Introduction

Cooperation can be defined as a behavior that benefits another individual and that is maintained because of its beneficial effects on the recipient (West et al., 2006, 2007). In microbes, cooperation generally involves the production of public goods in the form of extracellular products (such as toxins, proteases, surfactants and siderophores), which are expensive to produce for the individual but benefit all members of the local group (Keller and Surette, 2006; West et al., 2006). Behaviors that involve such public goods include foraging, virulence and biofilm formation. In bacteria, cooperative behaviors are often regulated by quorum sensing (QS), a cell density-dependent communication system mediated by small signaling molecules (Williams et al., 2007; Ng and Bassler, 2009).

Pseudomonas aeruginosa is a model organism used to study QS (Juhas et al., 2005; Schuster and Greenberg, 2006). It is known for its ability to cause disease in immunocompromised individuals, including those suffering from cystic fibrosis. P. aeruginosa uses QS to regulate the transcription of hundreds of genes, many of which encode extracellular virulence factors (Hentzer et al., 2003; Schuster et al., 2003; Wagner et al., 2003). There are three known QS systems in P. aeruginosa, namely las, rhl and pqs. These systems are arranged hierarchically with the las system positively regulating both the rhl (Latifi et al., 1996; Pesci et al., 1997) and pqs (Wade et al., 2005) systems. However, under certain conditions, the rhl and pqs systems can also be activated in the absence of las QS (Diggle et al., 2003; Medina et al., 2003; Dekimpe and Deziel, 2009). In addition, the rhl system negatively regulates the pqs system (McGrath et al., 2004; Wade et al., 2005). These three QS systems each comprise a diffusible autoinducer signaling molecule and a regulator protein. When population density is high, the autoinducer binds to its cognate receptor to form a complex that regulates gene transcription. In the las and rhl systems, autoinducer synthases LasI and RhlI produce the acyl-homoserine lactone (acyl-HSL) signals N-(3-oxododecanoyl)-HSL and N-butyryl-HSL (C4-HSL), respectively. These signals interact with their cognate receptors LasR and RhlR, respectively (Juhas et al., 2005; Schuster and Greenberg, 2006). Transcriptome analysis revealed that these two systems together regulate 315 genes, with the rhl system regulating a subset of ∼112 genes (⩾2-fold activation by both signals versus N-(3-oxododecanoyl)-HSL alone) (Schuster et al., 2003).

In the pqs system, the products of pqsABCD and pqsH are involved in the synthesis of the autoinducer 2-heptyl-3-hydroxy-4-quinolone (PQS) and other classes of alkyl quinolones (AQs) (Deziel et al., 2005); PQS binds to its cognate LysR-type receptor PqsR with high affinity (Wade et al., 2005; Xiao et al., 2006). The pqs system controls 141 genes, most of which are co-regulated by acyl-HSL QS, primarily the rhl system (Deziel et al., 2005).

Evolutionary theory predicts that the cost of performing a cooperative behavior leaves a population vulnerable to social cheating (Lehmann and Keller, 2006; West et al., 2006). Cheaters are individuals that cease (or reduce) the production of public goods and benefit from the cooperative actions of others. Therefore, QS populations are at risk of invasion by either signal-negative cheaters that do not produce signals or by signal-blind cheaters that avoid production of QS-controlled extracellular factors. In P. aeruginosa, the latter is predicted to be more favorable (Sandoz et al., 2007; Diggle et al., 2007b) because QS controls the expression of hundreds of genes. Signal-negative strains would not avoid expression of QS target genes as they would be triggered by autoinducers produced by surrounding wild-type cells. Consistent with this prediction, P. aeruginosa strains with mutations in lasR (signal-blind) have been predominately isolated from infections (Heurlier et al., 2006; Smith et al., 2006; D’Argenio et al., 2007; Fothergill et al., 2007; Tingpej et al., 2007; Hoffman et al., 2009; Kohler et al., 2009), and emerge during in vitro evolution (Sandoz et al., 2007). Other experimental studies have shown that both lasR and lasI (signal-negative) mutants enrich in vitro (Diggle et al., 2007b) and in a mouse infection model (Rumbaugh et al., 2009). These mutants have a growth advantage over wild-type cells; however, they exhibit negative frequency-dependent relative fitness because as their proportion increases in a population, there are less wild-type cells to exploit. Taken together, these results are indicative of las mutants arising as social cheaters.

