1. Evolutionary Biology
Download icon

Privatisation rescues function following loss of cooperation

  1. Sandra Breum Andersen  Is a corresponding author
  2. Melanie Ghoul
  3. Rasmus L Marvig
  4. Zhuo-Bin Lee
  5. Søren Molin
  6. Helle Krogh Johansen
  7. Ashleigh S Griffin  Is a corresponding author
  1. University of Oxford, United Kingdom
  2. Technical University of Denmark, Denmark
  3. Rigshospitalet, Denmark
  4. University of Copenhagen, Denmark
Research Article
  • Cited 5
  • Views 741
  • Annotations
Cite this article as: eLife 2018;7:e38594 doi: 10.7554/eLife.38594

Abstract

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.

https://doi.org/10.7554/eLife.38594.001

Introduction

Identifying mechanisms responsible for the maintenance of cooperation is a major achievement of evolutionary biology (Axelrod and Hamilton, 1981; Hamilton, 1963). We can predict in which conditions cooperation will thrive, and where it might pay to exploit cooperative neighbours. Evidence of tension between cooperative and cheating strategies are all around us in nature – for example, in the co-evolution between flowers and their pollinators (Jandér and Herre, 2010), the policing and counter-policing behaviours of social insects (Foster and Ratnieks, 2000) – and in human society (Mathew and Boyd, 2011; Rand et al., 2011). Cases where these dynamics have resulted in the loss of cooperation are much less well understood, primarily because long-term consequences of cheat invasions are unobserved and unreported (Sachs and Simms, 2006). This is for the obvious reason that it is inherently difficult to detect a history of something that no longer exists. And also because it is generally the case that once cheating has gone to fixation, it is almost impossible for cooperation to re-invade (Axelrod and Hamilton, 1981). This is where most studies of social dynamics end.

Individuals are nevertheless under selection to survive the loss of cooperation. Consider a population where cheating has gone to fixation. If cooperation fulfilled an important function, this population may now be at risk of extinction (Fiegna and Velicer, 2003). To escape this fate, selection may favour individuals that can restore function by employing a ‘privatisation’ strategy – replacement of a mechanism that was once performed cooperatively as a group, with a selfish one that only benefits the actor (Bel, 2006). As such, we use the term privatisation as it is normally understood in common language to describe a switch in strategy from one that helps a whole group to achieve a goal to one where a goal is achieved by the actor alone. This differs from its use for describing the acquisition of property for future benefit by a group or an individual (Strassmann and Queller, 2014), the retention of public goods for individual use (Asfahl and Schuster, 2017), and for when public goods only are useable to a part of a community (Niehus et al., 2017). Privatisation could be a common occurrence in nature that has nevertheless been overlooked because there is no reason, a priori, to interpret private function as a result of past social interaction. It may also have been missed for the practical reason that we are required to track behaviour over many generations post-cheat invasion in a natural population. We overcome these difficulties by studying the evolutionary dynamics of a cooperative trait for more than an estimated 1.3 million generations (Supplementary file 1A), in a population of bacterial cells. We report the first observation (to our knowledge) of a natural population responding to cheat invasion by adopting a privatisation strategy and, therefore, avoiding the possibility of extinction.

Our study population is comprised of Pseudomonas aeruginosa bacteria causing lung infection in patients with cystic fibrosis (CF). CF is a genetic disease that causes the build-up of dehydrated mucus in lungs and sinuses, which P. aeruginosa infects (Folkesson et al., 2012). The host actively withholds iron to limit infection, and, as a counter-measure, P. aeruginosa employ a range of different iron acquisition strategies (Poole and McKay, 2003). The primary mechanism relies on secretion of the siderophore pyoverdine (Haas et al., 1991; Konings et al., 2013). This has been demonstrated to be a cooperative trait, where iron-bound pyoverdine molecules are available for uptake not only by the producer, but also by neighbouring cells (Buckling et al., 2007; West and Buckling, 2003). P. aeruginosa also make a secondary siderophore, pyochelin, which is cheaper to produce but has a lower affinity for iron (Dumas et al., 2013). In contrast, the Pseudomonas heme uptake system (phu) is private, as the iron-rich compound heme is taken up directly without the secretion of exploitable exoproducts (Table 1). Additional mechanisms of uptake include the private ferrous iron transport system (feo; Table 1 [Cartron et al., 2006]), and the heme assimilation system (has). The has system produces a hemophore that can bind to heme, or hemoglobin that contains heme. In the closely related system of Serratia marcescens it is conditionally cooperative, as uptake is direct without the hemophore at high concentrations (Ghigo et al., 1997; Létoffé et al., 2004).

Table 1
Main iron uptake systems of P. aeruginosa.

The pyoverdine and pyochelin systems produce siderophores which can steal Fe3+from host carriers. The conditionally cooperative has system produces a hemophore that can bind to heme, or hemoglobin that contains heme. The phu and feo systems are private. All systems have a specific receptor and uptake feeds back to increase system expression. The illustration shows a receptor in the cell membrane taking up an iron source (star). In a cooperative system a secreted public good (circle) binds the target and is taken up by the receptor, whereas a private system takes up the iron source directly.

https://doi.org/10.7554/eLife.38594.002
SystemMechanismSocial statusTargetIllustration
PyoverdinePublic good; siderophore and specific receptorCooperativeFe3+,
host-bound
PyochelinPublic good; siderophore and specific receptorCooperativeFe3+,
host-bound
hasPublic good; hemophore and specific receptorConditionally cooperative or privateHeme, hemoglobin
phuSpecific receptorPrivateHeme, hemoglobin
feoSpecific receptorPrivateFe2+

P. aeruginosa iron metabolism evolves during long term infection. Pyoverdine production has repeatedly been found to be lost during CF infection (Andersen et al., 2015; Jeukens et al., 2014; Konings et al., 2013; Martin et al., 2011), and the phu system has been observed to be upregulated late in infection (Marvig et al., 2014; Nguyen et al., 2014). A major challenge is to identify the selective pressures experienced in situ that cause such changes. Host adaptation, to accommodate for example antibiotic pressure and resource limitations, is frequently predicted to be the primary force in pathogen evolution (Lieberman et al., 2011; Young et al., 2012). Experimentally recreating host conditions in vitro is challenging, given the complex interplay between spatial structure, host immunity and nutrient availability (Clevers, 2016), which we are only beginning to be able to measure (Koo et al., 2011). Clinical measurements may differ from experimental in vivo and in vitro findings (Cornforth et al., 2018), and studies using ‘simple’ experimental conditions do not always give comparable results to those that use more complex, and potentially more clinically relevant conditions (Harrison et al., 2017). In contrast, longitudinal sampling of patients to track in situ evolution provides an opportunity to infer selection without making a priori assumptions about the experienced environment. Evolutionary theory lets us make testable predictions to distinguish the effect of host adaptations and social interactions. With this approach we have shown that cheating is a major selective force in the loss of cooperative iron acquisition in the CF lung (Andersen et al., 2015).

Here, we identified changes in iron uptake strategies from genome sequences and correlated these with the social environment (the P. aeruginosa clonal lineage infecting a host), and duration of infection. Specifically, we test two hypotheses: first, we ask whether an increase in cooperation facilitates the breakdown of cooperation in a natural population, and second, can cells survive the loss of cooperation by switching to a private, and, therefore, unexploitable mechanism for acquiring iron. Two collections of P. aeruginosa samples were used, covering 551 whole genome sequenced isolates from 64 Danish patients, of 55 clone types (Andersen et al., 2015; Markussen et al., 2014; Marvig et al., 2013; Marvig et al., 2015; Rau et al., 2012; Yang et al., 2011); allowing for the identification of patterns of convergent evolution. While the P. aeruginosa populations within patients are remarkably uniform at the clone type level, colonizers diversify during infection (Mowat et al., 2011; Winstanley et al., 2016), and lineages may occupy distinct niches at late stages (Jorth et al., 2015). The isolate collection has been shown to represent the genetic diversity well, with multiple genotypes co-occurring within individual patient samples (Sommer et al., 2016). A key step in our analysis is to categorise isolates with respect to their social environment, for example, presence vs absence of cooperators (pyoverdine producers). This is necessary for testing our hypothesis that iron acquisition strategy is influenced by the social environment. The likelihood that we do not capture all co-infecting strains in our sample, for example leading to mis-classification of isolates as belonging to an environment without pyoverdine, will homogenise the sample groups and hence obscure any differences. Sampling errors of this kind will, therefore, hinder our ability to detect an effect (Type I error).

Materials and methods

Isolate collections

Request a detailed protocol

Two collections of whole genome sequenced P. aeruginosa isolates were used, as described previously (Andersen et al., 2015). The transmissible clonetypes DK1 (46 isolates) and DK2 (54 isolates) were sampled from 28 CF patients between 1972 and 2012 (Jelsbak et al., 2007; Markussen et al., 2014; Marvig et al., 2013; Rau et al., 2012; Yang et al., 2011). Six previously unanalysed DK1 isolates were included (from patients P55M3, P24M2 and P30M0, Supplementary file 1B). Further, 451 isolates of 54 clonetypes from 36 young CF patients were used (the ‘children’s collection’ (Andersen et al., 2015; Marvig et al., 2015), of which 10 were of the DK1 clone type). At each sampling time, 1–8 colonies were cultured from patient sputum, and stored at −80° C (Johansen et al., 2012; Sommer et al., 2016). The sampling regimen for each patient is described in Supplementary file 1B, and Appendix 1—figures 1 and 2. The two collections complement each other as one covers a long time period, with relatively few samples from each patient and of both early and late infection stages, whilst the other covers the first 10 years of infection by environmental clone types, with more extensive sampling within patients. In the children’s collection there is limited transmission between patients, but some clone types infect multiple patients, and some patients are infected by multiple clone types either at the same time or in succession (Marvig et al., 2015). All but two patients, however, harbour only one dominant clone type at a time and the social environment is thus made up of clonal lineages sharing the same ancestor. The analysis of increased pyoverdine production was only performed on the children’s collection, as described below.

The phylogeny of the 54 DK2 isolates has previously been reported by Marvig et al. (2013), and this was adapted for this study to phylogenetically map the order of mutations in regions of interest. To allow a similar analysis for the DK1 clone type, we used Snippy (Seeman, 2015) to identify SNPs in each of the 56 DK1 isolates relative to the PAO1 reference genome (RefSeq assembly accession GCF_000006765.1) and compared the SNPs of all the DK1 isolates to identify those that differed between the isolates. The SNPs were further filtered to only retain SNPs at positions covered by at least 10 reads in all isolates and exclude those where all isolates show more than 80% non-reference reads at the given position. Based on 4,639 SNPs within 2,614,892 sites, a maximum-likelihood phylogenetic tree was computed with RAxML version 8.2.11 (Stamatakis, 2014), with default settings using a general time reversible model of nucleotide substitution (option -m GTRCAT). The tree was rooted with the PAO1 reference genome.

