Pagotto et al., 2003; Mange et al., 2006; Grim et al., 2012; Almajed and Forsythe, 2016).

The aim of this work was to evaluate and compare the virulence and antibiotic resistance profiles of the Cronobacter sakazakii and Enterobacter spp. involved in the diarrheic hemorrhagic outbreak in Mexico in 2011.

### MATERIALS AND METHODS

### Bacterial Strains

All bacterial strains (n = 24) had been isolated and identified according to 7-loci MLST as previously described (Jackson et al., 2015) (**Figure 1**). They had been recovered from PIF (n = 14), fecal material (n = 6), and the PIF preparation area (n = 4).

### Sequencing of fusA Gene

The methodology described by Baldwin et al. (2009) was followed using PCR CORE Kit QIAGEN (Cat No. 201225) solutions. Amplified products were sent to MACROGEN in Korea for sequencing. The sequenced products were analyzed with the Gentle software and later aligned with the ClustalW software. A phylogenetic tree was constructed using the maximum likelihood method with the MEGA7 software. Identification was performed with the free access online database https://pubmlst. org/cronobacter/ and BLASTn (NCBI).

### Antibiotic Resistance Profile

The antibiograms of 24 strains were performed by the disk diffusion method (Clinical and Laboratory Standards Institute [CLSI], 2018). Disks with 12 commercial antibiotics were used (Bio-RadTM, United States): ampicillin (10 µg), amikacin (30 µg), levofloxacin (5 µg), cephalothin (30 µg), cefotaxime (30 µg), ceftriaxone (30 µg), chloramphenicol (30 µg), gentamicin (10 µg), netilmicin (30 µg), nitrofurantoin (300 µg), cefepime (30 µg), and sulfamethoxazole-trimethoprim (25 µg). The characterization of the strain resistance/susceptibility profiles was determined by measuring the inhibition area and interpreting the diameters according to the manufacturer's instructions. Escherichia coli ATCC 25922 was used as a reference.

## Virulence Determination of Cronobacter and Enterobacter spp.

### Adherence Assay

HEp-2 cells were cultured in Eagle's minimal medium (MEM) (In Vitro, Mexico) supplemented with 10% fetal bovine serum (FBS) (Gibco, United States) and without antibiotics. The cells were prepared in T75 cm<sup>2</sup> flasks (Sarstedt, Germany) and grown for 24 h at 37◦C and 5% CO2. Upon reaching

<sup>1</sup>www.PubMLST.org/cronobacter/

#### TABLE 1 | Resistant antibiotics of isolated strains by source and species.


confluency, cells were disaggregated with 0.25% trypsin (In Vitro, Mexico) and sown into 24 well plates from a 1 ml suspension containing 2.5 × 10<sup>6</sup> HEp-2 cells per ml (Sarstedt, Germany). The monolayers with 70–80% confluency were washed three times with phosphate-buffered saline (PBS) and 900 µl of MEM were added. Isolates were previously cultured overnight in 1% tryptone, and 100 µl bacterial suspensions (10<sup>8</sup> cells per ml) were added to each well. Plates were incubated for 3 h at 37◦C and 5% CO2. For quantitative assays, bacteria were removed by adding 1 ml 0.1% Triton X-100 (Amresco, United States), and serial 10-fold dilutions were plated onto tryptone soy agar (TSA) to determine the colony-forming units (CFU) of bacteria bound to HEp-2 cells. Triplicate assays were performed. Enteroaggregative E. coli O42 (EAEC) was used as the positive control. Escherichia coli K-12 HB101 was the negative control (Cruz et al., 2011). This assay was repeated three times and the results were expressed as the means ± SD of the data.

### Invasion Assays

The HEp-2 cell lines were prepared according to the procedure described in the adherence assay section. After 3-h incubation, the infected monolayers were washed three times with PBS and incubated with 1 ml MEM with lysozyme 300 µg/ml and gentamycin 100 µg ml−<sup>1</sup> (In Vitro, Mexico). Samples were washed once with PBS. For quantitative assays, cells were washed three times with PBS, detached with 1 ml 0.1% Triton X-100 and plated on TSA. Invasion frequencies were calculated as the number of bacteria surviving incubation with gentamycin divided by the total number of bacteria present in the absence of this antibiotic (bacterial adherence). Enteroinvasive E. coli 1192 and E. coli HB101 (K-12) were used as positive and negative controls, respectively. This assay was repeated three times and the results were expressed as the means ± standard deviation of the data (Cruz et al., 2011).

### RESULTS AND DISCUSSION

This study is an extension of a previous study by Flores et al. (2011) and Jackson et al. (2015), who suggested that, after the fusA sequence analysis, isolates from the outbreak in Mexico 2011 were a mixture of C. sakazakii, Enterobacter hormaechei, and Enterobacter spp. This was also corroborated by the phylogenetic analysis that clearly showed differences with other strains (**Figure 1**). Currently, the fusA sequencing method enables higher speciation accuracy because it follows the whole genome phylogeny and adjusts to taxonomic changes (Forsythe et al., 2014; Xu et al., 2014; Alsonosi et al., 2015; Jackson et al., 2015).

Several studies have confirmed that Cronobacter strains can be effectively eliminated by antibiotics, however, prolonged use of antibiotics, inappropriate dose, quantity and length of application are undesirable as it may result in the development of Cronobacter antibiotic resistance (Pérez et al., 2007; Langdon et al., 2016; Fei et al., 2017; Kardos, 2017). Therefore, it is interesting to determine some degree of association in the resistance profiles among strains from food products, environment, and fecal matter of colonized individuals exhibiting or not exhibiting symptoms or from a foodborne outbreak. This aspect is considered part of the objective of the present study because the strains were isolated from PIF, surfaces on which PIF was prepared (**Supplementary Figure S1**), and fecal matter of infants associated to an outbreak occurred in Mexico (Flores et al., 2011). This evaluation enables the design of treatment strategies for exposed individuals, especially those who are hypersensitive such as infants and the elderly. Although, there is considerable speculation about the source of PIF contamination. Some researchers suggest that the sources are either the environment of the production plants (Craven et al., 2010; Jacobs et al., 2011; Fei et al., 2015; Jing-Hua et al., 2015) or the ingredients (i.e., starch) used to prepare the PIF, which are the probable vehicles of transmission (Food Agriculture Organization of the United Nations [FAO] and World Health Organization [WHO], 2004; Jongenburger et al., 2011; Walsh et al., 2011). The Food Agriculture Organization of the United Nations [FAO] and World Health Organization [WHO] (2004, 2006) risk assessments on the microbiological safety of infant formula also recognized that other members of Enterobacteriaceae were recoverable from PIF and could put infants at risk even though no confirmed cases had been confirmed at that time.

In our study, the 24 strains were identified as C. sakazakii (5), Enterobacter hormaechei (3), and Enterobacter spp. (16). In general, 83% (20/24) of the isolated strains were resistant to 1–7 antibiotics. Eight percent (2/24) were resistant to 5 antibiotics and 37% (9/24) resistant to 3 antibiotics. Sixty-five percent (16/24) of the resistant strains were isolated from PIF (**Table 1**).

Eighty percent of C. sakazakii (4/5) strains were resistant to cephalothin (**Table 2**). It is important to assess the antibiotic resistance profile of Cronobacter spp., associated to those product (PIFs and infant products) consumed by high risk population whose are immunologically vulnerable. Molloy et al. (2009) reported that of 33 C. sakazakii strains isolated in the environment, 51% were resistant to cephalothin. Kleiman et al. (1981) also reported a moderate resistance to cephalothin in isolated strains in a case of meningoencephalitis.

For E. hormaechei, 100% (3/3) were resistant to cephalothin and ampicillin, 33% (1/3) to cefotaxime and ceftriaxone, and 66% (2/3) to nitrofurantoin. The Enterobacter spp. strains were resistant to cephalothin and ampicillin. The resistance values for ampicillin and cephalothin are higher than those previously reported (Kim et al., 2008; Molloy et al., 2009). Lai (2001) reported isolated strains were resistant to the first and second cephalosporin generation. The same situation was found in the present study with the Cronobacter sakazakii and Enterobacter strains isolated from PIF, milk kitchen surfaces, and fecal matter. Resistance was 26% (4/16) for cefotaxime, 13% (2/16) for ceftriaxone, and 26% (4/16) for cefepime (**Table 2**). This suggests that ß-lactamase production should be further monitored as recommended by the Food Agriculture Organization of the United Nations [FAO] and World Health

TABLE 2 | Resistant antibiotics profile of Cronobacter sakazakii, E. hormaechei, and Enterobacter spp.


LEV Levofloxacin, FEP Cefepime, CF Cephalothin, CTX Cefotaxime, SXT trimethoprim-Sulfamethoxazole, AM Ampicillin, CRO Ceftriaxone, NF Nitrofurantoin, NET Netilmicin, GE Gentamicin, AK Amikacin, CL Chloramphenicol. R: resistant; I: intermediate; S: susceptible.

Organization [WHO], 2008, especially since the resistant strains were isolated from PIF. Caubilla-Barron et al. (2007) analyzed Cronobacter sakazakii strains from an outbreak with fatalities in a neonatal intensive care unit in France; they found one pulsotype that was associated with the three fatal cases. These were later shown to be the pathovar C. sakazakii CC4 (Joseph and Forsythe, 2011; Masood et al., 2015). In addition, two of these isolates had extended-spectrum ß-lactamase activity. A recent study evaluated the antimicrobial and desiccation resistance of Cronobacter sakazakii (Caubilla Barron and Forsythe, 2007), and Cronobacter malonaticus isolates from powdered infant formula and processing environments showed that the 70 Cronobacter strains, representing 19 sequence types, were susceptible to the most of the antibiotics except for amoxicillin-clavulanate, ampicillin, and cefazolin (Fei et al., 2017) which is in accordance with our results.

Our findings indicate that hospitalized infants were unpurposed and accidentally exposed to Cronobacter and Enterobacter spp. for 2 months. This fact could increase the susceptibility to suffer an infection by this pathogen, especially if this pathogen has a variety of virulence factors which aid in tissue adhesion, invasion and host cell injury. In addition, the results of this study indicate the hospitalized infants were unpurposed and accidentally exposed to Cronobacter and Enterobacter spp. which were able to adhere and invade human cells (HEp-2 cell line) in vitro. This was shown using twelve selected strains which had been isolated from PIF, work surfaces, and fecal matter (**Figure 2**). Due to funding limitations it was impossible to carry out more strains analysis.

Adherence is one of the events that enables bacteria to colonize and invade the host cells; it is a property associated with bacterial pathogenesis, especially of intracellular pathogens (Pizarro-Cerdá and Cossart, 2006; Cruz et al., 2011). In our study, adherence mean values in Cronobacter spp., Enterobacter hormaechei, and Enterobacter spp. HEp-2 cells were 22, 23, and 19 × 10<sup>4</sup> CFU/mL and with no significant differences among them (p > 0.05). Mean invasion values were 3.3, 5.2, and 2.5%, respectively; E. hormaechei (p = 0.001) was significantly more invasive than C. sakazakii. In general, 100% of the evaluated strains had adherence capacity and 75% were invasive in HEpcells; these values were similar to the results reported by Mange et al. (2006) and Townsend et al. (2008).

Cruz et al. (2011) found five species of Cronobacter spp. (C. sakazakii, C. malonaticus, C. dublinensis, C. muytjensii, and C. genomospecies (current universalis), all of which had the capacity to adhere to HEp-2 cell lines. The C. sakazakii strains from a human source exhibited higher adherence values compared to strains of the same species isolated from other sources. Furthermore, when the invasion capacity of C. sakazakii was evaluated, it was found that 35% of the isolates were invasive and apparently more efficient than the other evaluated Cronobacter spp. species.

The C. sakazakii and E. hormaechei strains evaluated in our study were invasive; however, Enterobacter spp. only had 33% of invasive strains, which is of concern because the virulence trait is in isolated PIF strains. Reports of several outbreaks of sepsis in neonatal intensive care units in Brazil and the United States (Campos et al., 2007; Townsend et al., 2008) have shown that E. hormaechei is clinically significant, indeed an outbreak of E. hormaechei occurred among premature infants in the intensive care nursery (ICN) at the Hospital of the University of Pennsylvania between November 29, 1992 and March 17, 1993 (Wenger et al., 1997).

Cronobacter species adhered to HEp-2, Caco-2 and brain microvascular endothelial cells, producing two distinctive adherence patterns, a diffuse and a localized adhesion (Mange et al., 2006; Cruz et al., 2011). Moreover, it has been suggested that the outer membrane proteins OmpA and OmpX from C. sakazakii are involved in basolateral invasion of human enterocyte-like Caco-2 and intestinal epithelial cells (Townsend et al., 2007; Singamsetty et al., 2008).

In conclusion, all isolated strains showed resistant to more than one antibiotic (cephalothin, ampicillin, cefotaxime, and ceftriaxone) independent of the source of isolation. In addition, C. sakazakii strains isolated from PIF were significantly more invasive than Enterobacter spp. Individually; E. hormaechei was more invasive than C. sakazakii and Enterobacter spp.

The knowledge generated in the present work can be used to complement existing information to better control Cronobacter and Enterobacter spp. contamination in PIF production, prevent its transmission, and improve infant food safety. This information should support regulatory and health authorities in their microbial surveillance measures and improve neonatal and infant health.

### REFERENCES


### AUTHOR CONTRIBUTIONS

JP-F conceived the experiments. JP-F, JS-S, AC-C, and JA designed the experiments. JP-F, AC-C, and JS-S conducted the laboratory work. VJ and JA provided data analysis. JP-F, AC-C, JS-S, SF, VJ, and JA drafted the manuscript. EJ revised the manuscript and data analysis. All the authors reviewed and approved the final manuscript.

### FUNDING

Funding was provided by the Research Directorate of the Universidad del Bío-Bío, Projects 161720 3/R, 091824/R, and GI 171220/EF and Consejo Nacional de Ciencia y Tecnología, México project 98625.

### ACKNOWLEDGMENTS

We thank the Santander Research Mobility Grant for providing financial support. We also wish to thank Suzanne Théberge for technical assistance with the manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.02206/full#supplementary-material

FIGURE S1 | Milk kitchen floor plan: (A) PIF bottles preparation area (sterile), and (B) Washing and disinfection area.

by contaminated parenteral nutrition in Brazil. J. Hosp. Infect. 66, 95–97. doi: 10.1016/j.jhin.2007.02.013


from powdered infant formula and processing environments. Front. Microbiol. 8:316. doi: 10.3389/fmicb.2017.00316


resistance in a clinical isolate. Antimicrob. Agents Chemother. 51, 3247–3253. doi: 10.1128/AAC.00072-07


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

Copyright © 2018 Parra-Flores, Aguirre, Juneja, Jackson, Cruz-Córdova, Silva-Sanchez and Forsythe. 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.

# Discovery on Antibiotic Resistance Patterns of Vibrio parahaemolyticus in Selangor Reveals Carbapenemase Producing Vibrio parahaemolyticus in Marine and Freshwater Fish

Learn-Han Lee1,2,3 \* † , Nurul-Syakima Ab Mutalib<sup>4</sup> , Jodi Woan-Fei Law1,2 , Sunny Hei Wong<sup>5</sup> and Vengadesh Letchumanan1,2 \* †

#### Edited by:

Giovanna Suzzi, Università degli Studi di Teramo, Italy

#### Reviewed by:

Lanming Chen, Shanghai Ocean University, China Dapeng Wang, Shanghai Jiao Tong University, China

#### \*Correspondence:

Learn-Han Lee lee.learn.han@monash.edu; leelearnhan@yahoo.com Vengadesh Letchumanan lvengadesh@yahoo.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 13 June 2018 Accepted: 02 October 2018 Published: 25 October 2018

#### Citation:

Lee L-H, Ab Mutalib N-S, Law JW-F, Wong SH and Letchumanan V (2018) Discovery on Antibiotic Resistance Patterns of Vibrio parahaemolyticus in Selangor Reveals Carbapenemase Producing Vibrio parahaemolyticus in Marine and Freshwater Fish. Front. Microbiol. 9:2513. doi: 10.3389/fmicb.2018.02513 <sup>1</sup> Novel Bacteria and Drug Discovery Research Group, Biomedicine Research Advancement Centre, School of Pharmacy, Monash University Malaysia, Bandar Sunway, Malaysia, <sup>2</sup> Biomedical Research Laboratory, Jeffrey Cheah School of Medicine and Health Sciences, Monash University Malaysia, Bandar Sunway, Malaysia, <sup>3</sup> Center of Health Outcomes Research and Therapeutic Safety, School of Pharmaceutical Sciences, University of Phayao, Phayao, Thailand, <sup>4</sup> UKM Medical Molecular Biology Institute, UKM Medical Centre, National University of Malaysia, Bangi, Malaysia, <sup>5</sup> Department of Medicine and Therapeutics, Li Ka Shing Institute of Health Sciences, The Chinese University of Hong Kong, Shatin, Hong Kong

Vibrio parahaemolyticus, a Gram-negative halophilic bacterium is often associated with fish and fishery products, thus causing gastroenteritis in humans upon ingestion of contaminated food. V. parahaemolyticus has become a globally well-known pathogen with yearly reported cases in many countries. This study aimed to discover the antibiotic resistance patterns of V. parahaemolyticus as well as detect Carbapenem resistant isolates from marine and freshwater fish in Selangor. A total of 240 freshwater and marine fish samples collected from wet market and supermarket in Selangor were tested for the presence of V. parahaemolyticus. All the fish samples were determined positive for V. parahaemolyticus using conventional microbiological culture-based method. The toxR gene were detected via polymerase chain reaction (PCR) in 165/240 (69%) isolates. The two-virulence factor of V. parahaemolyticus, thermostable direct hemolysin (tdh) and TDH-related hemolysin (trh) was screened via PCR. As such, four isolates were trh+and none were tdh+. Majority of the isolates presented high resistance to ampicillin (88%), amikacin (64%), and kanamycin (50%). In addition, this study identified 19-imipenem resistant isolates isolated from freshwater and marine fish samples. Further analysis of these 19-imipenem resistant isolates revealed that the resistance toward imipenem was plasmid mediated after plasmid curing assay. The multiple antibiotics resistance index was >0.2 for 70% of the isolates. In summary, the results confirm the presence of V. parahaemolyticus in freshwater and marine fish samples in Selangor, Malaysia. To our best knowledge, this is the first report discovering the antibiotic resistant patterns and Carbapenem-resistant isolates of V. parahaemolyticus isolated from marine and freshwater fish samples in Selangor.

Keywords: Vibrio parahaemolyticus, carbapenem, freshwater, marine, antibiotic resistant, MAR index

### INTRODUCTION

fmicb-09-02513 October 23, 2018 Time: 14:30 # 2

Vibrionaceae family within the class of Gammaproteobacteria comprises of Gram-negative halophilic bacteria, straight or curved rods, ubiquitous and indigenous in aquatic environments (Tison and Kelly, 1984; Tantillo et al., 2004; Sawabe et al., 2013). The Vibrio genus consists of 142 species that are marine originated and its taxonomy is continuously been revised due to the discovery of new species (Sawabe et al., 2013). Vibrio parahaemolyticus is among the member of this genus that been regarded as important human pathogenic bacteria (Su and Liu, 2007; Iwamoto et al., 2010; Bier et al., 2015; Law et al., 2015). The species is widely distributed in marine and estuarine environments thus leading to gastrointestinal infections upon consumption of raw or undercooked seafood (Kubota et al., 2008; Letchumanan et al., 2014; Lee and Raghunath, 2018). Based on the published data by Centers for Disease Control and Prevention (CDC) in the United States during the year 2016, V. parahaemolyticus is considered as a major foodborne bacterium compared to other Vibrio species and accounted for nearly 34,664 foodborne cases annually in the United States (Scallan et al., 2011; Huang et al., 2016).

In terms of its pathogenicity, thermostable direct hemolysin (tdh) gene, TDH-related hemolysin (trh) gene, T3SS systems (T3SS1 and T3SS2) are among the virulence factors own by pathogenic V. parahaemolyticus in order to initiate an infection (Letchumanan et al., 2014, 2017). Usually, 99% of clinical V. parahaemolyticus isolates are known to be pathogenic because they carry tdh genes and/or trh genes, whereas majority of the environmental isolates are non-pathogenic (Sudha et al., 2012; Tsai et al., 2013). Nevertheless, around 0–6% of the environmental isolates are identified as pathogenic carrying tdh gene and/or trh gene (Letchumanan et al., 2014, 2015a).

The aquaculture industry in Malaysia is mainly associated with its economic gains from supplying domestic and foreign demands, and as well as generating a steady income for farmers (Witus and Vun, 2016). Fish is among the popular fishery products that been consumed in daily basis by consumers from Southeast Asian countries (Hajeb et al., 2009). Around 75% of the global fishery production is mainly for human consumptions (Teh, 2012). In Malaysia, the fish consumption has increased since 1970 and now its above 40 kg/capita/year (Teh, 2012). Professed has a healthy food, fish contains a high level of proteins, omega-3 fatty acids (n-3), essential vitamins and minerals that are required by an individual (Aremu and Ekunode, 2008; Hajeb et al., 2009). There are variety of fishes that been consumed by Malaysian in their daily life including the Indian mackerel, Spanish mackerel, black pomfret, silver pomfret, yellowstripe scad, catfish, fringe scale sardine, and tilapia (Osman et al., 2001; Hajeb et al., 2009; Taweel et al., 2013). The expanding and intense aquaculture industry has led to the suppression of immune systems and increases the susceptibility of fish to bacterial infections (Davies et al., 2001; Basti et al., 2006; Harikrishnan et al., 2011).

Intensified fish farming in order to meet consumers demand has prompted the use of antibiotics as treatment regime, prophylaxis and as growth promotion (Vaseeharan et al., 2005). Antibiotics are often been in-cooperated as feed additives or immersion bath in order to treat bacterial infections, promote fast growth of fish, and also prevent the growth of water plants (Abu Bakar et al., 2010). Oxytetracycline, tetracycline, quinolones, sulphonamides, trimethoprim, nalidixic acid, gentamicin, nitrofurazone, and trimethoprim-sulfamethoxazole are among the permitted antibiotics used in the Asian aquaculture industry (Harikrishnan et al., 2011; Manjusha and Sarita, 2011; Rico et al., 2012; Yano et al., 2014). Extensive use of antibiotics in aquaculture has resulted in the increase antibiotic resistance among bacteria including Vibrio species (Tendencia and de la Peña, 2001; Jerbi et al., 2011; Heng et al., 2017; Lee and Raghunath, 2018). Direct transmission of resistant bacteria through food to human, and transfer of resistance genes to other bacteria happens, thus causing a possible hazard to human wellbeing (Duran and Marshall, 2005; Guglielmetti et al., 2009; Kim et al., 2013).

Antimicrobial resistance (MDR) has been recognized as an important global threat issue to global public health and food safety (Food and Agriculture Organization [FAO], 2016). In hospitals, many clinical antibiotics are no longer effective to control bacterial infections (Tan et al., 2016). As a result of misuse of antibiotic to control infections during aquaculture production, V. parahaemolyticus has been reported to exhibit multidrug resistance, which raised the concern about public health and economic threat of this bacterium (Vaseeharan et al., 2005; Han et al., 2007; Lesley et al., 2011; Manjusha and Sarita, 2011; Noorlis et al., 2011). Carbapenems are always been regarded as the last treatment selection for Gram-positive and Gram-negative infections, and as well as infections caused by multidrug resistant bacteria (Nordmann et al., 2011; Martin et al., 2018). Nevertheless, their use has been compromised causing an increased incidence of carbapenem-resistant bacteria, and widely been discussed among medical practitioners, researchers, and public (Martin et al., 2018). A study by Nordmann and colleagues identified the novel New Delhi metallo-β-lactamase (NDM) encoded by the gene blaNDM−<sup>1</sup> in members of the family Enterobacteriaceae. This gene was reported to be not only present largely in Enterobacteriaceae, but also in Vibrionaceae (Nordmann et al., 2011). Over the years, Carbapenem-resistant Vibrio sp. has been detected and isolated from environmental and seafood samples (Walsh et al., 2011; Mandal et al., 2012; Gu et al., 2014; Bier et al., 2015). Recently, in Kolkata, NDM-1 producing Vibrio fluvialis strains has been isolated from diarrheal fecal samples from patients (Chowdhury et al., 2016).

The increase in bacterial resistance toward many clinical antibiotics affects many countries healthcare sector and food production sectors (World Health Organization [WHO], 2014). In view of previous reports and the possible severity of infections, continuous investigation on antimicrobial resistance of V. parahaemolyticus is needed for epidemiological purpose and guidance in healthcare treatment. For this reason, our study aimed to assess antimicrobial susceptibility profiles of V. parahaemolyticus from marine and freshwater fish in Selangor, Malaysia. In addition, we also report the identification and antibiotic resistant characterization of Carbapenem-resistant isolates isolated from marine and freshwater fish samples. To our knowledge, this is the first study examining the antibiotic resistant profiles and Carbapenem-resistant isolates of V. parahaemolyticus from both marine and freshwater fish samples in Selangor, Malaysia.

### MATERIALS AND METHODS

fmicb-09-02513 October 23, 2018 Time: 14:30 # 3

### Sampling

The study focused on two category fish – marine and freshwater fish. A total of 240 fish samples comprising of yellowstripe scad (Selaroides leptolepis) (n = 48), Indian mackerel (Rastrelliger kanagurta) (n = 48), black pomfret (Parastromateus niger) (n = 48), catfish (Clarias batrachus) (n = 48), and red tilapia (Oreochromis spp.) (n = 48) were collected from three wet markets and three supermarkets in Selangor (**Table 1**). From each sampling site, we collected eight fish samples and sampling was done weekly from January 2016 to May 2016. All the samples were kept in sterile sealed bags and transported to the laboratory in an ice box. Samples were analyzed immediately thereafter.

### Isolation of Vibrio sp. in Fish Samples

Isolation of Vibrio sp. was following Standard US Food and Drug Administration (FDA) protocol (Kaysner and DePaola, 2004) and FAO/WHO Risk Assessment of V. parahaemolyticus in Seafood (Food and Agriculture Organization/World Health Organization [FAO/WHO], 2011); this method was previously reported by Zarei et al. (2012) and Letchumanan et al. (2015a). 25 g of sample (gut and fish meat) was homogenized with 225 mL of alkaline peptone water (APW) with 2% w/v sodium chloride (NaCl), pH 8.5 for 60 s using a stomacher (BagMixer 400W, Interscience, Saint-Nom-la-Bretèche, France). The homogenate was enriched at 37◦C for 18 h. After 18 h of incubation, a loopful of enriched mixture was streaked onto selective media, Thiosulfate Citrate Bile Salts Sucrose (TCBS) agar (HiMedia, India) and incubated at 37◦C for 18 h. In each plate, one sucrose non-fermenting colony that has a green or bluish green color measuring about 3–5 mm suggestive of V. parahaemolyticus was selected from the TCBS plates. The isolate was purified by re-streaking onto Tryptic Soy Agar (TSA) (HiMedia, India) plates supplemented with 2% w/v sodium chloride (NaCl) (Vivantis, United States). The purified colony were inoculated into TSB semi-solid nutrient agar and stored until further identification.

## DNA Extraction

Bacterial lysate was prepared following established protocol (Suzita et al., 2010; Vengadesh et al., 2012; Letchumanan et al., 2015a,c). The isolates were revived in tryptic soy broth (TSB) (HiMedia) supplemented with 2% w/v sodium chloride NaCl (Vivantis, United States). Overnight suspension was transferred into 1.5 mL of microcentrifuge tube and centrifuged. The supernatant was discarded and 1 mL of sterile ultrapure water was added and vortexed. The suspension was heated at 100◦C for 7 min and then cooled on ice immediately into ice for 5 min. Cell debris from the cell lysate were pelleted by centrifugation at 13,000 rpm for 1 min. The supernatant was used as DNA templates for polymerase chain reaction (PCR) assays.

## Identification of Vibrio parahaemolyticus Using toxR-PCR Assay

Specific primers targeting toxR gene with the expected amplicon size of 368 bp were used to identify V. parahaemolyticus (Kim et al., 1999; Letchumanan et al., 2015a). The PCR assay was performed in 20 µL reaction mixture containing 2 µL of DNA template, 10 µL of 2× Taq PLUS PCR Smart mix 1 (SolGentTM, South Korea), 6 µL of ultrapure water and 1 µL of each primer. toxR-based PCR amplification was performed using PCR thermocycler (Kyratec, Super Cycler Thermal Cycler, Australia) with the following cycling conditions: initial denaturation at 95◦C for 4 min, 35 cycles of 94◦C for 1 min, 68◦C for 1 min and 72◦C for 30 s, and a final elongation at 72◦C for 5 min. PCR products was visualized by using 1.5% agarose gel and viewed under UV transilluminator using a Gel Documentation System (ChemiDocTM XRS, Bio-Rad, United States). The toxR-PCR results of a few presumptive V. parahaemolyticus isolates and type strain Vibrio parahaemolyticus NBRC 12711 were sequenced to confirm the identity of toxR gene (**Supplementary Table S1**). The Vibrio parahaemolyticus NBRC 12711 was used as the positive control and Vibrio vulnificus NBRC 15645 was used as the negative control.

### Detection of Virulence Gene

Molecular identification of thermostable direct haemolysin (tdh) and thermostable-related direct haemolysin (trh) was performed using a duplex PCR assay (Bej et al., 1999; Letchumanan et al., 2015c). The PCR assay was done in 20 µL of reaction mixture containing 2 µL of DNA template, 10 µL of 2× Taq PLUS PCR Smart mix 1 (SolGentTM, South Korea), 4 µL of sterile distilled water and 1 µL of each primer. The PCR amplifications was performed using a Thermocycler (Kyratec, Super Cycler Thermal Cycler, Australia) with the following cycling conditions: initial denaturation at 94◦C for 3 min, 30 cycles of 94◦C for 1 min, 58◦C for 1 min and 72◦C for 1 min, and a final elongation at 72◦C for 5 min. The PCR products was visualized by using 1.5% agarose gel and viewed under UV transilluminator using a Gel Documentation System (ChemiDocTM XRS, Bio-Rad, United States). The PCR results of a few presumptive V. parahaemolyticus isolates and type strain Vibrio parahaemolyticus NBRC 12711 were sequenced to confirm the identity of virulence gene (**Supplementary Table S1**). Vibrio parahaemolyticus NBRC 12711 was used as the positive control and Vibrio vulnificus NBRC 15645 was used as the negative control.

## Antibiotic Susceptibility Test

The antibiotic susceptibility of V. parahaemolyticus isolates was determined using Kirby-Bauer disk diffusion method (Yano et al., 2014). Fourteen type of antibiotics disks (Oxoid, United Kingdom) was used: amplicon (10 µg), ampicillin/sulbactam (30 µg), amikacin (30 µg), cefotaxime (30 µg), ceftazidime (30 µg), chloramphenicol (30 µg), gentamicin (30 µg), imipenem (10 µg), kanamycin (30 µg), levofloxacin (5 µg), nalidixic acid (30 µg), oxytetracycline (30 µg), sulfamethoxazole/trimethoprim (25 µg), and

tetracycline (30 µg). E. coli ATCC 25922 with known sensitivity pattern was included as a positive control in each test.

V. parahaemolyticus isolates was grown in tryptic soy broth (TSB) (HiMedia, India) 2% w/v sodium chloride (NaCl) (Vivantis, United States) at 37◦C for 18 h under constant agitation. The antibiotic disks were dispensed on Mueller Hilton agar (HiMedia, India) supplemented with 2% w/v NaCl (Vivantis, United States) plates with bacterial lawn. After incubation at 37◦C for 18 h, the inhibition zone was measured and interpreted based on guidelines of the Clinical and Laboratory Standards Institute (CLSI) M45-A2 (Clinical and Laboratory Standards Institute [CLSI], 2010). The multiple antibiotic resistance (MAR) index was determined based on the ratio of antibiotic resistance exhibited by the isolate to the number of antibiotics to which the isolates were exposed (Krumperman, 1983).

### Plasmid Curing

The 19-imipenem resistant V. parahaemolyticus isolate was subjected to plasmid curing assay to determine the antibiotic resistance mediation. The plasmid curing assay was performed using an intercalating agent, ethidium bromide (EB) (Lou et al., 2002; Molina-Aja et al., 2002; Letchumanan et al., 2015b). The isolates were revived in freshly prepared tryptic tryptic soy broth (TSB) supplemented with 0.2 mg/mL EB (Bio Basic, Canada), then incubated at 37◦C for 18 h under constant agitation. After treatment with the curing agent, the antibiotic resistance profiles were re-examined and compared with the antibiotic resistance phenotype on non-treated group.

### Genomic and Phylogenetic Analyses

Polymerase chain reaction amplification of the 16s rRNA gene for the 19-imipenem resistant V. parahaemolyticus was done according to the protocol described by Thomas et al. (2018) with slight modifications. The 16S rRNA gene sequence of each isolate was aligned with representative sequences of related type strains in the genus V. parahaemolyticus retrieved from the GenBank/EMBL/DDBJ databases using CLUSTAL-X software (Thompson et al., 1997). The alignment was first verified manually and adjusted, followed by construction of phylogenetic trees with neighbor-joining (Saitou and Nei, 1987; **Figure 1**) and maximum-likelihood algorithms (Felsenstein, 1981), utilizing the MEGA version 6.0 (Tamura et al., 2013). For neighborjoining algorithm, the evolutionary distances were computed using the Kimura's two-parameter model (Kimura, 1980). The calculations of level of sequence similarity were performed by GenBank server<sup>1</sup> . Bootstrap based on 1,000 resampling method of Felsenstein (1985) was used to analyze the stability of the resultant tree topologies.

### Statistical Analysis

Data analysis was performed with SPSS statistical analysis software version 20. Statistical analysis was performed in order to determine whether there was any significant difference in between two types of fish (marine and freshwater fish) and the MAR index of resistant isolates using the independent t-test. The significance level was set at p ≤ 0.05. Oneway analysis of variance (ANOVA) followed by appropriate post hoc text (Tukey) was performed to determine the significant differences between the type of fishes and MAR index of resistant isolates. A difference was considered statistically significant when p ≤ 0.05.

## RESULTS

### Prevalence of Vibrio parahaemolyticus in Fish Samples

The present study isolated V. parahaemolyticus from freshwater and marine fish. A total of 240 fish samples comprising of yellowstripe scad (Selaroides leptolepis) (n = 48), Indian mackerel (Rastrelliger kanagurta) (n = 48), black pomfret (Parastromateus niger) (n = 48), catfish (Clarias batrachus) (n = 48), and red tilapia (Oreochromis spp.) (n = 48) were collected from three wet market and three supermarkets. Based on the colony morphology on TCBS agar, a total of 240 isolates was picked and purified on TSA agar. The toxR-PCR assay exhibited positive amplification of toxR gene with 368 bp amplicon band in 69% (165/240) of the presumptive V. parahaemolyticus isolates. Based on the sampling location site, 47% (78/165) of the isolates originated from the wet market

<sup>1</sup>https://blast.ncbi.nlm.nih.gov



n = number of fish samples purchased from respective location. toxR<sup>+</sup> = number of positive Vibrio parahaemolyticus isolates harboring toxR gene. trh<sup>+</sup> = number of positive Vibrio parahaemolyticus isolates harboring trh gene.

and 53% (87/165) was from supermarket. A total of 96 (58%) of the isolates were isolated from marine fish samples and 69 (42%) of the isolates were isolated from freshwater fish samples.

### Detection of Thermostable Direct Hemolysin (tdh) and tdh-Related Hemolysin (trh)

A duplex PCR assay was performed to detect the presences of tdh and trh gene in all isolates (**Table 1**). None of the 165 V. parahaemolyticus isolates yielded tdh-positive PCR amplification. Only 4 (2.4%) out of the total 165 V. parahaemolyticus showed positive PCR amplification of the trh gene. The trh-positive V. parahaemolyticus isolates was isolated from black pomfret (wet market B) (FVP81), red tilapia (wet market B) (FVP92), and two from Indian mackerel (supermarket C) (FVP47 and FVP49). The presence of trhpositive V. parahaemolyticus isolates in both types of fish samples indicates possible high risk of foodborne gastroenteritis transmission to humans upon ingestion of the fish.

## Antimicrobial Susceptibilities of Vibrio parahaemolyticus Isolates

Most of the tested antibiotics in this study such as tetracycline, folate pathway inhibitors (trimethoprimsulfamethoxazole), third-generation cephalosporins (cefotaxime and ceftazidime), aminoglycosides (gentamicin and amikacin) and fluoroquinolones (ciprofloxacin and levofloxacin), are among the recommended antibiotics by CDC for the treatment of Vibrio sp. infections (Daniels and Shafaie, 2000; Shaw et al., 2014). **Table 2** summarizes the percentage of antibiotic resistant profiles of V. parahaemolyticus isolated from fish sample. Based on the results, the resistance rate of the 165 V. parahaemolyticus isolates in our study was 88% to ampicillin, 64% to amikacin, and 50% to kanamycin. A notable resistance pattern can be observed to the third generation cephalosporins (cefotaxime 52% and ceftazidime 28%). In contrast, high susceptibility rate was seen to chloramphenicol (93%), tetracycline (90%), imipenem (85%), levofloxacin (85%), gentamicin (84%), sulfamethoxazole/trimethoprim (80%), nalidixic acid (78%), oxytetracycline (72%), and ampicillin/sulbactam (70%).

TABLE 2 | Percentage of antibiotic susceptible, intermediate, and resistant of V. parahaemolyticus isolated from various fish samples.


<sup>a</sup>% = percentage (number of isolates/total number of isolates tested).

Interestingly, 19 isolates (12%) from this study exhibited resistance to imipenem, an antibiotic in Carbapenem class. The detection of imipenem resistant isolates is of concern as Carbapenems are among the beta-lactams that is the last line antibiotic used for bacterial infection treatment (Meletis, 2016). These 19 isolates had an MAR index of 0.14 to 0.50, resistant to more than two different type of antibiotics tested. Majority of the imipenem resistant isolates were isolate from freshwater fish sample (15/19) and the remaining 4 isolates were isolated from marine fish samples.

In this study, the values of MAR index ranged from 0.00 to 0.57 (**Table 3**). Forty-two different resistance patterns had a significant MAR index more than 0.2. Two of the isolates (FVP24 – yellowstripe scad, marine fish and FVP67 – red tilapia, freshwater fish) has the highest MAR index of 0.57, resistant to 8/14 antibiotics tested. Further analysis was performed by comparing the MAR index between source of sample (marine and freshwater) and MAR index. The mean MAR index of marine fish sample was 0.26 where else, freshwater fish sample was 0.25. The results showed that there was no significant difference between source of fish sample and MAR index. According to the one-way ANOVA analysis, there was no significant difference between the fish types on the MAR index of V. parahaemolyticus isolates. The results suggest that isolates from the fish samples may have similar level of antibiotic exposure, regardless there are marine or freshwater originated. As shown in **Figure 2**, 8% of V. parahaemolyticus isolates (13 isolates) did not exhibited MAR as they were susceptible to all of the antibiotics tested.

The antibiotic resistance patterns between freshwater and marine fish samples did not exhibit any significant profiles. Based on the analysis, both freshwater and marine fish samples were exposed to antibiotics and phenotypic assay showed a similar resistant profile to 0 to 8 types of antibiotics tested. For each fish sample, the mean MAR indices for V. parahaemolyticus isolates was 0.26 for yellowstripe scad, Indian mackerel was 0.24, black pomfret was 0.29, catfish was 0.25, and red tilapia was 0.24.

### Plasmid Curing

Plasmid curing may server as an effective assay to determine the antibiotic resistance mediation of bacteria. This assay enables to eliminate desired bacterial plasmid and subsequently reassess the antibiotic resistance phenotype by antibiotic disk diffusion method. **Table 4** summarizes the antibiotic resistance profile of the 19-imipenem resistant isolates before and after plasmid curing assay. All the phenotypically seen imipenem resistant isolates became susceptible to imipenem after curing assay, suggesting the resistance was plasmid mediated. All the isolates were still resistant to ampicillin and oxytetracycline, suggesting a possible chromosomal mediated resistance. Hence, the antibiotic resistance seen in 19-imipenem isolates are both plasmid and chromosomally mediated.

### Genomic and Phylogenetic Analyses

The nearly complete 16S rRNA gene sequence was determined for all the 19-imipenem resistant V. parahaemolyticus isolates and manual alignment of these sequences was performed

TABLE 3 | Antibiogram and multiple antimicrobial resistance (MAR) indices of V. parahaemolyticus isolates.


TABLE 3 | Continued


AMP, amplicon; OT, oxytetracycline; NA, nalidixic acid; C, chloramphenicol; CTX, cefotaxime; SXT, sulfamethoxazole/trimethoprim; IMP, imipenem; AK, amikacin; SAM, ampicillin/sulbactam; LEV, levofloxacin; CAZ, ceftazidime; K, kanamycin; CN, gentamicin; TE, tetracycline.

with the corresponding partial 16S rRNA gene sequences of the type strains of V. parahaemolyticus retrieved from GenBank/EMBL/DDBJ databases. Phylogenetic tree was constructed based on the 16S rRNA gene sequences to determine the phylogenetic position of the 19-imipenem resistant isolates (**Figure 1**). Phylogenetic analysis exhibited that closely related strains include Vibrio parahaemolyticus ATCC 17802 (NR 119058.1), Vibrio parahaemolyticus NBRC 12711 (NR 113604.1) and Vibrio parahaemolyticus ATCC 17802 (NR 114630.1), as the 19-imipenem resistant isolates form distinct five clades. The isolates within the same clade are closely related.

### DISCUSSION

The occurrence of pathogenic strains of V. parahaemolyticus in fish samples we studied does raise concern as this organism is known to cause foodborne gastroenteritis resulted from ingesting of uncooked or undercooked seafood (Ma et al., 2014; Romalde et al., 2014). However, while the microbiological culture-based method found all fish samples to be contaminated Vibrio sp., only 69% (165/240) of there were confirmed to be V. parahaemolyticus based on toxR PCR assay; and only 2.4% (4/165) of these were pathogenic strains (trh-positive) (**Table 1**). Our results came to an agreement with other researchers on the fact that the identity of V. parahaemolyticus could not


#### TABLE 4 | List of 19-imipenem resistant Vibrio parahaemolyticus isolates.

AMP, amplicon; OT, oxytetracycline; NA, nalidixic acid; C, chloramphenicol; CTX, cefotaxime; SXT, sulfamethoxazole/trimethoprim; IMP, imipenem; AK, amikacin; SAM, ampicillin/sulbactam; LEV, levofloxacin; CAZ, ceftazidime; K, kanamycin; CN, gentamicin; TE, tetracycline.

be fully confirmed by conventional microbiological culturebased method (Kim et al., 1999; Zulkifli et al., 2009; Fabbro et al., 2010; Ottaviani et al., 2013). Affirming with previous research, we found that toxR PCR assay was specific and reliable technique for the identification of both pathogenic and nonpathogenic V. parahaemolyticus (Kim et al., 1999; Dileep et al., 2003; Zulkifli et al., 2009). This reliable and specific toxR-PCR assay has resulted in many promising V. parahaemolyticus identifications studies (Deepanjali et al., 2005; Das et al., 2009; Vimila et al., 2010; Elamparithi and Ramanathan, 2011; Noorlis et al., 2011; Paydar et al., 2013). The remaining 75 isolates had the morphology of V. parahaemolyticus in TCBS agar, however, the toxR gene was not present in these isolates. This result demonstrates the detection of V. parahaemolyticus thru toxR PCR assay is highly sensitive, specific and accurate compared to microbiological culture-based technique (Mandal et al., 2011).

The tdh and trh genes are considered major virulence factors in V. parahaemolyticus, so in many clinically isolated strains of V. parahaemolyticus have hemolytic activity that is produced by these two genes (Ceccarelli et al., 2013; Raghunath, 2015). Our study reported the isolation of trh-positive isolates of V. parahaemolyticus at a very low prevalence rate, and none of the isolates have tdh-position genes. Our results follow the trends of worldwide dispersed studies that have reported low number of virulent V. parahaemolyticus strains from environmental sources (Fuenzalida et al., 2006; Nordstrom et al., 2007; Thongjun et al., 2013). Many studies have reported low prevalence rate (less than 5%) of environmental and food source have pathogenic V. parahaemolyticus isolates carrying tdh and/or trh genes (Parveen et al., 2008; Zulkifli et al., 2009; Tsai et al., 2013). In addition, it is strongly suggested that putative pathogenic environmental V. parahaemolyticus isolates may be less virulent than the clinical V. parahaemolyticus isolates (Vongxay et al., 2008; Tsai et al., 2013). The presences of tdh+ and/or trh+ V. parahaemolyticus in the marine and freshwater fish samples in Selangor is of concern due to several factors. Firstly, the fact that these pathogenic isolates could potentially cause gastroenteritis (Jun et al., 2012). Secondly, pathogenic V. parahaemolyticus not only contaminate seafood and transmit pathogenesis, but it also causes huge economic loss in the aquaculture sector (Fuenzalida et al., 2006; Thongjun et al., 2013). Hence, the study results need the importance for continuous monitoring of seafood for pathogen contamination.

Our antibiotic susceptibility test placed ampicillin at the top of the V. parahaemolyticus resistance scope (88%). This finding signifying that ampicillin may longer be an effective antibiotic to treat Vibrio sp. infections. In fact, V. parahaemolyticus resistance to ampicillin is well reported in many literatures (Joseph et al., 1978; Lesmana et al., 2001; Zulkifli et al., 2009; Melo et al., 2011; Oh et al., 2011; Al-Othrubi et al., 2014). Interestingly, ampicillin resistance was reported 100% in study by Devi et al. (2009) and Ottaviani et al. (2013). The chromosomally encoded β-lactamase is the cause for V. parahaemolyticus resistance to ampicillin and other penicillin (Devi et al., 2009). In addition, more that 70% of the V. parahaemolyticus isolates in this study remained susceptible to tetracycline, levofloxacin, gentamicin, sulfamethoxazole/trimethoprim, chloramphenicol, imipenem, nalidixic acid, oxytetracycline, and

ampicillin/sulbactam. Our findings are in line with previous studies that reported susceptibility of V. parahaemolyticus against chloramphenicol, tetracyclines, trimethoprim-sulfamethoxazole, nalidixic acid, and imipenem (Ottaviani et al., 2001; Devi et al., 2009; Melo et al., 2011; Al-Othrubi et al., 2014; Shaw et al., 2014). The MAR index values ranged from 0 to 0.57.

Forty-two different resistance patterns had a significant MAR value >0.2. Collectively, there were expressed by 70% of the V. parahaemolyticus isolates and resistant to 3 to 8 types of antibiotics tested. MAR index >0.2 are exposed to several antibiotics or isolated from contaminated sources as such dairy cattle, aquaculture, and agriculture farms. Where else, isolates with lesser than 0.2 MAR indices are lessen prone to antibiotic exposure (Noorlis et al., 2011; Subramani and Vignesh, 2012). In this study there was no significant difference been observed among the source of sample and MAR index. This result demonstrates that the isolates isolated from marine and freshwater samples are exposure of antibiotics. Our results came to an agreement with many studies that reported high percentage of V. parahaemolyticus isolated from seafood are resistant to more than one antibiotic tested (Zulkifli et al., 2009; Lesley et al., 2011; Manjusha and Sarita, 2011; Noorlis et al., 2011).

Imipenem, a member of Carbapenem class is an effective antibiotic used in the treatment of Gram-positive and Gram-negative infections (Papp-Wallace et al., 2011). Interestingly, in this study we detected 19-imipenem resistant V. parahaemolyticus strains isolated from marine and freshwater fish samples. Ever since the first detection case of carbapenemase producing Carbapenem-Resistant Enterobacteriaceae (Cp-Cre) in the United States, Cp-Cre have rapidly spread with more reported cases in another 50 states (Centers for Disease Control and Prevention [CDC], 2018). In fact, now carbapenem resistance is no longer associated with Enterobacteriaceae but also associated with other bacteria. As such, the resistance of Vibrio sp. to carbapenem has been reported by Bier et al. (2015) in Germany coastal line, Gu et al. (2014) in Southwest China, and Walsh et al. (2011) in India. Thus, our results agree with other findings and demonstrates the misuse of carbapenem that may cause a negative impact on the clinical treatment of Vibrio infections in future. Hence, a non-antibiotic approach is required in order to manage the occurrence of antibiotic resistance among Vibrio sp. in the environments (Tan et al., 2014; Letchumanan et al., 2016; Tan et al., 2016).

Further analysis on the 19-imipenem resistant isolates by plasmid curing assay exhibited interesting findings. The antibiotic resistance phenotype of these 19 isolates have been altered after plasmid curing. All the 19 isolate's phenotypically seen resistance to imipenem has changed to susceptible, suggesting the resistance was plasmid mediated. All the isolates were still resistant to ampicillin, suggesting the resistance was chromosomal mediated. In addition, the isolate FVP92, FVP96, FVP98, FVP100, FVP107, FVP112, FVP131, FVP154, FVP159 (**Table 3**) remained resistant to oxytetracycline even after plasmid curing and its chromosomally mediated. It is usual to find oxytetracycline resistant isolates from aquaculture products because this antibiotic is among the permitted antimicrobial used in the seafood production. In summary, plasmids are transferable between different bacteria and the presence of antibiotic resistant genes in the bacterial plasmid have facilitated the fast spreading of antibiotic resistance among bacteria (Wilson and Salyers, 2003; Stepanauskas et al., 2006; Manjusha and Sarita, 2011). Hence, the acquisition of imipenem resistance by the 19 isolates are possibly due to horizontal gene transfer from other environmental bacteria.

The results from phylogenetic and genomic analyses indicated that the 19-imipenem resistant isolates are closely related forming five clades (**Figure 1**). The isolates are closely related to each another within the same clade. Isolate FVP28 was closely related to Vibrio parahaemolyticus NBRC 12711 and Vibrio parahaemolyticus ATCC 17802 at 99% bootstrap value, indicating the high confident level of the association. The isolate FVP28 and both type strains are isolated from food source. The FVP28 was isolated from freshwater red tilapia where else, Vibrio parahaemolyticus Nbrd 12711 and Vibrio parahaemolyticus ATCC 17802 was originally isolated from shirasu food poisoning case in Japan. This result exhibits a close relationship between these strains isolated from different types of seafood. In addition, there was another clade with nine isolates (FVP30, FVP112, FVP98, FVP100, FVP131, FVP159, FVP96, FVP107, and FVP154) that were closely related to Vibrio parahaemolyticus ATCC 17802 at 51% bootstrap value. 7/9 of the isolates (FVP30, FVP112, FVP98, FVP100, FVP159, FVP96, and FVP107) was isolated from freshwater fish sample. Majority of the isolates within this clade were resistant to oxytetracycline, an antibiotic that is permitted in Asian aquaculture industry (Yano et al., 2014). In summary, phylogenetic tree analysis revealed that there was no distinctive grouping based on the antibiogram of each isolate, however, the16S rRNA sequencing had a high discriminating power to group the isolates into different clades (Hoffmann et al., 2010).

The global increase of antibiotic resistant bacteria is of great public health concern and warrants a continuous monitoring (Xie et al., 2017). In the case of V. parahaemolyticus, the situation is aggravated due to excessive use of antimicrobial agents in aquaculture to protect infectious diseases and huge production loses (Xu et al., 2016). In addition, antimicrobial resistance is likely caused by exposure to antibiotics via agriculture runoff or wastewater treatment plants, and thru mobile genetic elements or horizontal gene transfers among bacteria (Stepanauskas et al., 2006; Kümmerer, 2009; Al-Othrubi et al., 2014; Xu et al., 2016). Recently, the Food and Agriculture Organization (FAO) have drawn action plans to increase awareness and promote prudent use of antimicrobials (Food and Agriculture Organization [FAO], 2015).

## CONCLUSION

Our study confirms the presences of V. parahaemolyticus in freshwater and marine fish samples in Selangor by having use highly accurate detection and identification method (the combination of microbiological culture-based method and PCR).

To our best knowledge, this study represents the first evidence of Carbapenem resistant isolate and as well as antibiotic resistance patterns of V. parahaemolyticus isolated from freshwater and marine fish samples. The detection of tdh and trh genes provides better understanding regarding the distribution of pathogenic V. parahaemolyticus strains in fish samples. Despite the fact that majority most of the environmental V. parahaemolyticus isolates are non-pathogenic, consumer should still be aware and ensure that fish is cooked properly before consumption. Adequate cooking of fish before consumption is the main safety measure to prevent foodborne disease caused by V. parahaemolyticus associated with fish (Zulkifli et al., 2009). Furthermore, several important measures including good hygiene practices while handling the fish and the cleanliness of the handlers and display area are very crucial in order to prevent cross-contamination in wet market and supermarket. In conclusion, the information presented serves as a baseline on future microbiological risk assessment of V. parahaemolyticus associated with fish consumption in Selangor, Malaysia.

### REFERENCES


### AUTHOR CONTRIBUTIONS

L-HL, VL, and JL conducted the experiments and data analysis, and wrote the manuscript. N-SAM and SW provided vital insight, technical support, guidance, and proofreading for the project. L-HL founded the project.

### FUNDING

This work was supported by PVC Award Grant (Project No. PVC-ECR-2016) and External Industry Grant (Biotek Abadi – Vote Nos. GBA-808138 and GBA-808813) awarded to L-HL.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.02513/full#supplementary-material


and Vibrio parahaemolyticus from freshwater fish at retail level. Int. Food Res. J. 18, 1523–1530.



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

Copyright © 2018 Lee, Ab Mutalib, Law, Wong and Letchumanan. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Prevalence, Serotyping, Molecular Typing, and Antimicrobial Resistance of Salmonella Isolated From Conventional and Organic Retail Ground Poultry

, Usama H. Abo-Shama1,2, Katherine K. Harclerode<sup>1</sup> and

#### Edited by:

Ahmed H. Gad<sup>1</sup>†

Mohamed K. Fakhr<sup>1</sup>

\*

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Jianmin Zhang, South China Agricultural University, China Beatrix Stessl, Veterinärmedizinische Universität Wien, Austria

\*Correspondence:

Mohamed K. Fakhr mohamed-fakhr@utulsa.edu

†Present address:

Ahmed H. Gad, Northeastern State University, Broken Arrow, OK, United States

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 15 June 2018 Accepted: 17 October 2018 Published: 05 November 2018

#### Citation:

Gad AH, Abo-Shama UH, Harclerode KK and Fakhr MK (2018) Prevalence, Serotyping, Molecular Typing, and Antimicrobial Resistance of Salmonella Isolated From Conventional and Organic Retail Ground Poultry. Front. Microbiol. 9:2653. doi: 10.3389/fmicb.2018.02653 <sup>1</sup> Department of Biological Science, The University of Tulsa, Tulsa, OK, United States, <sup>2</sup> Microbiology and Immunology Department, Faculty of Veterinary Medicine, Sohag University, Sohag, Egypt

Ground poultry is marketed as a healthier alternative to ground beef despite the fact that poultry is a major source of foodborne Salmonella. The objectives of this study were to determine the prevalence of Salmonella in Oklahoma retail ground poultry and to characterize representative isolates by serotyping, antimicrobial resistance, PFGE patterns, and large plasmid profiling. A total of 199 retail ground poultry samples (150 ground turkey and 49 ground chicken) were investigated. The overall prevalence of Salmonella in ground poultry was 41% (82/199), and the incidence in conventional samples (47%, 66/141) was higher than in organic samples (27%, 16/58). The prevalence of Salmonella in organic ground chicken and organic ground turkey was 33% (3/9) and 26% (13/49), respectively. Twenty six Salmonella isolates (19 conventional and 7 organic) were chosen for further characterization. The following six serotypes and number of isolates per serotype were identified as follows: Tennessee, 8; Saintpaul, 4; Senftenberg, 4; Anatum, 4 (one was Anatum\_var.\_15+); Ouakam, 3; and Enteritidis, 3. Resistance to 16 tested antimicrobials was as follows: gentamycin, 100% (26/26); ceftiofur, 100% (26/26); amoxicillin/clavulanic acid, 96% (25/26); streptomycin, 92% (24/26); kanamycin, 88% (23/26); ampicillin, 85% (22/26); cephalothin, 81% (21/26); tetracycline, 35% (9/26); sulfisoxazole, 27% (7/26); nalidixic acid, 15% (4/26); and cefoxitin, 15% (4/26). All isolates were susceptible to amikacin, chloramphenicol, ceftriaxone, and trimethoprim/sulfamethoxazole. All screened isolates were multidrug resistant (MDR) and showed resistance to 4–10 antimicrobials; isolates from organic sources showed resistance to 5–7 antimicrobials. PFGE was successful in clustering the Salmonella isolates into distinct clusters that each represented one serotype. PFGE was also used to investigate the presence of large plasmids using S1 nuclease digestion. A total of 8/26 (31%) Salmonella isolates contained a ∼100 Kb plasmid that was present in all Anatum and Ouakam isolates. In conclusion, the presence of multidrug

**77**

resistant Salmonella with various serotypes, PFGE profiles, and large plasmids in ground poultry stresses the importance of seeking novel interventions to reduce the risk of this foodborne pathogen. Multidrug resistance (MDR) is considered a high additional risk and continued surveillance at the retail level could minimize the risk for the consumer.

Keywords: Salmonella, serotyping, antimicrobial resistance, PFGE, plasmids, ground poultry

### INTRODUCTION

fmicb-09-02653 November 1, 2018 Time: 15:9 # 2

Nontyphoidal Salmonella spp. is the primary bacterial pathogen causing foodborne illness and the leading cause of hospitalization among the top five foodborne pathogens in the United States (Scallan et al., 2011). Contaminated meats are the major foodborne sources of Salmonella, which has been recovered and characterized from retail beef, pork, bison, chicken, and turkey meats in several countries worldwide (Li et al., 2006; Cetinkaya et al., 2008; Nde et al., 2008; Ammari et al., 2009; Yang et al., 2010; Tafida et al., 2013; Maka et al., 2014; Sallam et al., 2014; Aslam et al., 2012; Soufi et al., 2012; Thai et al., 2012). Consumption of ground poultry has increased in the last few years, partially because it is marketed as a healthier alternative to ground beef. However, ground poultry, particularly ground turkey and chicken, is often contaminated with Salmonella (White et al., 2001; Fakhr et al., 2006a; Erol et al., 2013; Cui et al., 2015). A large, multistate-outbreak caused by an antimicrobial-resistant Salmonella enterica subsp. enterica serovar Heidelberg occurred in 2011 from the consumption of contaminated ground turkey and resulted in one death (Folster et al., 2012). Three other multistate-outbreaks caused by Salmonella Heidelberg occurred between 2013 and 2014 that were linked to chicken consumption (CDC, 2013a, 2014a,b; Gieraltowski et al., 2016). An outbreak of Salmonella enterica serovar Stanley infections associated with turkey meat was reported in 10 European countries between 2011 and early 2013 (Kinross et al., 2014). Comparative genomic analysis using Whole Genome Sequencing revealed that the S. Heidelberg isolates in the 2011 ground turkey outbreak clustered together when compared to isolates from human, animal, and retail meat sources (Hoffmann et al., 2014). Using an experimental oral challenge experiment in turkey, a recent study showed that the Salmonella isolate causing the 2011 outbreak was high in cecal colonization, dissemination to internal organs, and tissue deposition (Nair et al., 2018). Recently, a food-grade essential oil from pimento leaves was shown to reduce attachment of the 2011 S. Heidelberg isolate to turkey skin (Nair and Johny, 2017). By testing the host transcriptional response, a recent study showed that young commercial turkeys are susceptible to colonization by S. Heidelberg isolated from the 2011 ground turkey outbreak (Bearson et al., 2017).

The presence of antimicrobial resistant Salmonella in retail meats, particularly in poultry, is a major risk to the treatment of foodborne illnesses caused by this bacterial pathogen (Antunes et al., 2016; Chai et al., 2017). The presence of multidrug resistant (MDR), nontyphoidal Salmonella in retail meats has been reported in several studies (Cetinkaya et al., 2008; Zhao et al., 2008; Yildirim et al., 2011; Thai et al., 2012; Van et al., 2012; Maka et al., 2014; Yang et al., 2014; Iwamoto et al., 2017; Clothier et al., 2018). Most of the antimicrobial resistance genes in Salmonella are carried on conjugative plasmids that facilitate transfer between different isolates (Jones and Stanley, 1992; Rotgers and Casadesüs, 1999; Carattoli, 2003; Rychlik et al., 2006). Conjugation experiments showed that 95% of the β– lactamase genes (blaCMY) in Salmonella are plasmid-encoded (Folster et al., 2011). Quinolone resistance genes were also plasmid-borne in Salmonella isolated from human cases in the United States (Sjölund-Karlsson et al., 2010). Three emerging European clones of Salmonella enterica subsp. enterica serovar Typhimurium circulating in Europe were found to harbor MDR plasmids that encode additional virulence functions (García et al., 2014). Plasmid profiling is often used in epidemiological studies related to surveillance of disease outbreaks and in tracing the dissemination of antibiotic resistance (Mayer, 1988).

Pulsed field gel electrophoresis (PFGE) is considered the gold standard in typing Salmonella and is known for its ability to discriminate isolates and for tracking the source of outbreaks (Tenover et al., 1995; Fakhr et al., 2005; Foley et al., 2006, 2009; Folster et al., 2012). PFGE profiling has been used with relative success as a method to identify Salmonella serotypes (Gaul et al., 2007; Zou et al., 2010). A meta-analysis of PFGE fingerprints based on a constructed Salmonella database of 45,923 PFGE patterns indicated the presence of serotype-specific patterns that may potentially reduce the need to perform the laborious, traditional serotyping (Zou et al., 2013).

Despite the risk associated with the consumption of ground poultry contaminated with Salmonella, studies investigating the prevalence and characterization of Salmonella in retail ground poultry are relatively scarce. The objectives of this study were to determine the prevalence of Salmonella in retail ground poultry sold in the Tulsa, Oklahoma area and to characterize a selected number of the recovered strains by serotyping, antimicrobial resistance screening, plasmid profiling, and PFGE.

### MATERIALS AND METHODS

### Bacterial Sampling and Identification

Conventional methods were used to isolate Salmonella from ground turkey as described previously (Fakhr et al., 2006a; Nde et al., 2008). In the summer of 2009, 199 samples of ground poultry meat (150 and 49 from turkey and chicken, respectively) representing five brands were purchased at six retail stores representing six supermarket chains in Tulsa, Oklahoma. Ground poultry samples were stored in chilled containers, and transported to the laboratory within 4 h. Each sample (25 g)

was subjected to a pre-enrichment step by combining it with 225 mL of Buffered Peptone Water (BPW) (EMD, Gibbstown, NJ, United States) in sterile plastic bags (VWR Scientific, Radnor, PA, United States); the samples was massaged briefly by hand for 5 min. The pre-enrichment rinsate was then incubated at 37◦C for 24 h. To selectively enrich for Salmonella, 0.1 and 0.5 mL of each pre-enrichment broth sample was transferred to 10 mL of Rappaport-Vassiliadis broth (RVB; Difco, Becton Dickinson, Sparks, MD, United States) and tetrathionate broth (TTB; Difco, Becton Dickinson, Sparks, MD, United States), respectively, and incubated at 42◦C for 24 h. The pre-enrichment broths of duplicate samples were then artificially spiked with 10 µL of an overnight broth of two Salmonella strains (one H2S-positive and one H2S-negative); these served as positive controls. After selective enrichment was completed, a loopful of broth contains each of the enriched samples, including the two artificially-spiked Salmonella positive controls, were inoculated by dilution-streaking onto two selective agar media, XLT4 (Difco, Becton Dickinson, Sparks, MD, United States) and Brilliant Green Sulfide (BGS) (Difco, Becton Dickinson, Sparks, MD, United States) and incubated at 37◦C for 24 h. The identity of 4–6 suspected Salmonella colonies from each sample were confirmed biochemically by dilution streaking onto Triple Sugar Iron Agar (TSI) (Difco, Becton Dickinson, Sparks, MD, United States) and Lysine Iron Agar slants (Difco, Becton Dickinson, Sparks, MD, United States) and incubated at 37◦C for 24 h. Suspected Salmonella isolates were subjected to molecular confirmation by PCR using invA as described below.

The invA gene was amplified using the following PCR primers: forward, 5<sup>0</sup> **-** GTGAAATTATCGCCACGTTCGGGCAA-3<sup>0</sup> ; and reverse, 5<sup>0</sup> **-** TCATCGCACCGTCAAAGGAACC-3<sup>0</sup> as described previously (Rahn et al., 1992). PCR was conducted in 25 µL reaction volumes containing the following: 12.5 µL GoTaq <sup>R</sup> Green Master Mix (Promega, Madison, WI, United States), 3.5 µL sterile water (Promega, Madison, WI, United States), 1 µL (25 pmol) of each primer (IDT, Coralville, IA, United States), and 3 µL of template DNA. The cycling conditions were as follows: (1) 95◦C for 5 min; (2) 94◦C for 1 min; (3) 55◦C for 1 min; (4) 72◦C for 1 min; and (5) 72◦C for 10 min. Steps 2 through 4 were repeated for 35 cycles. PCR products were subjected to agarose gel electrophoresis, and a 1 kb plus DNA ladder (Bioneer, Alameda, CA, United States) was used as a molecular marker. Gel images were taken using a Bio-Rad Gel DocTM XR UV gel documentation system (Bio-Rad, Hercules, CA, United States). The presence of the 284 bp invA PCR product was considered to be positive for Salmonella molecular identification. Once confirmed as Salmonella, one isolate was kept as a representative for each ground poultry sample and further characterized by serotyping, antibiotic resistance profile, PFGE, and plasmid content.

### Serotyping

Salmonella isolates selected for serotyping were given a serial designation from GP001 to GP023 and from GP025 to GP027. Isolates were sent to the National Veterinary Service Laboratory (NVSL) in Ames, Iowa, for serotyping.

### Antibiotic Resistance Screening

Salmonella isolates were subjected to antimicrobial resistance profiling using the following 16 antimicrobials: cefoxitin (FOX), amikacin (AMI), chloramphenicol (CHL), tetracycline (TET), ceftriaxone (CTR), amoxicillin/clavulanic acid (AMC), ciprofloxacin (CIP), gentamycin (GEN), nalidixic acid (NAL), ceftiofur (TIO), sulfisoxazole (FIS), trimethoprim/sulfamethoxazole (SXT), cephalothin (CEP), kanamycin (KAN), ampicillin (AMP), and streptomycin (STR). Isolates were grown on Mueller-Hinton (MH) agar (Difco) and incubated for 24 h at 37◦C. Cultures were then added to Mueller-Hinton broth (Difco), and the turbidity was adjusted to a 0.5 McFarland standard, and inoculated onto 6-inch MH agar plates supplemented with the appropriate antimicrobials. Multiple antibiotic concentrations were tested including the breakpoint established for each antimicrobial according to the Clinical and Laboratory Standards Institute (CLSI) (Cockerill, 2011). The ranges of the concentrations used and the breakpoint of each of the 16 antimicrobials tested in this study were detailed previously (Fakhr et al., 2006b). Plates were then incubated at 37◦C for 48 h; results were read for growth or no growth and denoted as resistant or susceptible, respectively, according to the breakpoints for each antimicrobial.

### PFGE

Plug preparation for PFGE profiling was performed according to the PulseNet protocol and conditions established by the CDC (CDC, 2013b). Slices of the prepared PFGE plugs (2 mm wide) were incubated with XbaI (Promega, Madison, WI, United States) at a concentration of 50 U/plug for 3 h at 37◦C. Plug slices were then inserted into the wells of 1% Seakem Gold Agarose gels. XbaI-digested Salmonella serovar Braenderup H9812 was used as a sizing marker. PFGE was conducted in a CHEF Mapper PFGE system (Bio-Rad, Hercules, CA, United States) for 18 h following the electrophoresis conditions established for Salmonella by the PulseNet protocol; these included an initial switch time of 2.16 s, and a final switch time of 63.8 s (CDC, 2013b). After electrophoresis, gel images were captured using a Bio-Rad Gel DocTM XR UV gel documentation system (Bio-Rad, Hercules, CA, United States). Images were then imported and analyzed using the BioNumerics software v. 6.6 (Applied Maths, Austin, TX, United States). Similarity analysis and the banding patterns were analyzed using the Dice coefficient and clustered using the unweighted pair group method with arithmetic mean (UPGMA) and a 1.5% band tolerance.

### Plasmid Detection

Screening of large plasmids was performed by PFGE as described previously (Barton et al., 1995; Marasini and Fakhr, 2014). The PFGE plugs were prepared as described above; thin slices were cut and digested with S1 nuclease (17 IU/plug) for 45 min at 37◦C to linearize the plasmids. Plug slices were then inserted into the wells of 1% Seakem Gold Agarose gels, and XbaI-digested Salmonella serovar Braenderup H9812

was used as a sizing marker. PFGE was conducted using the CHEF Mapper PFGE system for 16 h using the conditions established for Salmonella by the PulseNet protocol (CDC, 2013b).

Large plasmids detected by PFGE were isolated by alkaline lysis using the Qiagen Miniprep kit and protocols established for Gram-negative bacteria (Qiagen Inc., Valencia, CA, United States). Isolated plasmids were analyzed by electrophoresis in 0.8% agarose gels at 120 V for 2 h. Gels were stained with ethidium bromide for 45 min, and images were captured using the Bio-Rad gel documentation system. DNA markers for sizing included plasmids preps of E. coli strains NCTC 50192 and NCTC 50193 and the 1 Kb plus DNA ladder (Bioneer). The isolated plasmids were also digested with EcoRI and HindIII (Promega, Madison, WI, United States) and subjected to agarose gel electrophoresis to determine variable restriction patterns.

### RESULTS

### Prevalence of Salmonella in Ground Poultry

A total of 199 retail ground poultry samples were investigated in this study (**Table 1**). Although only 6% (9/150) of the ground turkey samples were organic, all 49 ground chicken samples were organic. The overall prevalence of Salmonella in ground poultry was 41% (82/199), whereas the prevalence in conventional

TABLE 1 | Prevalence of Salmonella in ground poultry samples collected in this study.


<sup>∗</sup>np, number of positive samples; n, number of samples collected.

TABLE 2 | Ground poultry sources, serotypes and large plasmid profiles of the 26 Salmonella isolates characterized in this study.


<sup>∗</sup>Large plasmids in strains GP001, GP002, GP012, and GP023 share the same restriction pattern. ∗∗Large plasmids in strains GP009, GP013, and GP016 share the same restriction pattern.

samples (47%; 66/141) was higher than organic samples (27%; 16/58) **(Table 1**). The prevalence of Salmonella in organic ground chicken was 26% (13/49), whereas the incidence in ground turkey was 47% (66/141) and 33% (3/9) for conventional and organic samples, respectively (**Table 1**).

### Serotyping

To reduce the cost, twenty six isolates representing unique Salmonella-positive ground poultry samples (19 conventional, 7 organic) were selected for further characterization by serotyping, antibiotic resistance profiling, PFGE, and plasmid profiling. The twenty six isolates were carefully chosen to fairly represent the eighty two positive samples in this study in regards to variation in the collection and expiration date, brand, supermarket chain and location, and meat source (ground turkey or ground chicken). Six serotypes were identified: Tennessee (8 isolates), Saintpaul (4 isolates), Senftenberg (4 isolates), Anatum (4 isolates, including one Anatum\_var.\_15+), Ouakam (3 isolates), and Enteritidis (3 isolates) (**Table 2**). Serotypes Saintpaul, Ouakam, and Anatum were detected in conventional ground turkey, but not in organic ground chicken; the latter contained serotypes Tennessee (n = 4), Enteritidis (n = 2), and Senftenberg (n = 1) (**Table 2**).

### Antibiotic Resistance

The 26 serotyped Salmonella isolates were subjected to antibiotic resistance profiling to 16 antimicrobials (**Figure 1**). All 26 isolates were resistant to both gentamycin and ceftiofur. Resistance to the remaining antimicrobials was as follows: amoxicillin/clavulanic acid, 96% (25/26); streptomycin, 92% (24/26); kanamycin, 88% (23/26); ampicillin, 85% (22/26); cephalothin, 81% (21/26); tetracycline, 35% (9/26 ), sulfisoxazole 27% (7/26); nalidixic acid 15% (4/26), and cefoxitin, 15% (4/26). All isolates were susceptible to amikacin, chloramphenicol, ceftriaxone, and trimethoprim/sulfamethoxazole. All 26 tested isolates were

FIGURE 1 | Dendrogram of 26 Salmonella strains showing antibiotic resistance profiles, serotypes, and sources of ground poultry. Clustering was based on antibiotic resistance profiling, and the dendogram was created using BioNumerics software. Black squares indicate resistance. Abbreviations: FOX, cefoxitin; AMK, amikacin; CHL, chloramphenicol: TET, tetracycline; CTR, ceftiaxone; AMC, amoxicillin-clavulanic acid; CIP, ciprofloxac; GEN, gentamicin; NAL, nalidixic acid; TIO, ceftiofur; FIS, sulfisoxazole; SXT, trimethoprim/sulfamethoxazole; CEP, cephalothin, KAN, kanamycin; AMP, ampicillin; STR, streptomycin.

multidrug-resistant (MDR) and exhibited resistance to 4–10 antimicrobials (**Figure 1**). Isolates from organic sources also exhibited MDR to 5–7 antimicrobials. Sulfisoxazole resistance was observed only in the Anatum and Ouakam serotypes. Although there was some variability, some antibiotic profiles were common among a particular serotype (**Figure 1**).

### PFGE Analysis

fmicb-09-02653 November 1, 2018 Time: 15:9 # 6

All 26 serotyped Salmonella isolates were subjected to PFGE to determine XbaI restriction patterns. Although the four Saintpaul isolates were non-typable by XbaI-mediated PFGE, the remaining 22 isolates representing the other five serotypes were successfully analyzed (**Figure 2**). PFGE grouped the 22 Salmonella isolates into five distinct clusters each representing one of the following five serotypes: Enteritidis, Senftenberg, Ouakam, Anatum (including the Anatum\_var.\_15+), and Tennessee.

### Plasmid Profiling

PFGE was used to investigate the presence of large plasmids in the 26 serotyped isolates using S1 nuclease digestion. Eight of the 26 Salmonella isolates contained a large ∼100 Kb plasmid (**Table 2**). The four serotype Anatum strains, including the Anatum\_var.\_15 + isolate, contained a ∼100 Kb plasmid, as did the three Ouakam isolates and one of the four Saintpaul isolates (**Table 2**). A PFGE gel showing the 100 Kb plasmid in one of the Salmonella Ouakam isolates is presented in **Figure 3** (lane 3). All eight isolates harboring large plasmids were isolated from conventional samples. Restriction digestion analysis using EcoRI and/or HindIII revealed that the large plasmids harbored by the four Anatum isolates had similar restriction patterns (**Table 2**). Likewise, similar restriction patterns were obtained for large plasmids of the three Ouakam isolates (**Table 2**).

### DISCUSSION

Salmonella was recovered from 41% (82/199) of ground poultry samples collected in this study, which stresses the importance of monitoring this foodborne pathogen at the retail level. This result is similar to another study where Salmonella was recovered from 40% (30/74) of ground turkey samples in Fargo, North Dakota (Fakhr et al., 2006b). In a larger study conducted by researchers at the FDA, Salmonella prevalence was 52% in ground turkey after screening 1,499 ground turkey samples collected from grocery stores in several FoodNet sites across the United States

FIGURE 2 | Dendogram and pulsed field gel electrophoresis (PFGE) profiling of 22 Salmonella isolates using XbaI. Antibiotic resistance, serotypes, and the ground poultry sources are shown for the 22 PFGE-typable Salmonella strains. Similarity analysis was performed using the Dice coefficient, and clustering was performed using UPGMA based on PFGE profiles. Black squares indicate resistance. FOX, cefoxitin; AMK, amikacin; CHL, chloramphenicol: TET, tetracycline; CTR, ceftiaxone; AMC, amoxicillin-clavulanic acid; CIP, ciprofloxac; GEN, gentamicin; NAL, nalidixic acid; TIO, ceftiofur; FIS, sulfisoxazole; SXT, trimethoprim/sulfamethoxazole; CEP, cephalothin, KAN, kanamycin; AMP, ampicillin; STR, streptomycin.

large plasmids. Lane 1 contains the Salmonella serovar Braenderup H9812 molecular marker. Lane 2 contains E. coli NCTC 50192 treated with S1 nuclease; the three large plasmids (147, 63, and 43.5 kb) are indicated with red arrows. Lane 3 contains Salmonella Ouakam strain GP016 treated with S1 nuclease; the red arrow indicates the presence of a ∼100 kb plasmid.

(Zhao et al., 2006). A recent study indicated that the high incidence of Salmonella in turkey neck skin may predict a flock with greater potential for Salmonella contamination of ground turkey (Cui et al., 2015). Another study found that preharvest screening of Salmonella using a rapid protocol could potentially reduce Salmonella in ground turkey meat and possibly decrease future salmonellosis outbreaks (Evans et al., 2015). Recently, the Salmonella lytic bacteriophage preparation (SalmoFresh) was investigated for efficacy in reducing Salmonella populations on turkey breast cutlets and ground turkey (Sharma et al., 2015). While the bacteriophage preparation was effective in reducing Salmonella on turkey breast cutlets, it did not reduce the incidence of Salmonella Heidelberg in ground turkey (Sharma et al., 2015).

The detection of six serotypes in the 26 Salmonella-positive isolates indicates a high level of Salmonella diversity in ground turkey. This variability was also observed in other studies where different serotypes were detected depending on the geographic location and the date when studies were conducted (Fakhr et al., 2006b; Erol et al., 2013). All isolates were susceptible to amikacin, chloramphenicol, trimethoprim/sulfamethoxazole, and ceftriaxone; the latter is particularly important because ceftriaxone is the drug of choice for treating salmonellosis in children (White et al., 2001; Iwamoto et al., 2017). A recent study examining NARMS data between 1996 and 2013 showed that ceftriaxone resistance in Salmonella isolated from humans correlated with resistance in retail meats and food animals in the United States (Iwamoto et al., 2017). In the present study, the high percentages of resistance for gentamycin (100%), ceftiofur (100%), amoxicillin/clavulanic acid (96%), streptomycin (92%), kanamycin (88%), ampicillin (85%), and cephalothin (81%) is alarming. Similar high percentages of resistance ranging from 91.6 to 100% to several of these antimicrobials were reported in Salmonella isolated from chicken meat and giblets collected from Mansoura, Egypt (Abd-Elghany et al., 2015). The high incidence of resistance to aminoglycoside and β-lactam antibiotics is coincident with high prevalence of S. enterica resistance to these two classes of antibiotics in food animals (Foley and Lynne, 2008; Frye and Jackson, 2013). Ceftiofur has been used to prevent the death of 1-day old turkey poults, and its use in animal feed might select for the acquisition of plasmids with antibiotic resistance (Wittum, 2012). In our study, the moderate level of resistance in screened Salmonella isolates to tetracycline (35%), sulfisoxazole (27%), nalidixic acid (15%), and cefoxitin (15%) has been documented in other studies (Aslam et al., 2012; Thai et al., 2012; Nisar et al., 2017). A recent study analyzed the surveillance data of 18 years on antimicrobial resistance profiling showed higher level of resistance of chicken breast isolates toward thirdgeneration cephalosporins and tetracyclines when compared to human isolates (Paudyal et al., 2018). It is noteworthy that all 26 isolates in this study, including those isolated from organic sources, exhibited multidrug resistance (MDR). The presence of MDR Salmonella in retail meats in the United States, Canada, and the European Union is well- established (White et al., 2001; Zhao et al., 2006; Aslam et al., 2012; Florez-Cuadrado et al., 2018). In a recent study, 36% of Salmonella isolates were multidrug resistant to two to five antimicrobials despite being isolated from a antimicrobials free turkey production facility (Sanad et al., 2016). In a large study conducted in Spain, 41% of Salmonella isolates from meat products was resistant to three or more antibiotics (Doménech et al., 2015). The high number of MDR Salmonella detected in retail meats sold in Oklahoma is not surprising since previous studies have shown the high incidence of MDR Campylobacter spp. and Staphylococcus aureus in Oklahoma retail meats (Noormohamed and Fakhr, 2012, 2013, 2014; Abdalrahman and Fakhr, 2015; Abdalrahman et al., 2015a,b).

In this study, PFGE successfully grouped Salmonella isolates into distinct clusters that represented individual serotypes. PFGE previously showed discriminative ability for some Salmonella serotypes and antimicrobial resistance profiles (Fakhr et al., 2006b; Zhao et al., 2006). PFGE profiling was considered a possible alternative for identification of some Salmonella serotypes (Gaul et al., 2007; Zou et al., 2010, 2013). In a recent study, PFGE showed that Salmonella Heidelberg isolates from turkeys were more genetically diverse than those isolated from

chickens (Nisar et al., 2017). In this study, PFGE in combination with S1 nuclease digestion enabled successful detection of large plasmids in 8/26 (31%) of the Salmonella isolates. A large plasmid of ∼100 Kb was detected in all Anatum and Ouakam isolates and one Saintpaul isolate; furthermore, the Anatum and Ouakam were the only isolates with sulfisoxazole resistance, which might indicate a role for these plasmids in mediating resistance to this antimicrobial. Recently, we released the whole genome sequences of three isolates described in this study including Salmonella Ouakam, Anatum, and Anatum var. 15+; these isolates harbored large plasmids of 109,715, 112,176, and 112,176 bp, respectively (Marasini et al., 2016a,b). The large plasmids in the Anatum and Anatum\_var.\_15+ isolates were identical in size, which is consistent with the restriction patterns observed in this study. The large plasmids in Salmonella are known to encode genes for virulence and MDR, and their conjugative properties facilitates dissemination of virulence and antimicrobial resistance (Carattoli, 2003; Rychlik et al., 2006; Sajid and Schwarz, 2009; Folster et al., 2011; García et al., 2014).

### CONCLUSION

In conclusion, the presence of MDR Salmonella with various serotypes, PFGE profiles, and large plasmids in ground poultry is alarming. Intervention strategies to reduce this important

### REFERENCES


foodborne pathogen in retail meats are imperative, particularly in ground turkey. While ground poultry is being marketed as a healthier alternative to ground beef, consumers should apply strict food safety practices when handling ground turkey and consider cooking the meat thoroughly. The high prevalence of Salmonella strains recovered in this study with resistance to several antimicrobials can complicate the treatment of salmonellosis and increase the risk of this human illness. This is particularly critical for treating children with salmonellosis, since ceftriaxone is the drug of choice for pediatric salmonellosis and resistance to this compound would derail the efficacy of this antibiotic.

### AUTHOR CONTRIBUTIONS

MF performed the research design and provided the laboratory supplies. AG, U-AS, KH, and MF carried out the bench work and data analysis. MF and AG prepared the manuscript.

### ACKNOWLEDGMENTS

We thank the University of Tulsa for granting KH a Tulsa Undergraduate Research Challenge (TURC) grant.

and host transcriptional response. Front. Vet. Sci. 4:156. doi: 10.3389/fvets.2017. 00156


Salmonella typhimurium by polymerase chain reaction as a specific method of detection of Salmonella. Mol. Cell. Probes 6, 271–279. doi: 10.1016/0890- 8508(92)90002-F


isolated from food-producing animals, retail meat and humans in South East Asia. Int. J. Food Microbiol. 154, 98–106. doi: 10.1016/j.ijfoodmicro.2011.12.032


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

Copyright © 2018 Gad, Abo-Shama, Harclerode and Fakhr. 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.

# Pandemic GII.4 Sydney and Epidemic GII.17 Kawasaki308 Noroviruses Display Distinct Specificities for Histo-Blood Group Antigens Leading to Different Transmission Vector Dynamics in Pacific Oysters

#### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Gloria Sánchez Moragas, Instituto de Agroquímica y Tecnología de Alimentos (IATA), Spain Nigel Cook, Jorvik Food & Environmental Virology Ltd., United Kingdom

#### \*Correspondence:

Vasily Morozov vasily.morozov@medma.uniheidelberg.de †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 01 August 2018 Accepted: 02 November 2018 Published: 27 November 2018

#### Citation:

Morozov V, Hanisch F-G, Wegner KM and Schroten H (2018) Pandemic GII.4 Sydney and Epidemic GII.17 Kawasaki308 Noroviruses Display Distinct Specificities for Histo-Blood Group Antigens Leading to Different Transmission Vector Dynamics in Pacific Oysters. Front. Microbiol. 9:2826. doi: 10.3389/fmicb.2018.02826 Vasily Morozov<sup>1</sup> \*, Franz-Georg Hanisch<sup>2</sup> , K. Mathias Wegner<sup>3</sup>† and Horst Schroten<sup>1</sup>†

<sup>1</sup> Pediatric Infectious Diseases Unit, University Children's Hospital Mannheim, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany, <sup>2</sup> Institute of Biochemistry II, Medical Faculty, University of Cologne, Cologne, Germany, <sup>3</sup> Coastal Ecology, Wadden Sea Station Sylt, Alfred Wegener Institute – Helmholtz Centre for Polar and Marine Research, List auf Sylt, Germany

Noroviruses are the major cause of foodborne outbreaks of acute gastroenteritis, which are often linked to raw oyster consumption. Previous studies have suggested histo-blood group antigens (HBGA)-like structures in the oyster tissues as ligands for norovirus binding and persistence. To better understand how oysters function as vectors for the most common human noroviruses, we first tested the ability of the norovirus strains GI.1 West Chester, the pandemic GII.4 Sydney, and the epidemic GII.17 Kawasaki308 strains to interact with oyster tissues. Secondly, we explored how the HBGA preferences of these strains can affect their persistence in oyster tissues. We found limited HBGA expression in oyster tissues. HBGAs of A and H type 1 were present in the digestive tissues and palps of the Pacific oyster Crassostrea gigas, while the gills and mantle lacked any HBGA structures. By using Virus-like particles (VLPs), which are antigenically and morphologically similar to native virions, we were able to demonstrate that VLPs of GI.1 West Chester norovirus reacted with the digestive tissues and palps. Despite of the lack of HBGA expression in mantle, dominant GII.4 Sydney strain readily bound to all the oyster tissues, including the digestive tissues, gills, palps, and mantle. In contrast, no binding of the epidemic GII.17 Kawasaki308 VLPs to any of the investigated oyster tissues was observed. In synthetic HBGA and saliva-binding assays, GI.1 reacted with A type, H type, and Le<sup>b</sup> (Lewis b) HBGAs. GII.4 Sydney VLPs showed a broad binding pattern and interacted with various HBGA types. Compared to GI.1 and GII.4 VLPs, the GII.17 Kawasaki308 VLPs only weakly associated with longchain saccharides containing A type, B type, H type, and Le<sup>b</sup> blood group epitopes. Our findings indicate that GI.1 and GII.4 noroviruses are likely to be concentrated in oysters,

**87**

by binding to HBGA-like glycans, and therefore potentially leading to increased long term transmission. In regards to the GII.17 Kawasaki308 strain, we suggest that oysters can only function as short term transmission vector in periods of high environmental virus concentrations.

Keywords: norovirus, histo-blood group antigens (HBGA), oyster, norovirus outbreak, food pathogens, norovirus transmission, food safety

### INTRODUCTION

Noroviruses are non-enveloped single stranded RNA viruses of the Caliciviridae family. According to the most recent classification, there are seven genogroups (GI- GVII) in the norovirus genus (Vinjé, 2015). Noroviruses of the GI and GII genogroups have been detected in humans. However, most norovirus infections over the past decade have been caused by genogroup II genotype 4 (GII.4) noroviruses. The GII.4 variants are responsible for six pandemics and for the majority of sporadic outbreaks worldwide (Noel et al., 1999; Widdowson et al., 2004; Bull et al., 2006; Eden et al., 2010, 2014; Yen et al., 2011). Recently a novel GII.17 strain has emerged in Asia, causing an alarming number of infections and gradually replacing GII.4 strains (Chan et al., 2015a; Zhang et al., 2015; Koromyslova et al., 2017).

Norovirus are highly infectious, with as few as ten particles being able to cause disease (Teunis et al., 2008). Viral transmission occurs either via direct person-to-person contact or indirectly through contaminated water or food. In fact, noroviruses are considered one of the most common cause of foodborne infections associated with gastroenteritis outbreaks (Atmar and Estes, 2006; Moore et al., 2015; de Graaf et al., 2016). Many food-related disease outbreaks of acute gastroenteritis caused by noroviruses are associated with oyster consumption. Oysters are filter-feeding epibenthic bivalves, which filter up to 19 liters of water per hour and gram bodyweight (Jørgensen, 1996), and have been found to accumulate different pathogens from sea water. Oysters become contaminated with norovirus by exposure to discharging municipal water supply (Lees, 2000) and their raw consumption often results in outbreaks of acute gastroenteritis (Ang, 1998; Doré et al., 2010; Westrell et al., 2010).

It has been suggested that noroviruses are bound by oyster tissues through specific carbohydrate-mediated interactions, which involve the norovirus capsid protein (viral protein 1, VP1) and human-like histo-blood group antigens (HBGA). Human HBGAs are carbohydrate epitopes at the terminal end of O-glycans on glycoproteins and of glycolipids on the surface of red blood cells, the mucosal epithelium of the gastrointestinal, and the respiratory and the genitourinary tracts. Specific HBGAs, similar to blood group A type and H type, have also been identified in the digestive tracts of different oyster species (Tian et al., 2007; Ma et al., 2017).

The human noroviruses interact with HBGAs and oyster tissues in a strain-dependent manner. The GI.1 Norwalk noroviruses exhibit strong preferences toward A-type and H-type glycans and bind to the oyster digestive tissues (Le Guyader et al., 2006; Tian et al., 2006, 2007; Maalouf et al., 2010, 2011). Importantly, the GI.1 Norwalk VLP binding correlates with the expression of A-type HBGAs in the oyster digestive tract (Tian et al., 2007). Lastly, both GI.1 Norwalk VLPs and GI.1 virions from a stool sample were shown to be efficiently bioaccumulated by Crassostrea gigas (Maalouf et al., 2010, 2011). In contrast, the GII.4 Houston norovirus readily interacts with the digestive tissues, gills, and mantle tissue extracts through sialylated carbohydrates. Despite this, only low levels of GII.4 Houston norovirus VLPs were bioaccumulated by Crassostrea gigas, which was suggested to be due to an unknown degradation mechanism (Maalouf et al., 2011).

The dominant GII.4 Sydney 2012 noroviruses are most abundant among GII strains detected in samples from the oysterrelated outbreaks (Yu et al., 2015). Likewise, recent reports have found epidemic GII.17 noroviruses in oysters collected from the coasts of Italy, Japan, and Korea (Shin et al., 2013; Pu et al., 2016; La Rosa et al., 2017). Moreover, Rasmussen et al. directly linked a series of acute gastroenteritis infections with the consumption of GII.17-contaminated oysters (Rasmussen et al., 2016). In this work, we aimed at gaining a better understanding of how GII.4 Sydney and GII.17 Kawasaki308 strains are potentially transmitted to humans via oysters. To this end, we expressed the human norovirus capsid protein in insect cells. The expression resulted in the formation of VLPs, which are antigenically and morphologically analogous to native virions (Harrington et al., 2004; Hansman et al., 2006; Bok et al., 2009; Lindesmith et al., 2011, 2012, 2013; Ajami et al., 2012). Next, we measured oystertissue specificity of the GII.4 Sydney and GII.17 Kawasaki308 norovirus VLPs. We explored how binding preferences of these strains toward specific HBGAs can affect their persistence in oyster tissues. We compared the behavior of GII.4 and GII.17 noroviruses with the GI.1 norovirus, which is commonly found in oysters (Yu et al., 2015). The combination of these approaches will thus allow us to make specific predictions of the expected epidemiology of these norovirus strains.

### MATERIALS AND METHODS

### Reagents

We examined VLP binding to multivalent-HBGAs conjugates purchased from GlycoTech (Maryland, United States). They included Le<sup>y</sup> -PAA-biotin (01-043), Le<sup>b</sup> -PAA-biotin (01-042), H type1(tri)-PAA-biotin (01-037), Le<sup>x</sup> -PAA-biotin (01-036), Le<sup>a</sup> - PAA-biotin (01-035), H type 2 (tri)-PAA-biotin (01-034), Blood type A (tri)-PAA-biotin (01-032), Blood type B (tri)-PAA-biotin (01-033). The following monoclonal antibodies (MAbs) were purchased for the HBGA phenotyping: A type antibodies (clone Birma-1, Medtro GmbH), B type antibodies (clone LB-2, Medtro

GmbH), H type 1 antibodies (clone 17-206, Invitrogen), Lewis a antibody (clone 7LE, ThermoFisher), Lewis b antibody (clone 25LE, Abnova), Lewis x antibody (clone 73–30, TCI chemicals), Lewis y antibody (clone H18A, TCI chemicals). Porcine gastric mucin type III (PGM) was purchased from Sigma-Aldrich (M2378). Goat α-mice IgG HRP conjugated antibodies were purchased from ThermoFisher (62-6520).

### Norovirus Virus-Like Particles (VLPs)

The GI.1 West Chester (2001, AY502016), GII.4 Sydney (X459908), GII.17 Kawasaki308 (2015, LC037415) VLPs were produced in Spodoptera frugiperda (Sf9) cells using baculovirus expression system as described previously (Koromyslova et al., 2017). In short, the recombinant VP1 bacmids were transfected into the Sf9 cells using Effectene (Qiagen). After 5 days of incubations at 27◦C the cells were harvested, centrifuged, and the supernatant containing baculovirus was used to infect high five (H5) insect cells. The H5 cells were incubated at 27◦C for 6 days. At 6 days post infection, the supernatant containing secreted VLPs was collected. The VLPs were concentrated by ultracentrifugation at 35000 rpm (SW55 Ti rotor, Beckman Coulter) for 2 h at 4◦C and purified by the CsCl equilibrium gradient ultracentrifugation at 35000 rpm (SW55 Ti rotor, Beckman Coulter) for 18 h at 4◦C. The morphology of the VLPs was examined by electron microscopy.

### Oyster and Saliva Samples

All experiments were performed with the Pacific oysters Crassostrea gigas. The oysters were collected from an uncontaminated area on Sylt, North Sea, Germany, 54.937962◦ N, 8.313825◦ E, and shipped the same day (within 24 h) on ice to the laboratory in Mannheim. Upon arrival, the oysters were immediately dissected and the digestive tissues, mantles, gills, and pulps were collected from five to six different oysters. The tissues were homogenized in phosphate-buffered saline (PBS) pH 7.4, boiled for 10 min at 95◦C and centrifuged at 8000 rpm for 5 min (Mikro 200, Hettich). The supernatants were collected and used for ELISA experiments after the protein concentration was measured with the Bio-Rad protein assay kit. The tissue homogenate was tested for the presence of preexisting GI or GII norovirus contamination by a GeneXpert commercial assay according to the manufacturer's instructions (Cepheid Inc., Sunnyvale, CA, United States) (data not shown). The oysters which were artificially exposed to norovirus-containing stool samples were used as a positive control.

Saliva samples collected from 17 healthy adult persons were boiled for 10 min at 95◦C and centrifuged at 8000 rpm for 5 min (Mikro 200, Hettich). Approval was obtained from the ethics committee of the Medical Faculty of Mannheim, Heidelberg University, #2017-528N-MA.

### Detection of VLPs Binding to Oyster Tissues, Saliva Samples, and Synthetic HBGAs

The binding of the GI.1 West Chester, the pandemic GII.4 Sydney, and the epidemic GII.17 Kawasaki308 norovirus VLPs to oyster tissues was measured in triplicate for each oyster (Koromyslova et al., 2017). Briefly, NuncMaxisorp plates were covered with the digestive tract, mantle, gill, and pulp tissue extracts at a concertation of 40 µg/ml for 1 h at 37◦C. The plates were washed three times with PBS-Tween20 0.1% (PBS-T) and blocked with 5% bovine serum albumin (BSA) for 1 h at 37◦C. Subsequently, a serial dilution of VLPs (0–20 µg/mml, PBS) was applied to the plate for 1 h at 37◦C. Plates were washed and incubated for 1 h at 37◦C with either α-VLP-His<sup>6</sup> nanobodies (Nb85) for GI.1 VLPs, α-rabbit-polyclonal antibodies for GII.4 VLPs, or α-VLP-biotinylated nanobodies (Nb26) for GII.17 VLPs (Koromyslova and Hansman, 2017). After washing the plate, the secondary antibodies (horseradish peroxidase conjugated with α-His6-IgG, α-rabbit-IgG or streptavidin) were added. Finally, the washed plates were developed for 30 min at room temperature with o-phenylenediamine (OPD) and H2O2. The reaction was stopped after 30 min and absorbance was measured at 490 nm on a Tecan infinite M200 Multiwell reader (Tecan, Switzerland). Negative controls included a VLP-negative control and primary antibody-negative controls. PGM and secretor saliva samples, described below in the results, were used as positive controls.

For the synthetic HBGA assays, carbohydrate-polyacrylamide - biotin (HBGAx-PAA-biotin) conjugates were diluted to a final concentration of 20 µg/ml in PBS and applied on streptavidincoated plates overnight at 4◦C. The plates were blocked and a serial dilution of VLPs (0–20 µg/mml, PBS) was applied. The concentration of bound VLPs was detected as described above.

For the saliva binding assay, 100 µl of the 1:50 diluted saliva samples (each sample was measured in triplicate) was added to carbohydrate buffer pH 9.4, and were incubated overnight at 4 ◦C, then washed and blocked. The plates were incubated with 10 µg/ml VLPs in PBS and the concentration of bound VLPs was detected as described above.

## Detection of HBGA Present in Saliva and Oyster Tissues

The HBGAs phenotypes expressed in the oyster tissues and saliva samples were determined using MAbs as previously described (Tian et al., 2007). Each incubation step was followed by a washing step (3X PBS-T). The oyster tissues extracts were coated on NuncMaxisorp plates overnight at 4◦C. Plates were blocked with 5% BSA in PBS-T for 1 h at 37◦C. 100 µl of 1:400 α-A type MAbs, 1:100 α-B type MAbs, 1:250 α-H type 1MAbs, 1:50 α-Lewis a type MAbs, 1:1800 α-Lewis b type MAbs, 1:800 α-Lewis x type MAbs, and 1:3200 α-Lewis y type MAbs were incubated for 1 h at 37◦C. Goat α-mice IgG HRP conjugated Abs were used in the 1:800 – 1:3200 dilution range for 1 h at 37◦C. The signal was developed with the OPD/H2O<sup>2</sup> mixture for 30 min and the intesity was measured on the Tecan reader.

## Statistical Analysis

Means were compared by using the Student t-test, and a P-value of below 0.05 was considered as statistically significant (GraphPad Prisma Software, La Jolla, CA, United States).

### RESULTS

### GI.1, GII.4, and GII.17 VLP Binding to Oyster Tissues

GI.1 West Chester VLPs bound to the digestive tissues were easily detected even at the lowest concentration of 1.25 µg/ml, while no binding to mantle and gills was observed (**Figure 1A**). In contrast to previous studies (Maalouf et al., 2011) which used a different GI.1 strain, GI.1 West Chester VLPs also bound to palp tissue extracts. However, the binding was expected since A type HBGAs are present in palps (**Figure 3A**). This A type HBGA-dependency was indirectly confirmed by high standard deviations, which match the different levels of A type HBGA expression in the various oyster samples (**Supplementary Figure S1**).

Similar to the less prevalent GII.4 variant (Maalouf et al., 2011), the dominant GII.4 Sydney VLPs showed a strong binding efficiency to all tissues tested, even at the lowest VLP concentration (**Figure 1B**).

Surprisingly, no binding of the epidemic GII.17 Kawasaki308 VLPs to any of the oyster tissues was observed (**Figure 1C**). Although we obtained signals in the PGM coated wells (i.e., positive controls), all of the GII.17 VLPs assayed at a concentration of 20–1.25 µg/ml produced readings below the detection limit. The assay was repeated several times in parallel with the GI.1 and GII.4 binding experiments using the same tissue extracts. Additionally, testing higher VLP concentrations of 80 µg/ml, different coating procedures with higher concentrations of tissue extracts, and longer incubation times had no effect on the GII.17 VLP binding (**Supplementary Figure S2**). To exclude any seasonal factors which could have contributed to the GII.17 VLP binding, the assays were also conducted with oysters collected in December and June. However, these binding assays also resulted in qualitatively similar negative results (**Supplementary Figure S3**).

### Expression of HBGA in Oyster Tissues

The HBGA phenotypes present in oyster tissues were examined by eight MAbs which recognize A type, B type, H type 1 and Lea,b,x,<sup>y</sup> HBGA epitopes. The binding cut-off was set at OD<sup>490</sup> <sup>=</sup> 0.18, which was three times the value of the PBS control. Only A type and H type 1 carbohydrates were found to be present in the tissues (**Figure 2**). The A type carbohydrates were detected in the digestive tissues (OD<sup>490</sup> <sup>=</sup> 0.28) and palps (OD<sup>490</sup> <sup>=</sup> 0.27). Likewise, H type 1 HBGAs were present in the digestive tissues (OD<sup>490</sup> <sup>=</sup> 0.31), gills (OD<sup>490</sup> <sup>=</sup> 0.26), and palps (OD<sup>490</sup> <sup>=</sup> 0.27). None of the other tested HBGAs (B type and Lea,b,x,<sup>y</sup> ) were detected in the oyster tissue extracts.

### VLP Binding to Saliva Samples

To evaluate HBGA binding preferences of the norovirus VLPs in the human host, we performed saliva binding assays using a panel of saliva samples from 17 healthy adult individuals (subject A–Q). Prior to the VLP binding assay, the saliva samples were tested for the ABO, Lewis, and secretor phenotypes using the MAbs. Saliva containing Le<sup>b</sup> and Le<sup>y</sup> antigens were defined as

secretor positive probes, while the absence of the A, B, H, Le<sup>y</sup> , Le<sup>b</sup> antigens suggested a non-secretor status. Overall, thirteen samples were secretor positive and four samples were from nonsecretors (subjects B, E, K, P) (**Figure 3**).

The GI.1 West Chester VLPs bound strongly to the saliva probes J and L (OD<sup>490</sup> > 1.0), containing A/H1/Lea,b,x,<sup>y</sup> and A/Lea,b,x,<sup>y</sup> type glycans, respectively (**Figure 3A**). The moderate binding was also observed for the subjects A (A/H1/Lea,b,x,<sup>y</sup> ), C (H1/Lea,b,x,<sup>y</sup> ), D (A/Lea,b,x,<sup>y</sup> ), G (A/B/Lea,b,x,<sup>y</sup> ), I (A/Lea,b,x,<sup>y</sup> ), O (A/Lea,b,x,<sup>y</sup> ). Weak and no GI.1 VLP binding was observed for saliva samples B, P, E, F, K, which lack A type and H type 1 carbohydrates. Overall, the GI.1 West Chester VLP binding profile matched well-established preferences of GI.1 Norwalk VLPs toward H1, Le<sup>b</sup> and A type antigens (Harrington et al., 2002; Hutson et al., 2002; Huang et al., 2003; Shirato et al., 2008).

GII.4 VLPs bound to all secretor positive samples, and with an OD<sup>490</sup> in the 0.4–0.6 range (**Figure 3B**). Only weak or no binding was observed for the non-secretor saliva samples (B, E, K, P).

The GII.17 VLPs demonstrated the highest signal, with an OD<sup>490</sup> > 0.67, for subjects G (A/B/Lea,b,x,<sup>y</sup> ), J (A/Lea,b,x,<sup>y</sup> ), L(A/B/Lea,b,x,<sup>y</sup> ), N (A/B/Lea,b,x,<sup>y</sup> ) (**Figure 3B**). Moderate binding of OD<sup>490</sup> <sup>=</sup> 0.53–0.56 was observed for the subjects I (A/Lea,b,x,<sup>y</sup> ) and Q (B/Lea,b,x,<sup>y</sup> ). In contrast, subjects B, K, P, E

containing predominantly the Le<sup>a</sup> and Le<sup>b</sup> antigens displayed absorbance readings that were below or just above the cut-off limit (OD<sup>490</sup> < 0.2).

With some exceptions, the saliva binding pattern of GII.17 VLPs closely resembled the binding patterns of GI.1 VLPs (**Figure 3B**). However, in contrast to GI.1 VLPs, GII.17 VLPs demonstrated a stronger binding for the subjects F, N, Q. These subjects were B positive and A type negative phenotypes, suggesting that GII.17 VLPs recognize H1, Le<sup>b</sup> , A type, but also B type antigens.

### VLP Binding to Synthetic HBGAs

Saliva is a complex biofluid containing a mixture of HBGA glycans. To define specificities of GI.1 West Chester, GII.4 Sydney, and the GII.17 Kawasaki308 norovirus VLPs toward individual HBGA types, we further performed a synthetic oligosaccharide-based assay using polyacrylamide-conjugated multivalent HBGAs. As shown in **Figure 4A**, GI.1 VLPs recognized H type 1 trisaccharides, which was dose-dependent. A higher amounts of GII.4 VLPs bound to B type and lower to Le<sup>y</sup> and Le<sup>b</sup> types (**Figure 4B**). Interestingly, GII.17 VLPs did not bind to tested PAA-conjugated HBGAs, under the studied conditions.


TABLE 1 | Summary of the HBGA binding preferences combined from saliva and synthetic HBGA assays.

### DISCUSSION

Human noroviruses are genetically and antigenically diverse (Hansman, 2006), with a majority of the human norovirus infections being caused by the GII strains. According to the Noronet report April 2014, approoximately 91% of all norovirus sequences collected worldwide belong to the genogroup GII. 74% of these norovirus sequences were GII.4 variants. GI noroviruses compromised only approximately 9% of all outbreaks. Most of the norovirus genotypes which infect humans have also been found in oyster-related outbreak samples. However, the GI and GII proportions are significantly different in oyster-related outbreaks than in non-oyster related outbreaks. In oyster-related outbreaks, only 66% of all norovirus sequences found belonged to the GII strains (Noronet, 2014)<sup>1</sup> . Moreover, the dominant GII.4 strains only accounted for 30% of the GII sequences reported. This shift could have been driven by the more frequent occurrence of GI strains in coastal waters. However, influent and effluent waters still contain higher concentrations of GII noroviruses (Flannery et al., 2012), indicating that oysters are also exposed to higher GII concentrations. Consequently, GI noroviruses must be enriched in oyster tissues, where selection and persistence is dependent on specific molecular interactions between the virus and the oyster. Such specific viral enrichment might be mechanistically similar to HBGA-norovirus binding in humans (Tian et al., 2006, 2007; Maalouf et al., 2011).

Histo-blood group antigens located on epithelial cells of the human gastrointestinal tract are receptors or co-receptors for human noroviruses. Certain HBGA types are also expressed in oyster tissues. Tian et al. (2007) detected A type-like and H typelike HBGAs in the gastrointestinal tissues of Crassostrea virginica, Crassostrea gigas, and Crassostrea sikamea. Similarly, in our study we were able to demonstrate that both A type and H type 1 were present in the digestive tissues of Crassostrea gigas. Additionally, we demonstrated that the palps also contained these HBGAs, while gills and mantle showed undetectable levels of any HBGA types.

Norovirus interaction with HBGAs has been shown to occur via the protruding (P) domain of the major capsid VP1 protein (Cao et al., 2007; Choi et al., 2008; Hansman et al., 2011). Distinct structural features of the P domains from different norovirus strains define their binding preferences toward specific HBGA types. A summary of the HBGA binding preferences of the GI.1 West Chester, the pandemic GII.4 Sydney, and the epidemic GII.17 Kawasaki308 norovirus VLPs is presented in **Table 1**. We observed that GI.1 VLPs used in this study readily reacted with A type, AB type, and H type, but not with B type saliva. Moreover, these GI.1 VLPs also bound to synthetic PAA conjugated H type 1 HBGAs. This matches previous results of another GI.1 strain, which demonstrated that the prototype GI.1 Norwalk noroviruses preferentially recognize A-type and H-type 1 HBGAs, and might also bind Le<sup>b</sup> antigens (Harrington et al., 2002; Hutson et al., 2002; Huang et al., 2003; Shirato et al., 2008). As was expected from its HBGA binding capabilities, we found that GI.1 VLPs reacted with digestive tissues and palps of Crassostrea gigas, which predominantly contained A type and H type HBGAs. This further supports previous observations obtained from oysters (Maalouf et al., 2011).

The most prevalent GII.4 noroviruses seem to possess a broad and long-standing HBGA binding profile. Evolutionary studies revealed a highly conserved HBGA binding site in the GII.4 norovirus strains isolated throughout the 1974–2012 period (Bok et al., 2009). With the exception of the structural studies (Singh et al., 2015; Morozov et al., 2018), the HBGA binding preferences of the most virulent GII.4 Sydney strain were largely unexplored. Here, we demonstrated that the GII.4 Sydney VLPs interacted equally well to secretor and non-secretor saliva samples. Nonsecretors do not express H1 and Lewis-b epitopes, and hence the observed binding could be due to Lewis-y or H2 interference. In the synthetic HBGAs assays, the GII.4 Sydney VLPs reacted with B type trisaccharide-PAA, Le<sup>b</sup> type tetrasaccharide-PAA, Le<sup>y</sup> type tetrasaccharide-PAA conjugates. However, the latter results have to be treated with caution, due to that fact that the short carbohydrate chains are attached to a polymer linker, and are therefore less accessible for the bulky VLP structure. Our parallel work, using carbohydrate mixtures from breast milk (human milk oligosaccharides, HMOs), revealed the specificity of GII.4 Sydney VLPs toward longer saccharides containing 4–15 sugar units and the terminal blood group H type 1 or Lewis-b epitopes (Hanisch et al., 2018). Together these results suggest that GII.4 Sydney VLPs can recognize a broad spectrum of carbohydrates, including Le<sup>b</sup> , H type 1, Le<sup>y</sup> , and B type blood group antigens, although the affinities to each specific HBGA type remain to be elucidated. In this study, GII.4 Sydney bound well to all the investigated oyster tissues. In case of the digestive tissues this can be explained by recognition of the H type 1 HBGAs, while the binding to the mantle and gill tissues can most likely be explained by the presence of non-HBGA carbohydrates, as it was previously observed for GII.4 Houston VLPs (Maalouf et al., 2011).

Enhanced HBGA binding capabilities have been put forward as a factor to explain the rapid emergence of the GII.17 strains (Chan et al., 2015b). However, in our synthetic HBGA assay, only low concentrations of GII.17 Kawasaki308 VLPs associated with the sugars. As previously mentioned, this may have been due to steric effects, the affinity of GII.17 Kawasaki308 to short-chain HBGA-containing PAAs used on our plates was likely lower compared to GII.4, and GI.1 strains. GII.17 Kawasaki308 VLPs could nevertheless interact

<sup>1</sup>https://www.rivm.nl/en/Documents\_and\_publications/Common\_and\_Present/ Publications/Centre\_for\_Infectious\_Disease\_Control/Noronet\_updates/ Noronet\_update\_april\_2014

with longer saccharides and with glycoproteins present in saliva of secretor-positive individuals. Its binding patterns closely resembled those of GI.1 VLPs, but showed additional signals in B type subjects, which supports its enhanced binding capabilities. We observed no binding of GII.17 Kawasaki308 to the oyster tissues extracts, therefore we assumed that GII.17 Kawasaki308 exhibited low affinity interactions with endogenous blood group active oligosaccharides (A type, B type, H type, Le<sup>b</sup> blood group epitopes), and,thus, just weakly attach to the HBGA epitopes expressed in the digestive tissues of pacific oysters.

Considering these binding assay results it is surprising that oysters can be contaminated with GII.17 noroviruses (Shin et al., 2013; Pu et al., 2016; Rasmussen et al., 2016; La Rosa et al., 2017). One explanation could be that the epidemic GII.17 noroviruses do not actively accumulate through tissue binding in oysters and only use oysters as a short term transmission vector via passive accumulation during filter feeding. This might also apply to other norovirus genotypes detected in oyster related outbreaks. Oyster only express a limited array of HBGA types that, in some cases, cannot be recognized by every norovirus strain. It is therefore suggested, that many noroviruses can interact with tissues only non-specifically, while other noroviruses, i.e., GI.1, can be efficiently accumulated within oysters. Such bindingdependent accumulation could explain the shift toward the increased presence of GI genotypes in oysters (Yu et al., 2015). Alternatively, other unidentified proteins or, as it was shown for the GII.4 Houston strain, non-HBGA glycans could act as selective ligands for the norovirus accumulations in oysters (Maalouf et al., 2010). These molecules or a larger variety of HBGAs could also be provided by the diverse microbiome of the oyster digestive tract (Lokmer et al., 2016) that might therefore, serve as a substrate for norovirus recognition and accumulation (Miura et al., 2013; Li et al., 2015).

From the combination of binding assays applied in this study, we conclude that the pandemic GII.4 Sydney norovirus strains are likely to employ H type 1-mediated interactions to accumulate within oysters, which could potentially lead to long-term transmission. The epidemic GII.17 noroviruses, on the other hand are likely to have low persistence in oysters and transmission is suggested to be short-term, when high environmental concentration of these viruses is found. Understanding these strain specific characteristics of persistence will allow the prediction of norovirus epidemiology, which is vital for the development of control measures.

### REFERENCES


### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the ethics committee of the Medical Faculty of Mannheim, Heidelberg University, 2017-528N-MA, with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the ethics committee of the Medical Faculty of Mannheim, Heidelberg University, 2017-528N-MA.

### AUTHOR CONTRIBUTIONS

VM, KMW, and HS conceptualized the idea. VM, F-GH, KMW, and HS analyzed the data and wrote the original draft. VM performed the experiments. All authors had read and approved the final version of the manuscript.

### FUNDING

This research was funded by the Medical Faculty Mannheim, Heidelberg University. We acknowledge financial support by Deutsche Forschungsgemeinschaft within the funding program Open Access Publishing, by the Baden-Württemberg Ministry of Science, Research and the Arts and by Ruprecht-Karls-Universität Heidelberg.

### ACKNOWLEDGMENTS

The authors would like to thank Dr. Grant Hansman for providing the VLPs; Dr. med. Marlis Gerigk, Dr. med. Sybille Welker, Dr. med. Angela Petzold for testing the pre-existence of human norovirus contamination in the oyster samples; Prof. Christian Schwerk and Alexa Lauer for critical reading of the manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.02826/full#supplementary-material



clams, mussels, and oysters: a possible mechanism of bioaccumulation. J. Food Prot. 70, 2140–2147. doi: 10.4315/0362-028X-70.9.2140


diversity and temporal-geographical distribution from 1983 to 2014. Appl. Environ. Microbiol. 81, 7615–7624. doi: 10.1128/AEM.01729-15

Zhang, X.-F., Huang, Q., Long, Y., Jiang, X., Zhang, T., Tan, M., et al. (2015). An outbreak caused by GII. 17 norovirus with a wide spectrum of HBGA-associated susceptibility. Sci. Rep. 5:17687. doi: 10.1038/srep 17687

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

Copyright © 2018 Morozov, Hanisch, Wegner and Schroten. 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.

# Efficiency of Different Disinfectants on Bacillus cereus Sensu Stricto Biofilms on Stainless-Steel Surfaces in Contact With Milk

Higor Oliveira Silva<sup>1</sup> , Joyce Aparecida Santos Lima<sup>1</sup> , Carlos Eduardo Gamero Aguilar<sup>1</sup> , Gabriel Augusto Marques Rossi<sup>1</sup> , Luis Antonio Mathias<sup>1</sup> and Ana Maria Centola Vidal<sup>2</sup> \*

<sup>1</sup> Department of Preventive Veterinary Medicine and Animal Reproduction, School of Agrarian and Veterinarian Sciences, São Paulo State University, São Paulo, Brazil, <sup>2</sup> Department of Veterinary Medicine, School of Animal Science and Food Engineering, University of São Paulo, São Paulo, Brazil

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Jing He, Guangzhou Women and Children Medical Center, China Veronica Lazar, University of Bucharest, Romania

> \*Correspondence: Ana Maria Centola Vidal anavidal@usp.br

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 21 June 2018 Accepted: 14 November 2018 Published: 28 November 2018

#### Citation:

Silva HO, Lima JAS, Aguilar CEG, Rossi GAM, Mathias LA and Vidal AMC (2018) Efficiency of Different Disinfectants on Bacillus cereus Sensu Stricto Biofilms on Stainless-Steel Surfaces in Contact With Milk. Front. Microbiol. 9:2934. doi: 10.3389/fmicb.2018.02934 The species of the Bacillus cereus group have the ability to adhere to and form biofilms on solid surfaces, including stainless steel, a material widely used in food industries. Biofilms allow for recontamination during food processing, and the "clean-in-place" (CIP) system is largely used by industries to control them. This study thus proposes to evaluate the efficacy of peracetic acid and sodium hypochlorite against biofilms induced on stainless-steel surfaces. The SAMN07414939 isolate (BioProject PRJNA390851), a recognized biofilm producer, was selected for biofilm induction on AISI 304 stainless steel. Biofilm induction was performed and classified into three categories: TCP (Tindalized, Contaminated, and Pasteurized milk), TCS (Tindalized milk Contaminated with Spores), and TCV (Tindalized milk Contaminated with Vegetative cells). Subsequently, the coupons were sanitized simulating a CIP procedure, on a pilot scale, using alkaline and acid solutions followed by disinfectants (peracetic acid and sodium hypochlorite). Microorganism adhesion on the surfaces reached 6.3 × 10<sup>5</sup> to 3.1 × 10<sup>7</sup> CFU/cm−<sup>2</sup> . Results did not show significant differences (p > 0.05) for surface adhesion between the three tested categories (TCP, TCS, and TCV) or (p > 0.05) between the two disinfectants (peracetic acid and sodium hypochlorite). Microbial populations adhered to the stainless-steel coupons are equally reduced after treatment with peracetic acid and sodium hypochlorite, with no differences in the control of B. cereus s.s. biofilms on AISI 304 stainless-steel surfaces.

Keywords: peracetic acid, sodium hypochlorite, recontamination, food safety, biofilm formation

### INTRODUCTION

Biofilm formation is a complex process consisting of a string of molecular and physiological events that take place throughout several stages including adherence, formation of microcolonies, tridimensional structuring, and maturation (Watnick and Kolter, 2000). Bacteria of the Bacillus cereus group have the ability to adhere to and form biofilms on solid surfaces such as stainless steel (Salustiano et al., 2009; Kumari and Sarkar, 2016). In the industry, milk residues may occasionally persist on the surface of stainless-steel equipment forming a thin layer, rich in nutrients,

called conditioning film that renders these surfaces more prone to bacterial adhesion and the consequent formation of biofilms (Machado, 2005).

Reports of the presence of B. cereus microorganisms in commercial products are frequent (Vidal-Martins et al., 2005; Montanhini et al., 2015; Vidal et al., 2016), and biofilms play an essential role in the persistence of these microorganisms in processing lines. Biofilms have an extracellular matrix that constitutes a stable structure that protects bacteria against the action of sanitizing agents, making their control more difficult and allowing for persistent recontamination of food products (Simões et al., 2006; Salustiano et al., 2009; Majed et al., 2016). Another mechanism related to the biofilm persistence is considered the different gene expression in the multicellular population, cause some cells enter in a dormant or persistent state, manifesting a non-inherited resistance or tolerance to different antimicrobials (Wood et al., 2013; Singh et al., 2017).

The "clean-in-place" (CIP) system is commonly employed by industries to control biofilms in milk processing lines. It is a sanitization procedure that includes the regular cleaning of pipes and equipment by using acid and alkaline solutions applied at high temperatures (Bremer et al., 2006). It is used essentially to ensure the disinfection of clean surfaces and the elimination of organic residue through the action of sanitizers in hard-to-reach places (Shi and Zhu, 2009). However, product contamination and deterioration caused by biofilms are still recurring problems (Ostrov et al., 2016).

Microorganisms of the B. cereus group are among the most important deteriorating agents of the milk production chain and are also among those involved in foodborne diseases. Because they compromise the quality and microbiological safety of milk and dairy products, they represent a major concern for the dairy industry (Simões et al., 2010; Kumari and Sarkar, 2016).

In the last decade, the scientific community has been increasingly interested in research on biofilm formation by microorganisms of the B. cereus group (Majed et al., 2016). However, few studies have investigated the efficiency of processes to combat biofilms formed by B. cereus bacteria, mainly those adhered to stainless-steel surfaces. By contrast, these studies clearly show the importance of determining the efficiency of solutions and sanitizers used in CIP procedures for the removal of bacteria linked to stainless-steel surfaces (Parkar et al., 2003; Bremer et al., 2006).

The present study aimed to evaluate the efficiency of two types of disinfectants (peracetic acid and sodium hypochlorite) used in the CIP system on biofilms induced by spores and vegetative cells of Bacillus cereus s.s. formed on stainless-steel surfaces that were in contact with experimentally tindalized milk.

### MATERIALS AND METHODS

A strain of Bacillus cereus s.s. previously genetically sequenced (SAMN07414939 – BioProject PRJNA 390851) and phenotypically considered as biofilm producer was selected for the present study, which was performed in three separated experiments (**Figure 1**), each with three repetitions, as described in details in the following sessions.

Initially, the milk used in the study was tindalized, to allow the total elimination possible of sporulated bacteria. Then, the milk was experimentally contaminated with the strain of B. cereus s.s. selected. Coupons of stainless steel were submerged into the contaminated milk for 10 h to allow the biofilm formation on its surface. Posteriorly, the coupons were sanitized and the efficiency of the sanitization was evaluated based on the counts of the bacterial population adhered before and after the process.

### Experiment 1

An inoculum of vegetative cells of B. cereus s.s. (SAMN07414939 – BioProject PRJNA 390851) was produced to contaminate the tindalized milk. For this, 10 µL of the isolate kept in stock was transferred to test tubes containing 5mL of Brain-Heart Infusion Broth (BHI) and incubated at 30◦C for 12 h, performing the count of the population in 1 mL of the culture at the time of use.

To assure that the biofilm production was restricted to the vegetative cells of the B. cereus s.s. incubated a milk free of contamination was prepared. For this, 30 liters of raw milk were submitted to the tindalization process.

The tindalization was performed in the dairy industry of the City Campus Fernando Costa (PUSP-FC), located in Pirassununga, SP, Brazil. The equipment used include the

FIGURE 1 | Flowchart of the experimental design, describing all the steps of this study.

yogurt-making machine (composed by a Fermentation tank made by stainless steel – Mec Milk <sup>R</sup> ), with temperature control, coupled by tubes to a pasteurizer. Previously, the equipments were mechanically cleaned and sanitized with sodium hydroxide, nitric acid, and peracetic acid with concentrations at 0.05% at 25◦C for 1 h. Swabs were collected before and after the sanitization in order to evaluate the efficiency of the process in the equipment and pipes.

The tindalization was performed with a series of heatings: first, the milk was heated at 80◦C for 10 min, then cooled at 30◦C for 30 min, and then re-heated at 95◦C for 20 min. The entire process was repeated three times (Kim et al., 2012, adapted).

Thirty liters of raw milk were tindalized and pumped through the pipes from the raw-milk-receiving tank to the pasteurization machine. Of these, 10 liters were pasteurized and packed in polyethylene bags with a capacity of 1 liter. Posteriorly, the Tindalized Milk (TM) was transported to the Laboratory of Quality of Animal Products (Quali-POA) and stored under refrigeration until the moment of analysis.

An experimental prototype (**Figure 2**) was produced in stainless steel AISI 304, with coupons whose dimensions were 3.5 × 1.5 cm and 1 × 1 cm (coupons A and B, respectively), and prepared for the adhesion of B. cereus s.s., and a shelf with the capacity of 30 coupons (Oliveira et al., 2010, adapted). The coupons A were produced to perform the bacterial counts, presenting an area more adequate to the quantification so that the adhered population would not be underestimated. The coupons B were produced exclusively to perform the topography analysis after the biofilm induction, adapted to the size and capacity of the Scanning Electron Microscopy cannon. The prototype was previously sanitized by mechanical abrasion with sponge and neutral detergent, followed by a 1% sodium hydroxide, immersion in 70% alcohol and autoclaving (Ribeiro-Furtini, 2005).

To create a conditioning film in the surfaces of the coupons of stainless steel, similar to that found on food industries, 21 coupons (18 coupons A and 3 coupons B) were arranged in the shelf and were totally submerged in 1 liter of TM inside of a sterile beaker under stirring for 1 h at room temperature. The entire process was conducted in a biosafety cabinet. Posteriorly, to ensure the sterility after the procedure three coupons A were subjected to bacterial quantification with surface swabs. The coupons were

transferred aseptically to sterile petri dishes with filter paper with the sampling surface turned up and dried at 60◦C for 2 h.

In sequence, the milk contamination was performed transferring 1 liter of TM to a sterile beaker containing a prototype with 18 coupons (15 coupons A and 3 coupons B) to induce the bacterial adhesion. 2 mL of the inoculum were added to this milk at room temperature, remaining incubated under agitation for 10 h to induce the adhesion of the B. cereus s.s.

### Experiment 2

Ten micro liter of the culture of B. cereus s.s. in stock were transferred to the Tryptone Soy Agar (TSA) and incubated for 7 days at 30◦C to induce the spore production. In sequence, the agar surface was washed with 5 mL of sterile distilled water. The mixture was transferred to sterile tubes and then centrifuged for 20 min. The supernatant was discarded and resuspended with sterile distilled water, followed by a new centrifugation, repeating the process three times.

Posteriorly, the suspension was submitted to a thermal shock at 80◦C for 1 min followed by a cooling for 1 min to inactivate the vegetative cells (Giffel et al., 1997). The absence of vegetative cells was confirmed by the Wirtz-Conklin staining (Bier, 1975) and the count of the B. cereus s.s. was performed using 1 mL of the culture at the time of use.

In this experiment the tindalized milk free of contamination was also used, as described on the Experiment 1. The TM was transported to the laboratory Quali-POA and refrigerated until the moment of the analysis. Then, the contamination was performed by adding 2 mL of the inoculum in one liter of the TM and the rest of the experiment followed the same steps of the Experiment 1.

### Experiment 3

For the third experiment was used the same inoculum produced in the Experiment 1. After the tindalization of the raw milk, 10 liters of TM were pumped through to the raw-milk-receiving tank, where 15 mL of the inoculum of vegetative cells of B. cereus s.s. were added. The TM (at 30◦C) was homogenized, pasteurized and packed in polyethylene bags. After this, the milk was transported to the laboratory, where was refrigerated until the moment of the next analysis and the rest of the experiment followed the same steps of the Experiment 1.

### Sanitization of the Stainless Steel Coupons by the Clean-In-Place (CIP) System, Simulated in a Pilot Scale

The same sanitization process was repeated to three experiments.

After the incubation for 10 h, the prototype containing 15 coupons (12 coupons A and 3 coupons B) with biofilm adhered was submitted to the sanitization process. This stage was composed by seven steps, each involving one solution normally used in the dairy industry, as showed in **Figure 3**. The experimental prototype was submerged in the beakers, sterilized, and kept under agitation and heating on a magnetic stirrer.

The sanitization started with water at room temperature for 1 min, followed by an alkaline solution (2% sodium hydroxide) at 60◦C for 10 min, sterile distilled water at room temperature for 1 min, acid solution (2% nitric acid) at 60◦C for 10 min, and washed again with sterile distilled water at room temperature. In this step, the second lot of samples of the coupons A were collected.

The CIP process was completed after the sanitization with a 0.1% solution of peracetic acid and 0.02% sodium hypochlorite, both at room temperature for 10 min, followed by the rinse with sterile distilled water at room temperature for 1 min. For the negative control, the sanitization process was performed using only water, without the aid of the sanitizers.

At the end of this step, the third and last proceeding of collect of the coupons A was performed, in triplicates, and the coupons B, individually reserved in sterile Petri dishes with filter paper, where remained at 30◦C for 24 h.

### Quantification of the Bacterial Population and the Adhesion to the Coupon Surfaces

To quantify the bacterial population aliquots of 1 mL of the milk and surface swabs were collected. Serial dilutions were performed using 0.1% sterile peptone water and then 0.1 mL of the mixtures were transferred to TSA and Mannitol Egg Yolk Polymyxin Agar

(MYP) using the surface coating technique. The plates were incubated at 30◦C for 24–48 h.

To quantify the population adhered to the coupons, they were washed and then swabs were smeared on the surfaces.

The first lot of samples was washed with phosphate buffer to remove the residues of milk, followed by additional washes with sterile distilled water, saline solution of tween 80 peptone (H2Osp-Tween) and sterile distilled water again. On the second and third lot of samples the coupons were washed with sterile distilled water H2Osp-Tween and sterile distilled water again.

Three coupons were collected from each lot, using a sterile swab for each coupon, which posteriorly were submerged into tubes containing 9 mL of 0.1% sterile peptone water. Serial dilutions were performed, following by plating in TSA and MYP at 30◦C for 24–48 h.

### Topographic Analysis of the Coupons of Stainless Steel Surfaces

After incubation for 24 h, the coupons B were taken to the Electronic Microscopy Laboratory at the School of Agrarian and Veterinarian Sciences of the Sao Paulo State University (UNESP), located at Jaboticabal, SP, Brazil. The coupons were immersed on a fixing solution (Karnovsky modified: 2.5% glutaraldehyde, 2.5% formaldehyde in 0.05 M sodium cacodylate buffer at pH 7.2 and 0.001 M of CaC12) during at least 72 h.

Posteriorly the coupons were washed with phosphate buffer, fixed overnight in 1% osmium tetroxide, dehydrated using an ethanol gradient (25, 50, 70, 90, and 100%), dried at critical point and golden metallized (Bossola and Russell, 1998). Then, the topographic images of the coupons surfaces were obtained using a Zeiss EVO MA-10 electronic microscopy.

### Statistical Analyses

Population count results were analyzed via ANOVA and T test at the 5% significance level.

### RESULTS

In the experimental contamination, 2.4 × 10<sup>9</sup> to 7.2 × 10<sup>10</sup> CFU.L−<sup>1</sup> milk were inoculated, resulting in populations of 1.1 × 10<sup>8</sup> and 7.6 × 10<sup>9</sup> CFU.mL−<sup>1</sup> milk at the end of 10 h.

Biofilm formation and adhesion were observed in the three categories evaluated – TCP, TCS, and TCV –, as shown in **Table 1**. The average B. cereus s.s. populations in biofilms adhered to A coupons ranged from 6.3 × 10<sup>5</sup> to 3.1 × 10<sup>7</sup> CFU/cm<sup>2</sup> , with no significant differences (p > 0.05) between the three categories under study; i.e., there was no difference between adhesion of spores and vegetative cells.

Scanning electron microscopy revealed adhesion of vegetative cells of B. cereus s.s. to the stainless-steel surface (**Figure 4**) and production of supposed extracellular matrix (**Figure 5**).

After the coupons were sanitized by the CIP system, simulated on a pilot scale, despite the reduction of B. cereus counts, which varied between 3.6 × 10<sup>5</sup> and 1.1 × 10<sup>7</sup> CFU/cm<sup>2</sup> , biofilm removal was not complete with the disinfectants tested. The biofilms evolved to more-structured stages (**Figures 6**, **7**), and organic residues from milk were observed (**Figure 8**).

The B. cereus counts in coupons cleaned with water only, peracetic acid, and sodium hypochlorite ranged from 1.5 × 10<sup>5</sup> to 8.7 × 10<sup>6</sup> , 2.5 × 10<sup>5</sup> to 8.5 × 10<sup>6</sup> , and 3.0 × 10<sup>4</sup> to 4.6 × 10<sup>6</sup> , respectively. No statistically significant difference was observed (p > 0.05) between the populations after the sanitizers were applied and after the simple use of water in the final stage of the CIP procedure, which was simulated on a pilot scale.

### DISCUSSION

Research points to a greater adhesion capacity of spores as compared with vegetative cells (Faille et al., 2001; Branda et al., 2005; Jang et al., 2006). However, this was not observed in the results of the current study, where no significant difference was detected (p > 0.05) between adhesion and colonization in the studied categories.

This greater adhesion may not be related to multispecies biofilms, since spores of B. cereus microorganisms have a competitive advantage over other species due to the reduction of the competing load resulting from the thermal treatment applied to the milk. In the present case, where biofilm was induced in environments (due to their synergistic activities) totally restricted to B. cereus s.s. spores, adhesion at different intensities was not evident (Sharma and Anand, 2002).

Despite the lack of differences, the initial adhesion of spores, favored by the conditioning film, where the spores germinate and vegetative cells can synthesize the matrix components, multiply

TABLE 1 | Average counts of Bacillus cereus s.s. in biofilms adhered to AISI 304 stainless-steel surfaces in contact with tindalized, contaminated, and pasteurized milk; tindalized milk contaminated with spores; and tindalized milk contaminated with vegetative cells.


TCP, tindalized, contaminated, and pasteurized milk; TCS, tindalized milk contaminated with spores; TCV, tindalized milk contaminated with vegetative cells; CA, coupon with adhesion; CC, cleaned coupon; NC, negative control; CSP, treatment with peracetic acid; CSH, treatment with sodium hypochlorite. The counts of Bacillus cereus s.s. in biofilms are presented as mean ± standard deviation.

Frontiers in Microbiology | www.frontiersin.org

FIGURE 5 | Production of supposed extracellular matrix by vegetative cells of Bacillus cereus s.s. on AISI 304 stainless-steel surface.

**101**

FIGURE 4 | Adhesion of vegetative cells of Bacillus cereus s.s. on AISI 304 stainless-steel surface.

vegetative cells (Young and Setlow, 2003). Previous studies have shown that peracetic acid does not reduce the spores of Bacillus sp. to undetectable levels, and it is considered even less effective

fmicb-09-02934 November 28, 2018 Time: 12:3 # 6

FIGURE 6 | Beginning of the process of biofilm formation by vegetative cells of Bacillus cereus s.s. on AISI 304 stainless-steel surface with cell proliferation and formation of three dimensional structure.

in the decontamination of spores when compared with chlorine dioxide and ozone (Jang et al., 2006).

This protection may be associated with several factors, including enzymatic complementation and the organized spatial distribution of cells in the biofilm (Burmølle et al., 2006). The second factor was observed in this study (**Figures 6**, **7**).

The sanitization process is divided into cleaning and disinfection, whose purpose is to remove organic and mineral residues adhered to the surfaces, besides eliminating pathogenic microorganisms and reducing the microbial load to levels considered safe, respectively (Rossi, 2008). However, we observed that in spite of the reduction of microbial load, the biofilms evolved to more-structured stages (**Figures 6**, **7**), and milk residues were detected (**Figure 8**), demonstrating that, in addition to disinfectants, the acid and alkaline components also play an important limiting the factors for conditioning films.

In this study, even after sanitization, there was microorganism adhesion on the coupons' surfaces. Other studies have demonstrated that bacteria can survive and produce biofilms inside pipes even after a properly applied CIP procedure (Shi and Zhu, 2009). In China, Zhou et al. (2008) described that the powder milk contained less B. cereus isolates than raw milk, which can be associated to the presence of spores on the raw material and biofilms formed inside the pipes that are resistant to heat, drying and CIP procedure.

In this sense, the results of this study agreed with the study of Ribeiro et al. (2017) demonstrating that the conditioning of their functions is a determining factor in the adhesion degree of Bacillus cereus biofilms.

The milk residues render surfaces more prone to bacterial adhesion and consequent biofilm formation (Machado, 2005), as Surface conditioning changes its physical-chemical properties, may affect the order of adhesion events and biofilm formation and increase microbial fixation (Simões et al., 2010).

Biofilms associated with dairy processing industries are favored by the presence of conditioning films, generated mostly by residual milk, which allows for the accumulation of organic and inorganic milk compounds (Ribeiro, 2015). In addition, disinfectants are less effective when there are residues of organic materials such as fats, sugars and proteins (Wirtanen and Salo, 2004). Therefore, it is essential that hygiene procedures are conducted using detergents to remove residues of organic and inorganic food associated with sanitization by the use of physical or chemical agents to control microorganisms (Salustiano et al., 2010).

In this way, to prevent and combat the contamination of equipment by biofilms, especially given the trends of use of increasingly complex and automated equipment and plants, increasingly rigorous microbiological parameters must be adopted and required (Pasvolsky et al., 2014). Even under the condition of using automatic CIP plants, a constant microbiological control over the efficacy of the sanitization should be implemented in the dairy industries to ensure safe products (Kukhtyn et al., 2017).

Peracetic acid is considered effective against spores (Mohan et al., 2009; Buhr et al., 2013). The research has shown that despite their resistance, peracetic acid can inactivate spores. Apparently, this mechanism does not occur through damage to the DNA,

but probably as a result of alterations in the external layers of spores, more specifically the interim layer, such that when they germinate the membrane will break (Buhr et al., 2013; Park et al., 2014; Leggett et al., 2016). However, the mechanism of spore inactivation by peracetic acid is not fully known.

To improve the ability to destroy and remove biofilm from food processing facilities, the use of combined sanitizing treatments with other methods may be more effective than the use of any factor alone (Ban and Kang, 2016). It is possible that higher concentrations of peracetic acid are more effective against spores; however, using high concentrations of sanitizers in industrial environments is a challenge, as it may favor the corrosion process (Jang et al., 2006).

