The AcrAB-TolC Efflux Pump Impacts Persistence and Resistance Development in Stationary-Phase Escherichia coli following Delafloxacin Treatment

ABSTRACT

Bacteria have a repertoire of strategies to overcome antibiotics in clinical use, complicating our ability to treat and cure infectious diseases. In addition to evolving resistance, bacteria within genetically clonal cultures can undergo transient phenotypic changes and tolerate high doses of antibiotics. These cells, termed persisters, exhibit heterogeneous phenotypes; the strategies that a bacterial population deploys to overcome one class of antibiotics can be distinct from those needed to survive treatment with drugs with another mode of action. It was previously reported that fluoroquinolones, which target DNA topoisomerases, retain the capacity to kill nongrowing bacteria that tolerate other classes of antibiotics. Here, we show that in Escherichia coli stationary-phase cultures and colony biofilms, persisters that survive treatment with the anionic fluoroquinolone delafloxacin depend on the AcrAB-TolC efflux pump. In contrast, we did not detect this dependence on AcrAB-TolC in E. coli persisters that survive treatment with three other fluoroquinolone compounds. We found that the loss of AcrAB-TolC activity via genetic mutations or chemical inhibition not only reduces delafloxacin persistence in nongrowing E. coli MG1655 or EDL933 (an E. coli O157:H7 strain), but it limits resistance development in progenies derived from delafloxacin persisters that were given the opportunity to recover in nutritive medium following antibiotic treatment. Our findings highlight the heterogeneity in defense mechanisms that persisters use to overcome different compounds within the same class of antibiotics. They further indicate that efflux pump inhibitors can potentiate the activity of delafloxacin against stationary-phase E. coli and block resistance development in delafloxacin persister progenies.

INTRODUCTION

Antibiotic treatment failure is a growing public health concern that threatens to compromise our ability to treat infections (1, 2). Beyond the acquisition of genetic determinants that confer resistance to existing antibiotics, bacteria can reversibly reprogram their phenotypes stochastically or in response to environmental cues, thereby becoming tolerant to lethal antibiotic doses (3–6). These bacterial persisters—subpopulations of antibiotic-tolerant phenotypic variants within genetically clonal cultures—are thought to underlie infection relapse following antimicrobial therapy (4, 7). To further complicate the problem, antibiotic-refractory cells have been shown to accelerate the evolution of heritable antibiotic resistance in bacterial populations (8–11).

Many antibiotics at our disposal target cellular processes that bacteria depend on for growth (12). As such, bacteria in high-density or growth-inhibited cultures are refractory to some classes of antibiotics, including cell wall biogenesis-inhibiting β-lactams (13). Other classes retain their bactericidal activities against slow-/nongrowing populations; some of these drugs benefit from the coadministration of metabolites that stimulate metabolism and respiration without restoring bacterial growth (14–16). Fluoroquinolones (FQs), compounds that target type II topoisomerases in bacteria and are widely used to treat respiratory and urinary tract infections, have been shown to remain active against growth-inhibited bacteria, including those in stationary-phase cultures and biofilms (12, 17–19).

Despite their ability to survive lethal FQ doses, FQ persisters originating from slow-/nongrowing populations are not spared from antibiotic-induced DNA damage (11, 20–22). It was reported that starved Escherichia coli treated with FQs induced the expression of genes involved in DNA repair after the antibiotic had been removed and bacteria were inoculated into nutritive medium (20). During this postantibiotic treatment recovery period, the timing of DNA repair and the resumption of de novo DNA synthesis was shown to impact FQ persister levels in the population (21). Additionally, the induction of SOS response genes, including umuDC that encode error-prone polymerase V, following FQ treatment contributes to the development of antibiotic resistance in persister progenies after a single round of treatment (11). These findings suggest that the molecular events that take place before, during, and after exposure to antibiotics can potentially impact antibiotic persistence and resistance development.

In this study, we investigate whether FQ persistence in slow-/nongrowing E. coli cultures and resistance development following FQ treatment hinge on the expression of genes beyond those involved in the SOS response. Specifically, we focus on the expression of genes encoding the inner membrane protein AcrB and periplasmic linker AcrA, which together with the outer membrane protein TolC, compose the tripartite resistance-nodulation-division (RND) multidrug efflux pump, AcrAB-TolC (23). The genes encoding the AcrAB-TolC efflux pump are conserved in a number of Gram-negative pathogens; acrAB are often found in the same operon, and tolC is located at another locus (24, 25). This efflux pump is capable of expelling toxic metabolic intermediates and at least nine classes of antibiotics, including FQs, in a proton motive force-dependent manner (23, 26, 27). Mutations in genes encoding the efflux pump components and in genes encoding their local and global regulators, which lead to upregulation of pump production, are routinely detected in antibiotic-resistant isolates (28, 29). Deletion of genes encoding efflux pumps has also been shown to reduce resistance evolution in E. coli populations that were exposed to protein synthesis inhibitors (30). In genetically clonal bacterial populations, the expression of acrAB and tolC can be heterogeneous, and individual cells with higher efflux pump expression have transiently reduced antibiotic susceptibility and enhanced mutation frequency (31–33). Furthermore, elevated expression of AcrAB-TolC promoted persister formation and reduced intracellular antibiotic accumulation in growing E. coli populations treated with β-lactam drugs (34).

Here, we investigated the impact of acrB deletion on the persistence of stationary-phase E. coli toward four FQ antibiotics, levofloxacin (Levo), delafloxacin (Dela), ciprofloxacin (Cipro), and clinafloxacin (Clina). Although the MICs of all four drugs decreased in the acrB mutant, loss of efflux pump activity only decreased persistence toward Dela, a fourth-generation FQ. We found that inhibiting the efflux system potentiated the activity of Dela against E. coli in colony biofilms and pathogenic E. coli strains in stationary-phase cultures. We further discovered that when stationary-phase E. coli with impaired efflux systems were treated with lethal doses of Dela, resistance development in those persisters’ progeny was significantly reduced compared with persister populations with intact efflux functions.

RESULTS

Genes encoding efflux pumps are induced following FQ treatment.

