Is parallel evolution driven by broad-brush contrasts in selection?
This question can be informed by considering three possibilities. First, if colour divergence is driven by broad-brush contrasts in predation intensity, then adaptive divergence between low- and high-predation environments should be similar in all rivers. Secondly, if colour divergence is influenced by specific predator species, then divergence between low- and high-predation environments may differ between the north and south slopes – because these have very different suites of predators (Endler, 1983; Reznick et al., 1996). Thirdly, if colour evolution is sensitive to environmental factors other than predation, divergence between low- and high-predation environments may differ between rivers on the same slope, such as the small Damier River and large Yarra River (Fig. 1).
Previous work sets the stage for evaluating these possibilities. In general, Poeciliid fishes are less colourful in the presence of ‘dangerous’ predatory fishes than in their absence (Endler, 1978, 1980, 1982; Winemiller et al., 1990; Millar et al., 2006). And yet, male colour also varies considerably among populations within each of these two broad predation categories (Endler, 1978; Millar et al., 2006). Selective factors potentially influencing these nonparallel aspects of divergence are several, and we now consider them in turn. First, specific predator species do not appear particularly important for relative orange area, because this aspect of male colour is greater for low- than for high-predation populations on both slopes (Endler, 1978; Winemiller et al., 1990; Millar et al., 2006). In contrast, specific predator species may indeed be important for relative blue area, which is greater for low- than for high-predation populations on the south slope (Endler, 1978), whereas the opposite is true on the north slope (Millar et al., 2006; present study). Secondly, predator densities may be important because males have more orange and less blue at sites with more prawns (Millar et al., 2006). Thirdly, habitat features may be important because, for example, guppies at sites with more open canopies have less black (Millar et al., 2006).
Our results allow further insight. Most striking among these was that orange did not diverge between predation regimes in the Damier, a result that contrasts with the Yarra (Figs 2 and 5) and with other rivers (Endler, 1978, 1980, 1982; Winemiller et al., 1990; Millar et al., 2006). In contrast, the number and size of black spots on wild-caught males did diverge in parallel in the Damier and the Yarra, although these differences were lost in lab-reared males. Phenotypic divergence between predation regimes for black therefore seems to have a plastic basis. We can see several possible reasons for the apparent lack of genetic divergence in male colour between predation environments in the Damier. First, the intensity of predation in low- or high-predation sites may differ between the rivers because of currently unknown differences in predator densities. Secondly, any evolution of increased orange in the Damier low-predation environment might first require the evolution of increased female preference for orange (see below). Thirdly, the two predation environments in the Damier are reasonably similar in physical habitat features, such as width, depth, flow and canopy openness (S. Gordon, unpublished data). To the extent that these features influence colour divergence between predation regimes in other rivers (Grether et al., 1999; Millar et al., 2006), divergent selection may be weaker in the Damier. Fourthly, gene flow might constrain divergence of Damier high-predation males from low-predation males immediately upstream. Gene flow will not, however, explain why the Damier low-predation males have not evolved as expected for a low-predation population – because populations above waterfalls will only rarely receive migrants from below (Crispo et al., 2006). Fifthly, additive genetic variation may be low for male colour in some guppy populations (Hall et al., 2004).
The overall importance of the broad-brush contrast in predation environment vs. river-specific nuances of selection can be formally assessed on the north slope by comparing effect sizes (partial η2) for river, predation and their interaction (Langerhans & DeWitt, 2004). The main effect of river informs the importance of river-specific habitat features that are shared across predation regimes. The main effect of predation informs the importance of broad-brush contrasts in predation that are shared across rivers. The interaction between river and predation informs the importance of contrasts in predation environments that differ between rivers. Each of these components of diversification was sometimes more important than the others, although predation was rarely the most important. These results illustrate that evolution, even in the classic context of male guppy colour, is heavily contingent on local nuances. Many of these nuances likely relate to the details of local selection, whereas others may reflect founder effects, genetic drift, or gene flow. In truth, evolution in most systems will almost certainly be the result of both general and specific aspects of selection, as well as other evolutionary forces (Gilchrist et al., 2004; Langerhans & DeWitt, 2004; Langerhans et al., 2006).
