Multi-Year Persistence of Verotoxigenic Escherichia coli (VTEC) in a Closed Canadian Beef Herd: A Cohort Study

In this study, fecal samples were collected from a closed beef herd in Alberta, Canada from 2012 to 2015. To limit serotype bias, which was observed in enrichment broth cultures, Verotoxigenic Escherichia coli (VTEC) were isolated directly from samples using a hydrophobic grid-membrane filter verotoxin immunoblot assay. Overall VTEC isolation rates were similar for three different cohorts of yearling heifers on both an annual (68.5 to 71.8%) and seasonal basis (67.3 to 76.0%). Across all three cohorts, O139:H19 (37.1% of VTEC-positive samples), O22:H8 (15.8%) and O?(O108):H8 (15.4%) were among the most prevalent serotypes. However, isolation rates for serotypes O139:H19, O130:H38, O6:H34, O91:H21, and O113:H21 differed significantly between cohort-years, as did isolation rates for some serotypes within a single heifer cohort. There was a high level of VTEC serotype diversity with an average of 4.3 serotypes isolated per heifer and 65.8% of the heifers classified as “persistent shedders” of VTEC based on the criteria of >50% of samples positive and ≥4 consecutive samples positive. Only 26.8% (90/336) of the VTEC isolates from yearling heifers belonged to the human disease-associated seropathotypes A (O157:H7), B (O26:H11, O111:NM), and C (O22:H8, O91:H21, O113:H21, O137:H41, O2:H6). Conversely, seropathotypes B (O26:NM, O111:NM) and C (O91:H21, O2:H29) strains were dominant (76.0%, 19/25) among VTEC isolates from month-old calves from this herd. Among VTEC from heifers, carriage rates of vt1, vt2, vt1+vt2, eae, and hlyA were 10.7, 20.8, 68.5, 3.9, and 88.7%, respectively. The adhesin gene saa was present in 82.7% of heifer strains but absent from all of 13 eae+ve strains (from serotypes/intimin types O157:H7/γ1, O26:H11/β1, O111:NM/θ, O84:H2/ζ, and O182:H25/ζ). Phylogenetic relationships inferred from wgMLST and pan genome-derived core SNP analysis showed that strains clustered by phylotype and serotype. Further, VTEC strains of the same serotype usually shared the same suite of antibiotic resistance and virulence genes, suggesting the circulation of dominant clones within this distinct herd. This study provides insight into the diverse and dynamic nature of VTEC populations within groups of cattle and points to a broad spectrum of human health risks associated with these E. coli strains.

While cattle are recognized as important hosts for human VTEC infections (Caprioli et al., 2005;Bettelheim and Goldwater, 2014), knowledge gaps in the ecology of VTEC persist. For example, cattle density per unit area was positively correlated with an increased risk of human infection, but not for clinically associated serogroups O26 and O91 (Frank et al., 2008). Clinical cases of VTEC O157 have been shown to peak during the summer and fall months (Tarr et al., 2005;CDC, 2012), coinciding with elevated shedding of the organisms in cattle feces (Van Donkersgoed et al., 1999;Blanco et al., 2008). However, divergent seasonal shedding patterns in cattle have been reported for non-O157 serotypes (Barkocy-Gallagher et al., 2003) and whether any seasonality exists for non-O157 VTEC disease incidence remains unclear (Klein et al., 2002;Tarr et al., 2005;CDC, 2012;Rivero et al., 2012). Even for O157, seasonal trends have not been observed in all studies (Alam and Zurek, 2006;Ellis-Iversen et al., 2009;Lammers et al., 2015).
In this study, three consecutive cohorts of yearling heifers within a closed beef herd were monitored for VTEC using a hydrophobic grid-membrane filter (HGMF) verotoxin immunoblot method. Isolates were characterized by serotype, phylotype, and virulence-associated and antibiotic resistance gene content. Phylogenetic relationships and diversity were explored using whole genome sequence data. The objective was to examine the diversity of VTEC within a specific herd over time and the associated human health risks.

Herd Characteristics and Sample Collection
This study was carried out in accordance with the principles contained in the "Guide to the Care and Use of Experimental Animals, " Vols. I and II, by the Canadian Council on Animal Care (CCAC) and followed CCAC Guidelines. The protocol was approved by the Canadian Food Inspection Agency National Centre for Animal Disease (CFIA-NCAD) Animal Care Committee (protocol # 1201).
Three consecutive cohorts of yearling heifers (n = 10, 12, and 16) from the CFIA-NCAD (Lethbridge, AB, Canada) Angus-Hereford cross research herd were sampled monthly from April 2012 to March 2013, April 2013to March 2014and May 2014to March 2015. The closed herd has a size and management system similar to that found in beef cow-calf herds in western Canada, has been previously sampled for E. coli O157:H7 (Gannon et al., 2002), and shares high genetic homogeneity due to its restricted repopulation. A total of 368 samples were collected; 103, 111 and 154 in each year, respectively. Fecal samples were collected by rectal palpation, except in July and August 2012 and August 2014 when samples were collected fresh from the ground after defecation was observed. A single event