To improve the ability to destroy and remove biofilm from food processing facilities, the use of combined sanitizing treatments with other methods may be more effective than the use of any factor alone (Ban and Kang, 2016). Previous studies have shown that the bactericidal effect of peracetic acid is much more effective when combined with gaseous components such as carbon monoxide (White et al., 2006). The decontamination of spores with chlorine dioxide and ozone is more effective against spores of Bacillus sp., Because peracetic acid does not reduce to undetectable levels (Jang et al., 2006).

The combined treatment of disinfectant and steam is a very promising alternative technology to control biofilm cells from non-spore-forming pathogens. However, spore efficiency is still studied (Ban and Kang, 2016). In addition to these studies, a number of other approaches can be used to control biofilms in the dairy industry such as altering the chemical nature of the surface to prevent cell binding, treating surfaces with antimicrobial agents, and optimizing equipment design, processes and CIP cleaning regimes remain particularly important (Gopal et al., 2015).

The membrane of spores also provides resistance to hypochlorite (Sabli et al., 1996). In contrast, spores treated with hypochlorite have difficulty germinating, although those resisting treatments with hypochlorite do not present visible damage to the DNA (Young and Setlow, 2003). Both the nutrient receptors and the receptors of lytic enzymes from the cellular cortex of spores are possibly greatly damaged by treatment with hypochlorite. The authors believe hypochlorite damages the membranes by oxidizing fatty acids or by oxidizing membrane proteins, which can both occur (Young and Setlow, 2003).

Results point to a lack of differences between the use of the sanitizers applied in this study and the use of water during the final stage of the CIP procedure. This may be a consequence of the vortex effect caused by agitation, which is probably effective in the removal of microorganisms with weak adhesion.

The stage of bacterial adhesion is considered fundamental; without it, the process does not evolve (Wirtanen et al., 1996; Klemm et al., 2010). It is also the most vulnerable to the action of disinfectants (Ghigo, 2003; Webb et al., 2003).

A characteristic of CIP operations is their variable effectiveness in eliminating planktonic bacteria or surface-adhered biofilms (Faille et al., 2001; Marchand et al., 2012). Studies show that even using modern washing agents and disinfectants in the sanitization of dairy equipment, the equipment is not sterile after standard sanitization (Kukhtyn et al., 2017). This variability is due to an array of interfering factors; e.g., nature, time, and composition of biofilm; composition, concentration, time, and temperature of the cleaning agent; turbulence of the cleaning solution; among others. For this reason, the ideal CIP regime may vary across processing industries and also in specific points over time, in a given plant (Marchand et al., 2012).

Biofilm formation by B. cereus s.s. was not influenced by pasteurization in the study conditions, demonstrating that once installed in the processing line, biofilms can mature and disperse, colonizing also pasteurizing and packing machines (Eneroth et al., 2001; Faille et al., 2014).

Bacillus cereus s.s. survives pasteurization because of sporulation, and after germination, the cells are free from competition with other vegetative cells (Sharma and Anand, 2002). Although the biofilms formed by more than one bacterial species show more metabolic advantages in natural environments, biofilms formed by only one species of bacterium typically colonize the substrates and surfaces of equipment in dairy industries more effectively (Parkar et al., 2003).

### CONCLUSION

Peracetic acid and sodium hypochlorite were not effective in removing the B. cereus s.s. biofilms formed on stainlesssteel coupons submerged in milk, with no difference from the same sanitization process performed without the use of disinfectants. These results are of extreme importance for dairy industries to adopt the use of efficient disinfectants in biofilm removal.

### AUTHOR CONTRIBUTIONS

HS, GR, and CA performed the experimental stages. HS and JL wrote the article. JL wrote, adapted to the guidelines, reviewed, and finalized the article. LM performed the statistical analyzes. AV supervised all the steps, reviewed, and corrected the article.

### ACKNOWLEDGMENTS

The authors thank the São Paulo Research Foundation for the financial support (Grant No. 2015/20874-0) and the National Council for Scientific and Technological Development for the doctoral fellowship (Grant No. 2014/166512-1).

### REFERENCES

fmicb-09-02934 November 28, 2018 Time: 12:3 # 10


dissertação – Departamento de Engenharia Biológica, Universidade do Minho, Minho.


Salustiano, V. C., Andrade, N. J., Soares, N. F. F., Lima, J. C., Ernardes, L. M. P., and Fernandes, P. E. (2009). Contamination of milk with Bacillus cereus by port pasteurization surface exposure as evaluated by automated ribotyping. Food Control 20, 439–442. doi: 10.1016/j.foodcont.2008.07.004

Salustiano, V. C., Andrade, N. J., Ribeiro Junior, J. I., Fernandes, P. E., Lopes, J. P., Bernardes, P. C., et al. (2010). Controlling Bacillus cereus adherence to stainless steel with different cleaning and sanitizing procedures used in dairy plants. Arq. Bras. Med. Vet. 62, 1478–1483. doi: 10.1590/S0102-09352010000600026



and cleaning procedures in closed food-processing systems. J. Food Protect. 59, 727–733.


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

Copyright © 2018 Silva, Lima, Aguilar, Rossi, Mathias and Vidal. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# New Insights Into the Response of Metabolome of Escherichia coli O157:H7 to Ohmic Heating

Xiaojing Tian1,2, Qianqian Yu1,2, Donghao Yao1,2, Lele Shao1,2, Zhihong Liang1,2, Fei Jia1,2 , Xingmin Li1,2, Teng Hui1,2 and Ruitong Dai1,2 \*

<sup>1</sup> Beijing Advanced Innovation Center for Food Nutrition and Human Health, College of Food Science and Nutritional Engineering, China Agricultural University, Beijing, China, <sup>2</sup> Beijing Higher Institution Engineering Research Center of Animal Product, China Agricultural University, Beijing, China

The objective of this study was to investigate the effects of ohmic heating and water bath heating (WB) on the metabolome of Escherichia coli O157:H7 cells at the same inactivation levels. Compared to low voltage long time ohmic heating (5 V/cm, 8.50 min, LVLT) and WB (5.50 min), the high voltage short time ohmic heating (10 V/cm, 1.75 min, HVST) had much shorter heating time. Compared to the samples of control (CT), there were a total of 213 differential metabolites identified, among them, 73, 78, and 62 were presented in HVST, LVLT, and WB samples, revealing a stronger metabolomic response of E. coli cells to HVST and LVLT than WB. KEGG enrichment analysis indicated that the significantly enriched pathways were biosynthesis and metabolism of amino acids (alanine, arginine, aspartate, and glutamate, etc.), followed by aminoacyl-tRNA biosynthesis among the three treatments. This is the first metabolomic study of E. coli cells in response to ohmic heating and presents an important step toward understanding the mechanism of ohmic heating on microbial inactivation, and can serve as a theoretical basis for better application of ohmic heating in food products.

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Ilkin Yucel Sengun, Ege University, Turkey Brandon Luedtke, University of Nebraska at Kearney, United States

> \*Correspondence: Ruitong Dai dairuitong@hotmail.com

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 17 August 2018 Accepted: 15 November 2018 Published: 06 December 2018

#### Citation:

Tian X, Yu Q, Yao D, Shao L, Liang Z, Jia F, Li X, Hui T and Dai R (2018) New Insights Into the Response of Metabolome of Escherichia coli O157:H7 to Ohmic Heating. Front. Microbiol. 9:2936. doi: 10.3389/fmicb.2018.02936 Keywords: sublethal injury, untargeted metabolomic analysis, HPLC-MS/MS, lipid metabolism, amino acid metabolism

### INTRODUCTION

Thermal treatments, as conventional technology in food processing, are used for the pasteurization, sterilization, dehydration, evaporation, and blanching of foods. Generally, heat energy is generated externally and then transferred into the internal of food by conduction or convection in conventional thermal treatment methods. These methods are time consuming due to slow heat transfer through the product, particularly for larger diameter products, and may lead to overcooked surface and quality deterioration (Jaeger et al., 2016; Kanjanapongkul, 2017). Therefore, there is growing interest in alternative thermal treatment methods, which can avoid these shortcomings (Zell et al., 2010). Ohmic heating in particular is just one of these methods, where the heat is generated directly in the food when electric current passes through conductive food (Hradecky et al., 2017). Compared to conventional thermal treatment methods, the heat of ohmic heating is generated from the internal of the food, therefore it can prevent the surface of the solid food or particles from becoming overheated and preserve sensory attributes of food with a shorter heating time (Kanjanapongkul, 2017).

Besides the superior processing characteristics of ohmic heating, the inactivation effect on microorganisms also attracts many researchers' interest. Up to now, ohmic heating has been widely applied to inactivate vegetative cells and spores in various food, such as Salmonella in buffalo milk (Kumar et al., 2014), Escherichia coli O157:H7, Salmonella Typhimurium, and Listeria monocytogenes in orange and tomato juice (Sagong et al., 2011) and in salsa (Kim and Kang, 2017), Listeria innocua in meat (Zell et al., 2010), Alicyclobacillus acidoterrestris spores in apple juice (Kim et al., 2017), Bacillus licheniformis spores in cloudberry jam (Pereira et al., 2007), and Bacillus cereus spores in doenjang (Ryang et al., 2016). Most studies indicated that ohmic heating had a comparable or even better inactivation effect on microorganisms than conventional thermal treatment methods, and the possibility of non-thermal inactivation effect of ohmic heating was also referred (Pereira et al., 2007; Somavat et al., 2012; Tian X.J. et al., 2017). However, few studies have successfully proved the existence of the specific non-thermal effects and the effects of ohmic heating on intracellular content on the molecular level.

In recent years, transcriptomic, proteomic, and metabolomic collectively referred to as functional genomics techniques, are becoming increasingly important in the life sciences. These techniques can give new insights and a better understanding of the biological function of a cell or organism (Wu and Li, 2013). Metabolites (MW < 1000) are the building blocks of DNA, RNA, proteins, and lipids and play important roles in cell metabolism, signaling, and regulation for all organisms (Mousavi et al., 2016), their composition or contents can directly reflect the phenotype changes of living organisms (Tian J. et al., 2017). The metabolomics can generate the metabolic profiles of living systems at a specified time and specific environmental conditions. Recently, microbial metabolomics has received much attention due to its potential applications in a wide range of research areas, and several studies have shown metabolite changes when microorganisms were exposed to limited nutrient (Jozefczuk and Al, 2010; Lu et al., 2016), heat (Jozefczuk and Al, 2010), cold (Jozefczuk and Al, 2010; Alreshidi et al., 2015), bacteriostatic agent (Mousavi et al., 2016), toxic (Planchon et al., 2017), and oxidative stress (Jozefczuk and Al, 2010).

Untargeted metabolomics have the capacity to implicate previously unexplored biochemical pathways in a particular biological condition, because it can simultaneously detect as many metabolites as possible to maximize the opportunity of identifying compounds (Yanes et al., 2011). The purpose of this study was to investigate the metabolomic response of E. coli O157:H7 cells exposure to ohmic heating using untargeted metabolomic method. Furthermore, conventional water bath heating was performed in order to compare the metabolomic response of E. coli to ohmic heating.

### MATERIALS AND METHODS

### Bacterial Strain and Culture Conditions

E. coli O157:H7 (NCTC 12900) was used as experimental strain. Cell cultures were obtained by the same method as our previous study (Tian et al., 2018b). When cultures reached the mid-exponential growth phase (4 h, OD<sup>600</sup> = 0.314) they were centrifuged at 5000 × g, 4◦C for 10 min, washed twice with phosphate buffered saline (pH 7.2 ± 0.1, electrical conductivity 1.23 ± 0.09 S/m, 0.09 M NaCl, 0.03 M Na2HPO4, and 0.01 M NaH2PO4, PBS), and then re-suspended in the same PBS as above at final concentration approximately 2 × 10<sup>8</sup> colony-forming units per milliliter (CFU/mL) before treatment.

### Heat Treatments

### Water Bath Heating

Water bath heating (WB) was carried out by the same thermostatic water bath equipment, heating cell, and heating method as our previous study (Tian et al., 2018b). In brief, the heating cell was filled with approximately 160 mL E. coli cell suspension and was heated at 80◦C water bath accompanied by constant shakes (15 times/min) in order to achieve uniform heating. The same inactivation levels among WB and ohmic heating protocols for metabolomic analysis were designed to reduce less than 2 log CFU/mL of E. coli cell on tryptone soy agar (TSA, pH 7.3 ± 0.2, HB0177, Qingdao Hope Bio-Technology Co., Ltd). Finally, it required 5.50 min for WB to achieve this inactivation level. After heating, the samples were cooled to 4◦C in an icewater mixture immediately, then the samples were pelleted by centrifugation at 5000 × g for 12 min (4◦C), quenched using liquid nitrogen immediately, and stored at −80◦C until metabolomic analysis. Each treatment was performed in six biological replicates, and samples without any treatment were served as control (CT).

### Ohmic Heating

The ohmic heating equipment (frequency 50 Hz) used was the same as our previous study (Dai et al., 2013), and the experiment was carried out according to the method of Tian et al. (2018b). Approximately 160 mL E. coli cell suspension was assigned to the heating cell and was treated by 10 V/cm and 1.75 min (high voltage short time, HVST), and 5 V/cm and 8.50 min (low voltage long time, LVLT), respectively. After heating, samples were treated for subsequent analysis using the same method as WB samples.

### Measurement of Inactivated and Injured E. coli Cells

The inactivation of E. coli cells was measured by counting plates according to the methods described in our previous study (Tian et al., 2018b). TSA, thin agar layer (TAL), and selective medium Improved-MacConkey sorbitol agar (IMSA, pH 7.0 ± 0.2, 02–328, BeiJing AoBoXing Bio-Technology Co., Ltd) were used to assess the dead, alive, and sublethal status of E. coli cells after different treatments. Samples were diluted serially in sterile 0.85% physiological saline, and an appropriate dilution of 100 µL was spread on the three plates. The plates were incubated at 37◦C for 24–48 h before enumeration, and each dilution was performed in duplicate. The sublethal ratio was calculated according to the following equation (1) (Bi et al., 2015):

$$\begin{aligned} \text{The subtletal ratio (\%)} &= \\ 100 - \text{(CFU/mL\_{IMSA})/(CFU/mL\_{TAL})} &\times 100 \end{aligned} $$

Where CFU/mLIMSA represented the colony counts on the IMSA; CFU/mLTAL represented the colony counts on the TAL.

### Untargeted Metabolomic Analysis

### Sample Extraction and Preparation

fmicb-09-02936 December 5, 2018 Time: 10:15 # 3

The cell precipitate was added 20 µL L-2-chlorphenylalanine solution (0.3 mg/mL, dissolved in methanol) as internal standard, and then the mixture was transferred to the glass vial with pre-cooled methanol/water (4:1, v/v). The mixture was added 200 µL chloroform and resuspended. The suspension was kept on ice and the cells were disrupted by an ultrasonic cell disruptor (500 W, 6 min, 6 s on, 4 s off) (JY98-IIIN, NingBo XinZhi Bio-Technology Co., Ltd). Then metabolites inside the cells were released by an ultrasonic cleaner for 20 min (TYHD-600, Beijing TianYou HengDa Bio-Technology Co., Ltd). After ultrasound, the suspension was centrifuged at 10,000 × g, 4◦C for 15 min, and the supernatant was collected. One mL of supernatant was added into 1.5 mL centrifuge tube (twice, 0.5 mL each time) and was volatilized. Dried extracts were dissolved in 250 µL methanol/water (7:3, v/v) followed by vortex 30 s, ultrasonic 2 min. Next, the suspension was centrifuged at 10,000 × g, 4◦C for 15 min, and 180 µL of supernatant was transferred to the sample bottle with lining tube for HPLC-MS/MS analysis.

### HPLC-MS/MS Analysis

Metabolites of E. coli cells with different treatments were analyzed by instrument platform of ultra-high-performance liquid chromatograph-dual pressure linear well-electrostatic field orbital well tandem mass spectrometer (UHPLC-LTQ Orbitrap, Thermo Fisher Scientific, United States). An acquity BEH C18 column (100 mm × 2.1 mm, i.d., 1.7 µm; Waters, Milford, CT, United States) was used. The mixed mobile phase contained formic acid-aqueous solution (0.1%, v/v, A) and acetonitrile solution containing 0.1% formic acid (v/v, B), and gradient elution was as follows: 5–25% B over 0–1.5 min, 25–100% B over 1.5–10 min, holding at 100% B over 10–13 min, 100–5% B over 13–13.5 min, and holding at 5% B over 13.5–14.5 min. The column was maintained at 45◦C. Injection volume was 3.00 µL and flow rate was 0.40 mL/min.

An electrospray ionization (ESI) source in either positive or negative ion mode was used to acquire mass spectra profiles. The electrospray capillary voltage, input voltage, and collision voltage were 3.0 kV, 40 V, and 30 eV, respectively. The capillary and ion source temperature were all set at 350◦C, with a carrier gas flow rate of 45 L/h. The acquired mass data was collected from 50 to 1000 m/z with the resolution of 30,000.

Quality control (QC) sample was prepared by mixing all samples equivalently to be a pooled sample, and then analyzed using the same method with the analytic samples. The QC samples were injected at regular intervals (every 10 samples) throughout the analytical run to provide a set of data from which repeatability could be assessed.

### Data Processing

The metabolomics processing software progenesis QI (Waters Corporation,Milford, CT, United States) was used for baseline filtering, peak identification, integration, retention time correction, peak alignment, and normalization of the data sets. The retention time, mass ratio, and peak intensity was obtained.

The positive and negative data were combined to get a combine data set which was imported into SIMCA-P+ 14.0 software package (Umetrics, Umeå, Sweden). Firstly, the unsupervised principle component analysis (PCA) was carried out to visualize the overall distribution of the samples and the stability of the whole analysis process. Then, the supervised (orthogonal) partial least-squares-discriminant analysis [(O) PLS-DA] was performed to find the inter-group differential metabolites. Variable importance in projection (VIP) ranked the overall contribution of each variable to the (O) PLS-DA model, and those variables with VIP > 1 were considered relevant for group discrimination. In this study, the default 7-round crossvalidation was applied with 1/seventh of the samples being excluded from the mathematical model in each round, in order to guard against overfitting. R<sup>2</sup> and Q<sup>2</sup> values were used to evaluate the accuracy and predictive ability of the models.

Multidimensional analysis of (O) PLS-DA and singledimensional analysis (student t-test) were used to screen the inter-group differential metabolites (VIP > 1, P < 0.05). The metabolites were identified using human metabolome database<sup>1</sup> and METLIN database<sup>2</sup> . Then differential metabolites were annotated by KEGG database, including participating pathway and functional classification.

### Statistical Analysis

One-way variance (ANOVA) analysis was carried out by SPSS 21 software (IBM, United States), and results were considered to be statistically significant at P < 0.05. All experiments were performed in sextuplicate, and the values were given as means ± standard deviation of six replicates.

### RESULTS AND DISCUSSION

### Inactivation and Sublethal Injury of E. coli Cells

In order to obtain appropriate treatment conditions for the metabolomic analysis, E. coli cells were HVST, LVLT and WBtreated for 1.75, 8.50, and 5.50 min with final temperature of 57.90, 58.60, and 59.50◦C, respectively (**Table 1**). Under these treatment conditions, cell counts ranged from 6.81 to 6.83 log CFU/mL on TSA, ranged from 6.40 to 6.48 log CFU/mL on TAL, and ranged from 5.57 to 5.60 log CFU/mL on IMSA. There was no significant difference of logarithmic reduction on the same plate among HVST, LVLT, and WB-treated samples with similar final

<sup>1</sup>http://www.hmdb.ca

<sup>2</sup>https://metlin.scripps.edu


TABLE 1 | Inactivation of E. coli O157:H7 cells by HVST, LVLT, and WB treatments.

Values were means ± standard deviation of six replicates. Values in the same column with different letters (a-b) were significantly different (P < 0.05).

temperature (P < 0.05). The initial E. coli populations were 8.24, 8.16, and 8.07 log CFU/mL on TSA, TAL, and IMSA, respectively. There should be no significant difference of colony counts on TSA, TAL, and IMSA if no sublethally injured cell was induced by the three treatments (Chueca et al., 2015; Tian et al., 2018a). However, as shown in **Table 1**, the counts of E. coli cells on TSA and TAL were higher than that of IMSA after HVST, LVLT, and WB treatments, which indicated the existence of sublethal injury cells after the three treatments. Based on the difference of plate counts on TSA before and after treatments, there were more than 90% of the E. coli cells killed. According to Equation (1), more than 80% cells was sublethally injured after the three treatments (**Table 1**). At this inactivation degree, there were enough living cells and a large proportion of sublethally injured cells could respond to heat treatments. Because these treatments provided appropriate conditions for metabolomic analysis of E. coli cells.

In this study, the heating time for HVST was shorter than that of LVLT with the same inactivation levels of E. coli cells, which suggested that a higher voltage gradient could cause a comparable inactivation effect of E. coli cells at the similar final temperature with a shorter heating time. Our results were similar to those reported by other researchers. Kim et al. (2018) reported that increasing treatment voltage gradients (9.43–12.14 Vrms/cm) was an effective way to inactivate E. coli O157:H7, S. Typhimurium, and L. monocytogenes by continuous-type pulsed ohmic heating in buffered peptone water and tomato juice at final temperature of 80◦C. When 30, 40, and 50 V/cm voltage gradients were used to inactivate Alicyclobacillus acidoterrestris spores in orange juice, Baysal and Icıer (2010) found that the higher voltage gradient had a more effective inactivation effect on A. acidoterrestris spores at final temperature of 70◦C. Lee et al. (2012) also suggested that the most effective treatment voltage was 40 V/cm compared to 30 and 35 V/cm for inactivating E. coli, S. Typhimurium, and L. monocytogenes in orange juice and tomato juice at final temperature of 75.5◦C.

### Multivariate Statistical Analysis

Metabolome-based class separation was presented in the PCA score plot (**Figure 1**), the HVST, LVLT, and WB-treated samples were separated from the CT samples. (O) PLS-DA (**Figure 2**) models were developed for comparison of HVST vs. CT, LVLT vs. CT, and WB vs. CT treated samples. The models displayed good descriptive and predictive abilities, expressed as follows: R 2 (Y) = 0.992, and Q<sup>2</sup> = 0.937 in HVST vs. CT; R<sup>2</sup> (Y) = 0.997, and Q<sup>2</sup> = 0.931 in LVLT vs. CT; R<sup>2</sup> (Y) = 0.991, and Q<sup>2</sup> = 0.973 in WB vs. CT (data was not shown).

Statistical analysis indicated that after the three treatments, a total of 213 metabolites demonstrated the existence of metabolomic differences with VIP > 1 and P < 0.05, as compared with CT samples (**Table 2**). Of the 213 differentially expressed metabolites, 73, 78, and 62 metabolites belonged to HVST vs. CT, LVLT vs. CT, and WB vs. CT treated samples, respectively. These metabolites were distributed into 6 functional groups: lipid metabolism (20 metabolites), amino acid metabolism (30 metabolites), nucleotide metabolism (32 metabolites), energy metabolism (8 metabolites), carbohydrate metabolism (8 metabolites), and others (8 metabolites), and some metabolites might belong to more than one category. Moreover, there were 46 overlapping compounds among HVST, LVLT, and WB treatments.

## Classification Analysis of Differential Metabolites

### Lipid Metabolism

Lipid-based metabolism is vital to many biochemistry reactions and related to many biological functions, especially essential to the formation of cell membrane. There were 14, 12, and 14 lipid-metabolism-related differential compounds after HVST, LVLT, and WB treatments, respectively, and most of them were up-regulated, expect for sphinganine and lysoPE (0:0/14:1(9Z)) (POS mode). Among them, only arachidic acid, cPA (18:0/0:0), D-Glycerate 3-phosphate, and myristic acid belong to saturated fatty acids, all the others belong to unsaturated fatty acids. Lysophosphatidylcholine (lysoPC), derived from the hydrolysis of phosphatidylcholines (PC) by phospholipase A2, is of great importance to the cell and participates in many physiological functions (Liu et al., 2013). An increase of lysoPC, such as lysoPC [14:1(9Z)] and lysoPC (15:0) indicated a disturbance of phospholipid catabolism in E. coli cells. Lysophosphatidylethanolamine (lysoPE), a constituent of cell membranes, derived from the hydrolysis of phosphatidylethanolamines (PE), which is catalyzed by phospholipase A2 (Tepperman and Soper, 1999). The upregulation of lysoPE (0:0/14:0), lysoPE [0:0/14:1(9Z)], lysoPE (0:0/15:0), and lysoPE (0:0/16:0) indicated the changes of phospholipid metabolism or cellular damage. Sphinganine involved in the pathway of sphingolipid metabolism, was up-regulated after WB, down-regulated after HVST, but was not affected by LVLT. This indicated that WB promoted

sphingolipid metabolism, whereas HVST inhibited its metabolism, which meant that HVST and WB exerted greater damage to sphingolipid metabolism than LVLT. As indispensable components of cell membranes, sphingolipid might be among the first cell component to encounter extracellular stresses (Liu et al., 2013). Fatty acids, especially unsaturated fatty acids, are known to induce decrease in cell respiratory activity, membrane fluidity, and coagulation of cytoplasmic materials, and eventually lead to cell lysis followed by leakage of macromolecules (Mousavi et al., 2016), and play important roles in the environmental stress. The cell membrane should retain its structural integrity as much as possible to antagonize the heat shock stress (Tian J. et al., 2017). Tian J. et al. (2017) reported that the decreased oleic acid content of Saccharomyces cerevisiae might be a self-protection mechanism of ethanol-adapted strains to maintain membrane integrity through decreasing membrane fluidity. In this study, the increased fatty acid content, especially unsaturated fatty acid might mean that the membrane integrity was damaged and the membrane fluidity of E. coli cells increased. The increase of lipid metabolites could mainly attribute to the heat of the treatments, which might result in remodeling the composition and structure of the cell membrane. This result was consistent with previous study that microorganisms could manifest increased resistance to environmental stress and control strategies after sublethal injury (Liu et al., 2018).

### Amino Acid Metabolism

In general, stress could reduce membrane fluidity and accelerate the synthesis of some proteins (Zhang and Rock, 2008). The metabolomic analysis revealed that amino acid metabolism was strongly induced in E. coli cells by all the three treatments. There were 24, 24, and 20 differential compounds involved in amino acid metabolism after HVST, LVLT, and WB treatments, respectively. Among them, there were 15 amino acids overlapping the three treatments, and showed consistent trend of content change. However, in contrast to lipid-based metabolism, most of the differential amino acids were down-regulated, only indole and ketoleucine were upregulated after LVLT. The observed increase in levels of indole and ketoleucine could attributed to protein denaturing and inhibition of protein synthesis, which might be caused by a halt in the synthesis of essential enzymes (Mousavi et al., 2016). Accordingly, the decreased levels of most amino acids might indicate a weakening in enzymatic activity related to protein degradation or a strengthening in enzymatic activity related to protein synthesis. For instance, glutamate is the direct ammonia assimilation product by glutamate dehydrogenase with high external ammonia concentrations, which subsequently serves as a primary precursor in multiple biosynthesis pathways, and it is usually synthesized more in actively growing cells (Lu et al., 2016). Glutamate was down-regulated after the three treatments, indicating that cell activity was reduced. The reductions of amino acids might reflect their consumption to produce new essential proteins or to repair damaged or misfolded proteins, in order to facilitate acclimation to the changing environment. The microorganisms are prone to rapidly changing when the environment conditions change, such as shifts in temperature, osmotic pressure, pH, or nutrient availability, and they have developed many strategies to cope with such unfavorable conditions. Among these strategies, acquisition of

thermotolerance is mainly controlled by the activation and regulation of heat stress-related genes involved in the synthesis of specific compounds that protect the microorganism from thermal stress, which involves in induction of several proteins including stress proteins and chaperones (Paul et al., 2012; Dong et al., 2017). Therefore, in order to survive from the three treatments, E. coli cells might synthesize more molecular chaperones to regulate metabolism, which resulted in reduction of the most amino acids. Another reason for more downregulation of amino acids might due to the ATP deficiency in response to heat shock from the three treatments, where synthesis of amino acids was required ATP as the energy source (Li et al., 2015).

### Nucleotide Metabolism

As precursors of DNA and RNA, nucleotides participate in cell signaling and regulate many metabolic pathways, and play a vital role in stress response (Liu et al., 2013). Most purines and pyrimidines are present in the cell as nucleotides, and they are involved in the biosynthesis of genetic information carriers (DNA and RNA) or suppliers of energy (ATP and GTP) (Hu et al., 2017). The changes in purine and pyrimidine metabolism can suggest the increase of DNA damage and cell turnover (Zhou et al., 2017), where the nucleotide biosynthesis are direct indicators of DNA replication, cell division, and growth status, revealing a pronounced effect on cell proliferation (Bhat et al., 2015). The up-regulation and down-regulation changes of metabolites

#### TABLE 2 | List of differential metabolites from HVST vs. CT, LVLT vs. CT, and WB vs. CT treated E. coli O157:H7 cells.


(Continued)

#### TABLE 2 | Continued

fmicb-09-02936 December 5, 2018 Time: 10:15 # 8


(continued)

### TABLE 2 | Continued

fmicb-09-02936 December 5, 2018 Time: 10:15 # 9


involved in nucleotide metabolism did not behave like lipid metabolism or amino acid metabolism, although more than half of the metabolites showed significant increase after the three treatments. Adenine was up-regulated after the three treatments, but its derivatives (adenosine, cAMP, ADP, and ATP) changed differently among the three treatments. Adenosine was downregulated after HVST and WB, and was up-regulated after LVLT; cAMP was up-regulated after HVST and WB, and was downregulated after LVLT; as one of the five kind nucleotides of DNA or RNA synthesis, the change of cAMP meant that the DNA or RNA synthesis was suppressed by HVST and WB, but was promoted by LVLT. In the meantime, ADP and ATP were only up-regulated after LVLT, the reason might be that the more energy was required in DNA synthesis during LVLT. Cytosine and its derivatives (5<sup>0</sup> -CMP and CTP) were all up-regulated after LVLT, this change was similar to adenines. Hypoxanthine was down-regulated after the three treatments, and the derivative inosine (IMP) could be converted to AMP and GMP, its upregulation could promote generation of AMP and GMP during LVLT. In this study, guanine and its derivatives (cGMP and dGMP) were up-regulated after the three treatments, which meant that the IMP was mainly converted to AMP, and this result provided evidence that DNA or RNA synthesis were promoted. Uracil was down-regulated after HVST and WB, but was up-regulated after LVLT, which indicated that RNA synthesis was disturbed by the three treatments. However, UMP was down-regulated and UDP was up-regulated after LVLT, which meant that mRNA synthesis was promoted by LVLT; this result was consistent with the down-regulation of most amino acids, where some proteins were synthesized to resist stress. Previous study also proved that genes of Streptococcus agalactiae involved in purine metabolism were significantly up-regulated at 40◦C than 30◦C in the study of S. agalactiae transcriptomic analysis (Mereghetti et al., 2008).

### Carbohydrate Metabolism

Several changes were observed in the levels of the metabolites that were involved in carbohydrate metabolism. Specifically, significant changes of metabolites (citric acid, malic acid, succinic acid, and lactic acid) related to tricarboxylic acid cycle (TCA cycle) were observed. Citric acid was up-regulated after HVST and LVLT, which was synthesized from oxaloacetic acid; but malic acid, as the precursor of oxaloacetic acid, was down-regulated after HVST and WB, and up-regulated after LVLT. Succinic acid was down-regulated after the three treatments, but lactic acid was only down-regulated after WB. The changes of metabolites involved in TCA cycle might be due to the energy requirement during synthesis of proteins and nucleotides. Additionally, citric acid is reported as a powerful chelator and it may play a role in managing concentrations of cations such as Ca2<sup>+</sup> for survival, and there is evidence suggesting that Ca2<sup>+</sup> is involved in the regulation of cell division and gene expression in response to external stimulation in prokaryotes (Alreshidi et al., 2015). Citric acid was up-regulated after HVST and LVLT, and the reason might be that the combination of citric acid and cations was damaged by the electric current during HVST and LVLT. α-ketoisovaleric acid, a branched-chain organic acid, served as a precursor in leucine and valine synthesis, the up-regulation after LVLT might be required by L-Valine synthesis (Li et al., 2017).

### Energy Metabolism

Energy is required in basic metabolism, which includes syntheses of proteins, DNA, and RNA (Li et al., 2015). There were 8 metabolites involved in energy metabolism, 7 of them were presented after LVLT, but only 3 and 4 metabolites were presented after HVST and WB. The up-regulation of energy storage compounds (ATP and ADP) during LVLT indicated that they might be used to offset the negative effects of heat or the prolonged electric current stimulation from LVLT, thereby maintaining basic cellular reaction rates (Lu et al., 2016). Similar study reported that temperature variation perturbed the metabolic status of S. agalactiae, including energy metabolism processes, synthesis of proteins, contents of nucleotides, selective utilization of carbon sources, and some cellular materials (Hu et al., 2017). Another study on the metabolomic response of E. coli exposed to titanium dioxide nanoparticles also indicated that metabolites related to energy and growth were up-regulated (Planchon et al., 2017). The higher metabolomic changes of E. coli involved in energy metabolism might also be responsible for the increased heat tolerance of cells, also be partly responsible for the electric current from LVLT.

### Enrichment Analysis of the Differential Metabolites

The KEGG pathway enrichment analysis was performed by Fisher's exact test, and those with P < 0.05 were considered

significant pathways. This analysis could provide some additional clues about the complex identified metabolites. As shown in **Figure 3**, the statistical data revealed that differentially expressed metabolites were enriched to 28, 29, and 25 pathways from HVST vs. CT, LVLT vs. CT, and WB vs. CT treated E. coli cells, respectively. The top 5 pathways of enrichment ratio were aminoacyl-tRNA biosynthesis, 2-oxocarboxylic acid metabolism, alanine, aspartate and glutamate metabolism, arginine biosynthesis, beta-Alanine metabolism response to HVST; alanine, aspartate and glutamate metabolism, aminoacyltRNA biosynthesis, arginine biosynthesis, biosynthesis of amino acids, valine, leucine, and isoleucine biosynthesis response to LVLT; and aminoacyl-tRNA biosynthesis, biofilm formation, arginine biosynthesis, alanine, aspartate, and glutamate metabolism, and pantothenate and CoA biosynthesis response to WB, respectively. These results suggested that the most significantly changed metabolites mainly affect biosynthesis and metabolism of amino acid (alanine, arginine, aspartate, and glutamate, etc.) followed by aminoacyl-tRNA biosynthesis among the three treatments.

### CONCLUSION

In summary, in order to obtain similar inactivation levels of E. coli cells by HVST, LVLT, and WB, the required time for HVST (1.75 min) was shorter than LVLT (8.50 min) and WB (5.50 min). The major functional group of metabolites that displayed upregulation after the three treatments were metabolites involved in lipid metabolism, while a down-regulation was metabolites involved in amino acid metabolism. On the whole, a stronger metabolomic response was caused by HVST and LVLT compared

### REFERENCES


with WB, indicating that electric current might target partial metabolites during ohmic heating. This study provided a detailed description of overall metabolic responses of E. coli cells to ohmic heating, which would facilitate the understandings of ohmic heating on microbial inactivation on the molecular level. In addition, the results described here could provide a theoretical basis for ohmic heating on microbial inactivation in food products, and further facilitate the application of ohmic heating in food industry.

### AUTHOR CONTRIBUTIONS

RD was the fund manager of the grants received from the National Key R&D Program of China (2016YFD040040302) and National Natural Science Foundation of China (No. 31271894), and directed and supervised the whole experimental and writing process. XT performed the research plan, experimental process, data analysis, and manuscript writing. QY participated in part of the experimental process and conducted part of the data analysis. DY and FJ participated in part of the experimental process. LS participated in part of the data analysis. XL participated in part of the research plan and provided valuable advice. ZL helped in the metabolomic analysis of microbes. TH helped in the draft revision of the manuscript.

### FUNDING

This work was supported by grants from the National Key R&D Program of China (2016YFD040040302) and grants from the National Natural Science Foundation of China (No. 31271894).


buffered peptone water and tomato juice. LWT 93, 316–322. doi: 10.1016/j.fm. 2016.10.021


monocytogenes in orange and tomato juice using ohmic heating. J. Food Protect. 74, 899–904. doi: 10.4315/0362-028X.JFP-10-552


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

Copyright © 2018 Tian, Yu, Yao, Shao, Liang, Jia, Li, Hui and Dai. 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.

# Surface-Enhanced Raman Scattering (SERS) With Silver Nano Substrates Synthesized by Microwave for Rapid Detection of Foodborne Pathogens

Caijiao Wei, Mei Li and Xihong Zhao\*

Research Center for Environmental Ecology and Engineering, Key Laboratory for Green Chemical Process of Ministry of Education, Key Laboratory for Hubei Novel Reactor & Green Chemical Technology, School of Environmental Ecology and Biological Engineering, Wuhan Institute of Technology, Wuhan, China

### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

### Reviewed by:

Om V. Singh, Technology Sciences Group Inc., United States Lingxin Chen, Yantai Institute of Coastal Zone Research (CAS), China Simion Astilean, Babe ¸s-Bolyai University, Romania

> \*Correspondence: Xihong Zhao xhzhao2006@gmail.com

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 09 February 2018 Accepted: 06 November 2018 Published: 13 December 2018

#### Citation:

Wei C, Li M and Zhao X (2018) Surface-Enhanced Raman Scattering (SERS) With Silver Nano Substrates Synthesized by Microwave for Rapid Detection of Foodborne Pathogens. Front. Microbiol. 9:2857. doi: 10.3389/fmicb.2018.02857 Rapid and sensitive methods have been developed to detect foodborne pathogens, a development that is important for food safety. The aim of this study is to explore Surface-enhanced Raman scattering (SERS) with silver nano substrates to detect and identify the following three foodborne pathogens: Escherichia coli O157: H7, Staphylococcus aureus and Salmonella. All the cells were resuspended with 10 mL silver colloidal nanoparticles, making a concentration of 10<sup>7</sup> CFU/mL, and were then exposed to 785 nm laser excitation. In this study, the results showed that all the bacteria can be sensitively and reproducibly detected directly by SERS. The distinctive differences can be observed in the SERS spectral data of the three food-borne pathogens, and the silver colloidal nanoparticles can be used as highly sensitive SERS-active substrates. In addition, the assay time required only a few minutes, which indicated that SERS coupled with the silver colloidal nanoparticles is a promising method for the detection and characterization of food-borne pathogens. At the same time, principle component analysis (PCA) and hierarchical cluster analysis (HCA) made the different bacterial strains clearly differentiated based on the barcode spectral data reduction. Therefore, the SERS methods hold great promise for the detection and identification of food-borne pathogens and even for applications in food safety.

Keywords: Surface-enhanced Raman scattering, silver nanoparticles, foodborne pathogens, bioanalysis, food safety, rapid detection

### INTRODUCTION

For food safety management and monitoring, foodborne pathogens have always been an issue of concern as they can cause severe illness in humans via contaminated water or food (Zhao et al., 2017a). According to the data from the World Health Organization, there are many outbreaks and numerous deaths each year caused by Escherichia coli O157: H7, Staphylococcus aureus, Salmonella, Listeria monocytogenes, Campylobacter jejuni, and so on (Liu et al., 2017). Rapid identification and detection of pathogens are major issues for public health and food safety (Zhao et al., 2014, 2016). To date, the detection of foodborne pathogens mainly includes traditional methods, immunological methods, and molecular biology methods, but they are labor intensive,

time consuming and inconvenient for onsite detection (Erik and Anna, 2007; Roda et al., 2012; Zhong and Zhao, 2017, 2018; Wei et al., 2018). It is thus necessary to explore efficient, sensitive, fast, inexpensive and accurate methods to detect and identify pathogens (Zhao et al., 2017b).

Raman spectroscopy is a scattering spectrum discovered by the Indian physicist C.V. Raman (Krishnan, 1928). The spectrum mainly provides fingerprint information of the molecular structure according to vibrational and rotational information of matter. Due to the low intensity of conventional Raman scattering (RS), it has been severely limited in many applications. Meanwhile, Surface-enhanced Raman scattering (SERS) spectroscopy avoids the problems associated with conventional RS. When the target analyte approaches or adsorbs certain rough metal (gold, silver, etc.) nanoparticle surfaces, the signals can be enhanced by many orders of magnitude compared to normal RS (Sivanesan et al., 2014). In addition, this method handles samples easily and provides the basis for non-destructive and ultra-sensitive detection of samples (Wang and Irudayaraj, 2012). SERS, as an ultrasensitive vibrational spectroscopic technique, can detect molecules on or near the surface of plasmonic nanostructures and greatly extends the role of standard RS (Wang et al., 2012). Beyond that, SERS also inherits rich chemical fingerprint information from RS, which can be conveniently made under ambient and aqueous conditions. In particular, it also gains sensitivity by plasmonenhanced excitation and scattering and has a narrow width that is suitable for multiplex analysis (Zong et al., 2018). Therefore, SERS is rapidly emerging as a sensitive analytical tool and has been applied to many analyses and fields such as chemistry, biochemistry, microbiology, environmental sciences and so on. For example, Betz et al. (2012) detected whether melamine could be found in infant formula without the need for purification or additional equipment by using SERS. Kim et al. (2015) adopted silver nanoparticles as SERS substrates to detect C-reactive protein without using any labels; the minimum detection amounts in the buffer and in 1% serum were 0.01 and 0.1 ng mL−<sup>1</sup> , respectively. Sivanesan et al. (2014) developed nanostructured silver-gold bimetallic SERS substrates for selective identification of bacteria in human blood. Xu et al. (2015) applied label-free SERS to detect DNA with single-base sensitivity. SERS has the ability to identify single molecules by using their intrinsic vibrational fingerprint and can provide highly specific biochemical information about the components of bacterial cells, including proteins and peptides, polysaccharides, nucleic acids, phospholipids, etc. (Hou et al., 2014; Zheng et al., 2016; Zhao et al., 2018). Because each type of bacteria has its own specific biochemical information and can exhibit its characteristic peaks, this technique can rapidly identify good versus bad bacteria in the field based on its unique Raman fingerprint. Meisel et al. (2014) detected 19 species of the most important harmful bacteria via Raman microspectroscopy and built up a three-level classification model based on the whole amount of Raman data. After the first classifier differentiation of Gram-positive and Gram-negative bacteria by Raman spectra, two decision knots of the bacterial genus and species followed. The study showed that the accuracy of the identification results of each different step was in the range of 90.6–99.5%.

In the SERS study, it is important to bring target analyte or target molecular structure in contact with or in close proximity of the surface of metallic nanostructures (Dong et al., 2014; Huang et al., 2015; Li et al., 2017). For microorganisms and living cells, the colloidal nanoparticles are generally preferred as substrates (Culha et al., 2010). Gold and silver are two commonly used materials for the preparation of nano-metal substrates for SERS measurement. Compared with silver, gold is more expensive but produces weaker SERS enhancement than silver (Fan et al., 2011; Chuang et al., 2014). In addition, nanosilver has the following advantages: a high molar extinction coefficient, excellent optical properties and nanosilver aggregates having strong SERS effects (Wang et al., 2016). Therefore, the silver colloidal nanoparticles (AgNPs) were employed widely for bacteria detection. There are various methods for synthesizing AgNPs, such as reduction reaction methods, ultrasonic assisted reduction methods, electrolytic methods, light induction methods, thermal decomposition methods, microwave methods and so on (Huang et al., 2014). Among these, the microwave methods have the advantages of uniform reaction, easy nucleation and less pollution, and convenient and rapid synthesis; meanwhile, the prepared nanomaterials have high purity and uniform distribution (Du et al., 2015). Thus rapid microwave is an important technique for synthesizing metallic nanostructures. However, simple and green microwave methods for synthesizing highly SERS-active AgNPs have rarely been reported. Herein, the objective of this study is to evaluate the feasibility of adopting the microwave method to synthesize silver colloidal solutions as the SERS-active substrate for the detection and identification of foodborne pathogens.

### MATERIALS AND METHODS

### Preparation of the Silver Colloidal Nanoparticles

In this study, the microwave heating method was used to synthesize AgNPs due to the advantages of simple operation, uniform heating, fast heating and fast preparation that it holds over other methods. In brief, 1 × 10−<sup>3</sup> M of AgNO<sup>3</sup> was dissolved in 200 mL double distilled water. An aliquot of 1% sodium citrate solution (10 mL) was added dropwise with continuous stirring to an aqueous silver nitrate solution. The solution was put into a microwave oven at a power of 700 W until a yellow color solution was obtained. The prepared AgNps were characterized by UV-Vis spectrophotometry (UV-1800, Shimadzu Enterprise Management, Ltd., China) under a wavelength range of 300– 800 nm. At the same time, the particle size distribution of AgNps was determined by a ZEN3690 Malvern laser particle size analyzer (Malvern Instruments Ltd., United Kingdom). In order to more fully analyze the size, morphology and stability of nanoparticles, the as-synthesized AgNps was characterized by a transmission electron microscope (TEM, model JEOL JEM 1200EX) operated at 200 kV and the zeta potential of AgNps was measured by dynamic light scattering with Zetasizer Nano ZS (Malvern Instruments Ltd., United Kingdom). For TEM, samples were prepared by placing approximately 10 µL of the assynthesized Ag colloid dispersing in water onto a TEM grid and then drying under an IR lamp.

### Preparation of Bacterial Samples

fmicb-09-02857 December 12, 2018 Time: 15:7 # 3

All the strains from the American Type Culture Collection (ATCC, United States) were preserved at −80◦C in our laboratory until use, namely Escherichia coli O157:H7 (ATCC 43895), S. aureus (ATCC 27664) and Salmonella (ATCC 13076). Bacterial strains were stored at −80◦C in BactoTM Tryptic Soy Broth (TSB; Becton, Dickinson and Co., Sparks, MD, United States) containing 20% glycerol, were inoculated in LB agar plates and incubated at 37◦C overnight. Afterward, the pure culture was transferred to BactoTM TSB at 37◦C overnight with shaking (110 rpm). The bacterial concentration was about 10<sup>8</sup> CFU/mL. One milliliter of cell suspension was centrifuged at 6,000 × g for 5 min and the supernatant was discarded. The cell pellets were washed three times with double distilled water, centrifuged at 6,000 × g for 5 min and finally resuspended with 10 mL of the silver colloidal nanoparticles or double distilled water. Next, the solution was mixed with a vortex to obtain a mixture that was as homogeneous as possible and then stood for 3–5 min. This procedure was designed to make AgNps and bacteria adsorb each other. Transmission electron microscopy (TEM) measurements were also performed to investigate the distribution of the AgNps coated on the Escherichia coli O157: H7.

### SERS Measurement

In order to explore whether the AgNps that were prepared by the microwave heating method as the SERS substrate for detecting foodborne pathogens have an enhanced signal or not, the experiment was carried out as follows. All the Escherichia coli O157: H7, S. aureus and Salmonella samples were transferred to glass capillary tubes and then were detected using a Portable Raman Spectrometer (I0785MM0350MF, Ocean Optics Company, United States). The spectral coverage was from 400 - 2000 cm−<sup>1</sup> with a 785 nm excitation light. The laser power was 70 mW and the average scan time was 4 s. Bacteria samples without AgNps were also scanned. All the procedures were carried out in three replicates. The schematic diagram for the SERS detection of foodborne pathogens with AgNPs is shown in **Figure 1**.

### Data Analysis

The normal Raman and SERS spectra data were analyzed by Origin software version 9.0 (Origin Lab Corporation, Northampton, MA, United States) and SPSS multivariate data analysis software (version 11.5.0, SPSS Inc., Chicago, IL, United States). The baseline correction helped to determine any difference between spectra quickly (Sundaram et al., 2013), so the pre-processing algorithms were conducted to analyze the data, for instance smoothing, normalization and second-derivative transformation. For purposes of bacterial identification, principal component analysis (PCA) and hierarchical cluster analysis (HCA) were applied in this study. Prior to the two multivariate analyses, all the spectra were smoothed in order to eliminate

any high-frequency instrument background noise by averaging near data points and were then normalized to a range of 0 to 1; meanwhile, they were also subjected to second-derivative transformation to separate overlapping bands and remove baseline offsets.

### RESULTS AND ANALYSIS

### Characterization of the AgNps

The size and dispersion of AgNps have a certain effect on SERS detection, so absorption peak and particle size distribution are characterized by UV-Vis spectrophotometry and Malvern laser particle size analyzer, respectively. The characterization of the AgNPs, rapidly synthesized by green microwave technology, can be seen in **Figures 2A,B**. The UV-Vis spectrum of the AgNps reveals a maximum absorbance of 427 nm as is shown in **Figure 2A**. TEM and the corresponding size distribution revealed that the green microwave technology-synthesized AgNps have an average diameter of 65 ± 2 nm (**Figures 3A**, **2B**). In the Figures, there are no distinct peaks in the UV-Vis and DLS spectra and no aggregation in TEM images. Thus, it can also be shown that the distribution of AgNps in the solution is relatively uniform and without aggregation. In order to verify the stability of the prepared AgNps, the zeta potential of the as-synthesized Ag colloid dispersing in water was measured three times by dynamic light scattering with Zetasizer Nano ZS. The results were −40.0, −39.3, and −40.5 mV respectively, which indicated that the as-synthesized Ag colloid dispersal in water is evenly distributed and has certain stability. According to relevant literature, the bactericidal properties of the nanoparticles are size-dependent, since the only nanoparticles that present a direct interaction with the bacteria preferentially have a diameter of approximately 1– 10 nm (Morones et al., 2005; Shrivastava et al., 2007) However, in our study, the sizes of silver nanoparticles are mostly concentrated at 50–70 nm. Moreover, it takes only a few minutes from the mixing process of nano silver and bacteria to the whole process of detection. Thus, it can be stated that the proteins, purine and cell membrane of bacteria are not affected by AgNPs. Meanwhile, after placing the silver colloidal nanoparticles for 4 h, the solution was still a clear

and transparent light yellow with no deposition phenomenon, which also indicates that the silver colloid has good stability. However, in bacterial SERS experiments, whether the SERS effect can generate that depends on whether the bacteria can combine effectively with AgNps. Thus, to understand the formation of the AgNP-bacteria complex, the Escherichia coli O157: H7 coated with silver nanoparticles was performed by TEM. It can be clearly seen from **Figure 3B** that the AgNPs were successfully and uniformly synthesized around the bacterial cell wall.

FIGURE 2 | (A) UV-Vis absorption spectrum of AgNPs and (B) diameter distribution of AgNPs.

### SERS Spectra of Foodborne Pathogens and Structural Analysis

In order to determine how the AgNps enhance the SERS signal, equal concentrations of bacteria, both with and without AgNps, were tested and SERS spectra were recorded. As shown in **Figure 4A**, the foodborne pathogens without AgNps did not exhibit obvious Raman signal. However, when Escherichia coli O157: H7, Staphylococcus aureus, and Salmonella were coupled with AgNps, respectively, they exhibited a unique and significant Raman spectral signal (**Figure 4B**). These spectra were the average spectra obtained from the replicated samples. It is obvious that the SERS effect measured with the bacteria in AgNps is greatly enhanced and has stronger signal intensity than that without AgNps. It also demonstrates that SERS is capable of detecting these three foodborne pathogens with the aid of the silver colloidal nanoparticles. A comparison among the SERS spectra of Escherichia coli O157: H7, S. aureus and Salmonella is shown in **Figures 4a–c**, respectively. It can be seen from the figure that the bands of Raman vibration characteristic peaks of these three pathogens are mainly between 400 and 1650 cm−<sup>1</sup> . In order to further understand and distinguish these three different foodborne pathogens, all the characteristic peaks of the SERS signals are collected in **Table 1**. S. aureus has significant Raman vibrational peaks at 437, 545, 668, 727, 788, 915, 996, 1054, 1151, 1313, 1390, 1480, 1546, and 1621 cm−<sup>1</sup> . The Escherichia

fmicb-09-02857 December 12, 2018 Time: 15:7 # 4


coli O157: H7 SERS spectra exhibit characteristic peaks at 437, 542, 689, 792, 918, 999, 1054, 1108, 1161, 1228, 1304, 1390, 1422, 1474, 1543, and 1621 cm−<sup>1</sup> . Typical peaks at 437, 545, 668, 727, 788, 915, 996, 1054, 1151, 1313, 1390, 1480, 1546, and 1621 cm−<sup>1</sup> are observed in Salmonella. In the detection of microorganisms, Raman can provide phenotypic information on signatures from cell tissues, which are mainly attributed to proteins, lipopolysaccharides, carbohydrates, nucleic acids (DNA and RNA), peptidoglycan, quinones, cytochromes, phospholipids and some endogenous biomolecules (Nelson et al., 1992). It demonstrates the vibrational information from cell structural components. Therefore, specific information can be obtained based on the structures, conformations and attributions represented by these spectra, so that the bacteria can be classified. Peaks at about 550 cm−<sup>1</sup> are assigned to S-S stretching in proteins. Peaks at about 670 cm−<sup>1</sup> are for the cysteine stretch model present in the protein. Peaks near 690 cm−<sup>1</sup> belong to the guanine ring region of DNA/RNA.

Peaks at about 730 cm−<sup>1</sup> belong to the NAG component in the peptidoglycan structure of the bacterial cell wall (Jarvis and Goodacre, 2004). Peaks around at 918 cm−<sup>1</sup> are attributed to the vibration of the nucleic acid and the peaks between 1220 and 1640 cm−<sup>1</sup> are mainly attributed to amide I, amide II, amide III vibration and carboxylic acid stretching (Sundaram et al., 2013). The band at ∼1000 cm−<sup>1</sup> is assigned to phenylalanine according to the literature (Nelson et al., 1992). The SERS peak at 918 cm−<sup>1</sup> is due to C–COO- stretch (carbohydrates). A strong Raman band due to the COO-stretching vibration of proteins is observed at 1390 cm−<sup>1</sup> . The specific Raman peak assignments are shown in **Table 1**. Comparing the Raman peaks of Escherichia coli O157: H7, S. aureus and Salmonella, it can be seen that there are both similarities and differences between them. For example, the number of major spectral bands of Escherichia coli O157: H7, S. aureus and Salmonella exhibit clear similarities, such as bands at 1054 and 1621 cm−<sup>1</sup> , although their relative intensities are different. However, there are also obvious differences. For instance, bands at 542 and 918 cm−<sup>1</sup> are significant to Escherichia coli O157: H7 and Salmonella but not in S. aureus, while a band at 668 cm−<sup>1</sup> is unique for S. aureus and Salmonella. However, peaks at 689 and 1422 cm−<sup>1</sup> are only present in Escherichia coli O157: H7. Peaks at 885, 1256, and 1337 cm−<sup>1</sup> are only present in Salmonella. To distinguish the three foodborne pathogens, the ratio of intensities of the peaks and the unique peaks can be used. Therefore, the unique and distinct vibrational spectral information of SERS can be used to identify and discriminate between different foodborne pathogens.

### Reproducibility of SERS Spectra

According to the characterization of AgNps, substrates play a major role in the signal enhancement for SERS (Chu et al., 2008). Thus the reproducible SERS study using four different batches of silver nanoparticles was conducted to test Escherichia coli O157: H7, S. aureus and Salmonella, respectively, in this study. Each AgNps substrate was prepared independently. The three foodborne pathogens of SERS spectral reproducibility with substrates AgNps are illustrated in **Figures 5**–**7**, respectively. When bacterial samples were treated with different SERS substrates individually, the variability could be found to be common. However, in this study, the manufactured AgNps substrates did not show any change or noise in the spectra. From the figures, these three foodborne pathogens that combined four different batches of AgNps SERS spectra have a high degree of reproducibility. The results provide strong support for rapid SERS detection of foodborne pathogens.

### PCA and HCA Classification

Principal component analysis can greatly reduce the dimensionality of complex multivariate data to several principal components (PCs), eliminate random variation (noise) and can objectively capture the minimal spectral differences between the similar spectra (Zhang et al., 2012; Li et al., 2015). The SERS spectra are used as the inputs for the PCA model. Then the PCA projects data into the transformation space to maximize data variability, which makes it easier to observe the similarities and

FIGURE 5 | SERS spectra of Escherichia coli O157: H7 with four batches of prepared AgNPs.

FIGURE 7 | SERS spectra of Salmonella with four batches of prepared AgNPs.

differences in spectra. Therefore, PCA is widely used to analyze SERS spectral data variances and develop classification models to distinguish pathogens based on the SERS spectra. Hierarchical clustering analysis is performed using measurements of distances as the standard for classification. It utilizes multiple statistical values to decide the degree of affinity between different samples. A HCA dendrogram is constructed using the Ward-linkage algorithm and squared distances, which were used to evaluate the member dissimilarity (He et al., 2008). It also utilizes the corresponding dendrograms' multiple statistical results to discriminate and categorize the samples. At the same time, PCA results can be further corroborated. Considering that it was difficult to classify and identify microorganisms based on similar vibrational spectra, the PCA methods and HCA methods were employed to analyze and differentiate their SERS spectra acquired from bacterial samples in this study. **Figure 8A** shows a three-dimensional (3D) PCA plot of the first three PCs for the three bacterial strains. The PCA performed on the second derivative transformed SERS spectral data from the three foodborne pathogen samples (about 1 × 107–10<sup>8</sup> CFU/ml) over a range of 500–800 cm−<sup>1</sup> . The resulting HCA dendrogram presents a clear characterization at strain level of each analyzed foodborne pathogen. In **Figure 8B**, the three foodborne pathogen samples of Escherichia coli O157: H7, S. aureus and Salmonella can be effectively distinguished in the HCA dendrogram. The classification of HCA helps to determine the similarities and differences between groups. It can be seen from **Figure 8B** that the SERS spectra can mainly be categorized into three clusters and the result is consistent with that shown in PCA.

### DISCUSSION

The AgNps were prepared by a quick and easy microwave heating method, which was used as SERS substrates to detect and discriminate Escherichia coli O157: H7, S. aureus and Salmonella in this study. At this point, the molecular level interpretation about SERS' vibrational features has not been universally established, but we can make some general statements and identify these spectra based on the spectral position and varying intensities. In the given spectral region, all the foodborne pathogens have similar peaks, with some differences in frequency for certain peaks. Those results are consistent with previous reports (Zeiri et al., 2004; Sundaram et al., 2013). To allow for more precise distinction between the three foodborne pathogens SERS spectra, all the characteristic peaks in SERS spectra that can be attributed to different functional groups are collected in **Table 1**. Comparing the SERS spectra of these three kinds of bacteria, we found that the three Raman are similar to unique Raman peaks in themselves. S. aureus and Salmonella both have Raman peaks at 727 cm−<sup>1</sup> , which are attributed to N-acetylglucosamine (NAG) in peptidoglycans. This result also confirms that the NAG exists in both Gram-negative and Gram-positive bacteria, whereas for Escherichia coli O157: H7 this region is represented by a broad and weak peak. The similar results were obtained in the study of pathogen identification using a portable Raman spectroscopy system (Luo and Min, 2008). Moreover, all of them have many differences, which mainly come from the quantity and distribution of cellular components such as proteins, phospholipids, nucleic acids, and carbohydrates. For instance, Escherichia coli O157: H7 has unique Raman peaks at 689, 1107, 1161, 1304, and 1422 cm−<sup>1</sup> , S. aureus has its own unique Raman peaks at 966, 1151, 1313, and 1480 cm−<sup>1</sup> , while Salmonella has their own unique Raman peaks at 885, 1133, 1256, and 1337 cm−<sup>1</sup> . Meanwhile, there is also an obvious difference in the intensity of SERS vibrational peaks among these three foodborne pathogens. However, the spectral information of these three pathogens is not completely coincident with other studies. Comparing with other studies (Xie et al., 2013; Wang et al., 2016), they appear to be different peaks. It indicates that these spectra only presented ingredient information activated by metallic nanosilver.

In this study, a rapid SERS technique coupled with silver colloidal nanoparticles as substrates has been explored to identify the foodborne pathogens. Compared to traditional detection methods, this method is faster and easier to perform and has a high degree of reproducibility. All the foodborne pathogens have similar peaks, with some difference in frequency for certain peaks in the given spectral region. The spectra showed largely similar peaks, such as those at 542, 918, 1054, and 1621 cm−<sup>1</sup> , but they also have their unique peaks that can easily be distinguished. In addition, it is difficult to classify and identify microorganisms due to similar vibrational spectra, so multivariate statistical analysis of PCA and HCA were applied to explore the data and identify individual groups based on differences in the SERS spectra in this study.

### CONCLUSION

The SERS has great prospects for application in detection of foodborne pathogens. If this method is further applied, quantitative detection of viable cells from dead cells, identifying different foodborne pathogen species and subspecies level, and then establishing a bacterial SERS database could be next steps. We believe that SERS will be more universally applicable in the area of food microbiology and also provides a sensitive and efficient tool for food safety control. So far, it is still very challenging to apply SERS techniques to detect pathogens

### REFERENCES


for qualitative and quantitative analyses in complex media (food). If SERS can be integrated with chemometrics methods and spectral data analysis as well as with the development of micro-Raman spectrometers and nanosubstrates, and determine the best conditions to obtain reproducible results in complex food matrices or separate the bacteria from the complex food substrate via some filter membranes, there could be further advances not only in food, but also in the human health sectors.

### AUTHOR CONTRIBUTIONS

XZ designed this study and wrote the manuscript. CW and ML finished the experiments and collected the data. All authors read and approved the final manuscript.

### FUNDING

This work has been supported by the National Natural Science Foundation of China (31501582), Hubei Provincial Natural Science Foundation of China (2018CFB514), Science Research Fund of Wuhan Institute of Technology (17QD01 and 16QD04), and Open Project Program of Key Laboratory for Green Chemical Process of Ministry of Education in Wuhan Institute of Technology (2017007).


Escherichia coli 0157 under cryopreservation. Res. Microbiol. 168, 188–193. doi: 10.1016/j.resmic.2016.11.002


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

Copyright © 2018 Wei, Li and Zhao. 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-09-02857 December 12, 2018 Time: 15:7 # 9

# Quercetin Prevents Escherichia coli O157:H7 Adhesion to Epithelial Cells via Suppressing Focal Adhesions

Yansong Xue<sup>1</sup> , Min Du<sup>2</sup> and Mei-Jun Zhu<sup>1</sup> \*

<sup>1</sup> School of Food Science, Washington State University, Pullman, WA, United States, <sup>2</sup> Department of Animal Sciences, Washington State University, Pullman, WA, United States

The attachment of Escherichia coli O157:H7 to intestinal epithelial cells is indispensable for its pathogenesis. Besides translocated-intimin receptor (Tir), E. coli O157:H7 interacts with host cell surface receptors to promote intimate adhesion. This study showed that integrin β1 was increased in Caco-2 cells upon E. coli O157:H7 infection, while Caco-2 cells subjected to integrin β1 antibody blocking or CRISPR/Cas9 knockout had reduced bacterial attachment. Infection of E. coli O157:H7 inactivated focal adhesion kinase (FAK) and paxillin, increased focal adhesion (FA) and actin polymerization, and decreased cell migration in Caco-2 cells, which were rescued by integrin β1 antibody blocking or knockout. Pre-treatment with quercetin, known for its anti-oxidant and anti-inflammatory activity, reduced bacterial infection to Caco-2 cells, which might be partially via interfering integrin β1 and FAK association augmented by E. coli O157:H7. In addition, quercetin decreased FA formation induced by bacterial infection and recovered host cell motility. Taken together, data showed that E. coli O157:H7 interacts with integrin β1 to facilitate its adhesion to host cells. Quercetin inhibits bacterial infection possibly by blocking the interaction between E. coli O157:H7 and integrin β1. Collectively, these data indicate that quercetin provides an alternative antimicrobial to mitigate and control E. coli O157:H7 intestinal infection, and suggest potential broad benefits of quercetin and related polyphenols in fighting other enteric pathogen infections.

#### Keywords: E. coli O157:H7, quercetin, integrin β1, anti-adhesion, focal adhesion

### INTRODUCTION

Formation intestinal attaching and effacing (A/E) lesions is of necessary for the pathogenesis of Escherichia coli O157:H7 (Kaper, 2005). After attachment to intestinal epithelial cells, E. coli O157:H7 induces actin rearrangement to form pedestals (Knutton et al., 1989). Through this tight association with the host cell surface, E. coli O157:H7 utilizes various strategies to manipulate host signaling, leading to enhanced bacterial colonization and persistence, and host tissue damage (Xue et al., 2017). The host extracellular matrix (ECM) is composed of multiple macromolecules, which mediate multiple biological functions including cell to cell adhesion, migration, proliferation, and death (Meredith et al., 1993). Integrin β1, the most abundant cell surface integrin, is a transmembrane glycoprotein receptor that interacts with ECM components such as fibronectin, laminin, and collagen. Through interactions with ECM components, integrin β1 induces multiple

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Marina Sandra Palermo, Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Argentina Brandon Luedtke, University of Nebraska at Kearney, United States

> \*Correspondence: Mei-Jun Zhu meijun.zhu@wsu.edu

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 27 June 2018 Accepted: 17 December 2018 Published: 16 January 2019

#### Citation:

Xue Y, Du M and Zhu M-J (2019) Quercetin Prevents Escherichia coli O157:H7 Adhesion to Epithelial Cells via Suppressing Focal Adhesions. Front. Microbiol. 9:3278. doi: 10.3389/fmicb.2018.03278

bidirectional signal exchanges (Schwartz et al., 1995; Burridge and Chrzanowska-Wodnicka, 1996). In addition, integrin β1 recruits intracellular proteins such as talin, paxillin, and α-actinin, leading to the formation of the focal adhesion (FA) complex.

To tightly associate with host cells, pathogens utilize integrin β1 as an adhesion factor. Yersinia pseudotuberculosis interacts with integrin β1 via adhesin YadA to promote tight binding to the host cells (Eitel et al., 2005). Neisseria gonorrhoeae attaches to ECM substrate with the assistance of host integrin β1 (Muenzner et al., 2005). In response to infection, the rapid turnover and exfoliation of epithelial cells are innate defense mechanisms against pathogens (Mulvey et al., 2000). However, many pathogenic bacteria can circumvent host exfoliation and colonize the epithelium efficiently. Shigella flexneri reduces adhesion complex turnover and suppresses the detachment of infected cells from the basement membrane to manipulate host exfoliation (Kim et al., 2009). Integrins transduce extracellular signals into the host cells through association with intracellular adaptor proteins and protein kinases such as focal adhesion kinase (FAK) (Dia and Gonzalez de Mejia, 2011) and integrinlinked kinase (ILK) (Gagne et al., 2010). FAK deficiency increases the recruitment of FAs and reduces cell motility (Ilic et al., 1995), indicating FAK is involved in FA formation during cell migration. Thus, pathogens may manipulate FAK and associated kinases, which stabilize the FAs and ultimately enable them to colonize the host cells.

Quercetin is a polyphenol widely found in vegetables and fruits. Our previous study demonstrated that quercetin had anti-inflammatory and anti-oxidative properties that prevented E. coli O157:H7-induced inflammasome activation (Xue et al., 2017). However, the antimicrobial mechanism of quercetin has not been elucidated. We hypothesized that E. coli O157:H7 attaches to host cells via interacting with host integrin β1 and stabilizing FAs formation; quercetin inhibits integrin β1 expression and FA formation thus preventing E. coli O157:H7 infection.

### MATERIALS AND METHODS

### Cell Line, Media and Bacterial Strains

The human colonic epithelial cell line Caco-2 was obtained from the American Type Culture Collection (Manassas, VA, United States). Caco-2 cells were cultured in Dulbecco's Modified Eagle's medium (DMEM) (Sigma, St. Louis, MO, United States) supplemented with 10% fetal bovine serum (Sigma), 100 units/ml penicillin G, and 100 µg/ml of streptomycin (Sigma) at 37◦C with 5% CO2. The E. coli O157:H7 EDL933 wild type (EDL933) strain was obtained from the STEC center at Michigan State University. The E. coli O157:H7 EDL933 intimin (1eae) and tir (1tir) mutant strains were kindly provided by Dr. Carolyn H. Bohach's Lab at the University of Idaho. pEHEC tir plasmid was a generous gift from Dr. John M Leong at Tufts University (Campellone et al., 2002). EDL9331tir pEHEC tir strain was derived from E. coli O157:H7 EDL9331tir strain transformed with pEHEC tir plasmid. These strains were routinely grown in LB broth at 37◦C overnight with aeration.

### Infection of E. coli O157:H7 to Colonic Epithelial Cells

Caco-2 cells were seeded in a 24-well plate at 5 × 10<sup>5</sup> cells/ml for 12 h. Then the growth medium was replaced with fresh DMEM complete medium without antibiotics and supplemented with or without 200 µM quercetin (Sigma) for 12 h. Quercetin at this concentration did not impact the viability and growth of E. coli O157:H7 EDL933 (**Supplementary Figure S1**), nor did it decrease cell viability of Caco-2 cells (Xue et al., 2017). For integrin β1 blocking assay, cell monolayers were pretreated with integrin β1 antibody (rat IgG1, monoclonal, 1:200 dilution, DSHB) for 1 h prior infection, followed by 3 washes with PBS (pH 7.4). Then the cells were challenged with E. coli O157:H7 EDL933 at multiplicity of infection (MOI) of 10 for 4 h at 37◦C with 5% CO2.

### Quantitative Reverse Transcription PCR (qRT-PCR) Analysis

Total RNA was extracted from Caco-2 cells with TRI Reagent (Sigma) and reverse transcribed using an iScript kit (Bio-Rad, Hercules, California). cDNAs were used as templates for qRT-PCR analysis of selected genes using a CFX96 Real-Time PCR Detection System (Bio-Rad). SYBR green master mix (Bio-Rad) was used for all qRT-PCR reactions. β-actin was used as the housekeeping gene. Primers for qRT-PCR are listed in **Supplementary Table S1**. Amplification efficiency was 0.90 to 0.99 (Xue et al., 2017).

### Immunoblotting

Immunoblotting analysis was conducted according to the procedures described (Xue et al., 2017). Antibodies against vinculin (mouse monoclonal IgG1), talin (mouse monoclonal IgG3), and α-actinin (mouse monoclonal IgG1) were purchased from Santa Cruz (Dallas, TX, United States). Anti-p-FAK (rabbit polyclonal), FAK (rabbit polyclonal), p-paxillin (rabbit polyclonal), paxillin (rabbit polyclonal), and integrin β1 (rabbit monoclonal IgG) antibodies were from Cell Signaling Technology (Beverly, MA, United States). Antibody against β-actin (mouse monoclonal IgG1) was purchased from DSHB (Iowa City, IA, United States). Binding of antibodies was detected using HRP-coupled anti-rabbit or anti-mouse immunoglobulin (Cell Signaling) and visualized using Pierce ECL Western blotting substrate (ThermoFisher Scientific, Waltham, MA, United States). Density of bands was quantified by ImageQuant TL software (GE Healthcare Life Sciences, PA) and then normalized with reference to the β-actin content.

### Adhesion of E. coli O157:H7 to Colonic Epithelial Cells

Escherichia coli O157:H7 attachment to Caco-2 cells was conducted as previously reported (Xue et al., 2017). Briefly, Caco-2 cells were seeded at 5 × 10<sup>5</sup> cells/ml in a 24-well plate,

cultured until 80∼90% confluence and treated with 0 or 200 µM quercetin for 12 h. The cell monolayers were next challenged with E. coli O157:H7 EDL933 strain (MOI = 10) and co-cultured at 37◦C with 5% CO<sup>2</sup> for 4 h, followed by 3 washes with ice cold PBS and lysed with 0.2% Triton X-100. Lysates were serially diluted and appropriate dilutions were plated on LB agar plates.

The bacterial colonies were counted after 24 h incubation at 37◦C.

### Immunofluorescent Staining

fmicb-09-03278 January 16, 2019 Time: 13:5 # 4

Cell culture, quercetin treatment, and infection procedure were conducted as described above. Post-infection, the cell monolayers were washed 3 times with ice cold PBS and fixed in fresh prepared 4% paraformaldehyde for 30 min at room temperature. The fixed cells were then permeabilized with 0.5% Triton X-100 for 10 min, washed with PBS, and blocked with 5% normal goat serum for 60 min at room temperature (RT). Then the cells were incubated with anti-integrin β1 antibody (rat monoclonal IgG1, DSHB), vinculin antibody (Santa Cruz) or phalloidin (Sigma) overnight at 4◦C. The cells were rinsed with PBS and stained with Alexa Fluor 555 goat anti-rat IgG or Alexa Fluor 488 goat anti-mouse IgG (Cell Signaling) for 60 min at RT. These stained cells were washed 3 times with PBS and mounted with Fluoro-gel with DAPI (Electron Microscopy Sciences, Hatfield, PA). Fluorescence signal was visualized with EVOS FL fluorescence microscope (Life Technologies, Grand Island, NY).

### Co-immunoprecipitation

The post-infection cell monolayers were washed twice with ice-cold PBS and lysed in 200 µl IP buffer (50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% (v/v) Triton X-100, 0.1% (w/v) Na-deoxycholate, 1 mM EDTA, proteinase inhibitor cocktail) for 15 min on ice. The resulting cell lysates were transferred into pre-cooled 1.5 ml tubes, passed through a 29-gauge needle twice, and centrifuged for 10 min at 14,000 g, 4 ◦C. An aliquot of the supernatant was sampled for input protein content analysis. The remaining supernatants were pre-cleared with Protein G agarose beads (Thermo Scientific) with rotation for 30 min at 4◦C. The pre-cleared supernatants were incubated with anti-integrin β1 antibody (rat monoclonal IgG1, 1:100, DSHB) overnight with rotation at 4◦C. Then the Protein G magnetic agarose was added into the tubes and co-incubated overnight at 4◦C with rotation. The next day, tubes were placed on the magnetic stand to collect beads. The beads were washed with IP buffer 5 times, then resuspended in 100 µl of loading buffer and heated to 100◦C for 10 min to elute proteins. The supernatants collected were used for immunoblotting with anti-intimin-γ antibody (Gift from Dr. John M Leong) or anti-FAK antibody (Cell signaling), respectively.

### Integrin β1 CRISPR/Cas9 Knock Out (KO)

Caco-2 cells, at 70% confluence, were transfected with integrin β1 CRISPR/Cas9 KO plasmid (ITGB1 sgRNA/Cas9, GeneCopoeia, Rockville, MD, United States) or scramble control vector (Con

sgRNA, pCRISPR-SG01, GeneCopoeia) using X-tremeGENE HP DNA transfection reagent (Sigma) per manufacturer's instructions. Medium was changed 12 h post transfection, when 400 µg/ml G418 (Amresco, Solon, OH, United States) was added in the following 7 days to select cells with ITGB1 sgRNA.

## Cell Migration Activity

Cell culture, quercetin treatment, and infection procedure were conducted as described above. A scratch was introduced to the Caco-2 cell monolayer using a pipet tip. Then the cells were washed with PBS and infected with EDL933 strain or left uninfected for 4 h. Cells were washed with PBS and replaced with DMEM complete medium. Cells were migrated into the wound at 37◦C for 24 h. The migration was assessed by counting the number of Caco-2 cells that crossed the wound border as published previously (Kung et al., 2008).

## Statistical Analyses

Statistical analyses were conducted as previously described (Xue et al., 2017). Data were analyzed as a complete randomized design using GLM (General Linear Model of Statistical Analysis System, SAS, 2000). All data were analyzed by two-tailed Student's t-test. Means ± standard errors of mean (SEM) are reported. Statistical significance is considered as P ≤ 0.05.

## RESULTS

### Integrin β1 Was Involved in E. coli O157:H7 Attachment

Integrin β1 was expressed higher in infected cells than in control cells (**Figure 1A**). E. coli O157:H7 infection also increased surface level of integrin β1 (**Figure 1B**) as well as integrin α5 mRNA expression (**Supplementary Figure S2**). Neutralizing integrin β1 with anti-integrin β1 antibody reduced bacterial adhesion to Caco-2 cells (**Figure 1C**). To further explore the role of integrin β1 in bacterial adhesion, integrin β1 was knocked out with ITGB1 CRISPR/Cas9 sgRNA plasmid, which significantly attenuated EDL933 adherence to Caco-2 cells (**Figure 1D**). Integrin clustering is reported to be associated with FAK activation (Guan, 1997). Immunoprecipitation assay further showed FAK protein was associated with integrin β1 in Caco-2 cells infected with E. coli O157:H7, suggesting that infection induced FAK recruitment by integrin β1 (**Figure 1E**).

### Integrin β1 Was Implicated in Infection-Induced Dephosphorylation of FAK and Paxillin

FAK is a critical kinase that modulates FA activities (Zimerman et al., 2004). Phosphorylation of FAK and its downstream protein paxillin were markedly decreased in Caco-2 in response to

E. coli O157:H7 infection (**Figure 2**). Anti-integrin β1 antibody blocking prevented dephosphorylation of FAK and paxillin induced by E. coli O157:H7 (**Figures 2A–C**). Similarly, with integrin β1 KO, E. coli O157:H7 infection could no longer cause FAK and paxillin dephosphorylation as compared with uninfected control (**Figures 2D–F**). These results indicated that integrin β1 had a mediatory role in E. coli O157:H7-induced FAK inhibition.

### Integrin β1 Increased FA and Actin Polymerization in Response to E. coli O157:H7 Infection

FA is responsible for cell adhesion and migration (Hu et al., 2014). Enhanced FA assembly reduces cell mobility (Wozniak et al., 2004). E. coli O157:H7 infection increased FA proteins including talin, vinculin, and α-actinin in Caco-2 cells. However, integrin β1 antibody blocking or KO reduced the levels of these proteins in infected cells (**Figures 3A–D**). Immunofluorescence staining further showed that vinculin content was increased during E. coli O157:H7 infection, while integrin β1 KO impaired the accumulation of vinculin in response to infection (**Figure 3E**). These data collectively showed that integrin β1 was an important factor that mediated host FAs recruitment and assembly in response to E. coli O157:H7 infection.

The assembly of integrins and FAs serve as a platform for the organization of actin filaments. E. coli O157:H7 attachment to host cells is typically associated with actin rearrangement.

When integrin β1 was KO or blocked by antibody, the actin polymerization induced by infection was subsided (**Figure 4**), showing that integrin β1 was also implicated in infection-induced actin polymerization.

Enhanced FA assembly and decreased FAK activation could lower the ability of cell migration (Sieg et al., 1999; Kim et al., 2009). Consistent with enhanced FA assembly, E. coli O157:H7 infection significantly inhibited cell migration during wound healing. This inhibition phenomenon was attenuated by integrin β1 KO or antibody blocking (**Figure 5**).

### Intimin Is Involved in FAK Inhibition and FA Accumulation

Immunoprecipitation assay indicated that intimin was associated with integrin β1 (**Figure 6A**). To further understand the role of intimin in host FA formation, we infected Caco-2 cells with intimin mutant strain (1eae). Infection with 1eae strain did not suppress FAK and paxillin (**Figures 6B–D**), indicating a regulatory role of intimin in host signaling transduction. Consistently, FA proteins including talin, vinculin and α-actinin were not altered in cells infected with 1eae strain (**Figures 6E– H**). Immunofluorescent staining further showed that 1eae resulted in a lesser accumulation of vinculin as compared with EDL933 WT infected cells (**Figure 6I**). Interestingly, our data also showed that the tir deletion mutant (1tir) strain was incapable of causing dephosphorylation of FAK and paxillin (**Supplementary Figure S3**). The cytoplasmic C and N-terminus of Tir bind to FA proteins such as talin, vinculin, and α-actinin (Freeman et al., 2000; Huang et al., 2002), which might interfere with FAK activity. The interaction between Tir and host FA may

strengthen its association with host cell surface and facilitate colonization.

β1 and FAK induced by infection. FAK, focal adhesion kinase. → Activate or promote, Inhibit.

### Quercetin Inhibited E. coli O157:H7 Adherence Associated With Decreased Integrin β1 Expression and FA Formation

Quercetin reduced E. coli O157:H7-induced inflammasome activation (Xue et al., 2017). Here, we further showed that quercetin prevented integrin β1 expression (**Figure 7A**) and protein content (**Figures 7B,C**) in Caco-2 cells infected with EDL933, associated with decreased adhesion to Caco-2 cells (**Figure 7D**). Furthermore, quercetin attenuated the association of FAK with integrin β1 in EDL933 infected cells (**Figure 7E**). Additionally, quercetin reduced the protein contents of talin, vinculin, and α-actinin that were increased due to E. coli O157:H7 infection (**Figures 8A–D**), and rescued cell migration inhibited by E. coli O157:H7 (**Figure 8E**).

## DISCUSSION

### Integrin β1 Is a Potential Receptor for E. coli O157:H7 Adhesion

Integrins are a large family of heterodimeric receptors that are associated with a wide range of cell-to-cell interactions (Hynes, 1992). Integrin α5β1 is the most expressed and best characterized integrin heterodimer and functions as a receptor for many bacteria, such as Shigella flexneri and Pseudomonas aeruginosa (Watarai et al., 1996; Roger et al., 1999). The adhesin protein, CagL of Helicobacter pylori binds to and activates integrin α5β1 receptor and induces intracellular signaling (Kwok et al., 2007). Notably, many pathogenic bacteria enhance the surface level of integrins. H. pylori-infected gastric epithelial cells have a higher expression of both integrin α5 and β1 (Cho et al., 2006), and S. flexneri infection increases integrin β1 in HeLa cells (Kim et al., 2009). Consistently, our data also showed that both integrin α5 and β1 were upregulated in E. coli O157:H7-infected cells as compared to non-infected cells. Integrin β1 KO or blocking by integrin β1 antibody decreased bacterial attachment, indicating that integrin β1 was involved in E. coli O157:H7 adhesion.

## Inhibition of FAK May Strengthen Bacterial Colonization

Accumulating evidence shows that virulence factors of pathogens can utilize host kinases to manipulate host signaling. OspE, an effector of type III secretion system (T3SS) of Shigella (Miura et al., 2006), interacts ILK and subsequently reduces the phosphorylation of FAK and paxillin (Kim et al., 2009), resulting in stabilization of FAs and attenuated cell turnover (Miura et al., 2006). EspO1-1, a homolog of OspE in E. coli O157:H7 (Kim et al., 2009; Morita-Ishihara et al., 2013), similarly interacts with FAK to stabilize FA complex and inhibit the detachment of host cells from the ECM (Morita-Ishihara et al., 2013), indicating E. coli O157:H7 also has the ability to counteract the exfoliation of epithelial cells, which benefits its persistence. In our study, we found that intimin was co-immunoprecipitated with integrin β1, while intimin mutant strain was unable to induce FAK and paxillin dephosphorylation, suggesting that intimin mediates FAK and FA activity, and has ability to interact with integrin β1 to exploit host outside-in signaling.

Integrins transduce extracellular signals into the host cells through association with intracellular adaptor proteins and protein kinases such as FAK (Dia and Gonzalez de Mejia, 2011) and ILK (Gagne et al., 2010). These kinases serve as docking sites for recruitment of other kinases and FA components such as paxillin, talin, vinculin, and mediate cytoskeletal reorganization

(Schaller et al., 1995). FAK activation induces disassembly of FAs and correlates with enhanced cell turnover (Webb et al., 2004; Hamadi et al., 2010). In FAK−/<sup>−</sup> cells, the disassembly of FAs is significantly impaired with attenuated cell mobility (Webb et al., 2004). We found that infection enhanced interaction between FAK and integrin β1, which inhibits the phosphorylation of FAK and subsequently deactivates paxillin, thereby causing FA accumulation (**Figure 9**). As a result, cell migration was reduced in response to E. coli O157:H7 infection, which may inhibit host shedding and turnover. In support of our finding, FAK activation promotes migration of both endothelial cells and fibroblasts (Zhao and Guan, 2011), while FAK deficiency decreases cell migratory activity (Zhao and Guan, 2011) with an increased formation of FAs (Ilic et al., 1995).

### Quercetin Decreases Bacterial Infection by Regulating Integrin β1

Quercetin decreases ECM components such as collagen III productions and assembly in human corneal fibroblasts (McKay et al., 2015), and decreases cell surface level of integrin β1 in different cell types (He et al., 2015; Doersch and Newell-Rogers, 2017). In this study, although quercetin did not alter integrin β1 expression in uninfected cells, quercetin prevented the increase of both integrin β1 and integrin α5 expressions, as well as FA protein assembly induced by infection. Mechanisms for such preventive effects are twofold. Quercetin could directly interfere with integrin signaling elicited by bacteria to suppress FA accumulation and bacterial attachment, or the reduced bacterial attachment due to quercetin proportionally weakened intracellular signaling in comparison to untreated cells with more bacterial attachment. These data collectively suggested that quercetin prevented E. coli O157:H7 adhesion to epithelial cells through attenuation of integrin β1 accessibility to bacteria and/or suppression of intracellular signaling. The resultant effect may contribute to the reduced FA assembly.

In summary, E. coli O157:H7 attached to epithelial cells partially through the interaction with host integrin β1, which inhibited FAK phosphorylation and stabilized FA formation.

Quercetin inhibits bacterial infection likely via attenuated association between integrin β1 and FAK. Given that antibiotics

### REFERENCES


are not applicable for E. coli O157:H7 infection, these data provide a potential therapeutic application of quercetin for minimizing and eliminating E. coli O157:H7 infection. These data also suggest a broad application of polyphenolic compounds in the prevention of enteric pathogenic infection. However, additional in vivo studies to test the effects of quercetin on inhibiting E. coli O157:H7 infection will further strengthen our conclusions.

### AUTHOR CONTRIBUTIONS

YX, MD, and M-JZ designed the study, analyzed the data, and reviewed the manuscript. YX conducted the experiments. YX drafted the manuscript. MD and M-JZ revised the manuscript.

### FUNDING

This work was supported by USDA National Institute of Food and Agriculture (USDA-NIFA) (2018-67017-27517), Washington State University Emerging Research Issues Competitive Grant and BioAg Grant Program.

### ACKNOWLEDGMENTS

We acknowledge Dr. Carolyn H. Bohach' at University of Idaho for her generous gifts of E. coli O157:H7 tir and eae deletion mutant strains. We also thank Dr. John M. Leong at Tufts University and Dr. A. D. O'Brien at Uniformed Services University of the Health Sciences for their generous gifts of pEHEC tir plasmid and anti-imtimin-γ antibody, respectively.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2018.03278/full#supplementary-material

alpha5beta1 integrin and suppresses FAK/ERK/NF-kappaB signaling. Cancer Lett. 313, 167–180. doi: 10.1016/j.canlet.2011.09.002


intestinal cells under a fibronectin-dependent mechanism. J. Cell Physiol. 222, 387–400. doi: 10.1002/jcp.21963


**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 Xue, Du and Zhu. 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-09-03278 January 16, 2019 Time: 13:5 # 10

# From Pig Breeding Environment to Subsequently Produced Pork: Comparative Analysis of Antibiotic Resistance Genes and Bacterial Community Composition

#### Zongbao Liu1,2, Uli Klümper3,4, Lei Shi<sup>5</sup> , Lei Ye<sup>5</sup> \* and Meng Li<sup>1</sup> \*

1 Institute for Advanced Study, Shenzhen University, Shenzhen, China, <sup>2</sup> Key Laboratory of Optoelectronic Devices and Systems of Ministry of Education and Guangdong Province, College of Optoelectronic Engineering, Shenzhen University, Shenzhen, China, <sup>3</sup> ESI and CEC, Biosciences, University of Exeter, Cornwall, United Kingdom, <sup>4</sup> European Centre for Environment and Human Health, University of Exeter, Truro, United Kingdom, <sup>5</sup> Institute of Food Safety and Nutrition, Jinan University, Guangzhou, China

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

### Reviewed by:

Cristiana Garofalo, Polytechnical University of Marche, Italy Krassimira Hristova, Marquette University, United States

#### \*Correspondence:

Lei Ye 176003871@qq.com Meng Li limeng848@szu.edu.cn

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 19 September 2018 Accepted: 11 January 2019 Published: 29 January 2019

#### Citation:

Liu Z, Klümper U, Shi L, Ye L and Li M (2019) From Pig Breeding Environment to Subsequently Produced Pork: Comparative Analysis of Antibiotic Resistance Genes and Bacterial Community Composition. Front. Microbiol. 10:43. doi: 10.3389/fmicb.2019.00043 It is well verified that pig farms are an important reservoir and supplier of antibiotic resistance genes (ARGs). However, little is known about the transmission of ARGs between the breeding environment and subsequently produced pork. This study was conducted to investigate if ARGs and associated host bacteria spread from the breeding environment onto the meat through the food production chain. We thus analyzed the occurrence and abundance of ARGs, as well as comparing both ARG and bacterial community compositions in farm soil, pig feces and pork samples from a large-scale pig farm located in Xiamen, People's Republic of China. Among the 26 target ARGs, genes conferring resistance to sulfonamide, trimethoprim, aminoglycoside, chloramphenicol, macrolide, florfenicol, and tetracycline were observed at high frequency in both the pig breeding environment and pork. The prevalence of ARGs in pork was surprisingly consistent with breeding environments, especially between the pork and feces. The relative abundance of 10 representative ARGs conferring resistance to six classes of antibiotics ranged from 3.01 × 10−<sup>1</sup> to 1.55 × 10−<sup>6</sup> copies/16S rRNA copies. The ARGs conferring resistance to sulfanilamide (sulI and sulII), aminoglycoside (aadA), and tetracycline [tet(A) and tet(M)] were most highly abundant across most samples. Samples from feces and meat possessed a higher similarity in ARG compositions than samples from the farms soil. Enterobacteriaceae found on the meat samples were further identical with previously isolated multidrug-resistant bacteria from the same pig farm. Our results strongly indicate that ARGs can be potentially spreading from pig breeding environment to meat via the pork industry chain, such as feed supply, pig feeding and pork production.

Keywords: pig farm, antibiotic resistance genes, bacterial community composition, breeding environment, pork

## INTRODUCTION

fmicb-10-00043 January 25, 2019 Time: 17:49 # 2

The increasing prevalence and spread of antibiotic resistance genes (ARGs) from food animal sources has become a major public health concern (O'Neill, 2015). Livestock farm environments, such as farmed soils and animal waste, have been considered the most important reservoirs for environmental ARGs, as high abundances of various ARGs have frequently been detected in these environments (Cheng et al., 2013; Zhu et al., 2013; Li et al., 2015; He et al., 2016; Fang et al., 2018; Qian et al., 2018). It is generally accepted that the use of antibiotics in animal husbandry is one of the major drivers for the emergence of resistant bacteria and dissemination of resistance genes. The long-term and extensive use of antibiotics in food animals is not only a regional or national phenomenon, but part of a global problem. In 2010, global consumption of antimicrobials in food animal production was estimated at 63,151 (± 1,560) tons (Van Boeckel et al., 2015). In the United States, livestock producers used between 70% and 80% of all antibiotics sold across the country (Elliott et al., 2017). In Vietnam, more than 11 antibiotics were used for growth promotion, 25 for disease prevention, and 37 for therapeutic purposes in pig farming (Tao et al., 2014). As the largest producer and consumer of antibiotics in the world, China produced approximately 210,000 tons of antibiotics each year, 46.1% were used in the livestock industries (Liu et al., 2015). More than 85% of these administered antibiotics or their metabolites may be excreted through animal urine or feces and then discharged into the environment (Tao et al., 2014). Antibiotics will impose a widespread selective pressure on bacteria, leading to the enrichment of resistant strains, which are also capable of spreading between different environments (Andersson and Hughes, 2014). Furthermore, many ARGs are encoded on mobile genetic elements allowing their transmission upon entering a new environment independent of the original host to a multitude of bacteria from the indigenous community (Klümper et al., 2015). Consequently, bacteria with various ARGs are commonly found in food animal wastes and the ambient environment nearby livestock farms (Tao et al., 2014; Jia et al., 2017). A potential transmission route of these antibiotic-resistant bacteria and ARGs from food animal sources to humans is the meat industry chain.

Currently, main global monitoring efforts focusing on antibiotic consumption and antibiotic-resistant bacteria takes place in clinical and public health laboratories, while they are rarely focused on animal husbandry in most countries, especially in China. However, previous studies have revealed that an exchange of ARGs could occur between bacteria from farm animals/soils and clinical pathogens via horizontal gene transfer (Forsberg et al., 2012; Li et al., 2015). Thus, environments carrying drug-resistant bacteria are indeed potential reservoirs of clinical resistance genes. Therefore, investigating the prevalence, abundance and transmission of antibiotic-resistant bacteria and ARGs on livestock farms is essential for controlling antibiotic resistance. Many studies have examined the abundance of ARGs in pig farm environments using real-time polymerase chain reaction (real-time PCR) (Cheng et al., 2013; Zhu et al., 2013; Tao et al., 2014). However, few studies have determined the relative abundances of ARGs of bacteria residing in or on pork. As far as we know, no study has performed a comparative analysis of the abundances and similarities of ARGs in pig farm soils, pig feces and the subsequently produced meat products. Since the ARG composition is significantly correlated with microbial phylogenetic and taxonomic structure (Forsberg et al., 2014), we here combined the analysis of ARG composition and bacterial community composition to provide a better understanding of the dynamics of ARG transfer between environmental and meat samples.

The objectives of this study were (1) to determine the occurrence and abundance of ARGs in pig farm soil, fecal and meat samples collected from a large-scale pig farm based on PCR and real-time PCR methods; (2) to evaluate the similarity/difference of ARG compositions among different types of samples using non-metric multidimensional scaling (NMDS) analysis; and (3) to analyze the composition of the dominant bacterial community using PCR-denaturing gradient gel electrophoresis (DGGE) analysis.

### MATERIALS AND METHODS

### Sample Collection

A total of 68 farm soil, pig feces and fresh pork meat were collected from a large-scale pig farm over a period of more than one year (August 2012, April 2013, and November 2013) in Xiamen, China (longitude, 117◦ 590E; latitude, 24◦ 510N). On this farm sulfonamides/trimethoprim (trimethoprim is a potentiator that is often administered together with sulfonamides), tetracycline, gentamicin, streptomycin, chloramphenicol, florfenicol, and amoxicillin are used widely for the treatment of swine infections or as growth promoters. However, exact doses of each of the antibiotics were not available from the farm. Twenty-seven surface soil samples (0–8 cm) were collected nearby 27 independent houses of finishing pigs. For each soil sample, three replicates (each 100 g) were collected around one finishing pig house, homogenized and combined into one sample for DNA extraction. Nineteen pig fecal samples were collected from a total of six waste treatment pools approximately 30 m from the pig breeding area using sterile centrifuge tubes. Twenty-two meat samples (approximately 200 g) from different finishing pigs were collected in the slaughter room using aseptic methods and stored at 4◦C for a subsequent DNA extraction. All the samples were placed immediately on ice and transported to the laboratory for homogenization and DNA extraction.

### DNA Extraction

The bacterial genomic DNA of the meat samples was extracted according to the following procedures. First, the meat samples (∼200 g) were rinsed with 50 mL of sterile peptone water, and then ∼50 g of each sample was placed aseptically into a sterile lateral filter bag containing 100 mL of 0.1% sterile peptone water, and the following procedures were performed as described previously (Wang et al., 2006). Fifty milliliters of filtered meat homogenate was centrifuged at 500 × g for 10 min, and then 20 mL of the supernatant was transferred to another sterile centrifuge tube and centrifuged at 14,000 × g for 10 min; the precipitate was used for DNA extraction using the Mag-MK Bacterial Genomic DNA Extraction Kit (Sangon, China). The genomic DNA of the soil and fecal samples was extracted using the PowerSoil DNA Isolation Kit (Mo Bio, Germantown, MD, United States) according to the manufacturer's instructions. The quality and concentration of the DNA were determined by spectrophotometer analysis (NanoDrop ND-1000C, Thermo Fisher Scientific, United States), low-purity DNA (with A260/A<sup>280</sup> ratio < 1.6 or > 2.0, or A260/A<sup>230</sup> ratio < 1.8) was further purified using Dr. GenTLE Precipitation Carrier Kit (Takara, Shiga, Japan).

### PCR Detection of ARGs

fmicb-10-00043 January 25, 2019 Time: 17:49 # 3

Twenty-six ARGs were analyzed using a PCR assay; the primers used are listed in **Supplementary Table S1**. The PCRs were performed in a total volume of 25 µL including 1 µL of extracted DNA, 2.5 µL of Taq reaction buffer, 0.2 mM dNTPs, 0.2 µM primers, and 0.625 units of Hot Start Taq DNA polymerase (Takara, Shiga, Japan). The PCR conditions were as follows: 95◦C for 3 min, followed by 30 cycles of 94◦C for 0.5 min, 55–60◦C for 0.5 min, and 72◦C for 1 min, followed by one cycle of 72◦C for 10 min. The PCR products were analyzed with electrophoresis on 1.5% agarose gels in 1 × Tris–acetate–EDTA buffer (40 mM Tris, 20 mM acetic acid, and 1 mM EDTA, pH 8.0) at 100 V for 30 min.

### Real-Time PCR Detection of ARGs and 16S rRNA Genes

The real-time PCR analyses were performed on an ABI 7500 instrument (Applied Biosystems, Foster City, CA, United States) to quantify the copy number of the sulI, sulII, aadA, aphA-1, cmlA, ermB, floR, tet(A), tet(B), tet(M) genes, as well as the 16S rRNA V3 region. Standard curves for the real-time PCR assays were generated as described previously (Colomer-Lluch et al., 2011). Recombinant plasmids containing the target genes were used as positive controls. To construct the recombinant plasmids, the target ARGs and 16S rRNA V3 region gene were amplified with PCR and cloned into the pBackZero-T vector (Takara), and verified by sequencing at the Sangon Biological Engineering Technology & Service Company (Shanghai, China). The realtime PCRs were performed in a total volume of 25 µL using the SYBR Premix Ex Taq (Tli RNaseH Plus) Kit, including 1 µL of extracted DNA and 0.2 µM of each primer. The real-time PCR conditions were as follows: 95◦C for 2 min, followed by 40 cycles of 95◦C for 15 s, 57–60◦C for 30 s, and 72◦C for 45 s, followed by a melting curve stage.

### PCR-DGGE Analysis of Dominant Bacterial Community

The V3 variable region of 16S rRNA genes was used to analyze the composition of the dominant bacterial community. First, the 16S rRNA genes were amplified from the genomic DNA by PCR using primers 27F (5<sup>0</sup> -AGAGTTTGATCCTGGCTCAG-3 0 ) and 1492R (5<sup>0</sup> -GGTTACCTTGTTACGACTT-3<sup>0</sup> ) as described previously (Liu et al., 2015). Then, the PCR product was purified using TaKaRa MiniBEST DNA Fragment Purification Kit (Takara) according to the manufacturer's recommendations and diluted to 50 ng/µL with sterile double-distilled water. The concentration and purity of the DNA was checked with a NanoDrop 2000c spectrophotometer (Thermo Fisher Scientific, Waltham, MA, United States). Subsequently, the V3 variable region for the DGGE analysis was amplified from the purified 16S rRNA genes with PCR using the primers 338F-GC (5<sup>0</sup> -CCTACGGGAGGCAGCAG-3<sup>0</sup> ) and 518R (5<sup>0</sup> -ATTA CCGCGGCTGCTGG-3<sup>0</sup> ) (Zhang et al., 2016). To increase the stability of DGGE, a GC clamp (CGCCCGCCGCGCGCGGC GGGCGGGGCGGGGGCACGGGGGG) was added to the 5<sup>0</sup> end of the primer 338F (Muyzer et al., 1993). The PCR was performed in a total reaction volume of 50 µL containing 1 µL of 50 ng/µL purified 16S rRNA genes, 5 µL of Taq reaction buffer, 0.2 mM dNTPs, 0.2 µM primers, and 1.25 units of Hot Start Taq DNA polymerase. A touchdown PCR was used to amplify the 16S rRNA gene V3-GC region to increase the specificity of the amplification. The program was performed as follows: an initial denaturation at 94◦C for 3 min, followed by 10 cycles of 94◦C for 30 s, 65◦C for 30 s with a 1◦C decrease per cycle, and 72◦C for 1 min, followed by 25 cycles of 94◦C for 30 s, 55◦C for 30 s, and 72◦C for 1 min, followed by one cycle of 72◦C for 10 min. The amplified products were confirmed by gel electrophoresis.

The DGGE analysis of the 16S rRNA V3-GC regions was performed on a DCode System apparatus (Bio-Rad, Hercules, CA, United States) as described by Muyzer and Smalla (Muyzer et al., 1993; Muyzer and Smalla, 1998). PCR samples were separated on 8% acrylamide gels with an optimal denaturing gradient. To optimize the denaturing gradient, DGGE for each type of sample was performed using denaturant gradients of 35–65, 40–60, 45–60, and 40–55%. Based on these preliminary results, the linear gradient of 40–60% denaturant was chosen to analyze the meat samples. For the soil and fecal samples, denaturant gradients of 45–60 and 40–55% were used, respectively. Electrophoresis was performed in 1× Tris–acetate– EDTA buffer at a constant voltage of 60 V and 60◦C for 16 h. After electrophoresis, the gels were incubated in ethidium bromide solution for 30 min and rinsed with double-distilled water for 10 min. Images of the gels were obtained using the GelDoc XR System (Bio-Rad) according to the manufacturer's instructions. For each DGGE lane, band number and position were assessed for pattern similarity using Quantity One image analysis software (Bio-Rad, Hercules, CA, United States).

### Identification of DGGE Bands

The most detected and obvious DGGE bands were marked from each acrylamide gel. The bands were excised carefully from the acrylamide gels using a sterile scalpel. Every excised band was briefly washed tree times with 1 mL of double-distilled water in a 1.5-mL sterile centrifuge tube, and then crushed by a pipette tip. DNA fragments in crushed bands were eluted with 50 µL of double-distilled water by incubating overnight at 4 ◦C. The dissolved solution was centrifuged at 12,000 × g for 10 min, and the liquid supernatant was used as the template for reamplification of the PCR products with primer 338F without a GC clamp and primer 518R. The PCR conditions were as follows: 95◦C for 3 min, followed by 30 cycles of 94◦C for 30 s, 55◦C Liu et al. Antibiotic Resistance Genes in Pork

for 30 s, and 72◦C for 1 min, followed by one cycle of 72◦C for 10 min. The PCR products were cloned into the pBackZero-T vector and sequenced at the Sangon Biological Engineering Technology & Service Company. All DGGE band sequences were shown in the **Supplementary File**.

### Statistical Analysis

fmicb-10-00043 January 25, 2019 Time: 17:49 # 4

Non-metric multidimensional scaling was used to visualize the similarity of the ARG compositions in the 40 soil, fecal and meat samples. NMDS was performed using the abundance correlation matrix of the ARGs. Furthermore, differential abundance of ARGs between environmental and meat samples was tested by one-way analysis of variance (ANOVA). All statistical analyses were performed with Paleontological STatistics (PAST) software (version 3.16). Sequence identity was analyzed by comparison with GenBank sequences using the Basic Local Alignment Search Tool program<sup>1</sup> . Sequences with 97% or higher identity were considered to represent the same species. MEGA 6.06 (Center for Evolutionary Functional Genomics, Tempe, AZ, United States) was used to construct the neighbor-joining phylogenetic tree. A phylogenetic analysis based on the V3 region of 16S rRNA gene sequences used the maximum composite likelihood method. A bootstrap analysis was performed using 1000 replicates.

### RESULTS

### Distribution of ARGs

The prevalence of 26 resistance genes in 68 meat and environmental samples was determined by a PCR assay. Genes responsible for resistance to sulfonamide (sulI and sulII), trimethoprim (dfrA17), aminoglycoside (aadA and aphA-1), chloramphenicol (cmlA), a macrolide (ermB), florfenicol (floR), and tetracycline [tet(A), tet(B), and tet(M)] were distributed widely, as they were detected in 100, 100, 54.4, 100, 100, 100, 92.6, 100, 94.1, 80.9, and 92.6% of the samples, respectively (**Table 1**). Among these, sulI, sulII, aadA, aphA-1, cmlA, and floR were observed in every single sample. In contrast two genes, dfrA12 (trimethoprim) and aadB (aminoglycoside) had low detection rates. The dfrA12 gene was found only in meat (31.6%) and fecal samples (22.7%). In contrast, the aadB gene conferring resistance to an aminoglycoside was found exclusively in 22.2% of soil samples. Four tetracycline resistance genes were chosen as target ARGs in this study, of which, tet(A) and tet(B) were observed in all meat samples. Contrary, in fecal and soil samples, the resistance gene tet(M), instead of tet(A) or tet(B), was detected in every sample (**Table 1**).

In addition, the detection rate of individual ARGs differed significantly between different batches of samples. For example, among the fecal samples, the dfrA12 gene conferring resistance to trimethoprim was detected in 100% of the samples in the second batch (collected in April 2013). However, this gene was not observed in the other batches of samples. Similarly, in the meat samples, this gene was only detected in samples of the third batch (collected in November 2013), 71.4% of which tested positive.


The tetracycline resistance gene tet(M) was detected frequently, as it was found in 100 and 85.7% of the first and third batches of the meat samples, respectively, while it was only detected in 50% of the samples of the second batch (**Supplementary Table S2**). However, 13 further resistance genes [dhfrV, dhfrI, aac(3)-I, aac(3)-IV, catI,blaSHV, blaOXA, blaTEM, citM, moxM, dhaM, ereA, and tet(D)] were not detected in any meat or environmental samples.

### Quantification of ARGs and 16S rRNA Gene

Based on the previous prevalence testing, 10 representative ARGs were chosen in combination with the 16S rRNA gene and their copy number was determined in 40 representative soil, fecal and meat samples with qualitative real-time PCR assays. Relative ARG abundance (defined as the absolute number of ARG copies normalized to the absolute number of 16S rRNA) was used to compare the differences of 10 ARGs among the different samples. Ten ARGs conferring resistance to six classes of antibiotics were detected with abundances ranging from 1.55 × 10−<sup>6</sup> to 3.01 × 10−<sup>1</sup> copies of ARG per copy of the 16S rRNA gene (**Figure 1**). The ARG aadA, which is associated with resistance to

<sup>1</sup>http://www.ncbi.nlm.nih.gov/BLAST/

aminoglycosides, had the highest abundance ratio of 3.01 × 10−<sup>1</sup> in the soil samples. Similarly, the tetracycline resistance gene tet(B), which had the lowest abundance ratio of 1.55 × 10−<sup>6</sup> , was also found in the soil samples (**Supplementary Table S3**). For most of the samples, the abundance ratio range was between 10−<sup>4</sup> and 10−<sup>1</sup> .

In general, the resistance genes sulI, sulII, aadA, tet(A), and tet(M) had higher abundances than the other ARGs in the environmental samples (**Figure 2**). Moreover, the abundances of the resistance genes sulII and aadA were relatively high in the soil samples, with the average ratios of sulII/16S rRNA and aadA/16S rRNA reaching 1.08 × 10−<sup>1</sup> and 7.0 × 10−<sup>2</sup> , respectively. In the fecal samples, abundances of aadA and tet(M) were much higher, with average ratios of 5.54 × 10−<sup>2</sup> and 8.1 × 10−<sup>2</sup> , respectively. The abundances of all 10 ARGs detected in the fecal samples were in the order of: tet(M) > aadA > sulII > tet(A) > sulI > ermB > aphA-1 > tet(B) > cmlA > floR. Compared with the environmental samples, the abundances of the detected ARGs in the meat samples were lower, except for those of tet(A) and tet(B). Notably, the abundance of the tetracycline resistance gene tet(B) in the meat samples was much higher than in the environmental samples.

In summary, among the 10 representative ARGs, sulII, aadA, and tet(M), which confer resistance to sulfanilamide, aminoglycosides, and tetracycline were the most abundant genes in both the environmental and meat samples, respectively. In contrast, the ARGs aphA-1, ermB, and floR, which are associated resistance to aminoglycosides, macrolides, and florfenicol had much lower abundances in most of the environmental and meat samples, respectively (**Figure 2**).

### Similarity Analysis of ARG Compositions

The similarity of the ARG compositions in the 40 environmental and meat samples was evaluated using NMDS. Samples of the same type generally clustered more closely, which revealed that the grouping pattern was primarily influenced by sample type (**Figure 3**). For instance, the meat, feces, and most of the soil samples formed distinct clusters, especially the fecal samples which displayed high similarity in abundance (p-value = 0.96, evaluated by means of ANOVA statistical analysis). In addition, two fecal samples (F-10 and F-11) with the codes of 26 and 27, respectively, in the NMDS plot formed a cluster that was independent of the other fecal samples. Not surprisingly, these samples were collected in the same batch, which differed from those of the other fecal samples. Notably, among the three types of samples, the meat samples clustered more closely with the fecal samples, and this result was statistically supported by ANOVA analyses (p-value = 0.18).

### DGGE Analysis and Identification of DGGE Bands

The dominant bacterial community composition of the 40 soil, feces and meat samples (mentioned above) was analyzed with PCR-DGGE. The DGGE band patterns indicated complex dominant bacterial community composition across all sample types (**Figure 4**). Moreover, the band patterns of the soil samples displayed a higher degree of heterogeneity (**Supplementary Figure S1a**). In contrast, for the fecal samples, the composition of the bands in the DGGE profiles did not differ significantly among each other, especially among samples collected in the same batch (**Supplementary Figure S1b**). Compared with the soil samples, the band pattern of the meat samples indicated a lower bacterial diversity, but notably, the position and brightness of the bands among the meat samples was consistent (**Supplementary Figure S1c**), which indicated that the dominant bacteria in the meat samples were relatively stable.

To investigate the bacterial composition in the environmental and meat samples based on their PCR-DGGE bands, a total of 64 of the most frequent and obvious DGGE bands were marked and excised from the gels, and then purified and sequenced (**Figure 4**). As shown in **Supplementary Table S4**, 22, 18, and 24 bands were identified from the DGGE patterns for the soil, fecal, and meat samples, respectively. In the soil samples, among the 22 bands, 9 bands were identified as uncultured bacteria, which accounted for over 40% of the total bands. Moreover, the bacteria Bacillus sp. and Clostridium sp. were detected most frequently, both occupying three bands each. Similarly, in the fecal samples, 12 of a total 18 bands were identified as uncultured bacteria. Among the remaining six bands, two bands were identified as Clostridium spp., and two other bands were identified as Arcobacter spp. The last two bands were identified as a Desulfovibrio sp. and a Tissierella sp., respectively. In contrast to the soil and fecal samples, only one of the total 24 bands excised from the meat sample gel were identified as uncultured bacteria. The results showed that more than 14 species of known bacteria were identified from the meat sample gel, and four, three, three, two, and two bands were identified as Serratia spp., Aeromonas spp., Pantoea spp., Enterobacter spp., and Bacillus spp., respectively. Other species of bacteria, including common pathogens such as Klebsiella pneumoniae, were also identified from the meat samples (**Supplementary Table S4**).

### Phylogenetic Analysis

A neighbor-joining phylogenetic tree of the sequences of total 64 bands was constructed based on the maximum composite likelihood method. 17 of the 24 sequences of bands from the meat sample gel formed a very distinct and independent group, and showed high homology (**Figure 5**). Compared with the meat samples, the sequences of bands from the environmental sample gels showed relatively distant phylogenetic relationships. Furthermore, the bands MB12, MB13, MB14, and MB17 from the meat sample gel were identified as Serratia spp.; however, band MB13 was not found on the same branch of the phylogenetic tree as the other three bands. Band MB8, which was identified as Comamonas sp., shared one branch with band SB5 which was from the soil sample gel and shared 100% homology with Comamonas sp. ST18 (FJ982927.1). Moreover, bands FB4 and SB1 from the feces and soil sample gels shared 100% homology, respectively, and they were identified as Arcobacter cryaerophilus. Similarly, bands FB16 and SB15, which were identified as an uncultured Clostridium sp., also shared 100% homology. Five bands from the meat and soil sample gels were identified as Bacillus spp., though they shared relatively low homology. Generally, the sequences of the bands from the soil and fecal sample gels showed closer phylogenetic relationships. Notably, the band MB20 which was identified as Klebsiella pneumoniae, shared 100% homology with MDR strain M47 and M88 isolated from meat samples in the same pig farm.

### DISCUSSION

Antibiotics have been commonly used in veterinary medicine worldwide for therapeutic use and to increase production in animal husbandry. Numerous previous studies have focused on analyzing the abundance of ARGs in pig farm environments (Heuer et al., 2008; Cheng et al., 2013; Zhu et al., 2013; Ma et al., 2015). Nevertheless, rarely studies were also detecting the relative abundance of ARGs in the produced pork meat and connecting it to the surrounding environment (feces and farm soil).

In this study, we analyzed samples from a large-scale pig farm, on which sulfamethoxazole, trimethoprim, tetracycline, gentamicin, streptomycin, chloramphenicol, florfenicol, and amoxicillin were widely used for the treatment of swine infections or as growth promoters. Sulfonamides/trimethoprim, tetracyclines, macrolides, penicillins and aminoglycosides are the most widely used groups of antibiotics in animal husbandry (Committee, 1999; Haller et al., 2002; Economou and Gousia, 2015), and consequently ARGs associated to these antibiotics are generally detected most frequently in various livestock farms (Ho et al., 2010; Cheng et al., 2013; Zhu et al., 2013; Tao et al., 2014). While for this study no exact amounts of the corresponding antibiotic doses administered were available, the high abundance of ARGs conferring resistance to these

antibiotics is a good indicator that these antibiotics were consistently given on the farm.

In a previous study on this exact pig farm, we isolated 102 multidrug-resistant (MDR) enterobacterial strains, and identified MDR strains sharing 100% phylogenetic identity across the 3 different environments (meat, soil, and feces) (Liu et al., 2015). To further our understanding of the abundance and transfer of potentially antibiotic resistance bacteria on the pig farm we here moved from single isolates to a community wide detection of antibiotic resistance, as livestock farm environments are known to harbor a huge diversity of bacteria (McGarvey et al., 2004; Dowd et al., 2008). This approach involves detecting both, transmission of bacteria, as well as transfer of resistant genes from soil and fecal samples across the food production chain onto pork meat.

Since livestock farm environments harbored highly diverse bacteria (McGarvey et al., 2004; Dowd et al., 2008), the high throughput sequencing techniques could give much deeper insights into microbial community diversity compared with DGGE (Guo and Zhang, 2012). However, for the fresh meat samples, the PCR-DGGE technique remains a useful and economic tool to rapidly analyze the composition of dominant bacteria. In the past decade, the microbial diversity and main flora in fresh meat has been widely investigated using PCR-DGGE (Jiang et al., 2010, 2011; Osés et al., 2013; de Smidt, 2016; Koo et al., 2016). In this study, comparing with the soil and feces samples, the composition of bands from the meat samples showed high consistency across replicates, which indicated that dominant bacteria across meat samples were relatively stable. Unsurprisingly, the number of visible bands was lower than for both other sample types.

To investigate transfer of bacteria across environments we sequenced a total of 64 of the most frequent DGGE bands. Three bands from the meat sample were identified as Enterobacter sp. and Klebsiella sp., respectively, and bacteria of these two genera have previously been detected as the predominant MDR bacteria on the same pig farm (Liu et al., 2015). The detected bacteria of these groups showed close evolutionary relationship with the

gradient of the gels used for soil, feces, and meat samples were 45∼60%, 40∼55%, and 40∼60%, respectively.

bacteria identified in this study (**Figure 5**), indicating that transfer of these MDR bacteria from the pig farm onto the meat might be occurring. Additionally, in the Bacilli group, species from soil as well as from meat samples are found in close proximity. Further, Serratia, Aeromonas, and Pantoea were identified as apparent on meat. All these bacteria are widely distributed in environmental and pork samples (Jiang and Shi, 2013; Greig et al., 2015; Roberts and Schwarz, 2016; Møretrø and Langsrud, 2017), and various ARGs have been detected in antibiotics resistant strains belonging to these bacterial genera (Batah et al., 2015; Liu et al., 2015; Le et al., 2016; Carnelli et al., 2017). Contrary, over 40% of the bands from the soil and fecal sample gels were identified as uncultured bacteria (**Supplementary Table S4**), and accounted for the vast majority of the total bacteria as expected from various environmental samples (Rappé and Giovannoni, 2003). Based on analysis of the created phylogenetic tree we can conclude that the composition of the predominant bacterial community in pork differed significantly from that in soil or fecal samples, however, we found several species that were closely related and potentially spread across the environments, including the previously isolated and highly medically relevant multidrug resistant strain Klebsiella pneumonia, regularly involved in spreading ARGs from the environment to pathogens (Wyres and Holt, 2018).

There were more overlaps in bacterial community composition between meat and soil samples compared to meat and fecal samples. However, for the prevalence and composition of ARGs, a higher degree of similarity was detected among meat and fecal, rather than meat and soil samples. The prevalence of 26 ARGs in pork was surprisingly consistent with breeding environments, especially between the pork and feces. ARG composition of all 40 samples as detected using qPCR was subject to NMDS analysis using the Bray-Curtis distance. NMDS has been widely used in various environments to compare the bacterial communities of numerous samples (Guan et al., 2014; San Miguel et al., 2014; Xiong et al., 2015; Huo et al., 2017). But, it is also a useful tool to analyze the similarity of ARG compositions between different samples (Segawa et al., 2013; Li et al., 2015). Consistent with these previous studies, clustering in our study was mainly influenced by the sample origin. Further, among the three types of samples, the meat samples clustered more closely with the fecal samples (p-value = 0.18), combine the results mentioned above, strongly indicating that ARGs on meat samples can indeed originate from the fecal samples. This hypothesis can further be supported by the report that most bacterial genera detected on chilled pork are associated with fecal contamination during slaughtering (Zhao et al., 2015). And despite not detecting any immediate overlaps of sequenced DGGE bands between fecal and meat samples, identical MDR isolates found in both environments and mating experiments suggest that these bacteria furthermore harbor their resistance determinants on conjugative and thus selftransmissible plasmids that could spread to other bacteria on the meat (Liu et al., 2015).

Compared with fecal and meat samples, the soil samples did not only cluster further apart, but, consistent with the previously detected higher variance in bacterial composition also had a higher internal distance between replicates when analyzing the ARG content. The high abundance of ARGs in pig farm soils is generally assumed to primarily originate from the selection pressure of antibiotics originating from pig urine or feces (Tao et al., 2014). In this study, the soil sampling sites were widely distributed across the large-scale pig farm, therefore, the urine

previous study. M, pig meat; F, pig feces; and W, farm wastewater. The distinct clusters majorly formed by sequences from meat samples were highlighted with green and purple in background. Values on the branches represent the percentage of 1000 bootstrap replicates and bootstrap values over 50% are shown in the tree.

or feces pollution levels in soil samples did potentially differ substantially.

To complete this analysis the prevalence of 26 resistance genes was tested by amplification with commonly used ARG primers. Based on these results, the relative abundance of 10 representative ARGs, which were observed most frequently, was further detected with real-time PCR, allowing for a far more accurate and sensitive detection of ARGs than metagenomic sequencing analysis. To normalize the ARGs among the various samples, the relative abundance of the ARGs was expressed as copy of ARG per copy of 16S rRNA gene. The same calculation method has previously been used to estimate the overall bacterial abundance and to normalize ARGs to the bacterial population in samples from different sources (Gao et al., 2012; Cheng et al., 2013; Li et al., 2015; Subirats et al., 2017). The 10 tested ARGs, conferring resistance to six different classes of antibiotics, were detected with an abundance range between 3.01 × 10−<sup>1</sup> and 1.55 × 10−<sup>6</sup> per 16S rRNA copy in our samples.

Across all samples resistance to sulfanilamide (sulI and sulII), aminoglycoside (aadA) and tetracycline [tet(A) and tet(M)] were the most abundant ARGs. Based on information received from farm workers, these classes of antibiotics were consistently used in this large-scale pig farm. Consistent with our study, Cheng et al. (2013) reported detection of sulI, sulII and tetM with high relative abundance in livestock farms located in eastern China. Especially for sulfanilamide resistant genes, sulI and sulII, their relative abundance in this study was much higher than in other regions, such as United States (Munir and Xagoraraki, 2011) and Germany (Heuer et al., 2008), indicating a far increased and potential over-use of sulfanilamide antibiotics on our testing farm. Among our most frequently detected ARGs, sulI and aadA, are heavily associated with integron 1 gene cassettes (Binh et al., 2009; Byrne-Bailey et al., 2011; Liu et al., 2015), allowing their horizontal spread across communities and environments, and increasing their persistence as for example shown in manured soil (Zhang et al., 2015).

Tetracyclines are in general the most used antibiotics in pig farms, which are usually incorporated into animal feed to improve growth rate and feed efficiency (Sarmah et al., 2006). In this study, three tetracycline resistance genes [tet(A), tet(B), and tet(M)] were part of real-time PCR analysis. All three of these tetracycline resistance genes have been observed frequently in various livestock farms (Cheng et al., 2013; Kyselková et al., 2015; Li et al., 2015; Ma et al., 2015). While tet(A) and tet(M) were detected with high abundance across all samples, tet(B) was detected with far increased frequencies on the meat samples.

The other 4 tested ARGs, aphA-1, cmlA, ermB, and floR conferring resistance to aminoglycosides, chloramphenicol, macrolides and florfenicol were found across all samples, but at relatively low frequencies. Among all tested ARGs, floR had the lowest average relative abundance in both environmental and meat samples, with the average ratios ranging from 2.3 × 10−<sup>3</sup> to 4 × 10−<sup>4</sup> , consistent with a previous report (Li et al., 2015) where the relative abundance of floR in environmental samples from pig farms ranged from 2.02 × 10−<sup>5</sup> to 1.33 × 10−<sup>3</sup> copies/16S rRNA gene copies. ARG abundances were usually associated with the application of these antibiotics in livestock farms (Knapp et al., 2009; Gillings and Stokes, 2012; Zhu et al., 2013), therefore, the relatively low abundance of the above mentioned ARGs might be due to less use of the corresponding antibiotics on this pig farm.

### CONCLUSION

In conclusion, this study analyzed distribution and abundance of ARGs and dominate bacterial composition in environmental and pork samples from a large-scale pig farm where antibiotics were widely used. Our results demonstrated that there is a strong indication that ARGs and the associated MDR organisms potentially spread from the pig breeding environment to meat via the pork industry chain. These findings strongly indicate that the breeding environment is an important reservoir and breeding ground for antibiotic resistant bacteria and ARGs, which could be potentially transmitted to humans via the meat industry chain. Therefore, at the present time, the strategies for reasonable use of antibiotics, such as establishing regional management regimes for agricultural use of antibiotics, limiting the use of antibiotics as growth promoters and developing antibiotic substitutes, and establishment of scientific monitoring systems in animal husbandry are essential to limit the adverse effects of the abuse of antibiotics and to ensure the safety of animal-derived food and environment.

### ETHICS STATEMENT

The animals were processed according to the "Regulations for the administration of affairs concerning experimental animals" established by Guangdong Provincial Department of Science and Technology on the Use and Care of Animals. The experiments were approved by the Institutional Animal Care and Use Committee of Shenzhen University.

### AUTHOR CONTRIBUTIONS

LS and ML designed the experiments and provided experimental materials. ZL and LY carried out experiments. ZL and UK analyzed sequencing data and wrote the manuscript.

### FUNDING

This work was supported by the National Science and Technology Pillar Program during the Twelfth Five-Year Plan Period (2014BAD13B00) and the Science and Technology Innovation Committee of Shenzhen (Grant No. JCYJ201772796). United Kingdom received funding from the European Union's Horizon 2020 Research and Innovation Program under Marie Skłodowska-Curie grant agreement no. 751699 and is supported through an MRC/BBSRC grant (MR/N007174/1).

### ACKNOWLEDGMENTS

fmicb-10-00043 January 25, 2019 Time: 17:49 # 11

The authors are grateful to Zhigang Zhang, Miaorui Chen, and pig farm workers for their help during sample collection.

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.00043/full#supplementary-material


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

Copyright © 2019 Liu, Klümper, Shi, Ye and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fmicb-10-00043 January 25, 2019 Time: 17:49 # 12

# Comparative Genomic Analysis Reveals the Potential Risk of *Vibrio parahaemolyticus* Isolated From Ready-To-Eat Foods in China

Rui Pang1†, Tengfei Xie1†, Qingping Wu<sup>1</sup> \*, Yanping Li <sup>1</sup> , Tao Lei <sup>1</sup> , Jumei Zhang<sup>1</sup> , Yu Ding<sup>2</sup> , Juan Wang<sup>3</sup> , Liang Xue<sup>1</sup> , Moutong Chen<sup>1</sup> , Xianhu Wei <sup>1</sup> , Youxiong Zhang<sup>1</sup> , Shuhong Zhang<sup>1</sup> and Xiaojuan Yang<sup>1</sup>

*<sup>1</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 Institute of Microbiology, Guangzhou, China, <sup>2</sup> Department of Food Science and Technology, Jinan University, Guangzhou, China, <sup>3</sup> College of Food Science, South China Agricultural University, Guangzhou, China*

#### *Edited by:*

*Learn-Han Lee, Monash University Malaysia, Malaysia*

#### *Reviewed by:*

*Christopher Staley, University of Minnesota Twin Cities, United States Adrian Canizalez-Roman, Autonomous University of Sinaloa, Mexico*

> *\*Correspondence: Qingping Wu wuqp203@163.com*

*†These authors have contributed equally to this work*

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 11 September 2018 Accepted: 23 January 2019 Published: 07 February 2019*

#### *Citation:*

*Pang R, Xie T, Wu Q, Li Y, Lei T, Zhang J, Ding Y, Wang J, Xue L, Chen M, Wei X, Zhang Y, Zhang S and Yang X (2019) Comparative Genomic Analysis Reveals the Potential Risk of Vibrio parahaemolyticus Isolated From Ready-To-Eat Foods in China. Front. Microbiol. 10:186. doi: 10.3389/fmicb.2019.00186* *Vibrio parahaemolyticus* is a major foodborne pathogen associated with the consumption of aquatic products. The presence of this bacterium in ready-to-eat (RTE) foods has recently been reported. However, the genomic features and potential risks of *V. parahaemolyticus* isolated from RTE foods are poorly understood. To help understand the genome-wide characteristics of RTE food isolates, the complete genomes of 27 RTE food isolates were sequenced and compared to those of 20 clinical and 19 other environmental (e.g., water and aquatic product source) isolates using a comparative genomics approach. Analysis revealed that *V. parahaemolyticus* RTE food isolates had higher numbers of genes on average and possessed more accessory genes than isolates from other sources. Most RTE food isolates were positive for some known virulence-associated genes and pathogenicity islands (PAIs), and some of these isolates were genetically homologous to clinical isolates. Genome-wide association analysis revealed 79 accessory genes and 78 missense single-nucleotide polymorphisms that affected 11 protein-coding genes were significantly associated with RTE food sources. These genes were mostly involved in defense mechanisms and energy production and conversion according to functional annotation in the COG database. KEGG Pathway analysis showed that these genes mainly affected the biofilm formation of *V. parahaemolyticus*, and subsequent experiments confirmed that nearly all RTE food isolates possessed the ability to form biofilm. The biofilm formation can facilitate the persistence of *V. parahaemolyticus* in RTE foods, and the presence of virulence-associated genes poses a pathogenic potential to humans. Our findings highlight the potential risk of *V. parahaemolyticus* in Chinese RTE foods and illustrate the genomic basis for the persistence of these isolates. This study will aid in re-evaluating the food safety threats conferred by this bacterium.

Keywords: *Vibrio parahaemolyticus*, ready-to-eat foods, genomics, potential risk, biofilm

### INTRODUCTION

Vibrio parahaemolyticus is a gram-negative, halophilic bacterium that is commonly found in estuarine and marine environments worldwide. This microorganism is recognized as one of the most prevalent foodborne pathogens and typically causes acute gastroenteritis in humans (Letchumanan et al., 2014). This bacterium grows preferentially in warm and low-salinity marine water and sometimes colonizes aquatic hosts such as mollusks, shrimp, and fish (Depaola et al., 1990). Due to its frequent presence in aquatic products, V. parahaemolyticus infections are commonly associated with the consumption of raw or undercooked seafood (Ceccarelli et al., 2013). However, our recent report demonstrated the presence of this bacterium in Chinese ready-to-eat (RTE) foods (Xie et al., 2016), a specific type of source that was rarely associated with V. parahaemolyticus infections previously.

RTE foods, such as deli meat, roasted poultry, and cold vegetable dishes, are very popular in China because of their taste and convenience. Unlike in other types of food, no heat processing is needed for RTE foods before consumption. Therefore, these foods tend to be implicated in foodborne illnesses more than other types of food that must be cooked before eating (Tian et al., 2008). Previous studies have shown that RTE foods available in Chinese markets are contaminated by foodborne pathogens such as Listeria monocytogenes (Chen et al., 2014; Wu et al., 2015), Staphylococcus aureus (Yang et al., 2016a), Salmonella spp. (Yang et al., 2016b), and Cronobacter spp. (Xu et al., 2015). The contamination rate of V. parahaemolyticus in Chinese RTE foods can reach 7.63% (Xie et al., 2016). While most food industry processes in China include critical disinfection techniques, contamination with pathogens still occasionally occurs. One of the major reasons for this is that many bacteria possess the ability to form biofilms (Sun and Dong, 2009). By adhering to food surfaces and forming biofilms, bacteria may become a persistent source of contamination, threatening the microbiological quality and safety of food products and perhaps even resulting in foodborne disease and economic losses (Van Houdt and Michiels, 2010). However, there are no reports describing the biofilm formation ability of RTE food-isolated pathogens. In consideration of the mass sale of these foods in China, evaluating the potential pathogenicity of microbes in RTE foods is of critical importance for food security.

Bacterial pathogenicity is usually associated with the presence of virulence factors. The pathogenicity of V. parahaemolyticus is mainly attributed to the production of two major virulence factors: thermo-stable direct hemolysin (TDH), encoded by the tdh gene, and TDH-related hemolysin, encoded by the trh gene (Honda, 1993). TDH has hemolytic activity on Wagatsuma agar, designated the Kanagawa phenomenon (KP), and is involved in cytotoxicity (Miyamoto et al., 1969), while TRH is considered to have a similar action (Honda et al., 1988). The presence of multiple pathogenicity islands (PAIs) is also considered a feature of pathogenic V. parahaemolyticus. For example, the tdh gene is located in VPAI-7 (tdh-PAI), while the trh gene is located in the trh-PAI (Chen et al., 2011). In addition to the tdh-PAI, pandemic V. parahaemolyticus possesses six additional PAIs, VPAI-1 to VPAI-6 (Hurley et al., 2006). All of these PAIs are predominantly present among pandemic isolates and may have been acquired from other Vibrio species (VPAI-1 to VPAI-3) or Shewanella species (VPAI-5 and VPAI-6) by horizontal gene transmission (HGT). Moreover, comparative genomic analyses have revealed that pathogenic V. parahaemolyticus encodes two type III secretion systems (T3SS), while environmental isolates commonly encode only a single system (T3SS1) (Makino et al., 2003). T3SS1 contributes to the cytotoxicity of V. parahaemolyticus but does not appear to play a significant role in intestinal colonization or the induction of intestinal pathology (Park et al., 2004). In contrast, T3SS2 is essential for intestinal colonization and is derived from two separate lineages, one found on VPAI-7 with the tdh gene (T3SS2α), and the other found with the trh gene (T3SS2β) (Okada et al., 2009; Broberg et al., 2011). Similar to T3SS, V. parahaemolyticus also have two type VI secretion systems (T6SS). T6SS2 is found in all strains, while T6SS1 is mostly associated with pathogenic isolates and may contribute to virulence (Salomon et al., 2014).

Our previous studies found that none of the isolates from RTE foods carried the tdh or trh genes (Xie et al., 2016). However, further testing to detect the presence of other virulence factors is lacking, and the risk of V. parahaemolyticus in RTE foods still remains uncertain. Although molecular subtyping by enterobacterial repetitive intergenic consensus sequence PCR (ERIC-PCR) typing and multilocus sequence typing (MLST) has revealed the genetic diversity of V. parahaemolyticus isolates from RTE foods (Xie et al., 2016), the genetic relationship between these isolates and pathogenic isolates remains unknown owing to the lack of genome-wide information on V. parahaemolyticus RTE food isolates. Therefore, the aim of this study was to assess the risk of V. parahaemolyticus in RTE foods through a whole-genome sequencing strategy. We present a comparative genomic analysis of multiple isolates from RTE foods and clinical and other environmental sources (e.g., environmental water and aquatic products). Pan-genome analysis revealed that V. parahaemolyticus RTE food isolates possessed more accessory genes than isolates from other sources on average. At the same time, some RTE food isolates were found to carry several known virulence-associated genes. We also identified multiple genes and single-nucleotide polymorphisms (SNPs) that were closely correlated to RTE food sources, and these factors may contribute to defense processes and biofilm formation in V. parahaemolyticus. The results of this study provide critical insights into the genomic features of V. parahaemolyticus isolated from RTE foods and may aid in improving strategies for microbiological risk assessment.

### MATERIALS AND METHODS

### Bacterial Strains

Twenty-seven isolates of V. parahaemolyticus were collected from different cities and RTE foods in China (**Table S1**). Bacteria were grown overnight in 3% NaCl trypticase soy broth (TSB) before genomic DNA extraction. We also selected 20 clinical and 19 other environmental isolates for which genome sequences were available from the NCBI database (**Table S1**). All analyzed isolates were collected in Asia, and their serotypes, sources, and years of collection are listed in **Table S1**.

### Genome Sequencing and Assembly

Genomic DNA was obtained from V. parahaemolyticus isolates by lysing the bacteria with proteinase K followed by DNA extraction and purification with the Ezup Column Bacteria Genomic DNA Purification Kit (Sangon, Shanghai, China) according to the manufacturer's protocol. Each DNA sample was then fragmented into 400-bp fragments by a Covaris M200 sonicator and used to generate sequencing libraries. Whole genomes were sequenced with the Life Ion S5 platform to an average coverage of 100×. Clean reads were used for de novo assembly with SPAdes v3.6.2 (Bankevich et al., 2012).

### Pan-Genome Analysis

Genome annotation was performed on all analyzed isolates using Prokka v1.11 (Seemann, 2014). The output of Prokka was used to construct the pan-genome using Roary v3.11.2 (Page et al., 2015). A core genome was determined for each isolate using a 99% cutoff, with a BLASTP identity cutoff of 85%. To identify accessory genes overrepresented in RTE food isolates, we used Scoary (Brynildsrud et al., 2016). For this analysis, we used the isolate source as the trait of interest, and we adjusted the P-values for multiple comparisons using the Benjamini and Hochberg method.

### SNP Calling and Genome-Wide Association Analysis

Whole-genome alignments of all strains were constructed with Parsnp v1.2 (Treangen et al., 2014) using the RIMD2210633 genome as a reference, and PhiPack filtering (Bruen et al., 2006) was enabled to remove SNPs located in regions of recombination. SNP sites were then extracted by Harvesttools (Treangen et al., 2014) and annotated with SnpEff (Cingolani et al., 2012). We also used Gubbins (Croucher et al., 2015) to conduct recombination analysis on the core genome alignments generated by Harvesttools.

The core genome SNP alignment was used to estimate the genetic population structure using the hierBAPS module of the BAPS software program, which fits lineages to genome data using nested clustering (Cheng et al., 2013). The estimation used three independent interactions with 15, 30, and 45 clusters at levels 1–3 of the hierarchy, respectively.

To test for evidence of RTE food-associated SNP variation, we used the Cochran–Mantel–Haenszel (CMH) test as implemented in PLINK (Purcell et al., 2007). To account for the population structure, we used the BAPS level 3 clustering in the CMH test. Only SNPs with a minor allele frequency (MAF) > 0.01 across all isolates were used for association analysis. An association was considered statistically significant if the adjusted P-value (Bonferroni-corrected) of the SNP was less than 0.05. To enable that the SNPs were specifically associated with the RTE food isolates, we filtered out the non-significant SNPs from the above outliers in the comparisons of RTE food isolates to clinical and other environmental isolates separately (the Fisher's Exact test, P-value ≥ 0.05).

### Phylogenetic Analysis

Based on the SNP alignment, a maximum-likelihood (ML) phylogenetic tree was constructed using FastTree v. 2.1.10 with the general time-reversible (GTR) and gamma model of nucleotide substitution (Price et al., 2010). The ML phylogeny was visualized and annotated using iTOL (Letunic and Bork, 2016).

### Functional Analysis

To assess associations between RTE food-related accessory genes or missense SNP-containing genes and functional gene categories, we used BLAST to compare representative gene sequences with the NCBI Non-redundant (NR) and Clusters of Orthologous Groups (COG) protein database. Pathway annotation was conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) Automatic Annotation Server (KAAS).

