Dynamic Persistence of Antibiotic-Stressed Mycobacteria

See allHide authors and affiliations

Science  04 Jan 2013:
Vol. 339, Issue 6115, pp. 91-95
DOI: 10.1126/science.1229858


Exposure of an isogenic bacterial population to a cidal antibiotic typically fails to eliminate a small fraction of refractory cells. Historically, fractional killing has been attributed to infrequently dividing or nondividing “persisters.” Using microfluidic cultures and time-lapse microscopy, we found that Mycobacterium smegmatis persists by dividing in the presence of the drug isoniazid (INH). Although persistence in these studies was characterized by stable numbers of cells, this apparent stability was actually a dynamic state of balanced division and death. Single cells expressed catalase-peroxidase (KatG), which activates INH, in stochastic pulses that were negatively correlated with cell survival. These behaviors may reflect epigenetic effects, because KatG pulsing and death were correlated between sibling cells. Selection of lineages characterized by infrequent KatG pulsing could allow nonresponsive adaptation during prolonged drug exposure.

Mutations and genetic exchange are important drivers of microbial diversity, but these events are rare. Often cells display phenotypic diversity that is not due to genetic variations (1, 2). This diversity is critical for microbial persistence in fluctuating environments because it ensures that some individuals may survive a lethal stress that would otherwise extinguish the population. For example, fractional killing of bacterial populations by antibiotics is attributed to phenotypic variants called “persisters” (35). Recent work has also uncovered a similar role for persisters in fractional killing of cancer cells by cytotoxic drugs (6, 7).

Microbial persistence was identified by Bigger in early studies on penicillin, which inhibits bacterial cell wall biogenesis (8). The ineffectiveness of penicillin against stationary-phase bacteria (9, 10) suggested that persisters might constitute a subpopulation of nondividing cells. This concept was confirmed when preexisting subpopulations of infrequently dividing or nondividing Escherichia coli cells were shown to be refractory to ampicillin (11). Stochastic growth rate switching might be an evolved strategy for survival in fluctuating environments (12).

Reversible adoption of a slowly growing or nongrowing state is regarded as a general strategy for persistence. This assumption has been the rationale for persister-enrichment methods (13, 14), evaluation of antibiotic therapies (15), evolutionary models of genetic resistance (16) and social behaviors (17), and drug discovery strategies (18, 19). Here, we show that persistence is not correlated with single-cell growth rates in Mycobacterium smegmatis exposed to isoniazid (INH), which inhibits cell wall biogenesis (20). Instead, cell fate is linked to single-cell dynamics of the INH-activating enzyme catalase-peroxidase (KatG), which is expressed in stochastic pulses that are negatively correlated with survival. KatG pulsing and death are positively correlated between sibling cells, which may reflect epigenetic effects.

Fractional killing is characterized by multiphasic kinetics of population decay. We confirmed that INH killing of M. smegmatis was multiphasic, comprising delay (d), killing (k), and persistence (p) phases (fig. S1). We measured INH killing kinetics at the single-cell level by culturing the bacteria in a microfluidic device (fig. S2) and tracking individual cells by time-lapse microscopy. In antibiotic-free medium, cells grew and divided to form microcolonies (movie S1). Within ~30 min of adding INH to the flow medium, all cells markedly slowed their rates of growth and division. Cell lysis began after ~6 hours and continued throughout 144 hours of INH exposure (Fig. 1A and movie S2). When INH was withdrawn, one of the intact cells in Fig. 1A resumed rapid growth and division; a second exposure to INH killed the progeny with kinetics similar to those of the first exposure (Fig. 1A and movie S2). We conclude that persistence was due to reversible phenotypic tolerance rather than stable genetic mutations.

Fig. 1

Multiphasic dynamics of INH-mediated killing. Bacteria expressing GFP were imaged on fluorescence and phase channels for 212 hours at 15-min intervals and exposed to INH (50 μg/ml) at 12 to 156 hours and at 188 to 212 hours. This experiment was repeated 10 times. (A) Representative image series. Numbers indicate hours elapsed. Scale bar, 5 μm. 7H9, Middlebrook 7H9 medium. (B) Pedigree tree of cells descended from a single progenitor. Broken lines indicate INH addition (12 hours) or withdrawal (156 hours). Bifurcations indicate divisions. End points indicate deaths. Grayscale in the background corresponds to slope of population decay curve shown in (D). (C) Cell size was measured at 15-min intervals: (i) nonpersistent lineage, (ii) persistent lineage. (D to F) Population dynamics reconstructed from single-cell measurements. Broken lines indicate INH addition (12 hours). (D) Summed numbers of surviving cells. (E) Summed sizes (areas) of surviving cells in arbitrary units. (F) Mean cell size (black line) and SD (red lines).

