The Phosphorylation Flow of the Vibrio harveyi Quorum-Sensing Cascade Determines Levels of Phenotypic Heterogeneity in the Population

ABSTRACT

Quorum sensing (QS) is a communication process that enables a bacterial population to coordinate and synchronize specific behaviors. The bioluminescent marine bacterium Vibrio harveyi integrates three autoinducer (AI) signals into one quorum-sensing cascade comprising a phosphorelay involving three hybrid sensor kinases: LuxU; LuxO, an Hfq/small RNA (sRNA) switch; and the transcriptional regulator LuxR. Using a new set of V. harveyi mutants lacking genes for the AI synthases and/or sensors, we assayed the activity of the quorum-sensing cascade at the population and single-cell levels, with a specific focus on signal integration and noise levels. We found that the ratios of kinase activities to phosphatase activities of the three sensors and, hence, the extent of phosphorylation of LuxU/LuxO are important not only for the signaling output but also for the degree of noise in the system. The pools of phosphorylated LuxU/LuxO per cell directly determine the amounts of sRNAs produced and, consequently, the copy number of LuxR, generating heterogeneous quorum-sensing activation at the single-cell level. We conclude that the ability to drive the heterogeneous expression of QS-regulated genes in V. harveyi is an inherent feature of the architecture of the QS cascade.

IMPORTANCE V. harveyi possesses one of the most complex quorum-sensing (QS) cascades known, using three different autoinducers (AIs) to control the induction of, e.g., bioluminescence, virulence factors, and biofilm and exoprotease production. We constructed various V. harveyi mutants to study the impact of each component and subsystem of the QS signaling cascade on QS activation at the population and single-cell levels. We found that the output was homogeneous only in the presence of all AIs. In the absence of any one AI, QS activation varied from cell to cell, resulting in phenotypic heterogeneity. This study elucidates a molecular design principle which enables a tightly integrated signaling cascade to control the expression of diverse phenotypes within a genetically homogeneous population.

INTRODUCTION

Quorum sensing (QS) is a cell-to-cell communication process which relies on small diffusible molecules called autoinducers (AIs). This phenomenon was first described for luminescent bacteria such as the symbiotic marine bacterium Vibrio fischeri (1–3). Since then, many different QS systems have been discovered in Gram-negative as well as Gram-positive bacteria (4). AIs are synthesized and released into the environment, therefore accumulating in a cell-number-dependent manner. AIs are perceived by specific sensors. Once the AI concentration reaches a certain threshold, QS is triggered, and cells produce a phenotypic answer by activating genes for traits such as virulence, biofilm formation, luminescence, antibiotic production, or conjugation that benefit the population as a whole (5).

Vibrio harveyi strain BAA-1116 (recently reclassified as Vibrio campbellii [6]) uses a complex QS cascade and responds to three different classes of AIs: HAI-1, an acyl homoserine lactone [N-(3-hydroxybutyryl)-homoserine lactone]; AI-2 (furanosyl borate diester); and CAI-1, a long-chain amino ketone [(Z)-3-aminoundec-2-en-4-one] (Ea-C₈-CAI-1) (7–9). The distribution of these three AIs in the bacterial world is uneven: AI-2 is a widespread molecule produced by many different bacterial species (10); CAI-1 is produced by several Vibrio species, in particular Vibrio cholerae (11); and HAI-1 is largely specific to V. harveyi and its close relatives (12). Another striking difference among the three AIs lies in the timing of their production. We recently showed that in a growing wild-type (WT) V. harveyi culture, AI-2 is synthesized first, followed shortly by HAI-1, whereas CAI-1 was detected only at later time points in the late exponential growth phase (13).

Each AI is perceived by a specific membrane-integrated hybrid sensor kinase: HAI-1 is perceived by LuxN, AI-2 is perceived by LuxQ in combination with the periplasmic binding protein LuxP, and CAI-1 is perceived by CqsS (8, 11, 14, 15) (Fig. 1).

FIG 1.

The QS phosphorelay in V. harveyi. In V. harveyi, three synthases (LuxM, LuxS, and CqsA) produce three AIs (HAI-1, AI-2, and CAI-1). In the absence of AIs, the sensors autophosphorylate and transfer their phosphoryl groups to the phosphotransfer protein LuxU and to the transcriptional regulator LuxO. Phosphorylated LuxO, together with σ⁵⁴, induces the expression of five regulatory small RNAs, called Qrr sRNAs. In combination with the chaperone Hfq, the Qrr sRNAs destabilize the transcripts of luxR, which encodes the major QS regulator. When AIs are perceived by their cognate hybrid sensor kinases LuxN, LuxQ (in an interaction with the periplasmic binding protein LuxP), and CqsS, autophosphorylation is inhibited, thus interrupting phosphorelay. Draining of the phosphate from the cascade leads to a depletion of Qrr sRNAs and the expression of luxR. LuxR induces genes required for luminescence, biofilm formation, or proteases and represses the expression of the type III secretion system or siderophores. AphA is the counterpart of LuxR; it is present at low cell densities but absent at high cell densities and shares common targets with LuxR. Several feedback loops are depicted and are described in the text. Dotted lines indicate phosphotransfer reactions, and continuous lines indicate transcriptional or posttranscriptional regulatory interactions. H indicates histidine, and D indicates aspartate (they denote phosphorylation sites). CM, cytoplasmic membrane; CP, cytoplasm; PP, periplasm.

