Epigenetic cellular memory in Pseudomonas aeruginosa generates phenotypic variation in response to host environments

Significance

Traditional views consider bacteria as solitary single-cell organisms that can grow into clonal populations. However, recent research reveals complex levels of cooperation among bacterial cells. Within a bacterial community, cells can exhibit significant specialization rather than being homogeneous, thus conferring novel functionalities to the population that go beyond individual capabilities. In this study, we uncovered a regulatory system that controls the heterogeneous expression of a glycerol-metabolizing gene (glpD) in Pseudomonas aeruginosa populations. We demonstrate that differential glpD expression levels are associated with varying pathogenicity-associated phenotypes and propose that unique emergent collaborative behaviors enable P. aeruginosa populations to overcome their unicellular limitations and develop into an opportunistic pathogen.

Abstract

Phenotypic diversification within pathogen populations can enhance survival in stressful environments, broaden niche colonization, and expand the ecological range of infectious diseases due to emerging collective pathogenicity characteristics. We describe a gene regulatory network property in the opportunistic pathogen Pseudomonas aeruginosa that generates diversity of gene expression and pathogenicity behavior at the single-cell level and that is stabilized by epigenetic cellular memory. The resulting heterogeneity in the expression of the glpD gene—an indicator of host-derived glycerol metabolism and intra-host presence—shapes adaptive processes that are subject to natural selection. Our work on how epigenetics generates phenotypic variation in response to the environment and how these changes are inherited to the next generation provides insights into phenotypic diversity and the emergence of unique functionalities at higher levels of organization. These could be crucial for controlling infectious disease outcomes.

Clonal bacterial populations display a considerable level of phenotypic variability, referring to differences in physical and biological characteristics, despite growing in similar environmental conditions. This variability can result from adaptation to small variations in the microenvironment, as well as from mutations or reversible genetic changes in certain subgroups of individual cells within the population. However, phenotypic diversification unrelated to environmental conditions or DNA changes can also occur. Alterations in gene activity among otherwise identical bacteria within a single population may become apparent when a fluctuating gene expression signal is amplified by regulatory motifs (1–4). Consequently, an intermediate gene expression level can be transformed into a binary response, resulting in individuality or phenotypic variation among microbial cells (5, 6). In a bistable bacterial population, cells exhibit either high or low expression levels of specific genes. Moreover, cells can transmit their cellular state to the next generation through a mechanism called epigenetic inheritance (7).

Numerous documented instances of phenotypic variability are associated with responses to environmental stresses. Generating variable phenotypes appears advantageous for population survival during challenging conditions (8, 9). The establishment of a bistable population may represent a strategy of bet-hedging or risk-spreading (10) offering an alternative approach to cope with stressful conditions, as certain subgroups of bacteria may already be adapted to future environments (11). Alternatively, the diversification of phenotypes within a population could facilitate the colonization of broad niches by covering a wider range of possible responses to different environmental conditions.

Pathogenic bacteria face numerous obstacles in their quest to establish themselves within the human host. Preadaptation of certain subpopulations to adverse conditions could enhance their ability to invade humans, evade attacks by the immune system, and initiate infection. In two prominent examples, the importance of bacterial bistability for the establishment and maintenance of bacterial infections has been demonstrated. The expression of the type III secretion system (T3SS) has been described in Salmonella enterica serovar Typhimurium and enteropathogenic Escherichia coli to confer a fitness advantage to the pathogen within the human host (12). Furthermore, the formation of transiently antibiotic-tolerant subpopulations was shown to facilitate bacterial persistence (13).

Due to its remarkable adaptability across diverse environments, Pseudomonas aeruginosa is frequently employed as a model organism to investigate bacterial gene regulation and survival strategies in challenging habitats (14–17). This ubiquitous environmental bacterium is also notorious as an opportunistic pathogen, often displaying resistance to antibiotics. In this study, we have identified variable expression of glpD, a gene involved in glycerol metabolism, across different P. aeruginosa populations that were cultivated under the same environmental conditions. Interestingly, we observed that the expression of glpD was correlated with the expression of pathogenicity traits at the single-cell level. The occurrence of different subpopulations exhibiting distinct virulence traits that are linked to variant glpD gene expression may confer additional functionalities to this opportunistic pathogen, which could play a significant role in P. aeruginosa’s ability to survive and initiate infections within the human host.

Results

Variability of Gene Expression across P. aeruginosa Populations Grown under the Same Environmental Conditions.

To assess the extent of variation in individual P. aeruginosa gene expression values across multiple transcriptional profiles, we recorded replicate transcriptomes of overall 167 individual transposon (tn)-mutants retrieved from the Harvard mutant library (18) (Fig. 1A). These tn-mutants were cultured in LB medium until reaching an OD₆₀₀ of 2. A total of 337 transcriptomes, which had a minimum sequencing depth of one million reads mapping to protein-coding sequences and a minimum correlation of 0.9 between replicates, were included in this study (SI Appendix, Fig. S1 A and B). This sequencing depth was sufficient to capture the expression of the majority of the 5,905 analyzed P. aeruginosa PA14 genes (SI Appendix, Fig. S1C), for which we calculated the biological coefficient of variation (BCV) of gene expression.