We tracked the fates of individual lineages by generating pedigrees for 153 progenitor cells that were present at the time of INH addition (Fig. 1B and fig. S3A). After 72 or 144 hours of INH exposure, 114 or 136 (respectively) of the 153 lineages were extinguished because the founder cell and all of its progeny were lysed. Unexpectedly, divisions continued under INH exposure; 129 of 153 progenitor cells divided at least once. Figure 1C depicts representative trajectories of an individual that was killed and an individual that continued dividing throughout the entire period of INH exposure. In separate experiments we confirmed that growth, division, and death continued for at least 240 hours under INH exposure (fig. S3B).

Reconstruction of population behavior from single-cell data revealed that INH-mediated killing of microfluidic cultures followed first-order kinetics [y = α exp(–βt), where α is the initial cell number of each phase and y is the cell number at time t after entering each phase], with rate constants β = 3.50 (±0.02) × 10−2 hour−1 during the k phase and β = 0.38 (±0.02) × 10−2 hour−1 during the p phase (Fig. 1D). During the d phase, cell numbers increased (Fig. 1D) while aggregate cell area stayed the same (Fig. 1E), indicating that INH inhibits growth more rapidly than it inhibits division. Reductive divisions caused a sharp but transient decline in mean cell size (Fig. 1F, black line). Single-cell size variation increased with time (Fig. 1F, red lines), which suggested that INH might perturb mechanisms coupling division to cell size.

We tested the hypothesis that persisters are preexisting, either slowly growing or nongrowing, individuals by comparing the pre-INH growth rates of single cells. We found that single-cell growth was size-dependent, such that larger cells tended to grow faster (Fig. 2A). The frequency distribution of cell elongation rates was unimodal and could be viewed as a Gaussian distribution with an additional left tail (P = 0.28, Kolmogorov-Smirnov test for the Gaussian null hypothesis) (Fig. 2B). For cells exposed to INH for 72 or 144 hours, the mean pre-INH elongation rates of persistent and nonpersistent individuals were not significantly different (P = 0.17 or 0.20, respectively, Welch’s t test) (Fig. 2C). We conclude that single-cell growth rates and cell fates are not correlated in INH-stressed M. smegmatis, in contrast to ampicillin-stressed E. coli (11).

Fig. 2

Persistence of single cells under INH exposure. (A) Relationship between cell size A(t) and size increment ΔA(t), where Δt = 1 hour. Circles, single-cell measurements; squares, means ± SE of ΔA(t) in size windows nA(t) < (n + 1) μm2, where n is a positive integer. Histograms are frequency distributions of A(t) (top) and ΔA(t) (right). (B) Single-cell elongation rates were calculated by measuring cell length every 15 min from birth until next division and fitting to exponential curves. (C) Single-cell elongation rates and cell fates under INH exposure. A “persistent” cell produced one or more progeny that survived; a “nonpersistent” cell failed to produce any surviving progeny. Circles, single-cell elongation rates during the 1-hour interval before INH exposure; squares, means ± SD. (D) Rates of division and death under INH exposure. Divisions (squares) and deaths (circles) were counted at 15-min intervals, normalized by total cell number, and averaged over 10 hours ± SE (see supplementary materials).

Single-cell analysis revealed that the transition from fast (k phase) to slow (p phase) killing was due to a decreasing death rate rather than an increasing division rate (Fig. 2D). Consequently, the p phase is a dynamic state in which the rates of cell division and death are balanced, resulting in a deceptively static number of viable cells.

When bacteria divide, each newborn cell inherits a “new” pole from the last division and an “old” pole from a previous division (Fig. 3A). According to a recent report, old-pole siblings elongate faster and are more susceptible to antibiotics than new-pole siblings (21). Although we did not observe a significant difference in elongation rates of old- and new-pole siblings (P = 0.16, Welch’s t test) (Fig. 3B), pole age could potentially influence survival for other reasons.

Fig. 3

Positively correlated survival of sibling pairs. (A) Each individual inherits a “new” pole from the most recent division (age 0) and an “old” pole from a previous division (age ≥1). In time-lapse experiments, pole ages can be assigned from the third generation onward (white cells), excluding the poles of unknown age inherited from the progenitor cell (gray cells). (B) Elongation rates of old- and new-pole siblings. Circles, elongation rates of single cells pre-INH; squares, means ± SD, which were 0.22 ± 0.06 hour−1 for old-pole cells (n = 112) and 0.21 ± 0.07 hour−1 for new-pole cells (n = 112). (C) Four possible fates for a sibling pair. S = 1 if the cell survives until next division; S = 0 if the cell dies before dividing; So, old-pole sibling; Sn, new-pole sibling. (D) Number of single-cell observations of the four fate outcomes (n = 1764 cells). (E) Sibling pair correlation of cell fates. P(S = 1), total survival probability; P(So = 1), survival probability of old-pole cells; P(Sn = 1), survival probability of new-pole cells. Numbers indicate survival probabilities ± 95% confidence intervals.