The length of infection of each clone type in each patient was calculated. Clone types sampled only once from an individual patient were excluded (23 isolates of 22 clone types). For the transmissible clone types DK1 and DK2 the total length of infection across patients was used (39 and 36 years, respectively). This is a very conservative estimate of length of infection, as evolution occurs independently in different patients. To calculate the number of generations, a doubling time of 74 min was used for clonal lineages infecting for less than 6 years, 83 min for 6–24 years and 109 min for >24 years based on the measurements of Yang et al. (2011).

Changes in pyoverdine production during infection

Request a detailed protocol

We tested if the appearance of pyoverdine non-producers is correlated with preceding upregulation. Measurement of pyoverdine production was carried out by fluorescence readings at 400/460 nm excitation/emission with a 475 nm cut-off following growth in iron-limited media (Andersen et al., 2015; Kümmerli et al., 2009). For each clonal lineage the baseline pyoverdine production was determined as that of the first isolate(s) of a clone type from a patient, and the production of all subsequent isolates of this clone type was compared to this. An isolate was classified as an over-producer if its production was >30% higher than that of the first. This cut-off was chosen as ~95% of the isolates had a lower standard deviation between replicates. The data is available from Dryad (doi:10.5061/dryad.6963pj3). The analysis was performed on longitudinally sampled clone types from the children’s collection only, as the DK1 and DK2 isolates were not sampled frequently within patients. Only non-producers evolved from a clonal lineage that produced pyoverdine at the beginning of the infection period were included. Therefore clone types DK1 (10 isolates, P36F2), DK11 (two isolates, P51M5), DK40 (two isolates, P73M4) and DK53 (10 isolates, P36F2) were excluded from the analysis, as they were ‘chronic’ clone types with no pyoverdine producers sampled from the young patients (Appendix 1—figure 1; Andersen et al., 2015). As there were only three samples from P51M5, this whole patient was excluded from the analyses. In lineages with both over- and non-producers, over-producers were sampled before non-producers in all but three cases. In two cases there was <5 months between sampling, and we scored non-producers as occurring at the same time as overproducers. In one case the non-producer was sampled 2.95 years prior to the over-producer and scored as having occurred in the absence of an over-producer. The subsequent sampling of the over-producer, and an additional non-producer that shared the genotype with the first non-producer, was not included in the analyses (Appendix 1—figure 1, P53M4). In two clone types two independent transitions to non-producers were observed (non-producing isolates harboured different pyoverdine mutations) and these were included as independent events (P44F5 and P14M4). We focus here on within-clone type dynamics, as temporal overlap of longitudinally sampled clone types only occurred in two patients (Appendix 1—figure 1; Andersen et al., 2015).

We categorized each clone type in individual patients as harbouring over-producers or not. If an over-producer was present, the length of time from that sampling to a non-producer was observed, or till the last sample was collected from the patient, was noted. Clone types without non-producers were classified as right-censored. The same was done for patients without over-producers. In total, 45 clonal lineages were followed covering 404 isolates of 33 clonetypes from 35 patients (Appendix 1—figure 1; some clone types occur in multiple patients, and some patients are infected by multiple clone types [Marvig et al., 2015]). The timing of sampling of non-producers was analysed with the ‘survival’ package in R, which takes into account the sampling extent of a clonal lineage, that is both the timing and frequency of events (R Core Team, 2013; Therneau, 2015).

Identification of mutations in the pyoverdine, pyochelin, phu, has and feo systems

Request a detailed protocol

Mutations in the pyoverdine (pvdQ-pvdI +pvdS pvdG+pvcA ptxR+fpvB), the pyochelin (ampP-pchA), the phu (prrF1-phuR), the has (PA3404-hasI) and the feo systems (PA4357-feoA) were identified previously from Illumina GAIIx or HiSeq2000 reads (Andersen et al., 2015; Markussen et al., 2014; Marvig et al., 2013; Marvig et al., 2015; Rau et al., 2012; Yang et al., 2011) (Supplementary file 1C-D). Mutations of iron systems in the six new DK1 isolates were identified as described previously (Andersen et al., 2015), by mapping reads against a DK1 draft genome (Markussen et al., 2014). Mapping of reads revealed low sequencing depth of the hasD (required for hemophore secretion) and hasS (the has anti-sigma factor) genes, likely due to repetitive regions. These were unsuccessfully attempted amplified by PCR. Mutations were categorized as synonymous (syn) or non-synonymous (ns) SNPs, insertions or deletions (indels), or intergenic. We tested for a bias in which genes in each operon were mutated with [Poisson test, P(X ≥ changes observed)~Poisson distribution (pois) (X; changes expected) < 0.05].

We identified the order of mutation across the different iron uptake systems, with a system classified as mutated by the presence of ns SNPs or indels in any gene in the system. For pyoverdine, the presence of mutations was compared to the actual measurements of production; the two measures are highly consistent (Andersen et al., 2015) but when incongruent the phenotype was used. For the phu region intergenic phuS//phuR mutations predicted to cause receptor upregulation were included, and when occurring without other phu mutations these were recorded separately. Transitions were identified within clonal lineages, that is if an isolate for example had a pyoverdine mutation and measured loss of production, and a subsequent isolate had the same pyoverdine mutation in addition to a pyochelin mutation, this was considered a transition from wild-type (WT) to loss of pyoverdine, to loss of pyochelin. If for example a phu and pyochelin mutation was observed for the first time in the same isolate the exact order could not be inferred. In the transmissible clone type DK2 pyoverdine production was detected to have been lost in an isolate from 1973 and the mutation and phenotype was found in all later isolates. Subsequently, independent clonal lineages for example accumulate phu mutations. In the analyses this was counted as one loss of pyoverdine, and then multiple independent phu mutations. We calculated the expected order of mutation, equally weighted by the size and number of mutations of the different systems. We tested, as described above, whether this was different from the observed, using only cases where there was a clear order of mutations. The social environment, that is the genotypes of co-occurring isolates, was not considered in this analysis.

For the phu system, we located mutations in the phuR receptor gene to functional regions (the N-terminal-domain, periplasmic turns, transmembrane strands or extracellular loops) using a predicted structure of the receptor (http://bioinformatics.biol.uoa.gr//PRED-TMBB [Bagos et al., 2004]). For each domain in the receptor we calculated the expected number of non-synonymous amino acid changes, as the proportion the domain comprises of the entire protein times the observed number of changes. Whether the observed changes in each domain differed significantly from the expected was calculated as described above.

Transitions between public and private iron uptake

Request a detailed protocol

To test if the appearance of intergenic phuS//phuR mutations that cause upregulation of the phu system was correlated with overall loss of pyoverdine production in the social environment we categorized each independent acquisition of these mutations as having occurred in the presence or absence of measured pyoverdine production, by the focal isolate and co-occurring isolates. In the absence of pyoverdine production the time between loss of production and occurrence of mutations was estimated. The analysis was performed on longitudinally sampled clonal lineages from both collections of isolates, that is lineages with at least two isolates with or without pyoverdine production. Further, three cases where the loss of pyoverdine production and phuS//phuR mutations were observed in the same isolate were also included. Clonal lineages that did not acquire phuS//phuR mutations were classified as right-censored. For DK1 and DK2 we used monophyletic clades with and without mutations as independent clonal lineages (Blue lines in Appendix 1—figure 3 and 4). Branches with single isolates were not included unless they showed loss of pyoverdine from the social environment or phuS//phuR mutations (inclusion of these did not significantly affect the result of the test). We included isolates of DK53 where no WT pyoverdine producer was sampled, but phuS//phuR mutations were inferred by comparison to PAO1 (Supplementary file 1C).

The phuS//phuR mutations only occurred in isolates that had lost pyoverdine production (Appendix 1—figure 3 and 4). In three instances these, however, co-occurred in the patient with isolates that produced pyoverdine (Appendix 1—figure 5 and 6). The producers and non-producers with phu upregulation are unlikely to have interacted at the time when the iron system mutations occurred in two of these. In patient P30M0, infected with clone type DK1, the cooperating isolates were sampled from the patient’s sinuses only, while the non-producers with phu upregulation came from lung samples. In patient P28F1, two lineages of clone type DK1 co-occurred. One had been transmitted from another patient, where pyoverdine production was lost, and phu likely upregulated, prior to transmission and establishment of a co-infection. In patient P82M3 a pyoverdine producer and an isolate with a phuS//phuR mutation, both of clone type DK32, were found in the same lung sample. The latter harbours the most common phuS//phuR mutation (G 146 upstream//35 upstream A, Supplementary file 1C), which is located outside the phuR promoter (Marvig et al., 2014). It is the only isolate to have this phuS//phuR mutation in isolation, all others with it have an additional SNP, and the effect of it is unknown (but see below).

The timing of sampling of mutants was analysed with the ‘survival’ package in R (Therneau, 2015). We are likely to over-estimate the length of time to mutation, as patients chronically infected with the transmissible DK1 and DK2 clone types were infrequently sampled (Markussen et al., 2014; Marvig et al., 2013). In these cases, with several years between longitudinal samples, a mutation may only be detected years after occurring.

Uptake of heme following phuS//phuR mutations

Request a detailed protocol

We tested if the presence of phuS//phuR mutations gives a growth advantage in iron-limited media supplemented with heme at different concentrations, using an isogenic pair of DK2 isolates only differing in the presence of a 1 bp phuS//phuR deletion (Marvig et al., 2014). Three biological replicates of each isolate were cultured in liquid LB media for 24–48 hr, so that all cultures reached an OD600 above 0.1. Cultures were subsequently standardised to a starting OD600 of 0.001, in iron-limited CAA media following Kümmerli et al. (2009), without heme or supplemented with 1, 2.5, 5 or 10 μM heme as an iron source (BioXtra porcine hemin ≥97.0%, Sigma-Aldrich). Of each culture 200 μL were inoculated in 96 well plates. Plates were sealed with a breathable membrane to avoid evaporation and incubated in a plate reader measuring OD600 every 30 min for 48 hr. Growth curves were plotted in Excel and the average maximum OD600 of the three replicates calculated. The difference in max OD600 between the isolates with and without the phuS//phuR mutation at the various heme concentrations was analysed in R with a Two-Way ANOVA test with post hoc comparisons with Tukey HSD (R Core Team, 2013).

We also compared growth of pairs of clinical isolates with phuS//phuR mutations matched to their closest genetically related isolate without mutations. The phuS//phuR mutations occurred in three different clone types without additional phu mutations (DK1, DK2 and DK32, highlighted in yellow in Supplementary file 1C). Six isolates with mutations were matched with four closest relatives without, because DK2 in two instances had independent phuS//phuR mutations in two closely related isolates. The WT isolate for these pairs was used to make the isogenic pair described above, and the mutation of one of these clinical isolates moved to this background (P0M30-1979 and P173-2005; Appendix 1—figure 7; Marvig et al., 2014). One additional DK2 isolate with a mutation was lost from the collection, and only available as sequence reads (P80F1; Appendix 1—figure 4 and Supplementary file 1C). Growth was measured as described above with 2.5 or 5 μM heme as an iron source, and differences in max OD600 within a pair were compared with a t-test in R (R Core Team, 2013).

