Persister Escherichia coli Cells Have a Lower Intracellular pH than Susceptible Cells but Maintain Their pH in Response to Antibiotic Treatment

ABSTRACT Persister and viable but non-culturable (VBNC) cells are two clonal subpopulations that can survive multidrug exposure via a plethora of putative molecular mechanisms. Here, we combine microfluidics, time-lapse microscopy, and a plasmid-encoded fluorescent pH reporter to measure the dynamics of the intracellular pH of individual persister, VBNC, and susceptible Escherichia coli cells in response to ampicillin treatment. We found that even before antibiotic exposure, persisters have a lower intracellular pH than those of VBNC and susceptible cells. We then investigated the molecular mechanisms underlying the observed differential pH regulation in persister E. coli cells and found that this is linked to the activity of the enzyme tryptophanase, which is encoded by tnaA. In fact, in a ΔtnaA strain, we found no difference in intracellular pH between persister, VBNC, and susceptible E. coli cells. Whole-genome transcriptomic analysis revealed that, besides downregulating tryptophan metabolism, the ΔtnaA strain downregulated key pH homeostasis pathways, including the response to pH, oxidation reduction, and several carboxylic acid catabolism processes, compared to levels of expression in the parental strain. Our study sheds light on pH homeostasis, proving that the regulation of intracellular pH is not homogeneous within a clonal population, with a subset of cells displaying a differential pH regulation to perform dedicated functions, including survival after antibiotic treatment.

IMPORTANCE Persister and VBNC cells can phenotypically survive environmental stressors, such as antibiotic treatment, limitation of nutrients, and acid stress, and have been linked to chronic infections and antimicrobial resistance. It has recently been suggested that pH regulation might play a role in an organism's phenotypic survival to antibiotics; however, this hypothesis remains to be tested. Here, we demonstrate that even before antibiotic treatment, cells that will become persisters have a more acidic intracellular pH than clonal cells that will be either susceptible or VBNC upon antibiotic treatment. Moreover, after antibiotic treatment, persisters become more alkaline than VBNC and susceptible E. coli cells. This newly found phenotypic feature is remarkable because it distinguishes persister and VBNC cells that have often been thought to display the same dormant phenotype. We then show that this differential pH regulation is abolished in the absence of the enzyme tryptophanase via a major remodeling of bacterial metabolism and pH homeostasis. These new whole-genome transcriptome bacteria need to tightly control due to the sensitivity of their intracellular enzymatic reactions (47,49). In addition, metabolic reactions within the cytosol generate protons due to ATP production by glycolysis and oxidative phosphorylation. These processes cause an increase in acidity that the cell counteracts by expelling protons out of the cytosol via proton pumps and remodeling of key metabolic pathways (49)(50)(51). This helps maintain a stable electrochemical gradient across the cell membrane, which is vital for respiration (49). The cell regulates intracellular pH not only against internal pH variations but also in response to external stressors. For example, enteric bacteria, such as E. coli, must be able to survive in extreme pH ranges, as they need to pass through the stomach to reach a suitable environment for growth and infection (49). Shifting external pH values toward more extreme alkaline or acidic values induces the activation of rpoS, the SOS response, and heat shock-like responses, as well as cell-to-cell communication (52)(53)(54)(55).
Specifically, cell-to-cell signaling via indole, a ubiquitous signaling molecule produced by over 85 species of bacteria, has recently been linked to the regulation of intracellular pH at the population level (56)(57)(58). Indole is synthesized by the enzyme tryptophanase (TnaA), which catalyzes the reversible conversion of tryptophan to indole, ammonia, and pyruvate (59). Indole signaling is involved in a variety of biological processes, such as the regulation of the pathogenicity of cells (60), the tightening of epithelial cell junctions (61), biofilm formation (59,62), the activation of efflux pumps (44,45), responses to virulence (44,63), heat shock (101), and antibiotic stress (64). Two types of indole signaling have been identified: persistent indole signaling, when indole is present at a concentration of around 0.5 mM in the culture medium, and pulse indole signaling, occurring at the transition from the exponential to stationary phases with a maximum intracellular indole concentration of 50 mM, which can be mimicked experimentally by adding 5 mM indole to the culture medium (58). There is an ongoing debate regarding the role played by indole and tryptophan metabolism in antibiotic persistence, with some studies reporting that indole increases the fraction of persisters (65,66) and others reporting the opposite (42,43,67). It is conceivable that these opposing findings might be due to the fact that inhibiting tryptophan metabolism (by genetically knocking out tnaA in E. coli) might have other profound consequences on bacterial metabolism and pH homeostasis, besides indole production; however, this hypothesis remains to be tested.
In this paper, we set out to quantify the heterogeneity in intracellular pH within a clonal E. coli population, aiming to establish whether there is a link between the heterogeneity in intracellular pH and antibiotic survival, which ultimately can dictate the phenotypic structure within a clonal population. Furthermore, we set out to investigate whether the intracellular bacterial pH could be manipulated by interfering with tryptophan metabolism and to determine the genetic and metabolic consequences of the lack of tryptophanase. In order to achieve these aims, we use a microfluidics-microscopy approach (8,52) in combination with an intracellular pH reporter strain of E. coli to investigate the intracellular pH profiles of individual bacteria over time (8,52). This approach allows us to determine the differences in intracellular pH before, during, and after ampicillin treatment in persister, VBNC, and susceptible bacteria (8,52). Importantly, we considered bacteria to be VBNC if, after 3 h of antibiotic treatment and 21 h of exposure to antibiotic-free medium, they did not stain with propidium iodide, actively expressed fluorescent proteins, and did not divide within our experimental time frame. This does not exclude the possibility that these cells can start dividing if exposed to nutrients for longer periods of time (19,68), whereas we can exclude the possibility that these cells are dead (69) since they actively express fluorescent proteins. Next, we used a tryptophanase knockout mutant strain in combination with whole-genome transcriptome analysis and exogenous indole supplementation to investigate the role played by tryptophan metabolism as well as indole signaling on phenotypic survival after exposure to ampicillin and on the maintenance of intracellular pH in the different phenotypic responses to ampicillin. Our data show that cell-to-cell differences in the regulation of intracellular pH represents one of the possible bacterial strategies for escaping antibiotic treatment and therefore might open the way to future therapeutics to eradicate persister bacteria. Furthermore, this novel approach opens the way for investigating the role of the differential regulation of intracellular pH in individual cells on dynamic responses of microbial populations to external stressors.