Fig. 1.

Transcriptional variability of transposon mutant replicate transcriptome profiles. 337 transcriptional profiles of P. aeruginosa PA14 transposon mutants harboring transposon insertions in overall 167 different gene loci were included in the analysis. We determined BCV of individual gene expression across the replicates. (A) The BCV of the individual PA14 genes sorted by their BCV value is depicted. Selected genes are marked in red. The insert shows the BCV values as a function of the average gene expression values (log₂ counts per million). The red circle marks the BCV of the glpD gene. (B) BCV values of each gene were assigned to their corresponding PseudoCAP categories. The boxplots depict the biological variation of each category and are sorted from the lowest variation (Left) to the highest variation (Right). Solid black lines indicate the median value. Abbreviations are as on Pseudomonas.com (C) Bee-swarm plot depicting log₂(CPM) expression values of selected genes across the 337 tn-mutant transcriptional profiles and (D) across the previously recorded transcriptional profiles of 414 clinical isolates (19). Median and mean ± SD gene expression levels are depicted in orange and red, respectively.

In general, genes with low expression levels [log2(CPM) < 3] exhibited higher variability across the replicate populations compared to the average. However, most genes with expression levels of log2(CPM) > 3 maintained a stable expression, resulting in an overall low BCV (Fig. 1A). Notably, genes encoding proteins belonging to functional PseudoCAP (20) categories involved in fundamental cellular processes, along with chemotaxis genes, displayed the most stable expression patterns. Conversely, genes that are responsive to environmental changes exhibited more variable expression values (Fig. 1B).

Despite the generally low BCV among replicate P. aeruginosa cultures, we identified 21 genes [at log2(CPM) > 3] that displayed remarkably high BCV with values above 0.6, with an additional 58 genes exhibiting BCV values above 0.5 (Fig. 1A and Dataset S1). The gene with the highest level of transcriptional variation was glpD, boasting a BCV of 0.93, while maintaining a median gene expression level of log2(CPM) at 6.48. Two other genes in the glp locus, glpK and glpF, also demonstrated elevated BCV values of 0.55 and 0.43, respectively, with median expression levels of log2(CPM) at 6.25 and 7.27. The glpD gene encodes a glycerol-3-phosphate dehydrogenase involved in the oxidization of sn-glycerol-3-phosphate (G3P) to DHAP (dihydroxyacetone phosphate), which is shuttled into downstream metabolic reactions. A single repressor encoded by glpR, which maps adjacent to glpD, negatively regulates glpD and the glpFK operon. The latter encodes a membrane diffusion facilitator for glycerol (GlpF) and a cytoplasmic glycerol kinase (GlpK), involved in the conversion of glycerol to G3P. G3P binding inhibits GlpR activity, thus derepressing glycerol metabolism genes when a certain level of glycerol is available (SI Appendix, Fig. S2).

Aside from the glp gene locus, genes associated with denitrification pathways such as those found in the nar, nir, and nor operons, as well as chaperone-encoding genes (dnaKJ, groEL, and hslU), exhibited some of the highest BCV values. A correlation analysis further revealed a strong coregulation among glpD and glpK, while the correlation between the expression of glp genes and those encoding the denitrification pathway or chaperones was comparatively weaker (SI Appendix, Fig. S3).

Fig. 1C presents a beeswarm plot illustrating the distribution of gene expression variability in the tn-mutant dataset for the genes within the glp operon, dnaK, and three control genes (rpsG, uvrA, and uvrB). Notably, the control genes display a relatively uniform distribution of gene expression values. In contrast, glpD, in particular, exhibits a distinct pattern with high or low expression levels across the various P. aeruginosa tn-mutant cultures. To further investigate this, we conducted a similar analysis on a previously published dataset comprising transcriptome profiles of 414 nonisogenic clinical P. aeruginosa isolates (19) (Fig. 1D). Interestingly, the signal for glpD, glpK, and glpF displayed an even wider distribution, suggesting the existence of distinct states of gene expression levels within the different transcriptomes, despite all the isolates being cultivated under the same environmental condition (LB medium, OD₆₀₀ of 2).

Of note, we observed a high correlation between the expression of the glpT gene, located separately on the chromosome from the glp operon, and the expression of glpDKF within the clinical isolate dataset (SI Appendix, Fig. S4A). A shared DNA binding motif was identified upstream of glpT, glpD, and glpK, indicating that the DNA-binding regulator GlpR may bind not only to the promoter regions of glpD and glpK but also to that of glpT (SI Appendix, Fig. S4B). This finding suggests a coregulation of the glpDKF genes with glpT within different populations of clinical P. aeruginosa isolates.

P_glpD Promoter Heterogeneity within Populations of Isogenic Cells.

