A comparative analysis of predator-induced plasticity in larval Triturus newts

Authors


  • Present address: Josh Van Buskirk, Zoology Department, Melbourne University, Victoria 3010, Australia.

Benedikt R. Schmidt, Zoologisches Institut, Universität Zürich, Winterthurerstrasse 190, 8057 Zürich, Switzerland.
Tel.: ++41 1 635 49 84; fax: ++41 1 635 68 21;
e-mail: bschmidt@zool.unizh.ch

Abstract

Species that occupy similar habitats are expected to show convergent phenotypes. If habitats are defined by the presence of predators, then traits that modify vulnerability to predation, including predator-induced phenotypic plasticity, should be similar within habitats. We tested this idea using larvae of six syntopic newt species belonging to the two Triturus clades. Behavioural plasticity induced by odonate predators was strongly dissimilar between the two main clades but similar within them. Morphological plasticity was variable among species, even between one pair of closely related species. A predation experiment tested whether differences between clades could be caused by differences in body size. Size-specific vulnerability differed between newts in the small-bodied and large-bodied clades, indicating that similar predators may affect the two clades differently. The results showed both similarity and dissimilarity in predator-induced phenotypic plasticity in syntopic larval newts although theory suggests that divergence is unlikely in such ecologically similar species.

Introduction

Ecologists have long been fascinated by the evolutionary consequences of species coexistence. Both theory and field observations suggest that either phenotypic convergence or divergence can evolve when two species occupy the same habitat. In some cases, the conditions that favour convergence or divergence are well known. For example, when two species are syntopic and competing for resources, divergence in traits related to resource acquisition is expected (Lack, 1947; Brown & Wilson, 1956; Schluter, 2000). Convergent phenotypes are predicted when natural selection arises from environmental conditions that impose inescapable functional demands, which will operate regardless of whether the species are syntopic (MacArthur, 1972; Endler, 1982; Losos, 1990; Reznick et al., 1996). In this case, convergence is unlikely to affect trophic traits, but rather traits related to tolerance or performance with respect to the relevant features of the environment. Numerous empirical examples are available for both of these scenarios (Schoener, 1974; Losos et al., 1997; Schluter, 2000).

The outcome of community evolution is less obvious when habitats are defined by predator composition. The abundance and identity of predators delineate distinct habitat types for prey taxa (von Ende, 1979; Wellborn et al., 1996; McPeek, 1998). Evolutionary responses to predators are ubiquitous, but are convergent or divergent responses more likely? Studies of prey occurring within different predator communities have reported both kinds of response (Endler, 1982; McPeek, 1990; Skelly, 1995; Smith & Van Buskirk, 1995; Reznick et al., 1996; Richardson, 2001; Van Buskirk, 2002). Most models have concluded that convergence is the more likely outcome (Lively, 1986; Brown & Vincent, 1992; Van Tienderen, 1997; Doebeli & Dieckmann, 2000; Day et al., 2002) but McPeek (1996) and Abrams (2000) showed that convergent constitutive defences against predators should evolve in traits that affect vulnerability to predators in the same way for both prey species, whereas divergent defences are expected in traits that affect vulnerability in different ways. It is too early to determine whether and which of these models are empirically supported. Comparative tests of resource-based character displacement are common, but there have been very few comparable studies checking for convergence or divergence in antipredator traits (McPeek, 1990; Reznick et al., 1996; Richardson, 2002; Stoks et al., 2003). Here we offer such a comparison, focusing on predator-induced defensive traits in larvae of European Triturus newts.

