SEARCH

SEARCH BY CITATION

Keywords:

  • experimental evolution;
  • inclusive fitness;
  • iron availability;
  • Pseudomonas;
  • public goods;
  • siderophore recycling;
  • siderophores

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Bacteria secrete a large variety of beneficial metabolites into the environment, which can be shared as public goods among producing bacteria, but also be exploited by nonproducing cheats. Here, we focus on cooperative production of iron-chelating molecules (siderophores) in the bacterium Pseudomonas aeruginosa to study how relevant ecological factors influence selection for cheating. We designed patch-structured metapopulations that allowed us introducing among-patch ecological variation. We found that cheating readily evolved in uniform iron-limited environments. This finding is explained by severe iron limitation demanding high siderophore-production efforts, which results in high metabolic costs accruing to cooperators, and thereby facilitates the spread of cheats. In contrast, we observed a significant reduction or even negation of selection for cheating in metapopulations where we introduced patches with increased iron availability and/or opportunities to recycle siderophores. These findings are compatible with the view that cheats are less likely to invade in environments that allow bacteria to reduce siderophore-production efforts, as this lowers the overall metabolic costs accruing to cooperators. Because we increased iron availability and siderophore recycling opportunities moderately, and only in some patches, our findings demonstrate that already-small local variations in ecological conditions as occurring in nature can significantly affect selection for public-goods secretion in microbes. In addition, we found that most (84.6%) of the evolved cheats were partially deficient for siderophore production and not loss-of-function mutants. Genetic considerations indicate that mutations leading to partial deficiency occur more frequent than mutations leading to loss of function, but also suggest that partially deficient mutants might often be the more competitive cheats.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Research over the last decade has revealed that microbes secrete a large variety of metabolites into their local environment (West et al., 2007). These metabolites include quorum-sensing molecules for communication (Bassler & Losick, 2006), polysaccharides to build biofilms (Flemming & Wingender, 2010), proteases to establish infections (Henderson et al., 2004), biosurfactants for swarming (Daniels et al., 2004) and metal-chelating molecules to bind insoluble trace elements (Waldron & Robinson, 2009). These metabolites constitute public goods that are costly to produce for the individual cell, but provide benefits to all individuals in the vicinity, producers and nonproducers alike (West et al., 2006). Evolutionary theory predicts that public-goods secretion is vulnerable to cheating by mutants that refrain from making the costly contribution, but still reap the benefits by using the public goods produced by others (West et al., 2006). Indeed, cheating has been observed under clinical and laboratory conditions for various types of public goods (De Vos et al., 2001; Rainey & Rainey, 2003; Greig & Travisano, 2004; Griffin et al., 2004; Harrison & Buckling, 2005; Diggle et al., 2007; Ross-Gillespie et al., 2007; Sandoz et al., 2007; Brockhurst et al., 2008; Harrison et al., 2008; Köhler et al., 2009; Kümmerli et al., 2009c; Ross-Gillespie et al., 2009; Jiricny et al., 2010; Bachmann et al., 2011; Nadell & Bassler, 2011; Wilder et al., 2011). The seemingly ubiquitous risk of exploitation by cheats raises the question what factors contribute towards maintaining these cooperative behaviours as it is often found in natural settings? Several studies have addressed that issue by showing that increased relatedness among interacting individuals (Griffin et al., 2004; MacLean & Brandon, 2008; Kümmerli et al., 2009a,b) and reduced costs of cooperation (Brockhurst et al., 2008, 2010; Harrison et al., 2008; Kümmerli et al., 2009c; Kümmerli & Brown, 2010; Xavier et al., 2011) can lower selection for cheating – findings that are consistent with inclusive fitness theory (Hamilton, 1964).

In this study, we implement relevant natural conditions into laboratory settings to better understand how multiple ecological factors jointly influence the cost of cooperation and selection for cheating during experimental evolution. We use siderophore production in the opportunistic human pathogen Pseudomonas aeruginosa, as our cooperative model trait. Siderophores are iron-chelating molecules, which are secreted into the local environment under iron-limited conditions to scavenge insoluble or host-bound iron (Ratledge & Dover, 2000; Miethke & Marahiel, 2007). Pseudomonas aeruginosa mainly relies on pyoverdine, its primary siderophore, to scavenge iron (Poole & McKay, 2003). Pyoverdine production is a cooperative trait because the production cost accrues to the individual cell, whereas the benefit can be shared among cells (Buckling et al., 2007). The specific amount of pyoverdine produced is extremely fine-tuned in response to environmental factors such as iron availability, cell density, and frequency of producers and nonproducers (Kümmerli et al., 2009c). At the molecular level, pyoverdine production is regulated by the sigma factor PvdS (Tiburzi et al., 2008), which triggers transcription of pyoverdine-synthesis genes under iron limitation, but is silenced by the ferric uptake regulator (Fur) when intracellular iron concentration is high (Escolar et al., 1999). Because pyoverdine production is regulated facultatively depending on its need, we hypothesize that factors influencing the required pyoverdine-production effort in a given environment will determine the overall metabolic costs accruing to cooperators and selection for cheating.