We proceeded to investigate the heterogeneity of glpD and glpK promoter activity at the single-cell level within a population. To achieve this, we directly fused the natural PA14 promoter P_glpD and P_glpK to the green fluorescent protein encoding gene (gfp), cloned the construct into the pSEVA237 vector (21) and introduced it into PA14. Subsequently, we used flow cytometry to record the promoter activities from a total of 20,000 P. aeruginosa PA14 cells in a minimum of three independent bacterial populations cultured in LB to exponential (OD₆₀₀ ~ 0.6) and transition (OD₆₀₀ ~ 1.5) growth phases. As controls, we also examined the promoter-gfp fusion reporters from genes with lower variability in expression, namely glpR, dnaK, rpsG, uvrA, and uvrB.

We observed a higher coefficient of variation in the promoter activity of glpD and to a lesser extent of glpK, glpR, and dnaK, across isogenic populations of P. aeruginosa cells as compared to that of the control genes rpsG, uvrA, and uvrB (Fig. 2 A and B).

Fig. 2.

Transcriptional variability of genes with high or low BCVs. The natural PA14 promoter of selected genes was fused to gfp and introduced into PA14 on a pSEVA237 reporter plasmid. 20,000 bacterial cells of each population were subjected to flow cytometry analysis. Density plots showing GFP signal log₁₀ (arbitrary units, a.u.) intensities for at least three independent cultures depict promoter activities in the exponential and transition phases. The white line indicates median values. The promotor activities of the glpD, glpK, glpR, and dnaK genes that were identified to exhibit elevated BCV values in RNA-sequencing experiments (Fig. 1) is shown in (A), while the promotor activity of control genes that were identified to exhibit low BCV values is shown in (B).

Dynamics of P_glpD Promoter Activity on the Single-Cell Level.

To further investigate whether glpD expression heterogeneity within a population is static or whether cells dynamically switch glpD expression on and off, we conducted time-lapse microscopy for a duration of up to 5 h, capturing snapshots every 3 to 5 min (Fig. 3 A and B). A control strain that constitutively expressed the stable GFP variant exhibited a strong signal throughout the entire period (SI Appendix, Fig. S5). FACS analysis of control cells expressing the stable GFP version revealed that approximately 83.60% ± 23.02 of cells were fluorescent. To enable real-time monitoring of gene expression, we fused the glpD promoter sequence with the unstable GFP variant GFP(LVA) (22, 23). A much smaller fraction of cells (1.48% ± 3.34) expressing the unstable GFP(LVA) under the control of the P_glpD promoter were detected.

Fig. 3.

Time-lapse microscopy of glpD promoter activity. The glpD promoter was fused to an unstable GFP variant [GFP(LVA)] to ensure an unbiased signal and to observe the spatiotemporal transcription of glpD. (A and B) Multiple time-lapse experiments were carried out in LB medium. Selected shots recorded during a time course of 2.5 h are displayed. White arrows mark cells that lose their glpD activity after dividing from their glpD active mother cell (A) and a cell that spontaneously activates glpD (B). Time stamp and 5 µm scale are shown at the Bottom of each picture. (C–G) Maximum fluorescence intensity, which correlates to glpD promoter activity, is plotted for individual cells as a function of time (hours). Each line depicts the fate of a single cell. Gray dots (C and D) mark cell division events. (C) and (F) represent the same experiment as shown in (A) and (B), respectively.

As depicted in Fig. 3, the limited number of cells exhibiting an active glpD signal maintained this signal over an extended period. Interestingly, following cell division, the mother cell typically retained high expression, while the daughter cell lost signal activity, suggesting asymmetric inheritance (Fig. 3 B and C–E). We also observed a small fraction of cells spontaneously gaining and losing P_glpD activity (Fig. 3 F and G).

To further quantify this asymmetric division, we performed live-cell imaging on the population grown on agar pads, starting from single cells. Among glpD-positive cells (ON, n = 278), in more than one-quarter of instances (n = 72), the mother cells maintained the signal for up to five divisions, whereas the daughter cell rapidly lost the signal (Movie S1). Additionally, some cells retained glpD expression for a prolonged period without dividing (n = 11), while others lost expression before resuming growth (n = 195, see Movie S2). In populations initiated from single glpD-negative cells (OFF, n = 2,599), we observed stochastic expression of glpD in 5.7% of the cases (n = 148). Of these, the expression was short-lived in 55% of the cases (n = 81), whereas in 45% of the cases (n = 67), it led to the asymmetric division pattern previously described (Movie S3). The heterogeneity of glpD-expressing cells is summarized in SI Appendix, Fig. S6.

Heterogeneity of glpD Expression across Different P. aeruginosa Cultures Is Dependent on the Activity of GlpR and the Culture Inoculum.

We next measured the activity of the glpR, glpK, and glpD promoter in M9 minimal medium supplemented with either glucose or glycerol as the sole carbon sources (Fig. 4A). As expected, when glycerol was the sole carbon source, the P_glpD activity shifted toward higher values, while glucose as the sole carbon source resulted in low P_glpD activity. A similar pattern as for the glpD promoter was observed for the glpK and the glpR promoter. Consistent with the finding that GlpR represses glpD and glpK, inactivation of GlpR led to high expression of glpD across different P. aeruginosa cultures (Fig. 4B).

