Bacterial persistence is an active σS stress response to metabolic flux limitation

Abstract

While persisters are a health threat due to their transient antibiotic tolerance, little is known about their phenotype and what actually causes persistence. Using a new method for persister generation and high-throughput methods, we comprehensively mapped the molecular phenotype of Escherichia coli during the entry and in the state of persistence in nutrient-rich conditions. The persister proteome is characterized by σ(S)-mediated stress response and a shift to catabolism, a proteome that starved cells tried to but could not reach due to absence of a carbon and energy source. Metabolism of persisters is geared toward energy production, with depleted metabolite pools. We developed and experimentally verified a model, in which persistence is established through a system-level feedback: Strong perturbations of metabolic homeostasis cause metabolic fluxes to collapse, prohibiting adjustments toward restoring homeostasis. This vicious cycle is stabilized and modulated by high ppGpp levels, toxin/anti-toxin systems, and the σ(S)-mediated stress response. Our system-level model consistently integrates past findings with our new data, thereby providing an important basis for future research on persisters.

Keywords: Escherichia coli; metabolism; persistence; proteomics; stress response.

Figure 1. Non‐/slow‐growing cells and starved cells are antibiotic‐tolerant, accumulate ppGpp, and express TAS

1. Dynamics of establishing antibiotic tolerance during entry into non‐/slow growth or starvation. Fraction of antibiotic‐tolerant cells after treatment with ampicillin (2 h, 100 μg ml⁻¹) is shown at various times after the medium switch. Gray disks: non‐/slow‐growing cells, open circles: starved cells. Data from biological triplicate. Error bars represent one standard deviation.

2. Fractions of antibiotic‐tolerant cells after a 2‐h treatment of non‐/slow‐growing and starved cells with various antibiotics (ampicillin 100 μg ml⁻¹; chloramphenicol 140 μg ml⁻¹; kanamycin 100 μg ml⁻¹; ofloxacin 5 μg ml⁻¹; rifampicin 100 μg ml⁻¹; CCCP 50 μg ml⁻¹) 4 h after nutrient switch. Gray bars: non‐/slow‐growing cells, white bars: starved cells. Data from biological triplicate. Error bars represent one standard deviation. Statistical significance (t‐test or Wilcoxon rank sum test for kanamycin and ofloxacin, P‐value < 0.05) is marked with *.

3. ppGpp concentration in cells growing on glucose and in cells shifted from glucose‐to‐fumarate medium. Data from biological triplicate. Error bars represent one standard deviation.

4. Log2 fold change in transcript abundance of first genes in TAS operons compared to cells growing on glucose, normalized to housekeeping gene abundance. Green bars: 2 h after switch from glucose medium to glucose medium, gray bars: non‐/slow‐growing cells 2 h after switch from glucose medium to fumarate medium, white bars: starved cells 2 h after switch from glucose medium to medium without a carbon source. Data from triplicate experiments. Error bars represent one standard error of the mean.

Source data are available online for this figure.

Figure 2. Persisters grow and are metabolically active

1. A

   Images of a cell growing on glucose, a persister cell and a starved cell. Scale bars: 1 μm. Volumes of cells growing on fumarate, cells growing on glucose, cells entering persistence and cells entering starvation. See also Appendix Table S1.

2. B

   Evaporation‐corrected cell count development of cell populations entering persistence and starvation. Gray disks: persister cells, open circles: starved cells. Values from each replicate were normalized to t ₀. Error bars indicate one standard deviation. Green line (persister cells) and red line (starved cells) represent a prediction of a linear regression fitted to the log‐transformed data, where slopes are equal to the growth rate with the dotted lines error margins representing the 95% confidence intervals determined by the model. Vertical gray area covering the period from 0 to 2 h visualizes the period of reductive division. Data from at least ten biological replicates.

3. C–E

   Time course of the fumarate uptake rate (C), the oxygen transfer rate (OTR; D), and the carbon dioxide transfer rate (CTR; E) of persister cells. Points indicate time‐specific rate values, and lines indicate fits from generalized additive models with 95% confidence interval (indicated by areas). Vertical gray area covering the period from 0 to 2 h visualizes the period of reductive division. Data from at least three biological replicates.

Figure 3. Persisters maintain high energy charge levels through respiratory metabolism

1. A–C

   Persister cells utilize a higher proportion of the taken up carbon for ATP production through respiration and less on biomass formation than cells growing on fumarate. The yields (relative to the up‐taken fumarate) were calculated as ratios of physiological rates (cf. Materials and Methods) and in case of ATP, on the results of the flux balance analysis maximizing ATP production using the estimated physiological rates as constraints. Data from at least three biological replicates. Error bars indicate one standard error of the mean.

2. D

   Maximal possible ATP production rates in persister cells and in fumarate‐growing cells, estimated by flux balance analysis maximizing ATP production using the estimated physiological rates as constraints.

3. E, F

   Adenylate energy charge (E) and sum of adenylate nucleotide concentrations (F). Gray disks: persister cells, open circles: starved cells, green bars: growing cells. Error bars indicate 95% confidence interval of the mean, calculated with a mixed effects model based on multiple biological replicates in multiple measurement campaigns (cf. Table EV2).

Source data are available online for this figure.

