SUMMARY
Antibiotics can induce mutations that cause antibiotic resistance. Yet, despite their importance, mechanisms of antibiotic-promoted mutagenesis remain elusive. We report that the fluoroquinolone antibiotic ciprofloxacin (cipro) induces mutations that cause drug resistance by triggering differentiation of a mutant-generating cell subpopulation, using reactive oxygen species (ROS) to signal the sigma-S (σS) general-stress response. Cipro-generated DNA breaks activate the SOS DNA-damage response and error-prone DNA polymerases in all cells. However, mutagenesis is restricted to a cell subpopulation in which electron transfer and SOS induce ROS, which activate the σS response, allowing mutagenesis during DNA-break repair. When sorted, this small σS-response-“on” subpopulation produces most antibiotic cross-resistant mutants. An FDA-approved drug prevents σS induction specifically inhibiting antibiotic-promoted mutagenesis. Furthermore, SOS-inhibited cell division, causing multi-chromosome cells, is required for mutagenesis. The data support a model in which within-cell chromosome cooperation together with development of a “gambler” cell subpopulation promote resistance evolution without risking most cells.
INTRODUCTION
Antibiotic resistance is a world health threat with 700,000 deaths from resistant infections worldwide annually (O’Neill, 2014). Resistance occurs both by uptake of resistance genes from other bacteria, and de novo mutation of resident genes. Formation of new mutations underpins resistance to diverse antibiotics (Blair et al., 2014; Cannatelli et al., 2014; Palmer et al., 2011), and is a principle resistance route among the World Health Organization’s “priority pathogens” for which new antibiotics are needed (Magrini, 2017). Historically, the challenge of resistance has been met with development of new antibiotics. A complementary approach could be to discover, then inhibit, the molecular mechanisms that drive evolution of resistance (Al Mamun et al., 2012; Cirz et al., 2005; Rosenberg and Queitsch, 2014). Antibiotics not only select resistant mutants but can also induce their formation (Cirz et al., 2005; Gutierrez et al., 2013; Kohanski et al., 2010). Although detailed mechanisms are described by which antibiotics arrest cell growth, how antibiotics induce new mutations is poorly understood.
Fluoroquinolones are widely used antibiotics that inhibit bacterial type-II topoisomerases and kill cells via DNA double-strand breaks (DSBs), by arresting the topoisomerase at DNA double-strand cleavage (Drlica, 1999). Resistance to fluoroquinolones, including ciprofloxacin (cipro), the most used (Hicks et al., 2015), occurs primarily by de novo mutation (Jacoby, 2005). Cipro exposure at so-called “sub-inhibitory” concentrations (below minimal inhibitory concentration, MIC) occurs in ecosystems and during antibiotic therapies, and both induces and selects cipro resistance (Cirz et al., 2005). Another fluoroquinolone, norfloxacin, induced mutations that confer resistance to antibiotics not yet encountered or selected (Kohanski et al., 2010)—antibiotic “cross” resistance. The mutagenesis activity of norfloxacin required reactive oxygen species (ROS) induced by the antibiotic (Kohanski et al., 2010), as does its antibiotic (killing) activity (Kohanski et al., 2007). Yet, how the ROS might lead to mutagenesis—by what molecular mechanism—is unclear. The role of ROS in the antibiotic mechanism is to potentiate lethality by oxidizing DNA bases, the repair of which causes more lethal DNA breaks (Foti et al., 2012; Rasouly and Nudler, 2018; Zhao et al., 2015), but whether this is part of the ROS mutagenic activity is not known.
The general or starvation-stress response of Escherichia coli, activated by the sigma-S (σS) transcriptional activator (encoded by the rpoS gene), promotes mutagenesis induced by starvation stress (Fitzgerald et al., 2017), and also by beta-lactam (membrane-targeting) antibiotics (Gutierrez et al., 2013). The former occurs by allowing mutagenic repair of spontaneous DSBs, which also requires an SOS DNA-damage response for the upregulation of error-prone DNA polymerase (Pol) IV (Galhardo et al., 2009). The latter occurs by downregulating post-replication error correction (mismatch repair) via a different, SOS-independent mutation mechanism (Gutierrez et al., 2013). The σS response upregulates Pol IV about two-fold (Layton and Foster, 2003), which might be part of how it promotes mutagenesis during starvation (Fitzgerald et al., 2017).
Bacterial regulatory programs transiently differentiate phenotypically distinct cell subpopulations both stochastically and in response to environmental signals. A potential “bet-hedging” strategy, these subpopulations can allow phenotypes that may be advantageous under stress conditions but deleterious in more permissive environments (Norman et al., 2015; Veening et al., 2008). For example, bacterial “persisters” are a subpopulation of transiently nonproliferating or slowly growing cells, present at about 10-4 of the total, that can survive antibiotics without having a resistance mutation, and so lead to persistent infections by resuming growth once antibiotics have gone (Lewis, 2010). Persister formation can occur stochastically, leaving populations ready for a stress that they have not encountered (Balaban et al., 2004), and can also be induced responsively via stress-response regulons including the SOS- (Dorr et al., 2009) and σS-response (Radzikowski et al., 2016) regulons. It is unknown whether antibiotics induce transient differentiation that could promote resistance through mutagenesis, e.g., (Frenoy and Bonhoeffer, 2018).
Here we show that low, sub-inhibitory doses of cipro induce transient differentiation of a small cell subpopulation with high ROS and σS-response activity, that generates mutants, including cross-resistant mutants: a “gambler” subpopulation. We show that the ROS promote mutagenesis in gamblers by activating the σS response, which allows mutagenic repair of cipro-triggered DSBs—a novel signaling/differentiating role of ROS in mutagenesis. We elaborate the regulatory chain from cipro to ROS to σS response to mutant production, and also discover a requirement for SOS-induced inhibition of cell division, causing multiple chromosomes per cell. Mathematical analysis supports a model in which multiple chromosomes allow sharing of cellular resources (e.g., recombination, complementation), avoiding deleterious consequences of some mutations during mutagenesis and repair. Thus, multiple chromosomes allow higher mutation rates to be maintained – resulting in faster adaptation. The findings imply a highly regulated, novel transient differentiation process and support a model in which within-cell chromosome cooperation together with development of a gambler subpopulation drive evolution of resistance to new antibiotics without risk to most cells.
RESULTS
ROS-dependent Mutagenesis is σS-dependent MBR
We developed two assays to detect mutagenesis induced by cipro independently of cipro selection of the mutants (Figure 1A) (fluctuation tests, Methods), and use them to dissect the mechanism of mutagenesis. In both assays, E. coli are grown in liquid with low-dose cipro—at the minimum antibiotic concentration (MAC) at which final cfu are 10% of those observed without cipro (Lorian and De Freitas, 1979). The cells are then removed from cipro and plated selectively for colonies with resistance to rifampicin (RifR) or ampicillin (AmpR) antibiotics (Figure 1A), and mutation rates estimated (Methods). RifR arises by specific base-substitution mutations in the RNA-polymerase-encoding rpoB gene (Reynolds, 2000) (Figure S1A), and AmpR occurs in engineered E. coli by ampD loss-of-function mutation (Petrosino et al., 2002) (Figures S1B and C, Methods). Strikingly, cipro exposure increased apparent RifR and AmpR mutation rates by 26- and 18-fold, above the no-cipro rates, respectively (Figure 1B). The RifR or AmpR mutant cells are not selected in sub-inhibitory cipro, and in fact are at a slight but significant disadvantage (Figure 1C, legend) indicating that mutation rate increases are likely to be underestimates, and that mutation not selection of the mutants is elevated by low-dose cipro. Additional controls show negligible cell death in the low-dose cipro (Figure S1D), obviating potential concerns about death inflating apparent mutation rates (Frenoy and Bonhoeffer, 2018). Other controls for growth rate and colony formation are shown in Figure S2. We conclude that the RifR and AmpR mutations in rpoB and ampD are induced, and not selected by growth in low-dose cipro.
Like mutagenesis induced by norfloxacin (Kohanski et al., 2010), the cipro-induced mutagenesis is ROS dependent, and is inhibited by ROS scavenging/preventing agents thiourea (TU) and 2,2’-bipyridine (BP) (Figure 1D, and below). Whereas the mechanism(s) by which fluoroquinolone-induced ROS promote mutagenesis are unknown, the following data indicate that the ROS instigate a σS-licensed mutagenic DNA break-repair (MBR) mechanism triggered by cipro-induced DSBs.
