Whole Genome Sequencing of Extended-Spectrum- and AmpC- β-Lactamase-Positive Enterobacterales Isolated From Spinach Production in Gauteng Province, South Africa

The increasing occurrence of multidrug-resistant (MDR) extended-spectrum β-lactamase- (ESBL) and/or AmpC β-lactamase- (AmpC) producing Enterobacterales in irrigation water and associated irrigated fresh produce represents risks related to the environment, food safety, and public health. In South Africa, information about the presence of ESBL/AmpC-producing Enterobacterales from non-clinical sources is limited, particularly in the water–plant-food interface. This study aimed to characterize 19 selected MDR ESBL/AmpC-producing Escherichia coli (n=3), Klebsiella pneumoniae (n=5), Serratia fonticola (n=10), and Salmonella enterica (n=1) isolates from spinach and associated irrigation water samples from two commercial spinach production systems within South Africa, using whole genome sequencing (WGS). Antibiotic resistance genes potentially encoding resistance to eight different classes were present, with blaCTX-M-15 being the dominant ESBL encoding gene and blaACT-types being the dominant AmpC encoding gene detected. A greater number of resistance genes across more antibiotic classes were seen in all the K. pneumoniae strains, compared to the other genera tested. From one farm, blaCTX-M-15-positive K. pneumoniae strains of the same sequence type 985 (ST 985) were present in spinach at harvest and retail samples after processing, suggesting successful persistence of these MDR strains. In addition, ESBL-producing K. pneumoniae ST15, an emerging high-risk clone causing nosocomical outbreaks worldwide, was isolated from irrigation water. Known resistance plasmid replicon types of Enterobacterales including IncFIB, IncFIA, IncFII, IncB/O, and IncHI1B were observed in all strains following analysis with PlasmidFinder. However, blaCTX-M-15 was the only β-lactamase resistance gene associated with plasmids (IncFII and IncFIB) in K. pneumoniae (n=4) strains. In one E. coli and five K. pneumoniae strains, integron In191 was observed. Relevant similarities to human pathogens were predicted with PathogenFinder for all 19 strains, with a confidence of 0.635–0.721 in S. fonticola, 0.852–0.931 in E. coli, 0.796–0.899 in K. pneumoniae, and 0.939 in the S. enterica strain. The presence of MDR ESBL/AmpC-producing E. coli, K. pneumoniae, S. fonticola, and S. enterica with similarities to human pathogens in the agricultural production systems reflects environmental and food contamination mediated by anthropogenic activities, contributing to the spread of antibiotic resistance genes.


INTRODUCTION
The discovery of antibiotics in the 1940s led to a new age in medical care. However, the global increase in antimicrobial resistance (AMR) is reducing the effectiveness of clinically important antibiotics (Lobanovska and Pilla, 2017;Dandachi et al., 2019). Examples of shifting resistance profiles in bacteria within the β-lactam class of antibiotics, including penicillins and third generation cephalosporins, which are the most widely used in human and veterinary medicine and widely expressed AMR are being reported (Finton et al., 2020). Persistent exposure to these antibiotics has resulted in bacteria becoming resistant by evolving extended-spectrum β-lactamases (ESBLs), which hydrolyze the β-lactam ring within the antibiotic, thus rendering it inactive (Bush and Jacoby, 2010). Consequently, production of ESBLs is regarded as one of the most clinically significant resistance mechanisms (Bush and Jacoby, 2010), with ESBLproducing Enterobacterales (Escherichia coli, Klebsiella pneumoniae, and Serratia spp., among others) listed as priority pathogens for research and development in the new frontier of antibiotics (WHO, 2017).