Results and discussion

More cooperative populations are more vulnerable to exploitation

The environmental clones that infect CF patients produce pyoverdine (Andersen et al., 2015) but what happens as they transition to life in a host? We compared pyoverdine production between isolates sampled at different time points of infection in iron-limited media, in which cells must produce pyoverdine to avoid iron starvation. In some clonal lineages of P. aeruginosa, defined as longitudinally sampled isolates of the same clone type that share a recent ancestor, pyoverdine production was maintained for >30 years of CF infection: in the 12 clonal lineages of the transmissible clone type DK1 that were followed for >30 years there were five independent losses of pyoverdine production but three clonal lineages maintained production (Andersen et al., 2015); Appendix 1—figure 3, and below). Limited diffusion of secreted pyoverdine, as is likely in viscous CF sputum, impairs the ability of potential cheats to exploit and may contribute to stabilize cooperation in these lineages (Julou et al., 2013; Kummerli et al., 2009); although there may be significant exchange between spatially distinct microcolonies (Weigert and Kümmerli, 2017).

When analysing the isolates from 35 frequently sampled young patients of 33 clone types (some clone types infect multiple patients and some patients harbour multiple clone types), we found that pyoverdine production increased in 36% of clonal lineages within the first two years of infection. Overall, these ‘super-cooperator’ isolates, which had >30% higher production than isolates initiating infection, were detected in 25 of 45 independent clonal lineages (55%; Figure 1), and on average sampled 1.89 ± 1.79 st. dev. years into the infection. This suggests that upregulation is initially favoured, likely by iron limitation or inter-species competition. In CF patients, P. aeruginosa frequently co-infects with Staphylococcus aureus (Folkesson et al., 2012), and co-culture in vitro causes an upregulation of pyoverdine production (Harrison et al., 2008). When in co-infection, iron sequestering can have the additional benefit of making it unavailable for S. aureus or other competitors (Leinweber et al., 2018; Niehus et al., 2017). Increased production could also be part of a general acute infection phenotype, as either a pleiotropic side-effect or a beneficial trait (Coggan and Wolfgang, 2012). For six out of the 25 lineages with super-cooperators we identified candidate mutations in global regulatory and quorum sensing genes that likely cause increased production of pyoverdine as well as other virulence factors, such as pyochelin, pyocyanin and alginate (Supplementary file 1E). For the remainder, super-cooperators produced pyoverdine at a consistently higher rate than their ancestors but the genetic basis of upregulation was unclear.

Super-cooperators precede the sampling of non-producers.

Inverse Kaplan-Meier graph showing that pyoverdine non-producers are found more often in the presence of super-cooperators (dotted line, n = 25 representing 24 clonal lineages in patients, one of which had two independent losses of pyoverdine production), compared to when super-cooperators are absent (solid line, n = 22, representing 21 clonal lineages in patients, one of which had two independent losses). Circles indicate when sampling of clonal lineages stopped, without the observation of lost pyoverdine production. Insert shows that a transition from wild-type (WT) production to super-cooperation (pvd+) was observed in 24 clonal lineages, followed by nine independent occurrences of non-producers. Non-producers evolved in the absence of super-cooperators five times. In total, 33 clone types infecting 35 patients were followed.

https://doi.org/10.7554/eLife.38594.003

Despite the potential benefits of pyoverdine production, the presence of super-cooperators appears to weaken long-term stability of cooperation in the lung. In 12 out of the 35 patients, we previously observed the appearance of mutants that have lost the ability to synthesise pyoverdine (Andersen et al., 2015), sampled on average 3.06 ± 2.29 st. dev. years into infection. Here, we found that the likelihood of these non-producers arising was significantly higher in the presence of super-cooperators [Survival analysis, χ2 (2, N = 47) = 5.7; p < 0.05; Figure 1]. Non-producers were present in four out of 21 clonal lineages without (19%, on average 2.60 ± 2.45 st. dev years into infection) and in eight out of 24 lineages (disregarding one where the non-producer was sampled prior to the super-cooperator) with super-cooperators (33%, on average 3.32 ± 3.32 st. dev years into infection and 1.45 ± 1.39 st. dev. years after sampling of a super-cooperator). There was no significant difference between the time of sampling of non-producers between lineages with or without super-cooperators (Welch Two Sample t-test, t = −0.54, df = 8.01, p = 0.61). From the data set it is not clear whether cheats evolve from super-cooperators or directly from isolates with WT pyoverdine production, as the genetic basis of over-production is unknown.

Why is upregulation of pyoverdine production associated with the subsequent loss of this trait? Non-producers appear to be favoured as a result of exploiting the pyoverdine production of neighbouring cells – a strategy referred to as cheating (Ghoul et al., 2014a). We have previously shown that non-producers in the lung acquire mutations in a pattern consistent with a cheating strategy, as opposed to one expected from redundancy, in that they retained the pyoverdine receptor only if producers were present (Andersen et al., 2015). In other words, non-producers lose the capacity to contribute but retain the ability to benefit from pyoverdine, as long as it is being supplied by cooperative neighbours. The relative fitness of cheats is predicted by theory to be greater when there are higher levels of cooperation in a population, because their competitors are bearing a higher cost (Ross-Gillespie et al., 2007). This prediction has been shown to hold in empirical tests (Ghoul et al., 2014b; Harrison et al., 2008; Jiricny et al., 2010). Hence, super-cooperators are predicted to be especially vulnerable to exploitation and invasion of cheats. While isolates with a WT level of pyoverdine production may also benefit from the presence of super-cooperators, the larger production cost differential between non-producers and super-cooperators gives non-producers a greater advantage than WT producers. The pattern of upregulation, followed by loss, is likely to be applicable to the evolution of other exoproducts during infection.

Evolution of iron acquisition systems in the lung – the sequence of events

How do the changes in cooperative pyoverdine production relate to the overall iron metabolism of the cells? Comparison of the accumulation of mutations across five different iron uptake systems (Table 1) and measurement of pyoverdine production showed that the appearance of pyoverdine non-producers is the first change in iron metabolism observed in most clonal lineages, in both isolate collections. Loss-of-function of the cooperative pyoverdine system occurred earliest in 16 out of 25 clonal lineages where the phenotypic measurements and sequence analysis provided a clear order of systems affected (Figure 2; Supplementary file 1F). These were observed to occur on average 3.06 years after the first sampling of the clone type in the young patients (n = 11 clonal lineages in nine patients), and between 1–39 years after infection for the more rarely sampled transmissible clone types (n = 5 clonal lineages in five patients). When weighed by the high mutation rate of the system, reflecting a strong selection pressure, and its large size, this is not unexpected [Poisson test, P(X ≥ 16) ∼ pois (X; 14.69) > 0.05), Figure 2; Supplementary file 1D].

Order of mutations across iron acquisition systems.

Wild type-like (WT) isolates colonize patients, and subsequent loss of cooperative pyoverdine production (pvd-) is the most common change in iron metabolism (n = 25), compared to mutations expected to affect the pyochelin, phu and has systems negatively (pch-, phu-, has-). Loss of pyoverdine production in clonal lineages is followed by intergenic phuS//phuR mutations, predicted to upregulate the private phu system, significantly more often than expected by chance (phu+, n = 9). Colour indicates iron uptake system (green = pyoverdine, yellow = pyochelin, dark red = intergenic phuS//phuR, orange = phu (either a mix of phuS//phuR and other phu mutations, or only ns SNPs), blue = has). The occurrence of has mutations may be underestimated due to low sequencing depth of two genes. The figure shows only transitions where there was a clear order in which systems were affected, see Supplementary file 1F for all transitions.

https://doi.org/10.7554/eLife.38594.004

Mutation of the cooperative pyochelin system first was found not to differ from that expected given a random distribution of mutations [Poisson test, P(X ≥ 6) ∼ pois (X; 5.14) > 0.05), Figure 2; Supplementary file 1D]. The conditionally private or cooperative has heme uptake system was found to acquire few mutations during infection. It remains to be tested if this is because maintenance of the system is favoured by selection, or if selection for loss is relatively weak. The private feo system is rarely mutated (Supplementary file 1D) and has been suggested to be of increased importance as conditions turn microanaerobic during infection (Hunter et al., 2013).

Across both isolate collections we observed pyoverdine cheats appear 14 times without going to fixation within the patient during the sampling period, but further recorded eight cases where no producers were sampled a year prior to, or after emergence a non-producer, and two clonal lineages from where no producers are sampled at any point. This is the potential extinction scenario of cheat invasion. However, because we have continued to sample after this event, we can observe how the population responds to this potentially catastrophic situation. One of these events for example occurred in the transmitted clone type DK2 that was subsequently sampled from 16 patients over 35 years. And because cheat invasion is not inevitable, we can compare dynamics in lungs with and without availability of pyoverdine.

Cheat invasion is associated with a subsequent switch to private iron uptake mechanism

After cooperation was lost from a clonal lineage, and only if it was lost, we observed that iron uptake was privatised (Figure 3). This was achieved by upregulation of the private phu system. The phu system targets the iron-rich heme molecule, which is taken up directly through a membrane-bound receptor PhuR (Table 1; Ochsner et al., 2000). Increased expression of the phu system results from intergenic mutations between the phuS gene encoding a heme traffic protein, and the receptor gene phuR (phuS//phuR mutations; Marvig et al., 2014). In the isolate collections, a total of 29 SNPs and two indels accumulate in the phuS//phuR region. Eight of these mutations in the phuR promoter have been found to cause a significant upregulation of promoter activity, and one of these has further been shown, in an isogenic pair, to provide a growth benefit to the carrier when heme is available as an iron source (Marvig et al., 2014); see also below). These specific mutations were found to arise significantly more frequently than expected by chance following the loss of pyoverdine production [Poisson test, P(X ≥ 5) ∼ pois (X; 0.39) < 0.01); Figure 2]. In three patients, non-producing isolates with phuS//phuR mutations were sampled at the same time as pyoverdine producers, however in two of these the producer and non-producers were unlikely to interact, and in the third the phuS//phuR mutation occurred outside the phu promoter region with unknown effects on heme uptake (but see Appendix 1—figure 7). We thus conclude that phu upregulation takes place when pyoverdine production is lost from the social environment. Many of the isolates with phuS//phuR mutations also harboured pyoverdine receptor mutations that eliminate the possibility to cheat (6 out of 7 in DK1, 7 out of 7 in DK2, 1 out of 2 in DK53, 0 out of 1 in DK32; Appendix 1—figure 36; Andersen et al., 2015). Knock-out of the pyoverdine receptor is, however, not expected to be a selective force in phu upregulation. Rather, the data suggest that both occur because pyoverdine is no longer available in the environment, selecting for privatisation of iron uptake and elimination of the costly receptor (Andersen et al., 2015).

Loss of cooperation precedes privatisation of iron uptake.

