Privatisation rescues function following loss of cooperation

A single cheating mutant can lead to the invasion and eventual eradication of cooperation from a population. Consequently, cheat invasion is often considered equal to extinction in empirical and theoretical studies of cooperator-cheat dynamics. But does cheat invasion necessarily equate extinction in nature? By following the social dynamics of iron metabolism in Pseudomonas aeruginosa during cystic fibrosis lung infection, we observed that individuals evolved to replace cooperation with a ‘private’ behaviour. Phenotypic assays showed that cooperative iron acquisition frequently was upregulated early in infection, which, however, increased the risk of cheat invasion. With whole-genome sequencing we showed that if, and only if, cooperative iron acquisition is lost from the population, a private system was upregulated. The benefit of upregulation depended on iron availability. These findings highlight the importance of social dynamics of natural populations and emphasizes the potential impact of past social interaction on the evolution of private traits.


44
Identifying mechanisms responsible for the maintenance of cooperation is a major achievement of 45 evolutionary biology (Axelrod & Hamilton, 1981;Hamilton, 1963). We can predict in which 46 conditions cooperation will thrive, and where it might pay to exploit cooperative neighbours. 47 Evidence of tension between cooperative and cheating strategies are all around us in nature -for 48 example, the co-evolution between flowers and their pollinators , the policing and counter-policing 49 unobserved and unreported (Sachs & Simms, 2006). This is for the obvious reason that it is 53 inherently difficult to detect a history of something that no longer exists. And also because it is 54 generally the case that once cheating has gone to fixation, it is almost impossible for cooperation to 55 re-invade (Axelrod & Hamilton, 1981). The cheats won. End of story. 56 57 Life, however, goes on. Consider a population where cheating has gone to fixation. If cooperation 58 fulfilled an important function, this population may now be at risk of extinction (Fiegna & Velicer, 59 2003). To escape this fate, selection may favour individuals that can restore function by employing 60 a "privatisation" strategy -replacement of a mechanism that was once performed cooperatively as a 61 group, with a selfish one that only benefits the actor (Bel, 2006). As such, we use the term 62 privatisation as it is normally understood in common language to describe a switch in strategy from 63 one that helps a whole group to achieve a goal to one where a goal is achieved by the actor alone. interaction. It may also have been missed for the practical reason that we are required to track 70 behaviour over many generations post-cheat invasion in a natural population. We overcome these 71 difficulties by studying the evolutionary dynamics of a cooperative trait for more than an estimated 72 20 million generations (Table S1), in a population of bacterial cells. We report the first observation 73 (to our knowledge) of a natural population responding to cheat invasion by adopting a privatisation 74 strategy and, therefore, avoiding the possibility of extinction. 75 76 Our study population is comprised of Pseudomonas aeruginosa bacteria causing lung infection in 77 patients with cystic fibrosis (CF). CF is a genetic disease that causes the build-up of dehydrated 78 Here, we identified changes in iron uptake strategies from genome sequences and correlated these 114 A key step in our analysis is to categorise isolates with respect to their social 123 environment, for example, presence vs absence of cooperators (pyoverdine producers). This is 124 necessary for testing our hypotheses that iron acquisition strategy is influenced by the social 125 environment. While it is likely that we do not capture all co-infecting strains in our sample, the 126 effect of mis-classifying isolates will homogenise the sample groups and hence obscure rather than 127 amplify any differences, reducing our ability to detect an effect. 128 129 We found that pyoverdine production increased in 33% of clonal lineages within the first two years 155 of infection. Overall, these "super-cooperator" isolates, which had > 30% higher production than 156 isolates initiating infection, were detected in 24 of 45 independent clonal lineages (53%; Fig. 1 with super-cooperators we identified candidate mutations in global regulatory and quorum sensing 165 genes that likely cause increased production of pyoverdine as well as other virulence factors, such 166 as pyochelin, pyocyanin and alginate (Table S2). For the remainder, super-cooperators produced 167 pyoverdine at a consistently higher rate than their ancestors but the genetic basis of upregulation 168 was unclear. 169

