Summary
Many bacterial species use the MecA/ClpCP proteolytic system to block entry into genetic competence. In Streptococcus mutans, MecA/ClpCP degrades ComX (also called SigX), an alternative sigma factor for the comY operon and other late competence genes. Although the mechanism of MecA/ClpCP has been studied in multiple Streptococcus species, its role within noisy competence pathways is poorly understood. S. mutans competence can be triggered by two different peptides, CSP and XIP, but it is not known whether MecA/ClpCP acts similarly for both stimuli, how it affects competence heterogeneity, and how its regulation is overcome. We have studied the effect of MecA/ClpCP on the activation of comY in individual S. mutans cells. Our data show that MecA/ClpCP is active under both XIP and CSP stimulation, that it provides threshold control of comY, and that it adds noise in comY expression. Our data agree quantitatively with a model in which MecA/ClpCP prevents adventitious entry into competence by sequestering or intercepting low levels of ComX. Competence is permitted when ComX levels exceed a threshold, but cell-to-cell heterogeneity in MecA levels creates variability in that threshold. Therefore MecA/ClpCP provides a stochastic switch, located downstream of the already noisy comX, that enhances phenotypic diversity.
Introduction
Many species of streptococci can become naturally transformable by entering the transient physiological state known as genetic competence (Fontaine et al., 2014; Johnston et al., 2014). Competence plays a particularly important role for the oral pathogen Streptococcus mutans, influencing cell growth, death, interactions with other members of the oral flora and expression of known virulence traits. Bacteriocin production, biofilm formation, acid production and tolerance of acid and oxidative stresses by S. mutans all facilitate the competition, persistence and virulence of this organism in the human oral biofilm environment (J. A. Lemos; Burne, 2008). All of these traits are linked to the expression of ComX (also called SigX), an alternative sigma factor that activates competence genes required for DNA uptake and processing. ComX production is controlled by a pathway that integrates signals received from two quorum sensing peptides (Shanker; Federle, 2016) with environmental cues such as pH (Guo et al., 2014; Son et al., 2015b) and oxygen and reactive oxygen species (De Furio et al., 2017), intracellular noise (stochasticity) and positive and negative feedback (Smith; Spatafora, 2012; LeungDufour et al., 2015; Reck et al., 2015; Son et al., 2015a; Hagen; Son, 2017). As a result, S. mutans competence is a complex and heterogeneous behavior that can be exquisitely sensitive to the extracellular environment and that remains incompletely understood.
Population heterogeneity in S. mutans competence is evident from the low efficiency of natural genetic transformation (Y. Li et al., 2001), as well as from observations of cell-to-cell variability in comX gene expression (Lemme et al., 2011; Son et al., 2012; Reck et al., 2015; Hagen; Son, 2017). Transformation efficiency in biofilms is typically less than 0.1% (Y. Li et al., 2001), while even under very favorable conditions no more than 10-50% of cells naturally express comX (Lemme et al., 2011; Son et al., 2012). In addition, the expression of comX can be bimodal or unimodal in the population, depending on the exogenous signals present, the growth phase and the environment (Son et al., 2012; Shields; Burne, 2016). Post-translational regulation of ComX also appears to generate heterogeneity, as high levels of comX mRNA do not assure robust activation of comY (Seaton et al., 2011). As with many other bacterial regulatory proteins (Inobe; Matouschek, 2008), ComX levels in S. mutans are modulated post-translationally by an ATP-dependent protease system composed of MecA and ClpCP (Tian et al., 2013; Dong et al., 2014; Dufour et al., 2016). The MecA/ClpCP complex inhibits competence by targeting and degrading ComX, as it does in streptococci of the salivarius, mitis and pyogenic groups (Biornstad; Havarstein, 2011; Boutry et al., 2012; Wahl et al., 2014; Y. H. Li; Tian, 2017). However, the function of MecA/ClpCP within the S. mutans competence pathway, and particularly its role in cell-to-cell heterogeneity and the bimodal and unimodal competence behaviors, has not been explored in detail.
Figure 1 summarizes the competence regulatory pathway in S. mutans (Smith; Spatafora, 2012; Tian et al., 2013; Shanker; Federle, 2016). ComX activates late competence genes that include the nine-gene operon comYA-I, which contains seven genes that are required for transformation (Merritt et al., 2005).Transcription of comX can be triggered by either of two quorum sensing peptides: CSP (competence stimulating peptide) or XIP (SigX-inducing peptide). The efficacy of these peptides is sensitive to environmental factors, including pH, oxidative stress, carbohydrate source, and the peptide content of the medium.