HGMF Filtration of Unenriched Cattle Fecal Slurry on mTSAVC-BCIG Media
Ten grams of feces were homogenized in 10 mL Phosphate Buffer Saline (PBS). For each sample, 0.01 and 0.02 g of the 1:1 slurry, representing 0.005 and 0.01 g of fecal sample, respectively, were weighed and added to 8 mL aliquots of wash buffer (PBS, 0.1% Tween 20). Between May-August 2012, 0.01 and 0.005 g of unsuspended fecal samples were weighed and added to the wash buffer. The entire 8 mL volumes of both sample dilutions were filtered through sterile 0.45 µM HGMF ISO-GRID filters (Neogen, Lansing, MI, United States) using a HGMF Spreadfilter (FILTAFLEX, Ltd., Almonte, ON, Canada). Positive (isolate EC19920459; bovine O163:NM vt1+ vt2+) (PHAC NML Guelph) and negative (ATCC E. coli 25922; vt-) controls were streaked directly to the sterile HGMF ISO-GRID filter. Filters were overlaid on coated capture membranes and the mTSAVC-BCIG agar plates were incubated inverted at 37 • C for 18-24 h.
Capture Membrane Development, VT-ELISA, and PCR Confirmation of vt VT-capture membrane and HGMF grid filter assemblies were marked by needle punctures for re-orientation. The HGMFs were reserved and the VT-capture membrane was separated from the plate and probed with a mixture of four monoclonal antibodies (each at 2 µg/mL), (PHAC NML Guelph) followed by alkaline phosphatase-labeled rabbit anti-mouse IgG (0.6 mg/mL) (Jackson Immunoresearch, Cedarlane Laboratories, Burlington, ON, Canada) and substrates nitroblue tetrazolium and 5-bromo-4-choloro-3-indolyl-phosphate (Sigma) with 30 min incubations (10 min for final development) and 10 min washes (PBS, 0.1% Tween 20). Verotoxin producing colonies on the HGMFs corresponding in location to purple spots on the VT-immunoblot were picked and inoculated into 800 µL of modified Tryptic Soy Broth supplemented with 1.5 g/L bile salts No. 3, 10 µg/mL vancomycin and 10 µg/mL cefsulodin (Sigma) (mTSBVC) and incubated at 37 • C for 18-24 h. The VT-ELISA was performed on duplicate 50 µL aliquots of the overnight broth. Briefly, aliquots of the broth were added to the wells for 30 min then probed sequentially with 50 µL of a mixture of four monoclonal antibodies (2 µg/mL) (PHAC NML Guelph), horseradish-peroxidase-labeled rabbit anti-mouse IgG (0.2 µg/mL) (Jackson Immunoresearch, Cedarlane Laboratories) with 30 min incubations and 5 × 300 µL washes (PBS, 0.1% Tween 20) following each addition using the Asys Atlantis 2 microplate washer (Biochrom, Cambridge, United Kingdom). Plates were developed with 50 µL of substrate tetramethylbenzidine (Sigma) for 10 min with slow agitation, followed by 50 µL of 0.2 M sulfuric acid. Absorbance readings were measured immediately using the µQuant microplate spectrophotometer (BioTek, Winooski, VT, United States) at OD450 nm and 600 nm and the final reading was calculated by subtracting the reading at 600 nm from the reading at 450 nm and averaging the duplicate values. Verotoxin production was scored as positive if OD readings were >2× the mean of the negative toxin control and negative if <1.5× that of the negative control. Suspect broths were streaked for single colonies to mTSAVC-BCIG and incubated at 37 • C for 18-24 h. Based on the VT-ELISA OD reading, 3 to 9 colonies of varying morphologies from each test sample were inoculated into mTSBVC and incubated at 37 • C for 18-24 h. A secondary VT-ELISA was performed as above and presumptive isolates were streaked to a MacConkey agar lawn for DNA extraction using the EZ1 DNA Tissue Kit (Qiagen, Hilden, Germany) and the EZ1 Advanced XL or the BioRobot EZ1 (Qiagen) and assayed for the presence of vt1 and vt2 using PCR (Gannon et al., 1997). DNA and 30% glycerol stocks were stored at -70 • C. Johnson et al. (2014) demonstrated the efficacy of the VT-IB without enrichment for environmental water samples. In this study, a subset of samples was used to compare the efficacy of testing unenriched and enriched cattle fecal samples. Briefly, for the enriched protocol, the 1:1 fecal slurry was inoculated into 15 mL of mTSBVC and incubated at 37 • C for 18-24 h. Overnight broths (2 × 50 µL) were screened by VT-ELISA and based on the OD readings, 100 µL of up to three 10-fold dilutions ranging from 10 −2 to 10 −7 were suspended in 10 mL of wash buffer and filtered as per the direct method. A total of 64 samples with at least one successful isolate from either method were used for method comparison. 10/64 samples were also selected for in-depth sampling for serotype diversity by picking at least five colonies per method. Enrichment data was used for method validation only and not carried forward for prevalence studies.

