Mike Ritchie, Environmental and Evolutionary Biology, Dyers Brae House, University of St Andrews, St Andrews, Fife, KY16 9TH Scotland, UK. Tel.: +44-0-1334-463495; fax: +44-0-1334-463366; e-mail: firstname.lastname@example.org
Currently there is much interest in the potential for sexual selection or conflict to drive speciation. Theory proposes that speciation will be accelerated where sexual conflict is strong, particularly if females are ahead because mate choice will accentuate divergence by limiting gene flow. The Goodeinae are a monophyletic group of endemic Mexican fishes with an origin at least as old as the Miocene. Sexual selection is important in the Goodeinae and there is substantial interspecific variability in body morphology, which influences mate choice, allowing inference of the importance of female mate choice. We therefore used this group to test the relationship between sexual dimorphism and speciation rate. We quantified interspecific variation in sexual dimorphism amongst 25 species using a multivariate measure of total morphological differentiation between the sexes that accurately reflects sexual dimorphism driven by female mate choice and also used a mtDNA-based phylogeny to examine speciation rates. Comparative analyses failed to support a significant association between sexual dimorphism and speciation rate. In addition, variation in the time course of speciation throughout the whole clade was also examined using a similar tree containing 34 extant species. A constant rates model for the growth of this clade was rejected, but analyses instead indicated a decline in the rate of speciation over time. These results support the hypothesis of an early expansion of the group, perhaps due to an early radiation influenced by the key innovation of live bearing, or the prevalence of Miocene volcanism. In general, support for the role of sexual selection in generating patterns of speciation is proving equivocal and we argue that vicariance biogeography and adaptive radiations remain the most likely determinants of major patterns of diversification of continental organisms.
Organisms vary in the rate at which they diverge and produce new species, due both to intrinsic properties (e.g. specialization on host organisms may accelerate speciation, Ehrlich & Raven, 1964; Diehl & Bush, 1989) and for extrinsic reasons (e.g. those occupying island or peripheral habitats may speciate more rapidly, Mayr, 1963). Identifying factors that are consistently associated with speciation would help us to understand why groups vary in diversity and reveal important common influences on speciation. One intrinsic factor that is thought to accelerate speciation across many different groups of animals is the intensity of sexual selection. This hypothesis arises because of the observation that many species differ primarily in sexual signals and preferences (Butlin & Ritchie, 1994; Gray & Cade, 2000) and an apparent correlation between the number of species in a group and the importance of sexual selection, as indicated by sexual dimorphism or mating system. For example, Barraclough et al. (1995) demonstrated that bird clades with greater sexual dimorphism in plumage colour contained more species (but see Price, 1998) and Arnqvist et al. (2000) showed that polyandrous insect clades were on average four times more speciose than monandrous clades. Thus indicators of sexual selection seem to be associated with greater species diversity, encouraging the reputation of sexual selection as an ‘engine of speciation’ for a broad range of organisms (Arnqvist et al., 2000; Gavrilets et al., 2001; Martin & Hosken, 2003). However, while theory clearly predicts an association between sexual selection and speciation (Parker & Partridge, 1998; Partridge & Parker, 1999; Gavrilets, 2000; Gavrilets et al., 2001), there is an implicit problem in the empirical evidence that can be used to test this because species may be more likely to be described where sexual dimorphism or displays are present. Consequently there is the potential for circularity. In fact, it has been argued that in most of the cases held to support an association between sexual selection and speciation the evidence is too indirect to provide conclusive support (Panhuis et al., 2001).
