Peptide signaling without feedback in signal production operates as a true quorum sensing communication system in Bacillus subtilis

The concentration of signal molecule ComX increases by the square of bacterial density

In order to determine the dynamics of signal molecule (ComX) production, we have used experimental and mathematical modeling approaches. We quantified the ComX concentration over time in the spent medium of PS-216 (ΔcomP) with the biosensor strain BD2876 (for strain-description see Supplementary Table 1), which produces β-galactosidase in response to the exogenous addition of ComX³⁴. The assay included proper controls and calibrations to assure the biosensor-derived ComX concentrations are accurate (for details see Materials and methods). We found that the ComX concentration correlated positively with population density of PS-216 (ΔcomP) and remained constant at 10 nM after entering the stationary phase (Fig. 2a). Importantly, the representation of ComX concentration versus cell density (OD650) (Fig. 2b) showed a non-linear trend between the two parameters. The experimental data were fitted by an allometric function:

Where SM(t) is a signal molecule (ComX) concentration in time, N(t) is bacterial cell density in time, expressed as optical density of the bacterial suspension (OD650). The fitting results for parameters a and b were (9.6 ± 0.6) nM a.u.^−2.09 and 2.09 ± 0.10, respectively. The value of parameter a means that at OD650 = 1.0 a.u., which corresponds to the stationary growth phase in our experimental conditions and the bacterial density of 4 × 10⁸ cells mL^‒1, the ComX concentration is about 10 nM. In the early exponential growth phase the concentration was about 0.1 nM. Considering parameter b, the value obtained (2.09 ± 0.10) indicates that the ComX concentration increases by the square of bacterial density. This means that with increasing population density, the SM concentration (ComX) increases by the second power, while the amount of ComX per cell increases linearly. This relationship suggests that ComQXPA has an ultra-sensitive encoder module⁹, where signal molecule production is very sensitive to cell density. The same mathematical relationship can be obtained by assuming that SM production rate per cell corresponds to the product of a specific cell growth rate and a cell density (i.e., population growth rate, for details, see Supplementary Methods, Derivation of ComQXPA communication system model). The dependence of the SM production rate per cell on (a) cell density and (b) the specific cell growth rate can be seen as an alternative way to obtain the ultra-sensitivity of encoders, which is usually achieved by SM dependent positive feedback in many QS systems⁹. This makes ComX a true indicator of population density, which also encodes information about the cell growth rate.

The production of signal molecule ComX in native concentration does not present a substantial metabolic burden for the producer

Peptide signal molecules (SM) used by Gram-positive bacteria are metabolically costly to produce²⁹. We estimate here that single molecule of ComX produced by B. subtilis used in this study (pherotype 168) requires a considerable investment of 484 ATP units per single signal molecule (for calculation details see Supplementary Methods, Calculation of ATP requirements for synthesis of 1 SM, ComX, 168 pherotype). This drastically exceeds the estimated cost of typical Gram-negative bacterial QS signals, with butyryl-homoserine lactone, C4-HSL from Pseudomonas aeruginosa costing only 8 ATP units²⁹. However, the concentration of stationary growth phase signaling peptide ComX is 100–1000 times lower in B. subtilis (10 nM, Fig. 2a) compared to the typical concentrations of AHLs released by Gram-negative bacteria^35,36,37,38. This suggests that high cost per SM is buffered by low concentrations of SM, thereby reducing the fitness costs of SM production in peptide-based communication systems. In order to test the metabolic cost of ComX production, we first compared the growth curves of receptor-deficient PS-216 (ΔcomP), and signal and receptor-deficient PS-216 (ΔcomQXP) strains (Fig. 2c). The use of the strains without a receptor made it possible to separate the costs of signaling from the additional costs of the communication response.

Apparently, the maxima of growth curves of ΔcomP and ΔcomQXP, and their slopes (corresponding to the growth rate divided by log 2) were almost identical: PS-216 ΔcomP = (0.503 ± 0.008) h^‒1 and PS-216 ΔcomQXP = (0.496 ± 0.007) h^‒1, suggesting that ComX production does not represent a substantial metabolic burden in the observed system (Fig. 2c). The more direct fitness comparison between PS-216 (ΔcomQXP) and (ΔcomP), was carried out through a competition assay between PS-216 (ΔcomQXP) and PS-216 (ΔcomP) (Fig. 2d). In line with results in Fig. 2c, ratio of ComX producers and ComX non-producers did not changed considerably throughout the experiment, suggesting negligible costs for signal production (Fig. 2d).

