Non-social adaptation defers a tragedy of the commons in Pseudomonas aeruginosa quorum sensing

Abstract

In a process termed quorum sensing (QS), the opportunistic bacterial pathogen Pseudomonas aeruginosa uses diffusible signaling molecules to regulate the expression of numerous secreted factors or public goods that are shared within the population. But not all cells respond to QS signals. These social cheaters typically harbor a mutation in the QS receptor gene lasR and exploit the public goods produced by cooperators. Here we show that non-social adaptation under growth conditions that require QS-dependent public goods increases tolerance to cheating and defers a tragedy of the commons. The underlying mutation is in the transcriptional repressor gene psdR. This mutation has no effect on public goods expression but instead increases individual fitness by derepressing growth-limiting intracellular metabolism. Even though psdR mutant populations remain susceptible to invasion by isogenic psdR lasR cheaters, they bear a lower cheater load than do wild-type populations, and they are completely resistant to invasion by lasR cheaters with functional psdR. Mutations in psdR also sustain growth near wild-type levels when paired with certain partial loss-of-function lasR mutations. Targeted sequencing of multiple evolved isolates revealed that mutations in psdR arise before mutations in lasR, and rapidly sweep through the population. Our results indicate that a QS-favoring environment can lead to adaptations in non-social, intracellular traits that increase the fitness of cooperating individuals and thereby contribute to population-wide maintenance of QS and associated cooperative behaviors.

Introduction

Bacterial cell–cell signaling, termed quorum sensing (QS), often coordinates other cooperative behaviors such as nutrient acquisition, biofilm formation or virulence in a cell density-dependent manner (Waters and Bassler, 2005; Williams et al., 2007). In Gram-negative proteobacteria, QS generally comprises a LuxI-type signal synthase that produces a diffusible acyl-homoserine lactone (acyl-HSL) signal, and a cognate LuxR-type receptor that binds the signal and regulates transcription of target genes (Waters and Bassler, 2005; Williams et al., 2007; Schuster et al., 2013). The opportunistic pathogen Pseudomonas aeruginosa, a particularly well-understood example, employs two acyl-HSL signaling systems, LasI/R and RhlI/R, arranged in a hierarchical manner with LasR sitting atop the hierarchy (Schuster and Greenberg, 2006; Williams and Camara, 2009; Jimenez et al., 2012). Together, both systems regulate over 300 genes, many of which encode secreted public goods such as extracellular enzymes or secondary metabolites that have a role in virulence (Hentzer et al., 2003; Schuster et al., 2003; Wagner et al., 2003).

How social behaviors such as QS evolve and are maintained is of intense research and debate, as exploitation of common resources by selfish individuals should be favored and lead to a so-called ‘tragedy of the commons' (Keller and Surette, 2006; West et al., 2006). A tragedy of the commons results when the magnitude of selfish exploitation by cheaters exceeds the capacity of a cooperative system, resulting in the collapse of the entire population. Indeed, several studies have demonstrated the emergence of QS cheaters that reap the benefits of cooperative secretions without metabolic investment both in vitro (Diggle et al., 2007; Sandoz et al., 2007; Wilder et al., 2011; Dandekar et al., 2012) and in vivo (Kohler et al., 2009; Rumbaugh et al., 2009). These QS cheaters are defined by a loss-of-function mutation in the gene coding for the primary QS receptor LasR. We previously showed that P. aeruginosa lasR mutant cheaters consistently evolve in a minimal growth medium with casein as the sole carbon source that requires QS-dependent extracellular proteolysis (Sandoz et al., 2007). Using defined wild-type and lasR mutant co-cultures, we further showed that these cheaters do better when they are rare (that is, display negative frequency-dependent fitness), and that they impose a burden on population growth (Sandoz et al., 2007; Wilder et al., 2011). Intriguingly, however, this negative effect on group fitness was generally not observed during in vitro evolution experiments initiated solely with the wild-type strain, suggesting evolution of a mechanism that stabilized QS (Sandoz et al., 2007; Wilder et al., 2011; Dandekar et al., 2012).

To identify and characterize the underlying mechanism, we used a combination of whole-genome sequencing, genetic analysis and growth experiments. We found a single mutation in a transcriptional repressor, PsdR, that rapidly dominates the population, enhances intracellular dipeptide metabolism, increases both individual and group fitness, provides immunity against cheaters that do not themselves carry a psdR mutation and lessens the detrimental effect of certain lasR mutations on group fitness. Our results show that QS-favoring conditions can select for non-social adaptations that improve group fitness and defer a tragedy of the commons.

Materials and methods

Strains and culture conditions

P. aeruginosa PAO1 was used as the wild-type isogenic parent at the start of all original in vitro evolution experiments (Sandoz et al., 2007; Wilder et al., 2011). All mutants were created via allelic exchange using a suicide vector containing either evolved alleles or in-frame deletions constructed by splicing-overlap-extension PCR (Horton, 1995; Hoang et al., 1998) (see Table 1 for a comprehensive list of strains). For routine culturing, we grew strains at 37 °C on Lennox lysogeny broth (LB) agar or with shaking in Lennox LB broth buffered with 50 mM 3-(N-morpholino)-propanesulfonic acid (MOPS), pH 7.0. Plates were supplemented with 100 μg ml⁻¹ tetracycline when necessary for the selection of marked strains. For fitness, competition, substrate utilization and expression assays, M9 minimal medium supplemented with either 1% caseinate, 0.5% casamino acids (CAA) or 10 mM GlyGlu dipeptide was used (Sandoz et al., 2007; Kiely et al., 2008). In the case of caseinate fitness experiments with supplemented exoprotease, porcine elastase (Sigma, St. Louis, MO, USA) was added at the beginning of growth, in principle as described previously (Diggle et al., 2007). As determined with a fluorescein isothiocyanate (FITC)-casein assay (see below), the caseinolytic activity of the elastase concentration used (0.21 U ml⁻¹) was 20% of that found in the supernatant of wild-type cultures grown in M9-caseinate medium for 24 h. All experiments were performed using a minimum of three biological replicates with independent inocula.