Our experimental design consisted of a metapopulation with 24 patches, where dispersal of bacteria occurred between patches (Fig. 1). This design simulates the fact that bacteria often live in structured environments (Rubin, 2007; Nadell et al., 2008). It further allowed us varying environmental conditions among patches, which simulates naturally occurring variation in habitat quality (Or et al., 2007). The first environmental factor we manipulated was iron availability. Spatial heterogeneity in iron availability has been demonstrated in aquatic environments (Norrström, 1995; Mioni et al., 2003) and in cystic fibrosis lungs (Stites et al., 1999; Reid et al., 2004; Moreau-Marquis et al., 2008), where P. aeruginosa causes chronic infections (Lyczak et al., 2000). As pyoverdine-production effort and therefore the overall costs accruing to cooperators are determined by iron availability (Tiburzi et al., 2008; Kümmerli et al., 2009c), we expected selection for cheats to be reduced in patches with increased iron availability. The second environmental factor we manipulated was the opportunity for bacteria to recycle pyoverdine. Pyoverdine recycling allows bacteria to benefit from cooperative acts performed by ancestors (Lehmann, 2007, 2010) and results in the down-regulation of the individual pyoverdine-production effort (Kümmerli & Brown, 2010). Consequently, we expected selection for cheats to be reduced in patches with increased pyoverdine-recycling potentials. Opportunities to recycle pyoverdine are likely to vary spatially due to variations in exposure to factors that limit pyoverdine reusability, such as fluxes or exposure to oxygen (Kümmerli & Brown, 2010). Importantly, we supplemented relatively low quantities of iron and pyoverdine, such that pyoverdine investment was still required in all patch types (Table 1). We started the experiment by seeding metapopulation with wild-type pyoverdine producers and followed the evolution of cheating over time. We tested the general prediction that in metapopulations, where lower pyoverdine-production efforts are required, cheats are less likely to invade. We further assessed the frequency of evolved clones that have retained wild-type levels of pyoverdine investment, have showed partial deficiency in pyoverdine production or have completely lost the ability to produce pyoverdine. Based on the genetic architecture of pyoverdine synthesis, we provide mutually nonexclusive proximate (mechanistic) and ultimate (fitness-related) explanations for our finding that partially deficient cheats occurred much more often than loss-of-function cheats.

image

Figure 1.  Experimental evolution of Pseudomonas aeruginosa pyoverdine-investment levels in metapopulations that differ in their ecology (see Table 1). In T1, all patches are highly iron limited, which requires high pyoverdine-production efforts. In T2, iron is less limited as half the patches (grey shading) receive 0.5 μm FeCl3, which lowers pyoverdine-production effort. In T3, half the patches (black shading) are supplemented with 15% pyoverdine from a previous round of growth, which creates increased pyoverdine-recycling opportunities, thereby lowering pyoverdine-production efforts. In T4, half the patches are either supplemented with 0.5 μm FeCl3 or 15% pyoverdine, conditions resulting in the lowest pyoverdine-production effort in our experiment. The experimental procedure involved the following steps: (a) 24-h growth period under static conditions at 37 °C; (b) mixing aliquots from each patch within a metapopulation in an Eppendorf tube to impose global dispersal; (c) transfer of cultures to fresh medium to repeat the cycle; (d) prepare aliquot of mixture for long-term storage at −80 °C; and (e) for analyses of samples following experimental evolution. Altogether, we carried out 20 serial transfers for each metapopulation type in three-fold replication.

Download figure to PowerPoint

Table 1.   Growth characteristics (mean ± SE) and relative pyoverdine-production efforts by wild-type Pseudomonas aeruginosa in the different patch (t1–t3) and metapopulation (T1–T4) types.
 Mean growth rate (μmean)*Population growth after 24 h (OD600 nm)Number of generations in 24 hRelative pyoverdine-production effort
  1. CAA, casamino acids.