CSP is derived from the ComC precursor, processed to a final length of 18 aa and exported to the extracellular medium. S. mutans detects CSP through the ComDE two-component signal transduction system (TCS), which directly activates multiple genes involved in bacteriocin biogenesis, secretion and immunity. However, S. mutans ComDE does not directly activate comX. Instead, the ComRS system is the immediate regulator of comX in the mutans, salivarius, bovis, and pyogenes groups of streptococci (Mashburn-Warren et al., 2010). The ComRS system consists of the cytosolic receptor ComR and the 17-aa peptide ComS, which is processed by an unknown mechanism to form the 7-aa XIP. Extracellular XIP is imported by the oligopeptide permease Opp and interacts with ComR to form a complex that activates the transcription of comS and comX. Exogenous XIP induces comX efficiently in chemically defined media lacking small peptides (such as FMC or CDM (Mashburn-Warren et al., 2010; Son et al., 2012)), leading to population-wide induction of comX at saturating XIP levels. However, XIP elicits no induction of comX in complex growth media containing small peptides, possibly owing to peptide competition with XIP for uptake by Opp. Interestingly, the CSP peptide signal has a different action than XIP, as it activates S. mutans comX only in complex growth media containing small peptides. It elicits no activity from comX in defined media that lacks small peptides, even though CSP stimulates the ComDE TCS (leading to bacteriocin production) under these conditions. In addition, the comX response to CSP is bimodal in the population, with no more than 50% of cells expressing comX at saturating CSP concentrations (Son et al., 2012).
Consequently, the activation of comX in a population of S. mutans can exhibit two types of heterogeneity: a unimodal distribution when stimulated by exogenous XIP and a bimodal distribution when stimulated by exogenous CSP. Only the bimodal behavior requires an intact comS, whereas only the unimodal behavior requires the oligopeptide permease opp. We previously posited that these different behaviors are two modes of operation of the transcriptional feedback loop associated with comS, which encodes its own inducing signal. In the unimodal case the cells import and respond to exogenous XIP, whereas in the bimodal case XIP import is blocked, leaving each cell to respond to its intracellular ComS (or XIP). The first mode allows a generally uniform, population-wide activation of comX, but the second mode leads to noisy, positive feedback dynamics in both comS and comX (Son et al., 2012; Hagen; Son, 2017).
The mechanism of posttranslational control of ComX by MecA/ClpCP in S. mutans resembles that in pyogenic and salivarius streptococci, to which S. mutans MecA is closely homologous (Boutry et al., 2012; Wahl et al., 2014). S. mutans MecA is a 240 aa adapter protein that interacts with ComX and ClpC to form a ternary complex that sequesters ComX and targets it for ATP-dependent degradation by the ClpP protease (Tian et al., 2013; Dong et al., 2014). MecA/ClpCP similarly controls the master competence regulators ComW in S. pneumoniae (Wahl et al., 2014) and ComK in Bacillus subtilis (Turgay et al., 1998). In B. subtilis MecA was shown to facilitate the ATP-dependent formation of the ClpCP proteolytic complex, which unfolds and degrades both MecA and its ComK target, and then itself dissociates (Mei et al., 2009; Liu et al., 2013). Therefore MecA/ClpCP operates dynamically by continuously turning over MecA as well as its regulatory target if present.
Several studies in S. mutans have established that MecA/ClpCP suppresses the activation of comY under CSP stimulation, in complex media (Tian et al., 2013; Dong et al., 2014; Dufour et al., 2016). Deletion of mecA, clpC, or clpP increased ComX levels and transformability during growth in complex media and also prolonged the competent state. These studies imply that MecA/ClpCP serves either to suppress S. mutans competence or to switch it off as growth progresses, in complex media. Some studies have found the puzzling result that deletion of mecA or clpCP caused a weaker increase in ComX levels or transformability - or even had no effect at all – in chemically defined media (with added XIP) than in complex media (with CSP) (Boutry et al., 2012; Tian et al., 2013; Dong et al., 2014; Dufour et al., 2016). A subsequent study found that MecA deletion improved S. salivarius transformability in defined media, although the difference was attenuated at high levels of XIP stimulation (Wahl et al., 2014).
The possible significance of growth media and the presence of heterogeneity raise the question of how MecA/ClpCP functions within the full competence pathway, in which the XIP and CSP signaling pathways activate comX in defined and complex media respectively. Although it seems clear that MecA/ClpCP inhibits comY expression by sequestering and degrading ComX, a clearer model of how this regulation integrates with the known comX activation pathway, and how it may be overcome when competence is permitted, is still needed. Additional cell density signals (Dufour et al., 2016), as well as XIP-dependent feedback or additional gene products (Wahl et al., 2014), have been proposed as mechanisms for modulating ComX levels via MecA/ClpCP. We have used a single-cell, microfluidic approach to clarify some of these questions and to develop an explicit model of how MecA/ClpCP interacts with the noisy and bimodal mechanisms controlling S. mutans comX. Our data lead to a simple quantitative model that reproduces both the population average behavior and the cell-to-cell heterogeneity in comY activation.