Table 1. Bacterial strains and plasmids.

Strain or plasmid	Relevant properties	Reference or origin
Pseudomonas aeruginosa
PAO1	Wild-type (obtained from M Vasil and U Ochsner)	Holloway (1955)
PAO-HC	PAO1 derivative; evolved hybrid cooperator containing lasR5, psdR1 and abcB1 mutations	Sandoz et al. (2007)
PAO1 lasR5	PAO1 derivative; lasR5, unmarked mutant in which wild-type lasR was replaced with lasR5	This study
PAO1 psdR1	PAO1 derivative; psdR1, unmarked mutant in which wild-type psdR was replaced with psdR1	This study
PAO1 ΔlasR	PAO1 derivative; ΔlasR, unmarked in-frame deletion from amino acids 102–216	Wilder et al. (2011)
PAO1 ΔpsdR	PAO1 derivative; ΔpsdR, unmarked in-frame deletion from amino acids 11–124	This study
PAO1 psdR1 lasR5	PAO1 psdR1 derivative; psdR1 lasR5, unmarked mutant in which wild-type psdR and lasR were replaced with psdR1 and lasR5, respectively	This study
PAO1 psdR1 ΔlasR	PAO1 ΔlasR derivative; psdR1 ΔlasR, unmarked mutant in which wild-type psdR and lasR were replaced with psdR1 and ΔlasR, respectively	This study
Escherichia coli
DH5α	F-Φ80lacZYA-argF U169 recA1 hsdR17 (rk−, mk+) phoA supE44 λ- thi-1 gyrA96 relA1	Invitrogen
SM10	thi thr leu tonA lacY supE recA::RP4-2-Tc::Mu Km λpir	Simon et al. (1983)
Plasmids
pEX18Gm	Conjugative suicide plasmid; GmR	Hoang et al. (1998)
pEX18Gm.psdR1	pEX18Gm with the evolved psdR1 allele	This study
pEX18Gm.ΔpsdR	pEX18Gm with ΔpsdR containing an in-frame deletion from amino acids 11–124	This study
pEX18Gm.lasR5	pEX18Gm with the evolved lasR5 allele	This study
pUC18R6KT-mini-Tn7T-Tet	Broad host range mini-Tn7 vector with Tc resistance gene cassette	Courtesy of Herbert P Schweizer

Whole genome sequencing and targeted DNA sequencing

For genome sequencing, we selected an evolved ‘hybrid cooperator' (HC) isolate from our previous long-term growth experiment (Sandoz et al., 2007) and its wild-type PAO1 parent strain. The evolved isolate was dubbed HC because of its partially positive QS phenotypes (see the Results section for details). Both strains were grown individually overnight (18 h) in MOPS-buffered LB medium as described above. Genomic DNA was isolated using the Qiagen Puregene Yeast/Bacteria Kit B (Qiagen Sciences, Germantown, MD, USA) and assessed for quality on a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). The 454 pyrosequencing was carried out using unpaired reads on a Genome Sequencer FLX instrument with GS FLX Titanium series reagents (454 Life Sciences, Branford, CT, USA) by the Dhingra Genomics Lab at Washington State University in Pullman, Washington, USA. Sequencing of the HC isolate produced 507 094 reads covering ∼187 Mb, whereas the ancestral PAO1 produced 501 270 reads covering ∼200 Mb. Raw 454 reads were assembled using the Roche 454 Newbler assembler (Roche Diagnostics, Basel, Switzerland) with the PAO1 genome as a reference (Stover et al., 2000; Margulies et al., 2005; Winsor et al., 2011). The HC assembly utilized an average map length of 370 bp and average sequence depth of 29.5, whereas the ancestral PAO1 assembly utilized an average map length of 401 bp and average sequence depth of 31.7. Differences between the HC and ancestral PAO1 assemblies were discovered using SNP/INDEL calling in SAMtools (Li et al., 2009). To confirm the identified mutations, and to sequence specific loci of interest, standard dideoxy sequencing of PCR-amplified and purified chromosomal DNA was employed at the Center for Genome Research and Biocomputing at Oregon State University in Corvallis, OR, USA. Primers are listed in Supplementary Table S1.

Fitness assays and cheater load

Wild-type, evolved HC and defined mutants (tagged with an antibiotic-resistance marker where applicable) were grown in caseinate minimal medium. Overnight cultures of individual strains in MOPS-buffered LB were used as inocula for experiments and diluted to starting OD₆₀₀ values of 0.02 (1 cm path length, ∼2 × 10⁷ colony-forming units (CFUs) per ml). In the case of co-culture experiments, the combined total starting OD₆₀₀ was 0.02. All fitness experiments were allowed to proceed for 24 h with shaking at 37 °C. For rich media (LB+MOPS, M9-CAA) and dipeptide media co-cultures, conditions were kept identical to caseinate experiments with the exception that co-cultures in dipeptide media were grown for 7 days to allow the cultures to reach saturation. CFUs per ml were determined by dilution plating at t=0 and 24 h, with an additional plating at t=12 h during absolute fitness experiments. For enumeration in co-culture experiments, differential plating on tetracycline-supplemented LB agar was used. Fitness was calculated according to the Malthusian growth model (Lenski et al., 1991; Wilder et al., 2011). Absolute fitness is expressed as the average rate of increase or Malthusian parameter (m), with m=ln(N₁/N₀)/t, where N₁ and N₀ are the final and initial strain densities, respectively, and t is the culturing time in days. Relative fitness is expressed as the ratio of the Malthusian parameters (w) of two competing strains. Cheater-load experiments were performed as previously described (Sandoz et al., 2007), with the exception that for this set of experiments total starting OD₆₀₀ values of 0.02 were identical for all treatments.