Newt larvae display an array of inducible morphological and behavioural defences when exposed to predators (Van Buskirk & Schmidt, 2000; Schmidt & Van Buskirk, 2001; Orizaola & Braña, 2003). We expect these responses to be convergent, for several reasons. First, the mechanisms by which inducible defences protect newts seem likely to function for the larvae of many amphibians and against many predator species (Anholt & Werner, 1995; Doherty et al., 1998; Van Buskirk & McCollum, 2000; Van Buskirk et al., 2003, 2004). If that is the case, then the model by Abrams (2000) predicts convergence. Secondly, multiple Triturus species frequently occur together within the same ponds with strong overlap of their ecological niches (Arntzen & de Wijer, 1989; Kuzmin, 1991; Braz & Joly, 1994; Miaud, 1996; Jehle et al., 2000). Our own pond survey data show that sympatric newt larvae are exposed to very similar predation environments (Fig. 1). Thirdly, some closely related Triturus species are similar in body size (Griffiths, 1996), so even if predation risk is size dependent we expect similar defences in pairs of similar-sized species. Finally, European newts represent a fair test of the theory of convergence and divergence, because the more distantly related species diverged roughly 15 Ma (Zajc & Arntzen, 1999). This time frame provides ample opportunity for habitats and anti-predator traits to show evolutionary change (Losos et al., 1997; Price et al., 2000), especially because evolutionary response to predators can be relatively rapid and sister species often occupy different habitats (Wellborn et al., 1996; Reznick & Ghalambor, 2001). We therefore expect that effects of common phylogenetic history should be weak.

Figure 1.

Distributions of four sympatric Triturus with respect to predator density, in a survey of ponds in northern Switzerland (Van Buskirk & Schmidt, 2000; J. Van Buskirk, unpublished). Adult and larval newts were detected by dip netting and pipe-sampling, in May and June of 1997–2003. Dytiscid beetle larvae and aeshnid dragonflies were the most important predators of newt larvae, but other common taxa included libellulid dragonflies, larval hydrophylid beetles, and adult newts, dytiscids, and Notonecta. There were 534 pond records; sample sizes for each species were 288 T. alpestris, 57 T. cristatus, 55 T. helveticus, and 199 T. vulgaris. The four species were found in ponds with generally similar predator habitats.

Materials and methods

Study animals

We studied six species of Triturus newts (Amphibia: Salamandridae; for natural history information, see Griffiths, 1996), representing both major clades in the genus (Fig. 2). Triturus helveticus and T. vulgaris represent the ‘small brown newts’ and T. marmoratus, T. carnifex and T. cristatus represent the large-bodied ‘crested newts’. Triturus alpestris is usually positioned near the origin of the clade and often as a sister species to the crested newts (Macgregor et al., 1990) but recent analyses suggest that it may be a sister species to the brown newts (Zajc & Arntzen, 1999; Steinfartz et al., 2002). We represent this uncertainty as a trichotomy in Fig. 2.

Figure 2.

Phylogenetic hypothesis depicting relationships among the six species of Triturus, based on Macgregor et al. (1990) and Zajc & Arntzen (1999). Lengths of the branches do not reflect time since divergence.

The genus Triturus may not be monophyletic (Titus & Larson, 1995; Caccone et al., 1997; Zajc & Arntzen, 1999; Steinfartz et al., 2002). If polyphyly is confirmed, then the two main clades [represented in this study by T. vulgaris, T. helveticus, and T. alpestris in one clade (subgenus Palaeotriton) and the crested newts T. marmoratus, T. carnifex, and T. cristatus in the other clade (subgenus Triturus)] may not even be sister taxa but on rather different branches of the family Salamandridae. This would have no impact on our study because it leaves relationships among the six species in Fig. 1 unchanged.

We reared newts from hatching until near metamorphosis under standardized but semi-natural conditions, and recorded their behaviour and morphology in the absence of predators and in the presence of dragonfly larvae. Newt larvae hatched from eggs laid by adults collected in natural populations. Triturus vulgaris and T. alpestris were collected from populations north of Zürich, T. helveticus from a population near Basel, T. carnifex from a site near Geneva, and T. cristatus from two populations near Bern, all Switzerland, and a population near Jublains, France. The T. marmoratus were also collected near Jublains. All species except T. carnifex were syntopic with other Triturus species in the sites where we collected them, and T. carnifex is known to co-occur with T. alpestris.

The predators in our experiments were larval dragonflies (Aeshna cyanea; Odonata: Aeshnidae), collected from several ponds near Zürich. Aeshna was abundant and widespread everywhere we collected newts (B.R. Schmidt, personal observation).

