Eradication of bacterial persisters with antibiotic-generated hydroxyl radicals

Results

We initially sought to detect a subpopulation of persisters in a larger population of susceptible, logarithmically growing mycobacteria. Cultures of M. smegmatis were grown to stationary phase and then diluted in fresh media roughly 15-fold to an optical density (OD₆₀₀) of 0.2. Cells were next exposed to pairs of bactericidal antibiotics with different modes of action to prevent the emergence of genetically resistant clones. For example, the DNA gyrase inhibitor ciprofloxacin (CIP) at a concentration well above the minimum inhibitory concentration (MIC) was paired with a bacteriostatic concentration of the cell-wall biosynthesis inhibitor isoniazid (INH). M. smegmatis bacilli treated in this manner show approximately four logs of killing over the first 24 h but exhibit no further reduction in cell number during the following seven days (Fig. 1A and Fig. S1A). We did not observe significant differences in the mean number of persisters identified when CIP was used alone versus in combination with low-dose INH (Fig. S1B). Similarly, treated M. tuberculosis bacilli show three logs of killing over the first four days before stabilization in cell number is achieved (Fig. 1B). Along with surviving in the antibiotic in which they were initially challenged, M. smegmatis and M. tuberculosis persisters also demonstrated tolerance to subsequent challenges with bactericidal levels of rifampicin (RIF) or streptomycin (STM), antibiotics with completely unrelated mechanisms of action (Fig. 1 C and D and Fig. S1C).

During these initial experiments to characterize the persister subpopulation, we observed substantial experiment-to-experiment variability in the size of the final persister population. In particular we noted that perturbation of the culture during sampling reduced the size of the persister population compared with undisturbed cultures (Fig. S2). This observation led us to hypothesize that the sampling process, which required removing a foil cap, thus introducing ambient air into the flask headspace, may increase the DO saturation of the culture, thereby affecting the number of persisters identified. To test this hypothesis, we characterized the M. smegmatis persister population in two controlled conditions in which oxygen saturations would not be affected by sampling. In the first system, a highly aerobic condition that we will refer to as the “open system,” an oxygen-permeable membrane was used to cover a shaking baffled Erlenmeyer flask, increasing the oxygen transfer rate into the media and resulting in an equilibrium state of high DO. [Note: At 37 °C, the maximum dissolved oxygen in water is ∼6.7 mg/L (31)]. In the second system, the sealed or “closed system” condition, a rubber septum was used to seal a flat-bottom Erlenmeyer flask (headspace/media ratio 5:1), thus preventing ongoing oxygen diffusion from the atmosphere. Consequently, oxygen levels within the flask drop as cell metabolism consumes oxygen within the media and headspace. In both models, a needle and syringe were used to sample the cultures for bacterial enumeration to minimize perturbation of the systems. In the closed system model, we found that for RIF/INH, CIP/INH, and STM/INH, the number of viable M. smegmatis cells stabilized two days after the initial kill in a biphasic manner, as we had previously observed. In contrast, bacilli in the continuously aerated system continued to die (Fig. 2 A and B and Fig. S3).

To understand whether M. tuberculosis cultures were similarly sensitive to the DO system, we treated cells in tightly capped square media bottles with either a headspace ratio of 4:1, designed to approximate septum-sealed conditions, or a comparatively aerated 60:1 ratio. Similar to what we had observed with M. smegmatis, in M. tuberculosis the reduced oxygen condition exhibited slower killing and a stable persister population, in contrast to the more aerated condition where the number of survivors steadily declined (Fig. 2 C and D). To determine whether the functional relationship between DO and antibiotic susceptibility of persisters is specific to mycobacteria we tested the Gram-negative pathogen P. aeruginosa as well as E. coli using our model system (SI Materials and Methods). We found, similar to what was observed in mycobacteria, that fewer cells survive antibiotic treatment in the maximally aerated conditions compared with the septum-sealed conditions (Fig. S4). These results demonstrate that in vitro models used to identify persister populations are extremely sensitive to changes in oxygen tension. In addition, these observations suggest that high levels of DO may either prevent the formation or interfere with the maintenance of the persister population.

