A laboratory-acquired E. coli O157:H7 infection occurred in 2013 in an individual who was working in the laboratory with a mixture of six E. coli O157:H7 strains in a soil-biochar blend. Symptoms included bloody diarrhea, hemolytic uremic syndrome (HUS), complete kidney failure, encephalopathy, respiratory failure, and retinal hemorrhages. After symptoms developed, the researcher was hospitalized. Pulsed-field gel electrophoresis (PFGE) analysis of the two isolates recovered from the patient’s stool in the hospital and the six nalidixic acid-resistant (Nal^R) mutant strains the researcher used in the laboratory experiment showed that the two patient isolates had indistinguishable PFGE patterns from laboratory strains E. coli O157:H7 strains RM7386 (7386) and RM6535 (6535). Strains used in this study, their characteristics, and sources are shown in Table 1 .

Strains were cultured in LB broth o/n at 37°C with shaking at 180 rpm. Total genomic DNA was extracted from the o/n culture using the QIAamp DNA Mini Kit (Qiagen, Inc., Valencia, CA, USA) for Illumina MiSeq and Nanopore MinION sequencing. For PacBio RS II sequencing, genomic DNA was extracted using the Genomic-tip 500/G kit (Qiagen). To close the genomes, we pursued a hybrid approach combining short-read Illumina MiSeq, long-read PacBio RS II, and Minion (Oxford Nanopore Technologies) technologies. For Illumina sequencing, a paired-end library was prepared for all strains using the NxSeq AmpFREE Low DNA Library Kit (Lucigen) and sequenced with 250-bp read length using the MiSeq Reagent Kit v2 500-cycle (Illumina) following the manufacturer’s guidelines. For MinION sequencing, genomic DNA of E. coli O157:H7 strain 7386 wt was diluted to a concentration of 1.5 µg genomic DNA in 46 µl nuclease-free water. The library was prepared using the Ligation Sequencing Kit 1D (SQK-LSK108) in combination with the Native Barcoding Kit (EXP-NBD103) according to the manufacturer’s instructions, and sequencing was performed on a MinION Mk1B. MinION reads were assembled with Canu version 1.1 (Koren et al., 2017). In addition, PacBio long-reads were generated at the University of Delaware DNA Sequencing and Genotyping Center, Delaware Biotechnology Institute in Newark, DE and at the Drexel University Genomics Core Facility in Philadelphia, PA. Genomic DNA was sheared into approximately 10-kb fragments using g-TUBE (Covaris, Inc., Woburn, MA, USA). The library was prepared based on the 10-kb PacBio sample preparation protocol and sequenced using P6/C4 chemistry on four single-molecule real-time (SMRT) cells with a 180-min collection time. The continuous long-read data were assembled de novo using the PacBio hierarchical genome assembly process (HGAP version 2.3.0) (Chin et al., 2013). Contigs were merged and circularized using Circlator (v 1.0.2). Assemblies were polished, and motifs were detected using RS_Modification_and_Motif_Analysis (v 2.3.0). The integrity of the resulting PacBio assemblies was evaluated by Canu (v 1.1) assembly of the MinION (Koren et al., 2017) reads for strain 7386WT and further by using Illumina short-reads in combination with the available Nanopore and/or PacBio long-reads with the hybrid assembler Unicycler (Wick et al., 2017) with Pilon error correction (Walker et al., 2014). Finally, molecules were rotated to the oriC (Luo and Gao, 2019) or repA genes for the chromosome and pO157 plasmid and annotated using the NCBI Prokaryotic Genome Annotation Pipeline (PGAP) (Tatusova et al., 2016).

Virulence and resistance genes were identified in silico with VirulenceFinder (Joensen et al., 2014; Kleinheinz et al., 2014; Joensen et al., 2015), VFDB (Chen et al., 2016) and Card (Alcock et al., 2020). To determine the Shiga toxin subtype, a single colony from each strain was selected from tryptic soy agar plates and grown in Luria Bertani medium overnight (o/n). DNA template was prepared by incubating 100 μl of the bacterial culture in 900 μl of sterile H₂O at 100°C for 10 min. PCR assays to identify stx subtypes were performed according to Scheutz et al. (2012) using a ProFlex PCR system (Thermo Fisher, Waltham, MA, USA) with slight modifications as indicated in Baranzoni et al. (2016). The assays targeted stx2a, 2b, 2c, 2d, 2e, 2f, and 2g. Gel electrophoresis using 1 µl of amplified DNA was performed using 2.0% UltraPure Agarose (Invitrogen, Carlsbad, CA, USA) with 0.5X GelRed (Phenix Research Products, Candler, NC, USA) in 1X Tris-acetate-EDTA buffer at 100 V for 1 h, and products were visualized using an AlphaImager gel documentation system (Alpha Innotech, San Leandro, CA, USA). The allelic subtype of the intimin (eae) was determined in silico according to Lacher et al. (2006) and Yang et al. (2020b).