At low cell densities and, therefore, low AI concentrations, the respective sensors act as kinases. Each sensor autophosphorylates, subsequently transferring the phosphoryl group to the histidine phosphotransfer protein LuxU, which in turn phosphorylates the transcriptional regulator LuxO (16). Phosphorylated LuxO is activated and, together with σ⁵⁴, induces the expression of five Qrr regulatory small RNAs (sRNAs) (17, 18). These Qrrs in association with the RNA chaperone Hfq act to destabilize and degrade luxR transcripts, maintaining the QS phenotypes in an off state (19). At high cell densities and upon perception of their cognate AI, the sensors inhibit their kinase activity, resulting in the dephosphorylation of LuxU and draining of the phosphate from the phosphorelay cascade (20). Nonphosphorylated LuxO is inactive, and no Qrr sRNA is transcribed, releasing the brake on the production of the QS master regulator LuxR. LuxR then induces the genes necessary for the expression of phenotypes such as luminescence (21), biofilm formation (22), or proteolysis (23) and represses genes for type III secretion (11) and siderophore production (18).

Several feedback loops have been reported to influence the numbers of molecules of the different components of the QS signaling cascade: LuxR represses its own expression and induces the expression of Qrr2, Qrr3, and Qrr4 (24–26); LuxO represses its own expression, and luxO mRNA is targeted by the Qrr sRNAs (26, 27); and luxMN translation is also prevented by the Qrr sRNAs (26, 28). Finally, the transcription factor AphA, another master regulator of QS, is induced at low cell densities and likewise induces the expression of the Qrr sRNAs (26, 29). Thus, the Qrr sRNAs play a key role in both the activation and repression of the QS phenotypes in V. harveyi.

Despite detailed knowledge on signal integration in the QS cascade of V. harveyi, most studies have reduced the complex three-input cascade to a two-input system by deleting the CqsS sensor for CAI-1 on the grounds that CAI-1 is the weakest of the three signaling pathways under laboratory conditions (28, 30). In these strains, at low cell densities, one kinase activity is missing, and inversely, at high cell densities, the phosphorelay is deprived of one phosphatase activity, therefore unbalancing the natural phosphorylation flow. Moreover, it was shown that CAI-1 is produced and integrated into the QS cascade at late exponential phase (13). Therefore, we believe that all three inputs should be considered for a comprehensive analysis of QS signal integration. To explore these phenomena, we generated a new set of V. harveyi mutants, which are deleted for the genes encoding the AI sensors and/or synthases in every possible combination, and investigated QS activation at both the population and the single-cell levels. We demonstrate here that the CqsS sensor fulfills a role comparable to that of the LuxQ sensor, while LuxN and HAI-1 played only a minor role in QS-regulated bioluminescence induction. Our data further indicate that the ratio of kinase activities to phosphatase activities in the phosphorelay as well as the number of molecules of LuxR drive the level of intercellular heterogeneity of QS-regulated phenotypes in V. harveyi.

MATERIALS AND METHODS

Bacterial strains and growth conditions.

Strains and plasmids used in this study are listed in Table S1 in the supplemental material. Escherichia coli strains were grown aerobically in lysogenic broth (31) at 37°C on a rotary shaker. V. harveyi strains were cultivated in autoinducer bioassay (AB) medium (32) or LM medium (20 g/liter NaCl, 10 g/liter tryptone, 5 g/liter yeast extract) and incubated aerobically in a rotary shaker at 30°C. When required, media were solidified by using 1.5% (wt/vol) agar. If necessary, media were supplemented with 50 μg/ml kanamycin sulfate, 100 μg/ml ampicillin, and/or 20 μg/ml gentamicin. To allow the growth of the conjugation strain E. coli WM3064, meso-diaminopimelic acid (DAP) was added to a final concentration of 300 μM.

Strain construction.

Molecular methods were carried out according to standard protocols (31) or according to the manufacturer's instructions. Kits for the isolation of plasmids and purification of PCR products were purchased from Südlabor (Gauting, Germany). Enzymes were purchased from New England BioLabs (Frankfurt, Germany) and Fermentas (St. Leon-Rot, Germany). E. coli strains were transformed with replicative plasmids by using chemically competent cells (33).

In-frame deletion mutants of V. harveyi BAA-1116 were constructed as previously described, leaving terminal sections of the target genes (34–36). For this purpose, upstream and downstream fragments (ranging from 500 to 1,000 bp) of the desired gene regions were amplified by using the corresponding primers (see Table S2 in the supplemental material). After PCR product purification, the fragments were fused by overlap PCR (37). The overlap product was isolated from an agarose gel, digested with the corresponding restriction enzymes, and ligated into the suicide vector pNPTS138-R6KT (34). The resulting plasmid was introduced into V. harveyi by conjugative mating using E. coli WM3064 as the donor in LB medium containing DAP. Single-crossover integration mutants were selected on LB plates containing kanamycin but lacking DAP. Single colonies were grown over a day without antibiotics and plated onto LB plates containing 10% (wt/vol) sucrose to select for plasmid excision. Kanamycin-sensitive colonies were checked for targeted deletion by colony PCR using primers bracketing the site of the deletion.