Extracellular proteolysis

Extracellular caseinolytic activity was determined using an established FITC-casein assay (Twining, 1984; Wilder et al., 2011). Briefly, starter cultures of each strain were grown overnight in MOPS-buffered LB at 37 °C and diluted to an OD₆₀₀ of 0.02 in fresh CAA medium. Supernatants were harvested after 12 h of growth, sterile filtered and incubated with the FITC-conjugated casein substrate (Sigma). Digestion was allowed to proceed for 3 h at 37 °C. Fluorescence was measured at λ_ex=490 nm and λ_em=525 nm in a 96-well format on a Tecan Infinite M200 plate reader (Tecan Group Ltd, Männedorf, Switzerland).

To predict the cleavage pattern of our caseinate substrate (a mixture of α-s1-casein, α-s2-casein, β-casein and κ-casein) by LasB elastase, we employed ExPASy PeptideCutter (http://web.expasy.org/peptide_cutter/) for in silico digestion with thermolysin, the closest LasB elastase family member available in the database (Gasteiger et al., 2005).

Expression analysis

Strains were initially grown overnight in MOPS-buffered LB liquid culture, and then diluted to an OD₆₀₀ of 0.02 in fresh CAA medium. Expression cultures were harvested at OD₆₀₀ values of 0.5 and 1.5, corresponding to exponential and early stationary phases in this medium, respectively. Total RNA was isolated and complementary DNA synthesized as previously described (Schuster and Greenberg, 2007; Schuster, 2011). Quantitative reverse-transcriptase PCR was carried out according to established protocols (Schuster and Greenberg, 2007; Schuster, 2011) using an Applied Biosystems 7300 Real Time PCR System (Applied Biosystems, Foster City, CA, USA). Identical amounts of complementary DNA were used as template. Transcript levels were quantified using the relative standard curve method.

Results

Genome sequencing of in vitro-evolved P. aeruginosa

In our previous in vitro evolution studies, we cultured the PAO1 wild-type strain in caseinate medium for 20 days, subculturing into fresh medium each day (Sandoz et al., 2007; Wilder et al., 2011). We used two phenotypic screens as a proxy for QS proficiency, namely (1) protease production on skim-milk agar plates and (2) growth on minimal agar plates with adenosine as the sole carbon source. Negative results in both phenotypic screens correspond to mutations in the gene coding for the primary QS regulator, lasR, thereby conferring a cheater phenotype.

To confirm that QS-controlled extracellular proteolysis is solely responsible for the growth deficiency of the pleiotropic lasR mutant in caseinate medium, we cultured the ΔlasR single mutant in the presence of purified elastase. Addition of elastase restored ΔlasR mutant fitness, expressed as Malthusian growth parameter, m (Lenski et al., 1991), to a level indistinguishable from wild-type and significantly above that of the ΔlasR mutant without elastase (Figure 1).

Figure 1.

Effect of elastase addition on the absolute fitness of a lasR mutant. Absolute fitness of the P. aeruginosa ΔlasR mutant and its wild-type parent was calculated as Malthusian growth parameter (m) after 24 h of growth in caseinate medium. In the case of ΔlasR+elastase, 0.21 U ml⁻¹ porcine elastase was added. Bars represent means (n=3). *Significant difference as determined by unpaired t-test (P=0.00010). NS, difference not significant (unpaired t-test, P=0.096).

Although cheaters can exploit the public goods produced by the cooperating population in a way that eventually leads to a population crash, we only observed this outcome in one of our five replicate evolution experiments (Figure 2a–f). Instead, we found that lasR cheater frequencies of 25% on average are tolerated and do not significantly affect the growth yield of the population (Figure 2a). This was surprising because our previous co-culture experiments with specific initial frequencies of defined lasR-mutant cheaters and wild-type cooperators showed that comparable levels of cheaters impose a substantial burden on the productivity of a population (Sandoz et al., 2007).

Figure 2.

In vitro evolution of P. aeruginosa populations under conditions that require QS. (a–f) Population growth yield and phenotypic frequencies. OD₆₀₀ values measured daily before subculture are plotted on the left vertical axis (blue line). Frequencies of cooperator (black triangles), cheater (red circles) and hybrid cooperator (green squares) phenotypes are plotted on the right vertical axis. (a) Means and s.e.m. of all replicate experiments (n=5). In some cases, error bars are too small to be seen. (b–f) Individual, independent biological replicates. Based on raw data from Wilder et al. (2011; b, c) and Sandoz et al. (2007; d–f). (g) Schematic of evolutionary trajectories of individual mutations.

We also observed a subpopulation of isolates deficient in growth on adenosine, but not in skim-milk proteolysis. Based on their phenotypes, we dubbed these isolates HCs. The HC subpopulation rose in frequency similar to the cheaters, representing up to 20% of the total population (Figure 2a), and also harbored mutations in lasR (Sandoz et al., 2007). We reasoned that this HC phenotype had an important role in the maintenance of cooperative population growth. We hypothesized that the HC phenotype was caused by an independent second-site mutation that occurred either before or after mutation of lasR, and that this mutation partially restored QS proficiency.

