All the strains and plasmids used in the study are listed in Table 1. The O157:H7 strains with “PA” designations were from the Pennsylvania Department of Health collection and were characterized previously (Hartzell et al., 2011). The commensal E. coli strains were obtained from the E. coli Reference Center (ECRC) at The Pennsylvania State University. The bacteria were routinely grown in Lysogeny-Broth (LB) broth at 37°C, and their culture stocks were kept in 10% glycerol at −80°C. The modified LB broth and modified LB agar used for co-culture experiments were additionally supplemented with 10 mM CaCl₂. Working concentrations for antibiotics used in LB broth or agar were 100 μg/mL for ampicillin (Amp), 50 μg/mL for kanamycin (Kan), 30 μg/mL for nalixidic acid (Nal), 10 μg/mL for chloramphenicol (Cam), and 10 μg/ml for tetracycline (Tet). Spontaneous Nal resistant (Nal^R) mutants of C600 and 1.1954 were generated by spreading centrifuged cells harvested from 10 mL of overnight cultures onto LB agar plates with Nal at 37°C for 16 h. The Nal^R colonies were purified by re-streaking twice on similar media.

The co-culture assay was adapted from Gamage et al. (2003). Overnight cultures of PA2 or non-pathogenic E. coli strains were separately diluted in LB broth to an OD₆₀₀ of 0.05. One hundred and seventy microliters of each strain (OD₆₀₀ = 0.05) was mixed and added to modified LB broth to a final volume of 1,020 μL. The mixture was placed in a six-well plate (BD Biosciences Inc., Franklin Lakes, NJ). PA2 or non-pathogenic E. coli strains alone were used as controls. The six-well plates had 2 mL modified LB agar serving as the bottom base. Co-culture of C600 and O157:H7 was selected as the positive control (Gamage et al., 2003; Goswami et al., 2015). Stx2a level and cell density were determined after 16 h incubation at 37°C. Polymyxin B (PMB) was added to bacteria samples to final concentration of 6 mg/mL, and incubated at 37°C for 10 min for intracellular Stx2a release. PMB was used to ensure quantification of total Stx2a synthesized by bacteria (Shimizu et al., 2009; Laing et al., 2012; Ogura et al., 2015). After centrifuging at 8,000 × g for 2 min, the supernatants were collected for immediate usage or stored at −80°C for later Stx2a measurement. The Stx2a production was evaluated by a receptor based enzyme-linked immunosorbent assay (R-ELISA) as described below. Viable cell counts were calculated by spreading serial dilutions in phosphate buffer saline (PBS) onto Sorbitol MacConkey agar (SMAC). On this medium, non-O157:H7 and O157:H7 formed red and white colonies, respectively. Cell counts and toxin levels reported are the average from three biological replicates. The relative abundance was reported as percentage of commensal E. coli in the total population after co-culture, calculated by the equation:

For each R-ELISA run, supernatants from O157:H7 strain PA24 which produces only Stx1 was used as the negative control, while the lysate from high Stx2a-produing strain O157:H7 PA11 served as the positive control (Hartzell et al., 2011). The standard curves were generated using 2-fold serially diluted PA11 lysate or pure Stx2 (BEI resources, Manassas, VA). Any A₄₅₀ above 0.2 was considered positive. Total protein in each unknown sample was measured by the Bradford assay (VWR Life Science, Philadelphia, PA), following the manufacturer's recommended protocol. Stx2a quantities were reported as μg Stx2a/mg total protein.