More recently, attention has shifted from sexual selection per se to sexual conflict. Sexual conflict arises when the genetic interests of males and females diverge (reviewed in Chapman et al., 2003). Conflict underlies most models of sexual selection but is emphasized in recent models that address differences between the sexes in the preferred outcome of sexual reproduction (Holland & Rice, 1998; Partridge & Parker, 1999; Gavrilets et al., 2001). Sexual selection will almost always act strongly on males to increase mating rates or proximate female fecundity and female counter-adaptations to male sexual strategies will result in an evolutionary dynamic over the ensuing conflict (Arnqvist & Rowe, 2002). Incorporating the uncertain outcome of sexual conflict leads to a more complex prediction of the role of sexual selection in speciation. Partridge & Parker (1999) and Parker & Partridge (1998) have demonstrated that speciation will be more likely where the current conflict favours the females’ outcome, because successful mate choice by females is more likely to lead to restricted gene flow between diverging gene pools. However, if the male is ahead in the conflict, differentiation may be impeded as female mate choice is circumvented (see also Magurran, 1998; Gavrilets, 2000).
Often when a powerful new idea arises in science there is a period of widespread acceptance only subsequently followed by more critical evaluation (Simmons et al., 1999). More detailed comparative studies have recently begun to question the role of sexual selection in increased diversity. Gage et al. (2002) examined over 600 genera across a wide range of animal groups for an association between sexual size dimorphism or polyandry and speciation, but did not find any supporting evidence. They concluded that ‘well-recognized’ naturally selected processes were more likely to drive the evolution of biodiversity. Morrow et al. (2003) also failed to find an association between measures of pre and post-mating sexual selection and speciation in birds.
Rigorous comparative methods are allowing refined analyses of diversification rates and potential correlates such as sexual selection (Pybus & Harvey, 2000;Nee, 2001), sometimes with surprising results. Even traditional theories of speciation, such as a preponderance of allopatric over sympatric speciation, sometimes only receive equivocal support when tested using rigorous comparative analyses (Barraclough et al., 1998; Losos & Glor, 2003), so it is important to assess the importance of both traditional and more recent models of speciation using these techniques.
Here we measured morphological correlates of female-driven sexual selection and analyse a molecular phylogeny of the Mexican Goodeinae, an endemic monophyletic group of live-bearing fish, to test for a role of sexual selection in speciation. We predict that the time till speciation should be shorter in more dimorphic lineages if sexual selection by female mate choice is important, so our first test compares morphological measures of female driven sexual selection with speciation rate. We chose the Goodeinae as a model system because sexual selection is important in the group and there is considerable interspecific variation in sexual dimorphism. Males lack an intromittent gonopodium and so are incapable of inseminating females without female co-operation (Bisazza, 1997). Female choice is therefore very important and many species have elaborate fins that function as sexual ornaments. In one of the most elaborate species (Girardinichthys multiradiatus), the fins have been shown to be the targets of female choice and also to increase predation on males (Macías Garcia et al., 1994,1998). So female choice for elaborate fins leads to a departure from the naturally selected male body shape, incurring costs. Hence the extent of sexual dimorphism should reflect the extent to which sexual conflict operates in these species via female mate choice. However, in many other species sexual dimorphism is much reduced or almost absent, suggesting that female choice is less important in such species. Sexual dimorphism in fin size is related to complex male courtship behaviour, with less dimorphic species having more simple courtship (Fitzsimons, 1972; Macías Garcia, Unpublished). There are substantial population differences in sexual dimorphism in a few species (Fitzsimons 1972; Macías Garcia, Unpublished), which appears to reflect local differences in the outcome of sexual conflict over matings under differing ecological conditions (Macías Garcia, 1994; Moyaho et al., 2004). Thus it is highly likely that variation in sexual dimorphism between species reflects the extent to which female choice dominates the mating system, with the more dimorphic species being those in which female preference is more important to the outcome of mating attempts. Those with less dimorphism typically have morphologies lacking elaborate fins, more closely resembling the naturally selected body morphology.
We also examined the time course of speciation in the whole group. Previous studies of speciation in the Goodeinae have emphasized the role of vicariance or live bearing as an early key innovation (Hubbs & Turner, 1939). Webb et al. (2004) produced a phylogeny from mtDNA sequences and applied the Barraclough test (Barraclough et al., 1998) for modes of speciation. This gave a positive relationship between node depth and the degree of sympatry, as predicted for allopatric speciation, but the variance explained was very small. Webb et al. (2004) concluded that, although the pattern of speciation was likely to have been vicariance-driven, frequent range changes and river piracy had obscured the signal.