Comparison of VTEC Isolation From Enriched and Unenriched Fecal Samples
Detection of vt1, vt2, eae, hlyA, saa, and Serotyping PCR confirmation of vt1, vt2, eae, and hlyA (Paton and Paton, 1998), and O:H serotyping were completed by the Public Health Agency of Canada, National Microbiology Laboratory, at the E. coli Reference Laboratory in Guelph, Ontario. PCR screening for vt2 subtypes a, c, d was completed in-house as per (Scheutz et al., 2012) (Karmali et al., 2003). For serotypes not found in the original Karmali scheme, a modified SPT was used (EFSA, 2013) or otherwise assigned by the authors based on the literature, e.g., reported cases of HUS.

Whole Genome Sequencing and Analysis
One hundred and seventy-nine isolates spanning all serotypes were selected for whole genome sequencing. Genomic DNA was extracted from pure cultures using the MasterPure DNA Purification Kit (Epicentre, Madison, WI, United States) or the DNeasy Blood and Tissue Kits (Qiagen). 2 × 300 bp paired-end sequencing was performed on an Illumina MiSeq (Illumina, San Diego, CA, United States) using Nextera XT libraries and the MiSeq Reagent Kit v3 600 cycles (Illumina). Raw sequencing data has been deposited in the Sequence Read Archive (SRA) at the National Center for Biotechnology Information (NCBI) under BioProject PRJNA473796. Paired-end read data for 72 reference strains for serotypes isolated in this study were downloaded from the NCBI-SRA. A complete list of strains and corresponding meta data is available in Supplementary Table 10. De novo assembly was performed using the Shovill pipeline (v0.8.0) 1 which encompasses genome size estimation, FASTQ subsampling, adaptor trimming, read correction, merging of paired-end reads, genome assembly and polishing using Seqtk 2 , KMC 2 (Deorowicz et al., 2015), Trimmomatic (Bolger et al., 2014), Lighter (Song et al., 2014), FLASH (Magoc and Salzberg, 2011), SPAdes (Bankevich et al., 2012), SAMtools (Li et al., 2009), BWA MEM (Li, 2013), and Pilon (Walker et al., 2014).
In silico Clermont Phylotyping and Whole Genome Based Cluster Analysis DNA coding sequences for the arpA, chuA, yjaA, and TspEF.C2 genes (Clermont et al., 2013) were downloaded from NCBI GenBank and queried against a BLAST (v2.2.28) database generated from the 179 draft genomes. Phylogroup was assigned by scoring the presence/absence of the four markers (≥85% sequence identity, ≥60% query length). Cluster analysis was performed and visualized using the goeBURST Full MST algorithm in Phyloviz2 (Nascimento et al., 2017).

In silico Detection of Virulence-Associated and Antibiotic Resistance Genes
Draft assemblies for a subset (n = 115) of project and reference strains spanning all serotypes from this studyincluding, for each serotype, up to 6 strains from the study and up to 3 reference strains were interrogated for the presence of virulence-associated genes using Panseq (v.3.1.1) (Laing et al., 2010) with a curated list (n = 2,710) of virulence-associated gene sequences, including gene variants 6 , with the following parameters: 'percentIdentityCutoff ' = '85, ' 'coreGenomeThreshold' = '115.' Gene presence/absence was confirmed in ABRicate v0.7 7 using the VFDB database (2,597 sequences), with the following parameters: % coverage ≥60, % identity ≥80. The same dataset was further interrogated for the presence of antibiotic resistance genes in ABRicate v0.7 using the ResFinder database (2,228 sequences) and the CARD database (2,153 sequences), with the following parameters: % coverage ≥60, % identity ≥80. Gene presence/absence was confirmed using the RGI tool v4.0.3 (Jia et al., 2017) under "perfect" and "strict" match conditions. The presence/absence data was overlaid onto the core SNP phylogenetic tree and visualized using iTOL (Letunic and Bork, 2007).