The R-ELISA was performed as described previously (Goswami et al., 2015; Yin et al., 2015). Detachable 96-well polystyrene microtiter strip plates (Thermo Scientific, Waltham, MA) were coated with 2.5 μg per well of Gb3 analog, ceramide trihexoside (CTH), for Stx2a capture. The plate was stored at 4°C overnight with blocking buffer consisting of 4% bovine serum albumin (Sigma-Aldrich, St. Louis, MO) in 0.01 M PBS with 0.05% Tween20 (PBST). Samples were added in triplicate to wells and incubated for 1 h at room temperature (RT). Ten nanograms of monoclonal mouse anti-Stx2 (Santa Cruz Biotech, Santa Cruz CA) which specifically binds to the A subunit of Stx2 was added to each well and incubated at RT for 1 h. Then, 10 ng goat anti-mouse secondary antibody conjugated to horseradish peroxidase (Santa Cruz Biotech, Santa Cruz, CA) was added to each well and incubated at RT for 1 h. Detection was accomplished using the 1-Step Ultra Tetramethylbenzidine (TMB) (Thermo-Fischer, Waltham, MA), which was equilibrated to RT in a foil-wrapped tube for at least 30 min prior to use. Next, 100 μL TMB substrate was added into each well and incubated for 10 min to allow for color development. Finally, 100 μL of stop solution (2 M H₂SO₄) was added to each well. The reading values of A₄₅₀ were obtained using a DU®730 spectrophotometer (Beckman Coulter, Atlanta, GA). Between each addition of reagents above, the plate was washed with PBST for five times.

To generate bamA mutants in C600 or 1.1954, the approach from a previous study (Ruhe et al., 2013) was followed. The species source for bamA is indicated in the superscript and modifications of bamA loops are indicated in subscripts (Table 1). Target strains were first transformed with pZS21::bamA^E.coli(Kan^R). Next, the chromosomal bamA was deleted through one step recombination (Datsenko and Wanner, 2000) using the primer set of BamA-cam-For/Rev (Table 1). The transformants were selected on LB agar plates supplemented with Cam and Kan. Successful inactivation of chromosomal bamA was verified by PCR using primers BamA-VF/VR. The resulting E. coli ΔbamA::cam carrying plasmid pZS21::bamA^E.coli(Kan^R) was transformed with pZS21 (Amp^R) harboring the bamA^E.coli variants or bamA from other species (Enterobacter cloacae, Salmonella Typhimurium, Dickeya dadantii). Plasmid exchange was selected on LB agar supplemented with Amp. Amp^RKan^S colonies were chosen for later experiments.

The in-frame deletion of lamB or fadL in 1.1954 (Nal^R) was accomplished by marker exchange as previously reported (Chen et al., 2013). PCRs were designed using primer pairs LamB-UF/UR and LamB-DF/DR, which amplified 1,028 bp upstream and downstream of lamB, respectively. The two amplicions overlap by 28 bp including an XbaI site. About 20 ng of each PCR product and primers LamB-UF/DR were used in a second round of PCR. The final PCR product was digested with restriction enzyme XbaI, cloned into the suicide vector pDS132 (Philippe et al., 2004) and transformed into E. coli SM10λpir. Cam^R colonies were selected. Plasmid (pDS132::ΔlamB) was further transformed into E. coli S17λpir. Conjugation was conducted between E. coli S17λpir (pDS132::ΔlamB) and 1.1954 (Nal^R) as described before (Dudley et al., 2006). Transconjugants were selected on LB plates lacking NaCl, but supplemented with Cam, Nal and 5% (w/v) sucrose. Colonies were screened for Cam sensitivity and the correct deletion was confirmed by PCR using primers of LamB-VF/VR. The O157:H7 stx2 mutants were generated following the one step recombination method for enterohemorrhagic E. coli strains (Murphy and Campellone, 2003). Primers Stx2-tet-For/Rev were used to replace stx2 with a Tet cassette. The mutants were selected on corresponding LB agar plates and verified by PCR using primers Stx2-VF/VR.