Integration of the luxR-gfp fusion into V. harveyi was performed via homologous recombination (single crossover). The transfer of the conjugative plasmid from the donor strain E. coli WM3064 containing the required plasmid into V. harveyi was performed as described above. Therefore, the donor and the recipient strains were grown together in LB medium up to an optical density at 600 nm (OD₆₀₀) of 0.8 to 1.0, supplemented with additives if required. For Tn7 plasmid recombination, triparental mating was necessary, using an additional WM3064 strain containing plasmid pTNS2, which codes for the T7 transposase. Single colonies were checked for chromosomal integration by PCR and sequencing.

External AIs.

To study the impact of externally added AIs on the bioluminescence of different strains, AI-2 (Karina Xavier, Universidade Nova de Lisboa) and HAI-1 (University of Nottingham) were added to exponentially growing cells at a final concentration of 20 μM.

CAI-1 was extracted from cell-free supernatants by using dichloromethane. For this purpose, wild-type V. harveyi cells were grown in 2 liters of LM medium. After 8 h of growth, cells were pelleted, and the supernatant was filtered. CAI-1 was extracted from the 2-liter supernatant by using 400 ml dichloromethane and was further concentrated on a rotary evaporator to a final volume of 0.2 ml. The activity of CAI-1 was assessed via a reporter assay using a V. harveyi strain that can perceive only CAI-1 (AI negative [AI⁻] LuxN⁻ LuxQ⁻), and luminescence production of the reporter strain was used as a readout to assess CAI-1 activity. For use in induction assays, CAI-1 dichloromethane samples were diluted 1:500. This takes into account the instability of CAI-1 or the loss of active compound during the preparation process.

Bioluminescence assay.

For synthase mutants, cells from a culture grown overnight were diluted 1:5,000 in AB medium, and luminescence was measured every hour. For induction with autoinducers, cells were diluted 1:1,000, autoinducers were added after 3 h of growth, luminescence was measured every hour, and fluorescence pictures were taken after 4 and 6 h. Luminescence produced by V. harveyi strains and the OD₆₀₀ were determined in microtiter plates with a Tecan Infinite F500 system (Tecan) for 0.1 s. Data are reported as relative light units (RLU) in counts per second per milliliter per OD₆₀₀.

Single-cell fluorescence microscopy.

For phase-contrast and fluorescence microscopy, samples were analyzed on 0.5% (wt/vol) agar pads, which were placed onto microscope slides and covered with a coverslip. Images were taken on a Leica microscope (DMI 6000B) equipped with a Leica DFC 365 Fx camera. An excitation wavelength of 546 nm and a 605-nm emission filter with a 75-nm bandwidth were used for mCherry fluorescence.

Analysis of transcription levels via qRT-PCR.

V. harveyi strain BAA-1116 was cultivated as described above. Samples were withdrawn after 6 h of growth, and RNA was isolated as described previously (38). The RNA was then used as the template for random-primed first-strand cDNA synthesis according to the manufacturer's instructions. Quantitative real-time PCR (qRT-PCR) (iQ5 real-time PCR detection system; Bio-Rad) was performed by using the synthesized cDNA, a SYBR green detection system (Bio-Rad), and internal primers specific for luxO and luxU. The C_T (cycle threshold) value was determined after 40 cycles by using iQ software (Bio-Rad). Values were normalized with reference to recA values, and relative changes in transcript levels were calculated by using the comparative C_T method (39).

Single-cell analysis and statistical analysis.

A custom software application was developed to automate single-cell analysis of bright-field and fluorescence images. The analysis procedure involves the detection and segmentation of cells in the bright-field image, followed by fluorescence-level extraction for each individual cell. The graphical user interface allows visual inspection of the segmentation results and their manual correction if necessary. The software is open source, and binaries are available (see https://tmramalho.github.io/bigCellBrotherGUI/).

Briefly, the segmentation pipeline can be described as a series of steps. A preprocessing step enhances contrast and removes noise, background detection is done by the use of an adaptive threshold mechanism to identify background pixels, and cells are identified and segmented by using the watershed algorithm (40). Once cells have been segmented and labeled, mean fluorescence is calculated by averaging fluorescence values over all pixels assigned to each label.

To determine whether differences in the noise levels of various strain populations are statistically significant, we employed the following procedure. First, the log of the data was calculated, and the mean was subtracted; this allowed us to assume that the data were distributed as a zero-mean Gaussian distribution (41) with standard deviation σ:

$$p\left(d\right)=\frac{\exp \hspace{0.17em}\left(-\frac{d^2}{2{\sigma }^2}\right)}{\sqrt{2\pi {\sigma }^2}}$$

We then performed two-sample Kolmogorov-Smirnov tests (42) for all pairs of data sets, which tests the hypothesis that two strain populations are compatible. If the hypothesis was rejected, the coefficient of variation (\sigma/mean) of the original data was incompatible, and therefore, the difference in heterogeneity is statistically significant (P value of >0.05).

