Disruptive natural selection predicts divergence between the sexes during adaptive radiation

Abstract Evolution of sexual dimorphism in ecologically relevant traits, for example, via resource competition between the sexes, is traditionally envisioned to stall the progress of adaptive radiation. An alternative view is that evolution of ecological sexual dimorphism could in fact play an important positive role by facilitating sex‐specific adaptation. How competition‐driven disruptive selection, ecological sexual dimorphism, and speciation interact during real adaptive radiations is thus a critical and open empirical question. Here, we examine the relationships between these three processes in a clade of salamanders that has recently radiated into divergent niches associated with an aquatic life cycle. We find that morphological divergence between the sexes has occurred in a combination of head shape traits that are under disruptive natural selection within breeding ponds, while divergence among species means has occurred independently of this disruptive selection. Further, we find that adaptation to aquatic life is associated with increased sexual dimorphism across taxa, consistent with the hypothesis of clade‐wide character displacement between the sexes. Our results suggest the evolution of ecological sexual dimorphism may play a key role in niche divergence among nascent species and demonstrate that ecological sexual dimorphism and ecological speciation can and do evolve concurrently in the early stages of adaptive radiation.

thus stall the progress of speciation and adaptive radiation. This idea has been formalized for the case of speciation under sympatry (Bolnick & Doebeli, 2003;Cooper, Gilman, & Boughman, 2011;Van Dooren, Durinx, & Demon, 2004), and verbally generalized to adaptive radiation in general (Butler, Sawyer, & Losos, 2007). Theory predicts the dimorphism-as-constraint hypothesis to manifest as a correspondence between divergent selection and phenotypic divergence between the sexes and species over the course of diversification. This prediction would be expected even (or perhaps especially) if speciation has not occurred in sympatry; unless genetic constraints on sexual dimorphism are complete, sexual dimorphism is expected to evolve before sympatric speciation (Bolnick & Doebeli, 2003) in these models and thus the constraint hypothesis may be especially relevant in clades characterized by allopatric speciation.
An alternative view is that the evolution of ecological sexual dimorphism may often play an important positive role in adaptive radiation (De Lisle & Rowe, 2015b). For example, under the likely common conditions of allopatric divergence between nascent species (Coyne & Orr, 2004), and sex-specific phenotypic optima (Cox & Calsbeek, 2009), Lande's (1980) model of the evolution of sexual dimorphism leads to the prediction that population mean fitness (and thus probability of establishment of a nascent allopatric species) depends on the ability of males and females to reach their phenotypic optima. This suggests that rapid evolution of sexual dimorphism will play an important role in successful establishment of a nascent allopatric species whenever phenotypic optima differ for the sexes. In the case of sexually antagonistic (SA) natural selection driven by frequency-dependent resource competition (Slatkin, 1984), the predicted positive effect of dimorphism on species establishment may even be exacerbated as the strength of intraspecific resource competition is reduced during the course of ecological character displacement between the sexes.
Although a direct test of this extension of Lande's model is difficult, one signature would be the evolution of ecological sexual dimorphism that increases local adaptation during niche divergence associated with ecological speciation.
Direct empirical tests of the role of ecological sexual dimorphism in adaptive radiation are limited. Past studies have attempted to relate patterns of macroevolutionary diversification and community assembly to measures of sexual dimorphism (Butler et al., 2007;Dayan & Simberloff, 1994;De Lisle & Rowe, 2015b;Hendry, Guiher, & Pyron, 2014;Schoener, 1977;Stephens & Wiens, 2009). Although these studies are interesting, they are somewhat unsatisfying in that (1) all predictions for a role of ecological sexual dimorphism in adaptive radiation depend upon processes unfolding in the early stages of ecological speciation and (2) these past studies have no direct evidence of an ecological cause of sexual dimorphism.