170
Despite the potential benefits of pyoverdine production, the presence of super-cooperators appears 171 to weaken long-term stability of cooperation in the lung. In 12 out of 35 patients, we previously 172 observed the appearance of mutants that have lost the ability to synthesise pyoverdine (Andersen et  173 al., 2015), sampled on average 2.97 years after infection. Here, we found that the likelihood of these 174 non-producers arising was significantly higher in the presence of super-cooperators [c 2 (2, N = 47) 175 = 5.7; p < 0.05; fig. 1]. Non-producers were present in four out of 21 clonal lineages without (19%) 176 and in eight out of 24 lineages with super-cooperators (33%). 177 178 Why is upregulation of pyoverdine production associated with the subsequent loss of this trait? 179 Non-producers appear to be favoured as a result of exploiting the pyoverdine production of 180 neighbouring cells -a strategy referred to as cheating (Ghoul et al., 2013). We have previously 181 shown that non-producers in the lung acquire mutations in a pattern consistent with a cheating 182 strategy, as opposed to one expected from redundancy, in that they retained the pyoverdine receptor 183 only if producers were present (Andersen et al., 2015). In other words, non-producers lose the 184 capacity to contribute but retain the ability to benefit from pyoverdine, as long as it is being 185 supplied by cooperative neighbours. The relative fitness of cheats is predicted by theory to be 186 greater when there are higher levels of cooperation in a population, because their competitors are 187 showed that loss of pyoverdine production is the first change in iron metabolism observed in most 212 clonal lineages. When weighed by the high mutation rate of the system, reflecting a strong selection 213 pressure, and its large size, this is not unexpected [P(X ≥ 16) ∼ pois (X; 14.69) > 0.05), fig. 2, table 214 S3, SI text]. We observe cheats appear 14 times without going to fixation during the sampling 215 period, but further record nine cases where no producers are sampled a year prior to, or after 216 emergence a non-producer. One of these is in a transmitted clone type that is subsequently sampled 217 from 16 patients over 35 years. This is the "game over" scenario of cheat invasion. However, 218 because we have continued to sample after this event, we can observe how the population responds 219 to this potentially catastrophic situation. And because cheat invasion is not inevitable, we can 220 compare dynamics in lungs with and without availability of pyoverdine. Super-cooperators present

Super-cooperators absent
After cooperation was lost, and only if it was lost, we observed that iron uptake was privatised (Fig.  226 3). This was achieved by upregulation of the private phu system. The phu system targets the iron-227 rich heme molecule, which is taken up directly through a membrane-bound receptor (Table 1) 228 (Ochsner et al., 2000). Increased expression of the phu system results from intergenic mutations 229 between the phuS gene encoding a heme traffic protein, and the receptor gene phuR (phuS//phuR 230 mutations; . In the isolate collection, a total of 29 SNPs and two indels 231 accumulate in the phuS//phuR region. Eight of these mutations in the phuR promoter have been 232 found to cause a significant upregulation of promoter activity, and one of these has further been 233 shown, in an isogenic pair, to provide a growth benefit to the carrier when heme is available as an 234 iron source ) (see also below and fig. S1). These specific mutations were found 235 to occur significantly more frequently than expected by chance following the loss of pyoverdine 236 production [P(X ≥ 5) ∼ pois (X; 0.39) < 0.01); fig. 2]. In three patients, non-producing isolates with 237 phuS//phuR mutations were found to co-occur with pyoverdine producers, however in two of these 238 the producer and non-producers were unlikely to interact (SI text).
phu-has-phu-has-pvd- where iron concentrations fluctuate. We, therefore, tested whether the growth benefits of phu 302 upregulation (by phuS//phuR mutations) were dependent on iron availability, in the form of heme. 303 We examined the growth difference between a clinical isolate that had lost pyoverdine production, 304 and an isolate, isogenic but for a clinical phuS//phuR mutation that cause an upregulation of phuR 305  upregulated phu system achieved a higher density (Tukey HSD, padj < 0.01, fig. 5), consistent with 314 previous findings . In contrast, at 5 and 10 µM the isolate without upregulation 315 had an advantage (padj < 0.05, Fig. 5). No growth was observed in the absence of heme, and at 1 µM 316 there was no significant difference in growth between the isolates (padj = 0.073). This suggests that 317 phuS//phuR mutations lead to increased heme uptake that is beneficial at low external 318 concentrations (>1 and < 5 µM heme) but detrimental at high concentrations (> 5 µM heme). If iron 319 availability increases during infection, an initially beneficial upregulation of heme uptake would 320 turn toxic. In the phu system we found a significant bias towards mutations of the phu receptor gene   The sampling regime for each patient is described in Table  was determined as that of the first isolate(s) of a clone type from a patient, and the production of all 636 subsequent isolates of this clone type was compared to this. An isolate was classified as an over-637 producer if its production was > 30% higher than that of the first. This cut-off was chosen as ~95% 638 of the isolates had a lower standard deviation between replicates. Loss of pyoverdine production 639 was defined as sampling of a non-producing isolate from a clonal lineage that produced pyoverdine 640 at the beginning of the infection period. The analysis was performed on longitudinally sampled 641 clone types from the children's collection only, as the DK1 and DK2 isolates were not sampled 642 frequently within patients. In all but three cases, over-producers were sampled before non-643 producers. In two cases there was < 5 months between sampling, and we scored non-producers as 644 occurring at the same time as overproducers. In one case the non-producer was sampled 2.95 years 645 prior to the over-producer and scored as having occurred in the absence of an over-producer. In two 646 clone types two independent transitions to non-producers were observed and these were included as 647 independent events. 648

