Because phages T6 and U115 both rely on the Tsx porin for cell-binding, we predicted that evolution of resistance to one phage should often lead to cross-resistance against the other phage. To test this idea, we used a classic fluctuation analysis (see “Materials and Methods”) to obtain a collection of spontaneous mutants of E. coli that were resistant to each phage individually. We used a paired approach, where each of 10 independently-grown bacterial cultures were used to isolate one T6-resistant and one U115-resistant strain (20 mutants total). Results for efficiency of plaquing (EOP) assays (see “Materials and Methods”) confirmed that growth of phage T6 on each of the 10 T6-resistant mutants (T6R1 through T6R10) was below the limit of detection, compared to normal infectivity of the phage on wild-type bacteria (Supplementary Table 1). Similarly, EOP experiments showed that phage U115 was unable to grow on each of the 10 U115-resistant mutants (U115R1 through U115R10), relative to expected infectivity on the wild-type (Supplementary Table 1). Furthermore, we found positive support for our hypothesis; in all 20 cases, evolved bacterial resistance to phage T6 provided cross-resistance to phage U115 using EOP assays, and vice versa (Supplementary Table 1). We concluded that independent selection for E. coli resistance to one phage provided cross-resistance to the other phage.

We then used antibiotic-resistance assays (see “Materials and Methods”) to test the prediction that evolution of phage resistance should often lead to cross-resistance to the antibiotic albicidin. We first used E. coli strains BW25113 and BW25113ΔicdC as two positive controls, to confirm that these albicidin-sensitive bacteria fail to form confluent lawns (i.e., they show zones of inhibited growth) when exposed to albicidin-producing X. albilineans strain XA23, but grow normally on X. albilineans strain LS126 which does not produce albicidin. Results (Figure 1) showed that the controls behaved as expected, with zones of growth inhibition around XA23 indicating sensitivity of both bacterial strains to albicidin. We estimated that wild-type BW25113 had a mean clearing ratio (a ratio of the zone of clearing divided by the area of the X. albilineans spot) of 4.88 ± 0.47 s.d. in the presence of XA23 albicidin-producing bacteria, and that BW25113ΔicdC bacteria showed a mean clearing ratio of 5.094 ± 0.524 s.d. (Supplementary Figure 1A). As a negative control, the Tsx knockout, BW25113Δtsx, had no zone of inhibited growth on either XA23 or LS126 (Figure 1); in both cases the mean clearing ratio of BW25113Δtsx was 1.0 ± 0.0 s.d. (Supplementary Figure 1A). A test of the hypothesis confirmed our prediction was correct; all 20 phage-resistant mutants showed no zones of growth inhibition around XA23 or LS126 (Figure 1 and Supplementary Figure 1B) and presented mean clearing ratios of 1.0 ± 0.0 s.d. on XA23 and LS126 (Supplementary Figure 1A), indicating that all the phage-resistant mutants were completely resistant to albicidin as well.

We used albicidin selection (see “Materials and Methods”) to isolate nine independent spontaneous mutants of E. coli (albR1 through albR9). Using the above-described growth challenges on X. albilineans strains XA23 and LS126, our results confirmed that each mutant was albicidin resistant (Figure 1 and Supplementary Figures 1A,B). We then tested whether the albicidin-resistant mutants were cross-resistant to infection with phage T6 and phage U115. Results showed that in all 9 strains tested, acquisition of albicidin resistance conferred cross-resistance to both phages T6 and U115 (Supplementary Table 1). We concluded that the evolution of cross-resistance was absolute in this study system, such that evolution of resistance to one of the three antibacterials provided perfect cross-resistance to all of the selective agents.

To further examine the evolution of antibacterial cross-resistance, bacterial growth assays (see “Materials and Methods”) were performed with replication (n = 3) in LB medium. Controls confirmed that wild-type E. coli strain BW25113 showed no discernable growth in the presence of either phage T6 or U115 (Figure 2A), and that growth of the tsx-knockout strain, BW25113Δtsx, was similar in the presence and absence of each phage (Figure 2B). In contrast, all 10 of the T6-resistant mutants grew normally in the presence and absence of phage T6 (representative data shown in Figures 2C,D; see Supplementary Figures 2B–I for all results). Similarly, growth of all 10 U115-resistant mutants was unaffected by presence or absence of phage U115 (representative data in Figures 2E,F; see Supplementary Figures 2J–Q for all results). Lastly, the 9 albicidin-resistant mutants were capable of approximately equivalent growth in the presence and absence of either phage T6 or U115 (representative data in Figures 2G,H; see Supplementary Figures 2R–X for all results).

A visual comparison of the growth-curve results suggested that all the resistant mutants grew similarly to wild-type strain BW25113 in the absence of each phage (Figure 2 and Supplementary Figure 2). To examine this outcome more closely, we analyzed the growth data with GrowthCurver (Sprouffske and Wagner, 2016) to estimate intrinsic growth rate (r) as a proxy for bacterial fitness (Figure 2I). None of the resistant mutants had an r that differed significantly from that of BW25113. These results suggested that there are no appreciable growth costs associated with bacterial evolution of resistance to phage T6, phage U115 and albicidin. While resistance to any of the antibacterials did not affect bacterial fitness under the tested conditions, we could not eliminate the possibility that our growth assays failed to detect minor fitness differences.