Next, we tested whether the absence of prudent SM production induces measurable fitness cost. To test this, we overexpressed comX from the P_hycomX IPTG-inducible promoter (Supplementary Fig. 2a, b), which ensured the production of additional copies of ComX. As expected, the overproduction of ComX has a negative impact on the growth of B. subtilis (Supplementary Fig. 2a). The overexpression of ComX in E. coli had a similar negative fitness effect (Supplementary Fig. 2c). As it can be calculated from Supplementary Fig. 6b, the concentration of ComX in E. coli spent media was about 900 nM, corresponding to 200 nM a.u.^‒1, which is about 20 times more than we have measured in B. subtilis (Fig. 1a). The above results indicate that the costs of ComX synthesis under the native production regime are very low and can only be evaluated under non-native overexpressing conditions.

The ComQXPA communication system operates in strong correlation with the oxygen concentration

As already mentioned, the SM production rate per cell in ComQXPA is not controlled by the SM dependent positive feedback loop, but by cell density and specific cell growth rate (population growth rate), (eq S5-S6). The question is how bacteria then sense cell density and specific growth rate, which accelerate signal production. One of the key factors for the growth rate of B. subtilis is environmental oxygen content³⁹, which is believed to determine the survival strategies of this species⁴⁰. It was recently shown that surfactin, which is directly related to the response in ComQXPA (RM in this work) becomes critical for B. subtilis when oxygen is low⁴¹.

First, we did not allow any aeration of the batch culture, i.e., the oxygen supply to the growing culture was limited by diffusion of air through air filters on the inlets of the incubator. We monitored changes in oxygen concentration during growth in batch culture and observed an almost perfect negative correlation between the growth curve and the dissolved oxygen concentration (Fig. 3a, Supplementary Fig. 3). The strongest decrease in dissolved oxygen in the medium occurred during the exponential growth phase, exactly when the population growth rate reached its maximum. When spent medium of PS-216 wt was tested by the ComX biosensor BD2876 (ΔcomQ, srfA-lacZ), we could measure the significant response (for t > 1.75 h, P < 0.008) by the biosensor that increased with the growth of the culture (Fig. 3c), indicating ComX is being produced. This agrees with Fig. 2a, where we quantified the produced ComX in the spent medium of PS-216 (ΔcomP, producing ComX, but not responding to ComX). As expected for the proper ComX biosensor, it barely responded to the tested spent medium with no ComX (PS-216 ΔcomQ spent medium) and strongly responded to the same medium when purified ComX was added (Supplementary Fig. 4) confirming the ComX is the major factor being measured by the biosensor BD2876.

Next, we assessed the signal production in batch culture, where we assured continuous oxygen saturation. Under this condition the negative relationship between population size and dissolved oxygen concentration is broken. Surprisingly, oxygen saturation eliminated ComX production, which can be seen by very low biosensor BD2876 response that is indistinguishable from the spent medium with no ComX (PS-216 ΔcomQ spent medium) even in the late-stationary growth phase (Fig. 3d). This result indicates that the ComQXPA communication system has lost its functionality, when there is no ‘natural’ oxygen gradient, and that the oxygen content can be used as an indicator of cell density and growth rate.

The response model shows that the response of the cells to ComX is non-linear

In our model, the expression level of the srfA operon serves as a measure for the response (RM) to the signal molecule SM, represented by ComX. To study how RM depends on SM we evaluated promoter activity of srfA in the B. subtilis PS-216 (∆comQ, P_srfAA-yfp), which carries the markerless deletion of comQ⁴² and is therefore signal-deficient. Response level was assessed by incubating the PS-216 (∆comQ, P_srfAA-yfp) for 4 h in the presence of different ComX concentrations, which was the only factor that varied in this experiment. The response level was expressed as Yfp fluorescence per cell, normalized to the maximum response, W_max, which gives a relative measure, W(SM), of how strongly the cells respond to ComX and this is shown as a function of the exogenously added ComX in Fig. 4a. The response to SM was sigmoidal. In order to check whether the response curve had reached the final shape after 4 h of biosensor incubation, we performed the same experiment, but incubated biosensor with ComX for 3 or 6 h, respectively. As can be seen from the comparison of Fig. 4a with Supplementary Fig. 5, the response curve has not changed after extending the incubation over 4 h. We have therefore taken the 4 hours response curve (Fig. 4a) as a reference for further communication system analysis. The sigmoidal functions can typically describe the relationship between transcription factors and promoter activities, and can be modeled by the Hill equation^43,44. In the case of ComQXPA the ComX dependent transcription factor ComA-P acts directly on the P_srfAA promoter and induces its activity as the response to the signal. Assuming a linear relationship between ComX concentration (SM) and the active ComA-P one can expect that the experimental data can be fitted by the Hill equation:

