Primed cells are prepared for stress before antibiotic treatment
Bacterial populations are heterogeneous, even in log growth phase in a well-controlled lab environment, and especially in natural systems21, 22. The quantification of persister numbers often have huge variations with large error bars (SD or SEM), sometimes with 100s of fold differences23–25. We used defined media (MMB + 26, 27), which contains only chemically known components, to minimize variations between experiments. We optimized and standardized our experimental design, and reduced the internal error rate to below 2-fold. Much of this error comes from the compounding imprecisions from multiple pipetting (all pipets have an error rate, and these experiments require serial dilutions). Despite reducing the error rate, we noticed that there would occasionally be huge outliers (up to 100-fold higher or lower than the average).
We determined the typical persister variation in an E. coli population (Noise control: NC) is higher than our internal error. We grew a culture to mid-log phase and then divided it into 48 wells. We then treated it with ampicillin (Amp) or apramycin (Apr) for 3 h, followed by a persister assay (Fig. 1d). This resulted in a variation of ~ 6-fold for either antibiotic. These results do not explain the 100-fold changes we occasionally observed. Unable to explain the variation at the population level, we wondered if a "memory" was passed down over several generations and skewed our data. We set out to test for variations from a single cell using a FT13, 16, 17. Cultures were diluted to about 0.5 cell/well (Limiting Dilution assay28, 29) in a 96-well microplate; on average, 48 wells had growth, and 48 wells did not (Fig. 1e). Cells were then grown to an OD of 0.4–0.6 (log phase) and treated with Amp for 3 h. Most (but not all wells) reach the desired OD range simultaneously. We use two 96-well plates and pick 48 clones within the OD range to deal with this. After treatment, the cells were plated on Petri dishes, grown, and colonies were counted to get CFU/mL. We first tested lethal Amp dosages for 3 h, and the persister range was vast, ~ 60-100-fold. We tested this several times (Fig. 3d; 8 separate experiments with ~ 48 clones/experiment), and there is consistently an extensive range of persister variation among the clones. We wondered if this was specific to Amp, so we tested another class of antibiotic, Apr. Amp targets the cell wall30, while Apr targets the 30S ribosomal subunit31. Again, the persister range was vast, ~ 40-70-fold range with Apr (Fig. 2a). We hypothesize that the extensive range in persistence is due to either (1) mutation or (2) some cells are prepared for stress ("primed cells”), and these primed cells exhibited specific characteristics which allow them to prepare prior to the stress. To assess if a mutation causes this extensive persister range, we tested for antibiotic resistance by streaking clones on antibiotic plates, and no resistant colonies grew (Fig. S1a). To further test for mutations, we diluted the high persister clone (Hp clone) into 0.5 cell/well and repeated the fluctuation test. If a high persister clone was mutated, the average persister percentage would increase, and the range would decrease. However, they had the same average persister percentage and similar range (~ 60–100 fold) in both fluctuation experiments (Fig. 2b i). If the cells developed resistance, their minimum inhibitory concentration (MIC, the minimal antibiotic required to hinder growth) would change. But it remained the same (Fig. 2b ii). These results undoubtedly rule out mutations as the cause and led us to test hypothesis 2, that some cells are primed prior to stress.
Cell density is not the primary factor for variation in persister number
Some bacterial persistence levels are controlled (in part) via quorum sensing32, which is a density-dependent response. Thus, before testing hypothesis 2, we wanted to know how much cell density could skew our results by testing the overall persister range with different cell densities using FTs. For FTs in Fig. 2d, cells were harvested at OD 0.4–0.6 since the persister levels are remarkably similar at these ODs. We detected no correlation between cell density and % Survival in 15 FTs (7 Amp and 8 Apr) and a weak correlation in FT3 with Amp treatment (Fig. S1b i). Thus, to further determine how much cell density affects the persister levels, we tested persister levels from OD 0.3 to 0.7 in ~ 0.1 OD intervals (cells are in log phase and cell density ranges from ~ 2E07 to 2E08 CFU/mL) in a general population at 3 h Amp or Apr. No appreciable correlation was observed with either antibiotic (Fig. S1b ii).
