Skip to main content
NCBI home page
Proceedings of the National Academy of Sciences of the United States of America logoLink to Proceedings of the National Academy of Sciences of the United States of America
. 2021 Jul 30;118(31):e2103027118. doi: 10.1073/pnas.2103027118

Microbiota-derived metabolites inhibit Salmonella virulent subpopulation development by acting on single-cell behaviors

Alyson M Hockenberry a,b,1, Gabriele Micali a,b, Gabriella Takács a,b, Jessica Weng a,b,2, Wolf-Dietrich Hardt c, Martin Ackermann a,b
PMCID: PMC8346864  PMID: 34330831

Significance

Emergence of distinct cell types in populations of genetically identical bacteria is common. Furthermore, it is becoming increasingly clear that cooperation between cell types can be beneficial. This is the case during Salmonella infection, in which cooperation between inflammation-inducing virulent and fast-growing avirulent cell types occurs during infection to aid in colonization of the host gut. Here, we show gut microbiota–derived metabolites slow growth of the virulent cell type. Our study implies microbial metabolites shape cooperative interactions between the virulent and avirulent cell types, a finding that can help explain the wide array of clinical manifestations of Salmonella infection.

Keywords: Salmonella, single cell, pathogenesis

Abstract

Salmonella spp. express Salmonella pathogenicity island 1 Type III Secretion System 1 (T3SS-1) genes to mediate the initial phase of interaction with their host. Prior studies indicate short-chain fatty acids, microbial metabolites at high concentrations in the gastrointestinal tract, limit population-level T3SS-1 gene expression. However, only a subset of Salmonella cells in a population express these genes, suggesting short-chain fatty acids could decrease T3SS-1 population-level expression by acting on per-cell expression or the proportion of expressing cells. Here, we combine single-cell, theoretical, and molecular approaches to address the effect of short-chain fatty acids on T3SS-1 expression. Our in vitro results show short-chain fatty acids do not repress T3SS-1 expression by individual cells. Rather, these compounds act to selectively slow the growth of T3SS-1–expressing cells, ultimately decreasing their frequency in the population. Further experiments indicate slowed growth arises from short-chain fatty acid–mediated depletion of the proton motive force. By influencing the T3SS-1 cell-type proportions, our findings imply gut microbial metabolites act on cooperation between the two cell types and ultimately influence Salmonella’s capacity to establish within a host.

The mammalian gastrointestinal tract (GI) is chemically defined by resident bacteria metabolizing the host’s diet. Among the most abundant microbial metabolites are the short-chain fatty acids (SCFAs) acetate, butyrate, and propionate. Each are found at up to 100 millimolar concentrations in the human GI, with levels lowest in the ileum, increasing to high levels in the proximal colon, and tapering off in the distal colon (1). They also vary as a function of microbiota members and fluctuate over time, correlating with the timing and composition of meals (2–4). The chemical environment an enteric microbe finds itself in, therefore, varies by location in the gut, from person-to-person, and across time.

This holds true as well during the initial phases of infection by the enteric pathogen Salmonella enterica. Salmonella expresses virulence genes to inflame and subsequently colonize the host GI (reviewed in ref. 5). Inflammation is initially caused by invading into gut epithelial tissues using a Type III Secretion System encoded on Salmonella pathogenicity island 1 (T3SS-1, ref. 6, reviewed in ref. 7). This secretion system pumps effector proteins directly into host cells, leading to bacterial uptake and subsequent inflammation. Inflammation increases nutrient availability and killing of competitor resident microbiota, opening a niche for Salmonella establishment in the gut (8–12).

Recent evidence shows there is cell-to-cell variation in T3SS-1 expression by Salmonella cells. Each individual in the population is found in a discrete expressing (T3SS-1+) or nonexpressing (T3SS-1−) state (13, 14). The emergence of these two cell-types is beneficial, as they cooperate with each other to facilitate colonization of the host. T3SS-1+ cells invade host tissues and induce inflammation, while T3SS-1− cells replicate quickly in the intestinal lumen to exploit the niche cleared by the T3SS-1+ cells (15–17). It is not fully understood how these cooperative interactions are influenced by environmental signals; of particular interest are the dynamic conditions of the GI, where there is high variation in the frequency of T3SS-1+ cells (18).

Several reports have shown microbiota-derived metabolites impact Salmonella infectivity. Importantly, SCFAs at physiological levels decrease T3SS-1 expression by populations of Salmonella cells and limit their invasion into host cells (19, 20). The authors concluded from these studies that SCFAs repress T3SS-1 expression and result in decreased pathogenicity. Consistent with this idea, the SCFAs butyrate and propionate impact Salmonella pathogenicity in vivo (21, 22). However, the fact that T3SS-1 expression varies from cell to cell within a population prompts revisiting the effect of SCFAs on T3SS-1 expression from a single-cell perspective.

The previously observed SCFA-mediated reduction in population-level T3SS-1 expression could arise from a number of mechanisms scaling from behaviors by individual cells. Determining how SCFAs impact single-cell behaviors will inform on their mechanism of action and ultimately how the gut environment influences Salmonella pathogenicity. In this study, we address this by combining mathematical modeling with time-resolved single-cell measurements in microfluidic devices. We find that SCFAs decrease the population-level T3SS-1 expression not by transcriptional repression but by decreasing the growth rate of T3SS-1–expressing cells. The reduced growth rate of these cells arises from a selective decrease in the proton motive force (PMF) in T3SS-1–expressing cells.

Results and Discussion

SCFAs Reduce Population-Level T3SS-1 Expression through Decreasing the Proportion of T3SS-1–Expressing Cells.

Reduced population-level gene expression can result from a combination of mechanisms: either by fewer expressing cells in the population or by decreased expression per cell. To address how SCFAs impact T3SS-1 expression by individual cells during growth, we first quantified population-level growth and T3SS-1 gene expression using PprgH-gfp transcriptional reporter cells grown in a range of SCFAs at physiological concentrations (Fig. 1A and SI Appendix, Fig. S1 A and B and Table S1). Consistent with previous reports (19, 20), increasing SCFA concentration correlated with decreased population-level PprgH-gfp expression (Fig. 1A). SCFAs similarly decreased population-level expression of other T3SS-1 genes (sipC and sicA) and did not impact population-level constitutive gene expression (rpsM) (SI Appendix, Fig. S1 C and D).

