Mechanisms of bacterial persistence during stress and antibiotic exposure

See allHide authors and affiliations

Science  16 Dec 2016:
Vol. 354, Issue 6318, aaf4268
DOI: 10.1126/science.aaf4268

Structured Abstract


The escalating crisis of multidrug resistance is raising fears of untreatable infections caused by bacterial “superbugs.” However, many patients already suffer from infections that are effectively untreatable due to innate bacterial mechanisms for persistence. This phenomenon is caused by the formation of specialized persister cells that evade antibiotic killing and other stresses by entering a physiologically dormant state, irrespective of whether they possess genes enabling antibiotic resistance. The recalcitrance of persister cells is a major cause of prolonged and recurrent courses of infection that can eventually lead to complete antibiotic treatment failure. Regularly growing bacteria differentiate into persister cells stochastically at a basal rate, but this phenotypic conversion can also be induced by environmental cues indicative of imminent threats for the bacteria. Size and composition of the persister subpopulation in bacterial communities are largely controlled by stress signaling pathways, such as the general stress response or the SOS response, in conjunction with the second messenger (p)ppGpp that is almost always involved in persister formation. Consequently, persister formation is stimulated under conditions that favor the activation of these signaling pathways. Such conditions include bacterial biofilms and hostile host environments, as well as response to damage caused by sublethal concentrations of antibiotics.


The limited comprehensive understanding of persister formation and survival is a critical issue in controlling persistent infections. However, recent work in the field has uncovered the molecular architecture of several cellular pathways underlying bacterial persistence, as well as the functional interactions that generate heterogeneous populations of persister cells. These results confirm the long-standing notion that persistence is intimately connected to slow growth or dormancy in the sense that a certain level of physiological quiescence is attained. Most prominently, the central role of toxin-antitoxin (TA) modules has been explained in considerable detail. In the model organism Escherichia coli K-12, two major pathways of persister formation via TA modules are both controlled by (p)ppGpp and involve toxin HokB and a panel of mRNA endonuclease toxins, respectively. Whereas activation of the membrane-associated toxin HokB depends on the enigmatic (GTPase) guanosine triphosphatase Obg and causes persister formation by abolishing the proton-motive force, mRNA endonuclease toxins are activated through antitoxin degradation by protease Lon and globally inhibit translation. In addition to these two pathways, toxin TisB is activated in response to DNA damage by the SOS response and promotes persister formation in a manner similar to HokB. Beyond TA modules, many additional factors (such as cellular energy metabolism or drug efflux) have been found to contribute to persister formation and survival, but their position in particular molecular pathways is often unclear. Altogether, this diversity of mechanisms drives the formation of a highly heterogeneous ensemble of persister cells that displays multistress and multidrug tolerance as the root of the recalcitrance of persistent infections.


Though recent advances in the field have greatly expanded our understanding of the molecular mechanisms underlying persister formation, important facets have remained elusive and should be addressed in future studies. One example is the upstream signaling input into the pathways mediating bacterial persister formation (e.g., the nature of the pacemaker driving stochastic persister formation). Similarly, it is often not well understood how—beyond the general idea of dormancy—persister cells can survive the action of lethal antibiotics. Finally, one curious aspect of the persister field is recurrent inconsistency between the results obtained by different groups. We speculate that these variations may be linked to subtle differences in experimental procedures inducing separate yet partially redundant pathways of persister formation. It is evident that the elucidation of this phenomenon may not only consolidate progress in the field but also offer the chance to gain insights into the molecular basis and control of bacterial persistence.

Bacterial persisters defy antibiotic treatment.

Persister cells are phenotypic variants of regularly growing bacteria and survive lethal antibiotic treatment in a nongrowing, dormant state. Upon termination of treatment, the resuscitation of persister cells can replenish the population. Our Review focuses on the diverse molecular mechanisms that underlie bacterial persister formation and drive the heterogeneity of these cells. PMF, proton-motive force.


Bacterial persister cells avoid antibiotic-induced death by entering a physiologically dormant state and are considered a major cause of antibiotic treatment failure and relapsing infections. Such dormant cells form stochastically, but also in response to environmental cues, by various pathways that are usually controlled by the second messenger (p)ppGpp. For example, toxin-antitoxin modules have been shown to play a major role in persister formation in many model systems. More generally, the diversity of molecular mechanisms driving persister formation is increasingly recognized as the cause of physiological heterogeneity that underlies collective multistress and multidrug tolerance of persister subpopulations. In this Review, we summarize the current state of the field and highlight recent findings, with a focus on the molecular basis of persister formation and heterogeneity.

Antibiotic-resistant infections are presently a leading cause of anxiety surrounding public health policy (1). Typically, the current escalation in the acquisition of genetic resistance determinants has been seen as the main driver of this phenomenon. But this is only part of the story. Bacterial persistence is increasingly recognized as another major cause of antibiotic treatment failure and relapsing infections (24). Unlike antibiotic resistance that denotes the inherited ability of bacteria to grow in the presence of an antibiotic, bacterial persistence is based on the formation of rare persister cells that transiently display phenotypic tolerance to antibiotic treatment (5). These specialized survivor cells evade antibiotic killing by entering a physiologically dormant state—that is, at the cost of essentially complete abrogation of bacterial growth. The presence of persister cells can be easily detected by the “biphasic killing” phenomenon of bacterial cultures exposed to bactericidal antibiotics: Characteristically, an initial, rapid drop in bacterial counts represents the death of the majority of the population, followed by a second phase with much slower kinetics that reflects the poor killing of persister cells (Fig. 1) (6). This tolerance of persister cells and their ability to reinitiate growth after the termination of antibiotic treatment are commonly seen as main causes for the persistent and relapsing course of many bacterial infections (2, 3). Despite considerable efforts and recent advances in the field, major hurdles to efficient strategies against persisters remain, including a lack of comprehensive understanding of which physiological properties and molecular mechanisms underlie their formation, survival, and resuscitation (3, 5, 7, 8).