SM is ComX concentration and Km is the ComX concentration at which half of the maximum response is achieved; n (Hill coefficient) describes the cooperativity among transcriptional activators. Successful fits indicated by the low reduced χ² (see Supplementary Table 2), show that the biosensor sensitivity is maximized at 3–5 nM of ComX (Km), while the highest response value is reached at around 10 nM of the ComX. n > 1 values obtained for all fits indicate positive cooperativity (i.e., ultrasensitivity¹⁴) in the binding of the transcriptional factor ComA to the srfA promoter^43,44. This agrees with the research showing that two molecules of the ComA homodimer cooperatively bind to the two promoter regions located upstream of the RNAP binding sites of srfA^13,20,45. The inactivation of the second promoter region decreases the promoter activity of srfA by 100-fold (ref. ¹³), which underscores the importance of the second binding region, explains n ≥ 2 and the sharp increase in srfA promoter activity with ComX concentration. In addition, we show here that the critical concentration of ComX required to induce quantifiable response (designated here as lower limit response (LLR)) is 0.2–0.5 nM. These results, therefore, suggest that the response per cell depends cooperatively on the ComX concentration and that the cells respond to very low concentrations of ComX.

Fully functional ComQXPA communication system does not require a positive feedback loop–the validation of the ComQXPA communication system

The response curve in Fig. 4a is a function of the SM concentration only. In the more natural setting (i.e., during growth) the cells encounter growth-dependent changes in SM concentrations as well as changes in bacterial density and growth rate over time. We have therefore asked whether the response curve based on the modeling and results presented in Figs. 2b and  4a could fit the response data in the SM producing and responding strain exposed to changes in these three parameters.

We cultivated the SM producing and responding PS-216 strain carrying the response reporter (PsrfA-lacZ) in a large volume bioreactor system (Fig. 4b). This allowed sterile sampling of spent medium and cells (for response quantification) at several time points, without affecting growth conditions. Immediately after the inoculation of the fresh medium by overnight culture the β-galactosidase activity of PS-216 (PsrfA-lacZ) was high. We assumed that this was a consequence of the accumulation of the expressed PsrfA reporter (β-galactosidase, RM) during the overnight growth. As a consequence of the dilution of the intracellular β-galactosidase (RM) due to cell division, the activity of the β-galactosidase decreased sharply after 2 h incubation (Fig. 4b). Simultaneously, as predicted by (Eq. 1), the concentration of SM (ComX) in the medium was increased exponentially during growth, and soon reached a critical concentration to activate the srfA promoter. In particular, as elucidated by the fits of (Eq. 2) to the data in Fig. 4a, the lower limit of the response (LLR) is reached shortly before upturn of the cell response curve in Fig. 4b. At this point the culture is in exponential growth phase at the cell density of 3 to 8 × 10⁷ cells mL⁻¹. The steep slope of the response curve indicates that the rate by which the response molecule (RM) is synthesized now exceeds the dilution due to the cell division rate. From now on, the response per cell correlates approximately linearly with OD650, suggesting a strong coupling to cell growth. Taking these facts into account and considering that the response molecule (RM) concentration is sensitive to the concentration of the signal, SM, (Eq. 2) and that SM can be expressed in terms of cell density (Eq. 1), the concentration of a response molecule per cell, RM(t)/N(t), can be analytically described (see also Supplementary Methods, Derivation of ComQXPA communication system model) as:

where RM0/N(t) is the response per cell of overnight culture, i.e., the overnight accumulated β-galactosidase. The second term, RM1(t)/N(t) accounts for the synthesis of the β-galactosidase after inoculation of a fresh medium and comprises the parameters describing the sensitivity of the response to a signal molecule, Wmax, Km, n (Eq. 2), the signal production, a (Eq. 1), cell density, N(t) and proportionality constant, k that gives the magnitude of the response per cell when the potential to respond to the signal is maximally fulfilled (i.e., at Wmax) and the specific growth rate is 1 h⁻¹. The definition of RM1(t) is given in Supplementary Methods (eq S11). Note that for the derivation of Eq. (3) we assumed no degradation of SM occurs, as our experiments suggest SM was stable under the conditions studied (Supplementary Figs. 6a and 7, see also Supplementary Methods, Derivation of ComQXPA communication). All the parameters in (Eq. 3), except k in RM1(t) and RM0, were taken as constants obtained in the independent experiments by fits of (Eq. 1) and (Eq. 2). With k and RM0, as the only fitting parameters, we applied the mathematical model in (Eq. 3) (for details of the model equation see Supplementary Methods, Derivation of ComQXPA communication system model) to fit the experimental cell response data (Fig. 4b). The successful fit (see Supplementary Table 4 for details) indicates that the relationship assumed among cell density, cell growth, signal concentration and response in (Eq. 3) is valid and yields (760 ± 120) M.U. for k and (5.5 ± 1.5) M.U. a.u. for RM0. Again, we did not need to incorporate the SM feedback loop into our model, which is consistent with published results suggesting that this communication system lacks a feedback loop^10,11,12,13.