### Biofilm Formation of *V. parahaemolyticus* Isolates

A crystal violet staining method was applied to examine the biofilm-forming abilities of V. parahaemolyticus RTE food isolates, as described by Ye et al. (2014). Briefly, the isolates were inoculated into 5 mL TSB and grown at 37◦C with shaking at 150 rpm for 14 h. Thirty microliters of cultures with an optical density at 590 nm (OD590) of 0.65 were inoculated into 96-well polystyrene plates containing 90 µL fresh TSB and incubated at 37◦C for 24 h. The plates were rinsed three times with deionized water, and adherent bacterial cells were stained with 1% crystal violet for 30 min. After rinsing three times with deionized water, the crystal violet was liberated by acetic acid (30%). The OD<sup>590</sup> values of each well were measured. Each strain was assessed a minimum of three times. The OD values of the tested samples were normalized to that of the negative control (ODc), and biofilm formation ability was determined according to a previous metric: strong biofilm (OD > 4ODc), intermediate biofilm (2OD<sup>c</sup> < OD < 4ODc), weak biofilm (OD<sup>c</sup> < OD < 2ODc), and no biofilm (OD < ODc) (Ding et al., 2014). The formed biofilms were observed under a scanning electron microscope (S-3000N, Hitachi, Tokyo, Japan).

### Data Accessibility

The draft genome sequences of V. parahaemolyticus generated in this study are submitted to the NCBI database under BioProject PRJNA491373. The accession numbers were QYTA00000000, QYSZ00000000, QYSY00000000, QYSX00000000, QYSW00 000000, QYSV00000000, QYSU00000000, QYST00000000, QYSS00000000, QYSR00000000, QYSQ00000000, QYSP00 000000, QYSO00000000, QYSN00000000, QYSM00000000, QYSL00000000, QYSK00000000, QYSJ00000000, QYSI00 000000, and QYSH00000000.

### RESULTS

### General Genomic Features of *V. parahaemolyticus* RTE Food Isolates

We sequenced 27 isolates of V. parahaemolyticus collected from RTE foods sourced from different regions of China. As a comparison, we combined these data with 39 genome sequences of V. parahaemolyticus isolated from clinical and other environmental sources (**Table S1**). To exclude geographical influence, only isolates sampled from Asia were selected.

The size of the draft genomes of the 27 RTE food isolates (**Table S1**) ranged from 4.95 Mb (Vp34) to 5.99 Mb (Vp26). These isolates contained an average of 4,952 genes, which was significantly more genes than among other environmental (4,718 on average) or clinical isolates (4,580 on average) (**Figure 1**). This observation suggested that the persistence of V. parahaemolyticus in RTE foods depended on an increased abundance of accessory genes.

To validate the above inference, we analyzed the pangenome of all isolates. This revealed a pangenome consisting of 21,887 protein-coding genes (**Figure 2A**). Notably, most RTE food isolates (66.7% in the same clade) shared a similar pattern of accessory gene presence and absence, revealing the existence of potential gene clusters that are needed by V. parahaemolyticus to persist in RTE foods. Within the pangenome, 2,136 genes were present in all genomes (core genes) (**Figure 2B**), occupying 35–49% of each isolate's genome. A total of 7,421 accessory genes unique to RTE foods were identified, representing a much higher number than the number of genes specific to clinical and

FIGURE 1 | Genome size differences in *Vibrio parahaemolyticus* isolates from RTE foods and other sources. Boxes show the medians and upper and lower quartiles; whiskers show the most extreme values within 1.5 times the interquartile range. Clinical: *n* = 20; Other environmental: *n* = 19; RTE food sources: *n* = 27. *P*-values are obtained according to the one-way analysis of variance.

other environmental isolates (1,770 and 3,388 accessory genes respectively). This finding reinforced the above result that more accessory genes are presented in RTE foods isolates.

Therefore, we performed a pangenome-wide association analysis to identify accessory genes that are overrepresented in RTE food isolates using Scoary. We found that 109 genes were significantly associated with the RTE food trait, with 79 genes overrepresented in RTE food isolates (**Table S2**). The number of overrepresented genes was obviously smaller that the number of accessory genes unique to RTE foods. The reason was that most of those unique accessory genes were only presented in only one or two RTE food isolates. Instead, most of these overrepresented genes were present in over half of the RTE food isolates but were rarely present in other isolates. This reveals the potential key role of these genes in the persistence of V. parahaemolyticus in RTE foods.

We also determined the presence of known V. parahaemolyticus virulence-associated genes in all RTE foodsourced genomes using pangenome analysis. As reported by Xie et al. (2016), all RTE food isolates were tdh-negative (**Table 1**). Similarly, none of these isolates carried the trh gene except for one isolate, Vp19. This isolate also possessed most genes belonging to T3SSβ, as well as the complete complement of

food isolates, while red indicates clinical isolates and blue indicates other environmental isolates. (B) comparison of the unique accessory genes of RTE food-sourced, clinical, and other environmental isolates.


TABLE 1 | Detection of virulence-associated genes in the sequenced genomes of *V. parahaemolyticus* RTE foods isolates.

+*Indicates a gene or more than 80% genes within a region were detected,* ± *indicates that more than half and less than 80% of the genes within a region were detected.*

T6SS1 and T6SS2 genes. Genes belonging to T3SS2α and VPAI-5 to VPAI-7 were also absent from all RTE food-sourced genomes. However, all of these isolates carried complete or partial complements of T6SS2 genes, and over half of them carried T6SS1. In addition, an incomplete VPAI-4 was present in 26% of RTE food isolates, and several isolates showed the presence of VPAI-1, VPAI-2, or the filamentous vibriophage f237. These results highlighted the pathogenic potential of some V. parahaemolyticus RTE food isolates.

### Phylogenetic Structure of *V. parahaemolyticus* RTE Food Isolates

To provide insight into the genetic diversity of V. parahaemolyticus RTE food isolates, an ML phylogenetic tree was constructed using 68,410 non-recombining core genome SNPs (**Figure 3**). The clinical isolates used in our study were divided into two major lineages: one corresponding to the O3:K6 serotype pandemic strain (Nair et al., 2007), and the other containing two sub-clades was closed to the pathogenic environmental isolate BB22OP (Jensen et al., 2013). Most of the RTE food-sourced and environmental isolates were genetically distinct from these two pathogenic lineages. However, isolate Vp19 was genetically homologous to AQ4037, a pre-pandemic O3:K6 isolate that also possessed the ability to cause foodborne disease (Chen et al., 2011). AQ4037 possessed the same virulence-associated genes as Vp19 (trh, T3SSβ, T6SS1, and T6SS2), revealing that RTE food-sourced and clinical V. parahaemolyticus could share a similar genetic architecture. A similar inference was also obtained from isolate Vp43, which was genetically homologous to VpL83, a clinical tdh- and trh-negative isolate that may possess other uncharacterized virulence factors.

### Identification of RTE Food-Related SNPs

To clarify whether any SNP variants were consistently associated with RTE food isolates, we used the software program PLINK to analyze SNPs in the core genome. Before association analysis, we used BAPS to infer the population structure. Under the threshold of 30 clusters at the third level of the hierarchy, two distinct populations were identified (**Table S3**, **Figure S1**). The O3:K6 serotype isolates belonged to one population, and all other isolates were clustered together in the other population.

Association analysis was conducted after correction for population structure, and the results revealed 78 core genome SNPs that were significantly associated with RTE food sources (adjusted P-value < 0.05, Bonferroni method) (**Figure 4**). After

filtering by separately comparisons, 58 outlier SNPs were finally selected for further analysis (**Table S4**). Among these, 54 SNPs were located in protein-coding regions, resulting in 9 missense variants and 45 synonymous variants. The missense SNPs affected 8 genes (**Table 2**), including a glutathione Stransferase, a sodium/glutamate symporter, an outer membrane phospholipase, an ATP-dependent protease, and two regulators (LuxN and LysR).

### Functional Analysis of RTE Food-Related Genes

Functional analysis was performed on RTE food-related accessory genes and genes containing missense SNPs according to their COG annotation. Seventy-seven of the Eighty-Seven RTE food-related genes had got the COG annotation, and most of the annotated genes were classified as being involved in defense mechanisms and energy production and conversion (**Figure 5**). In addition, a proportion of genes were involved in RNA processing and modification and inorganic ion transport and metabolism. We also identified several functional categories that may contribute to the persistence of V. parahaemolyticus in RTE foods, including chromatin structure and dynamics, cell wall/membrane/envelope biogenesis, and posttranslational modification/protein turnover/chaperones.

We then analyzed the KEGG pathways of the RTE foodrelated genes. The biofilm formation pathway was found to be primarily affected by these genes (**Table S5**, **Figure S2**). In addition, the quorum sensing pathway was affected, revealing that RTE food-related genes may play an important role in the biofilm formation of V. parahaemolyticus in RTE foods. Some of RTE food-related genes were found to be involved in pathways related to drug resistance (e.g., drug metabolism and platinum drug resistance).

### Biofilm Formation of *V. parahaemolyticus* RTE Food Isolates

The ability of RTE food isolates to form biofilm was assessed using the crystal violet staining method. Twenty-three isolates were tested, among which only three isolates were unable to form biofilm (**Figure 6A**). Over half of RTE food isolates were able to form a strong or moderate biofilm, while 35% of isolates formed a relatively weak biofilm (**Figure 6B**). Notably, isolates that were unable to form biofilm or only formed a weak biofilm tended to possess fewer RTE food-related genes and alleles than those isolates that were able to form a strong or moderate biofilm (**Figure 6C**). Additionally, the ability to form biofilm seemed to have no connection with other features of isolates such as serotypes and virulence factors (**Table S6**). Taken together, these results indicate that a majority of V. parahaemolyticus RTE food isolates possessed biofilm formation ability and that this ability may be closely correlated with the number of RTE food-related genes and alleles present in bacteria.

### DISCUSSION

Bacteria isolated from RTE foods in China are mainly derived from the environment through the contamination of vegetables, incomplete heating, or cross-contamination from the environment (Wu et al., 2015). Our previous study showed that a moderate percentage of Chinese RTE foods are contaminated with V. parahaemolyticus, a major food-borne gastroenteritiscausing bacterium (Xie et al., 2016). This bacterium is usually isolated from aquatic products and previous studies were more often focused on these isolates (Letchumanan et al., 2015), while no previous study has explored the pathogenic potential of RTE food isolates. Thus, in this study, we analyzed the genomic features of these isolates for a full understanding of their potential risk. Using next-generation sequencing technology, we obtained the whole genome sequences of 27 V. parahaemolyticus RTE food isolates. Subsequent comparative genomics analysis revealed some genomic features specifically found in these isolates in comparison with other V. parahaemolyticus isolates.

One of the observed features of RTE food isolates is that they generally possess more protein-coding genes than clinical or environmental isolates (**Figure 1**). The phenomenon of clinical isolates having fewer genes on average than non-clinical isolates has been observed in both gram-positive and -negative bacteria (Merhej et al., 2013; Weinert et al., 2015) and is hypothesized to result from a reduction in regulatory complexity. However, V. parahaemolyticus isolates persisting in RTE foods tended to exhibit increases in genomic complexity instead. This difference may be the result of the diversity of environmental and nutritional stresses facing by different kinds of isolates. To ensure successful invasion and survival in host tissues, pathogenic isolates experience a passive loss of transcriptional regulators but in turn gain more genes encoding toxins, toxin-antitoxin


(TA) modules, and proteins involved in DNA replication and repair (Merhej et al., 2013). In contrast, RTE food isolates possess an overrepresentation of genes related to defense mechanisms and energy production and conversion according to their COG annotations (**Figure 5**), which may due to the specificity of this source. During the processes of RTE food preparation and packaging, bacteria experience multiple kinds of specific stresses, such as heat during pre-cooking, changes in nutritional substrates, drought during transportation, and even disinfectants (Lavieri et al., 2015). The acquisition of additional defense-related genes may therefore be a strategy that allows V. parahaemolyticus to tolerate the above stresses so as to persist in RTE foods, and frequent switching among different carriers may require the ability to utilize energy under different energy levels and metabolic yields (Schweinitzer and Josenhans, 2010).

Alternatively, bacteria can also use the strategy of biofilm formation to effectively overcome different environmental stresses (Kubota et al., 2008). Within a biofilm, bacteria are much more resistant to antibiotic treatment (Stewart, 1994; Desai et al., 1998), as well as other environmental stresses (Frank and Koffi, 1990; Kubota et al., 2009). The formation of biofilm is influenced by many factors (Greenberg, 2003), among which quorum sensing (QS) is thought to play a central role (Liaqat et al., 2014). Our analysis demonstrated that some RTE food-related genes are involved in both biofilm formation and QS pathways in V. parahaemolyticus (**Table S5**). In addition, subsequent experiments confirmed that a majority of RTE food isolates possessed the ability to form biofilm, and this ability was positively correlated with

the number of RTE food-related genes present (**Figure 6**). It can thus be seen that if RTE foods are contaminated with these isolates, elimination will be difficult owing to their persistence in the form of biofilms. Thus, the potential threat represented by V. parahaemolyticus in RTE foods deserves attention.

Clinical V. parahaemolyticus isolates are generally positive for some major virulence factors such as TDH, TRH, VPAIs, and T6SS1. Among which, the post-1995 V. parahaemolyticus O3:K6 serotype clone carrying TDH and VPAI-1 to VPAI-7 has disseminated worldwide and is considered to be pandemic (Vuddhakul et al., 2000). Our analysis revealed that only one RTE food isolate (Vp19) expressed TRH, while all other isolates

did not produce TDH or TRH, indicating that none of the RTE food isolates belonged to pandemic clone. This finding largely corresponds to the findings of a previous report (Xie et al., 2016). However, Vp19 showed genetic homology to a pre-pandemic O3:K6 isolate (AQ4037). The AQ4037 isolate is positive for the virulence factors trh, T3SSβ, and T6SS1 and is pathogenetic in humans (Hazen et al., 2015). The same factors were also present in the Vp19 isolate, suggesting the pathogenic potential of this RTE food isolate. In some other RTE food isolates, we also observed the presence of partial T6SS1 genes, which is mostly associated with pathogenic isolates and may contribute to virulence (Yu et al., 2012; Salomon et al., 2014). Notably, T6SS1 genes in different clinical isolates showed a variation range from 73 to 100% (Ronholm et al., 2016), suggesting that partial T6SS1 genes in RTE food isolates still have the pathogenic possibility. Moreover, some VPAIs were frequently present in various RTE food isolates, reflecting the occurrence of HGT of VPAIs among pathogenic isolates and RTE food isolates. The acquisition of multiple virulence factors by HGT can potentially cause the emergence of new pathogens (Espejo et al., 2017). Additionally, recent studies have reported that some clinical isolates do not possess the known PAIs or only carried part of them (Hazen et al., 2015), and some even showed the absence of the tdh and trh genes (Jones et al., 2012; Ottaviani et al., 2012; Hazen et al., 2015; Ronholm et al., 2016), as was observed for the VpL83 isolate. Thus, its homologous isolate, Vp43, is probably pathogenic in humans, even though it does not possess all known virulence factors. Together, these findings indicate the non-negligible potential for the pathogenicity of V. parahaemolyticus RTE food isolates, and further investigation should be performed to validate them.

In summary, this study illustrates the genomic features of V. parahaemolyticus isolated from RTE foods in China. Some of these isolates appear to share similar genetic architecture with clinical isolates and possess some of the known virulenceassociated genes, revealing considerable pathogenic potential. Moreover, most RTE food isolates tended to possess genes and alleles that contribute to defense mechanisms and increase biofilm formation in these isolates, and this may promote their persistence on the surfaces of RTE foods. In consideration of the fact that RTE foods do not require further processing before consumption, contamination with pathogens will pose more of a safety risk for consumers. Therefore, the persistence of V. parahaemolyticus in RTE foods deserves further assessment in the future, and more efforts should be made to develop effective control strategies.

### AUTHOR CONTRIBUTIONS

RP and QW conceived and designed the study. TX, JZ, YD, JW, LX, MC, XW, YZ, SZ, and XY performed the samples and data collection. RP, TX, YL, and TL performed the data analysis. RP and QW wrote and finalized the manuscript.

### ACKNOWLEDGMENTS

This work was supported by the Key Project of Natural Science Foundation of China (31730070), the Natural Science Foundation of Guangdong province (S2012030006235), GDAS' Special Project of Science and Technology Development (2018GDASCX-0911 and 2017GDASCX-0201), the National Key R&D Program of China (2017YFC1600100), and the Science and Technology Planning Project of Guangdong province (2017B020207007).

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.00186/full#supplementary-material


O3:K6 clone of a vibrio parahaemolyticus strain from environmental and clinical sources in thailand. Appl. Environ. Microbiol. 66, 2685–2689. doi: 10.1128/AEM.66.6.2685-2689.2000


**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 Pang, Xie, Wu, Li, Lei, Zhang, Ding, Wang, Xue, Chen, Wei, Zhang, Zhang 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.

# Lactobacillus plantarum ZS2058 and Lactobacillus rhamnosus GG Use Different Mechanisms to Prevent Salmonella Infection in vivo

Junsheng Liu1,2,3, Zhennan Gu1,2 \*, Fanfen Song1,2, Hao Zhang1,2,3, Jianxin Zhao1,2 and Wei Chen1,2,3,4

<sup>1</sup> State Key Laboratory of Food Science and Technology, Jiangnan University, Wuxi, China, <sup>2</sup> School of Food Science and Technology, Jiangnan University, Wuxi, China, <sup>3</sup> National Engineering Research Center for Functional Food, Jiangnan University, Wuxi, China, <sup>4</sup> Beijing Innovation Centre of Food Nutrition and Human Health, Beijing Technology and Business University, Beijing, China

#### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Maria Guadalupe Vizoso Pinto, Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), Argentina Jeanette Wagener, University of Aberdeen, United Kingdom

> \*Correspondence: Zhennan Gu

zhennangu@jiangnan.edu.cn

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 23 July 2018 Accepted: 04 February 2019 Published: 20 February 2019

#### Citation:

Liu J, Gu Z, Song F, Zhang H, Zhao J and Chen W (2019) Lactobacillus plantarum ZS2058 and Lactobacillus rhamnosus GG Use Different Mechanisms to Prevent Salmonella Infection in vivo. Front. Microbiol. 10:299. doi: 10.3389/fmicb.2019.00299 Pathogen-induced infectious diseases pose great threats to public health. Accordingly, many studies have investigated effective strategies targeting pathogenic infections. We previously reported the preventive effects of Lactobacillus plantarum ZS2058 (ZS2058) and L. rhamnosus GG (LGG) against Salmonella spp. in a murine model. Here, we compared the mechanisms underlying the preventive effects of these Lactobacillus strains in vivo. Notably, reduced C-reactive protein levels were observed with both ZS2058 and LGG, which suggests abrogated anti-infection and inflammatory responses. ZS2058 more efficiently reduced the pathogenicity of Salmonella by increasing the level of propionic acid in feces and production of mucin 2 in the mouse colon and activity through the interleukin (IL)-23/IL-22 and IL-23/IL-17 pathways. Meanwhile, LGG more strongly alleviated gut inflammation, as indicated by changes in the levels of tissue necrosis factor (TNF)-α, IL-10 and myeloperoxidase (MPO) in infected mice. Moreover, both ZS2058 and LGG restored the levels of interferon (INF)-γ, a cytokine suppressed by Salmonella, albeit through different pathways. Our results demonstrate that ZS2058 and LGG prevent Salmonella infection via different mechanisms.

#### Keywords: Lactobacillus, Salmonella, preventive effect, different, mechanism

### INTRODUCTION

Salmonella infection, or salmonellosis, is associated with high morbidity and mortality and is therefore a significant public health concern worldwide. This problem is much more severe in developing countries because of the presence of contaminated food and water and poor sanitation facilities (Castillo et al., 2012). Therefore, investigations of effective strategies for coping with infectious diseases are highly significant.

Salmonellosis is usually treated clinically with antibiotics. However, these drugs can cause side effects, including antibiotic resistance and enteric dysbacteriosis. To date, some strains of multi-antibiotic-resistant Salmonella enterica serovar Typhimurium (S. Typhimurium) have been identified in poultry (Rajashekara et al., 2000). Other work has found that antibiotic therapy can

exacerbate Salmonella-induced diarrhea and increase the period of pathogen shedding by at least 3 weeks (Neill et al., 1991). Accordingly, novel and safe strategies for salmonellosis prevention are vitally important.

Probiotics have been identified as a highly promising alternative treatment option for Salmonella infection because these products are associated with fewer side effects and better safety. Many studies have therefore investigated the ability of probiotics to prevent Salmonella infection, as well as the involved mechanisms. Several Lactobacillus strains were found to attenuate the intestinal epithelial barrier dysfunction induced by Salmonella lipopolysaccharide (LPS) (Yeung et al., 2013). L. rhamnosus S1K3 promotes the transcription of genes encoding Toll-like receptors in Peyer's patches (PPs) and modulates the levels of cytokines, which ultimately decreases the Salmonella load in mouse fecal matter and prevents bacterial invasion of internal organs (Kemgang et al., 2016). In a chicken model, L. salivarius CTC2197 was found to reduce Salmonella colonization (Pascual et al., 1999). Furthermore, the transcription of virulence genes identified as important contributors in Salmonella infection were reported to be modulated by Bifidobacterium thermophilum RBL67. This might indicate an important mechanism that could be targeted by probiotics to reduce pathogenicity and promote pathogen clearance (Tanner et al., 2016).

L. rhamnosus GG (LGG) is a well-established and widely recognized probiotic strain used extensively in scientific research and clinical applications. This strain exhibits strong antimicrobial activity against S. Typhimurium via the accumulation of lactic acid (De Keersmaecker et al., 2006). In a C3H/He/Oujco mouse model of infection, LGG reduces the population of S. Typhimurium, as well as the associated mortality (Hudault et al., 1997). Our previous studies demonstrated that both L. plantarum (L. plantarum) ZS2058 and LGG exhibited strong preventive effects against Salmonella-induced animal death in a mouse model (Liu et al., 2018). In this study, we investigated and compared the mechanisms by which ZS2058 and LGG prevent Salmonella infection. Notably, we found that ZS2058 and LGG each used several distinct pathways to prevent Salmonella infection.

### MATERIALS AND METHODS

### Bacteria and Culture Conditions

L. plantarum ZS2058 (ZS2058) and T (T), LGG and S. Typhimurium SL1344 (SL1344) were obtained from the Culture Collections of Food Microbiology (CCFM) at Jiangnan University (Wuxi, China). Lactobacilli and salmonellae were cultured, respectively, in MRS and LB broth (0.3 M NaCl) at 37◦C unless otherwise stated.

### Animal Experiments

Specific pathogen-free mice (SPF; C57BL/6, female, age: 6– 8 weeks) were obtained from the Model Animal Research Center of Nanjing University and housed in a controlled room (SPF, constant temperature of 22◦C ± 2 ◦C and humidity of 55% ± 5%) with a 12 h light-dark cycle at the Animal Experiment Center of Jiangnan University. This study was carried out in accordance with the recommendations of the European Community guidelines (Directive 2010/63/EU). The protocol was approved by the Ethics Committee of Jiangnan University.

Lactobacilli were washed with phosphate-buffered saline (PBS) and resuspended to a density of 5.0 × 10<sup>9</sup> CFU/ml. Each mouse was administered 0.1 ml of bacterial suspension or PBS (control and infection model groups) via gavage for 10 days. Subsequently, the mice were infected with 1.0 × 10<sup>6</sup> CFU of S. Typhimurium SL1344 (Liu et al., 2018). At 2 day post-infection, the mice were sacrificed and samples of sera and intestinal tissue were collected and stored properly (n = 5 or 6).

To study the potential effects of lactobacilli on SCFAs production, in vivo experiments were designed to analyze SCFAs changes in feces of healthy mice. Mice received PBS (control group) or lactobacilli for 10 days (n = 5) without Salmonella infection. Feces were collected at different times (see **Table 1**) and the content of SCFAs were analyzed.

### C-Reactive Protein (CRP) Determination

The serum levels of CRP were measured using enzyme-linked immunosorbent assay (ELISA) kits (Nanjing Senbeijia Biological Technology Co., Ltd., China).

### SCFAs Concentration in Feces

Fecal concentrations of SCFAs were measured as previously reported (Li et al., 2016). Briefly, feces were collected, weighed and freeze-dried. Subsequently, the fecal matter was soaked in saturated sodium chloride and treated with aqueous sulfuric acid and diethyl ether for acidification and extraction, respectively. The SCFA analysis was performed via gas chromatographymass spectrometry on a GCMS-QP2010 Ultra device (Shimadzu Co., Tokyo, Japan).


<sup>a</sup>Data were analyzed by using T-test, n = 5. <sup>∗</sup>p < 0.05 vs. Control.

### Determination of the Levels of Mucin 2 (MUC2), Myeloperoxidase (MPO), and Cytokines

The stored samples of intestines were homogenized at a 1:9 (m/v) dilution in cold PBS. The levels of MUC2, MPO and various cytokines [tissue necrosis factor (TNF)-α, interleukin (IL)-23, IL-22, IL-17, IL-10, interferon (IFN)-γ, IL-18, IL-2, IL-1α, IL-12, and transforming growth factor (TGF)-β] in these homogenates and the stored serum samples were then determined using ELISA kits (Nanjing Senbeijia Biological Technology Co., Ltd.).

### Body, Spleen and Liver Weights, and Organ Relative Ratios

Body weights (BWs) and organ weights were obtained at different time points or at dissection. The organ relative ratio was calculated as the g/g BW (Yoshimi et al., 2001).

### Statistics

Data are expressed as means ± standard errors of the means. Mean values of different groups were analyzed using a oneway variance analysis (one-way ANOVA) with Duncan's multiple range tests with SPSS 16.0 Statistical Software (IBM Corporation, Armonk, NY, United States). Data were considered to be statistically different at a P < 0.05, and were indicated by different superscript letters (such as a, b, and c). Differences between means that do not share a letter are statistically significant.

### RESULTS

### Both ZS2058 and LGG Reduced CRP Levels in Infected Mice

An infection is an important source of inflammatory stimuli and thus promotes the production of CRP (McDade et al., 2008), which can then be used clinically as a diagnostic parameter for infection and bacterial sepsis. As shown in **Figure 1**, Salmonella-infected mice exhibited significantly increased CRP levels (179.1 vs. 163.2 µg/L in the uninfected group). Both ZS2058 (161.4 µg/L) and LGG treatments (161.8 µg/L) were associated with reduced levels of CRP, indicating an alleviation of Salmonella infection in vivo. Our results demonstrate that pretreatment with either Lactobacillus strain could reduce CRP levels in the sera of infected mice.

### ZS2058 Showed More Efficiency in Reducing the Pathogenicity of Salmonella in Intestinal Phase ZS2058 Increased Propionic Acid Level in Feces

The increased research interest in probiotics and prebiotics has directed increasing levels of attention toward SCFAs. As shown in **Table 1**, PA levels in fecal matter were significantly increased after a 10 days course of ZS2058 gavage (18.63 µmol/g) when compared with the PBS-treated (control) group (10.66 µmol/g). Other treatments did not induce any significant changes in the SCFA profile at any time post-gavage (**Table 1**). At 7 and 14

days post-gavage, no changes in the tested SCFAs were observed after treatment with ZS2058, T or LGG. Our results therefore demonstrate that ZS2058 more strongly affected PA production. As this SCFA has been reported to limit Salmonella colonization in vivo (Alshawabkeh and Tabbaa, 2002), our finding suggests that ZS2058 might protect the host from infection by increasing the production of PA.

### ZS2058 Reduced the Level of MUC2 in Infected Mice

A mucus layer covers the intestinal surfaces to form a physiological barrier that excludes luminal bacteria. Several mucins form the gel-forming glycoprotein component of this barrier, of which MUC2 has been identified as the major contributor to the colonic mucus layer (Kumar et al., 2017). As shown in **Figure 2**, infection with S. Typhimurium SL1344 significantly increased the colonic MUC2 levels. However, pretreatment with LGG reduced the production of MUC2 (2.06 ng/g) to a level comparable with that in the uninfected group (2.06 ng/g). However, ZS2058 pretreatment significantly reduced the MUC2 level relative to the uninfected and model groups (1.93 ng/g) (**Figure 2**). In summary, ZS2058 more effectively reduced MUC2 production in Salmonellainfected mice.

### ZS2058 More Strongly Promoted Activity Through the IL-23/IL-22 and IL-23/IL-17 Axes

IL-23 is a critical cytokine associated with host innate immune responses against Salmonella. This cytokine can induce IL-17 and IL-22, which are involved in the rapid response to infectious agents (Valeri and Raffatellu, 2016). In the ileum, IL-23 levels decreased significantly in response to infection (0.528 vs. 0.594 ng/g in the uninfected group; **Figure 3A**), whereas ZS2058

pretreatment restored IL-23 levels in this organ (0.607 ng/g, **Figure 3A**). By contrast, SL1344 infection had no significant effects on the levels of IL-22 (**Figure 3B**) and IL-17 (**Figure 3C**). Both ZS2058 and T (**Figure 3B**) significantly promoted IL-22 production, whereas IL-17 expression was significantly decreased by T (**Figure 3C**) and slightly decreased by LGG (**Figure 3C**). These results demonstrate the ability of ZS2058 to promote the IL23/IL-22 axis in the mouse ileum.

In the colon, both Salmonella infection and ZS2058 pretreatment significantly increased the levels of IL-23, whereas LGG had no significant effect (**Figure 3D**). Regarding IL-22, the ZS2058- and LGG-treated mice, respectively, exhibited a slight increase and slight decrease (non-significant) in the levels of this cytokine (**Figure 3E**) when compared with PBS-treated and infected mice. Regarding IL-17, S. Typhimurium SL1344 infection significantly reduced the expression of this cytokine in the colon (**Figure 3F**). However, ZS2058 significantly increased the colonic level of IL-17 relative to the control group, whereas LGG only restored this cytokine to a comparable level with the control (**Figure 3F**). Our results demonstrate that ZS2058 more strongly promotes the IL-23/IL-17 axis in the mouse colon.

### LGG More Effectively Alleviated Gut Inflammation

Pathogenic infections usually cause gut inflammation, which can be detected by changes in the cytokine profile. As shown in **Figure 4A**, the levels of the pro-inflammatory cytokine TNF-α in the mice colon increased significantly in response to infection, but were significantly reduced by pretreatment with both ZS2058 and LGG. The levels of IL-10, an important anti-inflammatory cytokine, were slightly increased by infection (**Figure 4B**) and significantly reduced by ZS2058, but maintained at a comparable level with that in infected mice by LGG (**Figure 4B**). The level of MPO, a marker of neutrophil infiltration during inflammation, was not significantly affected by infection or pretreatment with ZS2058 or T, but was significantly reduced by LGG (**Figure 4C**). In summary, only LGG reduced the levels of TNF-α in the colon while maintaining IL-10 and significantly reducing MPO (**Figure 4C**). These results suggest that LGG more effectively alleviates gut inflammation.

## ZS2058 and LGG Restored IFN-γ Through Different Pathways

IFN-γ, a characteristic cytokine in a Th1-type response, may play a major role in enhancing the anti-bacterial activities of macrophages. An in vivo study found that IFN-γ deficiency led to an increased splenic Salmonella load and decreased survival rate in mice, while treatment with IFN**-**γ prevented the deterioration associated with infection (Eckmann and Kagnoff, 2001). As shown in **Figure 5A**, the uninfected mice in the control group had a serum IFN-γ level of 1319.4 ng/L. This level decreased significantly to 1114.9 ng/L in response to S. Typhimurium SL1344 infection. However, pretreatment with ZS2058 (1250.4 ng/L) or LGG (1271.0 ng/L) restored the IFN-γ levels to comparable with those in the control group (**Figure 5A**). We subsequently evaluated the levels of IL-12 and IL-18, which induce the production of IFN-γ. Notably, ZS2058 significantly increased the production of IL-18 (**Figure 5B**) while LGG significantly increased the production of IL-12 (**Figure 5C**), indicating that these probiotic strains restored IFN-γ levels through distinct pathways.

## DISCUSSION

Globally, salmonellosis is associated with high rates of morbidity, hospitalization, and mortality. A loss of body weight, splenomegaly and hepatomegaly are all typical symptoms of salmonellosis. In this study, we found that infected mice did not exhibit any of these symptoms (**Figure 6**) during the early stage of infection (2 days post-infection), indicating that S. Typhimurium SL1344 did not cause severe systemic infection at this time point.

As noted above, CRP is a useful diagnostic marker of infection. In effect, CRP can be produced in many types of inflammation (Ridker, 2003). We observed that the serum CRP level increased significantly in response to S. Typhimurium SL1344 infection but decreased in mice that were pretreated with ZS2058 or LGG. We previously reported that L. plantarum ZS2058 and L. rhamnosus GG, but not L. plantarum T, could significantly reduce Salmonella-related deaths in a murine model. Our results provide further evidence of the preventive effects of ZS2058 and LGG against Salmonella infection. The reduced levels of CRP in infected mice treated with both probiotic strain suggests that the inflammatory responses are alleviated systemically. Although LGG appeared to more strongly alleviate gut inflammation, ZS2058 appeared to be a more effective regulator of systemic inflammation resulting from Salmonella.

Following oral infection, pathogens pass through the gastrointestinal tract, where they colonize, survive, replicate and initiate the invasion process. In the gut lumen, the host microbiota plays a crucial role in the inhibition of invasive pathogens through processes including the fermentation of certain carbohydrates to produce SCFAs. In an analysis of the fecal SCFA contents, we found that after a 10-d treatment, only

ZS2058 led to a significant increase in the PA level (**Table 1**). However, ZS2058, T and LGG all tended to increase the levels of acetic acid, PA and butyric acid (**Table 1**). PA is reported to alter the expression of S. Typhimurium genes associated with invasiveness (Lawhon et al., 2002), and to reduce Salmonella colonization in the gastrointestinal tract (Levison, 1973). In vivo studies have shown that SCFAs promote host defense against Salmonella and reduce pathogen loads in the intestinal contents (Sunkara et al., 2011). Accordingly, the ability of ZS2058 to promote the production of PA might contribute to the preventive effects of this probiotic in vivo. Increased levels of SCFAs, such as PA, might comprise an important mechanism by which probiotics mediate anti-Salmonella functions. Although treatment with lactobacilli might alter the profile or increase the abundance of certain species of host microbiota, a limited substrate might not allow excessive changes in SCFAs. Additional

supplementation with substrates, especially prebiotics which are mostly fermented in the colon, might be an effective way to promote the production of SCFAs.

Besides the above-listed effects, SCFAs also contribute to gut immune homeostasis. For example, butyrate can modify the production of II-12 and IL-23 (Berndt et al., 2012). IL-23 induces the production of IL-22 and IL-17 in several cell types, including Th17 cells and NKT cells. In turn, IL-22 can induce the expression of iNOS, the siderophore lipocalin-2 and MUC4 (Raffatellu et al., 2009), while IL-17 recruits neutrophils that play a crucial role in host defense against extracellular bacteria (Santos et al., 2009). We found that in the ileum, LGG treatment did not significantly affect the IL-23/IL-22 (**Figures 3A,B**) and IL-23/IL-17 (**Figures 3A,C**) axes when compared with the Salmonellainfected group, whereas ZS2058 had a stronger positive effect on the IL-23/IL-22 (**Figures 3A,B**) axis when compared with T. LGG-treated mice exhibited slight, non-significant changes in the colon IL-23/IL-22 (**Figures 3D,E**) axis, while ZS2058 promoted the IL-23/IL-22 (**Figures 3D,E**) and IL-23/IL-17 (**Figures 3D,F**) axes. In summary, when compared with LGG, ZS2058 more strongly promoted the IL-23/IL-22 and IL-23/IL-17 axes in the host intestines. These axes might act synergistically to enhance host defense and promote pathogen clearance.

Although the host implements various anti-infection measures, Salmonella develops survival strategies such as unique respiration (Winter et al., 2010) and self-destructive cooperation (Ackermann et al., 2008). Consequently, surviving pathogens in the gut lumen adhere to the mucosal surface and initiate invasion. Salmonella attaches to the mucosal surface by binding to the mannose residues of glycoproteins with assistance from type 1 fimbriae in a process required for colonization of the host intestines (Erbslöh, 2013). As shown in **Figure 2**, the significant increase in MUC2 in response to Salmonella might contribute to this adhesion process. Compared with LGG, ZS2058 more efficiently reduced MUC2 production (**Figure 2**) in the colon, which might reduce the number of binding sites for Salmonella, thus discouraging adhesion and colonization and reducing the risks of invasion.

Taken together, the results listed above suggest that compared to LGG, ZS2058 more effectively reduces the pathogenicity of Salmonella in the intestinal phase. Therefore, pathogen control in the host gastrointestinal tract is required to prevent the further development of systemic infection and sepsis. Following intestinal infection, exogenous pathogens trigger a series of local inflammatory responses. Inflammation is a double-edged sword: an appropriate reaction favors the host's defense against infection, whereas an excessive response may cause unnecessary tissue damage. In the gut, inflammation was reported to provide a respiratory electron acceptor for Salmonella that would give it an advantage over the host microbiota (Winter et al., 2010). As shown in **Figure 4A**, Salmonella infection significantly increased the production of TNF-α in the mouse colon. This strong promoter of inflammatory responses suggests a tendency of Salmonella to provoke inflammation. LGG was previously reported to down-regulate TNF-α production in vivo (Mirpuri et al., 2012), and our observations with LGG and ZS2058 in the present study were consistent with those earlier findings (**Figure 4A**). Notably, Tedelind et al. (2007) reported that PA could inhibit LPS-induced TNF-α production. Therefore, the increased PA (**Table 1**) levels in the colon might enhance the ability of ZS2058 to suppress TNF-α production (**Figure 4A**).

IL-10 is an anti-inflammatory Th2 cytokine that primarily inhibits the production of inflammatory cytokines by innate cells. During infection, IL-10 is required to avoid excessive immune response and prevent development of colitis (Howes et al., 2014). The slight increase in IL-10 (**Figure 4B**) production in the colons of infected mice might indicate a strategy by which the host controls local inflammatory responses. Notably, IL-10 production in the colons of infected mice was suppressed by ZS2058 and maintained by LGG (**Figure 4B**), suggesting that the latter probiotic might better control inflammation in the infected gut. We also evaluated the production of MPO, an indicator of neutrophil filtration and colitis severity, in the colons of infected mice. As shown in **Figure 4C**, Salmonella infection and pretreatment with ZS2058 or T had no significant effect on MPO production in the colon. Consistent with a previous report in which LGG reduced MPO production in the lungs (Li et al., 2009), this study revealed a reduction in colonic MPO in mice treated with LGG (**Figure 4C**). In a murine model, LGG exacerbated the development of DSS-induced colitis and caused animal death (Mileti et al., 2009). However, a clinical study of patients with ulcerative colitis suggests that LGG can effectively and safely maintain remission (Zocco et al., 2006). Based on this study, we suggest that LGG is better able to alleviate gut inflammation during Salmonella infection. On one hand, these effects reduced the growth and competitive advantages of Salmonella vs. the gut microbiota in an inflammatory environment. On the other hand, excessive tissue damage was averted by the restriction of inflammatory responses.

After crossing the intestinal barrier, salmonellae are phagocytized by host immune cells such as macrophages. The bacteria replicate within these cells and are transported to other internal organs and into the blood (Broz et al., 2012), leading some invasive bacteria to shift from the intestinal phase to a systemic phase of infection. At this stage of infection, the in vivo response involves Th1-inducing cytokines that protect against Salmonella infection (Mizuno et al., 2003). Of these cytokines, IFN-γ is among the most powerful first-line defense agents against Salmonella. IFN-γ limits the availability of iron, withdraws iron from intracellular Salmonella (Nairz et al., 2008) and is reported to promote the intracellular killing of bacteria by activating neutrophils and macrophages (Van De Veerdonk et al., 2011). A reduced level of IFN-γ (**Figure 5A**) in response to infection might enhance the intracellular survival of Salmonella and contribute to the development of systemic infection. Pretreatment with ZS2058, T or LGG restored the production of IFN-γ (**Figure 5A**), which suggests that these lactobacillus strains help to promote the clearance of intracellular bacteria. As mentioned, IL-18 and IL-12 both induce IFN-γ production. Our analyses found that both ZS2058 and T restored IFN-γ production through an IL-18-dependent pathway (**Figure 5B**), whereas LGG induced the same effect through an IL-12 dependent pathway (**Figure 5C**). These results demonstrate

that the similar effects of probiotics against pathogenic invasion can be mediated by different mechanisms.

Based on the different effects of ZS2058 and LGG on IL-18 and IL-12 production, we hypothesized that these probiotics might modify host immune responses toward pathogens via distinct mechanisms. Our analysis of several cytokines involved in host defense against infection revealed distinct cytokine profiles (**Figure 7**) in mice pretreated with ZS2058, compared to those pretreated with LGG. Although we did not definitively determine the functions of these modifications in the development and prevention of salmonellosis, we were able to demonstrate the distinct immunoregulatory effects of ZS2058 and LGG to mediate their similar abilities to prevent Salmonella infection in vivo.

In this study, the previously reported preventive effects of ZS2058 and LGG against Salmonella were further verified by the observation of reduced CRP levels in infected mice. We additionally investigated and compared the mechanisms underlying the effects of these two probiotic strains. We found that ZS2058 significantly increased the fecal PA level, increased the colonic MUC2 level and promoted the IL-23/IL-22 and IL-23/IL-17 axes. Our results suggest that compared to LGG, ZS2058 more efficiently reduces the pathogenicity of Salmonella. An analysis of the cytokines in the infected mouse colon revealed

### REFERENCES


that LGG more effectively alleviated gut inflammation. Both probiotics restored the production of IFN-γ, a strong promoter of intracellular bacteria clearance, albeit by distinct pathways. Our results therefore demonstrated that ZS2058 and LGG use different mechanisms to exhibit similar anti-Salmonella effects in a murine model.

### AUTHOR CONTRIBUTIONS

JL designed the study, performed research, analyzed data, and prepared the manuscript. ZG supervised the study and revised the manuscript. FS performed research and interpreted results. HZ, JZ, and WC contributed to scientific idea and procedure designing of the study.

### FUNDING

This work was supported by the National Natural Science Foundation of China (Nos. 31470161 and 31530056, WC), the national first-class discipline program of Food Science and Technology (JUFSTR20180102) and collaborative innovation center of food safety and quality control in Jiangsu Province.


the developing murine colon through upregulation of the IL-10R2 receptor subunit. PLoS One 7:e51955. doi: 10.1371/journal.pone.0051955


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

Copyright © 2019 Liu, Gu, Song, Zhang, Zhao 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-10-00299 February 18, 2019 Time: 15:55 # 9

# Prevalence and Characterization of Food-Related Methicillin-Resistant Staphylococcus aureus (MRSA) in China

Shi Wu1,2† , Jiahui Huang1,2† , Feng Zhang1,2,3, Qingping Wu1,2 \*, Jumei Zhang1,2 , Rui Pang1,2, Haiyan Zeng1,2, Xiaojuan Yang1,2, Moutong Chen1,2, Juan Wang<sup>4</sup> , Jingsha Dai1,2, Liang Xue1,2, Tao Lei1,2 and Xianhu Wei1,2

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

### Reviewed by:

Santiago Castillo Ramírez, National Autonomous University of Mexico, Mexico Liang Li, Los Angeles Biomedical Research Institute, United States

> \*Correspondence: Qingping Wu wuqp203@163.com

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 12 November 2018 Accepted: 05 February 2019 Published: 20 February 2019

#### Citation:

Wu S, Huang J, Zhang F, Wu Q, Zhang J, Pang R, Zeng H, Yang X, Chen M, Wang J, Dai J, Xue L, Lei T and Wei X (2019) Prevalence and Characterization of Food-Related Methicillin-Resistant Staphylococcus aureus (MRSA) in China. Front. Microbiol. 10:304. doi: 10.3389/fmicb.2019.00304 <sup>1</sup> State Key Laboratory of Applied Microbiology Southern China, Guangdong Institute of Microbiology, Guangzhou, China, <sup>2</sup> Guangdong Provincial Key Laboratory of Microbial Culture Collection and Application, Guangdong Open Laboratory of Applied Microbiology, Guangzhou, China, <sup>3</sup> School of Bioscience and Bioengineering, South China University of Technology, Guangzhou, China, <sup>4</sup> College of Food Science, South China Agricultural University, Guangzhou, China

Methicillin-resistant Staphylococcus aureus (MRSA) is an emerging pathogen that is difficult to treat due to the multiresistance of the bacteria upon infection. From 2011 to 2016, 1581 S. aureus strains were isolated from 4300 samples from retail foods covering most provincial capitals in China. To determine the prevalence of food-related MRSA and its genetic background in China, antibiotic resistance, staphylococcal toxin genes, staphylococcal cassette chromosome mec (SCCmec) typing, spa-typing and MLST were carried out in this study. In total, 108 (7.4%) isolates were confirmed for MRSA by phenotyping (cefoxitin) and genotyping (mecA/mecC gene). A total of 52.8% (57/108) of the MRSA isolates belonged to clonal complex 59 (CC59) (ST59, ST338, and ST3355), which was the predominant clone in this study. These CC59 isolates carried SCCmec elements of type IV, V, or III and exhibited spa type t437, t441, t543, t163, t1785, or t3485, and half of them carried major virulence genes, such as the Panton-Valentine leucocidin (PVL) gene. The secondary clones belonged to ST9 (15.7%, 17/108) with a type of t899-SCCmec III and showed a broader range of antimicrobial resistance. The remaining MRSA isolates (31.5%, 34/108) were distributed in 12 different STs and 18 different spa types. All isolates harbored at least one of the enterotoxin genes, whereas only 4 isolates (3.70%) were positive for the toxic shock syndrome toxin tsst alleles. For antibiotic susceptibility testing, all isolates were resistant to more than three antibiotics, and 79.6% of the isolates were resistant to more than 10 antibiotics. Amoxycillin/clavulanic acid, ampicillin, cefoxitin, penicillin, ceftazidime, kanamycin, streptomycin, clindamycin, and telithromycin was the most common antibiotic resistance profile (55.6%, 60/108) in the study. In summary, the results of this study implied that the major food-related MRSA isolate in China was closer to community-associated MRSA, and some of the remaining isolates (ST9-t899- SCCmec III) were supposed to livestock-associated MRSA. In addition, most MRSA isolates showed resistance to multiple drugs and harbored staphylococcal toxin genes. Thus, the pathogenic potential of these isolates cannot be ignored. In addition, further studies are needed to elucidate the transmission routes of MRSA in relation to retail foods and to determine how to prevent the spread of MRSA.

Keywords: MRSA, antibiotic resistance, retail food, spa-typing, MLST

### INTRODUCTION

fmicb-10-00304 February 18, 2019 Time: 16:44 # 2

Methicillin-resistant Staphylococcus aureus (MRSA) is a pathogen of increasing importance in hospitals as well as in the community and livestock. It can be resistant to several antibiotics and quickly disseminates worldwide. In recent years, MRSA has been attributed to an estimated 5400 extra deaths and over one million extra days of hospitalization (Kobayashi et al., 2015). S. aureus became MRSA because of the acquisition of the gene mecA or mecC, which encodes the low-affinity penicillin-binding protein 2a (PBP2a), which, unlike other PBPs, remains active and allows for cell wall biosynthesis at otherwise lethal β-lactam concentrations. The mec operon is carried by the staphylococcal cassette chromosome mec (SCCmec) and most likely originated from horizontal transfer from coagulase-negative staphylococcal species (Banerjee et al., 2008; Garcíaálvarez et al., 2011).

MRSA strains have been reported from various sources. To distinguish epidemiological groups of MRSA, it was divided into hospital-associated MRSA (HA-MRSA), community-associated MRSA (CA-MRSA) and livestock-associated MRSA (LA-MRSA) (Petinaki and Spiliopoulou, 2012). MRSA was first recognized as HA-MRSA in 1961 and then spread into the community and, later on, into healthcare facilities; in 1990, this type became recognized as CA-MRSA. LA-MRSA has always been associated with animals and is linked to a jump from animals to humans (Petinaki and Spiliopoulou, 2012). In general, the genetic backgrounds differ among different types of MRSA. In previous research, CA-MRSA strains frequently harbor SCCmec type IV or V and often produce potent toxins/virulence factors, such as Panton-Valentine leucocidin (PVL), arginine catabolic mobile element (ACME) and phenol-soluble modulins, while HA-MRSA strains typically possess larger-size SCCmec type I-III and are more resistant to other classes of antibiotics (An and Otto, 2008; Chuang and Huang, 2013). For multilocus sequence typing (MLST), sequence type 5 (ST5), ST8, ST22, ST36, and ST45 spread successfully to different regions of the world and caused substantial nosocomial disease (Deleo et al., 2010), whereas CA-MRSA showed five lineages worldwide: ST1-IV (USA400), ST8-IV (USA300), ST30-IV (Pacific/Oceania), ST59-IV and V (USA1000, Taiwan) and ST80-IV (European) (Skov, 2009). In contrast, LA-MRSA strains exhibit co-resistance to many nonβ-lactam antimicrobials (e.g., antibiotics and metals), including those commonly used in animal production, and many of them belong to CC398 or CC9, as determined by MLST (Bens et al., 2006; Cui and Li, 2009; Neela et al., 2009). The SCCmec elements of LA-MRSA are different from those carried by other MRSA genotypes commonly found in the community and healthcare settings (Li et al., 2011). In addition, the majority of LA-MRSA isolates lack toxins, such as PVL and other enterotoxins (Hallin et al., 2011).

Nowadays, MRSA strains have been reported from various foods sources, such as poultry, pork, beef, milk and vegetables, suggesting that foods may serve as reservoirs (Wang et al., 2014; Wu et al., 2018). As is commonly known, foods have many different origins, and different types of MRSA are present in foods of different origin in different countries. In China, large-scale studies of the prevalence of S. aureus in food are scarce. From July 2011 to June 2016, we collected 4300 retail food samples from supermarkets, fairs and farmers' markets that covered most of the provincial capitals of China (**Supplementary Figure 1**) and found 1063 (24.7%) S. aureus-positive samples from all sampling sites. To better understand the genetic background among foodrelated MRSA isolates in China, this study aimed to identify the MRSA isolates from our previous study and to characterize these MRSA isolates for their antimicrobial resistance profiles, their virulence genes and their genotypic types (SCCmec-, MLST, and spa types).

### MATERIALS AND METHODS

### Bacterial Isolates

A total of 1581 S. aureus isolates were collected from 4300 retail food samples in 39 Chinese cities (**Supplementary Figure 1**), comprising 469 isolates from meat and meat products (bacon/sausage, poultry, pork, mutton and beef), 511 isolates from aquatic products (freshwater fish, shrimp and seafood), 368 isolates from quick-frozen products (frozen dumplings/steamed stuffed buns and frozen meat), 148 isolates from ready-to-eat food (cold vegetable/noodle dishes in sauce, fried rice/sushi, roast meat, sausage, and ham), 42 isolates from edible mushrooms, 30 isolates from vegetables and 13 isolates from pasteurized milk. These isolates were obtained between July 2011 and June 2016 according to the GB 4789.10-2010 food microbiological examination of S. aureus (National Food Safety Standards of China) and the most probable number (MPN) method (Gombas et al., 2003). They were identified by Gram stain, catalase and oxidase tests and API STAPH test strips (BioMerieux, Marcy-1'Etoile, France). Each isolate was incubated at 37◦C overnight in brain heart infusion (BHI) broth. Genomic DNA was extracted using a genomic DNA extraction kit (Magen Biotech, Guangzhou, China) according to

the manufacturer's instructions. The concentration of genomic DNA was determined at 260 nm using a NanoDrop-ND-1000 UV-Vis spectrophotometer (Thermo Fisher Scientific, MA, United States).

TABLE 1 | Prevalence of methicillin-resistant Staphylococcus aureus at different sampling sites.


<sup>∗</sup>These cities are direct-controlled municipalities.

### MRSA Confirmation

Cefoxitin disks (30 µg) were used for detecting methicillinresistant isolates. S. aureus ATCC 25923 was used as a control. The mecA/mecC gene, which has been shown to confer methicillin resistance to S. aureus (MRSA), was also detected by PCR using primers as described previously (Perez-Roth et al., 2001; Stegger et al., 2012).

### Antimicrobial Susceptibility Testing

A total of 25 antimicrobial agents were tested for antimicrobial susceptibility in all MRSA isolates. Amoxycillin/clavulanic acid (AMC, 30 µg), ampicillin (AMP, 10 µg), cefepime (FEP, 10 µg), penicillin G (P, 10 U), ceftazidime (CAZ, 30 µg), amikacin (AK, 30 µg), gentamicin (CN, 10 µg), kanamycin (K, 30 µg), streptomycin (S, 25 µg), chloramphenicol (C, 30 µg), clindamycin (DA, 2 µg), erythromycin (E, 15 µg), telithromycin (TEL, 15 µg), ciprofloxacin (CIP, 5 µg), norfloxacin (NOR, 10 µg), tetracycline (TE, 30 µg), linezolid (LZD, 30 µg), trimethoprim/sulphamethoxazole 1:19 (SXT, 25 µg), rifampicin (RD, 5 µg), quinupristin/dalfopristin (QD, 15 µg), teicoplanin (TEC, 30 µg), nitrofurantoin (F, 300 µg) and fusidic acid (FD, 10 µg) were tested by the disk diffusion method using Mueller–Hinton agar and commercially available discs (Oxoid, United Kingdom). Vancomycin and daptomycin were tested by broth microdilution according to the recommendations of the Clinical and Laboratory Standards Institute (CLSI) (CLSI, 2015). Staphylococcus aureus ATCC25923 and Escherichia coli ATCC25922 were used as quality control organisms (CLSI, 2015). The isolates with linezolid resistance determined by disk diffusion were also confirmed using a microdilution test according to the CLSI method for minimum inhibitory concentrations (MICs) (CLSI, 2015). The results of the antimicrobial susceptibilities of the analyzed strains were scored according to the guidelines of the CLSI (CLSI, 2015).

TABLE 2 | Prevalence of methicillin-resistant Staphylococcus aureus in different retail foods.


<sup>a</sup>All quick-frozen foods were stored at −10◦C before being sold.

### Detection of Staphylococcal Toxin Genes

All MRSA isolates were tested by PCR for the presence of 20 genes coding for staphylococcal enterotoxins (sea, seb, sec, sed, see, seg, seh, sei, sej, sek, sel, sem, sen, seo, sep, seq, ser, and seu), and the tsst-1 gene encoding the TSST (Varshney et al., 2009) and the lukSF-PV (pvl) genes were determined by PCR according to a previously published method (Jarraud et al., 2002). The primers and PCR conditions are presented in **Supplementary Table 1**. The amplicons were stained with Goldview, loaded, and electrophoresed in 1.5% agarose at 120 V for 0.5 h and visualized under a UV transilluminator gel imaging system (GE Healthcare, WI, United States). The images were saved as TIFF files for analysis.

### Molecular Typing of MRSA Isolates

All MRSA isolates were subjected to SCCmec typing, spa-typing and MLST. The SCCmec typing method was performed on the isolates by multiplex PCR as previously described (Zhang et al., 2005). The S. aureus protein A (spa) repeat region was amplified according to a published protocol (Shopsin et al., 1999). The MLST scheme used to characterize MRSA isolates is based on the sequence analysis of the following seven housekeeping genes: arcC, aroE, glpF, gmk, pta, tpi, and yqil (Enright et al., 2013).

The DNA fragments were purified using a PCR purification kit (Qiagen, Genman) and sequenced in each direction with Big Dye fluorescent terminators on an ABI 3730XL sequencer (Applied Biosystems). The spa types were randomly assigned using the SpaServer website<sup>1</sup> . For each MLST locus, an allele number was given to each distinct sequence variant, and a distinct ST number was attributed to each distinct combination of alleles at the seven genes. STs were determined using the S. aureus MLST database<sup>2</sup> . Clonal complex (CC) analysis was performed in eBURST v.3 as previously described (Feil et al., 2004). The minimum spanning tree (MST) was constructed with Bionumerics 7.6 software (Applied Maths, Sint-Martens-Latem, Belgium).

### RESULTS

### Prevalence of MRSA in Food

Overall, of the 1581 S. aureus isolates from retail food in China, 108 (6.83%) isolates from 89 positive samples (2.1%, 89/4300) were confirmed as MRSA, which exhibited cefoxitin resistance, 99.1% of isolates (107/108) were positive for mecA genes, and none were positive for the mecC gene. The distribution of MRSA among different sampling sites is shown in **Supplementary Figure 2**. In total, 29 of the 39 sampling cities (74.4%) had MRSA-positive samples, including 22 of the 24 southern cities and 7 of 15 the northern cities. Based on these results, there were 67 (2.4%) and 22 MRSA-positive samples (1.5%) in south China and north China, respectively. The most severe contamination level among the 29 cities was observed in Lasa (8%, 8/100), followed by Fuzhou (7%, 7/100), Nanchang (6%, 6/100), Shantou (5%, 5/100), Zhanjiang (4%, 4/100), Haikou (4%, 4/100), Nanning (4%, 4/100), Chengdu (4%, 4/100), Heyuan (3%, 3/100), Macao (3%, 3/100), Hangzhou (3%, 3/100), Xining (3%, 3/100), Huhehaote (3%, 3/100), Shijiazhuang (3%, 3/100), Shenzhen (2%, 2/100), Shanghai (2%, 2/100), Wuhan (2%, 2/100), Changchun (2%, 2/100), Zhengzhou (2%, 2/100), and Guangzhou (1.6%, 8/500); other cities were represented by one positive sample (**Table 1**). The analyzed food products were classified into seven categories, and the values of MRSA contamination in each sample were determined (**Table 2**). Among the analyzed categories, MRSA was detected in 4.8% (29/604) of raw meat, 3.0% (26/860) of aquatic products, 2.7% (16/601) of quick-frozen food, 1.1% (9/859) of ready-to-eat food, 1.0% (4/419) of vegetables and 0.7% (5/699) of edible mushrooms, whereas pasteurized milk was free of MRSA isolates. Of the 108 MRSA isolates, 34 (7.25%) of the 469 isolates were from raw meat, 31 (6.07%) of the 511 isolates were from aquatic products, 20 (5.43%) of the 368 isolates were from quick-frozen products, 11 (7.43%) of the 148 isolates were from ready-to-eat food, 7 (16.67%) of the 42 isolates were from edible mushrooms, and 5 (16.67%) of the 30 isolates were from vegetables, whereas pasteurized milk was free of MRSA isolates (**Table 2**).

### Antibiotic Resistance Profiles of MRSA Isolates

The antibiotic susceptibility results of 108 MRSA isolates are shown in **Table 3**. All MRSA isolates were resistant to more than three antibiotics, including 33.3% of the isolates that were resistant to 4–10 antibiotics, 46.3% of the isolates that were resistant to 11–15 antibiotics and 20.4% of the isolates that were resistant to 16–26 antibiotics. The isolates were susceptible to linezolid, vancomycin and daptomycin, and the frequencies of resistance to individual agents were 100% for ampicillin and penicillin G, followed by 97.2% for ceftazidime, 87.0% for amoxicillin/clavulanic acid, 83.3% for erythromycin, 79.6% for clindamycin, 75.9% for kanamycin, 74.1% for telithromycin, 67.6% for streptomycin, 65.7% for cefepime and tetracycline, 38.0% for chloramphenicol, 27.8% for gentamycin, ciprofloxacin and fusidic acid, 25.9% for norfloxacin, 22.2% for amikacin, 13.95% for quinupristin/dalfopristin, 13.0% for trimethoprim/sulphamethoxazole 1:19, 7.4% for rifampicin, 2.8% for nitrofurantoin and 0.9% for teicoplanin. In different types of food products, the resistance to most antibiotics were equally distributed. Norfloxacin, clindamycin, quinupristin/dalfopristin, gentamycin, and trimethoprim/sulphamethoxazole 1:19 resistance was rare in MRSA isolated from non-animal sources (i.e., edible mushrooms and vegetables). It is worth noting that amikacin was the most commonly observed resistance antibiotic, whereas quinupristin/dalfopristin, gentamycin, ciprofloxacin and norfloxacin resistance were observed in low frequencies in aquatic product-related MRSA isolates compared to other types of food isolates (**Supplementary Figure 2**). However, AMC-AMP-FOX-P-CAZ-K-S-DA-TEL was the most common antibiotic resistance profile (55.6%, 60/108) in this study. All selected antibiotics were grouped into 15 classes of agents. Five MRSA isolates exhibited resistance

<sup>1</sup>http://spaserver2.ridom.de

<sup>2</sup>https://pubmlst.org/saureus/

to only β-lactam antibiotics, whereas 97 isolates exhibited a multidrug resistance phenotype, with resistance to ≥3 classes of antimicrobial agents.

### Distribution of Virulence Genes

A total of 108 isolates of MRSA from retail food were detected for the presence of virulence genes. As shown in **Figure 1**, each isolate harboured at least one of the virulence genes, including 51 isolates carrying more than 10 genes. In total, 24.07% of isolates (26/108) were positive for the PVL gene lukSF-PV, whereas only 4 isolates (3.70%) were positive for the toxic shock syndrome toxin tsst alleles. Of the 18 investigated enterotoxin genes, the gene seg (82.41%, 88/108) was the most frequently detected, followed by the sei (80.56%, 86/108), seq (79.63%, 85/108), sek (77.78%, 82/108), sem (75.93%, 81/108), sec (75.00%, 79/108), sea (63.89%, 68/108), sep (63.89%, 67/108), sel (56.48%, 59/108), seb (50.93%, 54/108), ser (35.19%, 38/108), sej (34.26%, 37/108),seh (32.41%, 35/108), sen (27.78%, 30/108), seo (14.81%, 16/108), see (12.96%, 14/108), sed (12.96%, 14/108), and seu (4.63%, 5/108). In this study, 107 of 108 MRSA (99.07%) harboured one or more genes for classic SEs (sea, seb, sec, sed, and see), whereas 94 of 108 MRSA isolates (87.04%) harboured the genes of the egc cluster (seg, sei, sem, sen, seo, and seu). The classic SE genes showed 23.4% (195/833) of the detected genes, whereas the egc cluster accounted for 29.4% (245/833). Furthermore, some reported combinations of virulence genes of S. aureus were observed. The sec-sel gene combination, typical of the SaPIbov pathogenicity island, was harboured by 49.07% (53/108) of the isolates. The sea-sek-seq genes, which have been reported on 8Sa3ms and 8Sa3mw, were associated in 43.52% (47/108) of the isolates. In addition, sebsek-seqwas observed on SaIP3, andsed-sej-serwas observed on pIB485, which also showed 47.22% (51/108) and 11.11% (12/108) of the isolates, respectively.

### Molecular Types of MRSA Isolates

The molecular typing results are summarized in **Table 4**. For MLST, the STs of one strain (Sta2529-1) could not be determined. A total of 107 food-related MRSA isolates were assigned to 16 different STs, including ST1, ST5, ST6, ST7, ST9, ST10, ST45, ST59, ST88, ST188, ST338, ST398, ST630, ST943, ST3304, and ST3355. ST59 was the predominant clone and was observed in 47.7% of MRSA isolates (51/107). The remaining strains belonged to ST9 (15.9%, 17/107), ST1 (8.4%, 9/107), ST398 (6.5%, 7/107), ST7 (4.7%, 5/107), ST338 (4.7%, 5/107), ST630 (2.8%, 3/107), ST188 (1.9%, 2/107), and other STs. Based on eBURST analysis, three CCs were identified, including CC59 (ST59, ST338 and ST3355), CC7 (ST7, ST943), and CC1 (ST1, ST3304). spa-typing of all MRSA isolates yielded 24 spa types. t437, t899, t127, and t091 were the most predominant spa types, constituting 71.3% (78/108) of all of the isolates in this study. Other spa types, including t002, t034, t085, t114, t116, t163, t189, t377, t441, t528, t543, t571, t1751, t1764, t2874, t3485, t4549, t4792, t5554, and t9472, were distributed in fewer isolates (27.8%, 30/108). The MRSA isolates were also subjected to identified SCCmec types, and the types of 20 isolates could not be detected. The majority of MRSA isolates possessed SCCmec type IV, which was observed in 63.9% of isolates (69/108), including 50 for SCCmec type IVa, 17 for SCCmec type IVb and 2 for SCCmec type IVd. In addition,

9 isolates belonged to SCCmec III, and 10 isolates belonged to SCCmec V.

A phylogenetic tree based on the 7 concatenated MLST sequences (**Figure 2**) shows the relatedness between the MRSA strains. Two different clusters were observed in this study (designated as A and B). Cluster A included ST1, ST3304, ST188, ST9, ST15, ST6, ST5, ST7, ST943, ST630, and ST88, and cluster B included ST398, ST10, ST338, ST3355, and ST59, which showed distant genetic relationships. STs correlated well with spa types, such as ST1-t127, ST188-t189, ST9-t899, ST7 t091, ST59-t437, and ST338-t437. Overall, the genetic diversity among MRSA isolates was higher based on different cities and different food sources. For different food products, more than three subtypes were found in each type of food, except edible mushrooms and vegetables. CC59-t437 (45.4%, 49/108) was the predominant clone in this study, but ST9-t899-SCCmec IVb was the predominant clone in quick-frozen meat and was found only in animal-derived food (raw meat, quick-frozen meat and readyto-eat meat). In addition, 80% (4/5) of SCCmec III-ST338-t437 isolates were found in aquatic products from the city of Kunming. However, some strains showed correlations among geographical locations, such as Sta223-2 (isolated from Shenzhen), Sta251 (isolated from Guangzhou), Sta403 (isolated from Shantou), Sta487 (isolated from Heyuan), and Sta1753 (isolated from Zhanjiang), which were isolated from neighboring cities in this study and clustered into one subtype (ST9-t899-SCCmec IVb).

### DISCUSSION

MRSA is a significant public health concern in humans and animals. The rate of mortality due to MRSA infections has remained high in recent years. In hospitals, the prevalence rates of MRSA in some Asian countries, such as Taiwan, China, Japan, and South Korea, can reach 70–80% (Chuang and Huang, 2013). For CA-MRSA, the prevalence varies substantially worldwide and ranges from less than 1% to more than 50% in different countries (Tristan et al., 2007; Deurenberg and Stobberingh, 2008). MRSA


#### TABLE 4 | The STs, spa types and SCCmec types of the MRSA strains isolated from retail food in China.


<sup>a</sup>ND, not detected.

generated using S.T.A.R.T (version 2).

has also been isolated from animals, as reported in many previous studies (Leonard and Markey, 2008). It is important to identify the origin of food-related MRSA and to evaluate the potential pathogenicity of these MRSA isolates. From July 2011 to June 2016, 4300 retail food samples were collected from supermarkets, fairs and farmers' markets, covering most of the provincial capitals of China. This wide-scale and systematic investigation of S. aureus from retail food in China supplements nationwide qualitative and quantitative data on the prevalence and levels of S. aureus. In this study, we determined the MRSA isolates from these food-related S. aureus isolates and found 108 MRSA isolates (6.83%) in various types of food products (raw meat, aquatic products, quick-frozen products, ready-to-eat food, edible mushrooms and vegetables) from most of the sampling cities (29/39, 74.4%) in China, which suggested that retail food in China could be contaminated with MRSA.

Many studies have evaluated the presence of MRSA in retail food. In China, among studies that sampled retail food, 6.07% of MRSA isolates were found in quick-frozen dumpling samples of Shaanxi province (Hao et al., 2015) and in 1.7% of chicken samples (Wang et al., 2013), whereas MRSA was present in 29.5% of grain products, meat products and dairy products in southwest China. Overall, these rates varied from our results in this study. This incidence may be attributed to a number of factors, such as the sample size, sampling site, types of samples or isolation methods. In 2015, (Wang et al., 2017) collected

1150 S. aureus isolates from retail markets from 203 cities in 24 provinces in China and found 91 isolates (7.9%) that were identified as MRSA by PCR. The MRSA isolates were distributed in raw meat, rice and flour products, vegetable salads, sandwiches, meat products, and eggs. Compared with other countries' studies, the prevalence of MRSA from retail foods in China was not low. For instance, Ge et al. (2017) conducted a 1-year survey in 2010- 2011 from 3520 retail meats in eight U.S. states and found that 1.9% of samples were positive for MRSA. An Italian survey found that 6 of 160 (3.75%) foods of animal origin harbored MRSA (Normanno et al., 2007). MRSA was present in 3.6% (15/421) of retail meat in Korea (Lee, 2003), 1.6% (5/318) of food animals in Spain (Lozano et al., 2009), 0.75% (20/2662) in Switzerland (Huber et al., 2010), 3% (11/367) in Greece (Papadopoulos et al., 2018), and 1.9% (2/103) in the United Kingdom (Hadjirin et al., 2015). Therefore, Chinese food safety regulators should improve hygiene and supervision efforts.

Currently, MRSA isolated from food-producing animals has been recognized as LA-MRSA. The worldwide emergence of LA-MRSA since 2005 has prompted many of the surveys to focus on retail meat as a potential vehicle for this new MRSA clone (Ge et al., 2017). In general, ST398 was recognized as the most typical type in LA-MRSA. ST398 in swine, cattle and other animal species has been analyzed in several publications (Walther et al., 2009; Fessler et al., 2010; Graveland et al., 2011). It reported that 24–100% of pig farmers, 37% of poultry farmers, 30–38% of cattle farmers and up to 45% of veterinarians are colonized with MRSA CC398 in the nares (Köck et al., 2014). In this study, most MRSA isolates (88.9%, 96/108) were isolated from animal-derived food (raw meat, aquatic products, quickfrozen products and ready-to-eat food), but only 7 MRSA isolates belonged to this ST type. STs of MRSA isolated from retail food focused on CC59 (ST59, ST338, and ST3355). This showed significant genetic uniformity with the predominant Asian CA-MRSA lineage, which can reach 35.8–76.7% with CA-MRSA in China (Yang et al., 2017). In general, the CC59 clone always carried SCCmec IV/V and was concentrated in spa t437 and t441. In 2013, Li et al. (2013) analyzed the 110 CC59 isolates from invasive and noninvasive diseases in China by MLST, SCCmec typing and spa-typing and found that 65.5% of the clones were ST59-t437-IVa. From February 2016 to January 2017 Yang et al. (2017), collected S. aureus strains in Beijing Children's hospital from the respiratory tract, skin and soft tissue, and sterile sites in 104 child cases. Of these, 54.8% were categorized as community-associated SA (CA-SA) infections, and ST59-SCCmec IV-t437 (61.7%) was the most prevalent MRSA genotype. In addition to China, ST59 has also been reported in Vietnam, Japan, Australia and other countries with CA-MRSA infection (Tang et al., 2007; Coombs et al., 2010; Higuchi et al., 2010). In contrast, ST239 and ST5 were found in a nationwide study to be two major MRSA clones with unique geographic distributions in Chinese hospitals (Liu et al., 2009). Accordingly, these results implied that the major food-related MRSA in China was closer to CA-MRSA, a finding that should be brought to public attention.

In the present study, 15.7% (17/108) of isolates belonged to ST9, the secondary clones of which were found only in animal-derived food (raw meat, quick-frozen meat and readyto-eat meat). According to previous studies, ST9 was the predominant S. aureus and MRSA genotype in pigs and related workers in Asia (Chuang and Huang, 2015). In 2008, ST9 MRSA was first found in Chinese pigs, and farm workers carried ST9-t899-SCCmec III-PVL-negative (Cui et al., 2009). Studies in Taiwan, Hong Kong, Malaysia and Thailand have since reported the prevalence of this type of LA-MRSA (Neela et al., 2009; Graveland et al., 2011; Larsen et al., 2012; Lo et al., 2012). ST9 is the most prevalent LA-MRSA in most Asian countries and differs from the European pig-associated clone (ST398) with regard to clonal type, SCCmec content and resistance profile (Ye et al., 2016). In China, ST9 strains always showed SCCmec III with spa t899. These characteristics were in accordance with our results, which showed that the ST9 MRSA isolated from our study showed ST9-t899-SCCmec III. Therefore, this portion of the food-related MRSA isolates was supposed to be LA-MRSA. However, it is worth noting that in this study, all ST9 MRSA isolates were resistant to more than 15 antibiotics and showed a broader range of antimicrobial resistance than ST59 MRSA (**Figure 2**). More than 80% of ST9 MRSA was resistant to erythromycin, ciprofloxacin, gentamicin, tetracycline and clindamycin. Currently, more evidence has implicated animals as reservoirs of antimicrobialresistant bacteria and has indicated that animals can potentially transmit resistance genes to humans (Liu et al., 2018). Thus, more attention should be focused on this type of strain among food chains.

Except for CC59 and ST9, the remaining strains belonged to ST1 (8.4%, 9/107), ST398 (6.5%, 7/107), ST7 (4.7%, 5/107), ST630 (2.8%, 3/107), and ST188 (1.9%, 2/107). Of these strains, most STs correlated well with spa types, but seven ST398 isolates distributed in four different spa types. In addition, SCCmec types were not detected in most (**Figure 2**). This finding is consistent with the results of a previous study suggesting that divergent SCCmec elements were inserted into the (clonal) ST398 MSSA (van Duijkeren et al., 2008; Smith and Pearson, 2011). Therefore, further studies evaluating MSSA ST398 in retail food are needed to determine the reason for the correlation. In this study, ST1 t127 and ST7-t091 were also detected. These STs showed high genetic diversity among MRSA isolates based on different cities and different food sources, which is a common finding in isolates of human and animal origin (Franco et al., 2011; Hummerjohann et al., 2014). Thus, these types of S. aureus isolates have been relevant to a variety of clinical infections and have theoretical pathogenic potential.

As is usually observed with CA-MRSA strains, CA-MRSA ST59 isolates had significantly more pronounced virulence in various animal infection models than the geographically matched HA-MRSA clones ST5 and ST239 (Li et al., 2009, 2016). For the strains of ST59, the evolutionary acquisition of PVL, the higher expression of α-toxin and, possibly, the loss of a large portion of the β-haemolysin-converting prophage probably contribute to its higher pathogenic potential (Chen et al., 2013; Chen and Huang, 2014). In this study, we also investigated the PVL gene and found 24.07% PVL-positive MRSA isolates. All PVL-positive MRSA isolates belonged to CC59, including 22 ST59 isolates

and four ST338 isolates. For LA-MRSA ST9 isolates, certain important virulence factors, such as PVL, are absent. Despite a lack of virulence factors, ST9 strains have been found to cause disease in humans (van Loo et al., 2007; Liu et al., 2009; Chuang and Huang, 2015). Therefore, the hazards of these strains for consumers cannot be ignored.

As an important foodborne pathogen, S. aureus is involved in most staphylococcal food poisoning (SFP) events due to staphylococcal enterotoxins, the virulence factors that are heat stable and proteolytic or demonstrate emetic activity (Grumann et al., 2014). In this study, we also investigated most of the enterotoxin genes of S. aureus. All food-related MRSA isolates harbored at least one of the SE genes. The percentages of MRSA isolates containing sea, seb, and sec all exceeded 50%, whereas sed and see were detected in only fourteen MRSA isolates (12.96%). Except for the egc cluster, which accounted for 29.4% (245/833) of the detected genes in the present study, other SE genes showed 48.4% (403/833) of the detected genes. Generally, SEA, followed by SED, is the enterotoxin most frequently associated with SFP, although outbreaks caused by SEB, SEC, and SEE have also been reported (Argudín et al., 2012). In contrast to classical SEs, the relationship between the novel SEs and SFP is not fully understood, but most of them (SEG, SEH and SEI, SER, SES, and SET) have been shown to be emetic after oral administration in a primate model (Argudín et al., 2010). In this study, 94 of 108 MRSA isolates (87.04%) harbored egc cluster genes (seg, sei, sem, sen, seo, and seu), and 94.4% (102/108) of isolates harbored one or more genes for other novel SEs or tsst-1 genes. It is suggested that attention should not only be paid to classical enterotoxins but also to novel ones since an increasing number of foodborne outbreaks have been associated with novel enterotoxins.

Nowadays, multiple drug resistance is the most important characteristic of MRSA isolates. For the 26 clinically relevant antibiotics investigated, all MRSA isolates were resistant to more than three antibiotics. It seems higher than many previous studies in food-related MRSA isolates (Hanson et al., 2011; Shahraz et al., 2012; Ge et al., 2017; Tang et al., 2017). Interestingly, norfloxacin, clindamycin, trimethoprim/sulphamethoxazole 1:19, gentamycin and quinupristin/dalfopristin resistance were rare in MRSA isolated from non-animal sources (edible mushrooms and vegetables) (**Supplementary Figure 2**). The reason for this finding may be attributed to the food source. In general, most animal-derived food-related S. aureus came from animal farms that used these antibiotics as food supplements in animal feed (Wang et al., 2014), whereas S. aureus isolated from edible mushrooms or vegetables most likely originated in the environment. Furthermore, amikacin was the most commonly observed resistance antibiotic in aquatic product-related MRSA isolates, whereas gentamycin, quinupristin/dalfopristin, ciprofloxacin, and norfloxacin resistances were observed less frequently than other types of food isolates (**Supplementary Figure 2**). As we know, amikacin, gentamycin, kanamycin and streptomycin all belong to the aminoglycosides, which exert their bactericidal effects by irreversibly binding to the 30S ribosomal subunits of susceptible bacteria, inhibiting protein synthesis (Hammerberg et al., 1986). The most common clinical resistance mechanism to aminoglycosides is the structural modification of aminoglycosides by aminoglycosidemodifying enzymes (AMEs). Amikacin is a broad-spectrum semi-synthetic derivative of kanamycin and a poor substrate of many AMEs. Furthermore, even if amikacin is modified by AMEs, the modified amikacin can still bind to the 30S ribosomal subunit (Yuan et al., 2013). Therefore, amikacin is one of the most potent classes of antibiotics in S. aureus infection. In fact, the mechanism of MRSA resistance to amikacin is poorly understood. Thus, why amikacin resistance was higher in MRSA isolates from aquatic products should be further studied.

## CONCLUSION

Essential for human survival, food is the one of most basic necessities of life. In this study, we investigated food-related MRSA and determined its genetic background in China. MRSA isolates were found in most investigated cities and were observed in different types of food samples. ST59 was the predominant clone in food-related MRSA in China, which indicated that the major food-related MRSA isolates in China were closer to CA-MRSA. Moreover, as the major LA-MRSA in Asian populations, ST9 MRSA was the secondary clone and showed a broader range of antimicrobial resistance. Determination of staphylococcal toxin genes presented their virulence potential, and antimicrobial susceptibility testing further confirmed the severe situation of MRSA isolated in retail food in China. In addition, some antibiotics were also found to be higher in some types of food. However, further studies are required to determine the reason for this correlation and to elucidate the transmission routes of MRSA in relation to retail foods in order to provide the tools for preventing the spread of MRSA.

### AUTHOR CONTRIBUTIONS

QW, JZ, SW, and TL conceived and designed the experiments. JH, FZ, and JD performed the experiments. SW, RP, and HZ analyzed the data. XW, LX, MC, and XY contributed reagents, materials, and analysis tools. SW and JW contributed to the writing of the manuscript.

### FUNDING

We would like to acknowledge the financing support of National Natural Science Foundation of China (No. 31801657), China Postdoctoral Science Foundation (2017M612623), and GDAS' Special Project of Science and Technology Development (2017GDASCX-0817).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb.2019. 00304/full#supplementary-material

### REFERENCES

fmicb-10-00304 February 18, 2019 Time: 16:44 # 11


1 methicillin-resistant Staphylococcus aureus from pigs. J. Antimicrob. Chemother. 66, 1231–1235. doi: 10.1093/jac/dkr115


in pigs in Thailand. PLoS One 7:e31245. doi: 10.1371/journal.pone.003 1245


sequencing for typing of Staphylococcus aureus strains. J. Clin. Microbiol. 37, 3556–3563.


aureus CC9 in humans. Appl. Environ. Microbiol. 82, 3892–3899. doi: 10.1128/AEM.00091-16


**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 Wu, Huang, Zhang, Wu, Zhang, Pang, Zeng, Yang, Chen, Wang, Dai, Xue, Lei and Wei. 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.

# Survival and Environmental Stress Resistance of Cronobacter sakazakii Exposed to Vacuum or Air Packaging and Stored at Different Temperatures

Yichen Bai, Haibo Yu, Du Guo, Shengyi Fei and Chao Shi\*

College of Food Science and Engineering, Northwest A&F University, Yangling, China

The aim of this study was to evaluate the survival of Cronobacter sakazakii exposed to vacuum or air packaging, then stored at 4, 10, or 25◦C, and the environmental stress resistance of vacuum-packaged or air-packaged bacterial cells were determined by subjecting the cells to reconstituted infant formula at 50◦C, in acid (simulated gastric fluid, pH = 3.5), and in bile salt [bile salt solution, 5% (wt/vol)]. A cocktail culture of C. sakazakii desiccated on the bottom of sterile petri plates was air-packaged or vacuum-packaged and then stored at 4, 10, or 25◦C for 10 days. The viable cell populations during storage were examined, and the vacuum-packaged and airpackaged cells (stored at 10◦C for 4 days) were subsequently exposed to heat, acid, or bile salt. The results show that the populations of vacuum-packaged and airpackaged C. sakazakii were reduced by 1.6 and 0.9 log colony-forming units (CFU)/ml at 4 ◦C and by 1.6 and 1.3 log CFU/ml at 25◦C, respectively, in 10 days. At 10◦C, significant reductions of 3.1 and 2.4 log CFU/ml were observed for vacuum-packaged and airpackaged cells, respectively. Vacuum packaging followed by storage at 10◦C for 4 days caused significant decreases in the resistance of C. sakazakii to heat, acid, and bile salt conditions compared with air packaging. These results suggest that the application of vacuum packaging for powdered infant formula could be useful to minimize the risk of C. sakazakii.

Keywords: Cronobacter sakazakii, air packaging, vacuum packaging, survival, environmental stress

### INTRODUCTION

Cronobacter sakazakii (formerly known as Enterobacter sakazakii) is a Gram-negative, non-sporeforming bacillus that exists in the environment as well as in a wide variety of foods. It is regarded as a newly developing foodborne pathogen (Wan-Ling et al., 2010). C. sakazakii has been implicated in severe forms of neonatal infection, such as bacteremia, meningitis, and necrotizing enterocolitis, particularly in infants and premature babies; the mortality rates associated with these bacteria range from 50 to 80% (Li et al., 2013). Powdered infant formula (PIF), which is a main source of nutrition for neonates and infants, has been recognized as the major vehicle of transmission of C. sakazakii, and the consumption of contaminated PIF is associated with the majority of C. sakazakii outbreaks (Shukla et al., 2016). Due to the role of trehalose accumulation within its cells, C. sakazakii is remarkably resistant to desiccation (Breeuwer et al., 2010). Some capsulated strains of C. sakazakii

#### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Ariadnna Cruz-Córdova, Hospital Infantil de México Federico Gómez, Mexico Zhao Chen, University of Maryland, College Park, United States

> \*Correspondence: Chao Shi meilixinong@nwsuaf.edu.cn

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 23 November 2018 Accepted: 05 February 2019 Published: 20 February 2019

#### Citation:

Bai Y, Yu H, Guo D, Fei S and Shi C (2019) Survival and Environmental Stress Resistance of Cronobacter sakazakii Exposed to Vacuum or Air Packaging and Stored at Different Temperatures. Front. Microbiol. 10:303. doi: 10.3389/fmicb.2019.00303

**182**

in dehydrated PIF (with a low water activity of 0.2) were still recoverable after 2.5 years (Barron and Forsythe, 2007). Fei et al. (2017) determined the prevalence of C. sakazakii isolates from PIF collected from Chinese retail markets, and the contamination rate of C. sakazakii in that study was 2.8%. Heperkan et al. (2017) reported in a different study that the prevalence of C. sakazakii was 8.0% in milk powders and that the number of C. sakazakii was 7–15 MPN/g.

Foodborne pathogens face several hurdles after they enter the host. The strong acid condition in the stomach is the first line of host defense against these pathogens. Additionally, the presence of bile salts, antimicrobial peptides, and other hostile conditions also act as defenses against serious infections in humans (Anf et al., 2017). C. sakazakii exhibits unusual resistance to acid stress growth conditions, and they can grow at minimum pH values of ∼4.5, although this value varies depending on the strain and type of acid (Alvarezordóñez et al., 2014). Bile is an important antimicrobial component of the human digestive system, but growth has been described for some C. sakazakii isolates at bile salt concentrations as high as 5% (Fakruddin et al., 2014). Compared with other members of the Enterobacteriaceae family, C. sakazakii is reported to be significantly more thermotolerant (Amalaradjou and Venkitanarayanan, 2011). This feature represents a competitive advantage, facilitating its survival during improper PIF reconstitution (Shi et al., 2017).

Environmental stresses are known to induce adaptive responses within the bacterial cell. Bacterial pathogens have the ability to enhance their resistance to lethal stresses after their exposure to a sublethal one via genetic regulation or physiological adaptation (Yang et al., 2015). The phenomenon of one type of stress-response imparting auxiliary protection to cells subsequently stressed at higher levels is widely documented and may be referred to as "cross-protection" (Wesche et al., 2009). Cross-protection has been a growing concern in the microbiological food safety area.

To reduce the levels of C. sakazakii contamination in the finished products reconstituted from PIF, which are introduced during food processing, many thermal and nonthermal technologies for C. sakazakii inactivation have been proposed (Ha and Kang, 2014; Pina-Pérez et al., 2015; Shi et al., 2016). Additionally, the Food and Agriculture Organization of the United Nations/World Health Organization (2007) recommended that PIF reconstitution should be performed at 70◦C to reduce the risk of C. sakazakii survival during PIF preparation, and some studies have investigated the efficacy of antimicrobials for reducing the tolerance of C. sakazakii to environmental stresses to improve the safety of PIF (Amalaradjou and Venkitanarayanan, 2011). However, studies focusing on factors affecting the survival and environmental stress resistance of C. sakazakii during the packaging and storage processes are lacking.

The aim of this work was to evaluate the survival of C. sakazakii exposed to vacuum or air packaging and stored at 4, 10, or 25◦C. Additionally, the effects of vacuum packaging and air packaging on the thermotolerance and survival under simulated gastrointestinal conditions and bile salt conditions of C. sakazakii strains were also assessed. Desiccated C. sakazakii was used in this study to simulate the conditions of intrinsic PIF contamination.

## MATERIALS AND METHODS

## Bacterial Culture

Three strains of C. sakazakii (ATCC 29544, 14-15, and 18-8) were used for this study. ATCC 29544 was procured from the American Type Culture Collection (ATCC, Manassas, VA, United States). Strains 14-15 and 18-8 were isolated from PIF and baby formula, respectively (Li et al., 2016). All C. sakazakii strains were stored at −80◦C in Luria-Bertani (LB) broth with 30% (vol/vol) glycerin. To activate the frozen cultures, a loopful of each strain was streaked with a flamed loop onto Tryptic Soy Agar (TSA, Land Bridge, Beijing, China) and incubated at 37◦C for 24 h, followed by incubation at 37◦C in sterile Tryptic Soy Broth (TSB; Land Bridge) for 18 h.

## Preparation of Desiccated C. sakazakii

The method previously described by Al-Nabulsi et al. (2009) was used with slight modifications to prepare desiccated C. sakazakii. After centrifugation (8000 × g, 10 min, 4◦C), each 18-h culture of C. sakazakii was washed twice with sterile phosphate buffered saline (PBS, pH = 7.4). Each culture was then adjusted to an OD600 nm value of 1.0 with 0.2% (wt/vol) buffered peptone water (BPW). Equal volumes of three cultures were aseptically combined to produce a cocktail. Subsequently, 50 µl of this C. sakazakii cocktail was distributed evenly into the bottom of sterile petri plates. The plates were placed in a drying oven at 40◦C for 2 h without their lids. The plates were then covered with lids, transferred to a desiccator, and stored at 25◦C for 4 days to dry the cells.

### Preparation of Vacuum-Packaged and Air-Packaged C. sakazakii

For vacuum packaging, a vacuum sealer (Deli Group No. 14885, Zhejiang, China) was used. The plates coated with desiccated C. sakazakii were vacuum-packaged in thick plastic bags (Deli Group No. 14914) made of polyamide/polyethylene. For air packaging, the plates were placed in the same type of plastic bags and packed with air.

## Viability of Survivors During Storage

All packaged samples were stored at 4, 10, or 25◦C, and viable cell populations during storage were examined on day 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10, respectively. To collect the desiccated C. sakazakii, 3 ml of 0.2% (wt/vol) BPW was added to the bottom of each sterile plate. Serial dilutions of the collected cells made using PBS, and 100 µl of the diluents were plated evenly in triplicate on TSA plates. Five 200-µl aliquots of the undiluted samples also were spread plated onto five TSA plates to achieve a detection limit of 3 CFU/ml. The number of colonies was counted after a 24-h incubation at 37◦C.

### Effect of Vacuum and Air Packaging on the Environmental Stress Resistance of C. sakazakii

PIF was procured from a local supermarket (Yangling District, China). To detect any natural contamination, the method previously described by Yang et al. (2015) was employed with slight modifications, and background microflora (<25 CFU/g) was observed in the PIF, the colony morphology of background microbes was obviously bigger than that of C. sakazakii. The PIF (13.5 g) was then reconstituted with 90 ml of sterile deionized water as per the manufacturer's instructions on the label. Solutions of desiccated cells without storage (control), desiccated cells subjected to vacuum packaging and stored for 4 days, or desiccated cells subjected to air packaging and stored for 4 days were collected, the bacterial concentration of each sample was adjusted, and the samples were added to the reconstituted infant formula. If necessary, several samples treated in the same conditions were pooled together to reach the fixed bacterial concentration for the experiment.

### Determination of Heat Resistance of C. sakazakii

For heat resistance assay, the concentration of initial sample was about 3.0 log CFU/ml. All experimental samples were exposed to heat stress of 50◦C in a water bath for 0, 10, 20, 30, or 60 min.

### Determination of Acid Resistance of C. sakazakii

The acid resistance of C. sakazakii cells was determined by subjecting them to simulated gastric fluid (SGF, pH = 3.5). The SGF consisted of 8.3 g/L proteose-peptone, 2.05 g/L NaCl, 0.6 g/L KH2PO4, 0.11 g/L CaCl2, 0.37 g/L KCl, 0.05 g/L oxgall, 1 g/L lysozyme, and 13.3 mg/L of pepsin. All compounds were dissolved in deionized water and autoclaved together except for the oxgall, lysozyme, and pepsin, which were filter-sterilized (0.25 µm). The final pH was adjusted with a sterile 5.0 N HCl solution.

All experimental samples were separately mixed with SGF solution. To simulate the environment of the human body, the method previously described by Anf et al. (2017) was employed, the samples were placed in a shaker (130 rpm) at 37◦C for different lengths of time (0, 10, 20, 40, or 60 min). For acid resistance assay, the bacterial concentration of initial sample was about 2.7 log CFU/ml.

### Determination of Bile Salt Resistance of C. sakazakii

The bile salt resistance of C. sakazakii strains was determined according to the method previously described by Fakruddin et al. (2014) with slight modifications. The C. sakazakii solutions were transferred to 5% (wt/vol) bile salt solution and placed in a shaker (45 rpm) at 37◦C for different lengths of time (0, 20, 60, 90, or 120 min). For bile salt resistance assay, the bacterial concentration of initial sample was about 2.7 log CFU/ml.

For each sampling time, viable cell populations were determined on three TSA plates by a spread plate method with serial dilutions made using PBS. All plates were incubated at 37◦C for 24 h before the colonies were counted. The results were expressed as percentage of survived cells (%). Percentage of survived cells (%) = N1/N0, where N<sup>1</sup> represented the CFU after heat, acid or bile salt resistance and N<sup>0</sup> represented the initial contamination of inoculated reconstituted infant formula.

### Statistical Analysis

Mean values and standard deviations were obtained from three replicate experiments with duplicated plating (n = 3). The effects of different treatments during the storage period were appraised by a one-factor analysis of variance (ANOVA), and means were separated at the 95% confidence level (the difference was considered significant if p < 0.05) using SPSS software (Ver 12.0K, SPSS Inc., Chicago, IL, United States).

### RESULTS

### Effects of Vacuum and Air Packaging on C. sakazakii Populations at 4, 10, and 25◦C

**Figure 1A** shows the effects of air and vacuum packaging on the survival of C. sakazakii exposed to 4◦C for a period of 10 days. The initial mean population of C. sakazakii on day 0 (before storage) at 4◦C in both samples was approximately 4.0 log CFU/ml. The viable population of C. sakazakii, regardless of being subjected to vacuum packaging or air packaging, decreased over the length of the exposure period. However, a more drastic reduction in the viable population was found for the vacuumpackaged cells than for the air-packaged cells. After 10 days of exposure to 4◦C, the air-packaged C. sakazakii showed a population reduction of 0.9 log CFU/ml compared with 1.6 log CFU/ml for the vacuum-packaged cells.

As shown in **Figure 1B**, the survival of both air- and vacuumpackaged C. sakazakii cells dropped rapidly during the initial 10 days of exposure to 10◦C. After 5 days of storage, airand vacuum-packaged C. sakazakii were each strongly reduced, by approximately 1.0 and 1.7 log CFU/ml, respectively. At the end of the storage period, the air- and vacuum-packaged C. sakazakii cells exhibited a population reduction of 2.4 and 3.1 log CFU/ml, respectively.

As shown in **Figure 1C**, the survival rates of both air- and vacuum-packaged C. sakazakii cells at 25◦C were similar to those observed at 4◦C. There was a sustained reduction in the population of air-packaged cells over the 10-day period, and the survival of vacuum-packaged C. sakazakii cells dropped rapidly during the initial 4 days of storage. After 10 days of storage, a smaller population reduction of 1.3 log CFU/ml was noted for the air-packaged C. sakazakii compared with that for the vacuumpackaged cells (1.6 log CFU/ml). Based on all these results, 10◦C was chosen as the experiment temperature and 4 days was chosen as the storage time for all subsequent experiments.

### Effects of Vacuum and Air Packaging on the Heat Resistance of C. sakazakii

After 4 days of storage at 10◦C, C. sakazakii cells were subsequently subjected to 50◦C to determine if air packaging and/or vacuum packaging influences the survivability of the pathogens at the regular temperature of PIF reconstitution. The

number of C. sakazakii decreased steadily within 60 min at 50◦C for each of the three different treatments (**Figure 2A**). Throughout the entire 60-min period at 50◦C, the desiccated cells that had not been subjected to storage generally showed a significantly (p < 0.05) higher survival rate compared with the stored air- or vacuum-packaged cells. The vacuum-packaged cells were significantly more susceptible to heat (50◦C) than were the air-packaged cells. After 10 min of the heat treatment, the

vacuum-packaged and air-packaged C. sakazakii cells exhibited the survival rates of 39.0 and 63.8%, respectively. At the end of the 60-min incubation period, the vacuum-packaged and air-packaged C. sakazakii cells exhibited the percentage of survived cells (%) of 1.0 and 4.9%, respectively.

### Effects of Vacuum and Air Packaging on the Tolerance of C. sakazakii to Simulated Gastric Fluid (SGF)

SGF (pH = 3.5) was prepared for use in studying whether the different packaging methods affect the survival of C. sakazakii after host consumption. As shown in **Figure 2B**, the survival rate decreased steadily during the entire incubation period for each

treatment type, but the survival rate of the vacuum-packaged C. sakazakii was significantly (p < 0.05) lower than that of the air-packaged cells at each measurement interval. At the end of the 60-min incubation period, the air-packaged cells showed a survival rate reduction of 39.9% compared with the larger survival rate reduction of 54.8% for the vacuum-packaged cells.

### Effects of Vacuum and Air Packaging on the Bile Salt Tolerance of C. sakazakii

To determine the effect of vacuum and air packaging on the bile salt tolerance of C. sakazakii, the stored, packaged cells were exposed to a bile salt challenge for 120 min. As shown in **Figure 2C**, the survival rate after 20 min of bile salt exposure of the desiccated C. sakazakii that had not been subjected to storage was significantly higher (p < 0.05) than those of the vacuumpackaged and air-packaged cells. The difference in the percentage of survived cells (%) between the vacuum-packaged and airpackaged cells became pronounced as the exposure period was lengthened. At the end of the 120-min bile salt exposure period, the air-packaged C. sakazakii exhibited a survival rate reduction of 59.0%, whereas a larger survival rate reduction of 73.3% was noted for the vacuum-packaged cells.

### DISCUSSION

Powdered infant formula is not a sterile product, and infantile infections of C. sakazakii are epidemiologically related to the consumption of contaminated, reconstituted PIF (Tall et al., 2017). C. sakazakii has a competitive advantage in dry environments due to its high tolerance to desiccation (Breeuwer et al., 2010), and the persistence of C. sakazakii in PIF during a 2.5-year period was reported (Barron and Forsythe, 2007). Some decontamination methods have been found by researchers to control the C. sakazakii in PIF, including increases or improvements in the traditional PIF processing method and reductions in the level of C. sakazakii through the use of plantderived compounds during or after PIF reconstitution (Ha and Kang, 2014; Pina-Pérez et al., 2015; Shi et al., 2016); however, the impact of packaging and storage processes, important links in the PIF industrial chain, on the survival of C. sakazakii has not been explored. In this study, desiccated C. sakazakii, the intrinsic contamination of PIF in actual production processes, was simulated to study the influence of vacuum and air packaging followed by storage at different temperatures on the survival and environmental stress resistance of C. sakazakii.

We selected three different temperatures at which to detect the variation of the bacteria population within 10 days of storage: (1) 4◦C was chosen to simulate the food freezing temperature, (2) 10◦C was chosen as the transportation temperature, and (3) 25◦C was chosen as a typical indoor temperature. At each of these three temperatures, both the tested packaging methods (vacuum and air) can lead to a decline in the amount of C. sakazakii in PIF (**Figure 1**). However, the vacuum packaging method produced a negative effect on bacteria survival more quickly compared with air packaging. For facultative anaerobes like C. sakazakii, it is generally recognized that a reduction in the oxygen level causes metabolic modification and a corresponding reduction in growth rate. Thus, the oxygen concentration in air is likely optimal for growth, and a reduction in the oxygen concentration to below 21% would presumably cause slower growth (Couvert et al., 2019). In agreement with our results, Duffy et al. (2001) reported that the packaging atmosphere has a noticeable effect on the growth rate of Listeria spp. (L. monocytogenes or L. innocua); Listeria on minced beef did not grow in vacuum-packed samples stored at 0 or 10◦C, whereas Listeria spp. grew in aerobically packed samples at 10◦C. The effect of vacuum packaging and air packaging on the survival of bacteria may depend on the state of the bacteria. Williams and Golden (2001) reported that for heat-injured L. monocytogenes, vacuum packaging at 4◦C is not conducive to the survival of bacteria as compared with air packaging, but for uninjured cells, vacuum packaging at 4◦C is more conducive to L. monocytogenes survival as compared with air packaging. The storage temperature may also affect the survival of bacteria under vacuum and air packing. Williams and Golden (2001) additionally found that for heat-injured L. monocytogenes, the survival of these bacteria at 20◦C in vacuum packing is higher than that in air packing.

Our results show that the survival of C. sakazakii decreased at all three temperatures tested in this study (4, 10, and 25◦C), but the survival rate decreased the fastest at 10◦C. Lin and Beuchat (2007) monitored the survival characteristics of E. sakazakii in infant rice cereal for a period of 12 months and found that the population of E. sakazakii in cereal stored at 21◦C decreased more quickly than that in cereal stored at 4◦C. We saw a similar trend in the present study, where the amount of bacteria at 4 and 25◦C decreased by 0.99 and 1.34 log CFU/ml, respectively, within 10 days. Also in agreement with our results, Al-Nabulsi et al. (2009) found that 21◦C was more optimal for the survival of desiccated C. sakazakii compared with 10◦C; the number of desiccated cells dropped from 4.35 to 4.15 log CFU/ml after 4 h at 21◦C, whereas it decreased from 4.35 to 2.46 log CFU/ml at 10◦C.

Trehalose, a non-reducing disaccharide of glucose, is assumed to play a pivotal role in the protection of Cronobacter spp. In dried stationary cells, the trehalose concentration was increased more than fivefold (Breeuwer et al., 2010). The accumulation of trehalose can protect proteins and cellular membranes from inactivation caused by a variety of stress conditions, including desiccation and cold (Elbein et al., 2003). Kandror et al. (2002) demonstrated that enzymes for trehalose synthesis are induced in Escherichia coli under cold-shock conditions and that the resulting accumulation of trehalose increases the cell viability when the temperature falls to near freezing. Therefore, we speculate that the accumulation of trehalose in desiccated C. sakazakii may have made the bacteria more likely to survive at 4◦C than at 10◦C.

C. sakazakii has a strong capacity to adapt to elevated osmotic pressure, low pH, heat, oxidation, desiccation, and bile salt (Alvarezordóñez et al., 2014; Fakruddin et al., 2014). These characteristics enhance the overall survival of C. sakazakii and increase the risk of contamination with these bacteria in dairy products. In the present study, the heat, acid, and bile salt resistance of C. sakazakii were determined to comprehensively evaluate the effect of different packaging methods on reducing

PIF contamination with C. sakazakii. Some recent studies have assessed the effects of processing or pre-processing methods on the stress tolerance of bacteria. Exposure to 405 nm LED illumination was found to significantly enhance the susceptibility of L. monocytogenes and Salmonella spp. to simulated gastric acid (Li et al., 2018). Trans-cinnamaldehyde, the principal component present in cinnamon oil, reduced the tolerance of C. sakazakii to environmental stresses, such as heat, desiccation, acid, and osmolarity (Amalaradjou and Venkitanarayanan, 2011). Our results show that the percentage of survival for desiccated bacteria (without 10 days of storage) and for stored air-packaged or vacuum-packaged cells decreased during exposure to heat, acid, or bile salts. The vacuum packaging had the greatest impact on reducing the stress resistance of C. sakazakii, whereas desiccation without 10-day storage had the least impact. Our previous study demonstrated that desiccation stress significantly decreased the heat resistance of C. sakazakii in reconstituted infant formula (Shi et al., 2017). On the contrary, Chen and Jiang (2017) found that desiccated Salmonella cells in poultry litter showed enhanced heat resistance as compared to non-desiccated cells. And rpoS gene was involved in the cross-protection of desiccated Salmonella against high temperatures. Yang et al. (2015) reported that desiccation decreased the heat and acid resistance of C. sakazakii compared with unstressed cells. They speculated that the desiccation condition could induce the microorganisms to be metabolically exhausted, making it difficult for the cells to tolerate adverse conditions (Beales, 2004). The decreased environmental stress resistance of vacuum-packaged cells might be affected by the accompanying reduction in oxygen concentration, which could affect the normal metabolism of C. sakazakii.

### CONCLUSION

In conclusion, this study has analyzed the impact of storage temperature (4, 10, and 25◦C) and packaging methods (vacuum packaging and air packaging) on the survival of C. sakazakii in PIF. While the impact of vacuum and air packaging at 10◦C on the environmental stress resistance (heat, acid, and bile salt) of C. sakazakii has already been demonstrated, this is the first record of the impact of air and vacuum packaging on the survival

### REFERENCES


and environmental stress resistance of desiccated C. sakazakii. Our results show that vacuum packaging significantly decreases the survival of C. sakazakii compared with air packaging, and the populations of air- and vacuum-packaged C. sakazakii stored at 10◦C for 10 days decrease more than do similar populations stored at 4 or 25◦C. Vacuum packaging significantly decreased the tolerance of C. sakazakii to heat, acid, and bile salt. Thus, our results suggest that the application of vacuum packing for PIF during shelf life could have a beneficial effect in minimizing the risk of C. sakazakii contamination in the reconstituted product. Vacuum packaging could be applied in the packaging step during manufacture of PIF or as a novel hurdle in food preservation in combination with other preservative technologies. This study provides a new perspective on choosing food packaging and processing methods by evaluating their impact on the environmental stress resistance of foodborne pathogens. However, further research is needed to evaluate the virulence properties of air-packaged and vacuum-packaged C. sakazakii and the ability of C. sakazakii to cause infection.

### AUTHOR CONTRIBUTIONS

YB, HY, and CS conceived and designed the experiments. DG and SF performed the experiments. DG and YB analyzed the data. HY and DG contributed to reagents, materials, and analysis tools. YB and CS wrote the manuscript.

### FUNDING

This work was supported by the Fundamental Research Funds for the Central Universities (2452017228), National Natural Science Foundation of China (31801659), and General Financial Grant from the China Postdoctoral Science Foundation (No. 2017M623256).

### ACKNOWLEDGMENTS

We thank Dr. Baowei Yang and Dr. Guoyun Zhang in Northwest A&F University for technical assistance.

chicken breast meat after exposure to sequential stresses. Int. J. Food Microbiol. 251, 15–23. doi: 10.1016/j.ijfoodmicro.2017.03.022


**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 Bai, Yu, Guo, Fei and Shi. 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-10-00303 February 18, 2019 Time: 15:57 # 7

# Probiotic Properties of Enterococcus Isolated From Artisanal Dairy Products

Yousef Nami<sup>1</sup> , Reza Vaseghi Bakhshayesh<sup>1</sup> , Hossein Mohammadzadeh Jalaly<sup>1</sup> , Hajie Lotfi<sup>1</sup> , Solat Eslami1,2 and Mohammad Amin Hejazi<sup>1</sup> \*

<sup>1</sup> Department of Food Biotechnology, Branch for Northwest and West Region, Agricultural Biotechnology Research Institute, Education and Extension Organization (AREEO), Tabriz, Iran, <sup>2</sup> Dietary Supplements and Probiotic Research Center, Alborz University of Medical Sciences, Karaj, Iran

The present study focused on probiotic characterization and safety evaluation of Enterococcus isolates from different artisanal dairy products. All the isolates exhibited inhibitory activity against several food spoilage bacteria and food-borne pathogens, including Shigella flexneri, Staphylococcus aureus, Listeria monocytogenes, Yersinia enterocolitica, Klebsiella pneumoniae, Escherichia coli, and Bacillus subtilis. The PCR results indicated the presence of at least one enterocin structural gene in all the tested strains. The Enterococcus isolates were further evaluated regarding their safety properties and functional features. The isolates were susceptible to vancomycin, gentamycin, and chloramphenicol. The results of PCR amplification revealed that all the tested isolates harbored none of the tested virulence genes except E. faecalis (ES9), which showed the presence of esp gene. The Enterococcus isolates showed cholesterol lowering properties. The selected isolates showed a high tolerance to low pH, and toward bile salts. They also demonstrated hydrophobicity activity, auto-aggregation, and adhesion ability to the human intestinal Caco-2 cell line. These properties may contribute the bacteria colonizing the gut. This study revealed that the Enterococcus isolates, especially E. durans ES11, ES20 and ES32, might be excellent candidates for production of functional foods to promote health benefits.

Keywords: Enterococcus, probiotic properties, dairy products, low cholesterol, antimicrobial activity, safety evaluation, Enterococcus as probiotics, virulence factors

### INTRODUCTION

Enterococci are belonging to genera of lactic acid bacteria (LAB). They are Gram-positive, catalase negative, cocci-shaped, facultative anaerobe, and non-spore forming bacteria (Haghshenas et al., 2016). Based on phylogenetic evidence and molecular studies (16S-rDNA sequencing or DNA– DNA hybridization), more than 26 species were classified in this genus. These microorganisms are ubiquitous bacteria which present as common microbiota in the intestine of humans, mammals, and other animals gastrointestinal tracts, but they are also exist in soil, water, vegetable products, meats, fermented and cooked meat and dairy products (Li et al., 2018; Zommiti et al., 2018). This is due to their high tolerance to harsh conditions such as high temperatures, low pH and high salinity. Significant role of E. faecium, E. faecalis, and E. durans in the ripening of traditional cheeses have

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Sergio Enrique Pasteris, Universidad Nacional de Tucumán, Argentina Konstantinos Papadimitriou, Agricultural University of Athens, Greece

> \*Correspondence: Mohammad Amin Hejazi aminhejazi@abrii.ac.ir; aminhejazi@yahoo.com

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 26 September 2018 Accepted: 04 February 2019 Published: 26 February 2019

#### Citation:

Nami Y, Vaseghi Bakhshayesh R, Mohammadzadeh Jalaly H, Lotfi H, Eslami S and Hejazi MA (2019) Probiotic Properties of Enterococcus Isolated From Artisanal Dairy Products. Front. Microbiol. 10:300. doi: 10.3389/fmicb.2019.00300

indicated that enterococci play an important role in the ripening of these cheeses, probably through proteolysis, lipolysis, and citrate breakdown, hence contributing to their typical taste and flavor. A majority of works specify that Enterococcus isolates play a vital role in the development of the sensory properties of fermented foods like olives (they break down oleuropein in fermented olives), sausages and cheese (Moreno et al., 2006).

There are different arrays of probiotics, mostly Lactobacillus and Bifidobacteria groups and Enterococcus genera in recent years, are used in functional foods (Haghshenas et al., 2017). The claimed advantageous of probiotic enterococci are: (i) diarrhea or diarrhea treatment in association with antibiotic medication, viral contaminations, chemotherapy and diseases originated from food-borne pathogens (Lau and Chamberlain, 2016); (ii) curbing the pathogenic bacteria growth (Zorriehzahra et al., 2016); (iii) anti-mutagenic and anti-carcinogenic features; (iv); increased intestinal mucosal barrier (Ahl et al., 2016); (v) stimulation of the immune system (Sheikhi et al., 2016); (vi) prevention of ulcers related to Helicobacter pylori infection (Oh et al., 2016) and (vii) cholesterol assimilation in food and human intestine (Kobyliak et al., 2016). These microorganisms have antagonistic activities through pathogens by different antimicrobial compounds production comprises bacteriocins, lactic and acetic acids and hydrogen peroxide.

However, due to association of some enterococci with human infections, like urinary tract infections, bloodstream infections, bacteraemia, endocarditis and diarrhea and surgical site infections; concerns about the safety of these bacteria have raised the attention of health organizations to use as probiotic bacteria because their virulence aspects contribute in human infections (Brandão et al., 2010; Zommiti et al., 2018). Generally, vancomycin-resistant enterococci (VRE) in nosocomial infections is considered as a major problem (Arias and Murray, 2012). Furthermore, the action of many virulence genes have been elucidated in Enterococcus isolates (Carlos et al., 2010). The most important virulence factors are cylA, cylB and cylM, esp, agg, gelE, cpd, ccf, and cad genes. The gene cylA is responsible for the cytosilin transportation and activation. The genes cylB and cylM have an application in modification of post-translational, while a cell wall protein concerned in the immune evasion is associated to esp gene. Adherence to eukaryotic cells is associated to an aggregation protein which is encoded by agg gene. gelE is responsible for the production of toxin which hydrolyzes gelatin, and finally sex pheromones which are responsible for facilitating conjugation are encoded by cpd, ccf, and cad genes (Belgacem et al., 2010; Liu et al., 2015). The aim of this study was to isolate and identify Enterococcus isolates from traditional dairy products, evaluation of their safety, probiotic aptitudes, and antimicrobial properties due to high potential roles of enterococci in health and food.

### MATERIALS AND METHODS

### Sampling and Culture Conditions

Artisanal dairy products (yogurt, cheese, and curd) were collected from domestic producers (**Table 1**). The samples were TABLE 1 | Origin, region of samples prepared, and acid and bile tolerance of isolates.


a–zMeans in the same color with different lowercase letters differed significantly (p < 0.05). ∗∗Means are significantly different (p < 0.01).

transported to the laboratory in ice boxes and stored at 4◦C. For better separation of bacteria from solid particles of yogurt and cheese, initial homogenization took place by vortexing. To prepare the bacterial suspension of yogurt, 10 g of yogurt was transferred to 100 mL of sterile physiological peptone water and shaken gently. To prepare the bacterial suspension of cheese and curd, 20 g of each sample were suspended in 180 mL of trisodium citrate sterile solution, and after half an hour, 10 mL of prepared solution were added to 200 mL de Man Rogosa Sharpe (MRS) broth in order to enrich and enhance the initial bacterial population in anaerobic conditions and incubated at 37◦C for 24 h (Haghshenas et al., 2017).

### Isolation of Enterococcus Isolates

Enterococcus isolates were isolated by the streak-plate method on MRS agar and incubated aerobically at 37◦C for 24 h. The single colonies were routinely checked for purity by microscopic examination. The pure colonies were used to characterize Gram staining and catalase test. The colonies which were Gram-positive and catalase-negative were selected and inoculated in MRS broth containing 30% glycerol as cryo-protectant and stored at −80◦C (El Soda et al., 2003). The purified cultures were activated by sub-culturing twice in MRS broth before use.

### Assessment of Probiotic Properties

### Acid and Bile Salts Tolerance

fmicb-10-00300 February 26, 2019 Time: 13:35 # 3

To determine acid tolerance, 10 mL of bacterial culture of each sample were incubated for 24 h in MRS broth. Selected colonies were transferred into mineral medium phosphate-buffered saline (PBS, pH 2.5). The samples were incubated aerobically for 3 h at 37◦C. Afterward, the cells were diluted up to 10 times using sterile saline (sodium chloride: 5.8 g/L) and each dilution of 100 µL for MRS agar surface in culture medium was cultured. The samples were incubated aerobically for 48–72 h at 37◦C (Haghshenas et al., 2015).

Tolerance to bile salts was analyzed based on the method used previously by Nami et al. (2015b). Briefly, MRS broth culture medium, as a control, and MRS with 0.3% bile oxgall, used as a test medium (treatments), were inoculated simultaneously with 1% of active bacterial culture at 37◦C for 4 h. Optical densities of the control and treated cultures growth were measured by a spectrophotometer (Eppendorf, Germany) at 600 nm. The percentage of growth suppression was measured by using the following formula:

$$\begin{aligned} & \text{\% of suppression} \\ &= \frac{\text{Growth in Control growth} - \text{Growth in blue truth}}{\text{Growth in control breadth}} \times 100 \end{aligned}$$

### Antimicrobial Activity and Bacteriocin Detection

Well diffusion method was performed to conclude and recognize the inhibitory metabolites produced by Enterococcus isolates (Nami et al., 2015a). Overnight cultures of the selected isolates were cultured in MRS agar at 37◦C for 24 h. Indicator bacteria used in this study were Shigella flexneri PTCC 1234, Staphylococcus aureus ATCC 25923, Listeria monocytogenes ATCC 13932, Yersinia enterocolitica ATCC 23715, Klebsiella pneumoniae PTCC 1053, Escherichia coli PTCC, 1276 and Bacillus subtilis ATCC 19652. These pathogenic organisms were purchased from the Persian Type Culture Collection (PTCC) to detect the antagonistic substances. Half McFarland indicator bacteria (1.5 × 10<sup>8</sup> CFU/mL) were poured on Mueller-Hinton agar and the wells were cut on plates. Then, each well was filled by 50 µL of filtered supernatant and plates incubated overnight at 37◦C and finally, the inhibition zone around the wells was measured by digital calipers.

The proteinaceous nature of the inhibition was assessed. To this end, the active cell-free culture supernatants were obtained by centrifugation at 15000 RPM for 12 min at 4◦C. They were subjected to various enzyme treatments, including catalase, trypsin, α-chymotrypsin, and proteinase K, at 1 mg/mL at 37◦C for 2 h, after adjusting the pH at 6.2 with 1 M of NaOH. Then, the residual activity was assessed against pathogenic microorganisms. The protease sensitivity was determined by the absence of inhibition zones around the wells. To confirm the presence of hydrogen peroxide, the active supernatants were subjected to sterilized catalase (1 mg/mL) and incubated at 37◦C for 2 h and finally their activities were assessed by the well diffusion method.

### PCR Amplification of Known Enterocin Genes

All the structural genes concerned to the expression of wellknown enterocins EntA, EntB, EntP, EntL50A, EntL50B, Ent31 (Toit et al., 2000), EntQ (Belgacem et al., 2010), and Ent1071 (Omar et al., 2004) were amplified with specific PCR primers (**Table 2**). PCR amplification was performed at a final volume of 50 µL that comprised of 1 Taq polymerase buffer, 200 µM of dNTP's, 25 pM of each primer, 2 µL of template DNA (stock) and 1 U of Taq DNA polymerase (Thermo Fisher Scientific, United States). The PCR products were visualized by electrophoresis on 2% agarose gels.

### Exopolysaccharide (EPS) Production

The method used by Fguiri et al. (2016) was used for assessment of EPS production ability of isolates. Briefly, the cultures were streaked on m-MRS agar medium which was modified by replacing glucose with 100 g/L of sucrose and incubated at 37◦C for 24 h aerobically. Metal loop was used to drag up formed colonies. If the length of slime was above 1.5 mm, the isolate was considered positive slimy producers.

### Cell Surface Hydrophobicity

The adhesion ability of isolates to xylene was determined as previously described by Mishra and Prasad (2005).

### Auto-Aggregation and Co-Aggregation

The ability of the isolates to auto-aggregate was performed according to the method described by Angmo et al. (2016). Auto-aggregation percentage was determined using the following equation:

$$1 - \left(\text{At}/\text{A0}\right) \times 100$$

Where A0 represents absorbance at t = 0 and at represents absorbance at time t.

Co-aggregation of Enterococcus isolates against the seven pathogens was performed at 37◦C after 4 h of incubation based on method used by Zuo et al. (2016). Co-aggregation percentage was calculated based on equation:

$$\% = \frac{\text{A0} - \text{At}}{\text{At}} \times 100$$

### Adhesion Ability to Human Intestinal Cells

Adhesion ability to human colon carcinoma cells (Caco-2) was evaluated as reported previously by Nami et al. (2014). Briefly, the Roswell Park Memorial Institute (RPMI-1640; Sigma) medium, supplemented with 10% heat-inactivated fetal bovine serum, was used to culture the human cells. The cells were cultured on 24-well tissue culture plates and incubated at 37◦C in 5% CO<sup>2</sup> in a relatively humid atmosphere until a confluent monolayer was achieved. The viable Caco-2 cells were counted in a Burker haematocytometer chamber. Then, the cell suspension including

#### TABLE 2 | Primers used for PCR amplification of virulence factors and enterocin detection genes in Enterococcus strains.


Ta (◦C), Annealing temperature; bp, base pairs.

bacteria and Caco-2 cells was subjected to pure plate technique to determine C.F.U. bacteria adhesion was expressed as the total number of bacteria attached to viable Caco-2 cells.

### Cholesterol Assimilation

Cholesterol removal percentage was determined by o-phthalaldehyde method described by Rudel and Morris (1973) with some alteration. A freshly prepared MRS broth was supplemented with 0.3% oxgall (Merk Germany) as bile salt and water-soluble cholesterol (150 µg/mL) was added as the cholesterol source (sterilized by 0.2 µL filter), the mixture inoculated with each isolate at 1% level and incubated anaerobically at 37◦C for 20 h. The cells were removed by centrifugation (10000 rpm for 15 min) after the incubation period; subsequently, 1 mL of the cell-free broth was mixed with 1 mL KOH (33% W/V) and 2 mL ethanol 96%, vortexed for 2 min, followed by heating at 60◦C for 15 min. Mixes cooled in room temperature, 2 mL distilled water and 3 mL hexane were added and vortexed for 1 min. One mL the hexane layer was transferred into a glass tube and evaporated in water bath at 80◦C. The residue was immediately dissolved in 2 mL o-phthalaldehyde (Merck, Germany) reagent, Followed by 0.5 mL concentrated sulphuric acid and vortexed completely for 1 min. The samples were incubated at room temperature for 30 min and finally absorbance was read at 550 nm.

### β-Galactosidase Activity

β-galactosidase activity of Enterococcus isolates was assessed according to Angmo et al. (2016). Bacterial cultures were streaked on MRS agar plates containing 60 µL X-gal (5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside) and 10 µL of IPTG (isopropyl-thio-β-D-galactopyranoside) solution as inducer. The presence of β-galactosidase activity in strains was determined after 42 h of incubation at 37◦C.

### Safety Assessment Hemolytic Activity

Haemolytic activity of Enterococcus isolates was examined by culturing of fresh overnight cultures on Columbia agar plates (Oxoid) containing 7% (v/v) sheep blood (Oxoid) and incubated for 48 h at 37◦C. Finally, hemolytic activities were detected by 3 categories: appearance of a halo around the colony: greenish zone considered as α-hemolysis, clear zone for β-hemolysis and no halo for γ-haemolysis (Abedi et al., 2018).

### Bile Salts Hydrolysis

The bile salt hydrolysis was evaluated according to Argyri et al. (2013). The hydrolysis activity was indicated by 0.5% (w/v) taurodeoxycholic acid after 48 h of incubation at 37◦C. The hydrolysis activity was evaluated by the formation of opaque agar halos around the colonies.

### Detection of Virulence Factors

PCR amplification was performed to detect genes encoding potential virulence factors. Total bacterial DNA was isolated by the method described by Leenhouts et al. (1990). Virulence genes evaluated in this study were cylA, clyB, esp, gelE, asa1, Ace, efaAfs, and cpd. E. faecium ATCC 51299 was used as a control. PCR amplification was performed to detect genes encoding these factors using several primers (**Table 2**). The PCR amplification was carried out in 0.2 mL reaction tubes each with 25 µL of mixtures using 0.1 mM of deoxynucleoside triphosphates, 2.5 mM of MgCl2, 0.5 mM of each primer, PCR Buffer (1X), 2 U of Taq polymerase and 50 ng/µL of DNA template. PCR amplifications were performed with a cycle of initial denaturation (94◦C for 5 min), followed by 32 cycles of denaturation (94◦C for 60 s), annealing at an appropriate temperature (**Table 2**) for 60 s and elongation (72◦C for 5 min). The PCR products were analyzed by gel electrophoresis in 1.5% agarose stained with ethidium bromide (0.5 g/mL).

### Antibiotic Susceptibility

fmicb-10-00300 February 26, 2019 Time: 13:35 # 5

Antibiotic susceptibility test was carried out based on previously described method of Haghshenas et al. (2014). Briefly, disk diffusion assay was performed to determine antibiotic sensitivity of the isolates. MRS agar medium was used for this test. Antimicrobial disks (5 mm) were purchased from Padtan Teb Co, Iran. The tested antibiotics were vancomycin (30 µg), colistin (10 µg), streptomycin (10 µg), cefepime (30 µg), cefixime (15 µg), sulfamethoxazole (2 µg), kanamycin (30 µg), ciprofloxacin (5 µg), tetracycline (30 µg), erythromycin (15 µg), ampicillin (10 µg), gentamycin (10 µg), clindamycin (30 µg), ceftriaxon (30 µg), chloramphenicol (30 µg), and cefalexin (30 µg), using disk diffusion method. After overnight incubation at 37◦C, the diameter of inhibition (mm) around each disk was measured.

### 16S-rDNA Gene Sequencing

Amplification of 16S-rDNA gene (1500 bp) of the isolate was performed by using a pair of lactic acid bacteria (LAB)-specific universal primers (Hal6F/Hal6R) (F: 5<sup>0</sup> -AGAGTTTGATCMT GGCTCAG-3<sup>0</sup> and R: 5<sup>0</sup> -TACCTTGTTAGGACTTCACC-3<sup>0</sup> ) previously described by Haghshenas et al. (2016). The PCR amplification was fulfilled for the total volume 50 µL under the following conditions: initial denaturation at 94◦C for 5 min, followed by 35 cycles of denaturation at 94◦C for 45 s, annealing at 59◦C for 60 s, extension at 72◦C for 90 s, and a final extension step at 72◦C for 10 min. The PCR products were visualized through 1% (w/v) agarose gel (Sigma Chemical Co., Poole, United Kingdom) electrophoresis and stained via ethidium bromide. The PCR products were sent to the Macrogen DNA Sequencing Service (Korea) to be sequenced. Multiple sequence alignment of 16S rRNA genes was carried out using CLUSTAL W function (Thompson et al., 1994) with default parameters, and a phylogenetic tree of 16S rRNA genes was reconstructed using the neighbor-joining method (Saitou and Nei, 1987) implemented in the MEGA X software (Kumar et al., 2018) with p-distance parameter distance. Bootstrap values were calculated with 1,000 re-samples. The 16S rRNA gene sequence of Lactobacillus acidophilus (LC064893.1) was used as out-group for the analysis.

### Statistical Analysis

Statistical analysis of data was carried out using SPSS (Ver. 19.0 SPSS, Chicago, IL, United States). The comparisons of differences between the means of the treatments were analyzed by one-way ANOVA at a significance level of P < 0.05. All the experiments were performed in triplicate and data expressed as means ± standard deviations.

### RESULTS

### Characterization and Identification of Isolates

From 90 isolates from various dairy products of different regions of East Azerbaijan Province in Iran, 66 resulted isolates were gram-positive, catalase-negative and rod- or cocci-shaped bacteria. Among them, 32 isolates were cocci-shaped and exhibited optimum growth at 37◦C but not at 30◦C (data not shown). From 32 cocci-shaped isolates, 8 resulted isolates were from cheese, 8 from curd, and 16 from yogurt (**Table 1**). Based on morphological and biochemical assays, the authors assumed that these isolates are likely to be Enterococcus strains.

### Acid and Bile Salt Tolerance

A stimulated in vitro gastric juice (pH 2.5) was used to assess the acid tolerance profile of Enterococcus isolates (**Table 1**). The survival rate of isolates displayed a significant variability, ranging from 10.3 to 81.6%. The highest survival rate was observed for isolate ES20 (81.6 ± 0.3%), followed by isolates ES32 (77.4 ± 0.3%) and ES11 (76.3 ± 0.2%), while the lowest survival rate was observed for isolate ES16 with 10.3 ± 0.1% survival rate.

The percentage of viability of Enterococcus isolates was assessed after 4 h of incubation in M-17 broth supplemented with 0.3% oxgall (**Table 1**). The viability rate of isolates exhibited a significant variability ranging from 09.1 ± 0.1 to 79.8 ± 0.3%. The highest viability rate belonged to isolate ES11 (79.8 ± 0.3%), followed by isolates ES20 (79.1 ± 0.2%) and ES32 (78.2 ± 0.3%), while the lowest viability rate belonged to isolate ES7 (09.1 ± 0.1%). Among the 32 isolates, only isolates 4, 9, 11, 20, 27, 28, and 32 exhibited more than 50% acid and bile tolerance. Hence, these seven isolates were selected for further investigation.

### Antimicrobial Activity and Detection of Enterocin Genes

The inhibitory effect of isolated Enterococcus isolates against some important pathogenic microorganisms is shown in **Table 3**. The results showed that all the seven selected isolates were capable of inhibiting the growth of the majority of target pathogens. The target pathogens used in this study were Shigella flexneri, Staphylococcus aureus, Listeria monocytogenes, Yersinia enterocolitica, Klebsiella pneumoniae, Escherichia coli, and Bacillus subtilis (**Table 4**). Isolates ES20 and ES28 were able to inhibit the growth of all the tested pathogens. Moreover, isolates ES9, ES11, ES27, and ES32 were able to inhibit the growth of all the target pathogens except Shigella flexneri.

When pH adjusted to 6.2, the isolates ES4, ES9, ES11, and ES27 could not inhibit the growth of any pathogens. Also, the isolates ES20 and ES32 do not able to inhibit the growth of Shigella flexneri, Listeria monocytogenes, Klebsiella pneumoniae, and Bacillus subtilis. So, it resulted that the nature of inhibition of these isolates are because of acid production. Furthermore, after subjecting other isolates to catalase enzyme, none isolates were able to inhibit the growth of indicator pathogens, except isolates ES28 and ES32 against Staphylococcus aureus and Yersinia enterocolitica, respectively. To confirm the nature of inhibition of isolate ES28 against Staphylococcus aureus, and isolate ES32 against Yersinia enterocolitica, protease K enzyme was subjected. After applying this enzyme, the clear zone of inhibition was removed and it showed that the nature of inhibition is because of bacteriocin production.

TABLE 3 | The inhibitory effect of selected Enterococcus strains against pathogenic microorganisms.


#### TABLE 3 | Continued


Values are mean ± standard error of triplicates. (Strong ≥ 20 mm), (Moderate < 20 mm > 10 mm), and (Weak ≤ 10 mm).

Extracted DNA of Enterococcus isolates was subjected to PCR amplification to determine the existence of structural genes coding EntA, EntB, EntP, EntL50A, EntL50B, and Ent31 enterocins (**Table 3**). The PCR results indicated the presence of at least one enterocin structural genes in all the 7 isolates. The enterocins A and B structural genes were detected among all the isolates and the enterocins P, Q and L50A were found in 3 isolates. On the other hand, none of the evaluated isolates showed PCR amplification fragments for other tested enterocins (entL50B, ent1071, and bac31).

### Cell Surface Hydrophobicity

The cell surface hydrophobicity rate is illustrated in **Figure 1**. It ranged from 23.3 ± 1.6 to 58.6 ± 2.3%. The highest cell hydrophobicity rate was observed for isolates ES20, followed by ES11 and ES32 with 58.6 ± 2.3%, 54.2 ± 1.9% and 51.8 ± 1.4, respectively. Furthermore, isolates ES4 and ES27

TABLE 4 | The origin of indicator pathogenic bacteria used in this study.


showed the lowest cell hydrophobicity rates with 23.3 ± 1.6 and 24.7 ± 1.3%, respectively.

### Adhesion Capacity to Intestinal Cells

The adhesion capacity to human colon carcinoma cell line, Caco-2, was determined (**Figure 1**). Adhesion capacity to Caco-2 cells varied significantly among the tested bacteria, with adhesion ratio ranging from 22.1 ± 1.8 to 74.1 ± 1.9%. The highest adherence capacity belonged to isolates ES20, ES32, and ES11, with mean values of 74.1 ± 1.9, 66.4 ± 2.2, and 63.7 ± 1.8%, respectively.

### EPS Production Ability

The ability of isolates to produce EPS is illustrated in **Table 5**. The results demonstrated that all the isolates exhibited EPS production ability.

### Auto-aggregation and Co-aggregation

The results of cell auto-aggregation assay are shown in **Figure 1**. The cell auto-aggregation rates of the isolates ranged from 24.7 ± 2.3 to 81.2 ± 2.6%. The highest scores were obtained for isolates ES20, followed by ES32, and ES11 with 81.2 ± 2.6, 69.2 ± 2.1, and 67.9 ± 1.2%, respectively. Furthermore, isolate ES4 showed the lowest auto-aggregation rates with 24.7 ± 2.3%.

The results of co-aggregation of Enterococcus isolates in the presence of Staphylococcus aureus, Escherichia coli, Listeria monocytogenes, Shigella flexneri, Klebsiella pneumoniae, Yersinia enterocolitica, and Bacillus subtilis separately at 37◦C at 2 and 4 h of incubation are shown in **Table 6**. The results showed that isolates 11, 20, and 32 exhibited higher co-aggregation ability

differed significantly (p < 0.05).

TABLE 5 | Origin, Molecular identification, average cholesterol-removal ratio, BSH activity, EPS production, hemolytic activity and β-galactosidase activity of strains after 20 h of growth at 37◦C.


<sup>∗</sup>BSH activity was expressed based on the diameters of precipitation zones: –, no precipitation; +, >10 mm; ++, >15 mm; and +++, >20 mm. ∗∗ (−) no haemolysis; (α and β) haemolysis. ∗∗∗ (–) no EPS production; (+) EPS production.

compared to other isolates. The co-aggregation percentages increased (P < 0.05) during incubation. Co-aggregations of Enterococcus isolates with all the pathogens at 4 h of incubation were higher (P < 0.05) compared to 2 h of incubation. Isolates demonstrated lower co-aggregation (P < 0.05) toward grampositive pathogens (S. aureus, L. monocytogenes, and B. subtilis) compared to gram-negative ones (Y. enterocolitica, Sh. Flexneri, K. pneumoniae, and Escherichia coli).

### Cholesterol Assimilation and Bile Salt Hydrolysis

**Table 5** presents the levels of cholesterol assimilation by isolates in the presence of 0.3% bile oxgall at 37◦C for 20 h. The content of cholesterol removed varied (P < 0.05) and ranged from 99.93 to 216.45 µg/mL. The highest content of cholesterol assimilation was observed in isolates ES32, ES20, and ES11, which belonged to Enterococcus durans species with 216.45, 175.38, and 172.23 µg/mL. In contrast, the lowest content belonged to isolate ES28.

The bile salt hydrolysis of the Enterococcus isolates is shown in **Table 5**. The results indicated that isolates ES11, ES20, and ES32 showed the highest BSH activity (+++), whereas isolate ES27 exhibited moderate BSH activity (++). Furthermore, isolate ES4 showed less BSH activity (+), while isolates ES9 and ES28 showed no activity (−).

### Hemolytic Activity

Hemolytic activity of isolates is represented in **Table 5**. All the isolates showed no β-hemolytic activity.

### β-Galactosidase Activity

Isolates ES11, ES20, ES27, and ES32 indicated the presence of β-galactosidase activity, while isolates ES4, ES7, and ES28 did not show the presence of this enzyme.



Values are mean ± standard error of triplicates. <sup>a</sup>−iMeans in the same column with different lowercase letters differed significantly (p < 0.05).

TABLE 7 | Antibiotic susceptibility of strains.


V, vancomycin; CL, colistin; S, streptomycin; FEB, cefepime; CFM, cefixime; SXT, sulfamethoxazole; K, kanamycin; CP, ciprofloxacin; TE, tetracycline; E, erythromycin; AM, ampicillin; GM, gentamycin; CC, clindamycin; CRO, ceftriaxon; C, chloramphenicol; CN, cefalexin and CL, colistin. Erythromycin results based on R ≤ 13 mm; I: 13–23 mm; S ≥ 23 mm. Gentamycin results based on R ≤ 6 mm; I: 7–9 mm; S ≥ 10 mm. Vancomycin results based on R ≤ 12 mm; I: 12–13 mm; S ≥ 13 mm. I: intermediate (zone diameter, 12.5–17.4 mm); R: resistant (zone diameter, ≤12.4 mm); S: susceptible (zone diameter, ≥17.5).

### Detection of Virulence Factors

The presence of genes encoding eight known virulence factors in the Enterococcus isolates was assessed. The results of PCR amplification revealed that none of the isolates harbored any virulence factors except E. faecalis (ES9), which showed the presence of esp gene.

### Antibiotic Susceptibility

**Table 7** illustrates the antibiotic resistance of Enterococcus isolates against 16 tested antibiotics. Overall, all the isolates showed the ability to resist the impact of tetracycline and colistin, whilst all the isolates were susceptible to gentamycin, vancomycin and chloramphenicol. The impact of other antibiotics against isolates varied from susceptible to resistant. Among the tested Enterococcus isolates, isolate ES11 was susceptible to all the antibiotics, except for tetracycline and colistin.

### 16S-rDNA Sequencing

16S-rDNA sequencing was performed as molecular phylogeny analysis to identify selected Enterococcus isolates at the species level. Phylogenetic tree (**Figure 2**) was constructed based on the 16S-rDNA sequences from evolutionary distances by the neighbor-joining method. Analysis of the sequences depicted that isolate ES9 clustered with sequences of Enterococcus faecalis, isolates ES4 and ES27 clustered with sequences of Enterococcus faecium, isolate ES28 clustered with Enterococcus hirae and three isolates ES11, ES20 and ES32 clustered with Enterococcus durans.

### DISCUSSION

The ability of isolates to survive under high acidic conditions and to show acceptable tolerance against bile salts in the human intestine are two key properties for a candidate to be considered a probiotic (Kandylis et al., 2016; Ayyash et al., 2018). In this study, the survival rate of isolates in acidic conditions and bile salts displayed a significant variability, which might be due to the fact that mechanisms of acid and bile tolerance are species and strain-dependent. Isolates ES4, ES9, ES11, ES20, ES27, ES28, and ES32 showed favorable acid and bile tolerance compared to the other isolates. Therefore, only these seven isolates were

subjected to further tests. The acid and bile tolerances of the seven isolates are consistent with the results reported by Nami et al. (2014); El-Jeni et al. (2015), Haghshenas et al. (2016), and Ayyash et al. (2018).

Enterococcus and Lactobacillus species which were submitted in NCBI database as complete sequence.

The nature of inhibitory effect of isolates was assessed by adjusting pH to 6.2 and also using catalase and protease enzymes. After treating with protease enzymes, the clear zones around the halos were disappeared. It could be because of proteinaceous nature of secreted metabolites by isolates. It has been shown by some studies (Balla et al., 2000; Cintas et al., 2000) that bacteriocins secreted by Enterococcus isolates are strong inhibitors of food-borne pathogens such as S. aureus, L. monocytogenes and Clostridium tyrobutyricum. In our study, the inhibitory profile of the Enterococcus isolates under assessment tended to be active against a wide range of grampositive and gram-negative bacteria and food-borne pathogens, including Staphylococcus, Listeria, Yersinia, Bacillus, Shigella, Escherichia coli and Klebsiella. These recorded activities are in contrast with Stevens et al. (1991) and Zommiti et al. (2018), who theorized that bacteriocins of LAB are ineffective against gram-negative bacteria because the outer membrane blocks the bacteriocin target. Moreover, PCR amplification of genes coding for enterocins (EntA, EntB, EntQ, EntP, EntL50A, EntL50B, Ent1071, and Bac31) was investigated. All the isolates contained at least one enterocin gene and the enterocins A and B were detected in all the strains. This is consistent with the results reported by Cintas et al. (2000) and Belgacem et al. (2010), who detected these putative enterocin factors in Enterococcus isolates. Isolates ES11, ES20, and ES32, which showed the ability to compete against all the seven tested pathogens, contained a combination of four enterocins such as entA, entB, entQ and entL50A. This is consistent with Sánchez et al. (2007), who proposed the co-production of two or more enterocins by a strain generating supernatants with a higher antagonistic activity.