Thus, we conducted additional experiments, in an attempt to measure more subtle fitness differences among bacterial strains. To do so, we performed replicated (n = 3) competition assays (see “Materials and Methods”) for each test strain to gauge its fitness relative to a genetically-marked wild-type strain (BW25113ΔicdC) in the absence of phage and antibiotic selection. Competitive indexes (CIs) were calculated as the ratio of resistant-mutant colony forming units (CFU) to CFU of the common competitor strain, BW25113ΔicdC, to the ratio of BW25113ΔicdC CFU to wild-type, BW25113, CFU. Each ratio was normalized by the starting ratio of each competing strain. Results showed no significant differences in CIs of the resistant mutants as compared to the common competitor when competed in a resource-rich complex LB medium for 72 h (Figure 3). Furthermore, when competed in a resource-poor (minimal glucose) defined M9 medium for 24 h (Figure 3), all of the resistant mutants have CIs higher than the CI of BW25113ΔicdC with albR4 having the only statistically significant CI. Therefore, it appeared that all T6-resistant, U115-resistant and albicidin-resistant mutants suffered no growth deficits, compared to wild-type bacteria, in the absence of phage and antibiotic selection.

Owing to the importance of the Tsx porin in the interactions of all three antibacterials with cells, evolution of E. coli cross-resistance suggested that genetic changes in the tsx gene were likely involved. To examine this idea, we conducted whole genome sequencing (WGS) of all 29 strains. While some strains showed some single nucleotide variants (SNVs) and short insertions or deletions (indels) in tsx, many strains had no observable mutations in any genes using multiple variant calling pipelines including GATK (Van der Auwera et al., 2013) and breseq (Deatherage and Barrick, 2014). While WGS and variant calling pipelines are amenable for detecting SNVs and short indels, structural variants (SVs), including movement of transposable elements, are not easily detected (Ewing, 2015). Therefore, we conducted targeted Sanger sequencing of tsx in all 29 resistant mutants. As expected, results for the wild-type showed no mutations in tsx; however, mutations in tsx were identified in all 29 resistant mutants. For the 10 T6-resistant strains we observed the following mutations in the tsx gene: three insertion sequence (IS) elements, three deletions, three nonsense mutations and one missense mutation (Figure 4A and Table 1). For the 10 U115-resistant strains, we documented the following in tsx: four IS elements, two deletions, one insertion, two missense mutations, and two nonsense mutations (Figure 4A and Table 1). Interestingly, of the 10 pairs of phage-resistant mutants isolated from the same parent culture, only three pairs (T6R4/U115R4; T6R8/U115R8; T6R10/U115R10) showed identical mutations in tsx (Figure 4B and Table 1). For the nine albicidin-resistant mutants we observed six IS elements, two deletions, and one missense mutation in tsx (Figure 4A). Of the 29 mutations observed, three were in the promoter (tsxp2) region and four were in the signal sequence peptide of Tsx (Figure 4B and Table 1). In addition, our data showed that eight of the 29 mutations in tsx clustered within the region spanning base pairs 375–435 (Figure 4B).

The different missense mutations observed in this experiment were mapped (Figure 5) onto the crystal structure of Tsx (Ye and van den Berg, 2004) along with previously observed missense mutations (Figure 5 and Supplementary Table 2) that conferred resistance to either phage T6 or albicidin (Fsihi et al., 1993; Schneider et al., 1993). Mutant albR8 had a missense mutation R32P in the same surface exposed loop as three missense mutations (F27L, G28R, and G28E) observed by Fsihi et al. (1993) in mutants that were selected for albicidin resistance (Figure 5). Paired mutants T6R8 and U115R8 had the same missense mutation, L220R, in a beta strand that was in the region of a missense mutation (S217R) observed by Fsihi et al. (1993) in another mutant selected for resistance to albicidin (Figure 5). The third missense mutation observed in this study (V11E) in U115R5 is in the N-terminal signal peptide (Table 1).

While it is likely that these mutations are conferring resistance to phage T6, phage U115 and albicidin, this was not experimentally confirmed via recombinant genetics because the work fell outside the scope of the current study.