Inverse Kaplan-Meier graph showing that phuS//phuR mutations are observed 15 times when pyoverdine is absent from the social environment (solid line, n = 26), and only once when pyoverdine is available in the social environment (dashed line, n = 47, see text). The mutations are only found in clonal lineages that do not produce pyoverdine. Circles indicate when sampling of clonal lineages stopped, without the observation of phuS//phuR mutations. Insert shows that a transition from wild-type pyoverdine production (WT) to no production (pvd-) was observed 22 times in clonal lineages, while phuS//phuR mutations occurred 16 times independently (phu+), 13 of which were after loss of pyoverdine production from the environment. The order could not be inferred in three cases, in one of which the isolate co-occurred with pyoverdine producers.

https://doi.org/10.7554/eLife.38594.005

Social interactions are unlikely to be the only selection pressure influencing dynamics in iron acquisition strategies. Changes in iron uptake strategy might more readily be expected to reflect changes in the lung environment as the availability of iron in different forms increases with disease progression (Hunter et al., 2013). And while iron is found to be equally available in its oxidized ferric (Fe3+) and reduced ferrous forms (Fe2+) in early infection, there is a skew towards the reduced form later, which the feo system targets (Hunter et al., 2013). Heme and hemoglobin are also present in the sputum (Ghio et al., 2013), and availability has been speculated to increase as the lung tissue degrades (Marvig et al., 2014), as evidenced by increased coughing up of blood, termed hemoptysis, with age (Thompson et al., 2015). As such, pyoverdine may be most efficient early in infection, and phu only advantageous later as lung tissue degrades and/or alternative sources of iron become more or less abundant. However, the patterns we observe are not consistent with this explanation. If pyoverdine production is maintained, the phuS//phuR genotype remains unaltered from colonization, even after more than 30 years of infection (Figure 3). The phuS//phuR mutations, responsible for upregulating the private phu system, are only observed in non-producers, and in 15 out of 16 cases after the loss of cooperation in the population [Survival analysis, χ2 (2, N = 73) = 11.7; p < 0.01; Figures 3 and 4], occurring between 0 to 35 years after the loss of pyoverdine production. Upregulation of private iron uptake can, therefore, be predicted from the social environment, not duration of infection and associated changes in the lung environment.

Evolutionary trajectories of iron metabolism in the CF airways.

Following infection (n = 45 clonal lineages of 33 clone types), pyoverdine production is frequently upregulated (n = 25). Cheaters are significantly more likely to invade following upregulation. Following loss of pyoverdine production in the social environment (n = 26 non-producing clonal lineages of 6 clone types, from eight independent losses of production and two clonal lineages from which no pyoverdine producers were sampled), the phu system is frequently upregulated by intergenic phuS//phuR mutations (15 times), while this very rarely occurs (one time) when pyoverdine is produced in the social environment (n = 47 producing clonal lineages of 32 clone types).

https://doi.org/10.7554/eLife.38594.006

Why is private heme uptake only upregulated if cooperation is lost?

The fact that we only observe upregulation of the phu system if pyoverdine production has been lost suggests that privatisation is not universally beneficial but may instead represent a final recourse when other mechanisms are no longer functioning. P. aeruginosa iron uptake is finely controlled by feedback loops, in response to environmental iron concentrations and need (Vasil and Ochsner, 1999). As iron is toxic at high concentrations (Anzaldi and Skaar, 2010; Touati, 2000), an indiscriminate upregulation of one system could be a significant disadvantage in environments where iron concentrations fluctuate. We, therefore, tested whether the growth benefits of phu upregulation (by phuS//phuR mutations) were dependent on iron availability, in the form of heme which the phu system targets. We examined the growth difference between a clinical isolate that had lost pyoverdine production, and an isolate, isogenic but for a clinical phuS//phuR mutation that cause an upregulation of the heme receptor gene phuR (Marvig et al., 2014). The experiment was performed in iron-limited media supplemented with heme at biologically relevant concentrations, which ranged from <1 μM typical of healthy individuals, across 2.5 μM, 5 μM to 10 μM typical of CF patients (Ghio et al., 2013).

Consistent with our hypothesis, we find that phu upregulation is only beneficial at a narrow range of intermediate heme concentrations. There was a significant growth difference between the isolates and heme concentrations, and an interaction effect (Two-Way ANOVA; Interaction term: Isolate*Heme: F = 48.11, df = 3, p < 0.01; Figure 5; Supplementary file 1G). At 2.5 μM heme the isolate with an upregulated phu system achieved a higher density (Tukey HSD, padj <0.01, Figure 5), consistent with previous findings (Marvig et al., 2014). In contrast, at 5 and 10 μM the isolate without upregulation had an advantage (padj <0.05, Figure 5). No growth was observed in the absence of heme, and at 1 μM there was no significant difference in growth between the isolates but a trend for the WT to do better (padj = 0.073). This suggests that phuS//phuR mutations lead to increased heme uptake that is beneficial at low external concentrations (>1 and<5 μM heme) but detrimental at high concentrations (>5 μM heme). Analyses of pairs of clinical isolates with and without phuS//phuR mutations gave less clear results, but showed a similar pattern of a benefit of phuS//phuR mutations at low concentrations of added heme (Appendix 1—figure 7). If iron availability increases during infection, an initially beneficial upregulation of heme uptake would turn toxic. In the phu system we found a significant bias towards mutations of the phu receptor gene phuR (Poisson test, P(X ≥ 24) ∼ pois (X; 10.69) < 0.05; Supplementary file 1D). These were not randomly distributed but primarily found in the extracellular loops of the phu receptor, that initiate the uptake of heme (Noinaj et al., 2010; Smith et al., 2015) [Poisson test, P(X ≥ 15)~pois (X; 7.28) < 0.05; Figure 6]. If there is selection to reduce iron concentration within cells, these may, therefore, represent compensatory mutations.

Effect of phu upregulation depends on the heme concentration.

An isogenic pair of isolates differing only in a 1 bp phuS//phuR deletion was grown in iron-limited media supplemented with heme at 1–10 μM. The clinical isolate without the mutation, that does not produce pyoverdine (WT), had a significant higher maximum OD600 after 48 hr growth at 5 and 10 μM, whereas the mutant (phu+) grew best at 2.5 μM. There was no significant difference at 1 μM. Boxplots show median value, and the interquartile range ±25%. The * marks a significant difference at p < 0.05.

https://doi.org/10.7554/eLife.38594.007
Distribution of non-synonymous mutations in the PhuR heme receptor.

The figure shows the β-barrel structure spanning the cell membrane (horizontal lines) with transmembrane strands (grey bars), extracellular loops (labelled L1-L11) and periplasmic turns (bottom loops). Green circles indicate the location of non-synonymous SNPs, the smallest circles denoting one mutation and the largest four. Mutations were significantly biased towards the extracellular loops.

https://doi.org/10.7554/eLife.38594.008

Conclusions

Bacterial cells will be under selection to adapt to the host environment, including antibiotic pressure (Lieberman et al., 2011). The observations we report here, however, demonstrate the errors of interpretation that can arise from assuming that bacterial cells are driven entirely by the evolutionary imperative to survive in the host. As in all other organisms, fitness is also determined by the ability to compete with members of the same species, and these two evolutionary drivers – host adaptation and competition – do not always coincide. Our results show that the initial upregulation of cooperation, selected by either the environmental conditions, competition or as a pleiotropic side-effect, facilitates cheating. The subsequent privatisation of iron uptake is a response to the loss of cooperation, not simply an adaptation to iron availability in the host as previously suggested (Marvig et al., 2014), and as we show this carries its own costs. The effect of the social environment on iron acquisition mechanism remains statistically detectable despite the noise from variation in patient age, sex, infection and treatment. In a clinical system like this, failure to appreciate these factors can lead to misinterpretation of how bacterial cells are influenced by the host environment and, therefore, impair necessary understanding for the development of intervention strategies (Leggett et al., 2014).

Evolutionary theory clearly predicts that any adaptation involving cooperation can be vulnerable to exploitation, irrespective of its benefits (Axelrod and Hamilton, 1981). But this is often overlooked when it is difficult to characterize the selective environment, as for example, inside a human host. By longitudinal sampling of bacterial populations, we were able to observe the consequences of cheat invasion for an essential trait and show that switching to a private strategy can provide bacterial cells with an evolutionary alternative. This observation highlights the importance of considering potential constraints imposed by social interactions: if a private trait is expressed it may not be the most efficient in a given environment, but simply the final recourse in a population with a history of exploitation.

Appendix 1

The sampling regimen behind the two isolate collections.

We used two collections of P. aeruginosa isolates to study iron metabolism during CF infection. For the ‘children’s’ collection the young patients have been sampled frequently, for up to 10 years, and that the majority are infected with unique environmental clone types. When P. aeruginosa was found to transmit between CF patients (Cheng et al., 1996; Jelsbak et al., 2007), care was taken to avoid that younger patients become infected with adapted clone types in the clinic. In the collections we use this is clear as the children’s collection encompass 54 clone types while the collection of isolates from adults contain only two, highly transmissible, clone types. This collection covers >30 years. To give an overview of the sampling regimen we present first an overview of isolates sampled from the individual patients (Appendix 1—figures 1 and 2). To further illustrate how the DK1 and DK2 evolve and transmit between patients we present phylogenies of the two transmissible clone types with the primary changes in iron metabolism in the pyoverdine and phu systems highlighted (Appendix 1—figures 3 and 4).

Appendix 1—figure 1
Sampling regimen and standardised pyoverdine production of longitudinally sampled clonal lineages (>1 isolate) from the children’s collection.

Plots show the sampling time of isolates from individual patients denoted by the patient code and clone type. When more than one clone type was sampled they are ordered so that the first listed corresponds to the dot symbol, followed by the square and triangle. The y-axis shows pyoverdine production standardized by production of the first sampled isolate(s). Isolates with a production of 10–129% of the WT are green, non-producers with a production at <10% of the WT are red, and super-cooperators with a production >130% of the WT are blue. Dashed red lines mark these cut-off points. Four clonal lineages of a ‘chronic’ phenotype with no pyoverdine producers sampled were excluded from the analysis on super-cooperators (DK1 and DK53 in P36F2, DK40 in P73M4, and DK11 in P51M5). When sampling is shown to start later than year 0 (as e.g. for P67M4) there was a sample collected earlier from the patient that either was not sequenced or of a clone type only sampled once. Created with ggplot2 in R (R Core Team, 2013; Wickham, 2016).

https://doi.org/10.7554/eLife.38594.012
Appendix 1—figure 2
Sampling regimen for the transmissible clone types DK1 and DK2.

On the x-axis is sampling year and on the y-axis the patient ID. Red dots show DK1 and blue DK2. Dot size reflects the number of samples collected in a given year (1-4). Note that patient P36F2 from the children’s collection is included here as this patient harboured DK1 in addition to two other clone types (see also Appendix 1—figures 1 and 6). Created with ggplot2 in R (R Core Team, 2013; Wickham, 2016).

https://doi.org/10.7554/eLife.38594.013
Appendix 1—figure 3
Phylogenetic tree of DK1 samples.