To address our hypothesis, we sequenced the genome of a representative HC isolate from day 12 of one replicate experiment. We also sequenced the genome of the ancestral PAO1 wild-type strain for comparison. Genomes were assembled using the published PAO1 genome sequence as a reference (Stover et al., 2000; Winsor et al., 2011). In all, our analysis showed the HC harbored only three mutations when compared with the wild-type ancestor, including lasR. The mutations were single-nucleotide polymorphisms in lasR (PA1430) and in PA2408, as well as an 18 base-pair truncation in psdR (PA4499). PA2408 encodes a probable adenosine triphosphate (ATP)-binding component of an ABC (ATP-binding cassette) transporter, and psdR encodes a transcriptional regulator (Kiely et al., 2008; Winsor et al., 2011) (Table 2). Targeted Sanger sequencing of all three loci in two additional HC and cheater isolates from day 12 of two independent in vitro evolution experiments revealed that mutations in lasR and psdR are ubiquitous in isolates displaying both phenotypes, but mutations in PA2408 are not (Supplementary Table S2). We therefore concluded that the PA2408 mutation is not likely to be relevant to the HC phenotype. Additional evidence for this conclusion is presented below.

Table 2. Mutations in a sequenced Pseudomonas aeruginosa hybrid cooperator (HC) genome.

Gene (name)a	Functiona	Mutationb	Allele ID
PA1430 (lasR)	LuxR-type transcriptional regulator	C→T (683)	lasR5
PA4499 (psdR)	Putative transcriptional regulator	Δ18 bp (514)	psdR1
PA2408	Probable ATP-binding component of ABC transporter	T→C (337)	abcB1

^aGene names and functions as annotated in the Pseudomonas Genome Database.

^bNumbers in parentheses indicate position or beginning of a given mutation relative to the translational start site.

The presence of psdR mutations in cheater isolates in addition to the HC isolates indicated that a mutation in psdR is not a distinguishing feature of the HC phenotype, and that mutation of psdR may have arisen before mutation of lasR. To elucidate the evolutionary trajectories of the psdR mutation, and to assess whether psdR mutations are also present in cooperator phenotypes, we sequenced the psdR locus of evolved isolates positive for both skim-milk proteolysis and adenosine utilization from the day 4, 8 and 12 archives of two replicates (5 isolates per day, 30 total). Surprisingly, we found 100% of sequenced isolates harbored a nonsynonymous mutation in psdR (Table 3). Even as early as the first phenotypic screen at day 4, the entire sampled population that originally appeared to be ‘wild-type' with respect to QS actually had acquired point mutations, insertions or deletions in psdR. Isolates from the same replicate culture often harbored different psdR mutations, and some psdR mutations in early cooperator isolates were identical to those in later HC and cheater isolates. These results therefore suggest that the HC phenotype is primarily defined by the nature of the lasR mutation itself. In general, our sequencing data indicate a strong selection against a functional psdR during cooperative growth in caseinate medium, and show that the evolutionary trajectories of cheater, HC and cooperator phenotypes all start with a mutation in psdR (Figure 2g). This result is consistent with the presumed function of PsdR. PsdR has been characterized as a transcriptional repressor of genes involved in the uptake and intracellular degradation of dipeptides in P. aeruginosa (Kiely et al., 2008). Thus, derepression of dipeptide metabolism through mutation and inactivation of PsdR could potentially increase the fitness of P. aeruginosa during proteolytic growth in caseinate medium. Such a mutant would take up and process the dipeptides generated by the cocktail of secreted proteases (including LasB elastase, alkaline protease and protease IV) more rapidly. Consistent with this idea, in silico digestion of bovine casein by thermolysin, a homolog of P. aeruginosa LasB elastase with similar cleavage properties (Morihara and Tsuzuki, 1971; Jiang and Bond, 1992), indeed produces up to 6 dipeptides per casein molecule.

Table 3. The psdR mutations in evolved Pseudomonas aeruginosa isolates.

Mutationsa	Number of mutations (replicate)b			Changec
	Day 4	Day 8	Day 12
Cheater (4 sequences total)
Δ505-end			2 (2)	Deletion
T166C			2 (1)	S56P
HC (4 sequences total)
Δ145–148			2 (2)	Deletion
Δ261–422			1 (1)	Deletion
Δ261-end			1 (1)	Deletion
Cooperator (30 sequences total)
Δ261–422	1 (1)	1 (2)		Deletion
Δ505-end		1 (2)		Deletion
Δ147–159		1 (1)	1 (1)	Deletion
C74T	1 (2)	1 (2)		A25V
T100C	1 (2)			F34L
C109A	1 (1)		1 (1)	Q37K
T166C	1 (1)			S56P
G397A	2 (1)	2 (1)		G133R
C411A			3 (1)	STOP at 137
A431C	1 (2)	1 (1)	2 (2)	Y144S
Insert A at 3		1 (1)		Frameshift
Insert A at 378	1 (2)	1 (2)	1 (2)	Frameshift
No amplicon	1 (2)	1 (2)	2 (2)	Unknown

^aMutations are sorted by cheater, hybrid cooperator (HC) and cooperator phenotypes. Numbers indicate nucleotide position or beginning of a given mutation relative to the translational start site.

^bThe individual replicate of the in vitro evolution experiment is indicated in parentheses. Same-day isolates with identical mutations were always from the same replicate. Replicates 1 and 2 correspond to Figures 2b and c, respectively.

^cNumbers indicate amino-acid position relative to the translational start site.