Results
MecA/ClpCP affects transformation efficiency of S. mutans induced by XIP
Supporting Figure S4 shows how deletions in the MecA/ClpCP system affect transformation efficiency of S. mutans UA159. Transformability was measured in cells cultured in defined medium (FMC) containing various concentrations of XIP, as indicated. At the highest XIP concentration (1 μM), the transformation efficiencies of the mecA and clpC deletion mutants were similar to the wild type. This finding is consistent with previous reports for S. mutans (Tian et al., 2013; Dong et al., 2014) and S. thermophilus (Boutry et al., 2012), where little or no effect of mecA deletion on transformability in response to XIP was observed. However the behavior of the mutants diverged at lower XIP concentrations, where deficiency of MecA or ClpCP enhanced transformability. An effect of XIP concentration on the behavior of deletion mutants was also reported for S. salivarius (Wahl et al., 2014). Both ΔclpP and ΔclpC had higher transformation efficiency than the wild type at 10 nM XIP. Surprisingly the mecA deletion showed lower transformability than the wild type strain at 100 nM XIP; we note however this strain grows poorly and displays defects in cell morphology and decreased viability. Overall these data confirm that the MecA/ClpCP system interacts with XIP induction of transformability in defined medium. To obtain more detailed insight into XIP stimulation, MecA/ClpCP and comY activation, we turned to individual cell studies.
Activation of comX leads to heterogeneous induction of comY
We used dual fluorescent reporters (PcomX-gfp, PcomY-rfp) to compare the activation of PcomX and PcomY in individual S. mutans supplied with exogenous XIP. Figure 2A shows S. mutans UA159 growing in microfluidic channels under a constant flow of defined medium (FMC) that contains 0-2 µM XIP. PcomX is activated in all cells if the XIP concentration exceeds about 100 nM, and its activation saturates as XIP exceeds about 800 nM. However, very few cells activate PcomY at XIP concentrations of 400 nM or less, and cells that do activate PcomY vary widely in their red fluorescence intensity. Even at 1-2 μM XIP, many cells exhibit little PcomY activity.
Figures 2B-2C show the statistical distribution of PcomX (GFP, upper rows) and PcomY (RFP, lower rows) reporter fluorescence for cells in response to exogenous XIP or CSP. Reporter fluorescence was imaged while cells grew in microfluidic channels under continuous flow of defined medium for XIP (Figure 2B), or of complex medium for CSP (Figure 2C). As previously reported (Son et al., 2012), XIP in defined medium elicits a noisy but generally unimodal (population-wide) comX response. In contrast, CSP in complex medium elicits a much noisier, bimodal (double peaked distribution) comX response. For both CSP and XIP stimulation, the response of PcomY is highly heterogeneous. Even the highest concentrations of CSP and XIP, which saturate the response of PcomX, incompletely activate PcomY in the population; the PcomY expression levels in individual cells span 2-3 orders of magnitude above the baseline. These data suggest that post-translational regulation of ComX increases cell-to-cell heterogeneity in comY expression, which adds to the noise in the comX response to CSP or XIP stimulation.
The MecA/ClpCP system inhibits the comY response to XIP and increases its noise
To test whether the MecA/ClpCP proteolytic system affects ComX function in defined medium, and to assess its effect on noise in comY expression, we compared comX and comY expression in dual reporter strains in the wild-type (UA159) and ΔmecA genetic backgrounds. Figure 3 shows PcomY activity in individual cells that were stimulated by XIP in planktonic culture in defined medium and then imaged on glass slides. Similar results were obtained for cells growing in microfluidic flow channels. Deletion of mecA altered the PcomY response in two ways. First, the ΔmecA strain responded more strongly to XIP than did the wild type. Unlike the wild-type genetic background, the ΔmecA cells showed high median PcomY expression, exceeding the baseline level at XIP concentrations greater than about 200 nM. Second, deletion of mecA reduced noise in comY expression (Figure 3C, 3D). Although comY and comX expression correlated positively in UA159, the correlation was partially obscured by the noisy behavior of comY. In contrast, comY expression increased systematically as comX expression increased in the ΔmecA strain. Despite some noise in comY, a roughly proportional relationship can be discerned in the data of Figure 3D, but not in Figure 3B. (The upward curvature in Figure 3D results from the logarithmic horizontal axis.) The nearly linear correlation between comY and comX in the ΔmecA mutant suggests that, in the absence of MecA/ClpCP, ComX activates comY in a direct and predictive fashion.
The effect of MecA on noise in comY is also seen in histograms of comY expression at given comX expression levels. Supporting Figure S2 shows comY histograms for cells growing in microfluidic channels with flowing defined medium and XIP, binned according to their comX activity. Both at high and low comX activity, the shape of the comY histograms is qualitatively different in the two strains. The deletion of mecA qualitatively alters the relationship between comY and comX expression in defined medium with addition of XIP.
comY and comX expression are simply correlated in the absence of MecA
As is common for bacterial protein expression (Taniguchi et al., 2010), the histograms of PcomY expression (Supporting Figure S2) resemble a gamma distribution Γ(n | A,B), a two-parameter continuous probability distribution that can be interpreted in terms of sequential, stochastic processes of transcription and translation (see Methods). This finding, together with the roughly linear correlation between comY and comX activity in the ΔmecA strain (Figure 3D), motivates a simple mathematical model for comX/comY in the absence of MecA/ClpCP. The model is described in the Methods: comY is activated in a mostly linear (but saturating) fashion by comX on average, but is also subject to stochasticity. The comY activity in a given cell is thus a random variable drawn from a gamma distribution whose parameters are determined by the PcomX activity in the cell. The model has four parameters, which we obtained through a maximum likelihood fit to the ΔmecA individual cell RFP and GFP fluorescence data of Figure 3D. We then used these parameters to generate a stochastic simulation of the model for comparison to the data.
