Genomic Epidemiology of Multidrug-Resistant Nontyphoidal Salmonella in Young Children Hospitalized for Gastroenteritis

ABSTRACT Nontyphoidal Salmonella (NTS) gastroenteritis in children remains a significant burden on health care and constitutes a majority of all admissions for Salmonella infections in public hospitals in Hong Kong. In this prospective study, 41% of 241 children hospitalized with gastroenteritis from three public hospitals during 2019 were culture confirmed to have NTS infection. These Salmonella isolates were whole-genome sequenced and in silico predicted for their serovars/serotypes using the Salmonella In Silico Typing Resource (SISTR) and SeqSero1, and the antimicrobial resistance (AMR) genes were determined. Phylogenetic analysis revealed three major clades belonging to Salmonella enterica serovar Enteritidis sequence type 11 (ST11) (43%), multidrug-resistant (MDR) S. Typhimurium ST19 (12%) and its monophasic variant ST34 (25%), and mostly singletons of 15 other serovars. MDR S. Typhimurium and its variant were more common in infants <24 months of age and possessed genotypic resistance to five antimicrobial agents, including ampicillin (A), chloramphenicol (C), aminoglycosides (Am), sulfonamides (Su), and tetracyclines (T). Older children were more often infected with S. Enteritidis, which possessed distinct genotypic resistance to AAmSu and fluoroquinolones. In addition, 3% of the isolates possessed extended-spectrum beta-lactamase (ESBL) CTX-M genes, while one isolate (1%) harboring the carbapenemase gene blaNDM-1 was identified. Our findings provide a more complete genomic epidemiological insight into NTS causing gastroenteritis and identify a wider spectrum of determinants of resistance to third-generation beta-lactams and carbapenems, which are often not readily recognized. With high rates of multidrug-resistant NTS from studies in the Asia-Pacific region, the rapid and reliable determination of serovars and resistance determinants using whole-genome sequencing (WGS) is invaluable for enhancing public health interventions for infection prevention and control. IMPORTANCE Nontyphoidal Salmonella (NTS) gastroenteritis is a foodborne disease with a large global burden. Antimicrobial resistance (AMR) among foodborne pathogens is an important public health concern, and multidrug-resistant (MDR) Salmonella is prevalent in Southeast Asia and China. Using whole-genome sequencing, this study highlights the relationship of the MDR Salmonella serotypes and the diverse range of Salmonella genotypes that contaminate our food sources and contribute to disease in this locality. The findings update our understanding of Salmonella epidemiology and associated MDR determinants to enhance the tracking of foodborne pathogens for public health and food safety.

N ontyphoidal Salmonella (NTS) remains an important etiology of diarrhea and is responsible for an estimated 197. 35 (1). The burden disproportionately affects children under 5 years of age, and while most cases are mild, our review of 10-year admissions of patients with Salmonella infections from public hospitals of Hong Kong revealed that 88.1% of the 4,828 admissions were children aged #15 years (2). NTS gastroenteritis among children thus remains a significant burden on health care. Although antibiotics are not routinely recommended except in cases with invasive complications or bacteremia (3), our previous study showed that substantially longer stays and higher costs were associated with hospitalization with fluoroquinoloneresistant Salmonella spp. than with the cohort with infections caused by susceptible strains (4). Thus, interventions that reduce its burden or reduce antimicrobial resistance (AMR) may reduce the morbidity and health care costs of these infections.
Whole-genome sequencing (WGS)-based prediction of serovars is rapid and provides genotypic details for outbreak identification, source tracking, and diagnostic and epidemiological surveillance. With over 2,600 reported serovars (serotypes) to date (5), most diagnostic laboratories resort to only serogroup determination using polyclonal antibodies, without serovar information that facilitates interventions for outbreak investigations.
Accumulating evidence from performance comparisons of in silico serovar prediction tools (6)(7)(8) revealed that the Salmonella In Silico Typing Resource [SISTR] (9) has higher concordance than other methods, including SeqSero1 (10), SeqSero2 (11), multilocus sequence typing (MLST) (12), or metric-oriented sequence typing (MOST) (13). In addition, WGS allows the prediction of AMR genes, which further enhances the differentiation between clades of the same serotype. For example, two clades of Salmonella enterica serovar Typhimurium were described in Hong Kong (14), one of which was multidrug resistant (MDR) to ampicillin (AM), chloramphenicol (CHL), and trimethoprim. Recently, we identified a high prevalence of extended-spectrum betalactamase (ESBL)-and carbapenemase-producing Enterobacteriaceae from food animal products in our local wet markets (15). Such ESBLs and carbapenem resistance have been increasingly reported in Salmonella spp. from clinical cases, animals, and food samples (16,17). This highlights the importance of the monitoring and tracking of AMR in association with specific Salmonella serotypes.