Inducing and measuring predator-induced phenotypic plasticity

The experimental procedures were the same for all six species and are described in detail in our earlier publications (Van Buskirk & Schmidt, 2000; Schmidt & Van Buskirk, 2001), from which we also took the data for three of the six species (T. helveticus, T. alpestris and T. cristatus). Table 1 presents information on the timing of the experiments and densities of newt larvae.

Table 1.  Summary of the experiments measuring predator-induced plasticity in larval Triturus newts.
SpeciesStarting dateDensity (no./pond)ReplicatesReference
  1. *French population on 30 April, Swiss populations on 10 May.

Triturus vulgaris4 June 2001154B.R. Schmidt & J. Van Buskirk unpublished
Triturus helveticus4 June 1997154Van Buskirk & Schmidt (2000)
Triturus alpestris10 May 1998154Van Buskirk & Schmidt (2000)
Triturus marmoratus26 May 200192B.R. Schmidt & J. Van Buskirk unpublished
Triturus carnifex26 May 2000104B.R. Schmidt & J. Van Buskirk unpublished
Triturus cristatus30 April 1999; 10 May 1999*106Schmidt & Van Buskirk (2001)

The experiments took place in outdoor fibreglass stock tanks with a surface area of 1.35 m2 and filled to a depth of 60 cm (800 L volume). Tightly fitting lids, constructed of 35% shade cloth, prevented colonization by unwanted insects or frogs. We added leaf litter (0.5–1 kg; mainly Fagus and Quercus) to the tanks to provide structural heterogeneity and a source of nutrients. We also added 5 g commercially available rabbit food as a second source of nutrients. Artificial ponds were inoculated repeatedly during the first 2 weeks with aliquots of phytoplankton and zooplankton. We added three to five adult Lymnaea sp. snails to control growth of periphytic algae and to promote nutrient cycling. With these procedures, diverse and self-sustaining aquatic communities composed of algae, microbes, phyto- and zooplankton were established in the artificial ponds. We filled the ponds 1–2 months before the newt larvae and predators were introduced to allow enough time for the aquatic communities to develop. The large species T. marmoratus, T. carnifex and T. cristatus almost completely eliminated both zooplankton and juvenile snails. We therefore added 100 tadpoles of Rana ridibunda (Gosner stage 25) once as additional food to each artificial pond; only one tadpole was recovered in one pond at the end of the experiments.

The experiments included two treatments. In one, three late-instar Aeshna larvae were placed in each pond, enclosed within cages to prevent them from killing the newts. This dragonfly density (2.2/m2) is within the range of densities observed in nearby natural ponds (Schmidt & Van Buskirk, 2001). The cages were constructed of a 10 cm length of plastic tube (12 cm diameter), capped at both ends with fibreglass window screen. Artificial ponds in the second treatment were left without predators, but included empty cages. We fed the dragonfly larvae three times a week throughout the experiment, on a diet of 300 mg (per dragonfly larva) of live anuran larvae (mainly Rana sp., Bufo bufo, Bombina orientalis) and Triturus sp. at a ratio of 2 : 1. In every experiment, at least some conspecific newt larvae were fed to the predators.

We measured the behaviour of larval newts by conducting daytime counts of individuals that were visible above the leaf litter several times during the larval period (Van Buskirk & Schmidt, 2000). On each date, we repeated the counts three times for each pond and calculated the mean number visible. For analysis we used the mean proportion visible across all dates.

Body mass and morphology were measured when newt larvae were 6 weeks old. To measure mass, newts were blotted dry and weighed to the nearest 0.001 g on an electronic balance. Morphology was measured from photographs of five to six live individuals per pond, placed in a small water-filled plexiglass chamber. A system of mirrors was installed such that the side- and bottom-view of each larva appeared on the same negative. From digital copies of the photographs, we measured the three-dimensional coordinates of 31 landmarks using image analysis software. The landmarks were situated to identify size and shape, and to measure traits expected to respond to experimental treatments. Details of the digitizing process and two-dimensional maps of the landmarks are available in Van Buskirk & Schmidt (2000) and in the Electronic Archive of the Ecological Society of America: Ecological Archives E081-025 (http://www.esapubs.org/archive/ecol/E081/025/). We used pairs of landmarks to calculate four linear distances (torso length, tail length, tail depth and tail muscle width) that are known to affect predation mortality (Van Buskirk & Schmidt, 2000). Morphometric analysis focused on size-corrected traits, which were the residuals from a regression of the original measures against the square root of centroid size (the sum of the squared distances among all pairs of landmarks; Bookstein, 1991).