Statistical Analysis
A subset of 336 unique strains from the 948 heifer isolates and 25 unique strains from the 79 calf isolates, representing one serotype-virulotype (vt1, vt2, hlyA, eae, and saa) representative for each fecal sample was carried forward for all analyses unless otherwise stated. Using the R statistical software, the pairwise.fisher.test () function from the {fmsb} package (x, n, p.adjust.methods = "bonferroni") (α = 0.05) was used for pairwise comparisons of VTEC prevalence. The oddsratio function from the {fmsb} package (a, b, c, d, conf.level = 0.95) was used to calculate the odds ratio and 95% confidence interval for the VT-immunoblot methods comparison. 95% confidence intervals for proportions were calculated using the Wilson procedure with a correction for continuity 9 . Microsoft Excel was used to calculate the standard deviation (STDEV.S) and confidence intervals (CONFIDENCE.T) for the average VTEC isolation rates and average Simpson's Index of Diversity (SID) values at the herd and cohort levels. SID, where, n = number of isolates of a particular serotype and N = total number of isolates; 0 = no diversity, 1 = infinite diversity, was used to compare serotype diversity, taking into account the number of serotypes present ("serotype richness") and the relative abundance ("serotype evenness"). For Figure 1, to account for variation in the number of isolates recovered per sample, the contribution of a specific serotype to the serotype distribution in an individual cow was calculated as follows: for each sample (Y), no. of isolates of serotype X no. of isolates from sample Y no. of samples tested from cow Z followed by normalization to the isolation rate for that cohort-year. n = 277, 283, 388 for years 1, 2, and 3, respectively. For Figure 2, statistics source data are as in Supplementary  Table 9.

VT-IB of Bacterial Growth From Unenriched Cattle Fecal HGMF Filtrates on mTSAVC-BCIG Media
VTEC were detected in 74.7% (275/368) and isolated from 70.4% (259/368) of cattle fecal slurry samples plated directly onto HGMF filters on mTSAVC-BCIG media; representing an isolate recovery rate for immunoblot positive samples of 94.2% (Table 1). Among a subset of 64 samples tested by both unenriched/direct and enriched methods -60.9% (39/64), 32.8% (21/64), 6.3% (4/64) were positive by both methods, unenriched only and enriched only, respectively. That is, 93.8% (60/64) of samples were VTEC positive by isolation without a selective broth culture enrichment step. VTEC was 7.3 times more likely to be isolated from immunoblot-positive samples using the direct method than the enrichment method ( Table 2). The number of serotypes isolated from unenriched and enriched cultures of these samples were 18 and 15, respectively; with 11 serotypes shared between the two methods (Supplementary Table 2). Further, in-depth sampling of suspect colonies (≥5 colonies per method) showed only one case where the enrichment method recovered a serotype which was not also recovered directly from unenriched samples (Supplementary Table 3).

Genomic Diversity
Cluster analysis by wgMLST and pangenome-derived core SNP analysis delineated strains at the phylotype and serotype level,  Karmali et al. (2003): Unassigned: not previously identified as VTEC or not fully typed (n = 52); SPT A: high incidence, common in outbreaks, associated with HC and HUS (n = 4); SPT B: moderate incidence, uncommon in outbreaks, associated with HC and HUS (n = 5); SPT C: low incidence, rare in outbreaks, associated with HC and HUS (n = 81); SPT D: low incidence, rare in outbreaks, not associated with HC and HUS (n = 56); SPT E: non-human only, not implicated in outbreaks, not associated with HC and HUS (n = 138).
including epidemiologically unrelated reference strains (Figures  4, 5). Within several major serotype clusters, including O6:H34, O28ac/O42:H25, O22:H8 and O91:H21, reference strains could be further distinguished from the herd isolates based on the number of SNP differences (Figure 5). Core SNP analysis identified 161335 SNPs from the pan-genome derived from all 251 strains, including 179 from the herd and 72 NCBI reference strains (Figure 5).

Seropathotype Distribution
All five seropathotypes (SPT) as defined by Karmali et al. (2003) were identified in this study:

In silico analysis of additional virulence factors
Genetic sequences for previously described virulence factors with roles in adherence/colonization, invasion, iron uptake, type III secretion systems and toxin production were identified in all serotypes recovered in this study. Carriage of virulence factors was largely serotype-specific within the strains isolated in this study, relative to reference strains. Of interest, strains from human illness-associated serotypes (O157:H7, O26:H11 and O111:NM) shared similar virulence profiles with two serotypes which are not typically associated with human illness (O182:H25, O84:H2). Specifically, genes coding for components found in the bacterial type III secretion system were also found in the latter group. Colonization factor gene cfab was found in all strains from all serotypes except O2:H6, O137:H41, O132:H18, and O157:H7. Alternative adherence factor genes iha and lpfA were found in 88.9% (64/72) and 93.1% (67/72) of herd strains, respectively (Figure 7).

DISCUSSION
Despite a general recognition of cattle as an important reservoir for VTEC, there are still knowledge gaps pertaining to the ecology of the organism, including their prevalence, distribution and persistence in a herd, as well as that of particular serotypes and  virulence factors. In this study, we tried to address some of these questions through year-long surveillance of each of three consecutive cohorts of cattle within a closed herd.

