Increasing resource supply increased total bacterial population density, demonstrating that the manipulated resources were indeed limiting and non-toxic (data not shown; founding genotype, F_3,15 = 1.47, P = 0.2; log₂(resource supply), F_1,15 = 50.41, P < 0.0001). Crucially, biofilms formed under all treatments. To test whether selection for cooperation increased with increasing resource supply we calculated the proportion of biofilm-forming WS within each population (i.e., WS density/total density). Consistent with our predictions, the proportion of biofilm-forming WS increased with increasing resource supply (Figure 1; founding genotype, F_3,15 = 1.01, P = 0.4; log₂(resource supply), F_1,15 = 24.26, P < 0.0001). This suggests that selection to cease cooperative investment into biofilm production (i.e., bacteria inhabiting the broth and SM cheats within the biofilm) decreased with increasing resource supply.

Our verbal argument suggested that the mechanism underlying this pattern is decreasing physiological costs of cooperation with increasing resource supply. We tested this by growing WS and SM under conditions that do not allow biofilm formation such that the WS phenotype is purely costly (i.e., shaken tubes). In support of our predictions, the growth rate of WS relative to SM increased with increasing resource supply (Figure 2; WS genotype, F_3,15 = 5.29, P = 0.01; log₂(resource supply), F_1,15 = 51.94, P < 0.0001). This confirms that increasing resource supply did indeed reduce the cost of the cooperative WS phenotype.

Resource supply may also affect the benefits of cooperation: variable benefits of biofilm dwelling across the resource supply gradient could have altered the strength of selection upon SM for inhabiting the biofilm relative to selection for leaving the biofilm entirely. As density increased with increasing resource supply, it is likely that competition for oxygen within the microcosm also increased. This is likely to increase selection for inhabiting the biofilm. In support of this, the proportion of the SM population inhabiting the biofilm as cheats compared to asocial broth dwelling (i.e., biofilm SM density/total SM density) increased linearly with resource supply (data not shown; founding genotype, F_3,15 = 0.17, P = 0.9; log₂(resource supply), F_1,15 = 11.64, P = 0.004). However, opposing this, increasing resource supply reduced the physiological costs of cooperation (Figure 2), thus reducing the selective benefit of cheating relative to cooperating. These opposing selection pressures lead to the prediction of a unimodal relationship between resource supply and the proportion of cheating compared to cooperating within the biofilm. To test this we calculated the proportion of SM compared to WS within each biofilm (i.e., biofilm SM density/total biofilm density). In accordance with our verbal argument, the proportion of SM cheats peaked at intermediate resource supply (Figure 3; founding genotype, F_3,14 = 2.27, P = 0.125; linear log₂(resource supply), F_1,14 = 20.37, P < 0.0001; quadratic log₂(resource supply), F_1,14 = 12.15, P = 0.004).

As hypothesised, the fitness of cooperators relative to cheats increased with resource supply (Figure 4; F_1,10 = 18.74, P = 0.001). This confirms our prediction that selection for cooperation increases with increasing resource supply. It should be noted that the fitness of cooperators did not exceed that of cheats under any resource supply conditions. This is because our experiment involves local (i.e., within population) competition: previous studies with this system have shown that global (i.e., between populations) competition is required for cooperators to have a net fitness advantage over cheats 6. Because the media ingredients that were manipulated may have contained additional iron it is possible that increasing resource supply may have inadvertently increased iron availability if any additional iron could not be sufficiently chelated by apotransferrin. This may have resulted in down-regulation of siderophore production under high resource supply such that cooperators might have been fitter under high resources because they produce less public good. To test this, we grew pure cultures of cheats and cooperators along the resource gradient. Whereas pure cooperator and mixed cultures increased in density with increasing resource supply (F_1,10 = 15, 132, respectively; P < 0.01, in both cases) there was no effect of resource supply on cheat density (F_1,10 = 1.69, P > 0.2). This suggests that iron availability was not significantly altered across the resource supply gradient, and therefore that the cost of cooperation was indeed reduced through increased resource supply.

We have tested the effect on cooperation of the supply of resources that are not acquired directly through public-goods cooperation: a positive relationship was observed for two bacterial social traits. The crucial mechanism explaining the experimental results is simply that the cost of cooperation decreases with increasing resource supply. This arises because the benefit of investing resources into individual growth and reproduction is likely to show diminishing returns. Metabolic constraints (e.g., rate-yield trade-offs 25) would suggest that this assumption is frequently, if not always, correct.

