Wild-type ST131-H30R challenge strains for the mouse models were selected from ST131 subclones H30R1 (strains MVAST36 and MVAST392) and H30Rx (strains MVAST412, JJ2050, and JJ1886; Tables 1, 2). These strains, which exhibited O:H type O25b:H4, were selected in part based on the identification of bacteriophages active against them, as described below (Table 2), and for phylogenomic diversity within the H30R lineage (Price et al., 2013). Control strains for the murine sepsis model included UTI isolate CFT073 (positive control; Mobley et al., 1990a; Welch et al., 2002) and laboratory strain MG1655 (negative control; Welch et al., 2002).

For the intestinal colonization model, spontaneous streptomycin-resistant mutants of parent strains JJ1886 (ST131-H30Rx, which caused fatal urosepsis in the source patient; Owens et al., 2011) and EcN were selected by plating dense bacterial suspensions on streptomycin-containing agar (30 mg/L; Table 1). Colonies from these plates were re-purified, confirmed molecularly as corresponding clonally with the parent (Berg et al., 1994), and designated as JJ1886-S and EcN-S. A spontaneous rifampin-resistant mutant of EcN-S was derived similarly by using rifampin-containing agar (50 mg/L), confirmed molecularly as corresponding with the parent, and designated EcN-SR.

A preliminary screen of microcins C7, J25, L, N, and V for activity against ST131-H30R strains showed that microcin C7 was most broadly and potently active (not shown). Consequently, to create a microcin C7-producing derivative of EcN-SR we transformed EcN-SR with plasmid pp70 (generously provided by Dr. Konstantin Severinov, Rutger’s Wacksman Institute; Zukher et al., 2014). pp70 contains a microcin C7 expression/secretion/immunity cassette in a pBR322 backbone, and encodes ampicillin resistance (Table 1). Filter-sterilized supernatants of the transformant, which was designated EcN-SR/pp70, were confirmed as exhibiting potent in vitro inhibition (typically, by 5–6 log₁₀ CFU) of diverse ST131-H30R strains when combined with the target strains during log-phase growth, as described in a subsequent section.

A panel of diverse E. coli target strains (Tables 1, 2) was used to assess the spectrum of activity of presumptive bacteriophage-containing solutions, which was done as described below. These target strains included 12 ST131 strains [six H30R1 and six H30Rx; four of these were ESBL-positive (2 CTX-M-15, 2 CTX-M non-15)], plus representatives for three other STs within group B2, and one representative each for phylogroups A, B1, C, D, E, and F (Table 1).

To generate solutions containing microcin C7, EcN-SR/pp70 was incubated overnight at 37°C in static, non-antibiotic-supplemented LB broth, which was then centrifuged and filter-sterilized (0.45 micron filter). EcN-SR without pp70 was used in parallel as a no-microcin control. For activity testing, a 1:1 mixture of the sterilized supernatant and fresh LB broth was distributed into 24-well tray (900 μl per well). Separately, relevant target strains (e.g., the JJ1886-S control, or JJ1886-S isolates from mouse fecal samples, as described below) were grown at 37°C in LB broth to exponential phase. After turbidity adjustment of the broths to McFarland 1.0 (~10⁷ cfu/ml), 10X serial dilutions in PBS were used to inoculate microcin-containing trays (100 μl per well). After overnight incubation at 37C, wells were scored for growth/no growth. The limiting bacterial dilution that yielded growth was compared between test and control target strains (to assess different test strains’ susceptibility to microcin C7), or between microcin-containing and control wells (to assess the microcin-producing ability of different EcN-SR/pp70 isolates).