Here, we examine the joint evolution of phenotypic divergence between the sexes and among lineages in a relatively young clade of North American salamanders, the newts Notophthalmus. All three species of Notophthalmus are generalist semiaquatic predators that breed in ponds and lakes (Petranka, 1998). The most well studied species, the eastern newt N. viridescens has diverged rapidly as the last ice age into four subspecies adapted to unique niches associated with aquatic life (Takahashi & Parris, 2008;Takahashi et al., 2014). N. v. viridescens inhabits Appalachian upland forest and temporary ponds and is adapted to spend much of its life on land, with a terrestrial juvenile stage and adult migration out of the aquatic environment outside of the breeding season (Sever, 2006 (Croshaw et al., 2013;Petranka, 1998;Sever, 2006;Takahashi & Parris, 2008;Takahashi, Takahashi, & Parris, 2011;Takahashi et al., 2014). These differences in aquatic life cycle reflect adaptation to differences in both the quality of terrestrial habitat and breeding pond permanence (Sever, 2006;Takahashi & Parris, 2008). The natural histories of N. meridionalis and N. perstriatus, the two other species in the genus, have been characterized in much less detail although evidence suggests N. perstriatus may be adapted for predominately aquatic life (Petranka, 1998).
Past work shows a significant ecological component to sexual dimorphism in N. viridescens, in particular N. v. viridescens. Males and females have diverged subtly but significantly in head shape; females have wider gapes and shorter, shallower heads for their size than do males (Figure 1), and this morphological dimorphism is associated with divergence in diet (De Lisle & Rowe, 2015a). The sexes also exhibit divergent within-pond microhabitat use, parasite loads, and sensitivity to heterospecific resource competitors (Grayson, De Lisle, Jackson, Black, & Crespi, 2012;De Lisle & Rowe, 2014;De Lisle & Rowe, 2015a, 2015c. Using a series of artificial pond experiments, we have shown that this ecological sexual dimorphism is in part an outcome of resource competition-driven disruptive natural selection (De Lisle F I G U R E 1 Sexual dimorphism in head shape in N. viridescens (ventral view). Males (left) have narrower gapes and longer lower jaws than females (right), which tend to have shorter jaws for their size and wider gapes. A representative lateral view is provided at the top. The four linear traits measured are indicated by black lines. SVL, snout-vent length; HD, head depth; JL, jaw length & Rowe, 2015a). Fitness and selection within ponds is negative frequency (sex ratio)-dependent, and the strength of competition mediates the strength of disruptive, SA natural selection on morphology.
These results are consistent with models of disruptive selection driven by frequency-dependent resource competition and provide one of the most explicit tests of ecological character displacement between the sexes. Thus, past work suggests that morphological sexual dimorphism in N. v. viridescens is at least in part an outcome of SA natural selection, with other sources of SA selection more directly related to the sex roles also likely playing a role (De Lisle & Rowe, 2015a).
Our finding of disruptive natural selection within breeding ponds in N. v. viridescens, combined with recent adaptive ecological divergence among subspecies and species of Notophthalmus, allows for an empirical test of competition's potential role in both the evolution of sexual dimorphism and adaptation during speciation in a young adaptive radiation. Specifically, we can make three predictions: if similar processes of resource competition drive evolution of sexual dimorphism across the genus, then (1) we expect SA natural selection to align with morphological divergence between the sexes, and (2) the extent of sexual dimorphism to associate with an aquatic life history. This prediction can be made on the basis of biomechanical trade-offs between foraging in aquatic (via bidirectional suction) versus terrestrial (via tongue prohesion) habitat ; there is an a priori expectation that selection on multivariate feeding morphology would differ fundamentally across these habitats, as reflected in the intermediate feeding morphology that characterizes semiaquatic newts Heiss, Aerts, & Van Wassenbergh, 2013) and the morphological changes that underlie transitions between these habitats in salamanders and other vertebrates (Schwenk, 2000). Thus, if character displacement between the sexes prevails in aquatic habitat then we expect correlated evolution of sexual dimorphism and adaptation to aquatic life. Finally, (3) if the evolution of sexual dimorphism constrains diversification, then we would expect a correspondence between disruptive natural selection, phenotypic divergence between the sexes, and divergence between species. Note that this prediction holds regardless of how speciation has proceeded.

