Introduction

Antibiotics combat bacterial infections in humans and animals by either killing the bacteria or impeding their ability to thrive and reproduce. Bacterial cells employ various defense mechanisms to survive antibiotic treatment, primarily through resistance and tolerance. Some bacteria are naturally resistant, but others develop resistance by acquiring resistant genes1,2,3,4,5 or accumulating mutations in their genome, which enable them to counteract the antimicrobial drugs3,4,5. According to the Centers for Disease Control and Prevention (CDC), more than 2.8 million antibiotic-resistant infections occur annually in the US, claiming over 35,000 lives6. These numbers are likely underestimating the true scale of the problem, as not all infections are tested for resistance. Recent studies show that antibiotic-resistant bacteria cause more fatalities worldwide compared to other infectious diseases like AIDS, with global death projections to be at 10 million per year by 20506,7,8. The growing threat of antibiotic resistance cannot be overstated. For context, cancer, the second leading cause of death worldwide, claims around 9.5 million lives each year9,10. Without new tools to combat resistant bacteria, in just 25 years, antibiotic-resistant infections could claim a comparable number of lives as cancer does now. To prevent or minimize this looming crisis, it is crucial to deepen our understanding of the mechanisms that allow bacteria to survive antibiotic treatments, so we can develop more effective strategies to tackle this escalating threat.

In this study, we focus specifically on antibiotic persisters because they are a major driver of antibiotic resistance development and a known cause of recurrent bacterial infections11,12,13. When exposed to lethal antibiotic doses, bacterial populations display a biphasic death curve. In Phase 1, non-persister cells (susceptible cells and short-term tolerant cells) are rapidly killed off. In contrast, Phase 2 contains persister cells that take an exceptionally long time to be killed compared to their susceptible counterparts14,15 (Fig. 1A and Supplementary Fig. S2D). For example, treating Mycobacterium tuberculosis (the bacterium responsible for tuberculosis) can take up to a year, while Mycobacterium avium (which causes non-contagious lung infections) may require treatment lasting as long as two years. Despite these extended treatment durations to kill persisters, relapse rates remain notably high16,17.

Fig. 1: Phenotypic heterogeneity in bacterial populations leads to the formation of antibiotic persisters.
Fig. 1: Phenotypic heterogeneity in bacterial populations leads to the formation of antibiotic persisters.The alternative text for this image may have been generated using AI.
Full size image

A An isogenic bacterial population is heterogeneous, comprising diverse cell types, including antibiotic-susceptible (S), short-term tolerant (ST), primed (Pr), and persister (P) cells, with occasional antibiotic-resistant, or resister (R) cells. While S, ST, Pr, and P cells are genetically identical, R cells have acquired genetic mutations. When exposed to lethal doses of antibiotics, the population exhibits a typical biphasic death curve. In Phase 1, susceptible cells die off quickly, followed by short-term tolerant cells. In Phase 2, persister cells, both pre-existing and those derived from primed cells, survive significantly longer than other cell types. Resistant cells, if present, are not affected by antibiotics. They survive and divide even under lethal doses of antibiotics. B Interconversion of different cell states. i Primed cells divide actively and are pre-adapted for stress, even before encountering it. Their progenies can comprise primed cells, persister cells, or susceptible cells, but this may depend on the environment. The full extent of possible state transitions is not yet fully understood, but here we illustrate a few possibilities. ii Due to phenotypic heterogeneity, bacterial cells exist in interchangeable states. Our data shows that primed cells can reproduce more primed cells via epigenetic inheritance and can also transition into persisters or susceptible cells. We have shown that susceptible cells are converted to primed cells, but it remains unclear whether persisters transition directly into susceptible cells or via a primed cell intermediate. Similarly, it is uncertain if persisters can revert directly to primed cells. C Persisters (P) is a driving force of antibiotic resistance. Upon stress, only the persister (P) and resister (R) cells survive. With prolonged stress, the persisters eventually die out, but the resisters survive. Once the stress is removed, persisters revert to a normal state and resume dividing, and their offspring are more likely to accumulate mutations, potentially leading to the emergence of resistant cells.

Importantly, persisters and non-persisters are genetically identical. Although persisters are not inherently resistant to antibiotics, incomplete treatments—whether due to suboptimal dosing or insufficient duration—allow persisters to survive. Once antibiotic stress is removed, these surviving cells, often give rise to progenies with increased mutation rates, heightening the likelihood of developing resistance or acquiring resistance genes from their environment18,19,20.

For many years, it was widely assumed that clonal bacterial cultures would exhibit identical behaviors. However, there is a growing realization of the importance of heterogeneity within isogenic bacterial populations. Even in pure cultures during the log (exponential) phase of growth, bacteria display inherent variability (Fig. 1A), a phenomenon observed in both well-controlled laboratory settings and natural environments21,22. This variability manifests in gene expression differences—at either the transcriptional or translational level—resulting in phenotypic heterogeneity23. When this “noise” in gene expression is amplified through positive feedback mechanisms, then a graded response can shift to a binary response, establishing dynamic bistability where individual cells can exist in two different interchangeable states24,25,26 (Fig. 1B). Phenotypic variability has been characterized as both a chance occurrence stemming from stochastic fluctuations in gene expression and an evolvable trait, in different contexts24,27,28.

While heterogeneity may reduce fitness in an optimal, steady-state environment, it enhances survival during environmental fluctuations, a strategy known as bet-hedging29,30,31. Bacterial persistence is a classic example of a bet-hedging strategy, where a small fraction of cells in a growing culture, under normal conditions, either enter a state of arrested division with distinct gene expression profiles32,33 or are primed to transition into a persister state34. When exposed to antibiotics, these physiological adaptations enable persisters to outlive susceptible cells. Once the stressor is removed, persisters can revert to a dividing state (non-persister state), generating a population that includes both antibiotic-susceptible cells and persisters35,36 (Fig. 1B, C). This dynamic contributes to the re-establishment of bacterial infections after treatment, leading to the need for prolonged antibiotic use.