RESULTS
E. coli persisters display a lower intracellular pH than those of VBNC and susceptible cells before ampicillin treatment. Using time-lapse microscopy, we measured the temporal dynamics of the fluorescence levels of a pH-sensitive green fluorescent protein (GFP) (70,71), pHluorin, for thousands of individual bacteria as a proxy for intracellular pH (see Materials and Methods). We also measured pHluorin-fused mCherry fluorescence to allow for normalization of any cell-to-cell heterogeneity in protein expression and plasmid copy number. It is worth noting that the fluorescent signals from pHluorin and mCherry were measured in series rather than in parallel by using two different filter sets that allowed for the excitation and the recording of the emission intensity of one of the two fluorophores at a time. This minimizes any artifacts due to potential fluorescence resonance energy transfer between the two fluorophores (72). These microscopy measurements were carried out using a microfluidic mother machine device (73) to image and track each individual bacterium before (time zero [t = 0 h]) ( Fig. 1a to c) and during (0 , t , 3 h) exposure to ampicillin at a concentration of 25Â the MIC, as well as during 21 h of exposure to lysogeny broth (LB) (3 , t , 24 h) ( Fig. 1d to f and Materials and Methods). The mother machine device has previously been employed to investigate bacterial physiology and response to drugs (2,9,11,(73)(74)(75), and it has been demonstrated that this approach does not introduce cellular stress since both the doubling time and the fraction of persisters measured in the mother machine are in accordance with corresponding measurements performed on cultures growing in flasks or agar plates (9,73). This single-cell approach allows us to determine the phenotype of each individual bacterium by flowing propidium iodide through the mother machine at 24 h, as previously reported. These phenotypes included persisters (leftmost channel in Fig. 1g), VBNC cells (second channel from the left in Fig. 1g), and susceptible bacteria (two rightmost channels in Fig. 1g) (9). The bacterium in the rightmost channel of Fig. 1g has lysed and disappeared from the channel after ampicillin treatment, whereas the bacterium in the second channel from the right was still visible in the channel but stained with propidium iodide and displayed significantly lower fluorescent protein expression than the VBNC cell; therefore, it was classed as susceptible ( Fig. 1f and 1i, Materials and Methods; see also Fig. S1 in the supplemental material) (9). Moreover, the persister progeny were readily killed when exposed to ampicillin a second time at a concentration of 25Â MIC at 24 h, confirming that persisters were phenotypically but not genetically resistant to antibiotics.
Crucially, at t = 0 h, before ampicillin exposure, VBNC and susceptible E. coli subpopulations exhibited a wide range of intracellular pH values, from pH 6.5 to pH 8 (the number of VBNC cells [n VBNC cells ] = 104, n susceptible cells = 2,770; 3 biological replicates), forming a bimodal distribution with peaks around 7.0 and 7.6 ( Fig. 2b and c, respectively). In contrast, the persister subpopulation had a tighter regulation of intracellular pH, with a single peak at pH 7.0 (n persister cells = 30 in 3 biological replicates) (Fig. 2a). This translated into a more acidic intracellular pH in persisters than in VBNC and susceptible bacteria, with mean values of 7.04 6 0.04, 7.26 6 0.03, and 7.28 6 0.01, respectively, before antibiotic treatment. This differential pH regulation between different bacterial phenotypes suggests that the intracellular pH might play a role in survival to drug exposure and thus deserves further investigation.
Abolishing tryptophanase activity eliminates the difference in intracellular pH between persister, VBNC, and susceptible E. coli cells before ampicillin treatment. We then tested whether tryptophanase activity and indole production, which have previously been linked to persistence (42,43,(65)(66)(67), play a role in the regulation of intracellular pH in E. coli by carrying out the set of experiments described above with a DtnaA knockout strain (56). In the DtnaA strain, the initial intracellular pH values of persister, susceptible, and VBNC cells before ampicillin treatment were significantly higher than the corresponding values measured for the parental E. coli strain (P , 0.0001), in accordance with previously reported measurements at the population level (56). Crucially, for all three phenotypes of the DtnaA strain, the intracellular pH distribution displayed a single peak at around 7.5 ( Fig. 2e and f); thus, abolishing tryptophanase activity and indole production eliminated the difference in intracellular pH between persister, VBNC, and susceptible cells before antibiotic treatment.