Tree was generated as described in Methods. Patient ID is denoted by shape and sampling year is listed. When multiple samples were collected from the same patient the same year, they are labelled corresponding to labels used in Supplementary file 1C. Scale depends on branch thickness; the thickest branch is only used for samples from P36F2. A green empty circle shows the loss of pyoverdine production measured phenotypically. A green circle with a cross shows the occurrence of a non-synonymous mutation in the pyoverdine receptor gene fpvA or the receptor sigma-factor gene fpvI. A full red circle shows phuS//phuR mutations, a red circle with a cross non-synonymous mutations in the phu receptor gene phuR and an empty red circle additional non-synonymous phu mutations. Additional mutations in the pyoverdine, pyochelin, has and feo systems are not shown. Blue lines mark monophyletic clades used in the analysis on phu upregulation. For isolates from P36F2, one phuR mutation, and three in fpvA were found in multiple isolates on polyphyletic branches (marked with 1, 2 and 3). These are likely not independent, and were not treated as such. Differences in pyoverdine mutations between isolates from P36F2 and P46F0 and P28F1 suggest two independent losses of production..

https://doi.org/10.7554/eLife.38594.014
Appendix 1—figure 4
Phylogenetic tree of DK2 samples, adapted from (Marvig et al., 2013).

Pheno- and genotypes highlighted as in Appendix 1—figure 3.

https://doi.org/10.7554/eLife.38594.015

To clarify the within-patient dynamics and social environment we provide an overview of the phenotypes of isolates for the 11 patients with DK1 and DK2 that were sampled longitudinally (Appendix 1—figure 5). There are two patients (P28F1 and P30F0) where DK1 isolates that have lost pyoverdine production and have phu upregulated are sampled concurrently with WT pyoverdine producers. In P30F0 the isolates with phu upregulation were isolated from the sinuses, and the rest from the lungs, and these have therefore inhabited different niches. In P28F1, the pyoverdine and phu mutations occurred in another patient (likely P46F0, Appendix 1—figure 3) and were transmitted to patient P28F1 where another DK1 lineage also established. In three patients DK1 and DK2 overlapped in time (P24M2, P43M2 and P73F1, Appendix 1—figure 5). Could the two clone types have interacted, in regards to iron uptake, in these three patients? Patient P73F1 was first found to be colonized by DK2 in 1984, which had already lost pyoverdine production and uptake ability in 1973 (Appendix 1—figure 4). In the last two isolates there were two independent upregulations of the phu system (Appendix 1-fig. 4 and 5). A DK1 isolate was sampled in 2002 that also had pyoverdine production and uptake knocked-out, and carried phuS//phuR mutations for phu upregulation. From the DK1 phylogeny (Appendix 1—figure 3) it is unclear whether these mutations in DK1 occurred in patient P73F1 and were transmitted to patient P43M2 or vice versa. Given the simultaneous upregulation of phu in DK2 in P73F1 we speculate that they may have occurred in this patient. In patient P24M2 a WT-like DK1 clonal lineage is sampled over a timespan of nine years co-occurring with the non-cooperative and non-cheating DK2. The young patient P82M3 was infected by clone type DK32, and three of the late isolates had lost pyoverdine production (Appendix 1—figure 6). Two of these, co-occurring with pyoverdine producers, had a phuS//phuR mutation outside of the phuR promoter, as discussed in the main text. Patient P36F2 was first infected by DK53 and DK1, neither of which produced pyoverdine and DK53 further had phuS//phuR mutations. A WT- like pyoverdine producer, DK15, was later sampled (Appendix 1—figures 1 and 6).

Appendix 1—figure 5
Phenotypes of longitudinally sampled DK1 and DK2 lineages.

Size of circle denotes the number of samples (1-3). WT-like isolates that produce pyoverdine are shown as a full green circle, non-producers in white with a green stroke, pyoverdine receptor mutations with a green cross and phu-upregulation by phuS//phuR mutations is shown with red. Created with ggplot2 in R (R Core Team, 2013; Wickham, 2016).

https://doi.org/10.7554/eLife.38594.016
Appendix 1—figure 6
Phenotypes of co-occurring clone types in two patients from the children’s collection.

Phenotypes as in Appendix 1—figure 5. Size of circle denotes the number of samples (1, 2 and 4). On the x-axis years since colonization. Note that patient P36F2 is shown in both Appendix 1—figures 5 and 6.

https://doi.org/10.7554/eLife.38594.017

Growth effect of phuS//phuR mutations in clinical isolates.

We tested if the phuS//phuR mutations are beneficial in the clinical isolates, using six isolates of three different clone types that had acquired a variety of phuS//phuR mutations, and no additional phu mutations (Supplementary file 1C, marked in yellow). We compared fitness of the mutants with their closest genetically related isolate without phuS//phuR mutations. The isolates in a pair were, however, separated by numerous other mutations as they were sampled up to 26 years apart. The pairs were grown in iron-limited media with 2.5 and 5 μM heme added. Three mutants with intergenic SNPs had significantly higher (P73F1; 26% higher; t = −6.33, df = 3.48, p > 0.01 and P83M2; 16% higher, t = −4.48, df = 2.15, p < 0.05; isolates sampled 0 and 23 years apart) or a non-significant tendency towards a higher maximum OD600 in 5 μM heme compared to their closest relative (P66F0; 17% higher, t = −2.85, df = 2.67, p = 0.07; isolates sampled 10 years apart; Appendix 1—figure 7). A mutant with a three bp insertion, and an isolate with a one base pair deletion in phuS//phuR in addition to a SNP, showed no significant difference in growth compared to their closest relative (17% higher, t = −1.78, df = 2.95, p = 0.18; 16% lower, t = - 0.55, df = 2.56, p = 0.10; isolates sampled 3 and 26 years apart; Appendix 1—figure 7). These two isolates have, however, previously been shown to have respectively 25 and 116 fold increased expression of the phu receptor compared to the wild type (Marvig et al., 2014; Yang et al., 2011), and transfer of the one base pair deletion to its ancestor confirms that it infers growth benefits in heme media (Marvig et al., 2014; Figure 5). This suggest that other mutations across the genome affect the phenotype we are interested in, which is not unexpected given that we are comparing isolates sampled years apart, and the complex genome of P. aeruginosa where almost 10% of the genes are predicted to be regulatory (Stover et al., 2000). There were no significant growth differences with 2.5 μM heme added. Both isolates of the sixth pair, with a SNP, grew very slowly in rich LB media, and not at all in iron-limited media with or without heme. It is not clear why the beneficial effect of heme in these isolates is seen at a higher concentration compared to the isogenic pair (5 μM vs 2.5 μM, Figure 5).

Appendix 1—figure 7
Growth of pairs of clinical isolates with and without phuS//phuR mutations. 

Isolates were grown in iron-limited media with 5 μM heme added. Pairs where the same WT isolate is used because of two independent acquisitions of phuS//phuR mutations are highlighted in red and blue. Boxplots show median value of three biological replicates, and the interquartile range ±25%. * marks a significant difference. Plotted in R (R Core Team, 2013; Wickham, 2016).

https://doi.org/10.7554/eLife.38594.018
https://doi.org/10.7554/eLife.38594.011

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
    Global regulatory pathways and cross-talk control Pseudomonas aeruginosa environmental lifestyle and virulence phenotype
    1. KA Coggan
    2. MC Wolfgang
    (2012)
    Current Issues in Molecular Biology 14:47–70.
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
    Siderophore presence in sputa of cystic fibrosis patients
    1. B Haas
    2. J Kraut
    3. J Marks
    4. SC Zanker
    5. D Castignetti
    (1991)
    Infection and Immunity 59:3997–4000.
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
    War and peace: social interactions in infections
    1. HC Leggett
    2. SP Brown
    3. SE Reece
    (2014)
    Philosophical Transactions of the Royal Society B: Biological Sciences 369:20130365.
    https://doi.org/10.1098/rstb.2013.0365
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
    R: A Language and Environment for Statistical Computing
    1. R Core Team
    (2013)
    R: A Language and Environment for Statistical Computing.
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
    A package for Survival Analysis in S
    1. T Therneau
    (2015)
    A package for Survival Analysis in S, https://cran.r-project.org/web/packages/survival/index.html.
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
    ggplot 2: Elagant graphics for data analysis
    1. H Wickham
    (2016)
    ggplot 2: Elagant graphics for data analysis, https://doi.org/10.1007/978-0-387-98141-3.
  75. 75
  76. 76
  77. 77
  78. 78

Decision letter

  1. Ian T Baldwin
    Senior Editor; Max Planck Institute for Chemical Ecology, Germany
  2. Otto Cordero
    Reviewing Editor

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Privatisation rescues essential function following loss of cooperation" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Ian Baldwin as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Angela Wilks (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This manuscript by Andersen et al. describes the evolution of Pseudomonas aeruginosa in the lungs of people with CF with a particular focus on evolutionary contingency around cooperative iron acquisition systems. The authors analyze isolates obtained from CF patients sputum in two cohorts of patients with longitudinal sampling and conclude that a series of invasions caused by social evolution take place in the CF lung: first, social cheats invade in siderophore (public good) producing populations, leading to the demise of public good producers. Surprisingly, this event is followed by the raise in frequency of private iron acquisition strategies, which should in principle be able to compete against cheaters. This type evolutionary dynamics, reminiscent of a rock-paper-scissors game, are fundamentally interesting and potentially relevant in the context of pathogenesis, as they could shed light on the selective pressures that impact the evolution of Pseudomonas aeruginosa in CF patients.

All reviewers felt this was a potentially interesting and exciting study, however, a number of serious issues were raised by the reviewers (not by reviewer #3, who was brief but positive). These issues require your careful consideration before publication. Many of them revolve around the presentation of the data, which we felt lacked the required details. This lack of detail led to confusion by some of the reviewers, as you will see in their questions below. To be able to consider this paper for publication, we strongly encourage the authors to provide a better overview of the raw data in the main text (e.g. number of isolates, origin, time of isolation, numbers of isolates per patient, phenotypes tested, etc.). This could be done possibly through a dedicated figure or table.

We also found the style of writing to often be too colloquial. Specifically, frequent use of phrases like "game over", "dead end", "last resort" obscured the message of the paper. Other examples include "Life, however, goes on" and there are more. Along the same lines, the Abstract seems to sacrifice clarity for "impact". This needs to be corrected. See comment 6.

One of the reviewers also questioned the novelty of the results, relative to previous findings by the same group. The authors need to clarify their main new data relative to previous work.

We expect to see a detailed point by point reply to the comments you will read below. Given that many of the questions were related to a lack of clear scientific presentation, once we receive your comments we will have to reevaluate the novelty and significance of the findings.

Essential revisions:

1) Clarity of the data: It is very hard to follow the numbers in this work! Numbers are differently represented between figures and text and often do not seem to match. For example, In Figure 1 and the relevant text they claim they observed 14 pvd- cheating events (9+5), while later on they claim they had 23 events (14+9) (subsection “Evolution of iron acquisition systems in the lung – the sequence of events”). What is the relation between these two numbers? The authors need to find a better way to represent the data in a clearer fashion. We strongly suggest to have a new figure with a visualization of the number of strains per time point, per patient and their phenotypes/genotypes.