Figure 4 compares the ΔmecA experimental data (Figure 4A, 4B) with a simulation of the model (Figure 4C, 4D). The model accurately reproduces both the population-averaged comY response and its cell-to-cell variability. This result indicates that in the absence of MecA/ClpCP regulation of ComX, comY can be modeled as a typical noisy gene whose average activation is proportional to the concentration of active ComX protein.
A plausible alternative model is that extracellular XIP concentration, rather than PcomX activity per se, controls comY expression in ΔmecA. The simulation shown in Supporting Figure S3 indicates that the best fit of this model significantly overestimates the noise in PcomY. In short, modeling suggests that the PcomX activity of a ΔmecA cell is a straightforward predictor of its PcomY activity, and is also a better predictor than is the XIP concentration.
Different deletions in MecA/ClpCP produce different noise and threshold behaviors in comY
To determine which elements of the MecA/ClpCP system affect sensitivity and noise in comY, we measured PcomY and PcomX activity in the UA159, ΔmecA, ΔclpC and ΔclpP genetic backgrounds (Figure 5). All strains carried the dual fluorescent reporters and were imaged in microfluidic chambers while supplied with a continuous flow of defined medium containing XIP. In all strains, PcomY was more strongly activated at higher XIP concentrations where PcomX expression was higher, although noise and sensitivity varied among the different strains (Figure 5A). All strains showed a similar dependence of PcomX activity (GFP) on XIP concentration (Figure 5B). In the relation between comY and comX expression, the UA159 (wild type) showed a more pronounced threshold in the onset of comY activation, at a comX level near 300 units, and much greater noise in comY expression. The clpP deletion strain, in which the MecA/ClpC complex can presumably bind, but not degrade, ComX, showed slightly less noisy comY expression than the wild type and comY was somewhat more readily activated. Deletion of clpC, or especially mecA, reduced comY noise significantly, such that the population was almost uniformly activated when PcomX expression was strong, near 1 μM XIP. Therefore, the interaction between MecA/ClpC and ComX, as well as the proteolytic action of ClpP on that complex, contribute to noise in comY expression and also suppress the ability of comX expression to elicit the comY response. Similar data were obtained when cells were grown in static medium and image while dispersed on glass slides.
The role of MecA alone can be modeled by simple sequestration of ComX
A detailed model for the regulation of ComX by MecA/ClpCP must include the formation of the MecA/ClpC/ComX ternary complex, as well as the kinetics of ComX and MecA degradation by ClpP. Both of these mechanisms are absent in the ΔclpC strain, although the binary interaction of MecA with ComX is present. Therefore, we tested whether a binary sequestration (MecA + ComX) model could reproduce our data for the activation of comY by ComX in the ΔclpC strain. In this model, described in Methods, individual ComX molecules are presumed to be tightly sequestered by individual MecA molecules, leaving them unavailable to stimulate comY transcription. Then the probability distribution for the comY expression of a cell becomes determined not by its comX activity alone, but by the excess of ComX over MecA copy numbers. We modeled the MecA copy number as a random variable drawn from a gamma probability distribution; the activation of comY by the available (unsequestered) ComX is modeled as in Figure 4. The MecA probability distribution is presumed to be independent of XIP, consistent with our mRNA measurements showing no effect of XIP on mecA, clpC or clpP expression (Supporting Table ST1). Fitting this MecA model to the ΔclpC data therefore requires only a two-parameter fit for the gamma distribution parameters, which we obtained by maximum likelihood comparison of the data and model.
Figure 6 compares the ΔclpC data with a stochastic simulation of this model. The comY - comX correlation closely resembles the experimental data, both in its average trend and its noise. These results show that the higher comY expression noise that is observed in the ΔclpC strain, compared to the ΔmecA strain, is consistent with a mechanism where MecA suppresses comY response by sequestering ComX. Fitting the model to the data provides the probability distribution of the MecA copy number, Figure 6C, where MecA is measured in units of equivalent PcomX activity. Cell-to-cell variability in MecA copy number is then a source of variability in comY expression.
CSP and XIP stimulation produce similar correlations between comX and comY activation
Previous studies have demonstrated that deletions of mecA or clpCP enhance comY expression upon stimulation with CSP in complex media (Tian et al., 2013; Dong et al., 2014). Our data show with single cell resolution that the same deletions also affect the response to XIP in defined media. These findings raise the question of whether, in the presence of MecA/ClpCP, the activation of comY by ComX may be similar regardless of how comX transcription is induced, whether by XIP or CSP. Figure 7 compares single cell measurements of comX and comY activity with CSP and XIP respectively. Precise quantitative comparison of the two response curves is complicated by the stronger green auto-fluorescence of cells in complex medium, which shifts the horizontal axis of the CSP data. Further, CSP appears to induce a slightly noisier comY response than does XIP, possibly in connection with feedback behavior in the ComDE system (Son et al., 2015a). However the data verify a generally similar behavior in both conditions: comY responds in threshold fashion to activation of comX, and comY activation is highly heterogeneous in the population, even among cells with the highest comX activity.