649
We categorized each clone type in individual patients as harbouring over-producers or not. If an 650 over-producer was present, the length of time to a non-producer was observed, or till the last sample 651 was collected from the patient, was noted. Clone types without non-producers were classified as 652 right-censored. The same was done for patients without over-producers. In total, 45 clonal lineages 653 were followed (some clone types occur in multiple patients, and some patients are infected by 654 multiple clone types (Marvig et al., 2015)). Clone types DK1, DK11, DK40 and DK53 were 655 excluded from the analysis, as they were "chronic" clone types with no pyoverdine producers intergenic phuS//phuR mutations predicted to cause receptor upregulation were included, and when 678 occurring alone these were recorded separately. We calculated the expected order of mutation, 679 equally weighted by the size and number of mutations of the different systems. We tested, as 680 described above, whether this was different from the observed, using only cases where there is a 681 clear order of mutations (Fig. S1). To test if the appearance of intergenic phuS//phuR mutations was correlated with loss of pyoverdine 695 production we categorized each independent acquisition of these mutations as having occurred in 696 the presence or absence of measured pyoverdine production. In the absence of pyoverdine 697 production the time between loss of production and occurrence of mutations was estimated. The 698 analysis was performed on longitudinally sampled clonal lineages from both collections of isolates, 699 that is lineages with at least two isolates with or without pyoverdine production. Further, three cases 700 where the loss of pyoverdine production and phuS//phuR mutations were observed in the same 701 isolate were also included. Clonal lineages that did not acquire phuS//phuR mutations were 702 classified as right-censored. For DK1 and DK2 we used monophyletic clades with and without 703 mutations as independent clonal lineages. We included isolates of DK53 where no WT pyoverdine 704 producer was sampled, but phuS//phuR mutations were inferred by comparison to PA01 (Table S5). 705 The timing of sampling of mutants was analysed with the "survival" package in R (Therneau,  (Table S5). We compared fitness 789 of the mutants with their closest genetically related isolate without phuS//phuR mutations, to control 790 for phylogenetic differences. The isolates in a pair were, however, separated by multiple other 791 mutations as they were sampled up to 26 years apart. The pairs were grown in iron-limited media 792 with 2.5 and 5 µM heme added, which corresponds to medium concentrations observed in CF 793 patients (Ghio et al., 2013). Three mutants with intergenic SNPs had significantly higher (P73F1; 794 26% higher; t = -6.33, df = 3.48, p > 0.01 and P83M2; 16% higher, t = -4.48, df = 2.15, p < 0.05; 795 isolates sampled 0 and 23 years apart) or a tendency towards a higher maximum OD600 in 5 µM 796 heme compared to their closest relative (P66F0; 17% higher, t = -2.85, df = 2.67, p = 0.07; isolates 797 sampled 10 years apart; fig. S2). A mutant with a three bp insertion, and an isolate with a one base 798 pair deletion in phuS//phuR in addition to a SNP, showed no significant difference in growth 799 compared to their closest relative (17% higher, t = -1.78, df = 2.95, p = 0.   Table S2: Mutations identified in pyoverdine over-producers that may contribute to this phenotype. 846 For each mutation, the patient ID, clone type and gene is listed, as well as a reference to previous 847 work showing a link between the gene and pyoverdine production.     Table S7: Summary of the mutations found in five different iron uptake systems in P. aeruginosa. 902 The size of the respective operons is given in kb, and mutations are listed as non-synonymous (ns) 903 SNPs, synonymous (syn) SNPs, insertions and deletions (indels) and intergenic (IG). The overall 904 mutations, and ns SNPS and indels, per kb are listed. The phu operon experienced the highest 905 mutation rate, dominated by intergenic mutations. Two genes in the has operon were poorly 906 sequenced in the majority of isolates, and this is taken into account in the size and mutations listed 907