We tested whether the fates of sibling cells were correlated by counting the number of observations of each of the four possible combinations of cell fates for 1764 sibling pairs (Fig. 3C). A cell “died” if it underwent lysis and “survived” if it divided to produce two daughter cells. The numbers of events in each category (Fig. 3D) were used to calculate the conditional survival probabilities of old- and new-pole cells depending on the fate of the paired sibling. We found that the survival of sibling pairs was positively correlated (P = 2 × 10−7, χ2 test) (Fig. 3E), suggesting the possible contribution of epigenetic effects. This analysis also revealed a weak positive association between survival and inheritance of the old pole (P = 0.027, χ2 test) (Fig. 3E); this association is the opposite of that reported by Aldridge et al. (21).

INH is a prodrug that requires activation by bacterial catalase-peroxidase (KatG) (fig. S4A) (20). We investigated the relationship between KatG levels and INH killing with the use of a strain that expressed katG from an anhydrotetracycline (ATc)–inducible promoter. Cells grown with increasing concentrations of ATc expressed increasing levels of katG mRNA (fig. S4B) and protein (fig. S4C). ATc-induced cultures were exposed to INH and the fraction of surviving colony-forming units was measured. These experiments revealed a sensitive scaling relationship between KatG levels and INH killing, such that a factor of ~2 increase in KatG (fig. S4C) resulted in a factor of ~1000 enhancement of killing at 48 hours (fig. S4D). This effect was not apparent at 24 hours, indicating that overexpression of KatG enhanced killing only during the p phase (fig. S4D).

The sensitivity of INH killing to small changes in KatG expression suggested that cell-to-cell variation in KatG levels might influence cell fate under INH exposure. We tested this hypothesis in a strain in which chromosomal katG was replaced by a katG::dsRed2 fusion gene. The wild-type and katG::dsRed2 strains were equally sensitive to INH (fig. S5). Unexpectedly, single cells expressed KatG-DsRed2 in short-lived stochastic pulses (Fig. 4A and movie S3), separated by intervals in which KatG-DsRed2 fluorescence barely exceeded the detection threshold (Fig. 4B). KatG-DsRed2 pulsing was independent of INH, although cells were brighter in the presence of the drug (movie S3). Native KatG protein was less stable than green fluorescent protein (GFP), which we used as an internal control (Fig. 4C). These results suggest that active degradation contributed to the sharp decline in KatG-DsRed2 fluorescence after peak intensity in pulsing cells.

Fig. 4

Single-cell KatG pulsing and cell fate. Bacteria expressing KatG-DsRed were imaged on fluorescence and phase channels for 88 hours at 15-min intervals and exposed to INH (50 μg/ml) at 16 to 88 hours. This experiment was repeated five times. (A) Representative image series. Numbers indicate hours under INH exposure. Scale bar, 5 μm. (B) Single-cell pedigrees (upper panels) and time traces of KatG-DsRed fluorescence in arbitrary units (lower panels) for nonpersistent (left) and persistent (right) lineages. Black points in upper panels indicate pulse peaks. (C) KatG stability. Cells were pulse-labeled with [35S]Met and [35S]Cys and chased with unlabeled Met and Cys. Left: Pull-downs of cell lysates with antibody to KatG (upper panel) or GFP-Trap (lower panel) were analyzed by SDS–polyacrylamide gel electrophoresis. Right: Band intensities were quantified by scanning autoradiograms. Half-lives were ~0.7 hours for KatG versus ~5.7 hours for GFP. This experiment was repeated twice. (D) Negatively correlated KatG-DsRed pulsing and cell survival. K = 1, pulse; K = 0, no pulse. Values are survival probabilities ± 95% confidence intervals. (E) Sibling pair correlation of KatG-DsRed pulsing. Ko = 1, old-pole cell pulsed; Kn = 1, new-pole cell pulsed (pulsing indicated by red outlining). Values are pulse probabilities ± 95% confidence intervals.

We hypothesized that pulsing cells might be more vulnerable to INH killing as a consequence of increased activation of INH. We tested this idea by comparing the survival probabilities of pulsing (K = 1) and nonpulsing (K = 0) cells. This analysis confirmed that pulsing cells were less likely to survive (P = 7.9 × 10−5, χ2 test) (Fig. 4D). We also found that KatG-DsRed2 pulsing was positively correlated between sibling pairs (P = 7.2 × 10−8, χ2 test) (Fig. 4E), which might explain why cell fate was also correlated between siblings (Fig. 3E).

