Beyond metabolic dormancy: metabolic rewiring in bacterial persistence

Abstract

Bacterial persister cells that exhibit a transient state of antibiotic tolerance play a key role in chronic and recurring infections. Despite advances in our understanding of persisters, many aspects of their phenotypic plasticity, particularly their metabolism, are poorly characterized. In this Perspective, we examine the static and dynamic characteristics of persister metabolism that are shaped by genetics, environmental cues, and the intrinsic variability in cellular processes. These factors underlie much of the diversity observed among persister cells and largely explain the inconsistency in expression of classic persister hallmarks such as biphasic killing curves or metabolic dormancy. Further, the literature suggests that persisters are not uniformly metabolically dormant but represent a range of metabolic states. We will focus on the unique rewiring of the metabolic mechanisms in persisters, which depends on both internal and external factors.

Bacterial persister cells exhibit a transient state of antibiotic tolerance and are commonly assumed to be metabolically ‘dormant’. In this Perspective, Orman et al. re-examine common assumptions and emphasize that persisters represent a range of metabolic states, driven by internal and external factors, which may explain the inconsistent expression of classic persister hallmarks.

Introduction

In the early 1940s, Gladys Hobby and Joseph Bigger observed a small bacterial subpopulation that survived penicillin treatment, which they called persisters^1,2. However, despite eight decades of research, the physiology of these cells is poorly understood. Unlike resistant mutants with an altered genotype, persisters represent a small phenotypic subpopulation that exists in a transient, typically non-growing or slow-growing, antibiotic-tolerant state. Scientists in this field have long emphasized this distinction, noting an increased minimum inhibitory concentration (MIC) in genetically resistant mutants that is absent in persisters³. Antibiotic persistence differs from antibiotic tolerance, which represents population-level survival³. However, the boundaries for persistence, tolerance, and resistance may overlap (Box 1). Persisters can provide a reservoir from which resistant mutants eventually arise^4–6, and some mutations increase persistence without altering MIC. High-persister mutants can be responsible for refractory symptoms and frequent relapses in chronic and biofilm-associated diseases^7–10, which can result in recalcitrant infections that place a heavy burden on global healthcare systems.

Persisters are notoriously challenging to characterize because they are rare, have a transient tolerant state, and are difficult to isolate. Monitoring individual persister cells using fluorescent reporter proteins with single-cell imaging captures only a small number of cells¹¹, limiting the efficiency of these measurements. Moreover, distinguishing persisters is complicated by the greater number of viable but nonculturable cells (VBNC) in antibiotic-treated cultures^12–14. Like persisters, VBNCs are often non-growing cells with a basal level of metabolic activity and can survive antibiotic treatment; however, VBNCs cannot form colonies in a standard medium after the removal of the antibiotic¹⁵ (Box 2). Although persister cells form colonies when the antibiotic is removed, the mechanisms and the time required for their phenotypic switch vary with cell type, antibiotic, and environmental conditions.

Persister metabolism may provide a rich source of targets for drug development^16–18. However, understanding the metabolism of persisters is hampered not only by their low abundance and transient state, but also by their metabolic heterogeneity¹⁹. According to conventional wisdom, persisters avoid antibiotics by “playing dead,” i.e., shutting down their metabolism completely. However, this may not always be the case; some persisters can modify their metabolism to facilitate their entry into, maintenance of, and ultimate exit from their antibiotic-tolerant state¹⁷. Less a contradiction of “playing dead,” this modification likely reflects the context-dependent complexity of persister physiology, which makes persisters both elusive and intriguing.

Persister cells are not uniform, and the heterogeneity in metabolic states among persisters is key to understanding their behavior. As the persister literature has been reviewed extensively^3,20–22, we focus here on factors that address apparent conflicts in our understanding of persister metabolism; thus, we do not cite all prior studies. In line with this objective, we focus here on metabolic mechanisms in persisters, specifically alterations in core metabolic processes such as energy metabolism and macromolecule synthesis (e.g., DNA, protein, and cell wall synthesis). Based on our analysis of the extensive research by many labs in this field, we suggest that apparently contradictory findings may have resulted from the use of simplified or narrow definitions of persister hallmarks. As science advances, our definitions and frameworks must also evolve. Thus, re-examining established assumptions is a mark of the strength and maturity of the field.

Box 1: The spectrum of antibiotic survival: persistence, tolerance, resistance, and heteroresistance.

Antibiotic persistence describes subpopulation-level tolerance in a genetically susceptible bacterial population, where a small fraction of cells a) survives a bactericidal antibiotic by entering a transient tolerant state, b) has no increase in the MIC, and c) yields fully susceptible progeny after regrowth³.

Antibiotic tolerance describes a population-wide reduction in the rate of antibiotic killing without a change in MIC, allowing long-term survival of genetically susceptible cells to bactericidal treatment³. Unlike persistence, which affects only a small subpopulation of cells, tolerance slows killing uniformly across the entire population. Although the term “tolerance” has been used historically to broadly describe any form of bacterial survival under chemical or environmental stress, more recent studies^101,102 together with a consensus statement in 2019³ led to a formal definition of this distinct state. Prior to this definition, the term described cells with many different phenotypes. Web of Science records more than 13,000 articles before 2019 that include “tolerance” in the context of bacteria. In many of these articles, “tolerance” refers to stress endurance, adaptive survival responses, or reduced susceptibility following diverse stresses at population or subpopulation levels, often without distinguishing phenotypes based on specific killing kinetics. Therefore, in our opinion, the use of the term “population-level tolerance” instead of just “tolerance” in the consensus statement³ would have more clearly reflected the updated definition and maintained continuity with earlier literature.

Antibiotic resistance is the ability of bacteria to grow and reproduce in the presence of an antibiotic, typically through inherited genetic changes that raise the MIC³.