We next set out to determine whether supplementation with indole of a culture of the DtnaA strain would restore the persister parental-strain phenotype. We chose to perform chemical complementation with exogenous indole, instead of genetic complementation, in order to be able to contrast the effects between the loss of tryptophanase activity and the absence of indole signaling on the regulation of intracellular pH and on the phenotypic composition of the E. coli population. We mimicked persistent indole signaling by adding 0.5 mM extracellular indole when inoculating the DtnaA strain culture in LB or by progressively adding extracellular indole to the growing culture via a syringe pump to recapitulate the indole concentration produced by a (a to c) Bright-field, FITC, and TRITC fluorescence images of four representative bacteria injected in the microfluidic mother machine from a 2 ml aliquot that was prepared as follows. A 200 ml 17 h stationary phase E. coli BW25113 culture, with a typical OD 595 of 5, was spun down and resuspended in filtered medium (from the overnight culture) to be adjusted to an OD of 75. 2 ml were withdrawn from this sample and injected into the mother machine. (d to f) Corresponding images of the same bacteria after 3 h of incubation in ampicillin at a concentration of 25Â the MIC in an M9-LB (90:10, vol/vol) solution and a successive 21 h of incubation with fresh nutrients. (g to i) Corresponding images of the same bacteria after a 20 min incubation in propidium iodide (PI) to distinguish between susceptible bacteria (second channel from the right), in which PI addition increases the mCherry red fluorescence, and VBNC bacteria (third channel from the right), in which PI addition has no effect on mCherry red fluorescence (Fig. S1). The dynamics of intracellular pH for individual cells in each phenotype was measured by removing the background fluorescence, normalizing the pHluorin against the mCherry signal for each cell at each time point, and using a pH calibration standard (see Materials and Methods). Scale bar, 3 mm.
Unique Differential pH Regulation in Persister Cells ® growing parental-strain culture (56). We mimicked pulse indole signaling by adding 5 mM extracellular indole for 20 min directly before antibiotic treatment (see Materials and Methods) (56). Pulse indole signaling did not significantly affect the distribution of intracellular pH of the DtnaA strain, whereas both persistent indole signaling mimics returned similar intracellular pH distributions that were shifted toward alkaline values with respect to the DtnaA strain (Fig. S2). Therefore, we collated the experimental values gathered by using the former two approaches and found that the initial intracellular pH values of the persister, VBNC, and susceptible cells had alkalinized compared to that of the DtnaA mutant in the absence of indole ( Taken together, these results suggest that indole supplementation does not restore the intracellular pH profiles observed for persister, VBNC, and susceptible cells in the parental strain. In fact, the addition of indole exacerbates the mean alkalinity of all three phenotypes.
Genetic and metabolic consequences of the absence of tryptophanase. After demonstrating that restoring extracellular indole alone does not restore the intracellular pH regulation observed in the parental strain, we set out to determine which other pathways are affected by the lack of tryptophanase besides the well-known conversion of tryptophan to indole, pyruvate, and ammonia. In order to determine the consequences of the loss of tryptophanase on gene regulation in E. coli, we performed genome-wide comparative transcriptome analysis between the parental and the DtnaA strain by following our previously reported protocols for RNA extraction, sequencing, and Means of results from panels a to c and d to f are reproduced in panels g to i for comparison purposes only, showing that for all three phenotypes, the parental strain has the most acidic pH and the indole-supplemented DtnaA mutant has the most alkaline pH. Each strain was cultured at 37°C for 17 h in LB before injection into a microfluidic device for single-cell pH analysis and subsequent phenotype determination using fluorescence microscopy, as illustrated in Fig. 1 and described in Materials and Methods. For the parental strain, n persister cells was 30, n VBNC cells was 104, and n susceptible cells was 2,770. For the DtnaA strain, n persister cells was 97, n VBNC cells was 123, and n susceptible cells was 2,907. For the DtnaA strain with indole, n persister cells was 5, n VBNC cells was 222, and n susceptible cells was 4,729.
Goode et al.  76). Considering that tnaA expression is strongly upregulated at the transition from exponential to stationary phase (76), we performed RNA sequencing of biological triplicate samples after 3, 4, 5, and 17 h of growth in LB for both the parental and the DtnaA strain. At the whole-genome level, we found a weak correlation between the log 2 fold change in gene expression in the parental and the DtnaA strains between 3 and 4 h ( Fig. 3a) (R = 0.15), with the mutant displaying a significantly weaker regulation than the parental strain. Remarkably, this is the temporal window when tnaA is most strongly upregulated in the parental strain (6.1 6 1.1 log 2 fold change) (Data Set S1). In contrast, we found a stronger correlation between the log 2 fold change in gene expression in the parental and the DtnaA strain between 4 and 5 h and between 5 and 17 h ( Fig. 3b and c) (R = 0.23 and 0.44, respectively) when tnaA is downregulated in the parental strain (20.7 6 1.1 and 25.6 6 1.1, respectively). These data suggest a major remodeling of gene regulation in E. coli due to the absence of tryptophanase in the 3 to 4 h temporal window, a point on which we expand below.
In order to identify biological processes that were differentially regulated between the DtnaA and the parental strain, we clustered the transcriptomic data above, identifying 9 distinct patterns of gene regulation for the parental and DtnaA strains (top and bottom rows in Fig. 3d, respectively, and Data Set S1). Clusters 5 and 8 most closely resembled tnaA regulation in the parental strain (Data Set S1), containing genes with a significantly stronger upregulation at t = 4 h in the parental than in the DtnaA strain (3.1 versus 0.1 and 1.7 versus 0.4 mean log 2 fold changes, respectively; P , 0.0001 for both) (Fig. 3d). Gene ontology analysis revealed that cluster 8 was enriched for anaerobic respiration, oxidationreduction, carbohydrate transport, and several catabolic (including tnaA) processes (Fig. 4a). Unique Differential pH Regulation in Persister Cells ® It is worth noting that at the transition from exponential to stationary phase, the environment around E. coli BW25113 becomes more acidic, with the extracellular pH reaching a minimum of 6.2 at t = 4 h, followed by alkalinization of the extracellular environment, with a maximum pH of 7.2 at t = 17 h (76). Furthermore, cluster 8 was enriched for carboxylic acid catabolic processes (e.g., tnaA, ansB, garLR, pflB, and tdcB) ( Fig. 4a and Data Set S1) that at t = 4 h are upregulated in the parental but not in the DtnaA strain and that further lower the intracellular pH of the parental strain by producing organic acids, such as pyruvic acid (49).