MBR in starving E. coli is regulated mutagenesis during repair of DSBs, requiring the SOS and general (σS) stress responses (Figure 1E), reviewed (Fitzgerald et al., 2017), so that mutation formation, and potentially the ability to evolve, accelerate when cells are maladapted to their environments: when stressed. Spontaneous DSBs induce the SOS DNA-damage response and are repaired by homology-directed DSB repair (HR repair, Figure 1E). The SOS response transcriptionally upregulates error-prone DNA polymerases (Pols) IV, V and II; but repair synthesis does not become mutagenic unless the σS response is also induced (Ponder et al., 2005; Shee et al., 2011) by starvation (Al Mamun et al., 2012) (Figure 1E). The σS response, by unknown means, allows use of, or persistence of errors made by, Pols IV, V and II in DSB repair causing mutations (Frisch et al., 2010; Ponder et al., 2005; Shee et al., 2011) near DSBs (Shee et al., 2012).
We find that most cipro-induced ampD and rpoB mutagenesis requires proteins used in starvation-stress-induced MBR (Figure 1F): DSB-repair proteins RecA, RecB, and RuvC, the SOS and general-stress-response activators, and SOS-upregulated error-prone DNA Pols IV, V, and II, implying a MBR-like mutagenesis mechanism. Isogenic strains grown at their appropriate MACs (Table S1) showed 87% ± 3% and 70% ± 9% decreases in cipro-induction of mutagenesis (AmpR and RifR combined) when carrying an SOS non-inducible lexAInd mutation or ΔrpoS (σS) deletion, respectively (mean ± 95% CI, Figure 1F). Thus both stress responses are required for the mutagenesis. Moreover, double-defective mutant cells show no further reduction (Figure S1E), indicating their action in the same mutagenesis pathway. Further, ROS and the σS response also act in the same mutagenesis pathway, in that scavenging ROS with thiourea (TU) caused no further reduction of mutagenesis to ΔrpoS cells (Figure S1F, and additional controls Figures S1 D and S2). We conclude that cipro-induced ROS-dependent mutagenesis occurs by the σS-dependent MBR pathway.
Moreover, the mutagenesis also requires DSBs and their repair. We visualized and quantified cipro-induced DSBs as fluorescent foci of the GamGFP DSB-end-specific binding protein (Shee et al., 2013), under our low-dose exposure conditions (8.5 ng/ml) and found that cipro increased GamGFP (DSB) foci by 28 ± 9 times above the spontaneous DSB level (mean ± SEM Figures 1G, Figure S3A,B, additional controls Figure S4A). We also show that GamGFP protein, which binds DSB ends and prevents their repair (Shee et al., 2013), inhibited cipro-induction of mutagenesis (Figure 1H), indicating a requirement for reparable DSBs in mutagenesis. Additionally, RecBCD, interacts specifically with DSB ends in DSB repair (Kuzminov, 1999), and its requirement in cipro-induced RifR and AmpR mutagenesis (Figure 1F, recB) also implies the necessity of DSBs in the mutagenesis. The data indicate that DSBs and DSB repair are necessary for mutation formation, and support a MBR mutagenesis mechanism.
The following data implicate specifically cipro-induced DSBs in the MBR mutagenesis. First, cipro antibiotic activity results from DSBs caused by inhibition of E. coli type II topoisomerases gyrase and topo IV mid-reaction (Drlica, 1999). In dose-response experiments, we find a tight correlation between cipro antibiotic and mutagenic activities (Figures 1I and S1G, r2 = 0.87, 0.88, Pearson correlation), correlating the DSB-induction (antibiotic) activity of cipro with its role in mutagenesis. Second, we used special gyrA* and parC* mutants that produce functional gyrase- and topo IV-mutant proteins that are not bound by cipro (Khodursky et al., 1995), and find no induction of mutagenesis with cipro (Figure 1J). These data show that cipro action on its target topoisomerases is needed for induction of mutagenesis, eliminate possible “off-target” effects of cipro on mutagenesis, and indicate a role for the cipro-induced DSBs in mutagenesis. Finally, aE and R-loop-promoting proteins, which promote starvation-stress-induced MBR by promoting spontaneous DSBs (Gibson et al., 2010; Wimberly et al., 2013) (Figure 1E) are not required for cipro-induced MBR (Figure S1H), supporting an MBR mechanism like that in starvation, except with the DSBs resulting from cipro inhibition of topoisomerases, rather than spontaneous sources. The data indicate that DSBs generated via cipro trigger the MBR pathway.
Together, the data support a σS-dependent MBR mechanism instigated by cipro-induced ROS and DSBs, allowing “MBR” to fill the previous mechanism void between ROS and mutations. The role of ROS might be contributing to the DNA breakage, as ROS do in antibiotic-killing mechanisms (Foti et al., 2012; Rasouly and Nudler, 2018; Zhao et al., 2015), action after stress-response induction, as in starvation-stress-induced MBR (Moore et al., 2017), or at another stage. The data below show a novel role for ROS in mutagenesis as signaling molecules that activate the general stress response, and, surprisingly, that this is limited to a transiently differentiated cell subpopulation.
ROS Differentiate a Cell Subpopulation, Activate σS Response
We surveyed cipro-treated single cells for induction of ROS, and the SOS and σS stress responses using flow cytometry. We measured SOS induction using an SOS-response-reporter gene, PsulAmCherry, engineered at a non-genic chromosomal site (Nehring et al., 2015; Pennington and Rosenberg, 2007), and found that cipro promotes SOS dose-dependently and uniformly among cells (Figure 2A), with 208 ± 26 times more SOS-positive cells at the 8.5ng/mL mutagenic dose than without cipro (mean ± SEM Figure 2A). Auto-fluorescence, which is induced by bactericidal antibiotics (Renggli et al., 2013), is negligible compared with the SOS (or the ROS or σS)-activity fluorescence signals (Figures S4B-D).
Surprisingly, low-dose cipro induced both ROS and the σS general stress response strongly in only a discreet cell subpopulation(s). In flow-cytometric assays of ROS in log-phase cipro-treated cells using the peroxide (H2O2)-sensitive dye dihydrorhodamine 123 (DHR, Figure 2B), high ROS levels appeared in only in a distinct 25% ± 6% cell subpopulation (8.5ng/mL, Figure 2B, mean ± SEM). Similarly, measuring σS activity with the yiaG-yfp fluorescence reporter (Al Mamun et al., 2012) in log-phase cipro-treated cells also revealed high σS activity in a discreet subpopulation of 22% ± 3% of the cells (Figures 2C and S3B, additional controls Figure S3C). For both ROS and the σS response, induction occurred above a threshold with low ROS or σS activity at doses below the 8.5ng/mL dose at which mutagenesis is assayed (Figure 2B, C). Growth inhibition, a known ROS-dependent process (Kohanski et al., 2007), also occurred above an 8.5ng/ml threshold (Figure S3D). A threshold is also seen in the kinetics of induction of translation of the rpoS mRNA (to σS protein) by three small RNAs (sRNAs) (Soper et al., 2010), examined below. Thus, despite uniform/unimodal, dose-dependent induction of DBSs (Figure 1G) and SOS (Figure 2A), fluoroquinolone induction of high-ROS and the general stress response occurs in only a well separated subpopulation of ~20% of exposed log-phase cells, and these have very high ROS and σS-activity levels, respectively (Figure 2B, C). The subpopulation is smaller, near 10%, when the cells reach stationary phase (Figure 2D), when mutagenesis is assayed.
We found that ROS are required for, and promote mutagenesis by, activation of the σS response. First, ROS scavenging/preventing agents TU or BP blocked induction of σS-response activity removing the σS-high cell subpopulation (Figures 2D and S3E), reduced the accumulation of σS-protein (western blotting, Figure 2E), or of a σS-β-galactosidase fusion protein (Figure S3F) from an rpoS-lacZ reporter, indicating that ROS are required for induction of the σS response. Additional controls (Figure S5A and B) show that ROS are not generally needed for fluorescence, transcription, or protein accumulation, just for activation of the σS response by cipro. We conclude that σS-response induction by cipro requires ROS.