Probiotic capacity to remain alive in the gastrointestinal tract is one of the most desirable features of probiotics. To be colonized in the intestine, probiotics have to adhere to the intestinal mucosa to avoid being removed from the colon by peristalsis.

fmicb-10-00300 February 26, 2019 Time: 13:35 # 9

In this study, strains ES20, ES32, and ES11 exhibited favorable adherence capacity (Duary et al., 2011; Kumar et al., 2015). Similar to these results, high capability to adhere to Caco-2 cells was reported for Enterococcus isolates (Cebrián et al., 2012; Pimentel et al., 2012).

The ability of isolates to produce EPS was determined by the presence of ropy white mucus on skimmed milk plates containing ruthenium red. It has been shown that LAB is able to produce EPS, which improves the viscosity and texture of dairy products. Hence, EPS-producing LAB is widely used in the dairy industry. The presence of (glyco-) proteinaceous on the cell surface results in higher hydrophobicity, while the presence of polysaccharides leads to hydrophilic surfaces (Osmanagaoglu et al., 2010).

Hydrophobicity is one of the indicative parameters for cell surface properties of probiotics, which correlates with the adhesion ability of probiotics to epithelial cells (Duary et al., 2011; Zuo et al., 2016; Ayyash et al., 2018). Thus, the higher hydrophobicity resulted in higher ability of probiotics to attach to epithelial cells and promote health benefits. In the current study, isolates ES20, ES11, and ES32 exhibited better hydrophobicity percentages compared with the results reported by Ayyash et al. (2018). Moreover, Das et al. (2016) also reported that hydrophobicity of three LAB ranged from 22.2 to 25.0%, which is lower as compared with our findings.

The auto-aggregation and co-aggregation ability are two important properties of probiotics, which are defined as the bacterial accumulation of the same species and of different species, respectively (Campana et al., 2017). The autoaggregation and co-aggregation are fundamental for probiotics because it seems that auto-aggregation is correlated with adherence to epithelial cells (Collado et al., 2008), while coaggregation represents a defensive barrier for the colonization of pathogenic microorganisms (Kos et al., 2003; Abushelaibi et al., 2017). In addition, the bacterial equilibrium in the gastrointestinal tract is increased by aggregation of probiotics in the human gut (Tulumoglu et al., 2013) and the probiotic properties of the LAB are improved by their co-aggregation ability in the presence of gut pathogens. The formation of a defensive barrier because of co-aggregation of LAB in the presence of pathogens will not allow pathogens to colonize in the human gut (Vidhyasagar and Jeevaratnam, 2013). Tareb et al. (2013) reported that the ability of LAB isolates to co-aggregate with pathogens could be attributed to proteinaceous components present on the cell surface and interactions between carbohydrate and lectin. Nevertheless, Collado et al. (2008) revealed that the co-aggregation ability of LAB is time- and strain- dependent. Our results correspond with those of Angmo et al. (2016); Taheur et al. (2016), and Abushelaibi et al. (2017). Our study showed that the co-aggregation ability is significantly affected by incubation time and strain.

In vitro studies on cholesterol reduction by Enterococcus species have been considered as an important parameter for the selection of probiotic strains with diverse health-promoting benefits. The hypocholesterolemic effect on host is another important but not essential property of probiotics. Several mechanisms have been postulated for lowering cholesterol by probiotic bacteria (Miremadi et al., 2014), including conversion of cholesterol to coprostanol by reductase, cholesterol incorporation in the cell wall and disruption of cholesterol micelle in the intestine by deconjugated bile salts. Cholesterol removal results in the current study are in agreement with the results of studies by Ayyash et al. (2018). In the current study, the cholesterol removal could be attributed mainly to cholesterol micelle in the intestine by deconjugated bile salts.

Presently the role of BSH is controversial because it might act either positively in lowering of serum cholesterol or negatively in increasing the level of undesirable deconjugated bile salts (Xie et al., 2015; Fadda et al., 2017). On the other hand, BSH activity by probiotic bacteria might be desirable because it increases the intestinal survival and persistence of producing strains, which in turn increases the overall beneficial effects associated with the strain (Begley et al., 2006; Park et al., 2007).

The safety assessment is obligatory before a strain is qualified as beneficial for the host health for use in food industry. The absence of hemolytic activity and antibiotic resistance are the basic requirements for the selection of safe probiotic strains (Oh and Jung, 2015). No β-hemolytic activity was detected in the 7 tested strains.

The presence of β-galactosidase, which is a useful enzyme that hydrolyses lactose into glucose and galactose was performed. Lactose intolerance is due to the lack or shortage of this enzyme; hence lactose mal-digestion symptoms could be improved by consumption of probiotics that release β-galactosidase (Hussain et al., 2008). On the other hand, the products fermented with β-galactosidase producers play an essential role in the treatment of lactose intolerance (Vasiljevic and Jelen, 2001). Our findings proved that the tested enterococci isolates showed good β-galactosidase activity. Isolates ES32, ES20 and ES11 showed the highest β-galactosidase activity. Hence, these isolates may have an effective application in the dairy industry and also for the treatment of lactose intolerance.

A desirable characteristic for enterococcal bacteria used in food industry is the absence of cytolysin-encoding genes. Cytolysin is a bacterial toxin expressed by some E. faecalis isolates. In our study, only three isolates showed the presence of cytolysin-encoding genes, which belonged to E. faecalis isolates. None of the isolates belonging to E. durans, E. hirae and E. faecium showed the presence of these genes. In addition, only five isolates belonging to E. faecalis showed the presence of esp gene. The absence of esp in E. faecium was reported (Eaton and Gasson, 2001), which revealed the frequent presence of the esp gene in medical E. faecium isolates. Our results are in accordance with the results of Eaton and Gasson (2001) and Liu et al. (2015). Overall, the presence of virulence genes is higher in E. faecalis species than in E. faecium species, which is consistent with our results.

The resistance of Enterococcus species to various antibiotics has been reported by some studies (Vidhyasagar and Jeevaratnam, 2013). Because of transferring the resistance factors from probiotics to pathogenic microorganisms via the interchange of genetic materials, the intake of antibiotic-resistant strains disrupts the original flora in the intestine (Mathur and Singh, 2005; Hummel et al., 2007; Taheur et al., 2016). In the current study, all the isolates were resistant to colistin and tetracycline. This can be attributed to the overuse of these antibiotics in rural areas.

### CONCLUSION

fmicb-10-00300 February 26, 2019 Time: 13:35 # 11

The results of this study indicated that E. durans ES11, E. durans ES20 and E. durans ES32 are safe probiotic strains with the potential to assimilate total cholesterol. These strains fulfilled several criteria to be used as probiotic microorganisms, including auto- and co-aggregation ability, resistance to low pH and high bile salts, adherence to hydrocarbons, susceptibility to some antibiotics as well as EPS production. As these isolates are from the food sources which display a wide spectrum of capability against certain intestinal and food-borne pathogens, it could be used in functional foods since these probiotic strains adapt to the conditions and could provide protection against pathogens. Further studies will be required to determine the mechanisms underlying the cholesterol-lowering

### REFERENCES


effect and to evaluate the long-term probiotic potential of these strains.

### AUTHOR CONTRIBUTIONS

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

### FUNDING

This work was supported by the Agricultural Biotechnology Research Institute of Iran (ABRII) (Grant No. 3-05-0551-88020).

### ACKNOWLEDGMENTS

The financial support of the Agricultural Biotechnology Research Institute of Iran (ABRII) is gratefully acknowledged.

genes in Enterococcus spp. of human and animal origin isolated in Portugal. Arch. Microbiol. 192, 927–936. doi: 10.1007/s00203-010-0619-z




safety of Enterococcus faecium isolated from artisanal Tunisian meat "Dried Ossban". Front. Microbiol. 9:1685. doi: 10.3389/fmicb.2018. 01685


**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 Nami, Vaseghi Bakhshayesh, Mohammadzadeh Jalaly, Lotfi, Eslami and Hejazi. 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-10-00503 March 12, 2019 Time: 19:1 # 1

# Occurrence and Characterization of Methicillin Resistant Staphylococcus aureus in Processed Raw Foods and Ready-to-Eat Foods in an Urban Setting of a Developing Country

Mohammad Aminul Islam<sup>1</sup> \* † , Sahana Parveen2,3, Mahdia Rahman<sup>1</sup> , Mohsina Huq<sup>1</sup>† , Ashikun Nabi<sup>1</sup>† , Zahed Uddin Mahmood Khan<sup>3</sup> , Niyaz Ahmed<sup>4</sup> and Jaap A. Wagenaar5,6

<sup>1</sup> Food Microbiology Laboratory, Laboratory Sciences and Services Division, International Centre for Diarrhoeal Disease Research, Bangladesh, Dhaka, Bangladesh, <sup>2</sup> Institute of Food Science and Technology, Bangladesh Council of Scientific and Industrial Research, Dhaka, Bangladesh, <sup>3</sup> Department of Botany, Jahangirnagar University, Dhaka, Bangladesh, <sup>4</sup> Laboratory Sciences and Services Division, International Centre for Diarrhoeal Disease Research, Bangladesh, Dhaka, Bangladesh, <sup>5</sup> Department of Infectious Diseases and Immunology, Utrecht University, Utrecht, Netherlands, <sup>6</sup> Wageningen Bioveterinary Research, Lelystad, Netherlands

Infections by methicillin-resistant Staphylococcus aureus (MRSA) are gradually increasing in the community. In this study, we investigated a total of 162 food samples including 112 ready-to-eat (RTE) foods and 40 processed raw meat and fish samples collected from retail vendors in Dhaka, Bangladesh and determined the occurrence of toxigenic S. aureus and MRSA. Around 22% of samples were positive for S. aureus, RTE foods being more positive (23%) than the processed raw meat/fish samples (18%). Among 35 S. aureus isolates, 74% were resistant to erythromycin, 49% to ciprofloxacin and around 30% to oxacillin and cefoxitin. Around 37% of isolates were resistant to ≥3 classes of antibiotics and 26% of isolates (n = 9) were identified as MRSA. Majority of the isolates were positive for enterotoxin genes (74%), followed by pvl gene (71%), toxic shock syndrome toxin (tsst) gene (17%) and exfoliative toxin genes (11%). Multi locus sequence typing (MLST) of 9 MRSA isolates identified four different types such as ST80 (n = 3), ST6 (n = 2), ST239 (n = 2) and ST361 (n = 2). spa typing of MRSA isolates revealed seven different types including t1198 (n = 2), t315 (n = 2), t037 (n = 1), t275 (n = 1), t304 (n = 1), t8731 (n = 1) and t10546 (n = 1). To our knowledge, this is the first report entailing baseline data on the occurrence of MRSA in RTE foods in Dhaka highlighting a potential public health risk to street food consumers.

Keywords: methicillin resistant S. aureus, raw meat, ready-to-eat foods, MLST, spa typing

### INTRODUCTION

Staphylococcus aureus (SA) is present in up to 80% of healthy individuals as a commensal, yet it is one of the most common causes of skin and soft tissue infections sometimes leading to complicated infections, such as necrotizing pneumonia, septic arthritis, endocarditis, and osteomyelitis (Popovich and Hota, 2008; David and Daum, 2010). S. aureus produces various toxins

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Javad Sharifi-Rad, Shahid Beheshti University of Medical Sciences, Iran Floriana Campanile, Università degli Studi di Catania, Italy

#### \*Correspondence:

Mohammad Aminul Islam maislam@icddrb.org

#### †Present address:

Mohammad Aminul Islam, Paul G. Allen School for Global Animal Health, Washington State University, Pullman, WA, United States Mohsina Huq, School of Science, RMIT University, Melbourne, VIC, Australia Ashikun Nabi, Department of Biology, University of Vermont, Burlington, VT, United States

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 28 August 2018 Accepted: 27 February 2019 Published: 14 March 2019

#### Citation:

Islam MA, Parveen S, Rahman M, Huq M, Nabi A, Khan ZUM, Ahmed N and Wagenaar JA (2019) Occurrence and Characterization of Methicillin Resistant Staphylococcus aureus in Processed Raw Foods and Ready-to-Eat Foods in an Urban Setting of a Developing Country. Front. Microbiol. 10:503. doi: 10.3389/fmicb.2019.00503

**202**

fmicb-10-00503 March 12, 2019 Time: 19:1 # 2

which are often produced in the food, and consumption of intoxicated foods potentially leads to serious diseases (Abdolshahi et al., 2018). It has many cell-associated and secreted virulence factors; some of these virulence factors include Panton-Valentine leukocidin toxin (PVL), toxic shock syndrome toxin 1 (TSST-1), hemolysins, exfoliative toxins (ETs), and staphylococcal enterotoxins (SEs) (Tong et al., 2015). PVL is a cytotoxin, related to leukocyte destruction, tissue necrosis, diffuse cellulitis, skin and soft tissue infections, necrotizing pneumonia, and osteomyelitis (Lina et al., 1999). SEs cause staphylococcal food poisoning, whereas TSST-1 and ETs are responsible for toxic shock syndrome (TSS) and staphylococcal scalded-skin syndrome (SSSS), respectively (Tong et al., 2015).

Infections caused by S. aureus are difficult to treat due to its ability to acquire and develop resistance to multiple antibiotics. Over the past decades, the epidemiology of methicillin-resistant Staphylococcus aureus (MRSA) has changed significantly. MRSA has recently been listed as one of the high-priority antibiotic-resistant pathogens by the World Health Organization (Tacconelli et al., 2017). A majority of MRSA associated with disease in hospitalized patients is known as hospital-associated (HA)-MRSA. In the early 1990s, a new type of genetically different MRSA strains has been evolved in the community known as communityassociated (CA)-MRSA (Otto, 2010). Because of enhanced production of varieties of toxins, these CA-MRSA strains are exceptionally pathogenic (Cameron et al., 2011; Otto, 2012) compared to HA-MRSA. Furthermore, MRSA infections in the community caused by strains primarily associated with livestock is known as livestock-associated (LA)-MRSA (Nemati et al., 2008).

Apart from direct transmission to humans from animals, the latter being considered as a natural reservoir of this organism, transmission of MRSA might occur via exposure to or ingestion of contaminated foods. People having frequent contact with animal reservoirs or food contaminated with MRSA can become colonized with this organism and spread to the community. Food sampling and testing should be focused on foods of animal origin and especially the ready-to-eat (RTE) foods which require frequent manual handling for preparation and serving.

In Bangladesh, information on the prevalence of MRSA is currently scarce. Only a few surveys have been done in health care settings. One study among diabetic patients reported that around 37% of hospitalized and 22% of non-hospitalized patients were infected with MRSA (Jinnah et al., 1998). In a more recent study, the fraction of MRSA in hospitals of different cities in Bangladesh was shown to be 32–63%, which is much higher than in the United States and in European countries (Haq et al., 2005). There is substantial lack of information on the prevalence of MRSA in food sources in Bangladesh. Such information is useful for better understanding of the risk of exposure to MRSA through food, particularly the RTE foods.

In this study, we determined the occurrence of S. aureus and MRSA in retail food samples collected from local restaurants, superstores, and street vendors in Dhaka and characterized the isolates for antibiotic resistance, toxin genes, and genetic diversity using MLST and spa typing.

### MATERIALS AND METHODS

### Food Sample Collection

Between 2010 and 2013, a total of 162 retail food samples including 112 RTE foods and 40 processed raw meat and fish products were collected from different locations in Dhaka city (**Table 1**). At least 100g of each sample was bought from the vendors and collected in a sterile plastic bag. Samples were kept in an ice box (+4 to 8 ◦C) immediately after collection and transported to the laboratory within 3–4 h.

### Sample Processing

Twenty-five grams or ml (for liquid) of food sample were mixed with 225 ml of peptone saline water and homogenized. Diluted samples were spread on the Baird-Parker agar (BP) (Oxoid Ltd., Basingstoke, United Kingdom) and incubated at 37◦C for 24 to 48 h. After incubation, a maximum of 3 colonies showing typical characteristics of S. aureus (black/dark gray with lethicinase zone) were picked up and confirmed according to the procedure described earlier (International Organization for Standardization, 1999). All coagulase positive presumptive S. aureus isolates were confirmed with the API STAPH system (bioMérieux S.A., France) according to manufacturer's instructions.

### Antimicrobial Susceptibility Test

Susceptibility to antimicrobials was determined by an agar diffusion test using commercially available antibiotic disks (Oxoid Ltd., Basingstoke, United Kingdom) as described by the Clinical Laboratory Standards Institute (CLSI) guidelines (CLSI, 2012). The antimicrobial agents used were cefoxitin, chloramphenicol, ciprofloxacin, trimethoprimsulfamethoxazole, gentamicin, tetracycline, imipenem, erythromycin, amoxicillin-clavulanic acid, and oxacillin. Isolates that showed resistance to oxacillin in disk diffusion were tested for the minimum inhibitory concentration (MIC) for oxacillin by broth dilution method described by CLSI (2012). All MRSA isolates were tested for the MIC of vancomycin by E-test (bioMérieux S.A., France).

TABLE 1 | Prevalence of S. aureus and MRSA in ready-to-eat (RTE) food and processed raw meat, fish, milk samples in Dhaka, Bangladesh.


### Polymerase Chain Reaction Assays for Virulence Genes

All S. aureus isolates were tested for a panel of virulence and pathogenic genes including the S. aureus thermonuclease gene (nuc) (Brakstad et al., 1992), Panton-Valentine leukocidin toxin gene (pvl) (Lina et al., 1999), staphylococcal enterotoxin genes (sea, seb, sec, sed and see) (Sharma et al., 2000), TSS toxin-1 (tsst) gene, exfoliative toxin genes (eta and etb) and methicillin resistance gene (mecA) (Mehrotra et al., 2000). DNA was extracted from bacterial isolates according to the procedure described earlier (Bollet et al., 1955).

### MLST and Spa Typing

fmicb-10-00503 March 12, 2019 Time: 19:1 # 3

All MRSA isolates were characterized by multi locus sequence typing (MLST) according to the procedure described earlier (Enright et al., 2000). Sequence types (ST) were assigned according to the MLST database<sup>1</sup> .

For S. aureus protein A (spa) typing, the polymorphic X region of the spa gene (spa) was amplified by PCR using the primers 1095F and 1517R according to the procedure described earlier (Harmsen et al., 2003). spa types were assigned by using Ridom StaphType 1.4.1 software (Ridom GmbH, Würzburg, Germany<sup>2</sup> ).

### RESULTS

### Occurrence of S. aureus in Food Samples

Of the 162 samples, 35 (22%) were positive for S. aureus. Among these, 26 isolates were isolated from RTE foods and 9 from raw processed foods.

### Antibiotic Susceptibility of S. aureus

Antibiotic susceptibility test of the S. aureus isolates showed that 74% of isolates were resistant to erythromycin, 49% to ciprofloxacin, 31% to oxacillin, 26% to cefoxitin, 20% to amoxicillin-clavulanic acid, 20% to tetracycline, 11% to

<sup>1</sup>http://www.mlst.net/

<sup>2</sup>www.spaserver.ridom.de

trimethoprim-sulfamethoxazole, 6% to imipenem and 3% to gentamicin. None of the isolates were resistant to chloramphenicol (**Table 2**). Around 37% (n = 13) of isolates were multidrug resistant (MDR) (resistant to 3 or more classes of antibiotics). MIC for oxacillin was found ≥8 µg/ml for isolates that were identified as resistant in disk diffusion method. All MRSA isolates were found to be sensitive to vancomycin.

### Toxigenic Characteristics of S. aureus

All S. aureus isolates were positive for thermonuclease gene (nuc). About 71% of isolates (n = 25) were positive for pvl gene. More than 74% of isolates (n = 26) were positive for enterotoxin genes (sea = 26%, n = 9; seb = 11%, n = 4; sec = 49%, n = 17 and sed = 3%, n = 1) (**Table 3**). In each case, a PCR product of the expected size was generated (**Figure 1**). The frequencies of other genes are listed in **Table 3**. Nine (26%) isolates were positive for mecA gene either alone (2.8%, n = 1) or in tandem with other genes (sec-mecA-tsst1-pvl, seb-mecA, sea-mecA, sea-mecApvl, seb-mecA-pvl). None of the isolates were positive for see and etb genes.

### Identification and Characterization of MRSA

Of the 35 S. aureus isolates, 9 (26%) were detected as MRSA, which represents 6% of total number of food samples (n = 162) tested in the study. Of these 9 isolates, 6 were isolated from RTE foods mostly served in the road side small restaurants and street vendors, 2 from processed raw meat samples and 1 from processed fish sample. All but one MRSA isolates were resistant to both oxacillin and cefoxitin, with an MIC of oxacillin ≥16 µg/ml. All MRSA isolates were MDR. Of the 9 MRSA isolates, 4 (44%) were positive for pvl gene, 3 isolates of each were positive for sea and seb genes, respectively and 2 isolates were positive for sec gene. Isolates positive for pvl gene were positive for at least one additional enterotoxin gene (**Table 4**).

### Genotyping of MRSA

A total of 4 sequence types (ST) were identified among 9 MRSA isolates of which 3 isolates belonged to ST80 and 2 isolates in each belonged to ST6, ST239 and ST361. A total of 7 different spa types

TABLE 2 | Antimicrobial resistance of S. aureus strains isolated from RTE food and raw food samples (processed raw meat, fish and milk) in Dhaka, Bangladesh.


fmicb-10-00503 March 12, 2019 Time: 19:1 # 4

TABLE 3 | Prevalence of different toxin genes in S. aureus isolates from RTE food and raw food samples (processed raw meat, fish and milk) in Dhaka, Bangladesh.


were detected among 9 MRSA isolates, of which t1198 and t315 were the predominant one (2 isolates in each type), followed by t8731, t304, t275, t10546, t037 (1 isolate in each type).

### DISCUSSION

Foodborne transmission of MRSA is a global concern and therefore the prevalence and genetic characteristics of these organisms need to be thoroughly studied. This study provides the first evidence of the occurrence of MRSA in RTE food in Bangladesh. Around 23% (26/112) of RTE food samples collected from Dhaka city were found positive for S. aureus and 5% (6/112) were identified as MRSA. This rate is relatively higher than the reports from other countries, for example, the prevalence of MRSA in dairy products from Italy was 0.5% (Carfora et al., 2015) and 1.3% in retail foods from China (Yang et al., 2016).

Contamination of RTE foods with S. aureus can easily occur due to poor hygienic practices of food handlers during food preparation as it is known that 50–70% of healthy individuals serve as carriers of S. aureus (Solberg, 2000; Le Loir et al., 2003). Like many other resource poor settings, street foods in Dhaka city are often processed and served with bare hands. Although there is no data on the proportion of street food vendors in Dhaka city have S. aureus on their hands but a study in neighboring country India showed that 36% of hand rinse samples (n = 83) collected from workers responsible for food preparation, serving and cleaning, carried oxacillin resistant S. aureus (Kasturwar and Shafee, 2011). A similar study in Zimbabwe showed that 32% of food handlers carried S. aureus on their hands, while only 6.4% carried E. coli (Gadaga et al., 2008).

Clinical management of Staphylococcal infection is relied on antibiotic treatment which often fails due to aggressive resistance of organisms to antibiotics. We found that a high proportion of isolates in this study were resistant to erythromycin (74%) and ciprofloxacin (49%) while none of the isolates was resistant to chloramphenicol. fmicb-10-00503 March 12, 2019 Time: 19:1 # 5


TABLE 4 | Characteristics of MRSA isolated from RTE food and raw food samples (processed raw meat, fish and milk) in Dhaka, Bangladesh.

<sup>a</sup>AMC, Amoxicillin-clavulanic acid; E, Erythromycin; TE, Tetracycline; CIP, Ciprofloxacin; OX, Oxacillin; FOX, Cefoxitin; IPM, Imipenem; CN, Gentamicin; SXT, Trimethoprimsulfamethoxazole.

It indicates that this first generation antibiotic may serve as an alternative to the newer generation of more expensive antibiotics in resource poor settings, if infections are caused by these organisms.

Characterization of foodborne bacterial isolates for pathogenic properties provides important information on the ability of the isolates to cause human infection. We tested all S. aureus isolates for different pathogenic genes. We found that pvl was present in 71% of all S. aureus and in 44% of MRSA isolates. pvl is an important virulence gene of S. aureus, which is mainly found in clinical MRSA isolates, predominantly associated with community associated infections (CA-MRSA) (Pu et al., 2009; Hanson et al., 2011; Wang et al., 2014). The pvl gene is also considered to be a stable genetic marker for CA-MRSA (Deurenberg et al., 2007). The presence of pvl in large number of isolates in this study indicates the possible contamination of food via human sources and consequently contaminated food can serve as a source of CA-MRSA. Among classical enterotoxin genes, sec gene was predominantly found in S. aureus isolates (37%) while in case of MRSA isolates, sea and seb were more common (**Tables 3**, **4**). Epidemiological studies indicate that the majority of S. aureus infections and outbreaks have been caused by isolates with SEA type toxins, followed by isolates with SED, SEC and SEB toxin types (Asao et al., 2003; Ikeda et al., 2005; Cha et al., 2006; Kerouanton et al., 2007; Argudin et al., 2010). Among other toxin genes, tsst-1 (toxic shock syndrome toxin 1) and eta (an enterotoxin) were found in 17 and 9% of the S. aureus isolates, respectively. Although these toxins are mostly associated with human isolates, there are sporadic reports on the prevalence of S. aureus carrying these toxin genes from food sources (Hammad et al., 2012; Yang et al., 2018). Interestingly, one isolate was positive for multiple virulence genes including sec, tsst-1, eta and pvl indicating the potential ability of this isolate to cause human infection.

The genetic types of all MRSA isolates were characterized by MLST and spa typing. Of the 9 MRSA isolates, 3 belonged to the Sequence Type 80, two of these were pvl positive, had the same spa type (t1198) both isolated from RTE foods but of different types and from different locations (**Table 4**). pvl positive ST80 is predominantly found among CA-MRSA isolates in Europe and Middle East and they were associated with severe skin/soft tissue infections and necrotizing pneumonia (Budimir et al., 2010). The other genotypes found among MRSA isolates in this study were ST239-t037/t275, ST6-t304/t10546 and ST361-t315 (n = 2). All these genotypes were previously reported from clinical isolates of MRSA obtained from hospitalized patients. For example, the ST239-t037 was reported as the most common genotype among hospitalized burn patients in Iran and from hospitalized patients with wound/soft tissue infections and respiratory infections in Malaysia (Ghaznavi-Rad et al., 2010; Goudarzi et al., 2017). ST6 t304 was reported as the predominant genotype isolated from patients with wound/soft tissue infections at a tertiary hospital in the Sultanate of Oman (Udo et al., 2014). ST239-t037 and ST6 t304 clones of CA MRSA reported from Iran and Malaysia were pvl negative and a majority of ST239-t037 was positive for sea gene, which is similar to the characteristics of food isolates of the same genetic types found in this study (**Table 4**).

In conclusion, we report the first investigation of S. aureus from retail, RTE foods in Dhaka, Bangladesh. The contamination of S. aureus was common in RTE foods with a high prevalence of MRSA. All MRSA isolates were resistant to multiple antibiotics and a majority of these were positive for more than one toxin gene indicating their pathogenic potential. Genetic types of MRSA isolates in this study matched with the epidemic and pandemic clones of CA-MRSA. Our findings therefore strongly hint at the potential role of contaminated foods in the dissemination of multi-drug resistant S. aureus strains. A systematic surveillance of MRSA coupled with a focused educational and awareness campaign should be undertaken along the entire food production and supply chain, especially targeting the sectors involved with RTE foods. Furthermore, the findings described herein could also be generally relevant to the developing country settings of Asia, Africa and all other places where RTE food is sold and consumed.

### AUTHOR CONTRIBUTIONS

fmicb-10-00503 March 12, 2019 Time: 19:1 # 6

MI and SP developed the project and designed the research. MR, MH, AN, ZK, and JW performed the experiments. MI, MR, and MH wrote the manuscript. All authors analyzed and discussed the data, contributed to the writing of the statement and agreed

### REFERENCES


with its content and conclusions, and read and approved the final manuscript.

### FUNDING

This research study was funded by International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b). icddr,b gratefully acknowledge the following donors, who provided unrestricted support: Government of the People's Republic of Bangladesh, Canadian International Development Agency (CIDA), and the Department for International Development, United Kingdom (DFID).


fmicb-10-00503 March 12, 2019 Time: 19:1 # 7


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

Copyright © 2019 Islam, Parveen, Rahman, Huq, Nabi, Khan, Ahmed and Wagenaar. 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.

# Multi-Laboratory Validation of a Loop-Mediated Isothermal Amplification Method for Screening Salmonella in Animal Food

Beilei Ge<sup>1</sup> \*, Kelly J. Domesle<sup>1</sup> , Qianru Yang<sup>1</sup> , Thomas S. Hammack<sup>2</sup> , Shizhen S. Wang<sup>3</sup> , Xiaohong Deng<sup>2</sup> , Lijun Hu<sup>2</sup> , Guodong Zhang<sup>2</sup> , Yuan Hu<sup>4</sup> , Xiaokuang Lai<sup>4</sup> , Kyson X. Chou<sup>5</sup> , Jan Ryan Dollete<sup>5</sup> , Kirsten A. Hirneisen<sup>5</sup> , Sammie P. La<sup>5</sup> , Richelle S. Richter<sup>5</sup> , Diyo R. Rai<sup>6</sup> , Azadeh A. Yousefvand<sup>6</sup> , Paul K. Park<sup>7</sup> , Cindy H. Wu<sup>7</sup> , Tameji Eames<sup>7</sup> , David Kiang<sup>7</sup> , Ju Sheng<sup>8</sup> , Dancia Wu<sup>8</sup> , Lori Hahn<sup>9</sup> , Lisa Ledger<sup>9</sup> , Cynthia Logie<sup>9</sup> , Qiu You<sup>9</sup> , Durda Slavic<sup>9</sup> , Hugh Cai<sup>9</sup> , Sherry L. Ayers<sup>1</sup> , Shenia R. Young<sup>1</sup> and Ruiqing Pamboukian<sup>10</sup>

<sup>1</sup> Division of Animal and Food Microbiology, Office of Research, Center for Veterinary Medicine, United States Food and Drug Administration, Laurel, MD, United States, <sup>2</sup> Office of Regulatory Science, Center for Food Safety and Applied Nutrition, United States Food and Drug Administration, College Park, MD, United States, <sup>3</sup> Office of Analytics and Outreach, Center for Food Safety and Applied Nutrition, United States Food and Drug Administration, College Park, MD, United States, <sup>4</sup> Northeast Food and Feed Laboratory, Office of Regulatory Affairs, United States Food and Drug Administration, Jamaica, NY, United States, <sup>5</sup> Pacific Southwest Food and Feed Laboratory, Office of Regulatory Affairs, United States Food and Drug Administration, Irvine, CA, United States, <sup>6</sup> San Francisco Laboratory, Office of Regulatory Affairs, United States Food and Drug Administration, Alameda, CA, United States, <sup>7</sup> Food and Drug Laboratory Branch, California Department of Public Health, Richmond, CA, United States, <sup>8</sup> Office of Indiana State Chemist, Purdue University, West Lafayette, IN, United States, <sup>9</sup> Animal Health Laboratory, University of Guelph, Guelph, ON, Canada, <sup>10</sup> Office of Regulatory Science, Office of Regulatory Affairs, United States Food and Drug Administration, Rockville, MD, United States

Loop-mediated isothermal amplification (LAMP) has gained wide popularity in the detection of Salmonella in foods owing to its simplicity, rapidity, and robustness. This multi-laboratory validation (MLV) study aimed to validate a Salmonella LAMP-based method against the United States Food and Drug Administration (FDA) Bacteriological Analytical Manual (BAM) Chapter 5 Salmonella reference method in a representative animal food matrix (dry dog food). Fourteen independent collaborators from seven laboratories in the United States and Canada participated in the study. Each collaborator received two sets of 24 blind-coded dry dog food samples (eight uninoculated; eight inoculated at a low level, 0.65 MPN/25 g; and eight inoculated at a high level, 3.01 MPN/25 g) and initiated the testing on the same day. The MLV study used an unpaired design where different test portions were analyzed by the LAMP and BAM methods using different preenrichment protocols (buffered peptone water for LAMP and lactose broth for BAM). All LAMP samples were confirmed by culture using the BAM method. BAM samples were also tested by LAMP following lactose broth preenrichment (paired samples). Statistical analysis was carried out by the probability of detection (POD) per AOAC guidelines and by a random intercept logistic regression model. Overall, no significant differences in POD between the Salmonella LAMP and BAM methods were observed with either unpaired or paired samples, indicating the methods were

#### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Laura Lynn Montier, Baylor College of Medicine, United States Vikrant Dutta, BioMérieux, United States

#### \*Correspondence: Beilei Ge beilei.ge@fda.hhs.gov

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 21 December 2018 Accepted: 05 March 2019 Published: 28 March 2019

#### Citation:

Ge B, Domesle KJ, Yang Q, Hammack TS, Wang SS, Deng X, Hu L, Zhang G, Hu Y, Lai X, Chou KX, Dollete JR, Hirneisen KA, La SP, Richter RS, Rai DR, Yousefvand AA, Park PK, Wu CH, Eames T, Kiang D, Sheng J, Wu D, Hahn L, Ledger L, Logie C, You Q, Slavic D, Cai H, Ayers SL, Young SR and Pamboukian R (2019) Multi-Laboratory Validation of a Loop-Mediated Isothermal Amplification Method for Screening Salmonella in Animal Food. Front. Microbiol. 10:562. doi: 10.3389/fmicb.2019.00562

**209**

comparable. LAMP testing following preenrichment in buffered peptone water or lactose broth also resulted in insignificant POD differences (P > 0.05). The MLV study strongly supports the utility and applicability of this rapid and reliable LAMP method in routine regulatory screening of Salmonella in animal food.

Keywords: LAMP, Salmonella, multi-laboratory, validation, animal food

### INTRODUCTION

fmicb-10-00562 March 27, 2019 Time: 17:12 # 2

Salmonella is a ubiquitous human and animal pathogen, with human outbreak-related illnesses broadly attributed to multiple food categories of plant and animal origins (Interagency Food Safety Analytics Collaboration [IFSAC], 2018). The presence of Salmonella in animal food (e.g., pet food, animal feed, and raw materials and ingredients) is also well documented (Li et al., 2012; Ge et al., 2013; Hsieh et al., 2014; Nemser et al., 2014; Food and Agriculture Organization [FAO]/World Health Organization [WHO], 2015; Jiang, 2016; Molina et al., 2016; Magossi et al., 2018), which impacts not only animal health but also human food safety due to consumption of animal-derived food or direct contact with pet food (Crump et al., 2002; Food and Agriculture Organization [FAO]/World Health Organization [WHO], 2015). The FDA Food Safety Modernization Act (FSMA) prioritizes preventive controls for human and animal foods, emphasizing vigilant product testing and environmental monitoring for zoonotic pathogens such as Salmonella (Food and Drug Administration [FDA], 2017a,b). Rapid and reliable methods are thus in great need to effectively support such efforts.

Current Salmonella testing in foods relies on microbiological culturing, which consists of time-consuming and labor-intensive procedures that require days or weeks for a definitive result (International Organization for Standardization [ISO], 2017; United States Department of Agriculture [USDA], 2017; Andrews et al., 2018). Rapid, sensitive, and specific nucleic acid amplification tests (NAATs), including PCR, real-time quantitative PCR (qPCR), and loop-mediated isothermal amplification (LAMP), have been developed and applied in the detection and identification of Salmonella in foods (Malorny et al., 2009; Balachandran et al., 2012; Lofstrom and Hoorfar, 2012; Cheng et al., 2015; Yang et al., 2016; Domesle et al., 2018; Hu et al., 2018). The isothermal LAMP method, in particular, has gained wide popularity as highlighted in a recent comprehensive review (Yang et al., 2018). Two distinct advantages of LAMP over PCR are running at a constant temperature (Notomi et al., 2000) and high tolerance to matrix inhibitors (Kaneko et al., 2007), which obviate the need for a sophisticated thermocycler or a complicated DNA extraction protocol. These attractive features have led to the development of many new Salmonella LAMP assays, portable microfluidic devices, and commercially available systems (Yang et al., 2018).

Validation plays a critical role in the life cycle of a method from development to implementation. Despite the growing enthusiasm in developing new Salmonella LAMP assays, limited effort has been devoted to validate the assay performance against well-established reference methods (International Organization for Standardization [ISO], 2017; United States Department of Agriculture [USDA], 2017; Andrews et al., 2018). These validation studies, performed at single laboratory, independent laboratory, and collaborative study (multi-laboratory) levels, represent rigorous evaluations of an alternative method's performance compared with that of the reference method in a food matrix when conducted per international guidelines (Association of Official Analytical Chemists [AOAC], 2012; International Organization for Standardization [ISO], 2016). Similar FDA guidelines have been established for the validation of microbiological methods in foods (Food and Drug Administration [FDA], 2015). Methods that have successfully gone through multi-laboratory validation (MLV) are thus suitable for routine regulatory use.

We previously developed a LAMP assay specifically targeting the Salmonella invasion gene invA and showed it to be rapid, reliable, and robust in multiple food matrices (Chen et al., 2011; Yang et al., 2013, 2014, 2015, 2016; Domesle et al., 2018). The method was 100% specific among 300 strains (247 Salmonella of 185 serovars and 53 non-Salmonella) tested and was capable of detecting <1 CFU/25 g in animal food (Domesle et al., 2018). Following FDA guidelines (Food and Drug Administration [FDA], 2015), we recently completed a stringent single-laboratory validation of the method in six animal food matrices including cattle feed, chicken feed, horse feed, swine feed, dry cat food, and dry dog food (Domesle et al., 2018).

This MLV collaborative study aimed to validate the invA-based Salmonella LAMP assay as performed on the Genie II or Genie III platform (OptiGene Ltd., West Sussex, United Kingdom) (alternative method) against the FDA BAM Chapter 5 Salmonella (reference method) in a representative animal food matrix (dry dog food) for future incorporation into the FDA's compendium of analytical laboratory methods for food and feed safety. MLV participants included 14 independent collaborators from seven FDA, state, and academic laboratories in the United States and Canada. The MLV study also compared the effects of two preenrichment buffers used in LAMP and BAM on Salmonella detection in animal food.

### MATERIALS AND METHODS

### Study Design

**Figure 1** shows a diagram of the MLV study design. The main component (panels 2 and 3) used an unpaired design, where different test portions were analyzed by the reference FDA BAM method (panel 2) and the alternative LAMP method (panel 3) following preenrichment in different buffers (lactose broth [LB]

for BAM and buffered peptone water [BPW] for LAMP). All LAMP samples were confirmed by BAM culturing (panel 4, i.e., BPW-BAM). BAM samples were also tested by LAMP following LB preenrichment (panel 1, i.e., LB-LAMP), essentially forming paired samples (panels 1 and 2).

Fourteen independent collaborators (or independent teams), two each from seven FDA, state, and academic laboratories participated in the MLV. The FDA laboratories were from the Office of Regulatory Science at FDA's Center for Food Safety and Applied Nutrition, and Northeast Food and Feed Laboratory, Pacific Southwest Food and Feed Laboratory, and San Francisco Laboratory at FDA's Office of Regulatory Affairs. Other participants were the Food and Drug Laboratory Branch at California Department of Public Health, Office of Indiana State Chemist, and Animal Health Laboratory at University of Guelph (ON, Canada).

### Sample Inoculation, Storage, and Shipment

Inoculated samples were prepared by Q Laboratories (Cincinnati, OH, United States). Briefly, bulk dry dog food in kibble form was obtained from a local pet store and screened for the presence of Salmonella by the BAM Chapter 5 reference method (Andrews et al., 2018) and the iQ-Check Salmonella II Real-Time PCR detection kit (Bio-Rad, Hercules, CA, United States) to confirm negative results.

Dry dog food confirmed negative for Salmonella was separated into two sets and inoculated with a lyophilized culture of Salmonella enterica serovar Infantis ATCC 51741 at two target levels: a high level of ca. 2 to 5 CFU/25 g test portion and a low level of ca. 0.2 to 2 CFU/25 g test portion. An uninoculated control set (0 CFU/25 g test portion) was also included. After inoculation, the three sets of bulk samples were homogenized and held at room temperature for 2 weeks for aging to simulate storage. Replicate samples (5–10; 25 g each) from the two inoculated sets were evaluated at three time points (immediately after inoculation and homogenization, after 1 week of aging, and after 2 weeks of aging) by BAM and iQ-Check methods to verify the target levels and homogeneity.

On the day of shipment, a five-tube three-level most probable number (MPN) analysis was performed by evaluating 5 × 50 g replicates, 5 × 25 g replicates, and 5 × 10 g replicates to obtain final inoculation levels in the dry dog food sample sets. The samples were apportioned (25 g each), packaged, labeled (with randomized, blind-coded, three-digit numbers), and shipped overnight to the seven participating laboratories. For each laboratory, four sets of eight samples from each of the three inoculation levels were sent, along with two sets of samples from the uninoculated control set reserved for aerobic plate count (APC).

### Overview of Sample Analysis

All collaborators (or teams) began testing on the same day. APC was performed by the pour plate method according to the FDA BAM Chapter 3 (Maturin and Peeler, 2018) or using the CompactDry plates (Hardy Diagnostics, Santa Maria, CA, United States). On day 1, each collaborator processed 24 samples following the BAM method and 24 samples following the

LAMP method (**Figure 2**). Additionally, all LB preenrichment cultures from the BAM samples were tested by LAMP (i.e., LB-LAMP), and all BPW preenrichment cultures from the LAMP samples were processed with BAM for culture confirmation (i.e., BPW-BAM) from day 2. Therefore, a full data set from each collaborator consisted of 48 BAM and 48 LAMP results.

### The Loop-Mediated Isothermal Amplification (LAMP) Method

DNA extraction was performed by using the PrepMan Ultra sample preparation reagents (Thermo Fisher Scientific, Waltham, MA, United States). Briefly, aliquots (1 ml) of BPW or LB preenrichment cultures were first centrifuged at 900 × g for 1 min to remove large particles followed by another centrifugation at 16,000 × g for 2 min. The pellets were suspended in 100 µl of PrepMan Ultra reagent, heated at 100◦C for 10 min, cooled to room temperature, and centrifuged again at 16,000 × g for 2 min. The supernatants (sample DNA extracts) were stored at −20◦C until use.

The LAMP assay was carried out as described previously (Domesle et al., 2018). A positive control (S. enterica Typhimurium ATCC 19585 [LT2] at 1.7 × 10<sup>4</sup> CFU/reaction) and no template control (molecular grade water) were included in each LAMP run. Briefly, the reagent mixture in a total volume of 25 µl contained 1× isothermal master mix ISO-001 (consisting of a strand-displacing GspSSD DNA polymerase large fragment

from Geobacillus spp., thermostable inorganic pyrophosphatase, reaction buffer, MgSO4, dNTPs, and a double-stranded DNA binding dye; OptiGene Ltd.), 1× primer mix (0.1 µM each outer primer Sal4-F3 [GAACGTGTCGCGGAAGTC] and Sal4-B3 [CGGCAATAGCGTCACCTT], 1.8 µM each inner primer Sal4- FIP [GCGCGGCATCCGCATCAATATCTGGATGGTATGCC CGG] and Sal4-BIP [GCGAACGGCGAAGCGTACTGTCGCAC CGTCAAAGGAAC], and 1 µM each loop primer Sal4- LF [TCAAATCGGCATCAATACTCATCTG] and Sal4-LB [AAAGGGAAAGCCAGCTTTACG]; Integrated DNA Technologies, Coralville, IA, United States), and 2 µl of sample DNA extract. The LAMP reaction was run at 65◦C for 30 min followed by an annealing step from 98 to 80◦C with 0.05◦C decrement per second (**Figure 3C**) in the Genie II or Genie III real-time fluorometer (OptiGene Ltd.). Fluorescence readings were acquired using the 6-carboxyfluorescein (FAM) channel (**Figure 3A**) and time-to-peak values (Tmax; min) were determined when fluorescence ratios reached the maximum value of the amplification rate curve (**Figure 3B**). Corresponding annealing temperatures (Tm; ◦C) of LAMP products were obtained in the anneal derivative curve (**Figure 3D**). Both Tmax and T<sup>m</sup> values were displayed in the "Results" tab at the end of the run (**Figure 3E**). Testing was repeated once independently.

### The Bacteriological Analytical Manual (BAM) Method

Procedures described in the BAM Chapter 5 (Andrews et al., 2018) were followed. All media and reagents were from BD Diagnostic Systems (Sparks, MD, United States) unless specified otherwise. As outlined in **Figure 2**, samples were processed by preenrichment in LB (day 1), selective enrichment in Rappaport-Vassiliadis (RV) medium and tetrathionate (TT) broth (day 2), selective plating on bismuth sulfite (BS) agar, xylose lysine desoxycholate (XLD) agar, and Hektoen enteric (HE) agar (day 3), biochemical confirmation on triple sugar iron agar (TSI) slant and lysine iron agar (LIA) slant (day 4), and serological identification by Salmonella O antiserum poly B (day 5). Additional confirmation tests performed included VITEK 2 Gram-negative biochemical identification method (AOAC Official Method 2011.17), Bruker MALDI Gram-negative Biotyper method (AOAC Official Method 2017.09) (Association of Official Analytical Chemists [AOAC], 2018), or real-time qPCR as specified in the BAM Chapter 5 (Andrews et al., 2018).

### Statistical Analysis

fmicb-10-00562 March 27, 2019 Time: 17:12 # 6

MPNs were calculated for the low- and high-level inoculated dry dog food using the LCF MPN calculator version 1.6 (Least Cost Formulations, Ltd., 2008). BAM samples were considered positive when Salmonella isolates were recovered. LAMP samples with the correct T<sup>m</sup> (approximately 90◦C) and Tmax values between 5 and 30 min were considered positive. For this MLV, all LAMP testing results were reported as presumptive results (presumptive positive or presumptive negative). BAM and LAMP results for each contamination level (including uninoculated controls) were analyzed by using the probability of detection (POD) statistical model (Wehling et al., 2011) with the AOAC Binary Data Interlaboratory Study Workbook version 2.3 (Association of Official Analytical Chemists [AOAC], 2013). For each collaborator, PODs were calculated for the LAMP presumptive results (including false positive ones), LAMP confirmed by BAM results (including false negative ones), and LAMP final results (excluding false positive and false negative ones, i.e., only those LAMP samples tested positive by both LAMP and BAM confirmation), and the BAM reference results. LPOD values was determined by combining all valid collaborator-level POD data and the difference in LPOD (dLPOD) between two methods were calculated with a 95% confidence interval. The two methods were considered statistically significant when the dLPOD confidence interval did not contain zero.

Additionally, BAM and LAMP results were analyzed by using a random intercept logistic regression model for unpaired samples and Obuchowski's modified McNemar's test (Obuchowski, 1998) and a conditional logistic regression model for paired samples. Differences between the methods being compared were considered significant when P < 0.05.

### RESULTS

The average APC was 2.1 × 10<sup>1</sup> CFU/g (ranging from <1.0 × 10<sup>1</sup> to 1.6 × 10<sup>2</sup> CFU/g) for the uninoculated dry dog food controls. Salmonella MPNs obtained in the two inoculated sample sets, with a 95% confidence interval, were 0.65 MPN/25 g (0.30, 1.40) for the low level and 3.01 MPN/25 g (1.31, 6.89) for the high level. Two collaborators (9 and 10) mixed up sample bag sets among the four sets of samples received in their laboratory, resulting in uneven number (5–10) of samples tested per inoculation level. Nonetheless, the samples were not compromised as they were individually bagged and blindly coded, and their data were still included in the final statistical analysis for the MLV study. Data from another two collaborators (2 and 4) were excluded due to confirmed positive results among uninoculated controls. In total, there were 288 (12 collaborator × 24 samples/collaborator) data points each for LAMP and BAM in the final comparisons presented below, which include LAMP vs. BAM using unpaired samples, LAMP vs. BAM using paired samples, BPW vs. LB for use as LAMP preenrichment buffers, and BPW vs. LB for use as BAM preenrichment buffers.

### Unpaired Sample Statistical Analysis: LAMP Was Comparable to BAM

**Table 1** shows the collaborator-level comparative results for the detection of Salmonella Infantis ATCC 51741 in 25 g dry dog food test portions by the LAMP alternative method versus the FDA BAM Chapter 5 reference method in an unpaired study design, i.e., different portions were analyzed by LAMP and BAM using different preenrichment buffers. For the uninoculated controls, collaborators 2, 4, and 5 had LAMP presumptive positive results with the rate as high as 75% for both collaborators 2 and 4. The single sample for collaborator 5 did not confirm positive by BAM culturing, but several samples for collaborators 2 and 4 did (LAMP final). Based on these results, data from collaborators 2 and 4 were excluded from the MLV. In addition, collaborator 1 had one LAMP presumptive negative sample confirming positive by BAM and collaborator 4 had one BAM sample testing positive by BAM. Fractional recovery (i.e., 25 to 75% positive responses) was obtained for the low inoculation level by all collaborators although 2, 4, 9, and 10 achieved that by only one method (LAMP presumptive or BAM reference). All high-level inoculated samples tested positive, regardless of the method used (**Table 1**).

**Table 2** summarizes the statistics generated using the POD model and comparisons made using this model and a random intercept logistic regression model for unpaired samples (e.g., LAMP vs. BAM) and the Obuchowski's modified McNemar's test and a conditional logistic regression model for paired samples (e.g., LAMP presumptive vs. LAMP confirmed). For the low inoculation level, 51 out of 94 samples were LAMP presumptive positive (LPOD of 0.54) with 48 of them confirming positive (LPOD of 0.51). No false negative results were obtained (data not shown), therefore the LAMP final LPOD was also 0.51. Among 98 samples tested by BAM, 58 produced positive results (LPOD of 0.59). A dLPOD value of −0.08 with a 95% confidence interval (−0.24, 0.08) was obtained between LAMP final and BAM, indicating they were comparable. Similarly, for the high inoculation level and uninoculated controls, no significant differences were observed between LAMP final and BAM as confidence intervals for both dLPOD values contained zero. Based on dLPOD analysis, three other comparisons (i.e., LAMP presumptive vs. BAM, LAMP presumptive vs. LAMP confirmed, and LAMP presumptive vs. LAMP final) also showed no statistical significance. The statistical insignificance for all four comparisons at all three inoculation levels were separately confirmed by using aforementioned statistical models as indicated by P-values greater than 0.05 (**Table 2**).

### Paired Sample Statistical Analysis: LAMP Was Comparable to BAM

**Table 3** shows the summary statistics for the LAMP and BAM methods when paired samples were used (LB-LAMP vs. BAM), i.e., the same test portions were analyzed by LAMP and


TABLE 1 |

Comparative

 detection

of

Salmonella

Infantis ATCC 51741 in 25 g dry dog food test portions by the LAMP method versus the FDA BAM Chapter 5 reference method in an unpaired study design.

uninoculated

 test portions.

TABLE 2 | Summary of statistics generated using the POD model and other models for the detection of Salmonella Infantis ATCC 51741 in 25 g dry dog food test portions by the LAMP method versus the BAM reference method in an unpaired study design.


<sup>a</sup> LPOD is a composite POD across collaborators and includes between-collaborator variation in addition to variation inherent in the binomial nature of the binary probabilities. s<sup>r</sup> is repeatability standard deviation, s<sup>L</sup> is among-collaborator standard deviation, s<sup>R</sup> is reproducibility standard deviation. P-value is homogeneity test of collaborator PODs. <sup>b</sup> dLPOD is the difference in LPOD between two methods. The numbers in parenthesis are 95% confidence interval (lower control limit [LCL], upper control limit [UCL]) estimates on dLPOD. A confidence interval for dLPOD that does not contain 0 indicates a statistically significant difference between the two methods being compared. A random intercept logistic regression model was used for unpaired comparison and Obuchowski's modified McNemar's test and a conditional logistic regression model (numbers in parenthesis) were used for paired comparisons. f indicates that model fitting failed to converge. N/A, no test was done because of complete match of the results.

BAM following preenrichment in LB. For the low inoculation level, 58 out of 98 samples were LB-LAMP positive (LPOD of 0.59) while 58 out of 98 samples were positive by BAM (LPOD of 0.59). Collaborator 13 reported one positive sample by LB-LAMP only, while collaborator 8 had one sample positive by BAM only (data not shown). A dLPOD value of 0.00 with a 95% confidence interval (−0.18, 0.18) was obtained, indicating the two methods were not significantly different. Similarly, for the high inoculation level and uninoculated controls, no significant differences were observed as confidence intervals for both dLPOD values contained zero. One uninoculated sample from collaborator 14 was positive by LB-LAMP but not BAM (data not shown). The statistical insignificance between LB-LAMP and BAM at all three inoculation levels was separately


TABLE 3 | Summary of statistics generated using the POD model and other models for the detection of Salmonella Infantis ATCC 51741 in 25 g dry dog food test portions by the LAMP method (with LB preenrichment) versus the BAM reference method in a paired study design.

<sup>a</sup> LPOD is a composite POD across collaborators and includes between-collaborator variation in addition to variation inherent in the binomial nature of the binary probabilities. s<sup>r</sup> is repeatability standard deviation, s<sup>L</sup> is among-collaborator standard deviation, s<sup>R</sup> is reproducibility standard deviation. P-value is homogeneity test of laboratory (collaborator) PODs. <sup>b</sup> Collaborator 13 had one positive sample by LB-LAMP only, while collaborator 8 had one positive sample by BAM only. <sup>c</sup> dLPOD is the difference in LPOD between two methods. The numbers in parenthesis are 95% confidence interval (lower control limit [LCL], upper control limit [UCL]) estimates on dLPOD. A confidence interval for dLPOD that does not contain 0 indicates a statistically significant difference between the two methods being compared. Obuchowski's modified McNemar's test and a conditional logistic regression model (numbers in parenthesis) were used for LB-LAMP vs. BAM comparisons. f indicates that model fitting failed to converge. N/A, no test was done because of complete match of the results.

confirmed by using the Obuchowski's modified McNemar's test and the conditional logistic regression model (**Table 3**).

### Preenrichment With BPW vs. LB Did Not Affect Salmonella Detection

**Table 4** shows the statistics generated when unpaired samples were tested by either LAMP or BAM using different enrichment broths (LAMP vs. LB-LAMP and BPW-BAM vs. BAM). When tested by the LAMP method, 51 out of 94 low-level inoculated samples were positive (LPOD of 0.54) following BPW preenrichment, while 58 out of 98 samples produced positive results (LPOD of 0.59) following LB preenrichment. A dLPOD value of −0.05 with a 95% confidence interval (−0.22, 0.12) was obtained, indicating LAMP and LB-LAMP were not significantly different. Similarly, for the high inoculation level and uninoculated controls, no significant differences were observed as confidence intervals for both dLPOD values contained zero. Two different uninoculated samples were positive by either LAMP (for collaborator 5) or LB-LAMP (collaborator 14); neither was confirmed by BAM culturing (data not shown). The statistical insignificance at all three inoculation levels were separately confirmed by using the random intercept logistic regression model (**Table 4**). Therefore, preenrichment in BPW or LB did not significantly influence the LAMP results. The same held true for the BAM method when either BPW or LB were used as preenrichment buffers, i.e., there were no statistical significant differences for all three inoculation levels (**Table 4**).

### DISCUSSION

This collaborative study rigorously validated a LAMP-based method for the screening of Salmonella in dry dog food at the multi-laboratory level. FDA's current method validation guidelines for microbial pathogens in foods and feeds (Food and Drug Administration [FDA], 2015) were used, which align well with those from the AOAC and ISO (Association of Official Analytical Chemists [AOAC], 2012; International Organization for Standardization [ISO], 2016). In 2016, a United Kingdom study (D'Agostino et al., 2016) reported the validation of a LAMP/ISO 6579-based method for analyzing soya meal (an animal feed ingredient) for the presence of S. enterica in ten laboratories from eight European countries. For reasons of cost and logistics, that interlaboratory study did not use centrally prepared Salmonella-contaminated soya meal samples. Instead, commercially available certified Salmonella reference materials were used for inoculation by each participating laboratory, and no aging period was incorporated. Importantly, none of the three levels tested (0, 1–5, and 14–68 CFU per test portion) produced fractional positive results (25–75%) and three uninoculated control samples were confirmed positive for Salmonella (D'Agostino et al., 2016).


TABLE 4 | Summary of statistics generated using the POD model and other models for the detection of Salmonella Infantis ATCC 51741 in 25 g dry dog food test portions by LAMP or BAM when different preenrichment buffers were used for each one in an unpaired study design.

#### <sup>a</sup> LPOD is a composite POD across collaborators and includes between-collaborator variation in addition to variation inherent in the binomial nature of the binary probabilities. s<sup>r</sup> is repeatability standard deviation, s<sup>L</sup> is among-collaborator standard deviation, s<sup>R</sup> is reproducibility standard deviation. P-value is homogeneity test of collaborator PODs. <sup>b</sup> Collaborator 5 had one positive sample by LAMP and collaborator 14 had one positive sample by LB-LAMP; neither were confirmed by BAM culturing. <sup>c</sup> dLPOD is the difference in LPOD between two methods. The numbers in parenthesis are 95% confidence interval (lower control limit [LCL], upper control limit [UCL]) estimates on dLPOD. A confidence interval for dLPOD that does not contain 0 indicates a statistically significant difference between the two methods being compared. A random intercept logistic regression model was used for LAMP vs. LB-LAMP and BPW-BAM vs. BAM comparisons. N/A, no test was done because of complete match of the results.

Unlike the United Kingdom study which was a "paired" trial, this MLV used an unpaired study design, i.e., different test portions from the same bulk samples inoculated centrally and aged for 2 weeks were analyzed by the LAMP alternative method and the BAM reference method using different preenrichment buffers. All LAMP samples were confirmed by BAM culturing and the reported LAMP final positive results were for samples tested positive by both LAMP and BAM confirmation. For the low-level inoculation, the overall proportions of positive responses were 51% for LAMP final and 59% for BAM (**Table 2**), which clearly satisfies the criteria outlined in validation guidelines of the AOAC, FDA, and ISO (Association of Official Analytical Chemists [AOAC], 2012; Food and Drug Administration [FDA], 2015; International Organization for

Standardization [ISO], 2016). Multiple pairwise comparisons showed insignificant differences between LAMP and BAM by using either the POD analysis or other statistical models (**Table 2**), which highlights the success and rigor of this MLV study. Feedback from participating laboratories showed that the LAMP method was rapid, sensitive, practical, user-friendly, and easily adoptable.

A few false positive and/or false negative results were observed across the three testing levels in this MLV. For the lowlevel inoculated samples, there were three LAMP false positive results (one each for collaborators 5, 9, and 14 comparing LAMP presumptive and LAMP final) and no false negative results (**Table 1** excluding data from collaborators 2 and 4). In all three cases, the Tmax values were rather high (>15 min) compared to others (ca. 7 min) (data not shown), indicating the amount of target DNA in the sample DNA extracts was low. This may be attributed to low contamination levels and dead or injured cells in these samples, which failed to reach the detection limit of BAM even after enrichments. The samples were shipped without dry ice since pet food is usually stored and shipped at ambient temperature; however, this may have contributed to some of the variability observed in the study. Another possibility is there was cross-contamination introduced during DNA extraction or assay setup for LAMP. For the uninoculated controls, one false positive (collaborator 5) and one false negative (collaborators 1, noted in footnote) results were observed. The former had high Tmax values (average of 14 min), while the latter was technically true negative (false positive by BAM) as it was an uninoculated control. Cross-contamination may have occurred when the two collaborators processed the samples for LAMP or BAM. Prior to this MLV, the LAMP assay has been extensively evaluated and high specificity (100% inclusivity and exclusivity) and high sensitivity (a detection limit of <1 CFU/25 g in animal food) have been demonstrated (Chen et al., 2011; Domesle et al., 2018; Yang et al., 2013, 2014, 2015, 2016). During the MLV study, none of the positive control samples produced false negative results, and none of notemplate-control samples produced false positive results. These outcomes corroborate the high specificity and sensitivity of the Salmonella LAMP assay.

Besides the main component of the MLV study (unpaired design), we also compared the performance of LAMP and BAM using paired samples, i.e., the same test portions were analyzed by LAMP and BAM following LB preenrichment (LB-LAMP vs. BAM). One false positive (collaborator 13) and one false negative (collaborator 8) results were observed in lowlevel inoculated samples and one false positive was observed in one uninoculated sample (**Table 3**). Similar reasons described above may account for these false positive or false negative results observed. Nonetheless, paired samples also confirmed the statistically insignificant differences between the two methods using either POD analysis or other models.

As LAMP is gaining popularity in clinical diagnostics and food testing, many commercially available LAMP-based systems and assays have been developed and some were validated for Salmonella detection in food (Yang et al., 2018). These include the 3M Molecular Detection Assay (MDA) Salmonella (3M Food Safety, St. Paul, MN, United States) in raw ground beef and wet dog food (Bird et al., 2013, 2014), the 3M MDA 2 – Salmonella in raw ground beef and creamy peanut butter (Bird et al., 2016), and the SAS Molecular Tests Salmonella detection kit (SA Scientific Ltd., San Antonio, TX, United States) in ground beef, beef trim, ground turkey, chicken carcass rinses, bagged mixed lettuce, and fresh spinach (Bapanpally et al., 2014). In the two studies evaluating the 3M MDA Salmonella in wet dog food against the FDA BAM method, an unpaired study design was used and fractional positive results were obtained with POD analysis showing the methods were comparable (Bird et al., 2013, 2014). It is noteworthy that platforms used for the detection of LAMP amplicons were different in these studies as compared to our study. Bioluminescence was used for the 3M MDA assays, turbidity for the SAS kit, and fluorescence was used in our MLV (Genie II or Genie III). We previously tested the Salmonella LAMP assay on all three platforms (Chen et al., 2011; Yang et al., 2013, 2014, 2015, 2016; Domesle et al., 2018) and the fluorescence-based Genie II or Genie III was chosen for its simplicity, rapidity, portability, software interface, report format, and user-friendliness, with the annealing step offering an extra checkpoint to ensure the high specificity of the assay.

Another interesting outcome of the MLV study was the comparison of preenrichment buffers used for LAMP and BAM. We chose BPW as the default preenrichment buffer for the Salmonella LAMP method since preliminary data showed that shorter Tmax values were obtained for samples pre-enriched in BPW compared to those in LB (data not shown). Comparing Tmax values generated in this MLV for low- and high-level inoculated samples showed that BPW preenrichment generated Tmax values on average 1.9 min and 1.6 min shorter than those using LB preenrichment, suggesting the amount of DNA was higher when BPW was used as the preenrichment buffer. A recent study evaluating the 3M MDA Salmonella and the ANSR (stands for amplified nucleic single temperature reaction) detection system for Salmonella (Neogen Food Safety, Lansing, MI, United States) in egg products also showed that preenrichment in BPW improved the performance of both assays compared to LB (Hu et al., 2017). Nonetheless, using POD analysis for a qualitative method, statistically significant differences were not observed between BPW and LB for either LAMP or BAM, indicating they were comparable (**Table 4**).

It is worth noting that the S. Infantis ATCC 51741 used for inoculation in this MLV was a non-H2S producer with uncharacteristic serological reactions to Salmonella O antiserum poly B (data not shown). As a result, multiple confirmation methods besides serotyping were performed by participating laboratories, including VITEK 2, Bruker MALDI, and real-time qPCR, extending the time of sample testing by BAM to 2 weeks in contrast to 24 h by LAMP.

In our single-laboratory validation study (Domesle et al., 2018), five other animal food types (cattle feed, chicken feed, horse feed, swine feed, and dry cat food) besides dry dog food were successfully validated following FDA's guidelines. The LAMP method validated in this MLB study in dry dog food should be applicable to these and other animal food types, per guidelines of the AOAC, FDA, and ISO (Association of Official Analytical Chemists [AOAC], 2012; Food and Drug Administration [FDA], 2015; International Organization for Standardization [ISO], 2016). Additional matrix extension studies may be readily performed in a variety of food matrices.

### CONCLUSION

fmicb-10-00562 March 27, 2019 Time: 17:12 # 12

In conclusion, the MLV study clearly demonstrated the utility and applicability of this rapid and reliable LAMP method in routine regulatory screening of Salmonella in animal food. As only LAMP-positive samples should continue with the isolation of Salmonella by the FDA BAM culture method, the LAMP method holds great potential to significantly reduce the time and labor and improve efficiency in animal food testing.

### DATA AVAILABILITY

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

### AUTHOR CONTRIBUTIONS

BG, KD, QRY, TH, and RP contributed to design and coordination of the study. BG, KD, and QRY performed the data analysis and interpretation. SW contributed to statistical analysis. XD, LJH, GZ, YH, XL, KC, JD, KH, SL, RR, DR, AY, PP, CW, TE, DK, JS, DW, LH, LL, CL, QY, DS, and HC contributed to sample testing and coordination of the study in

### REFERENCES


seven participating laboratories. SA and SY assisted with sample analysis in the coordinating laboratory. All authors reviewed the final manuscript.

### ACKNOWLEDGMENTS

We are grateful to William B. Baroletti/Lawrence E. James, Teresa Lee, and Carrie Leach/Mark Moelhman for supporting the project at FDA's Northeast Food and Feed Laboratory, San Francisco Laboratory, and Office of Indiana State Chemist, respectively. We thank Andriy Tkachenko, Jake Guag, Sarah M. Nemser, and Renate Reimschuessel at the FDA's Veterinary Laboratory Investigation and Response Network (Vet-LIRN) for insights and guidelines on MLV studies and recruiting participating laboratories. We also thank Claudine Kabera and Heather Tate at FDA's National Antimicrobial Resistance Monitoring System (NARMS) for the help recruiting participating laboratories. We appreciate Maureen Davidson at the Division of Animal and Food Microbiology for supporting the project and Gordon Martin for assistance with the MLV study. Finally, we want to thank the FDA's Microbiological Methods Validation Subcommittee (MMVS) members for their critical review of the MLV study protocols and reports. The views expressed in this manuscript are those of the authors and do not necessarily reflect the official policy of the Department of Health and Human Services, the United States Food and Drug Administration, or the United States Government. Reference to any commercial materials, equipment, or process does not in any way constitute approval, endorsement, or recommendation by the Food and Drug Administration.


**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 Ge, Domesle, Yang, Hammack, Wang, Deng, Hu, Zhang, Hu, Lai, Chou, Dollete, Hirneisen, La, Richter, Rai, Yousefvand, Park, Wu, Eames, Kiang, Sheng, Wu, Hahn, Ledger, Logie, You, Slavic, Cai, Ayers, Young and Pamboukian. 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-10-00562 March 27, 2019 Time: 17:12 # 13

# A Rapid Method for Detection of *Salmonella* in Milk Based on Extraction of mRNA Using Magnetic Capture Probes and RT-qPCR

*Yalong Bai1,2 , Yan Cui1 , Yujuan Suo2 , Chunlei Shi1 , Dapeng Wang1 and Xianming Shi1 \**

*1 MOST-USDA Joint Research Center for Food Safety, School of Agriculture and Biology and State Key Laboratory of Microbial Metabolism, Shanghai Jiao Tong University, Shanghai, China, 2 Institute for Agri-food Standards and Testing Technology, Shanghai Academy of Agricultural Sciences, Shanghai, China*

#### *Edited by:*

*Om V. Singh, Technology Sciences Group Inc, United States*

#### *Reviewed by:*

*Dario De Medici, Istituto Superiore di Sanità (ISS), Italy Pendru Raghunath, Texila American University, Guyana*

> *\*Correspondence: Xianming Shi*

*xmshi@sjtu.edu.cn*

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 15 August 2018 Accepted: 26 March 2019 Published: 05 April 2019*

#### *Citation:*

*Bai Y, Cui Y, Suo Y, Shi C, Wang D and Shi X (2019) A Rapid Method for Detection of Salmonella in Milk Based on Extraction of mRNA Using Magnetic Capture Probes and RT-qPCR. Front. Microbiol. 10:770. doi: 10.3389/fmicb.2019.00770*

Magnetic separation is an efficient method for target enrichment and elimination of inhibitors in the molecular detection systems for foodborne pathogens. In this study, we prepared magnetic capture probes by modifying oligonucleotides complementary to target sequences on the surface of amino-modified silica-coated magnetic nanoparticles and optimized the conditions and parameters of probe synthesis and hybridization. We innovatively put the complexes of magnetic capture probes and target sequences into qPCR without any need for denaturation and purification steps. This strategy can reduce manual steps and save time. We used the magnetic capture probes to separate *invA* mRNA from *Salmonella* in artificially contaminated milk samples. The detection sensitivity was 104 CFU/ml, which could be increased to 10 CFU/ml after a 12 h enrichment step. The developed method is robust enough to detect live bacteria in a complex environmental matrix.

Keywords: amino-modified silica-coated magnetic nanoparticles, magnetic capture probes, *Salmonella*, RT-qPCR, milk

### INTRODUCTION

Magnetic nanoparticles and especially immunomagnetic nanoparticles have been widely used for foodborne pathogen detection (Escalante-Maldonado et al., 2015; Sun et al., 2015; Hwang et al., 2016; Luciani et al., 2016; Wang et al., 2016). The labeled antibody is a key point for a successful immunomagnetic detection method, and a limiting step is the quality of the anti-pathogen antibody used. The genus *Salmonella* is especially problematic because it has over 2,600 serotypes, so the probability of false negative may be high (Eng et al., 2015). In addition, the cost of superior anti-bacteria antibodies was always very high.

Developments in molecular biology, genomics, and bioinformatics now enable specific nucleotide sequences to be developed as barcodes for detection of target pathogens. Furthermore, the nucleotide sequence adjacent to the pathogen-specific barcode can also be used as a medium to purify the detection sequence. For example, magnetic nanoparticles labeled with complementary sequences were used to capture target DNA sequences containing barcodes of *Listeria monocytogenes* followed by amplification and identification by polymerase chain reaction (PCR) (Amagliani et al., 2004). They further used magnetic capture probes to simultaneously isolate *Salmonella* and *L. monocytogenes* DNA from seafood and detected the barcodes by triplex real-time PCR (Amagliani et al., 2010). However, since genomic DNA maintains a double helical structure, even when genomic DNA was denatured, the sensitivity would be compromised because of the large size of the genomic DNA.

Alternatively, the use of mRNA, which is single-stranded and much smaller than genomic DNA, for oligonucleotide hybridization may improve the procedure. The stability of DNA:RNA hybrids is also substantially greater than those of DNA:DNA duplexes (Casey and Davidson, 1977). Moreover, mRNA is considered as a more appropriate target than DNA to assess cell viability because mRNA have a short half-life, only a few minutes (Coutard et al., 2005; Liu et al., 2010), and is generally present only in viable cells (Garcia et al., 2010). That is, only viable cells could be detected using RNA-based detection methods (Zhou et al., 2014), which is the true harmful risk to food safety.

In previous research, we found nanoparticles affected PCR primarily *via* surface interactions with PCR components, and if the surface was blocked, the inhibition effect would be eliminated (Bai et al., 2015). Therefore, we proposed to directly add the complexes of magnetic capture probes and the captured target sequences in RT-qPCR to detect *Salmonella*, seeking to reduce operation steps and target losses, save time, and enhance sensitivity.

In this study, we prepared magnetic capture probes by modifying oligonucleotides complementary to target sequences on the surface of amino-modified silica-coated magnetic nanoparticles and optimized the conditions and parameters of probe synthesis and hybridization. We used the magnetic capture probes to separate *invA* mRNA, with the novel step of putting the complexes of magnetic capture probes and *invA* mRNA into RT-qPCR mixture without any denaturation and purification steps, to detect *Salmonella* in milk.

### MATERIALS AND METHODS

### Reagents

Glutaraldehyde, 4-(N-Maleimidomethyl) cyclohexane-1-carboxylic acid 3-sulfo-N-hydroxysuccinimide ester sodium salt (SMCC), Triton X-100, lysozyme, and proteinase K were purchased from Sigma-Aldrich (St. Louis, MO, USA). 2 × SYBR Green PCR mix was obtained from TaKaRa (Dalian, China). All preparations and measurements were carried out in sterilized Millipore water. PCR primer pairs and oligonucleotides were synthesized by Shanghai Biotech (Shanghai, China). The detailed information was shown in **Table 1**. All the oligonucleotides and primer pairs were designed for this study except the primers pair InvA-f/r used for *invA* of *Salmonella* that have been previously described (Jiang et al., 2013).

### Strains and Cultivation

The Guangdong Institute of Microbiology (Guangdong, China) provided *Salmonella enterica* (ATCC 13076). *Salmonella* was cultivated using Luria-Bertani medium (Becton Dickinson, MD, USA). Milk was obtained from a local dairy and tested negative for *Salmonella* by selective plating and PCR methods before use.

### Synthesis and Analysis of Amino-Modified Silica-Coated Fe3O4 Magnetic Nanoparticles

Fe3O4 magnetic nanoparticles, which were prepared using the co-precipitation method (Bai et al., 2015), were coated with silica and modified with amino groups by the reverse microemulsion method (Bai et al., 2016). ASMNPs morphologies were observed and analyzed by transmission electron microscopy (TEM) using a JEM-2010HT instrument (JEOL, Japan). The particles were sonicated for 1 min, and 10 μl of the solution was placed on a 200 mesh copper grid and then dried at room temperature. The grid was used for TEM analysis.

### Preparation of Magnetic Capture Probes

Two methods were used to label capture oligonucleotides on the ASMNPs surfaces based on the previous reports (Nam et al., 2004; Bruce and Sen, 2005). The schematic diagrams were shown in **Figure 1**.

*Method 1*: ASMNPs (2 mg) were dispersed in 1 ml phosphate buffered saline (PBS) buffer containing 200 μg SMCC, pH 7.4. After shaking for 8 h, the pellets were dispersed in 1 ml of Capture Oligonucleotide2 solution (10 μmol/L) and shaken overnight at room temperature. The pellets were blocked with 10% skimmed milk powder solution for 8 h and dispersed in


PBS buffer containing 2% BSA and 0.1% sodium azide. The final magnetic capture probe was named MP.

*Method 2*: ASMNPs (2 mg) were dispersed in 1 ml phosphate buffered saline (PBS) containing 5% glutaraldehyde (pH 7.4). The magnetic pellets were washed with PBS to remove free glutaraldehyde after shaking for 3 h, and the pellets were suspended in 500 μl PBS containing 200 μl Capture Oligonucleotide1 solution (10 μmol/L) and incubated overnight at room temperature. The magnetic pellets were blocked with 100 μl of 10% skimmed milk powder solution (Solarbio, Shanghai, China) and suspended in PBS containing 2% BSA and 0.1% sodium azide. The final magnetic capture probe was named MP′.

### Capture Using Magnetic Capture Probes

Samples were pretreated based on a modification method of published procedures (Amagliani et al., 2004; Bai et al., 2013). In brief, artificially contaminated milk samples (10 ml) were centrifuged at 6000 ×g for 20 min at 4°C, and the pellets were suspended in 1 ml of RNAprotect Bacteria Reagent (Qiagen, Germany). They were then incubated for 5 min and centrifuged at 9400 ×g for 10 min. The pellets were suspended in 50 μl of 50 mg/ml lysozyme, 50 μl of 20 mg/ml proteinase K, and 10 μl of Triton X-100 and incubated at 37°C for 15 min. Trizol (Invitrogen, Carlsbad, CA, USA) was added (1 ml) and the solution was incubated for 5 min at room temperature. Chloroform (250 μl) was added and the solution was vortexed for 15 s. After centrifugation at 9600 ×g for 10 min, the upper layer (aqueous phase) was transferred to a new 1.5 ml tube and incubated with 50 μg magnetic capture probes labeled with Capture Oligonucleotides3, which was partly complementary to *invA* mRNA, for 15 min at room temperature. The pellets were magnetically separated and mixed with 2.5 μl of 10X DNase buffer and 1 μl of DNase (Takara, Dalian, China) then incubated at 37°C for 20 min. The reaction was stopped by heating at 80°C for 2 min after the addition of 2.5 μl of 0.5 mol/L EDTA. The pellets were washed with DEPC-treated water and then used as templates in RT-qPCR.

### QPCR and RT-qPCR Amplification

Quantitative PCR (qPCR) was performed using 25 μl reaction volumes containing 1 μl of DNA template, 5 pmol of each primer, and 12.5 <sup>μ</sup>l of 2× SYBR® Green PCR master mix (TaKaRa, Dalian, China). PCR thermocycling was as follows: 2 min at 95°C, 40 cycles of 15 s at 95°C, 15 s at 60°C, and 20 s at 72°C. Amplifications were carried out using a Mastercycler® ep realplex instrument (Eppendorf, Germany). RT-qPCR was performed using a One Step PrimeScript RT-qPCR Kit (TaKaRa, Dalian, China) according to the manufacturer's instructions.

### RESULTS

### Optimization of Magnetic Capture Probes

The spherical amino-modified silica-coated Fe3O4 magnetic nanoparticles we synthesized appeared rough (**Figure 2**), different to the particles with smooth surfaces in previous reports (Bai et al., 2016). In our previous study, we found that this rough surface may be the reason for the large number of surface amino groups available for coupling (Bai et al., 2016). Using magnetic nanoparticles with many more amino groups was a basic strategy to maximize the capture efficiency of probes. Additionally, binding more capture oligonucleotides to the amino groups is another critical step.

There are two basic strategies to label the oligonucleotides on ASMNPs. One strategy is to use SMCC as a coupling agent to covalently immobilize the thiol-modified oligonucleotides (Capture Oligonucleotides1) on the surface of ASMNPs (the

FIGURE 2 | TEM image of ASMNPs synthesized for this study.

final magnetic capture probe was named MP); another strategy is to use glutaraldehyde as a coupling agent to covalently immobilize the amino-modified oligonucleotides (Capture Oligonucleotides2) to ASMNPs (the final magnetic capture probe was named MP′). The schematic diagrams were shown in **Figure 1**. We used Long Oligonucleotides1 whose 3′ end was complementary to Capture Oligonucleotides1 as target to compare the capture efficiency of these two types of probes.

Twenty micrograms of each of these two magnetic probes were used to capture the same amount of target (1 ml of Long Oligonucleotides1). And then the magnetic pellets were used as DNA templates for qPCR. The results of qPCR (**Figure 3**) showed that the magnetic capture probes prepared using SMCC had higher separation efficiency (*n* = 3, *p* < 0.05), and combined with the calibration curve (y = −3.1654x + 40.455, *R*<sup>2</sup> = 0.9923) which was established based on Ct values of qPCR using a set of Long Oligonucleotides1solutions of known concentration, the capture capability of MP was 67.4 times than MP′. Therefore, we used the probes based on SMCC-strategy in the following experiments.

### Evaluation of the Amount of Oligonucleotides Immobilized on the Surface of ASMNPs

It is inaccurate to evaluate the number of capture sequences immobilized on the surface of ASMNPs by counting the number of the captured target ssDNA because not all of the capture sequences would hybridize with the target sequences. To obtain more direct data, we immobilized the longer thiol-modified oligonucleotides (Long Oligonucleotides2) on the surface of ASMNPs. We could therefore roughly estimate the amount of immobilized capture sequences by qPCR. Twenty micrograms of magnetic capture probes were used as templates for qPCR resulting in a Ct value of 23.39 (*n* = 3). According to the calibration curve (y = −5.564x + 86.311, *R*<sup>2</sup> = 0.9988) which was obtained based on the serial dilution of Long Oligonucleotides 1, the amount of capture sequences was 1011.3 copies per 20 μg of magnetic capture probes. In general, the maximum amount of pathogen in culture media could reach 109 CFU/ml. Therefore, under ideal conditions and regardless of the hybridization rate and the recovery of the magnetic capture probes, even 20 μg of magnetic capture probes would be sufficient to separate the maximum amount of pathogenderived nucleic acid. In order to further increase the probability of capture, 50 μg was used in the practical application.

### Effects of Magnetic Capture Probes on Polymerase Chain Reaction

In previous studies, target sequences were always denatured from magnetic capture probes and tediously purified before PCR (Jacobsen and Holben, 2007). In an attempt to optimize sensitivity and detection speed, we planned to directly add the complexes of probes and target sequences in qPCR as templates. However, firstly we needed to identify whether the magnetic capture probes would inhibit qPCR. We varied the amounts to determine the maximum that we could add without inhibiting qPCR. Magnetic capture probes (0, 20, 40, 60, 80, and 100 μg) were added to qPCR mixtures. Though maximum fluorescence decreased with an increase of magnetic capture probes added, probably because the magnetic probes quenched part of fluorescence of SYBR Green, the Ct values were unaffected by the addition of 20, 40, and 60 μg (*n* = 3; *p* > 0.1). At the higher levels (80 and 100 μg), the Ct values increased slightly (*n* = 3, *p* > 0.05; **Figure 4**). The results showed that when the amount of magnetic capture probes added in qPCR was under 60 μg, the amplification was not affected.

### Effect of pH on Hybridization Rate

The pH of hybridization systems may vary with sample type so we investigated whether pH affected the hybridization rate between the capture sequences and target sequences. Before hybridization, the solutions containing the same amount of Long Oligonucleotides1 were adjusted to different pH with sodium hydroxide and hydrochloric acid. After hybridization and magnetic separation, the pellets were washed with TE buffer and then used as qPCR templates. The hybridization was severely affected only at low pH (pH 3), and there were only slight effects at other levels (**Figure 5**). That is, even when the solution was treated with Trizol (pH 5), the hybridization would not be much affected.

### Capture Capacity of Magnetic Capture Probes for Long Oligonucleotides1

In order to evaluate the capture capacity of the magnetic capture probes for isolating the target sequences, we used the Long Oligonucleotides1 as a model. These contained sequences for both hybridization capture and qPCR detection. The oligonucleotides were serially diluted 10-fold and 20 μg of magnetic capture probe was hybridized with the targets. All recovery rates were near 50% except for the solution whose original concentration was 108 copies/ml (recovery rate = 22%) (*n* = 3, *p* < 0.05; **Figure 6**). Although 20 μg of magnetic capture probes might contain more than 1011.3 copies of capture sequences based on the previous experiments, they were not sufficient for 108 copies of target sequences. The most probable reason for these results was steric hindrance.

### Detection of *Salmonella* in Milk

We used magnetic capture probes labeled with Capture Oligonucleotides3, which was partly complementary to *invA* mRNA, to separate the *invA* mRNA in the milk contaminated

artificially with 10-fold diluted *Salmonella* and then detected the *invA* mRNA by RT-qPCR. The schematic diagram was shown in **Figure 7**. The milk was processed according to 2.5; in this case, the target mRNA was released into the solution. When the magnetic capture probes were added, the labeled sequences would hybridize with *invA* mRNA. After magnetic separation and rinse, the magnetic pellets were used to do

contaminated milk.

TABLE 2 | Detection of *Salmonella* in artificially contaminated milk.


TABLE 3 | Specificity of the method based on extraction of mRNA using magnetic capture probes and RT-qPCR.


the RT-qPCR assay. The detection limit was 104 CFU/ml and log-linear relationships occurred from 104 to 107 CFU/ml (**Figure 8**).

Alternatively, we could enrich the bacteria to get a higher sensitivity. Milk (25 ml) contaminated with *Salmonella* was combined with 225 ml of Luria-Bertani broth and incubated at 37°C for 12 h, then we processed 10 ml of the solution as for the milk samples above. The detection limit reached 10 CFU/ml with a detection rate of 100% (**Table 2**).

In addition, 28 bacteria were used to determine the specificity of this method based on extraction of mRNA using magnetic capture probes and RT-qPCR (**Table 3**). Eighteen *Salmonella* (104 CFU/ml) which was separated from the food samples and identified by our labs were tested, and all showed positive results. Non-*Salmonella* bacteria (10 genera, 106 CFU/ml) showed negative results (no signal before 32 Ct). The results were expected because *invA* gene were proved previously many times to be a specific gene for *Salmonella* (Wang et al., 2015; Pande et al., 2016), and most of all, in this method, the target mRNA were separated firstly by using magnetic probes; thus, the specificity would be better than those methods in which all of the mRNA was used to convert into cDNA.

### DISCUSSION

*Salmonella* is the leading cause of bacterial food poisoning in humans worldwide (Zhang et al., 2016; Rahman, 2017). It is reported that more than 90% of human illness caused by *Salmonella* is foodborne and results from contaminated meat, eggs, and milk (Foley and Lynne, 2008). Thus, more rapid and reliable methods for the detection of *Salmonella* are required except the traditional culture methods which take 4–7 days. For culture-independent methods, magnetic separation has been widely used to enrich the targets to realize rapid detection (Brandão et al., 2015). In our previous research, we found ASMNPs could adsorb DNA by hydrogen bond and electrostatic interaction, and thus, we used ASMNPs to separate bacterial genomic DNA and combined it with PCR to rapidly detect *Salmonella* Enteritidis *and L. monocytogenes* (Bai et al., 2013). However, pathogen detection *via* DNA does not differentiate between viable and dead bacteria because DNA from non-viable bacteria also could produce signal. Since mRNA is very labile, it is considered that detection of mRNA is superior to detection based on DNA. Therefore, in this study, we prepared magnetic capture probes to separate mRNA sequences by hybridization and using RT-qPCR in which complexes of probes and target sequences were directly used as template to detect *Salmonella*.

For improving the separation capability of the magnetic capture probes, on the one hand, we prepared the magnetic nanoparticles with rich amino groups based on previous research (Bai et al., 2016); on the other hand, we selected a superior method, SSMCbased strategy, to couple the oligonucleotides with amino groups to make sure the probes contain more capture sequences.

Further, to reduce the operating steps, save time, and improve sensitivity, we tried to directly add the complexes of magnetic capture probes and target sequences to PCR. In previous research, we found that the bare ASMNPs would inhibit PCR

### REFERENCES

by adsorbing PCR components and the amplification would be normal after the surface of ASMNPs were blocked. The magnetic capture probes were also blocked by capture sequences and proteins; thus, appropriate amount of probes could not affect PCR amplification. Moreover, the magnetic capture probes separated mRNA sequences would work in the solution containing Trizol (pH 5.0) in order to simplify the mRNA isolation steps. In this case, the effect of pH on the hybridization rate should be evaluated. After comparing different pH conditions, we found there was only slight effect for hybridization at pH 5.

The magnetic capture probes were used to isolate the *invA* mRNA in milk artificially contaminated with *Salmonella*, and the *invA* mRNA was then detected by RT-qPCR. The detection limit was 104 CFU/ml (about 30 Ct). At the same time, we noted that a Ct value (about 32) existed with no *Salmonella* added to the milk. Though we could confirm this was caused by the magnetic capture probes, the detailed reason is still not clear and needed to be explored in future research. That is, if we clearly know that why the magnetic capture probes caused a false positive signal at 32 cycles in RT-qPCR without target mRNA, maybe we could realize a lower detection limits. But even so, the current detection limit was superior. Fey et al. reported that the detection limits were 5 × 104 and 5.5 × 104 copies (*invA* gene) in drinking and pond water, respectively (Fey et al., 2004). In this research, the sample was milk which was more complex than water. Most of all, this method was more rapid and simple to extract target mRNA. Alternatively, after cultivation for 12 h, the detection rate was 100% even though the milk contained only 10 CFU/ml of *Salmonella*. This result was also superior to our previous research extracting mRNA based on the traditional method. In our previous research (Zhou et al., 2014), the samples must be cultivated for 18 h to detect 10 CFU/ml of *Salmonella*.

### AUTHOR CONTRIBUTIONS

YB designed and initiated the study, interpreted the results, and wrote the manuscript. YC and XS contributed to improvement of the manuscript. YS, CS, and DW contributed to analysis of the data and discussion of the results. XS designed the outline of this study and manuscript and provided laboratory equipment and space.

### FUNDING

This study was supported by the National Key R&D Program of China (grant number 2016YFE0106100) and Shanghai Municipal Natural Science Foundation (grant number 18ZR1425500).

Amagliani, G., Brandi, G., Omiccioli, E., Casiere, A., Bruce, I., and Magnani, M. (2004). Direct detection of *Listeria monocytogenes* from milk by magnetic based DNA isolation and PCR. *Food Microbiol.* 21, 597–603. doi: 10.1016/j.fm.2003.10.008

Amagliani, G., Omiccioli, E., Brandi, G., Bruce, I. J., and Magnani, M. (2010). A multiplex magnetic capture hybridisation and multiplex real-time PCR protocol for pathogen detection in seafood. *Food Microbiol.* 27, 580–585. doi: 10.1016/j.fm.2010.01.007

Bai, Y., Cui, Y., Paoli, G. C., Shi, C., Wang, D., and Shi, X. (2015). Nanoparticles affect PCR primarily via surface interactions with PCR components: using

amino-modified silica-coated magnetic nanoparticles as a main model. *ACS Appl. Mater. Interfaces* 7, 13142–13153. doi: 10.1021/am508842v


hybridization RT-PCR. *J. Microbiol. Methods* 69, 315–321. doi: 10.1016/j. mimet.2007.02.001


**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 Bai, Cui, Suo, Shi, Wang and Shi. 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.*

# Antimicrobial Resistance Diversity Suggestive of Distinct Salmonella Typhimurium Sources or Selective Pressures in Food-Production Animals

#### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Pimlapas Leekitcharoenphon, Technical University of Denmark, Denmark Moussa S. Diarra, Agriculture and Agri-Food Canada (AAFC), Canada

#### \*Correspondence:

Kate C. Mellor kmellor5@rvc.ac.uk Alison E. Mather alison.mather@quadram.ac.uk

†Present address:

Alison E. Mather, University of East Anglia, Norwich, United Kingdom

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 25 January 2019 Accepted: 21 March 2019 Published: 12 April 2019

#### Citation:

Mellor KC, Petrovska L, Thomson NR, Harris K, Reid SWJ and Mather AE (2019) Antimicrobial Resistance Diversity Suggestive of Distinct Salmonella Typhimurium Sources or Selective Pressures in Food-Production Animals. Front. Microbiol. 10:708. doi: 10.3389/fmicb.2019.00708 Kate C. Mellor1,2 \*, Liljana Petrovska<sup>3</sup> , Nicholas R. Thomson2,4, Kate Harris<sup>3</sup> , Stuart W. J. Reid<sup>1</sup> and Alison E. Mather<sup>5</sup> \* †

<sup>1</sup> Royal Veterinary College, Hatfield, United Kingdom, <sup>2</sup> London School of Hygiene & Tropical Medicine, London, United Kingdom, <sup>3</sup> Animal and Plant Health Agency, Weybridge, United Kingdom, <sup>4</sup> Wellcome Trust Sanger Institute, Wellcome Genome Campus, Hinxton, United Kingdom, <sup>5</sup> Quadram Institute Bioscience, Norwich, United Kingdom

Salmonella enterica subsp. enterica serovar Typhimurium is a common cause of enterocolitis in humans globally, with multidrug resistant (MDR) strains posing an enhanced threat. S. Typhimurium is also a pathogen in food-production animals, and these populations can act as reservoirs of the bacterium. Therefore, surveillance and control measures within food-production animal populations are of importance both to animal and human health and have the potential to be enhanced though improved understanding of the epidemiology of S. Typhimurium within and between food-production animal populations. Here, data from Scotland and national surveillance England and Wales data for isolates from cattle (n = 1115), chickens (n = 248) and pigs (n = 2174) collected between 2003 and 2014 were analyzed. Ecological diversity analyses and rarefaction curves were used to compare the diversity of observed antimicrobial resistance (AMR) profiles between the host species, and within host species populations. Higher AMR profile diversity was observed in isolates from pigs compared to chickens across diversity measures and isolates from cattle for three of four diversity measures. Variation in AMR profile diversity between production sectors was noted, with higher AMR diversity of isolates from broiler compared to layer chickens, breeder compared to rearer and finisher pigs and beef compared to dairy cattle. Findings indicate variation in AMR profile diversity both within and between food-production animal host species. These observations suggest alternate sources of AMR bacteria and/or variation in selective evolutionary pressures within and between food-production animal host species populations.

Keywords: antimicrobial resistance, ecological diversity, surveillance, Salmonella Typhimurium, food-production animals

**Abbreviations:** AMR, antimicrobial resistance; APHA, Animal and Plant Health Agency; BP, Berger Parker; MDR, multidrug resistance; NCP, National Control Program; SD, Simpsons diversity; SE, Shannon entropy; SR, species richness.

### INTRODUCTION

fmicb-10-00708 April 11, 2019 Time: 17:43 # 2

Globally, drug resistant infections are projected to cause 10 million human deaths at a cost of 100 trillion USD annually by 2050 if current trends continue (O'Neill, 2014). Selection pressure caused by the widespread use of antimicrobials in human medicine, veterinary medicine and agriculture has increased the dissemination and prevalence of AMR in bacteria (Laxminarayan and Heymann, 2012). Improving our understanding of the emergence and spread of AMR bacteria within and between host species populations is essential to inform effective control policies to prevent or reduce dissemination.

In the EU, over 100,000 cases of enterocolitis, costing an estimated €3 billion annually, are attributed to nontyphoidal Salmonella infections, of which Salmonella enterica subsp. enterica serovar Typhimurium is the second most common serovar (European Food Safety Authority and European Centre for Disease Prevention and Control, 2017). Salmonella Typhimurium is a zoonotic pathogen and the primary reservoir is thought to be food-production animals, with the majority of human cases deriving through the food chain (Majowicz et al., 2010), although more recent studies have suggested a more nuanced situation (Mather et al., 2013). Compared to antimicrobial sensitive strains, those resistant to therapeutically relevant antimicrobials pose a greater threat to public health because they are associated with higher morbidity and mortality rates (Helms et al., 2002; Cosgrove, 2006; Depuydt et al., 2008). Multi-drug resistant (MDR) strains, most notably definitive type (DT)104, have been documented to disseminate globally, causing infections in multiple host species including humans and food-production animals (Leekitcharoenphon et al., 2016).

National surveillance of S. Typhimurium in United Kingdom animal populations is conducted primarily for outbreak identification control purposes to limit the public health risk. Mandatory active surveillance is conducted for poultry, whilst passive surveillance is conducted for other food producing animals (DEFRA, 2007, 2008, 2009). Passive surveillance relies upon submission of samples by veterinarians or farmers for clinical diagnostic purposes. Identification of Salmonella from an animal source is reportable to the APHA where confirmation and phenotypic antimicrobial susceptibility testing and phage typing are conducted (Zoonoses Order, 1989; Animal and Plant Health Agency, 2017). Phenotypic AMR profile may not consistently correlate with bacterial genetic lineage, due to the importance of horizontal gene transfer for the transmission of AMR, and phage types can be polyphyletic (Chen et al., 2005; Petrovska et al., 2016). The resolution afforded through analysis of phage type and AMR phenotypic profiles is therefore restricted, but the combination of phage type and phenotypic AMR profile have proved useful for distinguishing between isolates for outbreak detection purposes.

Variation in the prevalence of resistance of bacterial pathogens to individual antimicrobials between food-production animal groups has been attributed to differences in selective pressures between production systems and dissemination of AMR clones, e.g., DT104. However, our understanding of the evolution and dissemination of resistant bacteria within and between foodproduction animal populations is limited (Young et al., 2009). AMR profiles unique to individual host species would not be expected if there was transfer of strains between host species populations. The rate of transfer of strains between host species could be influenced by ecological or pathogen related barriers which could result in differences in the observed AMR profiles between host species (Rabsch et al., 2002). The presence of shared AMR profiles between multiple host species can arise through a common source of infection, transmission of bacteria or mobile genetic elements (MGEs) carrying AMR genes between the host species populations or through independent emergence or acquisition of resistance determinants.

Opportunities for transfer of non-host-restricted strains between host species are numerous and include mixed-species farms, markets, common grazing, contaminated feed products and movements of fomites, personnel or wildlife (Davies and Wray, 1997; Robinson and Christley, 2007; Skov et al., 2008). The risk of transfer between host species is likely to vary between food-production animal species due to the differences in industry structure and standard biosecurity protocols (Lindström et al., 2012). Enhanced surveillance and control measures for United Kingdom poultry through the National Control Program (NCP) were introduced to reduce the prevalence of Salmonella Typhimurium and Salmonella Enteritidis on farms (DEFRA, 2007, 2008, 2009). Combined with biosecurity practices and a pyramidal industry structure, the United Kingdom S. Typhimurium population in chickens could be predicted to be relatively isolated from S. Typhimurium circulating in other host species. However, free-range poultry could be exposed to Salmonella transferred by wild birds and flies (Wales et al., 2010; Andrés et al., 2013).

Variation in AMR profile diversity within or between host species could indicate a greater diversity of AMR bacteria sources entering a host species population or indicate continuous evolution in the host population in response to greater selective pressures. Additionally, distinct differences in AMR profiles between host species could indicate epidemiological barriers to transmission. Here, the available national surveillance data for S. Typhimurium isolates, collected over an 11-year period, have been used to determine whether or not there are detectable differences in circulating AMR profiles within and between S. Typhimurium isolates from cattle, chicken and pig populations.

### MATERIALS AND METHODS

### Data

Routine surveillance samples from cattle, chicken and pigs submitted to the APHA between 2003 and 2014, which had a recorded AMR profile, were eligible for inclusion in the study. The majority of samples were from England, with the remaining samples from Wales (n = 168, 4.8%) and Scotland (n = 40, 1.1%). United Kingdom region information was unavailable for 230 isolates (6.5%). Salmonella is a reportable pathogen in both animals and humans in the United Kingdom; samples are

typically submitted for clinical diagnostic purposes to regional laboratories. Samples from chickens are also collected through the National Control Program which requires sampling of commercial flocks at predetermined points in the production cycle (DEFRA, 2007, 2008, 2009). Due to the nature of the sample collection, NCP isolates are likely to be more representative of the microbial population in healthy birds compared to passive surveillance isolates obtained from symptomatic birds. Samples from food-producing animals which test positive for Salmonella are submitted to the APHA for confirmation and antimicrobial susceptibility testing in accordance with the Zoonoses Order (1989).

At the APHA, phage-typing was conducted (Anderson et al., 1977) and antimicrobial susceptibility testing performed using a disk diffusion technique (Animal and Plant Health Agency, 2017). Antimicrobials tested and breakpoints for classification as sensitive or resistant are detailed in **Table 1**. Both breakpoints and antimicrobial disk concentrations changed during the period

TABLE 1 | Antimicrobials used for resistance phenotyping of Salmonella Typhimurium isolates at the Animal and Plant Health Agency between 2003 and 2014.


Antimicrobial susceptibility testing was performed using the British Society for Antimicrobial Chemotherapy disk diffusion technique, with isolates classified as resistant or sensitive to the antimicrobial according to the zone size cut-off values. <sup>∗</sup>Testing for sensitivity to cefoperazone was replaced with cefotaxime in 2004. <sup>+</sup>10 µg until 2008. ♦from 2004 onwards. •2003 only. ◦25 µg from 2003 to 2008. Until 2007 a 13 mm breakpoint was used, with classification if growth inhibition zone ≤ 13 mm, BSAC recommended breakpoints were introduced for amikacin, cefotaxime, ceftazidime and ciprofloxacin from 2007 onward, and for gentamicin, sulphamethoxazole/trimethoprim, amoxicillin/clavulanic acid and chloramphenicol from 2008 onward. The historical veterinary breakpoints were used for the remaining antimicrobials throughout the period of study.

2003 to 2014 for multiple antimicrobials. Up to 2007 a 13 mm breakpoint was used, with classification as resistant if growth inhibition zone ≤ 13 mm. British Society for Antimicrobial Chemotherapy (BSAC) recommended breakpoints were adopted for amikacin, cefotaxime, ceftazidime and ciprofloxacin from 2007 onward, and subsequently for gentamicin, sulphamethoxazole/ trimethoprim, amoxicillin/clavulanic acid and chloramphenicol (British Society for Antimicrobial Chemotherapy, 2013). The historical veterinary breakpoints were used for the remaining antimicrobials, for which no BSAC breakpoints are available, throughout the period of study (Veterinary Medicines Directorate, 2013). Additionally, some changes to antimicrobials tested were made, these are detailed in **Table 1**.

### Diversity Analysis

Ecological measures of AMR phenotype diversity were calculated in R (R Core Team, 2016) using the "Vegan" package (Oksanen et al., 2011). Similar to the analysis described by Mather et al. (2012), four diversity measures were calculated: SR, SE, SD and reciprocal BP with 95% confidence intervals generated through resampling. Each measure differentially weights the importance of SR and species abundance. In this study, a "species" is defined as a unique AMR profile, including the profile corresponding to susceptibility to all tested antimicrobials. The AMR profile (or antibiogram) of an isolate is the combination of AMR phenotype (susceptible or non-susceptible) to each drug tested. SR reflects the richness of AMR profiles without weighting of abundance. SE and SD are measures in which both SR and relative abundance are taken into account. The BP diversity index is related to the proportion of isolates with the most common AMR profile.

Changes in antimicrobial testing protocol could impact upon results, particularly where the proportion of isolates from each host vary between time periods with different testing protocols. Four different 'testing periods' were identified within which antimicrobial sensitivity testing protocol was consistent (2003; 2004–2006; 2007; 2008–2014). As calculated diversity indices are dependent on sample size, for each 'testing period' the host species groups with larger sample sizes were randomly subsampled to the size of the smaller host species group. For each host species the combined data for the four testing periods was used as the input for the diversity analyses. Following 10,000 iterations the mean of iterations and confidence intervals of diversity indices were then calculated. The exponent of SE values and the reciprocal of SD and BP values were calculated to convert diversity indices into the effective number of profiles prior to plotting of results. The diversity measures were deemed to differ if the 95% confidence intervals of the values of two host species did not overlap.

Diversity analyses were conducted to compare between host species, and within host species based upon production type metadata.

### Network Analysis

The network of connectivity of AMR profiles in the host species was visualized using the "igraph" package (Csárdi and Nepusz, 2006) in R. The edges represent where a single change in the AMR profile occurred between two AMR profiles (nodes). Nodes are colored according to the host species, or host species combination.

### Rarefaction Curves

fmicb-10-00708 April 11, 2019 Time: 17:43 # 4

Rarefaction analyses were performed using the "iNEXT" package in R (Hsieh et al., 2018), which examines the SR of phenotypic profiles at each sample size. The rarefaction curves enable evaluation as to whether the total AMR profile diversity was captured by the sampling, and comparison of AMR profile diversity between host species. To account for changes in antimicrobial sensitivity testing procedure data is presented for the four time periods for which testing procedures were identical across the host species.

### Common AMR Profiles and Associated Phage Types

For each of the individual host species, the top five most frequent AMR profiles, and association with phage types (most common and total number) were determined and compared between host species.

### Compare AMR Profiles Between Host Species

All analyses were processed in R. To visualize AMR profiles common to multiple hosts and unique to individual hosts, the "eulerr" package was used to create a proportional Venn diagram of ellipses (Larsson, 2018).

### Comparison of Observed and Expected Numbers of AMR Profiles Shared by All Host Species

The expected number of AMR profiles shared by all host species was compared to the observed number. A subsample to the size of the smallest host species group for each 'testing period' without replacement was generated for each host species and data joined prior to source randomization for each isolate without replacement and the number of AMR profiles common to all host species recorded of the 10,000 iterations. The observed distribution was generated without source randomization. The observed number of phenotypic AMR profiles common to all host species groups (mean of 10,000 iterations with subsampling) was considered to be significantly different to the expected number if falling in the last or first 2.5th percentile of the distribution of expected number, equivalent to a two tailed at p < 0.05.

### Comparison of Observed and Expected Numbers of AMR Profiles Unique to Individual Host Species

The distributions of observed numbers of AMR profiles unique to each host were generated using subsampling to the smallest host species group for each time period without replacement. The number of AMR profiles unique to individual hosts were recorded for each of the 10,000 iterations.

### RESULTS

Between 2003 and 2014 a total of 3537 isolates (1115 from cattle, 248 from chickens and 2174 from pigs) were submitted to the APHA. The majority (95.8%) of submissions were obtained though passive surveillance with the remainder obtained through the NCP active surveillance system for poultry. Phenotypic antimicrobial test data were available for all except one isolate which was excluded from analysis. The data for these 3537 samples were used for the general analysis. A summary table of the antimicrobial resistance profiles observed in each host species is available in **Supplementary Table 1**. Production level data were available for 920 (82%) of isolates from cattle, 233 (94%) of isolates from chickens and 1180 (54.3%) of isolates from pigs.

Some isolates were collected on the same day from the same farm, as indicated by a shared submission reference. The proportions of isolates with shared submission reference varied between the host species 24.6% (274/1115) for cattle, 26.6% (66/248) for chickens and 48.9% (1062/2174) for pigs. The submission references indicate whether multiple isolates have been collected from the same farm on the same day, however, whether the isolates are from the same animal group or different animal groups on the farm is unknown. Multiple AMR profiles of isolates with a shared submission reference have been observed and these samples may have been taken from different animal groups (**Table 2**), however, this metadata was not available for analysis in this study. Isolates with the same submission reference which share an AMR profile could be the same strain, however, this cannot be determined given the resolution of the data.

### Comparison of AMR Profiles Between Host Species

A total of 129 AMR profiles, including full susceptibility, were observed amongst isolates from cattle, pigs and chickens, 18 (14%) of which were observed in all three host species groups; these 18 AMR profiles represented 89.9% of isolates from cattle, 96.8% of isolates from chickens and 76.6% of isolates from pigs (**Figure 1**). Separation between observed and expected distributions of AMR profiles common to all host species was observed (**Figure 2**).

TABLE 2 | Isolates with shared submission references indicating collection at the same farm location on the same day.


A total of 95 profiles were observed to be unique to a single host species. Comparison of the absolute number of observed profiles between host species groups does not account for differences in sample size for each host species group. Rarefaction curves were therefore used to compare SR of phenotypic AMR profiles between host species (**Figure 3**), showing a higher diversity of profiles observed from pigs compared to chickens despite the low sample number from chickens. The rarefaction curves also indicate that the full AMR profile diversity of isolates from cattle are closer to being captured than for isolates from pigs. The percentage of isolates with an AMR profile unique to the host species was almost 10-fold higher in isolates from pigs compared to chickens (**Table 3**). After controlling for sample size, distinct separation of the observed distribution of AMR profiles unique to individual hosts can be seen between isolates from chickens and pigs but overlap with distributions of observed number of AMR profiles of isolates from cattle and expected number of AMR profiles for a single host species (**Figure 4**).

Three out of the five most common AMR profiles are shared by all host species groups (**Table 4**). Apart from sensitive isolates, all top five most common AMR profiles include tetracycline resistance. The remainder of the profiles are resistance to ampicillin, streptomycin, sulphonamides and tetracycline ± resistance to chloramphenicol and/or trimethoprim-sulphonamides.

A greater proportion of isolates (42.7–79.8%) with a common MDR AMR profile, defined as resistance to three or more antimicrobial classes, are associated with a single phage type compared to isolates resistant to tetracycline only or sensitive to all tested antimicrobials where ≤ 25% of isolates are associated with a single phage type (**Table 4**). The AMR profile and most commonly associated phage type combination was consistent across host species for some AMR profiles (AmCSSuT and AmSSuT), but not others (AmCSSuSxtT and T). The most common phage type associated with the AmCSSuT profile was DT104 across all host species groups, although the percentage of isolates classified as DT104 varied from 42.7% of isolates from pigs to 76.9% of isolates from chickens. Phage type DT193 was associated with 2/5 of the most frequent AMR profiles for each of the food-production animal species groups, however, the associated AMR profiles varied between foodproduction animal species.

Red solid vertical line: mean expected number of AMR profiles common to all hosts (11.87). Bandwidth 0.5.

TABLE 3 | Summary of prevalence of resistance to one or more antimicrobial, multi-drug resistant (MDR), and comparison of AMR profiles observed for each host species.


MDR is defined as phenotypic resistance to three or more classes of antimicrobials.

### Network Analysis

The vast majority of AMR profiles are connected in a single complex (87.6%, 113/129), a further seven isolates form a small complex, two isolates are connected to one another, and seven (of which six are from pigs) isolates are not connected to any other isolate (**Figure 5**). The mean node degrees were significantly higher for AMR profiles observed in multiple host species, compared to AMR profiles observed in a single host species (Inset table, **Figure 5**).

### Ecological Diversity Analysis Diversity of AMR Profiles of Isolates From Cattle, Chicken and Pig Host Species

Ecological diversity calculations were performed to compare the diversity of phenotypic AMR profiles of isolates from cattle, pig and chickens (**Figure 6**). Across the diversity measures (SR, SE, SD, and BP) isolates from pigs had higher AMR profile diversity than those from cattle or chickens. The SR of AMR profiles of chickens isolates from was greater than isolates from cattle.

### Diversity of AMR Profiles of Isolates From Broiler and Layer Chickens

A total of 41 isolates from breeders, 68 isolates from broilers, 112 isolates from layers and 27 isolates from an unspecified production type were submitted. The diversity of AMR profiles of isolates from broilers and layers were compared. Across all diversity measures, the effective number of AMR profiles of isolates from broilers were higher than for isolates from layer chickens (**Figure 7**). Six AMR profiles were common to isolates from both broilers and layers, eight AMR profiles from broilers only and three AMR profiles identified in layers only. The percentage of isolates sensitive to all 16 tested antimicrobials was almost three-fold higher (66.1%) for isolates from layers compared to isolates from broilers (23.5%).

### Diversity of AMR Profiles of Isolates From Breeding, Rearing and Finishing Pigs

Isolates from breeding pigs had a higher diversity of AMR profiles compared to isolates from rearing pigs across all diversity measures, and compared to isolates from finishing pigs for SR and Shannon Entropy diversity measures (**Figure 8**). Analysis was based on 548 isolates from breeders, 595 from finishers, and 737 from rearing pigs.

### Diversity of AMR Profiles of Isolates From Dairy and Beef Cattle

Production sector data was available for the majority of isolates from cattle; n = 259 isolates were from beef cattle, and n = 808 isolates were from dairy cattle. Across the diversity measures (SR, SE, SD, and BP) AMR profiles of isolates from beef cattle were more diverse than isolates from dairy cattle (**Figure 9**).

## DISCUSSION

Salmonella Typhimurium is a common cause of foodborne illness globally, with MDR strains of particular public health concern due to association with poorer patient outcomes. Reduction of prevalence in the food-production animal reservoir is important for both animal and human health and improving understanding of S. Typhimurium epidemiology including AMR dynamics in the food-production animal populations and has the potential to inform interventions. Using national surveillance data collected over an 11-year time period, AMR profiles of S. Typhimurium were compared within and between food-production animal host species populations, as variation in observed AMR profiles and AMR profile diversity could indicate variation in AMR dynamics between the host species.

Active surveillance samples from chickens from 2007 onward were included, however, the majority of samples were obtained through passive surveillance and the data are therefore biased toward strains causing clinical disease in food-production animals or identified through diagnostic testing for other diseases. Asymptomatic/subclinical infection is common in some animal species and the full diversity of circulating S. Typhimurium at the population level is therefore unlikely to be captured (Rostagno et al., 2012; Troxell et al., 2015). Submissions are also influenced by variation in farmer and veterinarian behavior (Schwaber et al., 2004; Laupland et al., 2007; Rempel and Laupland, 2009). Passive surveillance systems have been found to capture

#### TABLE 4 | Top five most frequent AMR profiles for each host species.


Am, ampicillin; C, chloramphenicol; S, streptomycin; Su, sulphonamides; Sxt, trimethoprim-sulphonamides; T, tetracycline. Sensitive, sensitive to all tested antimicrobials. AMR profiles are colored.

FIGURE 5 | Network connectivity diagram of phenotypic antimicrobial resistance profiles. Nodes represent phenotypic antimicrobial resistance profiles. Edges represent loss or acquisition of resistance to a single antimicrobial. Inset table: mean node degrees and betweenness scores for AMR profiles observed in one, two, or three host species.

greater diversity of rare AMR profiles of Salmonella in food-production animals in Canada (Mather et al., 2016), and clinical isolates from food-production animals have been observed to have higher diversity of AMR phenotypes compared to asymptomatic food-production animals (Perron et al., 2007; Afema et al., 2015). Utilizing the national surveillance data should enable meaningful comparison of AMR profile diversity between and within food-production animal populations, although variations in surveillance between host populations could also contribute to any observed differences.

Ecological diversity analyses revealed variation in observed AMR profile diversity both between host species and between production types for chickens and pigs. The AMR profile diversity of isolates from pigs was greater than for isolates from chickens and cattle across SE, SD and BP diversity measures. The percentage of isolates resistant to one or more antimicrobials is almost two-fold higher for pigs (98.1%) compared to chickens (52.4%). This may in part account for the higher ecological diversity measures of isolates from pigs compared to chickens. The observation of 15.9% of isolates from pigs having an AMR profile unique to pigs suggests that some strains causing clinical disease in pigs are either not present, do not persist or are not causing clinical disease in cattle or chicken populations. The distinct separation of the distribution of the observed number of AMR profiles unique to pigs from that of chickens, controlling for sample size, indicates this finding is unlikely due to differential sampling intensity. Inference is limited due to the nature of passive surveillance; however, these observations suggest alternate sources of AMR bacteria and/or variation in selective evolutionary pressures between host species populations.

FIGURE 8 | Observed ecological diversities of phenotypic antimicrobial resistance profiles of S. Typhimurium isolates from breeder, finisher, and rearer pigs.

The separation of expected and observed distributions of AMR profiles shared by all food-production animal populations indicates that interchange of S. Typhimurium strains between host species is not complete, with differences in the circulating S. Typhimurium population between host species groups. Variation in AMR profile diversity between host species could be attributable to differences in observed lineages between host species. Bacterial typing information was limited to phage type, which affords limited resolution and does not necessarily correlate with lineage; multiple lineages can have the same phage type and AMR profile combination, and phage switching can occur (Baggesen et al., 2010; Pang et al., 2012). The consistency in AMR profile AmCSSuT and phage type DT104 combination could indicate the presence of a common single lineage between multiple host species; this particular AMR profile is also known to be chromosomally encoded in DT104 (Boyd et al., 2001). However, the most common phage type was inconsistent across host species for other AMR profiles including T and AmCSSuSxtT, suggesting different genetic lineages with the same phenotypic AMR profile may be circulating in different host species.

The selective pressures driving the maintenance of high proportions of MDR S. Typhimurium are not well understood and may not be due to ongoing selective pressure of antimicrobial use in food-production animal populations alone (Davis et al., 2002). Several factors, in addition to antimicrobial use, may influence variation in S. Typhimurium between host species including host immunity, vaccination status, biosecurity and industry structure. Vaccination can result in reduction in S. Typhimurium prevalence, reducing the risk of selection of AMR strains (Dagan and Klugman, 2008). Lower AMR profile

diversity of isolates from chickens compared to from cattle and pigs could be expected due to widespread vaccination of laying poultry (>85%) and broiler-breeder flocks against S. Typhimurium, maintenance of Salmonella free grandparent and parent flocks and a stringent active NCP with surveillance and control measures to contain outbreaks (DEFRA, 2007, 2008, 2009; British Egg Industry Council, 2013).

Investigation of association between AMR profile diversity and risk factors within host species populations were restricted by the metadata available for retrospective analysis and therefore limited to high level observations. Higher AMR diversity was observed in isolates from breeder compared to rearer and finisher pigs. No isolates from piglets or at farrowing stage were available. This is consistent with previous findings of acquisition of AMR with successive production stages and higher pulse-field gel electrophoresis cluster diversity of monophasic S. Typhimurium in sows compared to piglets (Fernandes et al., 2016). Compared to layer chicken flocks, broilers were observed to have higher AMR diversity. The relatively high percentage of isolates from layers sensitive to all tested antimicrobials compared to isolates from broilers contributes to the lower diversity of AMR profiles in isolates from layers. Stringent restrictions on antimicrobial use in layer poultry (NOAH, 2016), may place greater emphasis on the need to prevent incursion of infection and reduce selective pressure for AMR to a wider range of antimicrobials. In addition, the high vaccination rates of laying poultry (British Egg Industry Council, 2013) and host genetic differences (Lumpkins et al., 2010) may contribute to variation in AMR profile diversity between the poultry sectors.

Animals are a potential reservoir for AMR bacteria and AMR genes of public health concern, therefore it is important to understand the trends of AMR in animal populations. Despite the limited discriminatory ability of phenotypic AMR profiles and phage type, variation in AMR dynamics of S. Typhimurium were observed within and between host species populations. The transition to whole genome sequencing technology by the APHA provides an enhanced utility for surveillance over time and space, and importantly an opportunity for improved understanding of AMR dynamics between food-production animal populations and humans at the national level. If combined with enhanced metadata, this could ultimately result in higher food security by identifying where intervention strategies may be most effectively applied and provide the opportunity for delivery of an enhanced One Health strategy.

### REFERENCES


### ETHICS STATEMENT

Data was collected through routine national surveillance and therefore did not require study permissions.

### AUTHOR CONTRIBUTIONS

KM and AM contributed to the design of the study. KM performed the analyses. KH and LP were responsible for provision of data. KM wrote the first draft of the manuscript. All authors contributed to manuscript revision, read and approved the submitted version.

### FUNDING

KM is supported by the Bloomsbury Ph.D. Scholarship, Royal Veterinary College and London School of Hygiene and Tropical Medicine. AM is a Food Standards Agency Fellow and is supported by the Quadram Institute Bioscience BBSRC funded Strategic Program: Microbes in the Food Chain (Project No. BB/R012504/1) and its constituent projects BBS/E/F/000PR10348 (Theme 1, Epidemiology and Evolution of Pathogens in the Food Chain) and BBS/E/F/000PR10351 (Theme 3, Microbial Communities in the Food Chain). DEFRA project RDOZO347 funded data collection. The AMR data was collected under an AMR surveillance project (FZ2200) and the epidemiological and microbiological data through a Salmonella surveillance project (FZ2000) both of which were funded by DEFRA for the majority of the study period.

### ACKNOWLEDGMENTS

The authors are grateful for constructive comments received from Chris Teale (Animal and Plant Health Agency).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.00708/full#supplementary-material


and serovar agona. J. Bacteriol. 183, 5725–5732. doi: 10.1128/jb.183.19.5725- 5732.2001



**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 Mellor, Petrovska, Thomson, Harris, Reid and Mather. 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 Arcobacter butzleri Isolated From Shellfish: Novel Insight Into Antibiotic Resistance and Virulence Determinants

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Ariadnna Cruz-Córdova, Children's Hospital of Mexico Federico Gómez, Mexico Mariana Carmen Chifiriuc, University of Bucharest, Romania Carmen Wacher, National Autonomous University of Mexico, Mexico Giuseppe Blaiotta, Università degli Studi di Napoli Federico II, Italy

> \*Correspondence: Vincenzina Fusco vincenzina.fusco@ispa.cnr.it

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 24 October 2018 Accepted: 18 March 2019 Published: 16 April 2019

#### Citation:

Fanelli F, Di Pinto A, Mottola A, Mule G, Chieffi D, Baruzzi F, Tantillo G and Fusco V (2019) Genomic Characterization of Arcobacter butzleri Isolated From Shellfish: Novel Insight Into Antibiotic Resistance and Virulence Determinants. Front. Microbiol. 10:670. doi: 10.3389/fmicb.2019.00670 Francesca Fanelli<sup>1</sup> , Angela Di Pinto<sup>2</sup> , Anna Mottola<sup>2</sup> , Giuseppina Mule<sup>3</sup> , Daniele Chieffi<sup>1</sup> , Federico Baruzzi<sup>1</sup> , Giuseppina Tantillo<sup>2</sup> and Vincenzina Fusco<sup>1</sup> \*

1 Institute of Sciences of Food Production (CNR-ISPA), National Research Council of Italy, Bari, Italy, <sup>2</sup> Department of Veterinary Medicine, University of Bari Aldo Moro, Bari, Italy, <sup>3</sup> Institute of Biomembranes, Bioenergetics and Molecular Biotechnologies (CNR-IBIOM), National Research Council of Italy, Bari, Italy

Arcobacter (A.) butzleri is an emerging pathogenic microorganism, whose taxonomy has been recently suggested to be emended to the Aliarcobacter (Al.) butzleri comb. nov. Despite extensive taxonomic analysis, only few fragmented studies have investigated the occurrence and the prevalence of virulence and antibiotic resistance determinants of this species in strains isolated from shellfish. Herein we report for the first time the whole genome sequencing and genomic characterization of two A. butzleri strains isolated from shellfish, with particular reference to the antibiotic, heavy metals and virulence determinants. This study supported the taxonomic assignment of these strains to the Al. butzleri species, and allowed us to identify antibiotic and metal resistance along with virulence determinants, also additional to those previously reported for the only two A. butzleri strains from different environments genomically characterized. Moreover, both strains showed resistance to β-lactams, vanocomycin, tetracycline and erythromycin and susceptibility to aminoglycosides and ciprofloxacin. Beside enlarging the availability of genomic data to perform comparative studies aimed at correlating phenotypic differences associated with ecological niche and geographic distribution with the genetic diversity of A. butzleri spp., this study reports the endowment of antibiotic and heavy metal resistance and virulence determinants of these shellfish-isolated strains. This leads to hypothesize a relatively high virulence of A. butzleri isolated from shellfish and prompt the need for a wider genomic analysis and for in vitro and in vivo studies of more strains isolated from this and other ecological niches, to unravel the mechanism of pathogenicity of this species, and the potential risk associated to their consumption.

Keywords: Aliarcobacter butzleri, Arcobacter butzleri, antibiotic and heavy metal resistance, virulence genes, genomics, food safety, emerging foodborne pathogen, shellfish

### INTRODUCTION

fmicb-10-00670 April 16, 2019 Time: 11:29 # 2

The current validated taxonomy places the Arcobacter genus within the Campylobacteraceae family (belonging to the class Epsilonproteobacteria of the phylum Proteobacteria) together with Campylobacter and Sulfurospirillum genera. Recently, based on a wide comparative genomic analysis, Waite et al. (2017) proposed the re-classification of the class Epsilonproteobacteria with the Arcobacter genus ascribed to the Arcobacteraceae fam. nov. (type genus: Arcobacter, order: Campylobacterales, class: Campylobacteria, class. nov., phylum: Campylobacteraeota phyl. nov.) (Waite et al., 2017, 2018).

Arcobacter spp. have a wide diversity of hosts and habitats, with water being one of the main routes of transmission (Collado and Figueras, 2011). Arcobacter spp. have been indeed detected in environmental water sources, including rivers, lakes, sewage and plankton (Kim et al., 2010; Levican et al., 2013; Park et al., 2016), marine, domestic (Merga et al., 2014), and drinking water (Jalava et al., 2014; Talay et al., 2016), water distribution pipes (Phillips, 2001), groundwater (Fong et al., 2007), and recreational water (Lee et al., 2012). Four species (namely, A. butzleri, A. cryaerophilus, A. thereius, and A. skirrowii) have also been isolated in humans and animals (Banting and Figueras Salvat, 2017) and are able to cause human bacteraemia, endocarditis, peritonitis, gastroenteritis, and diarrhea (Jiang et al., 2010; Figueras et al., 2014; Arguello et al., 2015; Ferreira et al., 2016).

To date this genus includes 26 species (Ramees et al., 2017; Pérez-Cataluña et al., 2018a), which inhabit various ecological niches (Collado and Figueras, 2011; Ferreira et al., 2016). Recently, Pérez-Cataluña et al. (2018a) used a polyphasic approach to revisit the taxonomy of the genus. By setting specific cut-off values for each method (identity of 16S rRNA, genomic indexes such as ANI, AAI, and DDH, multilocus sequence analysis, etc.), they delimited genomic and phylogenetic groups and defined six different genera including Aliarcobacter gen. nov. This genus comprises seven species including also Aliarcobacter (Al.) butzleri comb. nov., whose type strain was confirmed to be LMG 10828<sup>T</sup> (ATCC 49616<sup>T</sup> ; RM4018; Miller et al., 2007), and the species description as the one given by Vandamme et al. (1992) and Pérez-Cataluña et al. (2018a,b).

However, it must be considered that both the species description of A. butzleri and Al. butzleri comb. nov., were obtained by analyzing an exiguous number of A. butzleri strains (12 by Vandamme et al., 1992 and only two by Pérez-Cataluña et al., 2018a, respectively). Thus, it might be expected that the likely increasing availability of A. butzleri genomic sequences could lead this species to a further redefinition.

Recent outcomes and progress related to the pathogenicity of A. butzleri have led to the allocation of this species in the list of microbes considered a serious hazard to human health by the International Commission on Microbiological Specifications for Foods (ICMSF, 2002). A. butzleri is considered an emerging food-borne enteropathogen and, in recent years, has been associated with enteritis, severe diarrhea, bacteraemia, and septicaemia in humans and enteritis, mastitis, abortion, and stillbirth in cows, sheep, and pigs (Engberg et al., 2000; Lau et al., 2002; Collado and Figueras, 2011; Collado et al., 2014; Ferreira et al., 2014a,b, 2016; Figueras et al., 2014; Van den Abeele et al., 2014; Heimesaat et al., 2015; Gölz et al., 2016; Webb et al., 2016; Franz et al., 2018; Fusco et al., 2018; Flynn et al., 2019). Enteritis due to the ingestion of food contaminated with Arcobacter spp. can be self-limiting. Nevertheless, the severity and protraction of the symptoms might require an antibiotic treatment, which might be affected by the (multiple) antibiotic resistance of the strain, thus complicating the treat of the relevant infections. A. butzleri is the most prevalent species of this genus in meat products (chicken, pork, beef, lamb), milk, cheese, and shellfish (Figueras et al., 2011a,b; Patyal et al., 2011; Shah et al., 2012; Hausdorf et al., 2013; Lee and Choi, 2013; Rahimi, 2014; Ramees et al., 2014; Lehmann et al., 2015; Ferreira et al., 2016). The presence and persistence in these niches, also endowed by the ability to form biofilms (Ferreira et al., 2013; Girbau et al., 2017), favor its spread and transmission to shellfish and farm animal, and increase the risk associated with food consumption. Although few fragmented studies have been carried out to assess the occurrence of this species in shellfish, A. butzleri has been found as the most common species in bivalve molluscs (mussels, clams, oysters, etc.) (Nieva-Echevarria et al., 2013; Levican et al., 2014; Mottola et al., 2016; Leoni et al., 2017). This is most likely due to capture by the filter feeding process of bivalves and to a fecal contamination of the relevant environment, so that Escherichia coli has been proposed by Leoni et al. (2017) as an index organism for A. butzleri contamination in bivalve molluscs, leading to suggest that these shellfishes could be a reservoir of A. butzleri (Leoni et al., 2017). For this reason, consumption of contaminated shellfish, especially if raw or undercooked, which is still a widespread practise (Schauer Weissfeld, 2014), may be source of A. butzleri infections in humans. The ability to survive in different environments, the endowment of antibiotic resistance genes and virulence potential found by genomic approaches (Miller et al., 2007; Pérez-Cataluña et al., 2018a), as well as the genetic plasticity conferred by the presence of mobile elements, which allow the transfer of genes (Douidah et al., 2014), are important determinants for the evolution and the fitness of this and the other species of this genus.

As far as we know, extensive genome-based characterization of the species has been carried out only on two A. butzleri strains, namely, RM4018 (Culture collection n. LMG 10828<sup>T</sup> ), isolated from human feces, and ED-1 isolated from microbial fuel cells (Miller et al., 2007; Pérez-Cataluña et al., 2018a). Considering that contaminated shellfish may be source of A. butzleri infection and given that the prevalence and expression of putative virulence and antibiotic resistance genes within this species may vary with the source of the strain (Douidah et al., 2012; Ferreira et al., 2014b; Girbau et al., 2015; Zacharow et al., 2015a), herein we report the antibiotic susceptibility and genomic-based characterization of two A. butzleri strains isolated from shellfish, with particular reference to the genetic determinants of the above mentioned traits of pathogenicity.

### MATERIALS AND METHODS

fmicb-10-00670 April 16, 2019 Time: 11:29 # 3

### Strains and Culture Condition

Arcobacter butzleri (Ab) strains 6V and 55 were isolated on 2016 from clams (Tapes philippinarum) (Mottola et al., 2016) and mussels (Mytilus galloprovincialis) obtained from local fish market in the Apulia region (Italy). These strains were previously identified and typed by MLST (Mottola et al., 2016; Mottola, 2017). Allelic profiles and sequences are available on the Arcobacter MLST database<sup>1</sup> under the ID numbers 717 (Ab 6V) and 839 (Ab 55).

Pure cultures were isolated and maintained in the microbial collection of the Institute of Sciences of Food Production, CNR, Bari<sup>2</sup> . Bacterial strains were maintained at −80◦C as pure stock cultures in Brain Heart Infusion (BHI) broth (Oxoid S.p.A., Rodano, Milan, Italy) supplemented with glycerol (30% vol/vol). Cultures were streaked on Agar blood plates (Oxoid, Milan, Italy) and grown at 37◦C for 48 h. Working cultures were obtained growing a single colony in 20 mL of BHI broth with 0.6% yeast extract (BHI-YE), at 37◦C for 48 h.

### Genome Sequencing and Assembly

DNA isolation was performed by using the Wizard <sup>R</sup> Genomic DNA Purification Kit (Promega), as previously described by Ercolini et al. (2005). The integrity, purity, and quantity of DNA were assessed as previously described by Fusco et al. (2011), by agarose gel electrophoresis, by NanoDrop-2000 (Thermo Fisher Scientific, Wilmington, DE, United States), and by Qubit 3.0 fluorometer (Life Technologies). DNA was then subjected to whole genome shotgun sequencing using the Ion S5TM library preparation workflow (Thermo Fisher Scientific, Waltman, MA, United States); 400 bp mate-paired reads were generated on the Ion S5TM System (Thermo Fisher Scientific). Duplicate reads were removed by FilterDuplicates (v5.0.0.0) Ionplugin. De novo assembly was performed by AssemblerSpades (v.5.0) IonpluginTM.

### Bioinformatic Methods

Genes were predicted and annotated using PROKKA pipeline implemented in the Galaxy platform (Galaxy Tool Version 1.0.0; Afgan et al., 2016). The predicted proteins were submitted to the PFAM annotator tool within the Galaxy platform in order to predict the pfam domains. Protein ID used in the manuscript indicated those obtained by NCBI (National Center for Biotechnology Information) Prokaryotic Genome Annotation Pipeline (Tatusova et al., 2016).

Predicted proteins were assigned to Clusters of Orthologous Groups (COG) functional categories by Web CD-Search Tool (Marchler-Bauer et al., 2017) using an Expected value threshold of 0.01. COG ID were then manually mapped into functional categories<sup>3</sup> .

All the proteins sequences used in this study were retrieved from GenBank (NCBI). The homology-based relationship of Ab 55 and Ab 6V predicted proteins toward selected proteins was determined by BLASTP algorithm on the NCBI site<sup>4</sup> . Gene models were manually determined, and clustering and orientation were subsequently deduced for the closely linked genes.

Antibiotic resistance genes were predicted by BLASTP search against the Antibiotic Resistance genes Database (ARDB; Liu and Pop, 2009) and beta lactamase database (Naas et al., 2017). Genes associated with antibiotic resistance were also retrieved by keywords terms search within UniProtID entry list obtained by functional annotation.

Functional annotation, subsystem prediction, and metabolic reconstruction comparison were also performed using the RAST server (Aziz et al., 2008). Genes involved in the mechanism of resistance to heavy metals were retrieved by Poole (2017) and used as queries for BLASTP search against Ab 55 and Ab 6V proteomes.

Genetic divergence was calculated by the ANI/AAI calculator (Goris et al., 2007; Rodriguez-R and Konstantinidis, 2016) which estimates the average nucleotide/aminoacid identity (ANI/AAI) using both best hits (one-way ANI) and reciprocal best hits (two-way ANI) between genomic datasets. The Genome-to-Genome Distance Calculator (GGDC) (Meier-Kolthoff et al., 2013, 2014) web service was used to report DDH for the accurate delineation of prokaryotic subspecies and to calculate differences in G+C genomic content (available at ggdc.dsmz.de). Formula 2 alone was used for analysis, providing an estimation of DDH independent from genome lengths, as recommended by the authors of GGDC for use with any incomplete genomes (Auch et al., 2010; Meier-Kolthoff et al., 2013).

### Antimicrobial Susceptibility Testing

The antimicrobial susceptibility tests for A. butzleri isolates were performed by disk diffusion and broth microdilution methods. The disk diffusion test was performed as described by Rathlavath et al. (2017) with modifications. Briefly, A. butzleri isolates were grown in 20 ml of BHI broth (Oxoid, United Kingdom) amended with 0.6% yeast extract (YE) (Biolife srl, Italy) under static condition for 48 h at 37◦C and then subcultured at 1% in BHI-YE broth and incubated at 37◦C for 48 h. Microbial cells were recovered after centrifugation (16,000 rcf × 6 min), washed in sterile 0.9% NaCl solution, adjusting optical density (600 nm) to 0.5. One hundred microliters of this cell suspension were then plated on 4 mm thick cation adjusted Muller Hinton agar (MHIIA, Liofilchem, Italy).

Antibiotic disks, soaked with ampicillin (10 µg/disk), cefotaxime (30 µg/disk), chloramphenicol (30 µg/disk), ciprofloxacin (5 µg/disk), erythromycin (15 µg/disk), gentamicin (10 µg/disk), kanamycin (30 µg/disk), nalidixic acid (30 µg/disk), streptomycin (10 µg/disk), tetracycline (30 µg/disk), vancomycin (30 µg/disk), and penicillin G (10 units/disk) (Biolab Zrt., Hungary), were placed onto inoculated plates and incubated at 37◦C under microaerophilic atmosphere (CampyGenTM Compact, Oxoid, United Kingdom),

<sup>1</sup>https://pubmlst.org/arcobacter

<sup>2</sup>www.ispa.cnr.it/Collection

<sup>3</sup>https://www.ncbi.nlm.nih.gov/COG/

<sup>4</sup>http://blast.ncbi.nlm.nih.gov/Blast.cgi

as recommended by BCCM/LMG Bacteria Collection<sup>5</sup> for A. butzleri type strain LMG10828 (RM4018). After 48 h of incubation, inhibition zone diameters were recorded.

For those antibiotics which the tested A. butzleri strains did not provide inhibition zone at all, the minimal inhibitory concentration (MIC) was calculated by broth microdilution method as described by Riesenberg et al. (2017). After 48 h incubation at 37◦C, microtitre plates were read spectrophotometrically at 600 nm using Varioskan Flash (Thermo Fisher Scientific, United States). To determine minimal bactericidal concentration (MBC), 10 µl of broth from three replicate of all wells without microbial growth was combined in a single sample spotted on MHIIA (Liofilchem, Italy) and incubated as described above up to 72 h. Since no breakpoint values are available for Arcobacter spp., classification of strains as susceptible (S), resistant (R), or intermediate (I) was defined according to zone diameter and MIC interpretive standards for Staphylococcus spp. (erythromycin, penicillin, and vancomycin) and Enterobacteriaceae (ampicillin, gentamicin, cefotaxime, ciprofloxacin, tetracycline, chloramphenicol, nalidixic acid, kanamycin, and streptomycin) (CLSI, 2015), as also reported by Elmali and Can (2017).

The A. butzleri LMG 10828<sup>T</sup> (RM4018) strain was used for comparison purposes.

### RESULTS AND DISCUSSION

### General Features of A. butzleri 6V and 55 Genomes

Ab 55 and Ab 6V genomes were sequenced using a whole genome shotgun approach on an Ion S5TM platform (Thermo Fisher Scientific) generating around 671,363 and 596,333 reads with a median length of 317 and 320 bp, respectively (**Table 1**). Genomes were assembled using the Spades v5.0 software for a

<sup>5</sup>http://bccm.belspo.be

TABLE 1 | Summary of A. butzleri 55 and 6V genome sequencing and assembly results.


total of 32 and 46 large contigs (>500 bp) and a GC% of 26.79 and 26.85, respectively. The overall contiguity of the assembly is good, with a N50 of 211 and 129 kbp for Ab 55 and Ab 6V, respectively; the longest assembled fragment is 403 kbp in length for Ab 55 and 251 kbp for Ab 6V (performed by QUAST, available at http://quast.sourceforge.net/quast) while the total length of the assembly was around of 2.3 Mb for both genomes. These Whole Genome Shotgun projects have been deposited at DDBJ/ENA/GenBank under the accessions QXMK00000000 (Ab 55) and QXNB0000000 (Ab 6V). The versions described in this paper are QXMK01000000 (Ab 55) and QXNB01000000 (Ab 6V).

### Genomic Analysis

The in silico MLST of the housekeeping genes retrieved from genomic sequences, confirmed in vitro results achieved by Mottola (2017), revealing that Ab 55 and Ab 6V belong to two novel sequence types, namely, ST675 and ST537, respectively, as they harbor 6/7 and 3/7 new alleles, respectively (**Table 2**).

Both Ab 6V and Ab 55 16S rRNA gene sequences show 100% identity with the type strain Ab RM4018 (Miller et al., 2007). ANI, AAI, and DDH analyses were performed with 22 strains within the Arcobacter group (**Supplementary Table S1**). Campylobacter jejuni subsp. jejuni NCTC 11168 and Helicobacter pylori 26695 were included as outgroups.