It was previously reported that persisters induced genes in the SOS regulon as they recovered from FQ treatment and that deletion or mutation of key DNA repair enzymes, including recA and lexA, drastically reduced persistence (11, 20–22). To investigate whether E. coli FQ persisters from slow-/nongrowing cultures depended on additional defensive mechanisms, we measured whether bacteria express genes encoding efflux pump components as they recover from FQ treatment. In our time-kill assays, FQs were administered at ∼10- to 100-fold above their MICs, so that persisters would be the only culturable cells in the population by the end of treatment (see Table S1 in the supplemental material). We initially focused on cells treated with Levo, the active (S)-enantiomer of ofloxacin. We monitored green fluorescence from a stationary-phase E. coli population bearing a transcriptional reporter of acrAB expression—composed of the acrAB promoter (P_acrAB) fused to green fluorescent protein (GFP)—during and after Levo treatment. We observed that in the treatment-free control and immediately after the population had been treated with a lethal dose of Levo for 5 h, acrAB expression was not detected (Fig. 1A, Fig. S1A). After the removal of the antibiotic, the Levo-treated cells were inoculated into Luria-Bertani (LB) broth with kanamycin (Kan) for plasmid maintenance. Within 2 h of recovery in LB-Kan broth, a small subpopulation of cells exhibited fluorescence above the negative control (Fig. S1A). After 4 h of recovery in LB-Kan broth, expression of the acrAB reporter was detected in 52% ± 6% of the population; 50% ± 3% of the population remained fluorescent after 6 h of recovery (Fig. 1B, Fig. S1A). Although the expression of acrAB is generally considered to be constitutive (35), in our experiments, we did not detect reporter expression in our FQ treatment-free controls as the cells grew in LB-Kan broth. This suggests that the fluorescent reporter reports on the upregulation of acrAB expression rather than its constitutive expression. Consistent with previous reports (20), stationary-phase E. coli treated with Levo also induced the expression of recA—a key SOS response protein—during recovery. Compared with expression of the recA reporter, where 95% ± 1% of cells were highly fluorescent within 2 h of antibiotic removal, acrAB induction was delayed and was observed in ∼52% of cells (Fig. S1B).

FIG 1.

acrAB expression is induced following FQ treatment in stationary-phase E. coli. (A) When stationary-phase E. coli cultures were treated with 5 μg/ml of Levo, >90% of the population was killed within 5 h of treatment. (B) As Levo-treated cells recovered in LB medium after treatment, a subpopulation of the culture induced the acrAB reporter within 4 h. (C) Stationary-phase E. coli cultures are also susceptible to 5 μg/ml of Dela, where >90% of the population was killed after 7 h of treatment. (D) Induction of the acrAB reporter was observed in E. coli recovering in LB medium following 7 h of Levo and Dela treatment. At least three biological replicates were performed for the kill curves. The histograms shown for the reporter assays are representative of data from at least three biological replicates.

To ensure that the induction of the acrAB reporter was not attributed to residual Levo present after the antibiotic had been removed, we monitored fluorescence from stationary-phase E. coli bearing the acrAB reporter inoculated into LB-Kan broth containing 0.6 ng/ml of Levo (the amount expected to remain after phosphate-buffered saline [PBS] washes and diluting the Levo-treated E. coli in LB broth). We did not observe green fluorescence in these cells after 4 h of growth in medium with trace Levo, indicating that residual antibiotic in the Levo-treated populations was not responsible for the fluorescence detected in cells harboring the acrAB reporter (Fig. S2).

Next, we asked whether treatment with other FQ antibiotics triggered acrAB expression following antibiotic treatment, and we turned to Dela—a novel, fourth-generation FQ that received regulatory approval in recent years (36). Compared with Levo, Dela killed stationary-phase E. coli more slowly, and the second phase of killing was not detected until after 5 h of treatment (Fig. 1C). Consequently, we treated E. coli bearing the acrAB reporter with Dela for 7 h before removing the antibiotic and recovering the cells in LB-Kan broth. For comparison, we further treated bacteria with Levo for 7 h before assessing acrAB reporter induction during recovery in LB-Kan broth. Similar to cells treated with Levo, we observed that 64% ± 3% of the Dela-treated E. coli induced expression of the acrAB reporter within 4 h after treatment and recovery in nutrient-rich medium (Fig. 1D; Fig. S3). As the MarA (multiple antibiotic resistance) transcriptional regulator participates in controlling the expression of acrAB and tolC, we constructed a transcriptional reporter with the promoter of marRAB fused to GFP. We observed that, similar to acrAB expression, fluorescence of the marRAB reporter was observed in 80% ± 7% of Levo-treated or Dela-treated cells 4 h postantibiotic treatment (Fig. S3C).

Deletion of acrB compromises persistence to Dela.

In agreement with stationary-phase E. coli treated with Levo and other FQs, cells treated with a lethal dose of Dela also engaged in the SOS response after antibiotic treatment and depended on recA to survive (Fig. S3 and S4). As we observed that both Levo- and Dela-treated cells induced expression of the acrAB reporter during recovery from treatment, we asked whether persistence toward Levo and Dela depended on AcrAB-TolC. We assessed Levo and Dela persistence in a ΔacrB mutant, as acrB deletion leads to inactivation of the entire efflux pump (32). While the deletion of acrB did not impact the persistence of stationary-phase cultures toward Levo (Fig. 2A), the ΔacrB mutant exhibited a 67-fold decrease in persistence toward Dela after 7 h of treatment (Fig. 2B). Complementation of acrB restored Dela persistence to wild-type levels (Fig. 2C). Deletion of acrB not only compromised Dela persistence of E. coli in planktonic stationary-phase cultures, but it also reduced persistence in colony biofilms, which is a more clinically relevant lifestyle. We observed significant decreases in Dela persistence in 24-h-old and 48-h-old biofilms, reducing persistence by 36-fold and 7-fold, respectively, compared with the survival of wild-type biofilms following 7 h of treatment (Fig. 2D to F). These data demonstrate that AcrAB-TolC function is necessary for the survival of E. coli in planktonic stationary-phase cultures and in biofilms upon exposure to lethal doses of Dela.

FIG 2.

Dela persisters from slow-/nongrowing cultures depend on the AcrAB-TolC efflux system for survival. (A) Deletion of acrB did not affect Levo persistence in stationary-phase planktonic E. coli cultures. (B) The loss of acrB reduced Dela persistence in stationary-phase planktonic cultures by 67-fold. (C) Complementation of acrAB expressed from the operon’s native promoter on pBAD33 restored Dela persistence to wild-type levels. (D) E. coli biofilms were cultured on a nutrient-permeable, bacterium-impermeable membrane overlaid on agar pads prepared with Gutnick-glucose. Deletion of acrB compromised Dela persistence of E. coli grown in (E) 24-h-old and (F) 48-h-old colony biofilms. At least three biological replicates were carried out for each experiment. The asterisks denote statistically significant differences (P ≤ 0.05) between designated mutants and the survival of the wild-type population at the same time point.