Fig. 1.

Fig. 1.

Open in a new tab

SCFAs decrease population-level prgH expression by decreasing the proportion of PprgH-gfp+ cells. (A) Mean ± SEM max growth rate and mean ± SEM GFP fluorescence intensity (i.e., prgH expression) normalized to OD600 by PprgH-gfp reporter cells in 0 (yellow), 75 (salmon), and 150 mM (dark orange) SCFAs. Average of triplicates from three independent plate-reader experiments, one-way ANOVA, expression by SCFA concentration, P < 0.0001; Tukey’s Honest Significance Test, 0 versus 75 mM, P < 0.0001, 75 versus 150 mM, P < 0.001. (B) The impact of SCFAs on single-cell GFP MFI ± SEM by PprgH-gfp cell-type and (C) mean ± SEM proportion of PprgH-gfp+ cells as measured by flow cytometry. A total of 50,000 cells quantified per timepoint and average of three independent experiments; MFI by treatment over time, two-way ANOVA, MFI by SCFA treatment, P > 0.2; PprgH-gfp cell type proportions by SCFA treatment over time, two-way ANOVA, proportion by SCFA treatment, P < 0.0001.

Using SCFAs at concentrations shown to reduce PprgH-gfp expression, we next quantified per-cell green fluorescent protein (GFP) expression. Reporter cells were cultured with 0, 75, or 150 mM SCFAs and examined by flow cytometry at different timepoints. The mean fluorescence intensity (MFI) of individual PprgH-gfp+ cells was unaffected by SCFA treatment (Fig. 1B and SI Appendix, Fig. S1E). Rather, SCFA treatment resulted in a dose-dependent decrease in the proportion of PprgH-gfp+ cells (Fig. 1C and SI Appendix, Fig. S1E). Untreated populations of cells consisted of roughly 40% PprgH-gfp+ cells by midexponential phase through stationary phase. Populations grown with 75 or 150 mM SCFA concentrations consisted of 25 and 10% PprgH-gfp+ cells at these timepoints, respectively. These observations demonstrate SCFAs do not decrease PprgH-gfp expression by individual cells, rather they act on the frequency of PprgH-gfp+ cells in the population.

The SCFA-Mediated Decrease in PprgH-gfp+ Cells Occurs Predominately through Limitations on PprgH-gfp+ Cell Growth.

How can SCFAs decrease the frequency of PprgH-gfp+ cells in the population? We reasoned PprgH-gfp+ population frequency is set by four parameters: the growth rates of PprgH-gfp− and PprgH-gfp+ cells and the rates of switching between the two cell types (SI Appendix, Fig. S2A). A minimal mathematical model describing the behavior of an exponentially growing population (expanded upon in SI Appendix, Supporting Materials) established decreased PprgH-gfp+ frequency can arise through multiple, nonmutually exclusive scenarios: increasing the PprgH-gfp+ to PprgH-gfp− switching rate, decreasing the PprgH-gfp− to PprgH-gfp+ switching rate, increasing the PprgH-gfp− growth rate, or decreasing the PprgH-gfp+ growth rate (SI Appendix, Fig. S2 B–D). Parameter space exploration shows the interaction of these four parameters (SI Appendix, Fig. S2 C–H). In general, smaller changes in growth rates led to large shifts in the cell-type proportions, whereas larger changes in switching rates were necessary to shift cell-type proportions.

Knowing how SCFAs change these parameters provides insight into their mechanism of action. Changes in cell-type growth rates imply SCFAs act on cellular properties. Meanwhile, SCFA action on cell-type switching rates would suggest they act on prgH regulation by individual cells. We measured the impact of SCFAs on these four single-cell parameters directly using quantitative time-lapse microscopy. We employed a “feeding culture” microfluidic approach to recapitulate our population-level experiments (23, Fig. 2A). Reporter cells under observation in a microfluidic chip were fed by an actively growing Salmonella culture with or without SCFAs. In this manner, reporter cells under observation are experiencing the dynamic environment of a growing culture, where nutrients are depleted and compounds are excreted. We observed individual reporter cells over 12 h of culture (Video S1). By analyzing the images, we quantified single-cell growth rates and cell-type switches for each cell type across growth phases.

Fig. 2.

Fig. 2.

Open in a new tab

SCFAs slow growth by PprgH-gfp+ cells. (A) Experimental setup for quantitative time-lapse microscopy using a feeding culture approach. An actively growing culture of Salmonella cells is pumped through a microfluidic chip and into a waste container using a peristaltic pump. Simultaneously, T3SS-1 reporter Salmonella cells loaded into the microfluidic chip are experiencing the same chemical environment as cells in the flask. Cells in the chip are imaged every 3 min. Time-lapse images are then analyzed to extract PprgH-gfp− and PprgH-gfp+ growth and switching rates. Data presented is from four independent experiments; we designated the first 0.5 h of growth lag phase, the next 6 h early exponential, and the last 5.5 h late exponential. (B) Mean growth rates (n = 2,502 cells, box = quartiles, whiskers = all observations) by cell type, SCFA treatment, and growth phase. Three-way ANOVA followed by Tukey’s Honest Significance Test, no SCFAs versus SCFAs, P < 0.001; PprgH-gfp+ versus PprgH-gfp−, P < 0.001. (C) Ratio ± SEM of PprgH-gfp+ and PprgH-gfp− cell growth rates in the absence of presence of SCFAs. Two-way ANOVA followed by Tukey’s Honest Significance test, no SCFAs versus SCFAs, P < 0.001. (D) Observed cell-type switching events plotted by SCFA treatment over time. Each dot indicates the time of an observed switching event. (E) Mean ± SEM probability of cell-type switching during each growth phase by treatment. Two-way ANOVA followed by Tukey’s Honest Significance test, no SCFAs versus SCFAs, P > 0.1. (F) Results from Gillespie simulations using our experimentally measured parameters. Average of 100 individual simulations per condition. Two-way ANOVA, no SCFA growth versus SCFA growth, P < 0.001; no SCFA switching versus SCFA switching, P > 0.1.

