• Xue-Xian Zhang,

    1. Institute of Natural and Mathematical Sciences, Massey University at Albany, Private Bag 102 904, North Shore Mail Centre, Auckland, New Zealand
    Search for more papers by this author
  • Paul B. Rainey

    1. New Zealand Institute for Advanced Study and Allan Wilson Centre for Molecular Ecology and Evolution, Massey University at Albany, Private Bag 102 904, North Shore Mail Centre, Auckland, New Zealand
    2. Max Planck Institute for Evolutionary Biology, August-Thienemann Strasse 2, Germany
    Search for more papers by this author


The idea that bacteria are social is a popular concept with implications for understanding the ecology and evolution of microbes. The view arises predominately from reasoning regarding extracellular products, which, it has been argued, can be considered “public goods.” Among the best studied is pyoverdin—a diffusible iron-chelating agent produced by bacteria of the genus Pseudomonas. Here we report the de novo evolution of pyoverdin nonproducing mutants, genetically characterize these types and then test the appropriateness of the sociobiology framework by performing growth and fitness assays in the same environment in which the nonproducing mutants evolved. Our data draw attention to discordance in the fit between social evolution theory and biological reality. We show that pyoverdin-defective genotypes can gain advantage by avoiding the cost of production under conditions where the molecule is not required; in some environments pyoverdin is personalized. By exploring the fitness consequences of nonproducing types under a range of conditions, we show complex genotype-by-environment interactions with outcomes that range from social to asocial. Together these findings give reason to question the generality of the conclusion that pyoverdin is a social trait.

Recognition of a social dimension to the life of microbes began more than a decade ago (Shapiro and Dworkin 1997; Crespi 2001; Velicer 2003). Persuasive data indicate that behaviors such as collective “hunting” by Myxococcus (Dworkin 1996), cell death in stalks of Dictyostelium (Cornillon et al. 1994), mat formation in Pseudomonas fluorescens (Rainey and Rainey 2003), and bacteriocin production in Escherichia coli (Chao and Levin 1981) are cooperative traits and their evolution attributable to benefits conferred on recipients. A large body of theory provides plausible evolutionary mechanisms (Sachs et al. 2004).

More recently, there has been a shift to cast all microbes as inherently cooperative and to do so under the umbrella of social evolution theory (West et al. 2006, 2007; Foster 2007; Nadell et al. 2008, 2009; Foster 2010). Much stems from reasoning regarding extracellular products, such as enzymes, polysaccharides, and secondary metabolites, which, it has been argued, can be considered costly “public goods,” the benefits of which are shared among a common pool (West et al. 2006). Initial suggestions that this might be true stem from the discovery (or in vitro construction) of nonproducing types that can invade populations of producers (Velicer et al. 1998; De Vos et al. 2001; Rainey and Rainey 2003; Greig and Travisano 2004; Griffin et al. 2004; Dugatkin et al. 2005; Smith et al. 2006; Ross-Gillespie et al. 2007; Jiricny et al. 2010; Lee et al. 2010; Nadell and Bassler 2011). These have been labeled as “social cheats”: types that exploit public goods, but make no return contribution (Velicer 2003; West and Buckling 2003; West et al. 2006, 2007). Their existence has been taken as fulfilment of a central prediction from theory, thus giving credence to the initial premise, namely, that extracellular products are public goods. In many instances such a conclusion has been taken as sufficient to warrant the claim that extracellular products are not only public goods, but that the behavior is a cooperative (West et al. 2006, 2007), or even an altruistic (Griffin et al. 2004; West et al. 2006), trait.

Given impact of the social evolution framework on how the ecology, evolution, and functioning of microbes is understood, the fit between theory and biology requires careful assessment. There are two major issues: the first is whether a particular extracellular product is a public good, that is, whether it is equally available to both producers and nonproducers (Driscoll and Pepper 2010), the second is why nonproducing types, which fail to make extracellular products, are selected. Both issues are experimentally tractable and testable via a set of established criteria (West et al. 2006).

An additional issue concerns the evolutionary causes for the origin of public good production in the first place. Although it is often assumed that a cell producing an extracellular product is a cooperator (West et al. 2006, 2007), that is, that production of the public good is the result of selection on recipient cells, this need not be the case: the public good could be a spandrel (Gould and Lewontin 1979; Redfield 2002; Driscoll and Pepper 2010). Testing the hypothesis that a public good evolved because of benefits to others (or is maintained because of benefits to recipients), poses significant challenges: the greatest of these stems from lack of knowledge concerning the ecological circumstances under which the trait evolved and/or is maintained.

The focus of our study is the fluorescent siderophore (pyoverdin)—an iron-chelating molecule—produced by bacteria of the genus Pseudomonas (Visca et al. 2007). Pyoverdin is the most well studied of the secreted metabolites with reports that it is both a public good and a cooperative trait (West and Buckling 2003; Harrison and Buckling 2005; Harrison et al. 2006, 2008; Buckling et al. 2007; Ross-Gillespie et al. 2007, 2009; Harrison and Buckling 2009; Kümmerli et al. 2009a,b; Jiricny et al. 2010; Kümmerli and Brown 2010; Kümmerli et al. 2010). However, there are reasons to consider alternative hypotheses. Foremost is uncertainty surrounding the ecophysiology of pyoverdin (Kraemer 2004). Although pyoverdin plays a primary role in uptake of ferric iron, it also functions as a heavy metal resistance system (Braud et al. 2010; Hannauer et al. 2011): it may also—by virtue of the redox status of the ferripyoverdin complex and interactions with other redox active secondary metabolites—aid survival in nutrient depleted environments by providing access to phosphates, trace metals, and organic compounds associated with mineral phases (Kraemer 2004; Price-Whelan et al. 2006).

Uncertainties regarding ecophysiology mean that choice of environmental conditions for the study of pyoverdin is crucial. What makes sense in the lab may have little relevance elsewhere. For example, there is no prima-facie reason to discount the possibility that pyoverdin directly benefits producing cells (Redfield 2002; Driscoll and Pepper 2010; Julou et al. 2013; sharing due to overproduction could be a laboratory artifact). If privatized, then nonproducing mutants are unlikely to be cheats. Alternate explanations for nonproducers include the possibility that they are cross-feeding types that take advantage of resource partitioning activities of pyoverdin (Helling et al. 1987); they may be mutualists (MacLean et al. 2010; Driscoll et al. 2011), or they may be adaptive mutants in their own right—their increase being attributable to advantages that stem from avoidance of, for example, iron-mediated potentiation of oxygen toxicity (Hassett et al. 1996; Dao et al. 1999).