Beyond Levo and Dela, we also assessed the dependence of stationary-phase E. coli on acrB to survive treatment with two other FQs, Cipro and Clina. In cells treated with either of these FQs, deletion of acrB did not further reduce persister levels (Fig. S5). Clina, similar to Dela, contains a chlorine substitution at the C8 position that is absent in other marketed FQs and acts as a strong electron-withdrawing group on the N1 aromatic ring (37–39). While stationary-phase E. coli was more susceptible to Clina treatment compared with treatment with the other three FQs, the deletion of acrB did not further reduce survival following Clina treatment. These observations suggest that loss of AcrAB-TolC function only impacts the persistence of slow-/nongrowing E. coli populations toward Dela, possibly due to the other unusual chemical properties, such as the absence of a basic substituent at position C7 and its large N1 ring, that affect Dela’s intracellular accumulation and action (40).

Next, we asked whether the difference in AcrAB-TolC dependence between Levo and Dela persisters is a consequence of the enhanced permeability of Dela across the outer membrane of E. coli. We hypothesize that if differences in AcrAB-TolC dependence were attributed to the faster accumulation of Dela, dependence on the efflux pump would be abolished when cells are treated with lower doses of Dela. Analogously, if the other FQs, such as Levo, were slower to accumulate in cells, we would expect to see the dependence of Levo persisters on AcrAB-TolC when the cells are treated with high doses of Levo. To test these hypotheses, we quantified persisters originating from stationary-phase E. coli cultures after 7 h of treatment with a range of Levo or Dela concentrations (Fig. S6). We found that wild-type E. coli and the ΔacrB mutant exhibited comparable survival after treatment with 0.5 to 50 μg/ml of Levo (Fig. S6A). By comparison, there were far fewer Dela persisters in ΔacrB cultures than in wild-type cultures across a range of concentrations (0.1 to 50 μg/ml). Notably, the ΔacrB mutant still had significantly impaired survival at 0.5 μg/ml Dela compared to the wild-type cells (Fig. S6B). These results suggest that in Dela persisters from slow-/nongrowing E. coli populations, the dependence on the AcrAB-TolC efflux pump involves additional factors beyond the enhanced permeability of this FQ.

The impact of tolC deletion on Dela persistence.

Although AcrAB-TolC is the predominant efflux pump involved in FQ expulsion in E. coli, FQs can be extruded by secondary efflux pumps, many of which employ TolC (e.g., AcrEF-TolC and MdtABC-TolC) (23, 41). To assess whether stationary-phase Dela persisters depend on additional TolC-associated efflux pumps, we deleted tolC in E. coli MG1655. Consistent with results generated with the ΔacrB mutant, stationary-phase cultures of the ΔtolC mutant were more susceptible to Dela treatment, but their ability to survive Levo treatment was not compromised across the range of drug doses investigated (Fig. S7A and B). We note that the ΔtolC mutant was more sensitive to Dela than to its ΔacrB counterpart, as the survival of the ΔtolC mutant significantly decreased by 4 orders of magnitude compared with the wild-type control when only 0.1 μg/ml of Dela was applied (Fig. S7B). By comparison, survival of the ΔacrB mutant was comparable to that of the wild-type population at this dose of Dela (Fig. S6B). We then carried out a time-kill assay with the ΔtolC mutant (Fig. S7C). Consistent with our concentration-dependent killing experiments, the ΔtolC mutant was more sensitive to Dela than the wild-type population. Complementation of tolC expressed in a low-copy plasmid under the control of its native promoter restored persistence to wild-type levels. These results suggest that additional TolC-associated efflux pumps may be involved in Dela persistence in stationary-phase E. coli cultures.

Dela persisters depend on efflux during antibiotic treatment.

After we found that the majority of Dela persisters originating from stationary-phase E. coli depended on AcrAB-TolC, we sought to determine whether the efflux pump is needed during or after Dela treatment. We utilized the efflux pump inhibitor phenylalanine-arginine β-naphthylamide (PAβN), which has been demonstrated to inhibit AcrAB-TolC (34, 42). We either coadministered PAβN with Dela or plated Dela-treated cells on LB agar with PAβN following Dela treatment. Dimethyl sulfoxide (DMSO), the solvent used to dissolve the inhibitor, was used in place of PAβN for negative controls. These treatment regimens enabled us to determine whether AcrAB-TolC activity was needed throughout the course of Dela treatment or during recovery from treatment.

We discovered that the coadministration of PAβN with Dela resulted in a 40-fold decrease in survival in stationary-phase E. coli populations after 7 h of treatment compared with cultures that were treated with Dela and DMSO (Fig. 3A). However, exposure of E. coli to the efflux pump inhibitor after Dela removal produced similar levels of survivors as populations that were treated with Dela alone (Fig. 3A). Exposing cells that were subjected to Dela and PAβN treatment to continued efflux inhibition during recovery from treatment did not enhance killing. Consistent with data from our genetic mutants, treating cells with Levo and PAβN did not increase killing compared with cells that were treated with Levo alone (Fig. S8A). To ensure that our results are attributed to efflux pump inhibition by PAβN and are not a consequence of its previously reported activity as an outer membrane permeabilizer (43), we treated the wild-type cells or the ΔacrB mutant with the efflux pump inhibitor at the time of Dela or Levo treatment (Fig. S8B). If PAβN were permeabilizing the membrane, we would expect to see a significant decrease in Dela or Levo persistence in the ΔacrB mutant treated with both the inhibitor and the FQs compared with mutants that were treated with FQs alone. In our experiments, we observed that addition of PAβN did not significantly reduce Levo persistence in wild-type or ΔacrB cells, nor did it further reduce Dela persistence in ΔacrB mutants. This indicates that our previous efflux inhibition results were not confounded by off-target membrane permeabilization by PAβN.

FIG 3.

Inhibition of AcrAB-TolC during Dela treatment compromises E. coli MG1655 and EDL933 survival. (A) Inhibition of AcrAB-TolC with PAβN during Dela treatment significantly reduced survival of stationary-phase E. coli MG1655 cultures compared with cells that were treated with Dela alone. Inhibition of the efflux system as cells recovered on LB agar after 7 h of Dela treatment did not enhance killing. (B) PAβN potentiated the activity of Dela against E. coli MG1655 grown in 24-h-old colony biofilms. (C) Coadministration of Dela and PAβN decreased the persistence of enterohemorrhagic E. coli (EHEC; strain EDL933) compared with cultures that were treated with Dela without the inhibitor. (D) Addition of PAβN did not enhance the bactericidal activity of Dela toward stationary-phase cultures of uropathogenic E. coli (UPEC; strain CFT073) after 7 h. In panels A, C, and D, the asterisks denote statistically significant differences (P ≤ 0.05) between designated data points and the survival of cells treated with Dela alone at the same time point. In panel B, the asterisk denotes a statistically significant decrease (P ≤ 0.05) in the survival fraction (SF) of the population treated with Dela and PAβN compared with the population treated with Dela alone, as a ratio of 1 would indicate no change.

