J. Van Buskirk, Institute of Zoology, University of Zurich, CH-8057 Zurich, Switzerland. Tel.: + 41 44 635 4985; fax: + 41 44 635 6821; e-mail: firstname.lastname@example.org
The external morphology of frog larvae is predicted to vary among habitats, for a variety of functional reasons. I performed a phylogenetic comparative study to test whether correlations between habitat and the shape of the tadpole and its oral disc are adaptive in 82 species from south-eastern Australia in the families Hylidae and Myobatrachidae. Habitat distributions and phylogenetic relationships were compiled from the literature and shape was quantified using geometric morphometric analysis of published drawings. Results indicate that shape evolved towards different optima in different habitats while also showing appreciable levels of phylogenetic inertia. Within myobatrachids, evolution of terrestrial tadpoles was associated with a short and shallow head/body and a shallow tail. In aquatic species, the use of benthic microhabitats was correlated with a long shallow tail, dorsal eye position, shallow head/body and ventral mouth with robust jaw sheaths. Even traits with evidence for adaptation evolved slowly in response to habitat shifts, usually requiring ≥10 million years to evolve half-way to a new optimum. Although these findings support adaptive evolution of tadpole body form, they also highlight a strong influence of ancestral character states and indicate that phenotypes in extant species are partly maladaptive.
A key assumption of this research programme is that the traits differentiating species have evolved as adaptations to known features of the environmental gradient. When a species’ habitat distribution shifts, its phenotype must also change. If this assumption is not true, then the gradient cannot be particularly informative about forces promoting species divergence, and trade-offs seem unlikely to explain habitat distributions. It must be said that many studies, a few of which are cited above, establish strong evidence for adaptation along many different environmental gradients. So the question is not whether adaptation occurs, but rather how rapid adaptive evolution is relative to processes that delay or counteract it. Furthermore, testing the assumption of adaptation can focus attention on the particular characters that are susceptible to trade-offs and the types of habitat gradients that enforce them.
One approach to detecting adaptive character divergence along environmental gradients is phylogenetic modelling that explicitly asks whether characters evolve in response to shifts in habitat use (Felsenstein, 1985; Harvey & Pagel, 1991; Martins, 2000). Recent methods are able to estimate simultaneously the tendency for characters to be influenced by ancestral character states and the rate at which they evolve towards adaptive optima (Butler & King, 2004; Hansen et al., 2008). In this study, I apply such models to a comparison between morphology and habitat use in larvae of the amphibian fauna of south-eastern Australia, and discover that both adaptive evolution and phylogenetic inertia are quantitatively important. My inclusion of all extant species in a large region, falling within families Hylidae and Myobatrachidae, is important because it ensures that species are sampled without bias, which could arise if accessibility or pattern of occurrence along the gradient influenced species inclusion.
The study included every species depicted in Anstis’s (2002) guide to the anuran larvae of the south-eastern Australian states of New South Wales, Victoria and Tasmania, except for the invasive cane toad (Bufo marinus) (Appendix S1). Using drawings published by Anstis, I quantified the shape of the oral disc and the tadpole itself when viewed from the side. Anstis collected eggs in the field, returned them to Sydney, Australia, and reared the tadpoles from hatching to metamorphosis in aquaria installed on an outdoor covered deck. The aquaria were filled with rainwater, partially exposed to morning sun, and contained sand, silt, leaf litter and vegetation. Specimens of a few species that Anstis could not locate were sent to her by Ross Alford of James Cook University. Anstis photographed and prepared detailed drawings of the lateral view of each species when at a mean Gosner (1960) stage of 35.7 (SD = 3.5). For most species, she also made drawings of the oral disc from preserved specimens collected in the field. Anstis included side-view drawings of 35 anuran species in the family Hylidae and 47 species in Myobatrachidae. There were drawings of the oral disc for 34 hylids and 42 myobatrachids; those not included have no jaw sheaths or tooth rows (Appendix S1). Although Anstis’s rearing protocol was consistent over the 20-year period of her study, it is inevitable that some phenotypic variation in the drawings is not due to species differences, but arose instead from unknown variation in conditions during development or idiosyncratic features of the populations that were sampled.
