Antagonistic coevolution across productivity gradients: an experimental test of the effects of dispersal


Laura Lopez-Pascua, Department of Biology and Biochemistry, University of Bath, Bath, Somerset BA2 7AY, UK.
Tel.: 00 44 (0)1225383382; fax: 00 44 (0)1225383382;


Coevolution commonly occurs in spatially heterogeneous environments, resulting in variable selection pressures acting on coevolving species. Dispersal across such environments is predicted to have a major impact on local coevolutionary dynamics. Here, we address how co-dispersal of coevolving populations of host and parasite across an environmental productivity gradient affected coevolution in experimental populations of bacteria and their parasitic viruses (phages). The rate of coevolution between bacteria and phages was greater in high-productivity environments. High-productivity immigrants (∼2% of the recipient population) caused coevolutionary dynamics (rates of coevolution and degree of generalist evolution) in low-productivity environments to be largely indistinguishable from high-productivity environments, whereas immigration from low-productivity environments (∼0.5% of the population) had no discernable impact. These results could not be explained by demography alone, but rather high-productivity immigrants had a selective advantage in low-productivity environments, but not vice versa. Coevolutionary interactions in high-productivity environments are therefore likely to have a disproportionate impact on coevolution across the landscape as a whole.


Antagonistic host–parasite coevolution, the reciprocal evolution of host defence and parasite counter-defence is pervasive in biological communities and thought to have a range of important ecological and evolutionary consequences, such as driving population dynamics (Thompson, 1998; Buckling & Hodgson, 2007), the evolution of diversity (Frank, 1993) and the evolution of parasite virulence (Anderson & May, 1982; Bull, 1994). Populations of hosts and parasites are often spatially structured, consisting of a set of patches connected by limited migration, and such spatial structuring is predicted to profoundly affect the coevolutionary process (Gandon et al., 1996; Gomulkiewicz et al., 2000; Nuismer et al., 2000, 2003; Thompson, 2005; Morgan et al., 2005; Gavrilets & Michalakis, 2008). The geographic mosaic theory of coevolution (GMT) (Thompson, 2005) suggests that variation in ecological conditions between patches can lead to differences in local selection pressures (selection mosaics), resulting in some patches where reciprocal selection is relatively strong (coevolutionary ‘hotspots’), whereas in others reciprocal selection is relatively weak or nonexistent (‘coldspots’). Gene flow between patches can then significantly alter local coevolutionary dynamics (Thompson, 2005; Nuismer et al., 2000; Gomulkiewicz et al., 2000; Nuismer et al., 2003; Thompson, 2005).

Environmental productivity is a ubiquitous abiotic variable (Rosenzweig, 1995) that is likely to generate variation in the strength of reciprocal selection between patches (Hochberg & van Baalen, 1998; Lopez-Pascua & Buckling, 2008), although other abiotic [such as climate (Toju, 2008)], and biotic factors [such as host-parasite encounter rate (Brockhurst et al., 2003; Laine, 2006; Lopez-Pascua & Buckling, 2008)], and the presence or absence of other interacting species (Benkman, 1999; Zangerl & Berenbaum, 2003) can also generate selection mosaics. High productivity is likely to result in high levels of reciprocal selection because of increasing host, and hence parasite, population densities (Hochberg & van Baalen, 1998). This in turn is predicted to increase investment into resistance and defence (if these traits are quantitative), or increase the degree of generalism where resistance and infectivity are qualitative traits in gene for gene (GFG) models of coevolution (Hochberg & van Baalen, 1998).

Migration across selection mosaics can lead to the domination of patches by either ‘hotspot’ or ‘coldspot’ adapted genotypes, depending on the strength of selection acting on at least one of the interacting species (Gomulkiewicz et al., 2000; Nuismer et al., 2003). Experiments with laboratory populations of bacteria and phages provide some empirical support for these theoretical predictions. Specifically, in populations of Pseudomonas fluorescens and phage SBW25Φ2, where the strength of reciprocal selection was determined by the bacteria–phage encounter rate, immigration from hotspots accelerated coevolution in coldspots, whereas immigration from coldspots decelerated coevolution in hotspots (Vogwill et al., 2009).