Initial social cheating studies focused on the las system because of its dominant position within the QS circuitry. In this study, we investigated the social behavior of signal-blind (rhlR, pqsR) and signal-negative (rhlI, pqsA) mutants from the rhl and pqs systems and compared them with lasR and lasI mutant strains. These experiments were performed under defined conditions, in a growth medium requiring and not requiring QS-regulated extracellular proteases, resembling ‘social’ and ‘non-social’ conditions, respectively. Although the common goods relevant to cheating are exoproteases supplied by the wild-type cooperator (Diggle et al., 2007a), variation in the relative fitness among mutant strains is expected to result from differences in the number and expression levels of all genes controlled by each system. We tested the hypothesis that both the pqsR and rhlR signal-blind mutants, in addition to the lasR mutant, would behave as cheaters under social conditions. In the simplest case, one might expect that the rhlR and pqsR mutants invade wild-type populations less than does the lasR mutant because rhlR and pqsR only control subsets of the QS regulon, although other outcomes are conceivable given the complexity of the QS network, including the conditionality of the regulatory hierarchy. Finally, we tested the hypothesis that the pqsA and rhlI mutants would cheat less or not at all compared with the respective pqsR and rhlR mutants.

Materials and methods

Bacterial strains, plasmids and culturing conditions

Bacterial strains and plasmids are shown in Table 1 . For general liquid culture, we used Lennox LB broth buffered with 50 mM MOPS (3-(N-morpholino)-propanesulfonic acid, pH=7.0). For QS assays, we used M9 minimal medium containing either 1% caseinate or 0.5% casamino acids (CAA) as the sole carbon source (Sandoz et al., 2007), designated M9-caseinate (‘QS medium’) or M9-CAA, respectively. All experiments were performed using true biological replicates on different days with different inocula. Further details are described in Supplementary Materials and Methods and in Supplementary Table S1.

Table 1 Bacterial strains and plasmids
Full size table

In vitro evolution and analysis of QS mutant phenotypes

In vitro evolution of the P. aeruginosa PAO1 wild type in liquid batch cultures containing M9-caseinate medium was performed as described previously (Sandoz et al., 2007). Culture aliquots were removed after days 4, 8, 12, 16 and 20 of culturing. To determine the colony-forming units (CFU) per milliliter, as well as the types of mutants arising, aliquots were appropriately diluted in 1 × M9 salts and subsequently plated onto LB agar plates. For 92 colonies from this plating, along with a PAO1-positive control and appropriate isogenic negative controls (Table 1), we determined growth on adenosine as the sole C source (Heurlier et al., 2005; Sandoz et al., 2007), proteolysis on skim-milk plates (Sandoz et al., 2007), rhamnolipid production on methylene blue plates (Kohler et al., 2000) and AQ production by bioassay (Fletcher et al., 2007), using a 96-well microplate format. The AQ bioassay detects both PQS and its precursor 2-heptyl-4-quinolone. AQ levels were normalized to the optical density at 600 nm (OD600) of the respective cultures at the time of harvest.

Results were scored as follows: For AQ production, <20% of the wild type was considered negative. For adenosine plates, growth was scored as positive, whereas the absence of growth was scored as negative. For skim-milk proteolysis and rhamnolipid production, the formation of a halo similar to the PAO1 control was scored as positive and the absence of a halo was scored as negative. These assays were performed in the same manner with all control strains shown in Table 2 .