Although persistence was first identified 80 years ago in 194437, the mechanisms underlying persister formation remain a topic of ongoing debate. We recently demonstrated the existence of “primed cells” that are pre-adapted to become persisters before the stressor hits, using the model Gram-negative bacterium E. coli34. We sought to investigate whether primed cells are exclusive to E. coli and specifically related to antibiotic stress. To explore this question, we examined primed cell formation in Priestia megaterium (formerly known as Bacillus megaterium), a bacterium that is very distantly related to E. coli. While E. coli is Gram-negative, P. megaterium is Gram-positive, providing an intriguing contrast for our study. Understanding this survival strategy sheds light on bacterial stress response mechanisms, which is crucial for developing novel drugs that target persisters and for the effective treatment of chronic recurrent infections16,38,39,40,41,42,43,44,45.

In this study, we demonstrated that Gram-positive bacteria produce a specialized subpopulation known as primed cells, which leads to increased antibiotic persisters. Notably, these primed cells exhibit a transient memory, with their prevalence increasing during the later bacterial growth phases.

Results and discussion

The Luria–Delbrück Fluctuation Test (FT) can be used to study the dynamics of antibiotic persisters

Advancements in single-cell technologies have unveiled notable variations in phenotype and expression patterns among individual cells within isogenic populations46,47,48,49,50,51,52. Recently, these methods have been employed to investigate cancer persistence53,54,55,56; though distinct from bacterial persistence, both phenomena involve a small fraction of a population capable of surviving lethal stress. However, traditional single-cell sequencing methods provide only a static picture of different cell states57.

To address this limitation, we leveraged a modified Luria–Delbrück Fluctuation Test (FT) to examine the dynamics of intercellular transitioning from one cell state to another over multiple generations on a temporal scale57. The classical FT devised by Luria and Delbrück demonstrated that genetic mutations associated with survival arise randomly rather than as a direct response to selection pressure58. Back at that time, there was a debate between two opposing theories: one stated that resistance mutations occur randomly before exposure to selection pressure (Darwinian theory), and the other stated that mutations are induced by selection pressure (Lamarckian theory). In their experiments, single E. coli cells were grown into clonal populations and subsequently infected by T1 phage (viruses that infect and kill E. coli cells). If each bacterial cell has a small, independent chance of acquiring phage-induced mutation, the distribution of resistant bacteria across the clonal population should adhere to a Poisson distribution. Conversely, if mutations arise randomly, the number of mutant cells would significantly fluctuate among clonal populations, depending on when the mutation emerged during colony amplification. Their experiments revealed a skewed non-Poissonian distribution in the number of mutant cells, confirming the validity of the Darwinian theory of evolution (Fig. 2A). This approach demonstrated the potential of fluctuation-based analyses to unveil latent random processes. Moreover, clone-to-clone fluctuations can also be utilized to calculate the rate of switching between two distinct cellular states59.

Fig. 2: Modified Luria–Delbrück Fluctuation Test (FT) can reveal phenotypic variations among individual bacterial cells.
Fig. 2: Modified Luria–Delbrück Fluctuation Test (FT) can reveal phenotypic variations among individual bacterial cells.The alternative text for this image may have been generated using AI.
Full size image

A The classical Luria–Delbrück Fluctuation Test (FT) showed mutations generate spontaneously. In the classical FT, each clone originates from a single E. coli cell and is subsequently infected by T1 phage. If resistance mutations are virus-induced, the distribution of resistant cells across the clones would follow a Poisson distribution. Conversely, if mutations arise spontaneously before viral exposure, there will be considerable clone-to-clone fluctuations in the number of resistant cells. The experimental results confirmed that the theory of spontaneous mutation is indeed correct in this context. B Modified FT showed that some bacterial cells are primed for stress. FT can also assess whether some bacterial cells are primed to combat stress. This modified approach retains a similar experimental setup as (A) but measures the variation in persister cells across the clones in response to antibiotic stress. If persisters are antibiotic-induced, their numbers would follow a Poisson distribution across clones. Alternatively, if primed cells exist that enable bacteria to prepare for stress before antibiotic exposure, considerable clone-to-clone fluctuations in the number of persisters would be evident among clones.

While Luria and Delbrück focused on irreversible genetic changes leading to phage resistance, we adapted a modified FT to investigate the transitory presence of “primed cells”—a state resulting from reversible, non-genetic processes—within a bacterial population prior to antibiotic treatment (Fig. 2B). Primed cells are a subpopulation of bacteria priorly prepared to combat stress, ultimately giving rise to persister cells34.

P. megaterium cultures contain “primed cells” that are pre-adapted to become persisters prior to lethal antibiotic stress

In our previous research, we demonstrated that primed cells enable E. coli to prepare for lethal antibiotic stress because primed cells are more likely to become persisters34. Building on this, we employed a similar approach to investigate whether P. megaterium also produces primed cells. Our methodology involved comparing the persister populations generated from the Noise Control (NC) with those derived from a modified Fluctuation Test (FT) (Fig. 3A, B). A P. megaterium culture was grown to mid-log phase (OD600 of about 0.4–0.5) and subsequently distributed into 48 wells of a 96-well optical bottom microplate. The bacterial cultures in these wells were treated with ciprofloxacin (Cip) for 3 h to kill non-persister cells. Cip eliminates non-persister cells within 2 h; however, we extended the duration to 3 h to further reduce any chance of non-persister survival. The remaining persister numbers were determined using the viable plate count method, resulting in distinct persister percentages across each of the 48 wells. This population-level persister percentage variation assay (or NC) yielded a ~2-fold difference. This fold difference is calculated by dividing the maximum persister percentage by the minimum across the 48 wells.