We used univariate and multivariate analyses (proc GLM in SAS 8.1, SAS Institute, Cary, NC, USA) to determine whether phenotypes of the six species differed and were affected by predators. Univariate tests were performed on mass and behaviour, while multivariate tests were used for the morphological responses. We could not employ formal phylogenetic analyses because quantitative data on habitat distributions are not available for these species, so comparisons between traits and habitats were not possible. Instead, we evaluated the extent of evolutionary convergence or divergence by inspecting patterns of phenotypic similarity and dissimilarity in the context of the phylogenetic hypothesis (Fig. 1). Adaptive divergence will be supported if species that share recent common ancestors exhibit strongly dissimilar phenotypes. Neutral divergence will be supported if the degree of phenotypic similarity is closely associated with phylogenetic relatedness. The outcome that we expected is that species show convergent phenotypes as a result of their shared habitat distributions, and this will be supported if the species are all very similar. Similar phenotypes could also occur if species share a habitat and do not diverge over long periods of time. Lack of divergence and convergence may reflect similar evolutionary processes, because both arise from selection acting within ecologically similar habitats. Our analyses do not test directly for convergence. If, however, species are significantly dissimilar then this would suggest divergence or lack of convergence. Both neutral and adaptive divergence or lack of convergence would run counter to our expectation.

Testing whether predation risk is size-dependent

The two major clades of Triturus differ in body size as larvae and adults (Griffiths, 1996). We therefore tested whether predation risk is size-dependent and whether body size differences can explain variation in predator-induced plasticity among species. We chose T. alpestris to represent the ‘small’ newts and T. carnifex as a representative of the crested newts.

Predation trials were conducted in outdoor tubs (0.28 m2, 80 L). We added a 50 cm length of the aquatic plant Myriophyllum sp. as a perch for the predator, a single dragonfly larva (Aeshna cyanea). Before each predation trial, both a newt larva and a dragonfly larva were weighed and put into a tub. After 24 h, we recorded whether the newt had survived or was killed. We ran 60 predation trials for T. alpestris on five sunny days (10 or 20 trials per day) from 2 July to 16 July, 2001, and 70 predation trials for T. carnifex on seven sunny days from 25 July to 8 August, 2000. On every day, we used larvae of the widest size range possible. The newt larvae were predator-naïve.

We used model selection to analyse the data in an explorative way (Burnham & Anderson, 2002). We were interested in which factors and combinations among them affected the probability that a newt larva was killed. The factors were species, predator size, and size of the newt larva. Candidate models are shown in Table 3. We then used the small-sample Akaike information criterion (AICc) to rank the candidate models. AICc ensures that the models neither under- nor overfit the data. We calculated the Akaike weight of each model, which reflects the relative support for a model and is scaled such that the sum of the Akaike weights of all candidate models is 1; an Akaike weight of 0 indicates no support whereas 1 indicates full support. The Akaike weight is a posterior model probability (the relationship to Bayesian posterior model probabilities is discussed in Burnham & Anderson, 2002, p. 302).

Table 3.  Model selection analysis of predation trials. K, number of parameters in the model; N, sample size.
ModelLog-likelihoodKNAICcΔAICcAkaike weight
  1. *Includes newt size, predator size, species and all interactions.

  2. †Main factors plus interaction.

  3. ‡Intercept only.

  4. §Newt size, predator size and interaction.