The increase in antimicrobial resistant strains and effective resistance mechanisms among Enterobacterales has led to numerous global reports of ESBLs, AmpC-, and more recently carbapenemase-producing Enterobacterales not only in clinical settings, but also in the agricultural environment (Ye et al., 2017;Al-Kharousi et al., 2019;Dandachi et al., 2019;Hassen et al., 2020;Richter et al., 2020). Although members of the Enterobacterales family occur naturally in human and animals' gastrointestinal tracts as well as in the environment (water, soil, and plants; Blaak et al., 2014;Ye et al., 2017), the occurrence of multidrug-resistant (MDR) strains in the different habitats is concerning. Inadequately treated or untreated effluents from industries, households, and zootechnical farms are reported as one of the main contamination causes of South African surfaceand ground water resources (Verlicchi and Grillini, 2020). It is also well-documented that the three principal antibiotic contamination channels in the environment are animal-, human-, and manufacturing waste (O'neill, 2016). Consequently, contamination of soil, irrigation-, and drinking water as well as crops can occur, adding additional exposure routes to humans (Finton et al., 2020;Lopes et al., 2021a).
Previous surveillance studies have shown prevalence of MDR ESBL/AmpC-producing Enterobacterales in fresh vegetables sold in South Africa (Richter et al., 2019) and in other countries e.g., the Netherlands, Switzerland, and Germany (Reuland et al., 2014;Zurfluh et al., 2015;Reid et al., 2020). Occurrence of ESBL-producing Enterobacterales has also been reported in corresponding irrigation water sources and cultivated crops (Blaak et al., 2014;Njage and Buys, 2014;Ye et al., 2017). Furthermore, Richter et al. (2020) reported the occurrence of ESBL/AmpC-producing Enterobacterales in different spinach supply chains from irrigation water and produce at harvest, throughout processing and at retail in the Gauteng Province of South Africa.
The high discriminatory power of whole genome sequencing (WGS) has led to an increase in use of this method for detecting points of contamination, source tracking, pathogen surveillance, and outbreak investigations (Oniciuc et al., 2018;CDC, 2019). WGS provides information regarding multiple AMR genes, genomic mutations, mobile genetic elements, and association with resistance genes, as well as molecular typing like multilocus sequence typing (MLST; Oniciuc et al., 2018;CDC, 2019;Kim et al., 2020). Consequently, the WGS results can aid in elucidating the genetic relationship among isolates from different environments and along the food chain (Adator et al., 2020). Surveillance of antimicrobial resistant strains through WGS is increasingly being used due to increasing accessibility and affordability (Adator et al., 2020). In South Africa, WGS has been used for characterization of clinical ESBL-producing K. pneumoniae strains among others (Founou et al., 2019), as well as typing of Listeria monocytogenes from environmental and clinical settings during the 2017 listeriosis outbreak (Thomas et al., 2020). However, the use of WGS for surveillance of antimicrobial resistant potential pathogenic Enterobacterales in retailed fresh produce and the production environment has not been reported locally.
The World Health Organization (WHO) developed the Global AMR Surveillance System (GLASS) in 2015 supporting research and surveillance as well as a global data sharing through a standardized analysis approach (WHO, 2020). Initially, the GLASS focus was mainly on surveillance of human priority pathogens, but has since shifted to include AMR in foodborne pathogens (WHO, 2020). Moreover, the One Health framework for understanding AMR in pathogenic gram-negative bacteria is increasingly attracting attention (Collignon and McEwen, 2019). In South Africa, information regarding AMR in fresh produce production systems and specifically focusing on the Enterobacterales is lacking. The aim of this study was thus to use WGS for analysis of AMR genes, associated mobile genetic elements, virulence factors, serotypes, multi-locus sequence types, and pathogenicity of selected, partially characterized, ESBL/ AmpC-producing environmental Enterobacterales from commercial spinach production systems (Richter et al., 2020). These isolates included four different species (E. coli, K. pneumoniae, S. fonticola, and S. enterica) listed by the WHO as a particular threat for gram-negative bacteria that are resistant to multiple antibiotics (WHO, 2017), while isolates harbouring integrons as described in Richter et al. (2020) were preferentially selected. The results of this study will contribute to address the problem of antimicrobial drug resistance at the water-plant-food interface and how it might impact human health and disease.

Isolation and DNA Extraction of ESBL/AmpC-Producing Enterobacterales
Irrigation water and fresh produce samples from spinach production systems were collected and ESBL-producing Enterobacterales were isolated as described (Richter et al., 2020). A selection of 19 isolates were further characterized ( Table 1). The genomic DNA of each isolate was extracted with the DNeasy PowerSoil kit (Qiagen, South Africa) according to the manufacturer's instructions. Following gDNA extraction, the concentrations were determined using the Qubit dsDNA Broad Range Assay and a Qubit 2.0 fluorometer (Life Technologies, Johannesburg) and quantification was determined on a Nanodrop 2000 (ThermoScientific, Johannesburg).