In stationary phase a subpopulation of bacteria also exhibit phenotypic antibiotic tolerance, thought to be due to nutrient exhaustion and changes in the pH of the media (5, 32). Given our finding that persisters identified within a freshly diluted culture are sensitive to antibiotic killing in the presence of sufficiently high DO levels, we next sought to determine if the antibiotic-tolerant population in a fully stationary phase, undiluted culture also becomes susceptible to antibiotic treatment when placed in highly aerated conditions. A culture of M. smegmatis was grown to stationary phase, transferred into either closed or open conditions, and then treated with bactericidal concentrations of CIP and INH. Over three weeks of treatment, stationary phase cultures in the closed system exhibited only a modest two-log drop in census, similar to what has been reported previously for stationary phase cultures (33). In contrast, the treatment of stationary phase cultures in the open system resulted in a reduction in census of more than five orders of magnitude to the lower limit of detection. Based on the observation that in a stationary phase culture the antibiotic-tolerant subpopulation loses antibiotic tolerance when oxygen levels are sufficiently increased, we conclude that tolerance is not due solely to nutrient limitation or changes in pH (Fig. 3A).

Hypoxia has previously been shown to play a role in inducing antibiotic tolerance in M. tuberculosis (34). In the most-studied hypoxia model, commonly referred to as the Wayne model, antibiotic tolerance in the bulk population appears as cells adapt to microaerophilic (DO saturation ∼ 1%) and anaerobic (DO saturation ≤ 0.06%) conditions. To determine the degree of oxygen depletion in the experiments described above (Fig. 2) we measured the dissolved-oxygen saturation in both the open and closed models. To our surprise we found only a 20% difference in DO saturation between the two conditions. In the closed system, the oxygen saturation fell over three days before stabilizing at 75%, but in the open system, oxygen saturations equilibrated around 95% (Fig. 3B). Thus, in contrast to the previously described Wayne model, our data demonstrate that very small decreases in DO are sufficient to promote bacterial survival, albeit of a small persister subpopulation, in the presence of bactericidal antibiotic concentrations. We next tested whether decreasing the inoculum, which should decrease the rate of oxygen consumption and therefore lead to higher DO saturation, would decrease the size of the persister subpopulation. Indeed, we found that with a 10-fold lower starting inoculum, we did not identify a stable persister population above the level of detection (Fig. S5), perhaps in part explaining the “inoculum effect” (35). To further confirm the effect that increasing oxygen levels has on the ability of the antibiotics to kill the persister population, we transferred M. smegmatis persisters generated in the closed system to the open system. This transfer into aerated conditions led to the subsequent death of persisters (Fig. 3C), thus demonstrating that persister cells are in fact vulnerable or can become vulnerable to antibiotics under conditions of higher DO saturation.

The observation that a 20% drop in DO saturation, the result of metabolic oxygen consumption, results in a stable persister population suggests that the reduction in oxygen saturation either stimulates persister formation or facilitates the survival of a preexisting bacterial population alive at the time of oxygen depletion. To differentiate between these scenarios, we tested whether preadapting a population of cells to lower DO levels might result in a higher percentage of persisters or even induce antibiotic tolerance in the entire population. We used gas displacement with nitrogen gas to preadapt cultures to 75% DO saturation for one hour before treating with high concentrations of CIP and INH (Fig. 4A). Following antibiotic treatment the oxygen-displaced flasks fell from 75% DO saturation to 60% DO saturation over the next four days, while the control flasks dropped from 100 to 80% oxygen saturation (Fig. 4A). Despite these differences in DO, the kill kinetics and mean size of the persister populations for the oxygen-displaced flasks were statistically indistinguishable from the control flasks seven days after treatment (Fig. 4B, t test P value ∼ 0.67). These data suggest that reduced oxygen levels are not driving increased persister cell formation before antibiotic exposure but instead are facilitating the survival of persisters. Furthermore, because the reduction of the DO level by nitrogen displacement did not change the initial first phase of killing compared with nondisplaced cultures, the 20% drop in DO saturation appears to have no effect on the survival of the dominant antibiotic-susceptible population. We therefore conclude that within the bulk culture there is a bacterial subpopulation (which forms the second-phase plateau) whose sensitivity to subtle changes in oxygen when challenged with antibiotics differs from the majority of the population.