  2. mean represents the mean growth rate between 4 h after inoculation (when cultures left lag phase) and 12 h (when the pyoverdine supplementation patch type reached stationary phase).

Patch type
 t1: CAA0.321 ± 0.0030.099 ± 0.0025.6 ± 0.031.00 ± 0.04
 t2: CAA + 0.5 μm FeCl30.388 ± 0.0030.138 ± 0.0016.1 ± 0.010.67 ± 0.01
 t3: CAA + 15% pyoverdine0.634 ± 0.0010.322 ± 0.0037.3 ± 0.010.28 ± 0.01
Metapopulation type
 T1 = t15.61.00
 T2 = (t1 + t2)/25.90.83
 T3 = (t1 + t3)/26.50.64
 T4 = (t2 + t3)/26.70.47

Material and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Culturing conditions

We cultured P. aeruginosa in 24-well plates, where each well represented an independent patch within a metapopulation (Fig. 1). All patches received 1.5 mL casamino acids (CAA) medium (CAA: 5 g casamino acids, 1.18 g K2HPO4·3H2O, 0.25 g MgSO4·7H2O, per litre). CAA is a mix of amino acids obtained through hydrolysis of the cow’s milk protein casein. The iron content of CAA has been estimated 2 μm (Höfte et al., 1993). To prevent nonsiderophore-mediated iron uptake, we added 100 μg mL−1 human apo-transferrin (Sigma), a strong natural iron chelator (Meyer et al., 1996). We further added 20 mm NaHCO3 (sodium bicarbonate) because bicarbonate is needed by transferrin as co-factor to bind iron efficiently (Schlabach & Bates, 1975).

We had four metapopulation types differing in whether patches received one of the following supplementations (Fig. 1): In metapopulation type T1, patches received no supplementation, which represents the standard laboratory conditions used in many of the previous experiments (e.g. Griffin et al., 2004; Harrison & Buckling, 2005; Ross-Gillespie et al., 2007; Kümmerli et al., 2009a). This metapopulation type served as control with which we compared the results from the other three metapopulation types. In metapopulation type T2, we manipulated heterogeneity in iron availability by supplementing 12 patches with 0.5 μm FeCl3 (i.e. increasing iron availability by 25%). In metapopulation type T3, we manipulated heterogeneity in pyoverdine-recycling opportunities by supplementing 12 patches with pyoverdine (15% of the amount that was produced within 24 h by bacteria from a previous round of growth), which simulates environmental conditions that allow bacteria to reuse pyoverdine produced by ancestors (Kümmerli & Brown, 2010). In metapopulation type T4, we studied the combined effects of increased iron availability and pyoverdine reusability by supplementing 12 patches with 0.5 μm FeCl3 and the other 12 patches with 15% pyoverdine.

To quantify the required pyoverdine-production efforts in the three patch types, we inoculated approximately 106 cells of the wild-type P. aeruginosa strain PAO1 (ATCC 15692) into 1.5 mL of the respective media in eight-fold replication. Following static incubation for 24 h at 37 °C, we quantified pyoverdine, which is a green fluorescent molecule, as relative fluorescence units (RFU) through excitation at 400 nm and emission measures at 460 nm (Ankenbauer et al., 1985) with a multimode microplate reader (Infinite 200 PRO, Tecan, Switzerland). To obtain a proxy for the average individual pyoverdine-production effort, we calculated RFU/OD (Kümmerli et al., 2009c), where optical density (OD) was measured at 600 nm. These measurements showed that compared with the standard CAA medium, supplementing 0.5 μm FeCl3 decreased required pyoverdine-production effort by 33%, whereas the addition of 15% pyoverdine decreased required pyoverdine-production effort by 72% (Table 1). At the metapopulation level, the supplementation regimes resulted in an average decrease in pyoverdine-production efforts by 17% in T2, 36% in T3 and 53% in T4, compared with T1 (Table 1). We further assessed growth differences in the three patch types and found that growth rates, yield and therefore the number of bacterial generations increased with iron and pyoverdine supplementations (Table 1). These growth differences most likely resulted in variations in the number of generations the different metapopulation types were subjected to experimental evolution. However, this was unproblematic because we were expecting faster spread of cheating mutants in treatments with fewer generations.

