Bacterial Quorum Sensing Stabilizes Cooperation by Optimizing Growth Strategies

ABSTRACT

Communication has been suggested as a mechanism to stabilize cooperation. In bacteria, chemical communication, termed quorum sensing (QS), has been hypothesized to fill this role, and extracellular public goods are often induced by QS at high cell densities. Here we show, with the bacterium Vibrio harveyi, that QS provides strong resistance against invasion of a QS defector strain by maximizing the cellular growth rate at low cell densities while achieving maximum productivity through protease upregulation at high cell densities. In contrast, QS mutants that act as defectors or unconditional cooperators maximize either the growth rate or the growth yield, respectively, and thus are less fit than the wild-type QS strain. Our findings provide experimental evidence that regulation mediated by microbial communication can optimize growth strategies and stabilize cooperative phenotypes by preventing defector invasion, even under well-mixed conditions. This effect is due to a combination of responsiveness to environmental conditions provided by QS, lowering of competitive costs when QS is not induced, and pleiotropic constraints imposed on defectors that do not perform QS.

IMPORTANCE Cooperation is a fundamental problem for evolutionary biology to explain. Conditional participation through phenotypic plasticity driven by communication is a potential solution to this dilemma. Thus, among bacteria, QS has been proposed to be a proximate stabilizing mechanism for cooperative behaviors. Here, we empirically demonstrate that QS in V. harveyi prevents cheating and subsequent invasion by nonproducing defectors by maximizing the growth rate at low cell densities and the growth yield at high cell densities, whereas an unconditional cooperator is rapidly driven to extinction by defectors. Our findings provide experimental evidence that QS regulation prevents the invasion of cooperative populations by QS defectors even under unstructured conditions, and they strongly support the role of communication in bacteria as a mechanism that stabilizes cooperative traits.

INTRODUCTION

Cooperative behavior is a widespread phenomenon that pervades all levels of biological organization and has helped to catalyze all major transitions in the history of life (1). Extensive cooperation is also prevalent among microbes and plays fundamental roles in many bacterial processes, including biofilm formation, virulence, bioenergy, host-microbe interactions, and the formation of stable communities that maintain essential ecosystem functions (2–7). One class of microbial cooperative behaviors is the production of public goods (PG), i.e., products that provide benefits to both producers and nonproducers within a community. In the context of PG, individuals that contribute by producing the good are defined as cooperators, while nonproducers are defined as defectors. Importantly, defectors are not always inherently cheaters but are capable of cheating if they reap fitness advantages by exploiting social behaviors (8–10). Conditional participation through phenotypic plasticity driven by communication could provide a potential solution to this dilemma (11).

Quorum sensing (QS) is a ubiquitous form of chemical communication in bacteria that allows cells to sense changes in local cell density to globally alter gene expression (12, 13). Many QS regulons appear to be enriched for secreted PG products (14), and it has been hypothesized that QS could act as a mechanism to stabilize cooperation (15). Additionally, QS has been shown to interact with other mechanisms that function to stabilize cooperative behaviors, such as metabolic constraint, metabolic prudence, and policing (16–19).

The postulated role of QS to innately stabilize cooperative traits has not been widely demonstrated, however. Most empirical studies of laboratory and natural populations suggest that QS systems can be invaded and in some cases destabilized, such as the invasion of both Pseudomonas and Vibrio species by QS defectors (20–22). In fact, the range of conditions under which cooperation is favored may be quite limited (23). This raises the question of why QS frequently regulates the expression of PG and whether it is in fact advantageous for maintaining cooperative traits.

Here we address this question by examining PG commonly regulated by QS, i.e., extracellular protease production in the bioluminescent marine bacterium Vibrio harveyi. This QS system is ideal for studying the role of communication in stabilizing cooperation because both defectors and unconditional cooperators (UCs) can be easily generated through directed mutagenesis of the QS pathway. By assessing the competitive outcomes of cooperators and defectors in an environment in which fitness is affected by QS-regulated protease production, we demonstrate that a wild-type (WT) strain with a functional QS system prevents defector invasion, whereas an UC mutant strain is driven to extinction. QS prevents cheating by regulating a switch in growth strategies from maximizing the growth rate at low cell densities by restraining protease production to increasing the growth yield at high cell densities, when available resources become more scarce. Whereas the WT strain can use QS to take advantage of these different strategies at the appropriate densities, defector and UC strains maximize only rate or yield, respectively, with a resulting trade-off in the other capacity. Our results suggest that prudent regulation of metabolism by QS is a mechanism by which this system promotes the maintenance of cooperation.