Antibiotic heteroresistance describes a bacterial population that appears susceptible to an antibiotic but contains a small subpopulation of cells with a higher MIC than the majority population⁷⁴. These high-MIC subpopulations may arise from genetic changes such as unstable gene amplifications or transient copy-number variants that have a fitness cost¹⁰³. Thus, these resistant cells expand only under antibiotic selection and collapse once the drug is removed. However, there are also subpopulations with transient, reversible elevations in MIC that proliferate in the presence of an antibiotic without genetic resistance. These cells may have phenotypic changes such as regulatory noise, bistable gene expression, fluctuations in the expression of efflux pumps and porins, or activation of alternative pathways that bypass the drug target. Although these nonmutational variants resemble heteroresistant subpopulations, key studies in two different types of cells have classified them as persisters: (i) In Mycobacterium smegmatis, isoniazid tolerance arises from stochastic, lineage-correlated fluctuations in KatG that limit prodrug activation, allowing a subset of persister cells to survive and proliferate in the presence of the drug⁹⁹. (ii) In cancer, rare cycling persister lineages persist under targeted therapy by enhancing NRF2-mediated antioxidant defenses and shifting metabolism toward mitochondrial fatty-acid β-oxidation¹⁰⁴.

Box 2: VBNC cells complicate the identification of persisters.

Viable but nonculturable (VBNC) cells are alive, but in a stable dormant state. They cannot form colonies on standard laboratory media and typically require specific conditions for resuscitation. VBNC cells arise under a wide range of stresses, including nutrient limitation, temperature fluctuations, oxidative and osmotic stress, pH shifts, heavy metals, and sublethal antibiotic exposure, and can persist for extended periods^12,13. Although VBNC cells have been described as dead cells or persisters^105,106, they are metabolically active, express genes, and maintain intact membranes^15,107. Further, depending on how the VBNC state was induced, these cells revive with the removal of the inducing stress, such as appropriate nutrient enrichment, temperature upshift, and catalase or pyruvate supplementation^12,13. Clinically, VBNC states are likely recognized more frequently than persisters, as many bacteria persist asymptomatically in the host for long periods without growth¹⁰⁸. Thus, these cells are not identified by culture-based diagnostic tests but are detected by techniques such as RT- qPCR, or metagenomic sequencing^109,110. VBNC cells are an important health concern because they survive in hostile host environments, persist in chronic or asymptomatic infections, and evade standard diagnostics¹³. However, they may then resume growth, resulting in relapse, culture-negative presentations, or underestimation of pathogen burden. Persisters can be distinguished from VBNCs in vitro by live-cell imaging, which tracks subsequent regrowth, and VBNC cells can be quantified by flow cytometry using viability staining or by β-lactam cell lysis assays¹⁵. However, VBNC cells still complicate persister research because (i) culture-based assays cannot distinguish VBNC cells from dead cells or from persisters, (ii) VBNC cells contaminate persister isolations, (iii) conditions that eliminate persisters may push cells into the VBNC state, and (iv) interpretation of molecular analyses is often complicated by the stress responses that are common to persisters and VBNC cells^{12,13,15,63,111}.

Metabolic rewiring: the driving force of the persister life cycle

Metabolism drives the persister life cycle. Persisters do not “turn off” metabolism; instead, as we describe here, they rewire it. By metabolic rewiring, we mean a shift in how cellular energy is allocated; while persisters may reduce energy-intensive anabolic and central dogma processes, they redirect energy toward life-sustaining activities. During persister formation, a subpopulation of cells slows or arrests proliferation in response to metabolic signaling pathways, including the stringent response, carbon catabolite repression, and stress-induced regulatory networks that detect shifts in nutrient availability, cellular damage, and other stresses^19,23–25. Given their reduced proliferation, persisters are expected to reduce energy-intensive anabolic activities such as ribosome biogenesis, cell wall synthesis, and DNA replication, but may maintain energy-producing processes such as respiration, proton motive force generation, and ATP synthesis²⁶, with an emphasis on stress protection and repair systems^27–29. These metabolic circuitries may sustain long-term survival under antibiotic exposure and later support the reversal of the persistence state by the reactivation of general metabolism, repair of protein and DNA, and restoration of macromolecular synthesis^14,28,30,31. Thus, metabolism is not an epiphenomenon of persisters but the central organizing principle driving their formation, their maintenance, and their reversal to an antibiotic-sensitive state.

Formation

Persisters arise either stochastically or in response to environmental cues, although in both instances the initiating events can be metabolic. Stochastic models reflect cell-to-cell variations in metabolic activity and energy dynamics. Normal intracellular fluctuations in metabolic states can lead to unstable cell growth with accumulation of ppGpp, activation of toxin–antitoxin systems, and changes in metabolic homeostasis that protect cells from stress and slow their growth^32–34.

Environmental and growth-dependent inputs also drive persister formation through metabolic signaling. Alterations in metabolic flux induced by rapid changes in the carbon source, nutrient shifts, or inhibition of transport can cause a rapid collapse in central carbon flow, elevation of ppGpp, activation of stress-responsive σ^S, and induction of multiple toxin–antitoxin modules^23,24. For example, in Escherichia coli, exhaustion of a primary carbon source and transition to a secondary source induce cAMP-CRP signaling and the stringent response, increasing ppGpp levels²³. This response reprograms transcription by reducing rRNA and ribosomal protein synthesis while activating amino acid and stress-response pathways³⁵. Importantly, the same regulatory framework supports different survival strategies depending on antibiotic context. In ofloxacin persisters, it contributes to reduced susceptibility through effects on DNA topology²³ whereas in ampicillin persisters, ppGpp-dependent signaling shifts cellular priorities away from cell wall biosynthesis¹⁹. These examples highlight how a common metabolic signaling network can be rewired in distinct ways, underscoring the heterogeneous and context-dependent nature of persister metabolism. Nutrient shifts and starvation also induce a σ^S-dominated proteome that is enriched for stress defense and catabolism and depleted in biosynthetic enzymes²⁴. Persisters in this state maintain a high adenylate energy charge by channeling limited carbon into oxidative phosphorylation rather than biomass synthesis²⁴.