DNA Sequencing and Whole Genome Analysis
Sequencing was performed on an Illumina MiSeq instrument (2 × 300 bp) with 100× coverage by the National Institute for Communicable Diseases Sequencing Core Facility, South Africa, following preparation of multiplexed paired-end libraries with the Nextera XT DNA sample preparation kit (Illumina, San Diego, CA, United States). The resultant reads were quality trimmed using CLC version 20 1 and de novo assembled. The contiguous sequences were then submitted to the National Centre for Biotechnology Information (NCBI) Prokaryotic Genome Annotation Pipeline. 2 The AMR gene presence was corroborated using ABRicate 3 that included the Comprehensive Antibiotic Resistance Database (CARD), ARG-ANNOT, ResFinder, NCBI AMRFinder Plus, and MEGARes databases (Zankari et al., 2012;Gupta et al., 2014;Jia et al., 2017;Feldgarden et al., 2019;Doster et al., 2020). Plasmid replicon types were determined with PlasmidFinder (version 2.1; Carattoli et al., 2014). Using the Centre for Genomic Epidemiology (CGE) platform 4 , mobile genetic elements for all four species, sequence types of E. coli, K. pneumoniae, and S. enterica as well as the E. coli serotypes based on lipopolysaccharide (O-antigen) and capsular flagella (protein; H-antigen), and virulence genes of E. coli were determined with MGEFinder, Multilocus Sequence Typing (MLST; version 2.2), SeroTypeFinder (version 2.0), and VirulenceFinder (version 2.0), respectively (Larsen et al., 2012;Joensen et al., 2014Joensen et al., , 2015Johansson et al., 2021). The following parameters were used in the Serotype Finder Web-based tool: 85% threshold for %ID and 60% minimum length (the number of nucleotides in a sequence of interest that must overlap a serotype gene to count as a hit for that gene; Joensen et al., 2015). The in silico serotyping based on the capsule polysaccharide (K-antigen) of K. pneumoniae strains was conducted using Kaptive Web (Wick et al., 2018), while the presence of virulence genes for K. pneumoniae was 3 https://github.com/tseemann/abricate 4 https://cge.cbs.dtu.dk/services/ identified by using the Institut Pasteur's Klebsiella database. 5 Additionally, paired reads of the WGS raw data files for the S. enterica strain were uploaded to the online SeroSeq tool version 1.0 which predicted the Salmonella serotype of the requested isolate (Zhang et al., 2015;Thompson et al., 2018). The Salmonella Pathogenicity Islands (SPI) were identified with SPIFinder 2.0 (Roer et al., 2016). Next, the existence of virulence factors in each SPI was analyzed by performing BLAST analysis on the predicted SPIs against the virulence factor database (VFDB; Chen et al., 2016;Ashari et al., 2019). The virulence factors of S. fonticola were determined using the VFDB with ABRicate (Chen et al., 2016). All sequences were submitted to the INTEGRALL database 6 for annotation and integron number assignment. Using PathogenFinder (version 1.1) on the CGE platform, the strains' pathogenicity towards humans was predicted (Cosentino et al., 2013).

Data Availability
The nucleotide sequences of the 19 Enterobacterales strains described in this paper were deposited in the National Center for Biotechnology Information GenBank database in the BioProject number: PRJNA642017, accession numbers NZ_JACAAL010000000, NZ_JACBIV000000000-NZ_JACBJE000000000, and NZ_JACNY M000000000-NZ_JACNYT000000000 ( Table 2).