MATERIALS AND METHODS

Bacterial strains, media, and growth conditions.

Bacterial strains used in this study included Vibrio harveyi strain ATCC BAA-1116 (24), which was recently reassigned as Vibrio campbellii (25) (for consistency with past literature, however, we use the V. harveyi designation in this study), and derivatives (Table 1). Bacteria were grown in LB broth (Accumedia) or in M9 salts (Sigma-Aldrich or Becton, Dickinson and Company) supplemented with sodium chloride (Macron Fine Chemicals) to a final concentration of 2% (wt/vol), with 0.5% (wt/vol) sodium caseinate (Sigma-Aldrich) (M9-casein medium), Casamino Acids (CAA) (Becton, Dickinson and Company) (M9-CAA medium), or tryptone (Becton, Dickinson and Company) (M9-tryptone medium) as the sole carbon source. Bacteria were routinely grown in a 30°C shaker at 250 rpm in 16-mm borosilicate glass tubes containing 3 ml of medium. Experimental cultures were initially grown in liquid LB cultures and subsequently acclimated by passaging in M9 liquid medium prior to the initiation of experiments. Cells were collected by centrifugation (10,000 × g) and washed in the equivalent growth medium prior to competition experiments. Supernatants were obtained from cultures grown in M9-casein medium for 24 h. Cells were removed by first pelleting at 10,000 × g and then filtered with 0.22-μm filters (Millipore). Supernatants were added back at the percentages of the final culture volume indicated in Fig. 2B.

TABLE 1.

Description of strains used

Strain	Genotype	Strain description
BB120	ATCC BAA-1116	Wild-type strain; has functional QS system
KM669	ΔluxR	Knockout mutant of gene encoding QS master regulator LuxR; acts as defector
JAF78	ΔluxOU	Knockout mutant of genes encoding LuxO and LuxU; acts as unconditional cooperator
KM83	luxO D47E	Point mutant, producing LuxO that constitutively mimics phosphorylated state; acts as defector
JMH634	ΔluxM ΔluxS ΔcqsA	Triple signal synthase mutant; unable to induce QS but responds to supplemented signal
CW2001	ΔluxAB	Strain with luciferase genes deleted; has functional QS system but is unable to produce light

FIG 2.

QS regulation of extracellular protease activity in V. harveyi. (A) Protease levels for cooperator and defector genotypes grown in M9-casein medium, plotted versus growth (measured as CFU per milliliter) (n = 3). Error bars, standard errors of the mean. (B) Effects of supernatant supplementation. Cultures of the ΔluxR strain were supplemented with supernatants from the strains indicated, at the indicated concentrations (n = 4). Error bars, 95% confidence intervals. ***, P < 0.001; N.S., not significant.

Competition fitness assays. (i) Competition design.

Competitors were assessed for their ability to invade when rare with the initiation of populations over a range of relative frequencies of defector and relevant cooperator strains, for single-growth-phase competition assays. Mixed populations were grown for 24 h and plated on LB agar, and population compositions were assessed. For serial competition assays, populations were initiated by mixing strains at a ratio of 99:1 on day zero. Populations were grown for 24 h and plated on LB agar for assessment of the population composition and productivity, and a subset was diluted 1/1,000 into new M9 liquid medium, followed by 24 h of growth (∼10 generations). During competition assays, bioluminescence phenotypes were determined in the dark by using an AlphaImager HP imaging system (ProteinSimple). Ancestral cooperator strains uniformly produce bright bioluminescent colonies, while all defector strains investigated were nonluminescent in appearance.

(ii) Fitness calculations.

When determined, defector relative fitness (w) was calculated as the ratio of Malthusian parameter (m) values for strains in pairwise competitions, which equates to the ratio of competitor doublings over the experimental time interval of 24 h (26). The Malthusian parameter is calculated as m_i = ln(x₁/x₀) for competitor i, where x₀ and x₁ are the densities (in CFU per milliliter) of that competitor at the start (x₀) and end (x₁) of the experimental period and w = m₁/m₂, comparing competitors 1 and 2.

Protease assays.