RESULTS

LuxQ and CqsS have greater impacts on luminescence induction than does LuxN.

In order to study the impact of each AI and its cognate sensor on the signaling output, we generated deletion mutants lacking one, two, or all AI sensors in all possible combinations. These deletions were introduced in the wild-type genetic background as well as in an AI⁻ mutant (MR15 [LuxM⁻ LuxS⁻ CqsA⁻]) lacking all three synthase genes (Fig. 2). The resulting mutant strains were assessed for growth and luminescence production. No growth defect was observed for any of the mutants (data not shown). As expected, mutants lacking all three sensors, either in the wild-type genetic background or in the AI⁻ strain, exhibited levels of luminescence similar to those observed for the constitutively QS-active mutant, designated the QS⁺ mutant (LP1 [ΔluxO]) (Fig. 2). The mutants devoid of every sensor are unable to perceive any AI signal; therefore, LuxU and LuxO are always in an unphosphorylated state, and luminescence is constitutively produced.

FIG 2.

Population-level luminescence produced by sensor mutants. Each AI⁻ mutant (LuxM⁻ LuxS⁻ CqsA⁻) was deleted for one or several hybrid sensor kinases. The quorum-sensing system of each mutant was activated by using different sets of AIs, and luminescence production was measured 3 h after the addition of the AIs. Error bars represent the standard deviations of data from three different experiments. RLU, relative light units in counts per second per milliliter per OD₆₀₀. The thick line represents background RLU, as measured in uninoculated AB medium.

The deletion of any one or any pair of sensors in the wild-type genetic background had little impact on luminescence production; the only notable changes were in the time of onset of luminescence production, as previously described (11, 43). Therefore, we focused on the AI⁻ mutant to assess directly the impact of various kinase/phosphatase ratios on the QS cascade in vivo.

Mutants lacking one of the three sensors in the AI⁻ background were unable to induce luminescence; i.e., measured luminescence levels were in the same range as those for the AI⁻ mutant (Fig. 2). This finding demonstrates that two sensors acting as kinases are sufficient to keep the QS cascade switched off.

Mutants with luxN luxQ (LuxN⁻ LuxQ⁻) and luxN cqsS (LuxN⁻ CqsS⁻) double deletions were also unable to induce luminescence production, indicating that LuxQ or CqsS kinase activities were strong enough to ensure sufficient phosphorylation of LuxU/LuxO to maintain the QS cascade in an off state. In contrast, the mutant with the luxQ cqsS double deletion (LuxQ⁻ CqsS⁻) produced ∼10% of wild-type levels of luminescence, suggesting that LuxN on its own is unable to maintain the cascade in an off state (Fig. 2). Negative feedback regulation of luxN mRNA by Qrr4 sRNA was described previously (28). Thus, in a mutant in which only LuxN is present, the number of LuxN molecules is apparently too low to allow a sufficient rate of phosphorylation of LuxU/LuxO.

To further investigate signal integration and the impact of each sensor on light production, AIs were externally added in different combinations, and luminescence induction was measured. When all three AIs were added to the AI⁻ mutant, luminescence production was restored to the wild-type level (Fig. 2), indicating that the three AIs were active and sufficient to mimic wild-type conditions. Luminescence levels close to those of the wild type were also measured for single-sensor-deletion mutants after the addition of the AIs recognized by the remaining sensors (LuxN⁻ plus AI-2 and CAI-1, LuxQ⁻ plus HAI-1 and CAI-1, and CqsS⁻ plus HAI-1 and AI-2) (Fig. 2). The addition of the two AIs resulted in a switch of the two remaining sensors into the phosphatase mode in these mutants. We also observed wild-type luminescence levels in double-sensor-deletion mutants upon exposure to the AI recognized by the remaining sensor (LuxN⁻ LuxQ⁻ plus CAI-1, LuxN⁻ CqsS⁻ plus AI-2, and LuxQ⁻ CqsS⁻ plus HAI-1). These results show that as long as no counteracting sensor in the kinase mode is present, the presence of one sensor in the phosphatase mode is sufficient to dephosphorylate LuxU/LuxO and to induce light production as in the wild type.

Further experiments confirmed our above-described observation that LuxN is the sensor that has the least impact on the signaling output. The addition of HAI-1 to the AI⁻ mutant hardly induced luminescence (1.4 times more induction than the AI⁻ mutant), indicating that LuxN phosphatase activity alone cannot overcome the effects of LuxQ and CqsS kinase activities. Only when both of the other sensors were deleted could HAI-1 and its receptor LuxN induce significant luminescence (5.8 times more luminescence upon HAI-1 addition in the LuxQ⁻ CqsS⁻ strain) (Fig. 2). This was not the case for the other sensors, LuxQ and CqsS, as the addition of AI-2 or CAI-1 to the AI⁻ mutant was enough to partially restore light production (261 and 6.9 times more luminescence upon AI-2 and CAI-1 addition in the AI⁻ mutant, respectively), revealing that LuxQ phosphatase activity can counteract LuxN and CqsS kinase activities and likewise that CqsS phosphatase activity can counteract LuxN and LuxQ kinase activities to a lesser extent.