VT-Immunoblot Method for Isolation of VTEC From Unenriched Cattle Fecal Samples
The VT-immunoblot method (VT-IB) has been successfully used for the isolation of VTEC from enrichment cultures of ground beef (Atalla et al., 2000), surface water samples (Johnson et al., 2014;Nadya et al., 2016;Falardeau et al., 2017) and enriched cattle feces (Karama et al., 2008). This study represents the first use for unenriched cattle feces. Used in conjunction with the VT-ELISA, recovery rates were relatively high (i.e., 94.2% of samples with a positive VT-immunoblot signal yielded a VTEC isolate) in contrast to methods which yielded high detection but low isolation rates for VTEC (Pradel et al., 2000;Jenkins et al., 2002;Renter et al., 2007;Bosilevac and Koohmaraie, 2011;Monaghan et al., 2011;Paddock et al., 2012). A similar study investigating the long term prevalence of VTEC in cattle using stxA-specific colony hybridization found that stx-positive colonies were isolated from 60.5% of PCR positive pre-screening cultures (Geue et al., 2002). In this study, we have demonstrated for the majority of samples, that pre-enrichment was neither necessary for isolate recovery nor for capturing the serotype diversity within samples. In fact, VTEC was more likely to be isolated from positive samples using the direct method. The elimination of enrichment reduces the turnaround time from sample to isolate by at least 6-24 h and may help prevent the enrichment culture bias observed during the isolation of VTEC, which is in agreement with other studies involving the isolation of pathogenic E. coli (Conrad et al., 2016), Listeria monocytogenes and Salmonella spp (Gorski, 2012).
shedders" as not just individuals shedding at least 10 4 cfu per gram of feces as per the original definition (Chase-Topping et al., 2008) but also those that shed persistently over a period of time (Lim et al., 2007;Carlson et al., 2009;Menrath et al., 2010). Specifically, Lim et al. (2007) and Carlson et al. (2009) defined this as shedding for more than three consecutive months and Menrath et al. (2010) as having at least half of the samples and ≥4 consecutive sampling being stx positive. According to the latter definition and in the absence of direct quantification of the level of shedding, almost two-thirds of the individuals in this study could be classified as a "super shedder." Indeed, the phenomenon of super or persistent shedding in non-O157 VTEC has been reported elsewhere (Menrath et al., 2010;Murphy et al., 2016). However, statistically significant differences in the frequency of VTEC shedding were not observed between any two individuals. This may be due to the small sample sizes, which may have been ameliorated with more frequent sampling. Additionally, the "non-super shedders" were generally within 1-2 positive samples of satisfying these requirements, demonstrating the potential pitfalls of categorically assigning individuals to groups using criteria derived from studies which may differ in experimental methods and design. Most notably, in Menrath et al. (2010), VTEC prevalence was based on the PCR detection of stx in fecal samples, whereas it was based on the isolation of VTEC in this study. Nevertheless, the regular detection of VTEC in clinically healthy cattle herds is well known (Geue et al., 2002).