Extracellular protease activity was assessed using a protease fluorescence assay (Sigma-Aldrich), with filtered supernatants (0.45-μm-pore filters; Ambion) from cultures grown in M9-casein medium. Conjugated casein molecules release the fluorophore fluorescein isothiocyanate (FITC) when cleaved, allowing sensitive assessment of protease activity. Fluorescence was measured with an EnVision multilabel plate reader (PerkinElmer), with an excitation wavelength of 485 nm and an emission wavelength of 535 nm. Measurements were normalized with respect to the dilution used for supernatant samples and the cell density of the cultures, as determined by viable plate counting.

Statistical analyses.

Statistical analyses were conducted using R 3.2.2. The growth of different strains in various M9 media was compared using analysis of variance (ANOVA) (population density = strain type [cooperator or defector] × carbon source) with Tukey's honestly significant difference (HSD) post hoc test to account for multiple comparisons (Fig. 1). Protease production was compared across three strains at 24 h of growth in M9-casein medium using ANOVA (per-capita protease activity = cell density + strain type) (Fig. 2A). The effect of supplementing different supernatants on ΔluxR strain growth in M9-casein medium was compared by using ANOVA (growth [measured as optical density at 600 nm {OD₆₀₀}] = supernatant source × supernatant concentration) with Tukey's HSD post hoc test (Fig. 2B). Linear models were constructed to analyze the effects of relative fitness versus frequency estimates in different M9 environments (competitor relative fitness = competition pairing × starting competitor frequency) (Fig. 3; also see Fig. S6 in the supplemental material). In experiments examining the dynamics of growth (Fig. 4 and 5), data points are bounded with 95% confidence intervals to allow statistical comparisons.

FIG 1.

QS requirement for maximal growth of V. harveyi in M9-casein medium. Six strains of V. harveyi were grown for 24 h in M9 medium with different sole carbon sources and were enumerated through viable cell counting (CFU per milliliter). (A) Bar graph showing genotype productivity. Bars, productivity of a given genotype at the culmination of a single growth cycle (24 h) (n = 4 separate biological replicates); error bars, 95% confidence intervals for the mean estimates. WT, wild-type strain; SN, signal-negative strain; D47E, luxO D47E strain. Differences that were statistically significant, compared with the WT strain in the given medium type, are reported. *, P < 0.05; ***, P < 0.001. (B) Images of cooperator and defector genotypes in M9-casein medium after 24 h of growth.

FIG 3.

Fitness outcomes from single-growth-cycle competitions between different V. harveyi genotypes. In pairwise competitions between different V. harveyi strains, the starting competitor frequency was altered and competition experiments were completed in M9-casein medium for 24 h. Each point corresponds to a single competition outcome, and fit lines consist of regression fits (solid lines) bounded by 95% confidence intervals (dashed lines) for each competition pairing, according to a linear model. For each pairing, the strain whose relative fitness is reported on the y axis and starting percentage on the x axis is underlined.

FIG 4.

Invasion of the ΔluxOU strain but not the WT strain by the ΔluxR defector. Populations contained one of the two cooperator strains, i.e., the WT strain or the ΔluxOU strain, and the ΔluxR defector at a 99:1 starting mixture, or vice versa. Cultures were grown for 24 h in M9-casein medium, diluted 1,000-fold, and repeated over multiple growth cycles. (A) Defector frequency (determined as nonluminescent colonies) plotted versus generations of growth. (B) Population productivity (measured as CFU per milliliter) plotted versus generations of growth. Error bars, 95% confidence intervals (n = 3 biological replicates per treatment).

FIG 5.

WT strain optimization of growth strategies at low and high cell densities. (A) Bioluminescence induction of the WT strain and the ΔluxOU strain in monoculture over the course of a growth cycle in M9-casein medium. RLU, relative light units. (B) Defector frequency over a single growth cycle for the ΔluxOU strain versus the ΔluxR strain, the ΔluxOU strain versus the WT strain, and the WT strain versus the ΔluxR strain (with the defector strains underlined), started at 1% of the total population and competed in M9-casein medium. (C) Growth of the WT strain, the ΔluxR strain, and the ΔluxOU strain in monoculture over the course of a single growth cycle. Error bars, 95% confidence intervals (n = 4 biological replicates per treatment).

RESULTS

QS is required for maximum growth of Vibrio harveyi utilizing casein.