Ab 55 and Ab 6V have 98.06% nucleotide identity (**Supplementary Table S2**) and are comprised in the clustering including all the A. butzleri species. According to ANI, the closest relatives for both Ab 55 and Ab 6V are Ab NCTC 12481 (97.80% and 97.81 ANI, respectively) and Ab RM4018 (97.79% and 97.80 ANI). The same clustering is obtained by using AAI (**Supplementary Table S3**) with 97.60% between Ab 55 and Ab 6V, and 97.67 and 97.36% with Ab RM4018, respectively. DDH analysis confirmed the clustering obtained by ANI and AAI analysis, with values of 83.50 between Ab 6V and Ab 55, 81.00% between Ab 6V and Ab RM4018, and 81.30% between Ab 55 and Ab NCTC 12481 (**Supplementary Table S4**). As proposed by Chun et al. (2018) and, more specifically for Arcobacter spp., by On et al. (2017), these ANI and DDH values are within the range suggested to include Ab 55 and Ab 6V into the A. butzleri species. Moreover, our results support those achieved by Pérez-Cataluña et al. (2018a). Therefore, Ab 55 and Ab 6V should be placed within the Aliarcobacter gen. nov. as Al. butzleri comb. nov. (Pérez-Cataluña et al., 2018a,b), while for the definition of subspecies, a phenotypic comparison should support the in silico analyses (On et al., 2017).

### Protein Functional Classification

For Ab 55, 1173 UniProtKB AC/ID identifiers retrieved by PFAM annotator tools (Galaxy Tool Version 1.0.0) were successfully mapped to 1190 UniProtKB IDs (The UniProt Consortium, 2017). In Ab 55, the retrieved list included 33 genes associated with antibiotic resistance, including beta-lactamase, multidrug efflux pump, DNA gyrase, and resistance protein, and three with antibiotic biosynthesis related to bacteriocin, six associated with drug transmembrane transporter activity, 19 with virulence, four with hemolysis, and two with quorum sensing (luxS and tqsA).


#### TABLE 2 | Allelic profiles of A. butzleri isolates.

fmicb-10-00670 April 16, 2019 Time: 11:29 # 5

Novel alleles and novel sequence type (ST) are indicated in bold.

For Ab 6V, 1189 out of 1189 UniProtKB AC/ID identifiers were mapped to 1207 UniProtKB. Based on their classification, we counted three genes associated with antibiotic biosynthesis, 32 related to antibiotic resistance, including beta-lactamase, multidrug efflux pump, DNA gyrase and resistance protein, six referred to drug transmembrane transporter activity, 22 associated with virulence, three with hemolysis, and two with quorum sensing (luxS and tqsA).

One of the few differences between these strains is the presence in Ab 6V of the hemolysin tylC gene which codes for a protein with a conserved protein/cyclin M (CNNM) transmembrane domain, while, in Ab 55, we found only a precursor of hemolysin C. In both genomes, we identified the RNA methyltransferase hemolysin A (tylA), which encodes for a 16S/23S rRNA (cytidine-2 0 -O)-methyltransferase that is considered a virulence factor in H. pylori infection and probably acts as a pore-forming toxin (Javadi and Katzenmeier, 2016).

Overall, 1551 for Ab 55 and 1592 for Ab 6V predicted genes were assigned to the COG classification (**Figure 1**). The few differences in the distribution of genes into clusters of COG functional categories across Ab 55 and Ab 6V genomes, which emerged from **Figure 1**, supported the limited functional variability among these two strains, even though isolated from different samples and in different harvest seasons (Mottola et al., 2016).

Among COG categories, the cluster signal transduction mechanisms represented the largest group for both organism (172 genes, 11.09% for Ab 55 and 187 genes, 11.75% for Ab 6V), followed by amino acid transport and metabolism cluster (147 genes, 9.48% for Ab 55 and 147 genes, 9.23% for Ab 6V), Cell wall/membrane/envelope biogenesis cluster (140 genes, 9.03% for Ab 55 and 147 genes, 9.23% for Ab 6V) and energy production and conversion cluster (133 genes, 8.58% for Ab 55 and 135 genes, 8.48% for Ab 6V). These findings suggest that our Arcobacter strains have a versatile sensory transduction system and that for energy and carbon mainly rely on amino acid catabolism rather than on sugar fermentation, which is consistent with the ecological niches they have been isolated from.

### Virulence Determinants

The ability to adhere to various surfaces as well as chemotaxis, motility, and signal transduction play a pivotal role in the microbial survival and colonization of diverse ecological niches and can be involved in pathogenesis and antibiotic resistance (Richards et al., 2013; Chaban et al., 2015; Tiwari et al., 2017; Matilla and Krell, 2018).

In both genomes, we identified orthologues of waaC and waaF genes, which are described as virulence determinant in A. thereius (Rovetto et al., 2017), but also in Pseudomonas aeruginosa, E. coli, Klebsiella pneumonia, and other Campylobacteraceae (Oldfield et al., 2002; DeLucia et al., 2011; Nilsson et al., 2018). waaF codes for a predicted ADP-heptose–LPS heptosyltransferase (Ab 55 D3M61\_00085; Ab 6V D3M75\_11180), involved in the biosynthesis of lipooligosaccharide (LOS); waaC codes for a lipopolysaccharide heptosyltransferase (Ab 55 D3M61\_00195; Ab 6V D3M75\_05185), which catalyzes the transfer of sugar moieties from activated donor molecules to specific acceptor molecules, forming glycosidic bonds. As in Ab RM4018 (Miller et al., 2007), which it shares the same content and organization with, in Ab 55, waaC and waaF genes are closely located suggesting the presence of a genetic cluster, whose composition is, however, different from A. thereius (**Figure 2**). In Ab 6V, the genetic locus comprised by waaC and waaF is similar to that of Ab ED-1 (26 genes) and contains several glycosyltransferases with no orthologues in Ab 55 and Ab RM4018. The outer genes of this locus are similar to A. thereius. However, in Ab 6V, these genes are located in three different contigs, which made only provisional, although likely, the reconstruction of the organization of the entire locus.

Arcobacter butzleri, as the other Arcobacter species, is motile by means of one single polar flagellum (Miller et al., 2007). The bacterial flagellar motor is comprised by a core structure (inner membrane stator complexes MotA4B2 and the C-ring), a dedicated type III secretion system (T3SS) export apparatus, the inner membrane MS-ring and the P- and L-rings. These core components are conserved across bacterial genera. Nevertheless, the architecture of flagellar motors in A. butzleri has a diverse (Rossmann and Beeby, 2018), "intermediate" (Chaban et al., 2018) motor structure. According to the analysis performed by Chaban et al. (2018), we retrieved homologous proteins of the flagellar system in Ab 55 and Ab 6V genomes (**Table 3**). In both strains, loci share the same genomic content and most of the flagellar genes are located within the same genomic locus. As reported by Chaban et al. (2018), the Arcobactertype motor accessory proteins did not contain homologues of the accessory proteins FlgP, FlgQ, FlgT, while we identified the homologue FlgO, an outer membrane protein required for flagellar motility in Vibrio cholerae, highly conserved in Vibrio spp. (Zhu et al., 2017).

In Ab 55, the flagellar biosynthetic protein FliP (an integral membrane component) (Ferris and Minamino, 2006) was predicted as a pseudogene (frameshifted) (D3M61\_02925), as the flagellar biosynthetic protein FlhB (a flagellar export component responsible for substrate specificity switching) (Minamino and

FIGURE 2 | Genomic organization of waaC/waaF gene cluster in A. thereius LMG 24486, A. butzleri ED-1, A. butlzeri 6V, A. butzleri 55, and A. butzleri RM4018. Gene clustering is represented by the arrows superposed on the black horizontal line. Intergenic spaces are not drawn in scale. For A. thereius LMG 24486, A. butzleri ED-1, and A. butzleri RM4018, the locus tag of each gene is indicated below the respective gene arrow; for A. butlzeri 6V and A. butzleri 55, protein ID is indicated below the respective gene arrow. Red arrows in A. butlzeri ED-1 and A. butzleri 6V indicate genes with no orthologue in A. butzleri 55 and A. butlzeri RM4018. D3M75\_05155 <sup>∗</sup> indicates pseudogene (frameshifted). lAaT: lipid A biosynthesis acylTransferase; dK: diacylglycerol kinase; yejM: inner membrane protein yejM; O-aL: O-antigen ligase; hP: hypotetical protein; gT: glycosyltransferase; pgaB: poly-beta-1,6-N-acetyl-D-glucosamine N-deacetylase; eptA: phosphoethanolamine transferase eptA; rfaF: lipopolysaccharide heptosyltransferase II; aT: acetyltransferase; sunS: glycosyltransferase sunS; aLP: alkaline phosphatase family protein; gPtT: glucose-1-phosphate thymidylyltransferase rfbA; yrbL-phoP: regulatory network protein; degT: DegT/DnrJ/EryC1/StrS family aminotransferase; rfbA: glucose-1-phosphate thymidyl transferase; rfbB: dTDP-glucose 4,6-dehydratase 1; rfbC: dTDP-4-dehydrorhamnose 3,5-epimerase; rfbD: dTDP-4-dehydrorhamnose reductase; mbOat: membrane bound O-acyl transferase; yrbL-phoP: YrbL-PhoP reg domain containing protein.

Macnab, 2000) in Ab 6V (D3M75\_09230). This might have occurred due to a homopolymeric region (AAAAAA) comprised in both the genomic loci which might have affected the base calling of the sequencing.

#### TABLE 3 | Flagellum proteins in Arcobacter butzleri genomes.

fmicb-10-00670 April 16, 2019 Time: 11:29 # 7


<sup>∗</sup>Pseudogene (frameshifted).

Both genomes harbor the genes cj1349 and cadF (coding for fibronectin-binding proteins CadF and Cj1349), mviN (encoding a protein essential for the peptidoglycan biosynthesis) and pldA (encoding phospholipase A), the gyrA, iroE, and irgA (iron-regulating outer membrane protein), ciaB (encoding the C. jejuni invasion antigen B) genes and the hemolysin gene tylA, while only Ab 55 harbors hecA [a member of the filamentous hemagglutinin (FHA) family], and hecB (encoding a hemolysin activation protein), as occurs in the Ab RM4018 (Miller et al., 2007).

Cia proteins (including CiaB, CiaC, and CiaD) have been suggested to be involved in promoting internalization of C. jejuni for host invasion and require a full-length flagellar filament for proper secretion (Chaban et al., 2015).

Several authors have screened A. butzleri strains for the presence of virulence genes such as ciaB, cadF, cj1349, hecA, and irgA, finding diverse prevalence of these genes in isolates from the same or different ecological niches (Collado et al., 2014; Ferreira et al., 2017; Rathlavath et al., 2017; Vicente-Martins et al., 2018). However, nucleotide sequence heterogeneity as well as PCR biases may provide false negative results thus underestimation of the actual prevalence of these genes. These drawbacks may be overcome by a genomic-based assessment, which, moreover, may allow the detection of novel (acquired) virulence genes.

Indeed, apart from the above-mentioned genes, recognized as putative virulence determinants in A. butzleri (Miller et al., 2007), we identified other genes coding for additional virulence associated protein in Ab 55 and Ab 6V genomes. Among these and present in Ab 55, Ab 6V, and Ab RM4018, we distinguished a DNA binding protein of the virulence factor B family, which contributes to the expression of virulence factors and to pathogenicity in Staphylococcus aureus (Matsumoto et al., 2010; Junecko et al., 2012), a VOC family virulence protein (glyoxalase/bleomycin resistance protein/dioxygenase superfamily domain), a conserved virulence factor B (DNA binding protein), and VirF of the AraC family of transcriptional regulators.

Only in Ab 55, with no orthologs in Ab 6V and Ab RM4018, we identified a virulence associated protein VirE (D3M61\_07785), which has an 87% identity with a hypothetical protein of Ab L353 (WP\_080952707.1); this was the only hit retrieved within Arcobacter genus (taxid: 28196) by BLASTP search in the NCBI database, while other results indicated identity of about 32% with a hypothetical protein of Dyella sp. 4G-K06 (WP\_115495989.1) and with a primase from the Escherichia phage vB\_EcoM-ep3. This sequence was also compared to metagenomic sequences comprised in the microbiome database MGNIFY (EMBL-EBI©; Mitchell et al., 2017) by BlastP analysis. Analysis showed 91% of identity with metagenomic sequence of a sample retrieved from Charlotte area wastewater treatment plants (North Carolina, United States). InterPro functional classification (Finn et al., 2017) assigned the protein the Virulence-associated E (IPR007936) family membership with a D-loop motif (Pfam:PF05272.5) and a related COG: 5545

Mobilome: prophage transposone category. The genomic locus in which the gene is located also comprises several hypothetical proteins, one Prophage CP4-57 integrase, several tRNA, one site-specific tyrosine recombinase XerC, putative DNA-invertase from lambdoid prophage Rac. The structure of this genomic locus is compatible with the presence of a genomic island (Juhas et al., 2009) and suggests the acquisition of this virulence element by a mechanism of horizontal gene transfer.

Furthermore, in Ab 6V genome, we also found one virulence sensor protein BvgS precursor and one virulence sensor histidine kinase PhoQ, which both have orthologues in Ab RM4018 (WP\_012012740.1 and WP\_012012731.1, respectively) but not in Ab 55.

Unique to Ab 6V, we identified two putative hemolysin activation/secretion proteins (D3M75\_11460 and D3M75\_03300), which have orthologues in A. butzleri ED-1, with a ShlB/FhaC/HecB family hemolysin secretion/ activation domain.

All together these findings lead to hypothesize a relatively high virulence of A. butzleri isolated from shellfish and prompt the need for a genomic analysis of more strains from this ecological niche, as well as for in vitro and in vivo studies, to unravel the mechanism of pathogenicity of this species.

### Antibiotic Susceptibility and Genetic Determinants

As shown in **Table 4**, both genomes harbor a wide endowment of genes involved in antibiotic resistance, including transporter, efflux pump, multidrug resistance protein and methyltransferase, although with some differences. As example, Ab 55 harbors a protein predicted as a bifunctional polymyxin resistance protein ArnA (D3M61\_11465), which is not present neither in Ab 6V nor in the A. butlzeri RM4018. BLASTP analysis retrieved as best hit a hypothetical protein of Campylobacter hyointestinalis (WP\_111949105.1).

The UDP-4-amino-4-deoxy-L-arabinose-oxoglutarate aminotransferase arnB gene was instead identified in both Ab 55 and Ab 6V, while it is not present in A. butlzeri RM4018. The coded protein belongs to the DegT/DnrJ/EryC1/StrS aminotransferase family protein, and it is required for resistance to polymyxin and cationic antimicrobial peptides (Lee and Sousa, 2014).

The predicted serine/threonine-protein kinase HipA belonging to the type II toxin-antitoxin system was only retrieved in Ab 6V predicted proteome (D3M75\_03575), and it is involved in the methicillin resistance. Only in Ab 55, we identified the mRNA interferase toxin RelE (D3M61\_04745), which is involved in ciprofloxacin and ampicillin resistance in E. coli (Harms et al., 2017).

**Table 5** shows metal resistance genes annotated in Ab 55 and Ab 6V genomes.

Observational and experimental studies have highlighted that exposure of bacteria to heavy metals (mainly zinc and copper), mainly due to anthropogenic environmental contamination, can induce or co-select resistance to them and to one or more antibiotics. In particular, resistance may be induced by metals (i) via co-selection resistance, when different genes coding for antibiotic and metal resistance share a close location (as in mobile genetic elements, such as integrin, plasmid, or transponson), (ii) via cross-selection, when the same genetic element encodes for both antibiotic and metal resistance, and (iii) via co-regulation, when antibiotic and metal resistance genes share the same regulatory system (Lupo et al., 2012; Seiler and Berendonk, 2012; Chenia and Jacobs, 2017; Li et al., 2017; Poole, 2017; Yu et al., 2017; Wu et al., 2018). Nevertheless, as far as we know, only Otth et al. (2005) investigated susceptibility of A. butzleri to heavy metals, finding that all the 50 tested strains were susceptible to mercury, silver, and chrome salts, whereas all were resistant to molybdenum, manganese, nickel, cobalt, lead, and iron.

Metal resistance genes share the same location in both Ab 55 and Ab 6V genomes. cadA is close to genes coding for the ferrous iron transport protein FeoA and FeoB, outer membrane efflux proteins, transcriptional regulators, and several tRNA genes, suggesting the presence of a genomic island (Juhas et al., 2009). czcB is close to gene coding for the multidrug resistance protein MstB, one permease and the gene coding for the sensor protein ZraS. One copA gene is close to the gene coding for outer membrane porin precursor and the copper sensing transcriptional repressor CsoR, while the other copA is close to the ferrous iron transport proteins FeoA and FeoB. cusS is close to the transcriptional activator protein czcR, macB, and several chaperonine, whereas arsC is located near to genes coding for plasmid stabilization system protein and genes coding for flagellum biosynthesis (fliK and flhB), and cell division protein FtsA and FtsZ. modA gene is located within a hypothetical operon including a transcriptional regulator GltR, a gene coding for a molybdenum-pterin binding protein, the regulator ModE, and the transport system permease protein ModB. copZ is close to merT gene, coding for a mercuric transport protein, a gene coding for a natural resistance-associated macrophage protein, a transcriptional regulatory protein ZraR and a sensor protein FixL.

The antimicrobial susceptibility of Arcobacter spp. isolated from various ecological niches has been investigated by several authors (Kabeya et al., 2004; Kayman et al., 2012; Ferreira et al., 2013, 2017; Scanlon et al., 2013; Collado et al., 2014; Rahimi, 2014; Yesilmen et al., 2014; Zacharow et al., 2015b; Aski et al., 2016; Van den Abeele et al., 2016; Elmali and Can, 2017; Rathlavath et al., 2017; Šilha et al., 2017; Soma et al., 2017; Vicente-Martins et al., 2018), but the lack of standardized protocols and interpretive criteria for the antimicrobial susceptibility testing (AST) of Arcobacter spp. is the major limitation for a comparable and univocal evaluation of antimicrobial resistance and susceptibility for these microorganisms (Ferreira et al., 2016). Results of the disk diffusion test are shown in **Table 6**, whereas in **Table 7**, are reported the MICs and MBCs assessed by broth microdilution method (section "Antimicrobial Susceptibility Testing"). The antimicrobial susceptibility pattern of Ab type strain LMG 10828<sup>T</sup> (ATCC 49616, RM4018), used as a reference strain for the AST in our study, was comparable to that reported by Miller et al. (2007), for 10 out of 12 tested antibiotics, i.e., gentamicin, kanamycin, streptomycin, ciprofloxacin, chloramphenicol, erythromycin,

TABLE 4 | Antibiotic resistance genes in Ab 55 and Ab 6V genomes.


TABLE 5 | Metal resistance genes in Ab 55 and Ab 6V genomes.


vancomycin, ampicillin, cefotaxime, and penicillin G. Slight differences were related to susceptibility and resistance toward tetracycline and nalidixic acid, reported by Miller et al. (2007); in particular, in our test, the type strain resulted intermediate resistant toward both antibiotics. As concerns MIC values, the type strain Ab LMG10828 showed results consistent with those find by Riesenberg et al. (2017). Additionally, in our study, using a wider range of concentrations for penicillin and vancomycin, we were able to assess MIC values for Ab LMG10828<sup>T</sup> toward these two antibiotics (i.e., 128 and 2048 µg/ml, respectively) that were previously reported by Riesenberg et al. (2017) as ≥64 µg/ml. According to MIC interpretive standards, A. butzleri LMG 10828<sup>T</sup> , Ab 55 and Ab 6V have been classified as resistant toward vancomycin and the three β-lactam antibiotics used in this study, i.e., cefotaxime (β-lactam cephalosporin), ampicillin, and penicillin G (β-lactam penicillins), confirming the results obtained by disk diffusion test (**Tables 6**, **7**).

For A. butzleri type strain, Ab 55 and Ab 6V, the highest MIC and MBC values were observed for vancomycin, which were ≥2048 µg/ml (**Table 7**).

As recently reviewed by Ahmed and Baptiste (2018) in enterococci, molecular basis for vancomycin resistance phenotypes are determined by the presence of the vancomycin resistance (Van) operons (described as vanA, -B, -C, -D, -E, -G, - L, -M, and N) which may be located on the chromosome or on



Classification as S (susceptible), I (intermediate), and R (resistant) was carried out according to zone diameter interpretive standards for Staphylococcus spp. (erytromycin and penicillin G) and Enterobacteriaceae (ampicillin, gentamicin, cefotaxime, ciprofloxacin, tetracycline, chloramphenicol, nalidixic acid, kanamycin, and streptomycin) (CLSI, 2015; Elmali and Can, 2017). To our knowledge, no vancomycin reference interpretive criteria are reported for A. butzleri. <sup>∗</sup>No inhibition zone detected.

a plasmid. No element of the described operons was found in Ab 55 and Ab 6V genomes, as well as in Ab RM4018. However, the resistance of Gram-negative bacteria toward vancomycin may be intrinsic due to the inability of glycopeptides [molecules with high molecular weight (1450–1500 Da) and size] to pass through porins, which govern the movement of hydrophilic molecules across their outer membrane (Quintiliani and Courvalin, 1995) to reach their site of action, i.e., the cell wall (Nicolosi et al., 2010). Indeed, 100% of the A. butzleri isolates tested by Aski et al. (2016), Rathlavath et al. (2017), and Soma et al. (2017) were resistant to vancomicyn.

Both strains are resistant to all the ß-lactam antibiotics tested (ampicillin, penicillin, and cefotaxime) (**Table 6**). The resistance of A. buzleri isolates to β-lactams is widespread in seafood and water sources (Collado et al., 2014; Rathlavath et al., 2017; Šilha et al., 2017) as well as in other environments (Kayman et al., 2012; Ferreira et al., 2013; Rahimi, 2014; Yesilmen et al., 2014; Zacharow et al., 2015b; Aski et al., 2016; Van den Abeele et al., 2016; Elmali and Can, 2017; Vicente-Martins et al., 2018). Rathlavath et al. (2017) found that on 40 A. butzleri isolated from shellfish, 100% resulted resistant to cefotaxime and 70% to ampicillin, while among 81 A. butzleri isolated from fish, 98.7% was resistant to cefotaxime and 72.8% to ampicillin. Moreover, of 26 A. butzleri isolated from coastal water, 100% was resistant to cefotaxime and 73% to ampicillin (Rathlavath et al., 2017). Šilha et al. (2017) found that 94.4% of 18 A. butzleri isolated from water sources were resistant to ampicillin and 100% were resistant to penicillin G, while Collado et al. (2014) found that 45.2% of 62 A. butzleri isolated from bivalve molluscs were resistant to ampicillin.

The ß-lactam resistance is generally due to the combined effects of the presence and activity of ß-lactamase genes, of the binding to targets (penicillin-binding proteins) and, in Gram-negative bacteria, of the outer-membrane permeability (Georgopapadakou, 1993). In the genomes of Ab 55 and Ab 6V, we identified three putative ß-lactamases orthologues to that of A. butlzeri RM4018 (Miller et al., 2007) [MBL fold metallo-hydrolase D3M61\_00510 and D3M61\_04375 (Ab 55), D3M75\_05505 and D3M75\_08430(Ab 6V); class D beta-lactamase D3M61\_10735, D3M75\_10375 (Ab 6V)] as well as penicillin binding proteins (Ab 6V D3M75\_03520, D3M75\_10675, D3M75\_10720; Ab 55: D3M61\_02930, D3M61\_10940, D3M61\_10985).

Furthermore, in the genomes of both strains, we retrieved the lrgAB operon, which modulates penicillin tolerance in Staphylococcus (Bayles, 2000; Groicher et al., 2000) and was previously suggested as the ß-lactam resistance enhancer in A. butzleri RM4018 (Miller et al., 2007).

**Figure 3** shows the multialignment of beta-lactamase protein in Ab 55 and Ab 6V and other Arcobacter species. In Ab 6V the predicted protein (D3M75\_10375) is truncated at N-terminal due to a nucleotide mutation in the genomic locus which leads to a premature stop codon. The DNA sequence translated by EMBOSS <sup>R</sup> Sixpack shows that changing the reading frame would recover the entire protein, identical to Ab 55 D3M61\_10735. The sequence of Ab 55 D3M61\_10735 share 100% of identity with OXA-464, which differs with the ß-lactamase of the type strain A. butleri RM4018 for a glutamine instead of glutamic acid in position 177.

Ab 55 and Ab 6V as well as the Ab type strain LMG 10828 (RM4018) were sensitive to the three aminoglycoside antibiotics used in this study namely gentamicin, kanamycin, and streptomycin (**Table 6**). Our results are in agreement with those obtained by Rathlavath et al. (2017) on the 147

TABLE 7 | Cefotaxime, ampicillin, penicillin G, and vancomycin MIC and MBC values for A. butzleri.


<sup>a</sup>Minimal inhibitory concentration. <sup>b</sup>Minimal bactericidal concentration.


A. butzleri isolates from seafood (40 isolates were from shellfish) and coastal water, which were all sensitive to these three antibiotics. Susceptibility to these antibiotics is also consistent to that found by other authors who reported susceptibility to streptomycin and/or gentamicin and/or kanamycin for the great majority (96.4–100%) of A. butzleri isolated from shellfish [gentamicin (Collado et al., 2014)], water sources [gentamicin and streptomycin (Šilha et al., 2017)] and other various sources [gentamicin (Kayman et al., 2012; Ferreira et al., 2013, 2017; Van den Abeele et al., 2016); gentamicin and streptomycin (Šilha et al., 2017); kanamycin and streptomycin (Kabeya et al., 2004); gentamicin, kanamycin and streptomycin (Rahimi, 2014)]. Aminoglycosides are proposed as antibiotics to be used in Arcobacter infections (Rahimi, 2014; Ferreira et al., 2016; Rathlavath et al., 2017). Nevertheless, a study reported high percentage (80%) of resistance to gentamicin and kanamycin in A. butzleri isolates from porcine samples (Scanlon et al., 2013) but only five isolates were tested. Mechanisms of bacterial resistance to this class of antibiotics are diverse including the inactivation by aminoglycoside modifying enzymes, mutation of the ribosome target, and modification of the ribosome by methyltransferase enzymes (Wilson, 2014; Garneau-Tsodikova and Labby, 2016). Resistance may also arise from mutation in the rrs encoding for 16S rRNA, even if these mutations are quite rare, as they would interfere with the vital cellular machinery.

Susceptibility to the hydrophilic fluoroquinolone ciprofloxacin or very low percentage of resistant isolates (0–3.2%) are widely reported in literature (Kayman et al., 2012; Collado et al., 2014; Rahimi, 2014; Ferreira et al., 2017; Rathlavath et al., 2017; Šilha et al., 2017; Soma et al., 2017), although some authors reported percentages ranging from 12.7 to 55.8% of A. butzleri resistant to ciprofloxacin isolated from patients with gastroenteritis, retail food products, and poultry slaughterhouse (Ferreira et al., 2013; Zacharow et al., 2015b; Van den Abeele et al., 2016; Vicente-Martins et al., 2018). The acquisition of fluoroquinolone resistance may represent a serious issue for the health care system since they are the first-choice antibiotic for treating Campylobacter infection in humans and they were suggested to be used in Arcobacter enteritis (Vandenberg et al., 2006; Collado et al., 2014; Ferreira et al., 2016).

Our strains are susceptible to ciprofloxacin. Indeed, we did not find in A. butzleri Ab 55 and Ab 6V the mutation in the quinolone resistance determining region in position 254 of the gyrA gene which causes a transition from cytosine to thymine leading to the substitution of a threonine to isoleucine (Abdelbaqi et al., 2007).

Ab LMG10828<sup>T</sup> (RM 4018) and Ab 6V were intermediate resistant to the hydrophobic quinolone nalidixic acid, whereas Ab 55 was resistant to this antibiotic. However, the absence of gyrA mutations in these strains suggests that the putative resistance could be due to the mechanisms of hydrophobic quinolones uptake as suggested by Miller et al. (2007). High percentage of resistance toward the nalidixic acid was reported for 77.5 and 83.4% of A. butzleri isolated from shellfish by Rathlavath et al. (2017) and Collado et al. (2014), respectively, for 71.6% of A. butzleri isolates from fish (Rathlavath et al., 2017) and for 57.6 and 88.9% of A. butzleri isolates from water sources in two different studies (Rathlavath et al., 2017; Šilha et al., 2017). One hundred percent of A. butzleri isolated from other various sources, namely, retail food products, porcine, slaughterhouse, and dairy plant samples, by different authors, were resistant to nalidixic acid (Scanlon et al., 2013; Elmali and Can, 2017; Ferreira et al., 2017; Vicente-Martins et al., 2018). By contrast, high percentage of susceptibility toward nalidixic acid, ranging from 50 to 77.8%, was reported by Yesilmen et al. (2014) and Kayman et al. (2012) for A. butzleri isolated from milk and dairy products and human gastroenteritis stool samples, respectively. However, no more than 10 A. butzleri isolates were tested in both studies.

Arcobacter butzleri Ab 55 and Ab 6V are intermediate resistant to chloramphenicol (**Table 6**). Differences about susceptibility against this antibiotic were reported among several studies (Ferreira et al., 2016); e.g., Rathlavath et al. (2017) reported high susceptibility rates ranging from 65.3 to 77.7% for 147 A. butzleri isolates from shellfish, fish, and coastal water whereas Šilha et al. (2017) reported that 66.6% of A. butzleri isolated from water sources was resistant. Also for A. butzleri isolated from various sources (poultry meat and other retail food products, poultry slaughterhouse, and animal and human stool samples) variable resistance percentages were reported ranging from 2.3 to 87.7% (Ferreira et al., 2013; Rahimi, 2014; Aski et al., 2016; Šilha et al., 2017; Soma et al., 2017; Vicente-Martins et al., 2018). Otth et al. (2004) suggested that it may depend on local differences in the usage of this antibiotic.

Chloramphenicol resistance commonly consists of its enzymatic inactivation mainly by acetyltransferases or occasionally by phosphotransferases, but additional mechanisms involve the presence of efflux pump which act as extrusion transporters, mutation or modification of the target site, and decreased outer membrane permeability. In the genomes of Ab 55 and Ab 6V, we retrieved the cat3 gene encoding for a type A chloramphenicol O-acetyltransferase (Ab 55 D3M61\_09345; Ab 6V D3M75\_06440), which catalyzes the acetyl-CoAdependent acetylation of chloramphenicol and they resulted as intermediate resistant.

Our A. butzleri strains Ab 55 and Ab 6V are resistant to tetracycline (**Table 6**). As for chloramphenicol, tetracycline susceptibility results vary among studies, even if tetracycline is proposed for treating Arcobacter infections by different authors (Ferreira et al., 2016; Rathlavath et al., 2017). Rathlavath et al. (2017) and Šilha et al. (2017) reported that 100 and 77.8% of A. butzleri isolated from shellfish, fish, coastal water, and water sources, respectively, were susceptible to tetracycline and high susceptibility rates, ranging from 78.2 to 100% were reported also for A. butzleri isolated from poultry meat, human and animal stool samples (Rahimi, 2014; Aski et al., 2016; Šilha et al., 2017). Conversely Vicente-Martins et al. (2018) reported that 95.4% of 65 A. butzleri strains isolated from retail food products was resistant to tetracycline and also Yesilmen et al. (2014) reported that 100% of A. butzleri isolated from milk and cheese was resistant toward this antibiotic, even if only 10 isolates were tested.

Tetracycline resistance may be due to several mechanisms: efflux, modification, protection from the ribosome binding, modification of 16S rRNA at the tetracycline binding site.

These mechanisms are mediated by different proteins, among which Tet(O) and Tet(M) are the most important. These proteins are paralogues of the translational GTPase EF-G which removes tetracycline from its inhibitory site on the ribosome through a GTP-dependent hydrolysis. Both are part of a larger group of proteins called ribosomal protection proteins (RPPs), which also includes Tet(Q), Tet(S), Tet(T), Tet(W), and OtrA (Chopra and Roberts, 2001). In the genomes of our strains, we retrieved proteins predicted as tetracycline resistance protein of class C (MFS transporter-multidrug efflux pump) with 27% identity with the orthologues in E. coli and elongation factors with the same domain found in the C terminus of RPPs Tet(M) and Tet(O), with 65% identity with EF-G of E. coli.

Concerning erythromycin, A. butzleri Ab 55 and Ab 6V are resistant to this antibioitc (**Table 6**). However, different percentages of erythromycin resistance were reported in several studies. In particular, A. butzleri isolated from seafood and water sources, as well as from poultry meat, animal and human stool samples, dairy plant, and cheese were found to be susceptible to this antibiotic at percentages ranging from 65 to 100% (Kayman et al., 2012; Collado et al., 2014; Aski et al., 2016; Ferreira et al., 2017; Rathlavath et al., 2017; Šilha et al., 2017). Zacharow et al. (2015b) and Soma et al. (2017) reported that 62 and 50%, respectively, of A. butzleri isolated from animals, humans, and foods of animal origin were resistant toward this antibiotic whereas Yesilmen et al. (2014) and Scanlon et al. (2013) reported that 80% of A. butzleri isolated from milk, cheese, and porcine samples, respectively, were resistant too, but no more than 10 isolates were tested.

Erythromycin resistance is due, besides the less common mutation in 23S rRNA or ribosomal proteins, to posttranscriptional methylation of an adenine residue in 23S caused by the action of erm class gene-coded methylases in Gram-positive bacteria (Kurincic et al., 2007). Ribosomal RNA small subunit methyltransferases were also present in Ab 55 (D3M61\_05905) and Ab 6V (D3M75\_01835) genomes with 26 and 27% identity with reference sequence of Enterococcus faecium (Accession: YP\_004172630.1), respectively; this protein is also present in the genome of A. butzleri RM4018, although it results intermediate resistant in response to erythromycin. Furthermore, we identified macA and macB genes encoding for macrolide exporter proteins in both Ab 55 and Ab 6V. The single copy of 23S rRNA of both Ab 55 and Ab 6V does not present any of the identified mutation responsible for erythromycin resistance.

### REFERENCES


### CONCLUSION

Genomic analyses herein performed allowed us to confirm the recently (Pérez-Cataluña et al., 2018a,b) suggested amendment of A. butzleri as Al. butzlerii, comb. nov.

Antimicrobial susceptibility tests defined Ab 55 and Ab 6V strains as resistant to vancomycin, tetracyclin, nalidixic acid (only Ab 55 whereas Ab 6V is intermediate resistant), erythromycin, and β-lactam antibiotics. Moreover, in our strains isolated from shellfish, we identified numerous virulence, antibiotic, and heavy metal resistance determinants, also additional to those previously found in the genome sequenced A. butzleri ED-1, isolated from fuel cell, and in the A. butzleri type strain RM 4018, isolated from human gastroenteritis (Miller et al., 2007; Pérez-Cataluña et al., 2018a), leading to hypothesize that shellfish strain may be potentially more virulent.

The findings of food-related A. butzleri support both epidemiological surveillance and food safety risk assessment and management in the shellfish industry.

Further analysis in our laboratories is ongoing to sequence and characterize other A. butleri strains isolated from shellfish and from other food matrices, in order to obtain an updated description of the species and to clarify the role of genetic endowment, as well as of the ecological niches the strain come from, in the pathogenesis of A. butzleri. The genomic sequences here presented, and the novel insights obtained in the present study appreciably contribute to achieve these goals.

### AUTHOR CONTRIBUTIONS

VF conceived the work, interpreted the data, and organized the bioinformatic work. FF performed the genomic sequencing and the bioinformatic work. DC and FB performed the antimicrobial susceptibility testing. VF and FF wrote the manuscript. All the authors contributed to the revision of the manuscript and read and approved the submitted manuscript.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.00670/full#supplementary-material

Ahmed, M. O., and Baptiste, K. E. (2018). Vancomycin-Resistant Enterococci: a review of antimicrobial resistance mechanisms and perspectives of human and animal health. Microbiol. Drug Resist. 24, 590–606. doi: 10.1089/mdr.2017.0147






agriculture production. Food Microbiol. 64, 23–32. doi: 10.1016/j.fm.2016. 12.009


**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 Fanelli, Di Pinto, Mottola, Mule, Chieffi, Baruzzi, Tantillo and Fusco. 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.

# Attenuation of Multiple *Vibrio parahaemolyticus* Virulence Factors by Citral

*Yi Sun1 , Du Guo1 , Zi Hua1 , Huihui Sun1 , Zhanwen Zheng1 , Xiaodong Xia1,2 and Chao Shi1 \**

*1 College of Food Science and Engineering, Northwest A&F University, Yangling, China, 2 Sino-US Joint Research Center for Food Safety, Northwest A&F University, Yangling, China*

Citral was known as a widely used food additive with antimicrobial activity; however, whether it can be a potential therapy for controlling bacterial virulence with less risk of antimicrobial resistance remains to be investigated. Herein, we demonstrated that *Vibrio parahaemolyticus* virulence factors that contribute to infection were effectively inhibited to different degrees by sub-inhibitory concentrations (3.125, 6.25, and 12.5 μg/ml) of citral. Citral exerted strong inhibition of autoinducer-2 production and adhesion to Caco-2 cells. Biofilm formation of *V. parahaemolyticus* was effectively decreased by citral at 30°C and 20°C. Moreover, citral repressed the transcription of genes related to flagella biosynthesis, biofilm formation, type III secretion effectors, and antibiotic resistance, as well as genes contributing to the regulation of quorum sensing and toxin production. Therefore, citral could effectively attenuate multiple virulence properties of *V. parahaemolyticus*, and its effect on *in vivo* infection by *V. parahaemolyticus* needs further investigation.

Keywords: citral, *Vibrio parahaemolyticus*, quorum sensing, biofilm, anti-virulence

### INTRODUCTION

The increasing occurrence of disease outbreaks caused by antibiotic-resistant pathogens is becoming a major cause of mortality worldwide (Letchumanan et al., 2015). Therefore, there is an urgent need to identify novel and alternative strategies to control disease outbreaks. Anti-virulence therapies aim to inhibit the specific functions of pathogens that are required to cause infection (Rasko and Sperandio, 2010). As virulence factors are not necessary for bacterial survival, anti-virulence therapies tend to be less prone to the development of resistance and have less impact on neutral and beneficial host bacteria compared with traditional antimicrobials (Rasko and Sperandio, 2010).

*Vibrio parahaemolyticus* is a halophilic, motile, curved gram-negative bacterium that is frequently associated with foodborne outbreaks of disease, usually following the consumption of contaminated seafood such as raw or improperly cooked shellfish and ready-to-eat foods (Wu et al., 2014; Xie et al., 2016). *V. parahaemolyticus* contains various virulence factors responsible for several distinct diseases, including wound infections, human acute gastroenteritis, septicemia, and even death and is therefore a significant cause for concern regarding seafood safety (Yeung et al., 2002).

The virulence factors of *V. parahaemolyticus* are complex and interactive. The bacterium readily forms biofilms on food-processing surfaces such as kitchen cutting boards and stainless steel and can adhere to human intestinal cell lines, contributing to cross-contamination and

#### *Edited by:*

*Learn-Han Lee, Monash University Malaysia, Malaysia*

#### *Reviewed by:*

*Qingping Zhong, South China Agricultural University, China Iddya Karunasagar, Nitte University, India Veronica Lazar, University of Bucharest, Romania Mariana Carmen Chifiriuc, University of Bucharest, Romania*

*\*Correspondence:* 

*Chao Shi meilixinong@nwsuaf.edu.cn*

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 17 October 2018 Accepted: 08 April 2019 Published: 25 April 2019*

#### *Citation:*

*Sun Y, Guo D, Hua Z, Sun H, Zheng Z, Xia X and Shi C (2019) Attenuation of Multiple Vibrio parahaemolyticus Virulence Factors by Citral. Front. Microbiol. 10:894. doi: 10.3389/fmicb.2019.00894*

Sun et al. Citral Attenuates *V. parahaemolyticus* Virulence

the ensuing diseases (Chiu et al., 2006; Mizan et al., 2016). During the establishment of infection, motility (swimming and swarming) is an important function of biofilm formation and adhesion (Zhu et al., 2013). Following intimate contact with the host cells, *V. parahaemolyticus* then secretes toxin proteins and delivers type III secretion system (T3SS) effectors into the host cell cytoplasm, inducing cytotoxicity and bacterial enterotoxicity (Makino et al., 2003; Park et al., 2004). In *V. parahaemolyticus*, quorum sensing (QS) is a cell-to-cell communication system that responds to fluctuations in cell population density through the secretion of autoinducer-2 (AI-2) (Mizan et al., 2016). Once AI-2 has reached a critical threshold concentration, the QS system begins to induce the expression of several virulence factors, including motility (Zhu et al., 2013), biofilm formation (Mizan et al., 2016), adhesion (LaSarre and Federle, 2013), T3SSs (Henke and Bassler, 2004), and toxin production (Zhu et al., 2002).

Essential oils have long been recognized as eco-friendly anti-microbial materials with low mammalian toxicities (Isman, 2000). Citral (3,7-dimethyl-2,6-octadienal) is the principal component of lemongrass oil and has strong antimicrobial activity (Adukwu et al., 2016). Additionally, citral is widely used as a health-promoting food additive for human and animal (FDA, GRAS, 21 CFR 182·60). Because of its strong antimicrobial activity and wide application in food products, the potential of citral to affect bacterial virulence factors should be recognized.

To date, anti-QS approach is one of the most intensively studied in anti-virulence therapies (LaSarre and Federle, 2013). In addition, therapies aimed at attenuating pili, secretion systems, or toxin production have also been reported, respectively (Rasko and Sperandio, 2010). However, the mechanism of *V. parahaemolyticus* pathogenesis is complicated and remains unclear; thus, alternative approaches that interfere with multiple virulence factors should be given more attention. The aim of the current study was to investigate the effect of citral at sub-inhibitory concentrations on the various virulence properties of *V. parahaemolyticus*. The effects of citral on QS, motility, biofilm formation, and adhesion of *V. parahaemolyticus* were investigated, and the effects of citral on the transcription of virulence-associated genes and antibiotic resistance genes were also examined.

### MATERIALS AND METHODS

### Reagents

Citral (CAS:5392-40-5) was obtained from the Chengdu Must Bio-technology Co., Ltd. (Chengdu, Sichuan, China) at a high-performance liquid chromatography purity of at least 99%. The desired concentrations of citral solution were freshly prepared in 0.1% dimethyl sulfoxide before use. All other chemicals were of analytical grade and were unaltered.

### Bacterial Strains and Growth Conditions

*V. parahaemolyticus* strains ATCC 17802 and ATCC 33847 (American Type Culture Collection, Manassas, USA) were used in this study. A further 46 *V. parahaemolyticus* isolates recovered from marine products collected by the Hong Kong Polytechnic University were also included. A loopful of each strain was inoculated into 30 ml of tryptic soy broth (TSB; Difco, Franklin Lakes, NJ, USA) containing 3% (wt/vol) NaCl and incubated at 37°C. Subsequently, the *V. parahaemolyticus* cultures were centrifuged at 8,000 × *g* for 5 min at 4°C, washed, and then re-suspended in fresh TSB (3% NaCl). *Vibrio harveyi* strains BB170 and BB120 (ATCC) were used for the detection of AI-2 in QS inhibition assays.

### Minimum Inhibitory Concentration Assay

The minimum inhibitory concentrations (MICs) of citral against two *V. parahaemolyticus* ATCC strains (17802 and 33847) and four isolates were determined using the agar dilution method (Li et al., 2014). Different concentrations of citral (from 25 to 400 μg/ml) were mixed with warm tryptone soya agar (TSA) supplemented with 3% NaCl and transferred into sterile 24-well plates. Aliquots (2 μl) of *V. parahaemolyticus* suspension (~105 colony-forming units, CFU) were then spotted onto the medium, and the samples were incubated at 37°C for 24 h. The MIC was defined as the minimum concentration of citral at which there was no visible growth of the test strain. Kanamycin (100 mg/L) was used as a positive control. Results were calculated as a mean of experiments performed in triplicate.

### Sub-inhibitory Concentration Assay

A broth dilution assay (Shi et al., 2017) with slight modifications was performed to identify the concentrations of citral that had no effect on the growth of *V. parahaemolyticus*. Equivalent volumes (125 μl) of bacterial suspension (~104 CFU) and citral solution were transferred into 96-well microtiter plates to give final citral concentrations of 0 (control), 3.125, 6.25, 12.5, 25, 50, 100, and 200 μg/ml. The samples were incubated for 24 h at 37°C, and cell density was measured as optical density (OD) at 600 nm using an automatic growth curve analyzer (Labsystems, Helsinki, Finland). Medium without bacteria was used as the negative control. OD measurements were taken in triplicate from three independent experiments to obtain a mean value for each citral concentration.

### Sub-inhibitory Concentration and Autoinducer-2 Determination Assay

To detect and quantify AI-2 production by *V. parahaemolyticus*, a bioluminescent bacterial reporter strain called *V. harveyi* BB170, which produces light in response to AI-2, was used in the assay (Han et al., 2016). The sub-inhibitory concentration (SIC) of citral for *V. harveyi* BB170 was first determined as described previously. The effect of citral on AI-2 production was then determined according to a previously described method (Han et al., 2016), with minor modifications. Briefly, a log-phase culture of *V. parahaemolyticus* ATCC 17802 was centrifuged (8,000 × *g*, 5 min, 4°C) and the cell pellet was re-suspended in TSB to an OD600 nm = 0.5. The suspension was then treated with 3.125 or 6.25 μg/ml of citral and incubated with shaking for 6 h at 30°C. Supernatant containing QS molecules was obtained by centrifuging the cultures at 8,000 × *g* for 10 min at 4°C. The supernatants were passed through 0.2-μm Tuffryn syringe filters and stored at −20°C. The cell-free supernatants were then tested for the presence of autoinducers that could induce luminescence in *V. harveyi* reporter strain BB170. For this bioassay, *V. harveyi* strain BB170 was grown overnight at 30°C with aeration in autoinducer bioassay (AB) broth and then diluted 1:1,000 in AB medium. Diluted strain BB170 was then added along with each individual cell-free supernatant to 50-ml tubes and incubated for 16 h at 30°C with shaking at 180 × *g* to allow the development of luminescence by the reporter strain. Aliquots (100 μl) of each of the samples were transferred to 96-well white microtiter plates and luminescence was measured using a microplate reader (Tecan, Infinite M200 PRO, Männedorf, Switzerland). *V. harveyi* strain BB120 (which produces AI-1 and AI-2) was used as a positive control and was grown overnight at 30°C with shaking in LB broth, following which, 1 ml of cell-free supernatant from the culture was prepared as described previously.

### Motility Assay

Swimming and swarming motility were assessed as described by Packiavathy et al. (2012). For the swimming motility assay, 20 ml of LB broth containing 0.3% (wt/vol) agar was used. Citral was added to the warm (45°C) semi-solid agar medium to obtain final concentrations of 0, 3.125, 6.25, and 12.5 μg/ml. After being dried for 1 h, the semi-solid agar plates were spotted with 5-μl volumes of *V. parahaemolyticus* culture (~1 × 106 CFU, at the center of the plate) and incubated at 37°C for 7 h. To examine swarming motility, 20 ml of LB broth was supplemented with 0.5% (wt/vol) agar and 0.5% (wt/vol) glucose and mixed with citral at final concentrations of 0, 3.125, 6.25, and 12.5 μg/ml. Aliquots (5 μl) of bacterial culture were then stabbed into the medium and plates were incubated upside down at 37°C for 20 h. The diameter of the motility area was measured using AutoCAD. Medium without citral was used as a control.

### Congo Red Agar Assay

To visually detect biofilm formation by the 46 *V. parahaemolyticus* isolates, a Congo red agar (CRA) assay was performed as previously described by Wojtyczka et al. (2014). The isolates were inoculated onto brain-heart infusion (BHI) agar medium supplemented with 5% (wt/vol) sucrose and Congo red, samples were incubated for 24 h at 37°C. Under these conditions, biofilm producers form black crusty colonies on the CRA-BHI plates, whereas non-producers form red colonies.

### Crystal Violet Assay

Biofilm formation was also examined with respect to biomass using a Crystal violet (CV) staining method described by Naves et al. (2008), with some modifications. Log-phase cultures of *V. parahaemolyticus* ATCC 17802, VP244, and VP253 were separately centrifuged (8,000 × *g*, 5 min, 4°C) and re-suspended in TSB (3% NaCl). Then, 250-μl aliquots of the cell suspensions (OD600 nm = 1) supplemented with citral (0, 3.125, 6.25, and 12.5 μg/ml) were inoculated into sterile 96-well polystyrene tissue culture plates and incubated for 24, 48, or 72 h without agitation. Uninoculated LB broth was used as a control. Six wells were used per strain. At each time point, bacterial growth was determined by measuring the OD at 630 nm using a microplate spectrophotometer (model 680; Bio-Rad, Hercules, CA, USA). Growth medium was then carefully removed, and each well was rinsed once with distilled water to remove any unbound bacteria. After being air-dried for 30 min, biofilms were stained with 250 μl of 1% (wt/vol) CV solution (Tianjin Kermel Chemical Regent Co., Tianjin, China) for 20 min at room temperature. The CV dye was then discarded, and the wells were rinsed three times with distilled water to remove any unbound colorant. After drying, the stained biofilm was solubilized in 250 μl of 33% (vol/vol) glacial acetic acid for 20 min and the OD570 nm was measured. The specific biofilm formation (SBF) was calculated by attaching and stained bacteria (OD570 nm) normalized with cell growth (OD630 nm). The experiment was repeated at least three times.

### Field Emission Scanning Electron Microscopy

Field emission scanning electron microscopy (FE-SEM) was used to assess the effect of citral on the biofilm morphology of *V. parahaemolyticus* ATCC 17802 as described previously (Li et al., 2014), with some modifications. Bacterial cells (OD600 nm = 1) were treated with citral (0, 3.125, 6.25, or 12.5 μg/ml) and then incubated at 30°C for 72 h to allow biofilm formation on coverslips. Bacterial suspensions were removed and then the coverslips were gently washed twice with phosphate-buffered saline (PBS, pH = 7.0) before the addition of 2.5% (vol/vol) glutaraldehyde and incubation overnight at 4°C to fix the cells. The samples were then serially dehydrated with ethanol (30, 50, 70, 80, 90, and 100%) for 10 min each. The samples were sputter-coated with gold under vacuum conditions and visualized using a scanning electron microscope (S-4800; Hitachi, Tokyo, Japan).

### Adhesion of Caco-2 Cells

Human colon adenocarcinoma cell line Caco-2 was maintained in Dulbecco's Modified Eagle Medium (DMEM) (Gibco, Grand Island, NY, USA) supplemented with 10% (vol/vol) fetal bovine serum (Hyclone, Logan, UT, USA), 1% (vol/vol) non-essential amino acids (Gibco), and 1% (vol/vol) double antibiotic solution (100 U/ml penicillin and 100 μg/ml streptomycin, Hyclone). Maintenance of the cell lines and subsequent experiments were carried out at 37°C in a humidified atmosphere containing 5% CO2.

To evaluate the effect of citral on the adhesion of *V. parahaemolyticus* ATCC 17802 to Caco-2 cells, an adhesion assay was performed as described previously (Amalaradjou et al., 2014). Caco-2 cells were seeded in 24-well tissue culture plates (105 cells/well) containing supplemented DMEM and incubated for 24 h. *V. parahaemolyticus* ATCC 17802 was grown to log phase with and without SICs of citral (3.125, 6.25, or 12.5 μg/ml), centrifuged, and then re-suspended in cell culture medium without antibiotics. The Caco-2 cells were rinsed then with PBS before the addition of culture medium containing the *V. parahaemolyticus* suspension (107 CFU, MOI = 10). The tissue culture plates were centrifuged at 600 × *g* for 5 min and incubated at 37°C in a humidified 5% CO2 incubator for 2 h. The culture medium was removed, then the infected monolayer cells were rinsed three times with PBS, and lysed with 0.1% Triton X-100 (Amresco, Solon, OH, USA). The number of viable adherent *V. parahaemolyticus* cells was determined by serial dilution and plating on TSA (3% NaCl) plates.

### Quantitative Real-Time Polymerase Chain Reaction

The effects of citral on the transcription of *V. parahaemolyticus* virulence genes (*flaA*, *flgM*, *flgL*, *ompW*, *VP0950*, *VP0952*, *VP0962*, *luxS*, *aphA*, *vopQ*, *vpA0450*, and *toxR*) and antimicrobial peptide (AMP)-resistant genes (*tolC*, *nusA*, *atpA*, *dld* and *fla*),were examined using a real-time quantitative polymerase chain reaction (RT-qPCR) assay. Total RNA was extracted from log-phase bacterial cultures grown with and without SICs of citral using a RNAprep Pure Bacteria Kit (Tiangen, Beijing, China) according to the manufacturer's protocol. After measuring RNA concentrations using a nucleic acid and protein spectrophotometer (Nano-200; Aosheng Instrument Co., Hangzhou, China), cDNA was synthesized using a PrimeScript RT Reagent Kit (Takara, Kyoto, Japan) according to the manufacturer's instructions. Primers used for RT-qPCR are listed in **Table 1**. RT-qPCR reactions were carried out in a 25-μl reaction volume using SYBR Premix Ex Taq II (Takara). The thermal cycler parameters were 95°C for 30 s, 40 cycles of

TABLE 1 | Effects of sub-inhibitory concentrations of citral on the transcription of virulence-associated genes and antibiotic resistance genes in *Vibrio parahaemolyticus* ATCC 17802.


*\*p ≤ 0.05; \*\*p ≤ 0.01; c F, forward; R, reverse.* 95°C for 5 s, and 60°C for 30 s, followed by dissociation steps of 95°C for 15 s and 60°C for 30 s. All samples were analyzed in triplicate and normalized to the endogenous control (*puvA*) gene (Coutard et al., 2007). Samples were run on an IQ5 system (Bio-Rad), and the transcription of target genes versus *puvA* was determined as previously described (Shi et al., 2017).

### Statistical Analysis

All experiments were performed at least in triplicate. Statistical analyses were performed using SPSS software (version 19.0; SPSS, Inc., Chicago, IL, USA). The data were presented as the mean values ± SD (*n* = 3) and differences between means were tested by Student's *t*-test. Differences were considered significant at *p* ≤ 0.05.

### RESULTS

### Determination of Minimum Inhibitory Concentrations

The MICs of citral for six *V. parahaemolyticus* strains ranged from 100 to 300 μg/ml (**Table 2**). *V. parahaemolyticus* ATCC 17802 was the most susceptible to citral (MIC = 100 μg/ml), while the four *V. parahaemolyticus* isolates showed a three-fold higher tolerance to citral (MIC = 300 μg/ml).

### Determination of Sub-inhibitory Concentration and Inhibition of Autoinducer-2 Quorum Sensing Signaling

The growth of *V. parahaemolyticus* was suppressed at concentrations of citral above 12.5 μg/ml (**Figure 1A**). Therefore, 3.125, 6.25, and 12.5 μg/ml were chosen as the SICs for further virulence-related assays.

At concentrations below 6.25 μg/ml, citral exhibited no inhibition of *V. harveyi* strain BB170 growth (**Figure 1B**). Production of AI-2 by *V. parahaemolyticus* ATCC 17802 was reduced by 42 and 58% following exposure to 3.125 and 6.25 μg/ml of citral, respectively (*p* ≤ 0.01) (**Figure 1C**).

### Inhibition of Swimming and Swarming Motility

Citral effectively reduced the swimming and swarming motility (**Figure 2**) of *V. parahaemolyticus* in a concentration-dependent manner. At 6.25 and 12.5 μg/ml, citral significantly decreased

TABLE 2 | Minimum inhibitory concentrations (MICs) of citral against different strains of *Vibrio parahaemolyticus*.


the diameter of the swimming area by 20 and 47%, respectively (*p* ≤ 0.01), while 3.125, 6.25, and 12.5 μg/ml of citral significantly decreased the swarming area by 22, 35, and 50%, respectively (*p* ≤ 0.01).

### Reduction in Biomass

Only two of the 46 tested marine product-derived *V. parahaemolyticus* isolates (VP244 and VP253) formed black crusty colonies on CRA medium, indicating biofilm formation.

The SBF of VP253 at 30°C was lower than that at 20°C, while the other two strains (ATCC 17802 and VP244) were not affected by the temperature change (**Figure 3**). At concentrations of 3.125, 6.25, and 12.5 μg/ml, citral significantly (*p* ≤ 0.05) reduced biofilm formation by *V. parahaemolyticus* strains ATCC 17802, VP244, and VP253 at both 20 and 30°C in both a concentration- and time-dependent manner. Following incubation for 72 h, citral (12.5 μg/ml) caused a greater decrease in *V. parahaemolyticus* ATCC 17802 biofilm density at 30°C (67.97% reduction compared with no-citral control) (**Figure 3D**) than at 20°C (55.73% reduction) (**Figure 3A**). In contrast, a greater decrease was observed for strains VP244 and VP253 at 20°C (**Figures 3B,C**) compared with at 30°C (**Figures 3E,F**).

### Observed Changes in Biofilm Structure Following Citral Treatment by Field Emission Scanning Electron Microscopy

The architectural integrity of biofilm and the aggregation of cells were significantly altered following exposure to citral (3.125, 6.25, or 12.5 μg/ml) (**Figure 4**). Biofilm formed by *V. parahaemolyticus* control cultures displayed firm threedimensional, multicellular, complex, self-assembled structures that contained extracellular polymeric substances (EPS) (**Figure 4A**). With increasing concentrations of citral, the *V. parahaemolyticus* cells secreted a lesser amount of EPS and the biofilm structure became loose (**Figures 4B,C**). At a citral concentration of 12.5 μg/ml, the biofilm structure was completely disrupted, with individual, dispersed cells (**Figure 4D**).

### Interruption of *V. parahaemolyticus* Adhesion to Caco-2 Cells

Citral significantly (*p* ≤ 0.01) inhibited the ability of *V. parahaemolyticus* to adhere to Caco-2 cells in a dose-dependent manner (**Figure 5**). The adherence of *V. parahaemolyticus* cells pre-exposed to 3.125, 6.25, or 12.5 μg/ml of citral was reduced by 35, 59, and 65%, respectively, compared with the control (*p* ≤ 0.01).

### Down-regulation of Virulence-Associated Genes and Antimicrobial Peptide-Resistant Genes

Citral downregulated the transcription of genes associated with flagella regulation (*flaA*, *flgM*, *flgL*), biofilm formation (*ompW*, *vp0950*, *vp0952*, *vp0962*), QS regulation (*luxS*, *aphA*), T3SS1 (*vopQ*, *vpA0450*), toxin production (*toxR*), and AMP-resistant genes (*tolC*, *nusA*, *atpA*, *dld* and *fla*) to various degrees (**Table 1**).

different concentrations of citral. Each value represents the average of three independent measurements. (B) Growth of *Vibrio harveyi* BB170 treated with different concentrations of citral. Each value represents the average of three independent measurements. (C) Inhibition of AI-2 production by *Vibrio harveyi* BB170 at sub-inhibitory concentrations of citral. Bars represent the standard deviation (*n* = 3). \*\**p* ≤ 0.01.

Among these, the most significant effects were observed following treatment with 12.5 μg/ml of citral, with greater than nine-fold decreases in *flaA*, *luxS*, and *vp0950* transcription.

### DISCUSSION

The global disease outbreaks and food contamination caused by *V. parahaemolyticus* underscore the importance of controlling the expression of virulence factors by this important foodborne pathogen. Citral has previously been shown to effectively control separate and specific virulence factors in different pathogens (Echeverrigaray et al., 2008; Ahmad et al., 2015). However, the effect of citral on virulence factors and antimicrobial resistance of *V. parahaemolyticus* still needs to be adequately investigated.

In previous studies, the MIC of citral was found to be 584 μg/ml for *Cronobacter sakazakii* (Shi et al., 2017), while curcumin inhibits *V. parahaemolyticus* and other *Vibrio* spp. with MICs exceeding 150 μg/ml (Packiavathy et al., 2013). In the current study, citral showed effective antibacterial activity against *V. parahaemolyticus* strains, with MICs ranging from 100 to 300 μg/ml.

*V. parahaemolyticus* can alternate between two cell types depending on the growth conditions (Zhu et al., 2013). When grown in low viscosity and liquid culture, the cells appear as short rods with a single-sheathed polar flagellum, which is used for swimming and overcoming repulsive forces between the bacteria and the host tissues. However, when grown on solid surfaces and in high viscosity medium, the cells switch to a swarmer cell type and utilize their lateral flagellum to increase the surface tension, allowing them to aggregate and form a biofilm. FlaA, a specific polar flagellin, can mediate the formation of the flagellar filament, resulting in swimming motility (Loh et al., 2004). FlgL is the hook-associated protein 2 that plays an important role in polar flagellation and adhesion to host cells (Kim et al., 2008). FlgM is an anti-σ factor, which can mirror the favorable conditions for swarming motility, resulting in an increase in flagellar filament numbers and switching the cell type to swarmer (Zhu et al., 2013). Echeverrigaray et al. (2008) demonstrated that citral effectively inhibited the swarming ability of *Proteus mirabilis*. In this study, citral effectively repressed both swarming and swimming motility of *V. parahaemolyticus*. Moreover, citral downregulated the transcription of *flaA*, *flgL*, and *flgM*. It was hypothesized that citral reduced the secretion of FlaA, FlgM, and FlgL, which impeded the ability of *V. parahaemolyticus* to recognize favorable attachment surfaces and the biosynthesis of polar and lateral flagella. Biofilm formation may be a major factor in the lowered shelf-life of *V. parahaemolyticus*-contaminated seafood and aid in the transmission of disease (Mizan et al., 2016). The CV assay showed that citral reduced the biofilm biomass of *V. parahaemolyticus* in a concentration-dependent manner at both 20 and 30°C within 3 days. A previous study showed that temperature influenced the production of EPS, which is related to biofilm formation (Garrett et al., 2008). In this study, the biofilm formation of isolate VP253 was decreased at 30°C, possibly because the isolate was more sensitive to the high temperature. Moreover, citral caused greater biofilm biomass reduction of *V. parahaemolyticus* isolates VP244 and VP253 at 20°C, while it showed more effective inhibition of biofilm formation by *V. parahaemolyticus* ATCC 17802 at 30°C. It could be due to the fact that the optimal temperature for EPS secretion of *V. parahaemolyticus* ATCC 17802 might be 30°C, while for the two other isolates, 20°C may be the optimal temperature. In line with the CV results, the FE-SEM images showed that

the structure of the *V. parahaemolyticus* biofilm was obviously affected by the treatment of citral. The VP0950 (encoding a lipoprotein-related protein), VP0952, and VP0962 (encoding hypothetical proteins) were parts of biofilm (Boyd et al., 2008). The outer membrane proteins (OMPs) play an important role in nutrient uptake and in interactions with the environment and host tissues (Ritter et al., 2012). The transcription of *ompW* was related to the biofilm formation in *Pseudoalteromonas* sp. D41 (Ritter et al., 2012). In this study, citral impeded biofilm development of *V. parahaemolyticus* strains, possibly by damaging the biosynthesis of biofilm-associated proteins, repressing the expression of OMP-associated genes (such as *ompW*) and therefore the transportation of substances associated with the biofilm formation. Bacterial adherence to epithelial cell surfaces is a key stage in their survival and colonization of the gastrointestinal tract (Rasko and Sperandio, 2010). We found that the number of *V. parahaemolyticus* cells adhered to Caco-2 cells was decreased after pretreatment with citral. This finding is in agreement with the results of Shi et al. (2017), who showed that citral effectively inhibited the adhesion of *C. sakazakii* ATCC 29544 to Caco-2 cells. Additionally, Kirov (2003) reported that lateral flagella of *Aeromonas* strains caused persistent and dysenteric infections in the gastrointestinal tract. It was likely that the inhibition of flagella biosynthesis by citral contributed to the attenuation of *V. parahaemolyticus* adherence to Caco-2 cells.

T3SS1 effectors help *V. parahaemolyticus* to evade the host immune response, inducing autophagy followed by efficient lysis of the infected host cells, as well as causing cytotoxicity in host cells (Park et al., 2004). Among these effectors, VopQ causes rapid induction of autophagy in target cells, while VPA0450 destabilizes the cell by interfering with the association between the actin cytoskeleton and the cell membrane (Zhang and Orth, 2013). In this study, citral effectively downregulated the transcription of genes coding for the VopQ and VPA0450 effectors. As a result of citral-induced inhibition of these important effectors, *V. parahaemolyticus* might be more easily eliminated by the host immune response and find it more difficult to invade host cells. Additionally, flagella contain a sophisticated export apparatus involved in the secretion of several virulence factors that is closely related to type III secretion pathways (Kirov, 2003). The damage to the flagella structure caused by citral may influence type III secretion pathways, thereby impeding the delivery of effectors.

ToxR coordinately regulates several virulence-associated genes, including the *tcp* genes (toxin-coregulated pilus) and the *ompU* and *ompT* genes (major outer membrane proteins) in *Vibrio cholerae* (Lee et al., 2000; Childers and Klose, 2007). Moreover, Vp-ToxR directly promoted the expression of *tdh2* and resulted in the development of Kanagawa phenomenon-positive virulent strains (Lin et al., 1993). Thermostable direct hemolysin (TDH)

FIGURE 3 | The effects of citral on biofilm formation by *Vibrio parahaemolyticus* ATCC 17802 (A,D), VP244 (B,E) and VP253 (C,F) at 20°C (A–C) and 30°C (D–F). Bars represent the standard deviation (*n* = 3). \**p* ≤ 0.05, \*\**p* ≤ 0.01.

FIGURE 4 | The effects of SICs of citral *Vibrio parahaemolyticus* ATCC 17802 as observed by field-emission scanning electron microscopy (4,000 × magnification). (A–D) Control, 3.125 μg/ml, 6.25 μg/ml, and 12.5 μg/ml citral, respectively.

is a protein toxin that has several biological functions, including hemolytic, enterotoxic, and cytotoxic activities (Lin et al., 1993). We observed that citral effectively downregulated the expression of *toxR* in *V. parahaemolyticus*, which possibly reduced the secretion of TDH or other toxins.

AI-2, a dihydroxy pentanedione-derived molecule synthesized by LuxS-like synthases, plays a role in inter-species communication in a wide variety of bacteria. At the low cell density, AphA is increasingly expressed to trigger the transcription of virulence genes which associated with infection (Wang et al., 2013). In this study, citral effectively repressed the biosynthesis of AI-2 and the transcription of *luxS* and *aphA* in *V. parahaemolyticus*. Similarly, a minimum citral concentration of 0.016 mg/ml inhibited QS by *Pseudomonas aeruginosa* (Ahmad et al., 2015). Low concentrations (100 μmol/L) of cinnamaldehyde were also effective at inhibiting AI-2-mediated QS in *V. harveyi* BB170 (Niu et al., 2006).

A *luxS* null mutation was reported to eliminate growth-phasedependent control of *flaA* in *Helicobacter pylori* and to downregulate *flgM* transcription in *Escherichia coli* K12 (Loh et al., 2004; Ling et al., 2010), while AphA is associated with the biofilm formation and motility in *V. parahaemolyticus* (Wang et al., 2013).

Moreover, QS appeared to repress ToxR-regulated virulence genes in *V. cholerae* (Zhu et al., 2002). In the present study, determining the effects of citral on multiple virulence targets could not exclude the influence of QS. The reduction of AI-2 may contribute to the changes in some traits during citral treatment, but whether it is the determining factor needs further investigation.

The growing emergence of antimicrobial-resistant *V. parahaemolyticus* becomes a challenge of controlling *V. parahaemolyticus* infections and food contamination (Xie et al., 2016). The OMPs (TolC and flagellin), transcription termination factor (NusA), ATP synthase F1, alpha subunit (F1-ATPa), and dihydrolipoamide dehydrogenase (DLD) were associated with AMP-resistance (Shen et al., 2010). This study showed that citral downregulated multiple AMP-resistant genes (*tolC*, *nusA*, *atpA*, *dld* and *fla*) to various levels. It could be speculated that citral played a role in damaging the multidrug efflux transporter and membranes of *V. parahaemolyticus*, therefore reduced the resistance of *V. parahaemolyticus*.

### REFERENCES


### CONCLUSIONS

In conclusion, this investigation indicated that citral attenuated multiple virulence factors of *V. parahaemolyticus*, including QS, motility, biofilm formation, the adhesion to Caco-2 cells, and repressed the expression of genes related to flagella (polar and lateral), biofilm and T3SS1 effectors, virulence regulators (*luxS*, *aphA* and *toxR*), and AMP resistance. The results obtained in this work demonstrate the inhibitory effect of citral on virulence factors of *V. parahaemolyticus*. However, the data reported in this study only demonstrate the anti-virulence effect of citral *in vitro*. Further research is needed to clarify the mode of the antivirulence action of citral and to investigate its effects in experimental animal models, aiming toward the application of citral as an alternative strategy to control the infections of *V. parahaemolyticus*.

### AUTHOR CONTRIBUTIONS

CS and YS conceived and designed the experiments. DG, ZH and HS performed the experiments. ZZ analyzed the data. XX contributed reagents, materials, and analysis tools. YS and CS wrote the manuscript.

### FUNDING

This work was supported by the Fundamental Research Funds for the Central Universities (2452017228); National Natural Science Foundation of China (31772084); and General Financial Grant from the China Postdoctoral Science Foundation (2017M623256).

### ACKNOWLEDGMENTS

We thank Dr. Baowei Yang and Dr. Guoyun Zhang in Northwest A&F University for technical assistance, Dr Hongyu Tian in Beijing Technology and Business University for writing assistance.

boards and food contact surfaces. *Lett. Appl. Microbiol.* 43, 666–672. doi: 10.1111/j.1472-765X.2006.02006.x


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

*Copyright © 2019 Sun, Guo, Hua, Sun, Zheng, Xia and Shi. 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.*

# Random Mutagenesis Applied to Reveal Factors Involved in Oxidative Tolerance and Biofilm Formation in Foodborne Cronobacter malonaticus

Maofeng Zhang1,2† , Xiyan Zhang<sup>1</sup>† , Liaowang Tong<sup>1</sup>† , Dexin Ou<sup>1</sup> , Yaping Wang<sup>1</sup> , Jumei Zhang<sup>2</sup> , Qingping Wu<sup>2</sup> \* and Yingwang Ye1,2 \*

<sup>1</sup> School of Food Science and Engineering, Hefei University of Technology, Hefei, China, <sup>2</sup> State Key Laboratory of Applied Microbiology Southern China, Guangdong Provincial Key Laboratory of Microbial Culture Collection and Application, Guangdong Institute of Microbiology, Guangzhou, China

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Chaoxin Man, Northeast Agricultural University, China Peng Fei, Henan University of Science and Technology, China

#### \*Correspondence:

Qingping Wu wuqp203@163.com Yingwang Ye yeyw04@mails.gucas.ac.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 19 January 2019 Accepted: 05 April 2019 Published: 01 May 2019

#### Citation:

Zhang M, Zhang X, Tong L, Ou D, Wang Y, Zhang J, Wu Q and Ye Y (2019) Random Mutagenesis Applied to Reveal Factors Involved in Oxidative Tolerance and Biofilm Formation in Foodborne Cronobacter malonaticus. Front. Microbiol. 10:877. doi: 10.3389/fmicb.2019.00877 Cronobacter species are linked with life-treating diseases in neonates and show strong tolerances to environmental stress. However, the information about factors involved in oxidative tolerance in Cronobacter remains elusive. Here, factors involved in oxidative tolerance in C. malonaticus were identified using a transposon mutagenesis. Eight mutants were successfully screened based on a comparison of the growth of strains from mutant library (n = 215) and wild type (WT) strain under 1.0 mM H2O2. Mutating sites including thioredoxin 2, glutaredoxin 3, pantothenate kinase, serine/threonine protein kinase, pyruvate kinase, phospholipase A, ferrous iron transport protein A, and alanine racemase 2 were successfully identified by arbitrary PCR and sequencing alignment. Furthermore, the comparison about quantity and structure of biofilms formation among eight mutants and WT was determined using crystal violet staining (CVS), scanning electron microscopy (SEM), and confocal laser scanning microscopy (CLSM). Results showed that the biofilms of eight mutants significantly decreased within 48 h compared to that of WT, suggesting that mutating genes play important roles in biofilm formation under oxidative stress. The findings provide valuable information for deeply understanding molecular mechanism about oxidative tolerance of C. malonaticus.

Keywords: Cronobacter malonaticus, random mutagenesis, arbitrary PCR, oxidative stress, biofilm formation

### INTRODUCTION

Cronobacter species are important foodborne pathogens causing life-threating infections in infants (Van Acker et al., 2001; Healy et al., 2010). Contaminated powdered infant formula (PIF) is considered to be the major transmission route of Cronobacter infections (Biering et al., 1989; Van Acker et al., 2001; Norberg et al., 2012; Ye et al., 2014). So, the high risks of Cronobacter strains in powdered infant formula on newborn's health has arouse public concerns.

Cronobacter spp. show unusual abilities to survive under environmental stress (Gurtler et al., 2005). To date, the genus of Cronobacter includes C. sakazakii, C. malonaticus, C. turicensis, C. muytjensii, C. dublinensis, C. universalis, and C. condiment (Iversen et al., 2008). The factors involved in oxidative stress in C. sakazakii have been reported. For example, polymorphisms in RpoS sequence and Significant heterogeneity of stress tolerance including oxidative stress among

**270**

natural isolates of C. sakazakii has been described (Alvarez-Ordóñez et al., 2012). Johler et al. (2010) demonstrated that genes including crtX, crtE, and crtY involved in yellow pigmenting of C. sakazakii ES5 affected tolerance to oxidative stress. In C. sakazkaii ATCC29544, Hfq, an RNA chaperone, has been found to increase the tolerance to oxidative stress (Kim et al., 2015). C. malonaticus has been implicated in infections in infant and adults (Forsythe et al., 2014; Alsonosi et al., 2015). PIF is the major source of C. malonaticus (Ogrodzki and Forsythe, 2015, 2017). Hydrogen peroxide (H2O2) is a well-studied sanitizer for inactivate foodborne pathogens. In addition, Ye et al. (2018) determined the inhibitory effects of H2O<sup>2</sup> on C. malonaticus cells and its biofilm formation. However, information about factors involved in oxidative tolerance in C. malonaticus is largely unknown.

In this study, a transposon mutagenesis approach was applied to reveal the factors involved in resistance to oxidative stress, and the biofilm formation among mutants and WT strains were further detected using crystal violet staining (CVS), scanning electron microscopy (SEM), and confocal laser scanning microscopy (CLSM) to reveal potential relationship between oxidative stress and biofilm formation.

### MATERIALS AND METHODS

### The Development of Mutants Library

The procedure of transposon mutagenesis approach was performed as described by Zhang et al. (2018).

### Screening of Mutants Tolerance to Oxidative Stress

For screening positive mutants tolerant to oxidative stress, overnight culture (OD<sup>600</sup> = 0.8, v/v, 1%) was inoculated into LB broth (Luqiao, Beijing) with 1.0 mM H2O<sup>2</sup> at 37◦C for 8 h. Growth of mutants (n = 215) and WT strain were measured spectrophotometrically in 96-well culture plates (Corning, New York, NY, United States) by determining the optical density at 600 nm (OD600). Each experiment was independently done in triplicate. Growth of strains were analyzed by the statistical analysis of t-tests using OriginPro 8.5.1 software. A significant difference was defined as a p-value (p < 0.05) between wild-type (WT) and mutants.

### Identification of Mutating Sites

The detailed procedure for identification of mutating sites and analysis of inserting sites was performed as described by Zhang et al. (2018). In brief, the mutating genes were amplified by arbitrary PCR, then the fragments were purified for being sequenced and aligned.

### Comparison of Biofilm Formation Among Mutants and Wild Type

Under oxidative stress (LB with 1.0 mM H2O2), biofilm formation using CVS was determined ranging from 24 to 72 h described previously by Zhang et al. (2018). In addition, the biofilms on the cell slips at 48 h was detected using SEM (Hitachi, Tokyo, Japan) and CLSM (Zeiss, Berlin, Germany) using LIVE/DEAD BacLight bacterial viability Kit according to instructions (Invitrogen, Carlsbad, CA, United States).

### RESULTS AND DISCUSSION

Based on the growth of mutants and WT strain under oxidative stress (1.0 mM H2O2), eight mutants were successfully screened, and the growth of eight mutants under oxidative stress was significantly (p < 0.05) decreased compared with that of WT shown in **Figure 1**. The mutating genes listed in **Table 1** encode thioredoxin 2 (Trx2), glutaredoxin 3 (Grx3), pantothenate kinase (Pank), serine/threonine protein kinase (STPK), pyruvate kinase (PK), phospholipase A (PLA), ferrous iron transport protein A (FeoA), and alanine racemase 2 (Alr2) which contributed to tolerance to oxidative stress in C. malonaticus.

In Escherichia coli, thioredoxin 2 (encoded by trxC) was identified on the basis of sequence similarity (Miranda-Vizuete et al., 1997), but trxC mutants do not show altered sensitivity to H2O<sup>2</sup> (Ritz et al., 2000). In addition, inactivity of thioredoxin 1 (encoded by trxA) and thioredoxin reductase (encoded by trxB) caused more sensitive to H2O<sup>2</sup> in stationary phase of E. coli (Takemoto et al., 1998). Glutaredoxin (Grx) is a thioldisulfide oxidoreductase widely distributed from bacteria to higher eukaryotes (Rouhier et al., 2008). In yeast, mutants lacking Grx are sensitive to oxidative stress (Luikenhuis et al., 1998). The OxyR and SoxR in E. coli, and the S. cerevisiae Yap1p transcriptional regulators were modulated by glutathioneand thioredoxin-dependent reduction systems for the adaptive responses to oxidative stress (Carmel-Harel and Storz, 2000). The inactivity of glutaredoxin 2 and glutaredoxin 3 encoded by grxB and grxC, respectively, were found in E. coli strains lacking glutaredoxin 1 and thioredoxin 1 still showed GSH



oxidoreductase activity (Aslund et al., 1994). The inactivity of glutaredoxin 2 in E. coli cells were more sensitive to hydrogen peroxide and other oxidants, and the interconnection between catalases and thioredoxin/glutaredoxin pathways in the antioxidant response was observed (Vlamis-Gardikas et al., 2002). Regulators including OxyR, SoxRS, and RpoS in E. coli were associated with the tolerance to oxidative stress (Chiang and Schellhorn, 2012). The redox proteins such as Grx A (Grx1)

FIGURE 3 | Biofilm formation of C. malonaticus wild strain (WT) and mutants at 48 h under 1.0 mM H20<sup>2</sup> using SEM.

required for maintaining redox status in bacteria also protect bacteria from oxidative stress (Caldas et al., 2006; Meyer et al., 2009). The pantothenate kinase is required for the biosynthesis of coenzyme A (CoA). In Bacillus anthracis, the type III pantothenate kinase plays important roles in maintenance of cytosolic redox balance and in adaptation to the oxidative stress in B. anthracis (Paige et al., 2008).

Ferrous iron (Fe2+) is one of the essential elements required for growth and virulence of the majority of pathogens (Hayrapetyan et al., 2016). Here, ferrous iron transport contributed to oxidative tolerance in C. malonaticus through the reduction reaction of Fe2<sup>+</sup> to attenuate the injuries from oxidation (H2O2). The ferrous iron transport (feo) operon was first discovered in E. coli K12 in 1987 through studies of a series of ferrous iron transport mutants, and the deletion of feo strains cause the failure to taking up ferrous iron (Hantke, 1987). In addition, in the absence of FeoB, H. pylori was unable to colonize the gastric mucosa of mice (Velayudhan et al., 2000). Naikare et al. (2006) found that FeoB is essential for the uptake of ferrous iron, gut colonization and intracellular survival. On the Contrary, feo deletions in V. cholerae do not seem to affect its colonization in the mouse model (Wyckoff et al., 2006).

Through 2-D method combined with MALDI-TOF-MS and database queries, pyruvate kinase was involved in enhancement of oxidative stress in Pichia caribbica (Zhang et al., 2017). In the mitochondrial, pyruvate kinase M2 isoform (PKM2) regulates oxidative stress-induced apoptosis by stabilizing B-cell lymphoma 2 (Bcl2) (Liang et al., 2017). Brien et al. reported that increased placental phospholipase A2 gene expression was implicated in oxidative stress in preeclampsia (Brien et al., 2017). Expression of serine/threonine protein kinase and peroxisomal catalase in P. caribbica were involved in the enhancement of oxidative stress tolerance and biocontrol efficacy of P. caribbica (Zhang et al., 2017). Serine/Threonine kinases activation was induced by oxidative stress in frontotemporal dementia (Palluzzi et al., 2017). S. mutans expresses a eukaryotic serine/threonine type kinase known as STPK which enhances resistance to oxidative stress (Zhu and Kreth, 2010).

Likewise, our results also found that inactivity of pantothenate kinase (Pank), serine/threonine protein kinase (STPK), pyruvate kinase (PK) caused sensitive to oxidative stress. To date, roles of Phospholipases (PLs) on tolerance to oxidative stress are not reported in other foodborne pathogens except for C. malonaticus.

Based on analysis of biofilms using CVS, the strong biofilm-formatting abilities among eight mutants and WT were observed, and biofilms of eight mutants significantly decreased at 48 h compared with that of wild type (WT) shown in **Figure 2**. Furthermore, the detection of spatial structure of biofilms was confirmed using SEM (**Figure 3**), and the mature biofilms were formed at 48 h among mutants and WT. From **Figure 4**, the viable cells and exopolysaccharides (blue) were more predominant at 48 h. Here, inactivity of eight factors caused weak biofilms compared with that of WT under oxidative stress, and a positive relationship between biofilm formation and oxidative tolerance was observed. Hartmann et al. (2010) demonstrated that cellulose and flagella facilitated biofilm formation in C. sakazakii. Using comparative proteomics analysis, genes including LuxS and TolB were found to contribute to biofilm formation in Cronobacter strains (Ye et al., 2016). In addition, the deoB, adh, and nlpD were involved in biofilm formation in C. sakazakii (Du et al., 2012). In addition, environmental conditions such as temperature and pH also greatly affected biofilm formation in C. sakazakii strains (Jung et al., 2013; Ye et al., 2015). In Haemophilus influenzae, expression abundance of peroxiredoxin–glutaredoxin increased in biofilms compared to planktonic cells (Gallaher et al., 2006). Similarly, thioredoxin, peroxidase, and thioredoxin were upregulated in biofilms in Candida albicans (Seneviratne et al., 2008). The biofilm formation in trxB mutant of Neisseria gonorrhoeae on human cervical epithelial cells was greatly reduced compared with wild-type strain (Potter et al., 2009). In S. typhimurium, the Feo system has been found to play important roles in colonization of the mouse intestine (Tsolis et al., 1996). Jiang et al. (2015) found that hydrolase and pantothenate kinase were detected in the Streptococcus mutans 593 biofilm only, indicating that pantothenate kinase was involved in the biofilm formation in S. mutans 593. The high pyruvate kinase activity in S. mutans contributed to the cariogenic biofilm formation in caries patents (Krzy´sciak et al., 2017). Pyruvate kinase activity in Staphylococcus aureus was regulated by serine/threonine protein kinase, which favors biofilm formation (Vasu et al., 2015). Serine/Threonine kinases (STPKs) have been implicated in biofilm formation of Bacillus subtilis (Madec et al., 2002). Ser/Thr protein kinase PrkC mediates biofilm formation in

### REFERENCES


B. anthracis by regulation of GroEL activity (Arora et al., 2017). Phospholipases (PLs) are considered important factors for C. parapsilosis adherence, tissue penetration, and host invasion (Junior et al., 2011). Meanwhile, the germination, adherence, biofilm formation, phospholipase and proteinase production were considered the virulence factors in Candida albicans (Larkin et al., 2017).

### CONCLUSION

In summary, the factors involved in tolerance to oxidative stress in C. malonaticus were identified including Trx2, Grx3, Pank, STPK, PK, PLA, FeoA, and Alr2. A positive relationship between biofilm-forming ability and oxidative tolerance was also observed, which might indicated that biofilm formation was related with environmental stress. The findings here provide valuable information for deeply understanding molecular mechanism about tolerance to oxidative stress.

### AUTHOR CONTRIBUTIONS

XZ carried out the experiments and analyzed the data. MZ carried out the experiments and analyzed the data. LT analyzed the data and carried out the experiments. DO carried out the partial experiments. YW analyzed the data. JZ modified the manuscript. QW and YY designed and modified the manuscript.

### FUNDING

The financial support of the National Key Research and Development program (2017YFC1601200 and 2017YFC1601202), National Natural Science Foundation of China (31671951), the Anhui provincial Grand Project special of Science and Technology (15czz03109), the Science and Technology Planning Project of Guangdong Province (2016A050502033), and Project of Science and Technology in Guangzhou (201604020036).

### ACKNOWLEDGMENTS

We gratefully acknowledge Prof. Xu, M.Y (Guangdong Institute of Microbiology) for presenting the E. coli WM3064.




**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 Zhang, Zhang, Tong, Ou, Wang, Zhang, Wu and Ye. 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.

# Isolation, Potential Virulence, and Population Diversity of Listeria monocytogenes From Meat and Meat Products in China

Moutong Chen<sup>1</sup>† , Jianheng Cheng<sup>1</sup>† , Jumei Zhang<sup>1</sup> , Yuetao Chen1,2, Haiyan Zeng<sup>1</sup> , Liang Xue<sup>1</sup> , Tao Lei<sup>1</sup> , Rui Pang<sup>1</sup> , Shi Wu<sup>1</sup> , Haoming Wu<sup>1</sup> , Shuhong Zhang<sup>1</sup> , Xianhu Wei<sup>1</sup> , Youxiong Zhang<sup>1</sup> , Yu Ding<sup>3</sup> and Qingping Wu<sup>1</sup> \*

<sup>1</sup> Guangdong Institute of Microbiology, State Key Laboratory of Applied Microbiology Southern China, Guangdong Open Laboratory of Applied Microbiology, Guangdong Provincial Key Laboratory of Microbial Culture Collection and Application, Guangzhou, China, <sup>2</sup> College of Food Science, South China Agricultural University, Guangzhou, China, <sup>3</sup> Department of Food Science and Technology, Jinan University, Guangzhou, China

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

### Reviewed by:

Jianmin Zhang, South China Agricultural University, China Peter Bergholz, North Dakota State University, United States

> \*Correspondence: Qingping Wu wuqp203@163.com †Co-first authors

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 04 December 2018 Accepted: 15 April 2019 Published: 07 May 2019

#### Citation:

Chen M, Cheng J, Zhang J, Chen Y, Zeng H, Xue L, Lei T, Pang R, Wu S, Wu H, Zhang S, Wei X, Zhang Y, Ding Y and Wu Q (2019) Isolation, Potential Virulence, and Population Diversity of Listeria monocytogenes From Meat and Meat Products in China. Front. Microbiol. 10:946. doi: 10.3389/fmicb.2019.00946 Listeria monocytogenes is a globally notorious foodborne pathogen. This study aimed to qualitatively and quantitatively detect L. monocytogenes from meat and meat products in China and to establish their virulence profiles and population diversity. From 1212 meat and meat product samples, 362 (29.9%) were positive for L. monocytogenes. Of these positive samples, 90.6% (328/362) had less than 10 MPN/g, 5.5% (20/364) samples had 10–110 MPN/g, and 3.9% (14/362) of the positive samples had over 110 MPN/g. Serogroup analysis showed that the most prevalent serogroup of L. monocytogenes was I.1 (1/2a-3a), which accounted for 45.0% (123/458) of the total, followed by serogroup I.2 (1/2c-3c) that comprised 26.9%, serogroup II.1 (4b-4d-4e) that comprised 4.8%, and serogroup II.2 (1/2b-3b-7) that comprised 23.3%. A total of 458 isolates were grouped into 35 sequence types (STs) that belonged to 25 clonal complexes (CCs) and one singleton (ST619) by multi-locus sequence typing. The most prevalent ST was ST9 (26.9%), followed by ST8 (17.9%), ST87 (15.3%), ST155 (9.4%), and ST121 (7.6%). Thirty-seven isolates harbored the llsX gene (representing LIPI-3), and they belonged to ST1/CC1, ST3/CC3, ST288/CC288, ST323/CC288, ST330/CC288, ST515/CC1, and ST619, among which ST323/CC288, ST330/CC288, and ST515/CC1 were newly reported to carry LIPI-3. Seventy-five isolates carried ptsA, and they belonged to ST87/CC87, ST88/CC88, and ST619, indicating that consumers may be exposed to potential hypervirulent L. monocytogenes. Antibiotics susceptibility tests revealed that over 90% of the isolates were susceptible to 11 antibiotics; however, 40.0% of the isolates exhibited resistance against ampicillin and 11.8% against tetracycline; further, 45.0 and 4.6% were intermediate resistant and resistant to ciprofloxacin, respectively. The rise of antibiotic resistance in L. monocytogenes suggests that stricter regulations should be formulated to restrict the use of antibiotic agents in human listeriosis treatment and livestock breeding.

Keywords: Listeria monocytogenes, animal-derived food products, LIPI-3, LIPI-4, multi-locus sequence typing, antibiotic resistance

### INTRODUCTION

fmicb-10-00946 May 4, 2019 Time: 16:17 # 2

Listeria monocytogenes is a facultative Gram-positive foodborne pathogen responsible for life-threatening listeriosis diseases, including septicaemia, meningitis, encephalitis, and miscarriage (Drevets and Bronze, 2008). The most susceptible groups are pregnant women, fetuses, elderly people, and immunocompromised individuals, who show a considerably high mortality rate (10–40%) (Hof, 2004). This pathogen can resist stressful conditions in foods and associated environments. It can grow in high salinity (10%), low temperatures (4◦C), low water activity (<0.9), and a wide pH range (4.1–9.6) (Guenther et al., 2009), resulting in a wide range of habitats during different stages of the food processing. Listeriosis cases occur mainly due to the consumption of L. monocytogenes-contaminated foods.

Previous studies report 0.27 and 0.39 reported cases of listeriosis annually per 100,000 individuals in the United States and France, respectively (Goulet et al., 2012; Silk et al., 2012). In China, 147 clinical cases and 82 outbreak-related cases were reported from 28 provinces between 1964 and 2010 (Feng et al., 2013). In recent years, the occurrence of listeriosis diseases have been increasing in China, especially in developed cities (Wang et al., 2015, 2018). A total of 253 invasive listeriosis cases were reported between 2011 and 2016 in 19 provinces, with a fatality rate of 25.7% (Li et al., 2018b). Listeriosis has therefore become a severe public health concern to consumers in China. Although a national human listeriosis pilot surveillance was started in 2013, a risk assessment on the prevalence and characteristics of L. monocytogenes in foods is an urgent necessity. Although several studies have reported the presence of L. monocytogenes in various foods, including ready-to-eat products, mushrooms, aquatic products, meat and meat products, and frozen foods, most such studies are regionally focused (Chen et al., 2014, 2015; Liu et al., 2017; Yang et al., 2017). A comprehensive surveillance of L. monocytogenes in foods throughout China is of crucial importance.

Among 13 serotypes of L. monocytogenes, serotypes 4b, 1/2a, 1/2b, and 1/2c account for over 95% of the isolates recovered from foods and clinical cases (Orsi et al., 2011). The pathogenicity of L. monocytogenes may depend on the presence of virulence genes. Several virulence genes and their encoded proteins have been described in previous studies. Listeria pathogenicity island-1 (LIPI-1), along with inlA and inlB, participate in the L. monocytogenes infection cycle in host cells (Kreft and Vazquez-Boland, 2001). The llsX gene (listeriolysin S, representing LIPI-3), which encodes a haemolytic cytotoxic factor associated with the destruction of gut microbiota during infection, is mainly present in a subset of lineage I (Cotter et al., 2008; Quereda et al., 2017a,b). LIPI-3 has been detected in several sequence types (STs) of L. monocytogenes strains, including ST1, ST3, ST4, ST6, ST77, ST79, ST191, ST213, ST217, ST224, ST288, ST363, ST380, ST382, ST389, ST489, ST554, ST581, ST619, ST778, ST999, ST1000, and ST1001 (Chen et al., 2018b; Kim et al., 2018; Wang et al., 2018). Furthermore, a cellobiose-family phosphotransferase system with a cluster of six genes was recently identified as LIPI-4 (Maury et al., 2016). LIPI-4, which is strongly associated with neural and placental infection, was first identified as a clonal complex (CC) 4-specific virulence factor. Several STs (ST87, ST213, ST217, ST363, ST382, ST388, ST663, ST1002, ST1166, and ST619) harboring the ptsA gene (representing LIPI-4) have been reported in recent studies (Chen et al., 2018a; Kim et al., 2018). The pathogenic potential of L. monocytogenes may differ though the presence of LIPI-1, inlA and inlB gene. However, LIPI-3 (llsX) and LIPI-4 (ptsA) are strongly associated with the L. monocytogenes infection.

The objectives of the present study were to (i) determine the qualitative and quantitative data on L. monocytogenes in meat and meat products; (ii) evaluate the potential virulence and antimicrobial resistance profiles of L. monocytogenes isolates; and (iii) characterize the molecular serotype and genetic diversity of L. monocytogenes isolates recovered from the Chinese retail aquatic system. This data will be invaluable for future risk assessments.

### MATERIALS AND METHODS

### Samples

Between July 2012 and April 2016, 1212 retail raw meat and meat product samples were collected from 43 cities of China, including beef (fresh = 108 samples, frozen = 46 samples), mutton (fresh = 17, frozen = 71), pork (fresh = 154, frozen = 14), minced meat (n = 99), preserved pork (n = 61), chicken (fresh = 103, frozen = 250), duck (fresh = 58, frozen = 2), dumplings (n = 166), steamed bun with meat (n = 29), wonton (n = 21), ham sausage (n = 6), and meat balls (n = 7). All samples were placed in insulated shipping coolers with frozen gel packs placed on the sides, middle, and top of the samples. All samples were kept below 4 ◦C during transportation, and testing was initiated within 4 h after receiving the samples.

### Qualitative and Quantitative Analysis

Qualitative detection of L. monocytogenes was performed based on the National Food Safety Standard of China (4789.30-2010) (Anonymous, 2010), with minor adaptations. Briefly, 25 g of homogenized samples were added to 225 mL Listeria enrichment broth 1 (LB1) (Guangdong Huankai Co., Ltd., Guangzhou, China). The cultures in LB1 media were incubated at 30◦C for 24 h. After incubation, 100◦µL of the LB1 enrichment culture was transferred to 10 mL Listeria enrichment broth 2 (LB2) and incubated at 30◦C for 24 h. A loopful (about 10◦µL) of the LB2 enrichment culture was streaked onto Listeria selective agar plates (Guangdong Huankai Co., Ltd.) and incubated at 37◦C for 48 h. Three to five (when possible) presumptive colonies were selected for the identification of L. monocytogenes using the Microgen ID Listeria identification system (Microgen, Camberley, United Kingdom), according to the manufacturer's instructions.

For quantitative detection, a nine-tube most probable number (MPN) method was used, based on a previous study (Gombas et al., 2003). Briefly, nine tubes were divided into three sets of three tubes each. Homogenized samples (25 g) were added to 225 mL half frasher broth. The first set of tubes contained 10 mL of the sample homogenate in 225 mL half Frasher broth,

while the second and third sets contained 10 mL of half Fraser broth (Guangdong Huankai Co., Ltd.) inoculated with 1 and 0.1 mL of the homogenate, respectively. These different volumes (10, 1, and 0.1 mL) of the sample homogenate represented 1.0, 0.1, and 0.01 g of the original sample, respectively. The nine tubes were incubated at 30 ± 2 ◦C for 24 ± 2 h. The darkened Fraser tubes were streaked onto Listeria Chromagar plates. If a Fraser broth tube did not darken, it was examined again after an additional 26 ± 2 h of incubation. The presumptive pure colonies were streaked onto TSA plates and identified using the Microgen ID Listeria identification system. The MPN value was determined based on the number of positive tube(s) in each of the three sets and the MPN table (United States Department of Agriculture, 1998).

### Serotyping and Virulotype Determination

Multiplex PCR was used for identifying the serotypes of the 458 isolates, as described previously (Doumith et al., 2004) (**Supplementary Table S1**). PCR was performed in a thermal cycler (Biometra, Gottingen, Germany) with the following conditions: an initial denaturation at 94◦C for 3 min; followed by 35 cycles of 94◦C for 35 s, 53◦C for 50 s, and 72◦C for 60 s; and a final cycle of 72◦C for 7 min. Two additional PCRs were performed to detect the llsX and ptsA genes (representing LIPI-3 and LIPI-4, respectively) in the L. monocytogenes isolates (Clayton et al., 2011; Maury et al., 2016). The PCR primers used are shown in **Supplementary Table S1**. The amplicons were separated on 1.5% agarose gels in TAE buffer and visualized by Goldview <sup>R</sup> staining (0.005%, v/v).

### Antimicrobial Susceptibility Test

The antibiotic susceptibility of the L. monocytogenes isolates was determined using the KB method, according to the breakpoints for Staphylococci spp., as recommended by the Clinical Laboratory Standards Institute (Clinical, and Laboratory Standards Institute [CLSI], 2014) for Staphylococcus. The breakpoints of ampicillin and penicillin G for specific Listeria have been defined (M45-A2 Vol. 30 No. 18). The following 17 common antibiotic agents (disk load), including those used to treat human listeriosis, were tested: kanamycin (30◦µg), gentamicin (10◦µg), ciprofloxacin (5◦µg), levofloxacin (5◦µg), ofloxacin (5◦µg), sulfamethoxazole with trimethoprim (23.75/1.25◦µg), streptomycin (10◦µg), rifampin (5◦µg), doxycycline (30◦µg), chloramphenicol (30◦µg), erythromycin (15◦µg), tetracycline (30◦µg), meropenem (10◦µg), vancomycin (30◦µg), linezolid (30◦µg), amoxycillin/clavulanic acid (10◦µg), and sulbactam/ampicillin (10/10◦µg) (Oxoid, Basingstoke, United Kingdom). Briefly, pure cultures were transferred to brain heart infusion (BHI) broth and incubated at 37◦C overnight. A cell suspension was adjusted to 0.5 MacFarland standards by 0.85% NaCl (w/v). The suspension was spread onto the surface of Mueller-Hinton agar (Huankai Co., Ltd., Guangzhou). The diameters of the inhibition zones were measured using precision calipers after 24◦h incubation. Staphylococcus aureus ATCC 25923 and Escherichia coli ATCC 25922 were used as quality control strains. Isolates exhibiting resistance to at least three classes of the tested antimicrobial agents were considered multidrug-resistant (Magiorakos et al., 2012).

### Multi-Locus Sequence Typing

Multi-locus sequence typing (MLST) analysis of L. monocytogenes was performed according to a previously published method (Ragon et al., 2008), which was based on seven house-keeping genes (abcZ, bglA, cat, dapE, dat, ldh, and lhkA) (**Supplementary Table S2**). A detailed protocol of MLST analysis, including primers, PCR conditions, STs and CCs assignments were performed according to the recommendation of Pasteur Institute website. PCR products were sequenced (Thermo Fisher Co., Ltd., Shanghai, China), and an allele number was assigned based on each variant locus of each housekeeping gene; STs and CCs were assigned via the Listeria MLST database at the Pasteur Institute website<sup>1</sup> . A minimum spanning tree (MST) was constructed to analyze the relationships between the isolates using the BioNumerics software Version 7.6 (Applied Maths, Belgium).

## RESULTS

### Occurrence and Contamination Levels of L. monocytogenes

A total of 1212 meat and meat products (12 types) were tested in this study. As shown in **Table 1**, the overall prevalence of L. monocytogenes in meat and meat products was 29.9% (362/1212); it was detected in 45 (51.1%) mutton samples, 143 (40.5%) chicken samples, 31.3% of both minced pork (31/99) and dumpling samples (52/166), 38 (24.7%) beef samples, 6 (28.6%) wonton samples, 7 (24.1%) steamed bun with meat stuffing samples, 7 (11.7%) duck samples, 31 (18.5%) pork samples, and 1

<sup>1</sup>http://bigsdb.pasteur.fr/listeria/listeria.html

TABLE 1 | Positive rate of Listeria monocytogenes in meat and meat products.


(16.7%) ham sausage sample. Only one preserved meat sample showed the presence of L. monocytogenes, while no positive samples were found in meatball samples. In addition, as risk identification required quantitative data to estimate the impact of L. monocytogenes on consumer health, the level of contamination was also determined in the meat and meat product samples. Most samples, 90.6% (328/362) had less than 10 MPN/g, 5.5% (20/364) samples had 10-110 MPN/g, and 3.9% (14/362) of the positive samples had over 110 MPN/g, 8 of which were from chicken samples (**Table 2**).

### Serogroups

Molecular serogrouping was performed by multiplex PCR on 458 L. monocytogenes strains isolated from all 364 positive samples. As shown in **Table 3**, serogroup I.1 (1/2a-3a) was the most prevalent (45.0%). As for the other serogroups, 26.9% (123/458) of the samples were in serogroup I.2 (1/2c-3c), 4.8% (22/458) were in serogroup II.1 (4b-4d-4e), 23.3% (107/458) were in serogroup II.2 (1/2b-3b-7), and none were in serogroup III (4a-4c).

### Antibiotic Susceptibility Test

All L. monocytogenes isolates recovered from meat and meat products were susceptible to vancomycin and amoxicillin/clavulanic acid. Over 90% of the isolates were susceptible to kanamycin, gentamicin, ofloxacin, sulfamethoxazole with trimethoprim, doxycycline, meropenem, linezolid, sulbactam/ampicillin, and penicillin. However, to some extent, L. monocytogenes isolates were resistant to some antibiotics, including ciprofloxacin, levofloxacin, streptomycin, rifampin, tetracycline, and ampicillin. Approximately 40.0% of the isolates exhibited resistance to ampicillin, while 8.1% were resistant to penicillin. Fifty-four (11.8%) isolates were resistant

TABLE 2 | Quantitative results of Listeria monocytogenes contamination in meat and meat products.


to tetracycline, and 11 exhibited intermediate resistance. In addition, 45.0 and 4.6% of the isolates were intermediateresistant or resistant to ciprofloxacin, respectively; further, 20.3% were also intermediate-resistant to levofloxacin (**Table 4**). In total, 27 multidrug-resistant strains were counted.

### MLST Analysis

The 458 L. monocytogenes isolates were grouped into 35 different STs belonging to 25 CCs and one singleton (ST619) by MLST analysis (**Figures 1**, **2**). Thirteen STs (35.1% of all STs) were represented by single isolates. Five CCs were the most prevalent: CC9 (n = 123 isolates; 26.9%) and CC8 (n = 82; 17.9%); followed by CC87 (n = 70; 15.3%), CC155 (n = 43, 9.4%), and CC121 (n = 35; 7.6%) (**Figure 2**). The remaining 20 CCs and one singleton (n = 105, 22.9%) were sporadically distributed (**Figure 2**). In addition, the presence of the llsX (representing LIPI-3) and ptsA genes (representing in LIPI-4) were also determined in the L. monocytogenes isolates. Thirty-seven isolates harbored llsX, and they belonged to ST1/CC1, ST3/CC3, ST288/CC288, ST323/CC288, ST330/CC288 ST515/CC1, and ST619; seventy-five isolates harbored ptsA, and they belonged to ST87/CC87, ST88/CC88, and ST619. Interestingly, isolates belonging to ST619 carried both llsX and ptsA.

## DISCUSSION

Listeriosis is a major public health concern worldwide, with a high morbidity rate. Surveillance for L. monocytogenes in food items is of utmost importance for risk assessment. In this study, the contamination levels of L. monocytogenes in different meat and meat products in China were determined, and the phenotypic and genotypic characteristics of isolates were analyzed by serotype, antibiotic resistance, and genetic diversity. The overall prevalence of L. monocytogenes in meat and meat products in China was 29.9% (362/1212), similar result (26.6%) was observed conducted in Beijing city by Ma (2015) and the results from Addis Ababa, Ethiopia (Derra et al., 2013); while the different contamination rate were reported in Changchun city by Zhu et al. (2016) (43.3%) and in Liaoning province by Wen et al. (2015) (8.88%). However, other countries have been reported to have different levels of L. monocytogenes prevalence in meat and meat products (Ndahi et al., 2014; Ristori et al., 2014). This variation maybe attributed to differences in sample size, sample constitution, or geographical location. Among the 12 types of meat and meat products analyzed, the highest prevalence was found in mutton (51.1%), followed by chicken (40.5%) and beef (24.7%), which was consistent with the contamination reported in retail-level beef meat in Poland (Wieczorek et al., 2012). The occurrence of L. monocytogenes in pork (18.5%) was higher than that reported in previous studies in several countries (2.6–12.8%) (Pesavento et al., 2010; Derra et al., 2013; Li et al., 2016; Hamidiyan et al., 2018). The high prevalence of L. monocytogenes in meat and meat products in some countries but not others suggest meat and meat products contaminated with L. monocytogenes may occur at processing level, which maybe associate with the hygiene conditions of



TABLE 4 | Antibiotic susceptibilities of Listeria monocytogenes isolates from meat and meat products.


retail environments of these products. Hoelzer et al. (2012) reported that transfer probabilities of L. monocytogenes may be from cutting boards, scales, deli cases, deli preparation sinks to product, floor drains, walk-in cooler floors, and knife racks to food contact surfaces. In addition, it should pay attention to the persistence of L. monocytogenes in food production process. Simmons et al. (2014) reported that one or more PFGE types were isolated on at least three separate occasions, suggesting that the persistence of a given L. monocytogenes subtype in the delis. The highest prevalence of persistent predominant genotypes of L. monocytogenes was also observed on the Finish dairy farm with the poorest production hygiene, such as feeding surfaces, water troughs, and floors (Castro et al., 2018). Due to the high prevalence of L. monocytogenes in meat and meat products, some

sanitization measures and regulations should be formulated to reduce the prevalence of contamination at the processing level.

Quantitative data are invaluable for estimating the impact of L. monocytogenes on consumer health. In the present study, the level of contamination in meat and meat products was also assessed using the MPN method. Most positive samples (90.5%) had less than 10 MPN/g, and only 3.9% of the samples had above 110 MPN/g, which were mainly chicken samples. These results were consistent with those of studies in other countries such as Poland (Modzelewska-Kapitula and Maj-Sobotka, 2014), Brazil (Ristori et al., 2014), and Ireland (Khen et al., 2015); similar results were also reported in China for other food items (Chen et al., 2018a,b), suggesting low levels of L. monocytogenes contamination in fresh food products. The low number of L. monocytogenes in most samples could still be problematic; being psychrotrophic, it may grow during the storage period. It was reported that an initial contamination by only 10 CFU/g of L. monocytogenes can make the food unsafe within 8 days (Salvat and Fravalo, 2004). These results demonstrate the need for further processing of meat and meat products after purchase. Additionally, cross-contamination of food items should be carefully avoided during storage to ensure food safety.

Serotyping is a classical method for L. monocytogenes subtyping. Our results showed that the most prevalent serotypes were serogroups I.1 (1/2a-3a), I.2 (1/2c-3c), and II.2 (1/2b-3b-7), which was consistent with the results of previous studies on L. monocytogenes isolated from food items (Korsak et al., 2012; Shen et al., 2013; Martin et al., 2014; Vallim et al., 2015). Interestingly, serogroup II.1 (4b-4d-4e), which we found to be scattered in all meat and meat products analyzed, has been reported to be the most predominant serogroup in ready-to-eat foods in China, including cooked meat (Chen et al., 2014; Wu et al., 2016). Serotypes 4b, 1/2b, and 1/2a have been shown to be predominant in human listeriosis cases (Orsi et al., 2011), suggesting that these isolates may exhibit pathogenicity against consumers. In fact, L. monocytogenes exhibit variable pathogenicity at the species-level, even though each isolate carries inlA, inlB, and virulence genes of LIPI-1.

Our MLST analysis results showed that five CCs were the most prevalent: CC9 (n = 123 isolates; 26.9%) and CC8 (n = 82; 17.9%), followed by CC87 (n = 70; 15.3%), CC155 (n = 43, 9.4%), and CC121 (n = 35; 7.6%). These results were consistent with those of previous studies in China (Wang et al., 2012; Li et al., 2018a), France (Felix et al., 2018), the European Union (Rychli et al., 2018), and Spain (Martin et al., 2014). The most prevalent ST was ST9 (26.9%), followed by ST8 (17.9%), ST87 (15.3%), ST155 (9.4%), and ST121 (7.6%) in this study. Thus, ST9 and ST121 may be dominant in food and food processing environments globally. In addition, ST87 has been found to be prevalent in other kinds of foods in China, including edible mushrooms and aquatic products (Chen et al., 2018a,b). ST2 and ST87 were reported to be persistent in prepacked smoked salmon in Singapore

(Chau et al., 2017). To the best of our knowledge, ST87 strains are rarely reported in western countries, indicating that ST87 isolates may have a geographically associated distribution in Asia. Further surveillance should be performed for the presence of ST87 strains in food, food processing environments, and clinic cases.

llsX (belonging to LIPI-3) encodes a bacteriocin-like haemolytic and cytotoxic virulence factor, which plays a role in the destruction of the gut microbiota. It is critical for the establishment of infection and for the survival of the pathogen in polymorphonuclear neutrophils (Cotter et al., 2008; Quereda et al., 2016, 2017a,b). We found that 37 isolates harbored llsX, and they belonged to serogroups II.1 (4b-4d-4e) and II.2 (1/2b-3b-7). These isolates were present in ST1/CC1, ST3/CC3, ST288/CC288, ST323/CC288, ST330/CC288, ST515/CC1, and ST619, implying that these lineage-I isolates carrying llsX may be responsible for epidemic listeriosis outbreaks. ST323, ST330, and ST515 were newly reported to be llsX-carrying isolates. Furthermore, LIPI-4, a cluster of six genes called the cellobiose-family phosphotransferase system, was mainly involved in neural and placental infection; it was first found only in CC4 strains (Maury et al., 2016). In our results, 75 isolates carried ptsA, and they belonged to ST87/CC87, ST88/CC88, and ST619. Interestingly, hypervirulent CC4 strains carrying both llsX and ptsA are known to be overrepresented in human isolates (Maury et al., 2016). ST619 isolates also carried both llsX and ptsA, suggesting that they may pose a hyper-pathogenic risk to public health. In addition, one of the predominant strains found in this study, ST87, is a known epidemiological hypervirulent ST in China (Li et al., 2018b; Wang et al., 2018). Thus, potential hypervirulent isolates are present in meat and meat products.

It was reported that 162,000 tons of antibiotics was used in China in 2013, of which 84,240 tons was used for livestock breeding and cultivation (Zhang et al., 2015). The extensive use of antibiotics has facilitated the emergence of antibiotic resistance in L. monocytogenes (Wilson et al., 2018). The first antibiotic-resistant L. monocytogenes strain was isolated in 1988. An increasing number of antibiotic-resistant L. monocytogenes strains are being reported worldwide. Ampicillin, amoxicillin with or without gentamicin, and trimethoprim-sulfamethoxazole are the first-line therapeutic antibiotics used for listeriosis treatment (Alonso-Hernando et al., 2012). Approximately 40.0% of the strains isolated in this study were ampicillin-resistant, indicating the necessity for regular surveillances for resistance against one of the oldest antibiotics used in livestock breeding and patient treatment. In addition, 11.8% of the isolates in this study were tetracycline-resistant; this result was consistent with previous studies, in which tetracycline-resistant isolates were frequently recovered from foods (Bertrand et al., 2016; Akrami-Mohajeri et al., 2018; Noll et al., 2018). A total of 1,540 tons of tetracycline was used for humans and livestock in 2013 in China (Zhang et al., 2015); the widespread resistance may be attributed to this excessive use of tetracycline. Fluoroquinolone antibiotics are also extensively used in both human and livestock; in 2013, 25,500 tons of these antibiotics were used in China (Zhang et al., 2015). The results of this study showed that 45.0 and 4.6% of the isolates were intermediate-resistant or resistant to ciprofloxacin, respectively; further, 20.3% of the isolates were intermediate-resistant to levofloxacin (**Table 4**). These results were consistent with the results of previous studies (Chen et al., 2015; Wieczorek and Osek, 2017; Wilson et al., 2018). It has to be noted that intermediate-resistant strains could develop into completely resistant strains under certain circumstances (Ruiz-Bolivar et al., 2011). Several molecular mechanisms for ciprofloxacin-resistance have been documented, including gene mutations (Godreuil et al., 2003), efflux pump (Godreuil et al., 2003; Guerin et al., 2014; Bertrand et al., 2016), and plasmid-mediated resistance (Wang et al., 2003; Jacoby et al., 2006; Jiang et al., 2008). In recent years, the high prevalence of ciprofloxacin, tetracycline, and streptomycin of L. monocytogenes isolated from foods were reported in China (Yan et al., 2010, Yan et al., 2014), suggesting that the abuse of these antibiotics may accelerate the emergence of antibiotic resistance in L. monocytogenes. Although we did detect multidrug-resistant L. monocytogenes strains, the majority of isolates were sensitive to antibiotics commonly used in listeriosis treatment. Over 90% of the isolates were susceptible to 11 antibiotics, including kanamycin, gentamicin, ofloxacin, sulfamethoxazole with trimethoprim, doxycycline, meropenem, linezolid, sulbactam/ampicillin, penicillin, vancomycin, and amoxicillin/claulanic acid, which are commonly used to treat human listeriosis (Olaimat et al., 2018). However, the emerging threat of antibiotic resistance highlights the necessity for continuous surveillance and elucidation of molecular mechanisms behind antibiotics resistance in L. monocytogenes from foods, environment, and clinical cases.

In conclusion, 458 L. monocytogenes strains were isolated from 1212 meat and meat product samples. These strains were characterized based on serogroup, antibiotic susceptibility, and MLST. Five STs (ST8, ST9, ST87, ST155, and ST121) were predominant in meat and meat products. Several isolates carried llsX and/or pstA virulence factors, which play an important role in human listeriosis diseases, posing a potential public health concern for consumers. In addition, the rising trend of antibiotics resistance in L. monocytogenes suggests that strict regulations to restrict the abuse of antibiotics should be formulated urgently.

## AUTHOR CONTRIBUTIONS

QW, JZ, and MC conceived and designed the experiments. MC, JC, and YC performed the experiments. LX, HZ, SW, RP, and HW conducted the bioinformatics analyses. MC, QW, SZ, TL, and XW drafted the manuscript. QW, YZ, and YD reviewed the manuscript. All authors read and approved the final manuscript.

## FUNDING

We would like to acknowledge the financial support from the National Natural Science Foundation of China (31701718 and 31501580), the Natural Science Foundation of Guangdong Province, China (2017A030313173), the Pearl River S&T Nova Program of Guangzhou (201710010018), and the GDAS' Special Project of Science and Technology Development (2017GDASCX-0201).

### ACKNOWLEDGMENTS

fmicb-10-00946 May 4, 2019 Time: 16:17 # 8

The authors would like to thank the team of curators of the Institute Pasteur MLST databases for curating the data and making them publicly available at http://bigsdb.pasteur.fr/.

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.00946/full#supplementary-material




resistance. Environ. Sci. Technol. 49, 6772–6782. doi: 10.1021/acs.est.5b0 0729

Zhu, J., Gong, Y., and Wu, Y. (2016). The investigation of Listeria monocytogenes contamination in foods. Chin. J. Public Health Eng. 201615, 491–493.

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

The reviewer JZ declared a shared affiliation, with no collaboration, with one of the authors, YC, to the handling editor at the time of review.

Copyright © 2019 Chen, Cheng, Zhang, Chen, Zeng, Xue, Lei, Pang, Wu, Wu, Zhang, Wei, Zhang, Ding 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.

# Antibiotic Resistance in Salmonella Typhimurium Isolates Recovered From the Food Chain Through National Antimicrobial Resistance Monitoring System Between 1996 and 2016

Xuchu Wang<sup>1</sup>† , Silpak Biswas<sup>2</sup>† , Narayan Paudyal<sup>2</sup> , Hang Pan<sup>2</sup> , Xiaoliang Li2,3 , Weihuan Fang2,3 and Min Yue2,3 \*

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Carlos Henrique Camargo, Instituto Adolfo Lutz, Brazil Sunil D. Saroj, Symbiosis International University, India

#### \*Correspondence:

Min Yue myue@zju.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 28 January 2019 Accepted: 18 April 2019 Published: 07 May 2019

#### Citation:

Wang X, Biswas S, Paudyal N, Pan H, Li X, Fang W and Yue M (2019) Antibiotic Resistance in Salmonella Typhimurium Isolates Recovered From the Food Chain Through National Antimicrobial Resistance Monitoring System Between 1996 and 2016. Front. Microbiol. 10:985. doi: 10.3389/fmicb.2019.00985 <sup>1</sup> Hangzhou Center for Disease Control and Prevention, Hangzhou, China, <sup>2</sup> CATG Microbiology and Food Safety Laboratory, Institute of Preventive Veterinary Medicine, College of Animal Sciences, Zhejiang University, Hangzhou, China, <sup>3</sup> Zhejiang Provincial Key Laboratory of Preventive Veterinary Medicine, Hangzhou, China

Salmonella is a major foodborne pathogen which causes widespread contamination and infection worldwide. Salmonella Typhimurium is one of the leading serovars responsible for human and animal salmonellosis, globally. The increasing rate of antibiotic resistance in Salmonella Typhimurium poses a significant global concern, and an improved understanding of the distribution of antibiotic resistance patterns in Salmonella Typhimurium is essential for choosing the suitable antibiotic for the treatment of infections. To evaluate the roles of animal and human in antibiotic resistance dissemination, this study aims to categorize 11,447 S. Typhimurium strains obtained across the food-chain, including food animals, retail meats and humans for 21 years in the United States by analyzing minimum inhibitory concentrations (MICs) values for 27 antibiotics. Random Forest Algorithm and Hierarchical Clustering statistics were used to group the strains according to their minimum inhibitory concentration values. Classification and Regression Tree analysis was used to identify the best classifier for human- and animal-populations' isolates. We found the persistent population or multi-drug resistant strains of S. Typhimurium across the four time periods (1996∼2000, 2001∼2005, 2006∼2010, 2011∼2016). Importantly, we also detected that there was more diversity in the MIC patterns among S. Typhimurium strains isolated between 2011 and 2016, which suggests significant emergence of diversified multi-drug resistant strains. The most frequently observed (43%) antibiotic resistance patterns found in S. Typhimurium were tetra-resistant pattern ASSuT (ampicillin, streptomycin, sulfonamides, and tetracycline) and the penta-resistant pattern ACSSuT (ampicillin, chloramphenicol, streptomycin, sulfonamides, and tetracycline). Animals (mainly swine and bovine) are the major source for these two frequently found antibiotic resistance patterns. The occurrence of antibiotic resistant strains from humans and chicken is

**287**

alarming. Strains were mostly susceptible to fluoroquinolones. Together, this study helped in understanding the expansion of dynamics of antibiotic resistance of S. Typhimurium and recommended fluoroquinolones as a possible treatment options against S. Typhimurium infection.

Keywords: Salmonella Typhimurium, foodborne pathogen, antibiotic resistance, minimum inhibitory concentration, population diversity, fluoroquinolones

### INTRODUCTION

fmicb-10-00985 August 14, 2020 Time: 13:41 # 2

The burden of foodborne illnesses is tremendous, affecting 10% of global population with 33 million deaths annually (Havelaar et al., 2015). Numerous factors contribute to diarrheal diseases, and Salmonella enterica causes foodborne illnesses with significant public health impact. Salmonella enterica has a group of ∼2600 closely related bacteria defined by their mosaic combination of surface O and H antigens, so-called serovar. Based on the different pathogenic behaviors there are two groups of Salmonella, these are, typhoid Salmonella and non-typhoidal Salmonella. While typhoidal Salmonella can result in systemic infections with high fatal capabilities, non-typhoidal Salmonella infections are generally self-limiting (Su and Chiu, 2007; Gal-Mor et al., 2014). The emergence of pathogenic Salmonella enterica serovar Typhimurium (S. Typhimurium), armed with multiple antibiotic resistance (AR) in particular, presents a considerable threat to public health and food safety.

Salmonella Typhimurium, with its broader range of host tropism, is one of the top two serovars responsible for causing infections in human and animal worldwide (Hohmann, 2001; Herikstad et al., 2002; Foley and Lynne, 2008; Lan et al., 2009; Majowicz et al., 2010; Hendriksen et al., 2011). Throughout the last few decades, S. Typhimurium DT104 (DT104) evolved and disseminated rapidly across the globe (Helms et al., 2005; Lan et al., 2009). Genomic island (SGI-1) was suggested to be an important factor for the multidrug resistance (MDR) phenotype in DT104 (Paul et al., 2016) and the acquisition of different resistance genes changed the bacterial fitness and virulence, which lead to emergence of highly virulent ST34 or ST313 clones (Okoro et al., 2012; Mather et al., 2018). The bacterial strains resistant to more than two antimicrobial drug classes are defined as multidrug resistant. Importantly, DT104 isolates of animal were suggested to be different from the isolates of human population by comparative population genomic analysis. Additionally, the antibiotic resistance pattern between human and food–animal isolates population has some overlaps.

Salmonella Typhimurium is usually isolated from human, retail meats, and animal origins (Zhang et al., 2003; Kingsley et al., 2009; Izumiya et al., 2011). S. Typhimurium is an excellent model to address antibiotic resistant bacterial persistence and transmittance through the food-chain. The National Antimicrobial Resistance Monitoring System (NARMS) in the United States tracks the information about AR in Salmonella including other foodborne pathogens by comparing different bacterial isolates obtained from human infections, animals, and their retail meats. Such information is essential for understanding the effectiveness of antibiotics for both humans and animals (McDermott et al., 2016).

Severe Salmonella infections require treatment, and AR is the most serious challenge to treat these bacterial infections (CDC, 2013a,b). The misuse of antibiotics is one of the important factors responsible for high resistance to antibiotics in various pathogenic bacteria, including S. Typhimurium (CDC, 2013b). AR in Salmonella is linked with horizontal gene transfer and these genes are found on mobile genetic elements. The expansion of antibiotic resistant Salmonella serovars are efficient in worldwide dissemination (Butaye et al., 2006; Michael et al., 2006; Alcaine et al., 2007; Mather et al., 2013). In vitro antibiotic susceptibility testing is an important basis for monitoring antibiotic susceptibility and resistance trends and for guiding effective anti-infective therapy.

Due to the lack of studies on how the antibiotic resistant S. Typhimurium isolates persist and distribute across the food-chain, the objective of this study was to analyze the minimum inhibitory concentrations (MIC) data of different clinically relevant classes of antibiotics of 11,447 S. Typhimurium strains from humans, food animals and their products in the United States collected between 1996 and 2016, to investigate the AR pattern through time, food-chain, host origins and also to recommend effective therapeutic options against S. Typhimurium infections.

### MATERIALS AND METHODS

### Salmonella Typhimurium Strains

We analyzed available data of S. Typhimurium strains obtained from the US NARMS (CDC and NARMS, 2019) database for Enteric Bacteria obtained from 1996 through 2016. Total 11,447 S. Typhimurium strains used to analyze in this study were from humans (n = 6381), animals (n = 2940), and retail meats (n = 2126). The S. Typhimurium strains from animals were from chickens (n = 1396), bovines (n = 810), porcine (n = 609) and turkeys (n = 125). No samplings from ready-to-eat foods and non-animal origin were reported in the current dataset. All the bacterial isolate has clear information of year of isolation, while only around 11% of isolate has clear location information, such as US state of isolation. The aim of this study is to understand the dynamic (time-scale) feature of S. Typhimurium across food-chain (animal, retail meat, and human).

The "food-chain" or "food-chain transmission" defined in this study is all about three steps from an infectious disease point of view. The recognized chain for transmission is

from animal to animal product, and then to humans, this is how we design the study. We did not study the spread of foodborne pathogens at different stages of industrial food production and processing.

### MIC Determination of Antimicrobial Agents

The minimum inhibitory concentration of the antimicrobial agents tested was recorded for each isolate and compared to breakpoints that were defined by the CLSI when available; otherwise, breakpoint interpretations from the National Antimicrobial Resistance Monitoring System were used as described in the NARMS 2012–2013 annual integrated report (FDA, 2015; Clinical and Laboratory Standards Institute [CLSI], 2017).

The MIC values interpreted according to the CLSI guidelines (Clinical and Laboratory Standards Institute [CLSI], 2017) for 27 antibiotics were evaluated and analyzed in this study. These 27 antibiotics were Amikacin (AMI), Apramycin (APR), Gentamicin (GEN), Kanamycin (KAN), Streptomycin (STR), Amoxicillin-clavulanic acid (AMC), Piperacillin-tazobactam (PTZ), Cephalothin (CEP), Cefoxitin (FOX), Ceftriaxone (AXO), Ceftiofur (TIO), Ceftazidime (CAZ), Cefotaxime (CTX), Cefotaxime/clavulanic acid (CTC), Cefquinome (CEQ), Cefepime (FEP), Sulfamethoxazole (SMX), Sulfisoxazole (FIS), Sulfamethoxazole-trimethoprim (COT), Azithromycin (AZM), Aztreonam (ATM), Imipenem (IMI), Ampicillin (AMP), Chloramphenicol (CHL), Ciprofloxacin (CIP), Nalidixic acid (NAL), Tetracycline (TET).

### Statistical Analysis

In this study, we used Random Forest Algorithm and Hierarchical Clustering analysis to classify Salmonella Typhimurium population based on their MIC patterns. Recent work demonstrated the usefulness of random forest method because of its unique advantages (Pan et al., 2018). Random Forest package 4.6 was used in this study for the classification. The data set for random forest analysis was classified based on three source (animal, meat, and humans) and five hosts (chicken, bovine, swine, turkey, and human). The input variable was the data set of MIC distribution. The test feature was the MIC values of each individual strains. The strains were categorized into four groups according to the time of their isolation as period 1 (Year 1996–2000), period 2 (Year 2001–2005), period 3 (Year 2006–2010) and period 4 (Year 2011–2016). Based on these four time periods, the diversity of S. Typhimurium population was segregated using Random Forest. The MIC values of 27 different antibiotics were used to group different population. Single dot depicts individual strains. The rule for each randomly created decision trees was based on the MIC cut-off values and the vote count was denoted by present (MIC higher than cut off) or absent (MIC lower than cut off). Proximity computation was done on three scales. To identify the best classifier, we have used Classification and Regression Tree (CART) method. Both Random Forest and CART were used to identify the most crucial variables. Using multinomial logistic regression (Afema et al., 2015), here we analyzed antibiotic-susceptibility and antibiotic resistant profiles of S. Typhimurium population.

### RESULTS

### Isolates From Different Hosts Behave Differently

A general trend of the AR of the S. Typhimurium isolates from animals, meat and humans to some commonly used antimicrobial agents is presented in **Figure 1**. In the early 2000s, a surge in resistance to tetracycline by isolates from humans coincided with a similar surge in the animals' isolates. These results apparently convince that food animals may serve as a pool of antibiotic resistant organisms to humans. But in absence of the records of the simultaneous meat samples a direct link between the animals and humans is missing. For the other years (after 2002) when data is available in all three hosts, the resistance pattern has remained in a generally steady state with a higher rate in animals/meat and a lower rate in humans. In this context, our major focus in this analysis was to show that the isolates from different hosts behave differently.

### Recent Diversification of Antibiotic Resistant Strains

Based on the four time periods (1996–2000, 2001–2005, 2006–2010, 2011–2016), the diversity of S. Typhimurium population was segregated using RandomForest (**Figure 2**). The MIC values of 27 different antibiotics were used to group different population. Single dot depicts individual strains, with four different colors pointing isolates from four different periods of time. The **Figure 2A** showed the temporal distribution of the S. Typhimurium strains for four time periods from 1996 till 2016. While the distribution based on the MIC of individual strains for until the year 2010 was clustered, the divergence was seen after 2011. For the period 2011– 2016, two distinct clusters of the strains are typically visible (**Figures 2A,B**). It shows that there could have been a shift in the MIC patterns of the isolates after 2011 so they started being divergent.

### Antibiotic Susceptibility and Resistance Analysis

Antibiotic resistance analysis with high discriminatory capability allowed differentiation of the 11,447 strains of Salmonella Typhimurium. Using hierarchical clustering, population diversity of S. Typhimurium was grouped in this study (**Figure 3**). The log values of the MIC of different antibiotics were used to group Typhimurium population (**Figure 3A**) which showed the diversification within S. Typhimurium strains. The blue to yellow color of the heatmap describing the MICs of individual S. Typhimurium strain for every antibiotic. The yellow color pointing the resistance, and blue color pointing the susceptibility. Gray

color pointing the strains lacking MIC value (**Figure 3B**). We found that the resistance was very low against the fluoroquinolones and the resistance was high against

Chloramphenicol (CHL), Sulfamethoxazole-trimethoprim (COT)].

ampicillin, chloramphenicol, streptomycin, sulfonamides, tetracycline, amoxicillin-clavulanic acid, ceftriaxone, ceftiofur antibiotics (**Figure 3B**).

### ASSuT and ACSSuT Are the Most Frequently Found Antibiotic Resistance Patterns

The AR patterns found in human isolates were compared with the AR patterns of isolates from animals and retail meats. Interestingly, the most frequently observed resistance patterns were tetra-resistant pattern ASSuT (Ampicillin, Streptomycin, Sulfonamides, and Tetracycline) and penta-resistant pattern ACSSuT (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, and Tetracycline). Other AR patterns such as ACSSuTAmc (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, Tetracycline, and Amoxicillin-clavulanic acid) and ACSSuTAmcAxoTio (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, Tetracycline, Amoxicillin-clavulanic acid, Ceftriaxone, and Ceftiofur) were also analyzed in this study. These four patterns comprised of 63% of the total 11,447 isolates. Among these, 39% of isolates showed ASSuT resistance pattern, 29% of isolates showed ACSSuT resistance pattern, 28% of isolates showed ACSSuTAmc and 4% of isolates showed ACSSuTAmcAxoTio resistance pattern. Four AR patterns found in S. Typhimurium isolates in this study are shown in **Figure 4**. **Figure 5** showed the graphical representations of antibiotic resistance patterns found in S. Typhimurium isolates obtained from five different hosts (chicken, bovine, swine, turkey, and human).

We found that, the ASSuT resistance in animal and human isolates of Typhimurium declined sharply during 2002–2008, but the resistance is on rise in the recent years. The ASSuT resistance in meat isolates showed increasing pattern with time (**Figure 4A**). S. Typhimurium isolates from bovine, swine, and turkey showed higher percentage of this resistance pattern than the isolates from chicken and human (**Figure 5A**). In this study, high resistance was recorded in S. Typhimurium strains to ampicillin, chloramphenicol, streptomycin, sulfonamides, and tetracycline (ACSSuT). The ACSSuT resistance in human and meat isolates of Typhimurium showed a decreasing but the same resistance in animal isolates showed an increasing trend with time (**Figure 4B**). Swine and bovine isolates showed more ACSSuT resistance pattern than other host isolates (**Figure 5B**). The ACSSuT pattern also found in different combinations with AMC, AXO, TIO antibiotics as additional resistances in our study. For ACSSuTAmc resistance, the isolates from human and meat showed decreasing pattern, but in animal isolates the resistance remained high with time (**Figure 4C**). The frequency of ACSSuTAmcAxoTio pattern remained constant over the years, though the resistance in isolates from animal showed a slight increase in the recent years (**Figure 4D**). ACSSuTAmc pattern was more prevalent in swine and bovine isolates and ACSSuTAmcAxoTio pattern in the bovine isolates (**Figures 5C,D**). **Figure 4E** showed the percentage of distribution of four different resistance patterns

of 11,447 Salmonella Typhimurium strains with MIC values for each bacterial isolate were used for producing the proximity matrices, where x-, y- and z- axes are the multidimensional scaling coordinates. Bacterial isolates with similar MIC values are represented by points close one to the other, whereas isolates with dissimilar antibiotic resistant MIC values are represented by separated points. Each bacterium was indicated as individual dot, where colors represented four different periods. (A) The dynamics of antibiotic resistance for four separate time periods. Time period 1 from the year 1996 to 2000, time period 2 from the year 2001 to 2005, time period 3 from the year 2006 to 2010 and time period 4 from the year 2011 to 2016. (B) The dynamics of antibiotic resistance for all four time periods describing together.

among 7237 strains of S. Typhimurium obtained from animals, meat and humans.

### Analysis of the Best Classifier for Antibiotic Resistance Divergence

Classification and regression tree analysis of individual strains based on the MIC for multiple antimicrobials reveal that the MIC value toward COT (Trimethoprim-sulfamethoxazole) could be the best classifier to identify the origin of the isolates. The isolates originating in the animals and meat had a COT MIC of >0.12250 whereas those originating in the humans were having COT ≤ 0.12250. This implies that in spite of being susceptible to COT, the effective MIC is higher for the animal or meat isolate as compared to the human isolate. Similarly, if the isolates are to be classified based on their host (as humans, chickens, bovine, porcine, and turkey), the chicken isolates could be best identified as those isolates with MIC for COT > 0.12250 (biological cut off is 2), GEN ≤ 6 (biological cut off is 4), and STR ≤ 48 (biological cut off is 16). That means strains susceptible to COT but resistant to gentamicin and streptomycin are likely to have originated from chickens.

### Sources for Human Infections and Dissemination of Resistance Genes

We found considerable variation in the resistance profile of Salmonella Typhimurium obtained from different sources. The AR profile abundance was higher in S. Typhimurium isolates obtained from humans. We found that S. Typhimurium strains from human and animal were emerged in easily distinguishable populations. Antibiogram of the strains derived from animals showed less diverse than the strains derived from humans. This significantly suggests that multiple sources could be involved in Salmonella infections in human. Our result also described that, epidemics of the human and animal are not similar and the diversity of AR is greater in the human isolates than the diversity of AR in isolates derived from animals and animal meat (**Figure 6**). We also found that S. Typhimurium strains from

FIGURE 3 | The population diversity of 11,447 S. Typhimurium strains from humans, animals, and retail meats. (A) Population diversity of S. Typhimurium strains grouped by hierarchical clustering. A hierarchical tree with 200 bootstrapping, by using the MIC value of 27 antibiotics, was used to group different population. (B) The antibiogram for individual S. Typhimurium strains were shown, with blue indicating the susceptibility, and yellow indicating the resistance, based on the MICs interpreted by the CLSI-2017 standards. Gray color indicates strains without MIC value.

FIGURE 4 | Graphical representations of four antibiotic resistance patterns (ASSuT, ACSSuT, ACSSuTAmc and ACSSuTAmcAxoTio) found in S. Typhimurium strains through 1996–2016 in United States. (A) ASSuT (Ampicillin, Streptomycin, Sulfonamides, and Tetracycline) resistance patterns found in S. Typhimurium isolates from animals, meat and humans. Though the ASSuT resistance in animal and human isolates of Typhimurium declined sharply during 2002–2008, the resistance is on rise in the recent years. The ASSuT resistance in meat isolates showed increasing pattern with time. (B) ACSSuT (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, and Tetracycline) resistance patterns found in S. Typhimurium isolates from animals, meat and humans. The ACSSuT resistance in human and meat isolates of Typhimurium showing decreasing pattern but the same resistance in animal isolates showing increasing trend with the time. (C) ACSSuTAmc (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, Tetracycline, and Amoxicillin-clavulanic acid) resistance patterns found in S. Typhimurium isolates from animals, meat and humans. The ACSSuTAmc resistance in human and meat isolates of Typhimurium showing decreasing pattern but in animal isolates the resistance remained high with time. (D) ACSSuTAmcAxoTio (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, Tetracycline, Amoxicillin-clavulanic acid, Ceftriaxone, and Ceftiofur) resistance patterns found in S. Typhimurium isolates from animals, meat and humans. ACSSuTAmcAxoTio resistance in animal isolates showed increasing tendency in the recent years. The XX' presents the time of data collection while YY' gives the percent of resistance. (E) Pie chart showing the percentage of distribution of four different resistance patterns among 7237 isolates of S. Typhimurium from animals, meat and humans.

isolates showed more percentage of resistance pattern among all host isolates. (D) ACSSuTAmcAxoTio (Ampicillin, Chloramphenicol, Streptomycin, Sulfonamides, Tetracycline, Amoxicillin-clavulanic acid, Ceftriaxone, and Ceftiofur) resistance patterns found in S. Typhimurium isolates from bovine, chicken, swine, turkey and human. Percentage of this resistance pattern found high in bovine Typhimurium isolates. The XX' presents the time of data collection while YY' gives the percent of resistance.

chicken showed higher AR diversity than the bovine, porcine or turkey strains (**Figure 7**).

### DISCUSSION

Continuous monitoring of the emergence of any bacterial serotype to detect in the food-chain is very important for public health point of view globally. The Salmonella Typhimurium strains from the recent years (2011–2016) showed much AR diversification as compared to previous year ranges (**Figure 2**). This is interesting and it could be explained by the fact that, the use as well as misuse of antibiotics is on rise in recent years and as a result antibiotic resistant strains of S. Typhimurium increased sharply and transmission of these strains occurred in the food-chain which ultimately affects humans. The immense use of antibiotics could be the main reason behind the emergence of AR in S. Typhimurium strains.

