Stochastic expression of a multiple antibiotic resistance activator confers transient resistance in single cells

Transient resistance can allow microorganisms to temporarily survive lethal concentrations of antibiotics. This can be accomplished through stochastic mechanisms, where individual cells within a population display diverse phenotypes to hedge against the appearance of an antibiotic. To date, research on transient stochastic resistance has focused primarily on mechanisms where a subpopulation of cells enters a dormant, drug-tolerant state. However, a fundamental question is whether stochastic gene expression can also generate variable resistance levels among growing cells in a population. We hypothesized that stochastic expression of antibiotic-inducible resistance mechanisms might play such a role. To investigate this, we focused on a prototypical example of such a system: the multiple antibiotic resistance activator MarA. Previous studies have shown that induction of MarA can lead to a multidrug resistant phenotype at the population level. We asked whether MarA expression also has a stochastic component, even when uninduced. Time lapse microscopy showed that isogenic cells express heterogeneous, dynamic levels of MarA, which were correlated with transient antibiotic survival. This finding has important clinical implications, as stochastic expression of resistance genes may be widespread, allowing populations to hedge against the sudden appearance of an antibiotic.

Bacteria can evade antibiotics through transient expression of resistance genes. By temporarily elevating resistance in a subset of cells, a population can undermine the efficacy of antibiotics resulting in chronic and recalcitrant infections 1,2 . For example, in bacterial persistence a small fraction of cells (≤ 1 in 100) called persisters stochastically enter a dormant, drug-tolerant state, allowing the population to hedge against the sudden appearance of an antibiotic 3,4 . It is important to recognize that transient resistance is not caused by genetic changes, rather cells use phenotypic variability or induce gene expression to generate a resistant phenotype 2,3,5,6 . To date, research on phenotypic variability in antibiotic resistance has focused primarily on dormancy. However, little is known about transient resistance strategies that generate a continuum of resistance levels within growing cells.
Isogenic bacterial populations are traditionally considered to be composed of identical cells. However, even though individual cells contain the same genetic material, protein levels between cells can vary due to stochastic events associated with gene expression and regulation 7 . Cell-to-cell heterogeneity has important implications, allowing populations of cells to diversify in order to survive environmental stress 8 , and evade the immune response 9,10 .
In contrast to permanent antibiotic resistance, usually associated with mutations or acquisition of resistance elements via horizontal or vertical transfer 11 , transient resistance allows cells to temporarily survive the appearance of an antibiotic. For instance, the pathogen Mycobacterium tuberculosis has subpopulations of non-replicating cells characterized by high antibiotic tolerance [12][13][14] . Bacterial persistence is a well-studied example of stochastic variation that results in a small fraction of cells that can survive antibiotic stress 3,15 . In persistence, cellular mechanisms such as DNA and protein synthesis are inhibited and consequently cells remain dormant and evade antibiotics that target cell growth processes 15 . An increasing number of studies have determined key factors involved in persistence. Toxin-antitoxin systems are often involved and quantitative measurements at the single cell level have shown that overexpression of the toxin can determine when and for how long a cell will remain in the dormant state 6 . In addition, stringent response via the mediator (p)ppGpp, reduced membrane potential, and extended single-cell lag times can confer transient antibiotic resistance [16][17][18] . These mechanisms have the common feature that antibiotic tolerance is achieved by temporarily placing cells in a dormant state.
Other studies suggest that mechanisms for transient antibiotic resistance exist beyond dormancy, however the molecular basis often remains unclear. A study in Escherichia coli showed that stress-induced variability within an isogenic population is transmissible between generations and plays a role in antibiotic survival at the single cell level, possibly by modifying membrane permeability 19 . Furthermore, cell populations can differentiate into resistant subpopulations with variable growth statuses due to cephalosporin hydrolase expression 20 . In Salmonella enterica, heterogeneous levels of porins and efflux pumps contribute to differential levels of antibiotic resistance 21 . In addition, single cell studies have shown that the probability of E. coli cell lysis correlates with the time since the last cell division 22 . Asymmetric cell division events 23 and stochastic pulses in the catalase-peroxidase KatG 24 in mycobacteria result in differences in antibiotic susceptibility. These studies hint at additional pathways by which cells can use stochastic, non-genetic variability to survive antibiotics.