Single-cell growth rates of untreated cells changed over time. Growth rates increased as they exited lag phase and entered exponential growth and decreased as they entered late exponential phase (Fig. 2B). All cells continued growth into stationary phase (SI Appendix, Fig. S3A). At all timepoints, PprgH-gfp− cells grew ∼25% faster than PprgH-gfp+ cells, in line with previous reports [Fig. 2 B and C (17)].

Cell-type switching rates of untreated cells also changed by growth phase (Fig. 2 D and E). Consistent with few PprgH-gfp+ cells during pre-exponential phase growth (Fig. 1C), PprgH-gfp+ to PprgH-gfp− switches were frequent during lag phase. Upon reaching early and late exponential phase, the probability of PprgH-gfp− to PprgH-gfp+ switches increased and PprgH-gfp+ to PprgH-gfp− switches decreased.

Salmonella cells grown in the presence of SCFAs behaved differently. Individual cells grown in the presence of SCFAs grew more slowly during lag phase, in line with our population-level observations (SI Appendix, Fig. S1 A and B). During the later phases of growth, SCFA treatment slowed single-cell growth rates by both cell types during each growth phase, however to different degrees (Fig. 2 B and C). While SCFAs resulted in PprgH-gfp− single-cell growth rates ∼25% lower than untreated cells, SCFAs slowed the growth of PprgH-gfp+ cells by roughly 50% (Fig. 2B). During each phase of growth, SCFA-treated PprgH-gfp+ cells grew roughly 50% slower than PprgH-gfp− cells (Fig. 2C).

Given that SCFAs were expected to repress PprgH-gfp expression by individual cells based on population-level measurements, it was surprising to see that this was not the case (Fig. 2 D and E and SI Appendix, Fig. S3 B and C and Video S1). Rather, PprgH-gfp+ cells frequently maintained expression for many generations, and switching off events were only observed during lag phase (Fig. 2 D and E). PprgH-gfp− cells seldom switched to PprgH-gfp+ in the presence of SCFAs, although these events were observed more frequently compared to PprgH-gfp+ to PprgH-gfp− switching during exponential growth. Although SCFAs reduced the number of cell-type switching events overall, the difference was not statistically significant (Fig. 2 D and E).

All of the measured changes in parameters can help to explain how SCFAs lower the frequency of PprgH-gfp+ cells. To understand the interplay of these parameters while mimicking population growth dynamics, we used stochastic simulations to predict the proportion of PprgH-gfp+ cells in a population growing under experimentally measured single-cell growth and switching values. Simulations using parameters extracted from SCFA-untreated and -treated cells yielded a final PprgH-gfp+ frequency of 18 and 6%, respectively (Fig. 2F). Both are roughly twofold lower than what is observed experimentally. This underestimation by the simulations suggests additional parameters are necessary to fully explain PprgH-gfp subpopulation dynamics (expanded upon in SI Appendix, Supporting Materials). Nonetheless, these simple simulations captured the SCFA-mediated reduction of PprgH-gfp+ frequency in the population observed during our experiments.

These simulations can also inform on the relative influence that experimentally measured parameter shifts have on the proportion of PprgH-gfp+ cells in a population. Populations of cells simulated to grow at untreated rates and cell-type switching at SCFA rates yielded a final PprgH-gfp+ proportion similar to untreated cells (Fig. 2F). Similarly, populations simulated to grow at SCFA-treated rates and switch with untreated rates more resembled SCFA-treated PprgH-gfp+ proportions. These simulations indicate growth rate changes during SCFA treatment is a dominant driver of decreased PprgH-gfp+ population frequency.

These results demonstrate SCFAs decrease PprgH-gfp expression at the population-level by acting on single-cell growth rates. We conclude PprgH-gfp expression is not inhibited by SCFAs, as cells maintain prgH expression in their presence for at least 12 h and over many generations (Video S1). Taken together, our results indicate SCFAs decrease the proportion of PprgH-gfp+ cells predominately through a selectively stronger reduction in growth by PprgH-gfp+ cells.

Dissipation of the PMF Slows Growth of PprgH-gfp+ Cells.

SCFAs can impact Salmonella physiology in multiple ways. Salmonella can use all three SCFAs present in our experiments as a carbon source, thus SCFAs could facilitate growth. More fundamentally, upon entry into the bacterial cytosol, SCFAs dissociate into the anion and a proton due to their relatively high pKa. An accumulation of the SCFA anion leads to an increase in turgor pressure, and each SCFA anion can be added to proteins as posttranslational modifications; of interest, the major T3SS-1 regulator, HilA, has been shown to be posttranslationally acetylated and butyrylated (24, 25). Accumulation of protons lead to a decrease in the intracellular pH (reviewed in ref. 25). A decrease in the intracellular pH ultimately lowers the PMF, leading to a slower rate of ATP production and by implication limited growth.

Because SCFAs slow single-cell growth rates, we examined how a global decrease in PMF impacts PprgH-gfp+ population frequency. We added carbonyl cyanide-m-chlorophenylhydrazone (CCCP), a chemical decoupler of the proton gradient, at concentrations that did not inhibit population-level growth (SI Appendix, Fig. S4A). Similar to SCFA treatment, the addition of CCCP decreased population-level PprgH-gfp expression (SI Appendix, Fig. S4B) and the proportion of PprgH-gfp+ cells (Fig. 3A). Other mechanisms of PMF disruption, including growth in minimal media and the compound nigericin, similarly decrease the proportion of PprgH-gfp+ cells (SI Appendix, Fig. S4D). Growth in the presence of both SCFAs and CCCP did not further reduce the proportion of PprgH-gfp+ cells compared to SCFAs or CCCP alone (Fig. 3A and SI Appendix, Fig. S4A). Thus, PMF dissipation at levels that do not limit population-level growth are sufficient to decrease PprgH-gfp+ subpopulation, consistent with the idea that SCFAs reduce the proportion of PprgH-gfp+ cells through PMF dissipation.