Furthermore, ROS promote cipro-induced MBR not by creation of DSBs, but instead, by induction of the σS response in a transiently differentiated cell subpopulation. The role of ROS in antibiotic (growth-inhibitory) activity (Table S1) is creation of DNA breaks via oxidized guanine (8-oxo-dG) in DNA (Dwyer et al., 2007; Foti et al., 2012; Kohanski et al., 2007). By contrast, we observed that reduction of cellular ROS levels with TU or BP, though profoundly inhibitory to the MBR mutagenesis (Figure 1D), did not reduce induction of DSBs by low-dose cipro, quantified as GamGFP foci (same TU and BP concentrations as mutation assays, Figure 2F and S3G). Neither did TU or BP diminish the SOS DNA-damage response (Figure 2G), implying that ROS promote mutagenesis independently of damage to DNA. Moreover, 8-oxo-dG incorporation appears not to be the principal role of ROS in the mutagenesis in that ROS-mediated 8-oxo-dG-signature mutations [G·C➝T·A and A·T➝C·G, (Schaaper and Dunn, 1987)] are a less important part of cipro-induced than spontaneous forward-mutations (ampD, Figure S1C). These data indicate that ROS promote mutagenesis other than by promoting DSBs or SOS-responsive DNA damage, or base misincorporation opposite oxidized guanine during DNA replication.
Further, the main or only role of ROS in cipro-induced MBR mutagenesis is σS induction, in that artificial/engineered upregulation of σS fully substituted for ROS in mutagenesis (Figures 2H and S5C). RifR mutagenesis was restored in the presence of TU by IPTG induction of σS (Figures 2H). ROS and the σS response also act in the same mutagenesis pathway (Figure S1E). The data indicate that ROS are needed for MBR only or mostly for induction of σS, such that if σS is otherwise supplied, ROS are no longer required for the mutagenesis (Figure 2H,I). Importantly, cipro induction of SOS, ROS, and σS activities all require cipro interaction with its topoisomerase targets, in that cells with active but cipro-non-binding mutant gyrA* and parC* alleles (Khodursky et al., 1995) showed no induction of SOS, ROS, or σS responses by cipro (Figure S3H-J, respectively). These data demonstrate that the events that lead to SOS, ROS, and σS induction begin with cipro interaction with its topoisomerase targets.
Together, these data show that cipro action on topoisomerases leads to induction of high ROS levels in a discreet cell subpopulation (Figure 2B), that the ROS activate σS in a subpopulation (Figure 2C-E), and that activation of the σS response is how ROS promote cipro-induced MBR (Figure 2H, I). This constitutes a novel role for ROS in mutagenesis—signaling induction of the σS general stress response—unlike those in antibiotic activity (Dwyer et al., 2007; Foti et al., 2012; Kohanski et al., 2007) or starvation-stress-induced MBR (Moore et al., 2017).
σS-active Gambler Cell Subpopulation Generates Mutants
We used fluorescence-activated cells sorting (FACS) to demonstrate that the small σS-response-high cell subpopulation, which encompasses 13% (± 1%) of the stationary-phase cells used in mutagenesis assays, produces most cipro-induced mutants (Figure 3). We sorted σS high- and low-activity cells to at least 97% enrichment (Figures S6A-C). Remarkably, whereas unsorted and mock-sorted cells show (mean) 25 ± 3-fold induction of RifR mutant frequencies by cipro (Figure 3A), the sorted σS-high cells displayed a 400 ± 7-fold induction of RifR mutagenesis—16 ± 3-times higher than unsorted and mock-sorted cells (Figure 3A, controls for the sorted populations Figures 3B, S6D, E, and S7A, B, Supplemental Discussion S1). The large σS-low-activity cell subpopulation, 87% ± 1% of cells, showed only 3 ± 1-fold induction of RifR mutagenesis by cipro, or 8 ± 2-times lower than unsorted and mock-sorted cells (Figure 3A), indicating that most mutants did not arise in the majority cell subpopulation. We can estimate the contribution of each subpopulation to total yield of mutants as follows. Because the σS low-activity cells display only a 3 ±1-fold increase in RifR mutants (Figure 3A), we can conclude that the σS-low cells produced about 12% of the mutants (3-fold increase / 25-fold increase in un/mock-sorted = 12%, Figure 3A). Because the σS-low cells will be contaminated with some σS-high cells, this means that at least 88% of RifR mutant yield originates in the σS high-activity cell subpopulation. The data demonstrate that most of the cipro-induced RifR mutants originate in the small σS-high cell subpopulation.
We excluded the possibility that the greater production of mutants by σS-high cells might result indirectly from their high fluorescence (possible high metabolic activity), using a fluorescence-reporter gene not controlled by σS: Placcfp (cyan fluorescent protein, Figure 3A, above autofluorescence, Figure S4E). Additional controls for sorted cells’ growth rates / colony formation are shown in Figures S2A,B.
Further, we find that the hypermutability of the σS-high cell subpopulation that generates RifR mutants appears to be transient, and not a heritable mutator state, as, for example, a “mutator” mutation would confer. RifR mutants recovered are not heritably mutator (Figure S5D).
Collectively, the data demonstrate that a small, transiently differentiated subpopulation of σS high-activity cells is transiently hypermutable and generates most cipro-induced mutants with de novo rifampicin-(cross)-resistance mutations. These data suggest a potential “bet-hedging” developmental strategy (Norman et al., 2015; Veening et al., 2008) that may allow evolution while risking mutagenesis in only some cells; we call these gambler cells. How the gambler cell subpopulation is differentiated, and a drug that prevents it, follow.
FDA-approved Drug Inhibits Evolvability
The σS-response high-activity cell subpopulation may be considered a novel therapeutic target for potential inhibition of cipro-induced mutagenesis to antibiotic cross resistance. We show that the drug edaravone, an ROS scavenger indicated for human use in ALS in the U.S. and cerebral infarction in Japan (Miyaji et al., 2015; Watanabe et al., 2018), inhibits cipro induction of mutagenesis but not its antibiotic (killing) activity. Edaravone, at concentrations seen in new formulations (100μM) (Corporation, 2014; Li et al., 2012; Parikh et al., 2016), inhibited the appearance of σS-high cells (Figure 3C), accumulation of σS-fusion protein (Figure S3F), appearance of ROS-high cells (Figure 3D), and most (82% ± 1% of) RifR mutagenesis (Figure 3E). Edaravone did not affect cipro induction of DSBs (Figure 3F), SOS activation (Figure 3G), cell growth (Figure S2A), colony formation (Figures S2B), or negative-control β-gal activity (Figure S5B), implying that its inhibition of mutagenesis reflects specific inhibition of σS-response activation (Figure 3I). Importantly, edaravone did not alter the ability of high-dose cipro to kill E. coli (Figure 3H), showing that edaravone can reduce mutagenesis induced by cipro without altering its utility as an antibiotic. These data serve as a proof-of-concept for small-molecule inhibitors that could be administered with antibiotics to reduce resistance evolution, by impeding differentiation of σS-response-active gambler cells, without harming antibiotic activity. We explored the basis of differentiation of the σS response-high cell subpopulation—how ROS activate the σS response in the subpopulation cells (Figure 3I)—as follows.
ROS-high Cells Become Gamblers via sRNAs
The ROS-high subpopulation cells could, in principle, induce σS activity in other cells, or themselves, or both. We distinguished these possibilities by following single cells over time through induction of ROS then the σS response, using fluorescence reporters, flow cytometry and time-lapse microscopy after cipro (Figure 4).
We found that ROS-high cells appear before σS-active cells. The high-ROS cell subpopulation is detectable with ROS dye DHR (green) and flow cytometry at 4 hours after cipro is added, when no σS activity from a σS-response fluorescence reporter (red) is detectable (4h, Figure 4A, Supplemental Discussion S2). Then, double-positive cells, dyed for ROS (green) and σS activity (red) develop between 8 and 16 hours (Figures 4A and S5E), implying that at least some σS-high cells begin as ROS-high cells. At 24h, the time at which cells were harvested for sorting (Figure 3A)/mutagenesis assays (Figures 1A,B), many double-positive ROS-/σS-high cells (upper right quadrant, Figure 4A 24h), and also some σS-high single-positive cells, were present (lower right quadrant, Figure 4A 24h). Whether the σS-single-positive cells at 24 hours originated from (had been) ROS-high cells before 24 hours was unclear. We used live-cell imaging with fluorescence-reporter genes (green) for two different oxidative stress responses, in cells that also carry the red σS-response reporter, to follow individual cells over time from their burst of ROS to σS-response induction. The reporters are transcriptional GFP fusions (Zaslaver et al., 2006) for oxyR (peroxide) and sodA (superoxide) responses, and both show double-positive and some σS-single-positive cells at 24 hours (Figure 4B, green, the peroxide control discussed Supplemental Discussion S3).
Using the sodA reporter with live-cell time-lapse deconvolution microscopy, we show that essentially all red σS-active cells arose from oxidative-stress-response-activated green cells (> 99%, Figure 4C and Movie S1). Some of the σS-response-induced (red) cells showed decreased ROS (green) after σS-response induction (Figure 4C and Movie S1), suggesting amelioration of high ROS levels by the σS response. The data demonstrate that cipro induces MBR by activating differentiation of a subpopulation of ROS-high cells that become mutable, σS-response-high gamblers that generate most of the antibiotic cross-resistant mutants (Figure 3A).