Detection of Mobile Genetic Elements and Association to Antimicrobial Resistance Genes
Known resistance plasmid replicon types of Enterobacterales including IncFIB, IncFIA, IncFII, IncB/O, and IncHI1B were observed in all strains following analysis with PlasmidFinder (data not shown). The β-lactamase gene, bla CTX-M-15 , was the only resistance gene associated with plasmids (IncFII_pKP91 and/or IncFIB(K)_1_Kpn3) in four K. pneumoniae strains upon further analysis ( Table 3). The IS6 family elements (IS6100) have been reported to play a pivotal role in the dissemination of resistance determinants in gram-negative bacteria (Partridge et al., 2018), and were observed in relation to the dfrA14b resistance gene in all five K. pneumoniae strains ( Table 3).
The bla CTX-M-14 and sul2 resistance genes were related to the ISEcp1 element within the IS1380 family in one E. coli and three K. pneumoniae strains, respectively, while one S. fonticola strain carried a sul2 gene that was related to IS110 (Table 3). One E. coli strain carried the qnrS1 resistance gene that was related to ISKra4. Other insertion sequences detected belonged predominantly to the IS3 and IS110 families (data not shown), with one K. pneumoniae strain carrying the bla SHV-80 broad spectrum β-lactamase that was related to IS3 (Table 3). In all K. pneumoniae strains (n = 5) where the qnrB1 resistance gene was present, association to Tn5403 was seen ( Table 1). In one E. coli and five K. pneumoniae strains, integron In191 was observed, with dfrA14 in the cassette array (Table 3).

In silico Analysis of Serotypes, Multi-Locus Sequence Types, and Virulence Factors
The in silico MLST analysis, predicted serotypes, and pathogenicity probability of all 19 strains are shown in Table 2. Three different sequence types (ST58, ST117, and ST10) and three different serotypes (O75:H9, O11:H4, and O8:H17) were observed in the three E. coli strains. The five K. pneumoniae strains belonged to three different sequence types and three different serotypes (KL27, KL24, and KL39) which were observed based on the K-antigen, while the O-serotype included O4 and O1 ( Table 2). The predicted antigenic profile of the S. enterica strain was O11:k:1,2. Furthermore, the S. enterica strain contained 11 Salmonella SPI, namely SPI-1, SPI-2, SPI-3, SPI-4, SPI-5, SPI-9, SPI-13, SPI-14, one unnamed, as well as the centisome 63 (C63PI) and 54 (CS54) pathogenicity islands, each harbouring between 20 and 60 virulence factors (Supplementary Table S1). A total of 42 virulence genes were identified in the E. coli and K. pneumoniae strains (Supplementary Tables S2, S3). Of these, 20 were detected in E. coli strains only and 20 in K. pneumoniae strains only, while fyuA (iron uptake associated with siderophores) and irp2 (iron uptake) virulence factors were detected in two E. coli strains from irrigation water as well as three K. pneumoniae strains from spinach samples. All three E. coli strains carried the terC (tellurite resistance) virulence gene (Supplementary Table S2) and in all five K. pneumoniae strains, the mrkA, mrkB, mrkC, mrkD, mrkE, (main structural subunit and assembly machinery for type 3 fimbriae) mrkH (regulatory protein), and mrkI (DNA binding protein) virulence factors were present (Supplementary Table S3). No shiga-toxin producing genes were present in the E. coli strains. A total of 89 virulence factors were identified in the S. fonticola FIGURE 1 | Antimicrobial resistance (AMR) genes present in Enterobacterales isolated from water and spinach from farm to retail. W, water; S, spinach.  Table S4). This included 25, 18, 16, and 6 of the virulence factors present in 100% (n = 10), 90, 80, and 70% of the selected S. fonticola strains, respectively, while the remaining 24 virulence factors were present in varying numbers in 1-6 of the strains (Supplementary Table S4). The iroN salmochelin siderophore receptor which plays a role in disease establishment was present in three S. fonticola strains (two from unwashed baby spinach samples at the retailer and one from the irrigation pivot point water), one E. coli strain from the ground water, as well as in the SPI-13 in the S. enterica strain from river irrigation water. Relevant similarities to human pathogens were predicted for all 19 strains with a confidence of 0.635-0.721 in the S. fonticola strains (n = 10), 0.852-0.931 in the E. coli strains (n = 3), 0.796-0.899 in the K. pneumoniae strains (n = 5), and 0.939 in the S. enterica strain ( Table 2).