Here we follow the evolution of a pyoverdin-producing genotype in a single environment, observe the emergence of nonproducers, genetically characterize these mutants and then test the appropriateness of the social evolutionary framework by performing growth and fitness assays in the same environment in which the nonproducing cells evolved. We find that in King's Medium B (KB), an environment that promotes the production of pyoverdin, nonproducers have a fitness advantage over producers. By exploring the fitness consequences of nonproducing types in a range of media, we show complex genotype-by-environment interactions—even personalization of pyoverdin—which together suggests the need for caution in concluding that pyoverdin is a social trait.



The wild-type strain of P. fluorescens SBW25 was isolated from sugar beet at the University of Oxford farm (Wytham, Oxford, U.K.; Silby et al. 2009). Eschericia coli DH5αλpir was used for gene cloning and as a donor for conjugation into Pseudomonas (Zhang et al. 2006). Bacterial strains and plasmids used in this study are provided in Table S1. Antibiotics and supplements were used at the following concentrations (μg mL−1): tetracycline (Tc), 10; spectinomycin (Sp), 50; kanamycin (Km), 50; ampicillin (Ap), 50; gentamycin (Gen) 25; nitrofurantoin (Nf), 100; 5-bromo-4-chloro-3- indolyl-β-d-galactopyranoside (X-gal), 40.

The two general Pseudomonas media, KB (King et al. 1954) and casamino acids medium (CAA) were used. King's Medium B contains (per liter) 20 g proteose peptone No. 3 (BD Diagnostic Systems, Auckland, New Zealand), 10 g glycerol, 1.5 g K2HPO4 and 1.5 g MgSO4.7H2O. Casamino acids medium contains (per liter): 5 g casamino acids (BD Diagnostic Systems), 1.18 g K2HPO4.3H2O and 0.25 g MgSO4.7H2O. Both media promote the production of pyoverdin: addition of iron (0.45 mM FeSO4) suppresses pyoverdin production indicating the media are deplete of biologically available iron.

Pseudomonas strains were cultured in 30 mL universal glass vials (microcosms). Where indicated, FeSO4 was used at the final concentration of 0.45 mM. The iron chelating agent 2,2′-dipyridyl (DP) was purchased from Sigma–Aldrich (Auckland, New Zealand) and a stock solution of 100 mM was prepared in ethanol.

Production of pyoverdin was estimated by measuring fluorescence of the supernatant at 460 nm with an excitation wavelength of 365 nm in a Synergy 2 multimode microplate reader (BioTek Instruments; Jiricny et al. 2010).


Standard DNA manipulation techniques were used. The sequences of primers for pvdS amplification are provided in Table S1. Additional primers were designed for fpvA, fpvI, and fpvR amplification and DNA sequencing (sequences available upon request).


The pvdS229 allele was introduced in P. fluorescens SBW25 by two-step allelic exchange (Zhang and Rainey 2007b). A primer pair (PvdS1/PvdS4) was designed to amplify a 720 bp DNA region with the pvdS229 mutation at its center (Table S1). The polymerase chain reaction (PCR) product, amplified from genomic DNA of the evolved Pvd strain ST60-1, was cloned into plasmid pCR8/GW/TOPO. DNA sequence was confirmed, the fragment retrieved by BglII digestion, and cloned into the BglII site of integration vector pUIC3 (Rainey 1999). The resulting recombinant plasmid pUIC3-68 was mobilized into ancestral P. fluorescens SBW25 with the help of pRK2013. Integration of pUIC3-68 into the SBW25 genome via homogenous recombination was selected on Lysogeny Broth (LB) plates containing Nf, Tc, and X-gal. Allelic exchange mutants were obtained by d-cycloserine enrichment (Zhang and Rainey 2007b).

For pvdS complementation and constitutive expression, the coding region of pvdS was amplified from wild-type SBW25 by PCR using primers pvdS_F and pvdS_R2. The PCR product (600 bp) was cloned into pCR8/GW/TOPO and sequenced. The fragment was then retrieved by BamHI/HindIII digestion and cloned into pUC8-mini-Tn7T-LAC. The resulting plasmid was mobilized into wild-type SBW25, and the pvdS229 mutant, by conjugation. The Tn7 element containing pvdS was integrated into its unique chromosomal site. Correct insertion was confirmed by PCR analysis using primers Tn7R109 and glmS-25 (Table S1).


Relative fitness of Pseudomonas strains was determined by direct competition. Cells were prepared from cultures stored at −80°C, first grown in LB and then subcultured once for ∼24 h in either KB or CAA. To initiate the competition, 6 μL of the mixed culture (∼105 cells) was inoculated into 6 mL fresh medium. A neutrally marked strain of SBW25-lacZ (Zhang and Rainey 2007a) was used as the competitor. For fitness of the pyoverdin constitutively expressing strain PBR852, ancestral SBW25 carrying a similar Tn7 element with a promoterless lacZ (strain PBR853) was used as the wild-type competitor to ensure costs associated with carriage of the Tn7 cassette were equivalent. Relative fitness was calculated as log ratio of Malthusian parameters of the two strains being compared (Lenski et al. 1991).


Treatment effects were assessed by analysis of variance (ANOVA; JMP 8.1). The experiment described under prediction 3 was first analyzed using a five-way crossed (fully factorial) ANOVA with medium (KB, CAA) genotype (PBR851, PBR840), mutant frequency (1%, 50%, and 99% initial founding frequency), environmental structure (structured, unstructured), and iron (medium supplemented with iron chelator and/or free iron), as fixed factors. Summary statistics from this overall ANOVA (F77,454 = 27.179, P < 0.0001) are shown in Table S9 and revealed highly significant effects of all treatments and the majority of interaction terms. Raw data are available in Table S10.