Here we tested whether a quantitative measure of sexual dimorphism in the Goodeinae that correlates with sexual conflict (being derived from multivariate discriminant analysis of sexually dimorphic body measurements, mainly fin morphology) is associated with clade diversity or internode intervals (speciation times). We also examined the variation in internode intervals through the clade to test whether speciation rate is constant or has been varying, as might be expected if there was an early radiation due to increased vicariance in the Miocene.
Materials and methods
Species sampling and morphological analysis
For the first analysis, samples of fish were collected in Mexico during the summers of 1999 and 2000. A species was included in the analysis if 20 individuals of each sex were sampled. The maximum sample sizes were 40 of each sex from up to four populations. If multiple populations were sampled, up to 10 individuals of each sex from each population were included in the multivariate analysis. Twenty-five of the 37 extant species were sampled (see Fig. 1). Fish were anaesthetized, pinned on styrofoam next to a ruler and photographed in the field on Kodak Ektachrome64 slide film with a Nikon N70 with 60 mm Micro Nikkor AF lens. Images were scanned into a computer and morphological measurements made using Scion software. Morphological traits measured were the standard length (snout to caudal fin base), mid-body depth (mid-line to body top), dorsal-, caudal- and anal-fin heights, bases and areas and finally body fin depth (mid-line to mid-point of dorsal-fin base), giving 12 measures in all. Two people made the measurements, but repeatabilities were very high within observers (average repeatability over 10 individuals each of two sexes and species = 0.987) and in a balanced nested anova, the average variance across traits attributable to observer was less than 0.12%.
Most similar studies use sexual size dimorphism as the key variable. However, size dimorphism could evolve for a variety of reasons, for example success in ‘sneaky matings’ may favour small size in male guppies (Bisazza & Marin, 1995), so a large sexual size dimorphism in this species is not necessarily indicative of sexual selection through female choice (also, some size assortative mating occurs, Bisazza, 1997). Targets of female choice in the Goodeinae are male fins, which are often colourful and used during courtship and copulation (Macías Garcia et al., 1994). We therefore took as our measure of sexual dimorphism the Mahalanobis distance between the sexes following a canonical (discriminant) multivariate analysis of all morphological traits (except body fin depth, which proved colinear with mid-body depth) with sex as the grouping factor. Prior to the analysis, all traits were standardized then size corrected (to the mean standard length of the species), so size dimorphism played no role in our measure of sexual dimorphism. In practise it was mainly determined by relative fin shape and size. This analysis was carried out separately for each species, as the precise targets of sexual selection might differ between species. Indeed, there is evidence that they do (Ritchie & Drummond, unpublished). This means that the morphologies that contribute to our measure of sexual dimorphism will differ but that the composite trait dimorphism can be standardized across species. For those species with multiple populations, analyses were carried out separately by population and the mean distance calculated. Usually, populations were fairly consistent, but for one species (Goodea attripinis), one population had an unexpectedly high dimorphism score, probably due to an unusual dorsal fin morphology in females. Distances were ln transformed before analysis. Statistical analyses were carried out using Minitab12.