Bacteriophages with activity against ST131-H30Rx were recovered from influent sewage newly obtained from municipal sewage facilities in St. Paul, MN, and Seattle, WA, United States. Raw sewage was centrifuged (2 min, 3,000 rpm) to pellet solids. Portions of the supernatant were filter-sterilized (0.45 micron filter), then spotted (10 μl) in 10x serial dilutions in PBS onto a series of Mueller-Hinton (MH) agar plates that had been overlain with 3 ml soft agar (7.5 gm/L) containing 10⁶ CFU/ml of a target E. coli strain (ST131-H30R: JJ1886, MVAST36, JJ2050, JJ228, and CU758; non-ST131: MG1655; Table 1).

After overnight incubation at 37°C, all plates showed complete lysis of the bacterial lawn within the supernatant-spotted areas out to the 10⁻² dilution; hazy growth or isolated breakthrough colonies at intermediate dilutions; and little or no lysis by the 10⁻⁷ and higher dilutions. From each plate containing an ST131-H30R target strain, the agar beneath the undiluted supernatant plaque was excised and incubated overnight with the cognate target strain at 37°C in Luria-Bertani (LB) broth. The following day the broth was centrifuged, filter-sterilized, and spread in serial 10x dilutions in PBS onto a series of MH or LB agar plates that had been overlain with soft agar containing 10⁶ CFU/ml of the cognate target strain.

After incubation overnight at 37°C, for each target strain, plaques were harvested (as described above) from the plate with the best-separated plaques, giving priority to different-appearing plaques, which were then processed in the same manner as the initial (sewage-supernatant-derived) plaques. Three rounds of such plaque purification, each involving serial dilutions from a single plaque, were done to obtain presumably pure bacteriophage preparations.

Bacteriophage preparations were assessed for spectrum of activity by spotting them individually onto a series of MH agar plates, each overlain with a different target E. coli strain (Table 2). Bacteriophages were presumed to be distinct if they exhibited a unique spectrum of activity or were from different sources.

For combined use in the mouse experiments, five distinct bacteriophages were selected based on potency against ST131-H30R, number of ST131-H30R target strains lysed, and lack of activity against non-ST131-H30R strains, especially EcN. Individually, the selected bacteriophages exhibited 3+ activity against from 1 to 4 (median, three) of the five target strains to be used in the sepsis model; conversely, each of the five target strains was lysed (3+ level) by from 1 to 4 (median, three) of the five selected bacteriophages (Table 2). Accordingly, the resulting pentavalent cocktail exhibited 3+ activity against each of the five mouse-model target strains. The five individual bacteriophages and the pentavalent cocktail were similarly active against JJ1886-S (used in the gut model) as against the JJ1886 parent (Table 2).

For administration to mice, preparations of individual phages were generated by inoculating appropriate volumes of LB broth (usually, 20 ml) with a purified bacteriophage preparation (usually, 50 μl) and the cognate ST131-H30R strain (Table 2), incubating this mixture overnight with shaking at 37°C, pelleting the solids, and filter-sterilizing the supernatant. Equal volumes of each individual phage preparation were combined to give the final five-phage cocktail. Phage titer was checked by spotting serial 10X dilutions of this preparation onto target-strain-overlain LB top agar plates, as described above.

The MVAHCS Institutional Animal Care and Use Committee (IACUC) approved the animal experimentation protocols.

Using an established murine subcutaneous sepsis model (Merino et al., 2020), female Swiss Webster mice (Envigo ND4; 7 weeks old; mean weight 23 g; and 10 mice for each of the 10 study arms) were inoculated subcutaneously with 100 μl of a suspension of one of the five ST131-H30R challenge strains (total dose: 1.3–2.0 × 10⁸ CFU; log-phase growth), which immediately (≤30 min) before inoculation was combined with 100 μl of either PBS (controls) or the bacteriophage cocktail [10⁷–10⁹ plaque-forming units (PFU)/ml; total dose, 10⁶–10⁸ PFU], as prepared within 48 h of use and stored at 4°C. In parallel, five mice each were challenged with reference strains CFT073 (positive control: lethal to four or five of five mice) and MG1655 (negative control: no observable mouse illness; Table 1), and with the bacteriophage cocktail alone (negative control: no observable mouse illness whatsoever). To minimize potential artifacts from cohort or temporal effects, mice from a given shipment were allocated randomly to the different treatment arms, which were run in parallel within each experiment.