(a) Triturus alpestris and T. carnifex in the size range of the former
 Global model*−54.388101126.330.000.535
 Predator size−61.642101127.411.080.312
 Newt size × predator size†−60.294101129.002.660.141
 Null model‡−66.881101135.809.470.005
 Newt size−65.952101136.029.690.004
 Species−66.702101137.5311.190.002
 Newt size × species†−65.624101139.6613.330.001
(b) Triturus carnifex (full size range)
 Global model§−33.6347075.840.000.995
 Predator size−41.1627086.5010.660.005
 Newt size−43.2027090.5714.730.000
 Null model‡−48.2817098.6222.780.000

The candidate models were fit to the data using proc LOGISTIC in SAS 8.1. We ran two separate analyses. In the first, we analysed the data from the two species together including all T. alpestris and the T. carnifex that were within the size range of T. alpestris. In the second, we analysed data from T. carnifex only, including the full size range of this species. Because there was model selection uncertainty (i.e. several models explained the data well), we used model-averaging techniques to estimate the survival probability of each newt larva (Burnham & Anderson, 2002). Model-averaged estimates are mean values of survival probabilities from all candidate models weighted by their Akaike weights.

Results

Phenotypes and predator-induced phenotypic plasticity

The experiments uncovered differences and divergence in body size, behaviour and morphology, and in predator-induced plasticity in these traits, among congeneric and often syntopic larval newts (Fig. 3). Mass at 6 weeks of age varied primarily among species and not between predator treatments; there was also no significant interaction (Table 2). The sizes of larvae were positively associated with adult body sizes. The three crested newt species are much larger as adults than the other three species, and their larvae were generally larger. The two brown newt species are small as adults, and their 6-week body sizes were also smaller. Early growth rate therefore seems correlated with adult body size in this group.

Figure 3.

Life history, behaviour, and morphology of six species of Triturus newt larvae in the absence (open symbols, mean ± SE) and presence of predators (closed symbols). Mass (g) and morphology (cm, residuals after size correction) were measured once after 6 weeks; behaviour (proportion in open water) is the mean of several measurements through the larval period. The six species are aligned in the same sequence shown in the phylogeny (Fig. 2; vul = Triturus vulgaris, hel = T. helveticus, alp = T. alpestris, mar = T. marmoratus, car = T. carnifex, cri = T. cristatus).

Table 2.  Summary of univariate and multivariate anova on plasticity in life history, behaviour, and morphology in larvae of six species of Triturus newts. Entries for individual morphological traits are coefficients of the dominant eigenvector; an asterisk indicates statistical significance in univariate anova at α = 0.05.
Source of variationd.f.Univariate anova on
MassBehaviour
FPFP
Species5, 3174.45<0.000164.41<0.0001
Predator treatment1, 310.040.83600.730.4007
Species × predator5, 310.670.647818.61<0.0001
Source of variationd.f.manova on morphology
FPTorso lengthTail lengthTail depthMuscle width
Species20, 93.8112.26<0.00017.10*3.21*4.57*2.11*
Predator treatment4, 283.650.01621.93*−0.30−6.17 *3.38*
Species × predator20, 93.81.920.01911.40−1.45*−0.945.35*

Behaviour was strongly affected by species identity and an interaction between predator treatment and species; again, there was no significant main effect of predator treatment (Table 2). Variation among species mostly arose from a difference between the two clades. The crested newts T. marmoratus, T. carnifex and T. cristatus were highly visible under both treatments and increased their visibility in the presence of predators, whereas the other three species were more often hiding and responded to predators by decreasing visibility. We tested the a posteriori hypothesis of a clade effect on behaviour and plasticity using a nested anova. In this analysis, the effects of clade, species nested within clade, and the clade-by-predator interaction were significant (clade: F1,4 = 236.0, P < 0.001, species: F4,42 = 12.0, P < 0.001, clade-by-predator treatment interaction: F1,4 = 34.5, P < 0.001). The main effect of predator treatment was not significant (F1,42 = 0.7, n.s.) and the species-by-predator treatment interaction was marginally nonsignificant (F1,42 = 2.5, P = 0.065). This suggests that the overall difference in behaviour and plasticity in behaviour evolved when the two major clades separated and may covary with body size (clade and body size are highly correlated).