Fig. 4.

Glp operon promoter activities measured via flow cytometry in defined carbon sources. (A) Activity of the glpR, glpK1F, and glpD promoter in exponential phase in minimal media supplemented either with 20 mM glucose (blue) or 40 mM glycerol (red). GFP fluorescence intensities of the respective promoter-gfp fusion reporters are displayed in log10 (a.u.) in biological duplicates. White lines indicate the median values. (B) glpD expression was monitored in LB medium for four independent cultures using a P_glpD-gfp fusion reporter in the PA14 wild-type (red) and a glpR deletion mutant (blue). (C) Activity of the glpD promoter in 60 individual LB-exponential-phase-grown PA14 cultures inoculated with high (500) or low (100) amounts of bacteria. (D) The activity of the glpD promoter in 60 individual LB-exponential-phase-grown PA14 cultures, which were inoculated with low (Left) amounts of bacteria, was recorded. From those 60 exponentially grown populations, 10 cells were then used to inoculate a new set of 60 PA14 cultures and the glpD promoter activity was remeasured in exponential growth. The pair-wise median GFP intensity of the first cultures was plotted against the median GFP intensity of the second culture and linear regression was applied. The statistical parameters indicate that 57% of the expression of glpD in the second culture can be attributed to its expression in the first culture.

We furthermore observed that the coefficient of variation of median glpD expression in independent populations of P. aeruginosa was influenced by the inoculation density. As illustrated in Fig. 4C, variations in glpD promoter activity were more diverse in populations inoculated with a low number of cells (100 cells). When high amounts of bacteria (500 cells) were used for the inoculation, the variations of glpD expression values between different PA14 cultures were almost completely diminished.

We next aimed to assess whether an inheritability of glpD expression levels is important in stabilizing this inoculation bias. We therefore examined the diversity of glpD promoter activity in 60 parallel cultures of P. aeruginosa that were initially inoculated with 10 cells. We then diluted the first cultures and grew them in second cultures until the exponential phase before measuring the glpD expression with flow cytometry again. Notably, as shown in Fig. 4D, the median and overall distribution of glpD activity in the second culture closely aligns with that of the first culture. When comparing the median fluorescence intensities between the two cultures through linear regression analysis, it becomes apparent that 57% of the expression of glpD in the second culture can be attributed to its expression in the first culture. This implies that the heterogeneous expression of glpD within clonal populations is an inheritable trait.

Since GlpR activity is inhibited by G3P, we conducted experiments where G3P was introduced to the medium and investigated the influence of exogenous G3P addition on glpD promoter heterogeneity. Indeed, the overall level as well as the heterogeneity in glpD promoter activity was strongly reduced when G3P was added (SI Appendix, Fig. S7).

Impact of the Activity of the glp Genes on PA14 Growth.

We proceeded to examine the growth behavior of P. aeruginosa strains with defects in genes associated with glycerol metabolism. To achieve this, we generated clean deletion mutants of glpR, glpD, and glpK. As anticipated, when these mutants were grown with glucose as the sole carbon source, their growth was comparable to that of the PA14 wild-type strain. Neither ∆glpD nor ∆glpK were able to grow in M9 minimal medium with glycerol as the sole carbon source, but both mutant strains grew in medium supplemented with glucose and glycerol (SI Appendix, Fig. S8). Notably, the growth of the ∆glpD mutant was more severely affected compared to ∆glpK. Consistent with the functional activity of GlpK, which phosphorylates glycerol to produce G3P, the substrate for GlpD (24), the growth defect of the ∆glpK mutant could be fully restored by the addition of exogenous G3P (SI Appendix, Fig. S8). Interestingly, the ∆glpD mutant was the only strain that exhibited impaired growth in LB medium, exhibiting a doubling time of 35.6 ± 0.69 min as compared to 29.2 ± 0.34 min of PA14. This correlated with the observation that the ∆glpD mutant formed smaller colonies on LB plates compared to PA14 or the ∆glpK and ∆glpR mutants. The ability of the ∆glpK mutant to grow in LB suggested the presence of G3P in the medium. It is worth noting that the ∆glpR mutant, which overexpresses glpD and glpK, initiated growth slightly earlier in glycerol-containing medium compared to the wild-type strain. However, this effect was considerably less pronounced compared to previous observations in Pseudomonas putida (25).

Stochastic Simulation.

Transcriptional activity in a single cell is intrinsically stochastic due to the low number of the key players involved in the processes (26). Therefore, to gain a qualitative understanding of the heterogeneity in glpD promoter expression when P. aeruginosa is cultivated in a medium with intermediate glycerol concentrations (LB medium) we used a stochastic algorithm (27) and developed a mathematical model that incorporates relevant reaction kinetics. Our model considers GlpR-dependent repression of glpD and glpK, which occurs when GlpR binds to the DNA binding motif upstream of the glpD and glpK promoter and takes into consideration a simplified representation of the combined impact of glycerol internalization and its conversion into G3P. All reactions are listed in Dataset S2 and are assumed to follow mass action kinetics. We considered increasing glycerol concentration in the medium and qualitatively captured experimental observations, such as the stable low/high activity of the glpD promoter when grown in glucose and glycerol medium, respectively (SI Appendix, Fig. S9). Importantly, the model qualitatively reproduced the diversity in glpD expression at intermediate glycerol levels (LB medium) reflected by intermediate glpD and glpK transcriptional activity.