Fig. 1 Biphasic killing kinetics of bactericidal antibiotic treatment.

A lethal dose of bactericidal antibiotic added at time zero rapidly eradicates the sensitive bulk of the population (blue) until only nongrowing persister cells (red) that are killed at a slower rate remain. The slower killing has been interpreted to reflect the persister resuscitation rate, but this remains to be substantiated experimentally. The termination of antibiotic treatment enables the population to be replenished by resuscitation of surviving persisters.

Clinical relevance of persisters

Bacterial infections have always been a scourge of mankind, but rates of morbidity and mortality related to these infections have declined over the past century, owing to lifestyle improvements as well as the discovery and extensive use of antibiotics. The prevailing crisis of antibiotic resistance has therefore raised fears that we are descending into a postantibiotic “era of untreatable infections” (9). Already today, however, many patients suffer from bacterial infections that defy massive, long-lasting, and repeated antimicrobial treatment regardless of genetically acquired antibiotic resistance. These infections are often chronic and are never fully cleared by antibiotic treatment because bacteria can persist in biofilms or other protected niches (10, 11). The recalcitrance of bacterial persister cells is well established as a cause for these prolonged and recurrent infections, leading to eventual treatment failure (24, 11). Clinical examples are urinary tract infections with uropathogenic strains of Escherichia coli; the notoriously recalcitrant infections with Mycobacterium tuberculosis; or opportunistic infections of implanted devices, open wounds, and other body lesions typically caused by biofilms of Pseudomonas aeruginosa or Staphylococcus aureus (10, 12).

Heterogeneous populations of persisters that display antibiotic tolerance, slow growth, and the ability to reinitiate infection after antibiotic treatment have been directly observed in diverse animal models of bacterial infection with Salmonella, uropathogenic E. coli, or M. tuberculosis (1217). In vitro, bacterial persisters inside host cells exhibit greatly increased tolerance to antibiotic treatment (12, 15, 18), and the hostile environment in phagocytic vacuoles of macrophages directly induces Salmonella persister formation (15). The standard treatment of chronic infections, which is based on the cyclic administration of high doses of antibiotics, is clearly linked to greatly increased persister levels of clinical isolates and has repeatedly been shown in vitro to rapidly select for mutants displaying particularly high rates of persister formation (1826). Apart from problems with antibiotic tolerance itself, bacterial persisters have also been described as a “catalyst” for the emergence of genetic resistance (19) because different stress signaling pathways commonly involved in persister formation are known to enhance mutation rates and activate mobile genetic elements (27, 28).

Dormancy and the physiology of persisters

Bactericidal antibiotics poison essential cellular processes to cause lethal damage (29). The survival of persister cells is usually explained by their transition into a “dormant” state hallmarked by a reversible and substantial reduction of growth rate and metabolism that protects the cellular processes otherwise poisoned by bactericidal antibiotics (5, 7). In 1944, Joseph Bigger initially discovered persisters and described them as cells surviving antibiotic treatment in a “dormant, non-dividing phase” (30). More globally, it has been proposed that bacterial communities in all ecosystems generate dormant cells as a perseverant “seed bank” that enables the population to recover and repopulate the habitat after catastrophic events (31). Therefore, persisters are likely just one face of a more general bacterial strategy to cope with dynamic and latently hostile environments. This hypothesis is in line with the finding that bacterial killing—not only by antibiotics but also by environmental hazards such as acid stress, toxic metals, or heat—displays biphasic kinetics (32).

However, it is clear that the nonreplicating state of persisters is not sufficient for survival because bacteriostatic conditions do not necessarily induce antibiotic tolerance (33, 34). One study using flow cytometry showed that most dormant cells were not antibiotic-tolerant, although persisters were strongly enriched in the physiologically dormant fraction of the population (35). It is therefore evident that “persisters are not simply non-growing cells” (36) but that persister formation involves specific qualitative changes of bacterial physiology to enable survival and resuscitation. Apart from persisters surviving in dormancy, the stochastic drop in expression of a prodrug-activating enzyme has been described as “dynamic persistence” characterized by transient antibiotic tolerance of growing bacteria (37). However, this mechanism of drug avoidance could also be seen as stochastically expressed resistance rather than as persister formation in the true sense.