I used geometric morphometric methods to describe variation in the shape of the tadpole and its oral disc, in two separate analyses. This approach defines shape in terms of the relative locations of landmarks after correcting for size and orientation (Zelditch et al., 2004). Image analysis software was used to measure the coordinates of the 24 side-view landmarks and 18 oral-disc landmarks shown in Fig. 1. A detailed description of landmarks is available in Appendix S2.
The first step in the morphometric analysis was to calculate for each specimen the centroid size, which is the mean distance between all landmarks and their centre of gravity. Next, I rescaled specimens to unit size, superimposed their centroids and rotated them using generalized least-squares Procrustes superimposition to minimize the sum of the squared distances among corresponding landmarks (Rohlf, 1990). Differences in form that remained after the Procrustes superimposition were due to variation in shape. From the sample of all specimens, I then calculated two kinds of variables representing shape: uniform shape components arose from variation affecting all landmarks due to compression and shear, and nonuniform components described variation affecting local subsets of landmarks. There were too many components to visualize separately, so I subjected them to a principal components analysis (PCA) and retained the first two PC components (termed ‘relative warps’) for subsequent analysis. Relative warps subsequent to the first two explained relatively little variation in total shape and showed no significant relationships with habitat. I carried out these procedures using IMP software (Sheets, 2004). Size was not included in the analyses because it was influenced by the age and developmental stage at which Anstis made the drawing.
A classification of the habitat distribution of each species was compiled from published descriptions (Tyler, 1994; Cogger, 2000; Anstis, 2002). Species were scored according to whether their tadpoles are aquatic, occur in streams, or in temporary, semi-permanent or permanent wetlands (yes/no in all five cases), with many species occurring in more than one of these categories (Appendix S1). I also recorded whether the position of tadpoles in the water column is typically benthic, mid-water or near the surface; species found at all three levels or with no preference were assigned intermediate values. I summarized correlations among the habitat variables with a PCA that accommodated a mixture of categorical and quantitative variables (Hill & Smith, 1976; Kiers, 1994), implemented in the ade4 package in R 2.8.1 (R Development Core Team, 2008). Position in the water column was the only quantitative variable. Axes were subjected to oblique promax rotation to improve correlations between the original variables and the derived habitat factors (SAS Institute, 1990). The first three components from the PCA were retained for further analyses of their association with tadpole morphology. The three habitat components together explained 80% of the variation in the original six variables, and were each readily interpretable in terms of original variables (Table 1).
Table 1. First three eigenvectors of a principal components analysis on six measures of tadpole habitat.
The first five variables are categorical (yes/no), with high scores on the habitat axes correlating positively with ‘yes’. Position in the water column is low for benthic species and high for pelagic species. The sample size was 82 species, for which habitat data are given in Appendix S1.
I estimated the phylogenetic signal in habitat and phenotypic data using a permuted linear regression technique (Legendre et al., 1994; Martins, 1994; Laurin, 2004). The question was whether related taxa exhibit more similar habitat distributions and traits than expected by chance. The method regressed the difference in habitat or phenotype between all pairs of species against the phylogenetic time separating them (summed lengths of branches). Statistical significance was judged by comparing the observed regression coefficient with the distribution of coefficients produced by repeating the procedure 10 000 times after randomly permuting the habitats and phenotypes of species.
I tested for relationships between tadpole shape and habitat using a phylogenetic comparative method developed by Hansen and colleagues (Hansen, 1997; Butler & King, 2004; Hansen et al., 2008). This method models evolution as an Ornstein–Uhlenbeck (OU) process evaluated within a phylogenetic context. The OU process describes a random walk in trait values (in this case, morphology) in the presence of an attractor. Random variation is created by drift and unmeasured ecological or developmental processes, and attraction is generated by adaptation towards an optimum that can vary with predictor variables (in this case, habitat). It is assumed that the optimal value of the trait is linearly related to predictors, and changes in predictors can occur throughout the phylogeny. Phenotypic evolution occurs in response to changes in predictor variables. This is a plausible model of the evolutionary process, and is derived from the quantitative genetic framework used in classic microevolutionary models (Lande, 1976, 1979). The form of Hansen et al.’s model is:
where z is a vector of trait values and x is a vector (or matrix) of one (or more) predictor variables, t is the time since the root of the tree, b is a vector of coefficients representing slopes of the optimal phenotype regressed on predictors, α is the rate of adaptation and k is a constant with no simple interpretation but related to the optimal phenotype in the ancestor (when x =0). Trait values and covariates are measured in extant species, and phylogenetic information and branch lengths are incorporated into the error structure of the regression model. In my analyses, the traits (z) were relative warps describing the shape of the side-view and oral disc, and the covariates (x) were the three habitat components.