In this study, we sought to use WGS to describe the epidemiology of NTS isolated from children hospitalized with Salmonella gastroenteritis from three acute-care hospitals in Hong Kong in 2019. The study infers detailed epidemiological insights into the Salmonella serovars, AMR, and characteristics of the common NTS serotypes causing gastroenteritis requiring hospitalization in children.

RESULTS
Concordance of serotypes and in silico serovar prediction. Salmonella serovars were predicted by SISTR and SeqSero1, and details of the serovars are listed in Data Set S1 in the supplemental material. One strain (isolate P144_06/19) did not have sufficient sequence reads for identification and was excluded from further analysis. SISTR predicted serovars with the full antigenic formula for all the strains, while SeqSero1 predicted only 95.0% with the full antigenic formula (Table S1). SeqSero1 could not distinguish five isolates due to the provision of only partial antigenic formulas lacking H1 or H2 antigens and for two isolates gave multiple serovar assignments to S. enterica serovar Goldcoast or S. Brikama (isolate A030_06/19) (Table 1) and S. Javiana or Salmonella serotype II 9,12:l,z28:1,5 (isolate A161_10/19) (Table 1). Disagreements on the traditional serogroup results were noted for three strains, whereby two strains belonged to groups C and P by both SISTR and SeqSero1, whereas the serogroup determination by agglutination misidentified both strains as belonging to group B, and a further isolate could not be serotyped by the traditional method (Table S2).
Overall, phenotypic resistance was highly correlated with the presence of known resistance determinants, with genotype agreeing with phenotype for 486 of 502 phenotypic tests for all nine antimicrobials from six different classes. This resulted in an overall concordance between the two of 96.8%. In total, 174 of the 502 antibiotic tests indicated resistance, and associated genes were predicted to cause resistance to all antimicrobials tested except meropenem. This resulted in an overall sensitivity of 96.6% (168/174) ( Table 2). Among the 328 phenotypic susceptibility test results, there were 10 occasions where resistance genes were detected by WGS. Among these resistance genes, 8 were related to ciprofloxacin, and 2 were related to chloramphenicol. This resulted in an overall specificity of 97.0% (318/328) ( Table 2).
For the mixed clade, almost all the serovars had point mutations in three distinct gyrA (n = 3) and parC (n = 15) regions, except for S. Singapore. S. Kentucky carried both gyrA and parC without PMQR genes.

DISCUSSION
To the best of our knowledge, this is one of the few studies using WGS to determine the serotypes and AMR of nontyphoidal Salmonella from children hospitalized for gastroenteritis in Hong Kong. This updates our epidemiology of the Salmonella serotypes and the associated AMR.
Our study showed that S. enterica serovar Typhimurium and its monophasic variant 4, [5],12:i:-were common in infants under 2 years of age, and S. Enteritidis was common in the older children. It is a custom of Chinese parents to feed a customized congee mixed with cooked minced pork during weaning for young infants. Raw and undercooked eggs or chicken (commonly associated with S. Enteritidis) is not an important source of protein until the children are older and gradually switch to an adult diet. In fact, the very high and extremely high MDR levels of S. Typhimurium and its monophasic variant 4, [5],12:i:-characterized by the multilocus sequence type 34 and the antimicrobial resistance profile ASSuT have been principally related to pork sources (20,21). These clones had been previously reported in many European countries (previously described as DT194) in the early 2000s.