Phosphorylation flow determines levels of heterogeneous QS activation.

To assess the impact of the phosphorylation flow on noise, we investigated QS activation with the new set of mutants at the single-cell level. A fluorescent reporter consisting of a fusion between the luciferase promoter (P_luxC) and mCherry was introduced at the attTn7 site (downstream of glmS; position 821.500 on chromosome 1) in the V. harveyi genome. The resulting strains expressed luminescence in a way similar to that of the parental strains, and a significant red fluorescence signal, as a reporter for luciferase induction, was observed at the single-cell level.

Fluorescence of the reporter in the wild-type background at the single-cell level followed essentially the same time course as the induction of luminescence. First, the signal slightly decreased due to inoculation and dilution of the AIs, and the signal intensity then increased (see Fig. S1 in the supplemental material). The maximal fluorescence level observed for the wild type was similar to the one observed for the constitutively QS-active mutant (QS⁺ [ΔluxO]) (see Fig. S1 in the supplemental material). In contrast, in the AI⁻ mutant, the reporter showed only a very faint fluorescence signal even at late time points, showing that there was no readthrough or unspecific transcription (Fig. 3G). The addition of all three AIs to the AI⁻ mutant restored the fluorescence signal to levels comparable to the ones observed for the wild-type (Fig. 3G). These preliminary observations confirmed that P_luxC-mCherry was suitable to monitor signal integration at the single-cell level. The use of the AI⁻ mutant deleted for sensor genes in every possible combination and the addition of external AIs allowed us to study every combination of kinase/phosphatase activity of the complex QS cascade in a few distinct V. harveyi mutants. This strategy also enabled us to control the pool of AIs present in the medium.

FIG 3.

Single-cell microscopy imaging and QS activation upon induction of sensor mutants with different AIs. V. harveyi mutants were grown to an OD₆₀₀ of 0.1 and induced with different AI combinations. Microscopy pictures (exposure times of 400 ms) were taken 6 h after induction and depict levels of QS activation as revealed by a P_luxC-mCherry reporter fusion. (A to C) AI⁻ mutants (LuxM⁻ LuxS⁻ CqsA⁻) lacking the indicated sensors and treated with one AI; (D to F) AI⁻ mutants lacking the indicated sensors and treated with AIs in different combinations; (G) AI⁻ mutant treated with AIs in different combinations. Arrows represent schematically the strengths of kinase (K) and phosphatase (P) activities. Average fluorescence and noise values (standard deviations divided by means from 500 cells [in boldface type]) are presented in the right panels.

The level of noise in the system was quantified by dividing the standard deviation for samples of 500 cells by the sample mean; the values are presented in Fig. 3 (right). The maximum noise level in an apparently homogeneous population was set to be 0.19. All noise values above this threshold (i.e., the standard deviation exceeds 19% of the mean value for that sample) were then considered to reflect heterogeneous gene expression. This threshold was usually not reached when mutants were grown in the absence of AIs (noise values ranging from 0.05 to 0.13) (Fig. 3, bottom images in panels). Under these conditions, cells produced low and homogeneous fluorescence signals. However, the LuxQ⁻ CqsS⁻ mutant behaved differently (Fig. 3C): even in the absence of HAI-1, some cells activated QS, and the noise value reached 0.58. Activation of QS in this mutant in the absence of any AI was already detected at the population level (Fig. 2). As expected, homogeneous activation of QS was observed for all mutants when all available sensors were saturated with their cognate AIs (noise values ranging from 0.10 to 0.19) (Fig. 3, top images in panels).

Homogeneous activation was also observed for mutants under the following distinct conditions: when AI-2 and CAI-1 were added to the AI⁻ mutant (Fig. 3G), when CAI-1 was added to the AI⁻ LuxQ⁻ mutant (Fig. 3E), and when AI-2 was added to the AI⁻ CqsS⁻ mutant (noise values ranging from 0.15 to 0.17) (Fig. 3F). Under these three conditions, QS was fully activated (average fluorescence values of >20,000 arbitrary units [AU]).