Following inoculation on d0, mice were assessed for illness severity daily for 3 days (AM and PM of d + 1 and d + 2; AM of d + 3), after which any surviving mice were euthanized. An experienced observer rated illness severity on a five-point scale (0, healthy; 1, mildly ill; 2, moderately ill; 3, severely ill; and 4, dead) according to standardized criteria (Merino-Velasco et al., 2017). Mice that reached stage 3 were euthanized and scored as stage 4 (dead) for any subsequent time points. Severity scores for a given mouse were averaged across the five time points. To confirm specificity, post-mortem spleen cultures were done for the two sickest mice per challenge strain. PCR-based genomic profiles (Berg et al., 1994) of E. coli colonies from spleen cultures uniformly corresponded with the inoculated strain (not shown).

An established streptomycin-treated mouse intestinal colonization model was used (Lescat et al., 2017). Female Swiss-Webster mice (Envigo ND4; 5–10 weeks old; mean weight 23.5 g; and 16 per study arm) were housed individually to avoid cross-contamination. Mice, randomly assigned within each mouse shipment to four parallel treatment groups (n = 16 each; divided as 4 per week × 4 weeks), began continuous streptomycin treatment via the drinking water (5 g/L) at d−3 before bacterial challenge, to eliminate endogenous aerobic gut bacteria, as confirmed by fecal cultures. They then underwent oral gavage with various test substances on d0, d + 3, and + d5.

Specifically, to establish colonization with JJ1886-S, on d0 all mice received by gavage H30Rx target strain JJ1886-S (mean dose, 2.3 × 10⁸ CFU). As potential decolonization interventions, on all three gavage days (d0, d + 3, and +d5) mice received by gavage one of four treatments: (i) PBS (negative control), (ii) bacteriophage alone (dose, 10⁶–10⁸ PFU), (iii) EcN-SR/pp70 alone (mean dose, 2.4 × 10⁸ CFU), or (iv) bacteriophage plus EcN-SR/pp70 (each dosed as when given alone). On d0, the decolonization intervention dose was co-administered with target strain JJ1886-S, immediately (≤ 30 min) after these suspensions were combined. Additionally, for continuous probiotic exposure, for mice assigned to receive EcN-SR/pp70 either alone or with bacteriophage, EcN-SR/pp70 was added to the drinking water (final concentration, 10⁸ CFU/ml), which was replenished on d0, d + 3, and + d5 (In pilot experiments, after 48 h in ambient-temperature water EcN-SR/pp70 exhibited 80% viability and remained ampicillin-resistant).

Fresh fecal pellets were collected from mice on days −3, 0, +1, +4, +7, and +10 by moving mice to a clean, solid-bottom cage until they defecated. Pellets were weighed and dispersed in PBS, which was then cultured quantitatively on agar plates supplemented, separately, with streptomycin (30 mg/L; selects for both JJ1886-S and EcN-SR/pp70), ciprofloxacin (4 mg/L; selects for JJ1886-S), or rifampin (50 mg/L; selects for EcN-SR/pp70). If the estimated concentrations of JJ1886-S and EcN-SR/pp70 differed by <10X, for greater precision 20 colonies from streptomycin-supplemented plates were replica-plated to both ciprofloxacin-supplemented and rifampin-supplemented plates, and the observed ratio or its inverse was applied to the total cell count from streptomycin-supplemented plates. Additionally, selected colonies of putative EcN-SR/pp70 were replica-plated to ampicillin-containing plates (50 mg/L) to confirm plasmid retention. For selected isolates from each study group, strain identity was confirmed by PCR-based genomic profiling.