DISCUSSION
To the authors knowledge, this is the first study to use WGS for in-depth molecular characterization of ESBL/AmpC-producing E. coli, K. pneumoniae, S. enterica, and S. fonticola isolates, previously identified and partially characterized, from spinach and irrigation water samples in commercial production chains (Richter et al., 2020). Characterization included AMR, mobile genetic elements (e.g., insertion sequences, plasmids, and integrons), serotypes, and determining the pathogenicity. All these factors are crucial in defining and attributing infection sources of food-related outbreaks caused by resistant microorganisms (Oniciuc et al., 2018). Overall, the results corresponded with main global findings where AMR genes and associated mobile genetic elements have been reported in Enterobacterales from fresh produce and irrigation water, with the potential to pose a health risk to humans upon exposure (Jones-Dias et al., 2016;Finton et al., 2020). Previously, the presence of intI3 was reported in a high percentage of isolates from the current study following conventional PCR and sequencing (Richter et al., 2020). However, in-depth WGS analysis showed that no attI fragment preceded the IntI3 genes; consequently, the IntI3 genes detected did not form part of complete integrons, which typically include an integrase intI gene encoding a site-specific recombinase, a recombination site attI as well as a promoter (P c ; Kaushik et al., 2018). Overall, six isolates in the current study were positive for Class 1 integrons (In191), similar to In191 positive clinical ESBL-producing Enterobacterales from an academic teaching hospital in Pretoria, SA . Additionally, these MDR environmental isolates harbored various virulence factors central to pathogenicity, including genes associated with urinary tract infections and iron sequestering systems crucial for disease establishment. All isolates had relevant similarity to human pathogens and form part of the WHO 3rd generation cephalosporin resistant critical priority pathogens (WHO, 2017). Two of the E. coli strains from the current study harboured plasmid-mediated AmpC bla CMY-2-like genes (bla  and bla CMY-101 ), which correspond to the phenotypic profile of resistance to expanded-spectrum cephalosporins previously reported for these isolates using traditional PCR analysis (Richter et al., 2020). The bla CMY-2 pAmpC genes are the most commonly reported in E. coli and other Enterobacterales species and have clinical relevance, as it inactivates third generation cephalosporins and mediates resistance to carbapenems (Jacoby, 2009;Bortolaia et al., 2014). Three different multi-locus sequence types, namely ST58, ST10, and ST117, were identified in the E. coli isolates. Isolated from the retailed unwashed spinach samples in the current study, ST58 E. coli have previously also been associated with human extra-intestinal infections including sepsis, and have emerged worldwide in wild and food-production animals (Reid et al., 2020). As an example, ST58 E. coli with serotype O75:H9 corresponded to an E. coli strain of bovine origin from Pakistan and also carried the IncFIB plasmid (Ali et al., 2020). Although the strain from the current study had less AMR genes than reported in ST58 E. coli with serotype O75:H9 by Ali et al. (2020), the trimethoprim (dfrA14), fluoroquinolone (qnrS1), and β-lactam (bla CTX-M-15 ) genes corresponded. Similarly, uropathogenic ST58 E. coli with resistance to fluoroquinolone and trimethoprim have previously been isolated from hospital patients in Australia (McKinnon et al., 2018). The bla CTX-M-15 gene identified in the ST58 E. coli strain from the current study was associated with the ISKra4 insertion sequence, previously identified in K. pneumoniae harbouring bla CTX-M-15 , and was responsible for the movement to different parts of the genome through a replicative transposition mechanism (Razavi et al., 2020). In contrast to Hauser et al. (2013) who identified food-associated shiga-toxin producing E. coli ST58, no stx genes were present in the strains. The E. coli ST58 from the current study harboured the gad (glutamate decarboxylase) virulence gene, similar to E. coli ST58 strains isolated from aragula (rocket; Reid et al., 2020). However, the presence of lpfA (long polar fimbriae) and terC (tellurium ion resistance protein) virulence factors in the strain from the current study contrasted the virulence gene profiles reported by Reid et al. (2020). Escherichia coli ST10 have previously been associated with human clinical infections and has been isolated from different sources including recreational and/or wastewater samples (Falgenhauer et al., 2019). From the current study, the E. coli ST10 with serotype O8:H17 was isolated from borehole water used