Questions pertaining to the evolutionary significance of specific phenotypes are powerfully addressed through the study of mutants that arise spontaneously under reproducible ecological conditions. To this end, eight replicate populations of ancestral P. fluorescens SBW25 were propagated by serial transfer (every 48 h) in spatially structured (nonmixed) and spatially unstructured (well-mixed) microcosms containing KB for 120 days.

After 600 generations, bacteria were plated onto KB agar plates and ∼500 single colonies from each microcosm were examined under UV light for evidence of pyoverdin production. Five of the eight nonshaken cultures contained non-pyoverdin-producing (Pvd) colonies (Table S2). The remaining three populations contained variants with visually reduced pyoverdin production. Of the eight unstructured microcosms, only one produced Pvd colonies and the remaining seven showed no evidence of variation. The effect of environmental structure on the evolution of Pvd mutants (based on the presence/absence of Pvd colonies) was statistically significant (likelihood ratio, χ2 = 4.56, P = 0.0328).

At 600 generations, a measure of pyoverdin production was obtained. Analysis of variance showed a significant effect of spatial structure on pyoverdin production (F1,47 = 86.744, P < 0.0001): mean pyoverdin per microcosm in structured microcosms was 199 relative fluorescence units [RFU; 95% confidence interval (CI) = 96–303], and 874 RFU (95% CI = 771–978) in unstructured environments (Table S2). Compared to the ancestral genotype, after 600 generations, the derived types produced significantly less pyoverdin per microcosm [F1,18 = 29.600, P < 0.0001 (structured microcosms) and F1,18 = 9.617, P = 0.006 (unstructured microcosms)].


One representative Pvd colony was chosen from each of the five structured microcosms (designated ST60-1, ST60-2, ST60-3, ST60-5, and ST60-8) and one unstructured microcosm (US60-3). The DNA sequence of candidate regulatory genes (fpvA, fpvI, fpvR, and pvdS) was obtained from each mutant and compared to the sequence of the ancestral genotype. Mutations in pvdS—a gene that encodes the extracytoplasmic family sigma factor PvdS and which directs expression of the pyoverdin biosynthetic genes (Cunliffe et al. 1995)—were detected in four of the six Pvd strains. The single nucleotide substitutions all generated nonsynonymous amino acid changes [G229A (D77N) in ST60-1; G503A (C168Y) in ST60-3; T470G (V157G) in ST60-5; G469A (V157M) in ST60-8].

To separate the effects of pyoverdin-specific mutations from other mutations that might have arisen during the course of the 600 generations of selection, mutation pvdS229 from ST60-1 was moved into ancestral P. fluorescens SBW25 by homologous recombination. The resulting mutant, PBR851 (pvdS229), was Pvd (pyoverdin production was restored upon introduction of a cloned copy of wild-type pvdS). Measures of pyoverdin production by PBR851 confirmed that the mutation is a “null,” that is, it produces no detectable pyoverdin (Fig. 1).

Figure 1.

Population density and production of pyoverdin in structured (A) and unstructured (B) microcosms. Circles show population density of Pvd+ (ancestral genotype), Pvd (PBR851), and a 1:1 mixture after 48 h growth in King's Medium B. The gray circle indicates the population density of PBR851 (containing the pvdS229 allele) in the environment in which the allele evolved. Triangles show pyoverdin production per microcosm estimated by measuring fluorescence of the supernatant at 460 nm and expressed as relative fluorescence units. Data are mean and SEs of three biological replicates.

Generation of PBR851 by placement of the pvdS229 allele in the ancestral genotype excluded the possibility of observing fitness effects attributable to additional mutations. Although such mutations might further enhance the competitive ability of ST60-1, the need to work with an isogenic mutant free of confounding effects due to mutations conferring niche specificity is of overriding importance.

PBR851 can use pyoverdin produced by the ancestral genotype. This was demonstrated by comparing the final cell density of PBR851 grown under iron-limiting conditions [KB supplemented with 500 μM 2,2′-dipyridyl (an iron-chelating agent added to decrease iron-availability)] in the presence and absence of the ancestral pyoverdin-producing genotype: in the presence of the pyoverdin-producing genotype, PBR851 reached a significantly greater (six-fold) final density after 48 h than when grown alone (F1,8 = 48.481, P = 0.0002). Together, these data show that Pvd mutants readily evolve in spatially structured KB microcosms, are frequently caused by mutations in the alternative sigma factor PvdS, and can use exogenously supplied pyoverdin.

Although our primary focus is the pvdS229 mutation carried by PBR851, a second Pvd genotype, PBR840 (Moon et al. 2008), was also included. PBR840 carries a deletion (constructed in vitro) of the entire pvdL open-reading frame (ΔpvdL), but is otherwise isogenic with the ancestral genotype. Like PBR851, PBR840 uses exogenously supplied pyoverdin (Moon et al. 2008). Inclusion of PBR840 allowed a test for genotype-specific effects: in PBR851 the pvdS mutation resides in a regulator of pyoverdin, whereas the mutation in PBR840 resides in a structural gene (pvdL) necessary for pyoverdin biosynthesis.


To investigate the social context of Pvd types, we performed a series of experiments to assess the fitness of pyoverdin-producing and nonproducing types under different scenarios. According to social evolution theory (West et al. 2006, 2007), if Pvd types are cheats and pyoverdin is a public good then: (1) populations of pyoverdin producers should outperform populations of nonproducers in monoculture; (2) populations of nonproducers should outperform populations of producers in monoculture when the environment is modified so that pyoverdin is not required; (3) nonproducers should outperform producers in mixed culture; (4) producers should not increase in frequency from rarity against a numerically dominant population of nonproducers.

An additional prediction concerns the effect of spatial structure. Producers are expected to be favored in structured environments (over what they would be in less structured environments), whereas nonproducers are expected to have an advantage in unstructured environments (Hamilton 1964; Chao and Levin 1981; Nowak and May 1992; Fletcher and Doebeli 2009; Driscoll and Pepper 2010). In structured environments, cooperation is favored (because public goods are localized and interactions among genetically related individuals are facilitated), whereas unstructured environments favor cheating types (because resources from producing types are equally available to nonproducing types). Although there is the potential for effects due to changes in spatial structure to be more complex, these simple expectations are likely to be valid for a product that is diffusible and where the product increases an essential resource available to the producing cells (Queller 1992; Wilson 1992; West et al. 2002).