Absolute fitness of evolved and defined strains

Next, we investigated the fitness contributions of each mutant allele to the cooperator, cheater and HC phenotypes. We constructed defined single and double mutants by transferring the evolved lasR5 and psdR1 alleles into the parental PAO1 strain background, and compared their phenotypes with those of in-frame deletion mutants. This approach also allowed us to assess whether the nature of the mutation in lasR distinguishes a HC from a cheater. As characterized in our previous study, all evolved cheaters deficient in skim-milk proteolysis and adenosine utilization were also fully deficient in other QS-dependent phenotypes, identical to a lasR in-frame deletion mutant (Sandoz et al., 2007).

We first assessed growth of individual strains by measuring their population densities during clonal growth in caseinate medium that requires QS-dependent proteolysis (Figure 3a). The corresponding absolute fitness values and statistical data are shown in Supplementary Figure S1. The wild-type is capable of two logs of growth within 24 h, from ∼10⁷ to 10⁹ CFUs per ml, whereas a lasR deletion mutant shows little growth. Interestingly, the defined lasR5 mutant displayed an intermediate level of growth and fitness significantly above that of the ΔlasR mutant at 24 h, indicating that lasR5 retains partial function. The defined psdR1 mutant and the ΔpsdR mutant displayed similar growth and fitness levels significantly above that of the wild-type at 12 h, indicating that the psdR1 mutation completely inactivates gene function. The defined psdR1 lasR5 double mutant and the evolved HC showed identical growth characteristics, similar to that of the wild-type at 24 h, supporting our sequencing data suggesting that the PA2408 mutation does not play a part in the HC phenotype. In contrast, the psdR1 mutation, when paired with the ΔlasR mutation, did not support growth to levels beyond the ΔlasR single mutant, strengthening the role of the lasR5 allele in the HC phenotype. Taken together, these results show that inactivation of psdR increases absolute fitness, and that this effect can compensate for the reduced level of cooperation from partial loss of function in lasR5.

Figure 3.

Growth and proteolysis in pure culture. (a) Growth in caseinate medium measured at 12 and 24 h, expressed as CFUs per ml. Means and s.e.m. are shown (n=3) and in some cases error bars are too small to be seen. Starting CFUs per ml are statistically the same (one-way analysis of variance (ANOVA), Tukey's multiple comparisons test, α=0.05). (b) Caseinolytic activity of cultures grown in CAA medium for 12 h, as measured by FITC-casein assay. Caseinolytic activity is shown per OD₆₀₀ to correct for slight variations in the final culture densities in CAA medium. Means and s.e.m. are shown (n=3). *Significant differences as determined by one-way ANOVA, Tukey's multiple comparisons test, α=0.05. Results of similar magnitude are grouped for clarity.

Exoprotease activity

To correlate the absolute fitness of each strain with its exoprotease activity, we quantified caseinolysis of culture supernatants using a FITC-casein assay (Twining, 1984; Wilder et al., 2011). This method is more precise than the qualitative skim-milk plate assay used previously (Sandoz et al., 2007). In order to uncouple exoprotease activity from its effect on growth, we replaced caseinate in our growth medium with CAA, a C-source that does not require QS-dependent proteolysis. All strains harboring the lasR5 allele showed intermediate levels of extracellular caseinolysis at half the levels of the wild-type lasR alleles and roughly three times higher than the ΔlasR alleles (Figure 3b). Strains containing psdR mutations did not show elevated caseinolysis compared with the wild-type. These results confirm that lasR5 is a partial loss-of-function mutation, and further show that psdR has no effect on QS-dependent exoprotease production, consistent with its role in regulating intracellular dipeptide metabolism.

Transcriptional regulation of dipeptide transport and processing

We have provided evidence that psdR mutations are explicitly linked to significant increases in the fitness of P. aeruginosa in a cooperative environment. As indicated above, PsdR is a transcriptional repressor of several neighboring genes involved in the transport and processing of dipeptides in P. aeruginosa (Kiely et al., 2008). Specifically, PsdR represses transcription of mdpA, which codes for the cytoplasmic dipeptidase MdpA, as well as dppA3, the first gene in a dipeptide transport gene cluster annotated dpp for a homologous region in the Escherichia coli K12 genome (Kiely et al., 2008). Associated with this gene cluster is a gene coding for the porin OpdP that is implicated in the uptake of single amino acids as well as dipeptides in P. aeruginosa (Tamber and Hancock, 2006).

Interestingly, our previous transcriptome analysis of P. aeruginosa grown in rich medium indicated that mdpA expression was affected by rhl-QS. Addition of the LasI-generated signal to a signal synthesis mutant only induced expression by 1.6-fold, but addition of both acyl-HSL signals induced expression by ninefold (Schuster et al., 2003). Thus, it was conceivable that lasR affected dipeptide transport and processing mainly indirectly, through its effect on the rhl system, although this regulation is nutritionally conditional (Medina et al., 2003; Dekimpe and Deziel, 2009; Mellbye and Schuster, 2014). To investigate a possible link between QS and dipeptide metabolism in our experimental system, we quantified the transcript levels of dppA3 and mdpA during growth in CAA medium for our set of eight P. aeruginosa strains used in the previous sections. Using quantitative reverse-transcriptase PCR, we assessed transcription in exponential and early stationary phases, corresponding to OD₆₀₀ values of 0.5 and 1.5, respectively. We found that for either gene at any growth phase tested, relative expression could be sorted by psdR allele, with at least an order of magnitude separating functional psdR alleles from those harboring psdR1 or ΔpsdR (Figure 4). Importantly, none of the lasR alleles substantially influenced expression of dppA3 or mdpA in our experiments. This result indicated that, under the growth conditions employed here, the regulation of dipeptide transport and processing is dependent on psdR but independent of lasR. Hence, control of the relevant ‘private' goods—the cellular dipeptide uptake and processing machinery—occurs independently of QS-mediated ‘public' goods in our system.