Discussion
The MecA/ClpCP proteolytic system is well conserved as a negative regulator of genetic competence across streptococcal groups and in other naturally competent species, including B. subtilis (Liu et al., 2013). However, while mechanistic studies of MecA/ClpCP have provided a clear description of its action, they have not fully resolved the question of how MecA/ClpCP contributes to competence regulation. Several authors have proposed that MecA/ClpCP serves either to suppress or terminate the competent state. For B. subtilis, Turgay et al. proposed that MecA/ClpCP degradation of the ComK competence regulator provides a ‘timing’ function by limiting synthesis of the auto-activating ComK regulator, thus permitting escape from the competent state (Turgay et al., 1998). Dufour et al. proposed a similar model for S. mutans, in which the sequestration and degradation of free ComX by MecA/ClpCP forces an exit from the competent state late in growth, when the transcription of comX is repressed (Dufour et al., 2016). Wahl et al. proposed that S. salivarius MecA/ClpCP serves a ‘locking’ function, preventing the cell from entering the competent state under inappropriate conditions, such as early in the growth phase (Wahl et al., 2014). Wahl et al. argued that at low XIP concentrations proteolytic degradation of ComX prevents competence, but that high XIP concentrations may alleviate this repression, possibly by overwhelming the proteolytic capacity or by activating another, unidentified gene product.
Both the ‘locking’ and ‘timing’ models interpret MecA/ClpCP as a mechanism for suppressing activation of comY when comX expression is weak. Our data are consistent with this description. Moreover, our data show that this suppression can be described by the simplest model in which an intracellular pool of MecA intercepts available ComX, sequestering it and blocking its otherwise straightforward activation of comY. Such a model quantitatively fits the data on the clpC mutant, in which MecA can sequester ComX but clpP proteolysis is absent. If the MecA copy number obeys a gamma probability distribution, as is common for bacterial proteins, then the model reproduces both the average relationship between comY and comX expression and the cell-to-cell variability in that expression. Therefore, the response of comY in individual clpC and mecA cells can be understood solely in terms of the PcomX activity and MecA copy number distribution. The behavior of the late competence genes in these mutants can be understood without positing any role for XIP other than as a stimulus for PcomX.
In addition, our single cell data show that the MecA/ClpCP system substantially enhances the noise (cell-to-cell heterogeneity) in comY expression when comX is activated. Even at high XIP concentrations that saturate comX expression, comY expression levels within the UA159 population span a range extending three orders of magnitude above the baseline; by contrast, the deletion mutants all express comY with far less noise at high XIP concentrations. Our modeling indicates that cell-to-cell variability in the MecA copy number in wild type cells, together with the proteolytic action of ClpP (which reduces MecA and ComX copy numbers) adds to noise that is generated upstream by the pathways that activate comX. The resulting noisy threshold effect is very similar to the toxin/antitoxin competition that generates phenotypic heterogeneity in bacterial persistence (Rotem et al., 2010), or to a sequestration-induced threshold model for non-linear gene regulation (Buchler; Cross, 2009).
A clear understanding of the role of MecA/ClpCP has perhaps been complicated by early reports that deletion of mecA or clpC increased transformability or ComX protein levels under CSP stimulation (in complex medium), but not under XIP stimulation (in defined medium). Our data confirm in detail that the MecA/ClpCP system affects signaling from comX to comY in defined medium. In fact, as the sequestration model described above is indifferent to whether comX is stimulated by exogenous XIP or CSP, we expect that signaling from comX to comY should be similar in both CSP/complex medium and in XIP/defined medium. Figure 7 suggests that the relationship is very similar.
This finding suggests that the MecA/ClpCP system acts continuously to suppress ComX levels, regardless of the extracellular inputs driving comX expression. A model where MecA/ClpCP performs this task in relatively steady fashion is consistent with findings that S. mutans MecA and ClpCP protein levels did not differ in complex and defined medium (Dong et al., 2014), that MecA induction showed little change during S. suis competence (Zaccaria et al., 2016), and that S. mutans mecA/clpCP mRNA levels are insensitive to XIP inputs (Supporting Table ST1). Thus competence will be suppressed when comX is weakly expressed due to insufficient CSP or XIP early in growth (‘locking’ behavior). Competence will also be suppressed when comX is weakly expressed late in growth due to inefficient CSP/XIP signaling. Falling extracellular pH late in the growth phase suppresses competence signaling by CSP and XIP (Guo et al., 2014; Son et al., 2015b), which may allow MecA/ClpCP to shut down the competent state (‘timing behavior’).