We note at this point two findings that run counter to expectations from social evolution theory: reduced production of pyoverdin in structured microcosms after 600 generations of evolution even when Pvd types are not present (Table S2), and a greater prevalence of Pvd mutants in structured microcosms.

Prediction 1

If pyoverdin is a public good then monocultures of pyoverdin-producing cells should outperform monocultures of pyoverdin-deficient types. We began by determining the final density of monocultures of the ancestral producer (Pvd+) and the Pvd mutant PBR851 (pvdS229) after 48 h growth in KB. Contrary to expectation, the cell density in monocultures of the Pvd mutant was greater than the producer, with the effect being most pronounced in structured microcosms (data not shown). The experiment was repeated and an intermediate data point (a 1:1 mixture of ancestral Pvd+ and Pvd) included. After 48 h growth, microcosms founded by the Pvd+ genotype had lowest cell density, followed by the mixture, followed by the Pvd genotype (Fig. 1A, B). Measurements of pyoverdin produced by the end of the 48 h growth period showed a marked negative correlation with cell density (Fig. 1A, B). The mixed cultures were intermediate for both cell density and pyoverdin production (Fig. 1A). This finding points to the possibility that pyoverdin production in KB is not only unnecessary, but is maladaptive.

A more extensive analysis was next performed in which growth of the ancestral Pvd+ genotype, along with both Pvd mutants [PBR851 (pvdS229) and PBR840 (ΔpvdL)], was determined after 48 h in monoculture. Once again, against prediction, populations of both Pvd genotypes reached significantly greater cell densities at 48 h in KB compared to populations of producer cells (Fig. 2A). Reduction of available iron by addition of 250 μM 2,2′-dipyridyl failed to reverse this result (the yield was equivalent). However, Pvd types were at a disadvantage in KB culture supplemented with 500 μM 2,2′-dipyridyl. Qualitatively similar results were obtained when experiments were repeated in unstructured KB microcosms (Fig. 2B). In an additional experiment, time-course data for the growth of PBR851, and the pyoverdin-producing ancestral genotype, was obtained in structured KB microcosms (Fig. 3).

Figure 2.

Population density of monocultures of Pvd+ (ancestral) and Pvd (PBR851 and PBR840) genotypes after 48 h growth in either structured (A) or unstructured (B) microcosms at different levels of iron availability. Microcosms contained King's Medium B supplemented (where appropriate) with either iron FeSO4 (0.45 mM) or the iron-chelating agent 2,2′-dipyridyl (250 and 500 μM). Data in A and B were collected in separate experiments and analyzed independently. In both instances, analysis of variance showed highly significant treatment effects (F11,35 = 26.626, P < 0.0001) and (F11,43 = 9.300, P < 0.0001), respectively. Bars show mean and SEs of three biological replicates. Gray lines with probability values show those environments in which the cell density of Pvd genotypes is significantly different to the cell density of the ancestral Pvd+ genotype as determined by LS means contrast tests. The gray bar highlights the population density of PBR851 (containing the pvdS229 allele) in the environment in which the allele evolved.

Figure 3.

Growth of Pvd+ (ancestral) and Pvd (PBR851) in structured microcosms. Microcosms contained King's Medium B. Data are mean and SEs of six biological replicates and were analyzed by repeated measures multivariate analysis of variance. The sphericity χ2-test was highly significant (χ2 = 182.21, P < 0.0001) necessitating use of the adjusted univariate F-test (Greenhouse–Geisser): main effects time (F2.14,21.4 = 346.36, P < 0.0001); time × genotype (F2.14,21.4 = 18.24, P < 0.0001). Differences between genotypes were highly significant at 30 h (F1,10 = 17.51, P = 0.0019) and 48 h (F1,10 = 21.83, P < 0.0001).

Together these data led us to reject the hypothesis that in structured KB microcosms (the environment in which the pvdS229 allele evolved) pyoverdin is a public good that confers a group-level benefit to producing cells: not only is pyoverdin unnecessary for growth in KB, but production comes at a cost to the producing cells. The finding is consistent with observations of previous long-term selection experiments with P. fluorescens in which the production of fluorescent pigment can be entirely lost from microcosms after ∼1000 generations (P. B. Rainey, unpubl. ms.).

Prediction 2

Cooperation is costly to individuals. It has been suggested that insight into costs and benefits can be obtained by examining fitness under conditions where, first, the trait is beneficial, and second where the trait is not beneficial (Griffin et al. 2004; West et al. 2006, 2007). We measured, in monoculture, growth of PBR851 and PBR840, and the pyoverdin-producing ancestral genotype, in both structured and unstructured microcosms supplemented with iron (Fig. 2A, B). No significant difference was observed between the population density of Pvd+ and Pvd types at 48 h in structured microcosms (Fig. 2A; F1,24 = 0.009, P = 0.930). The result was not affected by environmental structure (Fig. 2B; F1,32 = 0.732, P = 0.399).

Although this is at odds with previous findings (Griffin et al. 2004), it is consistent with the fact that pyoverdin is repressed in iron-replete media (Ochsner et al. 1995; Hassett et al. 1996). We further explored issues relating to cost via the construction of a genotype that produces pyoverdin even under iron-replete conditions: the ancestral genotype was modified so as to bring expression of the pyoverdin regulon under the control of an IPTG-inducible (Ptac) promoter fused to pvdS. The resulting genotype (PBR852) produces pyoverdin upon addition of IPTG—irrespective of the iron status of the environment (Fig. S1) and grows more slowly (in monoculture) than the ancestral genotype in KB (Fig. S2). A net cost associated with constitutive pyoverdin production was evident through comparisons with wild-type growth after 48 h in iron-replete KB (Fig. S3).