The core genome SNP discovery pipeline is implemented on Galaxy (Goecks et al., 2010; Afgan et al., 2018), an open-source web-based bioinformatics platform. The SNP discovery strategy, detailed in Rusconi et al. (2016), allows to determine strain-to-strain variation and to establish phylogenetic relationships within the genomes of various microbial pathogens (Eppinger et al., 2010; Eppinger et al., 2011; Eppinger et al., 2014; Nicholson et al., 2020; Petro et al., 2020). The core genome is defined as the set of genic and intragenic regions that are not repeated, do not contain phages, IS elements, plasmid regions, genomic islands, or other mobile genetic elements, which evolve at different rates and are not indicative of evolutionary relationships. These regions were determined for the complete annotated reference E. coli O157:H7 strain EC4115 chromosome (CP001164.1) as follows: repeats with NUCmer (Delcher et al., 2003) by running the reference against itself to find repeated regions, prophages with PHASTER (Zhou et al., 2011; Arndt et al., 2016; Arndt et al., 2017), IS elements using ISEScan (Xie and Tang, 2017) in Galaxy (Afgan et al., 2018), and plasmids using PlasmidFinder (Carattoli et al., 2014). The SNP discovery and verification pipeline contains the following modules:

To confirm the in silico predicted SNPs linked to nalidixic acid resistance (Saenz et al., 2003; Fabrega et al., 2009), we performed Sanger amplicon sequencing of the DNA gyrase and topoisomerase IV genes in clinical strain pairs 7386Nal/M1300706001A and 6535Nal/M1300706002 (Genewiz). Primers were designed with Primer Express (Applied Biosystems) and are listed along with PCR cycling conditions in Supplementary Table 1 . Each reaction was performed in a volume of 20 μl using the Phusion High-Fidelity PCR Master Mix (Thermo Scientific) followed by PCR product purification with GeneJET PCR Purification Kit (Thermo Scientific), prior to Sanger sequencing (Genewiz). To confirm SNP alleles, sequencing results were compared to the corresponding SNP positions in the reference strain EC4115.

Expression of the stx2a gene in the laboratory and clinical strains was quantified using a RT-qPCR assay under non-induced, MMC, and antibiotic-induced conditions. A total of six strains were grown from an overnight culture in Luria-Bertani broth (LB, Thermo Fisher Scientific) at 37°C and 180 rpm to an OD₆₀₀ of 0.3-0.5. Cultures were induced either with MMC at 0.5 µg/ml (Allué-Guardia et al., 2014) or a combination of the two hospital administered antibiotics, MET and LEV, at concentrations of 25 µg/ml and 50 ng/ml. After cultures were incubated for 18 h at 37°C at 180 rpm, total RNA was extracted using the PureLink RNA Mini Kit (Thermo Fisher Scientific) and treated with DNAse I (DNAse I, amplification grade, Invitrogen) following the manufacturer’s instructions. The RNA quantity and integrity were assessed at an absorbance of 260 nm using a UV-spectrophotometer and by agarose gel electrophoresis. RNA was then converted into cDNA (RevertAid H minus first strand cDNA Synthesis kit (Thermo Scientific), and stx2a was quantified using the Go-Taq^® qPCR Master Mix (Promega, Madison, WI) and MicroAmp Fast optical real-time 96-well PCR plates (Applied Biosystems, Grand Island, NY) on an ABI StepOne PLUS Real-Time PCR system (Applied Biosystems). The primers and the cycling program used are described in Supplementary Table 1 (Wang et al., 2002; Gobert et al., 2007). Specificity was checked by analyzing the melting curves. Expression levels of stx2a were normalized against the endogenous gene tufA using the ABI StepOne PLUS System SDS software (Applied Biosystems) and are shown in relative quantity (RQ). Two biological replicates were conducted for each strain tested.

Statistical significance was determined by two-way analysis of variance ANOVA using Prism (version 9.0.1) (GraphPad Software, San Diego, CA) for comparison of Shiga toxin production and expression among the different culture conditions within each strain and among the different strains for each culture condition. Differences among groups were performed by Tukey’s multiple comparison test. Kruskal Wallis-non-parametric ANOVA and Dunn’s multiple comparison test were used for analysis of log CFU/ml results. Statistical significance was considered when p < 0.05. A confidence level of 95% was applied.