The ComQXPA dependent signaling and response at the cellular level

So far, we have focused on the population averages, which is a traditional approach in studies on microbial communication systems^17,46. We here report results on communication dynamics of B. subtilis at the single cell level using fluorescence-based molecular tools. This approach provides the means to track temporal changes in expression of genes involved in signal synthesis (signaling) and in response and thus provides the insight into a phenotypic heterogeneity within the population.

We used the double-labeled fluorescencent strain B. subtilis PS-216 (comQ-yfp, srfA-cfp), in which fluorescent reporters were fused to the comQ and srfA promoters. The two genes code for the ComX signal-processing protein and the communication-activated operon, respectively. Since comQ and comX share the same promoter and their genetic sequences often overlap¹⁵ expression level of comQ corresponds to the expression level of comX. The fluorescence of individual cells was observed under the microscope in different growth phases and quantitatively analyzed (Fig. 5).

The observed expression pattern for the signaling gene (comQ-yfp) follows lognormal distribution (Fig. 5g). A small number of outliers in the comQ expression (on average, 10x brighter than the majority) were easily detected in the qualitative image analysis (Fig. 5a). These hyperproducers were not present in the overnight culture and began to occur during exponential growth, after 1-hour incubation in fresh medium, (Fig. 5d). In general, hyperproducers accounted for about 0.1–1% of the population, and their frequency increased during the first 6 hours. These data suggest that bulk of the ComX is nevertheless produced by the majority of the population as expected for the true QS communication systems. The contribution of the srfA hyperproducers to the total surfactin production is even less pronounced since their occurrence did not exceed 0.1% of the population (Fig. 5b, d).

The most heterogeneous expression of the communication signaling gene (comQ-yfp) was observed in overnight culture, immediately after its transfer to the fresh medium (Fig. 5g, Supplementary Fig. 8a), but hyperproducers where not detectable at this time (Fig. 5d). Once the cells begun to divide, the distribution shifted to lower fluorescence intensities with a simultaneous decrease in heterogeneity, but from 3 to 4 h onwards single cell fluorescence gradually increased, along with an increase in population heterogeneity (Fig. 5e, g, Supplementary Fig. 8a). This suggests that the expression rate is now higher than the division rate (i.e., the production overpowers the dilution due to cell division). A similar pattern was observed in the communication response (srfA-cfp) (Fig. 5f, h, Supplementary Fig. 8b) with two major differences. The minimum level in comQ-Yfp fluorescence was reached 1 h later than srfA-Cfp fluorescence, which suggests the expression of ComX is in first hours low compared to the srfA expression. Nevertheless, the entire cell population, with the exception of hyperproducers, which represented only a fraction of the comQ/srfA expressing cells, followed unimodal lognormal distribution expression pattern. This suggests that the ComQXPA communication phenomenon in B. subtilis, at least under the conditions in our experiment, is not restricted to individuals and can be studied at the population level, i.e., the averages represent well the population.

The comQ-yfp and srfA-cfp expression co-localization analysis (Supplementary Table 5) reveals the correlation coefficient of about 0.5, which is significantly (P = 0.01) higher than in the overnight culture. The presence of the correlation suggests that on average, cells that produce the signal more intensively, also respond to signal more intensively, supporting the idea of self-sensing⁴⁷. However, the correlation coefficient strength was only moderate, suggesting that sensing of the external signal (sensing-of others) still works as expected for a typical QS system.

The induced response in ComQXPA communication system is graded and almost switch-like

The perfect QS system does not produce a response until the threshold bacterial density is reached and then immediately switches to a full response. This minimum to maximum transition may be either a perfect switch or a graded induction. By combining the information from Figs. 2b and 4a in the form of eq S9 results in the normalized ComQXPA response curve as a function of bacterial density (Fig. 6a) that resembles a graded switch like induction (compare to Fig. 1a). The perfect switch like communication system is unrealistic, because it requires that all the cells are perfectly synchronized and immediately switch to maximal response, leaving no time for the adaptation to the signal stimuli. It is reasonable to expect that for the true quorum sensing system (QS) most of the response has to occur within the same generation of dividing cells (n_gen < 1). As can be seen in Fig. 6b, this is true for the ComQXPA communication system (our case), which achieves 50% of the response within the same generation of dividing cells (n_gen ≈ 0.7). On the other hand, if the communication system lacks the ultra-sensitivity in either the signal production (encoder module) or response production (decoder module), the achievement of 50% response shifts well over the same generation of dividing cells (n_gen ≈ 1.4 to 1.6), extending the cell density and time needed for substantial response to occur.

Source: Ecology - nature.com