We examined the fitness consequences of pyoverdin overproduction by performing competitive fitness assays in both structured and unstructured KB microcosms under different levels of iron availability (KB was supplemented with either iron, or the iron chelator 2,2′-dipyridyl; Fig. 4). In unsupplemented KB, the constitutive producer (PBR852) had lower fitness than the ancestral genotype. Analysis of variance revealed a highly significant difference among treatments (F11,24 = 17.945, P < 0.0001), with the effects of iron (F5,24 = 24.993, P < 0.0001), environmental structure (F1,24 = 30.048, P < 0.0001), and the interaction term (F5,24 = 8.477, P < 0.0001) being highly significant. The simple prediction, namely, that the fitness cost to overproduction would be alleviated in iron-limiting environments was borne out for assays performed in structured microcosms, however, in unstructured microcosms there was a significant net cost to pyoverdin production even under conditions of extreme iron limitation (Fig. 4A).

Figure 4.

Fitness (A) and population density (B) of pyoverdin overproducing Pvdc genotype (PBR852) in structured and unstructured microcosms with different levels of iron. Fitness of Pvdc was determined relative to the ancestral Pvd+ genotype (PBR853) after 48 h growth in King's Medium B supplemented (where appropriate) with either 0.45 mM FeSO4 (high iron), or with different levels of the iron-chelating agent 2,2′-dipyridyl (250–1000 μM). IPTG was included at the final concentration of 100 μM to induce pyoverdin production in PBR852. Fitness data (mean and SEs of three biological replicates) are log10 transformed ratios of the Malthusian parameters for each competitor. See text for results of analysis of variance (A). Density of Pvdc and Pvd+ was determined after 48 h growth in monoculture and mixed culture supplemented with 750 μM 2,2′-dipyridyl. Analysis of variance revealed significant differences among densities in both structured (F4,29 = 11. 238, P < 0.0001) and unstructured environments (F4,29 = 20.914, P < 0.0001). Density measures not connected by the same letter are significantly different (Tukey's honestly significant difference). Data are mean and SEs of six biological replicates (B).

One explanation for the reduced fitness of PBR852 under conditions of extreme iron limitation is that the ancestral genotype exploits pyoverdin produced by the constitutive type. If so, then the population density of the ancestral type should be greatest when grown in the presence of the constitutive type. The data in Figure 4B do not support this prediction. The ancestral type achieved a significantly higher cell density in monoculture, compared with mixed culture (in unstructured microcosms the opposite was true; Fig. 4B). This shows that the reduced fitness of the constitutive producer is a consequence of a net cost to overproduction of pyoverdin specifically in unstructured micrcosms, and is not due to “cheating” by the ancestral type. Although the deleterious effects are shared in mixed culture (the ancestral type is negatively influenced by presence of the constitutive type; Fig. 4B), the fitness data (Fig. 4A) indicates that the negative effect is felt most strongly by the constitutive producer. This indicates some personalization of this deleterious effect.

Prediction 3

Based on the expectation that there is a net cost to pyoverdin production that is avoided by the nonproducer, prediction 3 states that nonproducing types should outperform producers in mixed culture. From the perspective of the producing type, this amounts to a classic tragedy of the commons (Hardin 1968) in which the most basic form of a public goods game cooperation inexorably declines. Depending on the nature of the payoff matrix (and ecology), this may manifest as negative frequency-dependent selection (Killingback et al. 2010), whereby the fitness advantage to the nonproducer increases as a negative function of its frequency. The advantage to nonproducers is expected to be greatest in unstructured environments and reduced in structured environments. Populations of the ancestral producer genotype and nonproducer mutant types [both PBR851 (pvdS229) and PBR840 (ΔpvdL)] were founded at different initial frequencies (1%, 50%, and 99%). The two types were then allowed to compete and the final frequency of each type was determined after 48 h growth. Calculation of the ratio of Malthusian parameters provided a measure of the fitness of the nonproducing type relative to the producer. Although the focus of attention is the fitness of PBR851 in KB (in different social contexts), the ensuing experiments explored the fitness effects of both PBR851 and PBR840 in KB and KB supplemented with 2,2′-dipyridyl, in both structured and unstructured microcosms.

Data were first analyzed using a four-way crossed ANOVA design (with genotype, mutant frequency, environmental structure, and iron as fixed factors). Analysis of variance showed all factors to be highly significant along with the majority of interaction effects (Table S3). Given the significance of the genotype effect (F1,172 = 265.555, P < 0.0001; also evident by comparing Fig. 5A–C with Fig. 5E–G) and the need to understand more clearly the nature of the interaction effects, data were subsequently analyzed separately for each genotype using a three-way crossed ANOVA design with mutant frequency, environmental structure, and iron as fixed factors (Table S4). To unravel effects due to environmental structure at different mutant frequencies, specific comparisons were made using LS means contrast tests according to a priori predictions.

Figure 5.

Fitness of Pvd+ (ancestral) and Pvd (PBR851 and PBR840) genotypes in structured and unstructured microcosms at different initial founding frequencies. Fitness of Pvd (PBR851) relative to the ancestral Pvd+ genotype when the founding ratio of the Pvd genotype was 1% (A), 50% (B), and 99% (C). (D) Fitness of the ancestral Pvd+ genotype relative to the Pvd (PBR851) genotype when the founding ratio of the Pvd+ genotype was 1% (these data are the reciprocal of those presented by C). Fitness of Pvd (PBR840) relative to the ancestral Pvd+ genotype when the founding ratio of the Pvd genotype was 1% (E), 50% (F), and 99% (G). (H) Fitness of the ancestral Pvd+ genotype relative to the Pvd (PBR840) genotype when the founding ratio of the Pvd+ genotype was 1% (these data are the reciprocal of those presented by G). Microcosms contained King's Medium B (KB) supplemented (where appropriate) with the iron-chelating agent 2,2′-dipyridyl. Fitness data (mean and 95% confidence intervals of between five and six biological replicates) are log10 transformed ratios of the Malthusian parameters for each competitor. Gray circles depict fitness of PBR851 (containing the pvdS229 allele) in the structured KB environment in which the allele evolved. See text and Tables S4 and S5 for details of statistical analyses. Data are available as Table S10.

With focus first on PBR851 (Fig. 5A–C), ANOVA revealed a significant difference among treatments (F17,107 = 18.459, P < 0.0001; Table S4) and a pronounced effect of mutant frequency (F2, 107 = 11.954, P < 0.0001). The greater overall fitness of PBR851 when rare, compared to its fitness when common, is in accord with expectation (Fig. 5A–C). However, comparisons made within each environment reveal marginal (at best) effects. For example, although the mean fitness of PBR851 in unstructured KB microcosms at 1% initial mutant frequency [mean = 0.106 (5% CI = 0.100–0.112)] is greater than at 50% [mean = 0.048 (95% CI = 0.037 – 0.060)] and 99% [mean = 0.053 (95% CI = 0.002–0.104)], the difference in fitness at 50% and 99% is not significant.