Strains in this study are listed in Supplementary Table 4. E. coli strains were cultured for 24 h with shaking incubation at 37°C in lysogeny broth (Luria and Delbruck, 1943). M9 minimal medium (6 g Na₂HPO₄, 3 g KH₂PO₄, 0.5 g NaCl, 1 g NH₄Cl, 1 mM MgSO₄, 0.1 mM CaCl₂, 0.1% (w/v) glucose per liter) was used for competition assays (see below). The Tsx knockout, BW25113Δtsx, was used as a negative control in many experiments. A pseudogene knockout, BW25113ΔicdC, was used as a control for the Kan^R cassette and as a marked competitor in the competition assays. Where appropriate, LB was supplemented with kanamycin at 30 μg/mL (LB Kan30). X. albilineans strains (provided by D. Gabriel, U Florida) were cultured for 48 h with shaking at 30°C in Modified Wilbrinks (MW) medium (Rott et al., 1994). Lysates of phages T6 and U115 were obtained by mixing each phage with E. coli strain BW25113 in LB medium and culturing for 24 h at 37°C, followed by filtration (0.22 μm) to remove bacteria.

Ten cultures of E. coli strain BW25113 were grown independently as described above, and diluted samples were spread on LB agar (1.5%) plates, pre-saturated with either phage T6 or phage U115. After overnight incubation at 37°C, one colony was chosen randomly from each plate and colony-purified three times. For each of the 20 isolated mutants, phage resistance was confirmed via efficiency of plaquing (EOP) assays, which compared plaquing ability of phage T6 or U115 on the test mutant relative to growth on wild-type strain BW25113.

X. albilineans strain XA23 was cultured as described above, and a 10 μL sample was spotted onto each of ten Sucrose Peptone agar (SPA) plates that were incubated for 5 days at 30°C (Birch and Patil, 1985). A sample from each of ten independently-grown cultures of BW25113 was then overlaid on one of the SPA plates, and incubated overnight at 37°C. Colonies that appeared in the zone of growth inhibition around the XA23 spot represented spontaneous E. coli mutants that were resistant to albicidin. One plate was lost due to contamination; from each of the remaining nine plates, one colony was randomly chosen and colony-purified three times. Albicidin resistance of each mutant was verified by plating a sample of a grown-up culture of the isolate on an XA23 spot, as described above.

Xanthomonas albilineans strains XA23 and LS126 were each cultured for 48 h at 30°C, and their optical densities at wavelength λ = 600 nm (OD₆₀₀) were estimated via spectrophotometry. Each culture was then diluted in MW medium, to obtain OD₆₀₀ = 0.25. Then, a 10-μL sample of each diluted culture was spotted on a SPA plate, allowed to dry completely and then incubated for 5 days at 30°C. SPA plates were overlaid with 4 mL of LB top agar (0.75%) with 1 mL of a test E. coli strain at OD₆₀₀ = 0.25, and incubated overnight at 37°C. The next day, plates were imaged. Images were quantified using ImageJ, by thresholding until either the X. albilineans spots or E. coli clearings were outlined and the area within the outline was quantified. The ratio of the area of the zone of clearing to the area of the X. albilineans spot was used to define albicidin sensitivity: a ratio of greater than 1.0 indicates albicidin sensitivity, while a ratio of 1.0 indicates albicidin resistance.

Minimum inhibitory concentrations (MICs) of ampicillin and ciprofloxacin were determined using the twofold dilution method (MacLowry et al., 1970).

E. coli bacteria were cultured overnight as described above. Dilutions of a test strain in LB medium were placed in wells of a flat-bottomed 96-well plate, in the absence of phage or mixed with phage at a multiplicity-of-infection (MOI; ratio of phage particles to host cells) of ∼10. Plates were incubated with shaking for 18 h in an automated spectrophotometer at 37°C, and OD₆₀₀ measurements were obtained every 10 min. Growth curve data were fit to a logarithmic curve using GrowthCurver and growth parameters were extracted (Sprouffske and Wagner, 2016). All growth curve data are deposited on Dryad.

Replicated (n = 3) competition assays were conducted in both LB and M9 media by mixing two bacterial strains at a 1:1 initial ratio. Competition assays in LB were serially passaged (1:100 dilution) every 24 h for 72 h total. Competition assays in M9 were measured after 24 h. At the end of each experiment, a diluted sample of each competition was plated on LB agar to measure viable bacterial density (colony-forming units; CFU), and on LB Kan30 agar to estimate the CFU of a Kan^R competitor; the density of the Kan^S competitor was estimated by subtraction. Competitive indexes were calculated as the ratio of resistant-mutant CFU to the CFU of BW25113ΔicdC divided by the ratio of BW25113ΔicdC CFU to BW25113 CFU. Each ratio was normalized by the starting ratio of each competing strain.

Sequencing libraries were prepared using Illumina’s Nextera prep kit. Whole genome sequencing was done using paired-end 150 bp reads on the Illumina NextSeq platform or paired-end 300bp reads on the Illumina HiSeq platform. For targeted sequencing, primers (5′-CTGTGAAACGAAACATATTTTTG-3′ and 5′-CGTGCTTTTGTTGGC-3′) were designed ∼100 base pairs upstream and downstream of E. coli gene tsx, and used to amplify tsx of each resistant mutant and the ancestral BW25113 strain. Amplicons were Sanger sequenced at the Yale DNA Analysis Facility on Science Hill. Mutations were identified by comparing sequencing reads to tsx loci of the reference strain GenBank CP009273.1.