Antibiotics may also act as a metabolic trigger. Fluoroquinolone-induced double-stranded DNA breaks activate RecA, promote LexA cleavage, and induce the SOS network, which upregulates toxins such as TisB and growth inhibitors such as SulA, thereby shifting cells into a tolerant state²⁵. Because these persisters accumulate significant cellular damage²⁸, repair of this damage during and after treatment requires an active remodeling of cellular metabolism.

Maintenance

Although persister cells are reported to have reduced ATP levels or decreased energy flux relative to exponentially growing cells^36–38, persisters must have sufficient energy to survive prolonged antibiotic exposure. Antibiotic treatment of persisters results in double-stranded DNA breaks, stalled replication forks, protein aggregation, and membrane defects^28,39–41, which must be repaired in persister cells. DNA repair pathways such as homologous recombination, nucleotide excision repair, and translesion synthesis require ATP, as do protein refolding, proteolysis, and membrane repair. Consistent with this requirement, antibiotic-induced cell-wall-deficient L-forms and filamentation have been observed in persisters^41,42, phenotypes that depend on energy-intensive processes, including membrane remodeling, lipid biosynthesis, osmotic adaptation, and oxidative stress defenses¹⁸. In addition to these repair and maintenance processes, a subset of persisters can maintain energy-dependent efflux activity that lowers the intracellular antibiotic concentration⁴³. Thus, persisters are not strictly metabolically dormant; their viability depends on certain ATP-dependent reactions and pathways that are inconsistent with metabolic arrest.

Characterization of mutants supports the energy requirement of persisters. Mutants defective in genes for key nodes of central metabolism, including the tricarboxylic acid (TCA) cycle and ubiquinone biosynthesis^26,44–46, reduce persister survival, indicating that TCA cycle flux, NADH regeneration, and the electron transport chain (ETC) likely function in maintaining viability during antibiotic challenge. Persister cells retain or can rapidly restore the PMF through catabolism of specific carbon sources such as glucose and glycerol^15,47. This PMF-driven metabolism explains aminoglycoside potentiation, in which metabolic stimulation enhances aminoglycoside uptake and bacterial killing, which can be further rewired by the global regulatory complex cAMP-CRP in E. coli persisters⁴⁸. Further, certain persister subtypes exhibit highly specialized increases in metabolic activity. For example, growth-arrested MazF-derived persisters take up glucose and consume oxygen to support an ATP-consuming RNA futile cycle⁴⁹.

Although persister survival likely depends on selective metabolic reprogramming rather than metabolic shutdown, persisters do not have to exhibit higher metabolic activity or energy levels than drug-sensitive exponential-phase cells²⁶. Persisters likely achieve a balanced metabolic state that supports cell repair and maintenance functions without activating antibiotic targets or producing excessive energy and damaging reactive oxygen species. This model is supported by metabolically hyperactive phoU mutants, characterized by increased expression of energy production genes, flagellar components, and chemotaxis pathways, which produce few persisters⁵⁰. Similarly, atp synthase mutants exhibit elevated respiration during exponential growth but reduced survival under antibiotic treatment⁵¹. Notably, mutations and perturbations of energy metabolism also reduce the preexisting non-growing subpopulation of cells enriched for persisters^26,44; thus, the changes in persistence may result from multiple factors beyond elevated metabolism.

Persistence reversal

The resumption of growth by persisters is likely the result of active processes that reverse the stress-adapted state via defined nutrient-sensing, metabolic, and repair pathways rather than the passive relaxation of metabolic dormancy^28,30. Specific environmental cues likely drive the reversal process via sensing by chemotaxis receptors and phosphotransferase systems³⁰. These pathways may trigger a metabolic transition that restores central carbon flux, respiration, and ribosome function³⁰. Although ribosomes may be thought to be inactive in persisters because of hibernation and ribosome modulation factors⁵², persister-enriched populations may exhibit low but widespread translation⁵³ and retain dynamic transcriptional activity⁵⁴, indicating that these molecular programs remain engaged in persisters, potentially supporting their recovery⁵⁵.

Reversal of persistence also requires energy-dependent clearance of damaged cell components. Single-cell analyses show that the burden of DNA damage and the efficiency of its removal strongly affect how quickly individual persisters resume growth³¹. The diverse regrowth trajectories, including smooth elongation, filamentation before division, and markedly delayed or erratic recovery, are inconsistent with a purely spontaneous recovery model³¹. Instead, they suggest physiological diversity established during antibiotic exposure, especially in the extent of DNA, protein, and membrane damage, and in the capacity of each cell to repair the damage.

The damage incurred by persister cells must be repaired or removed before balanced growth can resume. Key ATP-dependent proteostasis pathways, including chaperone-mediated disaggregation and protease-driven turnover of damaged proteins, must restore a functional proteome⁵⁶. Energy-intensive DNA repair systems are strongly activated during the recovery window in quinolone persisters⁵⁷, directly linking reversal kinetics with damage resolution. For persistence programs driven by toxins, expression or reactivation of cognate antitoxins and auxiliary factors is required to relieve translational arrest and rebuild the translational apparatus²². Detoxification and repair processes require substantial energy supplied via the same metabolic reactivation that restores PMF and ATP production; therefore, persistence reversal is an actively programmed, metabolism-driven process. However, this does not preclude stochastic reversal of persistence that might be shaped by damage-dependent physiological variation.

Metabolism as a drug target in persister cells

Eradicating bacteria that resist or tolerate antibiotics is a major clinical challenge because various complex factors contribute to bacterial survival. However, identifying and targeting conserved metabolic pathways that are essential across diverse bacterial states provides a promising target for novel antibacterial therapies. Central energy metabolism may provide such a target⁵⁸, as it is essential in antibiotic-sensitive and resistant cells. However, this approach, which is effective in growing bacteria, is not considered applicable to non-growing or slow-growing persisters and other antibiotic-tolerant cells, which are typically viewed as metabolically inactive⁵⁹. The assumption that tolerance stems from suppressed metabolism and growth arrest is based on the fact that antibiotics work best against dividing cells⁵⁹. However, as discussed in the previous sections, there is growing evidence of a baseline level of metabolic activity in persister cells, especially in pathways that support energy maintenance and stress responses.