Contrary to expectation is the effect of environmental structure: although an effect is apparent at 1% and 50% initial founding frequencies (F1,90 = 93.669, P < 0.0001 and F1,90 = 66.319, P < 0.0001, respectively) it is opposite to the predicted trend. At 99% initial mutant frequency, the effect of environmental structure is not significant (F1,90 = 1.555, P = 0.216).

In general, and in accord with expectations from social evolution theory, within each mutant frequency class, fitness of PBR851 tends to be greatest when iron is scarcest. This is consistent with the fact that the cost of pyoverdin production increases with iron scarcity due to regulation of pyoverdin synthesis. The effect is most noticeable in unstructured environments.

Turning to the pvdL deletion mutant PBR840 (Fig. 5E–G), the effect of initial founding frequency on mutant fitness is much flatter and while the overall effect is significant, it is only moderately so (F2,64 = 4.680, P = 0.013; ANOVA results are summarized in Table S5). Moreover, LS means contrast tests within each environment showed no significant differences. A significant effect of environmental structure was evident at 1% initial mutant frequency (F1,53 = 7.500, P = 0.008), but as for PBR851, the trend is opposite to expectation. No effect of environmental structure was observed at 50% and 99% initial mutant frequency (F1,53 = 0.197, P = 0.659 and F1,53 = 0.878, P = 0.353, respectively).

Prediction 4

If pyoverdin is a public good, and equally accessible to producers and nonproducers alike, then the pyoverdin-producing (Pvd+) ancestral type should not invade against nonproducing types from rarity. This effect is expected to be pronounced in unstructured microcosms, and less evident in the structured environment.

The fitness data shown in Figure 5D and H are the reciprocal of those shown in Figure 5C and G and depict the fitness of the producer genotype when rare. Values below unity (the dotted line) indicate inability of the producer to invade from rarity. Figure 5D shows that the ancestral Pvd+ genotype has a fitness that is less than PBR851 in structured KB microcosms (it cannot invade from rarity), however the effect is marginal (fitness relative to PBR851 is −0.053 (95% CI = −0.104 to −0.002). Although consistent with predictions, this finding must be seen in light of the fact that in structured KB microcosms, PBR851 has an advantage due to the nonproduction of pyoverdin (see Prediction 1). In unstructured KB microcosms, there is a negative relationship between fitness of the Pvd+ type and iron scarcity as expected. However, this effect is erased in structured microcosms where there is indication of a weak positive effect on Pvd+ relative fitness. This suggests some personalization of pyoverdin in structured microcosms under conditions of low iron availability.

In the case of the pvdL deletion mutant PBR840 (Fig. 5H), the producer and nonproducer have equivalent fitness and there is no effect of spatial structure. In each case, the relative fitness of the ancestral genotype is greater than unity, once again indicating personalization of pyoverdin.

Effect of medium

The competition assays performed above were carried out in KB—the medium in which the pvdS229 allele evolved. However, because many investigations of the social effects of pyoverdin production are performed in CAA (a complex medium not dissimilar to KB), we repeated the measures of fitness of both PBR851 (pvdS229) and PBR840 (ΔpvdL) in CAA in different social contexts, as laid out in predictions 3 and 4. The simple expectation was that findings from KB would be mirrored in CAA, however, the assays revealed a striking effect of medium, capacity of the producing type to (under some conditions) personalize pyoverdin production, and many additional complex and contrasting effects (relative to effects in KB). Overall they showed that, contrary to expectations, the fitness of the pvdL deletion mutant (PBR840) was reduced in iron-deplete environments. Data are shown in Figure S4 and the summary statistics are in Tables S6–S8.


The causes of the success of PBR851 in structured KB microcosms beg explanation. The advantage appears to stem from not producing pyoverdin under conditions where it is not required. This is demonstrated by the enhanced growth of PBR851 in KB monoculture relative to the ancestral pyoverdin-producing genotype (Figs. 1-3). This raises the question as to why the wild type produces pyoverdin in the first place. A likely answer comes from recognizing the interplay between oxygen and iron availability and the fact that growth in structured KB microcosms generates a step oxygen gradient.

In the presence of oxygen, iron exists in the insoluble ferric (Fe3+) form and thus acquisition requires pyoverdin. Observation of cells growing in structured KB microcosms shows that pyoverdin produced during the earliest stages of growth is confined to the air–liquid interface—the region of oxygen abundance. Subsequently (between 24 and 48 h) pyoverdin diffuses throughout the microcosm (Fig. S5). Pyoverdin, when complexed with iron, activates PvdS [which directs expression of the pyoverdin biosynthetic genes (Cunliffe et al. 1995)] via the FpvA-FpvR-PvdS signaling pathway (Beare et al. 2003). Diffusion of the ferripyoverdin complex into the microaerobic broth phase thus stands to cause activation of pyoverdin synthesis in cells that do not require this molecule (under oxygen depleted conditions iron exists in the reduced ferrous (Fe2+) form which is soluble and readily available). Although in principle Fur-mediated repression of pyoverdin synthesis under anaerobic conditions should prevent this, our work shows that constitutive expression of PvdS overrides Fur-mediated repression, such that pyoverdin is produced even in the presence of an abundance of iron (see Prediction 2 and Fig. S1).

If pyoverdin synthesized by cells at the air–liquid interface subsequently diffuses through the microcosm causing Pvd+ cells in the broth phase to produce pyoverdin—even though this molecule is not required—then any fitness advantage to the mutant should be abolished in populations grown under microaerobic conditions where pyoverdin is not produced. To test this prediction, the Pvd genotype PBR851 was inoculated into structured KB microcosms along with the ancestral Pvd+ genotype at an equal founding ratio and a 15 mm layer of mineral oil laid above the broth. Control microcosms were included that were identical, but the oil overlay was omitted. Observation of microcosms overlaid with oil showed no evidence of the pyoverdin. The fitness of PBR851 was determined relative to Pvd+ after 48 h growth and ANOVA showed the effect of oil to be highly significant (F1,10 = 44.890, P < 0.0001). In the absence of oil, the mean fitness of PBR851 relative to the ancestral pyoverdin producing genotype was 0.075 (95% CI = 0.060–0.091), which is significantly greater than 0 (the fitness, by definition of the ancestral type), whereas in the presence of the oil layer the mean fitness of PBR851 was 0.008 (95% CI = −0.008 to 0.024), which is not significantly different from the fitness of the ancestral Pvd+ genotype.