| Theoretical background: geometry of sexual dimorphism
Sexual dimorphism can be described geometrically by a canonical discriminant analysis (De Lisle & Rowe, 2015a), where the direction through trait space that defines the maximum amount of divergence (relative to the phenotypic variance) between the sexes is described by the vector of canonical coefficients where P is the pooled phenotypic covariance matrix and z is a column vector of sex-specific phenotypic means (Campbell & Atchley, 1981;Mitteroecker & Bookstein, 2011). The canonical coefficient vector c is related to the magnitude of multivariate sexual dimorphism by where D 2 is the Mahalanobis distance between the sexes. Thus, the vector c is a measure of variance-standardized multivariate sexual dimorphism that captures an element of both the orientation and magnitude of divergence between the sexes that is comparable across populations and taxa (n.b. if the data are standardized to unit variance prior to calculating c (the total sample standardized canonical coefficients), such a comparison is mathematically equivalent and conceptually related to among-group comparisons of standardized linear selection gradients , as can be defined as the vector of canonical coefficients defining the population before and after selection (Mitteroecker & Bookstein, 2011)).
Evolutionary change in variance-standardized sexual dimorphism can then be described as where the primes denote parameters occurring in some set after a selective epidose(s). In a group of multiple independently evolving populations Equation 3 would extend to a variance and total evolutionary divergence in multivariate sexual dimorphism across a clade can be described by the second-order tensor where the diagonals of S describe the among-taxa variance in the standardized canonical coefficients (c 1−k , where k is the number of traits in the canonical discriminant analysis) and the off-diagonals describe the among-taxa covariance between standardized coefficients. Thus, S is a covariance matrix whose first eigenvector, s max , describes the direction through trait space where species have diverged the most in multivariate sexual dimorphism (for a discussion of the use of higher order tensors in comparative quantitative genetic analyses, see Hine, Chenoweth, Rundle, & Blows, 2009;Aquirre, Hine, McGuigan, & Blows, 2014). Importantly, Equations 1 and 3 imply that when selection is SA, one prediction is that sexual dimorphism may evolve so that Δ z m −z f and thus Δc and c � = P −1 z � m −z � f aligns with the direction of maximum SA selection (see, e.g., Lande, 1980;Wyman, Stinchcombe, & Rowe, 2013). In the case of competition-driven frequency-dependent disruptive natural selection (i.e., ecological character displacement between the sexes), the first eigenvector of the γ matrix of nonlinear selection is the theoretically appropriate description of the direction of maximum selection (De Lisle & Rowe, 2015a). If the same pattern of SA selection is conserved through the history of a radiation, as, for example, could be the case under phenotype-mediated frequency-dependent resource competition (Rueffler, Van Dooren, Leimar, & Abrams, 2006), then among-lineage lineage variation in c captured by the first eigenvector of S may be predicted to align with this pervasive SA selection. Alternatively, if speciation has occurred under sympatry and was driven by competition-driven disruptive selection, then conserved disruptive selection may instead be expected to align with among-species variance in phenotypic grand means, D. Geometric comparisons of variation in dimorphism S, selection, and total among-species phenotypic divergence D then allow for tests of how, and perhaps even why, the sexes and species diverge during adaptive radiation.

| Data collection
We examined and measured specimens of Notophthalmus from the Carnegie Museum of Natural History, the Smithsonian Museum of Natural History, and the American Museum of Natural History.