Agent-Based Simulation.

We next developed an agent-based framework in which each agent incorporates the aforementioned reaction framework, initialized in a state closely resembling the steady configuration of the LB medium. In our model, we assigned a random probability of switching from low to high (or vice versa) glpD state at the time of cell division. This reflects our experimental observation of a spontaneous gain of P_glpD activity (Fig. 3 F and G). To incorporate the effect of memory, we considered that one of the daughter cells will maintain the same growth and metabolic rate as it was for the mother cell and the other daughter cell may or may not maintain the same growth and metabolic rate and will be decided upon a coin toss. Therefore, some cells will maintain their glpD activity, reflecting the effect of memory. For additional information regarding the construction of the agent-based model refer to the Dataset S2.

With this framework in place, we grew populations starting with either 20 or 100 cells with the initial conditions similar to the LB medium (intermediate transcription of glpD/glpK) to mimic the experimental low and high inoculum scenario (Fig. 4C). We developed 50 populations until they reached a size of approximately 20,000 cells. Our initial analysis focused on examining the potential impact of switching and memory on inoculum density-associated bias. We found that inoculum density-associated bias diminishes over time, resulting in identical population distributions for both low and high inoculum cases. Therefore, this initial bias cannot explain the experimentally observed heterogeneity in the low inoculum case. When considering the switching effect, there is an overall mean shift in glpD expression within the population for both low and high inoculum case, but no statistical difference in population heterogeneity between low and high inoculum cases. However, when considering the memory effect, we observed greater heterogeneity in glpD-producing cells in the low inoculum case compared to the high inoculum case (SI Appendix, Fig. S9 A and B). It is interesting to note that the mean glpD expression remains the same in these two conditions.

Cells lacking glpD grow significantly slower in LB medium (SI Appendix, Fig. S8). To explore whether a disparity in growth rates has an additional impact on glpD expression heterogeneity within P. aeruginosa populations, we implemented an effect of glpD expression on the cell cycle duration into our model. High glpD-producing cells were assigned a shorter cell cycle duration (~20 min), while the other cells had longer cycle durations ranging from 30 to 40 min, as before. Indeed, we observed that variations in glpD expression became more evident when considering differences in growth between high and low glpD producers. Our results suggest that growth differences in combination with an inoculation bias due to cellular memory of glpD expression may contribute to the maximization of glpD expression diversity as observed in experimentally grown P. aeruginosa populations (SI Appendix, Fig. S9 A and B). Prior research has also underscored the significance of growth regulations across diverse contexts (28–33).

We verified our model by simulating conditions with exogenous G3P supplementation and compared the results to our experimental observations. Exogenous supply of G3P results in a condition highly similar to that induced by glycerol, as G3P binds to GlpR leading to an enhanced glpD expression. Furthermore, as G3P is a substrate of GlpD, cells will analogously produce GlpD to consume excess G3P and maintain its homeostasis. To model this, we implemented a sigmoidal function $\left(k_{\mathit{adj}}^{\mathit{trd}}=k_{\mathit{LB}}^{\mathit{trd}}+\frac{G3P}{k+G3P};k=100\right)$; to model the potential adjustment of glpD levels based on the intracellular G3P concentration. After implementing these adjustments, we cultivated 30 populations until they attained a size of around 20,000 cells. The outcomes are presented in SI Appendix, Fig. S11 A and B. Our model accurately replicates the features of the experimental distribution of glpD following the exogenous addition of G3P, which is characterized by high glpD states with lower heterogeneity.

Effect of glpD Expression on the Global Transcriptome of P. aeruginosa.

Our next objective was to assess whether the activity of the glp operon has broader implications on the global transcriptional profile beyond its role in regulating glycerol-metabolizing enzymes. The analysis of gene expression profiles of PA14 grown in M9 medium supplemented with glycerol and glucose, respectively, revealed the expected high expression of the glp genes in glycerol medium. However, especially during exponential growth, we found many more genes that were up-regulated (667) and down-regulated (230) under rich glucose versus glycerol conditions (Dataset S3). To identify genes exclusively regulated by glpD activity, we conducted a comparison between the gene expression profiles of a strain lacking glpD (ΔglpD) and a strain overexpressing glpD due to the absence of its transcriptional repressor (ΔglpR). We found 431 differentially regulated genes during exponential growth in LB and 237 differentially regulated genes in the transition phase (SI Appendix, Fig. S12A). As has been observed previously (34), deletion of glpD led to reduced transcription of central energy metabolism genes such as aceE and aceF in the exponential and transition phases (Dataset S4). Furthermore, we found a significantly decreased expression of biotin synthesis genes during the exponential phase (SI Appendix, Fig. S12B).