Salmonella species remains one of the top 10 causes of bacteremia in pediatric patients in China (22). Although bacteremia was uncommon in our cohort, a third of children complained of both fever and blood in stool, and half of the infants were under 3 months of age. MDR strains were mainly found in S. Typhimurium and its monophasic variant 4, [5],12:i:-. Our findings of ESBL-producing S. Typhimurium and S. Enteritidis and carbapenem-resistant S. Enteritidis are important in the trends of rising AMR in our community. A recent study from China demonstrated the emergence and prevalence of foodborne Salmonella harboring a chromosomally located bla CTX-M-55  (21) bla TEM-1B gyrA strA strB sul2 tetA (4) AM CIP (10) bla TEM-1B gyrA strA strB sul2 (10) AM T (9) bla TEM-1B strA strB sul2 tetB (9)  The overall percentage of Salmonella spp. resistant to ceftriaxone in the WHO GLASS report for Hong Kong was 5.18% in 2017 (25) and was similar to the rates of 5.9% (ceftriaxone) in Taiwan (26), 3% (ceftriaxone) in the United States (27), and 1.9% (cefotaxime) in Europe (21). None of the Salmonella spp. from Hong Kong were reported to be resistant to carbapenems (25). This is in contrast to a recent report from Zhejiang, China, of children hospitalized with NTS, with rates of resistance to ceftriaxone of 37.4% and to imipenem of 1.3% (28). However, our detection of a high prevalence of ESBL-producing Enterobacteriaceae of 52.8% and of carbapenem resistance from gut carriage among healthy subjects in our community in Hong Kong raises concern (29), while another study on food animal products from our wet market highlighted the high prevalence of both ESBL-and carbapenemase-producing E. coli due to NDM-1 (15). It has been noted that ESBL and carbapenem phenotypic susceptibilities are not routinely tested for extraintestinal NTS species in microbiology diagnostic laboratories, especially if first-line agents, e.g., ampicillin, fluoroquinolone, and trimethoprim-sulfamethoxazole, are susceptible according to the 2020 guidelines of the Clinical and Laboratory Standards Institute (30). The detection of the NDM gene suggests that our potential reservoir of AMR genes is much wider beyond our health care settings, and wider surveillance of AMR among Salmonella serotypes is urgently warranted. Our study further supports the robustness of WGS for in silico serotyping and predicting resistance phenotypes in Salmonella enterica. It is a valuable tool that could be more widely adopted in medical and public health laboratories in order to limit the propagation and dissemination of these "superbugs." Conclusion. MDR S. Enteritidis, S. Typhimurium, and its monophasic variant were the most common serotypes causing gastroenteritis requiring hospitalization among young children. WGS-based serotyping using SISTR combines the advantages of cgMLST and AMR determination to facilitate public health surveillance. Wider surveillance of ESBL-and carbapenemase-producing Salmonella serotypes is urgently warranted.
DNA extraction and WGS. DNA from cultures of the isolates grown overnight was prepared using the Wizard genomic DNA purification kit (Promega, USA), followed by DNA library preparation (Riptide DNA library preparation kit; iGenomx, USA), according to the manufacturers' instructions and as previously described (15). WGS was performed using the NextSeq platform (Illumina, USA) from the core facility to obtain paired-end reads of 150 bp. Sequence reads were demultiplexed according to the manufacturer's instructions. Quality control (QC) of the raw reads was performed using FastQC version 0.11.9 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) (33) and trimmed for quality using Trimmomatic version 0.39 (http://www.usadellab.org/cms/?page=trimmomatic) (34) with the default QC setting (QC30) and the TruSeq3 Illumina adapter (TruSeq3-PE-2.fa) to remove library adapters and lowquality short reads. High-quality short reads were de novo assembled for contigs using SPAdes version 3.15.2 (http://cab.spbu.ru/software/spades/) (35). Contigs that aligned to Salmonella genomes by Kraken2 version 2.1.2 (http://ccb.jhu.edu/software/kraken2/) (36), were longer than 500 bp, and had coverage of .10 served as the reference index to further filter short reads using Bowtie2 version 2.4.4 (http://bowtie-bio.sourceforge.net/bowtie2/index.shtml) (37). Filtered contigs with a depth of 10 and a length of $500 bp from each genome were included in the analyses.
WGS-based in silico serotype prediction. Serotyping of Salmonella spp. was predicted by using WGS-assembled contig files (in FASTA format) with SeqSero1 (https://cge.cbs.dtu.dk/services/SeqSero/) (10) and with assemblies using SISTR (https://lfz.corefacility.ca/sistr-app/) (9). In silico 7-gene MLST sequence types (STs) were generated by the SISTR platform. The genomes were also assigned to STs using MLST 2.0 (https://cge.cbs.dtu.dk/services/MLST/) (38). Serotype predictions were compared to laboratory serogroup results using antibody agglutination. Full agreement indicates that the in silico method was 100% concordant with laboratory results; disagreement indicates incorrect or incongruent results of the two. No/incomplete prediction indicates that the in silico method was not able to provide a serogroup.
Statistical analysis. Patient demographic and laboratory data were analyzed using SPSS v22.0. Categorical data were analyzed using a chi-square test and Fisher's exact test. A Kruskal-Wallis test was used for nonnormally distributed data that were presented as medians and interquartile ranges [IRs]. The Salmonella serotypes recognized by SISTR were grouped into multidrug-resistant (MDR) (resistant to $3 drugs) and non-multidrug-resistant (resistance to ,3 antimicrobial agents) serotypes. Statistical significance was considered a P value of ,0.05.
Data availability. All sequence data associated with this study have been deposited in the NCBI database under BioProject accession no. PRJNA678820.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, XLSX file, 0.1 MB. SUPPLEMENTAL FILE 2, PDF file, 0.2 MB.