Additional experiments were done to clarify selected aspects of the intestinal colonization model results. First, to confirm the presence and functionality of bacteriophage in the mouse gut, filter-sterilized supernatants from homogenized fecal pellets from bacteriophage-treated and concurrent control mice were tested (as described above) for lytic activity against the input JJ1886-R strain. Second, to confirm sustained microcin production by EcN-SR/pp70 after residence in the mouse gut, supernatants from broth-grown fecal isolates of EcN-SR/pp70 from mice treated with EcN-SR/pp70 were tested for inhibitory activity against the input JJ1886-S strain. Third, to assess for possible development of bacteriophage and/or microcin C7 resistance by JJ1886-S during gut residence, fecal isolates of JJ1886-S from bacteriophage-treated mice were tested (as described above) for susceptibility to the input bacteriophage cocktail; likewise, fecal isolates of JJ1886-S from EcN-SR/pp70-treated mice were tested (as described above) for inhibition by supernatants from the broth-grown input EcN-SR/pp70 strain.

For bacterial counts, between-group comparisons were tested using unpaired t-tests, assuming equal variance, and within-mouse comparisons between different strains were tested using paired t-tests. For mortality, between-group comparisons of proportions were tested using Fisher’s exact test. All p values were two-tailed. The significance criterion was p < 0.05.

To test whether the bacteriophage cocktail is active in vivo against various ST131-H30R strains, we used the murine subcutaneous sepsis model and five ST131-H30R test strains (two H30R1, three H30Rx). The bacteriophage cocktail by itself caused no sign of illness in control mice, which remained as healthy throughout as did the PBS controls (n = 5). When inoculated without bacteriophage, the test bacteria caused distinct patterns of disease, ranging from transient mild-to-moderate illness by H30R1 strains (MVAST36 and MVAST392), through sustained more severe illness (MVAST412, JJ2050) or rapidly progressive lethal sepsis (JJ1886) by the H30Rx bacteria (Figure 1). Results for a given strain were highly consistent across the corresponding 10 mice (not shown). For each strain the addition of bacteriophage cocktail to the inoculum immediately before inoculation significantly attenuated disease severity, as assessed for all five strains by the average illness severity score (Figures 1, 2) and, for strain JJ1886, by percent dead (0/10 with phage, vs. 10/10 without phage: p < 0.001).

In streptomycin-fed mice, JJ1886-S, when administered by gavage on d0, established sustained gut colonization, with mean fecal counts ranging from 2.6 × 10¹⁰ CFU/gm (d + 1) to 8.3 × 10⁹ CFU/gm (d + 10; Figure 3). Gavage administration of bacteriophage cocktail on d0 (co-administered with JJ1886-S), d + 3, and d + 5 reduced mean fecal counts of JJ1886-S significantly on d + 1 (by 0.5 log₁₀: p < 0.001); numerically but not statistically significantly on d + 4 and d + 7; and negligibly by d + 10. Administration of EcN-SR/pp70 by gavage on d0, d + 3, and d + 5, plus continuously in the drinking water, had no discernible effect on fecal counts of JJ1886-S at any time point (Figure 3), despite EcN-SR/pp70’s confirmed abundance in the gut (Figure 4).

By contrast, administration of the bacteriophage cocktail together with EcN-SR/pp70 yielded a marked synergistic effect (Figure 3). Specifically, fecal counts of JJ1886-S were reduced by 3.3 log₁₀ on d + 1 (p < 0.001), 1.0 log₁₀ on d + 4 (p = 0.002), and 0.5 log₁₀ on d + 7 (p = 0.02). However, by d + 10 the JJ1886-S counts were comparable to control.

To clarify the basis for the above findings from the gut model, additional experiments were done. Regarding the greater effect of combination therapy than of probiotic strain EcN-SR/pp70 alone, we compared over time, for mice that received EcN-SR/pp70 either alone or combined with bacteriophage cocktail, absolute fecal counts of JJ1886-S and EcN-SR/pp70, stratified by use of bacteriophage cocktail (Figure 4). Whereas in the absence of bacteriophage cocktail JJ1886-S significantly outnumbered EcN-SR/pp70 on most days (by approximately one log₁₀), with the addition of bacteriophage cocktail this ratio was reversed initially, with a significant excess of EcN-SR/pp70 on d + 1 (by 2.8 log₁₀: p < 0.001) and d + 4 (by 0.7 log₁₀: p = 0.001), whereas by d + 7 the two strains were approximately equally abundant, and by d + 10 JJ1886-S had become numerically predominant.