In energy-depleted cells, aminoglycosides are largely ineffective because they depend on an active PMF for uptake and killing^47,60. Thus, the susceptibility to aminoglycosides by persisters after specific carbon sources are added, thereby restoring their PMF, is particularly informative for persister metabolism⁴⁷. This effect occurs in persisters formed by several species (e.g., E. coli, Staphylococcus aureus, and Pseudomonas aeruginosa) and antibiotics (e.g., β-lactams and fluoroquinolones) in both exponential and stationary-phase cultures under diverse conditions^15,47,61,62. Persisters originated from antibiotic exposure, especially following fluoroquinolones treatment, often show extended recovery periods before dividing again⁶³. However, the effect of the added carbon source on aminoglycoside potentiation is rapid¹⁵, and the observed potentiation is not associated with the growth resumption of persisters⁴⁷. Therefore, these cells either (i) retain energy production reserves or (ii) rapidly activate key metabolic processes required for PMF generation, which likely requires transcription and translation. Either scenario suggests that persisters are metabolically capable cells that can maintain or rapidly restart critical metabolic functions.

Because energy metabolism is a universal requirement for life, its irreversible disruption should be detrimental to persister cells. Whether a bacterial cell is sensitive, resistant, or tolerant to an antibiotic, it cannot survive without some level of energy flux. Even persisters from stationary-phase cultures, in which cells are non-growing and metabolically reduced, need energy to maintain cellular integrity, manage stress, repair damage, and resume growth when conditions allow^26,28. If we can irreversibly disrupt persister energy metabolism, such as ATP synthase, in its most resilient, drug-tolerant state, we may be able to kill persisters. In fact, energy metabolism in mycobacteria, particularly the oxidative phosphorylation pathway, is a promising target for drug discovery⁶⁴, given the fact that these bacteria can persist in a non-growing, asymptomatic state for extended periods.

A controversy in persister metabolism: genetic and environmental complexity shape persisters

Persisters have been classically defined by their reduced metabolism⁶⁵; hence, their metabolic activity remains controversial. Active persisters are reported to engage in PMF generation, ATP-dependent maintenance, and futile metabolic cycles, as well as transcription and translation and cell growth^{24,41,42,44,49,54}. In contrast, other persisters show suppressed metabolism and low ATP levels^36,37,66,67. However, conflicting results across studies likely reflect differences in experimental conditions, growth states, and analytical approaches (Supplementary Table 1), rather than true contradictions, and also underscore the remarkable diversity and adaptability of microbial survival strategies. This adaptability, which is a defining feature that reflects the depth of nature’s evolutionary design, shaped over billions of years, presents a profound scientific challenge. The dynamic landscape of persistence is likely influenced by a range of interconnected factors (Fig. 1):

• Bacteria have evolved a rich repertoire of overlapping and redundant mechanisms to ensure their survival under fluctuating and often hostile conditions. This redundancy allows various subpopulations to adapt differently, depending on their environment and physiological state. For instance, nutrient-limited bacteria can recycle cellular resources to survive⁴⁴, similar to autophagy. Some persister types activate the stringent response, toxin–antitoxin systems, or the SOS network to transiently suppress growth and protect against damage^{23,25,33,68–70}, whereas others downregulate anabolic activity yet maintain energy homeostasis²⁴ (described in the preceding sections). These few examples illustrate the flexible, overlapping, adaptive pathways used by bacteria across diverse and challenging environments to fine-tune their physiology to enter a persistent state.

• Mutations that arise from replication errors, oxidative stress, or other internal and environmental pressures^71,72 may influence persistence by altering stress responses, metabolism, or regulatory circuits linked to growth and metabolic dormancy. Bacteria mutate frequently⁷³, and even minor genetic changes, particularly in regulatory or metabolic genes, which do not affect the phenotype, such as the MIC, may affect persistence. Moreover, coexisting, heteroresistant subpopulations with elevated MICs⁷⁴ may be inadvertently quantified as persisters (Box 1). As persister cell genomes are not typically sequenced, subtle genetic changes in the population that affect persistence as well as heteroresistant lineages are rarely detected.

• Persistence that is associated with mutations can vary not only between species but even between closely related laboratory strains. For example, defined gene knockouts (e.g., cyaA and crp) in E. coli MG1655 vs the BW25113-derived Keio strain can produce markedly different numbers of persisters under the same conditions²⁶ (Supplementary Table 1). These differences likely reflect the genetic variations between the strains, as sequencing of BW25113 revealed multiple genomic alterations that may contribute to phenotypic diversity⁷⁵. Additional mutations that may accumulate during strain maintenance, commercial growth and distribution, or routine laboratory handling may influence persistence. Further, unintended phage infections during strain maintenance or propagation that alter persistence phenotypes⁷⁶ illustrate how readily bacterial survival traits can change in laboratory settings. Similarly, clinical isolates of strains of the same species often respond very differently to antibiotic treatment⁷⁷. Thus, the genetic identity of a strain, shaped by its unique evolutionary history, plays a critical role in determining its persistence behavior.

• In addition to genetic variability, there is the stochastic nature of gene expression, such that random fluctuations in regulatory networks lead to heterogeneous cell-specific activation of persistence pathways^33,34,37,78. This inherent phenotypic diversity within a population complicates the interpretation and reproducibility of conventional experimental methods. The unpredictable fluctuations in the microenvironment that individual bacteria experience⁷⁹ can be a major source of this variability. Despite researchers’ best efforts to provide a uniform medium, tightly regulated temperature, and optimized oxygen levels, cells experience undetectable, localized differences in their microenvironment. These changes can result in low-frequency phenotypic variants, such as persisters, that arise stochastically. Hence, single-cell heterogeneity is not experimental noise; it is an inherent and important characteristic of microbial systems.

