Evolutionary dynamics of interlinked public goods traits: an experimental study of siderophore production in Pseudomonas aeruginosa

Introduction

Bacteria frequently cooperate by sharing secreted secondary metabolites (West et al., 2007). These shared products, though costly for the individual to produce, can benefit other individuals, or the bacterial collective in general, and in such cases, they constitute public goods (West et al., 2007). Bacterial public goods include structural materials for building biofilms (Nadell et al., 2009), signalling molecules for communication (Williams, 2007), enzymes to digest food (Diggle et al., 2007), biosurfactants for group motility (Kearns, 2010), toxins to fight competitors (Jousset, 2012), and chelating agents to scavenge essential metals (Griffin et al., 2004). Such cooperation can generate enormous group-level benefits, yet it is also famously vulnerable to being undermined by ‘cheat’ variants that contribute little or nothing to the collective stock of public goods and thereby escape their fair share of costs, while still benefitting from the cooperative efforts of others (West et al., 2006). By exploiting cooperators, cheats can increase in frequency – even if this ultimately harms the collective as a whole. However, as numerous studies have revealed, this ‘social dilemma’ can be modulated by a range of mechanisms that constrain cheat fitness and thereby maintain cooperation (e.g. Ross-Gillespie et al., 2009; Brockhurst et al., 2010; Kümmerli & Brown, 2010; Xavier et al., 2011; Dandekar et al., 2012; Drescher et al., 2014). For example, limited dispersal and viscous environments can both limit cheats' access to cooperators and their precious public goods (Kümmerli et al., 2009a; Julou et al., 2013). Also, cheats can lose their relative advantage as they become more common (negative frequency-dependent payoffs; Ross-Gillespie et al., 2007; Jousset et al., 2009; Raymond et al., 2012).

Although this body of work has greatly aided our understanding of the dynamics of public goods traits, the scenarios investigated in these studies remain, by and large, simplified approximations of what really goes on in natural microbial communities. In particular, most studies consider just one model trait at a time. Under natural conditions, however, bacteria have to juggle a portfolio of various different public goods, and hence, they could routinely find themselves simultaneously participating in multiple public good dilemmas on multiple fronts (Brown & Taylor, 2010; Mellbye & Schuster, 2014). Another important complication is that the production of different public goods is often linked, positively or negatively, at the regulatory level (Nadal Jimenez et al., 2012). Consequently, selection for or against one public good might have pleiotropic consequences for other public goods (Sandoz et al., 2007; Harrison & Buckling, 2009; Driscoll et al., 2011; Inglis et al., 2012; Friman et al., 2013; Jousset et al., 2013). Ultimately, if we want to understand the evolutionary dynamics of social traits in complex natural microbial communities, we will need to consider also the joint effects of superimposed public goods dilemmas and the nature of the links between them (Brown & Taylor, 2010).

Here, we take on this issue. We start by developing general predictions for any two-trait case, where traits, that may or may not be public goods, are linked positively, negatively or not at all (Fig. 1a). Next, we test a subset of our predictions in a specific empirical test case, tracking expression of two regulatorily linked public goods in the bacterium Pseudomonas aeruginosa during experimental evolution. P. aeruginosa, an opportunistic human pathogen, produces and secretes two siderophores to scavenge iron. The first, pyoverdine, is a more effective iron chelator but is metabolically more costly to produce, whereas the second, pyochelin, is less effective but cheaper (Cornelis, 2010). Importantly, both pyoverdine and pyochelin are freely diffusing exoproducts, and as we show below, both can function as public goods under suitable conditions. The relative investment into one or the other siderophore is dependent on relative iron availability: bacteria up-regulate pyoverdine and repress pyochelin when iron limitation is strong, but as it weakens, they reduce pyoverdine production and switch instead to pyochelin (Dumas et al., 2013). Mechanistically, this switch is mediated by (i) direct (albeit incomplete) suppression of pyochelin synthesis by pyoverdine and (ii) a strong negative feedback (operating via the ferric uptake regulator, Fur) that shuts down pyoverdine synthesis whenever iron is more readily available (Dumas et al., 2013).