The impact of migration across selection mosaics generated by variation in productivity is, however, predicted to be asymmetrical, with high-productivity patches having a greater impact on coevolutionary dynamics in low-quality patches, partly because high-productivity patches produce more migrants (Hochberg & van Baalen, 1998). Consistent with this prediction, resistance of Escherichia coli and phage T7 evolved more readily under high environmental productivity, and migration from high to low environmental productivity accelerated coevolution and increased the levels of resistance and infectivity in low-productivity patches (Forde et al., 2004, 2007). However, migration from low to high productivity has not been investigated. In this paper, we explicitly test the prediction that migration from high-productivity patches should have a greater impact on coevolution in low-productivity patches than vice versa, using coevolving populations of P. fluorescens and phage SBW25Φ2 (Buckling & Rainey, 2002).

Materials and methods

Selection experiment

Bacteria and phage were cultured in 6 mL of M9 salt solution supplemented with differing amounts of glycerol and proteose peptone [0.01 and 1 times the standard concentration for King’s Media B (KB) of 10 g L−1 glycerol and 20 g L−1 of proteose peptone no.3; low and high productivity, respectively] (Lopez-Pascua & Buckling, 2008), in 25 mL glass universals with loose plastic lids (microcosms). The high-productivity media has an approximately 10-fold greater carrying capacity in the absence of phages than the low-productivity media under these experimental conditions (Lopez-Pascua & Buckling, 2008). We employed four different experimental treatments: high and low productivity with no immigration, and high and low productivity with immigration from independent source populations evolved under the opposite productivity regime. The ‘no immigration’ treatments acted as the source populations for the immigration treatments. Six replicate microcosms for each treatment were inoculated with 108 bacterial cells [derived from a single P. fluorescens SBW25 clone grown overnight in King’s Media at 28 °C, in an orbital shaker at 200 rpm (0.9 g)], and 105 clonal phage particles, obtained from a single plaque of a clone of phage SBW25φ2 (Lopez-Pascua & Buckling, 2008). The bacteria and phage were then allowed to coevolve in a static incubator at 28 °C for a 48 h time period before 1% of the total population was transferred into a fresh microcosm, following vortex mixing to homogenize the culture. Migration was carried out immediately afterwards by transferring 1% of each source population to the recipient populations. To keep the nutrient ratios as constant as possible in treatments with immigration from the high to the low treatment, the M9 salt solution was not supplemented with glycerol or peptone (except when the microcosms were first established), as a 1% migration from the high treatment would result in the correct concentration for the low (1%) productivity treatment. This process was repeated for a total of 16 transfers (approximately 120 generations). Every second transfer, samples (600 μL) of the total populations were frozen at −86 °C in 20% v : v glycerol : KB solution, and a sample of phage isolated from bacteria by adding 100 μL of chloroform to 900 μL of the culture, and centrifuging at 13 000 rpm for 3 min, to lyse and pellet bacteria (Buckling & Rainey, 2002).

Estimating bacteria and phage densities

To determine approximate number of migrants, we measured densities of bacteria in the source populations at every second transfer. Densities of bacteria were estimated from the number of colony forming units on KB agar after 48 h growth at 28 °C. Phage densities were estimated from the number of plaque forming units produced on a soft (0.6%) agar lawn containing exponentially growing ancestral P. fluorescens SBW25 after 24 h incubation at 28 °C (Buckling & Rainey, 2002). Note that these phage density estimates were probably quite inaccurate because the very small plaques of some evolved phages made them difficult to count.