Persister formation in bacterial populations

Genetic basis of a phenotypic conversion

Strong experimental evidence supports the notion that the signaling that controls bacterial persistence and the mechanisms directly mediating persister formation are genetically encoded (5, 7, 20, 3842). An alternative model—“persistence as stuff happens”—explains persisters as accidentally nongrowing cells arising from “errors and glitches” (i.e., random cell malfunctioning) (43) but conflicts with the growing literature on defined molecular pathways as the basis of bacterial persistence. The genetic heritability of these pathways enables the frequency of persister formation to adapt to the incidence of antibiotic treatment. For example, recurrent antibiotic treatment is clearly linked to raised levels of persisters in clinical isolates and can rapidly select for increased persistence up to population-wide tolerance in vitro (1826, 44). However, persister formation also has a pleiotropic fitness cost that selects for reduced levels of persistence in the absence of antibiotic treatment, though the evolution along this path appears to be rather slow, at least under laboratory conditions (15, 22, 23). In natural populations, the forces driving the evolution of persister frequencies seem to be highly variable because persister levels differ widely between different species and strains (23, 4547).

Stochastic and responsive persister formation

Bacterial persister formation is driven by a combination of stochastic and responsive mechanisms that allow organisms to respond when harmful conditions are sometimes, but not always, preceded by a stress signal (48). Stochastic persister formation is typically interpreted as bet-hedging—that is, an evolutionary strategy relying on phenotypic heterogeneity to maximize the fitness of an isogenic population in dynamic environments (49, 50). This concept implies that some persister cells have formed before the onset of lethal antibiotic treatment. Direct observations from single-cell microfluidics and flow cytometry show that cells surviving antibiotic treatment had largely been part of a preexisting dormant subpopulation of exponentially growing E. coli (6, 34, 35, 51, 52). Furthermore, bacteria can respond to environmental cues by quantitatively and qualitatively modulating the rate of phenotypic conversion into persister cells, a concept known as responsive diversification (53). As a simple example, the application of sublethal levels of virtually any stress, including antibiotic treatment, has been shown to stimulate bacterial persister formation (26, 41, 54, 55).

Cellular signaling upstream of persister formation

Both stochastic and responsive persister formation are controlled by the same panel of signaling pathways. These include conserved components, such as (p)ppGpp signaling, that are integral to nearly all persister formation, whereas other pathways, such as the SOS response or hypoxia signaling, have mostly modulatory roles (Fig. 2).

Fig. 2 Environmental cues and cellular signals underlying persister formation.

Beyond a basis of stochastic persister formation (A), different environmental cues activate bacterial signaling to induce the formation of persister cells (red) from regular growing cells (blue). For example, the transition into bacterial persistence is strongly induced by the stationary phase (B), sublethal antibiotic treatment (C), and phagocytosis by immune cells (D), as well as in biofilms (E). GSR, general stress response; c, concentration; MIC, minimum inhibitory concentration.

Stringent response and (p)ppGpp signaling

Guanosine tetraphosphate (ppGpp) and guanosine pentaphosphate (pppGpp) collectively act in the “stringent response” as the ubiquitous second messenger and “alarmone” (p)ppGpp that is typically produced in response to nutrient starvation and other stresses to reprogram cellular physiology from growth to metabolic homeostasis and survival functions (56, 57). In E. coli, (p)ppGpp synthetase RelA is activated by amino acid starvation and heat shock, whereas carbon, nitrogen, phosphate, iron, and fatty acid starvation activate the (p)ppGpp synthetase activity of the bifunctional synthetase/hydrolase enzyme SpoT. Once produced, (p)ppGpp modulates bacterial physiology by transcriptional reprogramming and the direct adjustment of target protein activities (5658). Across diverse organisms and experimental models, mutants unable to produce (p)ppGpp often show reduced levels of persister formation. For example, relA spoT mutants of E. coli and Pseudomonas aeruginosa exhibit defective persister formation during exponential growth, in biofilms, and in the stationary phase (33, 40, 51, 59, 60). The magnitude of this phenotype is highly context-dependent, confirming that complementary signaling plays important roles in guiding persister formation. Similarly, (p)ppGpp signaling has been repeatedly implicated in persister formation and antibiotic tolerance in Gram-positive organisms such as Enterococcus faecalis or S. aureus (61, 62). However, the molecular basis of this contribution remains to be elucidated, and one recent study was unable to detect an effect of (p)ppGpp signaling on the persistence of S. aureus under their experimental conditions (63).

Bacterial persister formation can be triggered by a seemingly stochastic activation of (p)ppGpp signaling in rare cells during exponential growth (51) and by environmental conditions that stimulate the production of (p)ppGpp (e.g., in biofilms or during the stationary phase) (Fig. 2). Bacterial biofilms are surface-associated multicellular communities that are characteristically embedded in a thick extracellular matrix and are prevalent in the environment as well as during infections (10, 11). Though the physical structure of biofilms shields the bacterial community from hostile conditions, including immune responses, many antibiotics readily penetrate into the biofilm core (11, 64). Therefore, the notorious recalcitrance of biofilms to antibiotic treatment and their propensity to relapse have largely been attributed to the high amounts of persisters that form inside a structure (11, 64, 65). Biofilms typically harbor 100- to 1000-fold more persisters than planktonic cultures (2, 51, 65, 66). These high rates of persister formation depend on (p)ppGpp as well as on other signaling processes, including diverse stress responses such as SOS induction, hypoxia, or the sessility-associated second messenger cyclic diguanylate (commonly known as cyclic di-GMP) (51, 59, 6668). The stationary phase of bacterial cultures is also hallmarked by a quiescent and highly stress-tolerant physiology induced by the integration of (p)ppGpp signaling and various stress response pathways dominated by the RpoS regulon (69, 70). Consequently, stationary-phase cultures can easily harbor 103 to 104 times more persisters than exponentially growing cultures (26, 51, 65, 71, 72). A special case of persisters are those whose formation is induced by diauxic transitions between using a preferred carbon source and using an alternative one. During such lag periods, (p)ppGpp production is combined with catabolite derepression (i.e., signaling through the second messenger cyclic adenosine monophosphate) (38, 53, 73).