Consequently the sequestration mechanism can provide both ‘timing’ and ‘locking’ functions. The simulations in Figure 4 and Figure 6 are based on simple equilibrium models that address only the effects of sequestration by MecA on the pool of free ComX, omitting the kinetic effects of ClpP unfolding and degradation of ComX and MecA. A model that includes ClpP proteolysis is much more complicated, as it must include the sequential binding steps that are associated with the formation of the ternary complex, binding of ClpP, and the breakdown of both MecA and ComX. The binding and kinetic parameters of the model cannot be determined from our data; however we can construct a reasonably tractable model for the full system by simplifying the complex regulatory mechanism that is outlined in the literature (Mei et al., 2009). Supporting Figure S7 describes a simplified kinetic model that can rationalize some of the observations in our data, including the finding that deletion of clpC or clpP did not eliminate the comX threshold that is required for comY activation, and that only the mecA deletion eliminated the threshold and sharply reduced the noise in comY. Supporting Figure S7 shows that simulations from such rough models can reproduce key differences in comX-comY threshold behavior observed among the mutants studied here.
We note that a MecA copy number distribution that has higher mean but is narrower than that of Figure 6C would still provide the same ‘timing’ or ‘locking’ function without introducing as much noise in comY. The evident width of the distribution therefore suggests that the organism may benefit from greater noise. The competence pathway in S. mutans is linked to several stress-induced behaviors that are heterogeneous in the population, including competence, lysis and a persister phenotype (Perry et al., 2009; LeungAjdic et al., 2015; Leung et al., 2015). A link between quorum controlled behavior and phenotypic heterogeneity has often been noted in bacterial gene regulation. In other organisms, such as B. subtilis, complex pathways that integrate intracellular and extracellular signaling mechanisms with stochastic gene expression often generate phenotypic heterogeneity, distributing stress response behaviors such as competence and sporulation among different individuals in the population (Grote et al., 2015). Interestingly, propidium iodide staining of individual S. mutans indicates that comX-driven lysis is decoupled from comX-driven competence (Supporting Figure S5). While higher comX expression increases the probability of cell lysis, the most highly expressing cells (which are more likely to express comY) actually show less evidence of lysis. Accordingly, the MecA/ClpCP system may provide a bet-hedging advantage to an S. mutans population by providing an additional, stochastic switching point in the regulatory pathway from stress conditions to transformability.
Our data show that the action of the S. mutans MecA/ClpCP system can be quantitatively understood, at the level of individual cell behavior, within a very simple threshold mechanism. As the MecA/ClpCP system is widely conserved this finding raises the question of whether MecA/ClpCP also generates a heterogeneity advantage in other organisms such as S. pneumoniae, in which competence regulation is more straightforward and the comX bimodality mechanism is absent. Our data also highlight the long standing question of whether by combining cooperative behaviors of quorum signaling with deliberately noisy intracellular phenomena such as MecA and ComRS, S. mutans can achieve some form of optimum balance between socially-driven, environmentally-driven and purely stochastic behavior in competence regulation.
Experimental Procedures
Preparation of reporter strains
S. mutans strains and deletion mutants (Table 1) harboring green fluorescent protein (gfp) and/or red fluorescent protein (rfp) reporter genes fused to the promoter regions of comX (PcomX-gfp) and comY (PcomYA-rfp) were grown in brain heart infusion medium (BHI; Difco) at 37°C in a 5% CO2, aerobic atmosphere with either spectinomycin (1 mg mL−1), erythromycin (10 µg mL−1), or kanamycin (1 mg mL−1). PcomX-gfp was directly integrated into the chromosome of S. mutans (denoted XG) by amplifying a 0.2-kbp region comprising PcomX using primers that incorporated XbaI and SpeI sites (Table 2). This was fused to a gfp gene that had been amplified with primers engineered to contain SpeI and XbaI sites from the plasmid pCM11 (Lauderdale et al., 2010; Son et al., 2012), and inserted into the XbaI site on pBGE (Zeng; Burne, 2009). PcomYA-rfp was constructed in shuttle vector pDL278 (LeBlanc et al., 1992) by amplification of a 0.2-kbp region containing PcomY with HindIII and SpeI site-containing primers and fusing with the rfp gene reporter fragment amplified from plasmid pRFP (Bose et al., 2013), using primers that incorporated SpeI and EcoRI sites. The ligation mixtures were transformed into competent S. mutans (strain designated YR) and into the XG strain (denoted XG&YR). Additionally, to study the role of MecA/ClpCP on PcomY expression, both the XG integration vector and the YR shuttle vector were transformed into strains harboring non-polar (NPKmR) antibiotic resistance cassette replacements of mecA (this study), clpC or clpP (J. A. C. Lemos; Burne, 2002). Plasmid DNA was isolated from Escherichia coli using a QIAGEN (Chatsworth, Calif.) Plasmid Miniprep Kit. Restriction and DNA-modifying enzymes were obtained from Invitrogen (Gaithersburg, Md.) or New England Biolabs (Beverly, Mass.). PCRs were carried out with 100 ng of chromosomal DNA using Taq DNA polymerase. PCR products were purified with the QIAquick kit (QIAGEN). DNA was introduced into S. mutans by natural transformation and into E. coli by the calcium chloride method (Cosloy; Oishi, 1973).