VTEC Diversity and Serotype Persistence
VTEC diversity was assessed by cluster analysis using multiple methods with varying levels of resolution -specifically, serotyping, phylotyping, wgMLST and pangenome-derived core SNP analysis. From 38 heifers, 948 isolates or 336 unique strains (i.e., those with unique sample-serotype-virulotypes profiles) were classified into 32 serotypes, yielding high serotype diversity (SID 0.87). In a similar study involving longitudinal sampling of a single herd, 45 different serotypes were identified from only 94 strains (Jenkins et al., 2002). However, in that study, more than 700 samples were collected from random sampling of an extensive grazing beef cattle farm with a high turnover rate. Similarly, a 20-week study of a large abattoir turning over more than 1,000 cattle a day recovered 24 serotypes from 27 VTEC isolates (Karama et al., 2008). In contrast, Bettelheim et al. (2005) sampled a cohort of 30 steers on two separate occasions to obtain a total of 474 E. coli isolates, classified into 52 serotypes. This suggests that repeat sampling of individual animals within a closely related herd with a low turnover rate may contribute to decreased serotype diversity, particularly as the rumen microbiota stabilizes as the animal matures (Jami et al., 2013). The predominance of phylotype B1 in this study is consistent with findings that suggest it is prevalent in ruminants (Carlos et al., 2010), particularly among cattle isolates (Coura et al., 2015;Askari Badouei et al., 2016).
Genomic diversity assessed at the whole genome level, using both wgMLST and SNP methods, showed clear delineation at the serotype level, although herd strains from this study were distinguishable from epidemiologically unrelated reference strains of the same serotype, based on the number of allele and SNP differences. In most cases, virulence factor and antimicrobial gene profiles were also conserved within each serotype, among the herd strains. This may suggest that each serotype is actually represented by a limited number of clones which have been transmitted both within and between cohorts. Serotype-specific cluster analyses may elucidate any phylogenetic substructures within each serotype, which could suggest an evolution of these strains within groups of cattle over time.
Serotypes O139:H19, O22:H8 and O?(O108):H8 were persistently isolated throughout the study. E. coli O22:H8 has previously been isolated from ground beef and chicken products (Mora et al., 2007;Alonso et al., 2016;Cadona et al., 2016), domestic cats (Bentancor et al., 2007), cattle (Blanco et al., 2004;Farah et al., 2007;Gonzalez et al., 2016) and HUS patients  and was also repeatedly isolated in another study of non-O157 VTEC in cattle (Menrath et al., 2010). A recent study by Martorelli et al. (2017) showed that E. coli O22:H8 is able to interfere with E. coli O157:H7 in vitro and in vivo based on their superior growth rates, biofilm-forming abilities and adherence properties and may reduce the overall susceptibility of calves to O157:H7 colonization and shedding. The presence of this organism could potentially explain the low percentage of E. coli O157:H7 isolations from this herd. Further investigation of the association between E. coli O22:H8 and E. coli O157:H7 in cattle herds is clearly warranted. E. coli O139 is typically associated with animals and mild disease in humans (Sandhu et al., 1996;Wang et al., 2005;EFSA, 2013). However, Toro et al. (2015) showed that an insertion sequence (IS) excision enhancer (IEE) shown to promote genomic rearrangements and strain diversification in EHEC O157:H7, was identical in sequence to the IEE found in O139:H38/NM strains. In the former study, the simultaneous presence of the IEE and IS629 was proposed to be a marker for strains with enhanced pathogenic potential.
In the 1950s, Sears et al. (1950Sears et al. ( , 1956 described "resident" E. coli serotypes that were observed over several months in an individual but that were vulnerable to replacement by new resident isolates over time in humans and dogs. In our study, the potential "residency" of certain serotypes was reflected not just between cohort-years but also at the individual animal level; in all but 3/38 cows surveyed, the most prevalent serotypes in each year, including O139:H19, O22:H8 and O?(O108:H8), contributed at least 50% of the VTEC isolates for each animal. Apparent population turnover was also observed. Serotype O130:H38 was highly prevalent in the first two years but absent in year 3. Conversely, serotype O91:H21 was rarely isolated early in the study but became a dominant serotype in the last year. Temporal variability and persistence of serotypes within a herd has been described previously (Tokhi et al., 1993;Beutin et al., 1997;Jenkins et al., 2002), but to our knowledge, this is the first study showing that non-O157 VTEC serotypes can persist through multiple cohorts within a distinct population of cattle. Geue et al. (2002) monitored the long term prevalence of VTEC in five groups of cattle among four farms from birth to slaughter, over three years; in their study specific VTEC clones of serotypes O26:H11 and O165:H25 were found to have persisted in herds. Taken together, this may suggest that persistence of serotypes or clones may occur in the same cohorts over multiple years (Geue et al., 2002) or across multiple cohorts, as in the current study.

Human Health Risk
Seropathotypes Seropathotype classifications, based on reported occurrences of outbreak and sporadic illness and incidence of HUS (Karmali et al., 2003), showed that while SPT A and SPT B VTEC were isolated in this study, the majority of isolates were of SPT D, E or otherwise unassigned. This may suggest that there is a low risk to humans from most VTEC isolated from these cattle. However, since non-O157 VTEC are not routinely isolated in many public health labs, it may also reflect a knowledge gap in our understanding of the contribution of other VTEC serotypes to human illness. The remainder, accounting for almost 25% of isolates, belonged to five serotypes classified as SPT C, which have been shown to be associated with HUS and may represent emerging pathogens.