Table 2 Analysis of P. aeruginosa QS mutant phenotypes
Full size table

Single-culture and co-culture growth assays

The PAO1 wild-type and the respective antibiotic-tagged mutant strains were grown in M9-CAA or M9-caseinate media, in single or co-culture. Both media were inoculated at a starting OD600 of 0.015 with a single strain or appropriate mixtures from individual 18-h LB-MOPS overnight cultures. For single-culture growth assays, the CFU ml−1 of each culture was determined at 0, 2, 5, 8, 12, 18, 24 and 30 h after inoculation. For co-culturing experiments, M9-CAA and M9-caseinate cultures were harvested after 12 and 24 h, respectively. The number of mutant cells tagged with antibiotic resistance gene cassettes and the total number of cells in co-culture were distinguished by plating on media with and without the corresponding antibiotic, respectively. Further details are described in Supplementary Materials and Methods.

The relative fitness (v) of each mutant in co-culture was first calculated by comparing its initial and final frequencies over the duration of the experiment (Ross-Gillespie et al., 2007a; Diggle et al., 2007b). In particular, v=(x1(1−x0))/(x0(1−x1)), where x0 and x1 are the initial and final mutant frequencies, respectively. Therefore, values of v signify whether mutants increase in frequency (v>1), decrease in frequency (v<1) or remain at the same frequency (v=1) over the duration of the experiment. We also calculated an alternative measure of relative fitness (w), namely the ratio of Malthusian growth parameters, which is essentially the ratio of the number of doublings by mutant and wild-type populations (Lenski, 1991). Here, w=ln(X1/X0)/ln(Y1/Y0), where X0 and X1 are the initial and final mutant CFU ml−1, and Y0 and Y1 the initial and final wild-type CFU ml−1, respectively.

Complementation analysis, DNA sequencing and fluorescein isothiocyanate (FITC) casein assay

These procedures are described in Supplementary Materials and Methods.

Results

Both lasR and pqsR mutants arise during in vitro evolution, but rhlR mutants do not

Current evidence suggests that lasR mutants are social cheaters that cease production of QS factors and take advantage of their production by the surrounding cooperative wild-type population (Sandoz et al., 2007; Diggle et al., 2007b; Rumbaugh et al., 2009). The main objective of this study was to further analyze the social behavior of signal-blind and signal-negative mutants from the other two major QS systems in P. aeruginosa, namely rhl and pqs. We previously showed that lasR mutants emerge during in vitro evolution in a defined medium that favors QS (Sandoz et al., 2007), but we did not investigate the emergence of other QS mutants. For a more comprehensive assessment, we repeated this long-term growth experiment and screened for mutants in all three QS systems. We grew the PAO1 wild type in M9-caseinate medium for 20 24-h cycles, subculturing after each cycle into fresh medium. Growth in this medium requires the production of QS-dependent proteases such as LasB elastase, and thus cooperative behavior (Van Delden et al., 1998; Sandoz et al., 2007). Under standard growth conditions (LB liquid culture), las, rhl and pqs mutants all show reduced expression of LasB elastase (Brint and Ohman, 1995; Pearson et al., 1997; Diggle et al., 2003; Schuster et al., 2003; Deziel et al., 2005).

We used a screening scheme for distinct QS phenotypes that allowed us to distinguish between the different las-, rhl- and pqs-deficient variants (Table 2). After 8 days of in vitro evolution, we observed the emergence of AQ-deficient isolates. By day 12, isolates with other QS deficiencies arose, some with deficiencies for more than one phenotype (Figure 1, Supplementary Table S2). Our previous work had shown that all isolates that did not grow on adenosine were lasR mutants (Sandoz et al., 2007). Therefore, we did not further analyze candidate lasR mutant isolates from this in vitro evolution study.

Figure 1
figure 1

Emergence of QS-deficient isolates during in vitro evolution of the P. aeruginosa wild type. Deficiencies in skim-milk proteolysis (solid squares), growth on adenosine (solid triangles), rhamnolipid production (solid circles) and AQ production (solid diamonds) are shown. CFU ml−1 (open squares) are indicated on the right y axis. Skim-milk proteolysis and growth on adenosine are las dependent, rhamnolipid production is rhl dependent and AQ production is pqs dependent. The discrepancy in the number of adenosine and protease-negative isolates is due to the fact that some lasR mutants regain the ability to degrade skim milk (Sandoz et al., 2007). Data are averages of two independent in vitro evolution experiments.