If the rate of adaptation is very high, so that the inertial term (1 − e−αt)/αt approaches zero, then the model yields the optimal regression. This estimates the relationship between trait and habitat that would be observed if the phenotype closely tracked changes in habitat use. Alternatively, the rate of adaptation may be constrained by phylogenetic inertia, which is defined phenomenologically as resistance to evolutionary change in response to a changing adaptive optimum. Inertia can be caused by variational or selective constraints, or by the appearance of unknown traits in an ancestral taxon that bias the adaptive radiation of its descendents (Derrickson & Ricklefs, 1988; Blomberg & Garland, 2002; Hansen & Orzack, 2005). In this case, the inertial term in eqn 1 is appreciable and we obtain the evolutionary regression, which estimates both adaptation and phylogenetic inertia. This model discounts the slope (b) of the optimal phenotype against x by an amount that depends on phylogenetic inertia, and is equivalent to the phylogenetic generalized least squares model of Martins & Hansen (1997). The parameter α yields an intuitive estimate of the importance of phylogenetic inertia known as the phylogenetic half-life, t½ = ln(2)/α. This is the time required for a character to evolve half-way to a new optimum after a change in the environment. I implemented the analysis using an R program called ‘slouch’ (Hansen et al., 2008).
Uncertainty in branch lengths within the phylogeny was accommodated in two ways. First, I repeated the analysis using two recent estimates of divergence times for clades included in my sample, based on the fossil record, biogeography and molecular data (Marjanović & Laurin, 2007; Roelants et al., 2007). The two estimates differ by about 20–40%. For example, the divergence of myobatrachids and hylids was judged to occur about 105 Ma by Marjanović & Laurin (2007) and 145 Ma by Roelants et al. (2007). I used the two schemes to anchor dated divergence events, and then spaced undated events at equal intervals between them (Appendix S3). The second approach was to repeat each analysis on 30 phylogenies with the same topology but randomly generated branch lengths. Beginning with estimates from Marjanović & Laurin (2007), ages of the four clades listed in Appendix S3 were sampled at random from normal distributions having a mean equal to the estimated age and standard deviation equal to 20% of the estimate. Undated divergences were then placed at equal intervals between dated events. I did this only for the estimates of Marjanović & Laurin because the dates in Roelants et al. (2007) are probably too old for a variety of reasons (Marjanović & Laurin, 2007, 2008). Coefficients were significant if their 95% confidence intervals did not overlap zero averaged over all randomly generated phylogenies.
Variation in habitat and morphology
I begin by summarizing the PCA components used to describe habitat distributions and morphology (Table 1). The first two habitat components represented contrasts between anurans occurring as larvae in temporary or semi-permanent wetlands and those that are terrestrial (habitat 1), and between pelagic (also called nektonic; Orton, 1953) species in permanent ponds vs. benthic species in temporary habitats (habitat 2). The third habitat component differentiated species into three clusters: (i) stream-breeders with high values; (ii) terrestrial species with low values; and (iii) pond-breeders with intermediate values.
The morphological shape components are best appreciated in visual depictions (Figs 2–5; detailed drawings in Appendices S4 and S5). The first side-view relative warp was positively associated with a long and somewhat deep head/body, a dorso-anterior eye and a short tail fin. Side-view RW2 was correlated with a much larger gut mass, forward-pointed mouth and dorsal nares, lateral eye position, more anterior attachment of the dorsal fin to the head/body and a deep tail fin. The two relative warps explained 56.7% of all shape variation.
The first oral disc shape component represented variation in the length and curvature of tooth rows. Species with high RW1 values also had relatively few anterior tooth rows (r = −0.31, P =0.0057, N =76 species; data not shown). Oral disc RW2 increased in species with narrow mouths, thin lower jaw sheaths, widely spaced anterior tooth rows, and more tightly packed posterior tooth rows. The two components explained 66.3% of shape variation in oral disc landmarks.