Fig. 3.

Fig. 3.

Open in a new tab

PMF dissipation similarly reduces the proportion of PprgH-gfp+ cells and has a stronger effect on PprgH-gfp+ cells than on PprgH-gfp− cells. (A) Mean ± SEM PprgH-gfp+ proportion in the presence of no SCFAs, 150 mM SCFAs, 12.5 μM CCCP, and 150 mM SCFAs + 12.5 μM CCCP as determined by flow cytometry. Average of three independent experiments; two-way ANOVA followed by Tukey’s Honest Significance Test, no SCFAs versus other treatments, P < 0.001; SCFAs versus CCCP, P > 0.5; SCFAs versus SCFAs + CCCP, P > 0.2. (B) Cell-type switching events detected over time during feeding-culture microfluidics experiments in 12.5 μM CCCP. (C) Mean ± SEM single-cell growth rates by phenotype in the presence of no SCFAs, 150 mM SCFAs, and 12.5 μM CCCP. Three-way ANOVA followed by Tukey’s Honest Significant Difference (HSD) test; PprgH-gfp+ CCCP versus PprgH-gfp+ SCFA, P > 0.5; PprgH-gfp− CCCP versus PprgH-gfp− no SCFAs, P > 0.2; PprgH-gfp+ CCCP versus PprgH-gfp+ no SCFAs, P < 0.001. (D) Mean Mitoview 633 ± SEM staining intensity (higher intensity indicates higher PMF) by treatment and PprgH-gfp phenotype over time as determined by flow cytometry. Three-way ANOVA followed by Tukey’s HSD; interaction detected between cell-type and treatment, P < 0.001; for t = 5 and t = 7, PprgH-gfp+ no SCFAs versus PprgH-gfp+ all other treatments, P < 0.005; PprgH-gfp− no SCFAs versus PprgH-gfp− all treatments, P > 0.05; for t = 7, PprgH-gfp+ versus PprgH-gfp− no SCFAs, P < 0.01. (E) Ratio of PprgH-gfp+/PprgH-gfp− Mitoview 633 staining intensity per condition and timepoint; two-way ANOVA followed by Tukey’s HSD, no SCFAs versus all other treatment, P < 0.01; SCFAs versus CCCP or SCFAs + CCCP, P < 0.05; CCCP versus SCFAs + CCCP, P > 0.2.

We next quantified the impact of CCCP on single-cell growth and phenotypic switching rates using our microfluidic system (Video S2). CCCP treatment does not completely mimic SCFA treatment. In line with what we observed during SCFA treatment, the growth of PprgH-gfp+ cells was selectively slowed in the presence of CCCP (Fig. 3 B and C). In contrast to SCFA treatment, cells grown in the presence of CCCP show phenotypic switching behaviors similar to untreated cells (Fig. 3B). Because cells still phenotypically switch in the presence of CCCP, this suggests SCFAs have a dual role: as the SCFA dissociates into the proton and SCFA anion, the free proton reduces the PMF and the SCFA anion can exert other effects [e.g., posttranslational modification of HilA, the master-regulator of T3SS-1 (24)]. Nonetheless, these results indicate dissipation of the PMF selectively reduces the growth rate of PprgH-gfp+ cells and can decrease the proportion of PprgH-gfp+ cells.

A remaining question is how SCFAs and PMF dissipation differentially impacts the growth rates of PprgH-gfp+ and PprgH-gfp− cells. By combining single-cell analyses with a fluorescent PMF indicator, we observed PprgH-gfp+ cells maintain a higher membrane potential compared to PprgH-gfp− cells during log-phase growth (Fig. 3 D and E). SCFA and CCCP treatment led to an overall decrease in membrane potential (Fig. 3D). In particular, SCFA and CCCP treatment led to a statistically significant decrease in the PprgH-gfp+ cell PMF, whereas the SCFA-mediated decrease in in the PprgH-gfp− cell PMF was not statistically significant.

It is unclear how the two cell types are differentially susceptible to PMF dissipation. Previous work has shown HilD, the major regulator of T3SS-1, directly activates flagellar expression (26). Furthermore, coexpression of flagella and T3SS-1 is necessary for efficient invasion into host tissues (27, 28). Flagellar-based motility is fueled by the PMF. It is therefore possible that PprgH-gfp+ cells are more sensitive to PMF perturbation because flagellar biogenesis or activity depletes it more in this cell type.

We found that while SCFAs and CCCP reduced population-level PprgH-gfp expression and the proportion of PprgH-gfp+ cells in the nonflagellated mutant (Fig. 4 A and B), the reduction was less strong than that observed in wild-type cells (Fig. 4A). This suggests nonflagellated cells are less susceptible to PMF perturbation. This does, indeed, seem to be the case as nonflagellated cells maintain a higher PMF compared to wild-type cells in the presence of SCFAs and CCCP (Fig. 4C). However, expression of the flagella does not change the relative change in PMF by the two cell types (Fig. 4D), consistent with the observation that both cell types express flagellar machinery (29). Taken together, lack of flagellar activity decreases sensitivity to PMF perturbation overall but does not drive the differences between the two cell types.

Fig. 4.

Fig. 4.