To obtain pyoverdine for supplementation in a given round (ri), we inoculated evolved bacteria from two rounds in the past (ri-2) in 6 mL CAA for 24 h at 37 °C, conditions that matched the ones used during experimental evolution. We then centrifuged the culture for 10 min at 16060 g to separate cells (in the pellet) from pyoverdine (in the supernatant). We filter-sterilized the supernatant with a 0.22-μm filter and added 150 μL of the supernatant to the respective patches in T3 and T4. We used pyoverdine from ri-2 and not from ri-1 due to the time required to prepare pyoverdine supernatant. Note that we did not directly transfer supernatant containing pyoverdine from round to round because we aimed at adding fresh pyoverdine in every round, as otherwise degraded pyoverdine would accumulate over evolutionary time. Finally, we allowed the absolute amount of pyoverdine supplementation to evolve because it reflects the biologically realistic scenario that when pyoverdine production is selected against, lower quantities can be transmitted to subsequent generations.

Experimental evolution

At the beginning of the evolution experiment, we inoculated each patch in a metapopulation with approximately 106 cells of the wild-type pyoverdine producer strain PAO1. After a 24-h growth period, we mixed 50 μL culture from each patch within a metapopulation in a 2-mL Eppendorf tube (Fig. 1). We supplemented one part of this mix with glycerol (to a final concentration of 25% glycerol) for long-term storage and subsequent analysis at −80 °C. From the other part of the mix, we transferred 15 μL (corresponding to a 100-fold dilution) to each of the 24 wells of a new metapopulation plate containing 1.5 mL fresh CAA. This mixing prior to dispersal allowed mutants appearing in one patch to easily spread to other patches within the metapopulation. Hence, once cheating mutants appeared, this mode of dispersal led to low relatedness among individuals within patches. We allowed low relatedness to evolve in our design because we sought to focus on how the required pyoverdine-production effort, and thereby the overall costs accruing to cooperators, influences selection for cheating by ruling out the positive effects that increased relatedness can have on the evolution of cooperation (Griffin et al., 2004; Kümmerli et al., 2009a). Prior to every transfer, we randomly allocated patches to the different supplementation regimes incorporated in T2, T3 and T4. Altogether, we carried out 20 rounds of growth allowing for approximately 120 generations of experimental evolution, in three-fold replication for each metapopulation type. The number of rounds was determined prior to experimentation and was based on experience (Harrison et al., 2008).

Measuring pyoverdine investment at the metapopulation level

Following experimental evolution, we tested whether evolved cultures showed alterations in their ability to produce pyoverdine compared with the ancestral wild type. To obtain comparable measures across metapopulation types, we tested this ability in iron-limited CAA media, which provides data on the maximum ability to produce pyoverdine, and not the level that is necessarily produced in the metapopulation types where the bacteria evolved. Evolution of significantly reduced pyoverdine-investment levels would indicate the evolution of cheating. We simultaneously assessed pyoverdine investment of evolved cultures from all rounds and metapopulation types. We first inoculated evolved cultures from freezer stocks in wells on a 96-well plate with King’s B medium and incubated static overnight at 37 °C. Under these conditions, cultures grew to similar optical densities (OD600 nm = 1.323 ± 0.015; mean ± SE). We then added approximately 105 cells to 200 μL iron-limited CAA media (i.e. supplemented with transferrin and NaHCO3 as described previously) and let cultures grow for 24 h at 37 °C under static conditions in three-fold replication for each round and metapopulation type. After 24 h, we measured population growth (OD) and pyoverdine investment (RFU/OD) as described above.

Measuring pyoverdine investment of evolved clones

To characterize variation in pyoverdine-production abilities among evolved clones within metapopulations, we isolated and analysed evolved clones from all replicates and metapopulation types. We first plated evolved cultures from every other round (R2, R4, R6, R8, R10, R12, R14, R16, R18 and R20) onto KB plates and incubated them overnight at 37 °C. We then randomly chose 20 clones from each plate and inoculated them separately into iron-limited CAA medium. After 24-h growth at 37 °C, we assessed the growth and pyoverdine investment of each single clone in isolation using the multimode microplate reader as described previously. Overall, we analysed 2359 evolved clones and compared their pyoverdine investment with clones of the ancestral wild type. We allocated individual clones to eight different categories depending on their pyoverdine investment relative to the ancestral wild type (0–10%, 11–30%, 31–50%, 51–70%, 71–90%, 91–110%, 111–130% and more than 130% of wild-type pyoverdine investment). We defined phenotypes producing < 10% pyoverdine compared with the ancestral wild type as loss-of-function mutants, whereas we considered phenotypes producing < 70%, but more than 10%, pyoverdine compared with the ancestral wild type as partially deficient mutants. There are two practical reasons for using these thresholds. First, pyoverdine investment varies considerably across replicates even in an isogenic strain (i.e. PAO1 pyoverdine investment generally varies between 70% and 130%), such that only phenotypes with < 70% pyoverdine investment can reliably be considered as cheats. Second, pyoverdine-investment measures based on RFU give slightly positive values even for strains where pyoverdine-production genes are knocked out, with values typically ranging between 0% and 5% of the wild-type RFU (Kümmerli et al., 2009c). This background fluorescence is most likely due to other fluorescent components in cells or medium, such that a 10% threshold was meaningful to conservatively take this into account.