RpoS and the general stress response

The default pathway of a given bacterium to respond to adverse conditions is known as the general stress response. In many proteobacteria such as E. coli and P. aeruginosa, this response largely relies on transcriptional reprogramming via the alternative sigma factor RpoS (74). In E. coli, the general stress response can be induced in the stationary phase by nutrient deprivation and/or (p)ppGpp, temperature stress, biofilm formation, extreme pH, oxidative stress, and several other cues (70). Beyond globally enhancing stress tolerance of the population, RpoS is also a regulator of bacterial persister formation (Fig. 2). The dual role of RpoS in controlling population-wide stress tolerance and persister formation in single cells results in divergent phenotypes of rpoS knockouts: Under conditions that would induce the general stress response, these mutants can display increased or decreased tolerance, depending on the experimental setup (55, 66, 7577), whereas the low levels of RpoS in exponentially growing E. coli do not seem to relevantly contribute to persister formation (33, 74).

The SOS response

The SOS regulon comprises genes mostly involved in DNA repair and is induced in response to DNA damage caused by stochastic cell malfunctioning or various conditions such as oxidative stress, extreme pH, blocked DNA replication, and antibiotic treatment (78). More specifically, the SOS response is ultimately induced by the RecA protein that is activated by single-stranded DNA generated from the processing of damaged DNA (78). The role of the SOS response in bacterial persistence is twofold: first as a pathway of complementary stress signaling that modulates persister formation and second as a provider of diverse DNA repair functions that are important for persister resuscitation (see below). Consistently, SOS-deficient mutants usually exhibit a considerable drop in the number of surviving persisters, particularly upon treatment with DNA-damaging agents (26, 41, 67, 75, 79). However, it is often not clear which of the two roles of the SOS response is responsible for this phenotype.

Bacterial communication

Since bet-hedging and responsive diversification optimize the fitness of a clonal population by managing the investment of community resources into phenotypically distinct subpopulations, it is not surprising that bacterial communities coordinate persister formation with the help of intercellular signaling molecules (Fig. 2). For example, quorum sensing of pyocyanine or acyl-homoserine lactone and the stress-inducible quorum-sensing peptide CSP (competence-stimulating peptide) can induce persister formation of P. aeruginosa and Streptococcus mutans, respectively (80, 81). Similarly, the widespread bacterial signaling molecule indole can modulate persister formation of E. coli, though the molecular mechanism of indole sensing is still a matter of debate (8, 82).

Persister heterogeneity

The original concept of persister dormancy as a state in which antibiotic targets are generally inactive implied that persister cells should exhibit a rather homogeneous profile of multidrug tolerance. However, direct experimentation typically revealed divergent profiles of antibiotic tolerance and susceptibility among persisters in a given culture (38, 45, 46, 83). True multidrug-tolerance was detected for only a small proportion of persister cells, but one may speculate that the heterogeneity of persisters enables survival of the population under a wide range of harmful conditions, as expected from the concept of bet-hedging. The differential tolerance and cross-tolerance patterns of cells within the persister subpopulations result from physiological differences among the persister cells, probably reflecting different molecular mechanisms of their formation. Mutations affecting bacterial persister formation often compromise tolerance to different antibiotics unevenly, which has also been observed with persisters induced by special growth conditions or genetic manipulations (33, 38, 41, 66, 84). For example, persisters tolerant to β-lactams and fluoroquinolones are induced by broadly similar cellular signaling in the diauxic transition model of E. coli, but full β-lactam tolerance required higher levels of (p)ppGpp than fluoroquinolone tolerance (38).

Persister formation by TA modules

Biology of toxins and antitoxins

Bacterial toxin-antitoxin (TA) modules are genetic elements composed of a toxin protein that inhibits bacterial growth by interfering with essential cellular processes and an antitoxin that prevents or impairs the functionality of the toxin until this inhibition is abrogated in response to cellular signaling (85, 86). TA modules are classified according to the nature of the antitoxin and its mechanism of action, but thus far, only type I and type II TA modules have been widely studied in the context of persister formation. Whereas the antitoxins of type I TA modules inhibit toxin expression as antisense RNAs (87, 88), the antitoxins of type II TA modules are proteins that inactivate their cognate toxins in direct protein-protein interactions (85, 86). Type I toxins are usually small proteins that form pores in bacterial membranes to collapse the proton-motive force and halt adenosine triphosphate (ATP) synthesis, but they can also act via different mechanisms (87). The molecular activities of type II toxins are highly diverse, but most are translation inhibitors (85). For example, type II toxins can impair translation as ribosome-dependent or -independent mRNA endonucleases (RelE or MazF families, respectively); cleave rRNA or tRNA molecules (VapC family); or inactivate elongation factors, tRNA, and tRNA synthetases by posttranslational modification (Doc, TacT, and HipA families, respectively) (85, 86, 89, 90).