Open in a new tab

Nonflagellated cells are less susceptible to PMF perturbation; however, flagellation does not explain the difference in PMF perturbation sensitivity between the two cell-types. (A) Mean ± SEM population-level PprgH-gfp fluorescence in nonflagellated cells in the presence of a range of SCFAs and CCCP concentrations. Dashed lines indicate mean values measured in wt, flagellated cells (SI Appendix, Fig. S4B). Average of three independent experiments; two-way ANOVA; no SCFAs versus SCFAs, P < 0.01; no SCFAs versus CCCP, P < 0.005; three-way ANOVA, including comparison to wild-type reporter cells (SI Appendix, Fig. S4B), PprgH-gfp versus PprgH-gfp nonflagellated no SCFAs versus SCFAs, P < 0.05. (B) Mean ± SEM proportion of PprgH-gfp+ in the nonflagellated mutant over time in 150 mM SCFAs, 12.5 μM CCCP, or 150 mM SCFAs and 12.5 μM CCCP. Two-way ANOVA followed by Tukey’s HSD; no SCFAs versus all other treatments, P < 0.01; Three-way ANOVA comparing flagellated and nonflagellated cells by treatment; PprgH-gfp versus PprgH-gfp nonflagellated, P < 0.05; interaction detected between strain and treatment, P < 0.05. (C) Mean Mitoview 633 ± SEM staining intensity by treatment and PprgH-gfp phenotype over time in nonflagellated cells. Dashed lines indicate mean values measured in wt, flagellated cells (Fig 3D). Three-way ANOVA followed by Tukey’s HSD; interaction detected between treatment and cell phenotype, P < 0.001; all timepoints, PprgH-gfp− no SCFAs versus all treatments, P > 0.1; t = 3, PprgH-gfp+ no SCFAs versus other treatments, P > 0.05; t = 5, PprgH-gfp+ no SCFAs versus CCCP and SCFAs + CCCP, P < 0.001, PprgH-gfp+ no SCFAs versus SCFAs, P > 0.05; t = 7, PprgH-gfp+ no SCFAs versus other treatments, P < 0.01. Four-way ANOVA to compare flagellated versus nonflagellated, P < 0.001; interactions detected between strain and treatment, P < 0.05 but not between cell type, strain, and treatment, P > 0.5. (D) Ratio of PprgH-gfp+/PprgH-gfp− Mitoview 633 staining intensity per condition and time point in nonflagellated cells. Dashed lines indicate mean ratios measured in wt, flagellated cells (Fig 3E). Two-way ANOVA followed by Tukey’s HSD, no SCFAs versus all other treatment, P < 0.01; SCFAs versus CCCP or SCFAs + CCCP, P < 0.05; CCCP versus SCFAs + CCCP, P > 0.2; three-way ANOVA to compare flagellated versus nonflagellated, P > 0.5.

Conclusions

Our study demonstrates SCFAs decrease population-level T3SS-1 expression by differentially impacting T3SS-1 cell-type behaviors. Although not the major driver, we find SCFAs influence the molecular regulation of prgH by reducing cell-type switching events. However, because SCFA-treated PprgH-gfp+ cells maintain expression across many generations and PprgH-gfp− cells initiate prgH expression, we conclude SCFAs do not act as a canonical transcriptional corepressor of T3SS-1 transcription. Our data collectively indicate SCFAs shape population-level T3SS-1 expression predominately by acting on the growth rate of PprgH-gfp+ cells.

These findings also have implications for how Salmonella causes disease. That SCFAs act on cell-type proportions indicates they shape the cooperative interactions between T3SS-1− and T3SS-1+ subpopulations during infection (15–17, 30). Because SCFA levels fluctuate as a function of nutrient intake, microbiota metabolism, and location in the gut, these dynamic conditions likely influence Salmonella cell-type interactions over time. High SCFA levels would lead to fewer inflammation-inducing T3SS-1+ cells and thus limit the expansion of T3SS-1− cells, while low SCFA levels would enhance inflammation induction by increasing the number of T3SS-1+ cells and therefore aid in Salmonella colonization of the gut. As only protonated SCFAs are taken up and SCFAs have a relatively high pKa, changes in pH values across the gut will likely also contribute to these cooperative interactions (25, 31). SCFA levels and their dynamics, therefore, play a role in balancing infection outcomes between colonization resistance, asymptomatic carriage, and symptomatic disease.

Lastly, there is strong support for the idea that T3SS-1+ and T3SS-1− cell types are distinct in ways other than T3SS-1 expression. Previous reports show these two cell-types have different growth rates, susceptibility to antibiotics, and cell sizes (17, 32). We add to this body of evidence of Salmonella differentiation by showing that T3SS-1+ and T3SS-1− cells differ in response to an environmental stimulus (SCFAs) and PMF. We speculate these reflect further specialization by each cell type to best fulfill their function. For example, the higher PMF of T3SS-1+ cells could allow for more-efficient effector secretion during interaction with host cells, whereas the maintenance of a high PMF by T3SS-1− cells would be unconducive for fast growth (33). Identifying distinguishing characteristics of the two cell types will shed light on the specializations each make to fulfill their function, including how T3SS-1+ cells prepare for their intracellular future.

As single-cell analyses have become more common, the importance of bacterial cell types during pathogenesis is becoming clear (15–17, 34–39). Combining these efforts with detailed molecular maps of each cell type will help guide the development of therapeutics which inhibit a single cell type of interest (e.g., virulent cells) rather than the whole population. Cell-type–targeted therapeutics could impede disease development while not selecting against a given species as only a subset of the cells would be susceptible (40, 41), a valuable strategy in the fight against antibiotic resistance.

Materials and Methods

Bacterial Cultivation.

Throughout the study, we used derivatives of SB300 (a spontaneous Streptomycin-resistant SL1344 derivative) listed in SI Appendix, Table S1 (13). Frozen strains were streaked onto Lysogeny Broth (LB) Miller agar for single colonies. After 24 h, a single colony was transferred to 5 mL LB Miller in a 15-mL round bottom tube and incubated at 37 °C with shaking at 200 rpm. A 16- to 18-h liquid culture prepared in this way was the starting point for all experiments. All experiments were performed in LB Miller (pH = 6.8) and, when applicable, supplemented with sodium acetate, sodium butyrate, and sodium propionate at appropriate concentrations (SI Appendix, Table S2).

Quantification of Population-Level Growth and Gene Expression.