In order to counteract this intracellular acidification, at t = 4 h, the parental but not the DtnaA strain upregulates oxidation-reduction and anaerobic respiration components, including cytochrome bd (cydAB), which contributes to the proton motive force, sulfate reductase (cysH), and formate (fdnGHI), glycerol (gldA), sorbitol (srlD), and NADH (wrbA) dehydrogenases, hydrogenases (hyaA and hybABO), and other proton-dependent catabolic processes, including lysine (cadA) and glutamate (gadA) decarboxylases (cluster 8 in Fig. 4a and Data Set S1). Several other proton-independent metabolic pathways are upregulated at t = 4 h in the parental but not in the DtnaA strain, including the fumarate (frdBCD), dimethyl sulfoxide (dmsABC), and nitrate and nitrite (napA and narGHI, nirBD) reductases and the propionate kinase (tdcD), as well as anion transport processes, such as alanine (alaE) and sulfate (cysAPU) transporters (cluster 8 in Data Set S1). Furthermore, gene ontology analysis revealed that cluster 5, which was also upregulated at t = 4 h in the parental but not in the DtnaA strain, was enriched in genes for the cellular response to stress, the response to pH, the response to reactive oxygen species, and hyperosmotic response processes (e.g., the aldehyde dehydrogenase aldB, the GABA permease gabP, the transcriptional regulator gadX, the glycine betaine transporter osmF, the superoxide dismutase sodC, the glutaminase glsA, the chaperone hdeA, and the deglycases yhbO and hchA) (cluster 5 in Fig. 4a and Data Set S1). Taken together, these data demonstrate that, thanks to the availability of tryptophanase, the parental strain is capable of performing key metabolic processes that allow the maintenance of a lower intracellular pH than the one displayed by the DtnaA strain, thus providing a mechanistic explanation of our data in Fig. 2 and previous bulk measurements (56). Clusters 1 and 9 displayed a significantly stronger downregulation at t = 4 h in the parental than in the DtnaA strain (22.6 versus 20.6 and 20.7 versus 0.0 mean log 2 fold changes, respectively; P , 0.0001 for both) (Fig. 3d). Cluster 1 was primarily enriched in genes for metal ion transport and homeostasis (e.g., the fec, fep, and fhu operons, tonB, and efeU), secondary metabolism (e.g., the ent operon for the biosynthesis of the siderophore enterobactin), fatty acid metabolism (the fatty acid degradation regulon fad), and aerobic respiration (e.g., the cytochrome bo genes cyoABCD) processes, in contrast with cluster 8, which was enriched in genes for anaerobic respiration ( Fig. 4b and Data Set S1). Cluster 9 was primarily enriched in genes for ribonucleoprotein and ribosome assembly (e.g., the 50S ribosomal protein L genes rplABCDEFKP), translation (e.g., the 30S ribosomal protein S genes rpsDEGHKNPQS), peptide metabolism (e.g., the elongation factor tufA), lipopolysaccharide metabolism (e.g., the lipopolysaccharide glucosyltransferase genes waaBGJPQY), and macromolecule and oligosaccharide biosynthesis processes (e.g., the transcription elongation factor greA) ( Fig. 4b and Data Set S1). It is also worth noting that at t = 4 h, the DtnaA but not the parental strain upregulated carbohydrate transport processes in order to scavenge for carbon sources in the extracellular environment, especially the fructose-specific phosphotransferase system (fru operon) and, to a lesser extent, glucose (ptsG) and maltose (mal operon) import processes. Taken together, these data suggest that in order to compensate for the decreased availability of intracellular metabolites, such as pyruvate, which is a key intermediate in several metabolic pathways, at t = 4 h, the DtnaA strain employed alternative strategies (e.g., iron, carbohydrate, and peptide uptake and metabolism) for maintaining cellular metabolism and cell growth. Finally, clusters 2, 3, 4, 6, and 7 displayed a weak regulation at t = 4 h for both the parental and the DtnaA strain ( Fig. 5d; Data Set S1), but it is worth noting that the mutant (but not the parental strain) used the cytochrome bo (cyoABCD) and proton antiporters (chaA and mdfA) to maintain pH homeostasis in the presence of the acidic extracellular pH.
Overall, this genome-wide transcriptome analysis revealed that abolishing tryptophanase activity causes a major remodeling of the E. coli gene expression profile besides abolishing the conversion of tryptophan to pyruvate, indole, and ammonia. Such remodeling includes key processes governing pH homeostasis, thus explaining our findings that the DtnaA strain displays a higher intracellular pH than the parental strain and that indole supplementation alone is not sufficient for lowering the intracellular pH of the mutant.
Persister but not VBNC E. coli cells are capable of maintaining their intracellular pH during ampicillin treatment. The average intracellular pH of VBNC and susceptible cells in the parental E. coli strain decreased as soon as 1 h after ampicillin addition at a concentration of 25Â the MIC and reached a minimum of 7.0 after 3 h (Fig. 5b and c). Notably at t = 3 h, the second peak at pH 7.6 largely disappeared for susceptible and VBNC cells (Fig. S3). After this initial drop in intracellular pH, the VBNC and susceptible cells maintained an average intracellular pH at around 7.0 when incubated in fresh LB at t = 3 h, reaching a minimum pH of 6.8 at t = 24 h. In contrast, persisters maintained their intracellular pH during ampicillin treatment and then increased their pH during their successive growth on fresh nutrients, up to a maximum of 7.3 at t = 24 h (Fig. 5a and Fig. S3). Therefore, after treatment and incubation in LB, the distribution of intracellular pH within the E. coli population was completely reversed: persisters that before treatment had the most acidic mean pH were the most alkaline, and conversely, VBNC and susceptible E. coli cells that had the most alkaline pH were the most acidic.