Measuring resistance and infectivity

Both of the subsequent measures, rates of coevolution and resistance/infectivity ranges, employed the following methodology: resistance of a bacterial population against a given phage population was determined by streaking 20 independent bacterial colonies across a perpendicular line of phage (20 μL) that had previously been streaked and dried onto a KB agar plate. A colony was defined as resistant if there was no inhibition of growth; otherwise, it was defined as sensitive. Resistance of bacterial populations was measured as proportions, and infectivity of phage populations as (1 – proportion resistant bacteria).

Measuring coevolution

It is possible to estimate the extent of short-term phage infectivity evolution between two time points by measuring the resistance of a bacterial population from an intermediate time point to the phage populations from before and after. Specifically, we measured the resistance of a population of bacteria to phages from two transfers in the past and two transfers in the future. For example, the resistance of bacteria from transfer four was determined against phages from transfers two and six. The difference in resistance provides a measure of phage infectivity evolution over this four transfer period. There is a highly significant positive correlation between rates of infectivity and resistance evolution in this experimental system, hence rates of infectivity evolution correlate with the rate of coevolution of infectivity and resistance (Lopez-Pascua & Buckling, 2008). Rates of evolution of infectivity were estimated every second transfer, and these values were averaged across time for each replicate.

Measuring resistance and infectivity ranges

To determine whether migration across the productivity gradient affected mean resistance and infectivity ranges to all other populations (i.e. the extent of generalism), we streaked 20 bacterial clones of each treatment across 20 μL lines of phage from all populations from the same time point. This was performed for transfer 4, 8 and 12.

Statistical analyses

Our analyses were limited to two sample comparisons, because source and recipient populations were paired, whereas treatments of a given productivity, with or without immigration were unpaired. An inability to meet assumptions of mixed model repeated measures analyses (residual maximum likelihood) for some of these measures led us to simply calculate mean values of these measures through time for each replicate. The resultant small sample sizes (six per treatment) favour the use of nonparametric tests because of an inability to determine if model assumptions are met.

Results and discussion

We wanted to test the hypothesis that migration from high-productivity patches has a greater impact on coevolution in low-productivity patches than vice versa. We first briefly report the necessary data within high- and low-productivity environments in the absence of migration. Despite the high-productivity media supporting a 10-fold greater population size of bacteria in the absence of phages, population sizes were only two-fold greater in the high-productivity treatments in the presence of phages under the no migration treatments (Mann–Whitney: = 0.07). We found no significant difference in mean phage density through time (Mann–Whitney: > 0.2). Consistent with previous work (Lopez-Pascua & Buckling, 2008), we found the rates of infectivity evolution was approximately four times faster in the high-productivity environment (Fig. 1a; Mann–Whitney: = 0.02). Moreover, the degree of generalism of both bacteria and phages (resistance and infectivity ranges, measured against all communities from the same time point) was greater in high-productivity environments (Fig. 1b,c; Mann–Whitney: < 0.05, in both cases), as qualitatively predicted by theory (Hochberg & van Baalen, 1998).

Figure 1.

 Mean rate of (a) phage infectivity evolution; (b) resistance range; and (c) infectivity range across time for the six phage populations in each treatment. ‘High’ and ‘low’ refer to the productivity of the recipient population, and ‘+mig’ refers to immigration from a source population with different productivity. Phage infectivity evolution between two time points was calculated from the difference in infectivity of past and future phages to a focal bacterial population. These point estimates were calculated every second transfer for the 12 transfer experiment, and means calculated for each replicate through time. Resistance and infectivity ranges were calculated by measuring the resistance or infectivity of each population to all other populations in the experiment from the same time point. Ranges were calculated every fourth transfer, and means calculated through time.