In addition to stochastic effects, cells can evade antibiotics by transiently inducing antibiotic resistance at the population level 25,26 . A well-studied example of this is expression of MarA (the multiple antibiotic resistance activator), which plays a key role in multidrug resistance in enteric bacteria 27 . In E. coli, MarA expression can be induced by the addition of extracellular compounds, including antibiotics [28][29][30] . Thus, when antibiotics are detected, resistance genes are turned on, leading to population-wide resistance. MarA expression is regulated by a combination of positive and negative feedback loops (Fig. 1a). The marRAB operon is autoactivated by MarA and autorepressed by MarR 31,32 ; the periplasmic protein MarB also indirectly represses marRAB expression 33 . MarA activates over 40 downstream genes implicated in antibiotic resistance 34 . Examples include micF, an antisense RNA that represses the expression of the outer membrane porin OmpF, and the acrAB-tolC multidrug efflux pump genes 34 .
In this work, we focused on the role of MarA in transient resistance. Although it is well-known that MarA can induce antibiotic resistance at the population level, we asked whether stochastic expression of MarA could provide antibiotic resistance at the single cell level, even when uninduced. Using time-lapse microscopy, we studied MarA dynamics in isogenic cells and found cell-to-cell variability in MarA expression, which correlated with antibiotic susceptibility. This phenotypic variation has the potential to generate diverse resistance phenotypes within a population.

Results
MarA overexpression increases antibiotic resistance in population measurements. MarA's role in activating downstream multidrug resistance genes has been studied extensively at the population level [27][28][29][30][31][32] . In this work we used carbenicillin, a bactericidal antibiotic that inhibits cell-wall synthesis 35 . We first measured the minimum inhibitory concentration of carbenicillin in E. coli MG1655 and in the same strain with a plasmid overexpressing MarA (Fig. 1b). Consistent with previous reports, overexpression of MarA increased antibiotic resistance 36 . MarA expression is heterogeneous at the single cell level. Although inducible population-level resistance is well established, we wondered whether MarA expression is variable at the single cell level. Previous computational studies by our group have hypothesized that the feedback structure regulating MarA can produce stochastic MarA expression when the system is uninduced 37 . Motivated by these computational predictions, we experimentally measured the dynamics using a plasmid that reports MarA levels in the cell. To do this, we used a modified version of the marRAB promoter containing transversion mutations that inactivate the MarR binding sites in the operator, leaving the MarA binding site intact 29 . We fused this promoter to a cyan fluorescent protein gene (cfp) with an ssrA degradation tag to decrease the protein half-life and increase temporal resolution 38 . We conducted experiments with this plasmid in E. coli MG1655 (we refer to this strain as P marA -cfp). We note that MarR binding sites were only removed in the reporter plasmid; the chromosomal copy of the marRAB promoter remained unchanged. The promoter modification in the reporter was necessary to visualize CFP and allowed us to measure MarA independent of the action of MarR. In order to study dynamics and heterogeneity in MarA expression at the single cell level, we conducted time-lapse microscopy experiments with P marA -cfp. Within growing microcolonies we observed heterogeneous MarA expression that fluctuated over time ( Fig. 1c and Supplementary Movie 1). Therefore, MarA expression is stochastic within single cell lineages.

MarA variability is correlated with survival in the presence of carbenicillin within an isogenic
E. coli population. We next asked whether variability in MarA expression impacts survival under antibiotic treatment at the single cell level. Bacteria can transiently defend against antibiotic lethality by inducing the SOS response, which inhibits bacterial cell division but not elongation, enabling survival in the presence of lethal concentrations of antibiotics 39 . We exposed cells containing the MarA reporter P marA -cfp to lethal concentration of carbenicillin (50 μ g/ml) on agarose pads and observed the impact on individual cells using time-lapse microscopy ( Fig. 2a and Supplementary Movie 2). Cells lysis occurred rapidly after incubation with carbenicillin for a subset of cells in the population. As an indicator of cell death we used propidium iodide, which enters the cells and stains DNA if the membrane is depolarized 40 . Using P marA -cfp, we measured the initial fluorescence level of each cell at t = 0 mins. We then recorded the outcome of each cell after 400 mins of carbenicillin exposure (Fig. 2b). This duration, sufficient to kill a significant fraction of E. coli cells, allowed us to ensure that transient effects due to cell division time were not a factor in our analysis 22 . We primarily observed two outcomes: cell lysis, as indicated by propidium iodide staining, and filamentation, where cells elongate but do not lyse. A small fraction (~10%) of cells neither stained with propidium iodide nor formed filaments and were excluded from subsequent analysis. Each cell outcome was assigned to the initial CFP value reflecting the MarA expression level in the cell at t = 0 mins. As expected from our time-lapse microscopy experiments (Fig. 1c), we observed a distribution of initial fluorescence levels corresponding to cell-to-cell variability in MarA expression (Fig. 2b). We also observed a heterogeneous response to carbenicillin. Interestingly, heterogeneous outcomes were correlated with MarA variability between isogenic cells, where cells that filamented were more likely to have high initial MarA levels ( Fig. 2b).