Full size image

Of the AQ-deficient isolates that emerged, nine were selected for complementation and designated pqs1 to pqs9. We transformed each isolate with plasmids containing pqsR expressed either from its native promoter (pJN105.pqsR-N) or from the heterologous pBAD-araC promoter (pJN105.pqsR-H). Each isolate showed complementation with both the heterologous and the native promoter, indicating that the mutation resides in the pqsR coding or upstream regulatory region (Figure 2). Indeed, sequencing analysis revealed that seven isolates harbor an insertion, deletion or single-nucleotide change within the pqsR gene (Table 3 ).

Figure 2
figure 2

Complementation analysis of AQ-deficient in vitro evolution isolates. AQ production of P. aeruginosa AQ-deficient variants containing pJN105 (black), pJN105.pqsR-H controlling pqsR from a heterologous promoter (dark gray) or pJN105.pqsR-N controlling pqsR from the native P. aeruginosa promoter (light gray). ‘PQS’ and ‘HHQ’ are synthetic signal controls. Data are given as percentage of PAO1/pJN105. Error bars indicate s.d. of the mean of three replicates.

Full size image
Table 3 pqsR sequence analysis of selected AQ-deficient in vitro evolution isolates of P. aeruginosa
Full size table

The in vitro evolution experiment revealed only two rhamnolipid-deficient isolates (Figure 1), designated rhl1 and rhl2. Both isolates were transformed with plasmids containing rhlR expressed from the pBAD-araC promoter (pJN105.rhlR-H) or the native promoter (pJN105.rhlR-N) and were tested for complementation on rhamnolipid detection plates. Both isolates could not be complemented, nor were pyocyanin levels in rhl2 similar to a defined rhlR mutant (Table 4 ). Sequencing analysis of the rhlR gene from both rhamnolipid-deficient isolates further confirmed that these isolates are not rhlR mutants (Table 4). We did not test these isolates for mutations in rhlI, because this type of mutation does not cause a deficiency in rhamnolipid production (Table 2).

Table 4 Analysis of rhamnolipid-deficient in vitro evolution isolates of P. aeruginosa
Full size table

The pqsR mutant, but not the lasR and rhlR mutants, grows well in QS medium because of protease production

The emergence of lasR and pqsR mutants during in vitro evolution indicates that these mutant strains have a growth advantage compared with the PAO1 wild type. The lack of rhlR mutants, on the other hand, suggests that they do not have a growth advantage. To begin to test these predictions, we first grew defined signal-blind deletion mutants individually in growth media requiring (M9-caseinate) and not requiring QS (M9-CAA). In M9-caseinate, both the lasR and rhlR mutants grew to a maximum density of approximately 30–40-fold less than that of the PAO1 wild type after 24 h, indicating a severe growth deficiency (Figure 3). This is consistent with the fact that both lasR and rhlR mutants are impaired in the production of proteases (Brint and Ohman, 1995; Pearson et al., 1997; Schuster et al., 2003). Both mutants did exhibit some growth during the first 5 h of incubation, which might be attributed to the presence of impurities (amino acids, oligopeptides) in the caseinate stock used to prepare the M9 medium. Interestingly, the pqsR mutant strain grew at a similar rate as did the wild-type strain over the course of 24 h, indicating that the pqsR mutant has high intrinsic fitness in QS medium (Figure 3). All strains grew similarly in M9-CAA, reaching maximum density after 12 h.

Figure 3
figure 3

Single-culture growth of signal-blind P. aeruginosa strains. Growth of the wild-type (black), lasR (dark gray), rhlR (light gray) and pqsR (open) strains in M9-CAA and in M9-caseinate media (circle and square symbols, respectively). Error bars indicate s.d. of the mean of three replicates, and are too small to be visible in some cases.