Competence Peptides
Synthetic CSP (sCSP, aa sequence = SGSLSTFFRLFNRSFTQA), corresponding to the mature 18 aa peptide (Hossain; Biswas, 2012) was synthesized by the Interdisciplinary Center for Biotechnology Research (ICBR) facility at the University of Florida and its purity (95%) was confirmed by high performance liquid chromatography (HPLC). sCSP was reconstituted in water to a final concentration of 2 mM and stored in 100 μL aliquots at −20°C. Synthetic XIP (sXIP, aa sequence = GLDWWSL), corresponding to residues 11-17 of ComS, was synthesized and purified to 96% homogeneity by NeoBioSci (Cambridge, MA). The lyophilized sXIP was reconstituted with 99.7% dimethyl sulfoxide (DMSO) to a final concentration of 2 mM and stored in 100 μL aliquots at −20°C.
Microfluidic mixer design and fabrication
Microfluidic devices were fabricated by the soft lithography method of molding a transparent silicon elastomer (polydimethylsiloxane) on a silicon master (Sia; Whitesides, 2003). The master was made from a silicon wafer through conventional photolithographic processing. Details of the fabrication method and the devices were described previously (Jeon et al., 2000; Son et al., 2012; Son et al., 2015b). Our microfluidic device consisted of nine parallel flow chambers (each 15 µm deep and 400 µm wide), as shown in Supporting Figure S1. Three inlet channels supplied media containing different concentrations of signal peptides, delivered by syringe pumps into the device. The design has a mixing network that generates nine streams containing different admixtures of the three input solutions. These streams flow through the nine cell chambers in which S. mutans are adhered to the lower, glass window. The device also has two side channels: one for the control of fluid inside the device and the other for injection of different solutions into the cell chambers. Two-layer lithography allows air-pressure control of these side channels during cell loading and injection of different solutions (Unger et al., 2000).
Microfluidic experiments
Overnight cultures grown in BHI with antibiotic selection were washed and diluted 20-fold in fresh medium, which was either chemically defined medium (FMC) (Terleckyj et al., 1975; De Furio et al., 2017) or a complex medium that consisted of 1/3 of BHI (BD) and 2/3 of FMC by volume. Cultures were then incubated at 37°C in a 5% CO2, aerobic atmosphere. When OD600 reached 0.1 - 0.2, cells were sonicated at 30% amplitude for 10 sec (Fisher FB120) to separate cell chains and then loaded into the microfluidic device. Each cell chamber was continuously perfused with fresh medium containing different amounts of synthetic XIP (0 - 2 µM) or synthetic CSP (0 - 1 µM). The XIP or CSP concentration in each flow channel was generated by the mixture of three different inlet media in the mixing network in the device. A trace amount (0 - 10 ng/mL) of far-red fluorescent dye (Alexa Fluor 647) was added to each of the three inlet media in proportion to its signal molecule concentration, so that the concentration of signal molecule in each chamber could be calculated. After 2.5 h of incubation time, fresh medium containing 100 μg mL−1 of rifampicin was flowed through all cell chambers to halt GFP and RFP translation. Cell chambers were then incubated an additional 3 h to allow the full maturation of RFP. Cells were imaged in phase contrast and in green and red fluorescence using an inverted microscope (Nikon TE2000U) equipped with a computer controlled motorized stage and cooled CCD camera.
Single cell image analysis
Custom Matlab software was used to analyze the expression of the gfp and rfp reporters in individual cells from overlaid phase contrast, GFP, and RFP images (Kwak et al., 2012). The software first segments individual cells from the cell chain based on the phase contrast image, then finds the concentration of GFP and RFP by correlating the intensity of the phase contrast image with its GFP and RFP fluorescence intensity. This gives a unitless parameter (denoted R) that is proportional to the intracellular concentration of GFP or RFP. The GFP or RFP expression levels shown in the data figures are the R-values for green or red cell fluorescence respectively.
Transformation efficiency
Overnight cultures of selected strains were diluted 1:20 into 200 µL of FMC medium in polystyrene microtiter plates. Cells were grown to OD600 = 0.15 in a 5% CO2 atmosphere. When desired, 300, 500 or 1000 nM of sXIP was added and cells were incubated for 10 min. Then 0.5 µg of purified plasmid pIB184, which harbors an erythromycin resistance (ErmR) gene, was added to the culture. Following 2.5 h incubation at 37°C, transformants and total CFU were enumerated by plating appropriate dilutions on BHI agar plates with and without the addition of 1 mg mL−1 erythromycin, respectively. CFU were counted after 48 h of incubation. Transformation efficiency was expressed as the percentage of transformants among the total viable cells. The data presented are averages of two independent experiments that each included three biological replicates.
mRNA levels for mecA, clpCP, and com genes
Data for the analysis of relative mRNA levels for mecA, clpCP and com genes was taken from RNA-Seq analysis completed on strain UA159 (Kaspar et al 2018, in preparation). The wild-type strain was grown in FMC medium to OD600 = 0.2, at which time either a final concentration of 2 M XIP or vehicle control (1% DMSO) was added. The strains were then allowed to grow to mid-exponential phase (OD600 = 0.5) before harvesting. From the analyzed RNA-Seq data, total read counts for each selected gene were found from three biological replicates and RPKM (reads per kilobase per million) calculated under each condition. Finally, ratios for mRNA levels were found by using the normalized RPKM data and by setting mecA levels to 1.0. The data files used in this study are available from NCBI-GEO (Gene Expression Omnibus) under accession no. GSE110167.