We prepared a 96-well plate containing 2 μL overnight in 200 μL media containing or not SCFAs at indicated concentrations. Optical density at 600 nm and GFP fluorescence of cultures were measured every 3 min using a heated, automated microplate reader (Biotek Synergy). Note, the PprgH-gfp and PrpsM-gfp reporters are chromosomal, whereas the PsicA-gfp and PsipC-gfp reporters are plasmid based; the gain values were reduced in plasmid reporter strains to compensate for their inherently higher fluorescence. Three experiments were performed with each condition in triplicate. Blank (uninoculated) and autofluorescence (SB300) background control wells were run in each experiment. Resultant values were background subtracted by appropriate time-matched, control-well values. Maximum growth rate, lag time, yield, and max GFP fluorescence levels were determined by fitting a logistic equation to each time series using the GrowthCurver package in R (42).

Single-Cell Gene Expression Quantification Using Flow Cytometry.

Overnight cultures were diluted 1:100 into LB Miller ± SCFAs and incubated at 37 °C with shaking. At each timepoint, 10 mL culture was centrifuged (tabletop centrifuge, 3,000 × g, 15 min). Supernates were removed and pellets were resuspended in 1 mL phosphate buffered saline (PBS). Cells were centrifuged (microcentrifuge, 14,000 × g, 5 min), pellets resuspended in 1 mL 4% paraformaldehyde in PBS, and incubated at room temperature for 10 min. Fixed cells were pelleted, washed in 1 mL PBS, resuspended in 1 mL PBS, and stored at 4 °C. Fixed cell samples were analyzed by flow cytometry within 1 wk of collection. Forward scatter, side scatter, and GFP fluorescence of 50,000 events from each sample were measured using a Beckman Coulter Gallios flow cytometer. We performed three independent experiments.

Average per-cell gene expression (MFI) and subpopulation frequencies were calculated as follows. Distributions of GFP fluorescence values per sample were extracted from .fcs files using FlowCore (43). A test for bimodality was performed on each log10-transformed GFP fluorescence distribution (44). If distributions tested negativity for bimodality (e.g., all wild-type [wt] samples), a single normal distribution was fitted to the data. If distributions tested positively for bimodality (almost all reporter strain samples), the mixtools package was used to fit two normal distributions to the bimodal distribution (45). The mean of each distribution is the MFI of the subpopulation and the proportion of events assigned to each distribution using mixtools was used as the subpopulation frequency. The MFI of wt samples was used as a benchmark to determine which MFI represented the PprgH-gfp− subpopulation. There were no significant differences between wt and PprgH-gfp–reporter cell MFIs.

Mathematical Modeling and Simulations.

The simple differential equation was solved using Mathematica. Analytical solutions were run in MatLab. Gillespie simulations were run in RStudio. For more information, please see SI Appendix, Supporting Materials.

Feeding-Culture Microfluidics.

Microfluidic chips were fabricated using previously described methods (23, 32). All images were acquired with an automated Olympus IX81 inverted microscope using an oil-immersion 100× objective (Olympus), an ORCA-flash 4.0 version 2 sCMOS camera (Hamamatsu), X-Cite120 metal halide arc laboratory (Lumen Dynamics), and Chroma 49,000 fluorescent filter sets (Chroma, N49002).

A 16- to 18-h stationary culture of PprgH-gfp reporter cells were loaded into the microfluidic device. Cells were concentrated for loading into the chip by centrifuging 100 μL stationary phase culture, decanting excess media, and resuspending the cell pellet in the remaining culture media. To facilitate loading cells into observation channels, 1 μL 1% Tween-20 in PBS was added to the concentrated cells. The concentrated cells were then pipetted into the microfluidic device and examined by microscopy for sufficient loading. We then connected flasks of LB or LB + SCFAs (150 mM) to the microfluidic chip to feed the cells with a flow rate of 0.5 mL per hour.

To recapitulate the population-level experiments above, we used a feeding-culture approach. After fresh media was flowed through the device for ∼0.5 to 1 h (time used to set-up the automated time-lapse program for image acquisition), we inoculated the flasks feeding the microfluidic chip with stationary phase reporter cells at a ratio of 1:100. We then started image acquisition (t = 0). Phase contrast and GFP images were acquired for each position at 3-min intervals over 12 h.

Calculation of Growth Rates and Cell-Type Switching Probabilities.

Time-lapse microscopy images were segmented, tracked, and quantified using SuperSegger (46). Images were first deconvolved using a point-spread function (47). Segmentation of images (identification of individual cells) was performed using optimized segmentation constants to detect both PprgH-gfp+ and PprgH-gfp− cell types, which differ in size and curvature. All segmentation and tracking results were manually curated for erroneous boundary calling and tracking. For all lineages examined, we only tracked and quantified the “mother cell,” the cell at the closed end of the observation channel.

Cell-type switches were called by examining the pixel MFI of each mother cell over time. Per-lineage MFIs over the duration of the experiment were smoothed by fitting a local regression curve (α = 0.2) to reduce noise. The first derivative of the smoothed MFI trace was then examined to determine changes in PprgH-gfp levels: values of approximately 0 (±20, MFI change over 30 min) indicates equal GFP levels over time; values >20 indicate increases in MFI levels over time; and values <−20 indicate decreases in MFI levels over time. The time in which the derivative surpassed 20 or dropped below −20 was considered the time of cell-type switch. SI Appendix, Fig. S2D shows calling PprgH-gfp cell type by this method or by discriminating by cell size yields the same results in cell-type growth rates. In all data represented in this manuscript, we used GFP MFI changes over times as the discriminator of cell type in all experiments. SI Appendix, Fig. S2E shows tuning of the change in GFP over time parameters.

The time and direction of cell-type switch was then mapped to individual cells to call which cell type each cell was in at a given time. Manual inspection indicated good agreement between individual cell MFI levels and cell-type mapping using the above calling procedure. Growth rates of individual cells by cell-type assignment were then plotted and examined statistically using two-way ANOVA.

CCCP Treatment.