Figure 4.

Expression of dppA3 and mdpA. Relative transcript levels of dppA3 (a) and mdpA (b), as determined by quantitative reverse-transcriptase PCR (qRT-PCR). Relative expression during exponential (OD₆₀₀=0.5, empty bars) and early stationary (OD₆₀₀=1.5, filled bars) growth phases in CAA medium are shown. Means and s.e.m. are shown (n=3). *Significant differences as determined by one-way analysis of variance (ANOVA), Tukey's multiple comparisons test, α=0.05. Results of similar magnitude are grouped for clarity.

Relative fitness of evolved and defined strains in co-culture

Next, we measured the relative fitness of our set of strains through pairwise comparisons in co-culture, again employing caseinate medium that requires QS-dependent cooperation. We introduced an antibiotic resistance marker into one of the two strains at a neutral chromosomal site to allow differentiation in co-culture. The marker itself has no effect on growth (Wilder et al., 2011). With the marked strain at an initial frequency of 0.01 (1%) in starting populations of ∼2 × 10⁷ CFUs per ml total, we allowed competitions to proceed for 24 h, and calculated the relative fitness (w) as the ratio of the average growth rates (Malthusian growth parameters) (Lenski et al., 1991; Wilder et al., 2011).

First, we sought to ensure that selection for psdR mutants in the original in vitro evolution experiments was not just a general feature of prolonged growth but was tied to the specific growth medium. We therefore initiated defined co-cultures of the ΔpsdR mutant and the wild-type in different growth media, at a mutant frequency of 0.01. We used a complex medium (MOPS-buffered LB) and M9 minimal medium with essentially fully hydrolyzed casein (CAA) as the sole C-source. The ΔpsdR mutant did not enrich in either LB+MOPS or CAA media, confirming that adaptive mutation of psdR is linked to the cooperative media we employed (Figure 5a). To confirm that the increased absolute fitness of a psdR mutant can be attributed in part to increased uptake and metabolism of dipeptides, we also employed M9-minimal medium with the dipeptide GlyGlu as the sole carbon source (Kiely et al., 2008). The ΔpsdR mutant exhibited a high degree of relative fitness in this medium very similar to the psdR1 mutant in caseinate medium (Figure 5a and first column of Figure 5b), indicating that dipeptide uptake and metabolism is indeed a target of selection in our cooperative growth environment.

Figure 5.

Relative fitness. (a) Relative fitness of a ΔpsdR mutant in co-culture with its wild-type parent in rich and defined media, initiated at a mutant frequency of 0.01. LB, LB+MOPS; CAA, M9-CAA; GlyGlu, M9-GlyGlu. (b) Relative fitness of defined mutants in caseinate co-culture. Pairs of the respective rare and abundant strain were initiated at a ratio of 1:99. Relative fitness values were calculated as the ratio of Malthusian growth parameters (w) after 24 h, with the exception that experiments in M9-GlyGlu were allowed to proceed for 7 days to allow the co-cultures to reach saturation. Values of w signify whether the rare strains grow faster (w>1) or grow slower (w<1) than the respective abundant strains. Bars represent means (n=3), and means are significantly different from w=1 (one sample t-test, P<0.05), unless designated by §. NS, difference between two conditions not significant (unpaired t-test, P>0.05). Difference in mean relative fitness of ΔpsdR in M9-GlyGlu co-culture with WT (a) and psdR1 in caseinate co-culture with WT (b) is not significant (unpaired t-test, P=0.23).

Second, we sought to compare the relative fitness of mutant alleles (initial frequency of 0.01) against the wild-type ancestor to better understand the population dynamics at the beginning of our previous in vitro evolution experiment. The lasR5 mutant modestly enriched in wild-type co-culture, as did the ΔlasR mutant in accordance with previous studies (Figure 5b) (Sandoz et al., 2007; Wilder et al., 2011). This result was expected as an individual that decreases investment in a secreted ‘public good' while still taking advantage of its production by cooperators should exhibit higher relative fitness (West et al., 2006). A psdR1 mutant had a tremendous relative fitness advantage with respect to the wild-type, consistent with its high absolute fitness, and mirroring the early dominance of psdR mutants during in vitro evolution (Figure 5b). When we combined either of the lasR mutant alleles with psdR1, the average relative fitness again was well above that of the lasR mutants alone, demonstrating the independence of psdR fitness from LasR regulation (Figure 5b).

We then aimed to understand the relative fitness dynamics after the emergence and dominance of psdR mutations in the evolved populations. This time we initiated competitions with the psdR1-defined mutant in majority (initial frequency of 0.99). Both psdR1 lasR5 and psdR1 ΔlasR double mutants displayed relative fitness of >1.0, as would be required for their enrichment in the original in vitro evolution experiments (Figure 5b). The difference in the relative fitness between the two strains is reflected in their relative abundances during in vitro evolution (Figure 2a). Interestingly, a defined mutant with the ΔlasR allele alone was not able to enrich against the defined psdR1 mutant, further demonstrating the effect that a large increase in absolute fitness can have on relative fitness against an obligate cheater.

Fourth, to investigate resistance of the HC to obligate cheating, we initiated competitions with the HC genotype (psdR1 lasR5) in majority (initial frequency of 0.99). We observed resistance to invasion by the ΔlasR cheater, but when the obligate cheater allele is paired with the evolved psdR allele, the psdR1 ΔlasR relative fitness again rose above 1.0 (Figure 5b).