We investigated how ROS activate the σS response in subpopulation cells (Figure 5). σS is regulated at multiple levels (Battesti et al., 2011) including translational upregulation by small RNAs (sRNAs). The ArcZ, RprA, and DsrA sRNAs activate σS translation assisted by the Hfq RNA chaperone (Battesti et al., 2011). We found that DsrA and ArcZ, but not RprA, are required for both cipro induction of σS protein (Figure 5A), and differentiation of the σS-high gambler subpopulation (Figure 5B). The Hfq RNA chaperone is also required for cipro induction of σS protein (Figure 5A), σS-response activity (Figure 5B), and RifR mutagenesis (Figure 5C). Moreover, the requirement for Hfq in RifR mutagenesis can be substituted by artificial upregulation of σS from a plasmid, which restored 86% ± 10% of RifR mutagenesis to Δhfq cells (Figure 5C, controls Figure S2A,B). The data indicate that the Hfq RNA chaperone promotes cipro induction of mutagenesis mostly or wholly by promoting σS-response induction, presumably via the ArcZ and DsrA sRNAs. Knock out of hfq in ΔarcZ or ΔdsrA cells causes no further reduction in σS-β-galactosidase (Figure 5A), supporting this role of Hfq. Furthermore, we found that cipro induced dsrA and arcZ transcription by 2.3 ± 0.3- and 53 ± 3-fold, respectively in log phase (Figure 5D), shown with transcriptional lacZ fusions to the promoters of the dsrA and arcZ genes (Mandin and Gottesman, 2010; Sledjeski et al., 1996); and this transcriptional upregulation was ROS dependent, and was reduced by ROS reducers TU, BP, and edaravone (Figure 5D). The data demonstrate that cipro-induced ROS promote transcription of sRNAs DsrA and ArcZ, which, with RNA chaperone Hfq, upregulate σS, activating the general stress response. Thus, these sRNAs underlie the differentiation of ROS-high cells into the σS-active gambler subpopulation (Figure 5E) that generates antibiotic cross-resistant mutants.
There may be an additional component of σS upregulation by inhibition of σS-protein degradation. One of the multiple ways that the σS response is kept “off” in unstressed cells is via RssB, which delivers σS protein to the ClpXP protease for degradation. Using a rpoS-lacZ reporter that makes a σS-β-galactosidase fusion protein with an intact RssB-binding region (Zhou and Gottesman, 2006), we see that deletion of rssB increased σS-β-galactosidase activity in cells untreated with cipro, but not in cipro-treated cells (Figure 5A), implying that detectable RssB-mediated σS degradation ceases after cipro treatment. The data could mean either that σS-protein degradation may be inhibited by cipro/ROS, or that the upregulation of translation of σS by the DsrA and ArcZ sRNAs might cause saturation of RssB-mediated σS-protein degradation.
ROS Induced via SOS Response and Ubiquinone
Although antibiotics including fluoroquinolones induce ROS in cells (Dwyer et al., 2015), the ROS-induction pathway is only partly characterized (indicated by a “?” left of ROS in Figure 3I). Our data above show that cipro interaction with its target topoisomerases (DSB formation) is required for ROS formation (Figure S3I), that ROS arise in a cell subpopulation (Figure 2B), and previous work implicated Fenton chemistry and components of electron transfer (Dwyer et al., 2015). We discovered that the SOS response and ubiquinone oxidoreductase are required for induction of the ROS-high cell subpopulation by cipro.
First, we examined mutagenesis in cells defective for components of three electron-transfer chain (ETC) protein machines shown to promote the σS response during starvation-stress-induced MBR: NuoC (ubiquinone oxidoreductase I, an ETC “complex I” subuint), UbiD (biosynthesis of ubiquinone), and CyoD (a subunit of cytochrome bo’ oxidase, an ETC “complex II” subunit) (Al Mamun et al., 2012). Whereas CyoD and NuoC were not required for RifR or AmpR mutagenesis, UbiD (ubiquinone) was required for most of both (Figure 6A, controls Figure S2A,B). Moreover, UbiD (and ubiquinone) appear to act upstream of σS-response induction in mutagenesis, in that artificially induced production of σS substituted for UbiD, restoring most or all of RifR mutagenesis to ΔubiD mutant cells (87% ± 16% restored, Figure 6B). We observed reduction of σS accumulation and σS activity in ubiD null-mutant (ubinquinone-deficient) cells (Figure 6C and 6D), demonstrating that ubiquinone, and by implication, electron transfer, are required for cipro induction of the σS response. By contrast, UbiD/ubiquinone was not required for SOS-response activity (Figure 6E). Together, the data show that ubiquinone is required for cipro-induced mutagenesis, which it allows by promoting induction of the σS response.
We found that the ubiquinone role is specifically in the induction of ROS. Ubiquinone, or coenzyme Q, functions in the aerobic ETC by mediating oxido-reduction cycles required for ATP energy production (Meganathan and Kwon, 2009). We found that ubiD-defective cells showed severely reduced ROS generation in cipro compared with wild-type cells (32% ± 9% ROS-high cells in wild-type, and 8% ± 4% in ubiD-null cells, Figure 6F). ubiD-null-mutant cells also displayed reduced katG-lacZ activity, a reporter activated by H2O2 (Liu and Imlay, 2013) (Figure 6G). The data show ubiquinone-promoted induction of ROS, which are required for the cipro induction of the σS response.
Perhaps surprisingly, the SOS response is also required for cipro induction of ROS and the σS response. SOS-non-inducible (lexAind-) cells, and cells lacking RecB, which is required for SOS induction by DSBs (McPartland et al., 1980), showed reduced induction of ROS (Figure 6H) and the σS response (Figure 6I) by cipro, contributing to at least 70% ± 4% of ROS-high subpopulation cells (Figure 6H). Because ubiquinone was not needed for SOS-response induction by cipro (Figure 6E), we can infer that SOS acts upstream of, or in parallel with, ubiquinone in ROS induction (Figure 6H); not downstream of ubiquinone, which is not needed for SOS induction (Figure 6E). The SOS response, induced by UV light, was reported to inhibit aerobic respiration (Swenson and Schenley, 1974), and, also in assays without cipro, slowing of respiration increased autoxidation of quinols leading to superoxide production (Gonzalez-Flecha and Demple, 1995; Skulachev, 1998). Together with our data using cipro, these data support a model in which SOS activation may inhibit the ETC leading to ROS generation (Figure 6J). SOS action upstream of the ubiquinone contribution to ROS generation (Figure 6J) is harmonious with our data (Figure 6E). The data support a model in which the cipro-induced SOS response may inhibit/slow aerobic respiration, per (Swenson and Schenley, 1974), in only a subpopulation of cells, allowing autoxidation of ubiquinone to generate high ROS levels in those cells (Figure 6J).
Multi-Chromosome Cells Allow Evolvability
We prevented cipro from inducing multi-chromosome cells by deletion of the sulA gene (Figure 7), the product of which is induced by SOS and inhibits cell division, causing multichromosome cells during a DNA-damage response (Huisman and D’Ari, 1981). SulA inhibits polymerization of the microtubule-like “Z-ring”, which pinches off daughter cells (Bi and Lutkenhaus, 1993). Because SulA does not block DNA replication or elongation of the rodshaped E. coli cells, cells grow long and “filamentous” or snake-like with multiple chromosomes in them (Huisman and D’Ari, 1981) when SOS and SulA are induced (Figure 7A and B). We used microscopy and a protein that marks the chromosome as a fluorescent focus (Figure 7A-C) (Joshi et al., 2013), to show that ΔsulA cells do indeed make much shorter cells in low-dose cipro (Figure 7D-F), that these have fewer chromosomes per cell (Figure 7E,F), and are deficient in mutagenesis (Figure 7G). Without cipro only 1% ± 0.7% of exponential wild-type cells have four or more chromosome copies (Figure 7C), so we defined a multi-chromosome cell as those with ≤4 chromosome copies. With cipro, 33% ± 2% of wild-type cells have ≥4 chromosome copies (Figure 7B). By contrast, ΔsulA cells show much reduced cell length and chromosome content (Figure 7D-F). We find that these ΔsulA shorter cells show reduced cipro-induced RifR and AmpR mutant production (Figure 7G), showing 67% ± 5% and 70% ± 5% fewer mutants, respectively. Further, deletion of the ruvC HR-repair gene from ΔsulA mutant cells caused no further decrease in mutagenesis, indicating that SulA acts in the same HR-dependent MBR mutation pathway as RuvC. Thus, SulA, and the filamented, multi-chromosome state, are required for cipro-induced MBR.