Stochastic model for MecA regulation of comX
We used the gamma statistical distribution to model cell-to-cell variability (noise) in the activation of comY by ComX and the effect of the MecA/ClpCP system. Heterogeneity in bacterial protein copy number can be well-described by a physical model of transcription and translation as consecutive stochastic (Poisson) processes, characterized by rates kr (transcripts per unit time) and kp (protein copies per transcript per unit time), respectively (Friedman et al., 2006; Taniguchi et al., 2010). In this model the protein copy number n in each cell is a random variable drawn from a gamma distribution Γ(n | A, B). The two parameters A and B that determine the shape of the distribution are related to kr and kp, respectively (and to the mRNA and protein lifetimes) (Friedman et al., 2006). Gamma distribution fits to our PcomY reporter data are shown in Supporting Figure S2.
To model ComX activation of comY in the mecA deletion mutant (lacking post-translational regulation by MecA/ClpCP), we applied a simple quantitative model in which the PcomX activity of each cell, as reported by GFP fluorescence, determines the gamma distribution for its PcomY activity, measured by RFP fluorescence. Specifically, the PcomY-rfp reporter fluorescence Y of a cell is a random number drawn from a gamma distribution Γ(Y | A, B), for which the parameters are
Here, X is the PcomX-gfp reporter fluorescence of that cell. Thus Y is directly activated by X in a saturating but noisy fashion. We fit this model to a dataset of individual cell RFP and GFP fluorescence values collected on dual reporter (PcomX-gfp, PcomY-rfp) ΔmecA cells that were supplied with different concentrations of synthetic XIP (defined medium) and then imaged on glass slides. Maximum likelihood analysis gives the four model parameters a1, a2, b1, b2 for the ΔmecA strain as follows: We start with the experimental PcomX activity measured for each cell, then use the four parameters to define a PcomY gamma distribution for that cell. We find the probability of that cell’s actual PcomY activity, given that gamma distribution. The parameter values are then adjusted to maximize the likelihood of the total dataset. (The optimal values are given in Supporting Figure 3.) Given these model parameters we then generate a model simulation for comparison against the data as follows: We use the parameters and the experimental PcomX activity of each cell to generate its PcomY gamma distribution, draw a random number from that distribution to obtain a simulated PcomY activity for the cell, and then plot the resulting simulated PcomY vs PcomX values for all cells.
We also tested an alternative model in which environmental XIP concentration, rather than PcomX activity of a cell, is the determinant of that cell’s PcomY activity. In this model X in the above equations refers to the XIP concentration supplied to a cell. Again, using maximum likelihood, we found the parameters (a1, a2, b1, b2) that gave best agreement with the ΔmecA data in this alternative model. The scatterplot of Supporting Figure S3, generated by the above simulation procedure, compares the simulated PcomY to the experimental PcomY for the ΔmecA data.
For the dual-reporter ΔclpC mutant, we extended the above model by allowing MecA to sequester, but not degrade ComX. For simplicity we assume that (i) MecA and ComX bind with sufficiently high affinity that a cell can only activate comY to the extent that its ComX copy number exceeds its number of MecA copies, leaving some available ComX; ii) The activation of comY by the available ComX is as described in the ΔmecA model above (and with the same parameters); (iii) the MecA copy number M in a cell is a stochastic variable drawn from a gamma distribution Γ(M | A,B) whose A and B parameters are fixed, independent of XIP concentration. If X is the PcomX-gfp activity of a given cell, then X’ = X-M is the amount of ComX available after sequestration by MecA. Given a GFP measurement of X for a cell, the MecA gamma distribution Γ (M=X-X’ | A,B) determines the probability that X’ copies of ComX are available to activate comY. This X’ determines the probability distribution for Y (the PcomY-rfp response) by the above model. Averaging over the MecA distribution then gives a prediction for both the average behavior and cell-to-cell variability in the dependence of PcomY-rfp on PcomX-gfp, in the presence of MecA.
Taking the PcomY-rfp activation parameters obtained in the ΔmecA fit, we therefore analyzed individual cell PcomX/PcomY data that was collected on ΔclpC cells that were supplied with different XIP concentrations and imaged on glass slides. We then found the A and B values for the MecA distribution that maximize the likelihood of the PcomX/PcomY dataset, given the sequestration model. Using those parameters, we then generated a simulation of the PcomY versus PcomX activity. We compared these results to the experimental data for the ΔclpC strain. In plotting the simulation, we modelled the weak red auto-fluorescence background in the data by adding baseline Gaussian noise of 3 ± 0.8 red fluorescence units; this baseline is small compared to the typical red fluorescence (~102-104 units) of comY activated cells.
Acknowledgments
This work was supported by 1R01 DE023339 and T90 DE021990 from the National Institute of Dental and Craniofacial Research.