To evaluate the stability of QS and the cooperative behaviors it regulates against defector invasion in V. harveyi, we first established conditions under which fitness was dependent on QS activation of PG production. We found that growth in M9-casein medium was highly dependent on the ability to induce QS at high cell densities. In this environment, casein breakdown through extracellular protease production is required to reach the maximum growth yield. In V. harveyi, the accumulation of autoinducer increases expression of the gene encoding the master transcription factor LuxR, switching gene expression from the low cell density state to the high cell density state. We observed that only cooperator strains, including the WT conditional cooperator, which induces luxR in response to autoinducer as the cell density increases, and the UC ΔluxOU mutant, which constitutively expresses luxR and subsequently the high-cell-density QS regulon, reach maximum densities in this medium after 24 h of growth (Fig. 1). Defectors, including the ΔluxR strain (which cannot produce the LuxR master regulator), grew to a significantly lower density of around 4% of the WT density in M9-casein medium (P < 0.001) (Fig. 1). This outcome was observed for all defector strains tested that could not induce luxR expression, including the luxO D47E mutant and an autoinducer “signal-negative” (SN) production mutant (Fig. 1). The production of bioluminescence did not affect growth under these conditions, as the ΔluxAB dark mutant grew identically to the WT strain. In contrast, the use of predigested carbon sources such as CAA or tryptone significantly diminished the differences in growth between cooperator and defector strains (Fig. 1).

Functional and mutant QS strains differ in density-dependent protease activity.

To confirm that these growth results were due to extracellular protease activity, we monitored and quantified activity during the growth of cultures started at low cell density (Fig. 2). As expected, the ΔluxOU strain demonstrated high protease activity at all cell densities, as this strain is locked in the high-cell-density QS state. The WT strain exhibited QS regulation of protease production, with cells initially exhibiting high protease activity immediately upon back-dilution from high cell density. Protease activity decreased until a quorum was reached, but it ultimately matched the protease activity observed in the ΔluxOU strain at high cell density (Fig. 2A). The ΔluxR strain exhibited low or undetectable levels of protease activity at all densities examined. It should be noted that the ΔluxR mutant grew poorly in this medium and did not reach the cell density of the cooperator strains (Fig. 1). Our data showing that WT and ΔluxOU strains of V. harveyi have higher levels of protease activity than QS defector strains agree with earlier work in this and other QS systems (14, 27, 28).

The poor growth of the QS defector ΔluxR strain in M9-casein medium was not due to an inability to use peptides liberated from casein, as the growth of the ΔluxR strain in this environment could be rescued by the addition of supernatants from the WT or ΔluxOU cooperator strains, but growth was not improved by its own supernatant (P < 0.001) (Fig. 2B). It is also worth noting that supernatants from all examined defector strains induced bioluminescence expression of an autoinducer mutant of V. harveyi nearly 100- to 1,000-fold less than did supernatants from cooperator strains, a level lower than that predicted by growth differences in M9-casein medium alone, which suggests that these defectors produced low levels of autoinducer and would be unlikely to cheat by inducing cooperators to produce PG (see Fig. S1 in the supplemental material).

A functional QS circuit prevents defector invasion.

Although higher levels of protease production by the WT and ΔluxOU strains lead to increased growth in M9-casein medium, this behavior is potentially susceptible to cheating by defectors that could benefit from the nutrients liberated by these extracellular proteases (Fig. 2B). To test whether cooperators that produce higher levels of protease could be cheated, we competed WT and ΔluxOU strains against the ΔluxR defector. The competition experiments were performed in M9-casein medium, the initial frequencies of the competing strains were varied, and populations were started at combined densities significantly lower than the quorum threshold. The competitions consisted of one growth period, and the QS phenotype of isolated colonies was determined on the basis of colony bioluminescence phenotypes, to quantify defectors and cooperators.