This observation, combined with recent work demonstrating both the role of ROS in antibiotic killing and cellular antioxidant responses in antibiotic tolerance, led us to speculate that the sensitivity of the persister population to changes in oxygen levels may be linked to changes in the number of toxic free radicals generated with antibiotic treatment. Exposure to bactericidal antibiotics has been found to result in the intracellular generation of toxic free radicals, resulting in bacterial cell death. Oxygen diffuses freely across cell membranes (36), resulting in an intracellular concentration in equilibrium with that of the surrounding media, with previous measurements demonstrating that reactive oxygen species are generated in direct proportion to oxygen concentration (36, 37). The intracellular free-radical concentration following antibiotic exposure is therefore predicted to be higher in the open model compared with the closed model. To confirm this hypothesis we used the fluorescent reporter dye 3′-(p-hydroxy-phenyl) fluorescein (HPF) to measure relative hydroxyl-radical levels in norfloxacin-treated E. coli cells in maximally aerated media compared with media with a DO level of 75%. As expected, we found higher levels of hydroxyl radicals in maximally aerated media (P value below 0.0002, Fig. S6, SI Materials and Methods).

To test the role of free radicals in the killing of both the predominant antibiotic-susceptible population and in the smaller persister subpopulation, we blocked the effects of hydroxyl radicals after CIP treatment using the free-radical quencher thiourea (24) in the aerated, open system. We found that in high DO conditions, thiourea successfully restored the biphasic kill kinetic and an antibiotic-tolerant population was again identified (Fig. 5A). We also observed a reduced kill kinetic for the larger antibiotic-susceptible population, consistent with earlier work demonstrating the role of ROS in antibiotic killing in E. coli (24). From these data we conclude that under high DO conditions, the persister population is killed following antibiotic treatment as a consequence of an increased number of toxic hydroxyl radicals generated with antibiotic treatment.

We conversely hypothesized that increasing hydroxyl radicals within cells could successfully eradicate the persister population in the closed system. We used the antibiotic clofazimine, which results in increased intracellular ROS production due to redox cycling (38). CFZ, recently identified as a competing substrate for the enzyme NDH-2 within the electron transport chain, is reduced by NDH-2 and subsequently oxidized by O₂, resulting in increased ROS (38). When we challenged persisters with CFZ, the number of surviving persisters decreased by three orders of magnitude, contrasting with the tolerance observed for rifampin and streptomycin (Fig. 5B). In the presence of thiourea the addition of CFZ did not change the number of persisters in a statistically significant manner, suggesting that CFZ is killing the persister population through increased intracellular ROS generation (Fig. S7, t test P value ∼0.42). These data, combined with the nitrogen displacement data demonstrating that subtle changes in oxygen saturation do not affect the survival of the dominant antibiotic-susceptible population, suggest a model in which persisters respond differently to antibiotic-induced hydroxyl radicals compared with the larger antibiotic-susceptible population, except under maximal DO conditions.

Discussion

The important role that oxygen plays in antibiotic killing of bacteria has also long been recognized, bolstered by observations that in abundance it can improve antibiotic efficacy, and when absent can result in complete antibiotic tolerance (24, 34). We now find that much smaller changes in oxygen tension than previously described affect the antibiotic killing of bacterial persisters. Although these small reductions in oxygen tension do not alter the kill kinetic of the larger antibiotic-susceptible population, these changes dramatically affect the size and survival ability of the smaller persister subpopulation. In an environment where dissolved oxygen saturations are high, this subpopulation rapidly dies and can no longer be identified. We observe the same relationship between oxygen and antibiotic sensitivity in Gram-negative bacteria (P. aeruginosa and E. coli), which demonstrates that this phenomenon is not unique to mycobacteria. These results suggest that the persister subpopulation differs from the larger population in its ability to withstand hydroxyl radicals generated by antibiotic challenge. Previous work in E. coli has demonstrated that antibiotic killing in susceptible populations is mediated by hydroxyl radicals (24), in part through oxidation of the guanine nucleotide pool (39). With the addition of thiourea we find that the kill kinetic of the bulk population is reduced, suggesting that hydroxyl radicals may similarly mediate antibiotic killing in mycobacteria and that persisters may survive antibiotic challenge due to a unique capacity to withstand the generation of toxic free radicals. This ability of persisters to survive when the DO level is reduced is in contrast to the larger, susceptible population. When DO levels are near saturation, however, the protective mechanisms in persisters are insufficient and they succumb to antibiotic challenge. When decreased free-radical concentrations are achieved either through a reduction in oxygen saturation or the addition of thiourea, a free-radical quencher, the drug-tolerant persister subpopulation is able to survive in the presence of otherwise bactericidal antibiotic concentrations. In contrast, strategies to increase cellular ROS, such as the addition of the antibiotic CFZ, results in killing of the drug-tolerant population. Antibiotic-generated hydroxyl-radical–mediated killing may therefore represent an Achilles’ heel of persistence.