How quickly can selection drive adaptive divergence?
We here calculate rates of change (standard deviations per generation, ‘Haldanes’) for male colour when guppies are introduced to new environments. We separately calculated rates of ‘phenotypic’ divergence (based on wild-caught males) and ‘genetic’ divergence (based on lab-reared males) for the three relevant comparisons: DL vs. DH, DH vs. YH and DL vs. YH. Assuming 26 generations have passed, rates of phenotypic divergence were 0.001–0.031 (median = 0.010) and rates of genetic divergence were 0.0002–0.026 (median = 0.012). For black, the only colour that showed substantial evolution in the Damier, rates of phenotypic divergence were 0.001–0.031 (median = 0.018) and rates of genetic divergence were 0.0002–0.026 (median = 0.014). In comparison with other guppy introductions (Endler, 1980; Magurran et al., 1995; Reznick et al., 1997; see Hendry & Kinnison, 1999), these rates are similar to those for behaviour (0.002–0.032) and life history (0.014–0.149), but are much lower (even if we instead assume only 13 generations) than those for male colour (0.267–0.742). Substantial colour divergence is clearly possible over the time frame we examined, and yet divergence between the low- and high-predation populations in the Damier was minor. These comparisons suggest something special about Endler's (1980) experiment, or about ours. One difference was the much shorter time frame for divergence in Endler's (1980) experiment – because evolutionary rates are known to decrease with increasing time interval (Hendry & Kinnison, 1999; Kinnison & Hendry, 2001). We feel that a more important difference, however, may relate to differences in sexual selection among introduced populations, a topic to which we now turn.
How is parallel evolution influenced by sexual selection?
The evolution of greater male colour in low-predation environments is presumably the result of sexual selection imposed by females (Houde & Endler, 1990; Endler & Houde, 1995) in the absence of strong opposing natural selection (Haskins et al., 1961; Endler, 1980). This expectation becomes particularly interesting in the context of high-predation fish colonizing low-predation sites, which is both the expected direction of natural colonization (Crispo et al., 2006) and the direction imposed in the Damier introduction. In such cases, the evolution of male colour in the low-predation population will initially depend on female preferences in the introduced fish. So what are these preferences? Although it is widely assumed that females generally prefer greater male colour, substantial geographical variation in preference is evident (Houde & Endler, 1990; Endler & Houde, 1995; Brooks & Endler, 2001). Indeed, one might expect weaker preferences in high-predation populations – because exercising choice in the presence of predators may be costly (Houde & Endler, 1990). Indeed, experiments in our laboratory have found that females from the Yarra high-predation site do not generally favour males with greater colour (Schwartz & Hendry, 2007).
Unless female preference has evolved following colonization of the Damier, we therefore would not expect male colour to increase in this population. Weak or absent preference evolution does seem possible given that artificial selection on female choice does not always produce an evolutionary response (Hall et al., 2004), and that mate choice experiments in our laboratory have revealed similar preference functions in the ancestral Yarra and derived Damier populations (L. Easty, A. Schwartz and A. Hendry, unpublished data). The general lack of male colour divergence in the Damier is therefore consistent with what is known about female preference evolution. With greater time, however, female preferences may evolve in the Damier low-predation population so as to favour greater male colour, and we might then expect to see an increase in male colour. For the same reasons, it would useful to know the pattern of female preferences in the population that Endler (1980) introduced. These considerations show how the parallel evolution of secondary sexual traits can depend not only on changes in natural selection, but also on ancestral patterns of sexual selection, and on their rate of evolution in a new environment. We are not aware of any study that has yet examined the contemporary evolution of female preferences in the wild.