Morphological shape was significantly affected by species, predator treatment, and the species-by-predator interaction (manova in Table 2). Univariate tests demonstrated that the six species differed strongly in all morphological traits. The crested newts had relatively long torsos and deep tails. Morphological differences between the two brown newts were striking, because T. vulgaris had a much longer, wider, and deeper tail than did T. helveticus. The interaction between species and predator treatment was significant in univariate anovas on tail length and tail muscle width, indicating that plasticity in these two traits differed among species. Some species showed no plasticity in these traits, whereas others showed increases or decreases in trait values in response to predators. For example, T. vulgaris showed little plasticity in tail length, while tail length decreased in T. helveticus and increased in T. alpestris (Fig. 3). Plasticity in torso length and tail fin depth was always in the same direction for the species that showed plasticity.

Testing whether predation risk is size dependent

Larvae of Triturus newts differ in body size. If vulnerability to predation depends on size then larvae may experience predators differently even when they are syntopic, and this could select for dissimilar phenotypes. We therefore tested whether predation risk is size dependent.

The predation trials revealed complex interactions between species and size of the newt larva and predator. The best-supported model was the most complex one in all cases (Table 3). In the analysis of both species, the best model included species, the sizes of the newt larva and the predator, and all interactions. In the analysis of T. carnifex alone the best model included the sizes of the newt larva and the predator and their interaction. Of the three variables, predator size appeared to have the strongest effect (as evidenced by the fact that the ‘predator’ and ‘predator × newt size’ models were well supported). For Triturus alpestris, the model-averaged survival probabilities were between 20 and 75%; for T. carnifex they ranged from 10 to 90% (Fig. 4a,b). Small T. carnifex larvae were killed with high probability by predators, whereas they survived well when they were large and/or the predator was small (Fig. 5).

Figure 4.

Survival probabilities of larval T. alpestris (filled symbols) and T. carnifex (open symbols) in relation to (a) their size and (b) the size of the predator (Aeshna cyanea). Only T. carnifex that were within the size range of T. alpestris are included in this figure. Each symbol represents the model averaged survival probability of a single individual estimated from the logistic regressions in Table 3a.

Figure 5.

Survival of T. carnifex larvae in relation to their size and the size of the predator (Aeshna cyanea). Mass is given in grams. Open symbols represent newt larvae that died, closed symbols represent surviving newt larvae. The lines indicate survival probabilities estimated from the best model in Table 3b (model averaging was not done because there was no model selection uncertainty).

Discussion

Although we expected to find widespread similarity of Triturus phenotypes, we described several examples of evolutionary divergence in plasticity of traits related to escaping predators. The two clearest cases involve differences between the two closely related brown newt species, T. helveticus and T. vulgaris, and differences between the two main Triturus clades.

Although the two brown newts T. helveticus and T. vulgaris are closely related, they are not similar in their shape and morphological responses to predators. Triturus vulgaris is phenotypically more similar to the distantly related crested newts in its tail depth and muscle width, while T. helveticus resembles T. alpestris in torso length, tail shape, and plasticity in several traits. Differences between the brown newts (and their similarities with other taxa) evolved as they shared a common ancestor, and therefore represent evolution in different directions. Other good examples of divergence involve T. cristatus, which exhibited several traits that were more similar to T. alpestris than to the other crested newt species. For example, tail length and plasticity in tail length of T. cristatus were quite unlike those of its close relatives, and instead more similar to those of T. alpestris and T. helveticus. This suggests that tail shape has evolved in different directions since the origin of the crested newt clade.

The two main Triturus clades differed strongly in their behavioural response to dragonfly predators. All crested newts showed increased visibility in the presence of predators, whereas all other species reduced their activity with predators. This establishes crested newts as an exception to the almost universal rule of reduced activity under predation risk (Lima, 1998). Phenotypic differences between the clades could represent adaptive divergence in the distant past or gradual divergence under neutral evolution. In either case, extant variation among clades is not necessarily currently adaptive.