Similarly to the parental strain, the DtnaA mutant VBNC and susceptible cells exhibited a drop in intracellular pH after the addition of ampicillin (Fig. 5e and f). Both VBNC and susceptible cells exhibited two further phenotypic responses; a subset of VBNC and susceptible cells maintained a pH at around 7.6 at t = 3 h, whereas others exhibited a dramatically lower pH, at around 6.8 (Fig. S4). During the successive 21 h of incubation in fresh LB, VBNC and susceptible cells maintained an average intracellular pH of around 7.2 but still displayed two subpopulations at pHs 6.8 and 7.6 at t = 24 h (Fig. S4). The DtnaA persisters instead increased their pH after ampicillin addition, again as with the parental strain profile, up to a maximum of 7.7 at t = 24 h (Fig. 5d and Fig. S4).
Finally, when we added 0.5 mM extracellular indole to the DtnaA mutant culture, we found a steep decrease of more than 0.5 in the intracellular pHs of VBNC and susceptible cells after ampicillin addition, down to a minimum of 7 ( Fig. 5h and i). In contrast, we did not find a significant change in the intracellular pH of persisters throughout the experiment (Fig. 5g and Fig. S5).
Taken together, these data demonstrate a striking difference between the intracellular pH regulation of persister cells and that of VBNC cells before, during, and after ampicillin treatment, with the VBNC cell intracellular pH profile resembling more closely that of the susceptible cells. Furthermore, these findings suggest that tryptophanase activity and the biological processes associated with it play a role in the regulation of intracellular pH before drug treatment but do not affect the dynamic profile of intracellular pH during antibiotic treatment, which is instead affected by supplementation of extracellular indole.
Abolishing tryptophanase activity increases the fraction of persister but not VBNC E. coli cells. Since our microfluidics-microscopy assay allows us to determine the fate of each individual cell after treatment with ampicillin at 25Â the MIC, we were able to determine the composition of the microbial population in terms of the abundance of each phenotype in stationary phase, which is the phase investigated in all our single-cell measurements. We found that the fractions of persister, VBNC, and susceptible cells in the parental E. coli strain were 0.01, 0.04, and 0.95, respectively (filled bars in Fig. 6), in accordance with previously reported studies (9).
In comparison, the persister fraction of the DtnaA mutant was over three times higher than that of the parental strain, whereas the fraction of VBNC cells of the DtnaA mutant was similar to that measured for the parental strain (dotted and filled bars in Fig. 6, respectively). These data are in accordance with persister fractions found in bulk measurements showing that the DtnaA mutant contains more persisters than the parental strain in stationary phase, which is the growth phase after the indole pulse, which naturally occurs at the transition from exponential to stationary phase (Fig. S6a) (56,57). In contrast, we found that the opposite occurred in bulk cultures in exponential phase, with a higher fraction of persisters in the parental strain than in the DtnaA mutant (Fig. S6b).
Next, we set out to determine whether supplementing a stationary phase DtnaA culture with 0.5 mM indole would restore the persister parental strain phenotype. When we supplemented the DtnaA mutant culture with indole, we found a decrease in the persister fraction, down to 0.001, which is more than an order of magnitude lower than that measured for the DtnaA strain in the absence of indole; in contrast, indole addition did not have an impact on the fraction of VBNC cells (Fig. 4). These data suggest that in the absence of tryptophanase, the number of stationary phase E. coli cells that are persisters after ampicillin treatment increases; however, chemically restoring persistent indole signaling restores the low persister fraction observed in the parental Unique Differential pH Regulation in Persister Cells ® strain. Interestingly, we also note that neither tryptophanase activity nor indole signaling affected the composition of the microbial population in terms of the abundance of VBNC cells, which remained unchanged across the different experiments.

DISCUSSION
Maintenance of intracellular pH within a cell is vital for cellular structure, functioning, and pathogenesis (49,77). pH homeostasis allows correct protein folding, optimum enzymatic activity in metabolic reactions, and growth to take place (78), and it has recently been suggested that it might play a role in phenotypic survival after antibiotic treatments (46); however, this hypothesis remains to be tested. Therefore, we sought to investigate whether cell-to-cell differences in intracellular pH may represent a possible bacterial strategy to evade antibacterial activity.
By measuring the temporal changes in intracellular pH at the single-bacterium level, we found that before antibiotic exposure, E. coli persisters that survive ampicillin treatment display a lower intracellular pH than VBNC and susceptible E. coli cells. This difference resulted from a tighter regulation of pH in persisters, whereas VBNC and susceptible cells displayed a distribution of pH values, with two local maxima around 7.0 and 7.6. This novel finding represents, to the best of our knowledge, the first evidence of phenotypic differences between persister and VBNC cells before antibiotic treatment. Indeed, due to the difficulty of studying persisters and VBNC cells together, there is minimal literature comparing and contrasting these two phenotypes within the same experiment. The phenotypic differences that we found are notable considering that persister and VBNC cells are often thought to display the same dormant phenotype (26,79), whereas we demonstrate here that persister and VBNC cells are clearly distinguishable, suggesting that antimicrobials that selectively target persisters could be developed. These data also shed new light on the idea proposed by Bartek et al. that a contributing factor of antibiotic lethality is an antibiotic's effect on bacterial intracellular pH (77). Bartek et al. showed that a number of antibiotics cause an increase in the intracellular pH of Mycobacterium smegmatis, thought to disrupt essential processes, with the mechanism remaining unclear. This study also suggested that when a number of pathogens, including E. coli, were exposed to conditions which encouraged intracellular acidification, there was a decrease in their susceptibilities to antibiotics, including kanamycin, norfloxacin, and carbenicillin (77). This hypothesis proved to be true in our paper, at least with respect to persister E. coli cells displaying a lower pH than the population average before treatment and then surviving supra-MIC ampicillin treatment.