Legendre et al.’s (1994) regression with permutations on distance matrices detected a strong phylogenetic signal in most measures of habitat and morphological shape. This can be visualized to some extent in the distributions of species and genera along axes defined by the habitat and phenotypic measures (Fig. 2). Closely related species exhibited similar habitat distributions (habitat 1: β = 5.32 × 10−3, P =0.0146; habitat 2: β = 3.02 × 10−5, P =0.0001; habitat 3: β = 4.81 × 10−4, P =0.0277), side-view shape components (RW1: β = 3.61 × 10−5, P =0.0001; RW2: β = 1.40 × 10−5, P =0.0002), and oral disc shape (RW1: β = 3.72 × 10−4, P =0.0000; RW2: β = 1.54 × 10−5, P =0.14). These β values are estimated average rates of evolutionary change across the whole clade under a model of random drift (Martins, 1994). Distinctions at the level of the genus are easiest to visualize in the Hylidae, which includes only two genera in south-eastern Australia. Cyclorana was more benthic and in more temporary habitats than Litoria, and never in streams (Fig. 2a). Tadpoles of Cyclorana were characterized by a relatively short tail and large head/body (Fig. 2b), and by a wider mouth and robust keratinized jaw sheaths (Fig. 2c). Although myobatrachid genera showed equally great divergence in morphology and habitat, differences between families were less obvious (Fig. 2).
Associations between tadpole morphology and habitat
The dominant morphological contrasts within the Myobatrachidae were between species that develop terrestrially and those that have aquatic larvae (Fig. 3). Side-view shape components RW1 and RW2 showed significant evolutionary divergence along the terrestrial/aquatic habitat gradient, because terrestrial tadpoles had a short and shallow head/body with a shallow, long tail. These relationships were not sensitive to assumptions about branch lengths, because estimated coefficients were similar regardless of whether branch lengths were scaled according to Marjanovic and Laurin (2007) and Roelants et al. (2007), or generated randomly (Table 2).
Table 2. Phylogenetic multiple regressions for effects of habitat variables on morphological shape of south-eastern Australian tadpoles, performed separately for Myobatrachidae (including aquatic and terrestrial species) and all aquatic species (including hylids and nonterrestrial myobatrachids). Results are shown for branch lengths estimated by Marjanović & Laurin (2007) (a), Roelants et al. (2007) (b), and for 30 phylogenies with randomly generated branch lengths (c).
Phylogenetic half-life (t½)
The optimal regression estimates the relationship between trait and habitat that would be observed if there were no phylogenetic inertia; the evolutionary relationship results from both adaptation and inertia. The phylogenetic half-life (±1 SD) is expressed in millions of years. Regression coefficients are multiplied by 10.
Coefficients for which the 95% CI did not include zero are indicated in bold and with an asterisk (*).
Leaving aside species with terrestrial larvae, most shape variation was correlated with hydroperiod and foraging position in the water column. Figure 4 illustrates the three significant relationships that emerged from phylogenetic multiple regressions of relative warps on habitat (Table 2c). Tadpoles that are pelagic or surface-feeding had a deep head/body in the gut region, a deep tail that attached high on the head/body, more lateral eye position, dorsal nares and an anterior mouth. Benthic species had the opposite combination of traits, dominated by a shallow head/body and long, shallow tail. Species inhabiting streams also had a long and shallow tail, along with a short, shallow head/body and more lateral eyes. These relationships were significant under all three approaches to setting branch lengths.
The shape of the mouth was not closely associated with habitat (Table 2, Fig. 5). In analyses of the oral disc, the only significant relationship was the more robust jaw sheath of species in ponds compared with species in streams. This outcome did not differ among the assumptions about branch lengths.
Phylogenetic half-life (t½), defined as the time required for a character to evolve half-way to a new optimum after a change in the environment, was highest for oral disc RW1, the shape of the jaw sheaths (Table 2). For oral disc RW2 and the two side-view shape components, t½ was estimated at 6–20 million years, with longer times arising from the phylogeny with deeper branch lengths (Roelants et al., 2007). These estimates are roughly 8–14% of the depth of the entire phylogeny. When t½ was large the optimal regression had a substantially steeper slope than the evolutionary regression, illustrating that adaptation would have been stronger in the absence of phylogenetic inertia (e.g. Fig. 3).