Taken together, we confirmed our original predictions of the evolutionary trajectories (Figure 2g) of an evolving P. aeruginosa population. However, these relative fitness measurements do not fully explain the sustained cooperative growth of the evolved population. While even the psdR mutant was susceptible to subsequent invasion by psdR1 ΔlasR mutants, it is plausible that it would tolerate a higher proportion of cheaters, thereby maintaining high population growth in cooperative growth environments.

Cheater load

To finally determine the effect of increasing fractions of cheaters on the mean group fitness of the entire population, or cheater load, we again used defined co-culture experiments in caseinate medium. We varied the initial frequencies of an obligate cheater, ΔlasR or psdR1 ΔlasR, with respect to the cooperating parent strain, wild-type or psdR1, respectively, and quantified total population growth after 24 h. As expected, we found the burden of cheaters was significantly lower for the psdR1 mutant compared with the wild-type. Significant decreases in population productivity in the psdR1 background did not occur until the cheater was at a frequency of ⩾0.75, compared with 0.25 for the wild-type (Figure 6). This result demonstrated that psdR1 mutation helps stabilize cooperative, proteolytic growth as long as obligate cheaters are not the majority.

Figure 6.

Cheater load. Cheater load expressed as relative growth yield of the entire population. Co-cultures of a ΔlasR mutant cheater and its wild-type parent (filled bars), and of a psdR1 ΔlasR double mutant cheater and its psdR1 single mutant parent (empty bars) were grown for 24 h in caseinate medium. Starting cheater frequencies are indicated on the horizontal axis. Growth yield of each parent strain culture without cheater is set to 100%. Means and s.e.m. are shown (n=3). *Significant difference from respective parent strain without cheater as determined by unpaired t-test (P<0.05).

Discussion

In this paper, we identified and characterized a mechanism that helps transiently stabilize cooperative behavior in the QS pathway of the opportunistic pathogen P. aeruginosa. Growth in the QS-dependent cooperative environment described here strongly selects for non-social mutations that increase absolute fitness, thereby leading to increased tolerance to cheaters. Such adaptive mutation has been described in other microbial systems. Morgan et al. (2012) theoretically and experimentally showed that siderophore-producing populations of Pseudomonas fluorescens grown under iron-limiting conditions cannot be invaded by nonproducing mutants. The authors proposed that this occurred because the numerically dominant cooperators had a greater chance of obtaining a beneficial mutation that could sweep through the population. However, the underlying mutation was not identified. In a related study, Waite and Shou (2012) engineered a system with obligatory mutualistic cooperation between two nonmating yeast strains. The addition of an obligate cheater strain that exploits a common good shared between the two mutualistic cooperators lead to an adaptive race to either preserve cooperation or fail through population collapse. In the cases where cooperation was preserved, the cooperating subpopulation acquired a beneficial mutation that helped purge the cheater phenotype from the population. Here, the genetically engineered nature of the cooperative system raised questions about its relevance.

Our in vitro evolution experiments were initiated with pure cultures of wild-type bacteria. Under these conditions, there was essentially no adaptive race between cooperators and cheaters initially, because the non-social adaptation emerged first and quickly dominated the population. Of course, these mutants with non-social adaptations were then subject to invasion by cheaters that also carry the adaptation. It is possible that an adaptive race between these two evolved genotypes would eventually result in a second non-social mutation that further sustained cooperative growth, consistent with previous work on long-term microbial adaptation (Wiser et al., 2013). This stochastic scenario may explain why we observed a collapsing population in only one out of five in vitro evolution experiments (Figures 2b–f). Genome sequencing of late isolates beyond day 12 would be required to confirm this notion. As in the study by Waite and Shou (2012), we found that in defined co-cultures, non-social adaptation conferred resistance to cheaters with an otherwise wild-type background. However, in contrast to Waite and Shou (2012), evolved cheaters were still able to invade their cooperating parent strains. Our result is plausible in that cooperators, no matter how evolved, inevitably divert a portion of their resources into the secretion of public goods, resulting in an inherent growth disadvantage compared with nonproducing strains.

An increase in the absolute fitness of P. aeruginosa during proteolytic growth was realized through a loss-of-function mutation in the transcriptional repressor PsdR that, in turn, increases intracellular dipeptide transport and processing. This adaptation suggests that QS-dependent extracellular proteolysis is not growth-rate limiting during in vitro evolution, at least not exclusively. Presumably, proteolysis is only limiting during an initial lag period at the beginning of each growth cycle in caseinate medium. Abundant protease secretion during this period may lead to an excess in proteolytic break-down products that await uptake and processing later in growth. Here, psdR mutants would benefit. This effective separation in cooperative and non-cooperative selective targets during QS-dependent in vitro evolution of P. aeruginosa is illustrated in Figure 7. The psdR mutation proportionally increased the growth rates of cooperators and cheaters in co-culture, because the psdR lasR mutant showed the same relative fitness in psdR mutant co-culture as did the lasR mutant in wild-type co-culture (Figure 5b). This general impact on growth is nevertheless sufficient to explain its cooperation-stabilizing effect during in vitro evolution: mixed cooperator/cheater populations deficient in PsdR reach saturation faster than those with functional PsdR and are consequently more robust to cycles of dilution and regrowth.

Figure 7.

Targets of selection during P. aeruginosa QS-dependent in vitro evolution. Cooperative (left) and non-cooperative (right) targets of selection are illustrated in this schematic model. QS-controlled public goods, specifically extracellular proteases (red) that degrade polypeptides outside the cell (chains of yellow circles, each of which represent individual amino acids), constitute the cooperative target. PsdR-mediated repression of genes (green) coding for proteins (pink) that facilitate the uptake (DppA3) and intracellular processing (MdpA) of dipeptides constitutes the non-cooperative target. Temporal separation of these selective targets likely accounts for the evolutionary dynamics observed in this study.