Striking differences in defector invasion were observed depending on which cooperator strain was used. While the ΔluxR strain increased in frequency against the ΔluxOU strain at all starting frequencies tested (Fig. 3, red line), it was unable to invade the WT strain at any examined frequency (Fig. 3, black line; also see Fig. S2 in the supplemental material). As a result, the WT strain had fitness equivalent to that of the defector strain across all starting defector frequencies tested, whereas the ΔluxOU strain had much lower relative fitness than the defector. We similarly competed the WT strain against the ΔluxOU strain at various frequencies. Like the ΔluxR defector, the WT strain was able to invade the ΔluxOU strain, leading to higher fitness values at all frequencies examined (Fig. 3, blue line). Although WT-mixed populations were not invaded by the ΔluxR strain at any frequency tested (i.e., the relative fitness of the strains did not vary), these populations exhibited nearly identical frequency-dependent decreases in population yield, compared with the ΔluxOU-mixed populations (see Fig. S3 in the supplemental material), indicating that absolute fitness decreased with increasing defector frequency. Furthermore, this phenomenon persisted over a wide range of dilutions (and thus a wide range of starting densities), with the WT strain performing well against the ΔluxR strain while the ΔluxOU strain performed increasingly worse as the starting dilution (and thus the number of generations that occurred before the maximum density in M9-casein medium was reached) increased (see Fig. S4 in the supplemental material).

Based on these data, we predicted that the ΔluxR strain would drive the ΔluxOU strain to extinction over longer time periods but would be unable to invade the WT strain. We tested this prediction in extended pairwise competition experiments in the M9-casein environment, with either cooperative genotype (the WT strain or the ΔluxOU strain) paired against the defector (the ΔluxR strain); one competitor was seeded at a low starting frequency of ∼1% of the population, to evaluate its ability to invade. The competitions were extended over several growth cycles, with ongoing daily dilutions. From either starting condition, defectors rapidly swept and dominated the ΔluxOU strain in mixed populations (Fig. 4A). This resulted in underutilization of the available nutrient resources and a consequent “tragedy of the commons,” with the productivity of the population dropping to the levels of the defector strain grown in isolation (Fig. 4B) (29–31). The lack of stable coexistence with either frequency tested (defector rare or defector common) suggests that these competing strategies (ΔluxOU and ΔluxR) are unlikely to coexist in this environment.

When the WT cooperator strain was competed against the ΔluxR defector strain in casein-limited M9 medium, neither strain significantly invaded the other, as both genotypes maintained their starting frequencies (Fig. 4A). Unlike competitions against the UC ΔluxOU strain, the resulting yield of the WT-mixed populations mirrored the input frequencies and was stable over this time period (Fig. 4B). In fact, the resulting cell density arrived at in WT versus defector competitions seems to rely strongly on the initial WT frequency, and a high starting frequency of WT cooperators was able to prevent a tragedy of the commons over this time scale.

A functional QS system prevents defector invasion by modulating growth strategies at different cell densities.

As QS is a mechanism that can alter gene expression in response to changes in cell density, we sought to determine the dynamics of QS induction and the kinetics of defector invasion in the M9-casein environment. To determine the QS induction state at different densities in the M9-casein environment, we monitored per-capita bioluminescence as a proxy for activation of QS. As expected, the WT strain exhibited a strong drop in bioluminescence at low cell density; beyond 6 h of growth, however, a quorum was reached and bioluminescence was ultimately induced to maximal levels matching or even exceeding those of the ΔluxOU strain (Fig. 5A). In contrast, the ΔluxOU strain maintained a nearly constant level of bioluminescence, indicating that it was activating QS in an unconditional fashion (Fig. 5A). The regulation of bioluminescence in these strains mirrors the patterns seen for protease activity, as would be expected for two phenotypes similarly induced by QS in the high cell density state (Fig. 2A).

To determine the point during growth at which defectors invade cooperators, we measured the frequency of defectors started at 1% in three pairwise competitions, i.e., ΔluxOU strain versus WT strain, WT strain versus ΔluxR strain, and ΔluxOU strain versus ΔluxR strain (the defector strains are underlined), over the course of one growth cycle. We observed that significant invasion of the ΔluxOU strain by both the WT strain and the ΔluxR strain began to occur as the population reached quorum, and these strains experienced large increases in frequency over the next 10 to 12 h (Fig. 5B). At higher cell densities, the invasion of both of these strains plateaued and the ΔluxOU strain maintained a minor frequency in the population. These findings indicate that invasion occurred primarily at the transition from low to high cell densities. As seen in earlier results, the ΔluxR defector could not invade the WT strain at any density examined (Fig. 5B; also see Fig. S5 in the supplemental material).