When exposed to low-dose cipro, E. coli forms long, multi-chromosome cell “filaments” that “bud off” small, normal-length daughter cells that produce high frequencies of cipro-resistant mutants (Bos et al., 2015). These data suggested that multi-chromosome cells might promote adaptation by coupling mutagenesis, which can generate resistance but also deleterious mutations, with recombination or allele sharing, which might mitigate deleterious effects of many recessive mutations, allowing the multi-chromosome cell to produce resistant, surviving/adapted progeny (Bos et al., 2015). We tested directly whether the multi-chromosome state is required for cipro-induced mutant production, and explored whether it could promote adaptation via mutagenesis.
We used mathematical modeling to address the hypothesis of Bos et al. (2015) that multichromosome cell filaments have an advantage over normal cells in adaptation to changing (stress) environments via mutagenesis. We formulated a mathematical model to test possible benefits of multi-chromosome cell filaments for rapid adaptation (Methods). Results of the model, presented in Figure 7H, show that increasing filament mutation rate can greatly increase the probability of both adaptation and survival of a chromosome in multi-chromosome filaments relative to non-filamented cells. This supports recent work showing that cooperation can accelerate complex adaptations (Obolski et al., 2017). Further, the advantage of the multi-chromosome state increases with increasing selection coefficient (Figure 7H). Selection coefficient represents, for example, the lethality of cipro in cipro-treated cells, which are under selection for cipro-resistance mutations. This model demonstrates that the multi-chromosome state is capable of facilitating adaptation by mutagenesis, illustrated in a model in Figure 7I.
DISCUSSION
The mechanism of mutagenesis induced by the fluoroquinolone antibiotic cipro, revealed here (model, Figure 7I), demonstrates three new biological principles in mutagenesis (three headings below), and also unites quinolone-induced mutagenesis with a large canon of stress-induced mutagenesis mechanisms. Stress-induced mutagenesis mechanisms, from bacteria to human, are defined as mutation-producing mechanisms upregulated by stress responses (Fitzgerald et al., 2017). Activation of mutagenic DNA break repair (MBR, Figures 1E-J) by the general stress response (Figures 1F, 2C-E, H,I, 4, 5, 7I, couples mutagenesis to the σS response (Fitzgerald et al., 2017; Harris et al., 1994; Ponder et al., 2005; Rosenberg et al., 1994) causing mutations during stress. Fluoroquinolones were not known previously to provoke stress-induced/aS-dependent mutagenesis. Temporal regulation of mutagenesis by stress responses causes mutant generation preferentially when cells or organisms are maladapted to their environments—when stressed—potentially accelerating adaptation (Fitzgerald et al., 2017; Ram and Hadany, 2012, 2016).
A Regulatory Role for ROS in Mutagenesis
We discovered a novel role of ROS in mutagenesis—their activation of the general stress response (Figures 2D, E, H, 3C, D, 4, 5), which allows stress-inducible MBR (Figure 1E,F and model, Figure 7I). ROS have long been known to promote mutagenesis by other more direct mechanisms, including by oxidation of guanines in nucleotide pools and in DNA to 8-oxo-dG. 8-oxo-dG pairs with A, causing G-to-T and T-to-G (A-to-C) mutations (Schaaper and Dunn, 1987), a signature that is less important for cipro-induced than spontaneous forward mutations (ampD, Figure S1 B,C), indicating a minor role. DNA base mispairs of 8-oxo-dG with A are also attacked by base-excision repair, which can lead to DNA breaks that are part of various antibiotics’ killing mechanisms (Foti et al., 2012; Kohanski et al., 2007; Rasouly and Nudler, 2018; Zhao et al., 2015), but we show are not an important source of the DSBs that drive cipro-induced MBR (Figure 2F,G). By contrast, we found that, surprisingly, the mutagenic role of ROS can be fully substituted by engineered production of σS, the transcriptional activator (bacterial RNA-polymerase sigma factor) of the general/starvation stress response (Figure 2H), indicating that induction of the σS response is the predominant or only role of ROS in cipro-induced mutagenesis; if σS is supplied artificially, ROS are no longer needed. We found that ROS induce transcription of ArcZ and DsrA small RNAs (sRNAs, Figure 5), which, assisted by the Hfq RNA chaperone (Figure 5A-C), promote translation of the rpoS mRNA into σS protein (Battesti et al., 2011) (Figure 7I). This differs from a role of Hfq and another sRNA in mutagenesis via direct downregulation of translation of an mRNA encoding a mismatch-repair protein (Chen and Gottesman, 2017). We saw that induction of ROS by cipro precedes and is required for σS-response activation (Figure 4), and that cells with high levels of ROS become σS-response highly activated cells (Figure 4C, movie S1) that generate the mutants (Figure 3). These data highlight the centrality of stress-response-control of mutagenesis, even ROS-induced mutagenesis, and identify ROS as signaling molecules in this regulation. The DNA crosslinking agent mitomycin C also induces ROS that induce the σS response (Dapa et al., 2017), which might occur by similar mechanism with similar mutagenic consequences as shown here.
It is conceivable that several other ROS-promoted mutagenesis mechanisms may involve ROS upregulation of the general stress response, which can be mutagenic by allowing MBR (Lombardo et al., 2004; Ponder et al., 2005; Shee et al., 2011), downregulation of mismatch repair (Gutierrez et al., 2013), transposon movement (Ilves et al., 2001), and possibly other mechanisms. Because stress-response regulators, such as σS, are non-redundant hubs in the MBR network (Al Mamun et al., 2012), these are attractive candidates for targets of proposed drugs to slow evolution of pathogens, averting evolution of resistance and evasion of immune responses (Al Mamun et al., 2012; Fitzgerald et al., 2017; Rosenberg and Queitsch, 2014). Moreover, the FDA-approved antioxidant drug edaravone behaved as such an “anti-evolvability” drug in our experiments (Figure 3C-E) and did so without reducing the killing power of cipro as an antibiotic (Figure 3H), providing a promising proof-of-concept. Reactive oxygen affects many aspects of the biology of cells from lipid and protein oxidation (Pradenas et al., 2012; Tamarit et al., 1998) to DNA damage and mutagenesis (reviewed above). The addition of stress-response activation causing MBR to this list identifies ROS as evolution-promoting signaling molecules.
Mutagenesis in Transiently Differentiated Gamblers
Our data reveal that ROS, and then the general stress response, are induced strongly in a 1020% subpopulation of cipro-treated cells (Figures 2B-D and 4) that is transiently differentiated into a mutable state (Figure S5D), and produces most of the mutants (Figure 3A,C). Transient differentiation in bacterial subpopulations is a recognized potential evolutionary “bet-hedging” strategy, in which only some cells take the risk of “trying” a phenotype that may be advantageous in some environments and maladaptive in others (Norman et al., 2015; Veening et al., 2008). Transient phenotype examples include the persister state, which allows tolerance of lethal drugs but slows or halts proliferation (Balaban et al., 2004), competence for “natural transformation”— DNA uptake and incorporation into the genome (Chen and Dubnau, 2004), sporulation—a dormant but environmentally resistant state (Norman et al., 2015), and even programmed cell death (Amitai et al., 2009; Gonzalez-Pastor et al., 2003), hypothesized (Amitai et al., 2009; Rosenberg, 2009) or demonstrated (Gonzalez-Pastor et al., 2003) to benefit siblings of the sacrificed cells. The regulated/”programmed” limitation of mutagenesis itself—a major evolutionary driver—to a cell subpopulation appears to embed the apparent evolutionary strategy of environmentally tuned mutagenesis within another evolutionary strategy: “bet hedging” (Norman et al., 2015; Torkelson et al., 1997; Veening et al., 2008). Though transiently hypermutable cell subpopulations have been hypothesized (Hall, 1990; Ninio, 1991), supported by genetic evidence (Torkelson et al., 1997), and cells with stress responses linked to mutagenesis of unknown mechanism (Woo et al., 2018), our data provide the first isolation (Figure 3) of a hypermutable cell subpopulation in the act of mutagenesis, and show the defining, differentiating inputs: ROS and general stress-response activation (Figures 3 and 4) as well as the mutagenesis mechanism: MBR, illustrated in Figure 7I. We found that the subpopulation is relatively large at 10-20%, and that its transient differentiation (Figure S5D) is achieved by stress-response activation in the subpopulation cells, cell autonomously (Figure 4)—all novel mechanisms of potential means of promotion of the ability to evolve. Unlike “persisters,” these cells take the risk of inducing mutations, which can lead to heritable resistance to never-beforeencountered antibiotics, and so could be called “gamblers.”