The reduced susceptibility pattern showed by S. Typhimurium population (**Figure 3**) could be responsible for treatment failures in some clinical situations. Streptomycin is not regularly used for the treatment of salmonellosis; but it is commonly used as a growth promoter in animals. Due to this reason, streptomycin could serve as a marker for resistant isolates moving through the food-chain (McDermott et al., 2016). Among the multiple factors that confer the emergence of antibiotic resistant bacteria, the extensive and overuse of antibiotics in medical system and agriculture is believed as the most pivotal (NARMS, 2019). The mechanism of the AR in Salmonella at the cellular level is complex (Penesyan et al., 2015). Salmonella develops AR mechanisms by the production of enzymes that can destroy action of antibiotics, by activating efflux pumps, and by producing β-lactamase which can degrade the structure of antibiotic molecules (Foley and Lynne, 2008; Andino and Hanning, 2015). AR can also be achieved by biofilm production which can cause increased risk of food safety (Steenackers et al., 2012; Corona and Martinez, 2013).

A very important feature of DT104 is that it is bearing Salmonella genomic island 1 (SGI-1). SGI-1 contains different AR genes identified in several Salmonella enterica serovars

(Mossoro-Kpinde et al., 2015). Mulvey et al. (2006) reported that, variation in SGI-1 of the DT104 strains occurs through recombination. Another study using 359 strains demonstrated frequent and recurrent loss or gain of AR genes from the entire SGI-1 to single anti-microbial resistance genes (Mather et al., 2013). One of the characteristics of DT104 is that the strain typically shows resistance to ampicillin, chloramphenicol, streptomycin, sulfonamide, and tetracycline and it has the capability to gain resistance to other clinically important antibiotics (Helms et al., 2005; Mulvey et al., 2006). During 1990s MDR DT104 rapidly emerged globally and became the most common foodborne pathogen found in humans and animals (Ridley and Threlfall, 1998; Threlfall, 2000; Helms et al., 2005). Integrons, which contain AR genes, acquisition is also a very effective mechanism for DT104 to acquire resistance to different antibiotic classes (Stokes and Hall, 1989; Collis and Hall, 1995; Levesque et al., 1995).

Our analysis reveals that resistance was very low to an important group of antibiotics, the fluoroquinolones (**Figure 3B**). Due to its broad-spectrum activity, ciprofloxacin, which is a fluoroquinolone compound, is used for the treatment of Salmonella infections (Fàbrega and Vila, 2013). This same antibiotic is commonly recommended in sub-Saharan Africa, and prescribed with other important fluoroquinolone compounds such as levofloxacin, ofloxacin, and norfloxacin for patients with enteric diseases (Kagambèga et al., 2018). The antibiotic treatment regimen recommended by clinicians against Salmonella infection in human includes third generation cephalosporins, quinolones and macrolides (Guerrant et al., 2001). However, as a result of the high frequency of AR of animal-origin Salmonella isolates against cephalosporins, the recommended usage of cephalosporins became a great concern for the treatment of Salmonella derived from animals which can infect humans as well as other healthy animals. Another

two antibiotics such as tetracycline and streptomycin showed high level of AR in this study, suggesting that clinicians should consider avoiding these two antibiotics for clinical use as well as in animal husbandry. The AR analysis in this study suggested fluoroquinolones (i.e., Ciprofloxacin) still can be used as a therapeutic agent of choice against Salmonella Typhimurium infection.

Four important AR patterns (ASSuT, ACSSuT, ACSSuTAmc, and ACSSuTAmcAxoTio) were analyzed in S. Typhimurium strains in this study (**Figures 4**, **5**). ACSSuT pattern generally used when bacteria become resistance to ampicillin, chloramphenicol, streptomycin, sulfonamides, and tetracycline. A resembling resistance pattern ASSuT is used to show resistance to four antibiotics. Other important phenotypes (ACSSuTAmc and ACSSuTAmcAxoTio) which we found to be more prevalent along with ACSSuT and ASSuT were also analyzed here. The ASSuT resistance pattern was found related with S. Typhimurium DT193 strains which caused human infection in countries such as Spain, England and Wales (Prats et al., 2000; Threlfall et al., 2000). This same AR pattern was found associated with porcine population in United Kingdom (Maguire et al., 1993; Hampton et al., 1995). The ASSuT became more usual in Italy in S. Typhimurium since 2000 (Graziani et al., 2008; Dionisi et al., 2009). Here, this ASSuT is linked with S. Typhimurium strains of animals as well as strains of human origin. Busani et al. (2004) demonstrated a lower frequency of resistance to ACSSuT pattern in S. Typhimurium isolates obtained from humans than from non-human isolates. The ACSSuT is the typical resistance pattern factors of the DT104 (Adhikari et al., 2010; Mateva et al., 2018). The MDR S. Typhimurium strains in other studies showed similar AR pattern (ACSSuT); (Ribot et al., 2002; De Vito et al., 2015; Mossoro-Kpinde et al., 2015). Some reports described that the AR type ACSSuT may have evolved in Asia around early 1980s in other pathogens and were then horizontally transferred to DT104 (Angulo et al., 2000; Ribot et al., 2002).

Salmonella Typhimurium strains obtained from animal, meat and human were found to originate in distinct clusters. In our study, trimethoprim-sulfamethoxazole was found as the best classifier for AR differentiation of different population, which is also evidenced in other serovar such as S. Newport (Pan et al., 2018).

Based on antimicrobial susceptibility testing, our result suggests that multiple sources could be involved in Salmonella infections in human (**Figure 6**). Certain clusters of isolates act as the bridge between these human and animal isolates. From this perspective, the isolates from meat samples are at the interface of both animal and human (**Figure 6**). A study by Sun et al. (2014) demonstrated that S. Typhimurium infection in humans is linked with different clades than those which are prevalent in pet animals. Contaminated meat, milk and eggs are the probable source of Salmonella infection in humans. Humans are having

different diverse source of infections as compared to the local animals and it is possible that the AR diversity is more in humans than in animals (**Figure 6**). In their study, Mather et al. (2012) used epidemiological as well as ecological methods to analyze phenotypic AR data in DT104. The study found greater AR diversity in isolates derived from humans as compared to isolates obtained from animals. In another study, Wiesner et al. (2009), using MLST and PFGE methods found the existence of highly distributed S. Typhimurium strains that were derived in Mexico from human and food-animals, and from diverse geographic locations during different years.

The occurrence of antibiotic resistant strains in humans and chicken (**Figure 7**) warns of the possible risk of consumption of contaminated food. Through contamination of food, poultry may act as a very effective vector for transmission of Salmonella to human (Bemrah et al., 2003; Vandeplas et al., 2010; Paine et al., 2014; Li et al., 2018). Periodic outbreaks of live poultry-associated salmonellosis (LPAS) pointed the need of new alternative potential strategies to control human illnesses. LPAS outbreaks in the United States have been documented since 1955 (Anderson et al., 1955) and report showed that, the number of LPAS incidents is on rise significantly in last few years. Most of the LPAS outbreaks start in February (Basler et al., 2016). The DT104 strain which was correlated with cattle gradually became prevalent in poultry, pigs, and wild animals and showed a typical penta-resistance pattern of ACSSuT (Glynn et al., 1998). Study by Zhu et al. (2017) suggested the role of broiler chicken as the reservoir for multi-drug resistant Salmonella and in the processing plants cross-contamination is frequent between chicken and the environment including the workers working in the plants. This cross-contamination is of importance and these leads to the enhance of MDR Salmonella strains in the final stages which ultimately facilitate the spread of the AR genes widely to the consumers. Bovine salmonellosis is a significant threat to beef industry (Wray and Davies, 2000). Pigs are also responsible for the transmission of Salmonella to humans (Barilli et al., 2018). As the S. Typhimurium sometimes asymptomatic in pigs, most of the time the pathogen does not cause illness in pigs, but pigs could be an important source of contamination of the dead animals in the slaughter houses (Davies and Wray, 1997). This bacterial contamination could not be detected in the pigs in the farm, but this could finally cause the transmission to human (Kich et al., 2011; Almeida et al., 2016). Very few reports detected the antibiotic resistant Salmonella in turkey meat (Fakhr et al., 2006; Khaitsa et al., 2007) which is supporting our findings.

### CONCLUSION

Our results revealed some significant observations for Salmonella Typhimurium epidemiology describing AR pattern dynamics across the food-chain. The recent diversification of AR found among the S. Typhimurium isolates is of great importance and significantly draw attention about the misuse of the antibiotics in the recent years. Major resistance patterns found were tetra-resistance and penta-resistance. The population diversity analysis calculated based on the MIC value revealed typical patterns of host segregation suggesting the distinction between the isolates from different host and sources but with some cross-overs and common shared population. This study helped in understanding the expansion of dynamics of antibiotic resistance of S. Typhimurium and recommended fluoroquinolone antibiotics as a possible potential therapeutic option against S. Typhimurium infection. The antimicrobial susceptibility, resistance profiles and different AR patterns in Salmonella strains including other potential pathogenic and virulence factors is expanding globally and this phenomenon must be monitored cautiously and continuously for the betterment of the treatment of salmonellosis. Nevertheless, this study has some limitations. The sampling bias is the one, with over sampling in human and less in animals, including disequilibrium in samples of animal origins. Most of the data presenting in this study is based on the quantitative MICs data. Future study based on genetic information, particularly whole genome sequence, can accelerate our understanding of S. Typhimurium population diversity and transmission along the food-chain.

### AUTHOR CONTRIBUTIONS

XW and SB contributed equally to this work, conducted the data analysis, and wrote the manuscript. HP and NP made significant contribution to refine and reorganized the data used in the manuscript and its presentation. XL and WF provided critical comments and aided with the edition of the manuscript. MY conceived the idea, collected the data, and suggested the statistical interpretation of the data analyzed and aided with the writing. All authors have read and approved the manuscript.

### FUNDING

This study was supported by grants from National Program on Key Research Project (2017YFC1600103, 2018YFD0500501, and 2018YFD0701001), Fundamental Research Funds for the Central Universities (2-2050205-18-237), Zhejiang Provincial Natural Science Foundation of China (LR19C180001), Zhejiang University "Hundred Talent Program" (2016111), and the Recruitment Program of Global Youth Experts (13-313).

### ACKNOWLEDGMENTS

All the individuals involved in the United States NARMS are acknowledged for their contribution in data collection and deposition of those, on the World Wide Web.

### REFERENCES

fmicb-10-00985 August 14, 2020 Time: 13:41 # 11


typhimurium strains phage type DT120 in southern Italy. Biomed. Res. Int. 2015:265042. doi: 10.1155/2015/265042



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

Copyright © 2019 Wang, Biswas, Paudyal, Pan, Li, Fang and Yue. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Corrigendum: Antibiotic Resistance in Salmonella Typhimurium Isolates Recovered From the Food Chain Through National Antimicrobial Resistance Monitoring System Between 1996 and 2016

Xuchu Wang1†, Silpak Biswas 2†, Narayan Paudyal <sup>2</sup> , Hang Pan<sup>2</sup> , Xiaoliang Li 2,3 , Weihuan Fang2,3 and Min Yue2,3 \*

#### Approved by:

*Frontiers Editorial Office, Frontiers Media SA, Switzerland*

### \*Correspondence:

*Min Yue myue@zju.edu.cn*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

Received: *24 June 2020* Accepted: *02 July 2020* Published: *14 August 2020*

#### Citation:

*Wang X, Biswas S, Paudyal N, Pan H, Li X, Fang W and Yue M (2020) Corrigendum: Antibiotic Resistance in Salmonella Typhimurium Isolates Recovered From the Food Chain Through National Antimicrobial Resistance Monitoring System Between 1996 and 2016. Front. Microbiol. 11:1738. doi: 10.3389/fmicb.2020.01738* *<sup>1</sup> Hangzhou Center for Disease Control and Prevention, Hangzhou, China, <sup>2</sup> CATG Microbiology and Food Safety Laboratory, Institute of Preventive Veterinary Medicine, College of Animal Sciences, Zhejiang University, Hangzhou, China, <sup>3</sup> Zhejiang Provincial Key Laboratory of Preventive Veterinary Medicine, Hangzhou, China*

Keywords: Salmonella Typhimurium, foodborne pathogen, antibiotic resistance, minimum inhibitory concentration, population diversity, fluoroquinolones

### **A Corrigendum on**

**Antibiotic Resistance in Salmonella Typhimurium Isolates Recovered From the Food Chain Through National Antimicrobial Resistance Monitoring System Between 1996 and 2016** by Wang, X., Biswas, S., Paudyal, N., Pan, H., Li, X., Fang, W., et al. (2019). Front. Microbiol. 10:985. doi: 10.3389/fmicb.2019.00985

In the original article, the Acknowledgments section was not included. This section appears below.

### ACKNOWLEDGMENTS

All the individuals involved in the United States NARMS are acknowledged for their contribution in data collection and deposition of those, on the World Wide Web.

The authors apologize for this error and state that this does not change the scientific conclusions of the article in any way. The original article has been updated.

Copyright © 2020 Wang, Biswas, Paudyal, Pan, Li, Fang and Yue. 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.

# Isolation and Characterization of Clinical Listeria monocytogenes in Beijing, China, 2014–2016

Xiaoai Zhang1,2, Yanlin Niu1,2, Yuzhu Liu1,2, Zheng Lu1,2, Di Wang1,2, Xia Cui1,2 , Qian Chen1,2 \* and Xiaochen Ma1,2 \*

<sup>1</sup> Beijing Center for Disease Prevention and Control, Institute for Nutrition and Food Hygiene, Beijing, China, <sup>2</sup> Research Centre for Preventive Medicine of Beijing, Beijing, China

#### Edited by:

Learn-Han Lee, Monash University, Malaysia

### Reviewed by:

Edward M. Fox, Northumbria University, United Kingdom Beatrix Stessl, University of Veterinary Medicine Vienna, Austria

#### \*Correspondence:

Qian Chen cchenqian@263.net Xiaochen Ma xiaoch-ma@126.com

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 23 September 2018 Accepted: 18 April 2019 Published: 08 May 2019

#### Citation:

Zhang X, Niu Y, Liu Y, Lu Z, Wang D, Cui X, Chen Q and Ma X (2019) Isolation and Characterization of Clinical Listeria monocytogenes in Beijing, China, 2014–2016. Front. Microbiol. 10:981. doi: 10.3389/fmicb.2019.00981 Listeria monocytogenes is an important foodborne pathogen with a significant impact on public health worldwide. A great number of outbreaks caused by L. monocytogenes has been reported, especially in the United States, and European countries. However, listeriosis has not yet been included in notifiable disease in China, and thus information on this infection has been scarce among the Chinese population. In this study, we described a 3-year surveillance of listeriosis in Beijing, China. Fifty-six L. monocytogenes strains isolated from 49 clinical infectious cases (27 pregnancy-associated infections and 22 non-pregnancy-associated infections) were analyzed by serotyping, pulsed field gel electrophoresis (PFGE), multilocus sequence typing (MLST), and antimicrobial susceptibility testing between 2014 and 2016 in Beijing. The predominant serogroups were 1/2a,3a and 1/2b,3b,7 which accounted for 92% of the overall isolates. Four strains were serogroup 4b,4d,4e, isolated from patients with pregnancy-associated infections. Based on PFGE, these isolates were divided into 32 pulsotypes (PTs) and 3 clusters associated with serogroups. Ten PTs were represented by more than one isolate with PT09 containing the most number of isolates. MLST differentiated the isolates into 18 STs, without new ST designated. The three most common STs were ST8 (18.4%), ST5 (16.3%), and ST87 (12.2%), accounting for 46.9% of the isolates. STs prevalent in other parts of the world were also present in China such as ST1, ST2, ST5, ST8, and ST9 which caused maternal fetal infections or outbreaks. However, the STs and serogroup distribution of clinical L. monocytogenes in Beijing, China was different from those in other countries. Strains of ST1 and ST2 were isolated from patients with pregnancy-associated infection, whereas none of ST155 isolates caused pregnancyassociated cases. Surveillance of molecular characterization will provide important information for prevention of listeriosis. This study also enhances our understanding of genetic diversity of clinical L. monocytogenes in China.

Keywords: Listeria monocytogenes, human, MLST, PFGE, antimicrobial susceptibility, China

## INTRODUCTION

fmicb-10-00981 May 7, 2019 Time: 16:47 # 2

Listeria monocytogenes is an important foodborne pathogen with a significant impact on public health worldwide (Schlech, 2000; de Noordhout et al., 2014). It has the potential to cause human diseases ranging from self-limited gastroenteritis and spontaneous abortion in pregnant women to severe invasive infections (sepsis or meningitis) in immuno-compromised patients or older patients (Okutani et al., 2004; Guevara et al., 2009; McCollum et al., 2013; Wang et al., 2013).

Although L. monocytogenes is an uncommon human pathogen, it has a disproportionate share of the food borne disease burden. A previous study reported 1600 human cases of listeriosis annually in the United States, of which 260 resulted in death (Crim et al., 2014). In the European Union, a total of 1763 confirmed human cases of listeriosis (notification rate of 0.44 cases per 100,000 population) were reported with a fatality rate of 15.6% in 2013 (European Food Safety Authority [EFSA], 2015). In China, 253 invasive listeriosis cases were reported in 19 provinces from 2011 to 2016, with a fatality rate of 25.7% (Li et al., 2018). More importantly, the incidence is still increasing worldwide, despite antibiotic treatments (Goulet et al., 2008). A large listeriosis outbreak occurred in Canada in 1981 and provided the first evidence for transmission of listeriosis by foodborne L. monocytogenes (Schlech et al., 1983; Evans et al., 1985). Since then, a series of outbreaks caused by L. monocytogenes have been reported, especially in the United States (Fleming et al., 1985; Linnan et al., 1988; Dalton et al., 1997; Gottlieb et al., 2006; Centers for Disease Control and Prevention, 2008; Chen et al., 2017) and European countries (Bula et al., 1995; Ericsson et al., 1997; Goulet et al., 1998; Aureli et al., 2000; Lyytikainen et al., 2000; Althaus et al., 2017; Amato et al., 2017; Dahl et al., 2017; Kleta et al., 2017). However, listeriosis has not yet been regulated as a notifiable disease in China and therefore information on this infection has been largely scarce among the Chinese population.

Listeria contains multiple species. It has been subtyped using different methods, including serotyping, pulsed-field gel electrophoresis (PFGE), multilocus sequence typing (MLST), multi-virulence-locus sequence typing (MVLST), and restriction fragment length polymorphism (RFLP). L. monocytogenes can be divided into 13 serotypes, of which three serotypes (serotype 1/2a, 1/2b, and 4b) are believed to cause the majority of clinical cases (Gilot et al., 1996). In all of these methods, PFGE provides higher discrimination power than serotyping, making it an important tool in source tracking and outbreak investigation (Graves and Swaminathan, 2001; Centers for Disease Control and Prevention, 2008; Chen et al., 2017; Dahl et al., 2017; Kleta et al., 2017). MLST, based on the analysis of seven housekeeping genes, has been proved as a powerful tool in molecular epidemiological studies and population structure studies of L. monocytogenes (Salcedo et al., 2003; Jadhav et al., 2012). However, only a few studies have approached the molecular characterization of clinical L. monocytogenes in China (Lv et al., 2014; Huang et al., 2015; Wang Y. et al., 2015; Zhang et al., 2016). These studies suggested that listeriosis in China was caused by heterogeneous strains. MLST types (STs) prevalence in other parts of the world were also found in China (Lv et al., 2014; Huang et al., 2015; Wang Y. et al., 2015; Zhang et al., 2016). The objective of the present study was to determine the epidemiological characteristics of listeriosis cases, as well as the characteristics of clinical L. monocytogenes isolates in Beijing, China.

### MATERIALS AND METHODS

### Collection of the Clinical L. monocytogenes Isolates

The isolates analyzed in this study were collected from the Survey Project of Human Listeriosis in Beijing. This study was carried out in accordance with the recommendations of Manual of Foodborne Disease Surveillance, China issued by the National Center for Food safety Risk Assessment. The protocol was approved by the ethics committee of Beijing Center for Disease Prevention and Control (Beijing CDC). In this project, 12, 12, and 21 hospitals were covered in 2014, 2015, and 2016, respectively. All the suspected clinical cases of listeriosis were included in the survey. Samples were collected and used to isolate L. monocytogenes. We defined invasive listeriosis as isolation of L. monocytogenes strains from a normally sterile site or from products of conception (Li et al., 2018). All the L. monocytogenes isolates identified by clinical microbiology laboratories were sent to the lab in Beijing CDC.

A total of 56 isolates were isolated from 49 human patients who had a severe illness with serious suspicion of L. monocytogenes infection between 2014 and 2016. All the isolates were firstly identified using VITEK 2-compact system (bioMérieux, Lyons, France) or matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectrometry (Bruker, Leipzig, Germany); and were further checked by PCR targeting hly fragments specific to L. monocytogenes (Xu et al., 2009).

### Serotyping

All the strains were serotyped using multiplex PCR, which was based on the amplification of the following target genes: prs, lmo0737, lmo1118, ORF2110, and ORF2819 described by Doumith et al. (2004).

### Antimicrobial Susceptibility Testing

Antimicrobial susceptibility testing of the L. monocytogenes isolates was performed using broth dilution method. The minimum inhibitory concentrations (MICs) of following 13 antimicrobials were tested: ampicillin (AMP), oxacillin (OXA), vancomycin (VAN), clindamycin (CLI), tetracycline (TET), daptomycin (DAP), erythromycin (ERY), chloramphenicol (CHL), ciprofloxacin (CIP), trimethoprim-sulfamethoxazole (SXT), gentamicin (GEN), penicillin (PEN), and cefoxitin (FOX) (Xingbai, Shanghai, China). Since resistance criteria for AMP, PEN, and SXT of L. monocytogenes exists in the clinical and laboratory standard institute (CLSI) international guidelines, and resistance criteria for ERY exists in European Committee on Antimicrobial Susceptibility Testing (EUCAST) international

guidelines, the MICs of AMP, PEN, SXT were interpreted using CLSI international guidelines, and the MIC of ERY was interpreted according to EUCAST international guidelines. No resistance criteria exists for the other 11 antimicrobials of L. monocytogenes, so susceptibility to other tested antimicrobials was interpreted by those recommended for Staphylococcus spp. ATCC29213 was used as the reference strain.

### Pulsed-Field Gel Electrophoresis (PFGE)

Pulsed-field gel electrophoresis of the strains was processed in accordance with the PulseNet International protocol<sup>1</sup> . Slices of L. monocytogenes agarose plugs were digested using 50 U of AscI and ApaI (Takara, Dalian, China) per slice for 3 h at 37◦C; electrophoresis was performed using a CHEF-DRIII system (Bio-Rad Laboratories, Hercules, CA, United States); electrophoresis was conducted with a switch time of 4–40 s for 19 h; and images were captured using a Gel Doc 2000 system (Bio-Rad) then converted to TIFF files. Salmonella enterica serovar Braenderup strain H9812 restricted with XbaI was used for molecular weight determinations in all PFGE gels. The TIFF files were analyzed using BioNumerics software (version 7.6 Applied Maths, Kortrijk, Belgium). Clustering was performed using the unweighted pair group method with arithmetic mean (UPGMA).

### Multilocus Sequence Typing (MLST)

Multilocus sequence typing was performed on all the 56 isolates by amplification and sequencing of internal fragments of seven housekeeping genes (abcZ, bglA, cat, dapE, dat, ldh, and lhkA) (Salcedo et al., 2003). Sequencing was performed on an ABI 3770 automatic sequencer. The alleles and sequences types (STs) were determined by comparison with the allelic profiles for L. monocytogenes in MLST database<sup>2</sup> .

BioNumerics software was used to create a cluster tree and conduct minimum spanning trees (MST) based on the allelic profiles. In MST, a clonal complex (CC) was formed by STs with six of seven MLST alleles in common and at least two STs; the founder ST was defined as the ST with the highest number of single-locus variants (SLVs); single genotypes that did not correspond to any clone groups were defined as singletons. Types were represented by circles; size of a circle indicated the number

<sup>2</sup>http://bigsdb.pasteur.fr/listeria/listeria.html

of strains of this particular type. Lineages of isolates were assigned as Wiedmann et al. (1997) described.

### RESULTS

### Origin of the Strains

The origins of the strains were summarized in **Table 1**. Fiftysix strains were isolated from the 49 cases. Among them, 27 were pregnancy-associated infections, in which all mothers were cured; ten neonates were survived, whereas thirteen fetuses died in the womb or after birth. No data were available for the four fetuses. Among the other 22 non-pregnancy-associated patients, 11 (50%) were males. The median age of patients with nonpregnancy-associated infections was 42 years, five patients were >60 years, and five patients were ≤3 years.

### Characterization of the Isolates From the Same Patient

There were five groups of isolates from five patients (**Figure 1**). The isolates from the same patient had the same serogroup, antimicrobial susceptibilities, PFGE type, and ST. Therefore, only one isolate from each patient was used for the subsequent analysis.

### Lineages and Serogroups

Lineage was determined based on the MLST data. We found that 25 strains belonged to lineage I and 24 strains to lineage II. Almost half of the strains belonged to serogroup 1/2a,3a (n = 24,49%) followed by serogroup 1/2b,3b,7 (n = 21,43%), serogroup 4b,4d,4e (n = 4, 8%). Strains of serogroup 1/2b,3b,7 and serogroup 4b,4d,4e belonged to lineage I, whereas strains of serogroup 1/2a,3a belonged to lineage II. All 4 strains of serogroup 4b,4d,4e were isolated from patients with pregnancy-associated infections. More pregnancy-associated cases were caused by lineage I than lineage II strains (72 vs. 37.5%, respectively).

### PFGE

Pulsed-field gel electrophoresis analysis of the comprised AscI and ApaI divided 49 isolates into 32 pulsotypes (PT01-P32) (**Figure 1**). PT09 accounted for 16.3% (8/49) of isolates, followed by PT23 (3 isolates, 6.12%). Eight PTs (25%) were represented


#Pregnancy-associated infections comprise listeriosis in neonates in the first month of life and maternal-fetal infections. F, female; M, male; Y, yes; N, not; S, survival; D, death; UN, unknown.

<sup>1</sup>http://www.cdc.gov/pulsenet/pathogens/index.html

by two isolates. Twenty-two PTs (44.9%) were represented by only one single isolate. A UPGMA dendrogram was constructed for the 32 PTs based on presence or absence of bands. PTs were divided into 3 clusters, respectively (cluster A, B, and C) (**Figure 1**), corresponding to the three serogroups identified.

### MLST

Forty-nine isolates were divided into 18 STs, with no new ST designated. Lineage I included 8 STs of serogroups "1/2b,3b,7" and "4b,4d,4e"; lineage II included 10 STs of serogroups "1/2a,3a" (**Figure 2**). The most common STs were ST8 (9 isolates, 18.4%, "1/2a,3a"), ST5 (8 isolates, 16.3%, "1/2b,3b,7"), ST87 (6 isolates, 12.2%,"1/2b,3b,7"), followed by ST121 (4 isolates, 8.2%, "1/2a,3a"), ST155 (4 isolates, 8.2%, "1/2a,3a"), and ST619 (4 isolates, 8.2%, "1/2b,3b,7"). Other 12 STs contained one to two isolates, respectively (**Figure 2**). Among the 18 STs, 17 of them could be assigned clonal complexes and one (ST619) was a singleton based on querying the MLST database (**Figure 2**). Strains of ST1 and ST2 were isolated from patients with pregnancy-associated infection, whereas none of the ST155 isolates caused pregnancy-associated cases. Five of the nine ST8 strains were isolated from patients with pregnancy-associated infection among which three fetuses died. Six of the eight ST5 strains were isolated from patients with pregnancy-associated infection. Fetuses of pregnancy-associated cases caused by ST5 isolates died while those caused by ST87 isolates survived (except 1 lost to follow-up).

### Comparison of Isolates Obtained From This Study to Other Cities of China

The 49 isolates in this study were compared with 176 other isolates from cases of clinical illness in China reported in previous studies (Lv et al., 2014; Huang et al., 2015; Wang Y. et al., 2015; Zhang et al., 2016). These samples were isolated from Beijing (n = 14), Jiangsu (n = 3), Shandong (n = 4), Shanghai (n = 20), Shanxi (n = 3), and Taiwan (n = 132). Together, these 225 isolates were divided into 28 STs. Among them, 13 ST were found in different regions. There were no obvious regional characteristics of STs (**Figure 3**).

### Comparison of Isolates With Other Countries

The 49 isolates were compared with clinical L. monocytogenes isolates from rest of the world (**Figure 4**). A total of 1094 human L. monocytogenes, screened out from L. monocytogenes MLST database see text footnote 2 in October 08, 2017, were included in the analysis. The 1143 human isolates were divided into 38 CCs and 87 singletons. The most globally prevalent CC in the database were CC1 (224), CC2 (133), CC3 (118), CC9 (65), CC4 (49), CC7 (46), CC8 (45), CC155 (41), and CC101 (32). All isolates obtained in this study, excepting ST619, were co-clustered with foreign isolates. The 15 CCs detected in this study were also found in at least two other countries (**Supplementary Table S1**).

### Antimicrobial Susceptibilities of the Isolates

Antimicrobial susceptibility testing was conducted for the 49 L.monocytogenes isolates. Details were listed in **Table 2**. The highest resistance rate was observed for FOX, which reached 100%, followed by DAP (93.9%), OXA (85.7%), and CIP (36.7%). In contrast, three other antimicrobials (TET, 6.1%; PEN, 4.1%; GEN, 2.0%) had resistance rates lower than 10.0%. All the isolates were susceptible to AMP, VAN, CLI, ERY, CHL, and SXT. Among all the 49 isolates, 79.6% (39/49) were co-resistant at least to

OXA, DAP, and FOX; 28.6% (14/49) were co-resistant at least to OXA, DAP, CIP, and FOX. Seven isolates (ST5, ST8, ST7, ST155, and ST705) were sensitive to OXA, three isolates (ST5, ST378) were sensitive to DAP, while seventeen isolates (ST1, ST5, ST310, ST619, ST121, ST91, ST8, ST7, and ST155) were resistant to CIP, three isolates (ST155, ST705) were resistant to TET, two isolates (ST101, ST8) were resistant to PEN, one isolates (ST8) was resistant to GEN (**Table 2**).

### DISCUSSION

In China, a few studies have reported the prevalence of L. monocytogenes in food (Chen et al., 2009; Wang et al., 2012; Wu et al., 2016). However, the descriptions of clinical L. monocytogenes are very limited. In this study, we described the characteristics of molecular subtyping and antimicrobial susceptibilities of clinical L. monocytogenes strains in Beijing, the capital city of China. The clinical strains analyzed in this study were isolated from a systematic investigation, which provided a unique opportunity to investigate the characterization of this important pathogen in China.

Our findings, for the first time, revealed the predominant serogroups and STs of clinical L. monocytogenes in Beijing, China. The results showed that there were some differences in clinical L. monocytogenes serogroups distribution between Beijing and other countries. Among the 49 isolates in this study, the predominant serogroups were 1/2a,3a (49%) and 1/2b,3b (45%). The serogroup compositions were in agreement with that of previous study, showing 28 L. monocytogenes isolates collected from patients from four cities/provinces in China (Wang Y. et al., 2015). The predominant serogroups were also 1/2a,3a and 1/2b,3b in strains isolated from ready-to-eat foods, raw foods and edible mushrooms in China (Chen et al., 2015, 2018a; Wu et al., 2015, 2016). While, the predominant serogroups of clinical L. monocytogenes strains in our study were different from those from the other countries such as United States, Australia, Brazil, Italy and Portugal (Mammina et al., 2009; Bueno et al., 2010; Centers for Disease Control and Prevention, 2014; Magalhaes et al., 2014; Almeida et al., 2017; Jennison et al., 2017). In United States, serotype 4b (57%) was the most commonly identified serotype of L monocytogenes, followed by serotype 1/2a (26%) and serotype 1/2b (13%) (Centers for Disease Control and Prevention, 2014). In Australia, serogroup 4b,4d,4e was the predominant group, accounting for 56.6% of clinical isolates (Jennison et al., 2017). In Italy, 1/2a and 4b were the predominant serotypes, representing 46.3 and 42.6% of human isolates from sporadic cases (Mammina et al., 2009).

Previous molecular epidemiological studies have detected unreported human outbreaks of listeriosis, for example, Ariza-Miguel et al. (2015) identified an epidemiological connection among strains via analysis of the genomic relationships among isolates through PFGE and MLST subtyping. In our study, ten PFGE clusters possessed strains that were isolated from 2 or more different cases. The isolates with same PFGE patterns had no obvious link as the isolates were obtained from different time and places, with no evidence of epidemiological association. It should be noted that isolates from the case38 and case43

patients, which were identified in the same hospital for 6 days apart, shared an indistinguishable PT, suggesting a possible related source. On the other hand, five groups of isolates from same patients in this study had the same serogroups, molecular subtypes and antimicrobial susceptibility profiles, indicating dissemination of L. monocytogenes in body. In case21, we isolated a L. monocytogenes strain from a beef sandwich in the market, where the pregnant patient always bought foods. This isolate had the same PFGE pattern with the strain LM28, indicating that beef sandwich may be the source of infection (Wang and Chen, 2016). In case13, the PFGE pattern of strains LM15, and LM16 were consistent with a strain isolated from the patient's home environment swab, indicating possibility of cross contamination (unpublished data).

Most of STs identified in this study were largely distributed across world countries and regions, showing that "everything

size of the circle is proportional to the number of the isolates, and the sources of the isolates were colored as shown in figure. The shadow zones in different color represent different clonal complexes.

is everywhere" paradigm of L. monocytogenes clones (Chenal-Francisque et al., 2011). Many STs identified in this study have been associated with outbreaks in other countries, such as ST1 caused outbreaks in France in 1989 and in Sweden in 1995, ST2 caused an outbreak in Italy in 1997, and ST5 strains caused outbreaks in Canada in 1996 as well as in United States in 2011 (Ericsson et al., 1997; Aureli et al., 2000; Lomonaco et al., 2013). In addition, CC3, CC4, CC8, CC88, CC87, CC398, and CC403 were associated with outbreaks in MLST database.

In our study, CC8/ST8 clone was the most common ST, which was distributed globally For instance, in Switzerland, CC8 was the most prevalent clone during 2011–2013

#### TABLE 2 | The resistance rates of 49 L. monocytogenes isolates.


(Althaus et al., 2014); and in Denmark, a CC8/ST8 clone was associated with most of sporadic cases during 2004– 2012 (Jensen et al., 2016). Both CC8 clones in Denmark and Canada did not lead to pregnancy-associated infections but were mainly associated with elderly patients (Knabel et al., 2012; Jensen et al., 2016), which was different from our study. In our study, CC8 caused five pregnancy-associated infections. Some studies suggested that the CC8 strains from

Canada possessed a strong capacity of biofilm formation, which might support persistence within food production environments, resulting in subsequent contamination of foods (Verghese et al., 2011). In this study, eight CC8/ST8 strains had the indistinguishable pulsotypes; the fact that identical pulsotype had been found in the same period raised the possibility that contamination of food could originate from a common source.

CC5/ST5 clone was the second common ST in our study. Similarly, ST5 was the most predominate ST in ready-to-eat meat product in Nanjing, China (Wang G.Y. et al., 2015). In France, no obvious difference was observed between frequency distribution of CC5 in food samples and clinical samples (Maury et al., 2016). Although CC5 was not considered as a hypervirulent clone in the study of Maury et al. (2016) CC5 caused several outbreaks in the recent years, such as a multistate cantaloupe outbreak in US in 2011 (Lomonaco et al., 2013), a multistate ice cream outbreak in US in 2010– 2015 (Centers for Disease Control and Prevention, 2015), and a stone fruit recall in US in 2014 (Jackson et al., 2015). In our study, all fetuses of pregnancy-associated cases caused by ST5 isolates died. Further studies are needed to uncover the virulence of CC5.

Interestingly, ST87 was seldomly linked to human infections in other countries. In 2008, one ST87 isolate from water was reported by Ragon et al. (2008). In 2014, Perez-Trallero et al. (2014) reported two outbreak episodes caused by ST87 strains affecting 15 people in northern Spain. In 2012, ST87, a common MLST type, had been reported from foodborne strains in China (Wang et al., 2012). In 2015, ST87 was reported to be the most frequent ST from clinical strains in Taiwan from 2000 to 2013 (Huang et al., 2015). The ST87 were also found as the most common ST in isolates from rodents in nature environment in China (Wang et al., 2017). Besides, we searched CC87 in MLST database and found that there were a CC87 strain isolated from human in Japan in 1988 see text footnote 2. Our study showed that 12.2% of the human isolates were ST87, demonstrating that ST87 has been already circulating in Beijing. It is necessary to follow the dissemination of this clone to assess its potential emergence. Comparing with ST5, fetuses of pregnancy-associated cases caused by ST87 isolates survived. More studies will be required to further assess the virulent diversity of L. monocytogenes in different STs.

It is widely accepted that food is the source of human L. monocytogenes. The comparison of the population structure of the clinical strains in our study with that from foods, revealed that all STs, except ST621, were also reported in foodborne isolates in China (Wang et al., 2012; Wang Y. et al., 2015; unpublished data). However, there was a notable difference in prevalence of STs between human isolates and foodborne isolates. The most common types of isolates from food sources were ST9, ST8 and ST87 (Wang et al., 2012). ST87 and ST8 were predominant in fresh aquatic products and edible mushroom products in China (Chen et al., 2018a,b). Consistently, ST8 and ST87 were the most predominant STs in human isolates in our study. ST9 (72%) was the most common ST in strains isolated from 356 raw pork samples, 2104 raw pork retail environment swabs and 329 insects in China (Wang et al., 2017). MLST analysis of isolates from France showed a higher prevalence of ST9 and ST121 isolates among food sources when compared with those clinical original (Maury et al., 2016).

Some studies reported that reduced susceptibility of L. monocytogenes to many antibiotics in various countries and increases in the frequency of multidrug-resistant strains (Pesavento et al., 2010; Korsak et al., 2012). In our study, the prevalence of some common antibiotic resistance among L. monocytogenes isolated from patients in China was relatively low. PEN alone or with gentamycin is the prioritized antibiotic for treating human listeriosis. Sulfamethoxazole and trimethoprim is used in patients having PEN anaphylaxis. ERY is used in treating pregnant women. Meanwhile, VAN is used in treating L. monocytogenes bacteremia and endocarditis (Janakiraman, 2008). In this study, no resistance to SXT, ERY, VAN, AMP, CLI, and CHL was found, and there was a relatively low resistance rate to PEN (4.1%) and GEN (2.0%). Strains isolated from a Spanish hospital were found to be sensitive to AMP and ERY (Ariza-Miguel et al., 2015), which was consistent with our study. However, most L. monocytogenes isolated from food and food processing environments in China and other countries were resistant to PEN (Obaidat et al., 2015; Lotfollahi et al., 2017; Chen et al., 2018b), which may result from different resistance criteria use. L. monocytogenes is intrinsically resistant to FOX which had a resistance rate of 100% in our study. High resistance to OXA, as reported in previous studies (Khen et al., 2015; Wang et al., 2017), may be intrinsic. In the study, no correlation was found between antimicrobial susceptibility patterns and other characteristics, such as serogroups, PFGE, and MLST types. Further studies are required to survey the antibiotic susceptibility of clinical L. monocytogenes and to explore the potential molecular mechanisms of antibiotic resistance in L. monocytogenes.

## CONCLUSION

In summary, this study observes that more than half of listeria cases were pregnancy-associated infections in Beijing, and as such more attention should be paid to pregnant patients in future studies on this infection. The serogroup and ST distribution of clinical L. monocytogenes in Beijing was different from many other countries. This study enhances our understanding of genetic diversity of clinical L. monocytogenes in China. Continuous surveillance for this pathogen in clinical patients is necessary in China.

### AUTHOR CONTRIBUTIONS

YN and XM were involved in the collection of isolates and collected the clinical data. YL, ZL, DW, and XC performed

the molecular subtyping and antibiotic susceptibility tests. XZ performed the data analysis. XZ and QC designed the study, drafted, and revised this manuscript.

### FUNDING

This work was supported by The National Key Research and Development Program of China (Grant No. 2017YFC1601500), Young Talent Project of Beijing Excellent Talents Funding (Grant No. 2015000021469G186),

### REFERENCES


the Project from the Ministry of Health of the People's Republic of China (Grant No. 201302005), and Capitals Funds for Health Improvement and Research (Grant No. CFH2011-1013-02).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.00981/full#supplementary-material

isolates from edible mushrooms in chinese markets. Front. Microbiol. 9:1711. doi: 10.3389/fmicb.2018.01711



infections from butter in Finland. J. Infect. Dis. 181, 1838–1841. doi: 10.1086/ 315453



Zhang, J., Cao, G., Xu, X., Allard, M., Li, P., Brown, E., et al. (2016). Evolution and diversity of Listeria monocytogenes from clinical and food samples in shanghai China. Front. Microbiol. 7:1138. doi: 10.3389/fmicb.2016.01138

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

# A Novel Mathematical Model for Studying Antimicrobial Interactions Against Campylobacter jejuni

Mohammed J. Hakeem1,2, Khalid A. Asseri3,4, Luyao Ma<sup>1</sup> , Keng C. Chou<sup>5</sup> , Michael E. Konkel<sup>6</sup> and Xiaonan Lu<sup>1</sup> \*

<sup>1</sup> Food, Nutrition, and Health Program, Faculty of Land and Food Systems, The University of British Columbia, Vancouver, BC, Canada, <sup>2</sup> Department of Food Science and Nutrition, College of Food and Agriculture Sciences, King Saud University, Riyadh, Saudi Arabia, <sup>3</sup> Department of Anesthesiology, Pharmacology and Therapeutics, The University of British Columbia, Vancouver, BC, Canada, <sup>4</sup> Department of Pharmacology, College of Pharmacy, King Khalid University, Abha, Saudi Arabia, <sup>5</sup> Department of Chemistry, The University of British Columbia, Vancouver, BC, Canada, <sup>6</sup> School of Molecular Biosciences, College of Veterinary Medicine, Washington State University, Pullman, WA, United States

The aim of this study is to investigate the antimicrobial synergistic effect against Campylobacter jejuni, a leading foodborne pathogen that causes human gastroenteritis, by cinnamon oil, encapsulated curcumin, and zinc oxide nanoparticles (ZnO NPs). We compared three approaches to study the antimicrobial interactions, including the time-killing method, the fractional inhibitory concentration index (FICI) method, and a mathematical concentration-effect model. Isobologram analysis was performed to evaluate the synergy in different combinations, and a median-effect equation was applied to identify the combinations of synergistic effects at median, bacteriostatic, and bactericidal reduction levels. The time-killing method overestimated the synergistic interaction between antimicrobials, while the FICI method failed to detect an existing synergistic phenomenon. This lack of accuracy and sensitivity was mainly due to combining antimicrobials without a deep understanding of their concentration-effect relationships. Our results showed that each antimicrobial had a unique concentrationeffect curve. Specifically, encapsulated curcumin showed a sharp sigmoidal curve unlike cinnamon oil and ZnO NPs. A mathematical model was applied to study the interaction between antimicrobials with a different shape of concentration-effect curve. We observed an additive effect of cinnamon oil/ZnO NPs and synergistic interactions of other binary combinations (cinnamon oil/encapsulated curcumin and ZnO NPs/encapsulated curcumin). The tertiary combination of cinnamon oil/ZnO NPs/encapsulated curcumin at IC<sup>25</sup> (additive line <1-log CFU/mL) presented the greatest synergistic effect by reducing the bacterial population over 8-log CFU/mL. This mathematical model provided an alternative strategy to develop a new antimicrobial strategy.

Keywords: food safety, antimicrobials, encapsulation, FICI, isobologram, median-effect equation

### INTRODUCTION

Campylobacter is one of the leading bacterial causes of human infectious diseases worldwide. In Canada, this microorganism causes ∼145,350 cases of foodborne illness per year (Thomas et al., 2013). Campylobacter jejuni is the most common species that accounts for ∼80% of campylobacteriosis with a relatively low infectious dose (∼500–800 cells) (Nachamkin et al., 2008).

#### Edited by:

Om V. Singh, Technology Sciences Group Inc., United States

#### Reviewed by:

Koshy Philip, University of Malaya, Malaysia Nicolae Corcionivoschi, Agri-Food and Biosciences Institute (AFBI), United Kingdom

> \*Correspondence: Xiaonan Lu xiaonan.lu@ubc.ca

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 24 July 2018 Accepted: 25 April 2019 Published: 14 May 2019

#### Citation:

Hakeem MJ, Asseri KA, Ma L, Chou KC, Konkel ME and Lu X (2019) A Novel Mathematical Model for Studying Antimicrobial Interactions Against Campylobacter jejuni. Front. Microbiol. 10:1038. doi: 10.3389/fmicb.2019.01038

C. jejuni infections usually lead to non-fatal and self-limiting gastroenteritis, including watery diarrhea, nausea, and vomiting; however, severe autoimmune neurological disorders such as Guillain–Barré syndrome may occur in immunocompromised individuals (Kaakoush et al., 2015). Transmission of C. jejuni is commonly through the consumption of meat product (e.g., poultry and beef), raw milk, and/or contaminated drinking water. For all of these reasons, there is an urgent need to develop a new strategy of antimicrobial usage to reduce the prevalence of Campylobacter in the environment and agri-food products.

A combination of antimicrobials may target different bacterial sites and lead to synergistic interaction. This may form a novel and effective strategy of antimicrobial usage to reduce the prevalence of Campylobacter infections. Synergy is defined as an effect produced by two or more agents greater than the sum of their individual effects combined (i.e., additive effect) (Chou, 2006). Synergy requires a lower concentration of each agent to either increase or maintain the antimicrobial effect. Three methods have been used to study the synergism between antimicrobials, including disk diffusion, time-killing, and fractional inhibitory concentration index (FICI) methods (Odds, 2003; Zhou et al., 2016). Currently, there is no standard method for studying antimicrobial interactions. Over 60% of the dual antimicrobial studies used the FICI method and 36% applied the time-killing method during the past decade (Odds, 2003). Each method has advantages and disadvantages, and generates different outcomes that may not be comparable with each other (Rand et al., 1993; Odds, 2003; Chou, 2006; Noll et al., 2012; Zhou et al., 2016). For instance, the time-killing method investigates bactericidal effect over time, while the FICI method studies bacteriostatic effect after one time point (e.g., 24 h). Both methods are established on a linear concentrationeffect curve of antimicrobials, which can result in either over or under-estimation of interaction(s). According to a recent report, only 40 out of 86 studies published between 1999 and 2015 used rigid mathematical methods to accurately study the synergetic effects of Chinese herbal medicine (Zhou et al., 2016). Although advanced pharmacological methods are widely used to study drug combination effect in the pharmaceutical and biomedical sciences, few of these studies are related to antimicrobials research (Odds, 2003; Mora-Navarro et al., 2015; Zhou et al., 2016).

The isobologram is commonly conducted in drug combination studies to identify and evaluate drug interactions. The combined drugs are assumed to be equally effective, but the dose-response of each combined drug is not always similar (Chou, 2006). In-depth studies have indicated that even if two drugs have the same effect at the reference concentration [e.g., minimum inhibitory concentration (MIC)], this equivalence may not occur at their sub-concentrations (Tallarida, 2001; Chou, 2006). Quantitative assessment is therefore essential to identify dose-response of individual drugs and distinguish these situations when the shapes of dose-response curves are not similar. Methods used for the study of drug interactions can be valid only if both drugs have hyperbolic dose-response curves. Drawing an accurate additive line based on every pair of concentrations can overcome this limitation regardless the assay used, shape of dose-response curve, and the level of interaction (Chou, 2006). Even if an interaction exists, not every pair of concentration results in the same level of interaction. Additional analysis involving the use of a median-effect equation can provide a more comprehensive understanding of interactions between antimicrobials.

The objective of the current study was to compare the time-killing method and FICI method with a mathematical model and provide an easy and accurate approach to study the effect of dual antimicrobials. We used a non-linear mathematical concentration-effect model to evaluate the synergistic interactions of three representative antimicrobials (i.e., cinnamon oil, encapsulated curcumin and ZnO NPs) against C. jejuni as a foodborne pathogen model. The advantage of using the mathematical model is not only to identify or evaluate the synergy but also to avoid some of the possible mathematical errors and increase the sensitivity to detect an existing synergistic interaction. To the best of our knowledge, this was the first study to apply a mathematical model that could accurately evaluate the antimicrobial interactions at median (50% reduction), bacteriostatic, and bactericidal levels. This approach can be generalized to quantitatively evaluate any type of dual antimicrobial treatment against microorganisms for different applications.

### MATERIALS AND METHODS

### Chemicals and Reagents

Hydrophobically modified starch (HMS) HI-CAPTM 100 was donated from Ingredion Canada Inc., (Mississauga, ON, Canada). Cinnamon oil and curcumin purified from turmeric powder were purchased from Sigma-Aldrich (Oakville, ON, Canada). A powdered form of ZnO NPs (size: 40–100 nm, surface area: 12–24 m<sup>2</sup> /g) was obtained from Alfa Aesar (Haverhill, MA, United States). Acetic acid, acetonitrile, chloroform and dimethyl sulfoxide (DMSO) were purchased from Sigma-Aldrich (Oakville, ON, Canada).

### Preparation, Extraction and Quantification of Encapsulated Curcumin

To increase the water solubility, curcumin was encapsulated by HMS according to the protocol described in a previous study (Yu and Huang, 2010). Briefly, 1.8 g of curcumin was added into 1% (w/v) HMS solution, homogenized at 8,000 rpm for 10 min by using Omni Mixer Homogenizer (Omni, Kennesaw, GA, United States), followed by stirring for 24 h at room temperature. The suspension was centrifuged at 11,617 × g for 5 min. Supernatant was collected and filtered through a 0.45 µm nylon syringe membrane (Milliporesigma, Mississauga, ON, United States). Filtered encapsulated curcumin was placed at −80◦C for 4 h and then freeze-dried at −53◦C using a 12-L Console Labconco freeze-dryer (Kansas city, MO, United States) for 24 h. The freeze-dried product was packaged and stored at −20◦C for further use. The negative control was that of HMS processed without the addition of curcumin.

Encapsulated curcumin was extracted and quantified as follows. First, 1.5 mL of double-deionized water was added to 0.5 g of freeze-dried encapsulated curcumin. The curcumin aqueous solution was then mixed with an equal volume of chloroform in a 10-mL glass tube. Two-phase liquid-liquid emulsions were obtained and then stirred at 200 rpm for 10 min to remove any free curcumin. Next, the emulsion was mixed with chloroform at the same volume, followed by vortex for 10 min and stirring at 200 rpm for overnight. Chloroform paste was collected then filtered through a 0.45-µm membrane before quantification. The extracted curcumin was analyzed using an Agilent 1260 HPLC system coupled with a diode array detector (HPLC-DAD; Agilent Technology, Santa Clara, CA, United States). An aliquot (20 µL) of sample was injected into a SUPELCOSILTM C<sup>18</sup> column (300 × 4.6 mm, 3 µm; Sigma-Aldrich, Oakville, ON, Canada) and eluted at a flow rate of 1.0 mL/min and ambient temperature. The extracted curcumin was analyzed using an Agilent 1260 HPLC system coupled with a diode array detector (HPLC-DAD; Agilent Technology, Santa Clara, CA, United States). An aliquot (20 µL) of sample was injected into a SUPELCOSILTM C<sup>18</sup> column (300 × 4.6 mm, 3 µm; Sigma-Aldrich, Oakville, ON, Canada) according to the protocol described in a previous study (Jayaprakasha et al., 2002). The mobile phase consisted of (A) 2% acetic acid (v/v) in water and (B) acetonitrile. A linear gradient was conducted from 45% B to 65% B within 8 min, then to 90% B before 10 min, and returned to 45% B before 15 min. Chromatograms were obtained at 425 nm. The retention time of curcumin was 7.5 min.

### Bacterial Strains

The C. jejuni F38011 (clinical isolate), human10 (clinical isolate) (Li et al., 2017), and ATCC 33560 (bovine feces isolate) strains were routinely cultivated either on Campylobacter agar (OXOID, Nepean, ON, Canada) plates supplemented with 5% defibrinated sheep blood (Alere Inc., Stittsville, ON, Canada) for 48 h or in Mueller–Hinton (MH) broth with constant shaking for 18 h at 37◦C in a microaerobic environment (i.e., 10% CO2). Overnight C. jejuni cultures were individually prepared to 1 × 10<sup>9</sup> CFU/mL by adjusting OD<sup>600</sup> value. Then, a cocktail culture was prepared by combining equal volumes of each of the three cultures for further antimicrobial testing.

### Investigation of Synergistic Antimicrobial Effect Using Time-Killing Method

The minimum bactericidal concentrations (MBC) of cinnamon oil, ZnO NPs, and encapsulated curcumin against C. jejuni were separately determined using time-killing method (Wiegand et al., 2008). A series of two-fold dilutions from 1 to 100 ppm of antimicrobials were individually mixed with C. jejuni cocktail, with the initial cell count of 10<sup>5</sup> CFU/mL (**Supplementary Table S1**), and the survival cells were determined after different time intervals using the conventional platting assay. The MBC value was used as the reference concentration for synergistic antimicrobial testing. Different concentrations (2×, 1×, 0.5× and 0.25× MBC) of cinnamon oil were individually combined with the MBC of ZnO NPs and C. jejuni cocktail culture (**Supplementary Table S1**). To identify the interaction between antimicrobials, the results of individual antimicrobial treatments were compared to that of the combined antimicrobial treatment. Log reduction was calculated by subtracting the bacterial count in treated group from the control group at the same time point. The experiment was conducted in triplicates.

### Investigation of Synergistic Antimicrobial Effect Using Fractional Inhibitory Concentration Index Method

The interaction of antimicrobial effects between cinnamon oil, encapsulated curcumin, and ZnO NPs was also evaluated using the FICI method. The minimum bacteriostatic concentration (MIC) of single and dual antimicrobials was determined then the FICI of each combination was identified as follows: FICI = (MIC of A in combination / MIC of A alone) + (MIC of B in combination / MIC of B alone), where A and B represent a different antimicrobial agent. Tests were performed at FICI values of 0.3, 0.4, 0.5, and 1 in order to identify the type of interactions between antimicrobials. The minimum FICI value that inhibited bacterial growth was defined as synergistic (≤0.5), additive (0.5−4), or antagonistic (>4) (Odds, 2003). Antimicrobials were prepared in fresh MH medium containing C. jejuni cocktail with an initial cell density of 10<sup>5</sup> CFU/mL. The procedure was, similarly applied in all standardized susceptibility methods (Wiegand et al., 2008). The inhibitory effect was reported as either positive or negative based upon the clarity (+) or turbidity (−) of the tested bacterial cultures. The experiment was conducted in triplicates.

## Investigation of Synergistic Antimicrobial Effect Using Mathematical Modeling

Concentration-Effect Curves of Single Antimicrobials A broad range of concentrations of cinnamon oil, ZnO NPs, and encapsulated curcumin were tested against C. jejuni strain F38011 to identify the concentration-effect curve after a 3 h of treatment. Each antimicrobial was prepared at a series of concentrations (**Supplementary Table S2**) and then individually mixed with C. jejuni at an initial concentration of 10<sup>8</sup> CFU/mL, followed by incubation at 37◦C for 3 h in a microaerobic environment. Antimicrobial testing was performed using the conventional plating assay. The experiment was conducted in triplicates. Antimicrobial effect was reported as a percent and log reduction of the bacterial cells. Concentration-effect curves were generated using Prism 5 software (GraphPad, San Diego, CA, United States).

### Preparation of Antimicrobial Combinations

Concentration-effect data were used to prepare different pairs of concentrations based on their potencies (e.g., IC<sup>20</sup> + IC20). The inhibitory concentrations (e.g., IC20) used for preparing pairs of concentrations were determined by:

fmicb-10-01038 May 11, 2019 Time: 14:9 # 4

$$IC\_F = \left(\frac{F}{100 - F}\right)^{\frac{1}{H}} \times IC\_{50} \tag{1}$$

where IC<sup>F</sup> represents the inhibitory concentration (e.g., IC20), F denotes the cell percentage reduction (e.g., 20%), H stands for the Hill slope, and IC<sup>50</sup> is the inhibitory concentration that gives 50% reduction of the cells. The theoretical additive effect (e.g., the sum effect of IC<sup>20</sup> + IC20) was calculated by the fractional product method (Chou and Talalay, 1984):

$$(1 - f\_1)(1 - f\_2) = \,\_1V\_{1 \& 2}$$

Theoretical additive effect

$$(\%) = (1 - V\_{1\&2}) \times 100\tag{2}$$

where f represents the fraction of cell reduction (CFU/mL) by single antimicrobials (i.e., f<sup>1</sup> and f2) and V1&2 represents the theoretical fractional concentration of viable cells after treatment with two antimicrobials. Combined effect can be synergistic, additive, or antagonistic. Synergistic effect takes place when the combined effect is greater than the theoretical additive effect, while additive or antagonistic effects take place when the combined effect is equal to or lower than the theoretical additive effect, respectively. The maximum theoretical additive effect used in this study was 82% (<1 log) CFU/mL to ensure that the antimicrobial combinations do not completely inactivate the entire bacterial population. In this case, the 18% remaining population accounts for (>7 log) CFU/mL.

#### Isobologram

Isobologram analysis was used to investigate the interaction effect (synergistic, antagonistic, or additive interaction) between different binary combinations. Sets of equally effective concentrations were selected to generate isobolograms according to the method described in a previous study (Chou and Talalay, 1984). The IC<sup>50</sup> of each single antimicrobial was used to draw an additive line between the combined antimicrobials. Reduction of antimicrobials was calculated according to:

$$R = \frac{IC\_{50}\text{ in combination}}{IC\_{50}\text{ alone}} \times 100\tag{3}$$

where R represents the reduction of antimicrobial concentration. The isobolograms were generated using Prism 5.

#### Median-Effect Plot

A systematic analysis of the concentration-effect data of single and combined antimicrobials was conducted to generate a median effect plot. The data were normalized by Chou's median effect equation (Chou, 2006) as follows:

$$\log\frac{F}{100-F} = \log\left(\frac{C}{IC\_{50}}\right)^H \tag{4}$$

where C represents the concentration of antimicrobial(s). The median effect plot was generated using Prism 5.

### Statistical Analysis

Prism software (version 5.01: GraphPad Software Inc., San Diego, CA, United States) was used for statistical analysis and the graphs generation. The time-killing data were analyzed by oneway ANOVA, followed by post hoc Tukey's test for multiple comparisons. A P value was adjusted at 0.05 or less to define statistically significant differences between and within groups.

### RESULTS AND DISCUSSION

### Conventional Methods Used to Study the Antimicrobial Interactions Time-Killing Method

The time-killing method was used first to study single and combined antimicrobial treatments. Cinnamon oil, ZnO NPs, and encapsulated curcumin showed bactericidal activity at 1 × MBC and 2 × MBC after 3, 6, 12, and 24 h (P < 0.0001) (**Figure 1A**). The 0.5 × MBCs of all single antimicrobials had a mild effect with ≤1 log reduction and no significant effect compared to the control groups at all time points (P > 0.05) (**Figure 1**). The combination of cinnamon oil and ZnO NPs at low concentrations (≤0.5 × MBC) significantly enhanced the antimicrobial effect. For example, ZnO NPs at 0.5 × MBC and cinnamon oil at 0.25 × MBC resulted in a 6.24 log reduction of C. jejuni after 12 h of treatment (P < 0.0001) (**Figure 2**), while the same concentrations of both single antimicrobials showed no significant difference (P > 0.05) compared to the control group even after 24 h (**Figures 1A,B**).

Dimethyl sulfoxide (1%) did not significantly affect cell viability (P > 0.05) (**Figure 1A**), indicating that the antimicrobial effect of cinnamon oil solution was not due to the solvent. Similarly, up to 100 ppm of free curcumin (without encapsulation or HMS) and 1% HMS had no effect on the viability of C. jejuni (P > 0.05) (**Figure 1C**). The effects of 1% DMSO or 1% HMS with and without antimicrobial agents (4 × MIC) were significantly different after 3, 6, 12, and 24 h (P < 0.0007) indicating no interaction between these compounds and antimicrobials against cell viability (**Figures 1A,C**).

Synergism was identified when two antimicrobials resulted in a greater log reduction than the sum of their individual effects [34–36]. In the current study, the single treatment of cinnamon oil (0.5 × MBC) or ZnO NPs (0.5 × MBC) induced a ≤1 log reduction after 24 h (P > 0.05) (**Figures 1A,B**). An additive effect was considered if the combined treatment caused ∼2 log reduction after 24 h. Interestingly, the combination of cinnamon oil (0.5 × MBC)/ZnO NPs (1 × MIC) resulted in 5 log reduction within 6 h of treatment (P < 0.0001) (**Figure 2**), inducing additional log-reduction (CFU/mL) compared to the additive effect. Therefore, cinnamon oil and ZnO NPs were considered to have a synergistic effect against C. jejuni. Although previous studies used this approach to study antimicrobial synergic effect (Ghosh et al., 2013; Ha and Kang, 2015; Huq et al., 2015), it measures the effect on the basis of logarithmic scale, leading to potential overestimation of the interaction between the two antimicrobials. For example, the 0.5 × MBC of all

single antimicrobials showed ≤1 log reduction (≤90% of bacterial population) (**Figures 1**, **2**). If 90% of bacterial cells can be inactivated by one antimicrobial at the 0.5 × MBC, it is likely that the other antimicrobial can easily inactivate the remaining population (∼10% of the viable cells) at its 0.5 × MBC when both antimicrobials are applied at the same time. It is not reasonable to draw an additive line for two antimicrobials based on their log reductions because the sum of the two combined concentrations may fully inactivate the entire bacterial population. Although the time-killing method has many advantages including that the antimicrobial effects are monitored over time, our current results showed that this method overestimated the synergism between cinnamon oil and ZnO NPs. Thus, other methods were further applied to investigate the synergism between the selected antimicrobials.

### FICI Method

The FICI method has been frequently used to study antimicrobial interactions because it is simple and can test many concentrations simultaneously (Odds, 2003). The FICI method relies on combining multiple sub-MICs. The MIC at specific FICI value indicates the type of interaction. The additive effect takes place when the FICI is >0.5−1 (Odds, 2003). A lower FICI value indicates a higher synergism while a higher FICI value indicates antagonism. We used the FICI method to study the interactions between the binary combinations of cinnamon oil/ZnO NPs, cinnamon oil/encapsulated curcumin, and ZnO NPs/encapsulated curcumin. C. jejuni cocktails (5 log CFU/mL) were treated with antimicrobials for 24 h. All of the combinations did not show any inhibitory effect at a FICI value of 0.3, 0.4, or 0.5 (**Table 1**), indicating that no synergistic interaction occurred. In comparison, all combinations inhibited the growth of C. jejuni at a FICI value of 1, demonstrating an additive effect. Unlike the time-killing method, the FICI method did not show any synergistic interactions between tested antimicrobials (**Figure 2** and **Table 1**).

Both the time-killing method and the FICI method rely on the assumption that antimicrobials are equally effective at MICs/sub-MICs or MBC/sub-MBC. However, even when two antimicrobials at MICs cause similar effects, the antimicrobial effect at their sub-MICs might be totally different from each other depending on the shape of their concentration-effect curve [37]. It is difficult to predict the additive effect by combining two antimicrobials with different concentration-effect shapes. The extreme difference between two antimicrobials in

TABLE 1 | Antimicrobial effect of binary combinations of cinnamon oil, zinc oxide nanoparticle and encapsulated curcumin on C. jejuni cocktail.


The turbidity of the bacterial culture was observed after 24 h of incubation at 37◦C in a microaerobic condition. Initial bacterial count was 10<sup>5</sup> CFU/mL. The "+" denotes turbid and the "–" indicates clear (n = 3, duplicates). The average bacterial concentration in the turbid samples was 9.69 ± 0.27 log CFU/mL. <sup>∗</sup>FICI: fractional inhibitory concentration index.

their concentration-effect relationship highlights the challenge to predict the additive line of antimicrobial combinations in the conventional methods. Moreover, the potency of each antimicrobial can be unequal when they are mixed at 0.50 × or 0.25 × MICs (**Figure 3**). Friedman et al. (2002) tested the antimicrobial effects of 96 essential oils and 23 purified oil compounds against C. jejuni, Escherichia coli, Listeria monocytogenes, and Salmonella enterica and obtained different shapes of concentration-effect curve. Although this is a common problem, none of the conventional methods take this into consideration. Our results (**Table 1**) agreed with a previous study demonstrating the inability of FICI method to identify synergism between synergistic antimicrobials (Lambert and Lambert, 2003). The limitation is simply due to different dose-response of the combined antimicrobials. Therefore, it is important to use alternative methods to discover new synergistic combinations

that are not detected by the conventional methods. Taken together, FICI may not be able to identify an existing synergism between antimicrobials if they have different concentrationeffect curves.

### Mathematical Modeling Used to Study the Antimicrobial Interactions

A non-linear mathematical concentration-effect model was used to combine antimicrobials in a more detailed manner using a selected strain and a selected time point. The isobologram and median-effect equation were used to analyze the data (Chou and Talalay, 1984; Tallarida, 2001; Chou, 2006; Tallarida, 2006). The idea was to combine antimicrobials based on their quantitative potencies (i.e., IC<sup>40</sup> + IC40) instead of sub-MICs or sub-MBCs. Potency is the antimicrobial effect of a specific concentration (e.g., IC50) of an antimicrobial agent, while efficacy it the maximum effect induced by an antimicrobial. C. jejuni strain F38011 was selected because it was isolated from an individual with bloody diarrhea (O'Loughlin et al., 2015), effectively colonizes the chicken gastrointestinal tract (Konkel et al., 1998), and causes illness in mice and pigs (O'Loughlin et al., 2015). Only a single time point (i.e., 3 h) was selected to study the concentration-effect curve for each antimicrobial because 3 h was identified to be the minimum time period to cause Campylobacter reduction by single or dual antimicrobials (**Figures 1**, **2**). Similarly, another study tested the antimicrobial effect of ZnO NPs and identified that 3 h of treatment was the minimum time to reduce the count of C. jejuni (Xie et al., 2011). Taken together, different steps and considerations were used in a nonlinear concentration-effect mathematical model to study the antimicrobial interactions.

### Concentration-Effect Curves

fmicb-10-01038 May 11, 2019 Time: 14:9 # 7

All antimicrobials demonstrated a non-linear regression between the concentrations and antimicrobial effects (**Figure 4**). Cinnamon oil and ZnO NPs showed a similar shape of concentration-effect curve with a slight difference, while encapsulated curcumin presented a sharp sigmoidal curve. IC<sup>50</sup> is a representative concentration used to evaluate the potency of drugs or antimicrobials. The lower the IC<sup>50</sup> the more potent is the drug. Up to thirteen concentration-effect points were used to generate high quality concentration-effect curves. Some points showed no effect; some were on the slope; and some were on the plateau of their concentration-effect curves. Other methods, such as the FICI, rely on one reference concentration (MIC) for studying antimicrobial interactions. We used multiple data points and identify the IC<sup>50</sup> as a representative concentration (located on the middle of the concentration-effect slope) to determine nine other reference concentrations. The IC<sup>50</sup> of cinnamon oil, ZnO NPs, and encapsulated curcumin were identified to be 0.90, 1.20, and 1.48 log ppm, respectively (**Figure 4**). All three individual antimicrobials had the same efficacy because they were all able to reduce bacterial population over 99% at 1.39 log ppm for cinnamon oil, 1.54 log ppm for ZnO NPs, and 1.60 log ppm for encapsulated curcumin (**Figure 4**).

### Isobologram

The isobologram is a diagram that is commonly used to identify the type of interaction of a combination by comparing the concentrations of two single agents (x and y-axis) and their combination (axial point) (**Figure 5**). The IC<sup>50</sup> of single antimicrobial cinnamon oil (7.92 ppm), ZnO NPs (14.75 ppm) and encapsulated curcumin (29.63 ppm) were used to draw additive lines. The additive effect is indicated if the combined (IC50) data point is on the additive line and synergism is indicated if the combined IC<sup>50</sup> data point is below the additive line. The combination of cinnamon oil/ZnO NPs (4.14 + 10.37 ppm) indicated an additive interaction (**Figure 5A**). In contrast, cinnamon oil/encapsulated curcumin (0.523 + 11.70 ppm) (**Figure 5B**) and encapsulated curcumin/ZnO NPs (11.70 + 4.14 ppm) (**Figure 5C**) indicated synergistic interaction. Synergism between cinnamon oil and encapsulated curcumin resulted in the reduction of antimicrobial concentrations by 93.40% and 60.51%, respectively (**Figure 5B**). In comparison, ZnO NPs and encapsulated curcumin were reduced by 81.76% and 60.51%, respectively (**Figure 5C**). Taken together, up to 93.40% of the antimicrobial concentrations were reduced due to the synergistic interactions.

### Median-Effect Curves

The median-effect principle has been employed to analyze the dose-response data in enzymatic, cellular, and animal studies. A few studies have used this approach to study the antimicrobial effects, such as the ones of using the median-effect equation to investigate the antiviral synergistic effect against human immunodeficiency virus (HIV) and influenza virus A (Asin-Milan et al., 2014; Kulkarni et al., 2014; Belardo et al., 2015). Most studies used the CompuSyn software to generate the medianeffect plot (Chou, 2010). However, this software does not consider the shape of dose-response. Hence, it cannot overcome some of the possible mathematical errors in combinational studies. In the current study, we used non-linear concentrations-effect

(C) against C. jejuni strain F380I1 after 3 h treatment at 37◦C in a microaerobic condition. The additive lines connect the IC<sup>50</sup> of each single antimicrobial. Axial points show the IC<sup>50</sup> of antimicrobial combinations. An axial point on the additive line indicates additive interaction. An axial point under the additive line indicates synergistic interaction (n = 3, duplicates).

FIGURE 6 | Median-effect plots of cinnamon oil, ZnO NP, and encapsulated curcumin against C. jejuni F38011 strain after 3-h treatment at 37◦C in a microaerobic condition. Data were plotted for single or dual treatments as binary (A–C) or tertiary combinations (D). F represents the percentage reduction of bacterial cells. C means the antimicrobial concentration and H is the slope hill of concentration-effect curves. Each line represents the effect of single or combinational antimicrobial treatment. The "0" on the x-axis represents median-effect and the "0" on the y-axis represents a 50% reduction of viable cells. The two unconnected lines show the antimicrobial effects at MIC (90% reduction) and MBC (99.9% reduction) of viable cells. The line slope represents the potency of the antimicrobial treatment (n = 3, duplicates).

data to generate the median-effect plot and identified the overall effect of antimicrobials and their combinations in a wide range of concentrations. By using the median-effect principle, more meaningful data were provided because the IC<sup>50</sup> shown in the isobologram was not sufficient for treatments (**Figure 5**).

**Figure 6** shows the median-effect curves of single antimicrobials and their combinations at a broad range of concentration-effect relationship between 0 and >8 log reduction of bacterial cells. Cinnamon oil was identified to be the most potent antimicrobial because its median-effect curve reached to

TABLE 2 | Binary (A–C) and tertiary (D) combinations of cinnamon oil, encapsulated curcumin, and ZnO NP against C. jejuni F38011 after 3 h treatment at 37◦C in microaerobic condition (n = 3, duplicates).


the maximum effect (>8 log) first followed by ZnO NPs, and then the encapsulated curcumin (**Figure 6**). Although we did not study the mode of action of single or dual antimicrobials, the median-effect curves of cinnamon oil and ZnO NPs had the same slope, suggesting that they might work with a similar mechanism (**Figure 6A**). The primary antibacterial functions of cinnamon oil and ZnO NPs are both related to the disruption of bacterial cell wall and cell membrane (Davidson et al., 2005; Xie et al., 2011). In contrast, curcumin acts as a cytokinesis inhibitor by direct interaction with FtsZ (Rai et al., 2008), an essential bacterial cell-division protein. This might explain why the encapsulated curcumin showed a sharp sigmoidal concentration-effect curve (**Figure 4**) and a very steep median-effect curve (**Figures 6B–D**).

Several pairs of the fixed ratio concentrations (from IC<sup>10</sup> to IC40) were used to generate the median-effect plot. The median-effect line of cinnamon oil/ZnO NPs was almost horizontal and unable to cross the MIC (i.e., additive) line (**Figure 6A** and **Table 2A**). In comparison, all other combinations showed synergistic interactions and were able to cause antimicrobial reduction about three times greater (3 log-reduction) than the additive line (>1 log reduction) (**Figures 6B,C** and **Tables 2B,C**). The tertiary combination of cinnamon oil/ZnO NPs/encapsulated curcumin had the greatest synergistic interaction among the tested combinations because mixing three concentrations of IC<sup>25</sup> (additive line 57%) resulted in over 8 log reduction of bacterial viable cells (**Figure 6D** and **Table 2D**). The median-effect curves of all-synergistic combinations were parallel with the curve of the encapsulated curcumin (**Figures 6B,C**), suggesting that encapsulated curcumin played an important role in the synergistic interaction.

### CONCLUSION

In conclusion, conventional methods either overestimated or failed to detect the exiting synergic antimicrobial interactions

### REFERENCES


due to the undefined concentration-effect curves. We were able to identify reference concentrations and evaluate the synergism even between antimicrobials with different concentration-effect curves. Up to 93.40% of the antimicrobial concentrations were reduced while maintaining the same effect due to synergistic interactions. Cinnamon oil and ZnO NPs acted differently from the encapsulated curcumin. We propose that this mathematical modeling can aid in developing new synergistic combinations to potentially reduce the prevalence and survival of foodborne pathogens as well as open the door to discover new mechanisms of dual antimicrobials.

### AUTHOR CONTRIBUTIONS

MH provided the idea of the study, conducted the experiments, and wrote the manuscript. KA contributed to the idea of the study and analyzed the data. LM reviewed and edited the manuscript. KC reviewed and helped the mathematical modeling. MK reviewed and edited the manuscript. XL supervised, reviewed, and edited the manuscript.

### FUNDING

This work was supported by funds awarded to XL by Natural Sciences and Engineering Research Council of Canada (NSERC RGPIN-2014-05487). MH received a 4-year scholarship (2015– 2019) from King Saud University.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.01038/full#supplementary-material



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

Copyright © 2019 Hakeem, Asseri, Ma, Chou, Konkel and Lu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Effect of Previous Life Cycle Phase on the Growth Kinetics, Morphology, and Antibiotic Resistance of *Salmonella* Typhimurium DT104 in Brain Heart Infusion and Ground Chicken Extract

#### *Edited by:*

*Giovanna Suzzi, University of Teramo, Italy*

### *Reviewed by:*

*Abhinav Mishra, University of Georgia, United States Guillermo Tellez, University of Arkansas, United States*

> *\*Correspondence: Salina Parveen sparveen@umes.edu*

*† Present address: Jabari L. Hawkins, USDA FSIS, Washington, DC, United States*

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 31 August 2018 Accepted: 25 April 2019 Published: 24 May 2019*

#### *Citation:*

*Hawkins JL, Uknalis J, Oscar TP, Schwarz JG, Vimini B and Parveen S (2019) The Effect of Previous Life Cycle Phase on the Growth Kinetics, Morphology, and Antibiotic Resistance of Salmonella Typhimurium DT104 in Brain Heart Infusion and Ground Chicken Extract. Front. Microbiol. 10:1043. doi: 10.3389/fmicb.2019.01043*

*Jabari L. Hawkins1† , Joseph Uknalis 2 , Tom P. Oscar 3 , Jurgen G. Schwarz1 , Bob Vimini 4 and Salina Parveen1 \**

*1Food Science and Technology Program, Department of Agriculture, Food and Resource Sciences, University of Maryland Eastern Shore, Princess Anne, MD, United States, 2Molecular Characterization of Foodborne Pathogens Research Unit, USDA/ARS, Eastern Regional Research Center, Wyndmoor, PA, United States, 3Chemical Residue and Predictive Microbiology Research Unit, USDA/ARS, University of Maryland Eastern Shore, Princess Anne, MD, United States, 4Perdue Farms Inc., Salisbury, MD, United States*

Growth models are predominately used in the food industry to estimate the potential growth of selected microorganisms under environmental conditions. The growth kinetics, cellular morphology, and antibiotic resistance were studied throughout the life cycle of *Salmonella* Typhimurium. The effect of the previous life cycle phase [late log phase (LLP), early stationary phase (ESP), late stationary phase (LSP), and early death phase (EDP)] of *Salmonella* after reinoculation in brain heart infusion broth (BHI), ground chicken extract (GCE), and BHI at pH 5, 7, and 9 and salt concentrations 2, 3, and 4% was investigated. The growth media and previous life cycle phase had significant effects on the lag time (λ), specific growth rate (μmax), and maximum population density (*Y*max). At 2 and 4% salt concentration, the LLP had the significantly (*p* < 0.05) fastest μmax (1.07 and 0.69 log CFU/ml/h, respectively). As the cells transitioned from the late log phase (LLP) to the early death phase (EDP), the λ significantly (*p < 0.05*) increased. At pH 5 and 9, the EDP had a significantly (*p* < 0.05) lower *Y*max than the LLP, ESP, and LSP. As the cells transitioned from a rod shape to a coccoid shape in the EDP, the cells were more susceptible to antibiotics. The cells regained their resistance as they transitioned back to a rod shape from the EDP to the log and stationary phase. Our results revealed that growth kinetics, cell's length, shape, and antibiotic resistance were significantly affected by the previous life cycle phase. The results of this study also demonstrate that the previous life cycle should be considered when developing growth models of foodborne pathogens to better ensure the safety of poultry and poultry products.

Keywords: pH, salt, water activity, microscopy, *Salmonella*, growth, modeling

### INTRODUCTION

*Salmonella* is a Gram-negative, rod-shaped, and non-spore forming bacterium. Salmonellosis, an infection caused by *Salmonella*, is one of the most common and widely encountered foodborne diseases, with tens of millions of human cases occurring worldwide annually (WHO, 2013). In the United States, there is an estimated 1 million salmonellosis cases annually causing 19,000 hospitalizations and 380 deaths (Scallan et al., 2011). This has resulted in an increased interest to better understand the growth and changes of *Salmonella* under antibiotic and environmental stresses.

Wen et al. (2009) suggested that current food safety studies that base the effectiveness of interventions on the ability to reduce foodborne pathogens in the stationary phase may be overestimated. They found that *Listeria monocytogenes* in the long-term stationary (LTS) phase has more resistance to thermal and high-pressure processing than in the stationary phase. As the availability of nutrients dissipated and *L. monocytogenes* transitioned into the LTS phase, the morphology changed from a rod to a coccoid shape. The coccoid shape increased its survivability and resistance to treatments. Wen et al. (2013) also found that initial viable cell density (~106 to ~1010 CFU/ml) and pH (5.36–6.85) of *L. monocytogenes* affected the transition to the LTS phase.

Although many models exist for the growth of bacteria in model food systems (Oscar, 1997, 1999a, 2006, 2013; Mellefont et al., 2003; Ross et al., 2003; Parveen et al., 2007), there is limited information about the regrowth of bacteria at previous life cycle phases such as lag, log, stationary, death, and longterm stationary (LTS) phase. The objective of this study was to investigate how the phase of *S.* Typhimurium DT104 at the log, stationary, and death phase affects its transition back to the stationary phase. This study was conducted to observe the microbial behavior and the effect of previous life cycle phases at a constant temperature on the growth, cell morphology, and antibiotic resistance patterns of *Salmonella* in inoculated brain heart infusion broth and ground chicken extract.

### MATERIALS AND METHODS

### Preparation of Bacterial Inoculum

A multiple-antibiotic resistant strain (ATCC 700408, American Type Culture Collection, Manassas, VA) of *S.* Typhimurium definitive phage type 104 (DT104) was used in this study. This strain has an antibiotic resistance profile (chloramphenicol, ampicillin, tetracycline, and streptomycin) that is found in chicken isolates (Parveen et al., 2007) and has been used in similar growth modeling studies (Oscar, 2006, 2007). The stock culture of this strain was maintained at −80°C in brain heart infusion broth (BHI) (BBL, Difco, BD, Sparks, MD) that contained 15% (vol/vol) glycerol (Sigma, St. Louis, MO).

### Previous Life Cycle Phase

A growth curve was developed to identify the life cycle phases of DT104 (**Figure 1**). One microliter of the stock culture was transferred into sterile BHI (5 ml) in a centrifuge tube (15 ml)

followed by incubation at 37°C and 150 rpm from 0 to 770 h. This was carried out in three replicates and 50 μl of undiluted and appropriate 1:10 diluted (10<sup>−</sup><sup>1</sup> to 10<sup>−</sup><sup>5</sup> ) samples from cultures were spiral plated (Whitely Automated Spiral Plater, Microbiology International, Frederick, MD) onto tryptic soy agar (TSA) plates. The TSA plates were incubated at 37°C for 24 h and then bacterial colonies were enumerated. In **Figure 1**, after the stationary phase, the culture had a slow and extended death phase. Since there is no time period of a stable density, it is referred to as the early death phase (EDP) instead of the LTS phase throughout the paper.

### Starter Culture

One microliter of the stock culture was transferred into sterile BHI (5 ml) in a centrifuge tube (15 l) followed by incubation at 37°C and 150 rpm to obtain cells in the late log phase (LLP, 9 h), early stationary phase (ESP, 24 h), late stationary phase (LSP, 48 h) and early death phase (EDP, 720 h), respectively. The culture was centrifuged for 5 min at 2,100 × *g* and the pellet resuspended into 5 ml of the selected media as described below and adjusted to achieve a starting concentration of 104 CFU/ml.

### Growth Cultures

Analysis of cultures of DT104 for growth kinetic determinations were performed in centrifuge tubes (15 ml). Cultures grown under pH stress contained 5 ml of BHI adjusted to pH of 5, 7, or 9 with 1 N HCl or 1 N NaOH. Cultures grown under osmotic stress contained 5 ml of BHI adjusted to 2, 3 and 4% NaCl (Difco Lab, MI, USA). DT104 was also grown in ground chicken extract (GCE) prepared as previously described in Fratamico et al. (2011) with modifications. Fresh raw, 85% lean ground chicken was obtained from a local supermarket and high pressure processed (HPP) at 87,500 psi (603 MPa) for 2.5 min. Sterile water (15 ml) was added to ground chicken (50 g) in a stomacher bag and pummeled for 1 min, the liquid extract was removed by centrifugation at 2,100 × g for 5 min, and then filtered using a 0.22 μm filter, and frozen at −20°C until used. Growth cultures were incubated at 37°C and 150 rpm for 0 to 30 h.

### Microbial Analysis

At selected times post inoculation, depending on the age of the culture and growth culture medium of incubation, 50 μl of undiluted and appropriate 1:10 diluted (10<sup>−</sup><sup>1</sup> to 10<sup>−</sup><sup>5</sup> ) samples from growth cultures were spiral plated (Microbiology International, Frederick, MD) onto tryptic soy agar (TSA) plates. Sampling times for each growth condition (i.e. pH, salt concentration and GCE) were based on estimated lag time (*λ*), specific growth rate (*μ*max), and maximum population density (*Y*max) and were selected to produce a growth curve that accurately defined the lag, log and stationary phase over three log cycles of growth. The TSA plates were incubated at 37°C for 24 h.

### Predictive Growth Modeling

U.S. Department of Agriculture, Integrated Pathogen Modeling Program (IPMP) 2013 was used in this study. It is a free and easy-to-use data analysis platform for fitting primary models. Primary models include common growth and inactivation models, which can be used to analyze full growth curves, incomplete growth curves and inactivation/survival curves (Huang, 2014).

The data from the samples were assigned to IPMP 2013. The primary models evaluated were the full-growth Baranyi (Baranyi and Roberts, 1994), reparametrized Gompertz (Zwietering et al., 1990), Huang (Huang, 2014), and Buchanan (Buchanan et al., 1997) three-phase linear. The performance of the models is based on the production of the Akaike information criterion (AIC). The AIC is to estimate the likelihood of a model to predict/estimate the future values. A lower AIC value indicates a better fit. The Baranyi model, results not shown, was chosen because it was the best fit for growth data and had the lowest AIC value.

The Baranyi model is the following:

$$Y(t) = Y\_0 + \mu\_{\max} A\left(t\right) - \ln\left\{1 + \frac{\exp\left[\mu\_{\max} A\left(t\right)\right] - 1}{\exp\left(Y\_{\max} - Y\_0\right)}\right\}$$

$$A\left(t\right) = t + \frac{1}{\mu\_{\max}} \ln\left[\exp\left(-u\_{\max}t\right) + \exp\left(-h\_0\right) - \exp\left(-u\_{\max}t - h\_0\right)\right]$$

where *Y*0, *Y*max, and *Y*(*t*) are the bacterial population, in natural logarithm of bacteria counts, at initial, maximum and time *t*. *μ*max is the specific growth rate; *h*o is the physiological state of the microorganism under consideration.

### Scanning Electron Microscope and Transmission Electron Microscope

Scanning electron microscope (SEM) and transmission electron microscope (TEM) was performed on cultures (20 μl) of DT104 in different life cycle phases as described by Wen et al. (2009). A one-way analysis of variance was used to determine the effect of the previous life cycle phase on the cell length of *S.* Typhimurium DT104. Pairwise comparisons were made by Tukey's least significant difference test (*α* = 0.05) using Statistix 9 (Analytical Software, Tallahassee, FL).

### Antibiotic Susceptibility Disk Diffusion

Cultures of DT104 in different life cycle phases were tested for antibiotic susceptibility by the disc diffusion method on Mueller-Hinton agar (Sigma-Aldrich, Munich, Germany).

TABLE 1 | Antibiotic disk diffusion of *Salmonella* Typhimurium DT104 as a function of previous life cycle phase.


*R = Resistant*

*I = Intermediate*

*LLP, late log phase; ESP, early stationary phase; LSP, late stationary phase; and EDP, early death phase.*

*S = Susceptible*

The antibiotic sensitivity of the cultures of DT104 in different life cycle phases were evaluated according to classification guidelines suggested by the National Committee for Clinical Laboratory Standards (NCCLS). The cultures were tested for susceptibility to a panel of antibiotics including gentamicin (GM; 10 μg), sulphamethoxazole x trimethoprim (SXT; 25 μg), kanamycin (K; 30 μg), tetracycline (TE; 30 μg), nalidixic acid (NA; 30 μg), trimethoprim (TMP; 5 μg), ciprofloxacin (CIP; 5 μg), centriaxone (CRO; 30 μg), sulfisoxazole (G; 250 μg), chloramphenicol (C; 30 μg), and streptomycin (S; 10 μg) using the disk diffusion method (Briggs and Fratamico, 1999; O'Leary et al., 2015). The zones of inhibition (ZOI) were measured and the isolates were classified as being susceptible, intermediate or resistant to the antibiotic according to the NCCLS guidelines.

### Statistical Analysis

Two-way analysis of variance (ANOVA) was used to determine the effect of the previous life cycle phase, growth medium (independent variables) and their interaction on the *λ*, *μ*max, and *Y*max (dependent variables) of *S.* Typhimurium DT104 in various growth mediums (GCE and BHI) generated from the Baranyi model using GraphPad Prism 7 (GraphPad Software, San Diego, CA).

## RESULTS

### Growth Models

The experimental data obtained at various previous life cycle phases and under environmental conditions and stresses was fitted to the full Baranyi growth model (**Figures 2–4**). The Baranyi model was a good fit to the growth data to accurately estimate the growth parameters (*λ*, *μ*max, and *Y*max) for statistical analysis. The lower and upper 95% confidence intervals were defined for individual predictions.

### Brain Heart Infusion Broth vs. Ground Chicken Extract

The cultures responded differently depending upon the phase of the culture. There was a significant (*p <* 0.05) interaction between the medium and previous life cycle phase on the *Y*max. As seen in **Figure 5**, the EDP in BHI had a lower (*p* < 0.05) *Y*max (8.95 log CFU/ml) in comparison to the LLP, ESP, and LSP (9.19, 9.16, and 9.16 log CFU/ml, respectively). In the GCE there was no significant (*p* > 0.05) effect of the previous life cycle phases on the *Y*max. There was no significant interaction between the media and previous life cycle phase on the *λ* and *μ*max therefore only the main effects will be discussed. In BHI, the *λ* significantly (*p* < 0.05) increased as the age of the previous life cycle phase increased except that the ESP *λ* equaled the LSP *λ* (LLP < ESP = LSP < EDP). There was also a significant (*p* < 0.05) effect of the previous life cycle on the *λ* in GCE. The EDP had the significantly *(p* < 0.05) longest *λ* at 2.47 h and there was a significant (*p >* 0.05) difference between LLP and LSP of 0.19 h (**Figure 5**). The LLP had a significantly (*p <* 0.05) lower *μ*max than ESP in both BHI (1.03 vs. 1.18 h<sup>−</sup><sup>1</sup> , respectively) and GCE (0.96 vs. 1.10 h<sup>−</sup><sup>1</sup> ) (**Figure 5**).

### pH Levels

There was a significant (*p* < 0.05) interaction on the *Y*max between the previous life cycle phase and various pH levels (**Figure 6**). At pH 5 and 9, the EDP had a lower (*p* < 0.05) *Y*max than the LLP, ESP, and LSP. While at pH 7, the previous life cycle phase had no significant (*p* > 0.05) effect on the *Y*max. While there was a significant (*p* < 0.05) interaction on the *Y*max, there was none (*p* > 0.05) on the *λ* and *μ*max. Similar to BHI and GCE, the *λ* significantly (*p* < 0.05) increased as the age of the previous life cycle phase (LLP, ESP, LSP, and EDP) increased in both pH 5 (1.40, 1.90, 2.41, and 3.67 h, respectively) and pH 9 (1.04, 1.58, 2.13, and 3.56 h, respectively) (**Figure 6**). At pH 7, the previous life cycle phase also had a significant (*p <* 0.05) effect on the *λ* (LLP < ESP = LSP < EDP). The previous life cycle phase had no significant (*p >* 0.05) effect on the *μ*max regardless of the pH level (**Figure 6**).

### Salt Levels

In pertaining to the salt concentrations and the previous life cycle phase, there was a significant (*p <* 0.05) interaction on the *μ*max (**Figure 7**). At 2 and 4% salt concentration, the LLP had the highest (*p* < 0.05) *μ*max and the previous life cycle phase had no significant (*p* < 0.05) effect on the *μ*max at 3% salt concentration. The previous life cycle phase and salt concentration had no significant (*p >* 0.05) interaction on the *λ* and *Y*max. At 3 and 4% salt concentration the *λ* of LLP and ESP was not significantly (*p* > 0.05) different and then the *λ* significantly (*p* < 0.05) increased from the ESP to the EDP. However, at 2% salt the EDP had the longest (*p* > 0.05) *λ* (3.91 h). The salt concentrations had no significant (*p >* 0.05) effect on the *Y*max regardless of the previous life cycle phase (**Figure 7**).

### Scanning Electron Microscope and Transmission Electron Microscope

Representative photomicrographs of cells of DT104 in different phases are shown in **Figure 8**. The SEM photomicrographs depicts the cells decreasing in size as they transitioned from the log phase (LP) to the EDP and size increased (recovered) transitioning back to the LP and stationary phase (SP) previously from the EDP. In the LP and SP the cells were rod-shaped while a cocci (spherical) shape appeared in the EDP. ANOVA of the SEM data revealed that cell length was significantly (*p <* 0.05) affected by the growth phase with the LLP (2.79 μm) = LLP (previously EDP) (2.53 μm) > SP (previously EDP) (2.25 μm) = SP (2.14 μm) > EDP (1.81 μm).

### Antibiotic Susceptibility Disk Diffusion

Cells from the LP, SP, and EDP grown in BHI were examined for their resistance to antibiotic treatment using susceptibility disk diffusion. DT104 from all phases were resistant to tetracycline (TE), sulfisoxazole (G), chloramphenicol (C) and streptomycin (S). In **Table 1**, cells recovered from the EDP were more susceptible to the antimicrobial agents than the other phases especially for sulphamethoxazole x trimethoprim (SXT), in which the EDP was classified as susceptible to the antibiotic (ZOI 17.5 mm) but ESP, LSP and LLP (previously EDP) were intermediate. Cells from the LLP (previously EDP) are cells recovered from the EDP, regrown in fresh BHI to the LLP. Cells from the LLP (previously EDP) ZOI was similar to ESP and LSP when subjected to SXT, nalidixic acid (NA) and ciprofloxacin (CIP) except for gentamicin (GM), kanamycin (K) and ciprofloxacin (CIP) it was more susceptible and trimethoprim (TMP) and centriaxone (CRO) less susceptible. The LLP was more susceptible

FIGURE 4 | Baranyi model for *S*. Typhimurium DT104 in brain heart infusion (BHI) at salt concentration 2, 3, and 4% at 37°C. Observed, predicted, upper 95% confidence interval (U95CI) and lower 95% confidence interval (L95CI).

FIGURE 5 | The lag time (λ), growth rate (μmax), and maximum growth (*Y*max) of *S*. Typhimurium DT104 as a function of the previous life cycle phase, late log phase (LLP), early stationary phase (ESP), late stationary phase (LSP), and early death phase (EDP) in brain heart infusion broth (BHI) and ground chicken extract (GCE). Different lowercase and uppercase letters within a cluster of bars differed significantly (*p <* 0.05).

FIGURE 6 | The lag time (λ), growth rate (μmax), and maximum growth (*Y*max) of *S*. Typhimurium DT104 as a function of the previous life cycle phase, late log phase (LLP), early stationary phase (ESP), late stationary phase (LSP), and early death phase (EDP) in brain heart infusion broth (BHI) at pH 5, 7, and 9. Different lowercase and uppercase letters within a cluster of bars differed significantly (*p <* 0.05).

to GM, SXT, NA and CIP than ESP and LSP except for CRO, LLP was less susceptible.

### DISCUSSION

### Growth Phases and Model Evaluation

It has been reported that the previous growth pH of a culture has a significant effect on the growth kinetics of *S.* Typhimurium (Oscar, 1999c) while the previous temperature and growth salt concentration is not a major factor affecting *S.* Typhimurium growth kinetics (Oscar, 1999a,b). The model development phase of this study involved 32 growth curves conducted under combinations of pH, NaCl % and previous growth phase (LLP, ESP, LSP, and EDP) identified in **Figure 1**. IPMP 2013 was able to accurately fit the experimental data to the full Baranyi growth model. The smaller the value of the residual standard deviation, the better the fit of the regression curve to the data is (Lee et al., 2015). The *λ*, *μ*max, and *Y*max values presented in this study pertain to the Baranyi model. The growth parameters may vary to some extent dependent upon the model. Growth at pH 5 (0.078) was the most accurately predicted data of the medium and the ESP (0.085) of the previous life phases.

Previous studies have demonstrated that after a short and rapid decline in the death phase, the culture maintains a stable density for an extended period of time (long-term survival phase) (Wen et al., 2009, 2013; Llorens et al., 2010). Exhibiting a different growth curve, in this study, the stationary phase (**Figure 1**) lasted for a longer period of time in comparison of the stationary phase in Wen et al. (2009). DT104 was not able to establish a LTS phase whereas Llorens et al. (2010) and Wen et al. (2009) maintained their own viable population at a constant population by continuous growth and death of cells. Although not establishing a LTS phase, the DT104 cells in the EDP died at a slow rate to survive for an extended period of time. When entering the death phase of the life cycle, the bacteria encounter an environment deprived of essential nutrients, i.e. carbon, phosphate, and nitrogen. *Salmonella* may encounter periods of starvation in natural, host microenvironments and commercial environments. When deprived of carbon energy sources, *S.* Typhimurium undergoes morphological changes as illustrated in **Figure 8** of the EDP which at times is referred to as starvation-stress response (SSR) (Spector and Kenyon, 2012).

To survive for an extended period of time, bacteria develop survival skills and strategies enabling them to persist in the environment until optimum growth conditions are met (Watson et al., 1998). Gram-negative bacteria such as *Vibrio cholerae, Escherichia coli,* and *S.* Typhimurium may enter a state of dormancy or viable but not culturable (VBNC). In this state, cells are capable of metabolic activity but are unable to undergo cellular division to form a colony (Watson et al., 1998; Spector and Kenyon, 2012). Additional research is needed to determine whether the cells in the EDP of **Figure 1** are VBNC, which may explain the slow death rate. *S.* Typhimurium may also utilize nutrients that are not readily available in optimal growth conditions (Spector and Kenyon, 2012). In non-favorable growth conditions and environments, living cells utilize cryptic growth, which is the recycling of nutrients from dead cells for maintenance. In comparison to active and growing cells, cells transition from a rod shape to an efficient and smaller coccoid shape aiding in long term survival (Watson et al., 1998; Wen et al., 2009, 2013). During that transition, *S.* Typhimurium may use quorum sensing (QS) when responding to their own population density (Ahmer et al., 1998). When a high-density cell culture senses limited nutrients, they undergo a form of programmed cell death, also known as bacterial apoptosis. The majority of the population enters a "death mode" while the surviving cells exit the death program after another signal is released perhaps from the lysed cells and may proceed to reproduce (Finkel, 2006). This also might explain the survival of cells in the EDP. Quorum sensing is first initiated by small hormone-like molecules called autoinducers (AIs) that accumulate and monitor the environment (Bearson and Bearson, 2008). The LuxR homologue, SdiA achieves intercellular signaling and is capable of interspecies communication (Walters and Sperandio, 2006).

### Brain Heart Infusion and Ground Chicken Extract

BHI is a common nutrient rich broth used for the growth of *S.* Typhimurium and other fastidious and nonfastidious bacteria. The components (g/L) in BHI include: brain heart, infusion from solids (6), peptic digest of animal tissue (6), pancreatic digest of gelatin (14.5), dextrose (3), sodium chloride (5), and disodium phosphate (2.5). Chicken contains protein, including all of the essential amino acids, minerals (calcium, iron, magnesium, phosphorous, potassium, sodium, zinc and selenium), vitamins (niacin, pantothenic acid, vitamin B6 and B12, folate and folic acid (USDA/ARS, 2016). BHI contains 3.0 g/l of dextrose, whereas GCE does not and has a lower pH (6.4) than BHI (7.4). Overall, the GCE had a lower *Y*max than BHI (8.77 vs. 9.12 log CFU/ml, respectively). This may be due to the limited nutrients in the GCE, containing no carbohydrates, while BHI contains dextrose.

The present study and previous ones (Oscar, 1999a,b, 2006) demonstrated that chicken based mediums can be used to develop models for the growth of *S.* Typhimurium in the place of traditional lab media. The *λ* of BHI after reinoculation of the ESP cells was 1.47 h which is similar to Oscar (1997) at 1.25 h with similar growth conditions. Oscar (1997) reported a slower *μ*max at 0.86 log CFU/ml/h compared to 1.18 log CFU/ml/h in the present study. The *λ* of GCE after reinoculation of the ESP cells was 1.45 h while in Oscar (2007), a study that developed predictive growth models for DT104 in ground chicken, had a *λ* of 1.39 and 1.66 h at 34 and 40°C, respectively. Using IPMP 2013 we determined the significance (*p* < 0.05) of the previous life cycle phase and media on the growth kinetics of *S.* Typhimurium DT104.

A follow-up study will be performed on *S.* Typhimurium DT104 in GCE examining the global transcriptional profile of DT104 to provide insights on how this food matrix influences growth and survival and potentially identify specific genes that may be targeted for the development of control strategies. Previously, it has been difficult to perform transcriptomics in complex medium but due to technology advancement it has become possible instead of using a simple medium as BHI (Fratamico et al., 2011; Deng et al., 2012). GCE may be a better growth medium to mimic a poultry environment in place of BHI for transcriptomic analysis of *Salmonella*.

### pH

*Salmonella* may encounter various pH levels in food production. Acid-based antimicrobial solutions such as chlorine, organic acids, peroxyacetic and peracetic acid are commonly used in the poultry industry to reduce *Salmonella* and *Campylobacter* contamination on poultry and poultry products. Studies such as Oscar (1999c) and Gibson et al. (1988) developed *Salmonella* growth models in broth media and demonstrated *Salmonella* ability to grow at low pH levels of 5.2 and 5.63, respectively. This study was carried out at a lower pH of 5 and demonstrated an ability of DT104 to grow in an unfavorable environment. Preliminary results (data not shown) demonstrated DT104 ability to grow at pH ≥ 4.5 but not at less than <4.5, this may be due to inability to maintain an adequate internal pH level. To the best our knowledge this is the first study that examined the effect of the previous life cycle and pH on the growth kinetics of a food pathogen. The starved cells in the EDP spent 2.27 h longer in the *λ* than LLP taking more time to adjust and adapt to pH 5 (**Figure 3**).

The growth kinetics have not frequently been studied at an alkaline pH. Alkaline cleaning agents are used in the food industry to remove biofilm, contamination, food, fat, and protein from processing equipment (Gibson et al., 1999; Sharma et al., 2005). Cleaners such as alkaline chlorinated cleaners work by temporarily imparting hydrophilic properties to stainless steel that decrease attachment to surfaces. While Humphrey (1987) carried out a study demonstrating the inactivation of *Salmonella* at high pH and temperature, this is the first study to the best of our knowledge performing predictive growth models of *Salmonella* at high pH (pH 9).

### Salt

Salt is commonly added to food products to lower the water activity and inhibit microbial growth of pathogens and spoilage bacteria. In the presence of salt, the growth of bacteria is dependent upon the presence of osmoprotectants. In an abrupt shift of osmotic stress, cells accumulate harmful substrates such as K+ and glutamine. These substrates are replaced by proline, trehalose and glycine betaine, which are taken up from the medium or synthesized *de novo* by the cell to stimulate growth (Zhou et al., 2011). Various studies (Matches and Liston, 1972; Mellefont et al., 2003, 2004, 2005) have been carried out showing an extended *λ* of *Salmonella* as the salt concentration increases and the aw decreases. The study by Mellefont et al. (2004) carried out at a lower aw demonstrated that the cells from the log phase after reinoculation had a longer relative lag time than the cells from the stationary phase in contrast to our study, where cells from the LLP after reinoculation had a shorter *λ* than the ESP but was carried out a higher *a*w until at 4% NaCl cells from the LLP had a longer *λ*. These studies demonstrate that at concentrations higher than 4%, the cells from the LLP are not able to adjust to abrupt osmotic shifts. The cells with shorter *λ* may have better defense systems which prepares them to survive various environmental stresses without prior exposure. Further research is needed to better understand these mechanisms.

Similar to other studies (Ross et al., 2003), the *μ*max decreased with the addition of salt (**Figure 7**). This occurrence may be due to the initial response to osmotic shift with cells having three phases: dehydration, adjustment, and rehydration. After respiration and ion transport are stopped from exposure to the shock, the respiration will resume and accumulate K+ , glutamate and solutes. Unlike, Zhou et al. (2011) there was an initial decrease in cell density in basic minimal medium at high salt concentration, DT104 did not have a decrease due to essential proteins and enzymes present in the BHI.

### Scanning Electron Microscope and Transmission Electron Microscope

Like *S.* Typhimurium DT104 in this study, other bacteria such as *L. monocytogenes* and *S.* Virchow have also shown a decrease in cell size (McMahon et al., 2007; Wen et al., 2009). DT104 transitioning back to the rod shape from the coccoid shape suggests that the cell formation is dependent upon the available nutrients. Starvation is known to induce a change in cell shape from rods to coccoid in *Arthrobacter crystallopoietes, A. globiformis, Rhizobium leguminosarum* and marine vibrio (Novitsky and Morita, 1976; Hamilton et al., 1977; Watson et al., 1998; Demkina et al., 2000). The EDP cells have a condensed cytoplasmic network in comparison to the other phases. This may explain the extended lag time of the subsequent EDP when introduced to a new environment. The coccoid shape may be formed by cell shrinkage and cytoplasmic condensation illustrated in **Figure 8** TEM. This may explain why the EDP cells require additional time to uptake available and essential nutrients and replenish fluidity in the cytoplasm.

### Antibiotic Susceptibility Disk Diffusion

To the best of our knowledge, this is the first study to compare the effect of life cycle phases on the antibiotic resistance profiles of viable cells of cultures after the stationary phase. The DT104 cells in the EDP exhibited less resistance to antibiotics than the cells from the LP and SP. The SP exhibited higher resistance to the antibiotics than the LP. A previous study by Tabak et al. (2006) demonstrated *Salmonella* in a planktonic state was more resistant in the SP than the LP to triclosan. Although the aging cells in this study showed less resistance, a study by Wen et al. (2009) demonstrated *Listeria monocytogenes* barotolerance and thermotolerance significantly (*p <* 0.001) increased as the cells transitioned from the late log phase to the long term survival phase. However, a study by Juck et al. (2011) demonstrated that the cells in the log phase exhibited higher resistance to HPP than cells in the stationary phase.

DT104 in the EDP phase may have exhibited less resistance to antibiotics due to the loss of plasmids after prolong incubation time and depletion of nutrients. Various studies have reported a modification of the plasmid pattern after long periods of incubation in marine water (Mejri et al., 2012). It is possible that as the cultures continue to grow in the media the oxygen and nutrients become progressively more limited resulting in rendering the EDP cells more sensitive to antibiotics compared to early cells. *Salmonella* is capable of forming biofilms in environmental stresses for protection and to increase its longterm survival. Biofilms, communities of bacterial cells, attach to one another and a surface. This structure may provide a higher degree of resistance to antibiotic agents, sanitizers and biocides. In the EDP, surviving in a state of low nutrients, DT104 formed visible clusters of biofilm along the wall and top layer of meniscus along the glass tube. However, the biofilm derived cells from the EDP were treated with antibiotics after disruption of the biofilm. This may have had an influence on the increased susceptibility to antibiotics compared to the other phases. After disruption, the potentially damaged cells from the EDP are exposed and vulnerable to the antibiotics.

### CONCLUSION

Using IPMP 2013, we accurately fitted the experimental data to the Baranyi full growth model to determine the significance of the previous life cycle phase (LLP, ESP, LSP, and EDP) and media on the growth kinetics (*λ*, *μ*max, and *Y*max) of *S.* Typhimurium DT104. The results of this study showed that the previous life cycle phase has a significant effect on the *λ*, *μ*max, and *Y*max and was also dependent upon the media. Overall, the *λ* increased as the age of the previous life cycle phase increased. EDP may have spent a longer time in the lag phase repairing damaged cells, taking up available and essential nutrients and replenishing fluidity in the cytoplasm.

The results of this study suggest that not only the previous growth pH and temperature should be considered when

### REFERENCES


developing growth models but also the previous life cycle phase. If the culture from the log phase has a shorter *λ*, studies that base their food growth model on foodborne pathogens from the stationary phase may be underestimated. Therefore, establishments that base their prerequisite programs or sanitation standard operating procedures on these models may be inadequate. Additional research on various serotypes of *Salmonella* is needed to prove this theory.

### AUTHOR CONTRIBUTIONS

SP and JH contributed to the conception and design of the study. BV and JS contributed to the design of the study. JU contributed to the cellular morphology study. TO contributed to growth kineties study. JH performed experiments and wrote the draft first draft of the manuscript. SP, TO, and JU wrote sections of the manuscript. SP was responsible for the integrity of the work and overall supervision. All authors contributed to interpretation of the data, manuscript revision, read, and approved the manuscript.

### FUNDING

This project was supported by the U.S. Department of Education, Title III, USDA Evans Allen and Perdue Farms, Inc.

### ACKNOWLEDGMENTS

We would like to thank Joan Meredith for technical assistance. Earlier version of this manuscript has been released as a pre-print at https://www.biorxiv.org/content/early/2018/05/21/326579 (Hawkins et al., 2018).


**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 Hawkins, Uknalis, Oscar, Schwarz, Vimini and Parveen. 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.*

# Building of Pressure-Assisted Ultra-High Temperature System and Its Inactivation of Bacterial Spores

Dong Liang<sup>1</sup> , Liang Zhang1,2, Xu Wang<sup>3</sup> , Pan Wang<sup>1</sup> , Xiaojun Liao1,2, Xiaomeng Wu1,2 , Fang Chen1,2 and Xiaosong Hu1,2 \*

<sup>1</sup> College of Food Science and Nutritional Engineering, National Engineering Research Centre for Fruits and Vegetables Processing, China Agricultural University, Beijing, China, <sup>2</sup> Key Laboratory of Fruits and Vegetables Processing, Ministry of Agriculture, Beijing, China, <sup>3</sup> College of Food Science, Northeast Agricultural University, Harbin, China

### Edited by:

Om V. Singh, TSG Consulting, a Science Group Company, United States

#### Reviewed by:

Hui Li, Chinese Academy of Agricultural Sciences, China Christopher Doona, Massachusetts Institute of Technology, United States

> \*Correspondence: Xiaosong Hu huxiaos123@126.com

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 28 August 2018 Accepted: 22 May 2019 Published: 10 June 2019

#### Citation:

Liang D, Zhang L, Wang X, Wang P, Liao X, Wu X, Chen F and Hu X (2019) Building of Pressure-Assisted Ultra-High Temperature System and Its Inactivation of Bacterial Spores. Front. Microbiol. 10:1275. doi: 10.3389/fmicb.2019.01275 The pressure-assisted ultra-high temperature (PAUHT) system was built by using soybean oil as pressure-transmitting medium, and the multiple regression equation of soybean oil temperature change (1TP) during pressurization as a function of initial temperature (T<sup>i</sup> ) and set pressure (P) was developed: 1T<sup>P</sup> = −13.45 + 0.46 T<sup>i</sup> + 0.0799 P − 0.0037 T 2 <sup>i</sup> <sup>−</sup> <sup>2</sup>.<sup>83</sup> <sup>×</sup> <sup>10</sup>−<sup>5</sup> <sup>P</sup> 2 . The fitted model indicated that the temperature of the system would achieve ≥121◦C at 600 MPa when the initial temperature of soybean oil was ≥84◦C. The PAUHT system could effectively inactivate spores of Bacillus subtilis 168 and Clostridium sporogenes PA3679 (less than 1 min). Treatment of 600 MPa and 121◦C with no holding time resulted in a 6.75 log reductions of B. subtilis 168 spores, while treatment of 700 MPa and 121◦C with pressure holding time of 20 s achieved more than 5 log reductions of C. sporogenes PA3679 spores. By comparing the PAUHT treatment with high pressure or thermal treatment alone, and also studying the effect of compression on spore inactivation during PAUHT treatment, the inactivation mechanism was further discussed and could be concluded as follows: both B. subtilis 168 and C. sporogenes PA3679 spores were triggered to germinate firstly by high pressure, which was enhanced by increased temperature, then the germinated spores were inactivated by heat.

Keywords: pressure-assisted ultra-high temperature, mathematical model, bacterial spores, sterilization, inactivation mechanism

### INTRODUCTION

Pressure assisted thermal sterilization (PATS), a technique combining high pressure with elevated temperature, has been used to effectively inactivate bacterial spores in low-acid food (Peleg et al., 2008; Nguyen et al., 2010; Olivier et al., 2011). Adiabatic compression caused by pressurization results in quick heating of the product to process temperatures, and the subsequent decompression results in its rapid cooling. Therefore, PATS can be conducted in a short time based on quick heating/cooling (Paidhungat et al., 2002; Ardia et al., 2004; Zhu et al., 2008; Reineke et al., 2013c). The overall treatment conditions are less severe than with conventional sterilization (Reineke et al., 2011; Doona et al., 2016a). All these attributes can lead to a better and healthier product, because

unwanted reactions, like the Maillard reaction which are temperature and time dependent, might not occur or can be reduced by high pressure.

Generally, when the treatment was pressure of P = 600– 800 MPa and initial temperature of T<sup>i</sup> = 90–120◦C, all spores could be inactivated (Margosch et al., 2006; Wilson et al., 2008; Lopes et al., 2018). However, studies on PATS usually use water as pressure-transmitting medium and the final temperature is usually less than 121◦C after compression heating, and it still requires several minutes (> 3 min) to inactivate all spores (Margosch et al., 2004b; Ramaswamy and Balasubramaniam, 2007; Shao and Ramaswamy, 2008; Doona et al., 2017). In order to achieve higher ration of spore inactivation in shorter time, higher pressure and temperature could be a good choice for the PATS processing (Shao and Ramaswamy, 2008; Wang et al., 2017). Nowadays, studies on spore inactivation by PATS with final temperature more than 121◦C are relatively limited. In order to achieve higher temperature during PATS treatment, we supposed that the soybean oil could be used as the pressure-transmitting medium because of its higher compression heating coefficient (9.2◦C/100 MPa) (Patazca et al., 2007; Zhu et al., 2008). This PATS system with higher working temperature could be defined as pressure-assistant ultra-high temperature (PAUHT) system. This system is rarely built and its inactivation effect and mechanism of bacterial spores has never been investigated.

The mechanism of spore inactivation by PATS has been extensively studied (Mathys et al., 2009; Ramaswamy et al., 2013; Reddy et al., 2016). Generally, it is proposed that spores are firstly induced to germinate and lose resistance, and then inactivated by heat. However, spore germination induced by PATS is different due to different treatment conditions and species of spores. For spores of Bacillusspecies, the germination is different between moderate pressure (200–500 MPa) and higher pressure (> 500 MPa) (Reineke et al., 2013b). At moderate pressure, spores are induced to germinate by activating the germinant receptors (GRs) and ensuing the release of DPA (Paidhungat et al., 2002; Setlow, 2003; Vepachedu and Setlow, 2007; Doona et al., 2014). DPA release activates one of the cortex lytic enzymes (CLEs), CwlJ, and induces spore core partial hydration, which may activate another CLEs, SleB (Setlow et al., 2009; Paredes-Sabja et al., 2011). These two activated CLEs degrade the cortex, resulting in completion of spore germination (Li et al., 2013). However, the higher pressure (> 500 MPa) induced germination was triggered by directly opening the DPA channel, SpoVA, without activating the GRs (Elke and Wuytack, 2001; Vercammen et al., 2012; Reineke et al., 2013c; Sevenich et al., 2013; Kong et al., 2014; Sarker et al., 2015). Margosch and others proposed a spore inactivation mechanism by PATS that is not involved spore germination. These researchers suggest that the inactivation mechanism of B. subtilis and Bacillus licheniformis spores by PATS at temperature higher than 70◦C follows a two-stage strategy: (i) DPA is released by a short HHP pulse at high temperature and (ii) then spores are thermally inactivated-independent of pressure level upon depressurization (Margosch et al., 2004a). For Clostridial spores, Doona and others reported Clostridium perfringens spores with GRs exhibit similar germination-inactivation mechanism to spores of Bacillus species when treated by PATS (Doona et al., 2016a,b). However, for Clostridium difficile spores without GRs, spores were not germinated by pressure of 150 MPa. In contrast, the pressure of 550 MPa induced spores to release DPA, but the spores did not complete germination and remained heat resistant. As for C. sporogenes spores with GRs similar to C. perfringens spores (Setlow et al., 2017), they can theoretically germinate during PATS treatment and lose heat resistance, but it has never been verified.

The objectives of this study were to (1) build the PAUHT system with soybean oil as pressure-transmitting medium, (2) investigate the inactivation of PAUHT on B. subtilis 168 and C. sporogenes PA3679 spores, and (3) study the inactivation mechanism of spores under the condition of PAUHT.

### MATERIALS AND METHODS

### Bacterial Strains and Preparation of Spore Suspensions

The bacterial strains used in the study were B. subtilis 168, FB85 (without GRs, a derivative of strain 168) (Paidhungat and Setlow, 2000) and C. sporogenes PA3679. The B. subtilis 168 and C. sporogenes PA3679 were obtained from China General Microbiology Culture Collection Center and China Center of Industrial Culture Collection, respectively. FB85 was a gift from Prof. Peter Setlow. B. subtilis 168 and FB85 strains were grown in Luria Bertani broth medium. Spores were prepared at 37◦C by plating aliquots of 0.2 mL from the fresh overnight culture on the 2 × Schaeffer's glucose medium agar plates without antibiotics as described previously (Paidhungat and Setlow, 2000; Luu et al., 2015). The spores were harvested after 2 days of incubation when over 90% of the spores were released from the mother cell as observed by phase contrast microscopy (Axio Observer. A1, Carl Zeiss, Germany). The spore suspension was cleaned by repeated centrifugation (8000 × g, 10 min, 4◦C) for at least five times with cold distilled water (4◦C). Then, the spores were cleaned using the histodenz gradient centrifugation and washed three times with sterilized water. Finally, the spore suspensions were stored in the dark at 4◦C and were 98% free of growing or sporulating cells, germinated spores, and cell debris as shown in **Supplementary Figure S1** observed by phase-contrast microscope (Setlow, 2018).

Clostridium sporogenes PA3679 was grown anaerobically in the Reinforce Clostridial Medium (RCM) broth in anaerobic jars with an Oxoid GasPak sachet (Oxoid, Basingstoke, United Kingdom) for 2 days at 37◦C. An aliquot (0.2 mL) of C. sporogenes PA3679 culture was spread on the RCM plate and incubated at 37◦C anaerobically for more than 5 days. The spores were harvested when over 90% of the spores were released from the mother cell observed by phase contrast microscopy. Then, the spore suspension was cleaned by centrifugation same as spores of B. subtilis strains. Finally, the spores were suspended in the sterilized water and stored at 4◦C for the subsequently use. The concentration of the spore suspensions was adjusted to approximate 10<sup>8</sup> CFU/mL before treatments.

### High Pressure Equipment

fmicb-10-01275 June 10, 2019 Time: 13:0 # 3

High pressure treatment was carried out with an experimental setup as show in **Figure 1**. The HP device (FPG 7100, Stansted Fluid Power Ltd., Essex, United Kingdom) had a 2 L cylindrical pressure chamber (70 mm in diameter and 500 mm in height) with a maximum working pressure of 900 MPa. The pressure chamber was surrounded by a heating sleeve and the maximum temperature was 110◦C. A data logger connected with T-type thermocouples (Omega Engineering, CT, United States) was used to gather the in situ pressure and temperature data. The medium used for pressure transmission in the system was 30% propylene glycol.

### HHP Treatment

A polyoxymethylene (POM) plastic insulated chamber (18 mm inner diameter, 50 mm outer diameter, and 200 mm height) was used to hold soybean oil and the sample during the pressure treatment to provide the temperature stability as described elsewhere (Shao and Ramaswamy, 2008). The soybean oil was filled into the insulator and used as the pressurization medium to get desired temperature by compression heating.

Two mL of spore suspensions were packed in the sterilized polyamides/polyethylene plastic bags (4 cm × 4 cm) by heat sealer. The insulator and spore sample were preheated to the initial temperature prior HHP treatment. HHP treatment was carried out immediately after the spore was placed into the insulator. The compression rate was ∼350 MPa/min, and the depressurization time was less than 5 s at all pressure levels. Different pressure levels (600 and 700 MPa) combined with temperatures (121, 125, and 130◦C) were set for the treatments. The treated samples were then cooled in an ice bath and immediately analyzed.

### Thermal Treatment

Capillary tubes (0.9 mm inner diameter, 1 mm diameter, and 100 mm length) were used for the thermal treatments at 121, 125, and 130◦C in the methyl silicone oil bath by a heating circulator. Twenty µL of spore suspension was injected into the tubes by micropipette and sealed by flame as described elsewhere (Sapru et al., 1992). After heating, capillary tubes were cooled immediately in ice water, cleaned with 70% ethyl alcohol for 10 min, and washed with sterilized water. The spore suspension was flushed out with a pipette and analyzed.

### The Measurement of DPA Release

After different treatments, the DPA release was determined by measuring the fluorescence of DPA with Tb3<sup>+</sup> using a multi-well 96 fluorescence plate reader (Spark 10 M, Tecan) as described elsewhere (Yi and Setlow, 2010). After treatment, 50 µL spore suspension were mixed in the HEPES-TbCl<sup>3</sup> solution, which consisted of 50 mM HEPES (pH 7.4) buffer and 200 µM TbCl3, the volume was made to 200 µL. Analytic grade water was

used for the preparation of the buffer solution. DPA release was determined by the measurement of fluorescence emission at 545 nm with excitation at 270 nm. The maximum amount of DPA in the spore suspensions was determined by a thermal treatment for 20 min at 121◦C.

### Determination of Cell Count

Bacillus subtilis and C. sporogenes spore counts were determined by pour-plate enumeration in duplicate, using the nutrient agar and RCM agar, respectively. One milliliter dilution of spore suspension was added into each plate. The B. subtilis plate was incubated aerobically at 37◦C for 24 h, while the C. sporogenes plate was incubated anaerobically at 37◦C for 36 h before enumeration. The untreated samples were used as a control for each experiment in order to obtain the initial spore counts. The logarithm of survivors [log<sup>10</sup> (N0/Nt)] was used as the spore reduction after different treatments. N<sup>0</sup> and N<sup>t</sup> were the spore counts before and after treatment, respectively. The initial concentration of the spore suspensions was adjusted to approximate 10<sup>8</sup> CFU/mL before treatments.

### Microscope Analysis

After different treatments, the spore suspensions (OD<sup>600</sup> = 1.0) were double stained with 0.5 µM SYTO16 (Invitrogen, Carlsbad, CA, United States) and 15 µM propidium iodide (PI) (Invitrogen, Carlsbad, CA, United States). Suspensions were vigorously mixed and incubated in the dark for 15 min prior to analysis as described previously (Reineke et al., 2013a). Both fluorescent dyes are able to stain DNA, the membrane permeant SYTO 16 acts as an indicator for spore germination, whereas the membrane impermeant PI indicates membrane damage (Mathys et al., 2009). The spores were observed by phase contrast and fluorescence microscopy (Axio Observer. A1, Carl Zeiss, Germany).

### Statistical Analysis

All experiments were carried out in triplicate and data were presented as mean ± standard deviation. The oneway analysis of variance was used to test the statistically significant differences (P < 0.05) between treatments using software SPSS 17 for windows (SPSS Statistical Software, Inc., Chicago, IL, United States). Software Origin 7.5 (Origin Lab, MA, United States) was used for making plots and fitting mathematical model.

### RESULTS AND DISCUSSION

### The Pressure-Temperature Profile of the PAUHT System

Under compression during HHP treatment, the increased internal energy of the system resulted in a rapid rise in temperature (Ramaswamy and Balasubramaniam, 2007; Shao et al., 2010). In this PAUHT system, soybean oil was used as the pressure-transmitting medium to get the desired ultrahigh temperature by increasing the initial temperature with compression heating. **Figure 2** showed a typical temperature change of the soybean oil with different initial temperature during PAUHT treatment. When the initial temperature of soybean oil was set at 86◦C, the final temperature reached 121◦C with pressure of 600 MPa (**Figure 2A**). Under 700 MPa (**Figure 2B**), the final temperature of the medium reached to 130◦C when the initial temperature was set at 90◦C. The temperature increase (30–40◦C) of soybean oil during PAUHT treatment was much more than that (11–16◦C) of the 30% propylene glycol, indicating that the soybean oil has a better compression heating behavior than the 30% propylene glycol, which was consistent with the results previously reported (Houška et al., 2004; Min et al., 2010). Moreover, during the pressure holding time, the temperature of soybean oil remains almost unchanged in the insulator, indicating the POM material can be used for the insulator to remain the temperature stable. All of the results above suggested that the PAUHT constructed with soybean oil as pressure-transmitting medium has the capability to reach ultra-high temperature up to 130◦C with excellent stability during the process, hence it could be used as a reliable tool for the follow-up studies.

In order to predict the temperature change (1TP) of soybean oil during PAUHT treatment, 100 data points from the measurements were pooled to develop a multiple regression equation of 1T<sup>P</sup> of soybean oil during pressurization as a function of initial temperature (Ti) of soybean oil and set pressure (P). It could be expressed as a quadratic function of P and T<sup>i</sup> as follows:

$$
\Delta T\_P = -14.09 + 0.47 \ T\_i + 0.082 \ P - 3.1 \times 10^{-5} T\_i P
$$

$$
$$

(R 2 adj = 0.993, p < 0.05 for all items).

Where T<sup>i</sup> = initial temperature (50◦C < T<sup>i</sup> < 90◦C), P = pressure (100 MPa < P < 700 MPa), T<sup>p</sup> = target temperature ( ◦C), 1T<sup>P</sup> = temperature change (◦C), R 2 adj = adjusted R square, p = statistic parameter of significance.

When desired PAUHT conditions were given, the initial temperature of soybean oil could also be calculated by the following equation:

$$T\_i = 18.22 + 0.44 \ T\_P - 0.05 \ P - 4.61 \times 10^{-4} \ T\_P P$$
 
$$+ 0.0042 \ T\_P^2 + 4.25 \times 10^{-5} \ P^2 \tag{2}$$

(R 2 adj = 0.993, p < 0.05 for all items).

The parameters of these equations were determined by enter regression procedure. The adjusted regression coefficient (R 2 adj) value was used to measure the performance of the proposed model. Eq. (2) can be used to calculate the initial temperature under different combinations of T<sup>p</sup> and P-values.

The used conditions were listed in **Table 1**. The high temperature of 121◦C can be achieved under PAUHT at 600 MPa commonly used in industry, as long as the initial temperature of soybean oil was 84◦C. The verification experiment was done to test the proposed model. When the initial temperature was 84◦C at 600 MPa, the final temperature of soybean oil can reach to 120.86◦C, which was within the range of the confidence interval.

As a result, this equation can be used to predict the initial temperature of soybean oil for the PAUHT treatments.

### The Effect of PAUHT on Spore Inactivation

The inactivation of B. subtilis 168 and C. sporogenes PA3679 spores under different PAUHT treatments were shown in **Table 2**. The treatment of 600 MPa at 121◦C without pressure holding time resulted in 6.75 log and 2.56 log reductions of B. subtilis 168 and C. sporogenes PA 3679 spores, respectively, indicating that spores of C. sporogenes PA 3679 were more resistant to pressure and temperature than spores of B. subtilis 168, as reported previously (Wilson et al., 2008; Ramaswamy et al., 2013). Moreover, when the temperature increased to 125◦C or hold time up to 1 min, the amount of B. subtilis 168 spores was below the detection line, and the inactivation of C. sporogenes PA3679 spores was also increased. These results indicated the PAUHT treatments could effectively inactivate B. subtilis 168 and C. sporogenes PA3679 spores with less than 1 min of holding time. During pressure holding time for 1 min, the treatment of 600 MPa at 121◦C resulted in 3.58 log reductions of C. sporogenes PA3679 spores. Increasing the pressure to 700 MPa, the treatment at 121◦C for 1 min of holding time resulted in 6.31 log destructions of C. sporogenes PA3679 spores.


Moreover, increasing the temperature to 125◦C, the treatment of 600 MPa for 1 min of holding time induced 8.92 log C. sporogenes PA3679 spore inactivation. These results indicated that increase of pressure and temperature significantly accelerated the spore inactivation during pressure holding time of PAUHT treatments, which was consistent with previous report (Ahn et al., 2007). In addition, during PAUHT treatment, a compression of 120– 140 s (∼5 MPa/s) was needed to arrive the final pressure of P = 600–700 MPa (**Figure 2**). Even without pressure holding time, 6.75 log and 2.56 log reductions of B. subtilis 168 and C. sporogenes PA3679 spores were achieved at 121◦C and 600 MPa, respectively. Moreover, the treatment of 700 MPa at 121◦C without pressure holding time resulted in 8.85 log (under detection line) and 4.53 log destructions of B. subtilis 168 and C. sporogenes PA3679 spores, respectively (**Table 2**). These results indicated that the compression played an important role in spore inactivation and should be taken into consideration during PAUHT treatments. Notably, the treatment of 700 MPa for 20 s at 121◦C resulted in more than 5 log inactivation of C. sporogenes PA3679 spores. A 5 log reduction is regarded as the commercial sterilization standard for C. sporogenes PA3679 spores by food industry and regulatory agencies due to the fact that C. sporogenes spores are more heat resistant than C. botulinum spores (USFDA-FSIS, 2005; Ramaswamy et al., 2013; Reddy et al., 2016). Moreover, thermal sterilization required a process at 121.1◦C for over 2.52 min to reach the goal of commercial sterilization (Theron and Irving, 1977), while using the PAUHT processing (700 MPa, 121◦C) proposed here a processing time of 20 s was needed for the same effect, thus the PAUHT processing was more time efficient than thermal processing. However, although PAUHT treatment at temperature higher than 121◦C could be more efficient at inactivation spores (**Table 2**, 600 MPa/125◦C/0 s, 700 MPa/130◦C/0 s), this higher temperatures would also produce unwanted sidereactions and degrades food quality (Lau and Turek, 2007). Hence, the following studies were mainly focused on the PAUHT treatment at 121◦C.

TABLE 2 | The inactivation and DPA release of B. subtilis 168 and C. sporogenes PA 3679 spores after different treatments.


The initial total colony of B. subtilis spores was: 7.15 × 10<sup>8</sup> CFU/mL; the initial total colony of C. sporogenes spores was: 8.23 × 10<sup>8</sup> CFU/mL. 0 s represents decompression immediately after pressurization. "−" Represents spore counts were under the detection limit.

### The Inactivation Mechanism of Spore by PAUHT

Nowadays, the discussion about the mechanism of spore inactivation by PATS is mainly focused on whether spores undergo germination during inactivation by PATS treatment (Reineke et al., 2013b). Similarly, whether spores undergo germination during PAUHT treatment is the key point of figuring out how PAUHT inactivates them. In order to understand the mechanism of spore inactivation during PAUHT treatments, the structure changes of the B. subtilis 168 and C. sporogenes PA3679 spores during the process were investigated by staining the spores with PI and SYTO 16 after treatment, followed by observation using phase contrast microscopy and fluorescence microscopy. As shown in **Figure 3**, after PAUHT process, both B. subtilis 168 and C. sporogenes PA3679 spores turned from bright to gray, and could be stained by PI and SYTO 16 (**Figures 3C,F**). As PI and SYTO 16 staining were usually considered as the indicator of the integrity of spore's IM and cortex (Mathys et al., 2007; Kong et al., 2010), respectively, we could ensure that the PAUHT treated spores displayed damaged IM and cortex. Moreover, more than 90% DPA released after the PAUHT treatment (**Table 2**), further confirming that the IM was damaged (Reineke et al., 2013a). However, we noticed that PAUHT treated spores turned gray rather than dark, indicating that the cores of these spores were only partially hydrated (Katja et al., 2013). Since the cortex played a key role in maintaining the core rehydration and germinated spores would degrade their cortex and turn dark (**Figures 3B,E**), it could be concluded that the cortex of PAUHT treated spores were only partially damaged. From these results, we could conclude that PAUHT-inactivated spores had damaged IM and partially damaged cortex, but we couldn't tell whether spores went through germination because (1) heat-inactivated spores had a damaged IM and partially damaged cortex similar to PAUHTinactivated spores, but didn't go through germination (Setlow, 2006), and (2) spores that went through stage I of germination but didn't accomplish the stage II of germination could be

inactivated by PAUHT, and also exhibited damaged IM and partially damaged cortex.

Since the temperature during PAUHT was relatively high and it could play an important role in inactivation spores, we investigated the effect of heat treatment at temperatures same to PAUHT on spore inactivation. B. subtilis 168 and C. sporogenes PA3679 spores were treated at ambient pressure at 121◦C for different times (10, 20, and 60 s), then the spores were stained by PI and SYTO 16 and observed by microscopy as described above. Thermal treatment of 121◦C at 0.1 MPa for 1 min could induce 5.23 log and 2.51 log inactivation and DPA release (more than 90%) of B. subtilis 168 and C. sporogenes PA

3679 spores, respectively (**Figure 4**), indicating the temperature played an important role in spore inactivation during pressure holding time of PAUHT treatment. Moreover, compared to the untreated (**Figures 5A,D**) and germinated spores (**Figures 5C,F**), these heat treated spores were stained by both PI and SYTO 16 (**Figures 5B,E**), indicating the spore's IM and cortex were damaged as previously studied (Setlow et al., 2008). Besides, the thermal treated spores turned from bright (**Figures 5A,D**) to gray (**Figures 5B,E**) rather than dark (**Figures 5C,F**) in the phase contrast microscopy, indicating the spore core was partially rehydrated, which is probably due to partially damage of cortex (Luu-Thi et al., 2015). These results were similar with the PAUHT treated spores, but we still could not confirm that the spores whether germinated during PAUHT treatment, because more than 90% DPA of these spores was released (**Table 2**), which could also be the results from germination by HP during PAUHT treatment.

In order to further investigate whether spores underwent germination during PAUHT treatment, we investigated the effect of pressure treatment alone on spore germination. The pressure treatment was performed at 600 MPa at ambient temperature for 1 min of holding time. The DPA release was determined after high pressure treatment and used as the indicator for spore germination as other reports (Yi and Setlow, 2010). As shown in **Figure 6A**, pressure at 600 MPa for 1 min could not directly inactivate the B. subtilis 168 and FB 85 spores (data was not shown), but it could induce more than 25% DPA to release, similar as previous studies (Vercammen et al., 2012; Sarker et al., 2015), indicating spores were induced to germinate by HP of 600 MPa. Since FB85 spores has no GRs (Paidhungat et al., 2002), thus germination under pressure of 600 MPa were likely triggered by opening the DPA channel without activating the GRs, as reported previously (Black et al., 2005; Doona et al., 2014). For C. sporogenes PA3679 spores. Pressure at 600 MPa

used as control.

for 1 min couldn't induce spores to release DPA (**Figure 6A**), indicating C. sporogenes PA3679 spores could not germinate at this pressure. However, in the presence of SpoVA channels, the treatment of 600 MPa can theoretically induce C. sporogenes PA3679 spores to germinate similar to the C. perfringens spores (Paredes-Sabja et al., 2011; Doona et al., 2016b). It has been reported that the HP of 550 MPa triggered the C. perfringens spores to release their DPA and germinate, which happened after the cortex hydrolysis caused by activating the CLEs, SleC (Paredes-Sabja et al., 2009; Doona et al., 2016b). The C. sporogenes spores does not have the CSP and SleC (Paredes-Sabja et al., 2011; Doona et al., 2014), which are nececery for the cortex degradation of C. perfringens spores. Hence, the lack of CSP and SleC was probably the reason why HP at ambient temperature could not induce C. sporogenes PA 3679 spores to release DPA and germinate. In addition, previous reports indiccated that spore germination could be potentiated or activated by heat activation (Luu et al., 2015), hence we further investegate the effect of heat activation on spore germination triggered by HP (**Figure 6A**). Both B. subtilis 168 and C. sporogenes PA3679 spores were heat activated at 75◦C for 15 min prior HP treatment of 600 MPa at atmosphere for 1 min. for B. subtilis 168 spores, heat activated spores released DPA (68%) more than unactivated spores (26%) (**Figure 6A**), indicating HP triggered germination was enhanced by heat activation as previous reports (Luu et al., 2015; Doona et al., 2016b). Notably, for C. sporogenes PA3679 spores, after the heat activation (75◦C, 15 min), more than 20% DPA was released (**Figure 6A**), indicating the C. sporogenes PA3679 spores germinated after the HP of 600 MPa for 1 min. The effect of heat

activation on C. sporogenes PA3679 spore germination by HP was surprising, since this has not been reported previously.

(A) or PAUHT treatment of 600 MPa at 121◦C without pressure holding time (B).

Considering the spore germination triggered by HP could be enhanced by heat activation, we further investigate the effect of heat activation on spore inactivation during PAUHT treatment. Both B. subtilis 168 and C. sporogenes PA3679 spores were heat activated at 75◦C for 15 min prior PAUHT treatment of 600 MPa at 121◦C without pressure holding time. After PAUHT treatment, the counts of spore inactivation were determined, as shown in **Figure 6B**. For B. subtilis 168 after heat activation, 6.83 log spores were inactivated by the compression during PAUHT treatment of 600 MPa at 121◦C, which was comparative to unactivated spores (6.75 log inactivation), indicating that the heat activation has no effect on inactivation of B. subtilis 168 spores by PAUHT treatment. It could be attributed to the lack of GRs in B. subtilis 168 spores, since heat activation acted primarily on GRs (Luu et al., 2015). However, for C. sporogenes PA3679, 3.56 log of heat activated spores were inactivated by the same treatment, which was higher than unactivated spores (2.56 log inactivation), indicating that heat activation could significantly enhance the spore inactivation during compression of PAUHT treatment. These observations suggested that heat activation play different role on inactivation of different spores during compression. For C. sporogenes PA3679 spores, since the heat activation could not inactivate spores, but could enhance the spore germination by HP. Therefore, more heat activated spores were inactivated during compression, possibly because the HP triggered germination was enhanced by heat activation. Hence, the C. sporogenes PA3679 spores were probably triggered to germinate and then inactivated during compression of PAUHT treatment.

In order to further investigate whether spores undergo germination during compression of PAUHT treatment, the B. subtilis 168 and C. sporogenes PA3679 spores were treated with various conditions of T<sup>i</sup> = 65–84◦C and high pressure P = 500–600 MPa during compression (hold time = 0 s). At the completion of the compression (100–120 s), the pressure was released, samples were withdrawn from the HPP chamber, cooled on ice, and then recovered on LB and RCM agar. As shown in **Table 3**. During compression of the PAUHT treatment of 600 MPa at 121◦C (T<sup>i</sup> = 84 and P = 600 MPa), no increase of B. subtilis 168 and C. sporogenes PA3679 counts was discernible as above noted (**Table 2**), indicating the spores were not activated during compression of PAUHT treatment. Similar results were obtained for B. subtilis 168 spores with T<sup>i</sup> = 65–75◦C and P = 500– 600 MPa. However, for the C. sporogenes PA3679 spores, under the conditions of T<sup>i</sup> = 65◦C and P = 600 MPa, the viable counts increased after the compression (98% increase), indicating the C. sporogenes PA3679 spores were activated after compression. At T<sup>i</sup> = 75◦C and P = 600 MPa, a slight decrease (27%) in spore counts after the compression suggested that inactivation

TABLE 3 | Activation and germination and of B. subtilis 168 and C. sporogenes PA3679 spores during compression.