An overnight culture of PA2 was diluted to an OD₆₀₀ of 0.05 in LB broth. Ciprofloxacin was added to a final concentration of 45 ng/ml to promote stx2a-converting phage induction. After 8 h, the culture was centrifuged at 4,000 × g for 10 min and the supernatant was filtered through a 0.22 μm cellulose acetate filter (VWR, Radnor, PA). Phage was precipitated by adding one fourth volume of 20% PEG-8000/2.5 M NaCl buffer followed by overnight incubation at 4°C. The lysate was centrifuged at 4,000 × g for 1 h, and serial dilutions of phage suspensions were made in SM buffer [0.1 M NaCl, 50 mM Tris-HCl (pH 7.5), 8 mM MgSO₄, and 0.01% gelatin]. Two-hundred liters of the indicator strain C600 was added to 100 μL of phage, and further mixed with 6 mL modified LB soft agar (0.75% agar). This was poured on top of a modified LB agar petri dish, and incubated at 42°C for 16 h followed by plaque quantification.

C600 (Nal^R) or 1.1954 (Nal^R) was co-cultured with individual O157:H7 stx2 Tet^R mutant. After 16 h incubation, a 10-fold diluted culture was spread on LB agar plates containing only Nal to enumerate the total number of non-pathogenic E. coli, or onto plates containing both Tet and Nal to select for the lysogens. The rate of lysogen formation was calculated by using the equation:

Both C600 and 1.1954 were whole genome sequenced on an Illumina MiSeq (San Diego, CA, USA). The Illumina reads were de novo assembled using SPAdes v3.9 (Bankevich et al., 2012) into contigs to identify potential insertion sites. Previously described primer pairs (Serra-Moreno et al., 2007) were used to locate insertion sites within the assembled genomes, as well as E. coli MG1655 (accession no. CP027060). Visual comparison of these regions in C600 and MG1655, which are known to lack prophage at these sites, with corresponding sequences from 1.1954, was used to assess site occupancy.

The CDI assay followed a previously described protocol (Aoki et al., 2005). Polyethylene terephthalate (PET) track-etched membrane inserts (23 mm) of 0.4 μm pore size (Falcon, Corning, NY) were placed in six-well plates to create upper and lower culture wells. Overnight cultures of PA2 and non-pathogenic E. coli strains were diluted to an OD₆₀₀ of 0.05. Diluted PA2 (3.2 mL) and non-pathogenic E. coli (2.5 mL) were added to the bottom and top chambers, respectively. Plates were incubated at 37°C with shaking at 130 rpm for 6 h. Both top and bottom samples were 10-fold serially diluted in PBS and 100 μL aliquots were plated onto SMAC plates to ensure no cross contamination occurred. After harvesting the cells and treating them with PMB for 5 min at 37°C, samples from the bottom chamber were centrifuged at 10,000 × g for 1 min, and supernatants were stored for immediate use or at −80°C. Stx2a levels were evaluated by R-ELISA.

MS Excel was used to calculate the mean, standard deviation, and standard error; Minitab 18 was used for statistical analysis and GraphPad Prism 8 was used for generating figures.

We began by testing a small collection of non-pathogenic E. coli, including the laboratory strain C600 and five commensal E. coli strains from various O types, for the ability to increase toxin production of O157:H7 strain PA2 when grown in co-culture. Co-culture of PA2+C600 produced the highest amount of Stx2a, reaching to 95.6 ± 8.1 μg Stx2a/mg total protein. Additionally, 1.1954 increased the Stx2a production in co-culture of PA2, producing 40.3 ± 1.3 μg Stx2a/mg total protein, which was significantly higher than the amount of Stx2a that PA2 produced in monoculture of 6.1 ± 0.8 μg Stx2a/mg total protein. The other four commensal E. coli strains, namely, 1.0322, 1.0326, 1.0328, and 1.1968 did not show significantly enhanced toxin production in co-cultures, when compared to PA2 alone (Figure 1A).

In each co-culture, both PA2 and non-pathogenic E. coli were inoculated at the same starting cell density. As reported previously (Goswami et al., 2015), after a 16 h co-culture, C600 abundance was 2.2 ± 0.9% of the total bacterial population, likely due to killing by the stx2a-converting phages produced by PA2. In contrast, an increase in cell counts of 1.1954 to 83.0 ± 2.3% was seen after co-culture with PA2 (Figure 1B). This suggested that the mechanism by which 1.1954 enhances Stx2a production by PA2 differs from that previously described for C600.