To determine if variability in fluorescence levels was due to MarA expression itself, we introduced the same fluorescent reporter into a strain lacking the marRAB operon (we refer to this as P marA -cfp Δ marRAB). We recorded initial fluorescence levels and cell outcomes in the presence of carbenicillin as before (Fig. 2c,d). Cells exhibited higher lysis rates following carbenicillin exposure than in the strain with the intact marRAB operon. CFP levels for P marA -cfp Δ marRAB were lower than for P marA -cfp, as expected given the absence of MarA.
As a positive control, we also constructed a MarA-CFP protein fusion in order to produce a population with high, homogeneous expression of MarA. The translational fusion stabilizes MarA, increasing its half-life to ~30 mins ( Supplementary Fig. 1), in contrast to ~1 min for wildtype MarA 41 . As a result, cells exhibited homogeneous fluorescence levels (Fig. 2e). It is important to note that the CFP levels for this strain do not report the same levels of MarA as those strains with P marA -cfp. Instead, the MarA-CFP strain has markedly higher levels of MarA than either the P marA -cfp or P marA -cfp Δ marRAB strain due to the stabilized protein. When we exposed cells with MarA-CFP to carbenicillin, we observed a dramatic increase in the number of filamented cells relative to the P marA -cfp strain (Fig. 2f).
Filamented cells are able to regrow normally and are still susceptible to antibiotics. Are the filamented cells we observed following carbenicillin treatment able to resume growth after removal of carbenicillin? To test this, we used microfluidic chambers to trap cells while introducing and removing carbenicillin. Single cells were trapped in a microfluidic chamber and grown until the chambers were full. We then introduced a 90 min step of 50 μ g/ml carbenicillin. Following this, we returned to conditions without the antibiotic, then later introduced a second step of carbenicillin ( Fig. 3 and Supplementary Movie 3). As in our experiments on agarose pads with carbenicillin, we observed variability in MarA expression and heterogeneous responses, including both lysis and filamentation. Importantly, after carbenicillin was removed, the filamented cells were able to divide and regrow normally, suggesting the clinical relevance of transient antibiotic resistance. To confirm that these surviving cells were not resistant to antibiotics due to mutations or other non-transient mechanisms we introduced a second step of carbenicillin. We observed similar patterns of lysis and filamentation following this subsequent carbenicillin step, indicating that those cells that survived the first round of treatment were still susceptible to antibiotics.

MarA levels stochastically exceed a threshold that confers transient resistance to carbenicillin.
We were next interested in understanding how dynamic, heterogeneous MarA expression impacts survival. We first quantified the dynamics of P marA -cfp within microcolonies. We observed that MarA levels fluctuate in individual cell lineages (Fig. 4a). We also quantified fluorescence levels in the P marA -cfp Δ marRAB strain (Fig. 4b). Interestingly, the P marA -cfp Δ marRAB strain still exhibited fluctuations in MarA expression, though fluorescence levels were reduced relative to P marA -cfp. These residual dynamics could be due to the action of the MarA homologs Rob and SoxS 42 , other regulatory mechanisms that interact with P marA 43 , or dynamics intrinsic to the fluorescent reporter.