Regarding the time-limited effect of bacteriophage therapy in the gut model, we assessed several hypotheses. The possibility that bacteriophage was not being delivered to the colon, or that colonic bacteriophage lacked potency against JJ1886-S, was excluded by testing filtered extracts of fecal pellets (a surrogate for colonic contents) from d + 7 of bacteriophage therapy. These extracts inhibited the initial JJ1886-S, with a mean concentration reduction of 3.3 log₁₀ (range, 1.0–5.5 log₁₀), roughly comparable to the initial bacteriophage cocktail (mean reduction, 5.5 log₁₀), whereas fecal extracts from control mice were inactive. This presumptively confirmed the presence JJ1886-active bacteriophage in the colon of bacteriophage-treated mice.

By contrast, the possibility that JJ1886-S had developed bacteriophage resistance in vivo during bacteriophage exposure was confirmed. Specifically, for only four of 16 bacteriophage-treated mice did the initial bacteriophage cocktail inhibit fecal isolates of JJ1886-S from d + 7 of bacteriophage therapy, and for one of these four mice the inhibition was only by only 0.5 log₁₀. Notably, at no point did either JJ1886 or JJ1886-S exhibit evidence of acquired bacteriophage resistance after in vitro exposure to the bacteriophage cocktail.

To clarify the basis for the ineffectiveness of probiotic strain EcN-SR/pp70 against colonization with JJ1886-S when used alone, despite its substantial fecal abundance (Figure 4), combinatorial experiments were done. First, to assess for possible in vivo development of microcin resistance by JJ1886-S during residence in the mouse gut, isolates of JJ1886-S from 32 fecal samples (four each from eight EcN-SR/pp70-treated mice, as collected on d + 1, d + 4, d + 7, and d + 10 per mouse) were tested in duplicate (n = 16) or triplicate (n = 16) for inhibition in vitro by supernatants from the input EcN-SR/pp70 strain. Of the eight mice, four had received and four had not received concomitant bacteriophage cocktail. Of the 32 fecal JJ1886-S isolates, 11 yielded irreproducible results (i.e., >1 log10 difference between replicate determinations, or SE ≥ 1.0), hence were uninformative, whereas the remaining 21 isolates (66%) yielded reproducible results.

The 21 isolates with reproducible results were distributed fairly evenly both by study day (4–6 isolates per day) and by concomitant administration (n = 9) or not (n = 12) of bacteriophage cocktail to the corresponding mice. Nearly all isolates (20/21, 95%) showed wild type-level inhibition by EcN-SR/pp70, with a median effect size of 5.0 log10 (range, 3.5—6.0), whereas one (from d + 1) showed no effect. These data largely exclude acquired microcin resistance as an explanation for either the inefficacy of probiotic strain monotherapy or the time-limited effect of combined probiotic plus bacteriophage therapy.

Second, 62 fecal isolates of EcN-SR/pp70, including 3–4 isolates each for 16 EcN-SR/pp70-treated mice (both with and without bacteriophage therapy), as selected arbitrarily from all four sampling days, were tested in triplicate for activity against the input JJ1886-S strain. Fifty isolates yielded ≥4-log₁₀ inhibition and six others yielded 3.0–3.9-log₁₀ inhibition. The remaining six isolates, although non-inhibitory (<1.0-log₁₀ inhibition), remained ampicillin-resistant, suggesting plasmid retention. This largely excluded loss of bacteriocin production, including from plasmid segregation, by colonic-resident probiotic bacteria.