Sequence data are described in Webb et al. (2004) and consist of around 900 bp from cytochrome oxidase I and the control region of mtDNA (an indel from the latter region was excluded). Variation within species was minimal. Webb et al. (2004) constructed phylogenetic trees using a variety of approaches (parsimony and distance methods) and the tree shape is generally robust. The sequences saturate above a Tamura-Nei distance of around 0.2 and in all analyses here we used a transition : transversion ratio of 2, empirical base frequencies and other settings from the HKY85 model. The second analysis used data for the full 34 species, but for the first analysis species not sampled for morphological analysis (which were evenly distributed in the tree) were trimmed from the data set then the tree recalculated. Otherwise the same options as Webb et al. (2004) were used, with Empetrichthyslatos and Profunduluslabialis as designated outgroups. A single best tree was found for each dataset using a heuristic search with 20 replicates, using Paup*(4.0) (Swofford, 1998). The rate-based tests mainly used here require an ultrametric tree so that branch lengths are equivalent to time rather than to genetic distance (Barraclough & Nee, 2001), therefore sequences were reanalysed to produce phylogenetic trees using maximum likelihood with a molecular clock assumption. The trees used here differ only slightly from the parsimony trees described in Webb et al. (2004), due to the relative position of a single deep node. However, tests showed that the full data set is significantly longer when the clock assumption is on (–ln L with clock = 12006.1, no clock = 11922.2, χ2 = 85, d.f. = 35, P < 0.001). We therefore checked that our conclusions are independent of tree reconstruction method by repeating the analysis with the best maximum parsimony tree from Webb et al. (2004), which led to the same conclusion. In general, comparative methods allowing for tree uncertainty are poorly developed, but the consistency of our results with such different approaches is reassuring.
Ancestral dimorphism values were reconstructed using one parameter maximum likelihood, based on a Brownian motion model of trait evolution (Felsenstein, 1985), implemented using ACML (Schluter et al., 1997), which assumes a constant rate of change throughout the clade (necessary to test for a trait influencing branch lengths). There are a variety of potential methods for reconstructing ancestral states, but tests that assessed their effectiveness using fossil morphology suggest that there is little to choose among them, with similar levels of uncertainty in the reconstructed states (Webster & Purvis, 2002). The analysis used requires a rooted tree, but outgroup species were not analysed morphologically. Outgroups were therefore given an arbitrary score equal to the mean of the ingroups, which influenced the overall rate of change across the tree, but not the relative scores or any conclusions drawn.
Sexual dimorphism and speciation rate
Figure 1 shows the ultrametric tree for the species included in the morphological analysis, with the sexual dimorphism score (SD) of extant and the reconstructed score of ancestral taxa. The trait changes frequently, with reversals and clade-specific trends apparent, as has been found for other phylogenetic studies of sexually dimorphic traits (Baker & Wilkinson, 2001; Cox et al., 2003). If sexual dimorphism promotes speciation, the number of descendent species should be greater in clades with more dimorphic ancestors. The number of species per clade mid-way through the tree was not related to reconstructed ancestral SD (rank correlation = 0.565, n = 8, P = 0.14; the power in this analysis is weak, around 0.7).
A more powerful test examines the time till speciation at each nonterminal node. The prediction is that, if speciation is more rapid in more dimorphic lineages, the time till speciation will be shorter along more dimorphic branches (Nee et al., 1992; Nee, 2001). In total, 10 nodes support and six reject this hypothesis (Fig. 1). To assess the probability of this result given the phylogeny, the 16 paired branch lengths were each resampled 1000 times to give a null distribution of paired branch lengths. The ratio of nodes in each of these resampled data sets, which supported or rejected the hypothesis was calculated. Twenty-four percent of resampled datasets had the observed or a more extreme result, so there is no evidence for a significant association between SD and internode interval. Five per cent of the resampled datasets had a ratio of 12 : 4 or greater. Finally, repeating the analysis with an independent tree, the maximum parsimony tree from Webb et al. (2004) also failed to support the hypothesis (seven clades supported, eight did not, with one tie; data not presented).
Is speciation rate constant?