TA modules in stress and persister formation

Activation of TA modules requires sufficient expression of the toxin component and a disruption of the toxin/antitoxin balance in favor of the toxin. Type I and II TA modules are usually controlled by a combination of transcriptional and posttranscriptional regulation, as well as antitoxin degradation in response to cellular signaling (87, 91). In E. coli, the tisB/istR and hokB/sokB type I TA modules are activated by the SOS response and (p)ppGpp signaling, respectively (40, 41). Type II antitoxins are usually degraded by the protease Lon in response to (p)ppGpp signaling or oxidative stress, although degradation by ClpP has also been described (51, 9193).

Graded activation of TA modules in a way that does not completely abrogate bacterial growth enables cells to adjust growth rates and modify bacterial physiology to enhance stress tolerance (86, 94, 95). However, the activation of TA modules mostly exerts a robust switching into dormancy once free toxin levels have crossed a certain threshold, and this behavior is typically supported by transcriptional autoregulation (96, 97). The enforced expression of toxins of diverse type I or II TA modules often causes bacteriostatic growth inhibition that coincides with massively increased antibiotic tolerance, thus directly implicating TA modules in persister formation (3941, 52, 81, 84, 98). Consistently, the transcriptomes of persister cells reveal an up-regulation of TA modules in E. coli and M. tuberculosis (52, 72, 84, 99). Furthermore, a considerable jump in antibiotic tolerance has been obtained by the treatment of E. coli with drugs that mimic the action of type I or II TA modules (100), and clones selected for increased persister levels often carry known or suspected gain-of-function mutants of TA modules (18, 20, 21, 24, 44). However, genetic screens have mostly failed to detect any strong effect of TA module loss-of-function mutants on persister frequencies under standard conditions (77, 81, 101, 102), indicating that persister formation by TA modules exhibits considerable functional redundancy. This hypothesis has been experimentally confirmed in E. coli, where the consecutive deletion of 10 mRNA endonuclease TA modules showed the cumulative effects of different TA modules only after the first 5 had been knocked out (39).

Notably, single TA module knockouts occasionally display defects in persister formation, although these phenotypes are often very specific. For example, single knockouts in any mRNA endonuclease TA module had no effect on persister formation in exponentially growing E. coli (39). However, a mutant of yafQ (a RelE family mRNA endonuclease) displayed markedly reduced levels of persister cells in biofilms (103), and mqsR (another RelE homolog) or relE knockouts were impaired in the formation of long-time tolerant stationary-phase persisters (75). These results suggest that the roles of given TA modules in persister formation in natural environments may be highly context-dependent, and different TA modules may possibly be wired to distinct upstream signaling pathways (93, 104). Furthermore, it is clear that the different molecular mechanisms of distinct TA module toxins result in the formation of physiologically distinct persister cells. Consistently, the overexpression of different toxins induces distinct profiles of antibiotic tolerance (41, 52, 84, 105).

The observed accumulation of large numbers of TA modules—particularly in organisms that are adapted to dynamic environments including chronic, persistent infections—likely mediates both functional redundancy as well as the characteristic heterogeneity of persister cells. Typical examples for organisms that fit this pattern are E. coli, M. tuberculosis, and Salmonella enterica serovar Typhimurium (106). A recent dedicated study further showed that persister levels in different strains and species of the genus Pseudomonas correlate with the number of type II TA modules (47).

TA modules in the persistence of E. coli K-12

By far, the molecular basis of bacterial persister formation has been most intensively studied in the model organism E. coli K-12. This organism encodes at least ~30 TA modules: Of these, roughly half are type I and half are type II, with only few belonging to other types (85). A considerable number of these TA modules contribute to persister formation in three major pathways that have been elucidated in detail and provide useful examples for the molecular basis of bacterial persistence in general (Fig. 3).

Fig. 3 Toxin-antitoxin modules and persister formation in E. coli.

Three major pathways of persister formation in E. coli K-12 that rely on the activation of toxin-antitoxin (TA) modules are presented in detail, with mechanisms promoting and counteracting persister formation shown in red and blue, respectively. The Obg/HokB pathway (A) and the polyphosphate/Lon/mRNA interferase pathway (B) are under the control of (p)ppGpp signaling, whereas the TisB pathway (C) is activated upon strong SOS induction. Type I toxins HokB and TisB induce persister formation by abolishing the proton-motive force (PMF) as membrane-associated peptides, whereas the 10 mRNA endonuclease type II toxins interfere with ribosomal translation. Note that the latter pathway involves both positive feedback (via HipA) and negative feedback (via the drop of mRNA levels). E, P, and A denote ribosomal exit, peptidyl, and aminoacyl tRNA binding sites, respectively; Pi, inorganic phosphate.