Multi-chromosome Cells Promote Evolvability
We found that the multi-chromosome state, caused by the SOS-response-induced SulA inhibitor of cell division (Huisman and D’Ari, 1981), is required for cipro-induced mutagenesis (Figure 7). We observed SOS-dependent multi-chromosome cell “filaments” previously during low-dose cipro exposure and noted their “budding” off of small cells that produce high frequencies of cipro-resistant mutants (Bos et al., 2015). Here, we showed that “filamentation” is required for mutant production (Figure 7A-G). For cipro-induced mutagenesis by mutagenic repair of DNA doublestrand breaks (Figure 1E,F), induced here by cipro (Figure 1G-J, 2F and S1G), more than one chromosome is needed for DSB repair. Our modeling indicates that the multiple chromosomes may additionally facilitate survival and adaptation to stresses of highly mutating cells by cooperation (Obolski et al., 2017): sharing of gene products and/or alleles (recombination) between the chromosomes, masking deleterious recessive phenotypes (Figure 7H). Our data imply that “filamenting” cells may be biomarkers of rapid evolution. Bacteria like Bacillus subtilis undergo natural transformation—acquiring sibling cells’ DNA—simultaneously with, and activated by same Com stress response that upregulates a stress-induced mutagenesis mechanism (Sung and Yasbin, 2002). Thus, B. subtilis engages the adaptation-boosting combination of recombination and mutagenesis (Lenhart et al., 2012). Our data indicate that E. coli, which is famously incapable of natural transformation, may employ the same adaptation-accelerating strategy via multiple sibling chromosomes within one cell, rather than sibling DNA taken up exogenously. The data suggest that in addition to targeting stress-response regulators as an anti-evolvability drug strategy (Al Mamun et al., 2012; Fitzgerald et al., 2017; Rosenberg and Queitsch, 2014) and Figure 3C-H, dividing (and conquering) the multiple chromosomes might also prove to be an effective anti-evolvability, anti-pathogen therapeutic strategy.
AUTHOR CONTRIBUTIONS
JPP, LGV, OL-E, JB, RHA, CH, LH, and SMR conceived the project, advanced hypotheses and/or designed experimental approaches; JPP, LGV, YZ, AW, JL, JX, QM performed or guided the work; DB provided advice/assistance, JPP, PH and SMR wrote the manuscript.
STAR★Methods
Key Resources Table
CONTACT FOR REAGENTS AND RESOURCE SHARING
The corresponding author, Susan M. Rosenberg (smr{at}bcm.edu), is the contact for reagents and resource sharing.
EXPERIMENTAL MODEL AND SUBJECT DETAILS
Escherichia coli (strain MG1655) and isogenic derivatives were used for all experiments.
METHODS DETAILS
Strains, Media, and Growth
The Key Resources Table lists strains used in this study. Bacteria were grown in LBH rich medium (Torkelson et al., 1997) at 37°C with aeration, and additives where indicated at the following concentrations: ciprofloxacin (cipro, 1–24 ng/mL, Table S1), ampicillin (100 μg/ml), chloramphenicol (25 μg/ml), kanamycin (50 μg/ml), tetracycline (10 μg/ml), rifampicin (110 μg/ml), and sodium citrate (20 mM).
Assays for Ciprofloxacin-induced Mutagenesis
Saturated overnight LBH cultures, started each from a single colony, were diluted 1:4×106 into 25 ml in a 250ml flask in fresh LBH broth and incubated at 37°C with shaking for 3–3.5 h, then diluted 1:3 into fresh LBH broth (“no-cipro” controls) or into LBH with cipro at a final “sub-inhibitory” concentration minimal antibiotic concentration (MAC) that caused a final cfu titer of 10% of the titer observed in the no-cipro control, by the final 24h or 48h time point (fluctuation tests, below). This concentration was determined individually for each experimental strain. For dose-response fluctuation tests, the final cipro concentrations were 1, 2, 4, 8.5, and 12 ng/ml.
For all fluctuation tests, between 10 and 60 1-ml aliquots of cultures diluted 1:3 were dispensed into 96-deep-well plates or 14-ml tubes and incubated at 37°C with shaking. After 24h (RifR) or 48h incubation (AmpR), samples were plated onto LBH agar for determination of total viable cell titers or selective LBH-agar plates containing rifampicin (110 μg/ml) or ampicillin (100 μg/ml) to select mutants resistant to each drug. Total and resistant cfu were counted, and mutation rates (mutations per cell per generation) estimated with the MSS-MLE algorithm using the FALCOR calculator (Hall et al., 2009). The fold change in cipro-induced mutagenesis for each strain was determined as the ratio of the mutation rates of the treated divided by the untreated control samples.
For fluctuation tests performed with addition of reagents that reduce reactive oxygen, the final concentrations were 100 mM for thiourea, 0.25 mM for 2,2’-bipyridine, and 100 μM for edaravone. For assays in which GamGFP was produced to trap double-strand breaks (DSBs) (Shee et al., 2013), GamGFP was induced from the chromosome using 10 and 20 ng/mL doxycycline in LBH liquid or in plates, as used for determining cfu/ml. Plasmids for σS artificial upregulation and the empty-vector control were obtained from the mobile plasmid collection (Saka et al., 2005), and were induced with 30 μM isopropyl β-D-1-thiogalactopyranoside (IPTG) present throughout growth, except in the plates used to determine RifR or total cfu/ml. σS production was confirmed by western blotting.
Reconstruction Experiments
Reconstruction experiments were performed to verify that differences in cipro-induced mutant cfu titers observed between wild-type and various mutant strains were not caused by differences in colony-formation efficiency under exact reconstructions of selection conditions: selective plates with varying amounts of isogenic sensitive neighbor cells (108 or 109). About 100 cfu of ampicillin-resistant ampRC ΔampD cells or rifampicin-resistant rpoB Δ1687C, rpoB Δ1593-1598, rpoB A1547T mutant cells of each experimental strain genotype were mixed with ~109, ~108 or ~107 isogenic sensitive neighbor cells and plated onto ampicillin or rifampicin selective plates, respectively, and their numbers and speed of forming colonies scored. These platings reconstruct the experimental conditions in which mutant cells form colonies scored in our Assays for Ciprofloxacin-induced Mutagenesis. Resistant mutants were also plated alone for reference. For each strain, we quantified cfu observed after 24 h (ampicillin) or 48 h (rifampicin) at 37°C. Two replicates for each culture condition were performed per strain. See figure legends for numbers of independent experiments.
Competition Experiments
Cultures of sensitive and resistant mutants of each experimental strain genotype were mixed at a 50:50 ratio and grown per fluctuation tests, then plated at the end of the growth period on selective rifampicin or ampicillin medium and non-selectively, to obtain the final ratios of sensitive and resistant cfu after growth in competition. Pure cultures were also established as controls. These experiments showed that neither RifR not AmpR mutants is selected (wins the competition ending over 50% of cfu), and both are actually significantly counter-selected relative to their sensitive parent strains (e.g., Figure 1C and legend). These data indicate that all of our estimates of the induction of mutagenesis to RifR and AmpR are underestimates. See figure legends for number of independent experiments.
Flow Cytometric Assays for σS- and SOS-Response-Regulated Promoter Activity
Quantifications of cells that have induced their σS or SOS responses, and how much they have, were achieved using engineered chromosomal fluorescence reporter genes and flow cytometry, per (Nehring et al., 2015; Pennington and Rosenberg, 2007) for SOS, and per (Al Mamun et al., 2012) for σS-response activation. We used the yiaG-yfp σS-response reporter (Al Mamun et al., 2012) and the Δattλ::PsulAmCherry SOS reporter (Pennington and Rosenberg, 2007) modified by Nehring et al. (Nehring et al., 2015) in separate strains grown under fluctuation-test conditions as described for Assays for Ciprofloxacin-induced Mutagenesis, with or without cipro, at indicated concentration/s, and harvested the cells in late log phase or stationary phase. For quantification, flow cytometry “gates” were calibrated, for SOS, using the negative-control SOS-off lexA(Ind-), and SOS-response proficient cells, per (Pennington and Rosenberg, 2007) as the dividing place between peaks of the bimodal distribution of SOS-proficient cells at which most cells, diverge from the spontaneously SOS-induced fluorescent cell subpopulation, usually at between 0.5% and 1% of cells for cells cultured in LBH broth. Essentially all SOS-non-inducible recA or lexAInd- cells fall below this gate (~10-4 of them cross the gate). Cells that fell below this gate (less fluorescence) were scored as SOS negative, and above the gate as SOS positive. For the σS response, gates for σS-high activity cells were set to the point at which fewer than 0.5% of cells with cipro but without the reporter gene were positive. At this gate fewer than 10-3 of ΔrpoS cells, which are deficient in σS-response induction, cross the gate and would be scored as positive. For all, the percent of the population that scored as positive is reported.