• Many factors affect persister formation, including the growth phase of the culture, the growth medium, nutrient levels, and the antibiotic type, concentration, and duration of exposure (Supplementary Table 1). As different antibiotics target distinct cellular processes, the antibiotic class affects persister formation. β-Lactams primarily kill growing cells, thereby potentially selecting for cells with little cell wall biosynthesis⁸⁰. Fluoroquinolones may select for cells capable of managing DNA damage and replication stress⁸¹. Aminoglycosides may select for cells with reduced translation capacity and diminished PMF⁸². Thus, persister levels differ across these drug classes as each class eliminates a different subset of cells defined by their growth state, stress-response activity, and metabolic capacity. Further, the high degree of heterogeneity within persister subpopulations is at odds with the expected uniform pattern of survival of metabolically dormant persisters. For example, we demonstrated that prolonged aminoglycoside exposure of stationary-phase cultures diluted into fresh medium eradicated all the cells, whereas persisters to ampicillin and fluoroquinolones remained detectable under the same conditions, although at different levels²⁶. These observations suggest that β-lactam or fluoroquinolone persisters retain sufficient translation and energy metabolism for aminoglycoside uptake and killing, consistent with the measurable levels of transcription⁵⁴ and translation⁵³ found in persisters. Therefore, persistence is not a singular or uniform phenotype; it comprises a variety of context-dependent survival states shaped not only by the antibiotic used but also by broader environmental conditions and strategies that we further discuss below.

Fig. 1. Genetic and environmental complexity shape persister cells.

Each category highlights representative, and sometimes overlapping, factors that illustrate the critical topics requiring further study.

Culture age can confound metabolic dormancy

One such critical experimental condition is culture age, which affects persister formation and physiology via changes in cellular metabolic states, the proportion of non-growing cells, and the overall heterogeneity of the population (Supplementary Table 1). Kim Lewis’ group, which showed that deleting icd, a key TCA cycle gene, led to a modest increase in persistence to quinolone antibiotics during exponential growth, proposed that persisters arise stochastically as a low-energy subpopulation³⁶. The increased tolerance of icd mutants was also reported by other groups, including our own^26,83–85. As loss of icd also resulted in a small reduction in intracellular ATP concentration, the Lewis group posited an inverse correlation between ATP and persister levels³⁶. To further characterize this phenomenon, the Lewis group then used 48 h stationary-phase cultures treated with β-lactams and found that persisters had lower levels of ATP than the non-persisters that resumed growth³⁶. As non-growing cells (persisters, VBNCs, and others) typically form in aged cell cultures^45,86, aging of the culture may impact persister metabolism. For instance, Fan Bai’s group reported that persisters isolated from 24 h stationary-phase cultures have slightly reduced ATP levels, which overlap significantly with the levels in antibiotic-sensitive cells⁸⁷. Furthermore, Goormaghtigh and Van Melderen showed that growth rate, DNA content, and SOS induction vary among ofloxacin persisters, just as they do in the bulk population; therefore, these metabolic processes do not predict persistence⁴². These cells, obtained from 16 h overnight cultures, were still metabolically active at the time of antibiotic exposure. Further, a recent single-cell analysis showed that before antibiotic treatment in the exponential phase, persister cells display highly heterogeneous cell proliferation activities⁴¹. Notably, the persisters growing before the treatment exhibited diverse survival behaviors, including continued growth with L-form-like morphologies, transient growth arrest, and post-exposure filamentation⁴¹, indicating the heterogeneity in their metabolic activities. Similarly, the cultures used in this study were obtained after 15- or 16 h overnight growth. Although these last two studies did not measure ATP levels directly, cellular energetics are tightly linked to growth, transcription, and translation, which are among the most energy-intensive processes in the cells.

A careful analysis of the studies mentioned above reveals that metabolic activity at the point of antibiotic challenge does not preclude persistence, and that persister cells, once formed, may transition gradually into lower-energy states as cultures age. Thus, culture aging plays a key role in persistence⁸⁸ (Supplementary Table 1), and culture age should be carefully considered in any analysis of persister characteristics for the following reasons: 1) Many persisters in an aged culture likely formed long before the time of sampling, making it difficult to identify the key events that triggered their persistence. To determine whether persisters arise randomly or in response to specific signals, they must be evaluated as they transition to persistence, potentially early in the culture phase. 2) As persisters spend a prolonged time in a nutrient-deficient, stationary-phase environment, they will likely deplete internal energy stores and slow their metabolism. Thus, persisters from aged cultures would be expected to show lower metabolic activity than exponentially growing cells.

We performed several simple experiments to show how culture aging and growth status contribute to metabolic heterogeneity among persisters (Supplementary Methods). These experiments, based on fluorescent protein dilution and redox measurements, are not novel and have been used extensively in the literature. Although redox staining does not directly quantify ATP generation or energy-intensive processes such as transcription and translation, it is tightly coupled to cellular energy metabolism and therefore provides indirect insight into these activities (Supplementary Figs. 1 and 2). Our purpose here is simply to use them to reconcile the existing data discussed above for illustrative purposes. Hence, we grew cells carrying an inducible fluorescent protein (mCherry) expression system for up to 48 hours in the presence of the inducer. We collected samples at various time points, 16 hours (unaged), 24 h (moderately aged), and 48 h (aged), and transferred them into fresh medium without the inducer to monitor dividing (mCherry^low) and non-growing (mCherry^high) cells. Initially, all cells were red due to mCherry expression upon initial transfer to fresh medium (Fig. 2). As growing cells divided, their fluorescence intensity decreased due to protein dilution during cell division. In contrast, non-growing cells, enriched for persisters and VBNCs, retained high fluorescence, indicating a lack of division and protein dilution. In the exponential growth phase, bacterial populations from 24- or 48 h-old cultures contained a large fraction of non-growing cells (mCherry^high) (Fig. 2). In contrast, younger, 16 h cultures comprised dividing cells (mCherry^low) with few or no non-growing cells (Fig. 2).

Fig. 2. Effect of culture age on cell division status revealed by mCherry protein dilution.