Given these biological details, we can derive qualitative predictions for the evolutionary trajectories of pyoverdine and pyochelin production under different types of iron stress (Fig. 1b). Under strong iron limitation, where P. aeruginosa cells produce mainly pyoverdine, we expect cheats, defective for this siderophore to emerge and spread (Fig. 1b, upper right block). As a consequence of their lowered pyoverdine production, we then expect the pleiotropic derepression (i.e. increase) of pyochelin production in these mutants – turning pyoverdine cheats into pyochelin cooperators. Thus exposed to selection, pyochelin too should then decline, as pyochelin-defective cheats now also should emerge and spread. Double mutants, defective for production of either siderophore but still able to exploit both, could have the greatest relative advantage. Under more moderate iron limitation, meanwhile, where bacteria mainly produce pyochelin, we expect mutants defective for pyochelin production to emerge and to drag down the aggregate expression of pyochelin (Fig. 1b, lower right block). Unlike the scenario with strong iron limitation, however, we do not expect compensatory up-regulation of the other siderophore, because the regulatory link is strictly unidirectional – pyoverdine influences pyochelin production, but not the other way round (Dumas et al., 2013). Finally, in a fluctuating environment where iron is alternately strongly then moderately limited, we would expect selection for both pyoverdine and pyochelin cheats to be weaker, leading to slower and/or less pronounced shifts in the siderophore profiles of the population. This is because mutants defective for one or the other siderophore should have a selective advantage only part of the time. We tested these predictions by tracking siderophore production profiles in replicate populations of the P. aeruginosa PAO1 wildtype strain over approximately 300 generations (48 serial transfers) under strong, moderate, and alternating iron-limitation regimes.

Materials and methods

Results

Pyoverdine and pyochelin both function as public goods under appropriate conditions

Our competition assays confirmed that both of P. aeruginosa's siderophores function as beneficial public goods that can be exploited by cheats in a strongly iron-limited environment. Under more moderate iron stress, however, only pyochelin shows the characteristics of a public goods trait, whereas pyoverdine seems not to be required in this environment (Fig. 2 summarizes the competitive performance of nonproducers against producers under different levels of iron limitation; Fig. S1a contrasts the growth performance of all strains when grown as monocultures vs. in mixtures, and Fig. S1b compares per-producer siderophore production effort under monocultures vs. mixed cultures). In coculture under strongly iron-limited conditions, the siderophore nonproducer PAO1ΔpvdDΔpchEF significantly outperformed the pyoverdine-only-producer PAO1ΔpchEF, but also outperformed a pyochelin-only-producer PAO1ΔpvdD, indicating that cheats could invade either phenotype under these conditions (linear models, LMs, with H₀ that relative fitness = 1: t₄₀ = 13.83, P < 0.001 and t₄₀ = 9.60, P < 0.001, respectively). Under moderate iron limitation, meanwhile, the nonproducer outperformed only the pyochelin producer (t₄₀ = 3.27, P = 0.002) but not the pyoverdine producer (t₄₀ = −0.45, P = 0.657), indicating that only pyochelin is exploitable in this environment. Comparison of relative fitness values calculated using an alternative approach based on ratios of doublings (as per Lenski et al., 1991) yielded qualitatively similar results (Repeat of LM tests as ordered above: t₄₀ = 13.55, P < 0.001; t₄₀ = 12.36, P < 0.001; t₄₀ = 3.81, P < 0.001; t₄₀ = −0.90, P = 0.373).

Changes in siderophore production during experimental evolution

Isolates from lines evolved under strong iron limitation and assayed under the same conditions showed a decrease in per capita pyoverdine investment over time (Fig. 3a; LM: slope = −460.4, t₃₈ = −3.047, P = 0.004). Concomitantly, we saw an increase in pyochelin (LM: slope = 26.584, t₃₈ = 2.922, P = 0.005). In lines evolved under moderate iron limitation, pyoverdine investment also significant declined over time (LM: slope = −95.36, t₃₈ = −2.733, P = 0.009), although, given that the ancestor already showed low expression under these conditions, the absolute magnitude of the shift was relatively minor in this case. Pyochelin, meanwhile, showed a marked decrease under these conditions (LM: slope = −15.846, t₃₈ = −5.671, P < 0.001). Lines evolved under alternating iron limitation showed the same overall trends, but for the most part, the slopes here were nonsignificantly different from zero – that is towards less pyoverdine (slope = −274.8, t₃₈ = −1.779, P = 0.083) and more pyochelin (slope = 12.397, t₃₈ = 1.493, P = 0.144) under strong iron limitation, and less pyochelin (slope = −7.194, t₃₈ = −1.316, P = 0.196) and less pyoverdine (slope = −135.63, t₃₈ = −4.501, P < 0.001) under moderate iron limitation.

Assaying at a whole-population level (as opposed to averaging across a random sample of isolates, as above) yielded qualitatively similar patterns overall (Fig. 3a, dotted lines) with one notable difference: under either strong or alternating iron limitation, clonal sampling suggested a roughly linear decrease in pyoverdine over time, whereas at a population level, pyoverdine investment first increased before later declining (anova comparison of linear vs. negative quadratic fits: F_1,37 = 5.286, P = 0.027; F_2,36 = 7.662, P = 0.008 for the strong and alternating Fe-limitation regimes, respectively). No such difference was observed for pyochelin production patterns, in any of the conditions tested.