We then treated E. coli MG1655 24-h-old colony biofilms with Dela and PAβN and found that the survival of cells in the biofilms decreased 6-fold after 7 h of treatment relative to populations treated with the Dela without the efflux inhibitor (Fig. 3B). Although the acrB deletion significantly decreased Dela persistence in 48-h-old biofilms, coadministration of Dela and PAβN did not significantly decrease survival compared with biofilms that were solely treated with the FQ (Fig. S8C). However, it is unclear whether PAβN can freely diffuse into these more mature biofilms with more established extracellular matrices. Nevertheless, our data demonstrate that Dela persisters originating from stationary-phase E. coli cultures and 24-h-old biofilms depend on the AcrAB-TolC efflux systems during Dela treatment and not during the recovery phase.

Efflux inhibition reduces Dela persistence in stationary-phase enterohemorrhagic E. coli.

In addition to E. coli MG1655, we also assessed the effects of inhibiting the efflux machinery with PAβN during Dela treatment in stationary-phase cultures of two pathogenic E. coli strains, enterohemorrhagic E. coli (EHEC) strain EDL933 (an E. coli O157:H7 strain) and uropathogenic E. coli (UPEC) strain CFT073. Notably, we found that Dela killed stationary-phase EHEC and UPEC cultures more rapidly than E. coli MG1655 cultures. Within 7 h of Dela treatment, ∼99.9% of the EHEC culture and ∼99.99% of the UPEC culture were killed (Fig. 3C and D). By comparison, 7 h of Dela treatment reduced E. coli MG1655 populations by ∼99% (10-fold and 100-fold higher than EHEC and UPEC, respectively) in the absence of the efflux inhibitor (Fig. 3A). Under our experimental conditions, inhibiting AcrAB-TolC resulted in an 8-fold reduction in Dela persistence in EHEC populations after 7 h, whereas efflux inhibition decreased UPEC survival at earlier time points but did not further reduce UPEC survival after 7 h of treatment. These data demonstrate that similar to our findings in slow-/nongrowing E. coli MG1655 populations, efflux pump inhibition using PAβN can potentiate the activity of Dela against stationary-phase EHEC cultures.

Antibiotic resistance development increases following Dela treatment.

Previously, it was discovered that when FQ persisters are given the opportunity to recover from antibiotic treatment and engage in DNA repair, the frequency of antibiotic-resistant mutants in the resulting population significantly increases relative to that in untreated controls even after a single round of FQ exposure (11). In addition to the FQ that was used for treatment, the persister progenies exhibited enhanced resistance toward antibiotics with other modes of action, including RNA polymerase-targeting rifampicin (Rif) (11). Rif resistance enhancement was previously quantified in FQ persister progenies, and this antibiotic is an AcrAB-TolC substrate that is routinely used to determine mutation rates (11, 32, 44, 45). As such, we focused on determining Rif resistance development in our experiments.

We assessed whether antibiotic resistance is increased in populations stemming from stationary-phase Dela persisters by plating ∼10⁹ cells that had recovered from Dela treatment on LB agar plates containing 500 μg/ml of Rif, as described previously (11). We found that in populations derived from Dela persisters, the number of Rif-resistant (Rif^r) mutants per billion cells plated increased by 3.7- ± 1.0-fold relative to the untreated control (Fig. 4A; Fig. S9A). By comparison, in populations derived from Levo persisters, the level of Rif^r mutants increased by 10.8- ± 2.6-fold compared with that of the untreated control (Fig. 4A; Fig. S9A).

FIG 4.

Deletion of acrB reduces resistance development following Dela treatment in E. coli MG1655. (A) After stationary-phase cultures of E. coli MG1655 were treated with Levo or Dela for 7 h, surviving cells were recovered in LB in the absence of the antibiotic overnight, and the number of rifampicin-resistant (Rif^r) colonies per billion cells plated was enumerated. The bars depict the number of Rif^r mutants per 10⁹ cells in populations that had recovered from Levo or Dela treatment and controls that were not subjected to FQ treatment. (B) Deletion of acrB reduced Rif resistance in populations that recovered from Dela treatment. Complementation of acrAB restored resistance to wild-type levels. The asterisks (*) denote statistically significant differences (P ≤ 0.05) between Rif^r colonies per 10⁹ cells in the treatment-free control and the Levo-/Dela-treated population.

We next assessed the impact of deleting umuDC or acrB on Rif resistance development in the FQ persister progenies. Error-prone DNA polymerase V, encoded by the umuDC genes, was found to play a role in resistance enhancement in populations derived from FQ persisters (11). In agreement with previous reports, in the ΔumuDC mutant, the number of Rif^r mutants in progenies of Levo and Dela persisters was no longer significantly different compared with the number of Rif^r mutants in populations derived from the untreated controls (Fig. 4B; Fig. S9B and C). With the ΔacrB mutant, we observed that deletion of acrB compromised the recovery of Dela-treated cells following antibiotic removal. In two of our five experimental replicates, the number of CFU present after 16 h of recovery in LB broth decreased by over 3 orders of magnitude compared with the untreated control. In the other three experimental replicates, we were able to obtain ∼10⁹ ΔacrB mutants that had recovered from Dela treatment when we collected 10 ml of culture from each flask. In contrast, the numbers of CFU in the ΔacrB cultures following recovery from Levo treatment were similar to that of the untreated control. Remarkably, in cultures derived from the ΔacrB mutant that recovered from Dela treatment (from which we were able to recover ∼10⁹ cells), we either detected fewer Rif^r mutants compared with the untreated control or we did not detect any Rif^r colonies (Fig. 4B; Fig. S9B). Complementation of acrB restored the number of Rif-resistant mutants detected following recovery from Dela treatment to 5.2- ± 0.1-fold (Fig. 4B; Fig. S9B). Consistent with our observations that ΔacrB did not affect Levo persistence or culturability of subsequent Levo persister progenies, deletion of acrB did not reduce Rif-resistance development in progenies that recovered from Levo treatment (Fig. S9C).

In previous publications, it was reported that cells with higher acrAB expression have lower expression of the DNA mismatch repair gene mutS, which results in higher mutagenesis (32). Here, we asked whether efflux pump inhibition increases mutS expression in cells recovering from Dela treatment, resulting in decreased in Rif^r colonies in their progenies (Fig. S10A and B). Using a fluorescent reporter that reports on mutS expression, we found that in wild-type E. coli, 35% ± 2% of cells recovering from Levo and 45% ± 1% of cells recovering from Dela treatment induced mutS expression. However, in ΔacrB mutants, only 10% ± 1% of the population induced expression of the mutS reporter as they recovered from Dela treatment. By comparison, the percentage of ΔacrB mutants that induced the mutS reporter following Levo treatment were comparable to that of wild-type cells (39% ± 1%). These results suggest that the decrease in Rif^r observed in the Dela persister progenies is not associated with increased MutS activity during recovery following antibiotic treatment.