Fig. 3: Gram-positive bacteria are primed for surviving lethal Cip levels.
Fig. 3: Gram-positive bacteria are primed for surviving lethal Cip levels.The alternative text for this image may have been generated using AI.
Full size image

A Schematic of population-level persister variation or Noise Control (NC) methodology, which measures the variation in persister percentages across an entire bacterial population. B Schematic of single cell-level persister variation or Fluctuation Test (FT) methodology, which tracks persister percentage variation at the single-cell level in clonal populations. C Some P. megaterium cells are primed for survival before Cip treatment. NC shows ~2-fold variation, FT shows ~60-fold variation. Fold difference = Maximum persister percentage/Minimum persister percentage out of the 48 clones in each experiment. Mean values, standard deviations, and CVs were calculated using bootstrapping with a 95% confidence interval. A t-test comparing the persister percentages of 48 samples each of FT and NC during log phase showed a significant difference (p value = 3.10E-05). A t-test comparing their variances (CVs) also showed a significant difference (p value < 0.05). The error bars indicate standard deviation.

The Fluctuation Test (FT) was performed in a similar way to the NC but starting from a single cell. To perform the FT, we diluted the bacterial culture down to ~0.5 cells per well in a 96-well optical bottom microplate. Since cells cannot physically split and remain viable, this dilution ensures that, on average, about one cell is present in every two wells, resulting in growth in roughly 48 out of the 96 wells60,61. These clonal cultures were grown to an OD600 of about 0.4–0.5 before being treated with Cip for 3 h to assess persister levels. Since not all wells reached the target cell density simultaneously, two 96-well microplates were used for the assay, but only the 48 wells that achieved OD600 0.4–0.5 were treated with Cip.

Since both the FT and NC experiments contain genetically identical cells exposed to the same environmental conditions and at equal densities, we would generally expect no differences between them. However, we hypothesize that a subset of cells within the population may be pre-conditioned to become persisters (like primed cells), which should result in a noticeable difference in persister counts between the FT and NC. Indeed, the contrast between the FT and NC persister counts was striking. The FT results for P. megaterium treated with Cip showed a ~ 60-fold variation in persistence (Fig. 3C) compared to only ~2-fold variation in NC. These findings were replicated multiple times with consistent results. Since the bacterial culture was diluted down to single cells and subcultured to the log (exponential) phase, any carryover from the stationary phase was eliminated. The P. megaterium FT results are consistent with those observed in E. coli, where ~80- and ~60-fold differences in persister percentages were observed with ampicillin and apramycin, respectively34.

Two hypotheses could explain the high variation observed in the FT. First, though unlikely, the high fluctuations in a few clonal populations may be due to some mutation leading to resistance, rather than persistence. Second, some bacterial cells in a population have already been prepared or “primed” to become persisters that can withstand lethal doses of antibiotics. If the cells from the high persister culture were resistant, then their Minimum Inhibitory Concentration (MIC) is likely to increase62. To investigate this, we tested the MIC of Cip for P. megaterium wild-type culture and that of high-persister clones. However, the MIC of the high-persister clone was identical to that of a susceptible P. megaterium culture (0.064 µg/mL) (Supplementary Fig. S1A), making the mutation hypothesis highly unlikely. We further tested this by selecting multiple colonies from high-persister plates and streaking them on Cip plates containing 6.4 µg/mL Cip (100X MIC). No growth was observed (Supplementary Fig. S1B), further ruling out resistance. Additionally, an FT was performed with a high-persister clonal culture. If mutations were responsible for the increased persister numbers, we would expect the average persister percentage to rise compared to the wild-type clonal populations. However, repeating the persister assay with the high-persister clone revealed no apparent significant difference from the wild-type (Supplementary Fig. S1C). These experiments refute the first hypothesis and confirm that some cells in a bacterial population are primed to become persisters before exposure to antibiotic stress. These results align with our previous work with E. coli primed cells34, leading us to term these stress-prepared cells as “Priestia primed cells.”

In our previous work with E. coli primed cells, we used the antibiotics ampicillin (β-lactam targeting peptidoglycan synthesis) and apramycin (aminoglycoside targeting the 30S ribosomal subunit)34,63,64. In this study, we used Cip, a quinolone that inhibits DNA gyrase65. This reinforces that bacterial cells can prepare in advance for generalized antibiotic stress, rather than being primed specifically for any one class of antibiotics. In addition, since Gram-positive and Gram-negative bacteria are evolutionarily distant, these findings show that primed cells represent a widespread phenomenon, common across diverse bacterial lineages.

Primed cells exhibit short-term memory

We previously demonstrated that E. coli primed cells have a memory that is passed down to their progenies, with the high variation in persistence being non-random across generations34. Here, we sought to determine if P. megaterium exhibited similar behavior. Our null hypothesis was that no memory exists, and the variation range in persistence would be random. To test this, we employed a method to compare persister levels between replicas (Fig. 4A). A P. megaterium culture was diluted into single cells and grown to mid-log phase (OD600 of about 0.4–0.5). The culture was then divided into two 96-well microplates, further diluted (to 1:2, 1:10, 1:100, or 1:500), and allowed to grow. Persister assays (FTs) were performed with both replicates using Cip, with each plate (Replica Plate 1 and Replica Plate 2) grown independently from each other. If memory was present, corresponding well in Replica Plate 1 should exhibit similar persister levels to those in Replica Plate 2. Conversely, if no memory existed, corresponding well should show no correlation in persister levels between the two plates.