In recent years, the notion that pyoverdin is a cooperative trait and that nonproducers are cheats has moved from interesting idea to established fact (West and Buckling 2003; Griffin et al. 2004; Harrison and Buckling 2005, 2009; Harrison et al. 2006, 2008; Buckling et al. 2007; Ross-Gillespie et al. 2007; Kümmerli et al. 2009a,b, 2010; Ross-Gillespie et al. 2009; Jiricny et al. 2010; Kümmerli and Brown 2010; Dumas and Kümmerli 2012). However, few studies have explored the relationship between the social dimension of pyoverdin and environmental factors. Our work shows that under certain laboratory conditions (media supplemented with high levels of iron-chelating agent) pyoverdin behaves as expected of a public good (Fig. 2), however conformity to the social evolution framework is dependent on both genotype and environment (Fig. 5). In some environments, pyoverdin-defective types evolve because production of pyoverdin is maladaptive (Fig. 1). Under other conditions pyoverdin appears to be personalized (Figs. 4, 5D, H, S4D, H). A recent study shows pyoverdin production to benefit individual cells (Julou et al. 2013). The discovery of just one set of conditions under which pyoverdin is produced, and yet does not behave in accord with social evolution theory, gives reason to question the generality of the conclusion that pyoverdin is a public good, that producers are cooperators and that nonproducers are social cheats.

Sensitivity to environmental conditions is not surprising, but it is problematic. Given uncertainties as to the ecophysiology of pyoverdin, and the environmental conditions under which pyoverdin has ecological relevance, there is risk that experiments are performed under conditions chosen to suit theoretical predictions rather than to reflect the ecological context within which pyoverdin production makes sense. Solving this dilemma requires not only an improved understanding of the ecophysiology of pyoverdin, but also studies of the sociobiology of pyoverdin in environments that are the natural habitat of Pseudomonas (Dao et al. 1999). Although our study falls short of this need, demonstration of environmental and genetic sensitivity makes apparent the need to move future studies to environments beyond the lab.

Our study was initially motivated by curiosity as to the underlying genetic changes associated with the evolution of pyoverdin nonproducing types and their subsequent fate. For this work KB was chosen as the focal environment. King's Medium B is low in bioavailable iron and promotes production of pyoverdin (pyoverdin production is suppressed in KB upon addition of iron)—at least by cells at the air–liquid interface. We had not anticipated finding pyoverdin production to be maladaptive in this environment. To our knowledge only one other study has investigated spontaneously arising nonproducers in KB (Jiricny et al. 2010), however, in that study the fitness effects of the mutants were examined in a medium other than that in which the mutants evolved. Had the social context of nonproducer types been explored in KB it is likely that the authors would have reached conclusions analogous to those reported here.

Among the numerous departures from expectations under social evolution theory, the consistently reduced performance of PBR851, when rare, in unstructured, compared to structured, KB-containing microcosms (both with and without 2,2′-dipyridyl supplementation) warrants consideration (Fig. 5A). Although on one hand the result is against theoretical expectations, it points to an alternative explanation. From a theoretical perspective, secreted products are typically assumed to behave in a consistent manner across environments, but this need not be true and, indeed, for many reasons, this is unlikely to be true (Kümmerli et al. 2009a; Driscoll and Pepper 2010). One immediate consequence of manipulating environmental structure via microcosm disturbance—other than affecting the localization of pyoverdin—is to affect the availability of oxygen, with subsequent effects on the biological utility of pyoverdin. For example, in structured microcosms respiration rapidly depletes oxygen from all but the air–liquid interface, whereas unstructured cultures remain aerobic. Under anaerobic conditions iron exists in the reduced (soluble) ferrous (Fe2+) form and is freely available, whereas under oxygen replete conditions, iron exists in the oxidized (insoluble) ferric (Fe3+) state (Fig. S5). An additional factor, especially in unstructured microcosms, arises from the potentially harmful interplay between iron and oxygen, such that under aerobic conditions, iron, in the presence of reactive oxygen species (e.g., hydrogen peroxide), can lead to the formation of the hazardous hydroxyl radical OH• (Hassett et al. 1996; Dao et al. 1999). In fact it has been suggested (Dao et al. 1999)—and more recently shown (Vinckx et al. 2008)—that decreased production of pyoverdin and reduced uptake of ferripyoverdin is a mechanism for reducing the iron potentiation of oxygen toxicity. Data in Figures 1and 2 show that the production of pyoverdin in unstructured KB microcosms is unnecessary for growth (as is true in structured microcosms). Although this may be unproblematic in structured microcosms, excess pyoverdin could be harmful in unstructured microcosms due to the interplay with oxygen (Fig. 4B). Such a deleterious effect is likely to affect the performance of the mutant and producer cells alike when cocultured. Consistent with this hypothesis is the fact that the fitness advantage of the mutant increases in unstructured microcosms once KB is supplemented with 2,2′-dipyridyl (Fig. 5B), possibly because toxic effects arising from an oversupply of ferripyoverdin are reduced.

A previous report concerning the cost of pyoverdin production (Griffin et al. 2004) has been central to arguments that pyoverdin is a cooperative/altruistic trait. The experimental analysis performed here is identical, albeit with a different species of Pseudomonas; however, we found no evidence of a cost to pyoverdin production under iron-replete conditions. Nonetheless, in this study we genetically modified the ancestral cell causing it to constitutively produce pyoverdin. Not surprisingly, under iron replete conditions a net cost was observed. Interestingly, the cost was more pronounced in unstructured microcosms—even under conditions of extreme iron scarcity (Fig. 4A). This goes against the simple expectation that iron will be most scarce under aerated conditions and thus under these conditions, pyoverdin most useful. As shown (Fig. 4B), the effect is most likely attributable to the harmful interplay between oxygen and iron. That the deleterious effects were not shared equally with the wild type (the two genotypes were cocultured) reflects the fact that the overproducing genotype was disproportionately affected. This—along with data in Figures 5D, H, S4D, and S4H—points at the capacity of cells, under certain conditions, to personalize the effects of pyoverdin. It is possible that this is a consequence of overexpression of the ferripyoverdin (FpvA) uptake system, which is also activated by PvdS (Beare et al. 2003), and thus overexpressed in PBR852.