A researcher working with strains of E. coli O157:H7 developed sharp elbow pain with flu-like myalgias, followed by pain in multiple joints and eye pain two days after conducting an experiment with a six-strain cocktail of Nal^R E. coli O157:H7. His illness progressed to bloody diarrhea and severe abdominal pain that was alleviated with hydromorphone. Assessment following hospital admission was to refrain from antibiotic use until it was determined that infection was not due to STEC. Nonetheless, on the same day of admission, serologic testing for Campylobacter using the ImmunoCard STAT!^®CAMPY assay returned a positive result. The patient was then started on two cycles of LEV and MET on days 1 and 2, before laboratory tests indicated the presence of STEC. The antibiotics were stopped on day 3 and then recommenced on day 4 for a total treatment time of 16 days, and the patient was diagnosed with pancolitis.

The E. coli O157:H7 infection was reported to the PA State Health Department where stool samples were sent for confirmation. The patient’s clinical course continued to deteriorate, and one week after hospitalization, the patient developed HUS, followed by respiratory failure, encephalopathy with seizures, and retinal hemorrhages. After treatment with hemodialysis and plasmapheresis the condition gradually improved and the individual was discharged 24 days after admission. The researcher was working with a mixture of a total of six STEC strains in the laboratory. The two strains isolated from the patient had indistinguishable PFGE patterns to two laboratory strains 6535 and 7386 that were part of the STEC strain mixture, pointing to a likely source of the patient’s severe O157:H7 co-infection. These clinical isolates are linked to an outbreak caused by contaminated iceberg lettuce and bagged lettuce, respectively ( Table 1 ). Both strains carried the loci encoding the stx2a subtype and the eae-γ1 gene. The virulence profiles are shown in Supplementary Table 3 , neither strain encodes antibiotic resistance genes. Cases of laboratory-associated infections caused by E. coli O157:H7 have been reported (Booth and Rowe, 1993; Burnens et al., 1993; Rao et al., 1996; Coia, 1998; Salerno et al., 2004; Spina et al., 2005). The latter article describes four laboratory-associated cases that occurred in New York State from 1999 to 2004. The authors stated that standard laboratory biosafety practices had not been strictly followed, and in the four cases, the low infectious dose and ability of the pathogen to survive on surfaces for prolonged periods contributed to transmission. Regular assessment of laboratory safety protocols, adequate training, and ensuring compliance are important to avert exposure to potential hazards and for prevention of laboratory-acquired infections.

It has been reported that stressing E. coli O157:H7 with quinolone antibiotics such as LEV can induce a bacterial stress response due to perturbation of the DNA gyrase (topoisomerase IV) cascading to a Gram-negative “SOS response” (Nassar et al., 2013), leading to increased production and release of Shiga toxin 2 (Stx2) and escalating the likelihood of hemolytic uremic syndrome (Wong et al., 2000; Panos et al., 2006; McGannon et al., 2010; Smith et al., 2012; Wong et al., 2012). However, treatment with certain classes of antibiotics does not result in an increase in Stx production (McGannon et al., 2010; Bielaszewska et al., 2012). McGannon et al. (2010) reported that antibiotics that interfere with DNA synthesis, including ciprofloxacin and trimethoprim-sulfamethoxazole, increased Stx production, but other classes of antibiotics that target the cell wall, ribosome, or RNA polymerase did not. Antibiotics (LEV and MET) were administered to the patient described in this report due to a positive result with the stool antigen test for Campylobacter. LEV is a quinolone that interferes with DNA replication, and MET belongs to the nitroimidazole class and inhibits nucleic acid synthesis. The stool antigen test for Campylobacter used on the patient has been reported to have a positive predictive value (PPV) of only 36.6% (Fitzgerald et al., 2016). Later it was however determined that the patient was not infected with Campylobacter. This conclusion was based on a negative PCR result testing the patient’s stool for Campylobacter, thus the initial Campylobacter antigen test gave likely a false positive result. A similar incident occurred in June of 2013 in Virginia when a 69-year-old woman infected with STEC O111 was misdiagnosed with Campylobacter by the same immunoassay used on the patient in the current report (Operario et al., 2014). The woman was treated with azithromycin, ciprofloxacin, and metronidazole, and she subsequently developed hemolytic uremic syndrome. The authors stated that “Campylobacter was not confirmed by culture or PCR, suggesting that the initial Campylobacter EIA [enzyme immunoassay] was likely a false-positive result.”