Full size image

To further examine the unexpected growth of the pqsR mutant, we used a FITC-casein assay to quantify protease production of each signal-blind strain and the PAO1 wild type (Figure 4). This assay contains the same protein source as our growth medium. After 12 h of growth in M9-CAA, all signal-blind mutants produced significantly less proteases than did the wild type, consistent with previous findings (Brint and Ohman, 1995; Pearson et al., 1997; Diggle et al., 2003; Schuster et al., 2003; Deziel et al., 2005). After 24 h of growth in M9-caseinate, the pqsR mutant showed caseinolytic activity that was indistinguishable from the wild type, whereas the lasR and rhlR mutants showed a much lower activity. Thus, only the las and rhl systems but not the pqs system is required for the production of proteases to digest caseinate in QS medium.

Figure 4
figure 4

Caseinolytic activity of signal-blind P. aeruginosa strains. Degradation of FITC-casein by the indicated strains grown in either M9-CAA (light gray bars) or M9-caseinate (dark gray bars). Data are given as the percentage of the wild type grown in M9-caseinate. Error bars indicate s.d. of the mean of three replicates. Statistical significance of the data was determined using a two-tailed unpaired t-test with ‘*’ indicating P-values <0.05. Protease production of each mutant was compared with that of the wild type in the respective media.

Full size image

In addition, it is worth noting differences in protease production when bacteria are grown in solid versus liquid media. This is most apparent with the rhlR mutant, which showed high caseinolytic activity on plates (Table 2) but low activity in liquid culture (Figure 4). This difference may be due to upregulation of lasR or lasI on solid media, permitting enhanced exoprotease expression that would otherwise require both las and rhl QS.

In QS co-culture, both lasR and pqsR mutants have a growth advantage over the wild-type cooperator, but rhlR mutants do not

To determine each mutant's relative fitness when rare in a population, we co-cultured each defined mutant with the wild-type parent in M9-CAA and M9-caseinate at an initial mutant frequency of ∼1%. Each mutant contained an antibiotic resistance cassette at a neutral chromosomal site to distinguish it from the wild type by plating on selective media. The growth rates of these tagged strains were indistinguishable from those of the unmarked parent strains (data not shown). After 24 h of growth in M9-caseinate, the lasR and pqsR mutants enriched (Figures 5a and b), but the rhlR mutant did not (Figure 5c). To confirm that the rhlR mutant does not exhibit a subtle growth advantage that cannot be captured after a single 24-h incubation period, we co-cultured an unmarked rhlR mutant with the wild type for 3 consecutive 24 h cycles. This rhlR mutant also did not enrich (data not shown). The three signal-negative mutants, namely lasI, pqsA and rhlI, showed little to no enrichment in M9-caseinate (Figures 5a–c). The same held true for all signal-negative and signal-blind mutants in M9-CAA. In addition, we analyzed the enrichment of all signal-blind strains together with the PAO1 parent in one single M9-caseinate co-culture, with initial mutant frequencies of 1% (Figure 5d). Interestingly, both the lasR and pqsR mutants enriched, whereas the rhlR mutant did not. Overall, these results show that lasR and pqsR mutants can invade PAO1 wild type populations under conditions that require QS. This fitness advantage is due to loss of signal reception, not signal production. In the following, we therefore focus on investigating the signal-blind variants in more detail.

Figure 5
figure 5

Enrichment of defined QS mutant strains in co-culture. Cultures were grown in M9-CAA (light gray bars) and M9-caseinate (dark gray bars) media with initial mutant frequencies of ∼1%. Values above each bar indicate fold enrichment (the ratio of final frequency versus initial frequency). (a) The lasR or lasI mutants in wild-type co-culture. (b) The pqsR or pqsA mutants in wild-type co-culture. (c) The rhlR or rhlI mutants in wild-type co-culture. (d) The lasR, pqsR and rhlR mutants combined (each at an initial frequency of 1%) in wild-type co-culture. (e) The lasR or rhlR mutants (each at an initial frequency of 1%) in pqsR mutant co-culture. Error bars indicate s.d. of the mean of three replicates. Statistical significance of the data was determined using a two-tailed unpaired t-test, with ‘**’ indicating P-values <0.05 and ‘*’ indicating P-values <0.1. For each individual condition, the initial frequency was compared with the final frequency. For between-condition comparisons, fold change was compared. For panels a–c, brackets indicate P-values from comparisons between growth in M9-caseinate versus M9-CAA, as well as growth of the signal-blind versus signal-negative strains.