E. coli carrying an inducible mCherry expression system was grown for 16, 24, or 48 h (h) in the presence of inducer and then transferred to fresh medium without inducer. Flow cytometry histograms show mCherry fluorescence at 0, 1, and 2 h after transfer. At 0 h, all cells displayed high fluorescence. As cells grew and divided, their fluorescence decreased due to dilution of mCherry (mCherry^low; dashed lines), whereas non-growing cells retained high fluorescence (mCherry^high; solid lines). Number of biological replicates, n = 4. All raw data, including biological replicates, are available in the data repository Figshare¹⁰⁰.

Using a redox sensor green (RSG) dye, we also measured and compared the metabolic activities of dividing and non-growing cells by flow cytometry (Fig. 3a, b). In aged cultures (48-hour), non-growing cells generally exhibited reduced metabolic activity relative to dividing cells, although a fraction overlapped substantially with the dividing population (Fig. 3a, b). This pattern mirrors the ATP profiles of antibiotic-sensitive and tolerant cells reported by the Kim Lewis group in aged cultures³⁶.

Fig. 3. Metabolic activity of dividing and non-growing cells.

a E. coli cultures were grown for 16, 24, or 48 h, transferred to fresh medium, and analyzed after 2 h using RSG staining. These samples correspond to the 2-h time point shown in Fig. 2. Dot plots show RSG versus mCherry fluorescence, separating dividing and non-growing cells. b Normalized histograms highlight reduced metabolic activity in most non-growing cells, with partial overlap with dividing cells. Note that flow cytometry diagrams are representative of four independent biological replicates, which yielded reproducible results (data shown in Supplementary Fig. 3). c E. coli cultures grown for 16, 24, or 48 h were transferred to fresh medium and assayed for antibiotic tolerance after 2 h. Persister levels were determined based on CFU counts. Note that CFU-based assays detect culturable persisters but not VBNC cells; therefore, the 48 h population does not necessarily exhibit higher survival, as the increased non-growing fraction is largely due to VBNC cells (Supplementary Fig. 4, and Supplementary Table 2). Data are shown as mean ± standard deviation. Number of biological replicates, n = 4. All raw data, including biological replicates, are available in the data repository Figshare¹⁰⁰.

In the moderately aged (24 h) cultures, non-growing cells generally showed reduced metabolism; however, a distinct, reproducible subpopulation overlapped substantially with growing cells (Fig. 3b, and Supplementary Fig. 3). This raises the key question of whether this overlapping subpopulation is enriched for persisters. The study by Bai and colleagues, using similar culture conditions, may help address this question; they demonstrated that among VBNC, persister, and sensitive cells, the persisters exhibited ATP levels that overlapped substantially with those of sensitive cells, whereas VBNC cells showed much reduced ATP⁸⁷.

Persister levels in 24 h and 48 h aged cultures were similar despite the increased number of non-growing cells in aged cultures (Fig. 3c), since a larger fraction of these non-growing cells transition into VBNC cells, consistent with our previous findings⁸⁶ (Supplementary Fig. 4, and Supplementary Table 2). Because persisters constitute only a small subset of the non-growing population (Supplementary Table 2), and because some non-growing cells may resuscitate and be killed by antibiotics, non-growing status alone cannot be used as a proxy for persistence. Although we did not track the fate of individual non-growing cells here, persisters in 24 h and 48 h cultures may also occupy similar regions of the RSG distribution, which could explain their comparable survival (Fig. 3c).

But the story became more interesting when we analyzed 16 h cultures that lacked non-growing cells; although they contained a measurable number of persisters (about 100-fold fewer than in aged cultures) (Fig. 3c), these persisters must have arisen from dividing cells (Fig. 3a,b). This is consistent with our previous finding that the non-growing cell subpopulation contains many more persisters, often 100-fold higher than the growing subpopulation⁸⁹. In that study, fluorescence-activated cell sorting was used to separate dividing cells based on redox activity into high- and low-activity gates, and no significant difference in persister levels was observed between the two. This suggests that, in growing populations, metabolic activity or cell proliferation alone does not predict persistence, as was also evident in the single-cell studies of Goormaghtigh and Van Melderen⁴² and Umetani and coworkers⁴¹.

Culture aging and growth status create substantial heterogeneity among persisters, and it seems likely that persisters do not arise exclusively from preexisting dormant or low-energy state cells. There may be multiple routes by which persister cells arise in younger, growing cultures, which could include responses to 1) stress associated with high metabolic activity and proliferation, potentially through accumulated intracellular damage and other signaling pathways^90,91; 2) carbon depletion and elevated expression of persistence-related pathways in late exponential growth^19,23,24; and 3) direct induction of persister cells by antibiotics, or rapid physiological shifts upon antibiotic exposure (such as SOS activation or membrane stress)^25,41, triggering growth arrest during treatment and allowing survival without prior metabolic dormancy.

Experimental constraints may mask metabolic rewiring in persister cells

Beyond experimental conditions, the methodological strategies commonly used to study persisters may affect our interpretation of their metabolic mechanisms, as these approaches rely on indirect observations or restrictive perturbations that generate physiological states that may not reflect naturally occurring persisters. Three methodological constraints are particularly influential: (i) the widespread use of metabolic inhibitors, which artificially impose metabolic dormancy; (ii) the reliance on a limited number of gene deletions, which cannot capture the broader architecture of metabolic rewiring; and (iii) the comparison of persisters to exponentially growing cells, whose extremely high metabolic activity can obscure more subtle yet biologically meaningful metabolic programs in persisters:

• Inhibitors of ATP synthesis, protein synthesis, transcription, translation, or the ETC induce metabolic dormancy^38,92,93, which also increases antibiotic tolerance at the population level (Box 1). Such studies reveal links between population-level antibiotic tolerance and major cellular functions; however, metabolic dormancy is one feature of drug-tolerant cells, and the chemically induced changes may not represent the physiology of naturally occurring persisters (Box 3).