We also monitored induction of the recA reporter in the ΔacrB mutant recovering from Levo and Dela treatment (Fig. S10C). We found that while 90% ± 2% of Dela-treated mutants induced expression of the recA reporter during recovery, the fluorescence signal stemming from these cells was lower than that of ΔacrB mutants recovering from Levo or that of wild-type cells recovering from Dela or Levo treatment (Fig. S3). These data suggest that the loss of AcrAB-TolC function during Dela treatment leads to differential expression of genes in the SOS regulon as these cells recovered from treatment, resulting fewer Rif^r mutants in their persister progenies.

PAβN reduces resistance development following Dela treatment.

As we observed that inhibiting efflux with PAβN reduces Dela persistence in stationary-phase E. coli MG1655 and EHEC cultures, we asked whether treating these populations with Dela and the efflux inhibitor affects resistance development following antibiotic treatment and recovery in LB broth (Fig. 5). We observed that the E. coli MG1655 ΔacrB mutant exhibited reduced CFU counts following recovery from Dela treatment compared with the untreated control or the Dela-treated wild-type populations (Fig. 4B). In contrast, we found that the wild-type strain inhibited with PAβN during Dela treatment had comparable optical densities and CFU counts as populations that were treated with DMSO (the negative control) or with Dela alone after 16 h of recovery in LB broth. Notably, inhibiting E. coli MG1655 with PAβN during Dela treatment significantly reduced Rif resistance in progenies that recovered from treatment to levels that were indistinguishable from that of the DMSO-treated control (Fig. 5A). By comparison, in progenies derived from stationary-phase E. coli MG1655 cultures that were treated with Dela and DMSO (in place of the efflux inhibitor), Rif resistance was 15- ± 5-fold higher than for the DMSO control.

FIG 5.

PAβN reduces resistance development following Dela treatment. (A) In stationary-phase cultures of E. coli MG1655 treated with Dela along with PAβN that were given the opportunity to recover in LB medium following treatment, the number of rifampicin-resistant (Rif^r) mutants detected per billion cells plated decreased relative to the number of Rif^r mutants from populations that were treated with Dela alone. (B) Coadministration of PAβN with Dela also reduced Rif^r mutants in populations derived from stationary-phase E. coli EDL933 (an EHEC O157:H7 strain). (C) The addition of PAβN did not reduce Rif^r mutants derived from stationary-phase E. coli CFT073 (a UPEC strain) following Dela treatment and recovery. At least three biological replicates were performed for each experiment. The asterisks (*) denote statistically significant differences (P ≤ 0.05) between Rif^r colonies per 10⁹ cells in the treatment-free control and the Dela (±PAβN)-treated population.

In pathogenic E. coli EDL933 and CFT073, the number of Rif^r mutants detected in populations that were derived from the Dela persisters were comparable to the Rif^r mutants in the DMSO control. Nevertheless, we found that in EHEC strain EDL933, cotreatment with Dela and PAβN significantly reduced Rif-resistant mutants in the persister progenies compared with the DMSO control (Fig. 5B). In contrast, treating UPEC strain CFT073 with PAβN in conjunction with Dela did not reduce resistance development in the progenies derived from surviving persisters (Fig. 5C). Collectively, our results suggest that treating stationary-phase E. coli MG1655 and EHEC with PAβN together with Dela effectively reduces persistence and resistance development in progenies originating from the FQ persisters.

DISCUSSION

The persister phenotype is inherently heterogeneous, and bacteria within genetically clonal cultures can deploy different strategies to survive antibiotics targeting distinct cellular processes (46–49). Here, we report that E. coli persisters can engage in different defense mechanisms to survive treatment with FQs with different chemical properties. In our study, we discovered that Dela persisters originating from stationary-phase E. coli cultures and colony biofilms depend on the AcrAB-TolC efflux system, whereas the deletion of acrB does not impact persistence toward Levo, Cipro, or Clina in these E. coli populations (summarized in Fig. 6). Based on our results, we hypothesize that Dela persisters originating from stationary-phase E. coli cultures are more dependent on the action of AcrAB-TolC (compared with persisters that withstand other FQs) because of Dela’s unique chemical properties and equipotency toward type II DNA topoisomerases.

FIG 6.

Possible strategies to reduce Dela persistence. Our data suggest that targeting the AcrAB-TolC efflux pump can potentiate the activity of Dela toward slow-/nongrowing E. coli cultures during treatment, whereas efflux inhibition does not impact the activity of the other FQs examined in our study. Consistent with previous findings (11, 20, 21) inhibition of the SOS response can impair the recovery of stationary-phase cultures treated with Dela, Levo, Cipro, or Clina.

Dela is chemically distinct from other FQs currently on the market (including Levo, Cipro, and Clina), as it lacks a basic substituent at position C7 (38). Consequently, Dela remains predominately anionic at neutral pH, whereas other FQs are zwitterionic under similar conditions. Dela also carries a chlorine atom at position C8 that acts as a strong electron-withdrawing group, and it has a large molecular surface on its aromatic N1 ring (38–40). Collectively, these properties can promote its intracellular accumulation in stationary-phase E. coli and enhance its activity (38). Clina, a fourth-generation FQ that is no longer under clinical investigation due to phototoxicity and dysglycemia concerns, also contains chlorine at C8 (50). However, we did not observe a decrease in persistence in ΔacrB mutants following Clina treatment as seen after Dela treatment (Fig. S5B). These findings suggest that the combination of unusual chemical properties of Dela, besides the chlorine atom in the C8 position, contribute to its enhanced activity against E. coli MG1655 and EDL933 in planktonic stationary-phase cultures and in biofilms that lack intact efflux functions. Indeed, Dela is thought to exhibit enhanced accumulation in bacteria relative to other FQs, especially under acidic conditions where Dela is predominately in the neutral form (39). This property of Dela is advantageous, as acidic pH is prevalent in many infection milieux, including the urinary tract and abscess fluids (51).