Tracking temporal changes in siderophore production as a ‘walk’ in 2-dimensional phenotype space (Fig. 3b), we saw patterns broadly consistent with those predicted (Fig. 1). Under strong iron limitation, populations shifted from high-pyoverdine/low-pyochelin production towards low-pyoverdine/high-pyochelin production, although we did not see the final phase of the predicted trajectory (i.e towards low-pyoverdine/low-pyochelin production). Under moderate iron limitation, we observed, as predicted, a shift from high- to low-pyochelin production. Finally, under alternating iron limitations, we observed relatively smaller mean shifts in siderophore production profiles.

Inspection of individual isolates' phenotypes (Fig. 4) revealed relatively few ‘extreme’ phenotypes – that is those showing ‘maximal’ investment in one siderophore and complete suppression of the other. Moreover, no isolates were observed that were fully defective for both siderophores. Rather, the vast majority of isolates showed only partially reduced siderophore investment and thus continued to express both siderophores, although typically with altered relative investment in one vs. the other. As compared to lines selected under moderate iron limitation, those subjected to strong iron limitation (Fig. 4, left column) appear to have diversified into a broader set of phenotypes, with some degree of clustering.

Discussion

In this study, we have developed a conceptual framework to predict how the regulatory links between different social traits could influence their evolutionary trajectories in microbial populations. To test our ideas, we have investigated the evolutionary dynamics of two of P. aeruginosa's public goods traits – the iron-scavenging molecules pyoverdine and pyochelin – in situations where the traits are either hierarchically linked (strong iron limitation, where pyoverdine suppresses pyochelin) or unlinked (moderate iron limitation, where pyochelin is not suppressed). Under the latter scenario, we observed selection against pyochelin (Fig. 3, middle column), which is consistent with our expectation that pyochelin producers would be exploited by cheats under these conditions (Fig. 2). However, we also saw some (weak) selection against pyoverdine. This pattern, we posit, has arisen because pyoverdine provides negligible benefit under these conditions, yet still imposes net costs owing to maintenance of the molecular machinery and baseline expression (Zhang & Rainey, 2013; Kümmerli & Ross-Gillespie, 2014). Under strong iron limitation, where regulatory links between the siderophores are in effect, the evolutionary dynamics were more complex. Initially, cells produced predominantly pyoverdine, such that only pyoverdine was exposed to selection, whereas pyochelin was shielded from selection by the negative regulatory link (Fig. 3, left column). In time, mutants defective for pyoverdine production arose and spread (Fig. 4), consistent with our understanding that pyoverdine is an exploitable public good under strong iron limitation (Fig. 2). Selection against pyoverdine in turn weakened its suppression of pyochelin, causing pyoverdine-negative mutants to begin producing pyochelin. Accordingly, we saw that populations diversified into several distinct phenotypes, producing less pyoverdine and, in many cases, more pyochelin (Fig. 4). In our third selection regime – that of alternating exposure to strong and moderate iron limitation – the phenotypic slide towards lower siderophore production was qualitatively similar to, although less pronounced than, that seen under constant selection (Fig. 3). This result is consistent with the view that a fluctuating environment can, in moderation, reduce selection for cheats (Brockhurst et al., 2007) and instead favour generalists (i.e. an individual-level solution; e.g. Kassen, 2002; Dumas et al., 2013) or stabilize phenotypic polymorphism (i.e. a lineage- or population-level solution; Kassen, 2002; Venail et al., 2011).