The previous analyses do not provide support for an increased diversification rate in more sexually dimorphic clades. The complete phylogenetic tree allows inferences to be drawn about the rate of diversification independent of traits. Strictly, it is the ratio of speciation to extinction rates that determines clade shape (Nee et al., 1994a). However, if we assume that the probability of extinction is constant we can interpret deviations from constant growth models as reflecting variation in speciation rate (also see discussion). For example, if speciation in the Goodeinae was characterized by an early burst of speciation followed by a decline, the tree topology would have a characteristic deviation from a constantly bifurcating pattern. Constant rates lineage through time plots are straight if the ratio of speciation to extinction is constant, but upturn towards recent times because recent species are less likely to have gone extinct (Nee et al., 1994a,b; Harvey et al., 1994). Such a plot for the Goodeinae is convex (Fig. 2), rejecting the hypothesis of constant rates, in favour of either an early radiation or an increasing extinction probability in more recent species. A similar pattern is seen if we confine the analysis to only the species sampled morphologically. Furthermore, a similar convex curve is seen if the analysis is based on a parsimony tree (data not presented). Internode intervals, which reveal the time till speciation for extant lineages (Nee, 2001), are clearly increasing with time in this data set (Fig. 3).
Finally, the test statistic γ (Pybus & Harvey, 2000) indicates whether the rate of speciation is significantly increasing or declining (it tests whether the average internode interval is nearer the base or the tip of the clade than would be expected under a constant growth model). Figure 4 shows our observed γ. A Monte Carlo approach was used to generate an expected γ plus 95% CI for a tree of our size assuming a constant rate (Pybus & Harvey, 2000). The expected value varies with how completely the tree is sampled, as incomplete trees (either due to sampling or extinction) will have a lower γ. Our observed value is rather low, always less than the mean and is below the two-tailed 95% confidence interval if the proportion of all species sampled is around 60% or more.
We have assessed evidence for two hypotheses for variation in the rate of speciation within the Goodeinae, the role of sexual selection accelerating speciation and the existence of an early radiation. We found no evidence for a role of sexual selection in speciation rate. Clades with more sexually dimorphic ancestors did not have significantly more species (cf. Arnqvist et al., 2000;Gage et al., 2002) nor was the internode interval (time till speciation) shorter in more dimorphic lineages. Both results were in the appropriate direction, but not significant.
While these results do not suggest a large role for sexual selection in accelerating speciation in this group, some caveats demand discussion. First, reconstructing ancestral values requires caution (Webster & Purvis, 2002). The average standard error of our reconstructed values was 0.17 (around 7% of the mean trait value across species, but four times the SE among extant species). There is increased uncertainty about these values at deeper nodes, which is typical for ancestral state reconstruction. Second, the power of the tests is limited by the number of species involved. This is an inevitable consequence of the group's evolutionary history. Incorporating a deeper phylogeny into the tests is possible, but would involve including species with very different evolutionary histories (the Empetrichthyinae and the Goodeinae comprise the Goodeidae, but the former radiated in central North America) and different reproductive biologies (empetrichthyines are not live-bearers nor are they internally fertilized) that would introduce further phylogenetic noise and uncertainty about ancestral states into the compartive tests.
A third potential problem concerns our measure of the intensity of sexual selection. We assume that our sexual dimorphism score reflects female biased sexual conflict, as natural selection costs would prevent the elaboration of fin morphologies in males in the absence of female choice. This may not be a suitably broad explanation for sexual dimorphism across all the species. Some species are dimorphic in colour rather than morphology, for example, but an arbitrary score reflecting sexual colour dimorphism was strongly correlated across species with our measure of sexual dimorphism (rank correlation = 0.527, n = 25, P = 0.0007; colour was not included in our analysis due to problems of quantifying dimorphism). Another complication is that sexual conflict may be expressed most strongly through behaviours that are independent of obvious morphologies, such as post-copulatory female choice. Copulation in the Goodeinae requires female co-operation (Bisazza, 1997), so cryptic female choice is less likely to be prevalent in this group.
A final caveat worth considering is the interaction of speciation rate and extinction. If species in which sexual selection is most intense are more likely to go extinct (see Møller, 2000; Gavrilets et al., 2001), comparative tests such as ours will be conservative. There is unclear empirical evidence for this across bird species (Morrow & Pitcher, 2003), but more dimorphic species are less likely to establish on islands (McLain et al., 1995; Doherty et al., 2003), apparently because sexually dimorphic traits are costly for males and because mating skew reduces the effective population size, making the colonizers vulnerable to stochastic extinction (see also Parker & Waite, 1997). This may occur in the Goodeinae, where females of one species are known to refuse to mate with sub optimal males, even if this means forgoing reproduction altogether (Macías Garcia et al., 1998).