Several attempts to generate spontaneous phage resistant C600 derivatives were unsuccessful (data not shown), suggesting that disrupting phage adsorption requires changes to BamA beyond what can be achieved by those techniques. Using a previously published method (Ruhe et al., 2013), we generated three derivatives of C600 designated C600EC, C600ST, C600DD, in which a deletion of the chromosomal bamA was constructed through complementing in trans with plasmid-encoded bamA from E. cloacae, S. Typhimurium or D. dadantii, respectively. The deduced amino acid sequences of BamA from these species and E. coli are most divergent within the extracellular loops, which are the portions most likely involved in phage adsorption. All of these derivatives grew similarly to wild type C600 in LB broth (Figure 2A), and did not form any observable plaques when incubated with lysates containing the stx2a-converting phages (Figure 2B). Additionally, the concentrations of Stx2a produced during co-culture of these derivatives with PA2 were indistinguishable from that observed with PA2 alone, and significantly less than that measured in PA2+C600 (p < 0.05) (Figure 2C). Overall, expression of heterologous bamA in phage susceptible C600 rendered it resistant to phage lytic infection, providing further evidence that phage infection of C600 through BamA is required to enhance toxin production by PA2 in co-culture.

Using the tools developed by Ruhe et al. (2013), we could also address whether BamA loops 4, 6, and 7, which are the longest and least conserved of the extracellular loops, are needed for infection by stx2a-converting phages. Among the 12 variants we generated (Table 2), mutants with in-frame deletions in loop 4 or 6 (D4 and D6) as well as insertions in either loop 4 or 7 (I4 and I7) did not support the formation of detectable plaques by stx2a-converting phages (Figure 3A). Insertion of the HA epitope into loop 6 (I6) decreased plaque numbers by approximately 50% of that seen when C600 or EE was used as the host in the plaque assay. Strains expressing chimeric loops (C6, C7, C47) were also resistant to phage infection. The one exception was mutant C67, in which both loops 6 and 7 from bamA^E.cloacae were replaced with the corresponding sequences from bamA^E.coli. This restored susceptibility to phage infection, to approximately 25% of the number of plaques seen when either C600 or EE were used as host strains.

In accordance with these results, the co-cultures of PA2+C600 and PA2+EE produced similar levels of Stx2a, however expression of bamA^E.cloacae in place of bamA^E.coli (PA2+EC) decreased Stx2a expression to the baseline level (Figure 3B). An increase in Stx2a level was observed for PA2+C67 which had a chimeric bamA^E.cloacae with loop 6 and 7 replaced by those from bamA^E.coli, but this was not significantly different from the PA2 monoculture. The PA2+I6 combination produced half of Stx2a level in PA2+C600. Together, these results suggest that the three extracellular loops (4, 6, and 7) of bamA^E.coli are essential for optimal infection of C600 by stx2a-converting phages.

Since commensal 1.1954 was a Stx2a amplifier as well (Figure 1), we utilized the above approach to test whether bamA was necessary for 1.1954 mediated Stx2a amplification of PA2 in co-culture. Two bamA mutants were generated for 1.1954, in which the chromosomal bamA was deleted and complemented in trans by plasmid-encoded bamA from either E. coli (4EE) or S. Typhimurium (4ST). As expected, strains carrying bamA^E.coli (C600 or C600EE) produced significantly more Stx2a in co-cultures with PA2 than co-culture with the phage resistant strain C600ST (p < 0.05) (Figure 4). To the contrary, Stx2a concentrations in co-cultures of PA2+4ST and PA2+4EE were indistinguishable from that in PA2+1.1954, producing an average of 40 μg Stx2a/mg total protein (Figure 4). This suggests that commensal 1.1954 uses a bamA-independent mechanism for toxin enhancement in co-culture with PA2.