mRNA endonucleases under control of (p)ppGpp and Lon

A set of 10 mRNA endonuclease type II TA modules in E. coli K-12 is under the control of (p)ppGpp, inorganic polyphosphate, and protease Lon in a linear, hierarchical signaling pathway (Fig. 3) (51). During rapid growth, this pathway is triggered in rare cells that contain high levels of (p)ppGpp as a result of seemingly random physiological fluctuations (51). Through competitive inhibition of the polyphosphate hydrolase PPX, (p)ppGpp causes the accumulation of polyphosphate produced by polyphosphate kinase PPK which, in turn, stimulates protease Lon to degrade type II TA module antitoxins. Most importantly, the activation of 10 mRNA endonuclease TA modules inhibits global translation and thereby induces persister formation (39). Moreover, Lon degrades the HipB antitoxin of the HipBA type II TA module and hence frees the HipA toxin that inactivates aminoacyl-tRNA synthetase GltX by phosphorylation at its active site (92, 107). The resulting appearance of uncharged tRNAs at the ribosomal A site triggers RelA-dependent (p)ppGpp synthesis, presumably inducing a positive-feedback loop that causes sustained activation of the 10 mRNA endonucleases and HipA (34). In parallel, the mRNA endonuclease toxins limit the entry of uncharged tRNA into the ribosome through mRNA degradation and can thereby control the amplitude and duration of (p)ppGpp peaks via negative feedback (108). The critical role of the “master TA module” HipBA is highlighted by the repeated isolation of hipBA gain-of-function alleles upon selection for increased persister formation of E. coli in vitro, as well as in clinical strains. These alleles caused substantially increased levels of persister formation (e.g., by interference with transcriptional autoregulation of the TA module) (18, 20, 109).

The SOS response and SOS-regulated TA modules

Among a diverse array of functions predominantly mediating DNA repair, the SOS response of E. coli controls the activation of two type I TA modules (tisB/istR and symE/symR), as well as two type II TA modules (dinJ/yafQ and yafNO) (67, 78). Whereas the SOS-dependent DNA repair functions seem to be generally essential for the survival of fluoroquinolone treatment (see below), the TisB toxin was found to be additionally required for the formation of highly fluoroquinolone-tolerant persisters by exponentially growing E. coli (41). TisB is a small protein that forms pores in the bacterial inner membrane and thereby disrupts the proton-motive force that inhibits ATP synthesis and induces dormancy (Fig. 3). Endogenously expressed TisB only affected tolerance to fluoroquinolones, whereas the overexpression of TisB caused multidrug tolerance (41). It therefore appears that the levels of SOS induction achieved by random activation in exponentially growing E. coli are insufficient to trigger TisB expression, possibly due to the lack of a positive-feedback loop like the one implemented by HipBA for the formation of mRNA endonuclease persisters (see above).

HokB toxin under the control of Obg and (p)ppGpp

Obg is an enigmatic, yet essential and highly conserved guanosine triphosphatase (GTPase) believed to adapt ribosome functioning, DNA replication, and stress responses to the energy status of the cell. A recent study showed that cellular levels of Obg determine the proportion of persister cells in the population (40). This property of Obg depended on the production of (p)ppGpp, which caused up-regulation of the type I toxin HokB in accord with Obg levels (40). Like TisB, HokB is a membrane-targeted type I toxin that disrupts the proton-motive force to impair ATP synthesis and induce dormancy (Fig. 3) (40).

TA modules in persistence of other bacteria

In Salmonella Typhimurium, different subsets of the at least 5 type I and 13 type II TA modules of this organism were shown to promote survival inside fibroblasts and epithelial cells, respectively (110). Many of these TA modules were also activated by the acidic and nutrient-poor environment in vacuoles after phagocytosis by macrophages, which induced the formation of a heterogeneous set of multidrug-tolerant persisters in a manner dependent on (p)ppGpp signaling, Lon, and diverse type II TA modules (Fig. 3) (15). In contrast to the seemingly similar persister pathway of E. coli K-12, single TA module knockouts showed strong defects in persister formation in this model, suggesting that the activity of different TA modules may be linked by some kind of regulatory mechanism (15, 110). The notoriously persistent pathogen M. tuberculosis encodes at least ~80 TA modules, of which many have been shown to be functional in vivo and to differentially affect antibiotic tolerance or virulence in animal infection models (90, 111113). Though the interplay of different TA modules in this repertoire is not understood at the global level, the synergistic contribution of some subsets to stress and drug tolerance under given conditions suggests a certain degree of functional specialization (112, 113).

Unlike these organisms, P. aeruginosa encodes comparatively few easily detectable TA modules for a generalistic organism that is obviously adapted to highly dynamic environments [six and eight predicted type II TA modules in strains PAO1 and PA14, respectively (114)]. Though persister levels in Pseudomonas seem to be generally linked to the abundance of type II TA modules (47), none of these has ever been demonstrated to participate in persister formation. Instead, published work suggests that P. aeruginosa largely relies on other strategies, such as the enforced activity of antioxidative defenses (see below) and/or a pathway dependent on Obg but, unlike in E. coli, lacking an obvious homolog of HokB (40, 59). Similarly, it was recently shown that the deletion of known TA modules in S. aureus did not affect bacterial persistence and that persisters formed by this organism exhibit decreased levels of ATP (63). It will be interesting to see what the molecular basis of the latter phenomenon is and, more generally, whether future studies may uncover a role of TA modules also for persister formation of P. aeruginosa and S. aureus (possibly involving yet undetected TA loci).

Other mechanisms of persister formation

Many factors beyond TA modules have been proposed to participate in persister formation, but the evidence for many of these is largely genetic, so that the phenotypic basis of antibiotic tolerance remains elusive. As an example, mutants affected in amino acid biosynthesis or metabolism have repeatedly been found to display altered (usually decreased) persister formation in diverse setups (33, 67, 81, 101, 109). However, it is unclear whether these phenotypes are simply caused by direct or indirect effects on (p)ppGpp signaling by RelA or whether other aspects of bacterial physiology are involved. Furthermore, diverse genes of the flagellar machinery have recently been implicated in aminoglycoside tolerance, but the molecular basis of this finding has not been resolved (102).