Full size image

In the analysis of our long-term evolution study, we not only observed a large increase in pqsR mutants on day 12 but also a large decrease by day 16, coinciding with an increase in lasR mutants. To assess whether this pattern could be due to exploitation of the pqsR mutant by the lasR mutant, we initiated a co-culture of both strains with a lasR mutant frequency of 1%. After 24 h of growth in M9-caseinate, the lasR mutant enriched almost twofold (Figure 5e). For comparison, the rhlR mutant showed no enrichment (Figure 5e). These results indicated that the lasR mutant has a modest growth advantage over the pqsR mutant and can invade both the pqsR mutant and wild-type populations.

The relative fitness of the signal-blind strains shows negative and positive frequency dependence

To determine the frequency-dependent relative fitness of each signal-blind mutant, we co-cultured each mutant and its PAO1 parent in M9-caseinate for 24 h at starting mutant frequencies of ∼10% and 50%, in addition to 1%. Social evolution theory predicts that the relative fitness of cheaters will decrease with increasing frequency in a population as there are less cooperators to exploit (MacLean and Gudelj, 2006; Ross-Gillespie et al., 2007b). As observed previously, the lasR mutant exhibited negative frequency dependence (Diggle et al., 2007b) (Figure 6a). Two different established fitness calculations based either on mutant frequency (Ross-Gillespie et al., 2007a) or relative growth rates (Lenski, 1991) elicited similar trends. The pqsR mutant showed negative frequency-dependent fitness that it is much more pronounced at higher than at lower mutant frequencies, which we substantiated by an additional fitness measurement at 90% mutant frequency (Figure 6b). This confirms that the relative dominance of the pqsR mutant is entirely dependent on social interactions with the wild type. In contrast, the rhlR mutant seemed to exhibit positive frequency dependence (Figure 6c), indicating that its relative fitness increases with increasing abundance in a population (Molofsky et al., 2001). This result was surprising in light of the observation that the rhlR mutant displays low intrinsic fitness similar to the lasR mutant (Figure 3). Therefore, we further analyzed the fitness of the rhlR mutant in wild-type co-culture at even higher mutant frequencies of 90% and 99% (Figure 6d) and observed a decline to levels similar to those at low frequencies. Taken together, the relative fitness of the rhlR mutant centered around 1 with modest enrichment only at intermediate frequency.

Figure 6
figure 6

Frequency-dependent relative fitness of signal-blind strains. Relative fitness of (a) the lasR mutant, (b) the pqsR mutant and (c and d) the rhlR mutant calculated as the comparison of initial and final mutant frequencies (v, left panels) or as the ratio of mutant and wild-type Malthusian growth parameters (w, right panels). Data in the left panels a–c are on a double logarithmic scale and are fitted with either a power regression line (panels a and c) or an exponential regression line (panel b). Data in the right panels a–c are on a semi-logarithmic scale and are fitted with either a logarithmic regression line (panels a and c) or an exponential regression line (panel b). Panel d includes initial rhlR mutant frequencies of 90 and 99% (in addition to 1, 10 and 50%) plotted on either a semi-logarithmic scale (left) or a linear scale (right), resulting in a unimodal regression. Goodness of fit is indicated by R2. Fitness trends were considered significant by one-way ANOVA (using log-transformed values for v); P-values are indicated. Assays were performed in quadruplicate.