for irrigation. Although this sequence type has previously been associated with shiga-toxin-producing E. coli (STEC; Gonzalez-Escalona and Kase, 2018), no stx genes were detected in the current study. The virulence factors present were terC (tellurium ion resistance protein), astA (EAST-1 heat-stable toxin), fyuA (ferric yersiniabactin uptake receptor), irp2 (nonribosomal peptide synthetases), iss (increased serum survival), and sitA (iron transport protein). Previously, E. coli ST10 with similar virulence gene profiles were isolated from human blood cultures and reported as extra-intestinal pathogenic E. coli (ExPEC; Maluta et al., 2017). Additionally, ESBL-producing E. coli ST10 of the same serotype have been isolated from wastewater and are depicted as a probable environmental reservoir of bla CTX-M genetic determinants (Tanaka et al., 2019). In the current study, the ST58 E. coli strain harboured the bla CTX-M-15 genetic determinant, while bla CTX-M-14 was present in the ST10 E. coli strain. Globally, the CTX-M type ESBLs (especially bla CTX-M-14 and bla CTX-M-15 ) have become the dominant genotype and the most widely distributed (Cantón et al., 2012;Adamski et al., 2015). Escherichia coli bla CTX-M-14 positive strains have previously been isolated from store-bought produce in Germany and South Africa (Richter et al., 2019;Reid et al., 2020), food producing animals in China (Liao et al., 2015) and clinical settings in Brazil and South Africa (Cergole-Novella et al., 2010;Peirano et al., 2011).
The third E. coli sequence type (ST117) detected from irrigation source water in the current study has previously been reported as part of a group of multi-serotype extraintestinal pathogenic E. coli (ExPEC) and avian pathogenic E. coli (APEC) strains (Kim et al., 2017). The E. coli ST117 strain from the current study harboured 20 virulence factors including the ExPEC hlyF (Hemolysin F) virulence gene. In previous studies, stx genes were identified in E. coli strains with the same STs detected in the current study, yet the virulence gene content and serotypes differ from the strains in the current study (Gonzalez-Escalona and Kase, 2018). However, the three non-STEC E. coli strains (ST58, ST10, and ST117) from the current study had a 93, 89, and 85% probability of being human pathogens, based on the pathogenic protein families.
In addition to E. coli, other Enterobacterales isolates harbouring bla CTX-M-15 have also been detected in different environments. In the current study, all five K. pneumoniae strains harboured the bla CTX-M-15 genetic determinant. The prevalence and dissemination of bla CTX-M throughout various environments globally underlines the different contamination routes through which fresh produce may also become contaminated with these MDR organisms. For instance, Gekenidis et al. (2020) have demonstrated the long-term persistence of E. coli harbouring bla CTX-M-15 in soil and lettuce after its introduction via irrigation water. Similarly, bla CTX-M-15 positive ST985 K. pneumoniae strains were present in spinach at harvest on the farm as well as retail samples after processing in the current study, suggesting successful persistence of these MDR strains. In four K. pneumoniae strains (ST3559, n = 1 and ST985, n = 3), the bla CTX-M-15 genes were associated with IncF replicons (IncFII K and IncFIB) which have previously been linked to diverse K. pneumoniae outbreak strains (Dolejska et al., 2012(Dolejska et al., , 2013Löhr et al., 2015). Moreover, in K. pneumoniae ST3559, bla CTX-M-15 was also associated with ISEcp1 (also called ISEc9), a member of the widely reported IS1380 family, and can enable the independent transposition with insertion mutation and genetic relocations (Partridge, 2011). The K. pneumoniae strains in the current study also harboured bla SHV ESBL encoding genes (bla SHV-187 , bla SHV-106 , and bla SHV-178 ). Previously, SHV genetic determinants were reported in K. pneumoniae from hospitals and receiving wastewater treatment plants in Romania (Surleac et al., 2020) as well as irrigation water and agricultural soil in South Africa (Iwu et al., 2020;Richter et al., 2020). Interestingly, the K. pneumoniae ST15 strain isolated from water in the current study harboured bla SHV-106 which Liakopoulos et al. (2016) previously reported to be geographically constrained and have only been described in K. pneumoniae isolates from Portugal together with bla TEM-1 . Similarly, the K. pneumoniae ST15 strain from the current study also harboured bla SHV-106 together with bla TEM-1 . Klebsiella pneumoniae ST15 is regarded as an emerging international high-risk clone causing nosocomial outbreaks worldwide with high-levels of antibiotic resistance including production of ESBLs, mainly CTX-M-15 (Han et al., 2021).