In contrast, genes involved in phenazine biosynthesis and sulfur metabolism were up-regulated during the exponential phase in the glpD deletion mutant. The same was observed for genes involved in pathways involved in amino acid metabolism. This included the biosynthesis of hydrophobic amino acids such as valine, leucine, isoleucine, and the aromatic amino acids in the exponential growth phase, as well as glycine, serine, and threonine in the transition phase (SI Appendix, Fig. S12C). Genes involved in biofilm formation, bacterial secretion, and energy metabolism were also upregulated in ΔglpD. Interestingly, the nir, nor, and nos genes, which were shown to be coregulated with glpD in the transcriptional profiles of the tn-mutants (SI Appendix, Fig. S3) were also down-regulated in the ΔglpD mutant.

Link of glp Operon Activity with P. aeruginosa Phenotypic Traits in Populations.

We further examined relevant phenotypic traits of the PA14 ∆glpD, ∆glpK, and ∆glpR deletion mutants in comparison to the wild-type strain in whole populations. The ∆glpR mutant exhibited significantly reduced swimming and swarming motility, whereas ∆glpD and ∆glpK mutants did not show significantly altered motility (Fig. 5 A and B). Interestingly, overcomplementation of ∆glpD with glpD expressed on pSEVA551 resulted in impaired swimming and swarming motility, comparable to that of ∆glpR. This indicates that the reduced motility of the ΔglpR mutant might be caused by the increased expression of glpD. We furthermore observed a noticeable reduction in pyocyanin production after 18 h of growth in LB medium in the PA14 ∆glpD mutant (Fig. 5C), whereas high glpD expression in the complemented ∆glpD mutant revealed a higher pyocyanin production when grown in M9 medium with glucose as the sole carbon source (Fig. 5D). We next infected Galleria mellonella larvae with the glp gene deletion mutants. While the ∆glpK and ∆glpR mutants killed the larvae to the same extent as the PA14 WT, the ΔglpD mutant showed significant attenuation (Fig. 5E and SI Appendix, Fig. S13).

Fig. 5.

Phenotypic characterization of glp gene deletion mutants at the population level. (A) Swimming and (B) swarming motility measured on semisolid agar plates after 16 h. Pyocyanin production was measured following 18 h of growth in (C) LB-Medium and in (D) M9 medium supplemented with 20 mM glucose. Pyocyanin production of the glp mutants as well as of the ΔglpD strain complemented with glpD in trans (ΔglpD+) is shown relative to that of PA14 without (WT) or with (WT+) the empty vector control. Median and interquartile range of 4 to 5 replicates are plotted. Significant changes (*P ≤ 0.05; ***P ≤ 0.001, two-sided t test) as compared to the corresponding wild type are marked with asterisks. (E) Killing of G. mellonella larvae by the glp gene deletion mutants was assessed in five independent experiments with 10 larvae each. The fraction of killed larvae (in %) 24 h postinfection is depicted. Significant differences are marked with an asterisk (two-tailed t test *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001). (F) Phagocytic uptake 2 hpi by RAW 264.7 cells using a MOI of 1. Mean ± SD of four independent experiments with three technical replicates each are depicted, expressed as a ratio of CFU (Phagocyted/Inoculum) and normalized to WT values (****P < 0.0001 in one-way ANOVA with post hoc test: Dunnett). (G) Lactate Dehydrogenase (LDH) assay of infected RAW264.7 macrophages 6 h pi. PBS served as a negative control. Values expressed as a percentage of the WT LDH activity (***P <0.001, ****P < 0.0001 in one-way ANOVA with post hoc test: Dunnett). (H) Bacterial survival 6 h after infection in RAW 264.7 cells using a MOI of 1. Mean ± SD of four independent experiments with three technical replicates each are depicted, expressed as a ratio of CFU (6 h/2 h) and normalized to the phagocyted CFU recovered (*P < 0.05 in one-way ANOVA with post hoc test: Dunnett).

To further decipher the role of glycerol metabolism during interaction with the eukaryotic host, we used a RAW 264.7 murine macrophage-like cell line which we infected with a multiplicity of infection of one (MOI) and performed gentamycin protection assays. Uptake of a ΔglpD mutant was significantly reduced in comparison to the parental WT strain (Fig. 5F), the strain showed increased cytotoxicity (Fig. 5G), while exhibiting increased survival in macrophages after phagocytosis (Fig. 5H).

Phenotypic Impact of Glycerol Metabolism in Single Cells.

Given the observed heterogeneity in glpD expression within PA14 WT populations, we sought to investigate whether glpD expression levels could lead to distinct phenotypes at the individual cell level. As we observed a decreased swimming in the whole population on agar plates (Fig. 5A), we quantified the swimming velocity of individual cells grown in liquid medium, either LB or M9 with glucose as the sole carbon source using a microscope. Surprisingly, no changes in the swimming velocity of individual P. aeruginosa cells were observed (Fig. 6 A and B). Of note, we observed no different swimming velocity, regardless of the carbon source in the media (Fig. 6C).

Fig. 6.