Fig. 4: Primed cells have transient memory, and slow-growing cultures are not more prepared to survive Cip.
Fig. 4: Primed cells have transient memory, and slow-growing cultures are not more prepared to survive Cip.The alternative text for this image may have been generated using AI.
Full size image

A Schematic of Memory Test, which measures the variation in persister percentages at the single cell level among clonal populations across two replicates. B P. megaterium primed cells exhibit a transient memory before antibiotic addition, with this memory being passed down for ~9 generations. Statistical analysis using a t-test indicated no apparent significant differences between the FT of the two replicates for each dilution (n = 48 for each replicate). Both the x- and y-axes are log10 scaled. C P. megaterium cultures with varying cell densities (as measured by CFU/mL and OD600), grown over the same duration, exhibited no correlation with persister percentages (indicated as % Survival). Additionally, no correlation was observed between the growth rates of individual clonal cultures or their lag periods with persister percentages (n = 36). The y-axis is a log10 scale.

Persister levels in both replicates showed a strong correlation when cells were split and diluted 1:2 and 1:10, contradicting our null hypothesis (Fig. 4B). The correlation weakened at 1:100 dilution and was lost at 1:500, which was expected since the memory is transient. These findings further support that mutations are not responsible for the high variation in persister percentages in clonal populations because the survival advantage is only a transient one.

Since persisters are genetically identical to antibiotic-susceptible cells, the logical conclusion is that memory is epigenetic. Epigenetic memory refers to short-term behavioral changes that propagate this transient yet stable cellular phenotype from parent to daughter cells66. A primed cell divides into two primed cells, continuing the memory, but eventually, the non-genetic change fades. In our case, the memory persisted for ~9 generations.

Persister numbers do not always depend on the time of growth

In all previous experiments, we maintained similar cell densities (ODs around ∼0.5) to minimize potential variations between wells. However, when a P. megaterium culture is diluted to single cells in a 96-well microplate (~0.5 cells/well) and incubated, the wells do not all grow to the same density at the same time. We aimed to determine whether the overall cell density of a log culture influenced the number of primed or persister cells. To investigate this, we performed an FT with 36 clonal populations grown over the same time period, with OD600 values ranging from 0.3 to 0.8 (Supplementary Fig. S2A). Our results showed no correlation between cell density (measured by OD600 or CFU/ml) and persister levels (Fig. 4C).

According to well-established and well-cited literature, slower growth rates give rise to higher persister numbers67,68,69. Thus, we would expect clonal populations with lower OD600 values to exhibit greater persister percentages, but the results did not support this prediction. While our goal was to achieve 1 cell/well, this will not always be the case. Some wells may contain more than one cell, resulting in higher densities without necessarily faster growth rates. To effectively assess the relationship between slow growth and persister numbers, we compared growth rates across wells. Our analysis revealed no correlation between the growth rates of individual wells and their persister numbers (Fig. 4C), thereby contradicting existing literature, which suggests that slower growth rates contribute to increased persistence. Additionally, delays in cell division may influence persister levels. This delay can be quantified by measuring the lag time. Again, we found no correlation between the lag periods of individual wells and their persister numbers (Fig. 4C).

This FT result shows that slow growth does not lead to appreciable changes in persister or primed cell levels. One limitation of this experiment is that we do not directly visualize the growth rates of single cells. However, recent research has shown that antibiotic lethality is more closely tied to the metabolic state of cells rather than their growth rate70, which aligns with our findings. Based on our results, we concluded that the majority of persisters formed are not from slow-growing cells. If slow-growing cells do contribute to higher persister levels, they likely represent only a minor fraction of the overall population of persisters.

The number of primed cells increases in the post-log phases, leading to a higher prevalence of persisters during the transition and stationary phases than in the log phase

Previous studies have indicated that bacterial persistence levels tend to rise in the late-exponential phase in some bacteria14,36,71 (Fig. 5A). For example, Pseudomonas aeruginosa modulates persister frequency partly via quorum sensing72,73. These are density-dependent responses. We aimed to determine whether persister levels and primed cell numbers in P. megaterium varied based on cell density. Our findings indicated no apparent significant differences in persister percentages across OD600 values ranging from 0.25 to 0.8 (log phase) in the presence of Cip (Fig. 5B). This lack of variation was not unexpected, as during the log phase, a stable environment is maintained with abundant nutrients. However, we were more interested in what would happen to the persister and primed cell numbers when cells were in a more stressful environmental condition, such as those present during the transition and stationary phases (Fig. 5A). The average persister percentage of P. megaterium culture increased ~2-fold in the transition phase (OD600 ~ 1.0) and ~4-fold in the stationary phase (OD600 ~ 1.5) as compared to that in the log phase (Fig. 5C, D). The transition and early stationary phases were chosen as they are devoid of spores (Supplementary Fig. S2C) and yet are sufficiently more heterogeneous than the log phase. Despite the average persister percentage increasing in the post-log phases, for both NC and FT, the CVs, and the fold changes (the differences between the highest and lowest persister percentages) remained similar to that in the log phase (Fig. 5E. i). The only notable differences were in the mean and median values.