Fluorescence-Activated Cell Sorting
Cell sorting was performed using a FACS Aria II cell sorter (BD Biosciences, San Jose, CA) with a 70-μm nozzle. E. coli cells were identified using forward and side scatter parameters, and these were sorted using sterile 1X phosphate buffered saline (PBS) as sheath fluid. After treatment with cipro, yellow fluorescent protein-positive (σS activity, yiaG-yfp) and non-fluorescent cells were sorted into 14 mL conical tubes (20-30×106 negative cells and 3-8×106 positive cells) and plated on LB agar with and without rifampicin to determine cfu/mL (per Assays for Ciprofloxacin-induced Mutagenesis, above). These data were used to calculate RifR mutant frequencies in the sorted σS high-activity, σS low-activity, unsorted, and mock-sorted populations, the last being cells run through the machine and all cells collected. Control sorts for cyan fluorescent protein, encoded by the Placcfp gene, a negative control for metabolically active cells, and mutagenesis assays, were performed similarly in parallel with the experimental sorts.
HPII Catalase Activity
HPII (σS-dependent catalase) activity was measured as described (Iwase et al., 2013). The viable cell titers (cfu/mL) of cells growing in LBH broth were determined at appropriate time points in log or stationary phase. HPI catalase was inactivated by heating 100 μL culture aliquots at 55°C for 15 min. After inactivation, 100 μL 30% H2O2 and 1% Triton-X 100 (sigma) were added. After an additional 15 min incubation, the height of bubble formation was measured in millimeters. The millimeters of bubbles were then normalized to cfu/mL of cells. Controls in ΔrpoS cells demonstrated that these assays report on σS-response-dependent catalase activity.
Microscopy and Quantification of GamGFP (DSB) and TetR-mCherry (Chromosome) Foci
Cells containing the chromosomal inducible GamGFP cassette were diluted 1:4×106 into 250mL flasks and grown for 3 h. These were then diluted 1:3 in media with or without cipro (1-8.5 ng/ml). GamGFP, a DNA DSB-specific binding protein that traps DSBs and inhibits their repair (Shee et al., 2013), was induced in late log phase using 40 ng/mL of doxycycline. After 2 h of induction, cells were fixed with 1% paraformaldehyde and placed at 4°C until microscopy images were taken. Cells containing the inducible TetR-mCherry plasmids and the tetO chromosomal array were diluted 1:4×106 2.5μl into 10ml into 250mL flasks and grown for 3 h. These were then diluted 1:3 in media with or without cipro (MACs). The TetR-mCherry protein binds to the chromosomal tetO array labeling oriC-proximal chromosomal units as red foci, and was induced in late log-phase using 2 μM of sodium salicylate. After 4h of induction, cells were fixed with 1% paraformaldehyde and placed at 4°C until microscopy images were taken. Images were visualized with an inverted DeltaVision Core Image Restoration Microscope (GE Healthcare) with a 100X UPlan S Apochromat (numerical aperture, 1.4) objective lens (Olympus) and a CoolSNAP HQ2 camera (Photometries). Captured images for analysis were chosen randomly. The images were taken with Z-stacks (0.15-μm intervals) and then deconvoluted (DeltaVision SoftWoRx software) to visualize the whole cell for precise and accurate quantification of foci per (Xia et al., 2016; Xia et al., 2018). For each experiment, >400 cells were counted using ImageJ software (NIH) with visual inspection from each independent experiment. Only foci that overlapped with DAPI DNA stain were quantified (≥99% of all foci).
Live Cell Deconvolution Microscopy
Cells were grown as for Assays for Ciprofloxacin-induced Mutagenesis. At 8 hours after the addition of ciprofloxacin (8.5 ng/mL), 4 μL of culture were plated onto 35mm glass bottom cell culture plates. An agar pad containing spent medium from replicate cultures (8.5 ng/mL cipro in cells grown for 8h) was placed on top of the cells, and a glass cover slip placed over the agar pad and sealed with silicon grease to limit evaporation. Images were taken every 1-2 hours for 12 hours with an inverted DeltaVision Core Image Restoration Microscope (GE Healthcare) with a 100X UPlan S Apochromat (numerical aperture, 1.4) objective lens (Olympus) and a CoolSNAP HQ2 camera (Photometrics). Captured images for analysis were randomly chosen. The images were taken with Z-stacks (0.15-μm intervals) and then deconvoluted (DeltaVision SoftWoRx software) to visualize the whole cell. For each experiment, >250 cells were followed to track the activation of the GFP (PsodA-gfp oxidative stress response) and mCherry (σS activity) using ImageJ software (NIH) with visual inspection from each independent experiment.
rpoB and ampD Sequencing
A sole RifR or AmpR colony was isolated from each of 24 cipro-treated or 24 control independent cultures and the rpoB or ampD gene sequenced, respectively. RifR rpoB mutations occur mostly within two mutation clusters (Reynolds, 2000), and all isolated mutants contained mutations within these two clusters. ampD loss of function mutations confer ampicillin resistance in our E. coli assay strain due to the insertion of Enterobacter cloacae ampRC genes in the chromosome, as previously described (Petrosino et al., 2002). The rpoB cluster I and II were amplified, as described (Reynolds, 2000), see also STAR METHODS RESOURCE TABLE for primers (Reynolds, 2000). The ampD gene was amplified using primers described in STAR METHODS RESOURCE TABLE. All PCR fragments were subjected to Sanger sequencing (GeneWIZ, Massachusetts) to identify insertions, deletions, and/or base substitutions in the ampD or rpoB genes.
Western Blot Analyses of σS Protein Levels
Western blots for quantification of σS protein levels in cultures were performed as described (Barreto et al., 2016). Proteins were separated by SDS-PAGE and transferred to 200 polyvinylidine (PVDF) membranes (Amersham Biosciences), blocked with 2% blocking buffer, and probed with polyclonal mouse anti-aS antibody (1:700 dilution) (Neoclone) (85). Goat antimouse antibody conjugated to Cy5 fluorescent dye (1:5000 dilution) (Amersham Biosciences) was used to detect the antibody-bound σS protein. Fluorescence was quantified using a Typhoon scanner, with a PMT of 500 and 670BP 30Cy5emission filter, and the bands were quantified using ImageJ software (NIH). Quantifications from three separate western blots for σS are reported, each with band intensities normalized to the values from isogenic wild-type cells with no cipro treatment run in parallel, and the means ±SEM shown.
Beta-galactosidase Assays
Cells were grown as for Assays for Ciprofloxacin-induced Mutagenesis to equivalent ODs and frozen at −20°C until assays were carried out. Determination of the β-galactosidase activity of the ParcZ-lacZ, PdsrA-lacZ, rpoS-lacZ, and katG-lacZ fusions was accomplished using the standard assay described by Miller (Miller, 1992), except that the assays were carried out in 96-well plates to ease sample processing.
Flow Cytometric Detection of Intracellular ROS or GFP and σS Activity in Single Cells
Cells were grown in the absence or presence of cipro MAC (8.5 ng/mL) to early-, late-log, and stationary phase as for Assays for Ciprofloxacin-induced Mutagenesis (above). The ROS measurement protocol was modified from Gutierrez et al. and Xia et al. (Gutierrez et al., 2013; Xia et al., 2018). Cells were incubated with ROS-staining dye DHR123 (Invitrogen) for 30 min at 37°C in PBS. After washing twice with PBS buffer, flow cytometry analyses were performed immediately. Positive gates for ROS-positive cells were set so that <0.5% of cells treated with cipro without DHR dye were positive. For experiments in which ROS or GFP and σS activity were measured, cells were grown in the absence of presence of cipro MAC (8.5ng/mL) or with 0.5mM H2O2 as for Assays for Ciprofloxacin-induced Mutagenesis (above), then harvested serially from cultures at 4, 8, 12, 16, 24, and 48 hours for ROS detection using dihyrdorhodamine 123 (DHR), or at 12, 16, and 24 for hours for ROS detection using transcriptional fusions of the oxyR and sodA promoters to GFP (Zaslaver et al., 2006). For ROS detection using DHR, cells containing σS-activity reporter yiaG-mCherry were collected and ROS were detected as green fluorescence, and σS activity as red fluorescence. For ROS detection using PoxyR-gfp and PsodA-gfp, cells containing both σS-activity reporter yiaG-mCherry and plasmids carrying the PoxyR-gfp or PsodA-gfp transcriptional fusions, or a promoterless gfp parental plasmid Pvector-gfp, were maintained with 35μg/mL kanamycin, and used to detect both GFP and red fluorescence. Single color controls were also collected at time points for spectral compensation. For the PoxyR-gfp or PsodA-gfp transcriptional fusions, gates were drawn so that the promoterless-gfp vector Pvector-gfp had < 0.5% GFP-positive cells. σS high-activity-cell gates were drawn so that spontaneous σS activation in non-cipro-treated cells after growth (<0.5% of cells without cipro) were positive, and wild-type cells without the chromosomal σS-response reporter (autofluorescence) had fewer than 0.5% of their cipro treated cells scored as positive.