Expression of glpD affects single-cell phenotypes. Swimming velocity of exponentially grown cells (3.5 h) measured by tracking individual cells’ speed. Each value is the mean velocity of at least 30 individual cells, acquired independently, in (A) LB media and in (B) M9 with 20 mM glucose as carbon source (*P <0.05 in one-way ANOVA with post hoc test: Dunnett). (C) Swimming speed of PA14 WT with varying carbon source. Each reported value represents the mean velocity of at least 29 individual cells in a minimum of two independent experiments. The swimming velocity was compared in M9 minimal medium with six different carbon source compositions (From left to right: 20 mM glucose, 40 mM glycerol, 40 mM G3P, 20 mM glucose + 40 mM glycerol, 20 mM glucose + 40 mM G3P, 10 mM glucose + 90 mM G3P). ANOVA with Tukey correction indicates no statistical difference between conditions. (D) SD Confocal microscopy of SiR-actin stained RAW 264.7 macrophages and PA14 WT bearing the pSEVA237-P_glpD-GFP reporter at a MOI of 10 after 10 min of infection. Image representative of four independent biological replicates. (E) GFP intensity of PA14 WT pSEVA237-P_glpD-GFP single bacterial cell (y axis) in the presence of macrophages stained with SiR-Actin. The local Log10 SiR-actin fluorescence for each cell is indicated in the x-axis. Two subpopulations emerge, indicative of cells not in contact with macrophages (left side of the y axis) and cells in contact with macrophages (right side of the y axis). (F) K-mean automated clustering allows to discriminate cells in contact with macrophages (Actin positive, red) or not (Actin negative, gray). (G) Mean GFP fluorescence of PA14 WT cells harboring the three transcriptional GFP promoter fusions (P_glpD-GFP, P_rpsG-GFP, and P_EM7-GFP) without (−) or with (+) contact with macrophages of at least three independent biological replicates. *P < 0.05 in unpaired two-tailed t test with Welch’s correction. ns = not significant.

Due to the variability of glpD expression observed in the WT strain (Fig. 1), we hypothesized that the variability of glpD expression could itself modulate interaction with eukaryotic cells and that in an otherwise isogenic population, cells with a higher glpD expression would show a stronger macrophage association. To test this, we infected RAW 264.7 macrophages for 10 min with our WT strain bearing the P_glpD transcriptional reporter. We included two control constructs: one in which the GFP is under the control of a synthetic, strong, constitutive promoter in the same plasmid backbone as the P_gplD reporter and a second in which the GFP is under the control of the P_rpsG promoter, which displays a low biological coefficient of variation (Fig. 2B).

The samples were fixed with 4% paraformaldehyde and macrophages stained with SiR-actin, a marker for cellular actin that allows discrimination of macrophages by spinning disk (SD) confocal fluorescence imaging (Fig. 6D). To quantify the intensity of the transcriptional reporter, single bacterial cells were segmented using Fiji (Materials and Methods), and for each cell, the GFP and SIR-actin signal was extracted. The latter was then used to determine whether bacterial cells were in contact with macrophages (SiR-actin positive or negative, Fig. 6E). Overall, we observed that the bacterial cells harboring the P_gplD transcriptional reporter indeed exhibited increased GFP intensity, indicating a higher glpD expression, when in contact with macrophages (Fig. 6F). When considered across independent biological replicates and normalized to the mean intensity of bacteria not in contact with macrophages, the average GFP intensity of P_gplD transcriptional reporter harboring bacterial cells interacting with macrophages was significantly higher, while the control transcriptional constructs did not show varying GFP levels upon cell contact (Fig. 6G).

Discussion

Regulated expression of genes plays a crucial role in determining the cellular phenotype and the ability of bacteria to adapt to harsh and rapidly changing environments. Each gene must be expressed at the right time and level to ensure an appropriate functional outcome (35). In this context, the regulation and expression of genes encoding basic cellular functions must be robust and not disturbed by variations in the environment, whereas the expression of genes that play an important role in the physiological response to external influences and stress factors must follow changes in environmental conditions.

Robustness can be achieved through the coordinated effort of several regulatory factors that form gene regulatory networks, which ensure patterns that withstand perturbations. However, it has also been shown that the design of regulatory networks can lead to multistability in gene expression, thus generating phenotypic variation among a clonal population. This offers an interesting alternative strategy to adaptive gene regulation to cope with stressful situations, as the generation of phenotypic heterogeneity can endow organisms with new functions (36, 37).

In this study, we explored the heterogeneous expression of glpD, a key player in glycerol metabolism, within populations of P. aeruginosa. Our analysis revealed that individual cells exhibited varying levels of glpD transcription compared to their neighboring cells, and these differences persisted over extended periods. The regulation of glpD transcription is a complex process involving the inhibitor GlpR, which in turn is inhibited by G3P, the substrate of GlpD. When GlpD levels rise, G3P levels decrease, leading to an increase in active GlpR that is not bound to G3P. Consequently, GlpR acts as a potent inhibitor of glpD transcription. Notably, activated GlpR also hampers the transcription of glpK, encoding a glycerol kinase involved in G3P production. As a result, both enzymes responsible for G3P synthesis and degradation (GlpD and GlpK) are negatively regulated by GlpR, which is itself influenced by G3P levels.