We next consider how immigration affected local coevolutionary dynamics. Immigration from the low-productivity environment did not alter the rate of evolution in the high-productivity environment (Fig. 1a; comparison of high productivity environments with and without immigration; Mann–Whitney: > 0.2), nor did it alter resistance and infectivity ranges in the high-productivity treatments (Fig. 1b,c; Mann–Whitney: > 0.2, in both cases). By contrast, immigration from high-productivity environments resulted in a two-fold increase in the rates of evolution in low-productivity environments (Fig. 1a; Mann–Whitney: = 0.005), such that the rate in the low productivity recipient population was indistinguishable from that of its high productivity source population (Fig. 1a; Wilcoxon: > 0.2). Similarly, high-productivity immigrants increased the resistance and infectivity ranges in the low-productivity environments (Fig. 1b, c; Mann–Whitney: = 0.005, in both cases). Surprisingly, infectivity ranges in low-productivity environments were actually greater than that for their high productivity source populations (Fig. 1c; Wilcoxon: < 0.05), although there was no difference in resistance ranges (Fig. 1c; Wilcoxon: > 0.2); we do not know why. Taken together, these results unambiguously show that migration from high-productivity to low-productivity environments can greatly alter local coevolutionary dynamics, but not vice versa.

A general reason why high-productivity environments may have a greater impact on coevolutionary dynamics of low-productivity environments than vice versa is that high-productivity environments will typically support larger population sizes (Rosenzweig, 1995). As such, there will be greater gene flow from the high- to low-productivity environments, hence genotypes adapted to high productivity conditions will have a greater influence on the low-productivity population than vice versa (Hochberg & van Baalen, 1998; Garcia-Ramos & Kirkpatrick, 1997). Moreover, increasing immigration rates per se can accelerate coevolution if genetic variation within populations is otherwise limited (Gandon et al., 1996; Gandon & Michalakis, 2002; Morgan et al., 2005, 2007). In this study, bacterial population sizes in high-productivity environments were approximately twice that in low-productivity environments, hence these demographical effects are likely to have contributed to the greater impact of high-productivity immigrants in low-productivity environments.

However, differential immigration rates into high- and low-productivity environments alone cannot explain the massive asymmetry in the effects of immigration on local coevolutionary dynamics. Immigration rates were 1% per transfer, hence immigrants made up approximately 2% of the low-productivity populations (because high-productivity population sizes were approximately twice as large as low-productivity populations), and 0.5% of the high-productivity populations. High-productivity immigrant bacteria and phage must therefore have dramatically increased in frequency relative to low-productivity immigrants. This asymmetry in selection may result from the gene-for-gene like (Flor, 1956; Thrall & Burdon, 2003) specificity of the interaction between bacteria and phages, which results in the evolution of generalists. Under these conditions, many phages and bacteria from low-productivity environments are likely to have been strongly selected against in high-productivity environments because they were unable to infect or resist, respectively, the residents. By contrast, an influx of highly infective phages into the low-productivity environments may have favoured the more resistant immigrant bacteria.

Despite numerous recent examples of selection mosaics in natural populations (e.g. Laine, 2006; Toju, 2008; Benkman, 1999; Zangerl & Berenbaum, 2003; Hanifin et al., 2008), no studies to date have explicitly determined the importance of environmental productivity. This makes it difficult to compare our results with any natural systems. Surveys and experiments are currently under way to determine the role of productivity and migration in natural populations of bacteria and phages in a variety of habitats.

These results suggest that migration across selection mosaics imposed by variation in environmental productivity can play a major role in local coevolutionary dynamics. More specifically, coevolutionary dynamics at the landscape scale can be driven by a few high-productivity hotspots, as predicted by theory (Hochberg & van Baalen, 1998; Gomulkiewicz et al., 2000). By contrast, immigration from low-productivity coldspots had little effect on the coevolutionary dynamics within high-productivity hotspots. We conclude that coevolutionary interactions in high-productivity environments are therefore likely to have a disproportionate impact on coevolution across the whole landscape.


This work was funded by the European Research Council, The Royal Society and the Leverhulme Trust. We thank Allen Moore and two anonymous reviewers for comments that improved the manuscript.