The marked genotype effect [PBR851 (pvdS229) compared with PBR840 (ΔpvdL)] points to further subtleties and complexities surrounding investigations into the social biology of extracellular products. It also stresses the importance of working with nonproducing mutants that arise spontaneously in the environment under investigation. Although PBR851 and PBR840 are defective in pyoverdin production and both proficient in uptake of extracellular pyoverdin, the underlying genetic bases are different. The spontaneous pvdS229 mutation resides in the major regulator of pyoverdin biosynthesis, thus leading to a cessation of expression of the entire biosynthetic pathway (Cunliffe et al. 1995). However, in the pvdL mutant, regulation of biosynthesis remains intact, thus once activated, the absence of a critical component of biosynthesis is likely to result in the buildup of pathway intermediates with harmful consequences. Such pleiotropic effects are likely to minimal (even absent) in PBR851. As such, the overall flat response of PBR840 (Figs. 5, S4) is likely to reflect deleterious pleiotropic effects of the pvdL mutation.

A further complicating factor stems from the positive feedback regulation of pyoverdin synthesis: pyoverdin, when complexed with ferric iron, stimulates its own synthesis via the FpvA–FpvR–PvdS regulatory cascade (Lamont et al. 2002). Although arguments for the biological relevance of such control have been advanced, positive feedback regulation means that in a closed planktonic system pyoverdin production stands to increase unchecked. This makes poor biological sense and raises the possibility that the closed nature of the laboratory microcosm means that pyoverdin is produced at levels that would never occur in the natural environment. The effect is likely exacerbated by planktonic conditions that facilitate rapid diffusion of pyoverdin (Julou et al. 2013). A related issue is the PvdRT–OpmQ system that functions to recycle pyoverdin from the periplasm back into the extracellular milieu (Imperi et al. 2009). A dedicated system for recycling suggests that in the natural environment pyoverdin is scarce (and valuable) such that after removal of iron from ferripyoverdin, pyoverdin is put back to work. Both the autoregulation of pyoverdin synthesis, and the existence of a mechanism for recycling, suggest that what is observed in liquid laboratory culture (with high cell densities and an excess of nutrients) may be far removed from the relevant and appropriate ecological context within which the production of pyoverdin and its regulation evolved or is maintained.

If true, then an alternative explanation for the evolution of nonproducing mutants in structured KB microcosms becomes apparent. Given that pyoverdin has various physiological effects and is both recyclable and subject to autoregulation, then in a closed planktonic system where there is opportunity for diffusion, the production of pyoverdin, even if initially restricted to just a small number of cells (e.g., those at the oxygen replete air–liquid interface), will cause cells elsewhere in the microcosm (e.g., in the microaerobic broth phase), irrespective of need, to activate biosynthesis of pyoverdin (Fig. S5). The production of pyoverdin by cells that do not require the siderophore incurs an unnecessary metabolic burden (Fig. S3)—there is also the possibility of costs associated with uptake of excess iron, particularly in the presence of oxygen. It is thus understandable that mutants with defects in pyoverdin synthesis are favored by selection. The real problem seems to stem from a regulatory mechanism that appears ill-suited to tune expression of pyoverdin in structured KB microcosms to the needs of individual cells. Various results from our experiments support this model: for example, the greater yield of pyoverdin defective mutants compared to pyoverdin producers in KB monoculture (Figs. 1 and 2) and the enhanced advantage to the mutant during the second day of growth at which point pyoverdin levels are high (Fig. 3), but particularly the final experiment, in which the fitness advantage to PBR851 is eliminated under microaerobic conditions that suppress pyoverdin production.

Pyoverdin defective mutants are known to arise from Pseudomonas aeruginosa in the lungs of cystic fibrosis (CF) patients (De Vos et al. 2001; Smith et al. 2006). Their evolution has been viewed almost exclusively in the context of the social evolution framework (West et al. 2006, 2007; Harrison 2007; Lee et al. 2012). However, it is possible that the causes of the evolutionary success of nonproducing mutants in KB might also be relevant in the CF lung. The CF lung is not iron deplete: iron is available in sputum samples (Stites et al. 1998, 1999; Reid et al. 2007) and at levels (>10 μM) that are sufficient to suppress Pvd-Fe transport in vitro (Meyer et al. 1997). In addition, much microbial growth occurs within damaged and scarred lung tissue where oxygen is low or absent (and thus iron exists in the available ferrous form; Worlitzsch et al. 2002; Yoon et al. 2002). Despite the availability of iron, pyoverdin is also present in the majority of CF sputum samples (Haas et al. 1991; Martin et al. 2011). This suggests the possibility that pyoverdin production in the CF lung, as in KB microcosms, exceeds that required by the cells for growth. Pyoverdin-defective mutants may therefore reap an advantage by not producing an unnecessary and costly product, just as in structured KB microcosms. An alternate possibility is that nonproducing mutants are favored in the lung precisely because of this defect—a defect that slows growth and thus increases the chances of long-term persistence in the lung (Smith et al. 2006; Yang et al. 2011). Given the current emphasis on use of the social evolutionary framework as the basis for new therapeutic approaches for controlling disease (Harrison 2007; Brown et al. 2009), it is important that the fit between theory and biological reality be robust.

Associate Editor: S. Remold


The authors are indebted to D. Greig, B. Callaghan, A. Eldar, and W. Driscoll for detailed and insightful comments on drafts of this manuscript. The authors thank M. Anderson for assistance with statistical analyses, Y. Liu and H. Chang for technical support, and members of the Rainey Lab for valued discussion. PBR is a James Cook Research Fellow: he acknowledges support from the Royal Society of New Zealand.