Inactivation of antibiotic targets

Aminoglycosides: Ribosomal translation

Aminoglycoside drugs kill bacteria by corrupting ribosomal translation to cause the production of toxic peptides, and drug uptake is known to require the proton-motive force (29). Ribosome hibernation, the formation of translationally inactive ribosome dimers, affects the bactericidal activity of aminoglycosides on both levels by inhibiting translation and consequently reducing the proton-motive force (Fig. 4). Consistently, ribosome hibernation has been found to be critical for the formation of aminoglycoside-tolerant persisters by several bacteria (115). Ribosome modulation factor, the main player in the ribosome hibernation process, is strongly expressed in E. coli persister cells, as well as in dormant P. aeruginosa biofilm cells (66, 84).

Fig. 4 Mechanisms of persister formation beyond TA modules.

Diverse mechanisms beyond TA modules have been shown to contribute to bacterial persister formation. A nonexhaustive, exemplary set of factors involved in respiration and energy production (A), drug efflux pumps (B), and direct inactivation of antibiotic targets (C) are presented in this illustration. Note that mutants of complexes of the electron-transport chain (I to IV), ATP synthase (V), and different enzymes of the TCA cycle had divergent phenotypes in different studies or under different conditions (blue and red for increased and decreased persister formation, respectively). A nuoN mutant (at complex I of the electron-transport chain) displayed increased persister formation, but it is not known whether this phenotype is caused by a gain-of-function or loss-of-function mutation (22).

Fluoroquinolones: Topoisomerases and DNA replication

Fluoroquinolone antibiotics bind to DNA gyrase and topoisomerase IV in a way that promotes the formation of a ternary complex in which the enzyme is locked onto its DNA substrate and covalently bridges a DNA double-strand break, impairing DNA replication and gene expression. The bactericidal effect of fluoroquinolones is generally believed to be caused by the destabilization of these covalent complexes to release the free DNA ends—for example, by collision of these “roadblocks” with DNA-tracking complexes such as replisomes—resulting in chromosome fragmentation (116). Because ternary complex formation does not require enzymatic activity of the target enzyme, fluoroquinolone tolerance mechanisms may instead need to focus on the DNA side of antibiotic action. Consistently, several inhibitors of DNA replication (such as UmuDC or CspD) have been implicated in persister formation (42, 84), and (p)ppGpp itself also has dampening effects on transcription, as well as on DNA replication (56, 58, 117) (Fig. 4). Furthermore, the formation of fluoroquinolone-tolerant persisters has been studied in some detail for the carbon-source transition model (38, 73). These persisters are induced by (p)ppGpp signaling at a certain threshold level of this alarmone and require the transcriptional effects of (p)ppGpp and several nucleoid-associated proteins (73). However, the exact molecular basis of fluoroquinolone tolerance in this model remains unknown.

β-lactams: Peptidoglycan synthesis and cell growth

β-lactam antibiotics poison the cell wall synthesis machinery to induce a futile cycle of peptidoglycan synthesis and degradation that impairs cell wall integrity and results in osmotic lysis, particularly of growing cells (118). Apart from the general notion that persisters may survive β-lactam treatment because they do not replicate, a recent study unraveled a mechanism for the formation of ampicillin-tolerant persisters that links progressively increasing levels of (p)ppGpp to a broad shutdown of peptidoglycan synthesis (38) (Fig. 4).

Drug efflux pumps

A recent study showed that E. coli persister cells harbor higher levels of drug efflux pumps than the rest of the population and that these contribute to tolerance against β-lactams and quinolones by reducing intracellular antibiotic concentrations (Fig. 4) (119). The induction of efflux pumps has also been shown to contribute to antibiotic tolerance in other contexts; for instance, upon infection of host cells for mycobacteria (13) or as a component of indole-triggered persister formation in E. coli (8, 82).

Oxidant tolerance

The production of reactive oxygen species (ROS) and consequent damage to cellular macromolecules had been proposed as a common contributor to killing by diverse bactericidal antibiotics, but this view has been repeatedly challenged (29, 120, 121). Nevertheless, a study on P. aeruginosa showed that the survival of multidrug-tolerant persisters in biofilms of this organism was largely dependent on ROS detoxification enzymes under the control of (p)ppGpp signaling, and similar mechanisms may also operate in E. coli (59, 122).

Energy metabolism

A large body of evidence supports a connection between persister formation and the electron-transport chain, including the sources of electrons from the tricarboxylic acid (TCA) cycle and glycerol-3-phosphate dehydrogenase (GlpD) (Fig. 4). However, this notion is largely based on genetic experiments that detected usually decreased, but occasionally increased, frequencies of antibiotic-tolerant persisters in strains carrying mutations or insertions in genes such as glpD, ubiF, sdh, sucB, acnB, mdh, and diverse others (22, 83, 102, 123127). Phenotypes of decreased persistence were often very strong, consistently observed under diverse growth conditions, and frequently accompanied by increased sensitivity to a variety of stresses, ranging from extreme pH to heat or oxidants (123, 124). It therefore seems likely that these mutations generally impair cellular stress responses via metabolic deregulation, and defects in persister formation may be downstream of this phenomenon.