Fig. 5: The number of persisters increases in the post-log phases of bacterial growth (Log < Tran < Stat) with a proportional increase in the number of primed cells.
Fig. 5: The number of persisters increases in the post-log phases of bacterial growth (Log < Tran < Stat) with a proportional increase in the number of primed cells.The alternative text for this image may have been generated using AI.
Full size image

A P. megaterium growth over a 15 h period results in 3 different growth phases: Log, Transition (Tran), and Stationary (Stat) (n = 93). B The persister numbers remained constant during the log phase but increased in the transition and stationary phases: Stat >  Tran > Log. OD600 from 0.25 to 0.8, as indicated in the x-axis, belongs to the exponential phase of P. megaterium, while OD600 1.0 and 1.5 belong to the transition and stationary phases, respectively. (n = 3; the non-significant (ns) p values are in Supplementary Fig. S2B). Bacterial cultures were grown over time; samples were collected at respective optical densities and subjected to 6.4 µg/mL Cip for 3 h. Persister percentage was calculated by dividing the CFUs/mL post-antibiotic treatment by CFUs/mL pre-antibiotic treatment and then multiplying by 100. C, D At both the population level (NC) and single cell level (FT), the number of persisters increased significantly in the Tran and Stat phases under Cip treatment (6.4 µg/mL), with persister percentages averaging approximately 2-fold and 4-fold higher in the Tran and Stat phases, respectively compared to the Log phase (n = 48 per growth phase per experiment). However, the fold change (fold Δ = Max. persister %/Min. persister %) of NC or FT remained similar across all growth phases. Both the x- and y-axes are log10 scaled. Two more replicates (FT2, FT3) for both (C) and (D) are in Supplementary Fig. S3A, B. E Cip treatment results support that primed cells control persister levels in post-log phases. i The mean, median, CVs, and fold change (Fold Δ) of persister % of Tran/Log and Stat/Log phases in P. megaterium were calculated through bootstrapping with a 95% confidence interval. The mean and the median persister % increase post-Log phase, but the fold Δ and CV remain constant. The p values labeled ns for the comparison between Tran/Log and Stat/Log are 0.25 for Fold Δ and 0.29 for CV. ii From Log to Tran to Stat phases, the numbers of primed cells proportionally increase with the number of persisters (pictorial representation). For simplicity, in this figure, we have not shown the other non-persister cell types as well as how the primed cells are formed in the different phases of growth. All the error bars in (A, B, C, D, E) indicate standard deviation.

The overall findings indicate that the number of primed cells increases in the later growth phases, leading to a higher number of persisters (Fig. 5E. ii). It is often cited that there are two types of persisters: Type I persisters, which are non-growing cells formed in stationary phase, and Type II persisters, which originate from slow-growing cells that continually generate35. Our results demonstrate an increased presence of persisters in the stationary phase while providing evidence that persisters from both stationary and log phases do not represent distinct types (Type I or II). We base this conclusion on three key points we observed from our results:

  1. 1.

    The total number of persisters formed is regulated by the number of primed cells, regardless of the growth phase. If Type I and Type II persisters were formed through different mechanisms, it would be contradictory for primed cell numbers to influence both types.

  2. 2.

    Primed cells in the log phase are not characterized by slow growth, yet they play a crucial role in determining persister levels.

  3. 3.

    The “macro” growth rate does not appear to impact the levels of primed cells or persisters (Fig. 5B). It’s important to clarify that the “macro” growth rate does not accurately reflect the growth rates of individual cells. While our findings do not exclude the possibility that individual cell growth rates might be associated with persister formation, they do indicate that the number of primed cells is a more significant driver of cell survival than slower growth rates.

Primed cells are prepared for stress in general

Our experiments with P. megaterium and previous work with E. coli34 focused on antibiotic survival; however, persistence may be a general survival strategy. Notably, earlier studies have identified persisters in response to non-antibiotic stresses74. To investigate whether primed cells in a bacterial population are prepared for survival against other stressors, we exposed P. megaterium to lethal concentrations of fluoride. Commonly found in toothpaste and water supplies, fluoride is utilized for its ability to prevent dental cavities in humans28,75,76,77,78,79. However, in high concentrations, it becomes bactericidal, primarily by diminishing bacterial acid tolerance through a combination of enzyme inhibition, alterations in membrane permeability, and disruption of certain nutrient metabolic pathways80,81,82,83,84,85,86,87,88,89,90,91. The persister percentage fold differences for the NC and FT with fluoride were ~3-fold and ~78-fold, respectively (Fig. 6A). These results demonstrated that primed cells are indeed prepared to face a wide range of stressors. Maintaining elevated levels of primed cells confers a selective advantage to bacterial populations, enhancing their phenotypic plasticity in response to environmental stress.

Fig. 6: P. megaterium persister cells survive exposure to lethal levels of fluoride, a non-antibiotic chemical stressor.
Fig. 6: P. megaterium persister cells survive exposure to lethal levels of fluoride, a non-antibiotic chemical stressor.The alternative text for this image may have been generated using AI.
Full size image

A The fold change (fold Δ) in persister levels is ~3-fold for the total population (NC), but ~78-fold for clonal populations arising from single cells (FT) (n = 48 per experiment). The error bars indicate standard deviations. B The population-level average persister percentages of P. megaterium (NC) increase from log (OD600 0.5) to stationary phase (OD600 1.5) by ~2.5-fold when treated with lethal levels of fluoride. Bacterial cultures (n = 8) were grown over time; samples were collected at respective optical densities and subjected to a final concentration of 60 mM NaF for 3 h. Persister percentage was calculated by dividing the CFUs/mL post-chemical treatment by CFUs/mL pre-chemical treatment, multiplied by 100. The error bars indicate standard deviations. The results demonstrate that persistence is more pronounced in the stationary (Stat) phase compared to the Log growth phase because of increased primed cell numbers.

We have demonstrated that elevated levels of P. megaterium primed cells significantly increase persister numbers in the stationary phase compared to the log phase using Cip. To determine whether this increase is a general phenomenon or unique to antibiotics, we tested bacterial cell survival in stationary phase with lethal concentrations of fluoride. The results revealed that the number of primed cells in the stationary phase was higher than in the log phase with fluoride, demonstrating that this phenomenon is not limited to antibiotic exposure (Fig. 6B).