To further understand these dynamics, we measured the growth of the WT, ΔluxR, and ΔluxOU strains over the course of a single growth cycle in monoculture. The WT and ΔluxR strains exhibited equivalent rapid growth at low cell densities, compared with the UC ΔluxOU mutant (Fig. 5C). This difference was especially evident beyond the 6-hour time point, matching the time period during which the ΔluxOU strain was invaded and increases in defector frequency became readily apparent in competitions (Fig. 5B). As the WT strain grows to higher cell densities, it transitions to a new growth strategy, allowing growth to resume and cell numbers to increase until the maximum density is reached. This portion of growth more closely resembles that of the ΔluxOU mutant, which experiences an increase in growth rate at high cell density and ultimately reaches the same yield as the WT strain (Fig. 1 and 5C). This, in combination with the observation that the ΔluxR strain plateaus in growth before reaching high cell density, suggests that this phase of growth is largely dependent on QS-regulated PG. Because the ΔluxR mutant is unable to alter its growth strategy like the WT strain to produce the PG necessary for additional growth, it obtains a significantly lower population yield than either cooperator strain.

The low growth rate of the ΔluxOU strain at low cell densities is specific to the M9-casein environment, as the ΔluxOU strain exhibits fitness equivalent to that of the WT strain and the ΔluxR mutant when grown in M9 minimal medium supplemented with tryptone (see Fig. S6 in the supplemental material). Furthermore, the greater fitness of the ΔluxR strain, compared to the ΔluxOU strain, is not simply due to the higher individual growth rate of the former but rather is social in nature, as the ΔluxR strain grows to higher densities when competed with the ΔluxOU strain than it can by itself in monoculture, which demonstrates that cheating is occurring under those conditions (32) (see Fig. S2 and S5 in the supplemental material). Conversely, the ΔluxR strain does slightly worse in mixed populations with the WT strain (presumably due to resource competition) than in monoculture, in both M9-casein and M9-tryptone environments (see Fig. S5 and S6).

DISCUSSION

QS has been proposed as a mechanism to stabilize cooperation, but only modest experimental evidence exists to support this proposition (4, 16, 33–36). Because nonproducing defectors do not pay the cost to produce PG but can still reap the resulting benefits, a fundamental question is how QS systems, and the cooperative behaviors they regulate, are stably maintained in bacterial populations. Most evidence addressing this question is based on theory and simulations, with few experimental population dynamics data supporting it.

Although it has recently been shown that QS can serve to modulate investment in cooperative goods, our study demonstrates that QS allows cooperators to match the fitness of defectors even in well-mixed environments, in the absence of any apparent spatial structure or policing by cooperators (33). More specifically, while previous studies suggested the value of functional QS in comparison to a modeled or simulated constitutive producer (33, 37), this is the first study to explicitly and definitively show by experimental analysis the vastness of the difference between QS producers and UC producers. Moreover, comparing QS Pseudomonas aeruginosa and a simulated constitutive cooperator, Allen et al. concluded that there are environments in which all forms of cooperation are disfavored (33). However, our experimental results demonstrate this is not the case in V. harveyi. The generality of this effect is unknown, and any differences are likely to depend on specific features of the system, such as signal redundancy and stability, within-population heterogeneity, and QS system architecture (38–41). In this case, conditional PG production provided by QS would act as a proximate mechanism to stabilize cooperation, allowing ultimate stabilizing mechanisms of cooperation, such as kin selection, to act (42). Indeed, QS has been suggested to identify the proportion of kin in mixed-species communities, and our finding that defector strains produce less signal corroborates the ability to identify the number of cooperators in mixed single-species populations (37) (see Fig. S1 in the supplemental material).

We first established M9-casein medium as an environment in which absolute fitness outcomes were dependent on the ability of V. harveyi to activate the cell's QS system and produce regulated PG (Fig. 1). Interestingly and contrary to our expectations, the ΔluxR strain consistently and extensively exploited populations of the UC ΔluxOU mutant but was unable to do the same against the WT strain (Fig. 2 and 3). These results have ecological significance, as nonluminescent luxR defectors and UC luxO mutants have all been identified in natural Vibrio populations (43–45). Importantly, this occurred in a well-mixed environment in which no biofilms or cell aggregates were observed and neither assortment of competitors nor privatization of PG would be likely. The equivalent fitness of the WT strain and the ΔluxR defector at all frequencies examined demonstrates that functional QS can promote the maintenance of cooperative behaviors, providing a solution to the PG dilemma in mixed communities (46).