Figure 4. Persisters’ and starved cells’ proteome are shaped by the same cue

The experimental and data analysis procedure is described in the gray boxes step by step.

1. PCA plot of the Escherichia coli proteomes in different conditions and time points. Each point represents a proteome in a different state. The distances between points are inversely correlated with the similarity between proteomes (i.e. proteomes with higher correlation coefficient have a shorter distance between each other), calculated based on differences in the expression level of each quantified protein. Green disk: cells growing on glucose, green square: cells growing on fumarate, gray disks: cells entering persistence after glucose‐to‐fumarate switch, open circles: cells entering starvation from glucose, open squares: cells entering starvation from fumarate. Time series are indicated by gray color gradients.

2. The progression of changes upon entry into starvation and entry into persistence happens in the same direction in the two‐dimensional space, indicating that the same cue shapes these proteomes.

Figure 5. Proteome of persisters has enhanced catabolism and activation of stress response

1. A

   Projection of E. coli proteomes in various growth and stress conditions (Schmidt et al, 2016) on the PCA space created by proteomes generated in this study.

2. B

   PCA of proteomes of persister, fumarate‐growing, and glucose‐growing cells (upper panel); PCA of proteomes of persister cells and starved cells growing on glucose or fumarate before starvation (lower panel), markers as in (A). GOterms shared between the two analyses (i.e. persisters versus growing cells and persisters versus starved cells) are indicated in bold. GOterms specific to persisters versus starved cells analysis are indicated in italics. For a ranked list of assigned GOterms, see Appendix Table S3. See also Appendix Fig S2 showing expression levels of proteins involved in Escherichia coli central metabolic pathways.

3. C, D

   Time profiles of abundance of selected proteins that are significantly correlated with the persister phenotype in both PCA (i.e. proteins for which the correlation coefficient had P < 0.1). Abundance relative to cells growing on glucose (C) or relative to starved cells (D). Error bars indicate one standard deviation reflecting variation between technical replicates.

Source data are available online for this figure.

Figure 6. Persister and starved cells have depleted metabolite pools

Change in metabolite concentrations in persister, starved, and fumarate‐growing cells relative to glucose‐growing cells. For absolute concentrations with error estimates and numbers of replicates, see Table EV2.

Figure 7. Persistence is sustained through a system‐level feedback loop

1. A metabolic perturbation beyond the intrinsic buffering capacity of metabolism, which results in low metabolic flux, is the trigger for persistence. Cells with critically low metabolic fluxes get into a vicious cycle (feedback) and thus cannot restore metabolic homeostasis. The robustness of this primitive, system‐level feedback loop can be enhanced via various mechanisms, such as action of TAS, σ^S, or ppGpp, which lead to further inhibition of transcription and translation. The system‐level feedback loop is active until the vicious cycle is broken through restoration of metabolic homeostasis, for example, by addition of certain nutrients, stochastically higher expression of certain flux‐limiting enzymes or stochastically low expression of growth‐inhibiting mechanisms.

2. Fraction of growing cells (i.e. 1 − fraction of persister cells) in various knockout strains after a glucose‐to‐fumarate nutrient shift. White bars: BW25113‐derived strains. Gray bars: MG1655‐derived strains. Δ10: strain with 10 TAS knockout. Statistically significant difference (t‐test or Wilcoxon rank sum test for Δrmf, P‐value < 0.05) between the mutants and the respective wild type is indicated with an asterisk (*). Mean of at least three replicates and standard deviations are shown. See Appendix Table S6 for antibiotic tolerance assay results.

3. Fraction of growing cells after a glucose‐to‐fumarate shift decreases with higher induction of σ^S expression in ΔrpoS strain. Cells were induced with IPTG at the indicated concentrations after the nutrient shift.

4. Fraction of growing cells after a glucose‐to‐fumarate shift increases with higher induction of DctA fumarate transporter, and thus with higher metabolic flux, in the Δ10ΔrpoS strain. Cells were induced with the indicated IPTG concentrations for 16 h prior to the nutrient shift and after the nutrient shift. Mean of three replicates and standard deviations are shown.

5. Rapid increase in forward scatter and cell count upon addition of glucose (at 4 h after nutrient shift) indicates that persister cells can rapidly resume growth upon externally driven restoration of metabolic homeostasis. Forward scatter distribution from three replicates, cell count mean of three replicates, and standard deviations are shown.

Figure 8. Schematic model: Persistence and growth are two attractor states on a phenotypic landscape with the dimensions “metabolic flux” and “activity of growth‐inhibiting mechanisms”

The blue circle denotes the normal growth state, the red circle denotes the persister state, and the gray disk indicates the position of a cell directly after a perturbation. The magnitude of metabolic flux and the activity of growth‐inhibiting mechanisms determine a cell's position on the landscape. If a cell is on the right side of the watershed (i.e. the hill/dotted line), it will move toward the attractor indicated by the blue disk and achieve normal growth in metabolic homeostasis. If a cell happens to be on the left side of the watershed, it will become a persister cell. Both states are achieved through active mechanisms (that eventually also require resources/energy), as indicated by the finding that the persister state is not equal to the starved state. Various perturbations that were found to cause persistence (for instance, stochastic TAS induction, nutrient shift, or diauxie) move the cell on the landscape in different directions, but all of them push it from the state of metabolic homeostasis beyond the watershed.