Primed cells form in other Gram-positive bacteria as well

We also tested another Gram-positive bacterium, Bacillus subtilis, against lethal doses of Cip (4.7 µg/mL) and found a similar pattern of priming (Supplementary Fig. S4A). We have demonstrated that primed cells are common across many bacterial species by showing their form in two Gram-positive species in this work and a Gram-negative species in our previous work34 using different antibiotics.

Implications and limitations of this work

Bacterial populations employ multiple approaches to survive stressful environments, one of which is bet-hedging. This approach involves maintaining a variant phenotype that is maladjusted for the current environment but suited for a changed environment. This phenotypic plasticity arises from noise in gene expression23,29,92,93. Factors like cell size, division stage, and extracellular conditions also contribute to variability. While noise might seem like an unlikely driver of such a crucial survival trait, growing evidence underscores that stochastic gene expression, along with the inheritance of these altered gene expression patterns, is a key driver of these fluctuations94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109. This randomness fosters phenotypic diversity that enhances the population’s adaptability under stress. Understanding the fundamental mechanisms of heterogeneity is pivotal for developing effective therapies targeting bacterial persistence.

We did not address viable but nonculturable cells (VBNCs). While previous studies suggested that VBNCs exist in concentrations comparable to CFU/mL110,111, recent evidence indicates that some of the high reported VBNC levels in log cultures may stem from improper microscopy counts112. Regardless, we examined the potential revival of VBNCs through the ~0.5 cell/well method based on the CFU/mL. If our dilution technique is correct, then a ~0.5 cell/well dilution should yield growth in ~48 wells out of 96 in a 96-well microplate. If VBNCs exist, then we should get ~1 cell/well instead. We consistently observed growth in ~48 wells with 0.5 cell/well dilution, indicating non-interference of VBNCs. Our results were corroborated by growth in ~24, ~12, and ~6 wells when we performed 0.25, 0.125, and 0.0625 cell/well dilutions, respectively. The R2 is 0.99 (Supplementary Fig. S3C) when the number of wells with growth is plotted vs. cell/well based on CFU/mL. While this technique may not have encompassed all the intricacies of VBNCs, and not every well is likely to originate from 1 cell, we tested P. megaterium with ~2 cells/well and again observed a significant variability with Cip treatment (~60-fold difference) (Supplementary Fig. S3D). This indicates that even if VBNCs are present, they do not skew our results, besides reinforcing that the FT does not necessitate a 1-cell/well dilution.

Why do bacteria have both persisters and primed cells? As noted in the introduction, heterogeneity can reduce fitness in an optimal, steady-state environment. By limiting the number of cells exhibiting these alternative behaviors, the overall impact on population fitness is minimized. Having both primed and persister cells allows for diversity, with primed cells still growing and potentially reducing the fitness costs associated with non-growing persisters.

We previously provided evidence that no single gene controls persister formation; instead, multiple pathways are involved32. The same may be true for primed cells. We do not know if P. megaterium, B. subtilis, and E. coli primed cells arise through similar or different mechanisms. Additionally, we know little about the switching between primed cells, persisters, and susceptible cells. It seems unlikely that the gene expression changes leading to the primed cell state is being carried over to the non-dividing persister or after the persister emerges from persistence. We do not know if cells emerging from persistence transition directly to susceptible cells or have a primed cell intermediate. Though we do not have evidence for the latter (persister cell ←→ primed cell ←→ susceptible cell), this seems like a logical intermediate step. We know that primed cells originate from susceptible cells and can lead to more primed, susceptible, and persister cells (Fig. 1C). While previous studies indicate that extremely slow-growing cells can become persisters67, primed cells do not exhibit such slow growth34. This contradiction suggests that not all persister cells come from primed cells. Furthermore, this work does not have the sensitivity to test if primed cells are growing slightly slower or at the same rate as susceptible cells.

Our results indicate that while there are more persisters in stationary phase, distinct Type I and Type II persisters do not exist (with Type II forming in the log phase and Type I in the stationary phase)35,113. If Type I and Type II persisters were formed through different mechanisms, we would expect variations in CV (standard deviation/mean), range (upper and lower limits of persisters), and their fold difference across our FT replicates. Instead, our findings show that the increase in persisters during post-log phases is simply due to a higher number of primed cells compared to the log phase.

A cellular burden can affect growth and lead to various effects such as increased ATP use, higher expression of specific genes, and accumulation of intermediates, among others. As cells transition from the log to the stationary phase, a burden can be detected by measuring growth rates. While other micro-burdens (e.g., Reactive Oxygen Species) may arise in individual cells during the log phase, cells adapt to them with minimal impact on the population growth rate. These micro-burdens cause network disruptions, prompting an altered gene expression response that could temporarily lead to cells being more prepared for deadly stress (i.e., antibiotics). This prepared state (altered expression of specific genes) could be passed down several generations, which we think happens in primed cells. It is logical to predict that the transition and stationary phases harbor more of these micro-burdens than the log phase, resulting in an overall increased level of primed cells.

Primed cells and persisters should not be confused with heteroresistent cells because they differ in significant ways114. Heteroresistent cells are resistant to the antibiotic but can lose this ability without antibiotic selection in a few generations. Unlike heteroresistant cells, persisters neither mutate to survive nor grow under lethal antibiotic treatment.

We also considered reversible chromosomal rearrangements (RCRs) as a possible mechanism of primed cell formation and their transient memory115. RCRs involve structural changes in the chromosome—such as inversions, translocations, or duplications—that can revert to the original configuration. These rearrangements typically involve the reorganization or movement of large DNA segments within the chromosome. Unlike other forms of mutations, which permanently alter the DNA sequence, RCRs allow cells to switch back and forth between configurations, enabling rapid adaptation to changing environments. However, in our study, RCRs appear unlikely to be the cause of primed cells, as we observe a consistent loss of memory with 1:500 dilution (~9 generations) in P. megaterium and a similar pattern with 1:100 dilution in E. coli. If RCRs were responsible, we would expect a more random distribution of memory loss. Nonetheless, we cannot entirely rule out the potential role of RCRs in combination with epigenetic modifications.