Statistical analysis

We used linear models (LM) and linear mixed models (LMM) to test whether growth and pyoverdine investment changed over time and between metapopulation types. We treated growth and pyoverdine investment as dependent variables, and round number and metapopulation type as independent variables. Whenever appropriate, we implemented round number ID or clone ID as random factors to consider that measures from consecutive rounds and clones from the same round are not independent samples. We carried out and compared analyses with data from the metapopulation and clonal level. Such a comparison was important to verify whether changes in pyoverdine investment at the metapopulation level were based on the presence of evolved mutants, and not simply due to phenotypically plastic responses. With the clonal data, we further used an extended Fisher’s exact test to compare clone frequencies in the different pyoverdine-investment categories between metapopulation types. For pairwise comparisons between metapopulation types, we applied the false discovery rate control method (Benjamini & Hochberg, 1995) to adjust the nominal α = 0.05. All analyses were carried out with R 2.13.0 (http://www.r-project.org/ ).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Evolution of pyoverdine investment

We found a significant positive relationship between the reduction in the required pyoverdine-production efforts, induced by our experimental treatments, and the evolved pyoverdine-investment levels (Fig. 2, LM for metapopulation level: t11 = 4.49, = 0.0009; LMM for clonal level: t167 = 12.57, < 0.0001). This shows that selection for cheating was lower in environments with reduced pyoverdine-production efforts.

image

Figure 2.  Significant positive relationship between the experimentally induced reductions in pyoverdine-production efforts and the evolved pyoverdine-investment levels after 120 generations of experimental evolution in Pseudomonas aeruginosa metapopulations. This finding shows that selection for cheating was reduced in environments with lower pyoverdine-production efforts: analysis (a) at the metapopulation and (b) at the clonal level. Metapopulation types (T1–T4) differed in their ecology, which affected pyoverdine-production effort (see Table 1). In T1, all patches were highly iron limited. In T2, iron was less limited as half the patches received 0.5 μm FeCl3. In T3, half the patches were supplemented with 15% pyoverdine from a previous round of growth, which created increased opportunities to recycle pyoverdine. In T4, half the patches were either supplemented with 0.5 μm FeCl3 or 15% pyoverdine. Values of the evolved pyoverdine-investment levels were assessed in iron-limited casamino acids media following experimental evolution, with all values being scaled relative to the ancestral wild type (dashed line).

Download figure to PowerPoint

When looking at each metapopulation type independently, we found that pyoverdine-investment levels significantly decreased during the course of evolution in T1 (LMM: t18 = −5.7, < 0.0001) and T2 (t18 = −7.16, < 0.0001), but remained unaltered in T3 (t18 = 0.33, = 0.74), and even significantly increased in T4 (t18 = 2.28, = 0.035) (Fig. 3a, for analysis at the metapopulation level). These findings were mirrored at the clonal level (Fig. 3b) where, at the end of the experiment, clones from T1 and T2 had significantly reduced pyoverdine-investment levels compared with the ancestral wild type (T1: 0.44 ± 0.05: t32 = −11.51, < 0.0001; T2: 0.66 ± 0.06: t38 = −6.24, < 0.0001), whereas investment levels did not change (0.99 ± 0.01: t40 = −1.12, = 0.27) for clones in T3, and weakly but significantly increased (1.07 ± 0.03: t39 = 2.55, = 0.015) for clones in T4.

image

Figure 3.  Evolution of pyoverdine-investment levels in experimental Pseudomonas aeruginosa metapopulations. For analysis at the metapopulation (a) and clonal (b) level, pyoverdine-investment levels significantly decreased in metapopulation type T1 (where all patches were highly iron limited; filled circles), and in T2 (where half the patches received 0.5 μm FeCl3; open squares), but not in T3 (where half the patches received 15% of the pyoverdine from a previous round of growth; filled triangle), and even weakly but significantly increased in T4 (where half the patches were either supplemented with 0.5 μm FeCl3 or 15% pyoverdine; open diamonds). Values of the evolved pyoverdine-investment levels were assessed in iron-limited casamino acids media following experimental evolution, with all values being scaled relative to the ancestral wild type (dashed line).