Virulence Factors
Virulence profiles consisting of vt1, vt2, hlyA, eae, and saa were largely serotype-specific, suggesting the circulation of a single dominant clone within serotypes (Beutin et al., 1997;Jenkins et al., 2002). The most common vt type was vt1 vt2, comprising 68.5% (230/336) of isolates, which has been previously reported as the top vt type in surveys of dairy cattle (Irino et al., 2005;Gonzalez et al., 2016). In contrast, Mir et al. (2015) found 70.0% of all STEC encoded vt2 only, with vt1 vt2 being the least common genotype. Predominant vt types have been shown to vary between farms (Cobbold and Desmarchelier, 2001), differ between dairy and beef cattle (Cerqueira et al., 1999) and be serotype-specific in distinct cattle populations (Montenegro et al., 1990;Beutin et al., 1997); indeed, multiple vt types were only observed in serotype O22:H8. The vt2 subtypes a, c and/or d were identified in 98.6% of vt2 positive isolates -subtypes which have been associated with the development of haemorrhagic colitis or HUS (Orth et al., 2007;Kawano et al., 2008), as well as increased toxin potency in vitro and in vivo (Fuller et al., 2011). High levels of these subtypes in bovine isolates have been previously reported in Australia  and the US (Shridhar et al., 2017), and may represent a pool of transmissible virulence factors that could increase the pathogenicity of strains that acquire them. Additional studies should also be conducted to determine whether toxin-negative strain populations of the VTEC serotypes identified in this study co-exist with their toxin-positive counterparts; these could represent a reservoir of potential human pathogens, or alternatively could indicate that certain sub-populations are not likely to be associated with human disease.
In this study, 88.7% of VTEC carried the EHEC haemolysin gene (EHEC-hlyA), which is also known to be highly prevalent in human clinical VTEC (Beutin et al., 2004). A survey of cattle herds in Iran detected EHEC-hlyA in 11.9% (n = 452) of VTEC isolates (Askari Badouei et al., 2016) and in a longitudinal study of closed cattle herds in Germany, the prevalence of EHEC-hlyA was 26.5% (n = 1,647) (Geue et al., 2002). In our study, genetic diversity was also observed within the EHEC-hlyA gene itself. Notably, serotypes with known pathogenic potential and encoding eae, shared a variant that clustered distinctly from other serotype-specific variants. This supports previous findings where RFLP patterns for the EHEC-haemolysin gene were generally serotype-specific and also delineated by the presence/absence of eae (Boerlin et al., 1998;Askari Badouei et al., 2016). The identification of a number of other virulence-associated genes with roles in adherence, colonization, invasion, iron uptake and toxin production in the majority of strains suggests the potential to cause human disease exists amongst many of these commensal cattle isolates.
Intimin is an outer membrane adhesin encoded by the eae gene on the pathogenesis island called LEE (locus for enterocyte effacement). It mediates intimate attachment with host cells (McDaniel and Kaper, 1997) and is strongly associated with known pathogenic serotypes and severe disease in humans (Donnenberg et al., 1993;Boerlin et al., 1999;Blanco et al., 2004). Carriage of eae was observed in only 3.8% of isolates in this study and supports previous reports of low carriage of intimin in bovine isolates (Blanco et al., 1997;Geue et al., 2002;Irino et al., 2005;Farah et al., 2007;Karama et al., 2008;Menrath et al., 2010;Ennis et al., 2012) as well as the minor role played by intimin in the colonization of the bovine host by most VTEC (Ramachandran et al., 2003). Intimin may be essential for causing severe human illness by certain strains but may be replaced by alternatives, such as the saa-encoded STEC autoagglutinating adhesin, in other VTEC (de Azavedo et al., 1994;Paton et al., 1996;Paton et al., 2001). In this study, eae was absent from 100.0% of SPT C and 95.0% of SPT D isolates which are associated with sporadic but severe illness in humans and minor diarrheal illness, respectively. This further suggests alternative modes of colonization, including putative genetic determinants for host cell adherence iha (Tarr et al., 2000) and lpfA (Doughty et al., 2002), which were detected in 88.9% (64/72) and 93.1% (67/72) of isolates screened with the extended virulence factor panel in this study, respectively. Recent studies have shown that saa and eae may be mutually exclusive in single VTEC strains (Aidar-Ugrinovich et al., 2007;Farah et al., 2007;Ennis et al., 2012;Lee et al., 2017). In this study, saa was present in 82.7% of isolates (none of which also encoded eae), including several SPT C serotypes previously associated with severe human illness: O22:H8, O91:H21, O113:H21, O137:H41 and O2:H6. Similarly, all 13 intimin-positive isolates (O157:H7, O26:11 and O111:NM, O182:H25, O84:H2) were negative for saa.

Antibiotic Resistance
Owing to the dominance of lab-based antimicrobial susceptibility testing (van Belkum and Dunne, 2013) and relative novelty of whole-genome based predictions of antimicrobial resistance, it is difficult to assess the distribution of antibiotic resistance genes in this study in the context of similar studies, although high concordance between phenotypic and predicted susceptibility has been demonstrated in several bacterial species (Stoesser et al., 2013;Zankari et al., 2013;Tyson et al., 2015;Neuert et al., 2018). The detection of acquired antibiotic resistance genes in one strain (O113:H21, SPT C) in this study [tetracycline, tet(B); streptomycin/spectinomycin, aadA1; trimethoprim, dfrA1 and sulfonamide, sul2], suggests that cattle may contribute multidrug resistant VTEC of clinically associated serotypes albeit at low levels.