2) Relation to previous work by the same group: The main problem I see here is the relation between this work and the mBio paper by Marvig et al., 2014, and the PNAS paper by Andersen et al., 2015, both done by the same research groups involved in this work and the latter with the same first author as the current work. The authors often refer to these two papers in the current manuscript, but reading these two papers reveal that most of the basic observations mentioned in the current text were already made in these two papers. It is therefore unclear what is the novelty of this manuscript.

a) Cheating on pyoverdine production: Andersen et al., 2015, have shown that pvd mutations first arise in pvd production pathway (mainly pvdS) and later in the reception pathway, only once the pvd- production mutants are fixed in the population (based on other strains obtained in the same sample and in previous samples of the same patient). They hypothesize that production mutants are cheaters and that pvd reception mutants are favorable due to the cost of reception in the absence of pvd-related benefit. Several points should be made with respect to this manuscript.

i) First, it seems that the basic observation of cheating on pvd production has already been made. The only thing that is added in this manuscript is the observation regarding the relation between cheating and super-producers, to which I have several reservations (see separate point below).

ii) In the current work, only production mutants are discussed, while reception mutants are not mentioned at all. Since both pvd reception mutants and phu+ mutants are correlated with fixation of the pvd production mutants, it is surprising that reception mutants were not discussed and compared with phu+ mutants.

b) Marvig et al., has already identified the phu+ promoter mutation and the fact that they arise only in pvd production mutants, though their work was done on a smaller number of lineages. Therefore, it is not clear what is the added value of the current work? Is it the demonstration that phu+ alleles are only beneficial at low heme levels (see comments on the experiments in a separate point)? Is the claim for novelty in this work depends on the enlarged statistical power? Or is it the claim that phu+ promoter mutants arise only when pvd+ phenotype is completely eliminated from the social environment and not only from the specific lineage? See comments on sociality below for the last option.

Altogether, the distinction between the current work and previous works of the group are not sufficiently clear.

3) Social interaction: It is hard to understand when do the authors discuss social environment within the given lineage (i.e., when discussing interaction between a mutant and its ancestor) and when do they discuss social environment based on multiple strains sampled from the same patient at the same time. While the authors claim social explanations for their observation, much of the analysis in the text is not based on direct social observations, but on comparisons between a mutant and its ancestors (e.g., Figure 1, 2, 3). The two places where a social context is mentioned are:

a) When they discuss the social environment of pvd- cheats, the authors mention that 14 events are not fixed while 9 other are fixed (subsection “Evolution of iron acquisition systems in the lung – the sequence of events”).

b) Later, the authors mention that the phu+ mutations occurred significantly more when Pvd was lost, but here it is unclear whether they mean that Pvd was lost from the genome of the strains, or from the total social environment (subsection “Cheat invasion is associated with a subsequent switch to private iron uptake mechanism”), as they refer to Figure 2 which, to the best of my understanding, discussed only changes within a given genome and not the social context. On the other hand, they specifically mention that in all but three cases (I guess therefore that in 31-3=28 cases, but again – distinction between lineages and patients is unclear), phu+ arise when no pvd+ strains were observed at all.

In both cases there is insufficient information given on the social environment – how many strains were analyzed in any given sample. This data in neither presented nor analyzed in anyway whatsoever in the text. Now, this is obviously important – I would find the claim for fixation of a given genotype (or the elimination of another) to be much stronger if 10 isolates were sampled from the same patient at the same time than if only two samples were taken (or one!). Yet, the social data is completely absent, so I do not know how to interpret the claim for fixation of the cheater. As this is, to the best of my understanding, the most important point of the paper, the data supporting it should be presented and properly analyzed!

4) Pvd Super producers: the authors defined a subset of strains which produce more pvd than their ancestors (based on longitudinal sampling and genomic sequences). Super-production occurs in 50% of lineages and rather early (within ~2 years) in 33% of lineages. In a minority of super-producer lineages, the authors suggest that it is due to mutations in general regulators (e.g., lasR, mucA, gac). The authors suggest that this is an adaptation due to increased requirements for iron in the lungs. Finally, the author observed that a large fraction of super-producers became null producers (8/24) compared to strains that do not super-produce (4/21). We have several remarks on this part:

a) If super-production results from mutations in general regulators such as lasR or mucA, then pvd over-production might be a pleiotropic side effect of selection for a different trait dependent on the global regulator. This option is not mentioned in the text.

b) Super-producers by themselves are also super-cooperators which will be cheated by their wild-type pvd producer ancestors in their social environment. Something needs to explain the increase of cooperation. Selection for private pleiotropic effects of the mutation (section a) might explain this increased cooperation. Again, social evolution is not discussed in this context.

c) The observation that pvd- mutant phenotype arise more often in super-producers (8/24) than in wild-type producers (4/21) seems to be rather marginal to me. I'm not sure what is the statistical significance analysis, but if one simply assumes binomial distribution around the mean (12/45=26%), then with 21 samples, the chance of obtaining 4 or less events is ~30% and the chances of obtaining 8 events or more in 24 samples is ~40%. Therefore, I am not convinced that this difference is not occurring simply by chance.

5) Sensitivity to heme levels. If I understand correctly, the authors show in Figure 5 the average final OD for multiple related pairs which encode for the phu+ or phu-wt alleles (all in pvd- background) for different levels of heme (1, 2, 5, 10μM). I thought I got the picture clearly, but looking at the supplementary information and figure got me completely confused. The authors refer to six pairs of strains, for which they measured the growth in 2 and 5 μm of heme, but it is unclear on what was 1 and 10 μm measured. Also, while Figure 5 indicates that the phu-wt allele has a growth advantage in 5uM, Appendix 1—figure 7 shows that all the statistically significant (or marginally significant) effects in specific pairs, go on the other direction – phu+ grow better than phu-wt. I couldn't figure out how to resolve these two values in a satisfying manner.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Privatisation rescues essential function following loss of cooperation" for further consideration at eLife. Your revised article has been favorably evaluated by Ian Baldwin (Senior Editor), a Reviewing Editor, and one reviewer.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

Reviewer #2:

In this version, Andersen et al. tried to address the main criticism brought forward in the previous version. All in all, I find the data and figures added to the work to satisfy my main points of criticism. I still have few comments here and there, where I think analysis could be improved or representation could be simplified for the reader.

I would go over my previous main reservations and whether I am satisfied from the corrections done by the authors:

1) Connection between pvd receptor mutations and phu+ phenotypic change. The authors now added the analysis of pvd reception mutations and show a good correspondence between the phu+ promoter mutations and accumulation of pvd reception mutations. I think this supports the notion that cheating on pvd is not occurring upon the phu+ phenotypic change.

2) Social interactions. This is the main point of the paper – loss of pyoverdine production from the social environment and not only from a specific lineage explains the rise of phu+ phenotype. The authors now provide Appendix 1—figure 3A, B to directly show the data. I think this is a good start, but I would suggest that the authors would provide a more patient-centric view of the data and not lineage-centric. The problem with the current representation is that different lineages in the same patient are scattered throughout the two figures, complicating comparison. To look at the data of a specific patient, I had to scan the two figures to find all lineages sampled in this patient and then mentally order them by the sample time, to try and see whether the claim (pvd loss from all samples prior to phu+ mutation) can be verified. I suggest that in addition to the current representation, the authors should show a representation were the data samples for a single patient are clustered together with ascending years and the phenotypic/genetic status of each sample is clearly given (using the same mutation annotation as in the current figure). This would facilitate direct visual estimation of the social data in a given lineage. I don't think there is a need to show patients which have been sampled once with a single clone, as nothing much can be said about their social environment (many such patients are shown in Appendix 1—figure 2).

3) Super producers: the statistical claim of the author is that pvd cheaters are not only preferentially arising in the super-producers, but that they also arise more quickly (i.e., that the rate of appearance of the mutations is higher. This is clear from Figure 1 but is not mentioned appropriately in the text (subsection “More cooperative populations are more vulnerable to exploitation”, third paragraph). I think the authors should also mention that the mean time for appearance of a pvd- mutation was quicker for the eight mutations that arose in the presence of super producer than the mean time for the four that arose without it.

4) Number of generations. The reviewers have cut their estimates considerably, but still rely on an estimation of doubling time of ~70 minutes, which I still find unconvincing – it will either imply that the whole mocus volume of the lungs is replaced every 70 minutes (as in a chemostat, where growth is determined by clearance rate), or that there is a massive level of cell death alongside fast cell growth (and then the chemostat assumptions fail, as average growth does not reflect cell divisions). I am no expert in CF, but both things seem unlikely, though I might be completely wrong on that. In any case, I don't find this to be very important. I would just throw away this estimate, but if the authors choose to keep it, I would not object.

https://doi.org/10.7554/eLife.38594.032

Author response

Summary:

This manuscript by Andersen et al. describes the evolution of Pseudomonas aeruginosa in the lungs of people with CF with a particular focus on evolutionary contingency around cooperative iron acquisition systems. The authors analyze isolates obtained from CF patients sputum in two cohorts of patients with longitudinal sampling and conclude that a series of invasions caused by social evolution take place in the CF lung: first, social cheats invade in siderophore (public good) producing populations, leading to the demise of public good producers. Surprisingly, this event is followed by the raise in frequency of private iron acquisition strategies, which should in principle be able to compete against cheaters. This type evolutionary dynamics, reminiscent of a rock-paper-scissors game, are fundamentally interesting and potentially relevant in the context of pathogenesis, as they could shed light on the selective pressures that impact the evolution of Pseudomonas aeruginosa in CF patients.

Point of clarification: this mis-represents our findings. Cells adopting private iron uptake strategies may also be pyoverdine cheaters, in fact, we find that they always are. There are not three distinct strategies as in the rock-paper-scissors game and this is not a good model for our observations. Instead, the population “escapes” coop cheat dynamics completely by switching to a private mechanism. To our knowledge, we are the first to report this strategy as a response to cheat invasion.

[…] Essential revisions:

1) Clarity of the data: It is very hard to follow the numbers in this work! Numbers are differently represented between figures and text and often do not seem to match. For example, In Figure 1 and the relevant text they claim they observed 14 pvd- cheating events (9+5), while later on they claim they had 23 events (14+9) (subsection “Evolution of iron acquisition systems in the lung – the sequence of events”). What is the relation between these two numbers? The authors need to find a better way to represent the data in a clearer fashion. We strongly suggest to have a new figure with a visualization of the number of strains per time point, per patient and their phenotypes/genotypes.