Persister resuscitation

On top of mere tolerance to the direct action of antibiotics, persister survival critically depends on the maintenance of cellular integrity throughout persistence and the resuscitation of cell functions from dormancy. Persister resuscitation is typically seen as a stochastic event that occurs randomly in time and has been interpreted in the context of a “scout model,” where the randomness of persister awakening allows the population to continuously probe the environment for favorable conditions (128). However, the frequency of persister resuscitation is also influenced by resource availability and other environmental variables such that exposure to preferred carbon sources, for example, can induce pronounced persister awakening (129). Molecular mechanisms underlying the exit from dormancy have remained rather elusive, with the exception of persisters formed by the activation of type II TA modules. For these, a specific mechanism of transcriptional autoregulation, conditional cooperativity, has been suggested to modulate the toxin/antitoxin balance in favor of resuscitation (85, 97), and the posttranslational inactivation of toxins or the regeneration of corrupted targets may also contribute (89, 130).

Repair of DNA damage

Although the repair of damaged cellular macromolecules after antibiotic treatment is likely a general requirement of persister resuscitation, it has only been studied in the context of DNA damage. Two studies have shown that diverse components of the SOS regulon are involved in the repair of fluoroquinolone-induced DNA damage in stationary-phase persisters of E. coli (79, 131). Surprisingly, SOS induction caused by fluoroquinolone treatment was not different between persisters and nonpersisters, suggesting that the initial effects of the drug on the nucleoid were similar. Instead, survival of persisters depended on the abundance and activity of the DNA repair machinery primarily during the resuscitation phase, and enforced repression of the SOS response after fluoroquinolone treatment greatly reduced persister survival (79).

Overcoming persistence

Given that endurance is the key property of persisters as specialized survivor cells, it is unsurprising that their elimination is problematic for clinical situations. However, new insights into the central principles of persister cell formation and tolerance have enabled a recent surge of ideas about how to deal with the phenomenon by, for example, enhancing drug uptake and preventing or reversing persister formation (8). The observation that persister formation usually depends on (p)ppGpp signaling has inspired the use of (p)ppGpp synthesis inhibitors, such as relacin, as adjuvant drugs for antibiotic treatment (132). Furthermore, the acyldepsipeptide antibiotic ADEP4 is notable for its ability to kill persister cells, irrespective of their dormant physiology. ADEP4 can enforce activation of the ClpP protease by uncoupling it from the requirement to use ATP, thereby subverting ClpP protease activity to target a wide range of essential targets, such as ribosomal proteins (133).

Concluding remarks and prospects

Recent studies shed new light on the molecular mechanisms underlying bacterial persistence, and—although many details remain to be resolved—diverse cellular pathways, including TA modules and metabolic rearrangements, are well established as mediators of persister formation. However, important aspects of the cellular decision-making process upstream of these mechanisms are still unclear, particularly in the context of seemingly random persister formation where the pacemaker of this stochasticity has remained elusive. More specifically, the question of whether this behavior is truly stochastic (i.e., simply following an amplification of noise) or equivalent to the induction of persister formation by external cues experienced by some cells in their local environment [e.g., “microstarvation” as a trigger of (p)ppGpp signaling (51)] is still to be resolved. Furthermore, many open questions persist regarding the actual basis of antibiotic tolerance downstream of established mediators of persister formation (i.e., regarding those aspects that are most critical for failure or success of clinical treatment). For example, it remains a riddle how exactly persisters that arise through inhibition of protein translation by mRNA endonucleases can be protected from fluoroquinolones that corrupt DNA topoisomerases. Similarly, it is currently unclear why tolerant persisters are still slowly killed during antibiotic treatment (e.g., as apparent from the slope of the second phase of biphasic killing experiments). Furthermore, the question of which molecular mechanisms control the timing of persister resuscitation remains to be resolved.

One curious aspect of the research on persister cells is the frequent lack of congruence in results reported by different laboratories. As an example, the persister pathway driven by (p)ppGpp, Lon, and mRNA endonucleases in E. coli (Fig. 3) has been the topic of some controversy. Several groups noted that, in their hands, a lon knockout had no effect on persister formation that could be attributed to the 10 mRNA endonucleases (131, 134), though others could readily reproduce the lon phenotype (75, 119). Similarly, the formation of fluoroquinolone-tolerant persisters of E. coli has repeatedly been explicitly noted to depend on the TisB toxin only in exponentially growing cells (41, 79), but recent work in the stationary phase showed markedly decreased fluoroquinolone tolerance of a tisB mutant (75). Subtle differences in the experimental setup probably favor the formation of persisters by different pathways (83). Clearly, these discrepancies need to be resolved and understood, both as a means to learn more about the control of heterogeneous bacterial persister formation and also to facilitate the development of common concepts in the field.

References and Notes

  1. Acknowledgments: This work was supported by the Danish National Research Foundation–funded Centre of Excellence BASP (grant DNRF120), a Novo Nordisk Foundation Laureate Research grant, and the European Research Council Advanced Investigator grant PERSIST (294517). A.H. is grateful for the support of C. Dehio and a European Molecular Biology Organization Long-Term Fellowship (ALTF 564-2016). The authors do not declare any competing interests.
View Abstract

Stay Connected to Science

Navigate This Article