Cells are primed for persistence, not short-term tolerance
Short-term tolerance can mask persistence, and experimental evidence has shown that the phenotypes are distinct from each other33. Short-term tolerant cells are dividing and likely have different survival mechanisms than persisters. In the initial stage of antibiotic treatment, there are far more short-term tolerant cells than persister cells (Fig. 3i). We did a similar FT with clones grown from a single cell and treated them with a lethal Amp concentration (0.1 mg/mL). % Survival was determined for 1 h and 3 h of treatment using persister assays26, 27, 34, and short-term tolerance at 1 h does not indicate the level of persistence at 3 h with lethal Amp; no correlation at 1 h vs 3 h population (r2 = 0.02). If the primed cells are advantageous for long-term survival, we expect the high persister populations observed at 3 h treatment to stay high with more prolonged antibiotic exposure. As expected, % Survival at 3 h highly correlates with % Survival at 4 h (r2 = 0.85) and 5 h (r2 = 0.78) Amp treated population (Fig. 3). Therefore, our results demonstrate that cells are primed for persistence and not for short-term tolerance.
Primed cells have transient memory: Next, we determined whether high persister clones arise randomly due to noise in gene expression levels or in rare events where the gene expression level (memory) is passed down for several generations. We hypothesize that there is a transient memory at the transcriptomic level. The null hypothesis is that there is no memory and the variation range in persistence is random and solely due to noise. To test this, we divided and diluted the culture between 1:1 to 1:100 into separate microplates and allowed them to grow (Fig. 4a). If the null hypothesis is correct, there should be no correlation between divided cultures. However, persister levels were strongly correlated until the 1:20 dilution, and the memory is completely lost after a 1:100 dilution, supporting our transient memory hypothesis (Fig. 4b). We further confirmed the transient memory hypothesis with Apr (Fig. 4c). These results also add additional support that primed cells are not mutants, because 1:100 dilution led to no long-term (genetic) survival phenotype (in several different clones), as a resistance mutation would allow.
We hypothesize that the same primed cells will allow higher persister levels when using antibiotics from classes that target distinct cellular processes, e.g. Amp and Apr. If both Amp and Apr primed cells use an akin mechanism, we expect a reasonable correlation between their persister numbers/well. If they do not correlate, diverse types of primed cells likely exist. To understand this, we did experiments similar to Fig. 4a, but we tested Replica 1 with Amp and Replica 2 with Apr. In this case, both replicas had a transient memory, and the memory was lost by 1:20 dilutions (Fig. 4d).
Primed cells are not spontaneous persisters
Previous researchers proposed spontaneous persisters formation; persisters that are generated stochastically at a constant rate during exponential phase growth and switch to a dormant or a protected state (distinct slower growth rate than other cells, and this slowed growth rate is maintained for several generations and turn into persisters in the presence of stress)5, 20, 35. In addition, they proposed that during exponential growth, these phenotypic variants (e.g. persister formation) could also be induced by stress5. However, several research groups criticized the concept of spontaneous persister formation36, 37, questioning if it exists or is a proper terminology. In the original paper, where persistence was proposed in 1944, persisters were defined as non-dividing cells38. However, spontaneous persisters were defined as dividing cells39. Here we use the original definition of persisters; they do not divide. In addition, if persister formation is only induced by stress, all cells in the population should turn into persisters instead of a small percentage of the population. Also, induced persister formation (or sense-and-respond strategies) could be costly for the cells since it necessities constitutive expression of essential molecular machinery40.
On the contrary, primed cells (already prepared for stress in a population through heterogeneity) could provide a simple mechanism for adaptation to stresses they might or might not encounter. A key phenotype of spontaneous persisters is that they grow slowly. However, we did not observe any significant growth rate changes among the clones, and recent results demonstrate that persisters are not slow-growing before antibiotic treatment41. To further explore whether primed cells are reliant on slow growth, we constructed a simple mathematical model (explained in the Methods), where we plotted the fraction of persister cells \(\frac{y\left(t\right)}{x\left(t\right)}\) as a function of time for persister proliferation being 100%, 90%, 80%, 50%, and \(0\%\) of the proliferation rate of the drug-sensitive cells. We observed when persister cells do not proliferate (\({k}_{p}=0)\), their fraction rapidly dilutes back to the steady-state level in a few generations (Fig. 4e). These results show that a high correlation in persister maintenance, as seen in Fig. 4b for several generations, requires persister proliferation. For example, one requires \({k}_{p}=0.9{k}_{d}\) for it to take roughly 4 generations for the fraction of persisters to fall to 50% of its initial levels, similar to the drop in the correlation between replicates to 0.5 in Fig. 4b. Our evidence clearly shows that primed cells are not persisters (non-dividing cells) before antibiotic treatment since they grow and maintain a transient memory.