The study includes two collections of isolates. One from young patients infected primarily with environmental clone types, sampled extensively within patients for max 10 years, and one from patients infected by two transmissible clone types, sampled extensively between patients for >30 years. In the latter there are few samples from individual patients, and longer time between samples. In the analysis of upregulation of pyoverdine we only used samples from the frequently sampled young patients in order to track the pyoverdine dynamics within patients as closely as possible. This information is in the Materials and methods, but to clarify we have now mentioned the two isolate collections in the Introduction (last paragraph), provided more detail in the Materials and methods (subsection “Isolate collections”, first paragraph; subsection “Changes in pyoverdine production during infection”, first paragraph; subsection “Transitions between public and private iron uptake”, first paragraph), and further described in the Results and Discussion which isolates are used when (subsection “More cooperative populations are more vulnerable to exploitation”, first paragraph; subsection “Evolution of iron acquisition systems in the lung – the sequence of events”, first paragraph; subsection “Cheat invasion is associated with a subsequent switch to private iron uptake mechanism”, first paragraph).

2) Relation to previous work by the same group: The main problem I see here is the relation between this work and the mBio paper by Marvig et al., 2014, and the PNAS paper by Andersen et al., 2015, both done by the same research groups involved in this work and the latter with the same first author as the current work. The authors often refer to these two papers in the current manuscript, but reading these two papers reveal that most of the basic observations mentioned in the current text were already made in these two papers. It is therefore unclear what is the novelty of this manuscript.

We have reworded the Introduction (last paragraph), and the Discussion (subsection “Conclusions”) to clarify the novel contribution made by the current submission. This is the first study that we know of to reveal privatisation as a strategy to survive the loss of an essential trait through cheat invasion. Neither of the papers cited address this question, neither is it possible to address this question with the data they present. No other study has shown this in bacteria or in any other system. Bacterial infections offer unique opportunities to observe long-term evolutionary dynamics that are not possible in other taxonomic groups that are studied by the research community interested in social behaviours, e.g. the social insects, vertebrates. In no other system has it been possible to observe the invasion of cheats and the following evolutionary response in a natural population. We expect our results to be of general interest beyond the microbiology community.

Marvig et al., 2014, showed that intergenic mutations in the phu system are common, found in isolates that do not produce pyoverdine, and cause an increased growth rate in the presence of heme as an iron source. The results are interpreted as host adaptation to increased availability of heme as an iron source. In Andersen et al., 2015, we reported the first evidence that cheating is responsible for long-term loss of pyoverdine in a population of P.

aeruginosa infecting the CF lung. Importantly, we were able to present data that helped to reject the alternative, widely-held hypothesis: that pyoverdine loss was due to redundancy as iron levels changed through infection. We provided the first evidence of cheat invasion in a natural population of infectious bacteria.

The current paper reports the following novel findings: (1) increased cooperation facilitates the invasion of cheats, and (2) bacterial populations respond to the loss of an essential trait through cheating by switching to a private mechanism for iron uptake.

a) Cheating on pyoverdine production: Andersen et al., 2015, have shown that pvd mutations first arise in pvd production pathway (mainly pvdS) and later in the reception pathway, only once the pvd- production mutants are fixed in the population (based on other strains obtained in the same sample and in previous samples of the same patient). They hypothesize that production mutants are cheaters and that pvd reception mutants are favorable due to the cost of reception in the absence of pvd-related benefit. Several points should be made with respect to this manuscript.

i) First, it seems that the basic observation of cheating on pvd production has already been made. The only thing that is added in this manuscript is the observation regarding the relation between cheating and super-producers, to which I have several reservations (see separate point below).

The observation of cheating is not claimed as novel in this manuscript but stated as a finding of our previous article (Introduction, fourth paragraph).

ii) In the current work, only production mutants are discussed, while reception mutants are not mentioned at all. Since both pvd reception mutants and phu+ mutants are correlated with fixation of the pvd production mutants, it is surprising that reception mutants were not discussed and compared with phu+ mutants.

The Phu+ mutants are found only when pyoverdine production has been lost from the population, i.e. when cheating is no longer an option. We have elaborated: “Knock-out of the pyoverdine receptor is, however, not expected to be a selective force in phu upregulation. Rather, the data suggest that both occur because pyoverdine is no longer available in the environment, selecting for privatisation of iron uptake and elimination of the costly receptor”. The presence or absence of pyoverdine receptor mutations have been added to Appendix 1—figure 3.

b) Marvig et al., has already identified the phu+ promoter mutation and the fact that they arise only in pvd production mutants, though their work was done on a smaller number of lineages. Therefore, it is not clear what is the added value of the current work? Is it the demonstration that phu+ alleles are only beneficial at low heme levels (see comments on the experiments in a separate point)? Is the claim for novelty in this work depends on the enlarged statistical power? Or is it the claim that phu+ promoter mutants arise only when pvd+ phenotype is completely eliminated from the social environment and not only from the specific lineage? See comments on sociality below for the last option. Altogether, the distinction between the current work and previous works of the group are not sufficiently clear.

See comment above. The current manuscript is clearly distinct from Marvig et al. in the hypotheses tested, which concern selective forces responsible for the observed changes, i.e. social interactions and not solely host adaptation. It is significantly less challenging to report the presence of a mutation than it is to present evidence for why a mutation has arisen, and why it has spread. This is what we do here.

3) Social interaction: It is hard to understand when do the authors discuss social environment within the given lineage (i.e., when discussing interaction between a mutant and its ancestor) and when do they discuss social environment based on multiple strains sampled from the same patient at the same time.

The social environment is characterized as the co-occurring genotypes within a clone type, i.e. mutants and ancestor, as described (“the social environment (the P. aeruginosa clonal lineage infecting a host)”) and now as further clarified in the first paragraph of the subsection “Isolate collections”. We focus on within clone-type events as continuous overlap between clone types within patients occurred rarely. The distinction has been further highlighted in the first paragraph of the subsection “Changes in pyoverdine production during infection” and Appendix 1—figures 1 and 2.

While the authors claim social explanations for their observation, much of the analysis in the text is not based on direct social observations, but on comparisons between a mutant and its ancestors (e.g., Figure 1, 2, 3). The two places where a social context is mentioned are:

a) When they discuss the social environment of pvd- cheats, the authors mention that 14 events are not fixed while 9 other are fixed (subsection “Evolution of iron acquisition systems in the lung – the sequence of events”).

b) Later, the authors mention that the phu+ mutations occurred significantly more when Pvd was lost, but here it is unclear whether they mean that Pvd was lost from the genome of the strains, or from the total social environment (subsection “Cheat invasion is associated with a subsequent switch to private iron uptake mechanism”), as they refer to Figure 2 which, to the best of my understanding, discussed only changes within a given genome and not the social context. On the other hand, they specifically mention that in all but three cases (I guess therefore that in 31-3=28 cases, but again – distinction between lineages and patients is unclear), phu+ arise when no pvd+ strains were observed at all.

In both cases there is insufficient information given on the social environment – how many strains were analyzed in any given sample. This data in neither presented nor analyzed in anyway whatsoever in the text. Now, this is obviously important – I would find the claim for fixation of a given genotype (or the elimination of another) to be much stronger if 10 isolates were sampled from the same patient at the same time than if only two samples were taken (or one!). Yet, the social data is completely absent, so I do not know how to interpret the claim for fixation of the cheater. As this is, to the best of my understanding, the most important point of the paper, the data supporting it should be presented and properly analyzed!

The number of samples and extent of sampling from each patient is available in Supplementary file 2. To further clarify the sampling regimen we have included Appendix 1—figures 1 and 2 showing sampling time for the clone types that were longitudinally sampled from patients, highlighting over-producers and non-producers from the children’s collection (Appendix 1—figure 1; Clone types that were only sampled once are not included in the analysis or the figure, l. 183-184). For the transmissible clone types both the sampling time across patients and phylogenies with selected pheno – and genotypes are shown (Appendix 1—figures 2 and 3).

During patient sampling sputum was collected and cultured, and 1-8 colonies were selected and stored for further analyses, as clarified in the first paragraph of the subsection “Isolate collections”. Therefore, the sequencing effort was partly determined by how the samples were originally collected, and partly by prioritizing extent of sampling, rather than depth, for the children’s collection. As mentioned in the introduction, a subsequent metagenomic study showed that the longitudinal samples represent the standing diversity well, and as we explain “The likelihood that we do not capture all co-infecting strains in our sample, leading to mis-classification of isolates as belonging to an environment without pyoverdine, will homogenise the sample groups and hence obscure any differences. Sampling errors of this kind will, therefore, hinder our ability to detect an effect (Type I error).” While more samples are of course always desirable, we feel very confident in the presented result, in particular because of the high degree of convergent evolution of the genotypes we focus on, across clone types and patients.

Throughout the text we have clarified that some clone types occur in multiple patients, while some patients are infected by multiple clone types (subsection “Isolate collections”, first paragraph; subsection “Changes in pyoverdine production during infection”, last paragraph; subsection “More cooperative populations are more vulnerable to exploitation”, second paragraph; Supplementary file 2), which may have caused confusion in the presentation of results on clonal lineages in patients.

We define samples as co-occurring if they are sampled < 1 year apart (subsection “Evolution of iron acquisition systems in the lung – the sequence of events”, last paragraph). That means that if a non-producer is sampled and a producer sampled < 1 year prior, or subsequently, then the social environment of that non-producer is considered to be cooperative. When no more producers are sampled, cheating is considered to have gone to fixation.

4) Pvd Super producers: the authors defined a subset of strains which produce more pvd than their ancestors (based on longitudinal sampling and genomic sequences). Super-production occurs in 50% of lineages and rather early (within ~2 years) in 33% of lineages. In a minority of super-producer lineages, the authors suggest that it is due to mutations in general regulators (e.g., lasR, mucA, gac). The authors suggest that this is an adaptation due to increased requirements for iron in the lungs. Finally, the author observed that a large fraction of super-producers became null producers (8/24) compared to strains that do not super-produce (4/21). We have several remarks on this part:

We do not state that super-producers become cheats, rather that cheats are more likely to be sampled following the observation of super-cooperators (subsection 2 More cooperative populations are more vulnerable to exploitation”, third paragraph). As the genetic basis for increased production is not identified, we do not know if a sampled cheat evolved from a super-cooperator or from an isolate with WT pyoverdine production. We have clarified this in the third paragraph of the subsection “More cooperative populations are more vulnerable to exploitation”.

a) If super-production results from mutations in general regulators such as lasR or mucA, then pvd over-production might be a pleiotropic side effect of selection for a different trait dependent on the global regulator. This option is not mentioned in the text.

We agree that over-production of pyoverdine may be a part of an overall “acute infection” phenotype. We have further highlighted in the text that this may be as a pleiotropic effect or beneficial trait (subsection “More cooperative populations are more vulnerable to exploitation”, second paragraph). Regardless of whether the selective force is iron uptake or host colonization, the effects of upregulation on social interaction may be the same.

b) Super-producers by themselves are also super-cooperators which will be cheated by their wild-type pvd producer ancestors in their social environment. Something needs to explain the increase of cooperation. Selection for private pleiotropic effects of the mutation (section a) might explain this increased cooperation. Again, social evolution is not discussed in this context.