Activation of spores was measured after compression, as increases in viable counts before and after compression. "<sup>a</sup> " represents spore germination. After compression, spores were subsequently treated by wet heat (80◦C for 20 min) and then counted. The increase of viable counts before and after the wet heat treatment represents the germinated spores. All viability measurements are mean plate counts in triplicate.

was preceded by activation. Hence, during compression of the PAUHT treatment (T<sup>i</sup> = 84◦C and P = 600 MPa), the C. sporogenes PA3679 spores may also pass through the process of activation and inactivation.

The effect of compression (T<sup>i</sup> = 65–84◦C, P = 500–600 MPa) on the germination of B. subtilis 168 and C. sporogenes PA3679 spores were further investigated. After the completion of compression, spores were treated by wet heat (80◦C, 20 min) then enumerated. As shown in **Table 3**. At T<sup>i</sup> = 75◦C and P = 500 MPa or T<sup>i</sup> = 65◦C and P = 600 MPa, for B. subtilis 168, more than 90% spores were inactivated by subsequent wet heat treatment, indicating the spores were induced to germinate by compression. At T<sup>i</sup> = 75◦C and P = 600 MPa, the rate of germination decreased to 53% after compression, indicating some population of germinated spores were inactivated during compression. Similar results were obtained for C. sporogenes PA3679 spores. Hence, during compression of PAUHT treatment (T<sup>i</sup> = 84◦C and P = 600 MPa), both B. subtilis 168 and C. sporogenes PA3679 spores possibly underwent the process of germination and inactivation.

As a consequence, during the PAUHT treatment, the compression and pressure holding process played important role in spore inactivation. The mechanism of spore inactivation during PAUHT treatment could be concluded as follows: for B. subtilis 168, spores were firstly induced to germinate under HP, which was enhanced by increased temperature. Then the germinated spores were inactivated by heat as previous reports (Luu et al., 2015; Doona et al., 2016a, 2017). For C. sporogenes PA3679, spores were likely activated during the compression time of PAUHT, and then the activated spores were induced to germination by HP, followed by inactivation by heat. In addition, the PAUHT inactivated spores showed gray rather than dark observed by phase contrast microscopy, indicating that these spores only undergo the first stage of germination (DPA was released and spores lost partial resistance) and the cortex degradation was blocked, which was possibly due to the inactivation of CLEs by heat or pressure (Reineke et al., 2011).

### CONCLUSION

In this work, PAUHT system was established by using soybean oil as adiabatic medium, and a regression model was developed

### REFERENCES


to predict the initial temperature of soybean oil for PAUHT. The B. subtilis 168 and C. sporogenes PA3679 spores were inactivated during PAUHT treatment in short time (< 1 min). The inactivation mechanism of B. subtilis 168 and C. sporogenes PA 3679 spores by PAUHT treatment could be supposed as follows: during PAUHT treatment, spores were firstly triggered to germinate by HP, with completing stage I of germination, and then the germinated spores were inactivated by heat.

### INDUSTRIAL RELEVANCE

The inactivation of bacterial spores in the low-acid canned food remains the most serious problem in food sterilization because of the high resistance of spores to high pressure and thermal processing. In this work, we showed that the proposed PAUHT system with soybean oil as adiabatic medium was effective for inactivating bacterial spores in short time (<1 min), therefore it has the potential to be a promising alternative technique for food sterilization. Further understanding of spore inactivation mechanism by PAUHT treatment will help to make this technique available for studies in pilot and production scale.

### AUTHOR CONTRIBUTIONS

DL carried out the experiments and wrote the manuscript. LZ, XW, and PW gave the advice and assistance during the experiments. XL and FC reviewed the manuscript and gave the advice on the manuscript. XMW revised the manuscript. XH designed the experiments and reviewed the manuscript.

### FUNDING

This work was supported by Key Project of National Natural Science Foundation of China (NSFC) (No. 31530058).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.01275/full#supplementary-material




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

Copyright © 2019 Liang, Zhang, Wang, Wang, Liao, Wu, Chen and Hu. 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.

# Enhanced Efficacy of Peroxyacetic Acid Against *Listeria monocytogenes* on Fresh Apples at Elevated Temperature

*Xiaoye Shen1† , Lina Sheng1† , Hui Gao1,2 , Ines Hanrahan3 , Trevor V. Suslow4 and Mei-Jun Zhu1 \**

*1 School of Food Science, Washington State University, Pullman, WA, United States, 2 Department of Food Science, Zhengzhou University of Light Industry, Henan, China, 3 Washington Tree Fruit Research Commission, Wenatchee, WA, United States, 4 Department of Plant Sciences, University of California, Davis, Davis, CA, United States*

#### *Edited by:*

*Learn-Han Lee, Monash University Malaysia, Malaysia*

#### *Reviewed by:*

*Sunil D. Saroj, Symbiosis International University, India Xuming Deng, Jilin University, China*

> *\*Correspondence: Mei-Jun Zhu meijun.zhu@wsu.edu*

*† These authors have contributed equally to this work*

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 01 December 2018 Accepted: 13 May 2019 Published: 18 June 2019*

#### *Citation:*

*Shen X, Sheng L, Gao H, Hanrahan I, Suslow TV and Zhu M-J (2019) Enhanced Efficacy of Peroxyacetic Acid Against Listeria monocytogenes on Fresh Apples at Elevated Temperature. Front. Microbiol. 10:1196. doi: 10.3389/fmicb.2019.01196*

Peroxyacetic acid (PAA) is the most commonly used antimicrobial in spray bar antimicrobial treatment during fresh apple packing and processing. However, there are limited data regarding its practical efficacy against *Listeria monocytogenes* on fresh apples. This study evaluated the antimicrobial activity of PAA against *L. monocytogenes* on fresh apples applicable to current industry practice, and further examined practical parameters impacting its efficacy to maximize the biocidal effects. Apples were inoculated with a three-strain *L. monocytogenes* cocktail at ~6.0 Log10 CFU/apple and then subjected to comparative antimicrobial treatments after 48 h post-inoculation. An 80 ppm PAA treatment, at 30-s and 2-min exposure, reduced *L. monocytogenes* on fresh apples by ~1.3 or 1.7 Log10 CFU/apple, respectively. The anti-*Listeria* efficacy of PAA was not affected by the water hardness and pH of PAA solution, while it improved dramatically when applied at elevated temperature. A 2-min exposure of 80 ppm PAA at 43 and 46°C resulted in a 2.3 and 2.6 Log10 CFU/apple reduction, respectively. A 30-s contact time of 80 ppm PAA at 43–46°C reduced *L. monocytogenes* on apples by 2.2–2.4 Log10 CFU/apple. Similarly, PAA intervention at elevated temperatures significantly strengthened its effectiveness against naturally occurring apple microbiota. PAA treatment at 43–46°C can provide a vital method to improve antimicrobial efficacy against both *L. monocytogenes* and indigenous microbiota on fresh apples. Our data provide valuable information and reference points for the apple industry to further validate or verify process controls.

Keywords: apples, *Listeria monocytogenes*, peroxyacetic acid, antimicrobial, temperature

### INTRODUCTION

Apples are the second most commonly consumed fruit in the United States (US) and produced on more than 325,000 acres, yielding 33 billion apples annually (USAA, 2018). The average apple consumption in the United States is 22 kg per person annually and the major commercial varieties are Granny Smith, Fuji, Gala, and Red Delicious (USAA, 2017). The recent outbreak of *Listeria monocytogenes* linked to caramel apples (Angelo et al., 2017)

**344**

and multiple *L. monocytogenes* recalls associated with apple products (FDA, 2015, 2016, 2017b) have brought critical concerns to the apple industry and the general public regarding control of this pathogen on fresh apple fruit surfaces during production, storage, and packing. *L. monocytogenes* is an important foodborne pathogen that causes ~1,600 hospitalization and ~260 deaths in the US annually (FDA, 2012). It can survive on fresh apple surfaces for an extended period during cold storage (Sheng et al., 2017). If contaminated apples are used in the confectionary industry, *L. monocytogenes* proliferates in the microenvironment created between the apple surface and caramel coating layer (Glass et al., 2015).

During postharvest processing and handling, antimicrobial interventions have long been employed to reduce foodborne pathogens on apples and prevent or minimize crosscontamination during wash-processing. Chlorine is the most widely used antimicrobial in the fresh produce industry, but has limited efficacy against *L. monocytogenes* at the commonly used dose range of 50–200 ppm at 1–2 min exposure (Abadias et al., 2008). Chlorine wash at 100 ppm for ~1 min only provided ~1 log reduction of *L. monocytogenes* on apples (Beuchat et al., 1998; Rodgers et al., 2004). In addition, chlorine reacts with organic matter to form carcinogenic trihalomethanes (Brown et al., 2011) and chlorates (Gil et al., 2016), giving rise to health concerns. Therefore, the fresh produce industry is actively looking for alternative chemicals and/or intervention methods with a higher antimicrobial efficacy and a lower reactivity with organic matter.

Peroxyacetic acid (PAA) is generated as an equilibrium mixture between acetic acid and hydrogen peroxide in aqueous solution, and has a strong oxidation potential of 1.81 electronic volts (Dell'Erba et al., 2007; Carrasco and Urrestarazu, 2010; Hua et al., 2011). It is approved to be used at 80 ppm as a wash water processing aid on fresh produce without further rinse requirement (FDA, 2017a). PAA has a relatively low reactivity with organic matter, compared with chlorine (Banach et al., 2015), and the formed by-products have little or no toxicity (Monarca et al., 2002a,b). In addition, PAA decomposes to harmless acetic acid and oxygen (Gehr et al., 2003). PAA has been used in the fresh produce industry to control microbial contamination in iceberg lettuce, mung bean sprouts, cantaloupe, and others (Gonzalez et al., 2004; Hellstrom et al., 2006; Wang et al., 2006; Neo et al., 2013). PAA is currently the most commonly used antimicrobial in spray bar rinse-treatment during fresh apple packing and processing according to our survey of apple packers in Washington. In spite of its popularity, however, the sparse information available indicates that PAA has a limited efficacy against *L. monocytogenes* on fresh apples. PAA at 80 ppm, applied for 80-s contact time, resulted in ~1.0 Log reduction of *L. monocytogenes* on Golden Delicious apples (Rodgers et al., 2004). A 1-min treatment of apple plugs with 80 ppm PAA resulted in ~0.8 Log10 CFU/plug reduction of *L. monocytogenes* (Abadias et al., 2011).

The objectives of this study were to evaluate the antimicrobial efficacy of PAA against *L. monocytogenes* and resident microorganisms on fresh apples, and further optimize parameters that are applicable to the fresh apple industry to maximize its antimicrobial efficacy.

### MATERIALS AND METHODS

### Bacteria Strains

*L. monocytogenes* strains [NRRL B-57618 (1/2a), NRRL-33466 (1/2b) and NRRL B-33053 (4b)] were obtained from USDA-ARS culture collection [National Center for Agricultural Utilization Research (NRRL), Peoria, IL, US]. All strains were maintained at −80°C in Trypticase Soy Broth [Becton, Dickinson and Company (BD), Sparks, MD, US] supplemented with 0.6% yeast extract (Fisher Scientific, Fair Lawn, NJ, US; TSBYE) and 20% (v/v) glycerol.

### Preparation of Inoculum

Each *L. monocytogenes* strain was twice activated in TSBYE at 37°C for 24 h individually, then centrifuged at 8,000 × *g* for 5 min at 4°C. The resulting bacterial pellets were washed once and then resuspended in phosphate-buffered saline (PBS, pH 7.4) to achieve the target population. To prepare a 3-strain *L. monocytogenes* inoculum cocktail, each strain suspension at ~5 × 108 CFU/ml was mixed at 1:1:1 ratio to ~ 6.0 Log10 CFU/ml in PBS for apple inoculation.

### Apple Inoculation

Unwaxed mature Granny Smith apples (medium size, ~220 g/apple) without cuts, bruising, or scars were selected and rinsed with cold tap water and dried overnight to balance apple temperature to room temperature (22 ± 1°C, RT). Apples were then inoculated with *L. monocytogenes* by submerging into the inoculum solution prepared above and gently agitating for 8 min to let bacteria evenly distribute on each apple as described previously (Sheng et al., 2017). Inoculated apples were stored at RT under environmental relative humidity for 24 or 48 h before being subjected to the PAA treatments. Meanwhile, apples were sampled right after inoculation, 24 and 48 h post-inoculation to confirm the established *L. monocytogenes* population.

### Antimicrobial Immersion Procedure

Bioside HS (EnviroTech, Modesto, CA, US) containing 15% of PAA was used to prepare solutions of 40, 60, and 80 ppm of PAA. All PAA solutions were prepared with tap water, unless otherwise specified. The concentration of PAA was verified using a titration kit (Aquaphoenix Scientific, Hanover, PA, US). Apples at 24 and 48 h post-inoculation were immersed in respective antimicrobial solutions with agitation for 30 s or 2 min; 10 apples were used per treatment. All treatments were repeated independently three times. PAA solutions were used at RT unless otherwise specified.

To evaluate the influence of pH on the efficacy of PAA, the pH of PAA solution was adjusted with 6.0 M HCl to achieve a pH of 2.5 and 3.8, and PAA dissolved in tap water had a pH of 6.3. Chlorine solution with 100 ppm free available chlorine (FAC) was used as a reference control and prepared from Accu-Tab (Pace International, Wapato, WA, US) (Beuchat and Ryu, 1997). The pH of a chlorine solution was adjusted to 6.8 with 6 M HCl before being used in apple treatment. Water wash of apples was used as negative control to show the bacterial reduction due to factors other than antimicrobial activities. The pH and the oxidation reduction potential (ORP) of solutions were measured with an Orion Versa Star Pro advanced electrochemistry meter (Thermo Scientific, Waltham, WA, US) with an 8302Bnumd Ross Ultra pH/ATC Triode and ORP Triode. FAC was confirmed with a Taylor K-2006 complete test kit (Taylor Technologies, Sparks, MD, US).

### Water Hardness Determination and Adjustment

Water hardness was measured by a hardness test kit (Hach, Loveland, CO, US). Three levels of water hardness (20, 140, and 460 ppm) were selected in this study to determine the influence of water hardness on PAA antimicrobial efficacy. Deionized water and tap water were used as water with 20 and 140 ppm hardness, respectively. Water with a hardness of 460 ppm was achieved by adding calcium chloride (Sigma, St Louis, MO, US) to tap water.

### Antimicrobial Efficacy of Peroxyacetic Acid at Elevated Temperature

To evaluate antimicrobial efficacies of PAA at elevated temperature, PAA solution was made with water preheated to the target temperatures (~22–49°C) and used immediately after preparation. The temperature of PAA solutions was maintained for each setting throughout the experiment. The concentration and the temperature of PAA were verified using a PAA titration kit (Aquaphoenix Scientific, Hanover, PA, US) and a thermometer (Fisher Scientific, Hampton, NH, US), respectively, before and after treatment. The temperature of the apple surface was measured with a digital thermometer with a probe (Fisher Scientific).

### Microbial Analysis of Apples

Immediately after antimicrobial treatment, each apple was individually placed into a sterile stomacher bag with 10-ml sterile PBS and hand-rubbed for 1.5 min to detach microbiota from apple surfaces. The detached microbial suspension was 10-fold serially diluted with sterile PBS, and 0.1 or 1 ml (333 μl/plate, 3 plates) from appropriate dilutions was plated on TSAYE plates overlaid with Modified Oxford agar (MOX, BD), and incubated at 35 ± 2°C for 48 h. Non-inoculated apples were processed the same way as inoculated apples and plated onto TSAYE for total plate count (TPC, BD) and Potato Dextrose Agar (PDA, BD) for yeasts and molds count (Y/M), respectively. TSAYE and PDA plates were incubated at 35 ± 2°C for 48 h and at RT for 5 days, respectively. The detection limit of all microorganisms was 10 CFU/apple.

### Statistical Analysis

Data were analyzed with one-way Analysis of Variance (ANOVA) using IBM SPSS 19.0 (Chicago, IL, US). Mean difference was discerned by Least Significant Difference (LSD) multiple comparison. *p* < 0.05 was considered statistically significant. Each experiment was repeated three times independently. For a selected independent test, there are 10 apples per treatment, where each apple is an experimental unit. Data were reported as mean ± SEM (standard error mean), *n* = 3.

### RESULTS

### Influence of Concentration on the Antimicrobial Efficacy of Peroxyacetic Acid

Apples were inoculated with ~6.4 Log10 CFU/apple and then subjected to antimicrobial treatments after 24 and 48 h inoculation, respectively. At 24 h post-inoculation, 100 ppm chlorine at pH 6.8 caused 0.91 Log10 CFU/apple reduction and tap water wash led to 0.15 log reduction. PAA at 40 ppm reduced *L. monocytogenes* on fresh apples by 1.37 Log10 CFU/ apple at 2-min exposure, which was more effective than that of 100 ppm chlorine (**Figures 1A,B**). Increasing PAA concentration significantly increased its bactericidal effects. PAA at 80 ppm and 2-min contact time reduced *L. monocytogenes* on apples by 2.17 Log10 CFU/apple (**Figure 1B**). Extending the postinoculation time from 24 to 48 h significantly reduced 80 ppm PAA efficacy with a log reduction of 1.71 Log10 CFU/ apple at a 2-min treatment, though it had a minor influence on PAA efficacy at 40 and 60 ppm (**Figure 1B**). During the postharvest processing, foodborne pathogens can contaminate apples at any stage; thus, a bacterial attachment time of 48 h was used in the following study to mimic the harshest condition. PAA at 80 ppm was selected to mimic the current industry practice and to assess the maximal expected reduction.

### Impacts of Water Hardness and pH on Antimicrobial Efficacy of Peroxyacetic Acid

The hardness of wash water varies in the apple industry in Washington and ranges from 0 to 450 ppm (per our survey data). Thus, impacts of water hardness on PAA efficacy were further analyzed. PAA solutions made with water of different hardness had a similar efficacy against *L. monocytogenes* on fresh apples, ranging from 1.8 to 2.0 Log10 CFU/apple reduction (**Figures 2A,B**). Next, we examined the impact of pH on PAA antimicrobial efficacy and found that PAA exerted a similar bactericidal effect at pH 2.5–6.3, which reduced *L. monocytogenes* on apples by 1.7–1.8 Log10 CFU/apple (**Figures 2C,D**). In the subsequent studies, all PAA solutions were made with tap water with ~140 ppm hardness and a pH of 6.3.

FIGURE 1 | Antimicrobial efficacy of peroxyacetic acid (PAA) against *L. monocytogenes* on apples at a 2-min contact time at 22°C. (A) Representative bar graph of survival of *L. monocytogenes* on apples post-PAA treatment. (B) Log reduction of *L. monocytogenes* on apples, averaged from three independent experiments. a–dMeans within a column without common letter differ significantly (*p <* 0.05), A–Bmeans within a row without common letter differ significantly (*p <* 0.05). Mean ± SEM, *n* = 3. 24-h attachment: *L. monocytogenes* are allowed to attach to apples for 24 h before antimicrobial treatment; 48-h attachment: *L. monocytogenes* are allowed to attach to apples for 48 h before antimicrobial treatment.

### Improved Efficacy of Peroxyacetic Acid Against *L. monocytogenes* on Fresh Apples at Elevated Temperatures

In some commercial apple packing operations, apples are subjected to a hot-water (up to 38°C) rinse before sanitizer intervention as a necessary treatment to facilitate application of waxes or fruit lusters. Studies also showed that a 40-min exposure to 50°C water had no negative effect on apple quality (Hansen et al., 2006). This prompted us to assess the antimicrobial efficacy of PAA against *L. monocytogenes* on apples at elevated temperatures. Increasing PAA solution temperature from RT to 41°C had no significant influence on antimicrobial efficacy of PAA (**Figures 3A,B**). However, when the temperature was further increased to 43°C, the reduction of *L. monocytogenes* was significantly improved (**Figure 3B**). PAA at 43 and 46°C reduced *L. monocytogenes* on apples by 2.37 ± 0.06 and 2.63 ± 0.04 Log10 CFU/apple, respectively (**Figure 3B**). However, increasing PAA solution temperature to 49°C failed to further enhance its effectiveness (**Figure 3B**). Reducing contact time from 2 min to 30 s decreased its bactericidal effects (**Figures 3C,D**). The concentration of PAA at all the tested temperatures remained stable during the wash treatment, while pH and ORP of PAA solutions gradually decreased with increased temperature (**Table 1**). The surface temperatures of apples post 2-min PAA treatment at 43 and 46°C were 37.4 ± 0.3 and 38.4 ± 0.4°C, respectively (**Table 2**).

### Efficacy of Peroxyacetic Acid Against Background Microbiota at Elevated Temperatures

We further evaluated the effectiveness of PAA in reducing apple resident microbiota. PAA at 46°C significantly improved its antimicrobial activity compared with that at RT, and reduced TPC by 1.20 ± 0.02 and 1.54 ± 0.05 Log10 CFU/ apple at contact times of 30 s and 2 min, respectively (**Figures 4A,B**). Similarly, PAA at 46°C enhanced its

FIGURE 2 | Antimicrobial efficacy of peroxyacetic acid (PAA) against *L. monocytogenes* on apples under different water hardness and pH at 22°C. *L. monocytogenes* are allowed to attach to apples for 48 h before antimicrobial treatment. (A,C) Representative bar graphs of *L. monocytogenes* survival on apples; (B,D) Log reduction of *L. monocytogenes* on apples, averaged from three independent experiments. a Means within a column with common letter are not different significantly (*p <* 0.05), mean ± SEM, *n* = 3.

FIGURE 3 | Influence of temperature and contact time on antimicrobial efficacy of peroxyacetic acid (PAA) against *L. monocytogenes* on apples. (A,C) Representative bar graphs of *L. monocytogenes* survival on apples. (B,D) Log reduction of *L. monocytogenes* on apples, averaged from three independent experiments. a–cMeans within a column or a temperature without common letter differ significantly (*p <* 0.05). Mean ± SEM, *n* = 3.

effectiveness in reducing Y/M on apples and caused ~2.0 Log10 CFU/apple reduction at contact time of 2 min, which was almost double the log reduction when PAA was applied at RT (**Figures 4C,D**).

## DISCUSSION

In the fresh produce industry, especially, the fresh apple packing industry, PAA has become the preferred antimicrobial for microbial interventions. PAA is currently allowed under the National Organic Program (NOP) for organic food handling (USDA, 2016). The advantages of using PAA over commonly used chlorine are the unnecessity to adjust pH, low reactivity with organic matter, and safety of its reaction and residual breakdown products (Kitis, 2004; Buchholz and Matthews, 2010).

TABLE 1 | pH and oxygen reduction potential (ORP) of peroxyacetic acid (PAA) at different temperatures.


*Data are presented as mean ± SEM, n = 3.*

TABLE 2 | Temperature of apple surface and peroxyacetic acid (PAA) solution at pre- and post-PAA intervention.


*Data are presented as mean ± SEM, n = 3.*

### Antimicrobial Efficacy of Peroxyacetic Acid at Current Commercial Treatment Conditions

The efficacy of PAA against *L. monocytogenes* on apples is concentration dependent. Similarly, PAA at 25, 51, and 70 ppm for 3 min exposure resulted in 1.0, 1.4, and 1.8 Log10 CFU/g reduction of *L. monocytogenes* on bean sprouts, respectively (Neo et al., 2013). A PAA wash for 5 min at 80 and 250 ppm delivered a 0.4 and 1.3 Log10 CFU/g reduction of *L. monocytogenes* on iceberg lettuce, respectively (Baert et al., 2009). Under ambient temperature, 2-min wash with 80 ppm PAA delivered ~1.7 Log10 CFU/apple reduction of *L. monocytogenes* on apples, which was a little more effective than 80 ppm PAA against *L. monocytogenes* on Golden Delicious apples, where a log reduction time is about 80 s (Rodgers et al., 2004). *E. coli* O157:H7 on apples is less responsive to PAA, where 80 ppm PAA only reduced it by ~1.0 Log10 CFU/apple at 5-min contact time (Wisniewsky et al., 2000; Alcala et al., 2011). The difference in susceptibility could be due to difference in bacterial strains, surface attributes of apple varieties, as well as source of PAA solution.

The effectiveness of PAA against *L. monocytogenes* on apple surfaces was not measurably impacted by the hardness of water or pH condition. This is consistent with a previous publication that states that the stability of PAA solution was not affected by hardness of water (Artes et al., 2007). A 200 ppm PAA solution at both pH 2.8 and 4.3 reduced *Salmonella* Heidelberg on poultry product by ~1.0 Log10 CFU/ml at a 15-s contact time (Donabed, 2015). This might be due to the active compound of PAA solution, undissociated acid form of PAA, that was

FIGURE 4 | Efficacy of peroxyacetic acid (PAA) against background microbiota on apples treated at different temperatures. (A) Total plate count (TPC) of residential bacteria on apples. (C) Representative survival of yeast and mold (Y/M). (B,D) Log reduction of TPC (B) and YM (D) on apples, averaged from three independent experiments. a–cMeans within a column without common letter differ significantly (*p <* 0.05), mean ± SEM, *n* = 3.

stable at pH equal to or less than its pKa of 8.2 (Yuan et al., 1997; Wagner et al., 2002), thus exhibiting a similar antimicrobial efficacy at the tested pH range. Antimicrobial action of PAA is possibly attributed to its action on the lipoproteins in the cell membrane, which results in disruption of the lipoprotein cytoplasmic membrane or cell walls due to oxidative stress, and subsequently denaturation of intracellular enzymes and other important macromolecules (Leaper, 1984; Maris, 1995).

Antimicrobial efficacy of 80 ppm PAA at an ambient temperature against *L. monocytogenes* on apples increased with increased contact time. There was ~0.42 more log reduction at a 2-min contact time compared to that of 30-s contact time. Similarly, PAA at 80 ppm for 5 min reduced native microorganisms on iceberg lettuce by ~2.4 Log10 CFU/g, which was ~0.9 Log10 CFU/g more reduction than that of a 2-min contact time (Vandekinderen et al., 2009). However, 70 ppm PAA at either 1.5- or 3-min contact time reduced *L. monocytogenes* on mung bean sprouts by ~1.8 Log10 CFU/g (Neo et al., 2013). 1- and 2-min PAA treatments at 75 ppm showed a comparable efficacy (~2 Log10 CFU/produce reduction) against *Salmonella* on bell peppers and cucumbers (Yuk et al., 2006).

### Enhanced Antimicrobial Efficacy of Peroxyacetic Acid at Elevated Temperature

The biocidal effects of PAA significantly increased when the PAA solutions were applied at 43–46°C compared with that at an ambient temperature. A similar phenomenon was observed on beef carcasses. Even at 1000 ppm, PAA showed a minimal efficacy against *E. coli* O157:H7 on beef carcasses when applied at RT, while it resulted in ~0.9 Log10 CFU/cm2 at when applied at 45°C (King et al., 2005). The elevated temperature might increase transportation of PAA across bacteria membranes, impairing intracellular osmotic balance, and subsequently facilitate cell death (Laroche et al., 2001; McCutcheon and Elimelech, 2006). Additionally, increasing wash solution temperature reduced the surface tension between hydrophobic apple surfaces and hydrophilic PAA solution, thus exposing the entrapped *L. monocytogenes* cells to PAA (Vandekinderen et al., 2009). PAA delivered stronger antimicrobial efficacy at non-dissociated form (Luukkonen et al., 2014), and PAA at 43–46°C was likely maintained the non-dissociated form and enhanced its antimicrobial efficacy against *L. monocytogenes.* Though the decomposition rate of PAA was negatively affected by increased temperature at long-time exposure (Kunigk et al., 2001), it had no influence on PAA concentration within minutes of exposure. The concentration of PAA maintained stable after 2-min intervention at respective temperatures.

Apple resident microbiota including TPC and Y/M are reported to affect apple fruit quality and shelf life during storage (Doores, 1983; Palou et al., 2009). At an ambient temperature, PAA at 80 ppm and 2-min contact time showed a limited antimicrobial efficacy (~1.0 log reduction) against TPC or Y/M. Similarly, 80 ppm PAA at 5 min reduced Y/M by 1.0–1.5 Log10 CFU/ apple on apples when it was applied at an ambient temperature (Rodgers et al., 2004; Kreske et al., 2006). Similar to *L. monocytogenes*, elevation of PAA solution temperature significantly improved its biocidal effects against apple resident microbiota with more significant effect on Y/M lethality. Data indicate that PAA intervention at 43–46°C has a potential to increase apple shelf life in addition to improved microbial safety.

Elevated temperature slightly increased the surface temperature of apples to 35–38°C depending on treatment temperature and contact time. A previous study showed that 46°C treatment of apples for 12 h increased the firmness of fruits and reduced the development of superficial scald following subsequent 3 months under refrigerated storage at 0°C (Klein and Lurie, 1992). However, extended exposure time to 24 h resulted in fruit damage after storage (Klein and Lurie, 1992). Thus, temperate can have a negative effect for long-term exposure at evaluated temperature; however, PAA intervention at elevated temperatures used in this study was conducted in a short contact time, thus it has a minimal impact on apple fruit quality.

### CONCLUSION

PAA at 80 ppm and 2-min contact time reduced *L. monocytogenes* on fresh apples by ~1.7 Log10 CFU/apple when applied at an ambient temperature, which was not affected by the hardness or pH of PAA solution. PAA intervention at 43–46°C significantly enhanced its bactericidal effects, and reduced *L. monocytogenes* on fresh apples by 2.3–2.6 Log10 CFU/apple, and TPC and Y/M by ~1.5 and ~2.1 Log10 CFU/apple, respectively. These data provide valuable technical information and practical intervention methods for the apple packing and processing industry to support compliance with Food Safety Modernization Preventive Controls requirements. The study also provides important reference points for controlling other important foodborne pathogens such as *E. coli* O157:H7 and *Salmonella* on fresh apples, as well as other fresh produce with similar surface traits and postharvest handling systems.

### AUTHOR CONTRIBUTIONS

XS, LS and HG performed the experiment. XS wrote the manuscript. IH, LS and TS revised the manuscript. IH provided the survey information. M-JZ guided the experimental design and revised the manuscript.

### FUNDING

This activity was funded by the Center for Produce Safety 2017CPS10 and the Washington Tree Fruit Research Commission.

### ACKNOWLEDGMENTS

We would like to acknowledge Stemilt Growers LLC and Allan Brothers Inc. for their generous donation of fresh apples, and Pace international Inc. for providing Accu-Tab and Bioside HS. We thank Yuan Su, Tonia Green, and Zi Hua for assistance in sample preparation and microbial analyses.

### REFERENCES


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

*Copyright © 2019 Shen, Sheng, Gao, Hanrahan, Suslow and Zhu. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Embracing Diversity: Differences in Virulence Mechanisms, Disease Severity, and Host Adaptations Contribute to the Success of Nontyphoidal Salmonella as a Foodborne Pathogen

#### Rachel A. Cheng<sup>1</sup> \*, Colleen R. Eade2,3 and Martin Wiedmann<sup>1</sup>

<sup>1</sup> Department of Food Science, Cornell University, Ithaca, NY, United States, <sup>2</sup> Department of Population Medicine and Diagnostic Sciences, Cornell University, Ithaca, NY, United States, <sup>3</sup> Department of Chemistry, University of North Carolina at Charlotte, Charlotte, NC, United States

#### Edited by:

Marta López, Spanish National Research Council (CSIC), Spain

#### Reviewed by:

Min Yue, Zhejiang University, China Young Min Kwon, University of Arkansas, United States Chunlei Shi, Shanghai Jiao Tong University, China

> \*Correspondence: Rachel A. Cheng ram524@cornell.edu

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 30 November 2018 Accepted: 31 May 2019 Published: 26 June 2019

#### Citation:

Cheng RA, Eade CR and Wiedmann M (2019) Embracing Diversity: Differences in Virulence Mechanisms, Disease Severity, and Host Adaptations Contribute to the Success of Nontyphoidal Salmonella as a Foodborne Pathogen. Front. Microbiol. 10:1368. doi: 10.3389/fmicb.2019.01368 Not all Salmonella enterica serovars cause the same disease. S. enterica represents an incredibly diverse species comprising >2,600 unique serovars. While some S. enterica serovars are host-restricted, others infect a wide range of hosts. The diseases that nontyphoidal Salmonella (NTS) serovars cause vary considerably, with some serovars being significantly more likely to cause invasive disease in humans than others. Furthermore, while genomic analyses have advanced our understanding of the genetic diversity of these serovars, they have not been able to fully account for the observed clinical differences. One overarching challenge is that much of what is known about Salmonella's general biology and virulence strategies is concluded from studies examining a select few serovars, especially serovar Typhimurium. As targeted control strategies have been implemented to control select serovars, an increasing number of foodborne outbreaks involving serovars that are less frequently associated with human clinical illness are being detected. Harnessing what is known about the diversity of NTS serovars represents an important factor in achieving the ultimate goal of reducing salmonellosis-associated morbidity and mortality worldwide. In this review we summarize the current understanding of the differences and similarities among NTS serovars, highlighting the virulence mechanisms, genetic differences, and sources that characterize S. enterica diversity and contribute to its success as a foodborne pathogen.

Keywords: nontyphoidal Salmonella, virulence, serovars, foodborne pathogen, food safety

## INTRODUCTION

Salmonellae are Gram-negative, facultatively anaerobic bacteria. The genus Salmonella, named for Dr. Daniel Salmon, was first described in 1866 by Dr. Theobald Smith (Schultz, 2008). Initially, Salmonella was described as the causative agent of pig cholera (first named Salmonella choleraesuis). However, pig cholera was later discovered to be a viral disease, with Salmonella co-infection being common (Schultz, 2008).

FIGURE 1 | Phylogeny of Salmonella. Phylogenetic analysis reconstructed from Desai et al. (2013). Subspecies identifiers preceding the subspecies (ssp.) name are shown in parentheses. The number of currently recognized serovars (Issenhuth-Jeanjean et al., 2014) within each species or subspecies is listed in blue.

Today, the genus Salmonella includes 2 species: enterica and bongori (Brenner et al., 2000; **Figure 1**). Within the species enterica, there are 6 subspecies, which are characterized by both a name and a roman numeral: enterica (I), salamae (II), arizonae (IIIa), diarizonae (IIIb), houtenae (IV), and indica (VI) (Brenner et al., 2000). Salmonella bongori was originally classified as Salmonella enterica subspecies V, before being re-classified as a separate species (Brenner et al., 2000). A third species, Salmonella subterranean was proposed (Shelobolina et al., 2004), however, further DNA characterization revealed that this species did not belong in the genus Salmonella (Issenhuth-Jeanjean et al., 2014). Recent analyses suggest there are four major phylogenetic clades within S. enterica subsp. enterica (Worley et al., 2018). The majority of Salmonella isolates infecting warm-blooded hosts belong to subspecies enterica (I), while subspecies II-VI and S. bongori are primarily isolated from environmental sources or cold-blooded hosts, such as reptiles and amphibians.

Infection with typhoidal (S. enterica subsp. enterica serovar Typhi) and paratyphoidal serovars (i.e., serovars Paratyphi A, Paratyphi B, Paratyphi C, and Sendai) typically results in an invasive, extra-intestinal disease characterized by a high fever (>39◦C), malaise, vomiting, headache, and an elevated pulse rate (Dougan and Baker, 2014; Hiyoshi et al., 2018). Serovar Typhi is host-restricted, and is transmitted human-to-human primarily via fecal contamination of drinking water or improper food handling (Connor and Schwartz, 2005). In contrast, infections with NTS serovars (i.e., all serovars except Typhi, Paratyphi A, Paratyphi B, Paratyphi C, or Sendai) typically result in a selflimiting gastroenteritis that is cleared by the host within 4–7 days (Gal-Mor et al., 2014). Due to these stark discrepancies in disease manifestation, nontyphoidal and typhoid salmonellosis cases are frequently assessed and reported independently.

Serovar Typhimurium is the best-studied serovar for nontyphoidal salmonellosis and has become the "model serovar" for studying NTS (Sabbagh et al., 2010; Tsolis et al., 2011; Fàbrega and Vila, 2013; Rivera-Chavez and Baumler, 2015). The S. Typhimurium strain LT2 has been widely used since the 1940s and has been characterized extensively (Swords et al., 1997). In fact many studies have used LT2 as a model for NTS, despite the fact that this strain encodes a rare start codon (UUG) in its rpoS, rendering this strain avirulent (Lee et al., 1995;

Wilmes-Riesenberg et al., 1997). This suggests that conclusions drawn from experiments performed with S. Typhimurium LT2 should be cautiously considered, as this lab strain may not accurately represent NTS virulence. Furthermore, NTS vary significantly with respect to the severity of the illness that they cause, as infection with certain serovars is significantly more likely to result in human invasive disease, hospitalization, and death when compared to the rates of invasive disease observed for S. Typhimurium (Jones et al., 2008). This suggests that while S. Typhimurium was useful as a general model for NTS, there are likely additional virulence factors, adaptations, and other fitness determinants that may not be accurately represented in S. Typhimurium.

Despite extensive government and industry efforts aimed at reducing the incidence of nontyphoidal salmonellosis, little progress has been made in reducing the number of salmonellosis cases per year (Boore et al., 2015; Ford et al., 2016). The various hosts that NTS serovars encounter and colonize demonstrate their tremendous adaptability. Differences in food consumption, sanitation, cultural traditions, infrastructure, and food safety regulations all influence the global burden of nontyphoidal salmonellosis. Furthermore, differences in host susceptibility, virulence factors/mechanisms specific to select serovars, as well as serovar fitness, contribute to the severity, and outcome of salmonellosis (**Figure 2**). The role of antimicrobial resistance in the expansion of select NTS subtypes in a novel niche, as well as the potential implications associated with increasing the severity of the outcome of infection due to rapid outgrowth of an antibiotic-resistant Salmonella strain in the GI tract of antibiotictreated individuals, have been reviewed extensively elsewhere (Rabsch et al., 2001; Furuya and Lowy, 2006; Crump et al., 2015), and therefore will not be covered in this review.

A number of reviews have detailed the host-response to both nontyphoidal and typhoid salmonellosis (Gal-Mor et al., 2014; Gilchrist et al., 2015; Keestra-Gounder et al., 2015; LaRock et al., 2015; Rivera-Chavez and Baumler, 2015). However, a better understanding of the diversity of NTS serovars and which adaptations make them more suited to colonize different hosts or survive in different environments, is lacking. In this review, we have synthesized studies representing the diversity of NTS serovars to demonstrate how Salmonella uses a variety of virulence factors, and genetic and phenotypic adaptations to become one of the most successful foodborne pathogens worldwide.

### Classical and Modern Approaches to Defining and Studying NTS Diversity

Salmonellae are differentiated based on the reactivity of the O (somatic, or O polysaccharide component of lipopolysaccharide [LPS]) and H (flagellar) antigens to anti-sera, in a classification scheme known as the Kauffman-White-Le Minor scheme (**Figure 3**; Brenner et al., 2000; Liu et al., 2014). Salmonellae are

categorized into 46 unique O groups (Liu et al., 2014), with 114 unique H antigens (McQuiston et al., 2004). Together with the roman numeral abbreviation for the S. enterica subspecies, the combination of O- and H-antigens antigens is what defines the

FIGURE 3 | Illustration of NTS O- and H-antigens, comprising the serotyping scheme defined by Kauffman-White-Le Minor. The antigenic formula of a serovar is composed of the subspecies (I -VI), the O-antigen (numbered), and the flagellar antigens called phase 1, phase 2, and phase 3 (if present). Some serovars have lost their phase 1 and/or phase 2 antigen and others have gained additional flagellar antigens (phase 3). The O-antigen illustrated represents the terminal side chain of lipopolysaccharide (LPS) on the cell wall of Gram-negative bacteria, such as Salmonella, which is composed of an inner and outer core of primarily polysaccharides bound to Lipid A; the O side chain is composed of repeating saccharide units, and is the subunit of LPS that differs between serogroups. Salmonella enterica subsp. enterica serovar Heidelberg, the NTS serovar modeled here, is the common name of the serovar having antigenic formula I 1,4,[5],12:r:1,2.

serotype (also known as a serovar) of a given Salmonella strain. Salmonella serovars also have common names that are often the name of the geographical location where they were first isolated (Gossner et al., 2016), to accompany their antigenic formulae (Brenner et al., 2000).

Another traditional classification scheme, called phage typing, involves subjecting Salmonella strains to libraries of phage to determine which phage are able to lyse a given strain. Examples of historically important phage types include S. Typhimurium DT (definitive type) 104, which arose as an important multidrug resistant (MDR) strain in the United Kingdom in the early 1960s that subsequently spread and became a global epidemic (Threlfall, 2000), and S. Enteritidis PT (phage type) 4, which is associated with contamination of intact chicken eggs (Humphrey, 1994).

Although modern genetic analyses are now typically used to predict the phage type of a given Salmonella isolate, phage typing may still be reported as an additional level of discrimination. Likewise, other in silico typing schemes have increased in popularity, as they provide increased discriminatory power and negate the use for traditional serotyping. Rather, these methods utilize serotype prediction based on sequence typing of the O- and H-antigens (McQuiston et al., 2004), or more recently, using whole genome sequence (WGS) data (Zhang et al., 2015; Yoshida et al., 2016; Robertson et al., 2018). The sequence data are compared to a database of sequences, which compare the O- and H-antigens of known serotypes to predict which anti-sera an isolate would react with, and therefore, the serotype of the isolate. Other identification schemes are based on sequence comparisons

of housekeeping genes (Achtman et al., 2012), and more recently on core genome or whole genome multi-locus sequence typing (Alikhan et al., 2018).

Recent phylogenetic analyses have shed light on the fact that many commonly isolated serovars (serovars Newport, Montevideo, Kentucky, Paratyphi B, Derby, Nchanga, Cerro, Bareilly, Stanleyville, Dusseldorf, Livingstone, and others) are polyphyletic (Den Bakker et al., 2011; Cao et al., 2013; Timme et al., 2013; Yoshida et al., 2016; Sévellec et al., 2018; Worley et al., 2018). In fact, a recent study comparing 266 different serovars reported that ∼10% are polyphyletic or paraphyletic (Worley et al., 2018). The polyphyletic nature of some serovars may provide further epidemiological evidence that can assist in outbreak investigations by providing additional discriminatory power, as was recently proposed for S. enterica subsp. enterica serovar Derby among different regions in France (Sévellec et al., 2018). WGS subtyping adds an additional level of discriminatory power that can be used to aid epidemiologic investigations of traceback studies. There are also important clinical implications of polyphyletic Salmonella serovars that arise when different clades of the same serovar differ in the virulence factors that they encode, as is the case for S. Mississippi, where one clade encodes typhoid toxin genes, and the other does not (Miller and Wiedmann, 2016). Genomic analyses are just beginning to identify polyphyletic serovars, and to define key differences associated with the different clinical outcomes observed.

### Geographic Diversity – Distribution of NTS Serovars Is Regionally Associated

Infections with NTS account for just over one fifth of all bacterial foodborne illnesses worldwide, causing an estimated 78.7 million cases per year (Havelaar et al., 2015). In the United States NTS is the leading cause of bacterial foodborne illness, resulting in an estimated 1.2 million illnesses, 23,128 hospitalizations, and 452 deaths annually (Scallan et al., 2011). In the US, the 20 NTS serovars most commonly isolated from human clinical cases account for nearly 70% of all NTS clinical cases in humans (Centers for Disease Control and Prevention [CDC], 2016). According to WHO estimates of foodborne disease, the global incidence of nontyphoidal salmonellosis as of 2010 was 1,140 cases per 100,000 people (1.14% of people) (Kirk et al., 2015). However, the burden of nontyphoidal salmonellosis is not equally distributed among different geographic regions. Countries in the Eastern Mediterranean region (e.g., Iran, Kuwait, Saudi Arabia, Egypt, and others) had the highest estimated incidence of nontyphoidal salmonellosis (1,610 cases per 100,000), while the European region (e.g., European Union, Russia, Ukraine, Switzerland, and others) had the lowest (186 cases per 100,000) (Kirk et al., 2015). However, the African region had the highest death rate from nontyphoidal salmonellosis: 1 death per 100,000 persons compared to a global rate of 0.4 deaths per 100,000 persons (Kirk et al., 2015). In contrast to other foodborne bacterial pathogens such as E. coli and Campylobacter spp., the majority of NTS salmonellosis occurs in individuals above 5 years of age (Havelaar et al., 2015).

Despite the discrepancy in rates of illness per geographic region, Typhimurium and Enteritidis are consistently reported as the serovars contributing the highest proportion of human clinical salmonellosis worldwide (**Figure 4**; Hendriksen et al., 2009, 2011; Ran et al., 2011; OzFoodNet Working Group, 2015; Centers for Disease Control and Prevention [CDC], 2016; European Food Safety Authority [EFSA] and European Centre for Disease Prevention, and Control [ECDC], 2017). Interestingly, in some countries the proportion of human clinical cases are dominated by one serovar, such as Typhimurium in Australia or Enteritidis in Brazil, while other countries show a more balanced distribution of cases per serovar (e.g., Enteritidis, Newport, and Typhimurium in the United States). The serovars representing the 3rd, 4th, and 5th most commonly isolated serovars from human clinical infections appear to be more geographically associated. For example, serovars that are found among human cases in Australia (serovars Saintpaul and Virchow) are less common in other countries (**Figure 4**).

The regional differences observed among the incidence of infection by select serovars may be explained by differences in (i) animal hosts that populate the region, (ii) the quality of surveillance and reporting systems (Kirk et al., 2015), which may effectively underestimate the incidence of some serovars, (iii) concurrent immunodeficiencies such as HIV infection or cancer (Okoro et al., 2012), (iv) dietary intake, (v) farming practices or food production practices that might select for specific serovars, or (vi) environmental factors influencing the cultivation, survival, or routes of transmission. In support of differences in environmental exposures, within the United States some serovars are regionally distributed, with some serovars representing the majority of clinical isolates in some geographical regions, but being rarely isolated in other regions (e.g., S. Mississippi) (Boore et al., 2015). This suggests that the majority of human clinical NTS salmonellosis cases in the United States are the result of contamination events that happen at a local level, given that food is often distributed across the country, if not internationally.

### DIVERSITY OF NTS VIRULENCE FACTORS AND THEIR IMPLICATIONS IN DISEASE MANIFESTATION

Reflective of an extensively host-adapted lifestyle is the collection of virulence factors possessed by NTS. These include flagella, fimbriae, toxins, pathogenicity islands, and virulence-associated plasmids. These features do not occur in all NTS serovars, and thus their presence or absence influences the virulence and host range of a particular isolate or serovar.

### Salmonella Pathogenicity Islands (SPIs)

To date, 24 Salmonella pathogenicity islands have been identified. These horizontally acquired loci encode genes facilitating several virulence mechanisms, including (i) the expression of secretion systems, fimbriae, flagella, and capsules, (ii) serotype conversion, and (iii) host colonization and subsequent survival within the host (Van Asten and Van Dijk, 2005; Fàbrega and Vila, 2013).

Salmonella Pathogenicity Islands 1 and 2 are the best characterized in terms of genetic and phenotypic traits. Of the 24 SPIs (**Table 1**), SPI-1 is ubiquitous among all Salmonella species and subspecies (Fookes et al., 2011). SPI-1 encodes a type three secretion system (T3SS), which is essential for export of effector proteins required for invasion of host cells. In contrast, SPI-2 is found only in S. enterica, and SPI-22 is only found in S. bongori (Fookes et al., 2011). SPI-2 encodes an additional T3SS, harboring genes that are essential for intracellular survival and for preventing acidification of the Salmonella containing vacuole (SCV). The remaining 21 SPIs are variably present among S. enterica (Kingsley et al., 2003; Shah et al., 2005; Boyen et al., 2006; Saroj et al., 2008; Blondel et al., 2009, 2013; Den Bakker et al., 2011; Shah et al., 2012; Hayward et al., 2013; Lee et al., 2013; Tomljenovic-Berube et al., 2013), with all but three having been found in NTS serovars; SPI-20 and -21 have been found only in S. enterica subsp. arizonae (Blondel et al., 2009), and while SPI-15 occurs in S. enterica subsp. enterica, it has only been identified in S. Typhi to date (Vernikos and Parkhill, 2006; Sabbagh et al., 2010).

While SPIs are widespread among a number of S. enterica subsp. enterica serovars, some SPIs have been associated with select serovars, and are proposed to provide fitness advantages for these serovars. For example, SPI-7, encoding the Vi capsule, although classically thought to occur exclusively in S. Typhi (Faucher et al., 2005), has been found in strains of the NTS serovar Dublin as well (Morris et al., 2003; Pickard et al., 2003). It is interesting to note that both of these serovars are associated with invasive disease in humans. While SPI-11 is widespread among S. enterica subsp. enterica (**Table 1**), some serovars also encode genes for the typhoid toxin at this locus (Den Bakker et al., 2011). The presence of cdtB (a typhoid toxin gene) has been reported to be associated with significantly higher rates of invasive disease (Rodriguez-Rivera et al., 2015). Examples of SPIs associated with the ability to colonize a specific host have also been proposed. Mutation analyses of the T6SS encoded in SPI-19 in S. Gallinarum revealed a role for SPI-19 in the ability to colonize chickens (Blondel et al., 2010). However, introduction of SPI-19 (cloned from S. Gallinarum) in S. Enteritidis negatively impacted this serovar's ability to colonize chickens (Blondel et al., 2009), therefore suggesting that, depending on the serovar, this SPI may provide a fitness advantage for some serovars, but not for others.

### Toxins

Exotoxins constitute toxins that are secreted. Originally identified in serotype Typhi, the typhoid toxin (or Salmonella cytolethal distending toxin) is a genotoxin that has recently been identified in at least 41 NTS serovars as well (Den Bakker et al., 2011), however, it is not found in serovars Typhimurium, Enteritidis, and Newport, which cause many of the clinical infections in the United States, and worldwide. Experiments with S. Typhi


TABLE 1 | Current understanding of the size, function, and distribution of Salmonella pathogenicity islands (SPIs) among Salmonella.

(Continued)


<sup>1</sup>SPI-10 was initially defined in S. Typhi and was thought to be unique to S. Typhi (Sabbagh et al., 2010), although other studies have reported the presence of SPI-10 genes in NTS serovars as well (Den Bakker et al., 2011; Desai et al., 2013). <sup>2</sup> In some cases the size varies between serovars depending on the number of genes encoded. 3 "some" indicates that not all serovars, or strains within a serovar, carry a given gene(s). <sup>4</sup>Distribution based on screening of few representative strains, may not be representative of the entire subspecies as comprehensive studies for all serovars have not been conducted. For SPIs 5 and 11 no data were available for testing in subspecies other than enterica. <sup>5</sup> "Partial" indicates that strains tested in the labeled subspecies encoded part of the SPI, but not the full SPI.

have revealed a putative role for this toxin in fine-tuning the host response to infection by targeting specific cell types such as endothelial cells in the brain, and immune cells (Yang et al., 2018). Injection of purified typhoid toxin has been shown to recapitulate some signs of typhoid fever in a mouse model (Song et al., 2013). In NTS, this toxin has been shown to contribute to systemic host colonization by S. Javiana (Miller et al., 2018).

The spv operon, primarily present on the Salmonella virulence plasmid (described below), encodes a toxin that mediates ADPribosylation of actin and is implicated in host cell cytoskeletal rearrangements, and finally, apoptosis of the host cell (Libby et al., 1997, 2000; Otto et al., 2000; Lesnick and Guiney, 2001). The spv locus has also been reported to be chromosomally encoded for some isolates in subspecies II, IIIa, IV, and VII (Boyd and Hartl, 1998). SpvC has been shown to contribute to virulence, due to its phosphothreonine lyase activity, which inhibits host MAP kinases (Guiney and Fierer, 2011). SpvB and SpvC are only found in a select number of serovars (see **Table 2**). Isolates encoding the spv operon are proposed to have enhanced virulence, and are often associated with invasive disease (Libby et al., 1997; Guiney and Fierer, 2011), as spv genes appear to play an important role in suppression of the innate immune response by attenuating the intestinal inflammatory response (Haneda et al., 2012).

The ArtAB toxin (for ADP-ribosylating toxin) is encoded by S. Typhimurium DT-104 strains (Saitoh et al., 2005) as well as multiple other NTS serovars (Rodriguez-Rivera et al., 2015). The active subunit, ArtA, ADP-ribosylates host G proteins (Uchida et al., 2009), while ArtB forms the pentameric binding subunit (Tamamura et al., 2017). ArtA and ArtB share homology with the active subunit (S1) and one of the B monomers (S2) of the



Virulence plasmids are listed with their corresponding name and size in kilobase pairs (kb).

binding subunit of the pertussis toxin, respectively (Tamamura et al., 2017). In vitro, the ArtAB toxin results in increased production of cAMP in RAW 264.7 cells, and a cell clustering phenotype in CHO cells (Saitoh et al., 2005; Tamamura et al., 2017). In vivo, BALB/c mice injected with 0.5–2 µg of purified ArtAB have significantly higher insulin secretion, and a reduced survival rate compared to mice injected with heat-killed toxin (Saitoh et al., 2005; Tamamura et al., 2017).

Several reports describe heat-labile, trypsin-sensitive cytotoxins produced by several NTS serovars. In some instances, cytotoxic activity is associated with the outer membrane. Cytotoxin production has been reported for extracts of serovars Braenderup, Choleraesuis, Enteritidis, Indiana, Nchanga, Saintpaul, Typhimurium, and Virchow (Reitmeyer et al., 1986; Ashkenazi et al., 1988; Kita et al., 1993; Malik et al., 1996). Considerable work has been conducted to characterize toxic activity in extracts and culture filtrates of multiple serovars (Peterson, 1980; Houston et al., 1981; Gemmell, 1984; Hariharan et al., 1986; Chopra et al., 1987; Khurana et al., 1991; Chary et al., 1993). One gene implicated in the observed cytotoxic activity, stn, has been proposed, although contrasting results about its actual involvement in cytotoxic activity exist, as its protein product has also been implicated in bacterial membrane integrity (Nakano et al., 2012).

### Flagella

Most NTS are capable of expressing flagella, which confer motility (Fàbrega and Vila, 2013; Rivera-Chavez and Baumler, 2015). An important exception is serovar Gallinarum biovars Gallinarum and Pullorum, which do not express phase 1 or phase 2 flagella and are therefore non-motile (Foley et al., 2013). Flagella synthesis, assembly, and maintenance requires >50 genes (Bonifield and Hughes, 2003). However, the antigenic subunit, flagellin, is encoded by three genes, fliC (phase 1), fljB (phase 2), and flpA (phase 3; rare and often plasmid-encoded) (McQuiston et al., 2004). For most NTS serovars, 5-10 flagella of peritrichous organization may be observed (Van Asten and Van Dijk, 2005). Salmonella employ phase variation, a reversible genetic rearrangement, to switch between expression of fliC and fljB, a mechanism that is utilized by a number of important bacterial pathogens (Silverman et al., 1979; Bonifield and Hughes, 2003; Garcia-Pastor et al., 2018).

While flagella aid Salmonella in migrating toward host epithelial layers, and thus are important virulence determinants, they are also potent inducers of the host innate immune response (Dos Santos et al., 2018). Flagella have also been shown to allow Salmonella to taxi toward the host-derived nitrate and tetrathionate, which are used as alternate terminal electron acceptors (Rivera-Chavez and Baumler, 2015).

Although there is no obvious evidence linking specific flagellar antigens to differences in virulence or host-adaptations, regulation of flagella upon infection has been established as a mechanism to reduce or prevent activation of a host immune response. For example, in S. Typhimurium, flagellar expression is down-regulated 50–100 fold during infection of RAW 264.7 murine macrophage cells (Srikumar et al., 2015). Furthermore, Spöring et al. (2018) showed that this downregulation is a response to alterations of the cell envelope as a result of cell envelope stress (Spöring et al., 2018). Studies of flagellar regulation in other NTS serovars may reveal key differences in expression, and provide additional roles by which NTS serovars are able to either intentionally trigger an inflammatory immune response, which has been associated with providing a metabolic niche allowing NTS serovars to propagate in the intestine (Rivera-Chavez and Baumler, 2015), or to evade killing by immune cells.

### Fimbriae (Pili)

Fimbriae are thin appendages that aid in attachment and adhesion, and are produced by a number of Gram-negative and Gram-positive bacteria. Phenotypic and genomic analyses have identified 39 putative fimbrial operons in Salmonella. Of these, the agf operon is found among isolates of both S. enterica and S. bongori, and encodes the nucleator-dependent curli fimbriae, which are thin, aggregative fimbriae that may aid in bacterial adhesion and invasion (Van Asten and Van Dijk, 2005; Yue et al., 2012). The bfp and pil operons encode type IV fimbriae; the latter operon is found on SPI-7, and its presence is therefore restricted to S. Typhi, Paratyphi C, and NTS serovar Dublin (Morris et al., 2003; Van Asten and Van Dijk, 2005). The remaining 36 fimbrial operons encode chaperone-usher-dependent fimbrial pathways (Clouthier et al., 1994; Van Asten and Van Dijk, 2005; Yue et al., 2012). Of these, 27 have been identified in NTS, with typical serovars containing 5–14 fimbrial gene clusters (Yue et al., 2012). The fim operon is the only chaperone-usherdependent fimbrial operon found in all S. enterica isolates (Van Asten and Van Dijk, 2005; Yue et al., 2012). While the majority of fimbrial genes are not expressed under standard laboratory culturing conditions, most are expressed in vivo during infection (Laniewski et al., 2017), suggesting a role in adhesion to different tissues, or co-regulation with SPI-2 genes or other genes that are expressed when Salmonella is inside a host cell. The occurrence of other operons in various NTS has been reviewed elsewhere (Van Asten and Van Dijk, 2005). Of interest, the pef (plasmid encoded fimbriae) operon is located on a subset of Salmonella virulence plasmids (Rychlik et al., 2006), and is discussed later in further detail.

Differences in fimbrial gene clusters exist among serovars, and have been proposed as an additional mediator for allowing some serovars to colonize and persist in different hosts/environments

(Yue et al., 2012). The number of intact fimbrial gene clusters, as well as the type, varies among serovars. For example, Yue et al. (2012) demonstrated that host adapted/restricted serovars, such as Typhi, Dublin, Paratyphi A, Choleraesuis, and Gallinarum encode multiple non-functional fimbrial genes (Yue et al., 2012). In contrast, serovars demonstrating a broader host range (Typhimurium, Enteritidis, Montevideo, Newport, Tennessee, Kentucky, and others) have relatively few degraded fimbrial genes (Yue et al., 2012). Moreover, allelic variation of fimbrial genes has also been proposed as an adaptation associated with the colonization of certain hosts (Yue et al., 2015). For example, S. Newport isolated from bovine, porcine, equine, or avian hosts was more likely to have allelic variants A/A/A1 (representing allelic combinations of FimH/BcfD/StfH), while S. Newport isolated from humans and the environment had predominately B/B/B1 fimbrial alleles (De Masi et al., 2017). A similar observation was also reported for S. Typhimurium, where different alleles of FimH1 or FimH7 were significantly associated with adherence to different human and bovine intestinal epithelial cell lines (Yue et al., 2015). Furthermore, Yue et al. (2015) also demonstrated interserovar differences in the ability of E. coli strains expressing different Salmonella FimH alleles to adhere to host cell lines representing different mammalian species (Yue et al., 2015). For example, expression of porcine-associated Salmonella fimbrial alleles resulted in significantly higher proportions of recombinant E. coli bound to porcine epithelial cell types (Yue et al., 2015). Together, this suggests that maintenance of multiple different fimbrial genes, and allelic variants of these genes, represents a mechanism by which NTS serovars are able to adhere to a variety of surfaces, making them suited to many different environments.

### Virulence Plasmids

Nine NTS serovars have been described as harboring a low copynumber virulence plasmid. This plasmid varies among serovars, both in size and in genetic content (**Table 2**). Furthermore, carriage of the plasmid is not ubiquitous among all isolates of a given serovar (Chu and Chiu, 2006). Isolates that do carry the plasmid, though, generally exhibit increased virulence. The mediator of this increased virulence is the 7.8 kb spvRABCD (Salmonella plasmid virulence) operon, whose effectors alter the host cell cytoskeleton to enhance bacterial survival (Rotger and Casadesús, 1999). While spv is common to all Salmonella virulence plasmids, additional virulence factors or antimicrobial resistance genes may also be encoded (**Figure 5**).

The variation in content is attributed to the evolution of two distinct plasmid lineages, which have been characterized by sequence analysis, probe hybridization, and compatibility studies (Chu and Chiu, 2006). Serovars Paratyphi C (Liu et al., 2009), Sendai (Silva et al., 2017), Abortusequi (Akiba et al., 1999), and Abortusovis (Uzzau et al., 2000) also encode a virulence plasmid (**Table 2**), but the evolution of these plasmids is unclear.

The first lineage contains plasmids of S. Gallinarum (Pullorum; pSPV), and S. Dublin (pSDV) (Chu and Chiu, 2006). pSPV and pSDV harbor the replicon repB, which exhibits high similarity to the ancestral replicon IncFIIA (also called RepFIIA)

(Rotger and Casadesús, 1999; Rychlik et al., 2006). It has been proposed that the ancestral IncFIIA replicon contained the spv and tra loci, as all Salmonella virulence plasmids encode variants of these genes, although the tra genes in the current pSPV and pSDV virulence plasmids are remnants of a bacterial conjugation system that has undergone variable degradation in several plasmid lineages (Rychlik et al., 2006). By virtue of the tra locus, pSPV may be mobilized in the presence of a F plasmid (Chu and Chiu, 2006; Rychlik et al., 2006), whereas pSDV lacks homology to the tra locus (Rotger and Casadesús, 1999), and cannot be mobilized (Rychlik et al., 2006), suggesting substantial deterioration of the tra locus in pSDV.

The plasmids of serovars Typhimurium (pSTV), Enteritidis (pSEV), and Choleraesuis (pSCV) compose the second lineage (Chu and Chiu, 2006), and contain not only the repB replicon, but also elements of a second replicon, repC (Rotger and Casadesús, 1999). repC exhibits similarity to the IncFIB replicon (Rotger and Casadesús, 1999), which likely encoded the virulence factors pef, srgAB, and rck (Rychlik et al., 2006). The pef operon, or plasmidencoded fimbriae, is conserved in all three plasmids of this lineage (Rychlik et al., 2006). pef-encoded fimbriae mediate adhesion to cells within the small intestine, and exhibit preferential binding to certain host cells of various species (Baumler et al., 1996). The srgA gene product is important for the biogenesis of the pefencoded fimbriae (Bouwman et al., 2003), yet this locus has been lost in pSCV (Rychlik et al., 2006), as has rck (Rychlik et al., 2006). rck has been shown to contribute to resistance to complement

killing, one of the host's innate immune responses (Rychlik et al., 2006). As in the first lineage the tra locus exhibits variable degradation among the plasmids pSTV, pSCV, and pSEV. The tra operon of pSTV renders that plasmid capable of conjugative transfer (Rychlik et al., 2006), yet in the S. Enteritidis plasmid (pSEV), and S. Choleraesuis plasmid (pSCV) lineages the tra locus has degraded and is no longer functional (Rotger and Casadesús, 1999; Rychlik et al., 2006). The degradation of tra in pSCV and pSEV may indicate that they descended from pSTV via deletions (Chu and Chiu, 2006).

### DISEASE DIVERSITY-NONTYPHOIDAL SALMONELLOSIS DISEASE SEVERITY VARIES SIGNIFICANTLY BY SEROVAR

Although the complete host-pathogen relationship plays an important role in determining the resulting severity of salmonellosis (**Figure 2**), some NTS serovars excel at causing invasive human clinical infections that are reminiscent of the pathology exhibited by Typhi and paratyphoidal serovars of S. enterica. Invasive infections result from Salmonella's successful ability to escape the gastrointestinal tract, and subsequently spread to other tissues. These infections are considered to be the most severe, often resulting in hospitalization. Most commonly, invasive salmonellosis is defined as isolation of Salmonella from a "sterile site," usually blood, joint fluid, or cerebrospinal fluid (Jones et al., 2008). Global estimates of invasive nontyphoidal salmonellosis (regardless of serovar) range from 30 to 227 cases per 100,000 persons (Ao et al., 2015). With the exception of select NTS serovars, most invasive NTS infections are typically associated with individuals from so-called "high risk" populations (i.e.,<5 years old, >65 years old, having an immunodeficiency, or being pregnant). For example, invasive nontyphoidal salmonellosis is more frequently cited in Africa (Okoro et al., 2012; Ao et al., 2015) and Southeast Asia (Lan et al., 2016), and is associated with malnutrition, immunodeficiencies including HIV, and malaria co-infection (Scott et al., 2011; Feasey et al., 2012).

A study by Jones et al. (2008) found that, compared to S. Typhimurium infections, of which just 5.7% resulted in invasive disease, infections with serovars Choleraesuis (56.4%), Dublin (64.0%), Sandiego (18.9%), and Panama (18.0%) were associated with significantly higher rates of invasive disease among human clinical infections in the US (Jones et al., 2008). While the total number of salmonellosis cases contributed by these serovars is considerably lower than those of other serovars, the number of infections resulting in invasive disease is significantly higher, suggesting that either (i) most of these infections arise from individuals in high risk populations (i.e., only select sub-populations are susceptible to infection with these serovars, and when infection occurs it is often more severe), (ii) the majority of the cases are asymptomatic, and therefore only severe cases are reported, or (iii) these serovars have virulence factors or adaptations that make them inherently more invasive. While further research is needed to conclusively determine which components of the host-pathogen interaction specifically account for the observed invasive phenotype of these serovars, the current understanding is summarized below.

### NTS Serovars Associated With High Rates of Invasive Disease in Humans

Salmonella enterica subsp. enterica serovar Choleraesuis (S. Choleraesuis) is a host-adapted serovar, causing swine paratyphoidal disease (Chiu C.-H. et al., 2004; Pedersen et al., 2015). In humans, S. Choleraesuis often causes septicemia with minimal GI-tract inflammation (Chiu C.-H. et al., 2004), resulting in a disease more similar to typhoid fever (Sabbagh et al., 2010). This suggests that, similar to S. Typhi, the severity of the salmonellosis caused by this serovar results from successful evasion of host defenses in the gut, and thus failure of the immune system to detect and control S. Choleraesuis early in the infection. In support of this, most human clinical cases of S. Choleraesuis occur in patients with a pre-existing health condition, such as an immunosuppressive condition (Wang et al., 2006), or other chronic disease (Chiu S. et al., 2004). Like other serovars that are frequently associated with invasive human disease, S. Choleraesuis encodes a T6SS, which plays a role in virulence in vivo (Schroll et al., 2019). S. Choleraesuis strain SC-B67, originally isolated from a human patient with sepsis, has a high proportion of pseudogenes, most of which arose from SNPs resulting in premature stop codons in genes related to metabolism, fimbriae, and the chemotactic response (Chiu et al., 2005). Inactivation of several ancestral genes may limit the adaptability of S. Choleraesuis in several niches including animal hosts and environmental sites, which would account for the lack of clinical cases in animals other than pigs and humans for this serovar. While S. Choleraesuis is rarely isolated from human clinical cases in the US [∼10–20 confirmed cases per year (Centers for Disease Control and Prevention [CDC], 2016)] and the EU (Pedersen et al., 2015), cases of S. Choleraesuis are relatively more common in Southeast Asia such as in Taiwan and Thailand, although control efforts have been very successful at reducing the incidence of S. Choleraesuis infections in these areas (Hendriksen et al., 2009; Hendriksen et al., 2011).

Salmonella enterica subsp. enterica serovar Dublin (S. Dublin) is a bovine host-adapted serovar (Langridge et al., 2015). Genetic analyses have shown that S. Dublin shares a common ancestor with another host-adapted serovar, S. Gallinarum (Langridge et al., 2015), although S. Dublin has substantially fewer pseudogenes (Langridge et al., 2015). S. Dublin encodes two T6SSs located in SPI-6 and SPI-19 (Mohammed et al., 2017). In the United States, confirmed human clinical cases of S. Dublin salmonellosis contribute ∼150 reported cases per year (Centers for Disease Control and Prevention [CDC], 2016), with the majority of S. Dublin infections occurring in adults (median age 55 years, compared to median age of 23 years for infections with all other Salmonella) (Harvey et al., 2017). This suggests that either the incidence of exposure to S. Dublin is higher among older individuals, or S. Dublin is asymptomatic in younger individuals, but causes clinical disease when it encounters a host with a weakened immune system, as is common for other opportunistic pathogens. While S. Dublin infections in humans

are significantly more likely to result in hospitalization, invasive disease, prolonged hospital stays, and an increased likelihood of death (Harvey et al., 2017), the true incidence of exposure is unknown, and is likely much higher than the number of clinical infections reported each year. Nevertheless, it is intriguing that this serovar, when successful in colonizing humans, often results in invasive disease.

Infections with S. enterica subsp. enterica serovar Sandiego (S. Sandiego) and S. enterica subsp. enterica serovar Panama (S. Panama) are also associated with high rates of invasive disease (Jones et al., 2008; Crump et al., 2011). In contrast to S. Dublin and S. Choleraesuis, these serovars have not been characterized as being host-adapted. S. Sandiego has been linked to several outbreaks from handling small turtles (Walters et al., 2015), and has also been isolated from livestock including pigs (Oliveira et al., 2002), goats (El Tom et al., 1999), and cows (Boqvist and Vågsholm, 2005). S. Panama has been isolated from pigs, poultry, cows, and goats (Cordano and Virgilio, 1996). Both S. Panama and S. Sandiego encode S. Typhi-associated SPI-18 genes hlyE and taiA, and the typhoid toxin genes (cdtB, pltA, and pltB) (Den Bakker et al., 2011), although the role that these genes, and their corresponding gene products, play in invasive infections involving these serovars has not been assessed. Unlike S. Choleraesuis and S. Dublin, infections with S. Panama and S. Sandiego are more commonly diagnosed among children <5 years old (Boore et al., 2015). Therefore, either environmental exposure to S. Panama and S. Sandiego is more common in young children, perhaps as a result of animal contact, or these serovars possess virulence factors that enable them to successfully cause disease in young children, but not as frequently in adults, as is the case for serovars Choleraesuis and Dublin.

### NTS Serovars Associated With Reduced Rates of Human Clinical Disease: Evidence of Loss of Virulence

In contrast to NTS serovars associated with high rates of invasive human clinical disease, there also exist a number of serovars that are commonly isolated from agricultural reservoirs, but which cause a disproportionately low number of human clinical cases. The primary reasons for this include (i) effective kill steps such as heating or other inactivation techniques for contaminated food commodities (i.e., cooking of meat or pasteurization of milk), (ii) mutations in select serovars that effectively reduce their ability to cause disease in select hosts, and (iii) genetic adaptations that allow the serovar to colonize an animal host, but do not impart a fitness advantage allowing for disease manifestation in humans.

S. Cerro is one of the most frequently isolated serovars among dairy clinical isolates (Tewari et al., 2012; Valenzuela et al., 2017), although this serovar is responsible for just ∼30 reported human clinical cases per year in the United States (Centers for Disease Control and Prevention [CDC], 2016). In 2016, S. Cerro was the 4th most common clinical isolate among animal clinical cases reported to the USDA, and represented the 2nd most common serovar reported among clinical cases in cattle (Morningstar-Shaw et al., 2016). Despite S. Cerro being detected in bulk tank raw milk samples, outbreaks from Salmonellacontamination of raw milk in the United States usually involve serovars Typhimurium (Mungai et al., 2015), Montevideo, and Newport (Robinson et al., 2014), and the cow-adapted serovar Dublin (Harvey et al., 2017; Vignaud et al., 2017). S. Cerro isolates have a premature stop-codon in sopA (Rodriguez-Rivera et al., 2014; Kovac et al., 2017), a virulence factor shown to play an important role in Salmonella entry into host cells (Raffatellu et al., 2005). Indeed, S. Cerro isolates have a significantly lower rate of invasion in the Caco-2 cell line, compared to S. Typhimurium and S. Newport strains in a standard gentamicin protection assay (Rodriguez-Rivera et al., 2014). The comprehensive set of reasons behind the discrepant ability of S. Cerro to colonize and amplify within dairy cattle, and its low likelihood of causing clinical disease in both humans and most animals, remains unknown.

In the United States, S. Kentucky represents just 0.1% of reported human clinical cases of salmonellosis (Centers for Disease Control and Prevention [CDC], 2016), yet it is the most commonly isolated serovar from broiler chickens in the United States (United States Department of Agriculture [USDA], 2014). Although experimental evidence suggests that S. Kentucky may persist for longer periods of time in chickens than S. Typhimurium, the reasons why S. Kentucky is infrequently associated with human clinical cases of salmonellosis have yet to be confirmed (Cheng et al., 2015). One possibility for the lack of human clinical cases is that S. Kentucky isolates lack the virulence genes grvA, sseI, sopE, and sodCI (Beutlich et al., 2011; Cheng et al., 2015), which may play a role in human disease but are not necessary for colonizing chickens. Another study by Tasmin et al. (2017) using a phenotypic array found that S. Kentucky ST152 strain SK222\_32B differed significantly in its ability to utilize a number of common metabolites, namely 1,2-propanediol (Tasmin et al., 2017), which has previously been cited for enabling S. Typhimurium to successfully outcompete the resident microbiota in the mammalian gut (Faber et al., 2017). S. Kentucky SK222\_32B also failed to replicate in macrophages, implying that S. Kentucky has an impaired ability to resist immune-killing, possibly as a result of its lacking super oxide dismutase (sodCI) (Tasmin et al., 2017). Indeed, other studies have shown that, when exposed to media at pH 2.5, S. Kentucky was more sensitive to acid stress than serovars Enteritidis, Hadar, Mbandaka, Senftenberg, Typhimurium, and Worthington (Joerger et al., 2009). In contrast, human clinical cases caused by S. Kentucky are increasing in Europe. Interestingly, S. Kentucky type ST198 is more commonly isolated in human clinical cases (primarily in Europe and Northern Africa), but S. Kentucky ST152 is more commonly isolated from poultry (Le Hello et al., 2013; Haley et al., 2016). While studies suggest that ciprofloxacin resistance encoded by S. Kentucky ST198 explains its sudden expansion in human clinical cases (Le Hello et al., 2011, 2013), future studies examining the genetic and metabolic differences between the chicken-associated (ST152) and human associated (ST198) STs may reveal more conclusive evidence for why certain S. Kentucky STs are associated with different hosts.

Understanding why certain serovars are able to asymptomatically colonize a host represents an important gap in our current understanding of nontyphoidal salmonellosis. In the case

of S. Cerro, understanding the importance of mutations enabling this serovar to frequently colonize cattle, but rarely cause clinical symptoms in humans represents a key gap in our understanding of NTS virulence. For S. Kentucky, increases in human clinical infections involving a specific sequence type suggest either a new route of transmission, or possibly a novel adaptation of ST198 that may be absent in other S. Kentucky STs, which might be an important model for how other serovars that are frequently associated with animal reservoirs can adapt to cause human clinical disease.

### ADAPTING TO THE SITUATION – FOODBORNE, ZOONOTIC, AND ENVIRONMENTAL SOURCES OF NTS SEROVARS

While the vast majority of NTS cases are foodborne, infections resulting from animal contact and environmental exposures have been reported. Owing to their ability to successfully survive in a number of different environments, NTS serovars have adapted to achieve a number of different routes of transmission to cause human clinical infections. Traceback investigations aimed at linking a Salmonella isolate to its source represent a persistent challenge. Further complicating this, the majority of infections represent sporadic incidents, which cannot be linked to a common source. Here we review the foodborne, animalcontact, and environmental sources of NTS salmonellosis.

### Foodborne Nontyphoidal Salmonellosis

In the United States, approximately 94% of domestically acquired salmonellosis cases are acquired through the consumption of contaminated food (Scallan et al., 2011), with most representing sporadic cases (Ebel et al., 2016). In fact, it has been reported that just 5.9% of NTS infections in the United States are linked to an outbreak (Ebel et al., 2016). This is best represented by comparing the relative contribution of reported NTS serovars isolated from human clinical infections (**Figure 6A**) with the serovars causing outbreaks traced back to a single food item (**Figure 6B**). Data compiled between 2012 and 2018 show that in the United States, the primary NTS serovars associated with outbreaks vary significantly from year to year, and are often the result of a few large outbreaks (**Figure 6B**). The overall trend in human clinical infections remains constant however, as infections reported between 2010 and 2016 in the United States show very little fluctuation by serovar, with the top 6 serovars accounting for ∼50% of all infections each year. Therefore, the accumulation of NTS sporadic infections is likely the result of low-level Salmonella contamination from a wide range of foods, primarily by serovars Typhimurium, Newport, Enteritidis, Javiana, Infantis, and I 4,[5],12:i:-, and not due to large outbreaks involving these serovars.

Each food commodity presents its own challenges, including differences in pH, available water, temperature, nutrient composition, osmolarity, and the presence of antimicrobial compounds; combined these represent unique environmental stresses to which Salmonella must adapt. An analysis by the CDC of the food commodities most frequently associated with outbreaks of salmonellosis by serovar reveals that for certain food products, there is a strong association between select serovars and food commodities of animal origin (Jackson et al., 2013). For example, for serovars Enteritidis, Heidelberg, I 4,[5],12:i:-, and Hadar, the majority of outbreaks are associated with poultry (either chicken or turkey meat, or eggs), while serovars Infantis, Montevideo, and Uganda are more commonly associated with outbreaks linked to consumption of pork and beef (Jackson et al., 2013). This suggests that these serovars possess genetic and phenotypic traits that promote their ability to adapt to these environments. The USDA conducts routine surveillance in the United States to monitor retail-bound meat (United States Department of Agriculture [USDA], 2014), egg products (United States Department of Agriculture [USDA], 2017), and raw dairy foods (Sonnier et al., 2018) for contamination by NTS. An interesting observation is that the serovars most frequently isolated from these high-risk food sources do not mirror those predominantly responsible for infection in humans (**Figure 7**). This is readily apparent from the frequent isolation of S. Kentucky and S. Cerro from poultry meat products and dairy cows, respectively, which does not result in an equivalent predominance among human clinical infections (**Figure 7**).

A number of produce (fruits and vegetables) outbreaks have also occurred (Hanning et al., 2009), demonstrating NTS serovars' ability to survive in acidic and dry environments, as well as at a wide range of temperatures associated with the produce production chain. For example, tomatoes have been associated with several outbreaks involving serovars Montevideo, Newport, Braenderup, and Javiana (Hanning et al., 2009). Shi et al. (2007) observed differences among serovars that were able to grow on ripened tomatoes, likely due to the high acidity associated with this fruit (Shi et al., 2007). Differences in acid tolerance among NTS serovars, may therefore represent one mechanism allowing some serovars to contaminate and grow in select food commodities.

### Zoonotic Transmission of Salmonella

Several reports implicate a role for animal contact in human infection with NTS serovars. Not surprisingly, handling of chicks is responsible for a considerable number of salmonellosis cases (**Figure 8**), in which serovars Typhimurium, Johannesburg, Braenderup, Thompson, and Montevideo are frequently implicated with human outbreaks of salmonellosis (Loharikar et al., 2012; Pabilonia et al., 2014; Habing et al., 2015; Nakao et al., 2015; Anderson et al., 2016; Sharma et al., 2018). Zoonotic infection may also result from exposure to infected companion animals (kittens, guinea pigs, hedgehogs, and turtles) (**Figure 8**; Centers for Disease Control and Prevention [CDC], 2001; Wright et al., 2005; Bartholomew et al., 2014; Anderson et al., 2017; Gambino-Shirley et al., 2018). Interestingly, NTS infections arising from contact with mammals have been largely linked to serovars Typhimurium and Enteritidis (Hoelzer et al., 2011) (the two serovars causing the majority of foodborne salmonellosis in humans), while zoonoses contracted from turtles included infection with serovars Sandiego, Poona, and

FIGURE 6 | NTS causing overall vs. confirmed outbreak-associated infections in humans. The contribution of select NTS serovars to (A) overall human clinical infections vs. (B) human clinical infections confirmed from a foodborne outbreak are shown as percent infection for a given category. Data are modified from Centers for Disease Control and Prevention [CDC] (2016, 2018).

Pomona (Gambino-Shirley et al., 2018), which infrequently cause human clinical cases.

## Environmental Transmission of Salmonella

Although less common, nontyphoidal serovars have been isolated from soil, water, contaminated floors, carts, and other equipment-related sources (Cummings et al., 2014; Strawn et al., 2014; Liu et al., 2018). Environmental sources of NTS serovars often arise from the introduction of NTS-contaminated animal feces, either via contaminated water used to irrigate crops, or from direct contact with feces from the Salmonella-secreting animal. In the case of several teaching hospitals, nosocomial transmission has been confirmed between animals, including different species, implicating the ability of NTS serovars to survive and transit from different environmental sources, to a susceptible host (Cummings et al., 2014). Furthermore, albeit rare, human-human cases of NTS transmission have also been documented (Kariuki et al., 2006).

Outbreaks of foodborne salmonellosis have also been documented in which biofilms harboring NTS serovars were linked to contamination of food products. In 2004 and 2005, two outbreaks involving S. Agona in powdered infant formula were traced back to environmental contamination in the production plant where the formula was produced (Brouard et al., 2007). S. Agona has also been linked to two outbreaks in dry cereal, in 1998 and 2008 (Russo et al., 2013), resulting from S. Agona

persistence in the food processing plant. Importantly, these outbreaks highlight the potential for NTS serovars to persist in food processing plants for extended periods of time, and then contaminate food products.

Depending on the environment, transmission of NTS serovars from the environment onto a food surface represents an important, yet indirect, mechanism of transmission for a number of food commodities. While adaptations that allow NTS serovars to persist in different environments, such as Salmonella's ability to survive in dry foods/environments, are largely dependent on the specific food (i.e., fat content and moisture content, etc.), there are significant differences in the abilities of different serovars to survive in foods.

### REDUCING NTS INCIDENCE WILL REQUIRE A DIVERSIFIED APPROACH

The broad host-range displayed by a number of NTS serovars represents a key challenge in reducing nontyphoidal salmonellosis, as interventions aimed at targeting a single reservoir are often unsuccessful at eliminating Salmonella from other reservoirs, many of which are unknown. Furthermore, successful control of one serovar has historically been associated with increases in the number of human clinical cases and outbreaks caused by other serovars due to the vacancy of a previously occupied niche. For example, targeted control efforts aimed at eliminating S. Gallinarum (biovars Gallinarum and Pullorum) infections among chickens paralleled the sudden increase in S. Enteritidis, which is able to colonize chickens without causing overt clinical disease (Baumler et al., 2000). Therefore, successful control of one serovar in a given niche may enable the expansion of a new serovar. Complicating this, little is known about the natural reservoirs of the less-studied NTS serovars.

The vast majority of what is known about the virulence factors and mechanisms of NTS serovars is derived from experiments performed with NTS serovar Typhimurium, which have provided beneficial data that have expanded our understanding of how Salmonella cause disease. However, the genetic differences among serovars suggest that a number of mechanisms are serovarspecific, and therefore, improved approaches should consider the importance of using specific serovars (Jones et al., 2008; Suez et al., 2013). Identifying what makes certain serovars better adapted to persist in different environments also represents a key piece in resolving the NTS-contamination puzzle, as many serovars have yet to be associated with their reservoir(s).

The collective effort to reduce-foodborne nontyphoidal salmonellosis will require a collaboration of both laboratory science, generating experimental evidence about how diverse NTS serovars are able to overcome unfavorable environmental conditions to establish infection, and public health authorities utilizing novel methods for enhanced outbreak detection and source tracking. Furthermore, increases in the diversity and number of publicly available WGS data will enable the use of genetic tools (i.e., CRISPR, Lambda-Red mutagenesis, etc.) necessary to carry out genetic and phenotypic analyses among non-Typhimurium NTS serovars, allowing for the development of novel control and treatment strategies that will aid in reducing the overall morbidity and mortality associated with NTS on a global level.

As foodborne transmission represents the most common route of infection with NTS serovars, risk-based research approaches aimed at identifying and targeting specific serovars likely to contaminate a given food product represents an attractive approach to reducing human clinical cases. For example, the synthesis of data regarding contamination of a specific food commodity, such as S. Enteritidis and eggs, can be used to identify key sources of contamination so that research and regulatory initiatives may be focused to control contamination with specific serovars in these niches. Alternatively, assessments based on disease severity may be used to inform new regulatory decisions, such as imposing more stringent limits for some serovars (e.g., those associated with high rates of invasive disease), or relaxed limits for less virulent serovars (e.g., Cerro) or for specific strains showing virulence-attenuating mutations.

## CONCLUSION

While the global burden of nontyphoidal salmonellosis remains as one of the key challenges, novel methods for studying Salmonella have changed the way that the research community detects, investigates, and prevents salmonellosis. Reflecting the diversity of virulence factors, genetic and phenotypic adaptations among the more than 2,600 serovars, reducing contamination with NTS serovars represents a multi-faceted challenge that will require collaborations of industry, government, and academia to achieve the ultimate goal of reducing NTS human clinical infections worldwide.

### AUTHOR CONTRIBUTIONS

fmicb-10-01368 June 24, 2019 Time: 15:17 # 15

RC, CE, and MW conceived and revised the manuscript.

### FUNDING

CE was supported by the NIH/USDA-NIFA Dual Purpose with Dual Benefit program; grant number 2014-67015-21697 awarded to Dr. Craig Altier. This material is based upon work that was supported by the National Institute of Food and Agriculture,

### REFERENCES


United States Department of Agriculture, under the award number 2016-67012-25184 (to CE).

### ACKNOWLEDGMENTS

The authors would like to thank Dr. Craig Altier for his comments and suggestions during the preparation of this manuscript, Dr. Renato Orsi for assistance in the preparation of figures, and Dr. Laura Carroll for helpful discussions about S. Kentucky.

within infected macrophages. Infect. Immun. 81, 1207–1220. doi: 10.1128/IAI. 01165-12



foodborne illnesses. United States, 2004-2011. Emerg. Infect. Dis. 22, 1193–1200. doi: 10.3201/eid2207.150833



one major difference between serovar Kentucky and other serovars: response to acid. Foodborne Pathog. Dis. 6, 503–512. doi: 10.1089/fpd.2008.0144



Salmonella Typhimurium pathovariants in sub-Saharan Africa. Nat. Gen. 44, 1215–1221. doi: 10.1038/ng.2423




**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 Cheng, Eade and Wiedmann. 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.

## Antibiotic Resistance Profiles of *Salmonella* Recovered From Finishing Pigs and Slaughter Facilities in Henan, China

### *Edited by:*

*Om V. Singh, TSG Consulting, Washington, DC, United States*

#### *Reviewed by:*

*Xiang Yang, University of California, Davis, United States Burkhard Malorny, Federal Institute for Risk Assessment (BfR), Germany Silvia Bonardi, University of Parma, Italy*

#### *\*Correspondence:*

*Min Yue myue@zju.edu.cn Qigai He he628@mail.hzau.edu.cn*

*† These authors have contributed equally to this work*

#### *Specialty section:*

*This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology*

*Received: 22 January 2019 Accepted: 17 June 2019 Published: 04 July 2019*

#### *Citation:*

*Jiang Z, Paudyal N, Xu Y, Deng T, Li F, Pan H, Peng X, He Q and Yue M (2019) Antibiotic Resistance Profiles of Salmonella Recovered From Finishing Pigs and Slaughter Facilities in Henan, China. Front. Microbiol. 10:1513. doi: 10.3389/fmicb.2019.01513*

*Zenghai Jiang1,2† , Narayan Paudyal 3,4† , Yaohui Xu 2 , Tongwei Deng 2 , Fang Li 3 , Hang Pan3 , Xianqi Peng3 , Qigai He1 \* and Min Yue3,5 \**

*1 Division of Animal Infectious Diseases, State Key Laboratory of Agricultural Microbiology, College of Animal Sciences and Veterinary Medicine, Huazhong Agricultural University, Wuhan, China, 2 College of Veterinary Medicine, Henan University of Animal Husbandry and Economy, Zhengzhou, China, 3 College of Animal Sciences, Institute of Preventive Veterinary Medicine, Zhejiang University, Hangzhou, China, 4 Animal Health Research Division (AHRD), Nepal Agricultural Research Council (NARC), Kathmandu, Nepal, 5 Zhejiang Provincial Key Laboratory of Preventive Veterinary Medicine, Hangzhou, China*

With the increase in commercial pig farming, there is a simultaneous increase in the use of antibiotics for prophylaxis as well as therapeutics in China. In this study, we evaluated the prevalence and resistance diversity of salmonellae isolated from feces of asymptomatic, live and slaughtered pigs. We analyzed 1,732 pig fecal samples collected over 8 months, at Henan province of China. The salmonellae were isolated and identified by PCR. They were serotyped using commercial antisera and assayed for the MIC of 16 antibiotics by broth microdilution method. The average prevalence of *Salmonella* was 19.4% (95% CI: 17.6–21.4). Large farms (herd size ≥1,000) were found to have a higher prevalence as compared to the small- and medium-scale farms (*p* < 0.0001). The prevalence of salmonellae in samples collected from the farms [11.77% (95% CI: 10.1–13.6)] and from the slaughterhouse [45.23% (95% CI: 40.3–50.30)] was statistically different (*p* < 0.0001). Uncommon serovars of *Salmonella* such as Agama and common serovars such as Derby and Typhimurium were isolated. High resistance (>80%) was recorded toward ciprofloxacin (100%), tetracycline (99.4%), doxycycline (97%), sulfamethoxazole (85.8%), ampicillin (81.6%), and amoxicillin (80.4%). Multidrug resistance (MDR) to four, five, and seven classes of antibiotics was recorded to be approximately 25% in the most prevalent serovar like Derby. We conclude that the presence of alarmingly high resistance, toward the critical antibiotics such as fluoroquinolones and beta-lactams, in large swine farms in China, should draw public attention. These results highlight the need for continued antibiotic stewardship programs for judicious use of critical antibiotics in animal health as well as for producing safe pork.

Keywords: *Salmonella*, multidrug resistance, China, quinolones, beta-lactamase, prevalence

### INTRODUCTION

Fuelled by the rapid socio-economic growth and urbanization of China, the average intake of meat, especially pork, increased from 37.1 g/day in 1992 to 64.3 g/day per person in 2012 (He et al., 2016). With the population growth projected to reach 1.4 billion in 2020, the consumption of pork is also expected to surpass 100 g/day per person by 2020 (CIIN, 2018a). In 2017, Henan province alone contributed over 65 million live pigs, approximately 20% of the whole national consumption (CIIN, 2018b). The boost in pig production and consumption in China has indirectly increased the risk of foodborne zoonoses, including salmonellosis. Previous studies have shown that *Salmonella* is one of the leading foodborne pathogens on food commodities, meat in particular, and play a significant role for causing human diarrhea in China and elsewhere (Pan et al., 2018, 2019; Paudyal et al., 2018).

Routine surveillance of potential hazards is essential to minimize the risks of disease epidemics as well as other potential threats such as pathogens and antibiotic resistance (AR). The finishing herds usually ready for slaughter require special attention in the surveillance purpose because they could disseminate the pathogens, if any they carry in the gut. Studies in other parts of the world have investigated the carriage of salmonellae in apparently healthy, asymptomatic pigs in the finishing herds (Baptista et al., 2009, 2010, 2011; Wilhelm et al., 2012). There are numerous studies about antibiotic-resistant salmonellae in pigs in China. The prevalence of AR reported in these studies varies greatly but generally is reported toward critical drugs, including quinolones and cefotaxime (Bai et al., 2016), tetracyclines (Jiu et al., 2017), quinolones, and cephalosporins (Jiang et al., 2014) or toward trimethoprim-sulfamethoxazole, ampicillin, and tetracycline (Su et al., 2018).

Pathogens and AR can arise from multiple sources, for example, the endemic pathogens circulating in the farm or pathogens introduced through the feed, water, workers, and equipment. Slaughterhouse is one of the most important risk factors that can act as a mixing vessel for any kinds and numbers of pathogens including *Salmonella* present in animals collected from different unrelated farms. Under favorable circumstances, such pathogens can be disseminated *via* the meat to the consumers.

Previous studies on similar topics are essential for understanding the scope and magnitude of salmonellae prevalence, as well as the corresponding AR crisis in China. However, most of those studies have focused on clinically diseased pigs or retail-meat/pork. The overall burden of foodborne pathogens, including *Salmonella*, plays a significant threat to food safety and public health. Because Henan contributes a major segment of pork production in the Chinese national scenario, it is necessary to assess and inform the stakeholders on the possible prevalence of zoonotic pathogens like *Salmonella* recovered from swine finishing herds and slaughter facility. Additionally, co-existence of different models of farming systems, for example, small households manual farms with less than 100 animals per farm to large intensive automated commercial farms with more than 10,000 animals per farm, demand different strategies to prevent and curtail the prevalence of pathogens like *Salmonella* and AR. With an aim of minimizing these knowledge gaps, we examined the distribution, prevalence, and AR patterns of *Salmonella enterica* recovered from farms, and slaughterhouses in Henan province, one of the Chinese leading pig/pork producers.

### MATERIALS AND METHODS

This study was carried out in a cross-sectional observational design, to estimate the prevalence of various serovars of salmonellae in pig feces. The quantification of the minimum inhibitory concentration (MIC) of 16 molecules belonging to nine different classes of antibiotic agents was conducted simultaneously.

### Sample Collection

From March 2017 to November 2017, a total of 1,732 fecal samples (approximately 100 g per animal) were collected from randomly selected 36 small- and medium-scale farms (SMS-farms, housing <1,000 head of pigs), nine large-scale farms (housing ≥1,000 head of pigs), and two pig slaughtering facilities at Henan province in China (slaughtering capacity of 3,000 heads of pigs/day). In pig farms, 1,334 fresh fecal samples (499 from large-scale farms and 835 from SMS-farms) were collected from apparently healthy or asymptomatic finishing pigs by non-invasive sampling technique. In addition, a total of 398 rectal fecal samples were collected from two slaughterhouse facilities. The samples were collected aseptically and processed on the same day of collection.

### Isolation and Identification

The primary culture and isolation of the organism were done according to the protocol recommended by the World Organization for Animal Health Terrestrial Manual (OIE, 2016). This method consists of pre-enrichment in buffered peptone water (BPW) followed by enrichment in modified semisolid Rappaport Vassiliadis (MSRV) plate and final isolation on xylose lysine deoxycholate (XLD) agar plate. The obtained isolates were then confirmed by polymerase chain reaction (PCR). To elaborate, 10 g of feces were added to 90 ml of BPW and incubated at 37°C for 18–20 h. Three hundred microliters of thus incubated BPW was transferred to MSRV semisolid agar plate and incubated at 41.5°C for 24 h. An opaque halo in the MSRV medium is indicative of growth. A 10 μl loopful of the bacterial growth from MRSV medium was transferred to XLD agar plates using disposable sterile inoculating loops and these plates were incubated further for 18–20 h at 37°C. Typical round red colonies with black center on XLD media were considered presumptive of salmonellae. Other variant forms such as transparent single colonies, transparent pink colonies or transparent colorless colonies were sub-cultured on Brilliant Green Agar plates for two more times until they were declared negative and discarded. We picked up three isolated pure colonies for each sample and sub-cultured them for any mixed populations. Upon obtaining homogeneous colonies in all the three plates, only one pure isolated colony from the plate number one, was taken and purified further with one more passage in XLD plates. The pure colonies were then seeded in Luria-Bertani (LB) broth for serotyping and DNA extraction by boiling method. PCR using the primers for amplification of the enterotoxin *stn* gene for the confirmation of the *Salmonella* was performed as recommended (Zhu et al., 2015).

### Serotyping

The PCR confirmed salmonellae were serotyped according to White-Kauffmann-Le Minor scheme by slide agglutination to define O and H antigens using commercial antisera (S & A Reagents Lab, Bangkok, Thailand). Strains that did not express phase 2 flagellar antigen even after three attempts of phase inversion were considered monophasic.

### Antibiotic Susceptibility Test

The *Salmonella* isolates were subjected to antibiotic MIC assay using the broth microdilution method following the recommended procedure (CLSI, 2012). Sixteen different molecules of antibiotics belonging to nine classes were used for the MIC assay. The cutoff values recommended by CLSI (CLSI, 2016) except for enrofloxacin (Hao et al., 2013) and colistin (EUCAST, 2019) were used for the categorization of the results (**Figure 1**). The intermediate category was merged with the resistant ones for the analysis of our data. The classes of antibiotics and the MIC range (mg/L) used in our assay included penicillin (ampicillin or AMP, 0.5–64); beta-lactams (amoxicillin or AMC, 0.25–128); cephems (cefotaxime or TAX, 0.125–128; ceftriaxone or CRO, 0.125–128); aminoglycosides (gentamicin or GEN, 0.125–128; kanamycin or KAN 1–128; streptomycin or STR, 1–128); tetracyclines (tetracycline or TET, 0.25–128; doxycycline or DOX, 0.25–128); quinolones (ciprofloxacin or CIP, 0.06–16; enrofloxacin or ENR, 0.06–16); sulphonamides (cotrimoxazole or COT, 1–512; sulfamethoxazole or SMX, 8–1,024); polypeptide (colistin or CTE, 0.125–4); and phenicols (chloramphenicol or CHL, 0.5–128; florfenicol or FLO, 0.5–128).

### Data Analysis

The obtained numerical data were analyzed using Student's *t*-test and ANOVA in GraphPad Prism vs 7 for Windows platform. The alpha (*p*) level of less than 0.05 was considered significant and they are given as exact values. The difference in the prevalence of the two types of farms (SMS-farms and L-farms) was compared using a two-sample independent *t*-test. A Chi-square test was used to compare the prevalence among the L-farms, SMS-farms, and the slaughterhouse. The difference in the cumulative prevalence of resistance among the salmonellae, isolated from L-farms, SMS-farms, and the slaughterhouse with respect to the individual antibiotic agent was compared using two-way ordinary ANOVA.

### RESULTS

### Prevalence and Serovar Distribution

During the sampling period of 8 months (March to November 2017), a total of 1,732 fecal samples (398 from the rectum of slaughtered pigs and 1,334 from live pigs) were collected and analyzed. The spatial distribution of these sampling sites is shown in the geographical map of Henan in **Figure 2**. This distribution shows that our sampling frame covered a larger geographical area within Henan; hence, we believe that these results are representative of the total population. A total of 337 samples (180 rectal samples collected at the slaughterhouse,


FIGURE 1 | The distribution of the various minimum inhibitory concentration (MIC) levels of the *Salmonella* against 16 antibiotics. The names of the antibiotics (second column) are abbreviated as ampicillin (AMP), amoxicillin (AMC), cefotaxime (TAX), ceftriaxone (CRO), gentamicin (GEN), kanamycin (KAN), streptomycin (STR), tetracycline (TET), doxycycline (DOX), ciprofloxacin (CIP), enrofloxacin (ENR), sulfa-trimethoprim (COT), sulfamethoxazole (SMX), colistin (CTE), chloramphenicol (CHL), and florfenicol (FLO). The third column shows the average resistance (in percentage); the second row is the range of the MIC tested; and the values in each cell show the percentage of strains in that particular MIC dilution level. The vertical bar indicates the cutoff level of the minimum inhibitory concentration for each antibiotic at the highest value of that particular cell's dilution for susceptibility (equal to or less than) and resistance (greater than). For the sake of clarity and facilitate analysis, the intermediate category was merged with the resistant.

87 samples from large-scale farms, and 70 from SMS-farms' fresh fecal samples) tested positive for *Salmonella*. The average prevalence of *Salmonella* was 19.4% (95% CI: 17.6–21.4) in this study (337 positives out of 1,732 samples, **Table 1**). The isolates belonged to 10 different types of serovars.

The difference in the prevalence of salmonellae isolates from the fecal samples collected at the farms [11.77, 95% CI: 10.1–13.6 (*n* = 157, *N* = 1,334)] and at slaughterhouse [45.23 95% CI: 40.3–50.30 (*n* = 180, *N* = 398)] was compared using the two-sample independent *t*-test. The analysis showed that the higher prevalence in slaughterhouse samples was statistically significant (*p* < 0.0001). Among the 45 farms included in the study, nine farms were large commercial farms (L-farms) with ≥1,000 heads of live pigs while 36 were small- and medium-scale farms (SMS-farms) with <1,000 heads of live pigs. The prevalence of salmonellae in L-farms was 17.43% (95% CI: 14.2–21.1), whereas in SMS-farms, it was 8.38% (95% CI: 6.6–10.5). The results showed that the higher prevalence in L-farms is statistically highly significant (*p* < 0.0001). The comparison of prevalence among the L-farms, SMS-farms, and the slaughterhouses revealed that the difference in prevalence among these was statistically significant (*p* < 0.0001).

We differentiated 10 various serotypes of salmonellae in the total 337 positive isolates. The distribution of these serovars at different sampling sites is given in **Table 1**. Among the total samples collected (*n* = 1,732), Derby was the most prevalent serovar (8.14%, *n* = 141) while the least prevalent one was serovar Agama (0.06%, *n* = 1). Serovars such as Chester and London were isolated only from the slaughterhouse samples, while serovars like Agama and Agona, were isolated only from the farm samples. Serovars Bovismorbificans, Derby, Emek, Newlands, Rissen, and Typhimurium were isolated from both sample types. Among the total positive isolates (*n* = 337), serovar Derby was the most prevalent (26.1%, *n* = 88) in the samples from the slaughterhouse, Typhimurium was more prevalent (16%, *n* = 54) in the farms' samples. The spatial (farms and slaughterhouses) variation in the distribution of prevalence of these serovars is given in **Figure 3A**.

### Antibiotic Resistance and Multidrug Resistance Pattern

Broth microdilution MIC assay was performed for 16 antibiotic agents of nine different classes. The prevalence of resistance among *Salmonella* isolates against the different classes of antibiotics is shown in **Figure 3B**. Notable was the presence of high resistance toward some clinically important antibiotics such as fluoroquinolones and beta-lactams. Low resistance rates (<3%) against cephems and colistin were recorded. The overall distribution of the MIC values for these antibiotics is shown in the squashtogram in **Figure 1**. The results showed that this difference in the overall average resistance among the sampling sites (L-farms, SMS-farms or SH) was statistically significant (*p* = 0.0128), whereas the difference in resistance to individual antibiotic agents was statistically highly significant, *p* < 0.0001 (**Figure 4**).

The resistance among the serovars isolated from the slaughterhouse samples was compared for further analysis. It was seen that while the resistance among the serovars was statistically significant (*p* = 0.001), the difference among the antibiotics was highly significant (*p* < 0.0001) (**Figure 5A**). When the resistance between two types of farms were compared, variation in the resistance among the serovars was statistically non-significant (*p* = 0.2984), whereas the variation between the individual antibiotics was statistically highly significant (*p* < 0.0001) (**Figure 5B**).

The tetra-, penta-, and hepta-drug resistance patterns were analyzed for the isolates shown in **Figure 6A**. Tetra-resistance pattern (ASSuT, i.e., resistance to ampicillin, streptomycin, sulfamethoxazole, and tetracycline) was the most frequently (maximum of 25.8% for Derby) seen among the serovars. Penta-drug resistance (ACSSuT, i.e., ASSuT with chloramphenicol) was highest (23.4%) for serovar Derby. Heptadrug resistance (i.e., ACSSuT with amoxicillin and ceftriaxone) was recorded only in serovar Typhimurium (1.8%).

The multidrug resistance (MDR) patterns were also compared in relation to the origin of the isolates (L-farms, SMS-farms, and slaughterhouses). The tetra- and penta-drug resistance patterns were high (>50%) in L-farms for Typhimurium and high (43%) in SMS-farms for Derby. Uncommon serovars like Agama (only on L-farms) and Emek (only on SMS-farms) were 100% MDR (**Figure 6B**). The prevalence of tetra- and penta-resistance patterns in the serovars isolated from slaughterhouse was more than 40% except in serovar Rissen (**Figure 6C**).

TABLE 1 | Distribution of samples and various serovars at different sampling sites.


*SH, slaughterhouse; Prev., prevalence; FL, farm level; SL, site level; Ov., overall.*

percentage. Abbreviated names of the serovars are Bovis\*: Bovismorbificans and Typhi\*: Typhimurium. The names of the antibiotics (XX′) are abbreviated as ampicillin (AMP), amoxicillin (AMC), cefotaxime (TAX), ceftriaxone (CRO), gentamicin (GEN), kanamycin (KAN), streptomycin (STR), tetracycline (TET), doxycycline (DOX), ciprofloxacin (CIP), enrofloxacin (ENR), sulfa-trimethoprim (COT), sulfamethoxazole (SMX), colistin (CTE), chloramphenicol (CHL), and florfenicol (FLO).

### DISCUSSION

China produces and consumes more than half (at about 500 million heads per annum) of the global pork demand (Larson, 2015). Henan is one of the largest pig producing provinces of China. There are more than 3,000 farms (rearing 100–10,000 heads of pigs) contributing almost 62% of the production, while approximately 583 farms with a herd size of more than 10,000 contribute around 18% of the production (CIIN, 2018b). Pigs and pork from Henan are also distributed to other neighboring provinces like Zhejiang, Shanghai, Anhui, and Shanxi or to the south as far as Guangdong. These distribution channels also serve as a conduit for the transmission of pathogens as well as antibiotic-resistant strains from the farms to the distributors and the consumers. Thus, the prevalence of the pathogen in Henan pig/pork not only has a local but also widespread regional implication. Numerous earlier studies have reported on the spatiotemporal variation of the prevalence of salmonellae on different types of meat, foods, and animals across China (Yan et al., 2010; Li et al., 2013, 2014; Lai et al., 2014; Bai et al., 2015; Kuang et al., 2015; Cai et al., 2016; Jiu et al., 2017; Xu et al., 2017; Yi et al., 2017; Su et al., 2018). These studies suggested that pathogenic bacteria from animals can transmit AR determinants to strains isolated from humans.

In this study, the prevalence of the serovar Derby was the highest (41%, 141/337). An earlier study from Henan which analyzed eviscerated pig carcass swab samples reported that *S*. Typhimurium was the most prevalent serovar followed by Derby (Bai et al., 2015), whereas another study showed that Derby was the most prevalent serovars in pork (Zhu et al., 2019). For a comparison, serovar Typhimurium ranked first and Derby ranked the third most frequent cause of non-typhoidal salmonellosis in human patients in Henan (Xia et al., 2009). There are no reports on the presence of serovars such as Agama and Chester in Henan pigs, until date. Serovar Derby

FIGURE 4 | Distribution of the average resistance (in percent) of individual antibiotics in *Salmonella* isolates from samples collected across large farms (L-farms), small- and medium-scale farms (SMS-Farms), and slaughterhouses (SH).

has also been reported to have the highest prevalence in the pig/pork from Jiangsu (Li et al., 2014) or from Sichuan (Li et al., 2013). Jiangsu and Henan are the neighboring provinces and there is the trade of pig/pork between these two places. Sichuan is one of the top producers of pig/pork in China. It has multiple inter-provincial trades, which are likely linked to each other along the pork value chain.

The prevalence of salmonellae was higher in large farms as compared to smaller farms and the difference was statistically significant. Common knowledge dictates that larger farms have stronger biosecurity measures and generally implement good farming practices, which should have minimized the prevalence of pathogens. However, the farm level prevalence is a multifactorial outcome and factors such as stocking density, endemic contamination of the farm, histories of infection in the past, and flawed biosecurity measures could all be responsible for this increased prevalence. Antibiotics are frequently used to act as insurance to prevent a disease epidemic that is associated with overcrowding and poor sanitation in pig production. This can lead to lower apparent prevalence but probable longer shedding stage (Patterson et al., 2016) and increased resistance in small farms, which is evident in our analysis. The situation is further exacerbated because some antibiotics are widely used as a feed supplement primarily for growth stimulation. This could be the prime reason for the presence of almost similar prevalence rates of resistance to multiple antibiotics in the SMS- and large farms.

In our analysis, samples collected from the rectum of the slaughtered pigs had a higher prevalence of salmonellae as compared to the fresh fecal samples collected on the farm. The slaughterhouses from where the samples were collected for this study, provided service to multiple farms (even those beyond our sampling frame) throughout Henan. It was reported earlier that in apparently healthy flocks harboring *Salmonella*, 5–30% of the animals may still excrete the bacteria at the end of the finishing period which could probably double during transport and lairage (Berends et al., 1996). Stress due to transportation, along with a number of factors like environmental

chloramphenicol; S, streptomycin; Su, sulfamethoxazole; T, tetracycline; Amc, amoxicillin, and Cro, ceftriaxone. Abbreviated names of the serovars are Bovis\*: Bovismorbificans and Typhi\*: Typhimurium.

contamination and dose-response parameters, is mentioned to affect the pathogen secretion during lairage and subsequent slaughter (Simons et al., 2016).

The presence of less common serovars such as Agama (*n* = 1), Agona (*n* = 2), and Emek (*n* = 3) in our study shows that pigs not only contribute specific pathogenic serovars such as Typhimurium or Derby to humans but also harbor atypical strains. These atypical strains also showed higher levels of resistance for most of the common antibiotics used in pig husbandry in Henan. This is of concern because atypical strains, which are apparently harmless to pigs or humans, under the selective pressure of different antibiotics used in pig husbandry could play a significant role in the transmission of resistance to other pathogenic strains. Increase in AMR genes under selective pressure in swine fed with medicated feed has been previously reported (Looft et al., 2012). Currently, there are more than 20 classes of antibiotics with over 100 generic molecules but not all are equally used in animal husbandry (Zhang et al., 2015). Antibiotics are usually incorporated in animal feed at low concentrations (2.5–125 mg/kg) as growth promoters for an undefined time period (Marshall and Levy, 2011). An earlier report on a survey of 60 pig farms of various sizes during 2007–2008 in Henan province suggested that antibiotics were used in all stages of pig production, from weaning, to growing and fattening, as additives in animal feed, *via* drinking water or individual injection. The major classes of antibiotics used were tetracyclines, penicillins, sulfonamides, ceftiofur, and florfenicol (Zhang et al., 2015; Wu, 2018).

The occurrence of tetra- or penta-drug resistance in high frequency in the serovars Derby and Typhimurium decrease the chances of success of therapeutics at the farm level. The carriage of these MDR strains (ASSuT, ACSSuT, or ACSSuTAmcCro) by the pigs could potentially contaminate the downstream along the slaughter-chain. As the Henan pigs/ pork are distributed over wider geography in China, this can eventually lead to a spread of the resistance strains throughout the pig/pork food chain.

Our results showed that the effect of the sampling site had a significant influence on the prevalence of resistant bacteria. A study from southern China reported that in typical large-scale commercial pig farms, the most frequently used "in feed" antibiotics were tetracyclines, bacitracin, or sulfonamides (Zhou et al., 2013). Yet another analysis from 2015 reported that in the various regions of China, amoxicillin had the largest use in animals as well as in humans, whereas florfenicol and quinolones were the most common antibiotics used in pig farms (Zhang et al., 2015). We can, therefore, infer that there is a high selection pressure of antibiotics in the swine gut thereby causing the appearance of high resistance to common "in feed" or therapeutic antibiotics such as sulfa-trimethoprim, tetracyclines, amoxicillin, ampicillin, and chloramphenicol. The antibiotic use in animal production is a key contributor to the growth of animal production to ensuring food security in China over the last four decades (Wu, 2018). This usage rate has also increased with the expansion of animal production and with more intensive livestock production. This has inadvertently led to an increase in the risk of zoonotic pathogens such as *Salmonella* with high MDR capable of emergence and dissemination *via* the food systems.

Unlike earlier studies that reported the presence of increasing levels of resistance by salmonellae to cephems class of antibiotics (Jiu et al., 2017), the strains in our collection were highly susceptible to this class of antibiotic. This finding is in accordance with regional differences and variations in the selection pressure on different farms (Zhou et al., 2013; Collignon and Voss, 2015). The resistance of atypical strains such as Agama to all other classes of antibiotics except the cephems is also an interesting finding in the sense that this serovar is not generally associated with mammals but rather with reptiles and amphibians. Its presence and the general resistance is a matter of concern and further assessments.

### CONCLUSION

The prevalence of salmonellae isolates differed significantly according to the farm size, with large farms having more salmonellae as compared to the small farms with the maximum prevalence in the slaughterhouse samples. The presence of high level of resistance to critical antibiotics such as quinolones and beta-lactams in the various serovars of salmonellae isolated from apparently healthy/slaughtered pigs possibly illustrates the risk of acquisition and dissemination of such MDR *via* the pork food chain. Presence of highly MDR atypical serovars

### REFERENCES


exacerbates this risk even further as they might be acting as the "dark matters" for the resistance (Paudyal and Yue, 2019). This could eventually lead to therapeutic failures in animals as well as humans even for the clinically critical antibiotics.

### ETHICS STATEMENT

No ethical approval was deemed necessary for this study. Oral agreement and permission were obtained from the farmers as well as the slaughterhouse manager before sampling.

### AUTHOR CONTRIBUTIONS

ZJ, YX, and TD collected the samples. ZJ, NP, FL, and HP did the lab analysis. NP, XP, and QH did the data analysis and prepared a draft. MY conceived the project and provided critical comments for the draft. All the authors have read and agreed to the manuscript.

### FUNDING

This study was supported by the National Program on Key Research Project of China (2017YFC1600103; 2018YFD0500501), the Fundamental Research Funds for the Central Universities (2-2050205-18-237), Zhejiang Provincial Natural Science Foundation of China (LR19C180001), Zhejiang University of Hundred Talent Program, National Recruitment Program of Global Youth Experts, Key Discipline of Preventive Veterinary Medicine of Henan University of Animal Husbandry and Economy (MXK2016102), and China Agriculture Research System (CARS-35).

### ACKNOWLEDGMENTS

The farmers and the slaughterhouse staffs who assisted during the sample collection are duly acknowledged for their cooperation for this study.

status in pig herds. *Zoonoses Public Health* 57(Suppl. 1), 49–59. doi: 10.1111/j.1863-2378.2010.01354.x


Network. Available at: http://www.chyxx.com/industry/201801/603323.html (Accessed January 9, 2019).


**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 Jiang, Paudyal, Xu, Deng, Li, Pan, Peng, He and Yue. 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.*

# Assessment and Comparison of Molecular Subtyping and Characterization Methods for Salmonella

Silin Tang<sup>1</sup> \*, Renato H. Orsi<sup>2</sup> , Hao Luo<sup>1</sup> , Chongtao Ge<sup>1</sup> , Guangtao Zhang<sup>1</sup> , Robert C. Baker<sup>1</sup> , Abigail Stevenson<sup>1</sup> and Martin Wiedmann<sup>2</sup>

<sup>1</sup> Mars Global Food Safety Center, Beijing, China, <sup>2</sup> Department of Food Science, College of Agriculture and Life Sciences, Cornell University, Ithaca, NY, United States

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Min Yue, Zhejiang University, China Dapeng Wang, Shanghai Jiao Tong University, China Soohyoun Ahn, University of Florida, United States

> \*Correspondence: Silin Tang silin.tang@effem.com

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 20 March 2019 Accepted: 26 June 2019 Published: 12 July 2019

#### Citation:

Tang S, Orsi RH, Luo H, Ge C, Zhang G, Baker RC, Stevenson A and Wiedmann M (2019) Assessment and Comparison of Molecular Subtyping and Characterization Methods for Salmonella. Front. Microbiol. 10:1591. doi: 10.3389/fmicb.2019.01591 The food industry is facing a major transition regarding methods for confirmation, characterization, and subtyping of Salmonella. Whole-genome sequencing (WGS) is rapidly becoming both the method of choice and the gold standard for Salmonella subtyping; however, routine use of WGS by the food industry is often not feasible due to cost constraints or the need for rapid results. To facilitate selection of subtyping methods by the food industry, we present: (i) a comparison between classical serotyping and selected widely used molecular-based subtyping methods including pulsed-field gel electrophoresis, multilocus sequence typing, and WGS (including WGS-based serovar prediction) and (ii) a scoring system to evaluate and compare Salmonella subtyping assays. This literature-based assessment supports the superior discriminatory power of WGS for source tracking and root cause elimination in food safety incident; however, circumstances in which use of other subtyping methods may be warranted were also identified. This review provides practical guidance for the food industry and presents a starting point for further comparative evaluation of Salmonella characterization and subtyping methods.

Keywords: Salmonella, subtyping, serotyping, WGS, PFGE, MLST, food industry

### INTRODUCTION

A number of food safety incidents and recalls caused by Salmonella contamination have been associated with ready-to-eat low-moisture products (e.g., milk powder, raw almonds, dry seasonings, and peanut butter) (Pillai and Ricke, 2002; Maciorowski et al., 2004; Park et al., 2008; GMA, 2009; Hanning et al., 2009), and other food commodities (e.g., meat products, eggs, and vegetables) (Greig and Ravel, 2009; Wu et al., 2017; Ricke et al., 2018) in recent years. These cases highlight the need to reinforce Salmonella control measures in the food industry, including rapid and accurate tracking of pathogen contamination sources with appropriate subtyping tools. Tools used in incident investigations that can differentiate Salmonella beyond the species level (defined as Salmonella subtyping) are essential to improve control of this pathogen, as Salmonella contamination can occur from diverse sources at any stage of food production (Olaimat and Holley, 2012; Barco et al., 2013; Shi et al., 2015).

Conventional serotyping (White–Kauffmann–Le minor scheme) has been used as a Salmonella subtyping method for >80 years (Salmonella Subcommittee of the Nomenclature Committee of the International Society for, Microbiology, 1934; Grimont and Weill, 2007; Guibourdenche et al., 2010; Dera-Tomaszewska, 2012; Shi et al., 2015) and has been a certified approach for public health monitoring of Salmonella infections for over 50 years (CDC, 2015). This method classifies the genus Salmonella into serovars (also known as "serotypes") based on surface antigens including somatic (O), flagellar (H), and capsular (Vi) antigens (Brenner et al., 2000). More than 2,500 serovars of Salmonella enterica, the Salmonella species responsible for virtually all salmonellosis cases have been identified by conventional serotyping (Hadjinicolaou et al., 2009; Ferrari et al., 2017), but less than 100 serovars account for the vast majority of human infections (CDC, 2015). Due to the large variety of Salmonella serovars, a laboratory needs to maintain more than 250 different high-quality typing antisera and 350 different antigens for conventional serotyping of Salmonella (McQuiston et al., 2004; Fitzgerald et al., 2006). The turnaround time (i.e., time needed from isolate submission to a laboratory to receipt of the result) for serotyping a single isolate is usually >3 days. In some cases, it can take much longer (>12 days) as multiple antibody/agglutination reactions may be needed in a step-wise fashion to assign a final classification for complex serovars (Kim et al., 2006; Boxrud, 2010). Traditional serotyping is thus time-consuming and labor intensive requiring well-trained, experienced technicians (Boxrud, 2010; Shi et al., 2015). Unfortunately, it can also be imprecise (McQuiston et al., 2011). Moreover, the low discriminatory power of conventional serotyping may result in false-positive identification of relatedness between two unrelated isolates, as strains with the same serovar (such as the serovar Salmonella Enteritidis) may originate from multiple contamination sources. Further in-depth resolution beyond the serovar level is thus required for incident investigations (Ricke, 2017; Ricke et al., 2018). Various rapid molecular-based subtyping methods have been developed to provide faster, more discriminatory, and more accurate subtyping of Salmonella thus overcoming the limitations of traditional serotyping. Nevertheless, serovar data can still provide important historical epidemiological information, as certain serovars have specific virulence characteristics or may be associated with specific contamination sources (Ricke et al., 2018). Thus, it is important to link the subtypes identified by these molecular-based methods to Salmonella serovars.

There is no current global recommendation for the application of molecular characterization methods for Salmonella, although the food industry has applied both banding pattern-based and sequence-based subtyping methods for incident investigations. This review will provide (i) a comparison between classical serotyping and selected widely used molecular-based subtyping methods including pulsed-field gel electrophoresis (PFGE), multilocus sequence typing (MLST), and whole-genome sequencing (WGS, including WGS-based serovar prediction), and (ii) a scoring system to evaluate and compare Salmonella subtyping assays.

### BANDING PATTERN-BASED AND SEQUENCING-BASED CHARACTERIZATION METHODS FOR Salmonella

There are two major types of molecular-based subtyping methods: (i) nucleotide banding pattern-based subtyping methods, representing the banding patterns generated from the restriction digestion or polymerase chain reaction (PCR) amplification of genomic or plasmid DNA (Wachsmuth et al., 1991; Hartmann and West, 1997) and (ii) sequencing-based subtyping, identifying variants at the single-nucleotide level of the selected gene markers or the entire genome of an isolate. A comparison of the resolution, turnaround time, ability of serovar prediction, cost, and feasibility of these methods is given below (**Table 1**).

### Banding Pattern-Based Characterization Methods

### Pulsed-Field Gel Electrophoresis (PFGE)

Pulsed-field gel electrophoresis was first described in 1984 and developed as a subtyping method for Salmonella in the 1990s (Threlfall and Frost, 1990; **Figure 1**). PFGE is currently the gold standard for PulseNet International, and has been used by public health authorities and food regulators for outbreak investigations and source tracking globally (including USCDC, USFDA, USDA, and ECDC) (Zou et al., 2010; Wattiau et al., 2011; PulseNet, 2014; CDC, 2016a). Alternative methods for Salmonella subtyping are commonly compared against PFGE (Call et al., 2008). However, PulseNet is transitioning from using PFGE and multiple locus variable number of tandem repeats analysis (MLVA) toward using WGS as the standardized genotyping method for foodborne pathogens (CDC, 2017a; Nadon et al., 2017). PulseNet International has defined standard PFGE protocols (PulseNet, 2013; CDC, 2017b) and maintains a database of Salmonella PFGE profiles with >350,000 PFGE patterns representing >500 serovars. These PFGE patterns predominantly represent isolates collected since 1996 in North America and Europe (Zou et al., 2013). PFGE has relatively high concordance with epidemiological relatedness with two decades of data accumulation (CDC, 2018a). However, the PulseNet database for PFGE patterns is not publicly available and can only be accessed by PulseNet participating laboratories.

The PFGE approach uses restriction enzymes that recognize specific restriction sites along the genomic DNA and fragment the DNA to sizes normally ranging from 20 to 800 kb (up to 2,000 kb) (Schwartz and Cantor, 1984; Singh et al., 2006). These large fragments are separated in a flat agarose gel by constantly changing the direction of the electric current (pulsed field), which causes the DNA to separate by size, generating a specific "fingerprint pattern" for a given isolate (Foley et al., 2009). The restriction enzymes XbaI, NotI, SpeI, and SfiI have been typically used for Gram-negative bacteria including Salmonella (Barg and Goering, 1993). The primary restriction enzyme used for Salmonella PFGE is XbaI.


of June 2018, true costs may vary considerably

 based on number of isolates tested per run, labor costs, and

region/country,

 etc.

2NA, not available.

A public health laboratory usually has access to software [e.g., BioNumerics and GelCompar (Applied Maths, Sint-Martens-Latem, Belgium); Diversity Database Fingerprinting Software (Bio-Rad Laboratories, Hercules, CA, United States)], to analyze a PFGE pattern (Nsofor, 2016) and uploads PFGE patterns to a national database. PulseNet Central's database managers then analyze the uploaded pattern to see if a new outbreak has emerged or whether the isolate is part of an ongoing outbreak (CDC, 2018a). To make inter-laboratory comparison of DNA patterns possible, standardized protocols, molecular size standards (Salmonella Braenderup H2812, ATCC BAA-664), software, and nomenclature of PFGE patterns are required (PulseNet, 2015a). The approximate cost of the equipment and reagents required by PFGE can be accessed on the PulseNet International – PFGE site (PulseNet, 2015b).

Pulsed-field gel electrophoresis has been shown repeatedly to be more discriminatory than methods such as conventional serotyping, ribotyping, or MLST for many bacteria (Fakhr et al., 2005; Harbottle et al., 2006; Oloya et al., 2009; Soyer et al., 2010; Hauser et al., 2012). The combination of profiles generated by using additional restriction enzymes can enhance the value of this method for differentiating highly homogeneous Salmonella strains (Zheng et al., 2011); however, the cost increases as additional enzymes are used. PFGE can be used for subtyping of both Gram-positive (e.g., Listeria monocytogenes, Staphylococcus aureus) and Gram-negative (e.g., Salmonella, Escherichia coli, Shigella, Campylobacter jejuni) pathogenic bacteria. Typically, only the choice of the restriction enzyme and conditions for electrophoresis need to be optimized depending on the bacterial species investigated (PulseNet, 2015a).

Although various software platforms are available for PFGE pattern analysis, artifacts (e.g., brightly fluorescing spot) may lead to misidentification of bands. PFGE technology cannot usually be used to reliably visualize smaller fragments (e.g., <20.5 kb; Hunter et al., 2005) and has difficulty in differentiating bands differing by <5–10% in size due to the limited resolution of electrophoresis (Dijkshoorn et al., 2001; Persing et al., 2011). To address these issues, it has been recommended that users confirm PFGE pattern assignments using their experience and additional information to avoid incorrect band calling and systematic band shifts due to gel imperfections or imperfect reproducibility of electrophoretic conditions (Van Belkum et al., 2007). PFGE cannot be automated and requires high-level technical expertise and, thus, is hampered by low throughput, and may show low robustness and poor comparability of results between laboratories (Hyytia-Trees et al., 2007; Fabre et al., 2012; Kjeldsen et al., 2016).

No genetic information such as virulence potential or presence of antimicrobial resistance genes can be provided by PFGE, as the DNA fragments are separated by size rather than sequence (Ferrari et al., 2017). Observed bands of comparable size might not represent the same sequence of DNA, and a small mutation in a restriction site may result in changes in multiple bands. "Relatedness" determined by PFGE thus may

not represent a true phylogenetic relationship between isolates (CDC, 2018a). Typically, multiple distinct PFGE patterns can be identified among isolates classified into the same serovar. Polyphyletic serovars, which are derived from more than one common evolutionary ancestor or ancestral group (e.g., serovars Newport, Mississippi, Saintpaul, Kentucky), show high levels of PFGE diversity (Porwollik et al., 2004; Sukhnanand et al., 2005; Alcaine et al., 2006; Harbottle et al., 2006; Sangal et al., 2010). PFGE-based prediction of these serovars is unreliable if isolates in the database are not representative of all clades of the serovar. On the other hand, PFGE may cluster epidemiologically unrelated isolates into identical PFGE types (Barco et al., 2013) and may even provide similar or identical PFGE types for isolates that represent different, but genetically very similar serovars that have a common ancestor (Barco et al., 2013; Shi et al., 2015), such as Typhimurium (antigenic formula: 1,4,[5],12:i:1,2) versus Typhimurium var. Copenhagen (antigenic formula: 1,4,12:i:1,2) (Heisig et al., 1995; Hauser et al., 2011), and Typhimurium versus 4,5,12:i:- (Guerra et al., 2000; Soyer et al., 2009; Wiedmann and Nightingale, 2009; Hoelzer et al., 2010; Ranieri et al., 2013). Furthermore, differentiation of genetically homogeneous serovars such as serovar Enteritidis challenges the usefulness of PFGE in Salmonella subtyping activities (Olson et al., 2007; Zheng et al., 2007). Approximately 45% of serovar Enteritidis isolates reported to PulseNet display the same PFGE XbaI pattern (JEGX01.0004), although many of these isolates are not epidemiologically related (Zheng et al., 2007). It is important to mention that the serovars mentioned above (i.e., Enteritidis, Typhimurium, Newport, Mississippi, Saintpaul, and Kentucky) are ranked among the most common Salmonella serovars associated with human and animal salmonellosis globally (Galanis et al., 2006; CDC, 2009).

### Multiple Locus Variable Number of Tandem Repeats Analysis (MLVA)

Multiple locus variable number of tandem repeats analysis is a PCR-based typing method originating from forensic science where it is used for DNA "fingerprinting" samples of human origin. It has frequently been applied to scientific studies of prokaryotes as well as to microbial outbreak detection and source tracking (Lindstedt et al., 2003, 2013; **Figure 1**). MLVA is the second major genotyping tool (after PFGE) used in the PulseNet network (PulseNet, 2015c); prior to WGS, MLVA was one of the most popular subtyping methods used in public health surveillance and outbreak investigation of Salmonella, particularly in Europe (Torpdahl et al., 2007; Hopkins et al., 2011; Barco et al., 2013; Bauer et al., 2013; Lindstedt et al., 2013; Mughini-Gras et al., 2018). MLVA is usually performed following serotyping or PFGE for routine surveillance as a complementary technique for Salmonella subtyping (Torpdahl et al., 2007; Lienemann et al., 2015; Kjeldsen et al., 2016; CDC, 2017c; Ferrari et al., 2017), as it is challenging for PFGE to further differentiate isolates of genetically homogeneous serovars such as Salmonella Enteritidis (Kjeldsen et al., 2016). MVLA is especially used for typing Salmonella Typhimurium and Salmonella Enteritidis strains in reference or regulatory laboratories in Denmark, France, Germany, and United States [e.g., CDC, USDA – Food Safety and Inspection Service (FSIS) laboratories] (Barco et al., 2013; Bauer et al., 2013).

Multiple locus variable number of tandem repeats analysis is serovar specific, thus different Salmonella serovars usually require different MLVA schemes (Kruy et al., 2011). The first step toward uniform standardization of the MLVA profiles was collectively taken by PulseNet International and ECDC in defining the standard protocols of MLVA for Salmonella Typhimurium and Salmonella Enteritidis (ECDC, 2011, 2016b; PulseNet, 2015c). These serovars account for 26% of the culture-confirmed human Salmonella infections reported by US Laboratory-based Enteric Disease Surveillance (LEDS) and >60% of the salmonellosis cases reported by ECDC (ECDC, 2016a; Kjeldsen et al., 2016; CDC, 2018b). This uniform standardization of the MLVA profiles allowed direct comparison between laboratories irrespective of the platform used for MLVA (Larsson et al., 2009). Validated MLVA standard protocols for additional Salmonella serovars of clinical importance worldwide are largely missing, making MLVA use for serovars other than Enteritidis and Typhimurium difficult. However, with the advent of and transition into WGS, further development of MLVA may not occur.

Multiple locus variable number of tandem repeats analysis assesses the variation in the number of tandem repeated DNA sequences referred to as "variable-number tandem repeats" (VNTRs) in multiple regions of the bacterial genome to characterize bacterial isolates. The number of VNTRs in a given locus may vary between different microorganisms and even among bacterial isolates of the same species and serovar (Lindstedt et al., 2003; Torpdahl et al., 2007; Ngoi et al., 2015). The VNTR profiles vary in length between a few base pairs long to over 100 base pairs, enabling the development of techniques that utilize variation in the size of VNTR to discriminate closely related isolates (Lindstedt et al., 2003; Torpdahl et al., 2007; Fabre et al., 2012). The improved discriminatory power of MLVA varies with the serovar and phage type investigated (Torpdahl et al., 2007; Lienemann et al., 2015); e.g., in a study in Denmark, MLVA could differentiate distinct clusters within the most common phage types of Salmonella Typhimurium such as DT104, DT120, and DT12 even though these isolates displayed comparable PFGE patterns (Torpdahl et al., 2007). Public health laboratories usually have access to software (e.g., BioNumerics, GeneMapper, the free Peak Scanner) for analysis of MLVA patterns (ECDC, 2011; PulseNet, 2015c). Minimum spanning trees are frequently applied to MLVA profiles, yielding maps of predicted relationships among isolates based on single-locus and dual-locus variants (Van Belkum et al., 2007). However, webaccessible MLVA databases are not widely used for international collaboration (Guigon et al., 2008).

Multiple locus variable number of tandem repeats analysis is cheaper, faster, simpler to execute, and shows a relatively highthroughput compared with other molecular methods (Torpdahl et al., 2005, 2007; Lindstedt et al., 2007, 2013; Hopkins et al., 2011; Kruy et al., 2011). MLVA is less labor-intensive, time-consuming, and it is easier to perform than PFGE and MLST, as the protocol requires only a regular PCR step followed by capillary electrophoresis (Torpdahl et al., 2007; Lindstedt et al., 2013).

Reduced handling time of pathogenic bacteria is beneficial for large-scale investigations. MLVA is also suitable for automation using a pipetting robot work station, automated sequencer, and analytical software (Barco et al., 2013; Lindstedt et al., 2013; Ferrari et al., 2017). Moreover, MLVA demonstrates good international repeatability and reproducibility for specific serovars such as Salmonella Typhimurium and Salmonella Enteritidis (Larsson et al., 2013). The data generated by MLVA can be readily analyzed and standardized for inter-laboratory comparisons (Torpdahl et al., 2007; Hopkins et al., 2011; Lindstedt et al., 2013; Wuyts et al., 2013).

A major drawback of MLVA for Salmonella subtyping is that the most effective MLVA protocols described so far are serovarspecific (Barco et al., 2013; Ngoi et al., 2015; Kjeldsen et al., 2016); hence, isolates have to be serotyped prior to selecting a specific MLVA scheme for further subtyping (Kjeldsen et al., 2016). At least 27 MLVA schemes have been developed to subtype different Salmonella serovars, whereas only Salmonella Typhimurium and Salmonella Enteritidis MLVA assays have been standardized in Europe and in the PulseNet network (PulseNet, 2015c; Kjeldsen et al., 2016). Another drawback is that rapid evolution of the target loci may decrease the reliability of results provided by MLVA regarding the relationship between strains under investigation (Hopkins et al., 2007, 2011; Lindstedt et al., 2013). This might hamper the use of MLVA, particularly in longterm epidemiological studies (Lindstedt, 2005; Li et al., 2009).

### Repetitive Element PCR (Rep-PCR)

Repetitive element PCR targets the repetitive elements of genomic DNA to discriminate bacterial isolates. This method has been developed using three families of repeat sequences for subtyping Salmonella, including "enterobacterial repetitive intergenic consensus" (ERIC) sequences, "the repetitive extragenic palindromic" (REP) sequences, and the "BOX" sequences (Gilson et al., 1990; Hulton et al., 1991; Martin et al., 1992). The PCR products amplified from genome regions containing these repetitive elements are analyzed by agarose gel electrophoresis, and the banding patterns generated are used to investigate the genetic relatedness between bacterial isolates (Sabat et al., 2013). The DiversiLab system (bioMérieux, Marcy-l'Etoile, France) automated the whole process of the Rep-PCR subtyping approach after 2000 and has been used for subtyping pathogens in hospitals worldwide (Healy et al., 2005; Chenu et al., 2012; Sabat et al., 2013; **Figure 1**). As the low reproducibility of original Rep-PCR method may have resulted from variability in reagents and gel electrophoresis systems (Sabat et al., 2013), the application of the DiversiLab system with microfluidic capillary electrophoresis increased both the resolution and reproducibility of the Rep-PCR approach (Healy et al., 2005; Chenu et al., 2012; Sabat et al., 2013). However, the system has been discontinued, making Rep-PCR unavailable as a commercial platform.

The major advantages of this method include its relatively low cost (comparable to that of PFGE) and short turnaround time (within one day) (Sabat et al., 2013; Ngoi et al., 2015). However, the discriminatory power of Rep-PCR in subtyping Salmonella is reportedly lower than that of PFGE (Tiong et al., 2010; Thong and Ang, 2011; Elemfareji and Thong, 2013; Ngoi et al., 2015). Its relatively low reproducibility (which can at least be partially addressed by automation, such as in the DiversiLab system), and low accuracy of serovar prediction (Weigel et al., 2004; Wise et al., 2009) have limited its application in Salmonella subtyping.

### Sequencing-Based Characterization Methods

### Legacy Multilocus Sequence Typing (Legacy MLST)

Multilocus sequence typing is a nucleotide sequence-based approach that assesses DNA sequence variations (i.e., allelic type) of typically three, four, or seven selected well-conserved, housekeeping genes, usually using Sanger sequencing technology (Liu, 2010; Achtman et al., 2012). Schemes targeting seven genes are typically considered the "classical" MLST approach; this typing approach was originally proposed for isolates of Neisseria meningitidis (Liu, 2010). In this review, we focus on the most widely used Salmonella scheme targeting seven housekeeping genes [aroC, dnaN, hemD, hisD, thrA, sucA, and purE; hereafter denoted as legacy MLST to distinguish newer approaches (described below)] (Li et al., 2009; Yun et al., 2015). It was first introduced for Salmonella Typhi in 2002 (Kidgell et al., 2002), and extended to all Salmonella serovars in 2012 (Achtman et al., 2012; **Figure 1**). Legacy MLST is mainly used in research studies, assessing the population genetics and evolution of Salmonella. Public Health England (PHE) started adopting the seven-gene MLST (based on WGS data) approach as a replacement for traditional serotyping in 2015 (Ashton et al., 2016).