We conducted control experiments to eliminate the possibility that something about CFP expression or the ssrA tag was responsible for increasing antibiotic survival. To achieve this, we constructed a reporter strain that was independent of MarA where we could induce similar CFP expression levels to the P marA -cfp strain. We refer to this control strain as P lac -cfp. We observed variation in CFP expression across cell lineages, likely due to the intermediate induction levels (Supplementary Fig. 2a). We also measured the distribution of fluorescence levels for the P marA -cfp, P marA -cfp Δ marRAB, and P lac -cfp strains (Fig. 4c). CFP fluorescence for P marA -cfp had a long tail of high fluorescence values. For P marA -cfp Δ marRAB the distribution shape was similar, but the mean was slightly reduced and the tail of the distribution did not extend to CFP values that were as high as in the P marA -cfp strain.
Scientific RepoRts | 6:19538 | DOI: 10.1038/srep19538 By design, the CFP levels for P lac -cfp were similar to or higher than those for P marA -cfp, but notably, the shapes of the distributions were different, suggestive of differing underlying dynamic processes 44 . To show that MarA levels, and not a reporter artifact, were causing heterogeneity in antibiotic survival, we placed P lac -cfp cells on agarose pads containing carbenicillin and recorded cell lysis and filamentation outcomes as before. In contrast to results with P marA -cfp, we did not observe a correlation between higher fluorescence levels and filamented cells ( Supplementary Fig. 2b), confirming the contribution of MarA to heterogeneous antibiotic survival.
We measured the average autocorrelation of the CFP signal for P marA -cfp, P marA -cfp Δ marRAB, and P lac -cfp strains and observed no dominant periodicity in expression of any of the CFP signals (Fig. 4d). However, we note that although the average autocorrelation of CFP is similar between strains and is not indicative of a periodic signal, this does not preclude the possibility that stochastic properties differ, as these effects may be obscured by an average. Similar experiments with the MarA-CFP overexpression strain showed fewer fluctuations in fluorescence levels and slower dynamics, as expected due to the stabilized protein ( Supplementary Fig. 3). What is the relationship between cellular survival and fluorescence? Using the carbenicillin outcomes data ( Fig. 2b and Supplementary Fig. 2b), we measured the percentage of cells above a threshold fluorescence level that filament upon carbenicillin treatment (Fig. 4e). As we increased the threshold for the P marA -cfp strain, a larger percentage of cells filamented, demonstrating a relationship between CFP levels and survival in these strains. The overall rate of filamentation in the P lac -cfp strain was lower than P marA -cfp, possibly due to differences in the reporter plasmid or induction conditions. However, in sharp contrast to P marA -cfp, the P lac -cfp strain always exhibited a constant percentage of filamented cells, regardless of the threshold fluorescence level we set. These results demonstrate that MarA expression is responsible for the increase in filamentation. Furthermore, the differences in cell lysis versus filamentation we observed between P marA -cfp and P marA -cfp Δ marRAB indicate that wildtype cells routinely exceed threshold levels of MarA required to provide transient resistance to lethal concentrations of carbenicillin, while Δ marRAB cells are far less likely to cross this threshold.

Discussion
MarA activates a suite of downstream genes involved in antibiotic resistance 45 . Not all of these genes are activated at the same time or with the same number of MarA molecules [46][47][48] . Martin et al. estimated that, at a minimum, there is a 19-fold difference in the number of MarA proteins needed for half-saturation of the different downstream promoters 46 . Furthermore, physiological levels of MarA are far lower than those required to achieve saturation for the majority of these downstream genes 46 . Our findings suggest a possible mechanism for how cells within a population could achieve a gradient of resistance levels in a transient fashion. Cells that stochastically express higher levels of MarA may transiently turn on more downstream genes than those with low levels of MarA, leading to elevated resistance (Fig. 5). Because MarA expression is dynamic, a cell with high MarA that expresses many downstream genes will eventually revert to conditions with lower levels of resistance. Stochastic expression of a transcriptional factor could serve to coordinate expression of multiple downstream genes simultaneously 49 . When we conducted experiments in the Δ marRAB strain, cells still exhibited dynamic expression of P marA -cfp , however carbenicillin survival rates were reduced, suggesting that variability in MarA expression in wildtype cells is sufficient to allow a subset of the population to transiently achieve levels required for antibiotic resistance.