Epigenetic modifications and inheritance likely contribute to primed cell formation and their transient memory. Several mechanisms of epigenetic inheritance exist in bacteria116,117,118,119,120. While DNA methylation is the most extensively studied form of epigenetic memory in bacteria121,122, other mechanisms, such as multisite phosphorylation123,124 and ploidy125 have the potential to control primed cell memory. Transgenerational epigenetic inheritance is another possibility. It refers to the transmission of parental cellular/chemical states (e.g., protein or RNA or DNA modifications) to their offspring, thereby enhancing the progeny’s ability to adapt to environmental conditions previously encountered by their ancestors. This phenomenon has been mentioned in bacterial studies126,127, with the Lac operon being one of the most well-known examples of such inheritance. Toxin-antitoxin (TA) systems were once believed to be the primary drivers of persister formation128; however, recent research showed they are unessential for persister formation and play a secondary role in modulating persistence levels129.

A transient, heritable state may be evolutionarily favored as a survival strategy, as it offers adaptability without the need for a long-term commitment like DNA mutations94. Now that we have identified primed cells in both Gram-positive and Gram-negative bacterial species and demonstrated their crucial role in increasing persister numbers in response to various antibiotics, non-antibiotic chemical stress, and across different growth phases, the next step is to uncover the mechanisms underlying their formation and the processes that sustain their epigenetic memory. Since primed cells give rise to persisters, future work on finding a mechanism for primed cell formation can lead to the development of drugs that could prevent persister formation and eventually reduce the rate of antibiotic resistance.

Materials and methods

Microbial strains, media, and buffer

Bacillus megaterium (basionym) de Bary ATCC 14581, currently known as Priestia megaterium, and Bacillus subtilis strain 1A1 168 were used in this work. The bacterial cultures were grown in Miller’s modification Lennox lysogeny broth (LB) liquid media or LB with agar. All cultures were incubated at 37 °C and shaken at 250 or 300 rpm. Minimal Media B (MMB)32,130 was used to buffer dilutions.

Growth curve

P. megaterium culture was grown overnight in LB media at 37 °C and 250 rpm in a shaker incubator. The overnight culture was sub-cultured 1:100 and distributed evenly into the wells of a ThermoFisher Nunc MicroWell 96-Well Optical Bottom Plate, with at least 3 wells containing only LB media acting as blank. The plate was incubated in a FLUOstar Omega microplate reader at 37 °C and 300 rpm for about 20 h. Absorbance was measured at 600 nm at regular intervals. The blank was subtracted, and a semi-log graph of Log OD600 nm vs Time was generated.

Death curve

P. megaterium culture was grown overnight in LB media at 37 °C and 250 rpm in a shaker incubator. The overnight culture was sub-cultured at 1:100 in three replicates. The replicates were grown until the log phase (OD600 ~ 0.5), and a specific volume from each replicate was diluted and plated. This served as the CFU/mL for 0 h. The remaining volume of culture in replicates was treated with Cip at a final concentration of 6.4 µg/mL (100× MIC) and incubated at 37 °C, 250 rpm. A certain volume of the cultures was taken out after 1, 2, 3, 4, 5, 12, and 24 h post-antibiotic addition, diluted, and plated. These served as the CFU/mL for the respective time points. Persister percentages were calculated for each replicate at each time point (Persister percentage = ((CFU/mL at respective time point [e.g., 1 h/2 h/3 h, etc.] *100)/CFU/mL at 0 h)), and their logarithmic values were plotted against time to generate a semi-log death curve.

Population-level persister percentage variation assay: Noise Control (NC)

A log phase (OD600 ~ 0.5) P. megaterium culture, grown in LB media, having ~1E + 7 CFU/mL, was divided into 48 wells (to serve as 48 replicates) of a ThermoFisher Nunc MicroWell 96-Well Optical-Bottom Plate. A specific volume from each well-replicate was diluted and plated. The remaining culture was then treated with Cip at a final concentration of 6.4 µg/mL (100× MIC) for 3 h, at 37 °C and 300 rpm, in a FLUOstar Omega microplate reader. Post-treatment, cultures from each replicate were again diluted and plated. Plates were incubated for 10–12 h at 37 °C and then scanned using a flatbed Epson Perfection V370 Photo scanner. Custom scripts were run to count bacterial colonies from the scanned images131. Persister percentages were calculated for each replicate by ratioing their CFUs/mL post-antibiotic treatment to CFUs/mL pre-antibiotic treatment, multiplied by 100. This process yielded 48 different persister percentages from 48 different wells. The fold difference was calculated by dividing the maximum by the minimum persister percentage across the wells. The mean persister percentage and standard deviation (SD) of all the wells were calculated using bootstrapping in Python with R 1000 and Seed 42 parameters. A similar procedure was followed for cultures in later growth phases (post-log). P. megaterium cultures at the transition phase had an OD600 of ~1.0 (~2E + 7 CFU/mL), and at the stationary phase had an OD600 of ~1.5 (~4E + 7 CFU/mL). For B. subtilis, a log phase (OD600 ~ 0.5) culture, having ~1E + 7 CFU/mL, was used for the assay at a final concentration of 4.7 µg/mL of Cip (100× MIC).