An important consideration in our results is the role of pleiotropy, as the QS regulon of V. harveyi is extensive (47–49). Pleiotropy has been demonstrated as a potential mechanism to inhibit cheating, due to the costs imposed on defectors (17, 50–53). V. harveyi defectors do pay such costs, as all defector strains tested exhibited growth yield defects that spanned multiple environmental conditions, including those in which no proteolytic breakdown of nutrients was required (P < 0.001) (Fig. 1; also see Fig. S6 in the supplemental material). This is likely a result of the QS system's regulation of a large portion of the cell's genome and interconnectedness with the metabolism of the cell to appropriately tune growth across a wide range of environments (17, 34, 54). The results of competition between the WT strain and the ΔluxR strain in M9-tryptone medium and at lower densities in M9-casein medium suggest that the WT strain may indeed be a better nutrient competitor (see Fig. S5 and S6 in the supplemental material). The effect may be explained by the V. harveyi QS regulation of multiple genes related to transport (49). While the WT strain appeared to be neutrally competitive against the ΔluxR strain under the shaken conditions tested and was unable to invade when rare (Fig. 4A), it might be predicted that adding a variable such as a structure allowing assortment between competing genotypes could foster an advantage for the WT strain (55).

Pleiotropy is likely to play a role in the diminished growth rate of the ΔluxOU strain as well. Protease production is not the only trait upregulated by QS, and expression of other public and private goods is likely to affect growth performance. The WT strain may be appropriately expressing not only proteases but also other goods, both public and private. Indeed, regulation of cooperation by QS may even be an exaptation, with the original selection being on regulated private goods; regardless, it is significant for the persistence of these now-regulated cooperative behaviors. However, the ΔluxOU mutant is not maladaptive under all selective conditions, as this mutant excels in monoculture or when high nutrient levels are freely available.

The WT strain was also able to exploit and to invade the unconditional ΔluxOU cooperator (Fig. 3 and 5). This demonstrates that such unconditional cooperation can also be exploited by more restrained forms of cooperation. Indeed, the WT strategy resembles the defector strategy at low densities. Competitive exclusion of the ΔluxOU strain by both the ΔluxR strain and the WT strain exhibits negative frequency dependence, demonstrating some diminishing returns as less of the ΔluxOU strain is present. This is likely due to reduced opportunities for cheating, although this dependence does not elicit coexistence (Fig. 3) (56, 57). At no point does this cheating cease to be favored, characteristic of a prisoner's dilemma (58, 59). Similar experiments with Pseudomonas aeruginosa observed an intermediate frequency at which the cooperators and defectors had equivalent fitness levels (21), illustrating that these QS systems are not functioning equivalently.

In contrast, the WT strain appears to perform in a frequency-independent manner when competed with the ΔluxR defector, which suggests that the defector's performance here is not a result of cheating on PG (Fig. 3; also see Fig. S5 in the supplemental material). This is also supported by the inability of the ΔluxR strain to grow beyond monoculture yields in the presence of the WT strain (see Fig. S2 and S5 in the supplemental material). However, it should be noted that the ΔluxR strain can initially grow rapidly at low cell densities. This suggests that the casein used in our experiments contains nutrients that do not require protease for utilization by V. harveyi. The observation that the ΔluxR strain plateaus and cannot reach a high cell density is inconsistent with a model in which leaky protease production is responsible for the growth at low cell densities, as we would expect for the ΔluxR strain to maintain a lower growth rate throughout the experiment. Importantly, the ΔluxR strain cannot reach high cell densities on its own but is perfectly capable of doing so when provided with supernatants from cooperator strains, including the WT strain (Fig. 2B), and in cocultures with the ΔluxOU strain (see Fig. S2 and S5). Moreover, all signs of frequency dependence were removed when the same strains were competed in M9-tryptone medium, in which casein has been enzymatically digested (see Fig. S6 in the supplemental material). Under these conditions, the ΔluxOU strain exhibited fitness equivalent to that of the strains that invaded it in M9-casein medium. Therefore, our results are not simply due to the ΔluxOU strain exhibiting poor fitness in all environments examined.