The impact small changes in oxygen tension have on the maintenance and survival of persisters may explain inconsistencies within the M. tuberculosis persister literature. Recently, two studies were published with conflicting results concerning the enrichment of drug-tolerant persisters in late exponential and stationary phase cultures (12, 13), with one identifying an enrichment of persisters in these cultures while the other did not. Similarly, there are discrepancies regarding the cross-tolerance of M. tuberculosis persisters to heterologous drugs. Based on the data presented here, the origin of these divergent observations may lie in differing culture densities, which impacts the rate of oxygen depletion and oxygen saturation at the time of treatment, thereby changing antibiotic efficacy.

Within its human host, M. tuberculosis sometimes survives within low-oxygen physical niches, which may further protect M. tuberculosis from hydroxyl-radical-mediated antibiotic killing. The most extreme example is the granuloma, a collection of M. tuberculosis-infected macrophages and immune cells, which was recently demonstrated to possess a hypoxic core with an oxygen tension of 3 mm Hg (40). Because phenotypic drug tolerance is observed in vitro under hypoxic conditions, it has been suggested that standard TB therapy is likely ineffective in the granuloma. This work suggests that much more subtle changes in oxygen tension may dramatically reduce antibiotic efficacy with the goal of sterilization. During in vivo infections, M. tuberculosis likely encounters a gradient of oxygen tensions in tissue that range from 100 mm Hg in the alveolus to 60 mm Hg in normal lung tissue to 3 mm Hg in the center of a granuloma (40, 41). The data presented here suggest that effective sterilization of M. tuberculosis may be limited by the gradual decline in oxygen tension even before hypoxic conditions are obtained.

There is a growing body of evidence demonstrating the important role that ROS play in bactericidal antibiotic–mediated killing (24–26). This understanding of how bactericidal antibiotics result in cell death raises the hypothesis that drug tolerance may be mediated by increased abilities within a cell to detoxify ROS. Recent work has suggested that antibiotic tolerance of the bulk population, induced by changes in the environment, specifically starvation, may be the result of enhanced antioxidant strategies (28). Our work focuses on the small persister subpopulation within a larger culture, and we demonstrate that this subpopulation is differentiated from the larger antibiotic-susceptible population by different sensitivities to hydroxyl-radical-mediated damage from antibiotic exposure. Our findings are supported by recent work demonstrating that E. coli persisters produce fewer hydroxyl radicals following antibiotic challenge, and a possible mechanism is suggested by work demonstrating that E. coli secretes the signaling molecule indole that activates oxidative stress responses and increased drug tolerance (19, 42, 43). Although the lack of oxygen in anaerobic or microaerophilic conditions has long been known to affect the efficacy of antibiotics, we demonstrate that just a 20% change in oxygen saturation significantly affects antibiotic efficacy by allowing the survival of a subpopulation of cells. At the same time, we show that this subpopulation remains vulnerable to the common antibiotic-induced hydroxyl-radical–mediated death pathway if sufficiently high free-radical concentrations can be maintained, induced for example by the antibiotic clofazimine. Previous efforts toward eradicating persister populations have focused on identifying small molecules that specifically kill this population, or which potentiate antibiotic efficacy in this population (44, 45). Next-generation approaches to antibiotic discovery may lie in identifying small molecules that potentiate hydroxyl-radical formation or subvert cellular mechanisms that protect against hydroxyl-radical damage (38, 46). This work now argues that such approaches may have the important added benefit of eliminating persister populations, thus sterilizing an infection (46).

Materials and Methods

Supplementary Material