The K. pneumoniae ST3559 strain isolated from irrigation water in the current study was capsular type 27 and serotype O4, which is similar to an O4 serotype MDR K. pneumoniae outbreak strain from a neonatal care unit in sub-Saharan Africa (Cornick et al., 2020). In addition, K. pneumoniae ST3559 harboured the bla SHV-178 gene which, to the best of our knowledge, have previously only been reported in clinical Enterobacter hormaechei strains from the First Affiliated Hospital of Zhejiang University in Hangzhou (Gou et al., 2020). Apart from β-lactamase genes, the K. pneumoniae strains also harboured aminoglycoside, fosfomycin, fluoroquinolone, tetracyline, phenicol, trimethoprim, and sulfonomide resistance genes, which is a greater diversity of resistance genes than previously reported in Enterobacterales isolates from German surface waters (Falgenhauer et al., 2019). Similar to the results of clinical K. pneumoniae strains reported by Mbelle et al. (2020), In191, harbouring dfrA14 was identified in the three different K. pneumoniae sequence types of the current study, reiterating that it is not a narrow spectrum integron. In addition, dfrA14b was associated with IS6 that has previously been reported as having a vital role in the rearrangement and dissemination of antibiotic resistance (Varani et al., 2021). The presence of fosA and sul2 in all the K. pneumoniae strains of the current study also corresponds to the results reported by Mbelle et al. (2020) from clinical K. pneumoniae strains in Pretoria. The high-level of trimethoprim resistance globally has however led to trimethoprim-sulfamethoxazole no longer being recommended for outpatient treatment of urinary tract infections and similarly, the use of fosfomycin might not be efficacious anymore . Four MDR K. pneumoniae isolates from irrigation water (ST15, n = 1) and spinach (ST985, n = 3) had O1 serotypes, previously reported as the most commonly isolated serotypes from human hosts and dominant in human disease (Follador et al., 2016). However, it is noteworthy that no genes encoding carbapenamases nor resistance to colistin were identified in the current study. All five characterized K. pneumoniae strains also harbored several virulence factors including those that coded for an iron uptake system (kfu) and type 3 fimbrial adhesins (mrk) that play an important role in adhesion to medical devices such as catheters (Albasha et al., 2020;Finton et al., 2020).
Serratia spp. are opportunistic pathogens that may pose a health threat to immunocompromised and hospitalized patients (Petersen and Tisa, 2013). The S. marcescens species is most often associated with nosocomial infections; however, S. fonticola has been reported to function as a human pathogen when detected alone or may be a bystander and act as a carrier of resistance genes when discovered with other organisms (Petersen and Tisa, 2013;Aljorayid et al., 2016). Characterizing virulence genes of the MDR environmental strains therefore becomes important within the plant-food producing environment. In the current study, all S. fonticola strains harboured bla SFO-1 and numerous plasmid incompatibility (Inc) groups were identified in these S. fonticola strains (data not shown). However more in-depth plasmid typing and analysis will be required to fully understand the risk/probability of bla SFO-1 dissemination in the environment where S. fonticola naturally occurs. In certain Enterobacterales species, ESBL genes are inherently carried on chromosomes (Naas et al., 2008). This includes the bla SFO-1 ESBL gene from S. fonticola that differs from most class A ESBLs, as the β-lactamases' production can be induced by a high level of imipenem (Naas et al., 2008). The bla SFO-1 ESBL does not form part of the most clinically relevant ESBLs and are therefore rarely reported. Zhou et al. (2020) reported in contrast an increasing trend of the co-existence of plasmidborne bla SFO-1 and carbapenemase genes in clinical Enterobacter spp. in China. All the S. fonticola strains also harboured numerous fluoroquinolone resistance genes, raising a health concern for treatment options, as fluoroquinolones are often used for management of conditions including typhoid fever and MDR tuberculosis (Richards et al., 2019). Interestingly, one S. fonticola strain harboured an acquired trimethoprim (sul2) resistance gene associated with IS110, corresponding to K. pneumoniae from a German university hospital (Schwanbeck et al., 2021). The Serratia genus naturally lacks resistance genes for trimethoprim and sulfonamides (Sandner-Miranda et al., 2018). Previous reports of potential pathogenic S. fonticola primarily focused on the antibiotic resistance profiles (Tasić et al., 2013;Aljorayid et al., 2016;Hai et al., 2020). The strains from the current study additionally harboured various virulence factors. This included flagellar biosynthesis-and chemotaxisrelated genes as well as genes encoding iron uptake systems corresponding to those previously reported in important MDR nosocomial pathogenic S. marcescens (Iguchi et al., 2014).