While Levo and Dela are both good substrates of the AcrAB-TolC efflux pump, as our MIC assays indicate, our persister assays show that the deletion or chemical inhibition of acrB only reduces persistence toward Dela (Fig. 2 and 3). The MIC assay and the persister assay interrogate different aspects of E. coli’s response to FQs using populations grown under different conditions. The results of our MIC assays inform us of the minimal concentrations of each FQ needed to inhibit the growth of E. coli in mid-log phase. By comparison, our persister assays report on the bactericidal effects of each FQ, administered at doses that exceed the MIC, toward slow-/nongrowing E. coli. Loss of AcrAB-TolC action can lead to an increase in intracellular Dela accumulation in affected cells during the persister assay due to Dela’s unique chemical properties, and these cells may be more likely to die due to Dela’s cellular targets. While FQs target both type II DNA topoisomerases, most compounds preferentially inhibit DNA gyrase in Gram-negative bacteria or topoisomerase IV in Gram-positive bacteria (52). By comparison, Dela has been shown to exhibit a more balanced inhibition profile against DNA gyrase and topoisomerase IV in E. coli (52). We predict that increased intracellular abundance of Dela coupled with the inhibitory effects of this drug toward both gyrase and topoisomerase IV contribute to a two-pronged attack against stationary-phase E. coli lacking functional AcrAB-TolC.

In our experiments, we observed that in E. coli stationary-phase cultures and biofilms, coadministration of the efflux pump inhibitor PAβN at the onset of Dela treatment reduced persistence to the levels that we detected in the ΔacrB mutant. These data indicate that Dela persisters depend on efflux pumps for survival during the course of antibiotic treatment. Our data also suggest that these Dela persisters are not spared from antibiotic-induced DNA damage. Similar to the Levo persisters investigated here and ofloxacin persisters examined in previous studies (20), deletion of recA—a major SOS response enzyme—reduced Dela persistence by over 3 orders of magnitude. Dela- and Levo-treated E. coli also induce expression of the recA reporter after FQ removal as they were recovered in LB medium. These observations further indicate that these populations activate the SOS response to mediate DNA repair following FQ treatment.

Notably, we found that inhibition of AcrAB-TolC in stationary-phase E. coli following Dela and Levo treatment did not compromise persistence toward these antibiotics, despite our observations that E. coli recovering from treatment with these FQs induced acrAB expression. As the E. coli populations in our study recovered from FQ treatment, they also induced the expression of a marRAB reporter; it is well documented that the MarA transcription factor is involved in inducing the expression of acrAB and tolC (53). A recent chromatin immunoprecipitation and DNA sequencing study further showed that the mar regulon encompasses an array of genes, including those involved in lipid trafficking and DNA repair (54). In fact, the mar regulon was shown to be necessary for overcoming quinolone-induced DNA damage. We speculate that the acrAB expression reporter could have been induced as cells activated the mar regulon to permit DNA damage repair following FQ treatment. The role of the mar regulon in the resuscitation of FQ persisters originating from nongrowing populations warrants further investigation.

Not only did we find that efflux pump inhibition reduced Dela persistence, but we discovered that the loss of AcrAB-TolC function during Dela treatment decreased resistance development in persister progenies (Fig. 4 and 5). When we investigated gene expression in the ΔacrB mutant during recovery from Dela treatment, we found that fluorescence signal from mutants harboring the recA reporter was reduced compared with ΔacrB mutants recovering from Levo treatment or wild-type cells recovering from Levo or Dela treatment (Fig. S10C). Based on these observations, we hypothesize that the loss of AcrAB-TolC action during Dela treatment affects the expression genes in the SOS regulon during the post-Dela treatment recovery period, including genes that encode error-prone DNA polymerases (e.g., umuDC). This results in reductions in Rif^r mutants in the Dela persister-derived populations.

In addition to E. coli MG1655, we discovered that administrating Dela with PAβN decreased Rif^r development in Dela persister progenies of EHEC strain EDL933. By comparison, the loss of efflux pump action in UPEC strain CFT073 did not significantly decrease Dela persistence under our culture conditions, nor did it reduce resistance development in this strain’s Dela persister progenies. Whether the similar levels of Dela persistence in E. coli CFT073 (regardless of AcrAB-TolC function) is a consequence of increased Dela permeability or decreased AcrAB-TolC efficacy under our experimental conditions in this strain, compared with E. coli MG1655 or EDL933, remains to be investigated. Moreover, we speculate that these discrepancies could reflect differences in the coordination of error-prone DNA repair during the recovery period in E. coli CFT073 compared with E. coli MG1655 and EDL933. In our experiments, we observed that the number of Rif^r mutants in the Dela persister populations was significantly higher than the number of Rif^r colonies in the DMSO control in E. coli MG1655, whereas the frequency of Rif^r mutants in the progenies of Dela persisters and DMSO control were comparable in the pathogenic EHEC EDL933 and UPEC CFT073 strains. It was previously reported that the repertoire of genes in stress response regulons in pathogenic E. coli strains, such as EDL933, may be substantially different from that of laboratory K-12 strains (55). Whether the differences in the number of Rif^r colonies in Dela persister-derived cells in E. coli MG1655, EDL933, and CFT073 stem from divergence in the regulation of genes in their SOS regulons, which encompass those encoding error-prone DNA polymerases, will be the subject of future investigations.

Collectively, our findings emphasize the merit of investigating how different antimicrobial treatment regimens impact the molecular events that persisters from different strains of the same species depend on to survive. Understanding how efflux pump inhibition impacts cellular responses and repair following Dela treatment, which can lead to reductions in Dela persistence and resistance development in persister progenies, can potentially enhance our capacity to impair persister resuscitation and mitigate the impact of relapsing infections.

MATERIALS AND METHODS

Bacterial strains and plasmids.

E. coli MG1655 and its derivatives used in this study are listed in Table S2. The ΔacrB mutant was generated using P1 transduction with E. coli BW25113ΔacrB::KanR from the Keio Collection as the donor and E. coli MG1655 as the recipient (56). The ΔtolC mutant was generated using the Datsenko-Wanner method with the primers shown in Table S4 (57). Where indicated, the Kan resistance marker was removed using Flp recombinase expressed from pCP20 (57). All mutants were confirmed by colony PCR using the primers provided in Table S4.

The fluorescent gene expression reporter plasmids used in E. coli were constructed using low-copy vector pUA66 as the backbone (Table S3) (58). Inserts containing promoters from each gene of interest were amplified using E. coli MG1655 genomic DNA as the template and primers indicated in Table S4 with Phusion high-fidelity DNA polymerase (New England Biolabs, Ipswich, MA, USA). pUA66 was linearized, opening upstream of gfpmut2, and amplified by PCR. Promoters were cloned into pUA66 by Gibson assembly (New England Biolabs). For acrB and tolC complementation, the genes of interest under the control of their native promoters were amplified from E. coli MG1655 genomic DNA and cloned into pBAD33 by Gibson assembly (59). Cloned plasmids were verified using colony PCR and Sanger sequencing (Genewiz, South Plainfield, NJ, USA) before being transformed into E. coli MG1655.

Culture media and antibiotics.