• Genetic studies in persister research often rely on a relatively small number of gene deletions that may lead to an overgeneralization of the findings as evidence of a universal persistence mechanism. However, the results can be misleading, as gene knockouts often have pleiotropic effects. Changes in survival seen in a mutant may result from indirect or secondary effects rather than a direct role of the gene in persistence. For instance, compared to wild-type cells, strains with a deletion of ATP synthase genes can impair energy production and cell growth (Box 3), but in the lag or early exponential phases, the mutants have fewer persisters than the wild-type strain^26,94. Strains with deletions of a few TCA cycle or ETC genes (icd and nuo, respectively) show increased persisters^26,36; however, many related mutants (e.g., knockout strains of sdh, suc, mdh, or gltA genes) show reduced persisters in E. coli^26,44,45,94. Our genomic data demonstrated that most E. coli mutants lacking components of the TCA cycle, ETC, or ATP synthase exhibit reduced tolerance to ampicillin and ofloxacin²⁶, confirming that perturbing metabolic activity does not consistently lead to increased tolerance^26,44,94. However, differences in organisms, growth phase, or the antibiotic may also affect persister frequency. For instance, TCA cycle mutants can be highly sensitive to antibiotics during the lag phase in E. coli²⁶, but may exhibit increased tolerance in the late exponential phase in S. aureus³⁷. To gain a better understanding of the effects of metabolic mutants on persister formation, a systems-level approach using high-throughput analysis of a large collection of metabolic mutants may provide more reliable and broadly applicable insights.

• A key factor in characterizing persister metabolism is choosing an appropriate reference population. A limitation of studies of persisters has been the comparison of persister cells to exponential-phase cells, which may not reflect the physiological conditions of cells that give rise to persisters. Deviation from the population average is often the most informative comparison; however, using exponentially growing cells as the external reference can bias interpretation because their very high metabolic activity may obscure physiologically meaningful metabolic features of persisters. For instance, we found that persisters that formed in the stationary phase retained active energy metabolism, which was important for their survival⁴⁴. However, because the metabolic activity of persister cells is much lower than that of fast-growing cells, this difference is not obvious in exponential-phase cultures⁸⁹. Exponentially growing cells are in a high-energy state in an environment that differs substantially from those where persisters typically arise. Heinemann’s group has made a good point by normalizing energy generation to substrate consumption²⁴, thereby enabling more accurate comparison and interpretation of metabolic activity across different cells. However, physiological relevance is key. In many natural and clinical settings, such as biofilms, mucus layers, necrotic tissue, or the surfaces of medical devices, bacteria are likely to be in stationary phase or living under nutrient-limited, low-oxygen conditions^95–97. As persisters are most likely to form in these environments, it is not clear how much value is provided by comparisons with exponential-phase cells. Consideration of the context of the physiological conditions that support persistence may provide a more accurate characterization of the metabolic states that underlie persister formation.

Both newly formed persisters and those that preexist from aged cultures should have measurable metabolic activity, which is necessary for their survival. The studies of Lewis and Bai still show overlap in the metabolic profiles of persisters and antibiotic-sensitive cells that eventually resume growth^37,87. As Bai’s group found that persisters have higher metabolic activity than VBNC cells⁸⁷, it seems likely that persister cell metabolic activity lies between the metabolism of growing and unculturable cells, and may shift with changes in the environment or treatment. Describing persisters as “metabolically dormant” oversimplifies their metabolic state and fails to capture our growing understanding that persistence involves dynamic metabolic rewiring, not simply a full shutdown of activity.

Metabolic rewiring in eukaryotic persister cells, particularly cancer persisters, is well established (Box 4). However, the possibility that bacterial persisters also use a metabolic shift has not received much attention. Increasing evidence suggests that persisters with either low or high overall activity are metabolically altered or rewired to enhance survival under stress^{24,41,42,44,49,54}. This perspective of “metabolic rewiring rather than metabolic dormancy” provides a unifying framework for interpreting conflicting results.

Box 3: Cell growth vs persistence.

The relationship between bacterial growth and antibiotic persistence has been explored in many systems, demonstrating multiple aspects of this connection. Pontes and Groisman reported a high tolerance to antibiotics in Salmonella cultures whose growth is limited by environmental factors such as magnesium depletion or nutrient starvation or by chemical inhibition¹¹². Similarly, culture treatment with metabolic inhibitors induces metabolic dormancy, which is often accompanied by increased antibiotic tolerance at the population level. In contrast, we demonstrated that strains with knockouts in the TCA cycle, the ETC, or ATP synthase, which produce varying degrees of growth deficiency and drastic changes in persister levels, do not show a single pattern of tolerance^26,44,94. Therefore, it is not clear that reduced growth rate consistently correlates with higher persistence in these mutant strains. We recently found that E. coli cultures grown in environments with different osmolyte concentrations or extracellular pH produced fewer persisters when growth was restricted and more persisters in faster-growing cultures¹¹³. This finding is interesting because some of these environments (such as lower pH) are known to enhance tolerance or persistence¹¹⁴. However, in our study, cells were exposed to various environments during overnight growth and in the main culture, where persisters were quantified. This continuous exposure likely allowed cells to adapt physiologically. In contrast, sudden exposure to inhibitory conditions, which is common in persister experiments, often results in immediate growth arrest and increased drug tolerance. In natural environments or host cells, bacteria may experience osmotic stress and an acidic pH long before antibiotic treatment begins, so the adapted state may be more relevant in vivo. Nonetheless, further studies are required to better understand the correlation between slow growth and persistence.

Box 4: Parallels between persisters and cancer metabolic rewiring.