Download figure to PowerPoint

In an across-treatment comparison, evolved pyoverdine-investment levels at the end of the experiment were significantly lower in T1 and T2 than in T3 and T4 (LM, T1 vs. T3: t8 = 3.92, = 0.004; T1 vs. T4: t8 = 3.74, = 0.006; T2 vs. T3: t8 = 2.89, = 0.020; T2 vs. T4: t8 = 2.71, = 0.027), but not significantly different between T1 and T2 (t8 = 1.03, = 0.33) and T3 and T4 (t8 = 0.18, = 0.87) (Fig. 3a, for analysis at the metapopulation level). These findings partially contrast with the analysis at the clonal level, where evolved pyoverdine investment significantly differed between any two metapopulation types (LMM, pairwise comparisons: 2.05 ≤ t165 ≤ 11.10,  0.042) (Fig. 3b). The discrepancy is likely explained by the greater statistical power in the later analysis.

Evolution of growth

Theory and previous studies showed that increased prevalence of pyoverdine-deficient mutants went along with decreased population growth (Harrison & Buckling, 2005; Ross-Gillespie et al., 2007). The explanation for this pattern is that with fewer pyoverdine producers, less iron will be made available for growth. However, our data reveal mixed support for this hypothesis (Fig. 4). Although there was a significant positive relationship between evolved pyoverdine-investment levels and growth at the metapopulation level (LM: t11 = 2.57, = 0.026), this was due to only one replicate in T1, in which pyoverdine investment collapsed completely (Fig. 4a). In contrast, analysis at the clonal level showed no significant relationship between evolved pyoverdine investment and growth (LMM: t167 = −0.04, = 0.972, Fig. 4b). In fact, in 10 of the 12 replicates, evolution of increased growth was observed (Fig. 4a), suggesting that mutations occurred that allowed improved performance in CAA medium. Rapid adaptation to laboratory culturing conditions is a common phenomenon in experimental evolution with bacteria (e.g. Barrick et al., 2009). Improved growth could, for example, be caused by mutations that accelerate amino acid uptake rates (CAA is a mix of amino acids). Alternatively, mutations leading to the loss of the ability to degrade more complex nutrients such as proteins could be beneficial because they would abort investment into metabolic pathways not needed in CAA.

image

Figure 4.  Relationship between evolved pyoverdine-investment levels and growth after 120 generations of experimental evolution. (a) Significant positive relationship at the metapopulation level, which was due to one replicate (bottom left); (b) no significant relationship at the clonal level. Values of the evolved pyoverdine-investment levels and growth were assessed in iron-limited casamino acids media following experimental evolution, with all values being scaled relative to the ancestral wild type (dashed lines).

Download figure to PowerPoint

Characterization of evolved clones

After 20 rounds (approximately 120 generations) of experimental evolution, we found that 60.2% of all clones in T1 and T2 showed reduced pyoverdine investment (i.e. < 70% compared with the ancestral wild type) (Fig. 5a,c). However, only 15.4% of these were loss-of-function mutants, and they all occurred in the same one replicate in T1. In contrast, the great majority (84.6%) of mutants showed partial deficiency for pyoverdine production. While pyoverdine phenotype frequency did not change in T3 compared with the ancestral wild type, there was a considerable proportion (18.3%) of clones with increased pyoverdine investment (i.e. more than 130% compared with the ancestral wild type) in T4. When following phenotypes over time, we found that pyoverdine under- and over-producers were already present after 10 rounds of experimental evolution in the respective metapopulation types (Fig. 5b), albeit at relatively low frequency, whereas after 20 rounds, they substantially increased in frequency (Fig. 5c). At the end of the experiment, the phenotype frequency varied significantly between any two metapopulation types (pairwise comparisons with the extended Fisher’s exact test: probability ranges between < 0.0001 and = 0.029), confirming our previous results of significant differences in the selection for cheating between any two metapopulation types.