Full size image

Discussion

To understand the forces that select for QS deficiency, we chose a reductionist approach that compares the relative fitness of individual QS mutants in vitro under low viscosity growth conditions that do and do not require QS. In our previous long-term evolution study, we focused on the identification and characterization of the lasR mutants (Sandoz et al., 2007). In this study, we extended our analysis to include the rhl and pqs systems. A screening scheme for distinct QS phenotypes (Table 2) allowed us to detect variants with mutations in all three QS systems. Interestingly, we identified lasR and pqsR but not rhlR mutants during in vitro evolution of P. aeruginosa under growth conditions that require QS (Figures 1 and 2, Tables 3, 4). Of the nine complemented AQ-deficient variants, seven showed a mutation in pqsR (Table 3). Curiously, isolates pqs4 and pqs8 were not pqsR mutants, even though complementation analysis indicated a mutation within the upstream regulatory region or coding region of pqsR (Figure 2). Only pqs4 is predicted to harbor an additional mutation in the lasR gene (data not shown); however, this mutation would not fully diminish AQ production (Table 2) and thus does not explain the AQ-deficient phenotype. The reason for this phenotype is not ultimately clear. It is possible that these isolates are unable to produce AQ because of a constitutively active rhl system, which would repress pqsR (Wade et al., 2005). Consequently, the introduction of multiple plasmid-borne copies of pqsR might be sufficient to overcome the repressive effect by rhlR.

We found a good correlation between our long-term evolution study and co-culturing results in QS medium. In both experimental approaches, the lasR and pqsR but not the rhlR mutant had a growth advantage over the wild type. This was particularly apparent when we co-cultured the wild-type strain with all signal-blind isolates at once (Figure 5d). Indicative of social cheating and consistent with previous studies (Sandoz et al., 2007; Diggle et al., 2007b; Rumbaugh et al., 2009), our defined lasR mutant demonstrated substantial enrichment in wild-type co-culture (Figure 5a), had negative frequency dependence (Figure 6a) and displayed a growth deficiency when grown individually (Figure 3). The substantial fitness advantage of the lasR mutant in co-culture is probably a consequence of the energy saved from not expressing extracellular proteases resulting in cheating and from not expressing other dispensable lasR-controlled factors; if there is also a cost (for example, a metabolic imbalance) associated with the lack of certain, probably intracellular, lasR-controlled factors, it is relatively minor. In contrast to growth in M9-caseinate, the lasR mutant showed no enrichment in M9-CAA because in this medium QS-controlled genes are only induced at low-levels upon cessation of growth (Sandoz et al., 2007), and therefore do not impose a high metabolic burden on the wild type.

On the other hand, we found that our defined lasI mutant and other signal-negative strains showed virtually no enrichment in co-culture (Figures 5a–c). We had predicted that these signal-negative strains, in contrast to signal-blind strains, would not behave as social cheaters because the mutant phenotype can be compensated by surrounding autoinducer-producing cells, and because the cost of producing acyl-HSL or AQ is negligible compared with the cost of synthesizing hundreds of proteins. However, in previous studies by Diggle et al. (2007b) and Rumbaugh et al. (2009), lasI mutants displayed significantly higher relative fitness than did the wild type, similar to that observed for lasR mutants. In vivo competition experiments by Rumbaugh et al. (2009) were performed using an animal model mimicking either a chronic or an acute burn wound. Diggle et al. (2007b) used minimal medium supplemented with 1% bovine serum albumin as the sole carbon source and a starting inoculum that was ∼20-fold higher than ours. It is plausible that in both studies, lasI mutant enrichment resulted from a combination of social cheating and other mechanisms. The latter may include an intrinsic growth advantage on some carbon and nitrogen sources (D’Argenio et al., 2007), for example, increased levels of free amino acids associated with severe thermal injury (Fong et al., 1991), as well as enhanced survival under alkaline conditions during prolonged stationary phase (Heurlier et al., 2005). However, these benefits would only manifest in the absence of complementing acyl-HSL produced by wild-type cells. It is conceivable that reduced mixing during infection limits interactions between acyl-HSL-deficient and proficient cells, and that acyl-HSL levels are low in stationary phase cultures due to lactonolysis (Yates et al., 2002).