These findings run counter to our prediction that all species should have similar phenotypes. This expectation was based on both theory and natural history information. The six species occupy extremely similar habitats during the larval stage in most parts of Europe (Kuzmin, 1991; Braz & Joly, 1994; Miaud, 1996; Jehle et al., 2000; Fig. 1). Optimality and quantitative genetic models of evolution assume that species exposed to identical selection will converge upon similar phenotypes, as long as they are subject to similar constraints. If these conditions were upheld, we should have observed more similar traits and responses to predators in the six Triturus. The species differ in size, but the defensive mechanism of the antipredator phenotypes appear to be the same for all sizes.

Why do Triturus exhibit dissimilar phenotypes?

We can imagine three general explanations for the pattern of dissimilarity in traits related to predator avoidance and escape. The first is that there may be important but overlooked differences among species in habitat use and the constraints that govern their evolutionary dynamics. The second is that species may possess combinations of traits that represent multiple evolutionary equilibria, each of which is an adaptive response to the same set of habitat conditions. We will illustrate this explanation using the two brown newts. The third is that dissimilar phenotypes may have evolved in response to habitat differences in the past. We discuss this hypothesis using the divergence between the main Triturus clades.

We assumed that Triturus larvae experience the same predator habitats because many of these species occur syntopically throughout their ranges and our own data show broadly similar exposure to predators for four of the species (Fig. 1). But this assumption may not hold at the level of microhabitat within ponds. Triturus cristatus forages more pelagically than does T. alpestris (Braz & Joly, 1994; Schmidt & Van Buskirk, 2001), so the encounter rates of these species with different kinds of predator may be rather different even when they occur together within the same pond. Such microhabitat differences could create effectively distinct habitats for co-occurring newts, and such differences in other taxa are known to have caused the evolution of divergent syntopic forms (Schluter, 2000).

Even if multiple newt species occupy identical microhabitats, it is still possible that other features unique to each of them would prevent convergence upon similar phenotypes. For example, we do not expect similarity in morphology and behaviour if species differ in other traits important for escaping predators, such as palatability or body size. All Triturus larvae are palatable to predators until they approach metamorphosis. But the difference in body size between the crested newts and other species could influence the evolution of other responses to predators. The crested newts achieve large adult size in part by growing rapidly during the larval stage. In other taxa, rapid growth is known to have negative consequences for traits that affect vulnerability to predators, such as swimming speed, toughness of scales, and ossification of bone (Arendt & Wilson, 2000; Arendt et al., 2001; Lankford et al., 2001; Munch & Conover, 2003). The possibility that crested newt larvae are of ‘lower quality’ was supported by the predation experiment, in which small T. carnifex survived much less well than small T. alpestris when faced with large and dangerous dragonflies. This difference could be a consequence of rapid growth rate in T. carnifex. At the same time, it may account for the apparent constitutive morphological defence against predators possessed by the crested newts: all species had large tail fins and muscles regardless of the treatment.

Data from the predation experiment illustrate how syntopic newts can experience risk imposed by the same predators in very different ways. The survival of T. carnifex was much more sensitive than that of T. alpestris to the size of the predator (Fig. 4b). Although the causes of this difference are not known, its implications for our hypothesis of convergence are obvious. Triturus larvae that occupy similar habitats may not experience them similarly, and therefore may not undergo selection favouring convergent phenotypes. Credible reasons for this include differences both external and internal to the newts, involving differences in microhabitat and other components of the phenotype.

A second general explanation for the differences among species is that they experience comparable environments but have come to occupy distinct adaptive optima. An illustration of how this could work is provided by the two brown newts, which differed strongly in morphological plasticity but were similar in behavioural plasticity. Van Tienderen (1991) has modelled a scenario in which plastic and nonplastic outcomes are both optimal at the same time. The evolution of behavioural responses to predators, which function during the encounter and detection phases of the predation sequence, are probably best explained by Van Tienderen's soft selection model. Here, fitness depends in part on the phenotypes of other larvae in the population, as is true if individuals that are relatively active encounter predators first (Werner & Anholt, 1993). The model predicts a single adaptive peak exhibiting phenotypic plasticity, and this agrees with our observation of behavioural plasticity in both brown newt species. In contrast, morphological responses to predators may be under hard selection, as they function to protect the larva once an attack is underway and the consequence of an individual's morphology is unlikely to depend on the phenotypes of other individuals in the population. In this case Van Tienderen's (1991) model predicts, for a broad range of conditions, that both nonplastic habitat specialists and plastic generalists are favoured simultaneously. This agrees with our observation that T. helveticus exhibited morphological plasticity while T. vulgaris did not. More generally, this example illustrates that selection for dissimilar anti-predator phenotypes can occur even when two species occupy identical environments.