The competition results obtained with the V. harveyi QS system are different from those previously described for other systems, especially with regard to the apparent ability to withstand invasion by a defector strain that is unable to induce the high-density QS state. For example, a lasR mutant (another QS defector) of P. aeruginosa was able to successfully invade a QS WT strain under conditions that required proteolysis (17, 20). Invasion of the WT strain by defectors has even been reported for Vibrio cholerae, as a strain with deletion of hapR, a homolog of V. harveyi luxR, was able to invade and to supplant the WT strain in an M9-casein environment (22). Whereas in these other systems the WT strains are readily invaded by the corresponding defectors and in some cases are swept to extinction, the V. harveyi WT strain is able to withstand analogous invasion (Fig. 2 and 3). The results of this study demonstrate that QS systems and environmental conditions exist in which QS bacteria can resist the invasion of QS defector mutants. Thus, QS-regulated cooperation is shown here to be a more efficient and robust alternative to unconditional cooperation, allowing populations to resist invasion by within-species defectors (35, 36).

An examination of growth dynamics revealed that a functional QS system allows the WT strain to conditionally switch between low levels of investment in QS-regulated goods when nutrients are readily available and allowing cooperators to invest in the production of PG (such as proteases) when it is prudent, due to larger numbers of cooperator cells. This has the effect of maximizing growth rates at low cell densities while also achieving maximum growth yields by producing protease PG at high cell densities, presumably when surrounded by a large number of kin. The relatively high fitness afforded in both states preserves cooperative behaviors associated with QS. Such phenotypic plasticity has been shown to provide similar fitness benefits for other cooperative goods (11). Although the ΔluxR defector is able to rapidly invade the UC strain in pairwise competitions, because it is able to achieve a higher growth rate, it is unable to maximize yield in monocultures like a cooperator and it also draws down yields in mixed populations with cooperators as its frequency increases (Fig. 4B; also see Fig. S3 in the supplemental material). In contrast, the ΔluxOU mutant maximizes growth yield at the expense of a lower overall growth rate, particularly at low cell densities, which allows it to perform well in monoculture but renders it susceptible to defector invasion. This contrast in growth strategies has relevance to previous investigations into the trade-offs between growth rate and growth yield, and the plasticity conferred by QS may allow cells to evade this trade-off by matching growth strategies appropriately to the prevailing environmental conditions (60–62).

The delay observed in activating cooperation by QS regulation in the WT strain is consistent with proposed forms of restraint such as metabolic prudence and generalized reciprocity, in which investment in the expression of cooperative traits is tuned to be more economical for the producing organism (18, 33, 63). In our case, the cost is lowered for the WT strain during early growth, when signal density is too low to induce QS. In this state, the production of PG is not efficient, as the concentration of protease is too low to yield a net benefit (35). This situation could naturally occur when a pure culture of V. harveyi is at low cell density or when V. harveyi is in a mixed community surrounded by nonkin or defectors, when investment in competitive strategies is more beneficial than high levels of cooperation. However, we also note that invasion of the ΔluxR strain into the UC ΔluxOU strain ceases at higher densities (Fig. 5B). Following the reasoning of metabolic prudence (18), it is possible that, under these density conditions, it is less costly to express and to produce proteases. It is also possible that protease enzymes persist in the environment and thus measurement of their levels reflects past as well as current production, causing overestimation of the real costs to cooperator strains at high cell densities.

In summary, we have experimentally shown in V. harveyi that QS limits the production of cooperative goods to conditions where the benefits of making such products outweigh the production costs, and it can allow cooperation to resist defector invasion in environments where initial densities are low, which is likely common in the natural aquatic habitat of V. harveyi (64). QS regulation may be particularly advantageous for goods that exhibit accelerating benefits and would best be expressed at higher cell densities (35, 65). Therefore, it is advantageous to delay cooperation until cell density reaches sufficiently high levels when it is known that production is more beneficial (63), particularly at the transition from low to high density, precisely where QS is predicted to play an important role (36). However, as functional quorum sensors appear to perform well against defectors across a wide range of density conditions (see Fig. S4 in the supplemental material), QS may provide the environmental sensitivity to allow cells to act as density generalists, responding particularly to kin numbers in a given environment. The shifts in response to density that we observed resemble shifts between individual and group fitness maxima, with the expectation that organisms will behave to maximize their inclusive fitness (13, 66). QS could play an important role in effectively maximizing inclusive fitness. Together, these results suggest that cells with a fully functional QS system possess a significant advantage over systems that do not appropriately regulate cooperative behaviors and they are more resistant to cheating through optimization of growth strategies in a density-dependent manner, which provides support for the idea that communication can act as a form of cheater control (3, 15).