Our phylogenomic analyses revealed how the analyzed laboratory- and patient strains fit into the phylogenomic context of the O157:H7 lineage. Numerous genomic epidemiology studies of STEC have embraced whole genome sequence typing (Franz et al., 2014; Sadiq et al., 2014; Eppinger and Cebula, 2015; Pightling et al., 2018), which proved critical for strain attribution and source identification (Bono, 2009; Eppinger et al., 2011; Holmes et al., 2015; Strachan et al., 2015; Yin et al., 2015; Cowley et al., 2016; Lupolova et al., 2016; Lee et al., 2017). To further investigate the source of infection, we sequenced a total of six genomes associated with this clinical case using NGS short- and long-read technologies on the Illumina, Oxford Nanopore, and PacBio platforms. The two isolate sets were comprised of wt outbreak strains 7386 (Washington lettuce, clade 8) and 6535 (Taco John, clade 2), the two derived Nal^R mutant strains the patient was exposed to in the laboratory, and further the two isolates recovered from the patient. To investigate the genetic relatedness of the wt and respective derived Nal^R mutants and clinical strains recovered from the patient, we reconstructed a phylogenomic hypotheses inferred from whole genome alignment ( Figure 1 ). We included representative isolates from the stepwise evolutionary model of O157:H7, which has been refined by many groups (Whittam et al., 1988; Wick et al., 2005; Feng et al., 2007; Leopold et al., 2009; Zhou et al., 2010; Jung et al., 2013). The tree topology partitions the isolates into nine distinct clusters, as previously established by Manning (Manning et al., 2008). As evident from the tree topology, the Taco John (6535) and Washington lettuce outbreak (7386) strains, along with their laboratory-derived Nal^R mutants and isolates recovered from the patient, cluster with the representative clades 2 and 8 strains PA11 (Hartzell et al., 2011) and EC4115 (Eppinger et al., 2011), respectively. It is in the nature of isolates linked to single outbreaks to form tight clonal clusters and consequently outbreak investigations require the application of high-resolution subtyping strategies (Eppinger et al., 2011; Rusconi et al., 2016). To increase resolution and resolve the intimate relationship of the strains, we applied high resolution SNP typing ( Figure 2 ). Core genome SNPs are highly informative in the context of outbreak investigation to differentiate “near clonal” outbreak isolates in support of strain attribution and outbreak ex- and inclusion (Eppinger et al., 2014; Jenkins et al., 2015; Rusconi et al., 2016). Whole genome SNP discovery and typing provided the necessary resolution to resolve the genetically homogenous population structure and allowed to differentiate near clonal laboratory wt and Nal^R mutant and clinical strains. The SNP analysis yielded a total of 588 SNPs, of which 572 were parsimony informative. Phylogenetic reconstruction based on the curated high-quality SNP panel with PAUP (Wilgenbusch and Swofford, 2003) identified two genetically distinct clusters separated by 551 SNPs, each comprised of the respective laboratory wt strains, 7386 and 6535, derived Nal^R mutants, and clinical isolates recovered from the patient. Strains within clusters 7386 and 6535 are near clonal and are either indistinguishable on the SNP-level or separated by one or two SNPs ( Figure 2 and Supplementary Table 4 ). The Nal^R mutants feature non-synonymous SNPs in both the DNA gyrase and topoisomerase genes, which are known mutations conferring this particular resistance phenotype (Saenz et al., 2003; Fabrega et al., 2009). The in silico predicted SNPs for these two genes were further confirmed by Sanger amplicon sequencing. In STEC O157:H7 we previously demonstrated that such numbers of SNPs can arise during a single passage in the laboratory (Eppinger et al., 2011). The recorded SNP numbers in the two clusters are further in line with the numbers reported for serial patient-derived O157:H7 isolates that underwent short-term microevolutionary changes (Rusconi et al., 2016). Taken together, the phylogenomics data identified the laboratory derived Nal^R mutant strains as the source of the researcher’s life-threatening disease rather than a conceivable ingestion of unrelated STEC O157:H7 isolates circulating at the same time. The results also clearly support a simultaneous co-infection with laboratory-housed strains, which were the causative agents of previous O157:H7 outbreaks. Co-infections have been only observed rarely in STEC (Rivas et al., 1993; Gilmour et al., 2007; Cheung et al., 2020), and the potential impact on disease manifestation by such mixed infections of strains featuring distinct genome and virulence traits has not been evaluated.