Reporter cells in a range of concentrations of CCCP were analyzed by 96-well plate assays as described in the Quantification of Population-Level Growth and Gene Expression section. The flow cytometric analyses were performed and analyzed as described in the Single-Cell Gene Expression Quantification Using Flow Cytometry section by adding 12.5 μM CCCP to the culture medium. The time-lapse microfluidic experiments were performed as described in the Feeding Culture Microfluidics section other than including 12.5 μM CCCP into a feeding culture.

Single-Cell PMF Measurements.

Reporter cells were grown in LB, LB + 150 mM SCFAs, LB + 150 mM NaCl (ionic control for SCFAs), LB + 12.5 µM CCCP, and LB + dimethyl sulfoxide (DMSO) (CCCP vehicle control) and sampled at 0, 3, 5, and 7 h postinoculation. At the time of sampling, cells were centrifuged to pellet (table-top centrifuge, 3,000 × g, 15 min) and resuspended in PBS containing MitoView 633 (Biotium). MitoView 633 accumulates in the inner membrane as a function of membrane potential. Higher fluorescence signal indicates higher membrane potential (i.e., a higher charge differential across the membrane). Sampled reporter cells were stained with MitoView-633 for 20 min on ice and then analyzed using flow cytometry to assign PprgH-gfp cell type and MFI of MitoView accumulation. Wt Salmonella cells (no reporter) were used as a negative control to confirm no bleed through of GFP or MitoView into the other fluorescence channels. All MitoView values were background subtracted with the time-matched, unstained, reporter strain control. At all timepoints, the addition of 150 mM NaCl and DMSO lead to an approximate 20% increase in MitoView accumulation; an expected observation as the addition of ions will impact the charge differential across the membrane. Thus, a 20% correction was applied to SCFA- and CCCP-treated cells to account for the impact of 150 mM extra ions or DMSO in the environment.

Supplementary Material

Supplementary File
Supplementary File
Download video file (11MB, mov)
Supplementary File
Download video file (11.9MB, mov)

Acknowledgments

We would like to express gratitude to past and present members of the Microbial Systems Ecology group for the helpful discussions during all phases of this work. This work was supported by SEED-23 19-2 to A.M.H. and Grant Nos. 51NF40_180575, 31003A-169978, and 310030-188642 from NCCR Microbiomes and the Swiss NSF to M.A.

Footnotes

The authors declare no competing interest.

This article is a PNAS Direct Submission.

This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.2103027118/-/DCSupplemental.

Data Availability

All study data are included in the article and/or supporting information.

Change History

December 17, 2021: The Acknowledgments have been updated.