Persisters have been previously linked to indole and tryptophanase activities via a number of processes and stress responses, including multidrug exporters (44), dualfunction importers (67), quorum sensing (60), oxidative shock (43,65), and toxin-antitoxin modules (42). Therefore, we sought to determine the impact of the absence of tryptophanase activity and indole signaling on the intracellular pH values of persister, VBNC, and susceptible cells. Using a DtnaA strain, we found that the intracellular pH values of persister, susceptible, and VBNC cells before ampicillin treatment were significantly higher than the corresponding values measured for the parental E. coli strain, in accordance with population-level measurements (56). We then demonstrated that providing the DtnaA strain with mimics of persistent or pulse indole signaling could not restore the regulation of intracellular pH measured in the parental strain. Taken together, these data suggest that tryptophanase activity is key in maintaining pH homeostasis in response to changes in the pH of the extracellular environment (76) and that processes other than the conversion of tryptophan to indole underlie the observed acidification of stationary phase persisters, compared to VBNC or susceptible cells, preparing them to survive antibiotic challenge and promptly regrow after antibiotic removal.
Neutralophiles, such as E. coli, employ two major strategies to maintain pH homeostasis: the regulation of proton transporters, such as pumps and antiporters, and the remodeling of metabolic patterns, chiefly those involved in the production of organic acids (49). Accordingly, our whole-genome transcriptome and cluster analysis revealed that at the transition from exponential to stationary phase, the DtnaA strain upregulated a set of transporters that were orthogonal to those in the parental strain. Even more remarkably, the DtnaA strain significantly remodeled its metabolism compared to that of the parental strain, especially in terms of the use of amino acid decarboxylases, which are known to play a major role in acid resistance (80). Besides the loss of tryptophan conversion to pyruvic acid, which is known to decrease the intracellular pH (82), other key carboxylic acid catabolic processes, as well as oxidation-reduction, response to pH (83), response to reactive oxygen species, and hyperosmotic response components, were upregulated in the parental but not in the DtnaA strain. In contrast, the DtnaA but not the parental strain upregulated metal transport, secondary metabolism, fatty acid metabolism, ribosome assembly, translation, and carbohydrate transport processes. It is also worth noting that whereas the parental strain upregulated anaerobic respiration processes in this temporal window, the DtnaA strain, in contrast, upregulated aerobic respiration processes, and it is known that neutralophiles display different pH responses during anaerobic and aerobic respiration (83).
Overall, this novel knowledge about the genetic and metabolic consequences of the loss of tryptophanase is very important because it explains both our and previous findings that the DtnaA strain is more alkaline than the parental strain (56) and opens the way for the use of tryptophanase inhibitors (84) to manipulate bacterial metabolism and intracellular pH. Moreover, this novel genome-wide transcriptome analysis constitutes a very valuable resource when modeling bacterial metabolism in response to physicochemical changes in the extracellular environment (85)(86)(87), since we show that abolishing tryptophanase activity triggers a major remodeling of bacterial metabolism and pH homeostasis, whereas a previous study showed that indole supplementation alone affects protein expression to a lesser extent (88).
Finally, we sought to investigate the response to antibiotics in terms of both changes in intracellular pH in persister, VBNC, and susceptible cells as well as measuring the relative fractions of these three phenotypes after antibiotic treatment. We found that persisters increased their intracellular pH in the parental and DtnaA strains with and without indole supplementation, whereas VBNC and susceptible cells became more acidic in response to antibiotic treatment. Therefore, processes that are not linked to indole signaling or tryptophanase activity must underlie the alkalinization of persisters and acidification of VBNC cells in response to ampicillin treatment. It is worth noting that there are conflicting reports on the effect that indole has on the composition of a population in terms of the relative abundances of persister, VBNC, and susceptible cells. Vega et al. (65) and Kuczy nska-Wi snik et al. (66) found an increase in the abundance of persisters in the presence of indole, whereas Hu et al. (42), Kwan et al. (43), and Wang et al. (67) found the opposite. These inconsistencies are probably due to the different indole concentrations, modes of action of antibiotics applied, types of indole signaling, and phases of cells used between assays (58). Indeed, we also observed a change in persister fraction within our ensemble measurements, depending on certain criteria. In exponential phase, we found a higher proportion of persisters in the parental than in the DtnaA strain. However, in stationary phase, this phenomenon was reversed, with a higher proportion of persisters found in the DtnaA than in the parental strain, but mimics of persistent indole signaling restored the persister levels measured for the parental strain. Therefore, the DtnaA strain appears to be a high-persister-level mutant (21), exhibiting high persister levels in stationary but not in exponential phase, suggesting that persistent indole signaling increases the level of triggered persisters but not of spontaneous persisters (89). Surprisingly, we found that the abundance of VBNC cells within the E. coli population was not affected by the removal of either tryptophanase activity or indole signaling. This is important because it contradicts the hypothesis that persister and VBNC cells display the same dormant phenotype (79).
In conclusion, our findings demonstrate that the regulation of intracellular pH is not homogeneous within a clonal population, with persisters displaying differential pH regulation and regrowing after antibiotic treatment. We showed that tryptophanase activity and several associated metabolic processes that are upregulated at the transition from exponential to stationary phase underpin this differential pH regulation, whereas tryptophanase activity and its associated indole signaling control the fraction of persister cells that survive and regrow after ampicillin treatment. Considering that more than 85 bacterial species encode tryptophanases (90), our data open the way for the manipulation of the intracellular pH and phenotypic compositions of microbial populations by developing and using tryptophanase-targeting compounds.