We present three hypotheses for increased pyoverdine production, i) iron limitation, ii) inter-species iron competition, and iii) selection on an acute infection phenotype (subsection “More cooperative populations are more vulnerable to exploitation”, second paragraph). Organisms with WT level production of pyoverdine may also exploit super-cooperators but the benefit of exploitation increases with an increased difference in pyoverdine production. Hence, cheats will have a greater benefit exploiting super-cooperators than WT, and than WT exploiting super-cooperators. We have clarified this in the last paragraph of the aforementioned subsection. As such, this paragraph is all about social evolution.

c) The observation that pvd- mutant phenotype arise more often in super-producers (8/24) than in wild-type producers (4/21) seems to be rather marginal to me. I'm not sure what is the statistical significance analysis, but if one simply assumes binomial distribution around the mean (12/45=26%), then with 21 samples, the chance of obtaining 4 or less events is ~30% and the chances of obtaining 8 events or more in 24 samples is ~40%. Therefore, I am not convinced that this difference is not occurring simply by chance.

The results of the statistical analysis are presented in the text (subsection “More cooperative populations are more vulnerable to exploitation”, third paragraph) and graphically in the inverse Kaplan-Meier plot in Figure 1. We have used a Survival analysis to take into account the timing of events and not simply correlate presence / absence. This has been further clarified in the Materials and methods and Results and Discussion (subsection “Changes in pyoverdine production during infection”, last paragraph; subsection “Cheat invasion is associated with a subsequent switch to private iron uptake mechanism”, last paragraph).

5) Sensitivity to heme levels. If I understand correctly, the authors show in Figure 5 the average final OD for multiple related pairs which encode for the phu+ or phu-wt alleles (all in pvd- background) for different levels of heme (1, 2, 5, 10μM). I thought I got the picture clearly, but looking at the supplementary information and figure got me completely confused. The authors refer to six pairs of strains, for which they measured the growth in 2 and 5 μm of heme, but it is unclear on what was 1 and 10μm measured. Also, while Figure 5 indicates that the phu-wt allele has a growth advantage in 5uM, Appendix 1—figure 7 shows that all the statistically significant (or marginally significant) effects in specific pairs, go on the other direction – phu+ grow better than phu-wt. I couldn't figure out how to resolve these two values in a satisfying manner.

As stated in the text Figure 5 shows growth measurements of an isogenic pair of clinical isolates, differing only in the presence of a phuS//phuR mutation (subsection “Uptake of heme following phuS//phuR mutations”, first paragraph; subsection “Why is private heme uptake only upregulated if cooperation is lost?”, first paragraph). Here we see a clear difference in growth dependent on the heme availability. In Appendix 1—figure 4 we show measurements from pairs of clinical isolates that differ in the presence of phuS//phuR mutations (subsection “Uptake of heme following phuS//phuR mutations”, last paragraph). Although the isolates in a pair are the closest related we have, they have been sampled many years apart and differ by many additional mutations that likely affect growth (Appendix). These isolates overall show higher growth at 5 μm heme added, compared to at 2.5 μm for the isogenic pair. We have added text to highlight this difference (subsection “Why is private heme uptake only upregulated if cooperation is lost?”, last paragraph; Appendix). In the Methods we have switched the order of presenting the isogenic pair and the clinical isolates for clarity.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

Reviewer #2:

In this version, Andersen et al. tried to address the main criticism brought forward in the previous version. All in all, I find the data and figures added to the work to satisfy my main points of criticism. I still have few comments here and there, where I think analysis could be improved or representation could be simplified for the reader.

I would go over my previous main reservations and whether I am satisfied from the corrections done by the authors:

1) Connection between pvd receptor mutations and phu+ phenotypic change. The authors now added the analysis of pvd reception mutations and show a good correspondence between the phu+ promoter mutations and accumulation of pvd reception mutations. I think this supports the notion that cheating on pvd is not occurring upon the phu+ phenotypic change.

Yes, thank you for the suggestion. We agree that this has helped to clarify our results.

2) Social interactions. This is the main point of the paper – loss of pyoverdine production from the social environment and not only from a specific lineage explains the rise of phu+ phenotype. The authors now provide Appendix 1—figure 3A, B to directly show the data. I think this is a good start, but I would suggest that the authors would provide a more patient-centric view of the data and not lineage-centric. The problem with the current representation is that different lineages in the same patient are scattered throughout the two figures, complicating comparison. To look at the data of a specific patient, I had to scan the two figures to find all lineages sampled in this patient and then mentally order them by the sample time, to try and see whether the claim (pvd loss from all samples prior to phu+ mutation) can be verified. I suggest that in addition to the current representation, the authors should show a representation were the data samples for a single patient are clustered together with ascending years and the phenotypic/genetic status of each sample is clearly given (using the same mutation annotation as in the current figure). This would facilitate direct visual estimation of the social data in a given lineage. I don't think there is a need to show patients which have been sampled once with a single clone, as nothing much can be said about their social environment (many such patients are shown in Appendix 1—figure 2).

We agree with the reviewer that this is a useful way to present the data and have provided a patient-centric overview in Appendix 1—figure 4, showing the phenotypes and sampling time for the transmissible clone types DK1 and DK2. The two additional clone types that acquire phu mutations have also been included. We discuss the within-patient social environment in the Appendix text.

3) Super producers: the statistical claim of the author is that pvd cheaters are not only preferentially arising in the super-producers, but that they also arise more quickly (i.e., that the rate of appearance of the mutations is higher. This is clear from Figure 1 but is not mentioned appropriately in the text (subsection “More cooperative populations are more vulnerable to exploitation”, third paragraph). I think the authors should also mention that the mean time for appearance of a pvd- mutation was quicker for the eight mutations that arose in the presence of super producer than the mean time for the four that arose without it.

To clarify, the cleanest prediction that we can make is that cheats are more likely to arise in the presence of super-cooperators, compared to when they are absent. This is the prediction we test with our “survival analysis”. Predicting a difference in time before a mutation arises is more tenuous: what to compare exactly? Time since super-producer first arrives and appearance of a cheat, or time since first colonization and the appearance of a cheat in both cases? As we are interested in the potential effect of super-cooperators in the analysis, we compare the time since first sampling of super-cooperator and appearance of a cheat, to time from colonisation to cheat appearance in the absence of super-cooperators. The reviewer highlights that the difference in time axes on the figure can lead to the mistaken impression that cheats arise quicker in the presence of super-producers. In order to avoid misleading readers, we now explicitly state in the text that there is no significant difference in time to first appearance (subsection “More cooperative populations are more vulnerable to exploitation”, third paragraph), and in the Materials and methods we further clarify “If an over-producer was present, the length of time from that sampling to a non-producer was observed […]”. Thank you to the reviewer for pointing out that this would be helpful.

4) Number of generations. The reviewers have cut their estimates considerably, but still rely on an estimation of doubling time of ~70 minutes, which I still find unconvincing – it will either imply that the whole mocus volume of the lungs is replaced every 70 minutes (as in a chemostat, where growth is determined by clearance rate), or that there is a massive level of cell death alongside fast cell growth (and then the chemostat assumptions fail, as average growth does not reflect cell divisions). I am no expert in CF, but both things seem unlikely, though I might be completely wrong on that. In any case, I don't find this to be very important. I would just throw away this estimate, but if the authors choose to keep it, I would not object.

Our estimate for generation times is based on the experimental work by Yang et al., 2011 on the very isolate collection that we work on. It is, to our knowledge, the best available for the CF environment. This is highlighted in the legend (Supplementary file 1A legend).

https://doi.org/10.7554/eLife.38594.033

Article and author information

Author details

  1. Sandra Breum Andersen

    1. Department of Zoology, University of Oxford, Oxford, United Kingdom
    2. Novo Nordisk Foundation Center for Biosustainability, Technical University of Denmark, Lyngby, Denmark
    Present address
    Langone Medical Center, New York University, New York, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    sandrabreumandersen@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7030-040X
  2. Melanie Ghoul

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    Investigation, Methodology, Writing—review and editing
    Competing interests
    No competing interests declared
  3. Rasmus L Marvig

    Center for Genomic Medicine, Rigshospitalet, Copenhagen, Denmark
    Contribution
    Formal analysis, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5267-3173
  4. Zhuo-Bin Lee

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  5. Søren Molin

    Novo Nordisk Foundation Center for Biosustainability, Technical University of Denmark, Lyngby, Denmark
    Contribution
    Resources, Funding acquisition, Writing—review and editing
    Competing interests
    No competing interests declared
  6. Helle Krogh Johansen

    1. Department of Clinical Microbiology, Rigshospitalet, Copenhagen, Denmark
    2. Department of Clinical Medicine, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Contribution
    Resources, Funding acquisition, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0268-3717
  7. Ashleigh S Griffin

    Department of Zoology, University of Oxford, Oxford, United Kingdom
    Contribution
    Conceptualization, Supervision, Funding acquisition, Methodology, Writing—original draft, Project administration, Writing—review and editing
    For correspondence
    ashleigh.griffin@zoo.ox.ac.uk
    Competing interests
    No competing interests declared

Funding

Villum Fonden (95-300-13894)

  • Sandra Breum Andersen

Lundbeckfonden (r108-a10094)

  • Sandra Breum Andersen

Danmarks Grundforskningsfond (126)

  • Rasmus L Marvig

Rigshospitalet (R88-A3537)

  • Helle Krogh Johansen

Novo Nordisk (NNF12OC1015920)

  • Helle Krogh Johansen

Lundbeckfonden (R167-2013-15229)

  • Helle Krogh Johansen

H2020 European Research Council (SESE)

  • Ashleigh S Griffin

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Stuart West, Lars Jelsbak and Rolf Kümmerli for comments on the manuscript. Trine Markussen and Lars Jelsbak provided analysed sequences of additional DK1 isolates, and S M Hossein Khademi provided valuable discussion of the phu mutations. This work was supported by grants from Villum Fonden (grant 95-300-13894) and Lundbeckfonden (grant R108-2012-10094) to SBA, the European Research Council (grant SESE) to ASG, and Rigshospitalet (grant R88-A3537), Novo Nordisk Fonden (grant NNF12OC1015920) and Lundbeckfonden (grant R167-2013-15229) to HKJ.

Ethics

Human subjects: The CF isolates used for sequencing were obtained from the Department of Clinical Microbiology, Rigshospitalet, Copenhagen, where they had been isolated from CF patients treated at the Copenhagen CF clinic. The use of isolates was approved by the local ethics committee at the Capital Region of Denmark Region Hovedstaden: registration number H-1-2013-032. All patients have given informed consent. In patients under 18 years of age, informed consent was obtained from their parents. All patient isolates were anonymized prior to any experimental use or analysis.

Senior Editor

  1. Ian T Baldwin, Max Planck Institute for Chemical Ecology, Germany

Reviewing Editor

  1. Otto Cordero

Publication history

  1. Received: May 23, 2018
  2. Accepted: November 17, 2018
  3. Version of Record published: December 18, 2018 (version 1)

Copyright

© 2018, Andersen et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 741
    Page views
  • 94
    Downloads
  • 5
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Evolutionary Biology
    Christian Damsgaard et al.
    Research Article
    1. Evolutionary Biology
    2. Neuroscience
    Lucie Bastin-Héline et al.
    Research Article