Historically, fractional killing has been attributed to subpopulations of nonreplicating bacteria on the assumption that antibiotics targeting growth processes should not damage growth-arrested cells (8). Recently, this interpretation has been challenged by the observation that growth arrest per se is not sufficient to explain the refractoriness of stationary-phase Pseudomonas bacteria to antibiotics (22). In populations of exponentially growing mycobacteria, we found no correlation between single-cell growth rates and cell fates, inasmuch as slowly growing individuals were as likely to die (or persist) as rapidly growing individuals. These observations do not rule out the possibility that in other systems, persistence may be linked to single-cell growth rates, as in E. coli exposed to ampicillin (11).

Although bacterial numbers are relatively constant during the p phase of INH killing, we found that this stability masks a dynamic state of balanced division and death. Ongoing division could promote the emergence of resistant genetic variants because the rate of spontaneous mutation increases in proportion to the division rate (23). It has been argued that resistant mutants are unlikely to arise de novo under antibiotic exposure on the assumption that persistent cells are nonreplicating (16, 24). Our discovery that persistent cells may grow and divide in the presence of an antibiotic necessitates a reexamination of how persistence might enhance the generation of resistant mutants.

The link between stochastic expression of katG and the fate of single cells under antibiotic stress might provide insight into the generality of persistence. In principle, stochastic expression of any factor that facilitates or opposes the action of an antibiotic could influence the fate of single cells. The additive contribution of independently fluctuating factors might explain why the link between KatG-DsRed2 pulsing and death, although significant, was not absolute. Consistent with this view, a recent study showed that expression of E. coli HipA toxin from a “noisy” promoter generates supra-threshold fluctuations that protect a fraction of cells from killing by ampicillin (25). These observations raise the possibility that natural cell-to-cell variation in HipA levels might influence survival probability under antibiotic pressure.

Because KatG activates INH, the observed low frequency of KatG pulsing should increase cell-to-cell variation of INH activation (fig. S6A) relative to other modes of KatG expression, such as high-frequency pulsing (fig. S6B) or constitutive expression (fig. S6C). This prediction is consistent with our observation that pulsing and nonpulsing individuals have different survival probabilities. Because low-frequency, high-amplitude pulsing should broaden the single-cell frequency distribution of activated INH concentrations (fig. S6D), this regime is more likely to generate lineages that persist because they fail to activate INH to lethal levels.

Although the mechanistic basis of KatG pulsing is unknown, we envisage several possibilities. First, recent work in E. coli has established that single-cell gene expression is inherently stochastic and “bursty” (26). Burst frequency and amplitude can be tuned by nucleotide changes within the nucleic acid sequences that specify the efficiencies of transcription and translation (27). Thus, KatG pulsing may simply reflect low-frequency, high-amplitude stochastic bursting. Second, recent work in Bacillus subtilis demonstrated that stochastic pulses of transcription are generated by a network architecture comprising a noise-triggered ultrasensitive switch coupled to mixed (positive and negative) feedback loops operating on different time scales (28). Similarly, KatG pulsing might reflect the operation of network motifs that randomly switch the katG promoter between “on” and “off” states. Third, in many bacteria, transcription of katG is induced by oxidative stress (29). Thus, stochastic production of reactive oxygen species during respiratory metabolism (30) could trigger transient pulses of katG transcription. In each of these hypothetical scenarios, instability of KatG would contribute to a sharp rise of KatG levels at the initiation of a transcriptional pulse and a sharp fall of KatG levels at the termination of such a pulse.

INH-mediated selection of cell lineages characterized by infrequent pulsing of KatG could result in gradual adaptation of the population as less-fit phenotypic variants are eliminated. It is also plausible that antibiotics or other extrinsic stimuli might trigger adaptive responses; indeed, environmental factors have been shown to modify the efficiency of antibiotic killing in several systems (3138). However, sense-and-respond adaptive strategies are expensive because the requisite molecular machinery must be expressed constitutively (39). In contrast, nonresponsive, selection-mediated adaptation could provide a simple mechanism for cell populations to adapt to stresses that they had never encountered in their evolutionary histories, or stresses that were encountered too infrequently to repay the investment in responsive mechanisms.

Supplementary Materials

Materials and Methods

Figs. S1 to S6

Movies S1 to S3

References (4043)

References and Notes

  1. Acknowledgments: Supported by fellowships from the Charles H. Revson Foundation and the Heiser Program for Research in Leprosy and Tuberculosis of the New York Community Trust (N.D.); a Harvey L. Karp Discovery Fellowship and the JST PRESTO Program (Y.W.); and the Bill and Melinda Gates Foundation, NIH grant HL088906, and Swiss National Science Foundation grant 310030_135639 (J.D.M.).
View Abstract

Navigate This Article