Attempts to test whether 1.1954 bamA^{S.Typhimurium} is phage-susceptible by standard plaque assays were unsuccessful, as 1.1954 does not form a bacterial lawn when grown on the antibiotic-containing medium (data not shown). Others have suggested LamB and FadL could serve as alternative receptors for stx-converting phages (Watarai et al., 1998). Therefore, we generated mutants of 1.1954 with in-frame deletion in fadL or lamB (4F, 4L), or in combination with either homologous (4FEE, 4LEE) or heterologous bamA (4FST, 4LST). In the absence of FadL or LamB or BamA, the single knockouts (4F, 4L, 4ST) produced statistically indistinguishable levels of Stx2a when compared to PA2+1.1954 (Figure 5). Similarly, the double knockouts (4FST, 4LST) lacking BamA plus either LamB or FadL, still exhibited the toxin amplification phenotype as wild type 1.1954. These results indicate that 1.1954 does not require BamA, FadL or LamB to enhance toxin production of O157:H7.

As an indirect measure of whether stx2a-converting phage infect the strain 1.1954, the PA2 mutant (PA2T) whose stx2 was replaced with a tetracycline resistance marker was used to monitor lysogenized rates when co-culturing with either C600 or 1.1954. The average lysogen forming rate in C600 was 0.016% (Figure 6), however, no lysogen formation was observed in 1.1954. We also monitored the rate for 1.1954 at different time points during the 16 h co-culture, and no lysogens were observed at any time point (data not shown). In order to test if phage type affected lysogen formation during co-culture, several other E. coli O157:H7 strains carrying genetically diverse stx2a-converting phages (Yin et al., 2015) were also tested. In the C600 background, SakaiT had the lowest average lysogen forming rate at <0.008%, while EDL933T gave the highest of 0.021%. No difference was observed for the lysogen forming rates among PA2T, PA8T, PA28T, and EDL933T (p < 0.05). However, no lysogens formed in the 1.1954 background by any tested stx2a-converting phage (data not shown). This suggested that 1.1954 does not undergo lysogenic conversion during co-culture with O157:H7.

It was reported earlier that if the primary phage insertion site in the host strain is occupied, the stx2-converting phages will integrate at alternative sites (Serra-Moreno et al., 2007). Five stx2a-converting phage insertion sites (sbcB, yehV, argW, yecE, and z2577) were checked for occupancy in both C600 and 1.1954, the first three of which are preferred by stx-converting phages. Four out of five were available in 1.1954, with only z2577 occupied. For C600, all five phage insertion sites were unoccupied. Collectively, these data suggest that stx2a-converting phages do not infect 1.1954.

To test whether secreted factors produced by 1.1954 could trigger toxin amplification of PA2, we used a modified CDI assay (Aoki et al., 2005), where non-pathogenic E. coli strains and PA2 were grown together while separated by a membrane. The E. coli strain ZK1526, which produces DNA gyrase inhibitor—microcin B17 (Genilloud et al., 1989), was selected as the positive control. As shown in Figure 7, ZK1526 promoted significantly more Stx2a production of PA2 than the negative control, PA2 alone. The toxin levels for PA2 in setups of PA2+C600 or PA2+1.1954 were as similar as the baseline level in PA2+LB. This indicated that neither C600 or 1.1954 secreted a soluble enhancer for Stx2a production of PA2.

With this result, physical contact between 1.1954 and PA2 seemed to be required for toxin amplification. Thus, we also considered a role for cdiA-encoding CDI systems (Aoki et al., 2005). Using BLAST, we found that 1.1954 harbors a typical cdiBAI operon while PA2 does not (data not shown). The deduced amino acid sequences of the carbon terminal (CT) of CdiA in 1.1954 shares 99% identity to that of uropathogenic E. coli (UPEC) 536, and the immunity protein CdiI shared 100% homology. Given the close relationship, we speculated the CdiA of 1.1954 is a tRNA anticodon nuclease as it is in UPEC 536 (Diner et al., 2012). If true, it seems unlikely that a tRNase is involved in increasing toxin expression (Toshima et al., 2007), and additionally this CDI system in UPEC 536 is repressed when grown in LB broth at 37°C (Aoki et al., 2010).

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