The adaptationist programme has long emphasized associations between phenotypes of organisms and the habitats they occupy. Indeed, this association can be impressive even on the level of populations or closely related species (Lack, 1961; Hespenheide, 1973; Benkman, 1993; Losos et al., 1999; Schluter, 2000; Grant & Grant, 2002). The results of this study could be viewed as supporting the notion that morphological differences among species represent adaptations to the habitats they occupy. Although comparative analyses cannot establish causality, my results do suggest that the body form and mouth structure of tadpoles have evolved in response to changes in habitat use. The direction of evolutionary change was in most cases consistent with functional information about tadpole morphology. However, the opposite interpretation is also possible, because the amount of morphological variation associated with habitat in this study was small, and estimates of phylogenetic inertia were surprisingly high. Before evaluating these alternative interpretations of the results and their general implications, it is perhaps advisable to consider the strengths and weaknesses of the data set.
Anstis (2002) performed what was essentially a common garden experiment with 83 species. This is an obvious strength of her work. My comparison of morphology with habitat depends on all species having been raised under comparable conditions. If Anstis had made drawings of tadpoles sampled from nature, it would have been impossible to distinguish inherent phenotypic variation among species from plastic variation induced early in development by unknown features of the site. Plasticity in morphological shape should not be underestimated in these species; it is frequently large enough to overwhelm average differences between species (Relyea, 2001; Van Buskirk, 2002; Kraft et al., 2005). It is nevertheless fair to ask whether Anstis’s single drawing of each species accurately represented the form typical of that species. Her ‘experiment’ had no replication at the level of populations or sampled tadpoles, so we cannot assign confidence intervals to measurements from individual species. This is not a serious problem for my analysis, which draws its strength from the inclusion of nearly all extant species in a regional fauna, rather than precise estimation of variation within species. Therefore, the key remaining issue is the accuracy of Anstis’s drawings, to which it can simply be noted that she drew specimens that were in her experience typical of each species.
The results themselves also inspire confidence in the quality of the data set. Most significant relationships between tadpole morphology and habitat were consistent with previous comparative and functional studies. Terrestrial tadpoles have evolved three times within south-eastern Australian myobatrachids: genus Assa within Geocrinia, Bryobatrachus within Crinia, and Philoria within Limnodynastes. In each case, the body form has become shallower and more elongate, especially in the tail (Table 2, Fig. 3). This is consistent with earlier descriptions of semi-terrestrial tadpoles as having an elongate head, a shallow tail and relatively large eyes (Altig & Johnston, 1989; Altig & McDiarmid, 1999). The modified form of the body may represent adaptation to a life style in which vermiform locomotion has replaced aquatic swimming. The size of the eye did not contribute to any of the shape components in my analyses (Appendix S4). Bryobatrachus nimbus and Assa darlingtoni have relatively large eyes; the four Philoria do not, but these species occasionally develop in aquatic microhabitats (Anstis, 2002).
Morphological divergence among larvae of the 75 aquatic species has occurred along the habitat dimensions of stream vs. pond, benthic vs. pelagic, and hydroperiod. Two of these habitat dimensions, stream/pond and benthic/pelagic, have been emphasized in descriptions of tadpoles elsewhere in the world (reviewed by Altig & Johnston, 1989; Altig & McDiarmid, 1999). Lotic species are reported to have small eyes and shallow, stiff tail fins with massive tail muscles for maintaining position against the current. The suite of traits associated with occurrence in streams in my analyses, a shallow head/body and long tail (Fig. 4), was slightly different from that reported in the literature, perhaps because few of the south-eastern Australian species inhabit high-gradient streams. Benthic species are reportedly characterized by a depressed head with dorsal eyes and a shallow tail fin, whereas pelagic species can be cylindrical or compressed, with lateral eyes and a tail that is tapered and sometimes very large. Any adaptive significance of these traits is unknown, although it may be important for benthic species to minimize their shadow by maintain a low profile of the head and tail (Altig & Johnston, 1989). Nevertheless, the data in Fig. 4 confirm these patterns in Australian taxa. Use of benthic microhabitats was associated with the evolution of higher values of the first relative warp and lower values of the second, corresponding to an elongate and shallow head/body, more dorsal eyes, a ventral mouth, and a shallow tail.
My results suggest a possible adaptive interpretation of the deep tail in pelagic tadpoles (Fig. 4). The strong positive correlation between pelagic position in the water column and occurrence in permanent wetlands, described by the second habitat axis in Table 1, means that morphological adaptation along this axis could involve selection imposed by either factor. Recent demonstrations of the utility of a deep tail in predator escape raises the possibility that tadpole shape in pelagic/permanent habitats arises from adaptation to high predation risk (Van Buskirk & McCollum, 2000; Van Buskirk et al., 2003). Indeed, in comparison with ephemeral wetlands, permanent ponds often contain high densities of large predators (Woodward, 1983; Wellborn et al., 1996; Schneider, 1997). Thus, the correlation between tail shape and the second habitat axis could reflect adaptation to predators in permanent wetlands rather than adaptation to the pelagic microhabitat.