Our agent-based model confirmed that this complex regulatory system allows for the generation and maintenance of diverse glpD expression levels within clonal populations. This heterogeneity in glpD expression was shown to be maximized by a combination of an inoculation bias due to cellular memory of glpD expression with glpD-dependent differences in growth. The resulting notable differences in glpD levels across separate bacterial cultures could be crucial during infection initiation, as infections often start from small bacterial populations resulting from an initial bottleneck. Introducing a sampling bias in the bacterial populations, such as in those transitioning from colonization to bloodstream entry, could thus lead to increased diversity during the infection’s onset phase. This diversity may play a pivotal role in the pathogen’s success.

It was previously shown that glycerol catabolism is not only essential for growth on glycerol (32, 33) but also for virulence factor production in P. aeruginosa PAO1 (34). The glycerol dehydrogenase GlpD seems to affect virulence and persister formation also in other pathogenic bacteria such as Borrelia burgdorferi (38), Vibrio splendidus (39), and E. coli (40). In agreement, we observed increased pyocyanin production and greater virulence in the G. mellonella infection model in P. aeruginosa populations with higher glpD expression levels. Additionally, we demonstrate that glpD gene expression levels are correlated with increased interaction with host cells at the single-cell level. This suggests that host-derived glycerol may act as a recognition factor, triggering an appropriate adaptive response in individual cells.

While the GlpR-mediated regulation of glp genes was extensively studied in multiple organisms (25, 41, 42), recent publications found that another transcriptional regulator, ANR, regulates the expression of glpT in P. aeruginosa PAO1 (43). This transcription factor also regulates the expression of dissimilatory nitrate pathways (44), providing a possible explanation for the coregulation of the glp genes with genes involved in pathways of dissimilatory nitrate reduction.

Of note, we did not identify genes, which have previously been described to influence virulence, motility, attachment to the host cell, or pyocyanin production, to be transcriptionally regulated in a glpD-dependent manner. Instead, it seems that a shift in glycerol metabolism could play a more direct role in shaping virulence phenotypes possibly by altering bioenergetics or availability of precursors e.g., for pyocyanin production or cell surface structures. This could lead to the development of diverse virulence phenotypes, differing from the well-known regulatory mechanisms where environmental factors activate transcriptional regulators to control virulence gene expression.

In conclusion, our finding of glpD expression heterogeneity under conditions of intermediate glycerol levels suggests that the subsequent variant P. aeruginosa virulence phenotype might confer clonal P. aeruginosa populations with additional functionalities. The simultaneous presence of a virulent subpopulation, which triggers an inflammatory response, and a noninvasive subpopulation of Salmonella typhimurium, has been demonstrated previously to benefit the entire population through a division of labor (45–47). Given that infections often arise from the clonal expansion of small initial populations, the generation of P. aeruginosa populations exhibiting varying levels of glpD expression and associated phenotypes responsible for either expanding or maintaining the infection might be a strategy whose significance for the success of P. aeruginosa as an opportunistic pathogen has not been fully recognized.

Materials and Methods

Detailed information on the materials and methods used for bacterial culture, media, genetic engineering, primers, phenotypic characterization (motility, single-cell swimming speed, and measure of pyocyanin), and infection assay (G. mellonella and RAW 264.7 cells) can be found in SI Appendix. We also detail the methods used for RNA sequencing, flow cytometry, mathematic modeling, and microscopy approaches with the corresponding data analyses supporting codes.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (PDF)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (XLSX)

Dataset S06 (XLSX)

Movie S1.

Asymmetric inheritance of glpD expression. P. aeruginosa PA14 pSEVA237V-glpD immobilized under an agar pad were recorded every 10 minutes with brightfield and GFP fluorescence by spinning disk microscopy. The timelapse shows brightfield (left), GFP (middle) and merges of these channels (right) over a time period of 5 hours. Scale bar, 5 μm.

Movie S2.

Spontaneous loss of glpD expression. P. aeruginosa PA14 pSEVA237V-glpD immobilized under an agar pad were recorded every 10 minutes with brightfield and GFP fluorescence by spinning disk microscopy. The timelapse shows brightfield (left), GFP (middle) and merges of these channels (right) over a time period of 5 hours. Scale bar, 5 μm.

Movie S3.

Stochastic activation of glpD expression. P. aeruginosa PA14 pSEVA237V-glpD immobilized under an agar pad were recorded every 10 minutes with brightfield and GFP fluorescence by spinning disk microscopy. The timelapse shows brightfield (left), GFP (middle) and merges of these channels (right) over a time period of 5 hours. Scale bar, 5 μm.

Data, Materials, and Software Availability

The codes underlying the mathematical modeling are available at https://zenodo.org/records/14280669 (48) and https://github.com/arnabbandyopadhyay/transcriptional_noise (49). All study data are included in the article and/or in the supporting information.