The culture medium components, chemicals, and antibiotics that were used in this study were purchased from either Thermo Fisher Scientific (Waltham, MA, USA) or Sigma-Aldrich (St. Louis, MO, USA) unless otherwise indicated. All media, carbon sources, and nitrogen sources were prepared using MilliQ-purified and deionized water. LB medium was prepared from individual components (10 g/liter tryptone, 5 g/liter yeast extract, and 10 g/liter NaCl), whereas LB agar was prepared using a Difco premixed LB Miller broth (25 g/liter) and agar (15 g/liter). Gutnick-glucose medium was prepared with 10× Gutnick minimal salts (47 g/liter KH₂PO₄, 135 g/liter K₂HPO₄, 10 g/liter K₂SO₄, and 1 g/liter MgSO₄·7H₂O), 10 mM NH₄Cl, and 10 mM glucose (60). LB medium, LB agar, and Gutnick salts were sterilized by autoclaving for 30 min. Gutnick-glucose agar pads used to culture colony biofilms were prepared with autoclaved 1.5% agar, which was supplemented with Gutnick medium, 10 mM NH₄Cl, and 10 mM glucose once the agar had cooled to ∼50°C. Mueller-Hinton Broth (MHB) was prepared using a BD Difco mix (21g/liter). Nitrogen and carbon sources, as well as salts, were filter-sterilized using 0.22-μm filters before being added to the growth medium. Gutnick-glucose medium was further filter-sterilized after preparation.

Stock solutions of ampicillin (Amp; 100 mg/ml) or Kan (50 mg/ml) were prepared with MilliQ water. Stock solutions of chloramphenicol (Cam; 25 mg/ml) were prepared in 200-proof ethanol. Transformant and mutant selection was performed with medium and agar containing 100 μg/ml Amp, 50 μg/ml Kan, or 25 μg/ml Cam. Levo was prepared as a 5-mg/ml stock solution in MilliQ water titrated with 1 M NaOH until soluble. Dela and Clina were prepared as 5-mg/ml stock solutions in DMSO. Cipro was prepared as a 5-mg/ml stock solution in 0.2 N HCl. The efflux pump inhibitor phenylalanine-arginine β-naphthylamide (PAβN) was prepared as a 50 mg/ml stock in DMSO, and the final concentration in cultures was 50 μg/ml, in accordance with previously published protocols (34). Rifampicin (Rif) was prepared as a 100-mg/ml stock solution in DMSO, and E. coli mutants were selected on LB agar containing 500 μg/ml Rif (11). All of the antibiotic stock solutions that were prepared with water were filter-sterilized using 0.22-μm filters.

MIC assays.

MICs of FQs against E. coli were determined following previously established protocols (61). Briefly, E. coli MG1655, the ΔacrB mutant, and the ΔtolC mutant were inoculated into 2 ml of MHB and cultured at 37°C with shaking at 250 rpm for 16 h. Following overnight growth, the optical density at 600 nm (OD₆₀₀) of each culture was quantified using a BioTek Synergy H1 multimode plate reader (BioTek Instruments, Inc., Winooski, VT, USA), and cells were diluted to an OD₆₀₀ of 0.01 in fresh MHB. Cells were grown to an OD₆₀₀ of ∼0.2 to 0.4 before being diluted to a density of ∼5 × 10⁵ cells/ml in 5 ml of MHB. Then, 100 ml of these diluted cultures were inoculated into individual wells of a 96-well plate containing 100 μl of FQ (prepared as a 2-fold dilution series). The plates were sealed with Breathe-Easy sealing membranes (Sigma-Aldrich) and incubated at 37°C for 20 h. Following the incubation period, the OD₆₀₀ of each well was quantified to determine the MIC of each FQ for each strain. At least two biological replicates were performed for these assays.

Antibiotic persistence assays.

E. coli MG1655 or its plasmid-bearing derivatives were inoculated from –80°C frozen stocks stored in 25% glycerol into 2 ml of LB medium or LB medium with Kan/Cam for plasmid maintenance. The cells were grown at 37°C with shaking at 250 rpm for 4 h before being diluted 200-fold into 25 ml of Gutnick medium with 10 mM glucose in a 250-ml baffled Erlenmeyer flask. The bacteria were grown for 16 h at 37°C (shaking at 250 rpm). Following overnight growth, the OD₆₀₀ of each culture was measured. Then, 500 μl of culture was collected in a microcentrifuge tube for CFU enumeration prior to antibiotic treatment, where the cells were pelleted by centrifugation at 21,000 relative centrafugal force (rcf) for 3 min. After removing 450 μl of supernatant, cell pellets were resuspended in 450 μl of PBS, serially diluted, and plated on LB agar as detailed below. To assess the effects of different concentrations of Levo or Dela on the survival of wild-type E. coli, the ΔacrB, or the ΔtolC mutant, 1-ml aliquots of the overnight cultures were distributed in test tubes and treated with no drug or different concentrations of Levo (ranging from 0.5 to 50 μg/ml) or Dela (ranging from 0.1 to 10 μg/ml) for 7 h before persisters were enumerated. To assess time-dependent killing, the culture in each flask was treated with 5 μg/ml of Levo, 5 μg/ml of Dela, 1 μg/ml of Clina, or 1 μg/ml of Cipro. These antibiotics were administered at ∼10- to 100-fold above their MICs, as defined for wild-type E. coli MG1655 (Table S1). Where indicated, 50 μg/ml of PAβN (25 μl of the 50 mg/ml stock) was added at the time of antibiotic treatment. For experiments involving PAβN, negative controls were treated with Levo or Dela along with 25 μl of DMSO instead of the efflux pump inhibitor.

To quantify persisters at designated times throughout the course of FQ treatment, 500 μl of culture was removed and pelleted by centrifugation. After removing 450 μl of supernatant, pellets were washed with 450 μl of sterile PBS and pelleted again. Once again, 450 μl of supernatant was removed and pellets were resuspended with 450 μl of sterile PBS. After these washes, 10 μl of the sample was diluted in 90 μl of PBS. These steps effectively diluted the concentration of antibiotics by 1,000-fold, reducing the antibiotic concentration below its MIC (Table S1). As we observed that the residual concentration of Dela inhibited the outgrowth of the ΔtolC mutant in our initial experiments, this mutant was washed once more in subsequent assays. The cells were further subjected to five 10-fold serial dilutions in PBS, and 10 μl of each dilution was spotted onto LB agar. The cells were incubated at 37°C for 16 h. Dilutions that enabled 10 to 100 colonies to be counted were used for CFU enumeration.

Persistence in colony biofilms.