Historical MLST data including legacy MLST sequence types are maintained on EnteroBase (Alikhan et al., 2018). As of November 2017, the number of legacy MLST sequence types for Salmonella has reached 3,929 (Alikhan et al., 2018). Legacy MLST analysis can be conducted online by entering the sequences of amplified genes. Allelic variation at each locus is cataloged and a sequence type is assigned by comparing the allele set. The strains are characterized by their unique sequence type. With the advent of next-generation-sequencing technologies, legacy MLST data can also be extracted directly from WGS data using bioinformatics pipelines (Achtman et al., 2012; Ashton et al., 2016). The relatedness of isolates subtyped by legacy MLST can be displayed as a dendrogram or a minimum spanning tree based on the matrix of pairwise differences between their allelic profiles (Francisco et al., 2009), or as a phylogenetic tree built directly from the nucleotide alignment of the seven genes.

Legacy MLST can deliver results more rapidly than PFGE (Shi et al., 2015; Yun et al., 2015; **Table 1**), and the publicly available databases and online query system enable legacy MLST results to be highly reproducible and exchangeable between laboratories. However, legacy MLST shows lower discriminatory power than PFGE and MLVA, which limits its application to further discriminate isolates within a given serovar (Torpdahl et al., 2005; Alcaine et al., 2006; Foley et al., 2006; Harbottle et al., 2006; Hauser et al., 2012; Ngoi et al., 2015), and for source attribution (Barco et al., 2013). Protocols targeting sequences in

genes that change more rapidly than housekeeping genes have been developed to improve the discriminatory power of legacy MLST (Ross and Heuzenroeder, 2005, 2008).

### Clustered Regularly Interspaced Short Palindromic Repeat-Based Subtyping (CRISPR-Based Subtyping)

The clustered regularly interspaced short palindromic repeat (CRISPR) typing method uses the diversity of the content of CRISPR loci to distinguish bacterial strains. The application of the CRISPR system for subtyping foodborne pathogens is discussed in detail elsewhere (Shariat and Dudley, 2014; Shi et al., 2015; Barrangou and Dudley, 2016; Ferrari et al., 2017; Ricke et al., 2018). Although the CRISPR system has been applied to the subtyping of at least 100 Salmonella serovars (Shariat and Dudley, 2014; Barrangou and Dudley, 2016), this approach is not widely used by public health authorities or food regulators (Ferrari et al., 2017).

Clustered regularly interspaced short palindromic repeat loci contain variable lengths of CRISPR spacers obtained from foreign nucleic acids of plasmids or bacteriophages (Shariat and Dudley, 2014; Wright et al., 2017). These CRISPR spacers are acquired or lost during evolution of the pathogen over time in a sequential manner (Ricke et al., 2018), thus constructing a unique set of DNA sequence patterns that may provide sufficient resolution for pathogen subtyping (Fricke et al., 2011; Barrangou and Horvath, 2012; Shariat and Dudley, 2014; Wright et al., 2017). For subtyping, amplified CRISPR loci PCR products are sequenced by Sanger sequencing technology (Liu et al., 2011). The CRISPR spacer sequences are analyzed to assign each locus with an allelic type. The combination of the allelic types of analyzed CRISPR loci determine the isolate's allelic profile (also referred to as the isolate's sequence type) and is used to investigate the relationships between isolates (Liu et al., 2011).

The CRISPR approach has been shown to be feasible for subtyping of Salmonella (Liu et al., 2011; Fabre et al., 2012; DiMarzio et al., 2013; Shariat et al., 2013a,b,c; Almeida et al., 2017). Liu et al. (2011) developed a CRISPR–multi-virulencelocus sequence typing (MVLST) approach using virulence genes sseL and fimH with CRISPR1 and CRISPR2 loci; this approach was used to compare 171 isolates representing nine serovars (Typhimurium, Enteritidis, Newport, Heidelberg, Javiana, I 4,[5],12:i:-, Montevideo, Muenchen, Saintpaul) and was reported to be able to subtype Salmonella with resolution at the outbreak level. CRISPR–MVLST using different schemes of virulence genes has also been applied by others for subtyping Salmonella (DiMarzio et al., 2013; Shariat et al., 2013a; Almeida et al., 2017). The results from these studies suggest that CRISPR– MVLST has a higher discriminatory power than legacy MLST (Ferrari et al., 2017); however, discrimination is lower than PFGE in some cases (Almeida et al., 2015). While CRISPR typing has a relatively short turnaround time (comparable to MLST), current major drawbacks include high cost (Almeida et al., 2017; Ferrari et al., 2017), unstandardized protocol, and database, as well as limited research on the concordance between the diversity of Salmonella isolates reflected by CRISPR loci content and by the other standard subtyping methods (Shi et al., 2015).

### Whole-Genome Sequencing (WGS)

Whole-genome sequencing captures DNA sequence changes across the entire genome of single microbial isolates. The data are useful to assess evolution, allowing accurate description of the genetic relatedness of isolates. The use of WGS for Salmonella subtyping in outbreak investigation and pathogen source tracking has proven effective by a rapidly increasing number of studies (den Bakker et al., 2011, 2014; Allard et al., 2012; Leekitcharoenphon et al., 2014; Deng et al., 2015; Taylor et al., 2015; Hoffmann et al., 2016; Inns et al., 2016). WGS was first used to trace a Salmonella multistate outbreak in the United States in 2009 (CDC, 2019), and has been used for pathogen subtyping by the public health surveillance systems in the United States (Allard et al., 2018), Canada (Vincent et al., 2018), the United Kingdom (Ashton et al., 2016), Denmark (Kvistholm Jensen et al., 2016), and France (Moura et al., 2016). PulseNet international is also making efforts to implement WGS within the PulseNet network as a routine tool to replace PFGE and MLVA (Nadon et al., 2017; **Figure 1**). Both PHE (Ashton et al., 2016) and the US FDA (2018) have started using "real-time" WGS to subtype Salmonella isolates. CDC is also using WGS in state laboratories for Salmonella outbreak investigations (CDC, 2016b). WGS will be used increasingly for contamination incident investigations in the food industry, particularly as cost continues to shrink and ease of use increases. WGS (as well as other sequencing approaches that use the same next-generation sequencing technologies used for WGS) also have a number of additional applications in the food industry, which will further drive implementation of these tools. Examples of other applications include (i) monitoring ingredient supplies, (ii) identification of microbial persistence in processing environments, and (iii) prediction of antimicrobial resistance (including in Salmonella) and other relevant phenotypes, facilitating the improvement of sanitary management, microbial hazard control, and microbiological risk assessment (Allard et al., 2018; Rantsiou et al., 2018; Ricke et al., 2018).

Sequenced Salmonella genomes can be deposited and made publicly available on the National Center for Biotechnology Information site<sup>1</sup> , the European Bioinformatics Institute site<sup>2</sup> , or the DNA Data Bank of Japan site<sup>3</sup> with data shared between all three (Kodama et al., 2012; Jagadeesan et al., 2019). NCBI provides phylogenetic tree-based clustering of all publicly available sequence data at the NCBI pathogen detection site<sup>4</sup> . These phylogenetic trees show the closest matches to any newly submitted data (Allard et al., 2018). NCBI also houses the data using GenomeTrakr Network (FDA, 2018). This was developed by the US FDA and NCBI as the first distributed network of laboratories to utilize WGS, with both genomic and geographic data, for foodborne pathogen characterization. This network includes the WGS laboratories of the CDC and USDA (Allard et al., 2016; Jackson et al., 2016). As of February 2019, there are over 184,000 genome sequences or raw sequencing data

<sup>1</sup>https://www.ncbi.nlm.nih.gov/sra

<sup>2</sup>https://www.ebi.ac.uk/ena/

<sup>3</sup>https://www.ddbj.nig.ac.jp/index-e.html

<sup>4</sup>https://www.ncbi.nlm.nih.gov/pathogens/

of S. enterica available on NCBI. WGS data analysis can also be performed off-line without using any public databases, an approach that may sometimes be preferred by industry.

Sequencing platforms that can be used currently for WGS include Illumina, Ion Torrent, Oxford Nanopore Technologies, and Pacific Biosciences (PacBio). Procedures to validate the complete workflow for S. enterica WGS with Illumina (MiSeq and HiSeq) and PacBio platforms from subculture of isolates to bioinformatics analysis have been reported by Portmann et al. (2018). The Illumina sequencing system is one of the most widely used sequencing platforms; it produces DNA-sequence reads with the length of 50–300 bp using sequencing-by-synthesis (SBS). This process uses fragmented DNA templates to detect single bases as they are incorporated during a DNA replication reaction on a solid surface flow cell (Illumina (2019)). For applications including comparative genomics and phylogeny, these short reads of DNA sequences can be aligned to a reference genome or de novo assembled into longer sequences called contigs (Loman and Pallen, 2015). The large amount of data generated by WGS combined with a complex data analysis process generally requires expertise in bioinformatics to deploy and run (Wyres et al., 2014; Deurenberg et al., 2017). Software with a more user-friendly interface, such as CLC Genomics Workbench<sup>5</sup> , BioNumerics, and Geneious (Biomatters, New Zealand), however, is available, including for industry users with limited bioinformatics expertise and an increasing number of user-friendly bioinformatics tools are being developed.

The rapid growth of WGS data in the publicly available databases allows industry to compare isolates with global entries of pathogen sequences used by food regulators and public health authorities (Allard et al., 2018; Rantsiou et al., 2018). Despite increasing availability of data analysis software, it is still challenging to generate consistent analytical reports due to the lack of standardized approaches to data analysis and interpretation (Clooney et al., 2016); for example, even with a standard software, choice of reference genomes can have considerable effects on the data analyses (Pightling et al., 2014). Furthermore, there are currently no clearly outlined safeguards to protect companies from regulatory action if shared WGS data show a relationship between pathogen isolates identified by a company and an outbreak isolate. Development of a mechanism for sharing data through anonymous hubs may allay concerns on confidentiality and encourage data sharing (FAO, 2016). This mechanism may also enable more effective data capture and analysis for monitoring trends and identifying related incidents.

The current cost of the entire WGS process, including DNA library preparation, sequencing, data analysis, and storage, is relatively high compared with the other molecular-based subtyping methods. The cost difference is more apparent when a small number of isolates are sequenced (as could be typical for the food industry). The cost of maintaining data analysis tools and bioinformatics personnel needs to be taken into consideration (Leekitcharoenphon et al., 2014; Ferrari et al., 2017; Nadon et al., 2017).

### WGS-Analysis Procedures

Interpretation of WGS data for source tracking or outbreak investigation typically uses two approaches to represent results: (i) single-nucleotide polymorphism (SNP) or allelic differences (often presented as distance matrix tables), and (ii) phylogeny or clustering of the isolates. SNP or allelic differences show objectively the genetic distance between two isolates. Hence, if isolate A shows three SNPs or allelic differences to isolate B, and 26 SNPs or allelic differences to isolate C, then we can say that isolate A is more similar to isolate B than to isolate C. If one assumes that all three isolates evolved at the same rate, then we can say that isolates A and B are evolutionarily more closely related to each other than they are to C. However, this assumption (i.e., all isolates evolve at the same rate) may not always be true. Environmental conditions or mutations in the DNA repair system may influence the rate of genetic change accumulated in a genome; e.g., a Salmonella isolate persisting in a humid, nutritious environment such as in a chicken farm may multiply much faster than an isolate persisting in a dry food processing environment. This environmental difference will allow the "chicken farm" isolate to accumulate more mutations (per year or any other time unit) than the dry food processing environment isolate, because the "chicken farm" isolate will multiply more times during the same period than the dry food processing environment isolate. Moreover, mutations in genes involved in DNA repair may result in the so-called "mutator phenotypes" (also sometime referred to as "hypermutators"). Mutator isolates accumulate mutations at a higher rate than non-mutator isolates (Muteeb and Sen, 2010). Hence, analyzing the number of SNP or allelic differences alone may result in misinterpretation of the results if the assumption that isolates evolved at the same rate does not hold true. Phylogenetic or clustering analyses are thus better suited to an investigation, as these analyses group isolates by their similarities instead of their differences (Pightling et al., 2018). To infer the evolutionary relationship of the isolates within a data set, therefore, a phylogeny must be constructed. For more detailed and technical information on reconstructing bacterial phylogenies from WGS data, the reader is referred to two in-depth reviews on this subject (Collins and Xavier, 2017; Patané et al., 2018).

### WGS Analysis Approaches for Serotyping

Genetic-based approaches have been developed for in silico determination of serovars, because the phenotypic determination of Salmonella serovars is costly, time-consuming, and laborintensive. These in silico methods have relied on two main approaches: (i) indirect determination using genetic markers associated with particular serovars and (ii) direct determination using genes responsible for the expression of the somatic O (rfb gene cluster) and flagellar H (fljB and fliC) antigens. The latter method has the advantage of relying on the same genetic information that results in the phenotype assessed by traditional serotyping, while the former method may require validation for new described serovars. These two approaches can also be combined for more reliable serovar prediction.

With the advent of whole-genome sequencing (WGS), in silico direct serovar determination has become the most used approach,

<sup>5</sup>https://www.qiagenbioinformatics.com

and at least two Salmonella serovar databases and programs have been routinely used for in silico serotyping of Salmonella: SeqSero (Zhang et al., 2015) and SISTR (Yoshida et al., 2016a). SeqSero uses a database of 473 alleles representing 56 fliC antigenic types and 190 alleles representing 18 fljB antigenic types in a combined H-antigen database (Zhang et al., 2015). The somatic O-antigen database associated with SeqSero consists of 46 rfb gene cluster sequences corresponding to the 46 O-antigens identified in Salmonella (Zhang et al., 2015). The rfb database was specifically designed to be used with genome assemblies (as opposed to raw sequencing reads). A third database was specifically built for determination of the somatic O-antigen using raw sequencing reads (as opposed to genome assemblies). This third database consists of the genes wzx (encoding the O-antigen flipase), wzy (encoding the O-antigen polymerase), and other targets, all of which are found within the rfb gene cluster. In total, the authors claimed that the SeqSero scheme can theoretically identify 2,389 of the 2,577 serovars that were described in the White–Kauffmann–Le minor scheme by the end of 2014 (Zhang et al., 2015). The inability to predict 188 serovars is due to the absence of the DNA sequences for the antigen-encoding genes corresponding to these serovars in the SeqSero database. Empirical data showed that the SeqSero database has an accuracy of 91.5–92.6% for serotype prediction (Zhang et al., 2015).

SISTR is a platform for in silico analysis of Salmonella draft genome assemblies. SISTR includes the Salmonella Genoserotyping Array (SGSA) tool among other resources. SGSA relies on the allelic differences found within the rfb gene cluster for determination of 18 of the 46 somatic O-antigens, and fljB and fliC for determination of 41 flagellar H antigens (Yoshida et al., 2014). SGSA targets the identification of 90% (n = 2,190) of Salmonella serovars. When serovar determination using genoserotyping is not possible or is incomplete, SISTR also has the option to use the core genome MLST (cgMLST) scheme to infer the serovar based on phylogenetic context. The accuracy of SISTR in predicting Salmonella serovars has been assessed to be close to 95% (Yoshida et al., 2016a,b; Robertson et al., 2018).

Since SISTR can use genoserotyping and the cgMLST scheme to infer the serovar, higher confidence should be attributed to assignments where both genoserotyping and cgMLST agree on the serovar designation. Moderate confidence should be attributed to serovar assignments when only cgMLST is able to identify the serovar. When neither the genoserotyping nor cgMLST can identify the serovar, SeqSero may be used and may allow for serovar prediction.

### WGS Analyses for Subtype Characterization

### **Overview of WGS data analysis approaches**

Different approaches can be used for analysis of WGS data for subtyping characterization related to source incident tracking. The most common approaches are based on (i) high-quality single-nucleotide polymorphism (hqSNP) identification and pairwise comparison of hqSNP differences, or (ii) whole-genome (wg)/cgMLST typing using pre-defined schemes (i.e., databases) containing allelic differences for either the pan (wg) or core (cg) genomes of Salmonella and subsequent pairwise comparison for assessing the number of allelic differences.

### **High-quality SNP analyses**

High-quality SNP analyses rely on identification of SNP differences across a set of closely related isolates using raw sequence data, which are mapped to a closed or draft genome assembly (also referred to as the "reference genome"). Only SNPs that have been vertically transferred from an ancestral isolate to the current isolates are subject to the hqSNP analysis, while SNPs that were supposedly horizontally transferred are filtered out from the results. The reference can be a closely related genome outside the dataset, or a genome within the dataset. The analysis consists of two main steps: (i) mapping the raw sequence reads against the reference genome and (ii) SNP calling using stringent criteria to prevent the misidentification of sequencing errors or misaligned regions as SNPs (Davis et al., 2015; Katz et al., 2017). The choice of a closely related reference has been shown to be a key step in the analysis. Reference genomes that are not closely related to the set of isolates under investigation may result in underestimation of the number of SNPs, due to specific regions of the genome that may be present in the dataset under investigation, but that are missing in the reference genome (Pightling et al., 2014). There are at least two publicly available approaches that have been commonly used for hqSNP analysis: (i) the US FDA CFSAN (The Center for Food Safety and Applied Nutrition) SNP pipeline (Davis et al., 2015) and (ii) the US CDC-developed Lyve-SET hqSNP pipeline (Katz et al., 2017). These two pipelines rely on publicly available software to carry out the mapping and SNP calling steps and offer similar results despite some methodological differences, including different criteria for filtering out lowquality SNPs and masking regions supposedly acquired through horizontal gene transfer.

High-quality SNP analysis has been applied in several outbreak investigations in the United States, Canada, and some European countries, including a Salmonella Enteritidis outbreak in the United Kingdom that was linked to a German egg producer (Inns et al., 2015). Historical Salmonella Typhimurium isolates from humans and foods involved in five outbreaks and consisting of five distinct MLVA subtypes were re-analyzed using hqSNP analysis by Octavia et al. (2015); in this study at least 11 isolates not previously linked to the outbreaks were ruled in based on less than two SNP differences to the isolates previously linked to the outbreaks. Another retrospective study used hqSNP to analyze a collection of 55 Salmonella Enteritidis from seven epidemiologically characterized outbreaks and sporadic cases. One isolate not previously linked to any outbreak (i.e., sporadic) was identified to be part of one outbreak ("ruled in") (Taylor et al., 2015). An investigation into a multi-state outbreak caused by Salmonella Poona was carried out in 2015 using PFGE and hqSNP analysis. Analysis by PFGE demonstrated three different patterns. However, WGS results showed that isolates with different PFGE patterns were genetically linked with less than six SNP differences (Kozyreva et al., 2016). Subtyping of Salmonella Dublin with PFGE was shown to have limited value in a recent outbreak investigation

due to its low discriminatory power for this Salmonella serovar (Mohammed et al., 2016). The nine clinical isolates associated with the outbreak were indistinguishable by PFGE, but they were also indistinguishable from other unrelated Salmonella Dublin isolates. The nine isolates linked to the outbreak clustered together with one to nine SNP differences when analyzed using hqSNP, and they could be distinguished from other isolates that shared the same PFGE pattern with epidemiologically unrelated isolates showing more than 50 SNP differences when compared to the outbreak isolates (Mohammed et al., 2016). These studies show that public health agencies are increasingly relying on hqSNP analysis for outbreak investigation, including tracking the source of outbreaks. High-quality SNP analysis clearly improves subtype accuracy and outbreak investigations by not only allowing for increased discriminatory power, but also reducing instances where closely related isolates are being classified as "different."

### **wgMLST**

Whole-genome MLST (wgMLST) analysis relies on the comparison of individual genomes against a database containing all known alleles for all the genes representing the pan genome of a defined group of strains (i.e., serovar, subspecies, species, and genus). The pan genome is defined as all the genes present in at least one genome from a defined group. Two main approaches can be used, and these are often used in combination: (i) assembly free mapping and (ii) assembly based mapping. Raw sequencing reads are directly mapped against the database in an assembly free approach. Hence, this approach does not require de novo assembly of the genome prior to its utilization. SRST2 (Inouye et al., 2014) and BWA-MEM (Li, 2013) are the most commonly used programs to carry out this task. Because this approach deals directly with the raw sequence reads, it allows filtering low-quality reads or specific nucleotides with low quality within a good-quality read. In an assembly based approach the raw sequence reads are first used to generate a highquality draft genome (i.e., usually not a closed genome) using a genome assembler. Later, the draft genome (i.e., assembly) is used to find matches against the database. The program most commonly used to map the draft genome against the database is BLASTN (Altschul et al., 1990), although other options also exist. Independently from the approach used (i.e., assembly free or assembly based), the result of mapping a genome against a database is a list of the alleles found in the analyzed genome. When more than one genome is analyzed, the list of alleles from each genome can be compared and the number of allele differences can be computed. Assembly free and assembly based wgMLST allele assignment should match for high confidence. Results are often shown as a distance matrix of allele differences and a dendrogram constructed from this distance matrix. The wgMLST methods allow for comparison of non-closely related isolates from different groups since all genomes are compared against the same database, which is a great advantage of this method over hqSNP (Maiden et al., 2013; Nadon et al., 2017). A disadvantage of the method is that the database must be constructed and shared across different groups, who must agree in using the same database in order to make their results comparable (Nadon et al., 2017). Construction of such databases is also time-consuming and labor-intensive, with the difficulty increasing with the diversity of the organisms included in the same database (e.g., a database for S. enterica subspecies enterica serovar Agona will require less time and labor than a database for all S. enterica).

### **Core genome MLST (cgMLST)**

The cgMLST method is very similar to the wgMLST method. The major difference is the size and nature of the database. While the wgMLST database contains alleles for all genes in the pan genome of the defined group, the cgMLST only contains alleles for those genes that are present in all (or almost all) genomes of the defined group (i.e., the "core genome"). Hence, a cgMLST database will not capture the genetic diversity present in the accessory genes (i.e., genes that are not present in all isolates) and hence tends to be much smaller than a corresponding wgMLST database. The advantages of using the cgMLST are: (i) speed; because the cgMLST database is smaller than the wgMLST database, results can be obtained faster, and (ii) construction of the cgMLST database is generally easier than the wgMLST database, as typically less genomes are needed to identify the core genome than the pan genome of a group (den Bakker et al., 2010). While allele code schemes are used by some groups to summarize the differences observed among isolates subtyped by both cgMLST and wgMLST (Nadon et al., 2017), it generally is easier to define standard, stable, cgMLST allele codes. This allele code scheme can be easily transferred in a spreadsheet and can be interpreted similarly to what has been in use for PFGE. An allele code scheme may not, however, be fully stable and may need to be revised as new cg- or wgMLST types are identified (Nadon et al., 2017). A disadvantage of cgMLST is that it may show reduced discriminatory power over wgMLST, as shown in a comparison between the Salmonella cgMLST and wgMLST schemes defined in EnteroBase (Alikhan et al., 2018), carried out using Salmonella Enteritidis historical isolates from a UK egg-associated outbreak (Inns et al., 2015), as well as closely related non-outbreak isolates identified previously (Dallman et al., 2016). The 177 isolates from this dataset resulted in 177 unique sequence types by wgMLST (Simpson's diversity index = 1.00) and 137 unique sequence types by cgMLST (Simpson's diversity index = 0.98) (P < 0.05), showing the superior discriminatory power of wgMLST over cgMLST. However, both approaches grouped the isolates into identical clusters (Pearce et al., 2018).

### **Comparison of hqSNP-based analysis and genomic MLST analysis**

Theoretically, hqSNP analysis is the most discriminatory approach for molecular subtyping, as it investigates all possible SNPs between each pair of isolates in the dataset. The second most discriminatory approach is wgMLST, which is designed to investigate virtually all genes in the genomes; intergenic regions and genes not present in the wgMLST scheme will not be investigated and polymorphisms present in these regions will be missed. The cgMLST approach is the least discriminatory of the three as it relies on only a subset of the genes present

in the wgMLST scheme. Hence, similarly to the wgMLST approach, polymorphisms present in intergenic regions and in genes not included in the cgMLST scheme will not be assessed (Chen et al., 2017). Both wgMLST and cgMLST are referenceindependent which makes the results more reproducible and transferable than hqSNP analysis (Nadon et al., 2017). In order to reproduce the results obtained from hqSNP analysis, one needs to use the same reference and parameters that were used in the original analysis (Nadon et al., 2017). This is not an issue with wgMLST or cgMLST analysis as long as analyses use the same scheme containing the same genes and alleles to allow for comparisons. Transference and communication of the results also seem to be more complicated for hqSNP analysis than for cgMLST or wgMLST (Nadon et al., 2017). This is because hqSNP analysis, as compared to cgMLST or wgMLST analyses, requires more parameter settings, which must be communicated for better interpretation. wgMLST and cgMLST analyses are also typically integrated into commercially available software, while the hqSNP pipelines are available as free open software or integrated into commercial software. Free-of-charge hqSNP pipelines require UNIX-based systems and are run through the command line, which may require specialized expertise (Nadon et al., 2017). Commercially available software, which can run cgMLST and wgMLST (e.g., BioNumerics) tends to be more user-friendly. BioNumerics uses a graphical user interface and can be installed in Microsoft Windows computers. The hqSNP analysis can easily be kept private as the analysis can be run within a closed dataset of genomes. The cgMLST and wgMLST can also be kept private; however, it may require some additional infrastructure (i.e., a private cloud) to be built around the commercial software.

### Comparison of Molecular Methods for Predicting the Serovar of Salmonella

A comparison of different molecular methods for predicting the serovar of Salmonella is shown **Table 2**. Acceptable correlation between PFGE patterns and serovars has been described by several researchers (Weigel et al., 2004; Nde et al., 2006; Gaul et al., 2007; Kerouanton et al., 2007; Zou et al., 2010; Shi et al., 2015; Bopp et al., 2016). Shi et al. (2015) summarized the serovar-prediction accuracy of different molecular serotyping methods with studies from 1993 to 2013. The proportion of isolates that may not be accurately serotyped with PFGE is generally comparable to the proportion that is not typeable, or that requires extensive additional labor and reagents using conventional serotyping (Bopp et al., 2016). Examples of serovars incorrectly predicted by PFGE are summarized below (**Table 3**). Overall, with PFGE patterns for approx. 500 Salmonella serovars in the PFGE pattern database (Ranieri et al., 2013; Shi et al., 2015) and the reported good correlation between PFGE patterns and serovars, PFGE-based serovar prediction should be possible for a large proportion of these serovars, but will not be possible for a large number of less common serovars not represented in the database.

Multiple locus variable number of tandem repeats analysis is not widely used for serovar prediction even though efforts have been made to develop MLVA subtyping schemes to subtype multiple serovars of Salmonella with one protocol (Van Cuyck et al., 2011; Kjeldsen et al., 2016). A universal MLVA scheme for most frequently isolated Salmonella serovars (accounting for 80% of the clinical isolates from humans in Europe) has been developed by Kjeldsen et al. (2016). In another study, an MLVA scheme identified 31 serovars (Van Cuyck et al., 2011). Nevertheless, further development of multiple-serovar MLVA schemes and robust MLVA profile databases is unlikely to occur given the benefits offered by WGS.

The serovar-prediction accuracy of Rep-PCR has been reported to range between 0 and 100%, indicating some limitations of this method (Shi et al., 2015). Ranieri et al. (2013) showed that Rep-PCR accurately predicted the serovar of 30 out of 46 isolates representing the top 40 Salmonella serovars isolated from human and non-human sources, with an accuracy of 65%. This accuracy was relatively lower than that obtained with PFGE or MLST, when the same set of isolates were evaluated.

Ashton et al. (2016) compared the serovars predicted by using legacy MLST sequences extracted from WGS data to the results generated by conventional serotyping, for 7,338 isolates representing 263 serovars of Salmonella enterica subspecies I. The 10 most common serovars in this S. enterica subspecies I dataset were serovars Enteritidis, Typhimurium, Infantis, Typhi, Newport, Virchow, Kentucky, Stanley, Paratyphi A, and Java. They found that the serovar prediction accuracy of legacy MLST was 96%.

The overall serovar-prediction accuracy for the CRISPR subtyping approach has been reported to range from 78 to 90% (Liu et al., 2011; Fabre et al., 2012; Shi et al., 2015). More studies are needed to further assess serovar-prediction accuracy using CRISPR.

Given the range of serovars represented in the SeqSero and SISTR databases, WGS can be used to theoretically predict 2,389 and 2,190 of the 2,577 serovars described in the White–Kauffmann–Le minor when using the serovar prediction programs SeqSero (Zhang et al., 2015) and SISTR (Yoshida et al., 2016a), respectively. Using empirical data, the accuracy of serotype prediction with SeqSero and SISTR has been reported to be approx. 92 and 95%, respectively (Zhang et al., 2015; Yoshida et al., 2016a,b; Robertson et al., 2018). By comparison, traditional Salmonella serotyping had an accuracy of 73% when 33–36 independent laboratories performed serotyping of the same eight Salmonella strains representing seven different serovars (Petersen et al., 2002), suggesting that WGS-based methods may be more reliable than traditional serotyping to assign Salmonella isolates to serovars. Nevertheless, further experimental studies are needed to continue to quantify the ability of WGS-based methods to identify Salmonella serovars.

### Comparison of Molecular Methods for Subtype Differentiation of Salmonella

Molecular methods are used for subtyping Salmonella isolates that belong to the same serovar, as well as being used for serovar prediction. This section briefly provides some examples of comparative studies of subtyping methods. In one study, PFGE

TABLE 2 | Comparison of molecular characterization methods for prediction of Salmonella<sup>1</sup> serovars.


<sup>1</sup>This table is revised from the information provided by the review of Shi et al. (2015).

TABLE 3 | Examples of serovars incorrectly predicted by PFGE.


was compared to MLVA to subtype 163 non-typhoidal Salmonella isolates representing 15 serovars; MLVA differentiated the isolates into 79 MLVA subtypes while PFGE differentiated the same isolates into 87 subtypes. The Nei's diversity index for MLVA was 0.979 compared to 0.999 for PFGE (Kjeldsen et al., 2016). However, for specific serovars (e.g., Salmonella Enteritidis)

MLVA has been reported to provide improved discriminatory power over PFGE (Boxrud et al., 2007; Beranek et al., 2009; De Cesare et al., 2015). MLST has the advantage of being highly reproducible and easily transferable among laboratories. However, in a study of 110 Salmonella isolates from 25 serovars (Torpdahl et al., 2005), MLST resulted in 43 sequence types, while PFGE was able to differentiate the isolates into 73 PFGE subtypes. The downside of PFGE in this study was the inability to type 11 of the 110 (10%) isolates. In a study comparing different molecular methods to differentiate 52 Salmonella Enteritidis isolates, PFGE resulted in eight subtypes, while MLVA resulted in 18 subtypes and WGS resulted in 34 subtypes. The discriminatory power of PFGE, MLVA, and WGS was 0.81, 0.92, and 0.97 (Simpson's index of diversity), respectively (Deng et al., 2015). In another study, PFGE and WGS were used to differentiate 55 Salmonella Enteritidis isolates; PFGE resulted in 10 subtypes; however, WGS was able to further differentiate the isolates into 45 unique subtypes (Taylor et al., 2015), showing the greater discriminatory power of WGS over PFGE. In a study of isolates from a Salmonella Poona outbreak (Kozyreva et al., 2016), 4 PFGE subtypes and 7 WGS subtypes were observed among the 16 isolates; in silico MLST using the WGS data resulted in one MLST sequence type. Phylogenetic analysis using WGS data showed that the distinct PFGE types did not necessarily correlate with increased genetic distance between isolates. Isolates that differed by 0 SNPs showed distinct PFGE subtypes, suggesting that PFGE results would be misleading for these isolates (Kozyreva et al., 2016). While the relative discriminatory power of different subtyping methods depends on the strains and serovars tested, WGS methods were consistently found to be most discriminatory, followed by PFGE. While some MLVA schemes provide enhanced discriminatory power over PFGE for some serovars, for other serovars PFGE may be more discriminatory than MLVA.

### CRITERIA TO EVALUATE AND VALIDATE DIFFERENT Salmonella CHARACTERIZATION METHODS

Molecular-based Salmonella characterization methods including WGS are evolving very fast. Many of the characterization methods and technologies, as well as data analysis pipelines, are operated as research tools, and are under continuous development. Evaluation of these tools for Salmonella investigation, especially for those serovars/strains highly relevant to food products and processing environments, is pre-requisite for the implementation of these methods. Methods that can be used by the food industry must be thoroughly validated before implementation to ensure reliability and consistency of the method when it is used across different laboratories. Validation should cover the end-to-end workflow for source tracking from isolate subculture to bioinformatic analysis, articulating the key quality requirements and criteria (Ferrari et al., 2017; Nadon et al., 2017; Portmann et al., 2018). Proposed criteria for evaluation of Salmonella characterization methods for potential routine use in the food industry are shown below (**Table 4**).

### IMPLEMENTATION OF MOLECULAR-BASED Salmonella SUBTYPING METHODS BY THE FOOD INDUSTRY

We consider that WGS is the most suitable method to characterize Salmonella for incident investigation at production facilities in the food industry. This opinion is based on comparison of the resolution, turnaround time, ability of serovar prediction, cost, and feasibility of the available methods. Bioinformatics is a key capability required for WGS. The food industry may choose to invest in in-house capability that can interface with outside resources (e.g., academic partners, industry partners, government agencies), however, there are also opportunities to outsource data analyses to commercial or academic partner labs. Both the CFSAN pipeline and the Lyve-SET pipeline have been widely tested and seem to provide comparable and reliable results for hqSNP analysis. Implementation of wgMLST and cgMLST within BioNumerics has been successfully completed for L. monocytogenes in the United States. A cgMLST scheme is publicly available from EnteroBase (EnteroBase URL: https://enterobase.warwick.ac.uk/) and it is likely to be implemented within BioNumerics in the future. Other data analysis methods such as genome distance analysis (Pinho et al., 2009; Auch et al., 2010) can also become possible future approaches that allow for the food industry to develop data analysis capabilities for contamination source tracking.

The turnaround time of in-house WGS subtyping can be comparable to many conventional subtyping methods including conventional serotyping and PFGE (**Table 1**). WGS, however, provides much more information about an isolate with one single experimental procedure, enabling full characterization of the pathogen (including in silico serovar prediction and antimicrobial resistance gene identification) and more accurate clustering/discrimination of the isolates investigated. This is faster than using multiple conventional subtyping approaches in a stepwise approach to get equal information. The cost of WGS is also comparable to that of the conventional subtyping tools, considering the high quality and volume of information provided by WGS within one experimental procedure. In silico serotyping should be performed instead of traditional serotyping for determination of serovars once WGS is implemented as the subtyping method for Salmonella. This approach will greatly reduce the costs and time associated with serotyping.

Legacy MLST targeting variants of seven housekeeping genes of Salmonella can be used in combination with WGS. While legacy MLST classification can be obtained using Sanger sequencing technology (also known as first-generation sequencing technology) within 1 week, it can also be obtained by using the sequence information extracted from WGS data. Although legacy MLST has relatively lower discriminatory


 industry

1

.

fmicb-10-01591 July 11, 2019 Time: 13:8 # 14


TABLE 4 |

Continued


power compared with PFGE and MLVA, it is faster than PFGE when using an in-house Sanger sequencer such as Applied Biosystems Genetic Analysis Systems (Thermo Fisher Scientific). It is also more universal to all Salmonella serovars than MLVA which usually requires a specific scheme for each serovar. In addition, the serovar prediction ability of legacy MLST has been demonstrated to be comparable to that of PFGE (**Tables 1**, **3**).

PFGE is currently still the "gold standard" and most widely used Salmonella DNA fingerprinting method used by public health authorities and food regulators to characterize and track this pathogen in outbreaks, although it is being replaced by WGS. PFGE remains a valuable tool for foodborne pathogen characterization by the food industry, while a transition to WGS occurs. PFGE has been repeatedly shown to be more discriminatory than methods such as conventional serotyping, automated ribotyping, or MLST for many bacteria including Salmonella. In addition to these methods, single-plex or multiplex PCR assays that can detect and identify specific Salmonella serotypes have been described (Kim et al., 2006; Akiba et al., 2011; Zhu et al., 2015; Xiong et al., 2018; Xu L. et al., 2018; Xu Y. et al., 2018); these tools provide an alternative approach for detection and identification of specific Salmonella serovars.

The results of any subtyping approach can be used to assess the relationship of isolates in an investigation. Nevertheless, the epidemiological context is indispensable in final decision making in incident investigation and to determine further actions for food safety management improvement. High-resolution WGS subtyping results should not be interpreted in the absence of epidemiological information.

The raw sequence data generated by molecular-based subtyping methods, especially WGS, require both physical and virtual space for storage. It is desirable to retain the original sequence reads (usually files with >200 MB for each Salmonella isolate) for potential future analysis using alternative data analysis methods or for a retrospective investigation. Commercial clouds can provide a storage solution, provided that special attention is paid to data security. A robust Internet connection and high band-width is needed to transfer WGS data if data storage is outsourced. Subtyping analysis needs to be supported by complete metadata providing the relevant epidemiological context to identify the root cause of the contamination. Thus, the capability for metadata collection, organization, and storage is needed together with building the capability for WGS. The metadata should include information such as the geographic and temporal background of the isolates, the sample type, and sample source (e.g., raw ingredients, finished products, environment), etc. The Consortium for Sequencing the Food Supply Chain, founded by IBM and Mars Incorporated, represents industrial groups putting effort into collecting genomic information on pathogenic bacteria across the food supply chain<sup>6</sup> . This consortium represents one part of the broader goal to increase knowledge of foodborne pathogens at the genomic level.

fmicb-10-01591 July 11, 2019 Time: 13:8 # 16

TABLE 4 |

Continued (Kaufmann–White

 Le Minor scheme) is around 90% taking the typeability and accuracy of it into consideration

 (Bopp et al., 2016).

<sup>6</sup>https://researcher.watson.ibm.com/researcher/view\_group.php?id=9635

## CONCLUSION

fmicb-10-01591 July 11, 2019 Time: 13:8 # 17

The application of DNA-based methods for characterization of pathogens such as Salmonella has become common practice. Our literature-based assessment supports the superior discriminatory power of WGS and its advantages compared with other methods for Salmonella subtyping and source tracking for the food industry. We also identified circumstances under which use of other subtyping methods may be warranted. Implementation of molecular-based Salmonella characterization methods, including WGS, provides improvement of source tracking and root cause elimination; however, these methods require investment in bioinformatics capability. Routine use of WGS or complete replacement of current subtyping methods by WGS will require attention to key issues including standardization, robustness, and validation of the analytical methodology. High resolution WGS subtyping of Salmonella promises to vastly improve the ability of the food industry to track and control Salmonella and is poised to become standard methodology in food safety for characterization of foodborne pathogens by public health authorities and food regulators. Nevertheless, standardization of WGS operation and data analysis, in particularly source tracking analysis, is required at a global level. A common agreement of understanding and the application of WGS between the food industry, public health, and

### REFERENCES


food safety regulators are expected to guide the implementation of WGS in food safety management.

### AUTHOR CONTRIBUTIONS

ST and MW conceived and designed the work. ST, RO, and HL collected the data, conducted data analysis, and interpreted it. ST, RO, HL, and MW drafted the article. MW, AS, RB, CG, and GZ critically revised the article.

### FUNDING

The authors declare that this study received funding from Mars Global Food Safety Center. The funder was not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

### ACKNOWLEDGMENTS

We thank Peter Markwell, Dr. Bala Ganesan, and Dr. Kristel Hauben for comments that greatly improved the manuscript.




acquired infections in Finland by phage typing, antimicrobial susceptibility testing, PFGE and MLVA. BMC Microbiol. 15:131. doi: 10.1186/s12866-015- 0467-8


assets/PulseNet/uploads/pfge/PNL05\_Ec-Sal-ShigPFGEprotocol.pdf (accessed March 24, 2013).


fmicb-10-01591 July 11, 2019 Time: 13:8 # 21


probes in epidemiologic investigations of foodborne, diarrheal diseases. Int. J. Food Microbiol. 12, 77–89. doi: 10.1016/0168-1605(91)90049-U


fmicb-10-01591 July 11, 2019 Time: 13:8 # 22


**Conflict of Interest Statement:** ST, HL, CG, GZ, RB, and AS were employed by the Mars Global Food Safety Center. MW serves as a compensated scientific advisor for BioMérieux, Mérieux NutriSciences, Mars, and Neogen and has served as a paid speaker for 3M and IBM.

The remaining 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 © 2019 Tang, Orsi, Luo, Ge, Zhang, Baker, Stevenson and Wiedmann. 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.

# Composition and Dynamics of Bacterial Communities in a Full-Scale Mineral Water Treatment Plant

Lei Wei1,2, Qingping Wu<sup>2</sup> \*, Jumei Zhang<sup>2</sup> , Weipeng Guo<sup>2</sup> , Qihui Gu<sup>2</sup> , Huiqing Wu<sup>2</sup> , Juan Wang<sup>3</sup> , Tao Lei<sup>2</sup> , Moutong Chen<sup>2</sup> , Musheng Wu<sup>4</sup> and Aimei Li<sup>4</sup>

<sup>1</sup> School of Bioscience and Bioengineering, South China University of Technology, Guangzhou, China, <sup>2</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 Institute of Microbiology, Guangdong Academy of Science, Guangzhou, China, <sup>3</sup> College of Food Science, South China Agricultural University, Guangzhou, China, <sup>4</sup> Guangdong Dinghu Mountain Spring Company Limited, Zhaoqing, China

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

### Reviewed by:

Margarita Kambourova, Institute of Microbiology (BAS), Bulgaria James Scott Maki, Marquette University, United States Carmen Wacher, National Autonomous University of Mexico, Mexico

> \*Correspondence: Qingping Wu wuqp203@163.com

### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 07 January 2019 Accepted: 20 June 2019 Published: 24 July 2019

#### Citation:

Wei L, Wu Q, Zhang J, Guo W, Gu Q, Wu H, Wang J, Lei T, Chen M, Wu M and Li A (2019) Composition and Dynamics of Bacterial Communities in a Full-Scale Mineral Water Treatment Plant. Front. Microbiol. 10:1542. doi: 10.3389/fmicb.2019.01542 The aim of this study was to gain insight into the bacterial composition and dynamics in a mineral water treatment system (MWTS). The bacterial community of a fullscale mineral water treatment plant in the Maofeng Mountain, South China, was studied using high-throughput sequencing combined with cultivation-based techniques in both the dry and wet season. Overall, adenosine tri-phosphate (ATP) concentration (6.47 × 10−<sup>11</sup> – 3.32 × 10−<sup>8</sup> M) and heterotrophic plate counts (HPC) (3 – 1.29 × 10<sup>3</sup> CFU/mL) of water samples in the wet season were lower than those (ATP concentration 5.10 × 10−<sup>11</sup> – 6.96 × 10−<sup>8</sup> M, HPC 2 – 1.97 × 10<sup>3</sup> CFU/mL) in the dry season throughout the whole MWTS. The microbial activity and biomass of water samples obviously changed along with treatment process. All 300 isolates obtained using cultivation-based techniques were distributed in 5 phyla, 7 classes, and 19 genera. Proteobacteria accounted for 55.7% (167) of the total isolates, among which predominant genus was Pseudomonas (19.3%). Illumina sequencing analysis of 16s rRNA genes revealed 15 bacterial phyla (relative abundance >0.1%) as being identified in all water samples. Among these, Proteobacteria constituted the dominant bacteria microbiota in all water samples. A large shift in the proportion of Bacteroidetes, Actinobacteria, and Firmicutes was obtained during the treatment process, with the proportion of Bacteroidetes, Actinobacteria decreasing sharply, whereas that of Firmicutes increased and predominated in the final water product. The core microbiome, which was still present in whole MWTS comprised several genera including Pseudomonas, Acinetobacter, Clostridium, and Mycobacterium, that contain species that are opportunistic pathogens, suggesting a potential threat for mineral water microbiology safety. This study is the first to investigate the bacterial community of a fullscale mineral water treatment plant in China. The results provided data regarding the bacteria composition and dynamics in an MWTS, which will contribute to the beneficial manipulation of the mineral water microbiome.

Keywords: mineral water treatment system, bacterial community, high-throughput sequencing, core microbiome, opportunistic pathogens

### INTRODUCTION

fmicb-10-01542 July 23, 2019 Time: 17:16 # 2

Large-scale outbreaks of infectious diseases via waterborne pathogens constitute a public serious health problem (Faria et al., 2009; Liu et al., 2018). It is cautiously estimated that approximately 350 million people worldwide are infected with waterborne pathogens and 25 million children annually die from diseases caused by drinking contaminated water (Eichler et al., 2006; Deere et al., 2017). Although numerous strictly regulated microbial indicators exist for mineral water, the production of mineral water that meets quality standards does not necessarily guarantee microbial safety (Varga, 2011; Zhang et al., 2015). In particular, several recent studies have found that mineral water contains pathogens and opportunistic pathogens, which can adversely affect the health of consumers (Herath et al., 2012; Wei et al., 2017). The bacterial community is ubiquitous throughout the entire mineral water treatment system (MWTS) (Bharath et al., 2003; Pontara et al., 2011). Therefore, a comprehensive understanding of the composition and dynamics of bacterial communities in the MWTS is essential to guarantee the microbial safety of mineral water.

The spatiotemporal heterogeneity of microbial communities in drinking water treatment system (DWTS) has been investigated in several previous studies. For example, Pinto et al. (2014) found that the microbial community structure exhibits a large temporal change in DWTS. However, some studies have also shown that the primary dynamic pattern of variations in bacterial communities occur during the treatment process rather than via temporal fluctuations (El-Chakhtoura et al., 2015; Li et al., 2017). Hou et al. (2018) reported that microbial community structure had temporal change in DWTS and the treatment processes exerted profound influence on the microbial community composition. However, compared to DWTS, little information is available regarding the spatiotemporal variations of multi-step treatment processes in MWTS, which cannot guarantee the microbial safety of the final mineral water product.

The production process of mineral water mainly includes three levels of filter (quartz sand, activated carbon, and fine filter), ozone sterilization, filling and capping, and light inspection of the finished product (Zhang et al., 2011; Wei et al., 2017). Although some studies have focused on the raw water and the final water product in MWTS, the information is still lacking regarding on the spatiotemporal changes of the bacterial communities in the whole treatment process. Moreover, how each treatment process and season affects the bacterial community structure of mineral water remains unclear, and it is unknown whether common microbial patterns exist in the MWTS. This information was very important for ensuring the microbial safety of mineral water production. In addition, most studies regarding the bacterial communities have been conducted using a single method, such as cultivation-based techniques or molecular biological methods. However, as 90% of microorganisms are uncultured, the results of cultivation-based techniques cannot accurately reflect the actual information of bacterial communities (Griebler and Lueders, 2009; Albertsen et al., 2013). Molecular biology technology which avoids the technical difficulty of obtaining uncultured microorganisms, has fundamental disadvantages. Molecular biology technology does not allow an accurate understanding of the physiological and metabolic characteristics of microorganisms, which is unable to be effectively applied for the control and utilization of microorganisms (Hahn et al., 2016; Pulschen et al., 2017). Therefore, the actual composition and characteristics of the mineral water microbial community needs to be explored in more detail through the use of high-throughput sequencing analysis combined with cultivation techniques.

To our knowledge, a systematic survey of the bacterial community in MWTS has not previously been conducted in China. In the present study, the bacterial community of a fullscale mineral water treatment plant in the Maofeng Mountain, South China was studied using Illumina HiSeq sequencing analyses combined with cultivation-based techniques in both dry and wet season. The aims of this study were to understand the composition and dynamics of the bacterial community in whole MWTS and investigate whether potential pathogens and opportunistic pathogens exist across the mineral water. The information generated in this study will provide strong basis for ensuring the microbial safety of mineral water.

### MATERIALS AND METHODS

### Study Site and Sample Collection

The mineral water treatment plant is located in the Maofeng Mountain, south China. Its raw water originates from the groundwater at a depth of 200 meter. The production process of mineral water is shown in **Figure 1**. Water samples were collected in sterile containers from raw water (A), quartz sand filtered water (B), activated carbon filtered water (C), fine filtered water (D), and final water product (E) in the wet season and dry season in 2017. For each sampling point, three replicate samples were collected. All water samples were maintained at 4◦C during transportation and were transported to laboratory within 4 h after sample collection. After all the samples arrived in the laboratory, 2 L of raw water, 5 L of quartz sand filtered water, 5 L of activated carbon filtered water, 10 L of fine filtered water and 10 L of final water product were filtered through sterile 0.22 µm pore-sized polycarbonate membranes (Millipore, Billerica, MA, United States). Membranes were transferred to sterile tubes and stored at −80◦C. Each water sample was retested three times. The remainders of the sample were stored at 4◦C, until all assayed parameters were confirmed.

### Water Quality Analysis

To understand the water quality of all water samples, several standard physicochemical parameters were measured. Turbidity was measured using a turbidimeter (Mettler Toledo, Zurich, Switzerland). Temperature and pH was measured in situ using a multi-parameter water quality monitoring sonde (INESA Scientific, Shanghai, China). Other physicochemical parameters, including ammonia nitrogen, chemical oxygen demand (COD), total organic carbon (TOC) were measured according to standard examination methods for drinking water (GB/T 5750-2006, China). Ammonia nitrogen was determined by spectrophotometry. COD was measured by titration. TOC was

measured using a total machine carbon analyzer (Shimadzu, Kyoto, Japan). All determinations were performed in triplicate. As groundwater was confined water and the dissolved oxygen content is very low, the parameter of dissolved oxygen (DO) was not measured.

### Microbial Biomass and Activity

Heterotrophic plate counts (HPC) of all water samples were determined using a spread plating method on Reasoner's 2A (R2A) agar (Haibo Co., Qingdao, China) (Lehtola et al., 2004). Briefly, all water samples were diluted to the appropriate concentration and 0.2 mL diluent was spread on R2A agar plate. The plates were cultured at 20◦C for 7 days before colony-forming units (CFU) counting. The microbial activity was characterized by measuring adenosine tri-phosphate (ATP). ATP concentration was detected using a bioluminescence assay kit (Huankai Co., Guangzhou, China) described by Zhang et al. (2011). Briefly, membranes that filter water samples were fully washed with 0.5 mL ddH2O, and were reacted with 0.5 mL ATP-releasing reagent for 2 min. 0.1 mL reaction solution, 0.2 mL buffer solution, 0.1 mL luminescent reagent were added into a cuvette in sequence and luminous pulse value was detected by Glomax 20/20 luminometer (Promega Co., Madison, WI, United States). The ATP concentration of each sample was calculated according to a calibration curve generated using standards of known ATP. All determinations were performed in triplicate.

### Isolation and Identification of Bacteria From Raw Water

Three kinds of medium, nutrient agar (NA) (Huankai Co., Guangzhou, China), trypticase soy agar (TSA) (Huankai Co., Guangzhou, China), and R2A (Haibo Co., Qingdao, China) were used as separate medium for the isolation of bacteria from raw water. Following incubation in 20◦C culture for 7 days, colonies with different morphologies (color, size, edge, and transparency) were purified using the streaking method. After purification, all the isolates were cultured with R2A liquid medium to preserve and extract the genomic DNA. The genomic DNA of the isolate was extracted according to the specifications of the Bacterial Genomic DNA Purification Kit (Dongsheng Biotech, Guangzhou, China). Primers 27F (5<sup>0</sup> - GTGCTGCAGAGAGTTTGATCCTGGCTCAG-3<sup>0</sup> ) and 1492R (50 -CACGGATCCTACGGGTACCTTGTTACGACTT-3<sup>0</sup> ) were used to amplify the 16S rRNA gene of each isolate (Dymock et al., 1996). The polymerase chain reaction (PCR) mixture contained 2 µL of template, 12.5 µL PCR Taq-mix, 0.5 µL of each primer, and 8.5 µL of ddH2O. Amplifications were performed using a

DNA thermocycler under the following temperature profiles: 5 min of denaturation at 95◦C and 35 cycles of 30 s at 95◦C, 45 s at 56◦C, and 1.5 min at 72◦C, followed by an additional 10 min at 72◦C at the end of the 35th cycle for repair and extension. The PCR products were electrophoresed using 1.0% (W/V) agarose gel, and the positive PCR products were selected and sent to Huada Gene Co., (Guangzhou, China) for sequencing using the PCR primers. According to a previously described method (Kim et al., 2012), 16S rDNA sequences were divided into different Operational Taxonomic Units (OTUs) and representative 16S rDNA sequences were selected in each OTU. Representative 16S rDNA sequences were submitted to the National Center for Biotechnology Information (NCBI) and GenBank databases for comparison and identification using BLAST software.

### Microbial Community Analysis

The polycarbonate membranes that had been used to analyze the microbial community were cut into pieces with sterilized scissors. Total genomic DNA of all the samples was extracted according to the specifications of the Powersoil DNA extraction kit (Mobio Laboratories, Carlsbad, CA, United States). PCR amplicon libraries were constructed for Illumina HiSeq sequencing using bacterial primers 341F (5<sup>0</sup> - CCTAYGGGRBGCASCAG-3 0 ) and 806R (5<sup>0</sup> -GGACTACHVGGGTWTCTAAT-3<sup>0</sup> ) targeting the V3+V4 hypervariable regions of the 16S rRNA genes (Wu et al., 2015). The PCR mixture (20 µL) contained 10 ng of template DNA, 4 mL of 5× FastPfu Buffer, 2.5 mM dNTPs, 0.4 mL FastPfu Polymerase, 5 M of each primer, and ddH2O. Amplifications were performed as follows: an initial denaturation at 95◦C for 5 min; followed by 27 cycles at 95◦C for 30 s, 55◦C for 30 s, and 72◦C for 45 s; and a final extension at 72◦C for 10 min. The obtained PCR products were purified and joined using a sequencing linker. After constructing the gene library, the modified products were subjected to highthroughput sequencing using Illumina Hiseq2500 (San Diego, CA, United States). The reads from the original DNA fragments were merged using FLASH<sup>1</sup> (Magoc and Salzberg, 2011 ˇ ). Quality filtering was carried out on the initial data obtained by sequencing to obtain more accurate and high-quality DNA sequence. To obtain high-quality sequences (primer mismatch base number <1%, sequence length 200–500 bp, no chimeras), the original sequences were processed using the protocol described by Caporaso et al. (2010). Sequences were grouped into OTUs at 97% sequence similarity using Mothur version 1.34.0<sup>2</sup> . To obtain the species classification information of each OTU, the representative sequences for each OTU were selected and submitted for taxonomic identification using the Ribosomal Database Project (RDP) classifier (see text footnote 2) (Wang et al., 2007). Mothur version 1.34.0 was used as the processing pipeline for calculation of alpha-diversity indices including the Chao1, Simpson, and Shannon indices. Based on the relative abundance of bacterial phyla, unweighted UniFrac with QIIME<sup>3</sup> was used for the unweighted pair-group method with arithmetic mean clustering. Principal component analysis (PCA) and canonical correspondence analysis (CCA) were conducted using Canoco 4.5 software<sup>4</sup> .

### Data Analysis

All the data were processed using Microsoft Excel 2010 software (Redmond, WA, United States) and statistical analysis was performed using IBM SPSSV20.0 software (Armonk, NY, United States). The differences between groups were analyzed using one-way analysis of variance (ANOVA) and the statistical significance level was set at p = 0.05.

### RESULTS

### Water Quality

As shown in **Table 1**, a series of specific physicochemical parameters were determined for all water samples. The raw water temperature in the wet season and the dry season were approximately 20.0◦C and 20.9◦C, respectively, and has no obvious change in the follow-up process treatment. The pH of all the water samples was stable in the range of 7.3 to 7.5 across the whole MWTS in both seasons. The concentrations of ammonia in whole MWTS were <0.02 mg/L in both seasons. The turbidity, TOC, and CODMn of raw water samples were 0.60 nephelometric turbidity units (NTU), 0.61 and 0.41 mg/L in the wet season, and 0.70 NTU, 0.74, and 0.45 mg/L in the dry season, respectively. As the seasons changed, the turbidity, TOC, and CODMn of the water samples showed no significant difference in whole MWTS (p > 0.05). However, the turbidity, TOC, and CODMn of the water samples were changed by most treatment processes (p < 0.05).

### Microbial Biomass and Microbial Activity

The HPC from raw water in the wet and dry season was 1.29 × 10<sup>3</sup> and 1.97 × 10<sup>3</sup> CFU/mL, respectively (**Table 2**). Along with the treatment process in whole MWTS, the HPC showed a downward trend in both seasons (**Figure 2A**). After filtration by quartz sand and fine filter, the microbial biomass decreased sharply (p < 0.01). However, after activated carbon filtration, the HPC showed no significant change. The HPC of all finished water in both seasons was <4 CFU/mL. The microbial activity of water samples in the wet season was lower than that in the dry season in whole MWTS, especially in the raw water and activated carbon filtered water (p < 0.01). ATP concentration during the treatment process ranged from 3.32 × 10−<sup>8</sup> to 6.47 × 10−<sup>11</sup> M in the wet season and 6.96 × 10−<sup>8</sup> to 5.10 × 10−<sup>11</sup> M in the dry season (**Figure 2B**). Along with treatment process, the ATP concentration was obviously reduced after quartz sand filter, and then exhibited a significant increase after activated carbon filter in both season (p < 0.01). The ATP concentration of activated carbon filtered water was 4.20 × 10−<sup>9</sup> M in the wet season and 6.81 × 10−<sup>9</sup> M in the dry season.

<sup>1</sup>http://ccb.jhu.edu/software/FLASH/

<sup>2</sup>https://mothur.org

<sup>3</sup>http://qiime.org/scripts/make\_phylogeny.html

<sup>4</sup>http://www.canoco5.com/


A, represents raw water; B, represents quartz sand filtered water; C, represents activated carbon filtered water; D, represents fine filtered water; E, represents final water product. ∗∗represents sample in dry season significantly different from sample in wet season (p < 0.01); #represents the sample significantly different from its front sample (p < 0.05); ##represents the sample significantly different from its front sample (p < 0.01).

### Isolation and Identification of Bacteria From Groundwater

In total, 141 isolates from the wet season and 159 isolates from the dry season were isolated in raw water samples (**Supplementary Table S1**). A total of 139 isolates were cultured from R2A medium, 93 isolates from NA medium, and 68 isolates from TSA medium. According to 16s rRNA gene sequence, all the isolates could be divided into 23 OTUs (**Table 3**). All 300 isolates were distributed in 5 phyla (Proteobacteria, Chloroflexi, Firmicutes, Bacteroides, and Actinobacteria), 7 classes (α-proteobacteria, β-proteobacteria, γ-proteobacteria, Chloroflexi, Bacilli, Flavobacteriia, and Actinomycete) and 19 genera (Pseudomonas, Chryseobacterium,

TABLE 2 | Heterotrophic plate counts (HPC) and bacterial adenosine tri-phosphate (ATP) concentrations of 5 water samples in whole MWTS.


A, represents raw water; B, represents quartz sand filtered water; C, represents activated carbon filtered water; D, represents fine filtered water; E, represents final water product. <sup>∗</sup> represents sample in dry season significantly different from sample in wet season (p < 0.05); ∗∗represents sample in dry season significantly different from sample in wet season (p < 0.01); # represents the sample significantly different from its front sample (p < 0.05); ##represents the sample significantly different from its front sample (p < 0.01).

Porphyrobacter, Xanthomonas, Brevundimonas, Caulobacter, Acinetobacter, Sphingomonas, Bacillus, Brevibacterium, Arthrobacter, Aquimonas, Chloroflexus, Mycobacterium, Roseomonas, Chromobacterium, Pimelobacter, Microbacterium, and Hydrogenophaga). Proteobacteria accounted for 55.7% (167) of the total isolates, followed by Actinobacteria accounting for 23.7% (71). Among all the isolates, the predominant genera were Pseudomonas (19.3%) and Sphingomonas (7.7%) at genus level. All the isolates cultured from R2A medium were distributed in 18 different genera, and Porphyrobacter, Bacillus, Chromobacterium, Roseomonas, and Pimelobacter were isolated only on R2A medium. Mycobacterium and Pimelobacter were isolated only from the sample in the wet season.

### Diversity of Bacterial Community

A range of 20,693 to 39,643 original bacterial 16S rRNA gene sequences were extracted from each sample using Illumina HiSeq sequencing analysis (**Table 4**). The OTUs of the water samples in the wet season ranged from 2411 to 6607, and ranged 2524 to 7391 for sample in the dry season. As shown in **Figure 3**, In general, the diversity indices of water samples in the dry season were comparable to those in the wet season. Among all the water samples in both seasons, the diversity indices of raw water samples were the highest. Chao1 richness index ranged from 15421 to 38316 in the wet season, and ranged 15567 to 33647 in the dry season. Shannon index ranged from 3.19 to 5.55 in the wet season, and ranged 3.46 to 6.49 in the dry season. Along with treatment process in whole MWTS, the richness index of Chao1 and the Shannon index gradually decreased to the lowest level and were significantly changed by the fine filtration and ozone sterilization treatments. Simpson diversity index of finished water was 0.1872 in the wet season and 0.1819 in the dry season. However, those of raw water reached 0.0245 and 0.0158 in the wet and dry seasons, respectively.

### Composition of Bacterial Community

The bacterial community composition of all the water samples was identified at the phylum level (**Figure 4A**). Overall, 15 bacterial phyla (relative abundance >0.1%) including Proteobacteria, Verrucomicrobia, Planctomycetes, Nitrospirae, Gemmatimonadetes, Firmicutes, Cyanobacteria, Chloroflexi,

TABLE 3 | Blast results and classify information of representative isolates cultured from raw water sample.


The operational taxonomic units (OTUs) were defined with 97% similarity.

Bacteroidetes, Actinobacteria, Dependentiae (Tm6), Deinococcus– Thermus, Saccharibacteria, Chlamydiae, and Acidobacteria were identified in all the water samples. No obvious difference was observed upon comparison of the bacterial microbiota of water samples in different seasons. Proteobacteria comprised the dominant bacteria microbiota in all the water samples, accounting for approximately 60% of all the samples, especially in quartz filtered water (69.7%) and activated carbon filtered water (67.5%). The relative abundance of Proteobacteria from samples in the wet season was higher than those in the dry season. Along with the treatment processes, significant change in the proportion of Bacteroidetes, Actinobacteria, and Firmicutes was obtained in the MWTS. The relative abundances of Bacteroidetes and Actinobacteria in raw water were 22.3 and 16.1%, respectively. The proportion of Bacteroidetes decreased initially in quartz sand filtered samples, followed by an increase upon activated carbon filtration; however, Bacteroidetes from final water product were present a very low levels (<1%) in both seasons. The proportion of Actinobacteria gradually decreased in both seasons, being only <1% in final water product. Among all bacterial microbiota, the largest change of relative abundance was observed for Firmicutes. The proportion of Firmicutes in raw water was 1.4% in the wet season and 3.4% in the dry season, but was highly abundant in finished water in both seasons (41.8 and 42.7%, respectively). The final water product in both seasons mainly comprised Proteobacteria (50.8%), Firmicutes (42.2%), and Verrucomicrobia (1.4%).

The bacterial community composition of all the water samples was identified at the genus level (**Figure 4B**). Overall, 25 bacterial genera (relative abundance >0.1%), including Citrobacter, Pseudomonas, Enhydrobacter, Acinetobacter, Sphingomonas, and Novosphingobium were detected in all the water samples. The proportion of unclassified genera was range from 60 to 70% in all the samples. The relative abundance of each genus was <3.5% in raw water. Quartz filtered water in both seasons mainly comprised Novosphingobium (5.5%) and Sphingomonas (4.6%), whereas activated carbon water was dominated by Pseudomonas

TABLE 4 | Bacterial community richness and diversity indices for 5 water samples in whole MWTS.


A, represents raw water; B, represents quartz sand filtered water; C, represents activated carbon filtered water; D, represents fine filtered water; E, represents final water product; <sup>∗</sup> represents sample in dry season significantly different from sample in wet season (p < 0.05); ∗∗represents sample in dry season significantly different from sample in wet season (p < 0.01); # represents the sample significantly different from its front sample (p < 0.05); ##represents the sample significantly different from its front sample (p < 0.01). <sup>a</sup> represents the operational taxonomic units (OTUs) were defined with 97% similarity.

(7.9%), Citrobacter (7.8%), and Acinetobacter (4.7%). The final water product in both seasons mainly comprised Pseudomonas (8.6%), followed by Acinetobacter (2.4%).

### Core Microbiome

The bacteria that were still present in whole MWTS were defined as the core microbiome (Shade and Handelsman, 2012). Venn diagrams were constructed to investigate the core microbiome in whole MWTS. As shown in **Figure 5**, 178 (1.9%) OTUs were shared in all the samples from the wet season and 265 (2.9%) from the dry season. Among these shared OTUs, 94 were shared in both seasons and were identified as Pseudomonadaceae (n = 20), Sphingomonadaceae (n = 9), Enterococcaceae (n = 8), unclassified AB024 (n = 7), AJ234 (n = 6), Moraxellaceae (n = 6), Xanthomonadaceae (n = 6), Comamonadaceae (n = 6), Citrobacteriaceae (n = 5), Caulobacteraceae (n = 5), Bacillaceae (n = 5), Micrococcaceae (n = 4), Microbacteriaceae (n = 3), Mycobacteriaceae (n = 3), and Nocardiaceae (n = 1). The number of shared OTUs between adjacent treatment steps ranged from 370 to 2127 in samples collected during the wet season and from 397 to 2601 in those collected during the dry season, accounting for 4.1 to 23.4% and 4.4 to 29.2% of overall OTUs, respectively.

### DISCUSSION

In this study, we identified that the microbial activity (**Figure 2B**) (ATP concentration 5.10 × 10−<sup>11</sup> – 6.96 × 10−<sup>8</sup> M) and biomass (**Figure 2A**) (HPC 2 – 1.97 × 10<sup>3</sup> CFU/mL) of mineral water samples from an MWTS in the dry season were higher than those (ATP concentration 6.47 × 10−<sup>11</sup> – 3.32 × 10−<sup>8</sup> M, HPC 3 – 1.29 × 10<sup>3</sup> CFU/mL) in the wet season. The seasonal fluctuation of microbial activity and biomass may be attributed to environmental factors (**Table 1**); specifically, the concentration of TOC. As TOC represents a carbon source, the changing of seasons might alter the nutrient composition of the water environment in the MWTS. The rise of TOC potentially promoting microbial activity and biomass shifts in the dry season in whole MWTS, which is consistent with the description of Li in DWTS (Li et al., 2016). The seasonal fluctuation of microbial activity and biomass had a positive correlation with turbidity and COD. Turbidity is related to suspended, which can adsorb and enrich bacteria. COD is an indicator of organic content and has direct relationship with bacterial biomass. In addition, previous studies in DWTS have shown that environmental parameters such as phosphate, sulfate, pH and water depth were the major factors controlling the temporal variation of bacterial activity and biomass (Lautenschlager et al., 2013; Ji et al., 2015). The temporal pattern of bacteria diversity in DWTS in particular has been thoroughly studied, whereas comparatively little research has been conducted on MWTS. Henne et al. (2013) found obvious seasonal variations of bacterial community composition in surface water reservoirs and Kelly et al. (2014) reported that the diversity of bacteria community in biofilms in the summer months was higher than that observed in the winter. Compared with previous studies in DWTS, the results of the present study showed that the bacterial diversity of water samples have no obvious seasonal variations and bacterial microbiota of water

samples in the different season exhibited no obvious difference (**Figure 3**). The differences in seasonal variations of microbial diversity in the DWTS and MWTS may have been driven by the strong seasonal changes in the DWTS. Previous study also suggested that physicochemical parameters, temperature, and pH, were correlated with bacterial diversity (Chen et al., 2017). In comparison, the stability of the groundwater environment from which the MWTS samples were derived resulted in no obvious seasonal difference of microbial diversity (Fovet et al., 2018). In addition, the groundwater that serves as the raw water of mineral water is not influenced by other external factors as is surface water. For example, after, rainfall or pollution, the surface water bacterial diversity demonstrates significant change (França et al., 2015). In our study, the kinds of bacterial phylum are significantly different from the results of studies in DWTS, which is due to different raw water in DWTS and MWTS.

Distinct differences were identified in the microbial activity, biomass, and diversity at each step of the MWTS. Along with treatment process in whole MWTS, the microbial activity, biomass, and diversity exhibited an obviously downward trend, especially, following the fine filtration and ozone sterilization treatments, which indicated that the water quality was gradually improved. However, after the activated carbon filter treatment process, the activity and diversity of bacteria showed a clear increase in both seasons. Activated carbon filter systems are commonly used for mineral water treatment to ensure equipment life, improve water quality, and prevent pollution (Feng et al., 2013). Activated carbon filters, which contain numerous pores and a large surface area, possess a strong physical adsorption capacity to absorb organic pollutants and microbes (Liao et al., 2013). Our previous study had shown that activated carbon filters could serve as a gathering place for opportunistic pathogens and is the most serious microbial contamination site throughout the whole production process of the MWTS (Wei et al., 2017). The results of the present study thus showed that there was dynamic bacterial community change among the samples of the different treatment processes and that specific treatment processes in the MWTS could affect the bacterial communities in unique ways (**Figure 4**). In comparison, Poitelon et al. (2010) reported that the influence of water treatment by coagulation-flocculation and sedimentation on bacterial communities is relatively small, whereas Li et al. (2017) observed that bacterial communities significant changed during these steps. Consistent with previous report, the present analysis revealed that Proteobacteria predominated in the water samples, which could mainly be related to the capacity of Proteobacteria with regard to the biodegradation or biotransformation of various organic compounds (Galvão et al., 2005). In addition, significant change in the proportions of the Bacteroidetes, Actinobacteria, and Firmicutes was also obtained in the MWTS. The activated carbon filtered water was mainly composed of Proteobacteria and Citrobacter, which differed from the result of DWTS (Lin et al., 2014). Furthermore, cluster analysis indicated that quartz sand filtration apparently changed the bacterial community structure, with Sphingomonas and Novosphingobium increasing to a higher level in the quartz sand filtered water, which demonstrated that quartz sand filter promoted the proportion of Sphingomonas and Novosphingobium. However, after ozone sterilization treatment, the proportion of Firmicutes exhibited a sharp increase. This result is consistent with the study of Mi et al. (2015) which indicated that disinfection strategies could have a profound impact on the proportion of Firmicutes in microbial communities in MWTS. The disinfection resistance mechanism of Firmicutes remains unresolved, which may be connected

with higher spore resistance. Therefore, additional studies are necessary to validate the effect of disinfection on Firmicutes content in mineral water treatment plant bacterial communities.

Despite the spatial-temporal variations of bacterial diversity was detected in MWTS, shared OTUs were found along the treatment processes in different seasons. The bacteria that were present throughout whole MWTS were defined as the core microbiome (Shade and Handelsman, 2012). In the present study, the core microbiome comprised 15 families, which indicated that the core microbiome in MWTS is generally difficult to remove completely and may gradually become a healthy threat (**Figure 5**). Therefore it is critical to comprehensively understand the core microbiome in MWTS. In particular, some genera, such as Pseudomonas and Acinetobacter, which are highly likely to contain pathogens and opportunistic pathogens, were retrieved from Illumina amplicons of final water product. The occurrence of these genera in mineral water may increase the risks of waterborne diseases and health problems. For example, some Acinetobacter species were important opportunistic pathogens that underlie hospital infections and can cause respiratory tract infection, septicemia, meningitis, endocarditis, and urogenital tract infection (Wong et al., 2017). Some Mycobacterium species were the pathogen that causes tuberculosis (Pfyffer, 2015). Among all 300 isolates obtained using cultivation-based techniques in raw water, Pseudomonas was predominant species (19.3%). Additionally, Illumina sequencing analysis of 16s rRNA genes showed that the final water product in both seasons mainly comprised Pseudomonas (8.6%). Pseudomonas aeruginosa serves as an important opportunistic pathogen that is frequently detected in mineral water for human consumption and causes human urinary tract infections (Lu et al., 2016). Our previous research showed that the Pseudomonas aeruginosa contamination rate of final water product was 2.3%, which can adversely affect the health of consumers (Wei et al., 2014). Moreover, many waterborne bacteria can form biofilms, which are difficult to remove (Flemming and Wingender, 2010). Although it is possible that PCR amplification might amplify DNA fragment from dead cells, there remains a possibility that some pathogens are able to survive the water treatments considering their resistances to disinfectants. Some pathogen may also cause diseases even at relatively low concentrations (Kirk et al., 2015). Illumina sequencing analysis of 16s rRNA genes in final water product showed that some opportunistic pathogens exist in the final water product that was disinfected by ozone. Hence, manufacturers of mineral water need to established measures for monitoring the MWTS.

This study is the first to investigate the composition and dynamics of bacterial community in a full-scale mineral water treatment plant in China during both the wet and dry seasons. In general, the microbial activity (ATP concentration 5.10 × 10−<sup>11</sup> – 6.96 × 10−<sup>8</sup> M) and biomass (HPC 2 – 1.97 × 10<sup>3</sup> CFU/mL)

### REFERENCES

Albertsen, M., Hugenholtz, P., Skarshewski, A., Nielsen, K. L., Tyson, G. W., and Nielsen, P. H. (2013). Genome sequences of rare, uncultured bacteria obtained of mineral water samples from an MWTS in the dry season were higher than those (ATP concentration 6.47 × 10−<sup>11</sup> – 3.32 × 10−<sup>8</sup> M, HPC 3 – 1.29 × 10<sup>3</sup> CFU/mL) in the wet season. All 300 isolates obtained using cultivation-based techniques were distributed in 5 phyla, 7 classes, and 19 genera. Proteobacteria accounted for 55.7% (167) of the total isolates and the predominant species comprised Pseudomonas (19.3%) at the genus level. Illumina sequencing analysis of 16s rRNA genes revealed that 15 bacterial phyla (relative abundance >0.1%) were detected in all the water samples and that Proteobacteria constituted the dominant bacteria microbiota. The bacterial diversity of water samples has no obvious seasonal variations and bacterial microbiota of water samples in the different season exhibited no obvious difference. Distinct differences were identified in the microbial activity, biomass, and diversity at each step of the MWTS. Along with treatment process in whole MWTS, the microbial activity, biomass, and diversity exhibited an obviously downward trend, especially, following the fine filtration and ozone sterilization treatments. In particular, some core microbiome that persisted throughout the treatment process requires special consideration. Notably, opportunistic pathogens including Pseudomonas, Acinetobacter, Clostridium, and Mycobacterium, which can adversely affect the health of consumers, were detected throughout the treatment process. These data may therefore provide useful information for the development of public health policies and effective strategies to ensure the safety of mineral water products.

### AUTHOR CONTRIBUTIONS

QW, JZ, and LW conceived and designed the experiments. LW and WG performed the experiments. LW, QG, and JW analyzed the data. MC, MW, and AL contributed reagents, materials, and analysis tools. LW, HW, and TL contributed to the writing of the manuscript.

### FUNDING

We would like to acknowledge the financial support of the Science and Technology Planning Project of Guangdong Province (2017A07070218) and the GDAS' Project of Science and Technology Development (2017GDASCX-0201).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.01542/full#supplementary-material

Bharath, J., Mosodeen, M., Motilal, S., Sandy, S., Sharma, S., Tessaro, T., et al. (2003). Microbial quality of domestic and imported

by differential coverage binning of multiple metagenomes. Nat. Biotechnol. 31:533. doi: 10.1038/nbt.2579

brands of bottled water in Trinidad. Int. J. Food Microbiol. 81, 53–62.


fmicb-10-01542 July 23, 2019 Time: 17:16 # 11


**Conflict of Interest Statement:** MW and AL were employed by Guangdong Dinghu Mountain Spring Company Limited.

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

# Prevalence, Genotypic Characteristics and Antibiotic Resistance of Listeria monocytogenes From Retail Foods in Bulk in Zhejiang Province, China

Yunyi Zhang<sup>1</sup>† , Shilei Dong<sup>2</sup>† , Honghu Chen<sup>1</sup> , Jiancai Chen<sup>1</sup> , Junyan Zhang<sup>1</sup> , Zhen Zhang<sup>1</sup> , Yong Yang<sup>1</sup> , Ziyan Xu<sup>3</sup> , Li Zhan<sup>1</sup> \* and Lingling Mei<sup>1</sup> \*

<sup>1</sup> Department of Microbiology, Zhejiang Provincial Center for Disease Control and Prevention, Hangzhou, China, <sup>2</sup> Department of Clinical Laboratory, Zhejiang Hospital, Hangzhou, China, <sup>3</sup> Department of Biotechnology, Wenzhou Medical University, Wenzhou, China

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Marc Heyndrickx, Institute for Agricultural and Fisheries Research (ILVO), Belgium Gonçalo Nieto Almeida, Instituto Nacional de Investigação Agrária e Veterinária, Portugal

#### \*Correspondence:

Li Zhan lzhan@cdc.zj.cn Lingling Mei llmei@cdc.zj.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 09 October 2018 Accepted: 11 July 2019 Published: 25 July 2019

#### Citation:

Zhang Y, Dong S, Chen H, Chen J, Zhang J, Zhang Z, Yang Y, Xu Z, Zhan L and Mei L (2019) Prevalence, Genotypic Characteristics and Antibiotic Resistance of Listeria monocytogenes From Retail Foods in Bulk in Zhejiang Province, China. Front. Microbiol. 10:1710. doi: 10.3389/fmicb.2019.01710 Listeria monocytogenes is an important foodborne pathogen causing public concern. A total of 3354 retail foods in bulk were sampled and screened for L. monocytogenes. Seventy-three (2.2%) samples including 21 ready-to-eat (RTE) foods and 52 raw foods were confirmed positive for L. monocytogenes. Sushi and salmon sashimi occupied the top two slots in RTE foods with relatively high presence rate of 12.9 and 6.9%, respectively. Meanwhile, L. monocytogenes was found to be distributed unequally in raw foods; the presence rates in raw meat (3.5%) and poultry (3.8%) were significantly higher than that in raw seafood (1.3%). Notably, L. monocytogenes was not detected in raw freshwater food. The L. monocytogenes isolates belonged to four serotypes, 1/2a, 1/2b, 1/2c, and 4b, with the most prevalent serotype being 1/2a (47.9%). Eighteen sequence types (STs) and eighteen virulence types (VTs) containing four newly assigned VTs (VT180, VT181, VT182, and VT183) were determined via multilocus sequence typing (MLST) and multi-virulence-locus sequence typing (MVLST). Among the 73 L. monocytogenes isolates, 23 (31.5%) belonged to epidemic clones (ECs) including ECI, ECIV, ECV, ECVI, ECVIII and ECXI among which ECV was predominant. Antibiotic susceptibility tests revealed a high resistance rate (11.0%) to tetracycline. Moreover, we identified the distribution patterns of virulence genes of four Listeria pathogenicity islands (LIPI) in L. monocytogenes isolates. prfA, hly, plcA, plcB, mpl, actA genes in LIPI-1 and inlA, inlB, inlC, inlJ genes in LIPI-2 were detected in approximately all L. monocytogenes isolates. The distribution of both LIPI-3 genes and LIPI-4 genes exhibited association with lineage and ST. LIPI-4 genes were present exclusively in ST87 isolates. Relatedness analysis revealed the absence of distinct association between STs, ECs, LIPI-3 and LIPI-4 distribution and specific food groups. This study provided fundamental data for Chinese food safety authorities to grasp the contamination status of L. monocytogenes in foods, assess the potential risk of this pathogen and further address the safety issue of retail foods in bulk in China.

Keywords: Listeria monocytogenes, prevalence, retail foods in bulk, antibiotic resistance, multilocus sequence typing (MLST), multi-virulence-locus sequence typing (MVLST), epidemic clone (EC)

## INTRODUCTION

fmicb-10-01710 July 24, 2019 Time: 14:54 # 2

Listeria monocytogenes is an important foodborne pathogen which can cause severe human listeriosis, particularly in older adults, newborns, pregnant women and immunecompromised individuals, resulting in septicemia, abortion, preterm delivery, stillbirth, meningitis, encephalomyelitis, or even death (Lomonaco et al., 2009; Lamont et al., 2011). In some cases, this bacterium's capability to infect the central nervous system (CNS) of immunocompetent adults was also revealed (Guo and Liang, 2014; Giménez-Muñoz et al., 2015). The case-fatality rate of listeriosis attains 20–30% in several regions around the world (Centers for Disease Control and Prevention, 2017; European Food Safety Authority [EFSA] and European Centre for Disease Prevention and Control [ECDC], 2017; Food Standards Australia New Zealand, 2017; Li et al., 2018; Salama et al., 2018). Although not common, listeriosis shares a large proportion of the public health and economic burden resulting from foodborne pathogens. Almost 1500 listeriosis cases were reported annually in the United States, only accounting for 0.02% among major foodborne pathogens infection cases. However, 4% of hospitalization and 19% of deaths were caused by L. monocytogenes infection (Scallan et al., 2011).

Listeria monocytogenes extensively inhabits environments including surface soil, rivers, decaying vegetation, effluents from sewage treatments, wild animal intestines and domestic animal feces, due to its robust survival ability under various unfavorable conditions such as high salinity, low pH and low temperature (Donnelly, 2001; Vermeulen et al., 2007; Soni et al., 2013; Ribeiro and Destro, 2014; Ribeiro et al., 2014; Haley et al., 2015; Wang et al., 2017b). Wide distribution of L. monocytogenes offers opportunities to contaminate diverse foods, raw material and the environments of processing plants for food production (Heisick et al., 1989; Bula et al., 1995; Zhu et al., 2012; Wang et al., 2017b; Bergholz et al., 2018). Intake of contaminated food by L. monocytogenes has been widely considered to be the entry point of infection. In China, retail foods in bulk account for a large percentage of food consumption. Foods in bulk are defined as foods that are not divided into parts and/or packaged in separate units when sold in various sales locations. Compared with prepackaged foods, foods in bulk are prone to be directly exposed to the environment during the sales process, which makes it easier for them to be contaminated by pathogens. Surveying the prevalence, phenotypic and genotypic characteristics of L. monocytogenes in food products in bulk can provide valuable information on the distribution profiles of this pathogen, allowing Chinese food safety authorities to assess the potential risk of L. monocytogenes in bulk foods, and take appropriate hygiene measures to improve microbial safety. Epidemic clones (ECs) of L. monocytogenes are defined as genetically similar isolates descended from a common ancestor, involved in either a single large outbreak or geographically and temporally unrelated outbreaks (Kathariou, 2002; Cantinelli et al., 2013). Eleven

known ECs including ECI - ECXI have been identified thus far (Chen et al., 2007, 2016; Lomonaco et al., 2013; Filipello et al., 2017). EC isolates were usually closely associated with listeriosis massive outbreaks (Haley et al., 2015). Since the first determined ECI outbreak in 1981, outbreaks of ECI were reported in western Switzerland during the time period 1983– 1987, in California, United States in 1985, and in France in 1992 (Kathariou, 2002; Cantinelli et al., 2013). ECII caused an United States multistate hotdog outbreak and another turkey deli meat outbreak (Chen et al., 2011). In 2000, ECIII caused an United States outbreak attributed to intake of turkey deli meat. ECIV caused the United Kingdom and Ireland outbreak in 1987–1988, and an outbreak in Italy in 1997. ECV caused several major outbreaks occurring in the United Kingdom, Italy, Canada, and United States (Chen et al., 2011). ECVI and ECVII were newly defined in 2011, and these two epidemic clones led to 147 cases in 28 U.S. states, causing 33 deaths and one miscarriage (Lomonaco et al., 2013). ECVIII caused an United States milk chocolate outbreak in 1994, and an outbreak in Pennsylvania in 1987 (Chen et al., 2016). ECIX caused an United States multistate caramel apple outbreak in 2014–2015 (Chen et al., 2016). ECX caused a France pork rillettes outbreak in 1999–2000 (Chen et al., 2016). In 2012, ECXI caused an United States ricotta salata outbreak (Filipello et al., 2017). Clearly, ECs concealed in food products have the potential risk to cause L. monocytogenes infection, and even listeriosis outbreaks. Initially, ECs were confirmed based on the pulsed-field gel electrophoresis (PFGE) typing, ribotyping, and multilocus enzyme electrophoresis (MEE) (Kathariou, 2002; Kathariou et al., 2006). In subsequent studies, multi-virulencelocus sequence typing (MVLST) was developed and verified to discriminate ECs accurately (Chen et al., 2007; Lomonaco et al., 2013). Additionally, in some previous studies, the PCR assay was used to rapidly screen specific ECs markers to determine epidemic clones. However, these results were presumptive and should be confirmed using MVLST (Chen and Knabel, 2007; Knabel et al., 2012; Lomonaco et al., 2012). To date, ECs among L. monocytogenes strains isolated from foods in China were screened by means of PCR, and limited from ECI to ECIII (Wu et al., 2015, 2016). Accurate and comprehensive data for ECs distribution in foodborne L. monocytogenes isolates in China were still sparse. The aims of this study were (i) to reveal the prevalence of L. monocytogenes in retail foods in bulk sampled in Zhejiang Province, and (ii) to gain information on the phenotypic and genotypic characteristics, especially the EC profiles of L. monocytogenes isolates. Zhejiang Province is located on the southeast coast of China with a population of 57 million people. In 2018, the annual GDP (Gross Domestic Product) in Zhejiang Province was 81 billion USD. According to 2017 statistical data, the production of raw meat including pork, beef, and mutton was 869.900 tons, the total fishery production was 594.45 million tons and the number of poultry raised was 25,139,100 (Department of Agriculture of Zhejiang Province, 2017). The annual per capita raw meat, poultry and seafood consumption was 27.3, 11.5, and 26.4 kg respectively in 2017 (China Statistical Yearbook Committee, 2017).

## MATERIALS AND METHODS

### Sampling and Isolation of L. monocytogenes

fmicb-10-01710 July 24, 2019 Time: 14:54 # 3

A total of 3354 retail foods in bulk including 1196 RTE foods and 2158 raw foods were sampled from supermarkets, farm fairs, fish markets, and restaurants in eleven geographic regions of Zhejiang Province, China from March 2016 to December 2016. The detailed information for food samples is shown in **Table 1**. Raw foods in this study were defined as not-ready-toeat raw foods. During the sampling process, sterile containers were used to transfer representative units or portions of foods to the laboratory by aseptic practices. Frozen or chilled foods were transported in thermal insulated containers. Frozen foods including ice cream, frozen meat, frozen poultry and frozen seafood were stored at about −18◦C and kept frozen prior to sampling. Chilled foods included chilled meat and chilled poultry stored at 0 – 5◦C, as well as chilled seafood and salmon sashimi, were stored in crushed ice prior to sampling. Other categories of foods were stored at ambient temperature prior to sampling. Sample analysis was initiated as soon as the sample was received. All RTE foods except ice cream were sampled on the day of production. Ice cream was sampled during the shelf life period. Isolation of L. monocytogenes was carried out following the recommendation of the National Food Safety Standard of China – Food microbiological examination: L. monocytogenes (GB4789.30-2010; National Standard of the People's Republic of China, 2010). Twenty-five grams of each sample was suspended in 225 mL Listeria enrichment broth LB1 (Huankai, Guangzhou, China) and homogenized for 2 min. The homogenate was incubated at 30◦C for 24 h. 0.1 mL of each culture was transferred into 10 mL of listeria enrichment broth LB2 (Huankai, Guangzhou, China) and incubated at 30◦C for 24 h, followed by spreading on CHROMagar Listeria plates (CHROMAgar, Paris, France). After incubation at 36◦C for 48 h, five suspect colonies on each plate were selected for identification. If less than five colonies grew on any one plate, all suspect colonies were selected for identification. VITEK 2 compact system (Bio Mérieux, Lyon, France) and the following PCR method (Huang et al., 2007) were performed to identify each suspect colony. PCR was applied to discriminate L. monocytogenes from other Listeria species. L. monocytogenes strains EGDe (ATCC BAA-679) and 10403S were included as positive control strains for PCR. Serotyping was conducted adopting serum agglutination according to the manufacture's instruction (Denka Seiken, Tokyo, Japan). All confirmed strains were stored at −70◦C for further analysis.

### Multi-Locus Sequence Typing (MLST) Analysis

The total DNA of L. monocytogenes isolates was extracted using Bacterial DNA Kit (Omega, United States) according to the manufacturer's protocol. DNA quality was tested by electrophoresis on 0.8% agarose gel and the concentration was determined by using Spectrophotometer NanoDrop 1000 (Thermo, United States). Prepared DNA was stored at −20◦C for MLST, MVLST and virulence genes analysis. Seven housekeeping genes abcZ, bglA, cat, dapE, dat, ldh, lhkA were amplified according to the scheme advised in Institute Pasteur MLST database<sup>1</sup> (Moura et al., 2016). Each amplicon was sequenced bidirectionally using 3730XL Genetic Analyser (Applied Biosystems, United States). Assembled nucleotide sequences were queried against Institute Pasteur MLST database. Sequence type (ST), clonal complex (CC) and lineage were then designated to each isolate. The phylogenic tree was constructed harnessing MEGA 5.1 (Tamura et al., 2011). Based on the concatenated sequence (3288 bp) of seven housekeeping genes, a neighborjoining statistical method was employed. The number of Bootstrap replications was 1,000. The Genbank database accession numbers for abcZ, bglA, cat, dapE, dat, ldh, lhkA genes sequences of L. monocytogenes isolates in this study were MK368885 – MK368957, MK368812 – MK368884, MK368958 – MK369030, MK369031 – MK369103, MK369104 – MK369176, MK369177 – MK369249 and MK369250 – MK369322, respectively.

### Multi-Virulence-Locus Sequence Typing (MVLST) Analysis

Multi-virulence-locus sequence typing analysis was carried out following the method developed by Zhang et al. (2004). Six virulence genes including prfA, inlB, inlC, dal, clpP, and lisR were amplified and then sequenced. The nucleotide sequences were trimmed to the correct length and concatenated in a specific order as mentioned in the MVLST database<sup>2</sup> . Alignment was performed between the concatenated sequence of each isolate and the reference sequence of each virulence type (VT) available in the MVLST database by using BLAST 2.2.28+. VT and EC of each isolate were determined according to the alignment results. New VTs were designated by the MVLST database. The Genbank database accession numbers for prfA, inlB, inlC, dal, clpP, and lisR gene sequences of L. monocytogenes isolates in the current study were MK392393–MK392465, MK369469–MK369541, MK369542–MK369614, MK369396–MK369468, MK369323– MK369395, and MK369615–MK369687. The polymorphism of both MLST and MVLST genes was analyzed using DnaSP 5.10 software (Librado and Rozas, 2009). Analysis parameters included number of polymorphic sites, G + C content, Ks (the number of synonymous substitutions per synonymous site), Ka (the number of non-synonymous substitutions per non-synonymous site), π (average pairwise nucleotide difference per site) and Tajima's D test for neutrality. Tajima's D is a population genetic test statistic to distinguish between neutral evolution and non-neutral evolution of DNA sequences (Tajima, 1989).

### Antibiotic Susceptibility Tests

Antibiotic susceptibility of L. monocytogenes isolates was assayed using the broth micro-dilution minimum inhibitory concentrations (MICs) method according to the Clinical

<sup>1</sup>https://bigsdb.pasteur.fr/listeria/listeria.html

<sup>2</sup>https://sites.google.com/site/mvlstdatabase/home

#### TABLE 1 | Prevalence of L. monocytogenes in different retail foods in bulk.


and Laboratory Standards Institute guidelines (Clinical and Laboratory Standards Institute, 2015). Breakpoints for Penicillin and Ampicillin were found in CLSI documents M45-A3 (Clinical and Laboratory Standards Institute, 2015). Since there is no relevant criteria for Erythromycin, Clindamycin, Quinupristin/dalfopristin, Vancomycin, Tetracycline, Gentamicin, Rifampin, Levofloxacin, Ciprofloxacin, Gatifloxacin, and Oxacillin, the susceptibility results for these antibiotics were interpreted based on the breakpoints of Staphylococcus spp (Clinical and Laboratory Standards Institute, 2017) as reported previously (Conter et al., 2009; Lungu et al., 2011; Chen et al., 2015; Khen et al., 2015; Tahoun et al., 2017). L. monocytogenes isolates resistant to three or more types of antibiotics belonging to different antibiotic classes were defined as multi-drug resistant (Magiorakos et al., 2012). Escherichia coli ATCC29522 and S. aureus ATCC29213 were used as quality control strains.

### Detection of Virulence Genes

The detailed information of primers used for virulence genes detection is shown in **Table 2**. Primers for llsA, llsG, llsH, llsB, llsY, llsD, llsP in LIPI-3 and all LIPI-4 genes were designed using PrimerPremier 5.0. The genome sequences of LIPI-3 positive L. monocytogenes strains including FSL N1- 017, FSL R2-503, H7858 and NCTC 11994 were obtained from the NCBI Refseq genome database. Multiple alignment of nucleotide sequences of genes llsA, llsG, llsH, llsB, llsY, llsD, llsP mining from the above genomes were performed using clustal v1.83. The conserved region of each gene was used as a template for the primer's design. Primers for LIPI-4 genes were designed according to the LIPI-4 sequence of L. monocytogenes strain LM09-00558 (Maury et al., 2016). PCR was conducted employing the performance system and conditions described previously (Zhang et al., 2017). The amplicons were analyzed with electrophoresis on 1% agarose gel. To validate the new primers for LIPI-3 and LIPI-4 genes, all associated positive PCR products were sequenced. The DNA sequences were analyzed using the BLAST algorithm on the website http://www.ncbi.nlm. nih.gov/BLAST.

### Statistical Analysis

Chi-square analysis or Fisher's exact test was performed to determine if a significant difference in the prevalence and serotype distribution of L. monocytogenes among different food groups could be found. The significance level was set at a p-value of <0.05. All analyses were performed using the SPSS v 21.0 software package. Association between food groups and genotyping data, antibiotic resistance profiles, and Listeria pathogenicity islands (LIPI) distribution was calculated based on the Gini coefficient (Henri et al., 2016) using the Stata software. A coefficient value of smaller than 0.4 reveals no association between genotypes, antibiotic resistance profiles or LIPI distribution and food types, a value between 0.4 and 0.6 indicates moderate association, and a value greater than 0.6 indicates unequal dispersion of genotypes, antibiotic resistance profiles or LIPI within food groups.

## RESULTS AND DISCUSSION

### Prevalence of L. monocytogenes in Retail Foods in Bulk

The prevalence of L. monocytogenes in various analyzed food samples was shown in **Table 1**. Seventy-three out of 3354 samples were confirmed to be L. monocytogenes positive with the total rate of 2.2%. Among them, 21 strains (1.8%) were isolated from 1196 ready-to-eat (RTE) foods and 52 strains (2.4%) were isolated from 2158 raw foods. No statistically significant difference in the prevalence of L. monocytogenes was present between RTE foods and raw foods (χ <sup>2</sup> = 1.545, p > 0.05). RTE foods contaminated with pathogenic microorganisms have been widely considered to be the major source of foodborne pathogen infections due to the absence of further cooking, baking or pasteurizing processes prior to consumption (European Food Safety Authority [EFSA], 2013; Pouillot et al., 2015; Luchansky et al., 2017). Here L. monocytogenes was isolated in five RTE food categories including sushi, salmon sashimi, salad, vegetables in sauce and cooked meat. The occurrence of L. monocytogenes in sushi and salmon sashimi exhibited the top two highest rates (12.9 and 6.9%) among RTE foods and significantly higher than that in all screened RTE foods (1.8%) (χ <sup>2</sup> = 27.762, p = 0; χ <sup>2</sup> = 5.082, p = 0.024) and cooked meat (1.1%) (χ <sup>2</sup> = 14.076, p = 0; χ <sup>2</sup> = 4.123, p = 0.042). Relatively high presence rates of L. monocytogenes in sushi and salmon sashimi poses a potential risk of causing human L. monocytogenes infection. RTE meat products are considered one of the important food sources of human L. monocytogenes infection throughout the world (European Food Safety Authority [EFSA], 2013; Iannetti et al., 2016; Raheem, 2016). According to our data, the prevalence rate of L. monocytogenes in cooked meat was close to that in prepackaged heated meat at the end of its shelf-life in the European Union as well as meat products in Italy (2.07 and 1.66% respectively) (European Food Safety Authority [EFSA], 2013; Iannetti et al., 2016).

Listeria monocytogenes was distributed unequally in raw foods. Presence rates in raw meat (3.5%) and raw poultry (3.8%) were significantly higher than that in raw seafood (1.3%) (χ <sup>2</sup> = 5.663, p = 0.017; χ <sup>2</sup> = 6.415, p = 0.011). It is worthy to note that L. monocytogenes was not detected in 396 samples of raw freshwater food. We further analyzed if rates of L. monocytogenes in raw foods under different storage conditions were different. The results did not show a significant difference among fresh, chilled and frozen foods. Due to the common eating habit in China of cooking foods completely, raw foods were not predicted to have a high risk of causing L. monocytogenes infection in China. However, transfer of L. monocytogenes might occur from contaminated raw foods to RTE foods during the process of storage after purchase. In particular, cross contamination may be more likely to occur due to lack of tight packaging for foods in bulk.

### Serotyping

Four different serotypes, 1/2a, 1/2b, 1/2c, and 4b were identified in the 73 L. monocytogenes isolates (**Figure 1**). 1/2a was the most

#### TABLE 2 | Primers used for L. monocytogenes virulence genes.


prevalent serotype, accounting for 47.9% of the identified isolates. Serotype 1/2a strains have exhibited extensive distribution in various foods around the world, which might be due to superior adaptability in different environments compared to other serotype strains (Korsak et al., 2012; Schmitz-Esser et al., 2015; Wu et al., 2015). Serotypes 1/2b and 1/2c had similar presence rates at 21.9 and 23.3%, respectively. Among the 73 isolated strains, only five (6.8%) were confirmed as serotype 4b. Serotype distribution difference among different food groups was analyzed based on Chi-square or Fisher analysis, as appropriate. The prevalence of serotype 1/2a in raw poultry was significantly higher than that in raw seafood (χ <sup>2</sup> = 8.478, p = 0.004) and RTE food (χ <sup>2</sup> = 6.025, p = 0.014). Meanwhile, there was no obvious distribution difference of serotypes 1/2b, 1/2c, and 4b among different food groups (p > 0.05). The current study showed that a total of 52 (71.2%) isolates were grouped into 1/2a, 1/2b, and 4b serotypes, which were the main causative agents of clinical cases worldwide and were responsible for 98% of listeriosis outbreaks (Buchrieser et al., 1993; Raheem, 2016). Among all the confirmed serotypes, no serotypes 3a, 3b, 3c, 4a, 4c, 4d, 4e, or 7 were detected, which was consistent with a previous report (Wu et al., 2015) in China indicating that those serotypes were rarely identified from food and clinical samples.

### MLST Analysis

Eighteen different sequence types (STs) were classified among all 73 L. monocytogenes isolates, which were further assigned to sixteen clonal complexes (CCs) and one singleton by using MLST analysis, with the Simpson's index of diversity (DI) at 0.892 (**Table 3**). Among the 18 STs, ST9 (23.3%) was predominant, followed by ST155 (16.4%), ST8 (12.3%), and ST121 (9.6%). Many studies have shown that hypervirulence exists in L. monocytogenes with different STs. Although ST9, ST155, ST8, and ST121 were all revealed to be involved in L. monocytogenes infection, they tended to be isolated from food products and food processing environments. Only a small number of clinical cases were induced by these ST types (Ebner et al., 2015; Schmitz-Esser et al., 2015; Wu et al., 2016; Maury et al., 2016; Wang et al., 2018). For instance, while ST9 and ST121 were two major STs among L. monocytogenes foodborne isolates during 1999–2014 in France, ST121 included the fewest clinical isolates (Henri et al., 2016). ST1 and ST2 clones have been verified to be strongly associated with clinical origin particularly in human CNS or maternal-neonatal (MN) listeriosis cases (Maury et al., 2016). According to our data, one isolate from fresh pork and three isolates from sushi, chilled beef and chilled seafood were genotyped into ST1 and ST2, respectively. Although the presence rates (1.4 and 4.1%) were relatively low, the existence of ST1 and ST2 in foods, particularly in RTE foods, poses a potential risk on food safety due to their high virulence. ST87 clones were detected in four samples including two RTE foods. ST87 strains were scarcely isolated from food, environmental or clinical samples in North America, South America, Europe, and Australia (Maury et al., 2016). In contrast, ST87 was the predominant ST in clinical L. monocytogenes isolates and closely related to CNS and MN infection in China (Wang et al., 2015, 2018; Wu et al., 2016).

### MVLST and EC Analysis

A total of eighteen VTs including four newly assigned VTs (VT180, VT181, VT182, and VT183) were determined in all L. monocytogenes isolates (**Figure 2**). Among 73 strains, 23 (31.5%) of them were epidemic clones including ECI, ECIV, ECV, ECVI, ECVIII, and ECXI, which is different from previous reports indicating that ECI and ECIII are the only detected ECs in China (Wu et al., 2015, 2016). ECV was most common; 10 out of 23 (43.5%) EC isolates were classified into this group.

To explore the polymorphisms of MLST and MVLST genes in L. monocytogenes population investigated in this study, the number of polymorphic sites, Ka/Ks ratio, π and Tajima's D of associated genes in 73 L. monocytogenes isolates were calculated (**Table 4**). Seven MLST genes contained a total of 151 polymorphic sites (4.6%, range 4.2–12.1% per gene). Six MVLST virulence genes contained a total of 150 polymorphic sites (5.7%, range 1.1–14.2% per gene). Dat and dal had the two highest percentages of polymorphic sites, respectively. Ka/Ks of all MLST genes as well as MVLST genes dal, inlB, inlC, prfA were less than one, indicating these genes evolved under purifying (negative) selection. Meanwhile, Ka/Ks of clpP was 3.551, which indicated clpP was under positive selection during the course of genetic evolution (Zhang and Yu, 2006). Through comparison analysis of Ka/Ks between our data and previous reports (Cantinelli et al., 2013; Wu et al., 2016), MLST genes in different L. monocytogenes populations demonstrated similar selective strength. However, the selective strength for MVLST genes presented to be variable in L. monocytogenes populations from different ecological niches. For instance, inlB in L. monocytogenes outbreak strains in France was under neutral selection. Meanwhile, clpP was under purifying (negative) selection (Cantinelli et al., 2013). The average nucleotide diversity was close between MLST genes and MVLST genes, with π equaling 1.8% (range 1.0–5.6% per gene) and 2.1% (range 0.3–6.1% per gene) respectively. Tajima's D test illustrated that abcZ, bglA, ldh, lhkA, inlB, inlC, lisR, and prfA


TABLE 3 | MLST characteristics of L. monocytogenes isolates.

<sup>∗</sup>Number in parentheses indicates the number of strains isolated from the given food category.

evolved neutrally, whereas cat, dapE, dat, clpP, dal evolved under balancing selection.

### Antibiotic Susceptibility of L. monocytogenes Isolates

The susceptibilities of L. monocytogenes isolates to thirteen antibiotics were shown in **Table 5**. All 73 strains were susceptible to penicillin, ampicillin, gentamicin and rifampin. More than 95% of strains were susceptible to erythromycin (97.3%), quinupristin/dalfopristin (98.6%), vancomycin (98.6%), levofloxacin (98.6%) and ciprofloxacin (95.9%). The results were similar to previous reports from China, the United States, Ireland, and Poland (Wieczorek et al., 2012; Shen et al., 2013; Chen et al., 2015; Khen et al., 2015; Wieczorek and Osek, 2017) and consistent with the general thought that the Listeria genus is naturally susceptible to ampicillin, penicillin, gentamicin and erythromycin, which are usually active against Grampositive bacteria (Wang et al., 2013; Chen et al., 2015; Wu et al., 2015, 2016). On the other hand, L. monocytogenes strains isolated from raw milk, milk equipment and farm workers in Egypt showed lower susceptibility rates to gentamicin (19.0%), rifampin (0%) and ciprofloxacin (42.9%) (Tahoun et al., 2017). The most prevalent detected antibiotic resistance type was resistance to oxacillin. 86.3% of L. monocytogenes isolates were resistant to this antibiotic. Compared with previously reported resistance rates (1.0–4.5%) to tetracycline of L. monocytogenes isolated from various types of raw foods in China and beef chain in Ireland (Wieczorek et al., 2012; Shen et al., 2013; Wang et al., 2013; Chen et al., 2015; Khen et al., 2015), a much higher resistance rate of 11.0% to tetracycline was observed among our isolates. In the reference listeriosis therapy scheme, ampicillin or penicillin G combined with gentamicin was recommended as the treatment of choice. Meanwhile, vancomycin, trimethoprim-sulfamethoxazole and erythromycin were usually used as alternatives, especially for pregnant women (Hof, 2004). Both our results and previous reports (Rodas-Suarez et al., 2006; Morvan et al., 2010; Lungu et al., 2011; Wang et al., 2013; Chen et al., 2015; Tahoun et al., 2017) showed relatively low resistance rates to ampicillin, penicillin, gentamicin, vancomycin, trimethoprim-sulfamethoxazole and erythromycin, which revealed that the antibiotic treatment might be efficient for most of the L. monocytogenes strains. Including three (4.1%) multidrug-resistant strains, resistant to Erythromycin/Clindamycin/Quinupristin/dalfopristin/ Vancomycin/Tetracycline, Clindamycin/Tetracycline/Oxacillin and Erythromycin/Tetracycline/Oxacillin, respectively, all 73 isolates were grouped into seven antibiotic resistance patterns. It is worthy to note that one strain was identified as resistant to six types of antibiotics (**Figure 2**).

### Virulence Genes Profile of L. monocytogenes Isolates

The molecular determinants of L. monocytogenes virulence has been investigated for many years (Tilney and Portnoy, 1989;


quinupristin/dalfopristin; VAN, vancomycin.

Cossart, 2011). Four Listeria pathogenicity islands (LIPI) have been verified thus far, which are involved in invasion, survival and colonization of L. monocytogenes in host tissues. LIPI-1 contains six genes including hly, prfA, plcA, plcB, mpl, and actA (Hamon and Cossart, 2011; Travier and Lecuit, 2014; Mitchell et al., 2015; Rupp et al., 2015; Wang et al., 2017a; Hadjilouka et al., 2018; Poimenidou et al., 2018). LIPI-2 encodes a series of internalin family proteins which interact with the molecular cell surface and are essential for host cell adherence and virulence (McGann et al., 2007). LIPI-3 contributes to the expression of listeriolysin S (LLS), which is a post-translationally modified hemolytic peptide acting as a bacteriocin to alter the host intestinal microbiota, and plays an important role in the survival of L. monocytogenes in polymorphonucleocytes (PMNs) and virulence in the murine model (Cotter et al., 2008; Quereda et al., 2016). LIPI-4 encodes a cellobiose-family phosphotransfer system (PTS) and is involved in neural and placental infection (Maury et al., 2016). The distribution of virulence genes of these LIPIs in L. monocytogenes isolates were tested in this study (**Figure 2**). LIPI-1 genes were detected in approximately all L. monocytogenes isolates except one ST101 strain isolated from sushi, in which mpl was absent. inlA, inlB, inlC and inlJ of LIPI-2 existed in all 73 strains. 11.0% of isolates harbored all LIPI-3 genes and four isolates (5.5%) were determined LIPI-4 genes positive.

Remarkably, the distribution of both LIPI-3 and LIPI-4 exhibited apparent association with L. monocytogenes lineage and ST. LIPI-3 positive strains belonged to ST1, ST3, ST224, ST330, and ST1047; these STs were grouped into lineage I, in accordance with the finding that LIPI-3 was identified exclusively in a subset of lineage I (Clayton et al., 2014; Quereda et al., 2016). LIPI-4 was verified to exist uniquely in CC4 L. monocytogenes and closely linked to high virulence in CNS and MN listeriosis



Ks, the number of synonymous substitutions per synonymous site; Ka, the number of non-synonymous substitutions per non-synonymous site. π, average pairwise nucleotide difference per site. <sup>∗</sup> : Statistic difference of p < 0.05; ∗∗: Statistic difference of p < 0.01.

TABLE 5 | Antibiotic susceptibility of L. monocytogenes isolates.


<sup>a</sup>Breakpoints for Listeria monocytogenes. <sup>b</sup>Breakpoints for Staphylococcus spp.

(Maury et al., 2016). A subsequent study pointed out that ST619, CC87 strains carried LIPI-4 fragment ptsA (Wang et al., 2018). In this study, we detected six genes encoding PTS sugar transporter subunit EIIC (licC), EIIB (licB), EIIA (licA), PTS systems associated protein (gene locus in Genbank, lm900558- 70012), transcriptional antiterminator (gene locus in Genbank, lm900558-70013) and maltose-6<sup>0</sup> -phosphate-glucosidase (glvA) (Maury et al., 2016) of LIPI-4 in all ST87 isolates, which confirmed that this L. monocytogenes group carried all LIPI -4 genes. Meanwhile, the test of virulence genes was based on the PCR method employing target gene specific primers. There are some disadvantages to this method, including (i) polymorphism might exist in the primer annealing regions in the genomes of certain L. monocytogenes strains, leading to invalid or inefficient binding to primers and then false negative results, (ii) specificity of primers is not sufficient to avoid amplification of non-target regions. In recent years, whole genome sequencing (WGS) was utilized in many studies on L. monocytogenes, including the determination of virulence genes profiles (Maury et al., 2016; Fox et al., 2017; Rychli et al., 2017; Pightling et al., 2018), which can avoid false negative or positive results significantly. Genetic diversity of virulence genes can be further analyzed based on the nucleotide sequence.

Additionally, Gini coefficient analysis did not find an association between STs, ECs, LIPI-3 and LIPI-4 distribution of L. monocytogenes isolates and food groups. Values of the

coefficient for eighteen STs ranged from 0.0 to 0.38. Seven ECs ranged from 0.0 to 0.25: 0.25 for LIPI-3 and 0.17 for LIPI-4, which indicated that STs, ECs, LIPI-3, and LIPI-4 were distributed uniformly within the four food groups: RTE food, raw meat, raw poultry and raw seafood. Meanwhile, moderate association was demonstrated between antibiotic resistance to oxacillin and raw poultry (coefficient value of 0.46), indicating that L. monocytogenes isolated from RTE food and raw poultry tend to be resistant to oxacillin.

### CONCLUSION

In summary, a comprehensive study of prevalence and characteristics of L. monocytogenes isolated from retail foods in bulk in Zhejiang Province, China was performed. Both RTE foods and raw foods were included, showing a wide range of food categories. The potential risk of causing human L. monocytogenes infection by certain foods with relatively high contamination rates, including sushi and salmon sashimi, should arouse public concern. Distribution differences of serotype 1/2a among different food groups revealed that this serotype of L. monocytogenes might have specific ecological niches. To the best of our knowledge, this is the first time the distribution of ECs (ECI-ECXI) has been investigated in foods sampled in China. Furthermore, the discovery of multidrug-resistant strains and the particularly high resistance rate (11.0%) to tetracycline among L. monocytogenes isolates indicates a potential public health problem. According to our relatedness analysis, ECs and LIPI-3 or LIPI-4 positive isolates were distributed equally among various food groups. The present study provides initial data for Chinese food safety authorities to address the issue of microbial safety of retail bulk foods in China. One recommendation is to strengthen the monitoring of retail foods in bulk with relatively

### REFERENCES


high detection rates of L. monocytogenes, including sushi and salmon sashimi. In addition, the public must recognize the potential risks of certain foods with high contamination rate of L. monocytogenes, and a national standard should be developed for the detection limit of L. monocytogenes for bulk foods with high risk of causing L. monocytogenes infection.

### AUTHOR CONTRIBUTIONS

LZ, LM, and YZ designed the experiments. YZ, LZ, HC, JC, JZ, ZZ, YY, and ZX carried out the experiments. YZ and SD analyzed the experimental results and wrote the manuscript.

### FUNDING

This work was supported by the National Key Research and Development Program of China (2017YFC1601503), the National Research Project of the 13th Five-Year Plan (2018ZX10714002), and the Medical Scientific Research Foundation of Zhejiang Province (2018KY034 and 2016KYB065).

### ACKNOWLEDGMENTS

The authors would like to thank Dr. Hongliang Yang of Houston Methodist Research Institute and Kaity Mussio of University of California, Berkeley, for their help to revise and polish the language throughout the manuscript, and Drs. Mingbin Liang and Rong Zhang at Zhejiang Provincial Center for Disease Control and Prevention for their contribution on the statistical analysis.

identification of globally distributed clonal groups and differentiation of outbreak strains of Listeria monocytogenes. Appl. Environ. Microbiol. 82, 6258– 6272. doi: 10.1128/AEM.01532-16



genes suggests functional diversity of these proteins among the listeriae. Appl. Environ. Microbiol. 73, 2806–2814. doi: 10.1128/AEM.02923-06



**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 Zhang, Dong, Chen, Chen, Zhang, Zhang, Yang, Xu, Zhan and Mei. 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.

# Epidemiological and Molecular Investigations on Salmonella Responsible for Gastrointestinal Infections in the Southwest of Shanghai From 1998 to 2017

#### Edited by:

Learn-Han Lee, Monash University Malaysia, Malaysia

#### Reviewed by:

Jianmin Zhang, South China Agricultural University, China Iddya Karunasagar, Nitte University, India Chunlei Shi, Shanghai Jiao Tong University, China

#### \*Correspondence:

Shurui Bu 18930819609@163.com; Qixl051816@163.com Dongfang Lin lindongfang@fudan.edu.cn

#### Specialty section:

This article was submitted to Food Microbiology, a section of the journal Frontiers in Microbiology

Received: 07 August 2018 Accepted: 19 August 2019 Published: 18 September 2019

#### Citation:

Qi X, Li P, Xu X, Yuan Y, Bu S and Lin D (2019) Epidemiological and Molecular Investigations on Salmonella Responsible for Gastrointestinal Infections in the Southwest of Shanghai From 1998 to 2017. Front. Microbiol. 10:2025. doi: 10.3389/fmicb.2019.02025 Xulin Qi<sup>1</sup> , Pei Li<sup>2</sup> , Xiaogang Xu<sup>2</sup> , Yiqun Yuan<sup>1</sup> , Shurui Bu<sup>1</sup> \* and Dongfang Lin<sup>2</sup> \*

<sup>1</sup> Department of Infection, Jinshan Hospital Affiliated to Fudan University, Shanghai, China, <sup>2</sup> Institute of Antibiotics, Huashan Hospital Affiliated to Fudan University, Shanghai, China

Purpose: To investigate the characteristics of gastrointestinal infections in Southwest Shanghai.

Methods: Clinical and epidemiological characteristics of patients with Salmonella infections between 1998 and 2017 admitted to the Jinshan Hospital in the Southwest of Shanghai were retrospectively analyzed. A total of 565 isolated Salmonella strains were classified by serotyping and pulsed field gel electrophoresis (PFGE).

Results: From 1998 to 2006, diarrhea was mainly caused by Vibrio parahaemolyticus followed by Shigella and Salmonella. From 2007 to 2010, Vibrio parahaemolyticus infection was the major cause of diarrhea followed by Salmonella and Shigella. From 2011 to 2017, Salmonella infections became the main cause of diarrhea after Vibrio parahaemolyticus. Salmonella infections increased from 2006 on and peaked between May and October, accounting for 82.48% of yearly infections. Patients with Salmonella infections (90.5%) had a history of eating unclean food, abdominal pain (58.05%), diarrhea ≥5 times a day (50.44%), moderate fever (24.96%) and increased fecal leukocytes (41.42%). From 1998 to 2017, infected specimens from clinical cases were dominated by Salmonella enterica serovar Typhimurium (S. Typhimurium) (21.59%) followed by Salmonella enterica serovar Enteritis (S. Enteritidis) (16.81%), Salmonella enterica serotype London (6.55%) and Salmonella group B (13.10%). Other species included Salmonella enterica serovar Thompson, Salmonella enterica serovar Saintpaul, Salmonella group D, Salmonella group C, Salmonella enterica serovar Choleraesuis and Salmonella enterica serovar Aberdeen. The PFGE classification of Salmonella serovars in 2008–2017 demonstrated that S. Enteritidis had 9 PFGE banding patterns and S. Typhimurium 16 with varying degrees of similarity among S. Enteritidis and S. Typhimurium. The results of antibiotic susceptibility tests for the 330 Salmonella strains revealed that fosfomycin had the highest sensitivity rate (97.5%) followed by levofloxacin

**431**

and ceftriaxone (81%), and ampicillin/sulbactam (78.2%). The resistance to piperacillin and ciprofloxacin was 60.9 and 50.61%, respectively.

Conclusion: The features of onset, epidemiological characteristics and molecular subtyping of Salmonella were conducive to clinical diagnosis, rational use of antibiotics and improved therapeutic efficacy.

Keywords: diarrhea, Salmonella strains, food poisoning, epidemiology, PFGE, Chinese patients

### INTRODUCTION

fmicb-10-02025 September 13, 2019 Time: 16:58 # 2

Infectious diarrhea is a common and frequently occurring disease worldwide. The WHO reported that the incidence rate is ranked second compared to all infectious diseases (Lamberti et al., 2014). The pathogens that cause infectious diarrhea are widespread and include viruses, bacteria, fungi and parasites. The bacterial pathogens commonly involved include Shigella, Salmonella, enteropathogenic Escherichia coli (EPEC), Vibrio parahaemolyticus and other organisms (Evyline Isingoma et al., 2018).

Salmonella is an important pathogen that poses a significant threat to human health, with more than 2,600 species of serotypes discovered to date (Bugarel et al., 2018). Salmonella infection is an infectious intestinal disease causing symptoms that include enteric typhoid-like and paratyphoid-like fever, gastroenteritis and various forms of extra-intestinal inflammation such as bacteremia, cholecystitis, and pyelonephritis (Brenner et al., 2000; Fàbrega and Vila, 2013). Moreover, the virulence of different Salmonella types can be quite different. For example, Salmonella enterica serovar Anatis presents as an asymptomatic phenotype, while septicemia caused by Salmonella enterica serovar Choleraesuis may lead to death (Guo et al., 2018). S. Typhimurium produces the symptoms of dysentery and S. Enteritidis often causes gastroenteritis (Kurtz et al., 2017). Therefore, Salmonella infections produce digestive symptoms among other general ailments including fever. Furthermore, there are many strain types, which makes it difficult to diagnose rapidly and accurately the condition (Bugarel et al., 2018). In a previous study conducted in 28 sentinel hospitals located in five geographic regions of the Henan Province six different main serotypes were detected (Xia et al., 2009). Similar results from studies in Africa reported infections in patients with comorbidities affecting the immune system, the predominance of S. Typhimurium and other Salmonella serovars, and the presence of drug-resistance in isolates (Albert et al., 2019).

Therefore, in order to establish unequivocally the epidemiological characteristics and clinical features of Salmonella in Southwest of Shanghai, and to provide assistance with judging the condition, diagnosis and treatment options, 20 years worth of cases of Salmonella intestinal infections in our hospital were collected to analyze their clinical features and epidemiological characteristics.

Pulsed field gel electrophoresis (PFGE) can be used to isolate long linear DNA sequences, with banding patterns appearing on the agarose gel based on the different lengths of chromosome segments. PFGE has the highest discriminatory power among the different molecular typing methods for the investigation of clonal relationships between bacteria (Ersoy Omeroglu, 2015). Therefore, we used PFGE for homogenous bacterial typing, which directly or indirectly reflected the variation and differentiation of the pathogens studied. In particular, PFGE was used to analyze changes in Salmonella intestinal infections, using fecal routine examinations and feces culture investigations. The cases of positive Salmonella infections in Shanghai were analyzed and classified.

### PATIENTS AND METHODS

### Materials and Methods Inclusion Criteria

The study was a retrospective study, but was conducted in accordance with good clinical practice, the Principles of the Declaration of Helsinki and the requirements of the local hospital (Jinshan Hospital) ethics committee, which waived informed consent. Between 1998 and 2017, patients who visited the Enteric Disease Clinic of Jinshan Hospital with two or more of the following symptoms were selected for inclusion in the study. The inclusion criteria were: diarrhea 3 or more times in 24 h with abnormal feces; fever with a temperature ≥38◦C accompanied by headache, chills and general fatigue; vomiting, diarrhea, abdominal pain, and abnormal feces such as diluted feces, water-like feces, mucus-like feces and bloody feces. Patients who exhibited at least two of the symptoms described above were included.

### **Inclusion**

This study analyzed all demographic data and the clinical performance records of all enrolled patients. Pathogenic Salmonella-positive cases and laboratory data were also analyzed.

### **Exclusion criteria**

From 1998 to 2017, the same strain from the same location in the same patient was excluded from all types of bacterial culture specimens examined.

### **Fecal sampling method**

Patients were first informed of the purpose of fecal culture and then advised to defecate in a clean toilet bowl after emptying their bladders. A sterile bamboo stick was used to obtain a small amount of excrement from the central part or from purulent blood or mucus, which was then placed in a sterile culture bottle. The specimen was then sent for detailed laboratory analysis. The collection of all stool specimens was performed in accordance with the Institutional Review Board guidelines of Jinshan Hospital.

## Isolation, Culture and Identification of Bacteria

### Bacterial Isolation

fmicb-10-02025 September 13, 2019 Time: 16:58 # 3

Freshly collected fecal samples were preserved in Cary-Blair transport medium and inoculated on SS (for selective separation of Shigella and Salmonella) and TCBS plates (thiosulfate citrate bile salts-sucrose agar culture medium) for selective separation of Vibrio parahaemolyticus). After 6–10 h of culture at 35◦C, suspicious colonies were selected for preliminary biochemical identification using the serum agglutination method and then further cultured to obtain a pure strain.

### Bacteriological Examination

An automatic bacteria identification instrument (VITEK 2 Compact, Biomérieux, France) was used to identify pathogenic bacteria from pure cultured strains.

### Antimicrobial Susceptibility Test Method

Minimum inhibitory concentrations (MICs) were determined using custom dehydrated MicroScan broth microdilution (Siemens Medical Solutions Diagnostics), following Clinical and Laboratory Standards Institute (CLSI) guidelines (Cockerill, 2012) and relative susceptibility interpretations were based on CLSI clinical breakpoints (CLSI, 2017).

Thirteen antimicrobial agents commonly used to treat diarrhea were tested. Results were only included in the analysis when corresponding quality control isolate test results were in accordance with CLSI, 2017 guidelines, which were unchanged between 2014 and 2017 and therefore within an acceptable range.

### PFGE Subtype Method

Salmonella strains were classified according to a previous publication (Ribot et al., 2002). A standard Salmonella strain H9812 was used as the molecular weight control marker. The prepared DNA plugs with Salmonella strains were digested with XbaI for 3 h at 37◦C. The electrophoresis conditions were: voltage 6 V/cm; pulse duration 2.2 s to 63.8 s; linear transformation with a convert angle of 120◦ ; electrophoresis period and temperature 19 h at 14◦C, respectively, which was a modified method from the report of Laconcha et al. (2000). In a PFGE electrophoretogram, each lane represents one isolated strain. With the effect of endonuclease, different strains present as different numbers and band sizes. After calibration by a unified molecular weight standard based on the H9812 strain, the strains which showed a high degree of resemblance over 60% were classified as identical PFGE types.

### Statistical Methods

Data was analyzed by using SPSS Statistics for Windows (version 10.0, SPSS Inc., United States). A t-test was used for continuous quantitative data that was normally distributed and a chisquared test to compare qualitative data. Results of PFGE were processed through BioNumerics software version 3.3. Clustering analysis was based on the unweighted pair group method using arithmetic averages (UPGMA). The dice coefficient, based on bands comparison, was used to determine the similarity of PFGE bands. Those in which the homology was over 60% were considered to be the same PFGE type.

### RESULTS

### Patients' Basic Information

In total, 32,544 patients with acute diarrhea as the first symptom, who were admitted to our enteric disease clinic located in the Southwest of Shanghai between 1998 and 2017, were retrospectively analyzed. Most of the patients came from Shanghai, but some were from the neighbor city Jiaxing. A total of 29,589 case specimens were sent for laboratory examination, which represented an inspection rate of circa 90%. A total of 2,849 pathogen-positive cases were collected and the bacterial isolation rate was 9.63%.

From 1998 to 2010, Vibrio parahaemolyticus infection (59.49– 99.5%) was the major cause of acute diarrhea in Southwest Shanghai, followed by Salmonella (0.00–37.97%) and to a lesser extent Shigella infection (0.0–11.8%). However, significant changes occurred between 2011 and 2017. The Salmonella infection rate became the number one cause of diarrhea (53.13–85.00%) followed by Vibrio parahaemolyticus (15.00– 46.88%) with almost no infections due to Shigella; only 13 strains of Shigella were found in 2011. In recent years, the age range of patients with Salmonella infection as the major infective pathogen was 6 months to 83 years (average 35.55 ± 12.82 years). The isolation rate of positive patients with pathogenic Salmonella infection between 1998 and 2017 was 0∼85.00%. From 1998 to 2006 <10 strains were isolated each year, with the total incidence being 0–0.90%. From 2007 to 2015, isolated strains began to increase (18–58 strains) with the total incidence rate reaching 23.08% to 83.87%. Between 2016 and 2017, 102 and 120 strains were isolated with an 85.0 and 75.47% incidence rate, respectively. It is worth noting that the incidence rate of Shigella infection in Southwest Shanghai from 2012 was very low and practically zero, while Salmonella infections became the dominant infectious pathogens (**Supplementary Table S1**).

### Correlation Between Infections and Time

**Figure 1** shows a histogram of the pathogens that caused intestinal infections between 1998 and 2017.

In terms of years, Salmonella infections began to increase from 2006 and were mainly centralized in the third quarter, and then in the second and fourth quarters of each year. From the monthly distribution, May to October was the peak period of Salmonella infections, accounting for 82.48% but it should be noted that infections still occurred throughout the year (**Figure 2**).

### Clinical Characteristics of Salmonella Infection

Among 565 patients with Salmonella infection, 58.05% had abdominal pain, 50.44% diarrhea >5 times daily and 24.96%

moderate fever with an average oral temperature of 38.7 ± 0.7◦C. Patients with upper gastrointestinal symptoms accompanied by nausea and emesis accounted for 30.97 and 18.05% of cases, respectively. A total of 41.42% of patients had increasing fecal leukocyte counts and 90.5% mentioned they had consumed unclean food. Normally after having eaten unclean food, the minimum time to onset of symptoms was 1 h and the maximum time 72 h (**Table 1** and **Supplementary File S1**).

### Results of Serotyping

fmicb-10-02025 September 13, 2019 Time: 16:58 # 5

From 1998 to 2017, infected specimens from clinical cases in Southwest Shanghai were dominated by S. Typhimurium (21.59%), followed by S. Enteritidis (16.81%), Salmonella enterica serotype London (6.55%) and Salmonella group B (13.10%) (**Figure 3**). Other species included Salmonella enterica serovar Thompson, Salmonella enterica serovar Saintpaul, Salmonella group D, Salmonella group C, Salmonella enterica serovar Choleraesuis and Salmonella enterica serovar Aberdeen.

### The Susceptibility of Common Antimicrobial Agents to Salmonella

Commonly used antibiotics susceptibility tests for the 330 Salmonella strains collected between 2014 and 2017 revealed that fosfomycin had the highest sensitivity rate, reaching 97.5%, followed by levofloxacin and ceftriaxone (81%) and ampicillin/sulbactam (78.2%) (**Figure 4**). The sensitivity rates of piperacillin and ciprofloxacin were 39.1 and 49.39%, decreasing by 30.26 and 3.96%, respectively, results comparable to previous studies (Demirtas et al., 2012; Shubert et al., 2014; **Figure 4**).

### PFGE Subtypes of Salmonella S. Enteritidis and Salmonella S. Typhimurium From 2014 to 2017

A total of 33 S. Enteritidis strains and 60 S. Typhimurium strains were collected from 2014–2017 and digested with XbaI enzyme and analyzed using PFGE. The results showed that the DNA bands had a good separation. The size of the bands ranged from 20.5 to 1,135 kb and the number of bands from 12 to

TABLE 1 | Clinical characteristics of Salmonella infections in the southwest of Shanghai in 1998 to 2017.


16. Cluster analysis, using the BioNumerics module, classified S. Enteritidis into 9 PFGE types (named A'–J') and S. Typhimurium into 16 types (named A–P) based on a 60% similarity. These results suggested a polymorphism distribution of PFGE types of S. Enteritidis and S. Typhimurium in the Southwest of Shanghai (**Figures 5**, **6**).

### Annual Similarity of Salmonella Enteritidis From 2014 to 2017

The homologous proportion of 12 strains in 2014 was 41.7, 16.7, 16.7, 8.3, 8.3, and 8.3% (F' type 5/12, D' type 2/12, E' type 2/12, A' type 1/12, H' type 1/12, J' type 1/12). The homologous proportion of 7 strains in 2015 was 42.8, 14.3, 28.6, and 14.3% (A' type 3/7, B' type 1/7, C' type 2/7, G' type 1/7). The homologous proportion of 8 strains in 2016 was 75, 12.5, and 12.5% (A' type 6/8, I' type 1/8, J' type 1/8). The homologous proportion in 2017 was 100% (D' type 6/6).

### The Annual Similarity of S. Typhimurium From 2014 to 2017

The homologous proportion of 12 strains in 2014 was 66.7, 8.3, 8.3, 8.3, and 8.3% (N type 8/12, F type 1/12, J type 1/12, K type 1/12, M type 1/12) and the proportion of 9 strains in 2015 was 44.4, 33.3, and 22.2% (O type 4/9, D type 3/9, L type 2/9). The homologous proportion of 9 strains in 2016 was 75, 12.5, and 12.5% (A type 6/8, C type 1/8, F type 1/8) and the proportion for 31 strains in 2017 29.0, 19.3, 12.9, 12.9, 3.2, 3.2, 3.2, 3.2, 3.2, and 3.2% (C type 9/31, E type 6/31, A type 4/31, K type 3/31, P type 3/31, B type 1/31, G type 1/31, H type 1/31, I type 1/31, J type 1/31, N type 1/31).

**Figures 5**, **6** show varying degrees of similarity among S. Enteritidis and S. Typhimurium strains collected each year. In addition, the figures show that in the same year, there was an epidemic of the same strain in Southwest Shanghai. This phenomenon can probably be attributed to the rich seafood resources in Southwest Shanghai, which is located close to the East China Sea. In terms of the inspection time, most samples were sent for laboratory analysis between May and September, which implies that the prevalence of a particular strain may be related to local seasonal eating habits.

### DISCUSSION

The strain types responsible for bacterial diarrhea have been shown to change over time and according to the district analyzed (Xu and Zhang, 2012; Lamberti et al., 2014). The present study clearly demonstrated that Vibrio parahaemolyticus was the dominant infectious strain in the Southwest of Shanghai between 1998 and 2007. After 2007, there was a remarkable increase in Salmonella as the dominant infectious strain. Our findings suggest that infectious factors for diarrhea have undergone great changes in the Shanghai region over the last decade. Due to improvements in the sanitary management of water resources in 2007, especially in rural areas of Shanghai, the rate of water infection has clearly decreased, with dietary infection becoming the main cause of infectious diarrhea (Shen, 2008).

From 2006, Shanghai residents have participated in the global testing program for non-typhoid Salmonella led by the WHO (Zhang et al., 2014). S. Typhimurium was found to be the dominant strain responsible for non-typhoid Salmonella infectious diarrhea in Shanghai (Zhang et al., 2014). In our study, S. Typhimurium accounted for 21.59% of the total Salmonella infections from 1998 to 2017, followed by S. Enteritidis (16.81%). PFGE analysis was used to characterize 330 Salmonella strains from 421 strains collected between 2014 and 2017. The results demonstrated that 33 S. Enteritidis strains could be classified into 9 PFGE types, and 60 S. Typhimurium organisms into 16 types. The results showed that S. Typhimurium was the dominant group in recent years in Shanghai, followed by S. Enteritidis, which presented with a polymorphism distribution. In addition, Salmonella enterica serovar Typhi was often found in blood culture specimens, while S. Enteritidis was mostly isolated from feces samples, and S. Typhimurium was widely present in urine (Kwambana-Adams et al., 2015). Our study used feces from diarrhea patients as the main specimen.

Annual and monthly distributions of Salmonella infections (**Figure 2**) revealed that they occurred throughout the year, frequently between April and November, but were much more prevalent between May and September, findings in agreement with a previous survey and related published papers (Qi et al., 2008; Lin et al., 2010; Cao et al., 2012). However, the number of patients and the bacterial isolation rate during the last 2 years has decreased according to the annual infectious diarrhea survey. This decrease is probably related to the strengthened supervision of food hygiene in ChinaDaily.com. (2017). However, it is noteworthy that the detected number of Salmonella infections during the last 4 years has significantly increased, which might be related to an alteration in the diet of individuals (Matsuoka et al., 2004).

Major suspicious diet history for Salmonella infection in this study included ice milk, ice watermelon, seafood and meat. It has also been reported (Centers for Disease Control and Prevention [CDC], 2008; Raguenaud et al., 2012; Smith et al., 2012; Zielicka-Hardy et al., 2012) that the main cause of infectious diarrhea is the consumption of eggs and meat contaminated with Salmonella. The findings suggest that Salmonella infection is closely associated with food contamination.

In addition, the clinical symptoms of patients with Salmonella infection in the present study concurs with previously published research (Qu et al., 2011; Shen et al., 2014a,b), and our study has also shown that Vibrio parahaemolyticus infection has decreased, but that Salmonella has become the major strain associated with diarrhea since 2011 in Shanghai (Liu et al., 2004, 2008; Chen et al., 2010).

Serious foodborne Salmonella outbreaks have been reported many times in the past. Hence, we investigated the antibiotic susceptibility of various pathogens during the last 4 years, with the results revealing that the sensitivity rate of Salmonella to commonly used antibiotics was >70%. It is noteworthy that the sensitivity rate to quinolones was decreased compared with a previous study (Qi et al., 2016), so it will be important to continue monitoring the sensitivity of Salmonella to these agents. Also, antibiotics should be employed to monitor susceptibility and perhaps prophylactically to prevent the outbreak of serious diarrhea symptoms in patients.

Currently, third-generation cephalosporins and quinolones are the most common antibiotics used to treat Salmonella infections in the clinic. With increasing usage and the frequency of prescriptions (Xu and Zhang, 2012; Lamberti et al., 2014), unfortunately many Salmonella strains have now become resistant to quinolones, which makes treatment of infection much more difficult and costly. A recent report has shown that about 22.5% of non-typhoid Salmonella strains are resistant to at least one antibiotic. The most common multi-antibiotic-resistant phenotypes are to ampicillin, chloramphenicol, streptomycin, sulfonamides and tetracycline (9.35%) (Liu et al., 2004). Patients with diarrhea take non-standard broad-spectrum antibiotics (nalidixic acid, ampicillin, sulfamethoxazole, streptomycin and tetracycline) but resistance is clearly adversely affecting the control and treatment of Salmonella infections.

Our study unfortunately did not analyze the antibiotic resistance of the same or different PFGE type strains. However, for identical PFGE strains, they had very similar antibioticresistant profiles. For those strains sharing the same PFGE type but different antibiotic-resistant profiles, the antibiotics resistance gene might mutate and also the mutation site is

FIGURE 6 | The 60 S. Typhimurium strains collected from 2014 to 2017 were classify into 16 types (A–P) based on a >60% similarity (Endonuclease XbaI).

not located at the PFGE restriction enzyme cutting site. As a result, antibiotic resistance cannot be reflected fully in the PFGE classification (Lin et al., 2010). Therefore, the PFGE method can be used as a sensitive method to study genotyping in the molecular epidemiology of Salmonella infection. Furthermore, it can help to establish the relationships among cases. Clearly, further research will be necessary to characterize the resistant strains identified by the PFGE method. Nevertheless, the PFGE method can be used clinically to identify the pathogen and to investigate the origin of the outbreak of Salmonella poisoning.

A limitation of the present study is that all samples came from a single hospital and therefore our findings may not reflect the situation in other regions of China.

### CONCLUSION

The proportion of gastrointestinal infectious diarrhea cases caused by Salmonella has increased in the Southwest of Shanghai. The types of Salmonella bacteria are numerous and are widely distributed. Our study provides a basis for the early clinical identification and diagnosis of pathogens causing intestinal infectious diseases.

### REFERENCES


### AUTHOR CONTRIBUTIONS

XQ, XX, SB, and DL conceived and design the study. All authors were responsible for acquisition and analysis of data, commented on the draft, and approved the final version of the manuscript. XQ and DL were in charge of statistical analysis. XQ, PL, SB, and DL drafted the manuscript.

### FUNDING

This work was supported by the Science and Technology Innovation Fund Program of Jinshan District (grant number: 2017-3-06) and Science and Technology Commission of Shanghai Municipality (grant number: 16411972100).

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmicb. 2019.02025/full#supplementary-material


associated with lyophilized enteral nutrition. J. Hosp. Infect. 58, 122–127. doi: 10.1016/j.jhin.2004.05.003


agar dilution testing. Antimicrob. Agents Chemother. 58, 1779–1781. doi: 10. 1128/AAC.02181-13


**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 Qi, Li, Xu, Yuan, Bu and Lin. 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-10-02025 September 13, 2019 Time: 16:58 # 9