Although the population-level evolutionary dynamics largely followed the predicted patterns (Fig. 1), the level of interclonal variation in siderophore phenotypes that emerged under strong iron limitation was somewhat surprising (Fig. 4). For instance, why did pyoverdine-negative mutants not spread to fixation, and why did we not see the rise of ‘double cheat’ pyoverdine- and pyochelin-defective mutants as we had expected? After all, defined mutants with these phenotypes have previously been shown to outcompete the wildtype across a range of conditions (Ross-Gillespie et al., 2007; Kümmerli et al., 2009b; Jiricny et al., 2010). We can posit a number of possible explanations, and they are neither exhaustive nor mutually exclusive. First, it may be that some of the isolates we assayed indeed carried mutations reducing expression of both siderophores, yet such mutants showed little net change in pyochelin production because their reduced production potential (owing to mutations in pyochelin genes) was masked by a concomitant boost in production effort [because mutations reducing pyoverdine derepress pyochelin; (Dumas et al., 2013)]. Second, it may be that we simply did not continue the experimental evolution for long enough to see pyoverdine cheats fix and/or double mutants rise and spread to detectable frequencies. This seems plausible in the light of a recent study which demonstrated that when bacteria are forced to adapt simultaneously to changes in both their abiotic and social environment, the rise and spread of cheats is relatively slower (Morgan et al., 2012). In other words, what we see in our evolved lines could represent a transient situation, rather than some sort of stable equilibrium. Third, it may be that populations have indeed reached equilibria whereby multiple siderophore strategies can stably coexist. Such coexistence could be facilitated by negative frequency-dependent selection, whereby the fitness of cheats decreases as they become more common, and eventually drops below that of the wildtype at some threshold frequency. Such dynamics have been previously inferred from short-term invasion experiments using this same model system (Ross-Gillespie et al., 2007) and are known to occur with other microbial social traits too (Kerr et al., 2002; Jousset et al., 2009; Barrett et al., 2011; Raymond et al., 2012). Finally, coexistence of different social strategies could also potentially come about through mutualistic interactions between community members (MacLean et al., 2010; Driscoll et al., 2011). This scenario would involve a division of labour whereby strains reciprocally swap different public goods at the population level. Although mutualism could – in principal at least – evolve in the long run, for it to be stable, it would need to be accompanied by the coevolution of mechanism(s) to repress competition and keep cheats in check (Frank, 2003).

Whether stable or transient, it is nonetheless interesting to reflect on why we see greater phenotypic diversity in lines selected under strong iron limitation as compared to those selected under moderate iron limitation. First, let's consider the relative strength of selection. Moderate iron limitation imposes weaker selection and permits the persistence of larger populations, which can shelter a diversity of marginally maladapted phenotypes – variants that can serve as a backdrop for further mutations. Consequently, under this regime, we might expect evolution to ultimately arrive at sophisticated phenotypes that are close to theoretical optima (Weissman et al., 2009; de Visser & Krug, 2014). Under strong iron limitation, meanwhile, conditions are harsher, population size is lower, and fewer variants can survive concurrently. Consequently, evolution could proceed more erratically and may be more likely to converge on proximate, rather than ultimate, fitness optima (de Visser & Krug, 2014). Such a scenario could generate discrete clusters, rather than a cloud of phenotypes. Second, differences in the genetic architecture underlying the two traits could potentially impose different constraints on the evolutionary routes available to each trait (Koonin & Wolf, 2010) and thus could lead to different patterns of diversification. However, we consider this unlikely to matter much in the present case, because the genetic architecture underlying pyoverdine and pyochelin is not strikingly different (Visca et al., 2007; Youard et al., 2011): in both cases, synthesis is effected by nonribosomal peptide synthetases and is controlled by a single major regulator (PvdS, in the case of pyoverdine, and PchR, in the case of pyochelin). Moreover, targeted mutagenesis of these regulators often leads to incremental reduction in siderophore investment (Michel et al., 2005; Wilson & Lamont, 2006). Most of the altered phenotypes we observe are incremental and therefore probably caused by changes in these two functionally analogous genes.

Taken together, our results support the view that pleiotropy caused by regulatory cross-links between traits can be an important modulator of the evolutionary dynamics of cooperative traits in microbes. Indeed, some types of regulatory links may have explicitly evolved because of their stabilizing effects on cooperation (Foster et al., 2004). As an example, let's consider quorum sensing (QS) (Williams et al., 2007), a form of cell-to-cell communication that facilitates synchronized expression of costly cooperative traits only when it is most beneficial to do so – at high cell density (Darch et al., 2012; Ross-Gillespie & Kümmerli, 2014). Cheating mutants – that do not respond to other cells' QS signals yet still benefit from the QS-controlled public goods these others produce – have repeatedly been observed to arise and spread (Sandoz et al., 2007; Rumbaugh et al., 2009; Wilder et al., 2011). However, as QS coregulates many traits (i.e. an example of positive hierarchical regulation – see Fig. 1a), it may be that in some settings, cheats that would otherwise benefit by avoiding investment in one trait are kept in check by costs arising from the pleiotropic down-regulation of another trait. Such pleiotropic costs can include, for example, increased vulnerability to protist grazers (Jousset et al., 2009) or toxins (Jousset et al., 2013), or reduced metabolic capability (Dandekar et al., 2012). In our system, the negative regulatory linkage between pyoverdine and pyochelin fine-tunes the siderophore profile to prevailing environmental conditions (Dumas et al., 2013) and has probably evolved primarily for this reason. However, here too linkage may help to stabilize cooperation: because pyoverdine-defective cheats pleiotropically become pyochelin-producing cooperators, their relative advantage is reduced and their spread may thus be hindered.