Colony biofilms were cultured on sterile 25-mm Supor hydrophilic polyethersulfone (PES) filter discs with 0.2-μm pores (Pall Corporation) overlaid on Gutnick-glucose agar pads, as described previously (21, 62). E. coli MG1655 and its ΔacrB derivative were inoculated from –80°C stocks into 2 ml of LB medium. After 4 h of culturing at 37°C, cells were diluted 1:100 into 2 ml of Gutnick-glucose and cultured for 16 h. Following overnight growth, the OD₆₀₀ of the cultures were quantified, cells were diluted to an OD₆₀₀ of ∼0.01 in Gutnick-glucose, and 100 μl of the culture was inoculated onto the PES filter discs. Five colony biofilms were seeded for each time course. After culturing the colony biofilms at 37°C for 24 h or 48 h, one of the filters containing a biofilm was transferred into a 15-ml Falcon tube with 1 ml of PBS and vortexed at maximum speed for 1 min, dislodging any cells still adhering to the filter. Cells were then serially diluted and spotted on LB agar for CFU enumeration. The other colony biofilms were treated with 200 μl of 10 μg/ml Dela or 200 μl of 10 μg/ml Dela with 100 μg/ml of PAβN. At designated time points posttreatment, cells from one biofilm were dislodged from the filter by vortexing, and 500 μl of cells was transferred into microcentrifuge tubes so that they could be washed with PBS before serial dilution and plating. CFU were enumerated after the agar plates were incubated at 37°C for 16 h.

Assessing resistance development in fluoroquinolone-treated cells.

E. coli MG1655 or its derivatives were treated with Levo or Dela as described above and recovered in LB medium as described previously (11). All experiments included a negative-control population that was treated with sterile water or DMSO instead of antibiotics and was cultured following the same procedures. For experiments involving PAβN, 25 μl of the inhibitor was added with the FQ, and 25 μl of DMSO was added with the FQ in the controls. Briefly, after cells were treated with each FQ±PAβN for 7 h, 1-ml samples were recovered from each flask. The cells were pelleted by centrifugation at 21,000 rcf for 3 min. To wash the pellets, 900 μl of supernatant was removed, and the cells were resuspended in 900 μl of PBS. The cells were pelleted and washed with PBS again, resulting in an ∼100-fold dilution of the antibiotic. Following another round of centrifugation, 700 μl of supernatant was removed, and the cells were resuspended in the remaining volume then transferred into 25 ml of LB medium, further diluting the antibiotics by ∼84-fold. The OD₆₀₀ of each culture was measured at the time of inoculation, and the cells were incubated at 37°C with shaking at 250 rpm for 16 h.

After 16 h of recovery, the OD₆₀₀ of each culture was quantified and used to determine a volume corresponding to ∼10⁹ cells, assuming that an OD₆₀₀ of 1 corresponds to approximately 8.8 × 10⁸ cells/ml. From this calculation, ∼10⁹ cells from each culture were transferred to microcentrifuge tubes and centrifuged for 3 min at 21,000 rcf. After removing all but ∼100 μl of the supernatant, the cells were plated on LB agar with Rif to determine the number of resistant mutants per billion cells. As we observed a decrease in culturability in progenies of the ΔacrB mutant and the mutant bearing pBAD33 (the empty-vector control) following Dela treatment, we also collected 10 ml of these cultures for plating on Rif agar. This allowed us to plate close to 10⁹ cells for these populations. At the same time, another aliquot of each sample was collected, serially diluted, and plated on LB agar to determine the total CFU plated on LB-Rif agar. Colonies on these plates were enumerated after 24 h of incubation at 37°C. Rif^r mutants originating from the FQ-treated populations were compared with the number of resistant colonies from the negative control, which was treated with sterile deionized water or DMSO instead of FQ in the absence or presence of PAβN during the persistence assay.

Quantifying gene expression following FQ treatment.

E. coli MG1655 or its ΔacrB mutant bearing fluorescent gene expression reporters was cultured and treated with Levo or Dela as described above for persister assays. Prior to antibiotic treatment, two 500-μl aliquots were collected from each flask; one of these aliquots was used for CFU enumeration, and the other was fixed with 4% paraformaldehyde (PFA) on ice for 30 min, allowing the cells to be preserved until their fluorescence was quantified. Following FQ treatment, three 1-ml aliquots were recovered from each flask and pelleted by centrifugation. Following two rounds of washes with PBS and a final round of centrifugation, one of the aliquots was used for serial dilution and plating, the second aliquot was fixed with 4% PFA, and the third aliquot was inoculated into 25 ml of LB-Kan. Immediately after inoculation into the fresh LB-Kan, and after 2, 4, and 6 h of recovery, the OD₆₀₀ of each sample was measured, and 1 ml of cells was collected for fixation with 4% PFA.

To fix the samples, cells were pelleted and the supernatant was removed. The pellets were then resuspended in 1 ml of 4% PFA. Following 30 min of incubation on ice, the cells were pelleted by centrifugation again. Upon removing the supernatant, they were resuspended in 1 ml of PBS. The fixed cells were stored at 4°C until analysis by flow cytometry.

We further carried out a control experiment to ensure that residual Levo that remained in the culture following PBS washes and inoculation into 25 ml of LB-Kan was not responsible for inducing expression of the acrB reporter. The PBS washes diluted the drugs ∼8,400-fold. As such, we cultured E. coli MG1655 bearing the reporter plasmid overnight and treated the culture with deionized water. Following 5 h of treatment, cells from the culture were collected, washed with PBS, and inoculated into 25 ml of LB-Kan as described above. To this culture, 0.6 ng/ml of Levo (∼8,400-fold less than the 5 μg/ml of Levo that was used for treatment) was added. At 0, 2, and 4 h following inoculation, 1 ml of cells was collected from this culture and fixed with 4% PFA.

To assess gene expression during recovery, PFA-fixed cells were diluted to an OD₆₀₀ of ∼0.01 with PBS and analyzed using an LSRII flow cytometer (BD Biosciences, San Jose, CA). Bacteria were identified using forward scatter (FSC) and side scatter (SSC) parameters that were established using an untreated control. GFP fluorescence was detected upon exciting the cells with a laser emitting at 488 nm, and fluorescence emission was detected using a green fluorescence filter (525/50-nm-band-pass filter). Fluorescence data were acquired using the FACSDiVa and analyzed using FlowJo (software from BD Bioscience). To analyze the data, we identified single cells using FSC versus SSC gating. Using the histogram of the gated untreated (negative control) population, we set the threshold for fluorescent cells with a GFP-negative gate that captured >99% of the nonfluorescent control. Single cells with fluorescence intensities above this threshold were considered fluorescent.

Statistical analysis.

Unless otherwise stated, at least three biological replicates were performed for each experiment. The graphs were generated using GraphPad Prism. The error bars for each data point depict the standard error of the mean (SEM). Two-tailed Student’s t tests with unequal variance were performed for pairwise comparison, and P values of ≤0.05 were considered statistically significant.