References

  • 1.Cummings J. H., Pomare E. W., Branch H. W. J., Naylor E., Macfarlane G. T., Short chain fatty acids in human large intestine, portal, hepatic and venous blood. Gut 28, 1221–1227 (1987). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Filkins L. M., et al., Prevalence of streptococci and increased polymicrobial diversity associated with cystic fibrosis patient stability. J. Bacteriol. 194, 4709–4717 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Ríos-Covián D., et al., Intestinal short chain fatty acids and their link with diet and human health. Front. Microbiol. 7, 185 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Zoetendal E. G., et al., The human small intestinal microbiota is driven by rapid uptake and conversion of simple carbohydrates. ISME J. 6, 1415–1426 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.LaRock D. L., Chaudhary A., Miller S. I., Salmonellae interactions with host processes. Nat. Rev. Microbiol. 13, 191–205 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Galán J. E., Curtiss R. III, Cloning and molecular characterization of genes whose products allow Salmonella typhimurium to penetrate tissue culture cells. Proc. Natl. Acad. Sci. U.S.A. 86, 6383–6387 (1989). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Wotzka S. Y., Nguyen B. D., Hardt W.-D., Salmonella typhimurium diarrhea reveals basic principles of enteropathogen infection and disease-promoted DNA exchange. Cell Host Microbe 21, 443–454 (2017). [DOI] [PubMed] [Google Scholar]
  • 8.McLaughlin P. A., et al., Inflammatory monocytes provide a niche for Salmonella expansion in the lumen of the inflamed intestine. PLoS Pathog. 15, e1007847 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Stecher B., et al., Salmonella enterica serovar typhimurium exploits inflammation to compete with the intestinal microbiota. PLoS Biol. 5, 2177–2189 (2007). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Winter S. E., et al., Gut inflammation provides a respiratory electron acceptor for Salmonella. Nature 467, 426–429 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Thiennimitr P., et al., Intestinal inflammation allows Salmonella to use ethanolamine to compete with the microbiota. Proc. Natl. Acad. Sci. U.S.A. 108, 17480–17485 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Maier L., et al., Microbiota-derived hydrogen fuels Salmonella typhimurium invasion of the gut ecosystem. Cell Host Microbe 14, 641–651 (2013). [DOI] [PubMed] [Google Scholar]
  • 13.Hautefort I., Proença M. J., Hinton J. C. D., Single-copy green fluorescent protein gene fusions allow accurate measurement of Salmonella gene expression in vitro and during infection of mammalian cells. Appl. Environ. Microbiol. 69, 7480–7491 (2003). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Bumann D., Examination of Salmonella gene expression in an infected mammalian host using the green fluorescent protein and two-colour flow cytometry. Mol. Microbiol. 43, 1269–1283 (2002). [DOI] [PubMed] [Google Scholar]
  • 15.Diard M., et al., Stabilization of cooperative virulence by the expression of an avirulent phenotype. Nature 494, 353–356 (2013). [DOI] [PubMed] [Google Scholar]
  • 16.Ackermann M., et al., Self-destructive cooperation mediated by phenotypic noise. Nature 454, 987–990 (2008). [DOI] [PubMed] [Google Scholar]
  • 17.Sturm A., et al., The cost of virulence: Retarded growth of Salmonella Typhimurium cells expressing type III secretion system 1. PLoS Pathog. 7, e1002143 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Wotzka S. Y., et al., Escherichia coli limits Salmonella Typhimurium infections after diet shifts and fat-mediated microbiota perturbation in mice. Nat. Microbiol. 4, 2164–2174 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Lawhon S. D., Maurer R., Suyemoto M., Altier C., Intestinal short-chain fatty acids alter Salmonella typhimurium invasion gene expression and virulence through BarA/SirA. Mol. Microbiol. 46, 1451–1464 (2002). [DOI] [PubMed] [Google Scholar]
  • 20.Gantois I., et al., Butyrate specifically down-regulates Salmonella pathogenicity island 1 gene expression. Appl. Environ. Microbiol. 72, 946–949 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Rivera-Chávez F., et al., Depletion of butyrate-producing Clostridia from the gut microbiota drives an aerobic luminal expansion of Salmonella. Cell Host Microbe 19, 443–454 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Jacobson A., et al., A gut commensal-produced metabolite mediates colonization resistance to Salmonella infection. Cell Host Microbe 24, 296–307.e7 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Moreno-Gámez S., et al., Wide lag time distributions break a trade-off between reproduction and survival in bacteria. Proc. Natl. Acad. Sci. U.S.A. 117, 18729–18736 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Zhang Z. J., Pedicord V. A., Peng T., Hang H. C., Site-specific acylation of a bacterial virulence regulator attenuates infection. Nat. Chem. Biol. 16, 95–103 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Wolfe A. J., The acetate switch. Microbiol. Mol. Biol. Rev. 69, 12–50 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Singer H. M., Kühne C., Deditius J. A., Hughes K. T., Erhardt M., The Salmonella Spi1 virulence regulatory protein HilD directly activates transcription of the flagellar master operon flhDC. J. Bacteriol. 196, 1448–1457 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Hausmann A., et al., Intestinal epithelial NAIP/NLRC4 restricts systemic dissemination of the adapted pathogen Salmonella Typhimurium due to site-specific bacterial PAMP expression. Mucosal Immunol. 13, 530–544 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Furter M., Sellin M. E., Hansson G. C., Hardt W.-D., Mucus architecture and near-surface swimming affect distinct Salmonella Typhimurium infection patterns along the murine intestinal tract. Cell Rep. 27, 2665–2678.e3 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Sánchez-Romero M. A., Casadesús J., Single cell analysis of bistable expression of pathogenicity island 1 and the flagellar regulon in Salmonella enterica. Microorganisms 9, 210 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Sánchez-Romero M. A., Casadesús J., Contribution of SPI-1 bistability to Salmonella enterica cooperative virulence: Insights from single cell analysis. Sci. Rep. 8, 14875 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Pinhal S., Ropers D., Geiselmann J., de Jong H., Acetate metabolism and the inhibition of bacterial growth by acetate. J. Bacteriol. 201, e00147-19 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Arnoldini M., et al., Bistable expression of virulence genes in Salmonella leads to the formation of an antibiotic-tolerant subpopulation. PLoS Biol. 12, e1001928 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Erhardt M., Mertens M. E., Fabiani F. D., Hughes K. T., ATPase-independent type-III protein secretion in Salmonella enterica. PLoS Genet. 10, e1004800 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Davis K. M., Mohammadi S., Isberg R. R., Community behavior and spatial regulation within a bacterial microcolony in deep tissue sites serves to protect against host attack. Cell Host Microbe 17, 21–31 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Davis K. M., Isberg R. R., Defining heterogeneity within bacterial populations via single cell approaches. BioEssays 38, 782–790 (2016). [DOI] [PubMed] [Google Scholar]
  • 36.Davis K. M., For the greater (bacterial) good: Heterogeneous expression of energetically costly virulence factors. Infect. Immun. 88, e00911-19 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Rego E. H., Audette R. E., Rubin E. J., Deletion of a mycobacterial divisome factor collapses single-cell phenotypic heterogeneity. Nature 546, 153–157 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Diard M., et al., Antibiotic treatment selects for cooperative virulence of Salmonella typhimurium. Curr. Biol. 24, 2000–2005 (2014). [DOI] [PubMed] [Google Scholar]
  • 39.Ronin I., Katsowich N., Rosenshine I., Balaban N. Q., A long-term epigenetic memory switch controls bacterial virulence bimodality. eLife 6, 7808–7818 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Rasko D. A., Sperandio V., Anti-virulence strategies to combat bacteria-mediated disease. Nat. Rev. Drug Discov. 9, 117–128 (2010). [DOI] [PubMed] [Google Scholar]
  • 41.Bell G., MacLean C., The search for, “evolution-proof” antibiotics. Trends Microbiol. 26, 471–483 (2018). [DOI] [PubMed] [Google Scholar]
  • 42.Sprouffske K., Wagner A., Growthcurver: An R package for obtaining interpretable metrics from microbial growth curves. BMC Bioinformatics 17, 172 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Hahne F., et al., flowCore: A Bioconductor package for high throughput flow cytometry. BMC Bioinformatics 10, 106 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Maechler M., Package ‘diptest’ Title Hartigan’s Dip Test Statistic for Unimodality-Corrected. https://CRAN.R-project.org/package=diptest. Accessed 5 February 2021.
  • 45.Benaglia T., Chauveau D., Hunter D. R., Young D., mixtools: An R package for analyzing finite mixture models. J. Stat. Softw. 32, 1–29 (2009). [Google Scholar]
  • 46.Stylianidou S., Brennan C., Nissen S. B., Kuwada N. J., Wiggins P. A., SuperSegger: Robust image segmentation, analysis and lineage tracking of bacterial cells. Mol. Microbiol. 102, 690–700 (2016). [DOI] [PubMed] [Google Scholar]
  • 47.van Vliet S., et al., Spatially correlated gene expression in bacterial groups: The role of lineage history, spatial gradients, and cell-cell interactions. Cell Syst. 6, 496–507.e6 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary File
pnas.2103027118.sapp.pdf (3.4MB, pdf)
Supplementary File
Download video file (11MB, mov)
Supplementary File
Download video file (11.9MB, mov)

Data Availability Statement

All study data are included in the article and/or supporting information.

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences

RESOURCES