Increased noise values were measured for the AI⁻ mutant induced with AI-2 or CAI-1 (noise values of 0.39 and 0.41, respectively) (Fig. 3G) or the AI⁻ LuxN⁻ mutant supplemented with AI-2 or CAI-1 (noise values of 0.33 and 0.31, respectively) (Fig. 3D). In these four cases, one sensor (LuxQ or CqsS) was in the kinase mode and the other was switched into the phosphatase mode by the addition of the corresponding AI. The increase in noise was statistically evaluated by comparing the fluorescence signal distributions in the mutant populations: the fluorescence distributions between the AI⁻ and AI⁻ plus AI-2 or the AI⁻ and AI⁻ plus CAI-1 populations were significantly different, with P values of 7.69 × 10⁻²⁷ and 1.02 × 10⁻³², respectively, as were those between the AI⁻ LuxN⁻ and AI⁻ LuxN⁻ plus AI-2 or AI⁻ LuxN⁻ plus CAI-1 populations, with P values of 1.05 × 10⁻²⁸ and 5.50 × 10⁻³⁴, respectively. In contrast, the addition of HAI-1 to the AI⁻ mutant did not increase the noise in fluorescence (noise values of 0.06) (Fig. 3G), and the fluorescence distributions between the AI⁻ and AI⁻ plus HAI-1 populations were not significantly different (P value of 0.22).

In summary, when the QS system was saturated, meaning that every sensor present received its cognate AI and switched into the phosphatase mode, QS was homogeneously activated, with little cell-to-cell variability. When the QS system was not saturated and when one sensor(s) was in the phosphatase mode and the other(s) was in the kinase mode, heterogeneity in the activation of the QS cascade was observed at the single-cell level. Thus, the level of heterogeneity was determined by the ratio of the kinase activity to the phosphatase activity of the three sensors.

Synthase mutants also display heterogeneity in QS activation.

To ensure that the heterogeneous behavior observed with the addition of external AIs was not caused by artificial effects (unnatural time point or an artificial AI concentration, etc.), we introduced the P_luxC-mCherry reporter into synthase mutants that could produce only one or two AIs (Fig. 4; see also Table S1 in the supplemental material). These mutants were assessed for growth and luminescence production. None of the mutants showed a growth defect (data not shown). Pairwise comparisons of noise distributions between each mutant showed that there was no significant difference between the WT and QS⁺ strains (P value of 0.76) but that all other distributions were significantly different (P values ranging from 2.48 × 10⁻³⁴ to 0.02). As reported previously, CAI-1 is a late signal and is produced only upon the transition into stationary phase (13). Thus, after 6 h of growth in AB medium, the CAI-1 concentration was insufficient to induce any fluorescence signal (LuxM⁻ LuxS⁻ CqsA⁺) (Fig. 4). The only hint of any effect of CAI-1 at this time point came from the ΔcqsA mutant (CqsA⁻ LuxM⁺ LuxS⁺), which exhibited a noisier distribution of the fluorescence signal (noise value of 0.27) than did the wild type (0.13), although its mean fluorescence level was similar to that observed for the wild type (Fig. 4). After 24 h of growth, CAI-1 activated QS in some cells of the luxM luxS (LuxM⁻ LuxS⁻ CqsA⁺) double-deletion mutant, and noise increased (see Fig. S2 in the supplemental material). The AI with the greatest impact on QS activation was AI-2. The deletion of luxS (LuxS⁻ LuxM⁺ CqsA⁺) markedly reduced luminescence, and the fluorescence signal was heterogeneous, ranging from nonactivated cells to fully activated cells (Fig. 4). Similarly, in the case where only AI-2 was synthesized (LuxM⁻ CqsA⁻ LuxS⁺), luminescence was induced, and some cells displayed relatively high levels of fluorescence (noise value, 0.24).

FIG 4.

Single-cell microscopy imaging and mCherry fluorescence distributions in synthase mutants. The box plots represent the distributions of mCherry fluorescence signals from 500 cells under each condition after 6 h of growth (left axis). Noise values (standard deviations divided by means from 500 cells) are presented below the plot. For the wild type, the QS⁺ mutant (LuxO⁻), and single-synthase mutants, pictures were taken with a 200-ms exposure time. For the AI⁻ mutant (LuxM⁻ LuxS⁻ CqsA⁻) and double-deletion mutants, pictures were taken with a 400-ms exposure time. The overall luminescence of the population is indicated below the graph. RLU, relative light units in counts per second per milliliter per OD₆₀₀.

The noise of each mutant compared to that of the wild type or to the AI⁻ mutant was evaluated at different time points (see Fig. S2 in the supplemental material). Our results indicate that even after 24 h of growth, the distribution of the fluorescence signal remained very noisy, excluding the possibility that some cells would be in a form of latency and would switch on their QS system with a certain delay. The characteristic feature common to these synthase mutants is that they always have a certain ratio of counteracting kinase and phosphatase activities of the QS sensors at all times.

The balance of QS cascade components influences heterogeneous QS activation.

Our results thus far indicate that the phosphorylation cascade determines heterogeneous QS activation. Phosphorylated LuxO regulates the expression of the Qrr sRNAs, which in turn regulate multiple mRNA targets, including luxR, luxO, luxM, and aphA (Fig. 1). With the next experimental setup, we demonstrated how a slight change in the balance of the QS cascade components is sufficient to affect the heterogeneity of the QS output. In this experiment, we analyzed the effect of a second copy of the luxR gene (under the control of its native promoter), encoding the high-density master regulator, on the induction of the P_luxC-mCherry promoter in the wild type and the synthase mutants at the single-cell level.