We initially predicted that all signal-blind isolates would be social cheaters that invade a wild-type population to varying degrees. Although our work with the lasR mutant was consistent with previous studies and our hypothesis (Sandoz et al., 2007; Diggle et al., 2007b; Rumbaugh et al., 2009), the social behaviors of the pqsR and rhlR mutants were more complex. The pqsR mutant enriched substantially in co-culture (Figure 5b) and showed overall negative frequency dependence (Figure 6b), a pattern characteristic of social cheating. Individually, however, the pqsR mutant grew as well as did the wild-type parent (Figure 3). The production of wild-type levels of protease accounts for its high intrinsic fitness (Figure 4), but does not explain its frequency-dependent dominance over the wild type. We further determined that the enrichment of the pqsR mutant is due to the lack of AQ reception rather than the lack of AQ production (Figure 5b). Therefore, we propose that the fitness pattern of the pqsR mutant is a combination of the following: this strain saves energy as it does not express pqsR-dependent factors, including those that are under control of pqsE (Deziel et al., 2005; Rampioni et al., 2010). Most of these factors would be dispensable for growth. On the other hand, the loss of PQS production itself incurs a fitness cost through the loss of the involvement in the oxidative stress response (Bredenbruch et al., 2006; Diggle et al., 2007c; Haussler and Becker, 2008). As an anti-oxidant, PQS decreases the oxidative stress induced by oxidation reactions through activity comparable with that of ascorbate (Haussler and Becker, 2008). Therefore, when the pqsR mutant exceeds a certain frequency in co-culture, the quantity of PQS produced by the wild type would be insufficient to provide adequate oxidative stress protection, resulting in decreased fitness. If the selective forces are similar in vivo, then pqsR mutants might abound in infections and other natural environments, although they would eventually be superseded by lasR mutants (Figures 1 and 5e). To our knowledge, a systematic analysis on the prevalence of natural pqs mutants has not been published.

Similar to the pqsR mutant, the behavior of the rhlR mutant did not conform to our prediction of social cheating. Although it displayed a similar individual growth pattern as did the lasR mutant in QS medium, the rhlR mutant was unable to invade a wild-type population (Figures 3 and 5c). The non-invasive nature of the rhlR mutant could be due to a large cost associated with the upregulation of pqs-dependent factors (those not also co-regulated by the rhl system) that outweighs the benefits from not producing rhl-dependent factors (Table 2, Figure 4) (McGrath et al., 2004). In particular, synthesis of PQS itself and transcription of pqsA have been shown to be upregulated sixfold and sevenfold, respectively, in rhl mutant strains (McGrath et al., 2004). In addition, certain rhl-controlled factors might also be beneficial for the individual, and their lack might be contributing to an overall decrease in fitness.

Furthermore, the rhlR mutant displayed both positive and negative frequency dependence with maximal fitness at intermediate frequency (Figure 6d). Such a pattern could result from a frequency-dependent effect of the wild type on the upregulation of costly pqs-dependent products in the rhlR mutant, or, perhaps more intriguingly, from spiteful behavior of the rhlR mutant towards the wild type. Spiteful behaviors, such as the production of bacteriocins, are costly to both the producer and the susceptible recipient. Consequently, they are predicted to evolve to maximal levels when the spiteful phenotype is at an intermediate frequency in the population (West and Buckling, 2003; Inglis et al., 2009). If the behavior of the rhlR mutant was indeed spiteful, it could be in relation to the upregulation of PQS (McGrath et al., 2004). In addition to its beneficial effects at low concentration, PQS can be deleterious and can promote autolysis when overproduced (D’Argenio et al., 2002; Haussler and Becker, 2008). If the rhlR mutant was to gain an advantage through this process, the wild type would have to be more sensitive to PQS-mediated killing than the rhlR mutant. However, regardless of the mechanism, the positive frequency dependence of the rhlR mutant when rare, together with its low relative fitness overall, may explain why it is not often found in nature.

Taken together, our data represent a comprehensive analysis of QS from an evolutionary point of view. They show that the inactivation of global regulators, as integral parts of an intricate regulatory network, can lead to complex social phenotypes. Pleiotropic effects likely have a role in modulating the behaviors of the pqsR and rhlR mutant strains.