Only one S. enterica strain isolated from river irrigation water was characterized in the current study. Irrigation water is well documented as a source for fresh produce contamination of foodborne pathogens including Salmonella spp. (Liu et al., 2018). The strain harboured an AmpC resistance gene, similar to S. enterica characterized from surface water in the United States (Li et al., 2014). In addition, the S. enterica from the current study carried aminoglycoside resistance genes [aac(6')-Iaa and aac(6')-Iy], similar to results reported by Nair et al. (2016) for non-typhoidal Salmonella spp. isolated from a United Kingdom population. Of the 23 known Salmonella SPIs previously described (Mansour et al., 2020), the isolate from the current study carried 11 SPIs. This included SPIs that are commonly reported in S. enterica and encode genes responsible for enabling invasion of epithelial cells (SPI1), facilitating the replication of intracellular bacteria (SPI2), adhesion to epithelial cells (SPI3, 4, 5, and 9;Waterman and Holden, 2003;Velásquez et al., 2016;Mansour et al., 2020), as well as SPI13 and 14 which corresponds to being part of the core genome of invasive non-typhoidal Salmonella spp. (Suez et al., 2013). Additionally, pathogenicity islands C63PI and CS54 were present in the S. enterica strain in this study, which has previously been found in the S. Typhimurium and S. Typhi genomes (Sabbagh et al., 2010;Jibril et al., 2021). Since no phenotypic indication of virulence was investigated, the prediction of virulence genes using in silico tools should be regarded with care; however, using PathogenFinder, the S. enterica strain from the current study showed 94% probability of being a human pathogen.

CONCLUSION
This is the first WGS analysis study of MDR ESBL/AmpCproducing E. coli, K. pneumoniae, S. fonticola, and S. enterica isolates from spinach production systems within South Africa. The selected isolates represent potential pathogenic genera listed by the WHO as a priority for surveillance of AMR screening. Numerous clinically relevant resistance genes were detected in the screened samples. This study showed the potential of using WGS in metadata studies for detailed molecular characterization of potential pathogenic Enterobacterales. Furthermore, the study highlighted the importance of the agricultural production environment as a source of antibiotic resistance genes within Enterobacterales at the water-plant-food interface. A more in-depth and controlled analysis, with a greater number of sequenced isolates from the farm-to-retail supply chain is required to better understand the prevalence and resistance gene transmission through the supply chain. The results from this study further highlight the need for expanded surveillance in agricultural systems.

DATA AVAILABILITY STATEMENT
The nucleotide sequences of the 19 Enterobacteriaceae strains described in this paper were deposited in the National Center for Biotechnology Information GenBank database in the BioProject number: PRJNA642017, accession numbers NZ_JAC AAL010000000, NZ_JACBIV000000000-NZ_JACBJE000000000, and NZ_JACNYM000000000-NZ_JACNYT000000000 ( Table 2).

AUTHOR CONTRIBUTIONS
EP, SD, LR, and LK contributed to the conception and design of the study. LR performed the experiments. LR, SD, MA, and AI analyzed the data. LR, EP, and SD contributed to interpretation and presentation. SD, EP, and LK were involved in funding acquisition. All authors contributed to the article and approved the submitted version.