Single cell-level persister percentage variation assay: Fluctuation Test (FT)

A log phase (OD600 ~ 0.5) P. megaterium culture, grown in LB media (freshly sub-cultured or frozen at −80°C), having ~1E + 7 CFU/mL, was diluted to ~0.5 cells per well (or 1 cell per 2 wells) in a ThermoFisher Nunc MicroWell 96-Well Optical Bottom Plate and typically about 50% of the wells had growth (ideally 48, but generally 46–50 out of 96 wells). For each single cell-level persister percentage variation assay, the dilution procedure was done twice to confirm accuracy. Plates were incubated in FLUOstar Omega microplate reader at 37 °C, 300 rpm for 6–8 h, allowing the cultures to grow to log phase (~1E + 7 CFU/mL, OD600 ~ 0.5). As per the dilution, each of the 48 wells represents a clonal population likely proliferating from a single cell. A specific volume of the cultures from each clonal population was diluted and plated. The remaining culture in each well was then treated with Cip at a final concentration of 6.4 µg/mL (100× MIC) for 3 h, at 37 °C and 300 rpm, in a FLUOstar Omega microplate reader. Post-treatment, cultures from each clonal population were again diluted and plated. Plates were incubated for 10–12 h at 37 °C and then scanned using a flatbed Epson Perfection V370 Photo scanner. Custom scripts were run to count bacterial colonies from the scanned images. Persister percentages were calculated for each clonal population by ratioing their CFUs/mL post-antibiotic treatment to CFUs/mL pre-antibiotic treatment, multiplied by 100. This process yielded 48 different persister percentages from 48 different wells. The fold difference was calculated by dividing the maximum by the minimum persister percentage across the wells. The mean persister percentage and standard deviation (SD) of all the wells were calculated using bootstrapping in Python with R 1000 and Seed 42 parameters. A similar procedure was followed while doing this assay at later growth phases (post-log). P. megaterium cultures at the transition phase had an OD600 of ~1.0 (~2E + 7 CFU/mL), and at the stationary phase had an OD600 of ~1.5 (~4E + 7 CFU/mL). For B. subtilis, a log phase (OD600 ~ 0.5) culture, having ~1E + 7 CFU/mL (freshly subcultured or frozen at −80°C), was used for the assay at a final concentration of 4.7 µg/mL of Cip (100× MIC).

Antibiotic resistance assay

After each fluctuation test, multiple colonies from P. megaterium high-persister wells were picked and streaked with and without Cip-containing LB agar plates and incubated at 37 °C for 10–12 h. The final concentration of Cip in the LB agar plates was 6.4 µg/mL.

Spore detection assay

P. megaterium stationary phase cultures (OD600 ~ 1.5) having ~4E + 7 CFU/mL were subjected to heat shock (100 °C for 1 min) and plated to check for the presence of spores. The plates were incubated at 37 °C for 72 h. No growth was detected on LB plates, indicating no viable spore formation.

Memory assay

A log phase (OD600 ~ 0.5) P. megaterium culture having ~1E + 7 CFU/mL was diluted down to 0.5 cells per well, as described above in the FT procedure. Cultures were incubated in a microplate reader at 37 °C, 300 rpm for 6–8 h, and grown to log phase ~1E + 7 CFU/mL (OD600 ~ 0.5). 48 wells serving 48 different clonal populations were selected for further experimentation. Then, they were diluted in prewarmed LB (37 °C) to 1:2, 1:10, 1:100, and 1:500, respectively. Each dilution was performed in 2 replicates. Fluctuation Test (FT) was performed with the 2 replicates simultaneously for each dilution using Cip following the FT procedure. Since we had two 48-well replicates, we got a total of 96 persister percentages, 48 from each replicate. The R2 values were determined from a regression line for the persister percentages of each well in both the replicates for all the dilutions.

Fluoride assays

We chose sodium fluoride (NaF) as a non-antibiotic chemical stressor to test for primed cells in P. megaterium. Both population-level (NC) and single cell-level (FT) persister percentage variation assays were performed in an identical way, as mentioned above, but at a final concentration of 60 mM (Minimal Bactericidal Concentration) NaF instead of Cip. The MIC of P. megaterium with NaF was 21.4 mM. MMB was used to buffer the reduction of pH by NaF in the culture media.

Minimal inhibitory concentration (MIC) assay

A log phase (OD600 ~ 0.5) P. megaterium or B. subtilis culture (Supplementary Fig. S4B), grown in LB media, was used to make a lawn on a LB agar plate. A Liofilchem Cip MIC Test Strip132 was then placed on the agar surface, and the plate was incubated at 37 °C for 8–10 h. The MIC value was determined as indicated by the tail of the zone of exclusion.

Colonies from a high-persister plate of P. megaterium obtained through FT were re-cultured in LB media up to the log phase (OD600 ~ 0.5) and were used to make a lawn on an LB agar plate. The MIC was determined similarly, using a Cip test strip. MIC of NaF was determined using the serial microdilution technique.

Statistics and reproducibility

Microsoft Excel, Spyder (Python version 3.12), and RStudio (version 2024.04.2 + 764) programming applications were employed for post-hoc statistical data analysis and plotting. Statistical significance (two-tailed p value) was determined using a t-test with either similar or dissimilar variance (based on F-test results). For all experiments, the sample size (n) and p values are mentioned in the figures and/or legends in both the manuscript and the supplementary materials. Experiments (especially the modified fluctuation tests) were repeated to ensure reproducibility (Fig. 3C and Supplementary Fig. S3A, B). The growth rate in Fig. 4C has been calculated using the formula:

$${Growth\; rate}=\frac{{Ln}\left(\frac{{OD}0.3}{{OD}0.1}\right)}{({Time\; taken\; to\; reach\; OD}0.3-{Time\; taken\; to\; reach\; OD} \, 0.1)}$$

Reporting summary

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