The insertion of an extra copy of luxR in the genome of the wild type had no major impact on the fluorescence or level of noise, indicating that when all sensors are switched into the phosphatase mode, luxR mRNA is in excess compared to Qrr sRNAs. Conversely, in the absence of AIs, when all sensors act as kinases, Qrr sRNAs are in excess compared to luxR mRNA despite the additional luxR copy, since no fluorescence is induced under this condition (compare Fig. 4 and 5).

FIG 5.

Single-cell microscopy imaging and mCherry fluorescence distributions in synthase mutants harboring two copies of the luxR gene. The box plots represent the distributions of the mCherry fluorescence signals of 250 cells under each condition after 6 h of growth. Noise values (standard deviations divided by means from 250 cells) are presented below the plot. For wild-type and single-synthase mutants, pictures were taken with a 200-ms exposure time. For the AI⁻ mutant (LuxM⁻ LuxS⁻ CqsA⁻) and double-deletion mutants, pictures were taken with a 400-ms exposure time.

However, the most striking phenotype was observed for the double-synthase mutants harboring two genomic copies of luxR. These mutants produced higher levels of fluorescence than did those that have only one copy of luxR (compare Fig. 4 and 5). In addition, the noise of QS activation increased in the mutants harboring two luxR copies, as can be seen by both noise quantification and comparing the sizes of the box plots (compare Fig. 4 and 5). For example, the average fluorescence of the LuxS⁻ CqsA⁻ mutant increased from 4,759 AU (single luxR copy) to 7,315 AU (two luxR copies), and the noise value increased from 0.08 to 0.20 in these mutants (Fig. 4 and 5).

The single-synthase mutants showed a different phenotype, with a tendency for noise reduction. For example, the mutant that produces HAI-1 and CAI-1 (LuxS⁻) exhibited the highest noise level (noise value of 0.44) when only one copy of luxR was present (Fig. 4). The corresponding mutant with two luxR copies produced higher levels of fluorescence, but the intercellular fluorescence variability was 2-fold reduced (noise value of 0.21) (Fig. 5). It is hypothesized that the low level of phosphorylated LuxO in the single-synthase mutants causes a low level of Qrr sRNAs, which is certainly outcompeted by the higher level of luxR mRNA, and LuxR, as well as its target genes, is expressed at a higher and less noisy level.

These results underline the correlation between the design of the QS cascade, with several feedback regulation mechanisms, and the noise of the output at the single-cell level. In this regard, it is worth mentioning that according to our qRT-PCR results, luxR transcripts were found to be in excess compared to luxO or luxU transcripts (transcript levels as a percentage of recA levels of 20% for luxO, 61% for luxU, and 6,987% for luxR). According to these results, LuxO seems to be expressed at a low level, and therefore, it is conceivable that the degree of its phosphorylation is critical for the level of heterogeneous QS output.

DISCUSSION

Several bacterial species are known to produce and perceive more than one AI. For example, the plant pathogen Ralstonia solanacearum (44) and the human pathogen Vibrio cholerae (45) produce and respond to two AIs; others, such as Pseudomonas aeruginosa (46) and Vibrio harveyi, produce and respond to four and three AIs, respectively. In each case, the circuits in which the different AIs are integrated are interdependent: one system can be inducing a second one (as in R. solanacearum), some systems can induce or repress other systems (as in P. aeruginosa), or the different systems are integrating all AIs into the same cascade. The latter case holds for V. cholerae and V. harveyi, in which two/three signals are integrated into one signaling cascade. In order to fully understand why such a complex QS cascade with several AIs and several feedback loops has evolved, it is necessary to study the QS system with all its inputs. Therefore, we generated a complete set of mutants to study the impact of each AI and sensor individually and in combination on the activation of the QS cascade in V. harveyi at the population level, using luminescence as a natural readout, and at the single-cell level, using a P_luxC-mCherry reporter fusion.

To avoid intercellular variability due to fluctuations in plasmid copy numbers, we introduced Tn7-based reporter fusions into the genome of V. harveyi. Under these conditions, and when all three AIs are present, QS activation occurs homogeneously across the population, with cell-to-cell variability (standard deviation) not being higher than 19% of the mean (Fig. 3 and 4). However, when one or two AIs were absent, some sensors were in the kinase mode, while others were in the phosphatase mode. In such a situation, full activation of the QS cascade is prevented, leading to a highly heterogeneous answer at the single-cell level, with a cell-to-cell standard deviation reaching up to 44% of the mean, e.g., when AI-2 was absent. Under all the conditions tested, the mCherry distribution was always Gaussian or single peaked; only the width of the distribution was significantly affected. Thus, a bimodal distribution could not be detected in our setup of experiments. In some instances, such as the regulation of chemotaxis in Escherichia coli, it was shown that a noisy distribution of gene expression leads to a bimodal response; the switch from noise to bistability in this case arises from the presence of a positive feedback loop in the regulatory circuits governing chemotaxis (47). However, while several negative feedback loops govern the fate of the different components of the V. harveyi QS cascade, none of them is involved in the reactivation of the QS system.