The shape of structures within the oral disc was weakly associated with habitat (Table 2). The literature supports only a few relationships between tooth rows or jaw sheaths and habitat. For example, species occurring in streams seem to have more and longer tooth rows, which could be useful for clinging to the substrate (Johnston, 1990; Altig & McDiarmid, 1999). In ponds, tadpoles have higher tooth rows with fewer teeth, which cannot generate much force but are sufficiently flexible to secure the oral disc to uneven substrates while the jaw sheaths are removing food (Altig & Johnston, 1989; Johnston, 1990; Wassersug & Yamashita, 2001). Species that eat large prey have wide mouths and well-developed jaw sheaths, often with serrations (Vera Candioti, 2005, 2007). In my study, only the comparison between stream and pond tadpoles was significant (Fig. 5), with stream species having arched anterior tooth rows, a narrow mouth and a fine lower jaw sheath. The comparison between this pattern and observations from the literature is not clear, although the wide mouth and robust jaw sheaths of pond species may indicate that their diet consists of larger, or at least more variable, prey items (Diaz-Paniagua, 1985; Inger, 1986; Hoff et al., 1999).
The previous paragraphs outline possible adaptive interpretations of associations between evolutionary change in habitat and morphology. However, my results also demonstrate that adaptation in this group proceeds slowly. Most habitats and traits showed significant phylogenetic signal, and genera were characterized by recognizable differences in phenotype even though many of them diverged at least 20 million years ago (Marjanović & Laurin, 2007; Roelants et al., 2007). It appears that the shape of the tadpole body and oral disc have resisted evolutionary change over vast periods of time. A possible explanation is that clades exhibit consistent long-term distinctions in habitat use, perhaps enforced by ecological interactions among species; if the ecology of a clade remains constant over many millions of years, then the morphology of its species will as well. But this explanation is inconsistent with the high values of phylogenetic inertia in Table 2. The side-view shape components had the best evidence of correlated evolution with habitat. But even in these traits, estimates of phylogenetic half-life implied that roughly 6–14 million years elapsed while these traits evolved half the distance to a new optimum following an evolutionary shift in habitat use. Simulations in Hansen et al. (2008) suggest that these numbers should not be taken too seriously because estimates of t½ become inaccurate when phylogenetic inertia is high. But we can at least conclude that phylogenetic inertia in tadpole morphology is large relative to the optimizing force of selection. Although comparative studies such as this tend to emphasize evidence for adaptive evolution, it is clear in this case that maladaptation resulting from resistance to adaptation is a major part of the story.
These results sound a cautionary note for studies of adaptive resolution of trade-offs along habitat gradients (Tilman & Wedin, 1991; Smith & Van Buskirk, 1995; Wellborn, 2002). Even as phenotypes of extant species show plenty of evidence for adaptation, they can simultaneously contain the signature of inherited maladaptation. This applies even to characters that might be expected to exhibit evolutionary lability, such as the proportions of a tadpole’s tail or head. Of course, maladaptation in current phenotypes could reflect adaptation to other unmeasured features of the environment. But it would be unwise to simply assume adaptation to unknown factors whenever phylogenetic inertia is encountered, because this makes the adaptive hypothesis impossible to disprove or even qualify (Gould & Lewontin, 1979), and it implies that too little natural history information was known beforehand to formulate testable hypotheses. In the case of amphibian larvae, it would be difficult to argue that we do not yet know which habitat variables correlate with species distributions (Wellborn et al., 1996; Skelly, 2001; Van Buskirk, 2005; Werner et al., 2007). This leaves us with another possibility that may be unpopular in this age of renewed adaptationism, but is at least worth considering: the phenotypes of species occurring along environmental gradients are not usually optimal for the habitat they currently occupy.
Special thanks are due to Marion Anstis for her careful and beautiful drawings of tadpoles, to T. Kawecki and M. Laurin for helpful comments on the manuscript, and to the Australian Research Council and Swiss National Science Foundation for funding.