One of the more dramatic examples of dissimilar phenotypes, that in behavioural plasticity between the two Triturus clades, may have evolved in response to different habitats occupied in the past. Recent studies of salamandrids suggest that Triturus may be polyphyletic (Titus & Larson, 1995; Caccone et al., 1997; Zajc & Arntzen, 1999; Steinfartz et al., 2002), with crested newts forming a clade of pond-dwelling salamanders within a larger clade of stream-dwelling salamanders (e.g. Chioglossa, Euproctus, Salamandra and Salamandrina). A common antipredator response of salamander larvae that inhabit streams is not to hide but rather to actively drift out of pools with predators (e.g. Sih et al., 1992). When crested newts invaded ponds, they may have maintained the ancestral behavioural response to predators, which may be adaptive in ponds as well because dragonfly larvae are not found in open water. In contrast, the small brown newts may have evolved or retained an ancestral hiding response when exposed to predators. Interestingly, the behavioural differences evolved in concert with differences between clades in growth rate and larval body size; theory suggests that different life histories may evolve in prey that are strongly coupled to different predators (Day et al., 2002). Whatever the true relationships among Triturus newts, differences between the two main clades contradict our expectation of similar phenotypes in species that are often syntopic. If crested newts evolved in the same habitat as the other species, then they diverged. If they shift from a stream habitat to a pond habitat, then selection imposed by predators may have caused some convergence, but the species remain quite dissimilar nevertheless.

Conclusions

The results of our comparative analysis imply that the major components of the phenotype evolve relatively independently. Figure 2 depicts a patchwork of life historical, behavioural and morphological traits that show nearly no correlation across species. The morphological traits, and especially the three reflecting tail shape, tended to co-vary among species, but were not correlated with mass or visibility. The two brown newts were the shyest species, but could not have been more different in morphology. Although T. marmoratus and T. alpestris had similar growth rates, their behaviour and morphology were vastly different. Thus, species possess traits that have experienced divergent evolution (e.g. morphology in the brown newts) and convergence (behaviour within the two clades). At the outset we expected to see primarily convergence because the traits we studied are known to be under selection, the protective mechanism general, and antipredator phenotypes can evolve rapidly.

The evolution of convergence and divergence in antipredator traits has received much interest from theoreticians (Lively, 1986; Brown & Vincent, 1992; Van Tienderen, 1997; Abrams, 2000; Doebeli & Dieckmann, 2000; Abrams & Chen, 2002; Day et al., 2002). Yet, none of these models could have predicted the mixture of convergence and divergence that we observed (Fig. 3). We need models that explicitly include all trophic levels from resources to predators (Day et al., 2002) and that allow for the joint evolution of multiple antipredator traits that may trade-off or act in a compensatory way. If such theory is mathematically tractable and experimentally testable, we will gain a better mechanistic understanding of the evolutionary ecology of predator–prey communities.

Acknowledgments

We thank H. Scheuber, M. Arioli, G. Guex, R. Jehle and J. W. Arntzen for their help with collecting adult newts or providing larvae, P. Ramseier, M. Arioli, R. Altwegg, and C. Vorburger for their help with the experiments, and U. Reyer and two reviewers for comments on the manuscript. We thank the Veterinäramt des Kantons Zürich for permits to conduct the experiments (nos. 128/97, 66/98, 52/99, 84/00, 76/01). We were supported by the Schweizerischer Nationalfonds and a grant from the Forschungskredit of the University of Zürich.

Ancillary