In agreement with previous studies (30), we show that the V. harveyi QS system employs gradient integration involving all three AIs. The more AI types that are engaged, the greater the level of activation of the QS cascade and the more homogeneous the response becomes. This is true even though, due to the negative feedback loop acting on LuxN, HAI-1 had a markedly lesser impact on the activation of QS than did either of the other AIs. Our data show that the LuxN/HAI-1 system has the least impact, CqsS/CAI-1 has a moderate impact, and LuxQ/AI-2 has the greatest impact on QS activation (considering bioluminescence as a readout). This hierarchy is not in full agreement with data from previous studies where HAI-1 was shown to have an important impact on QS activation (11). This discrepancy could be explained by different environmental and growth conditions, such as differences in the medium composition or in shaking conditions, both of which influence AI production. Notably, the small impact of the LuxN/HAI-1 system was confirmed with the AI⁻ LuxQ⁻ CqsS⁻ mutant, in which the kinase activity of LuxN was not sufficient to keep the QS cascade strictly switched off and luminescence or fluorescence was produced. These results support data from previous studies suggesting that HAI-1 becomes important at later time points when AI-2 is also present (13). Indeed, on its own, AI-2 is able to partially switch the QS cascade on, therefore reducing the number of sRNAs and the feedback regulation of luxNM. Thus, LuxN and HAI-1 could play a more prominent role once this cascade has been triggered by AI-2 or CAI-1, perhaps in the maintenance of the QS output or induction of late QS phenotypic traits. Additionally, we show not only that the presence of one AI is important but also that the “state” (phosphatase or kinase) of its cognate sensor drives the integrity and homogeneity of the QS response. For example, it is possible to induce the QS cascade (up to 10% of the wild-type level, based on luminescence measurements) in the absence of all AIs by deleting luxQ and cqsS. Indeed, in this situation, the presence of LuxN as the only sensor, negatively feedback regulated, is sufficient to turn the QS system on in some cells. This heterogeneous activation of the QS cascade was triggered independently from any external cue, arguing that the noisy activation of QS in the different mutants and under the different conditions is not dependent on a different AI perception from one cell to another but rather is dependent on the internal state of the different cells.

Single-synthase mutants, in which a low number of Qrr sRNAs is expected due to a low level of phosphorylated LuxO, produced a less noisy output when they contained a second luxR copy in the genome. In contrast, the level of noise increased in the double-synthase mutants harboring a second luxR copy. These results underline the correlation between the tight regulation of the components of the QS cascade and the degree of noise. Furthermore, in view of the high level of luxR transcripts in comparison to those of the luxU and luxO transcripts, the importance of the components of the phosphorylation cascade for the degree of heterogeneity in output is further supported. Interestingly, several of the known feedback loops tightly control luxO expression, e.g., autorepression of luxO and degradation of luxO and luxN mRNAs by the Qrr sRNAs. The extremely low LuxO protein levels also prevented our efforts to quantitatively analyze the phosphorylated state of LuxO in the different subpopulations. Thus, the design of the QS cascade in V. harveyi clearly has the ability to drive the heterogeneous expression of QS-regulated genes.

Our observation that mutants that could not produce CAI-1 or AI-2 still displayed a high level of heterogeneity even after 24 h of cultivation indicates that cells that were in an off state did not turn to an on state, and no tendency toward a more homogeneous QS answer was observed. This hysteresis can be explained by the fact that AIs are taken up (in the case of AI-2) (4) or are unstable (in case of CAI-1), and their production is transient (13). Therefore, it seems likely that a cell that possesses a high level of phosphorylated LuxO and consequently a high number of Qrr sRNAs has no chance to perceive any further signal and is unable to switch into the on state.

As it is necessary to consider the QS cascade of V. harveyi with its three inputs, it is also important to remember that this bacterium is found in different environmental niches: as a free-living microbe in the ocean, as a pathogen of several crustaceans and fish, or at the surface of algae. In addition, the production of AIs strongly depends on the metabolic state of the bacterium. Thus, besides QS, the production and perception of AIs might also reflect prevailing environmental conditions. Under laboratory conditions, in a test tube with all AIs available, the wild-type strain of V. harveyi induces its QS system in a homogeneous way. When one AI is missing, QS activation becomes heterogeneous. We believe that under natural conditions, V. harveyi has to face completely different environmental situations, and we can envision several scenarios: inside a biofilm, when all nutrients are available and/or all AIs are concentrated, cells switch on their QS system fully, but when cells are situated at the surface of the biofilms, where nutrients and AIs are washed away, division of labor is necessary for the survival of the population, with some cells expressing proteases (activated by QS) and some cells expressing siderophores or motility (repressed by QS). An alternative hypothesis is that under challenging conditions (e.g., low nutrient and therefore low AI concentrations), heterogeneity in QS activation is a form of bet-hedging strategy to ensure that at least part of the population adopts the phenotype that best promotes its survival. Furthermore, it was previously described with a theoretical model that in a changing environment, a heterogeneous population is more fit than a homogeneous one (48).