MATERIALS AND METHODS
Strains and culture conditions. All chemicals were purchased from Fisher Scientific or Sigma-Aldrich unless otherwise stated. Cells were cultured using LB medium (10 g/liter tryptone, 5 g/liter yeast extract, and 10 g/liter NaCl; Melford) and LB agar plates (15 g/liter agar). Minimal medium (M9) was made from 1Â M9 salts, 2 mM MgSO 4 , and 0.1 mM CaCl 2 in deionized water. The medium was autoclaved before the addition of sterile thiamine hydrochloride, for a final concentration of 3 mM. E. coli BW25113 parental and DtnaA strains were obtained from the Keio collection (91,92). Both strains were transformed with the pSCM002 plasmid, which expresses the mCherry-pHluorin translational fusion protein under an arabinose-inducible promoter and carries a chloramphenicol resistance marker. pSCM002 was constructed from pSCM001 (Addgene; 124605) (56) by swapping the ampicillin resistance marker with a chloramphenicol resistance marker. Briefly, PCR was used to amplify two fragments: a chloramphenicol resistance marker from pT2SC (Addgene 59382) and pSCM001, excluding the ampicillin resistance marker (93). Each fragment was amplified with flanking regions homologous to the other fragment. The two fragments were ligated using a Gibson assembly kit (NEB) (94) to generate a pBAD TOPO plasmid containing an mCherry-pHluorin translational fusion and a chloramphenicol resistance marker, pSCM002. The sequences of primers used to generate the pSCM002 plasmid are available in the supplemental material (Table S1). pHluorin (superecliptic pHluorin) is a green fluorescent reporter for intracellular pH derived from GFP with 8 amino acid changes (F64L, S65T, S147D, N149Q, T161I, S202F, Q204T, and A206T) (70,71). mCherry fluorescence is not affected by pH and was therefore used to normalize changes in fluorescence caused by cell-to-cell variation in protein expression and differences in plasmid copy numbers between the exponential and stationary phases. Cultures were streaked on LB agar supplemented with chloramphenicol (25 mg/ml) for plasmid maintenance and incubated overnight at 37°C to obtain single colonies. Overnight liquid cultures were prepared with a single colony from a streak plate and grown in 200 ml of LB medium supplemented with chloramphenicol (25 mg/ml) and arabinose (5 mg/ml) for 17 h at 37°C on a shaking incubator at 200 rpm. When required, 0.5 mM indole (dissolved in absolute ethanol) was supplemented at the beginning of an overnight culture of the DtnaA strain.
Microfluidic device fabrication. The mother machine microfluidic device was fabricated by pouring a 9:1 (base-to-curing agent) polydimethylsiloxane (PDMS) mixture into an epoxy mold kindly provided by S. Jun, as previously described (95). The PDMS was cured at 70°C for 2 h and then peeled from the epoxy mold to obtain 12 individual chips. Fluidic access was achieved using a 0.75 mm biopsy punch (Harris Uni-Core; WPI). The microfluidic device was assembled by bonding the PDMS chip to a glass coverslip, both of which were exposed to an air plasma treatment for 10 s at a 30 W plasma power (plasma etcher; Diener, Royal Oak, MI, USA) and then placed in contact to irreversibly bond the PDMS chip to the glass coverslip. The plasma treatment temporarily renders the device hydrophilic, so within 5 min postbonding, the device was filled with 2 ml of 50 mg/ml bovine serum albumin (BSA) solution and incubated at 37°C for an hour to prevent cell adhesion to the internal surfaces of the device.
Microfluidics-microscopy assay to measure intracellular pH in individual cells. An overnight culture was prepared by incubating E. coli in LB at 37°C for 17 h with agitation at 200 rpm. Bacteria were harvested from the overnight culture via centrifugation for 10 min at 4,000 rpm at room temperature (Eppendorf 5810 R). The supernatant was filtered twice (medical Millex-GS filter, 0.22 mm; Millipore Corp.) and used to resuspend the bacteria to an optical density at 595 nm (OD 595 ) of 75; additionally, BSA was added to the bacterial suspension at a concentration of 0.5 mg/ml. Two microliters of the bacterial suspension was injected into the microfluidic device and incubated at 37°C until there were 1 to 2 bacteria in the lateral side channels. Fluorinated ethylene propylene tubing (1/32 in. by 0.008 in.) was connected to the inlet and outlet holes and connected to a flow rate measuring device (Flow Unit S; Fluigent, Paris, France) and outlet reservoir, respectively. The flow rate measuring device was regulated by a computerized pressure-based flow control system (MFCS-4C; Fluigent) controlled by MAESFLO software (Fluigent). Spent medium was flushed through the device to wash excess bacteria out of the main channel at 300 ml/h for 8 min to completely exchange the fluid in the device and tubing. The chip was mounted on an inverted microscope (IX73 Olympus, Tokyo, Japan), and images were acquired in bright-field and two fluorescence modes. Images were collected via a 60Â, 1.2 numerical aperture (NA) objective (UPLSAPO60XW; Olympus) and an sCMOS camera (Zyla 4.2; Andor, Belfast, UK). The fluorescence modes used a fluorescein isothiocyanate (FITC) and tetramethyl rhodamine isothiocyanate (TRITC) filter, exposing the bacteria to blue and green excitation bands of broadspectrum LEDs at 20% and 5% of their intensity, respectively, for 0.03 s. The microfluidic device was moved by two automated stages (M-545.USC and P-545.3C7 [Physik Instrumente, Karlsruhe, Germany] for coarse and fine movements, respectively) to image multiple fields of view in a sequential manner. The imaging setup was controlled by LabView, with automated shutter and filter switching. After acquiring the first set of images, we allowed 25Â the MIC of ampicillin to flow in a 90:10, vol/vol, M9-LB solution at 300 ml/h for 8 min before lowering the flow rate to 100 ml/h for 3 h. After 3 h of treatment, LB medium was flushed through the chip at 300 ml/h for 8 min before the rate was lowered to 50 ml/h for 21 h. Images were taken hourly between t = 0 and t = 6 h and at 24 h. At 24 h, propidium iodide (PI) was used to flow through the microfluidic device for 15 min according to the manufacturer's specifications. Bright-field and fluorescence images were acquired to distinguish between VBNC and SNL cells using 5% green LED intensity with a TRITC filter for 0.03 s. All experiments were carried out at room temperature.