image

Figure 5.  Pyoverdine-investment levels of (a) 20 ancestral wild-type clones; (b) 240 evolved clones after 10 rounds (approximately 60 generations) of evolution; (c) 228 clones after 20 rounds (approximately 120 generations) of evolution in four different metapopulation types (T1–T4), which differed in their ecology. In T1, all patches were highly iron limited. In T2, iron was less limited as half the patches received 0.5 μm FeCl3. In T3, half the patches were supplemented with 15% pyoverdine from a previous round of growth, which created increased opportunities to recycle pyoverdine. In T4, half the patches were either supplemented with 0.5 μm FeCl3 or 15% pyoverdine. Values of the evolved pyoverdine-investment levels were assessed in iron-limited casamino acids media following experimental evolution, with all values being scaled relative to the ancestral wild type. The different shadings correspond to wild-type clones (white), clones from replicate A (light grey), replicate B (dark grey) and replicate C (black).

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

By using experimental evolution, we show that ecological factors, which diminish the required pyoverdine-production effort, significantly contribute to the evolutionary maintenance of this cooperative behaviour. In patch-structured metapopulations, it was sufficient to make iron less stringently limited and/or increase the opportunity for pyoverdine recycling in some patches to significantly reduce or even negate selection for exploitative cheats. This leads to the seemingly paradoxical result that pyoverdine production is most strongly selected against under conditions where it is needed most, but is evolutionarily more stable in environments with reduced pyoverdine requirement. This paradox is, however, fully compatible with inclusive fitness theory (Hamilton, 1964), which predicts that Hamilton’s rule for the evolution of cooperation, rb c, is unlikely to be satisfied when relatedness (r) among pyoverdine producers and nonproducers is low, as occurring in our experiment in liquid culture where bacteria were motile within patches and could disperse between patches, and when the overall cost (c) of cooperation relative to the generated benefit (b) is high, as occurring when severe iron limitation demands extensive pyoverdine-production efforts.

Our findings show that the supplementation of relatively small amounts of iron to only half of the patches within a metapopulation already resulted in a reduction in the required pyoverdine-production effort and significantly lowered selection for cheating. Obviously, supplementing higher concentrations of iron to all patches would have affected pyoverdine-production effort and selection for cheating more dramatically (see Harrison et al., 2008; Kümmerli et al., 2009c). However, the aim of our study was to investigate whether relatively little local variation in iron concentration can significantly affect selection for public-goods secretion in microbes. This seems important because environmental heterogeneity occurs at the microscale and is ubiquitous (Or et al., 2007). Representatively, local heterogeneity in iron concentrations has been observed in lake water (Mioni et al., 2003), soil water (Norrström, 1995) and among different locations in the cystic fibrosis lung (Stites et al., 1999; Reid et al., 2004; Moreau-Marquis et al., 2008). This heterogeneity presumably determines local siderophore requirements as simulated in our experiment. The same argument can be developed for the reusability of siderophores (or any other public goods). Apart from biochemical stability, the reusability potential of public goods is determined by its probability to stay in the local environment. This probability can, for example, be decreased by flux that washes the public goods away from the producer or by the presence of micro-organisms that feed upon the public goods produced by others (Kümmerli & Brown, 2010). Fluxes are ubiquitous in most habitats with local variations in water flux strength occurring in soils (Or et al., 2007). This suggests that the proportion of public-goods molecules that can be reused is probably often small and varies locally, conditions we simulated in our experiment. In addition, other studies have identified further ecological factors that also impacted bacterial public-goods cooperation. For example, increased resource availability has been found to reduce selection for siderophore cheating in P. aeruginosa (Brockhurst et al., 2008). Similarly, increased resource availability and intermediate disturbance rates both decreased selection for cheating in Pseudomonas fluorescence biofilm formation (Brockhurst et al., 2007, 2008, 2010). These factors are likely to have similar effects on public-goods-production efforts and selection for cheating as found in our study. Altogether, although our data show that the patchy occurrence of factors reducing public-goods-production efforts is sufficient to decrease selection for cheating, it is important to note that no conclusions can be drawn on how strong a patchy, as opposed to a uniform, resource distribution affects the maintenance of cooperation. Experiments disentangling the total resource input and heterogeneity would be needed to address this question.