We recorded production traits of Stx2a, the most potent allelic subtype associated with human disease (Tesh et al., 1993; Orth et al., 2007; Fuller et al., 2011), in response to MET and LEV in wt strains, laboratory-derived Nal^R mutants, and clinical isolates. For a comprehensive readout, we recorded both toxin transcript and protein levels under spontaneous (non-induced) and MMC-inducing conditions ( Figure 3 ). Expression and production levels of the stx2a gene and Stx2 were measured by RT-qPCR ( Figures 3A, B ) and ELISA ( Figures 3C, D ), respectively. MMC is a potent Stx2 prophage-inducing agent that triggers toxin production via the SOS response mechanism (Raya and H’bert, 2009; Pacheco and Sperandio, 2012; Allué-Guardia et al., 2014). Bacterial counts (log CFU/ml) after 18 h of growth at 37°C in TSB medium alone were very similar among the six strains analyzed ( Table 2 ). Statistically significant reductions in CFU/ml, compared to TSB, were observed in the presence of MET in 7386 wt, 7386 Nal^R, 6535 wt, and patient isolates M1300706001A and M1300706002 ( Table 2 ). The addition of both MET and LEV to the cultures resulted in statistically significant reductions in patient-derived strain M1300706001A and in 7386 Nal^R compared to counts observed in TSB alone ( Table 2 ). LEV had no significant effect on the reduction of CFU/ml in any strain compared to TSB ( Table 2 ). Therefore, the reduction in bacterial growth observed in cultures with the combination of MET and LEV were primarily due to MET.

Several biotic and abiotic cues can induce the Stx converting prophage complement in STEC hosts, including antibiotics (Pacheco and Sperandio, 2012). Therapeutic use of antibiotics for STEC infections is thus controversial, as certain antibiotics are known to induce Stx2 phages and consequently toxin production through SOS-dependent activation of the phage lytic cycle (Kimmitt et al., 2000; Wong et al., 2000; Zhang et al., 2000; McGannon et al., 2010). Elevated toxin levels and stx2a transcripts were observed when compared to spontaneous conditions after exposure to MET and LEV ( Tables 3 , 4 ). Significantly increased production of Stx2 was observed in each of the six strains analyzed with combined MET and LEV compared to TSB (p<0.05), except for 7386 Nal^R, which showed high levels of toxin production in the presence of TSB alone. The highest Stx2 production with combined MET and LEV was observed in 6535 wt (average 4717 ng toxin/ml) with significantly higher Stx2 production compared to 7386 wt, 7386 Nal^R, and both patient isolates (p<0.05) ( Figure 3D ). Significantly higher production of Stx2 was observed in 6535 wt compared to the 6335 Nal^R patient isolate M1300706002 ( Table 3 ). The 6535 wt strain produced significantly less Stx2 under MET-induction when compared to the combination of the two antibiotics.

Stx2 production levels with exposure to MMC were higher than those induced by combined MET and LEV in the two Nal^R mutant strains and in the two patient isolates ( Table 3 ), but statistically significantly increased Stx2 production was only observed in 7386 Nal^R. Lower Stx2 production by MMC was observed in both wt strains, with significantly lower production in 6535 wt compared to exposure to LEV and the combination of MET and LEV.

Stx2 transcript levels were examined only in the presence of LEV and MET combined, and not with the antibiotics individually ( Figures 3A, B and Table 4 ). Results indicated that the induction effect of these antibiotics was not as pronounced as in samples treated with MMC. Results with MET+LEV treatments were not significantly different than treatment with TSB alone with the exception of strain 7386 wt. Another notable exception was that results with strain 7386 wt did not show a statistical difference between MMC and MET+LEV treatments.

When comparing stx2 expression versus Stx2 production results, a similar trend was observed in most strains. 7386 wt showed no significant differences in both expression and production of toxins in MMC versus MET+LEV, while the other strains showed significant differences in both toxin expression and production in MMC versus MET+LEV ( Tables 3 , 4 ). Host derived strains M1300706001A/7386-Nal^R and M1300706002/6535-Nal^R showed significant differences in stx2a gene expression under MMC versus MET+LEV conditions, but not in toxin production. This may indicate a delayed production of toxins in this strain; however, this would require further study. We note here a disconnection between transcript and Stx2 protein levels has been previously reported in O157:H7 strains, though the reason for this apparent lack of correlation remains to be elucidated (Leenanon et al., 2003; Neupane et al., 2011).