Whilst it is obviously impossible to rule out all these potential caveats, our results do not support a major role for sexual selection influencing speciation rate continuously throughout the Goodeinae. However, we do find evidence for variation in speciation rate through the clade. The lineage through time plot is convex, most probably due to an early burst of speciation. Difficulty in resolving the basal nodes in the molecular phylogeny (cf. Webb et al., 2004) is consistent with this finding. Internode intervals decline through the extant clade and Pybus & Harvey's γ is significantly negative if we have sampled more than 60% of the total clade. The phylogenetic tree used in this analysis is known to include virtually all described extant species (Webb et al., 2004). It is, of course, difficult to estimate how many species have gone extinct, but five of the six known fossil Goodeids represent extant species, with the one extinct species being of Miocene origin (Alvarez & Arriola-Longoria, 1972; Smith, 1980). We are therefore confident that the patterns found here are unlikely to be seriously confounded by high levels of extinction. Whether extinction rate has varied is another question but for our conclusion that speciation rate declines to be invalidated, extinction rate would have to vary so that more recently evolved species had a higher probability of going extinct.
There are a few examples of similar results of lineage though time analyses such as ours. Lovette & Bermingham (1999) interpreted a convex LTT plot as supporting a late Miocene or early Pliocene radiation in new world Dendrioca warblers as a result of climate-driven forest fragmentation. Within species, jaguars (Panthera onca) show a similar relationship between mtDNA lineages and time, thought to reflect a recent demographic expansion (Eizirik et al., 2001). American 7-spined gobies (Gobiosomatini) show this and a significant negative γ, indicating an early radiation followed by a decline in speciation rate through a similar time period as the Goodeinae (Rüber et al., 2003). This occurred in a marine habitat and is associated with changes in ecology (specialization or habitat) rather than biogeography, although the precise factors accelerating speciation are probably diverse. Causes of other early fish radiations remain unclear (Johns & Avise, 1998). African cichlids provide an example where sexual selection is implicated in a very rapid evolutionary radiation (Turner, 1999). While there is much evidence to suggest that behaviour plays a role in the patterns of reproductive isolation amongst these very closely related fishes (Knight & Turner, 1999; Couldridge & Alexander, 2002), strict comparative analyses such as those carried out here are so far lacking for this important fish radiation.
Adaptive radiation is characterized by an early burst of radiation upon exploiting vacant niches and the subsequent slow-down is thought to reflect the filling up of niches (Schluter, 2000; Rüber et al., 2003). Radiations could also slow-down if vicariance became less common, though it is not clear if this happened over the relevant time period in Mexico (Ferrari et al., 1999; Webb et al., 2004). Intriguingly, specialist fish predators in the form of Thamnophis snakes entered the area around 5-7 MYA (Malnate, 1960). There seems little reason to suppose that radiations driven by sexual selection should show such slow-downs, as constraints on the range of behavioural phenotypes are more difficult to imagine.
Detecting a role for sexual selection in evolutionary radiations may be extremely difficult. Many of our caveats will apply to most such studies and our results are marginal (one or two key differences among extant species could make the trends seen with the ultramteric tree here significant). Disentangling the relative roles of geographic and ecologically driven speciation from sexual selection is a particular challenge. Behaviour may be particularly important in influencing contemporary levels of gene flow and determining the outcome of secondary contact between closely related taxa, but over an evolutionary time scale (necessary for most comparative analyses), biogeography and ecological specialisation may swamp the more subtle phylogenetic signal due to sexual selection.
Sean Nee provided advice on analysis, J. M. Artigas-Azas, O. Dominguez, E. Avila Luna & E. Smart helped with field work and Jillyan Drummond provided helpful technical assistance. This work was funded by the N. E. R. C.