Because pyoverdine production was required in all patch types (Table 1), it was surprising to find no detectable selection for cheating in T3 and T4. Specifically, we expected cheats to experience fitness benefits (although differing in their magnitude) in all patch types, with metapopulation only differing in how fast cheats can spread. As we observed high frequencies of cheating mutants in all replicates of T1 and T2, the metapopulations with the fewest number of generations (Table 1), it seems unlikely that there was insufficient time for cheats to rise to significant levels in T3 and T4. More likely is the explanation that the significant reduction in the required pyoverdine-production efforts could have contributed towards satisfying Hamilton’s rule in these environments, such that cheating mutants could not spread. This, however, implies that relatedness among interacting individuals must have been somewhat greater than zero. This would have been possible if some form of population structure was established during static growth within patches, such that a higher proportion of the cooperative benefit accrued to cooperating relatives than to cheating mutants. Even more surprising was our finding that some evolved clones from T4 showed pyoverdine over-production when transferred back to iron-limited media. A possible explanation is that these mutants have lost a regulatory element controlling pyoverdine production under severe iron limitation (conditions bacteria did not experience in T4), which then resulted in the phenotypic up-regulation of pyoverdine investment in iron-limited CAA media.

Our data from the clonal analysis indicate that partially deficient mutants were much more frequent (85%) than loss-of-function mutants in T1 and T2. This is an unexpected result because a previous study showed a positive relationship between the level of deficiency in pyoverdine production and the fitness of cheats (Jiricny et al., 2010), such that we expected loss-of-function cheat to spread most successfully. We can offer a proximate and an ultimate explanation for this observation, which are mutually nonexclusive. At the proximate level, pyoverdine is built via nonribosomal peptide synthesis (Visca et al., 2007), which means that the pyoverdine locus does not code for the pyoverdine molecule itself, but for a number of enzymes (nonribosomal peptide synthetases and accessory enzymes) that are involved in synthesizing pyoverdine (Visca et al., 2007). The expression of these genes is regulated by the sigma factor PvdS (Tiburzi et al., 2008). At least three types of mutations can lead to deficiency in pyoverdine production, which are (i) point mutations in one of the pyoverdine genes or in its regulatory sequences – mutations that often lead to partial deficiency in pyoverdine production (Wilson et al., 2001); (ii) point mutations in the sequence coding for PvdS – mutations that also often lead to partial deficiency in pyoverdine production (Wilson & Lamont, 2006); (iii) insertions/deletions in the pyoverdine locus, which ultimately leads to a loss of pyoverdine-production ability (Tiburzi et al., 2008). Hence, if point mutations occur more frequently than insertion/deletions, then this could explain our data. In support of this view, whole-genome mutation analysis in P. aeruginosa revealed more (63%) nonsynonymous point mutations than (29%) insertion/deletion mutations (Smith et al., 2006).

At the ultimate level, partially deficient mutants could be more common because they have higher relative fitness than loss-of-function mutations. In support of this, we previously found that constructed pyoverdine knockouts, which are deletion mutations (Ghysels et al., 2004), perform relatively poorly even in iron-rich media most likely because they suffer from pleiotropic effects (Kümmerli et al., 2009c). Furthermore, a deletion in one pyoverdine gene only entirely interrupts pyoverdine synthesis if the mutant grows in monoculture. In contrast, when growing with the wild type, all non-knocked-out genes involved in pyoverdine synthesis are normally transcribed in the mutant, due to a positive feedback loop associated with pyoverdine uptake (Tiburzi et al., 2008). Consequently, deletion mutants in co-culture with the wild type experience considerable metabolic costs, which might slow down their spread. In contrast, point mutations within the PvdS coding sequence often result in mutants that have a modified sigma factor (Wilson & Lamont, 2006). A modified PvdS simultaneously reduces transcription of all genes at the pyoverdine locus, thereby likely leading to a partially deficient, highly competitive cheat.

In conclusion, our work contributes towards understanding how ecological factors determine public-goods-production effort and thereby influence selection for or against public-goods secretion in microbes. Furthermore, our findings are relevant for bacterial pathogenicity as many public-goods traits are involved with virulence, and it has been suggested that evolutionary dynamics between public-goods producers and cheats could be used for medical purposes (Brown et al., 2009). Specifically, Brown et al. (2009) suggested that engineered cheats, harbouring a drug-susceptible construct, could be used to first replace pathogenic strains within hosts (due to cheating) and then be targeted by medical intervention (due to their drug susceptibility). However, our data indicate that such a Trojan cheater invasion might not be possible across a wide range of ecological conditions and that partially deficient public-goods cheats might be more successful in doing so than knockout cheats.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

We thank Fred Inglis and four anonymous reviewers for their constructive comments. This work was funded by an Ambizione grant from the Swiss National Science Foundation and a Marie-Curie Reintegration grant (No. 256435) by the European Commission.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Data deposited at Dryad: doi: 10.5061/dryad.2f4g2h90