The high number of independent psdR mutations during in vitro evolution was a surprise, raising the possibility that this locus is a mutational hot spot. However, this notion is not supported by the analysis of published P. aeruginosa genomes. Out of 18 genomes in the NCBI (National Center for Biotechnology Information) database, 17 contain a psdR homolog with ⩾99% identity, and 9 of which show 100% identity (Altschul et al., 1990) (http://blast.ncbi.nlm.nih.gov/). This ubiquity and sequence conservation implies that a functional PsdR is likely necessary for the evolutionary success of P. aeruginosa in its natural environment, although the situation may be different for other Pseudomonas species with dpp operons. P. protegens Pf-5 (formerly P. fluorescens) contains a truncated, presumably inactive psdR allele, whereas several other Pseudomonas spp. do not carry psdR at all (Kiely et al., 2008). The maintenance of a functional PsdR in natural P. aeruginosa isolates suggests that proteolysis may limit growth more often than subsequent peptide processing, or that PsdR activity may be modulated through a natural ligand and derepression would sufficiently increase dipeptide metabolism. PsdR is a Mer-type regulator with a helix-turn-helix DNA-binding domain and a cupin sensor domain that has the potential to respond to a variety of environmental stimuli (Brown et al., 2003; Kiely et al., 2008).

A second, beneficial effect of the psdR mutation was that it was able to promote cooperative growth near wild-type levels when paired with a partial loss-of-function lasR allele, lasR5. This mutation in lasR, by itself, conferred intermediate levels of proteolysis and proteolytic growth in culture. Thus, the random emergence of certain lasR mutations, particularly in an adapted parent, is not detrimental to cooperative growth. Although lasR5 affects LasR-dependent phenotypes other than caseinolysis (Sandoz et al., 2007), the precise impact on the entire regulon is not clear. The lasR5 mutation substitutes a valine for a nonconserved alanine at position 228 in the DNA-binding domain of LasR, presumably weakening, but not completely eliminating, interaction with target promoters.

Given the properties of the lasR5 mutation, one might expect a HC to exhibit lower relative fitness than a fully lasR-deficient cheater when paired with a cooperator. We observed this difference with strains harboring the psdR1 mutation, but not with those harboring functional PsdR. A possible explanation for this discrepancy could be the large size of the lasR-controlled QS regulon and the difference in the relative burdens it imposes on cooperators with and without the psdR mutation. Potential fitness differences between lasR5 and ΔlasR alleles stemming from the variable costs of cooperative extracellular proteolysis (and other LasR-dependent behaviors) could be effectively masked by PsdR-mediated repression of dipeptide uptake and processing, and only manifest after this rate-limiting step has been removed through mutation of psdR. Thus, genetic context may be important when considering the relative fitness contributions from a cooperative allele. This idea is also supported when interpreting our results in the framework of kin selection theory.

Kin selection theory, encapsulated in Hamilton's rule, states that cooperation evolves if rb – c>0, where b is the benefit of cooperation, c is the cost of cooperation and r is the genetic relatedness between actors and recipients (Hamilton, 1964a, 1964b). It has been shown that the cost c of bacterial cooperation may decrease with increased resource supply (Brockhurst et al., 2008). Analogously, psdR mutation appears to alleviate c by increased use of the products of protease digestion. This reduction in c does not require a direct mechanism, that is, a direct effect of psdR on the cooperative trait itself, but merely reflects the context in which the behavior is performed. Given the nonlinearity of fitness effects in our system (including a synergistic effect from QS induction and a saturating effect from protease secretion), further analysis of frequency-dependent relative fitness in the context of a generalized form of Hamilton's rule would be required to precisely quantify b and c (Smith et al., 2010).

More broadly, cycles of non-social, genetic adaptation and cheating are unlikely to maintain cooperative behavior in the long term as environmental adaptation is expected to eventually reach an optimum. Non-social adaptation through mutation likely works in concert with other mechanisms that stabilize cooperative behavior, and may be particularly beneficial early in the evolution of cooperative behavior. The generally high phenotypic plasticity of present-day microbes with unpredictable life histories would appear sufficient for coping with most changes in their natural environment. In microbes, other stabilizing mechanisms include positive assortment of cooperating individuals through, for example, colonial growth (Fletcher and Doebeli, 2009), the linkage of cooperative behaviors with other essential traits through pleiotropy (Foster et al., 2004) and metabolically prudent regulation of public goods such that their production is only initiated if the limiting nutrient is not also a building block of the good (Xavier et al., 2011; Mellbye and Schuster, 2014). It seems that pleiotropic control of extracellular proteolysis and subsequent intracellular metabolism via QS would be a reasonable strategy to curtail cheating, and a recent investigation using a similar in vitro evolution system has provided support for this notion (Dandekar et al., 2012). QS cheaters that do not contribute to proteolysis would be punished with reduced nutrient uptake and processing. We found that lasR indeed controls mdpA transcription in rich medium (Schuster et al., 2003), but not in the minimal medium used in this study. Of course, the relative fitness advantage of lasR mutant cheaters is consistent with the latter. Even if pleiotropic control played a role here, our results suggest that QS regulation of mdpA or related genes would be subject to strong counterselection whenever dipeptide uptake and processing was growth-rate limiting.

In summary, we have shown that non-social, genetic adaptation to a new growth environment that requires QS can help maintain cooperative behavior in P. aeruginosa populations. The adapted population is still vulnerable to invasion by cheaters that also carry the adaptation. However, a higher intrinsic growth rate affords higher tolerance to these cheaters, and some lasR mutations even contribute to cooperative growth.

The authors declare no conflict of interest.