The genetic mechanism that underlies stochastic expression of MarA is an interesting area for future study. A computational model from our group proposed a mechanism involving feedback control of marRAB expression that can lead to stochastic dynamics 37 . However, our results here suggest that this model is incomplete, as cells retain dynamic behavior even when the marRAB operon is deleted. It is possible that other regulatory proteins, such as SoxS and Rob play a complementary role, or that degradation of these proteins by Lon protease may introduce additional dynamics 41,42 . Indeed, MarA does not function alone in the cell but together with SoxS and Rob. These three regulators control a common set of downstream genes and significant interactions have been identified between them 42 . Although we have focused here on the uninduced case, a recent stochastic modeling study suggests that induction of MarA expression may have interesting dynamics 50 . Time-lapse experiments could explore the effect of induction on the network at the single cell level. In addition, it will be interesting to explore the genetic basis for the long tail of MarA levels observed in the P marA -cfp strain (Fig. 4c), which may be the result of the feedback architecture controlling expression of the marRAB operon.
Expressing downstream genes involved in antibiotic resistance imposes a burden on cells 51 . Generating diverse resistance phenotypes within a population could serve as a bet hedging strategy to allow populations to survive antibiotic treatment without requiring that all cells express these costly genes. Similar strategies have been demonstrated at the single cell level through studies on bacterial persistence 3,6,24 . Our findings present an alternative to dormancy, where cells temporarily grow into filaments and continue producing cellular components 39 . Other studies have pointed to alternative mechanisms by which cells can generate a continuum of transient resistance levels. For example, resistance has been shown to be negatively correlated with RpoH expression in E. coli 19 , positively correlated with the cephalosporin hydrolase gene in E. coli 20 , and with the porin gene ompC in S. typhimurium 21 . These studies, coupled with our findings, suggest that it is important to consider stochastic, single cell level effects associated with expression of antibiotic resistance genes and their regulators.  Stochastic gene expression can facilitate evolutionary adaptation 52 . A recent study showed that antibiotic resistance can emerge from multinucleated bacterial filaments via the SOS response 53 . A strategy where a subset of cells have high MarA levels and grow into filaments could generate a favorable environment for mutations, especially over long time or cyclic exposure to antibiotics. We have identified a new stochastic role for the multiple resistance activator MarA, where even without induction, a subset of cells achieve expression levels sufficient to achieve transient resistance to antibiotics.

Methods
Plasmids and Strains. To construct P marA -cfp we placed a modified marRAB promoter upstream of a degradation tagged cfp. We used a copy of the wildtype marRAB promoter and introduced transversion mutations to inactivate the two MarR binding sites 29 . This plasmid was transformed into E. coli MG1655.
The P marA -cfp Δ marRAB strain is E. coli MG1655 Δ marRAB transformed with the plasmid described above. The MarA-CFP translational fusion strain is E. coli MG1655 transformed with a plasmid where marA is fused to an untagged version of cfp using a (Gly 2 -Ser) 2 linker downstream the lacUV5 promoter.
The P lac -cfp strain is E. coli MG1655 transformed with a plasmid where degradation tagged cfp is fused to the lacUV5 promoter.
Further details on plasmid and strain construction are provided in Supplementary Information.

Time-lapse Microscopy.
Overnight cultures were grown from single colonies in LB medium with 30 μ g/ml kanamycin. From these cultures a 1:100 dilution was used to inoculate fresh LB containing 30 μ g/ml kanamycin (for MarA-CFP and P lac -cfp experiments, this was supplemented with 100 μ M or 50 μ M IPTG, respectively). Cultures were incubated for 3 hrs at 37 °C with shaking. Cells were then diluted 1:100 in M9 minimal medium containing 0.2% glycerol, 0.01% casamino acids, 0.15 μ g/ml biotin, and 1.5 μ M thiamine (which we denote MGC medium). Cells were then placed on 1.5% MGC low melting temperature agarose pads containing kanamycin and IPTG as described above. Cells were imaged at 100× using a Nikon Instruments Ti-E microscope. The temperature of the microscope chamber was held at 32 °C for the duration of the movies. For the single cell carbenicillin assays, cells were prepared as described above, but were diluted 1:3 in MGC following the 3 hr incubation. 50 μ g/ml carbenicillin and 10 μ g/ml propidium iodide were added to the agarose pads.
For the microfluidic chip experiments, we used P marA -cfp E. coli MG1655 Δ fliC; the fliC deletion makes the strain non-motile. 50 μ g/ml carbenicillin and 10 μ g/ml propidium iodide were introduced at the times shown in Fig. 3.
Image analysis was performed using custom MATLAB software. See Supplementary Information for full methods.