Indole supplementation. Three indole supplementation protocols were tested to recover the parental strain in the DtnaA strain. The first protocol used addition of indole to a final concentration of 0.5 mM at the start of a culture and then incubation overnight as previously described. The second protocol mimicked the previously specified indole pulse (56,57) by supplementing indole at a final concentration of 5 mM for 20 min once the 17 h overnight DtnaA culture was loaded in the microfluidic device. The third protocol used a portable syringe pump (Chemyx Inc.) to dynamically drip indole into a growing DtnaA culture for 200 min to mimic the external indole concentration previously recorded by Zarkan et al. (56). The results from all three indole supplementation methodologies are reported in Fig. S2.
Intracellular pH determination via single-cell fluorescence. pHluorin fluorescence intensity has a linear relationship with pH between pH 6.5 and pH 8 (71). The fluorescence intensity of mCherry is unaffected by pH and was used to normalize the pHluorin fluorescence due to variations in protein expression and plasmid copy number between individual cells. To equate the relative fluorescence of each cell to a pH value, the intensity ratio of pHluorin to mCherry for each cell was calculated and, with a calibration curve, converted to a pH value. The standard curve was generated by exposing the parental strain and DtnaA mutant to 50 mM carbonyl cyanide m-chlorophenyl hydrazine (CCCP) at 300 ml/h for 8 min and then at 100 ml/h for 20 min, which causes the intracellular pH to equalize to the pH of the medium by allowing the translocation of protons across the cytoplasmic membrane (56). PBS solutions whose pHs were manually adjusted to 6.5, 7.0, 7.5, and 8.0 were used to generate the standard curve using the same imaging/experimental settings described above. The relationship between fluorescence and pH obtained from the standard curve was used to calibrate all pH values in our experiments.
Image and data analysis. All image analysis was performed using a custom Python module, MMHelper (96). In brief, channels and bacteria were detected by using automated thresholding algorithms and were assigned unique numeric labels. Each bacterium in a frame was tracked, and a lineage was established and then manually verified. Note the increase in red fluorescence of the susceptible but not of the VBNC cells, thus indicating the susceptibility of the cell. By normalizing the pH-sensitive pHluorin intensity against the mCherry intensity, which is unaffected by pH, for each bacterium at each time point, we accounted for variation in protein expression and plasmid copy number between individual cells. We determined the intracellular pH of individual cells from their relative fluorescence by using a calibration curve.
For each lineage, the corresponding masks were used to extract the width, length, area, and fluorescence intensities of the individual bacteria from the corresponding fluorescence images. The background fluorescence was established from the average fluorescence for the areas of the channels that did not contain bacteria and subtracted from the fluorescence intensity measured on each bacterium. The information for all bacteria within a frame across the different time points was written into .csv files. Each bacterium was labeled manually as a persister, a VBNC cell, a susceptible lysed cell, or a susceptible nonlysed cell. The data presented in this paper were plotted using GraphPad Prism 8 and represent means and standard errors of the means from at least 3 biological replicates. Due to the large sample sizes, error bars are small compared to the corresponding mean values and are hidden behind the data points in some of the graphs. Statistical significance was tested by an unpaired t test with Welch's correction.
Transcriptomic analysis. RNA isolation, library preparation, sequencing, and transcriptomic data processing were performed as previously reported (76). Briefly, parental and DtnaA E. coli cultures were prepared as described above, and 500 ml aliquots were taken after t = 3, 4, 5, and 17 h after inoculation in LB in biological triplicates. RNAprotect bacterial reagent (Qiagen) was added to each aliquot. Extractions were performed using an RNeasy minikit (Qiagen) by following the manufacturer's instructions. DNA removal during extraction was carried out by using RNase-Free DNase I (Qiagen). RNA concentration and quality were measured using a Qubit 1.0 fluorometer (ThermoFisher Scientific) and 2200 TapeStation (Agilent), respectively, and only samples with an RNA integrity number larger than 8 were taken forward. Transcript abundance was quantified using Salmon for each gene in all samples. Subsequent differential analysis was performed using DEseq2 in R software to quantify the log 2 fold change in transcript reads (97) for each gene.
Cluster and gene ontology analysis. Clustering analysis of the transcriptome data was performed using the mclust 5 package (version 5.4.7) for R (98). Using this method, a Gaussian finite mixture model was estimated for all transcript log 2 fold change data, with 6 total dimensions (3 time point comparisons across 2 genotypes). The method was configured to test all possible models (e.g., hyperspherical versus ellipsoidal cluster shapes, equal volume versus various volume clusters, etc.), for a range of 2 to 20 total clusters. The minimal, best-fitting model was identified by the Bayes information criterion as 9 ellipsoidal clusters with equal shapes and orientations. Gene Ontology enrichment analysis was performed using the clusterProfiler package (version 3.16.1) for R (99). Enrichment in terms belonging to the "Biological Process" ontology was calculated for each gene cluster, relative to the set of all genes quantified in the experiment, via a one-sided Fisher exact test (hypergeometric test). P values were adjusted for false discovery by using the method of Benjamini and Hochberg (100). Finally, the lists of significantly enriched terms were simplified to remove redundant terms, as assessed via their semantic similarity to other enriched terms, using clusterProfiler's simplify function.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.