VTEC in Calves
Minimal interpretation of the calf VTEC data can be made as sampling occurred only once per year. However, the prevalence of VTEC falls within the wide range of rates previously observed in calves; from 20 to 95.6% (Tokhi et al., 1993;Shaw et al., 2004;Baltasar et al., 2014). Overall detection and isolation rates (78.6 and 71.4%) were similar to those observed in yearling heifers, which supports the findings of Baltasar et al. (2014) but are contrary to those of Mir et al. (2016), who found that VTEC colonization decreased with age and that this was correlated with lower gut microflora diversity in younger animals. Low serotype diversity and a high prevalence of SPT B isolates, specifically O26:NM and O111:NM were observed in calves. These serotypes have previously been associated with severe illness and occasional outbreaks (Karmali et al., 2003) and were rarely isolated from the heifers in this study. Previous studies have also reported a low concordance between dam and calf serotype populations Shaw et al., 2004), suggesting that the calf microbiome differs from that of the yearlings (Jami et al., 2013;Mir et al., 2016), and may allow different VTEC types to flourish. The early introduction and high prevalence of serotype O26 in calves have also been reported in Scotland and New Zealand Shaw et al., 2004;Jaros et al., 2016), with a possible decline in frequency as the animals age (Shaw et al., 2004). Baltasar et al. (2014) suggested that particular VTEC strains may have a propensity for the colonization of immature gastrointestinal systems. Direct sampling of dam-calf pairs would be required to elucidate the transmission dynamics of VTEC in calves.

CONCLUSION
In summary, we have demonstrated the efficacy of the VT-immunoblot method for the recovery of VTEC from unenriched cattle feces; VTEC was successfully recovered from 94.2% of samples with a positive immunoblot signal. VTEC prevalence and serotype diversity were relatively high among the yearling heifers and calves within this closed herd and patterns of multi-year persistence were observed for certain serotypes, particularly O139:H19, O?(O108):H8, and O22:H8. The majority of cattle could be described as "persistent VTEC shedders, " although significant differences in VTEC prevalence were not observed among animals. Statistically significant temporal trends were observed among several prevalent serotypes but not for overall VTEC prevalence. Intra-cohort contributions of individuals to serotype prevalence were not significantly different. The majority of isolates were classified as phylotype B1 and although all seropathotypes were identified, the majority of isolates belonged to SPT C (24.0%) or E (40.9%). Higher risk pathogenic serotypes known to cause illness did not dominate in the yearling heifers but based on limited sampling, may be prevalent in calves, with implications for cohort-specific mitigation efforts for VTEC transfer. Carriage of virulence factors vt1, vt2, eae, hlyA, and saa was largely serotype-specific; the majority of strains carried vt1 vt2 hlyA and saa, while eae was restricted to recognized pathogenic serotypes O157:H7, O26:H11, and O111:NM but also O84:H2 and O182:H25. Elevated prevalence of EHEC-hlyA in the predominately non-EHEC strain collection and the carriage of ETEC specific virulence factors supports the potential for intermediate or hybrid pathotypes and/or the exchange of virulence factors between potentially pathogenic E. coli. In silico assessment of an extended panel of virulence factors identified the presence of genes associated with adherence, colonization, invasion, iron uptake and toxin production. Acquired antibiotic resistance genes identified by ResFinder were observed in one strain belonging to a serotype of clinical significance (O113:H21). High levels of clonality within serotypes at the whole genome level, combined with the serotype-specificity of virulence and AR gene content, suggests the presence of single dominant clones that are transmitted within and among cohorts. This may indicate the existence of common farm-level sources of contamination that persist across years, that some VTEC occupy certain niches, and that the introduction of new members to distinct cattle populations is limited. Importantly, while strains generally clustered phylogenetically by serotype, differences at the gene and SNP level within each serotype enables the examination of a distinct VTEC population at two levels beyond serotype. We have demonstrated the utility of multi-year cohort sampling for investigating the extent of VTEC and serotype persistence and generated a relatively unique collection of isolates to support future analyses of the rate of genomic change over time, among different serotypes. Future work should also focus on the genetic factors which promote intra-herd persistence and exclusion of particular serotypes and the applicability of these results to the broader ecology of VTEC in cattle.

AUTHOR CONTRIBUTIONS
RJ developed the original VT-immunoblot assays used for screening environmental and food samples for VTEC and provided training and advice on experimental procedures and necessary reagents. KZ's laboratory carried out wet lab serotyping and toxin typing of all VTEC isolates and provided reference strains. CJ adapted the original VT-immunoblot methods for the Lethbridge laboratory, designed and established the sampling strategy and oversaw all aspects of the study prior to June 2013. LYRW and CJ contributed to the collection of samples, isolation of VTEC, and curation of the strain collection. LYRW contributed to the whole genome sequencing. LYRW and CL performed the analyses. LYRW wrote the first draft of the manuscript. All authors contributed to the conception and design of the study, manuscript revision, and read and approved the submitted version.