Cancer cells may provide insights into the rewiring of cellular metabolism to adapt to environmental stress. Nearly a century ago, Otto Warburg observed that cancer cells favor glycolysis for energy production, even when oxygen is available, a metabolic feature later known as “aerobic glycolysis^115,116.” This shift allows tumor cells to funnel glycolytic intermediates into biosynthetic pathways to support rapid growth and cell division. Interestingly, non-dividing, drug-tolerant, or persister-like cancer cells rely more heavily on oxidative phosphorylation than drug-sensitive cells^117,118. This alteration supports survival rather than growth, reflecting different priorities for cells under stress. This is likely an adaptive strategy, as oxidative phosphorylation is far more efficient than glycolysis, generating roughly 15–16 times more ATP per molecule of substrate¹¹⁹. That efficiency allows cells to produce sufficient energy with limited carbon resources, an obvious advantage in the face of nutrient scarcity, therapeutic pressure, or other forms of stress.

Beyond the biphasic model: defining persistence through physiological diversity

For decades, persistence has been identified primarily through the biphasic killing model, which assumes that metabolically active, dividing cells are sensitive to antibiotics, and persisters are metabolically dormant survivors that generate the slow second phase of killing. With a better understanding of the metabolic diversity within bacterial populations, we know that the biphasic model does not capture the range of physiological states that influence persister behavior. After antibiotic treatment, actively growing, exponential-phase cultures may show a sharp initial decline in viability followed by a slower decline, producing the classic biphasic pattern. In contrast, for lag-phase cells, which resume growth or exit lag at different times⁹⁸, susceptibility can be asynchronous or multiphasic, masking the separation between sensitive and persister cells. Similarly, the slow killing and staggered recovery of deeply dormant cells in aged cultures can mask biphasic behavior, whereas newly formed persisters in younger cultures may retain measurable activity and partial susceptibility, reducing the sharpness of the plateau phase. Persisters are not uniform; some persisters preexist as non-growing cells in aged cultures, whereas others arise during active metabolism or in response to antibiotic stress or cellular damage, with slow or continuing growth^{11,23–25,36,45,89,99}. This heterogeneity further depends on growth phase, strain background, culture conditions, and the specific survival pathways involved, as discussed above. These variations in physiological states in a population may obscure the canonical biphasic pattern, limiting the usefulness of a single killing curve to define persister cells.

The biphasic killing model, with its assumption that persisters are metabolically dormant, fails to capture the range of physiological states and the biological diversity of persister cells. Current evidence suggests that persistence is better viewed as a reversible, non-heritable survival mechanism involving altered or rewired metabolism. Relaxing strict criteria for persisters may reconcile conflicting findings across studies and better reflect the physiological diversity that defines bacterial persistence. Further, other experimental approaches may be necessary to accurately quantify and characterize the diversity of persister populations (Box 5).

Box 5: Complementary and alternative approaches to study persisters.

In the standard assay of persister survival after antibiotic treatment, samples are 1) collected at various time points from a culture in the desired growth phase that has been treated with antibiotics at high concentrations, 2) washed to remove the antibiotic, and 3) plated to quantify CFUs of surviving cells. A killing curve is determined from a plot of CFUs vs. time. In addition to biphasic killing curves, the reversible nature of persistence can be confirmed by washing, regrowing, and re-treating antibiotic-exposed cells to determine whether sensitivity is restored and the persister fraction remains constant. Measuring the MIC of the treated population determines whether survivors acquired heritable resistance. Further, this treatment–regrowth–retreatment procedure provides a useful complement to killing curves, particularly when killing is not biphasic.

However, CFU measurements do not provide information on which cells survive or why they survive. This type of information is provided by more advanced approaches, such as time-lapse microscopy of individual cells expressing reporter proteins associated with diverse phenotypic variants¹²⁰. These time-consuming, labor-intensive methods are not always feasible for routine use, and we still lack the perfect, persister-specific biomarker. Fluorescent reporters for candidate persister mechanisms can track single cells before, during, and after antibiotic treatment, but images of large numbers of cells are required for statistically significant conclusions. As persisters occur at a frequency of about 10⁻⁶ for cells in exponential-phase obtained from unaged cultures (Supplementary Table 2), tens of millions of cells must be analyzed to measure persisters. Although aged cultures can be used alternatively as they have more persisters, they also have a large fraction of VBNC cells (Supplementary Table 2), complicating interpretation. Further, persisters in these cultures may have formed long before antibiotic treatment, limiting the ability of reporters to reveal the dynamics of persister formation. Single-cell RNA sequencing has been used widely in eukaryotic systems¹²¹. While, more recently, it has been tested on bacterial persisters¹²², its application to bacteria is still limited by technical challenges, including small cell size, low mRNA abundance, rapid RNA turnover, and the lack of polyadenylated transcripts. Nevertheless, single-cell RNA sequencing of a cell population at multiple time points before, during, and after antibiotic treatment can reveal distinct transcriptional subpopulations. Each subpopulation should, in principle, exhibit unique molecular signatures that could serve as biomarkers. Fluorescent reporters for these biomarkers could be used in single-cell microscopy experiments to track the behavior of each subpopulation to determine which correspond to persisters, VBNC cells, or antibiotic-sensitive cells.

Conclusion

The classic view of bacterial persistence as a dormant, metabolically inactive state is challenged by the evidence of substantial physiological and metabolic heterogeneity among persister cells. Persisters represent a diverse spectrum of survival strategies that are influenced by genetic redundancy, environmental variability, stochastic gene expression, and metabolic flexibility rather than a single, uniform phenotype. Metabolic dormancy is one facet of persistence. However, although metabolic dormancy is neither universal nor sufficient to explain all tolerant states, it may discourage scientists from exploring persister metabolism. In contrast, characterizing persisters as metabolically rewired rather than dormant provides possible targets for effective therapeutic strategies against chronic and recurrent infections.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.A.O. conceived, designed, and wrote the manuscript. H.N. and S.G.M. performed the experiments, prepared the figures, and curated the references. All authors read and approved the final manuscript.

Peer review

Peer review information

Nature Communications thanks Tatsuya Akiyama and the other, anonymous, reviewers for their contribution to the peer review of this work.

Data availability

All data generated or analyzed during this study, including raw numerical data and flow cytometry datasets, are available in Figshare at 10.6084/m9.figshare.31416419¹⁰⁰.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71427-7.