Situations where public-goods cooperation is either not directly involved in resource acquisition, or supply of the limited resource targeted by social action is not strongly correlated with the supply of other essential resources, are likely to be common in nature. A relevant example is iron availability to bacterial pathogens within host organisms, the supply of which is often more restricted than that of other resources due to active sequestering by the host 26. However, the relationship between cooperation and resource supply is unlikely to be monotonic if the manipulated resource is itself acquired through public-goods cooperation. For example, if the impact of oxygen availability on biofilm formation was investigated, it is likely that biofilm cooperation would be reduced at high oxygen supply rates because biofilms would not be necessary to obtain oxygen. Thus the net effect of resource supply is likely to depend upon the covariance of different types of resource, both those targeted by the public-good and those that provide raw materials for its production.

In summary, we have shown that the success of cooperation is crucially dependent on resource supply rate. Our experiments suggest that cooperation increases with increasing resource supply rate, due to decreasing costs of cooperation. These results along with observations from termite societies, where cooperative behaviours were highest in colonies with abundant food 10, highlight that the role for resource supply in mediating the costs and benefits of kin-selected traits may apply very generally indeed.

Four replicate microcosms (30 mL glass universal containing 6 mL of King's B nutrient media) were inoculated with Pseudomonas fluorescens SBW25 to a total of approximately 10⁷ cells. These were statically incubated for 6 days at 28°C, after which time all populations were vortexed and an aliquot diluted and plated onto KB agar. A single wrinkly-spreader colony was then isolated from each population for further study and stored at -80°C in 20% glycerol.

Populations were initiated with 10⁷ cells of one of the isolated WS genotypes grown for 18 h under shaken conditions. These were then propagated under one of the following resource supply regimes: 0.125×, 0.25×, 0.5×, 1× and 2× standard KB, generated by serial dilution of KB medium into M-9 salt solution. These resource supply rates were selected following pilot studies to ascertain that, under all levels of resource supply, biofilm formation was possible and media was non-toxic. 6 uL of each culture was transferred to a fresh microcosm every 4 days over a 16-day period. After 16 days the broth phase of each population was sampled, then populations were homogenised and sampled. Samples were then plated onto agar and the frequencies of WS and SM colonies counted.

WS genotypes and the SM ancestor were grown in KB for 18 hours in shaken conditions to attain the same physiological state. A microcosm at each resource supply level was inoculated with 10⁷ cells of a 50:50 mixture of one of the WS genotypes and the SM ancestor. Mixtures were plated onto KB agar to determine WS and SM densities. Microcosms were then incubated for 24 hours under shaken conditions, and plated onto KB agar to determine WS and SM densities. The fitness of WS relative to SM was then calculated.

We inoculated approximately 10⁶ cells of overnight cultures of PA01, a pyoverdin-negative mutant, PA6609, or 1:1 mixtures of the two into wells of 96-well microtitre plates containing 150 ul of media. Four resource supply rates were used: M9 salts containing 2.5, 1.25, 0.625 & 0.3125 g of Casamino acids (sigma). All media were supplemented with 100 mg/ml human apo-transferrin and 20 mM NAHCO₃ to chelate iron. Three replicates were established for each strain-resource supply combination. Cultures were propagated for 72 h in a 37°C static incubator. At the start and end of the experiment, cultures were pleated onto KB agar and the frequency of cooperators (green colonies) and cheats (white colonies) elucidated, and relative fitness calculated. Final densities were estimated from optical densities (OD⁶⁰⁰).

All proportion data were arcsin-square-root transformed, density data were log₁₀ transformed and relative fitness data were cubed prior to analysis to conform with GLM assumptions of homogeneity of variance and normality of residuals. Data were then analysed in GLMs fitting founding genotype as a random factor and log₂ multiple of standard resource concentration as a linear (and in certain cases quadratic) covariate. Relative fitness (W) of cooperators was calculated from the ratio of the estimated Malthusian parameters (m) of the cooperators:cheats, m = ln (N_f/N₀), where N₀ is the starting density and N_f the final density. For the siderophore experiment resource supply was fitted as a covariate in a GLM.