Mathematical Modeling of Cipro-Induced Multi-chromosome Cell Filaments
In our model, a population of microbes is exposed to severe external stress (e.g., antibiotics), and two strategies are available: either growing into “filament” cells, that can contain multiple DNA copies, or reproducing individually. We consider a case in which resistance to the external stress can be acquired by a single mutation, with baseline rate μ, and deleterious mutations occur at many other loci, with the number of deleterious mutations per replication following a Poisson distribution with average λ. We assume that during the external stress the basic mutation rates of all cells (both μ and λ) increase A-Fold, and mutation rates in filament cells are further increased B-fold relative to non-filament cells.
We denote by s and δ the selection coefficients against the external stress and each deleterious mutation, respectively. We denote by Ia the level of adaptation to the external stress, where . The fitness (modeled here as the probability to replicate) of an individual that possess n deleterious mutations is thus ω(Ia, n) = (1 − s)1−Ia · (1 − δ)n. In the filament population, we assume that DNA copies in the same cell filament share gene products, and that deleterious mutations are recessive. Once a genome copy within the filament acquires the beneficial mutation that confers resistance to the major stress, it buds out of the filament, and begins to duplicate regularly (in proportion to the number of deleterious mutations it possesses).
We follow the two strategies for k replication cycles, starting from a population that doesn’t carry any deleterious mutations nor is adapted to the external stress. In the filament population the cells duplicate their genome without dividing and have up to 2C DNA copies. Because the populations begin without any deleterious mutations, we neglect filaments in which all DNA copies share the same deleterious mutation. Therefore, the fitness of DNA copies in the filament population is affected only by the external stress, while in the non-filament population the fitness of each DNA copy (or cell) is affected both by the external stress and by the number of deleterious mutations it carries. After k replication cycles the filaments divide to cells, each containing a single DNA copy. We then compare the population size and fitness, the proportion of adapted individuals, and the distribution of deleterious mutations, between the filament population and the nonfilament population.
Parameter values in figure 7H: λ = 0.003, μ = 6 · 10−7, δ = 0.03, A = 100, = 4. In the left and middle panels we use B = 4 and s = 0.9, whereas in the right panel B is the value on the x-axis. The value B = 4 is derived from empirical results presented in Figure 7G, in which we see that during antibiotic stress the mutation rate of cells that do filament (WT) have a fold-increase of ~4 relative to non-filamented cells.
The model tests the effect of filaments on evolvability, where mutation serves as the variation mechanism. However, if chromosomes in filaments also experience recombination, then the system corresponds to the case of Fitness-Associate Recombination (FAR) (Hadany and Beker, 2003b) - the less fit chromosomes experience higher recombination rate then the fitter ones. Previous work has shown that this mode of recombination results in increased mean fitness and improved adaptability (Hadany and Beker, 2003a).
Parameters:
Beneficial mutation rate ~Ber(μ)
Deleterious mutation rate ~Poisson(X)
A – stress-induced increase in mutation rate
B – filament cells fold increase in mutation rate relative to non-filament cells
s – selection coefficient of the antibiotic
δ - selection coefficient of each deleterious mutation (multiplicative model)
k – number of replication cycles
Measurement of High-Dose Cipro Antibiotic Activity
Cells were grown to log phase OD600 ~0.5, then cipro (1.5 μg/mL) with or without edaravone (100μM) was added, and cells were harvested 0.75, 1.25, 2.25, and 3 hours later to determine cfu/mL. Cells were washed twice with PBS and then assayed for viable cfu.
Nalidixic-Acid Test for Heritable Hypermutability
Tests for heritable mutator phenotype were as described (Torkelson et al., 1997). Ten independent (different mutations in rpoB) cipro-induced RifR mutant isolates were grown in parallel with control wild-type (non-mutator) and mutS mismatch repair-defective (mutator) strains each in duplicate independent cultures. 100μL of each saturated overnight culture was spread onto an LBH agar plate. After 10 minutes, dry nalidixic acid powder was spotted onto each plate using a capillary tube. The plates were incubated for 24 hours at 37°C, after which the number of microcolonies in the zones of inhibition were counted, and compared with the positive (mutS) and negative (isogenic wild-type) controls.
Flow-Cytometric Detection of Dead Cells
Cells were grown in the presence of cipro MAC per Assays for Ciprofloxacin-induced Mutagenesis (above), and harvested serially from cultures at log phase (4 and 12 hours) and stationary phase (24 hours) for dead cell detection using SYTOX blue dead cell stain. Cells were stained according to manufactures recommendation. Cells were incubated with SYTOX blue dye (1μM) for 30 minutes at room temperature and flow cytometry analyses were performed immediately. As a positive control, cells were incubated in 95% ethanol for 10 minutes before staining. Positive gates for dead cells were set so that <0.2% of undyed cipro-treated cells were positive, at which 90% ± 5% of the SYTOX-blue dyed positive-control ethanol-treated cells were positive.
Statistics
All statistics were performed in Microsoft Excel or GraphPad PRISM. For comparisons of two groups, a two-tailed Students t-test was used if data were normally distributed and homoscedastic. For comparisons of 3 or more groups, ANOVA with Tukey post-hoc test was used if data were normally distributed and homoscedastic, otherwise a Kruskal-Wallis non-parametric test was used. For mutation rates and ratios, which are not normally distributed, natural-logarithm transformed data were used to calculate 95% confidence intervals as well as performing ANOVAs.
GRAPHICAL ABSTRACT
In Brief
Bacteria exposed to antibiotic transiently differentiate a small subpopulation of gambler cells that increase mutation rate and evolve resistance, while most cells avoid the risk. The gamblers are differentiated beginning with the antibiotic inducing reactive oxygen only in subpopulation cells. The reactive oxygen activates the general stress response, which allows mutagenic DNA break repair in the gambler cells. Multi-chromosome cells are required and, modeling shows, can allow high mutation rates and rapid evolution by chromosome cooperation buffering deleterious mutations.
Highlights
Antibiotic-induced mutable cell subpopulation generates resistant mutants
Mitigates risk to most cells; reactive oxygen ➔ σS stress response ➔ gamblers
FDA-approved drug blocks σS response and mutagenesis: anti-evolvability drug
Multiple chromosomes needed: chromosome cooperation can allow rapid adaptation
ACKNOWLEDGEMENTS
We thank S Gottesman, J Imlay, I Matic, and L Zechiedrich for kind gifts of E. coli strains, N Majdalani for guidance on sRNAs and σS activation, S Kozmin for advice on dose-dependent induction of mutagenesis, S Henikoff for helpful conversation, and KM Miller, and Meng Wang for improving the manuscript. This work was supported by the NIH Grants R35-GM122598 (SMR), R01-GM088653 (CH), R01-GM102679 (DB), R01-GM106373 (PJH), the Israeli Science Fund ISF 1568/13 (LH), and by the Integrated Microscopy Core at Baylor College of Medicine with funding from the NIH (dK56338, and CA125123), the Dan L. Duncan Comprehensive Cancer Center, RP160283 Baylor College of Medicine Comprehensive Cancer Training Program Postdoctoral Fellowship (DMF) and American Cancer Society Postdoctoral Fellowship 132206-PF-18-035-01-DMC (DMF); and the John S. Dunn Gulf Coast Consortium for Chemical Genomics. This project was supported by the Cytometry and Cell Sorting Core at Baylor College of Medicine with funding from the NIH (P30 AI036211, P30 CA125123, and S10 RR024574) and the expert assistance of Joel M. Sederstrom.