Subspecies of N. viridescens were identified by locality and coloration (i.e., dorsal spot/stripe pattern Petranka, 1998). We avoided measuring individuals from phenotypically intergraded populations at the subspecies range margins because assigning subspecies status to such individuals/populations is tenuous. We sexed adult specimens based on cloacal morphology and secondary sex traits when present (e.g., nuptial pads) and confirmed sex by examination of gonads for any specimens whose abdomens had been previously incised. For each specimen, we measured snout-vent length, gape, lower jaw length, and head depth (from the lower right jaw to the top of the orbital skel-

| Statistical analysis
Our analysis entailed two general approaches. First, we fit multivariate and univariate mixed models with fixed effects of taxa, sex, and their interactions to test the hypotheses that sexual dimorphism and among-species differences are statistically significant. We then fit a second series of mixed models to estimate S and D matrices, which were used for geometric comparisons with selection that are most consistent with theory (e.g., see above; Theoretical background section). This approach allows us to both leverage well-developed and powerful fixed-effect hypothesis tests in addition to exploring the geometry of among-taxa variation. Note it is not feasible to estimate both the hypothesis tests of fixed effects and the random effect covariance matrices in a common model; for example, an amongspecies phenotypic covariance matrix D would be conditioned on any fixed effect of sex, making geometric comparisons with sexual dimorphism a trivial outcome of the model specification.
To assess whether sexual dimorphism exists across the genus and differed across taxa, we fit a multivariate mixed model with trait values as the response vector, and trait type, sex, taxon (N. perstriatus, N. meridionalis, N. v. ssp), and all two-and three-way interactions as fixed effects. We included trait as an R-side (repeated measures) random effect with individual as the subject, modelling a separate phenotypic covariance matrix for each taxon. This residual covariance structure fit better than a common phenotypic covariance (∆AIC = 59.66).
To compare trait-specific effects, we also present univariate mixedmodel analyses. These models included sex, taxon, and their interaction as fixed effects, and the taxon-specific residual variance as a random effect. We performed univariate tests on both the raw trait values and size-corrected head traits (residuals from a least-squares regression of gape, jaw length, or head depth on the first principle component of the total phenotypic correlation matrix for all four traits) to illustrate patterns in the data that are independent of size; note that multivariate analyses can accommodate such size-independent effects because the covariance structure of the data is explicitly modeled.
We repeated our multivariate analysis, but limited to subspecies of N. viridescens, to test the hypothesis that dimorphism differs across subspecies of N. viridescens. We quantified the extent of sexual dimorphism for visual presentation by Mahalanobis distance between the sexes, which is a standardized estimate of multivariate effect size.
Confidence intervals for Mahalanobis distance were obtained from percentiles of the sampling distribution constructed from a nonparametric bootstrap (1,000×).
We estimated the among taxa divergence matrix, D (i.e., the among-species/subspecies phenotypic covariance matrix (Lande, 1979)), by fitting a multivariate mixed model with trait values (both sexes pooled in the same analysis) as the response vector and and trait type as a fixed effect to center the data. This model included a G-side random effect of trait type with taxon as the subject to estimate D. We where the intercept and sex*trait effect are included in the same random term) or fitting two random terms to the same set of subjects, which failed to converge.
Our approach to data standardization aimed to be as consistent as possible across analyses; in no case did we standardize separately for males and females, as this was not performed in past work (De Lisle & Rowe, 2015a) and would result in nonlinear discriminant analyses.
None the less, we explored a variety of standardizations and conclusions remain unchanged in all cases. We also obtained the same results using the mean discriminant vector (rather than s max ) to describe divergence between the sexes.
We calculated the association between divergence in sexual dimorphism, s max , or divergence among species, quantified as d max ,   This approach of resampling the information matrix is a robust method for constructing confidence intervals for arbitrary linear or nonlinear functions of random-effect covariance parameters estimated from mixed models (Houle & Meyer, 2013, 2015, and has a major computa-  Kelly & Price, 2004). Nonetheless, we present an analysis of phylogenetically informed, Brownian motion analogs of D and S in the Appendix S1 for completeness. Conclusions on the relationship with γ max remained qualitatively equivalent and quantitatively similar.
All statistical analyses were performed in SAS/IML 9.3 (SAS institute, Cary, NC, USA), with the exception of supplementary analysis performed in R. Mixed models were fit by REML in the glimmix procedure.
All random effect covariance matrices were modeled as the Cholesky parameterization of an unstructured covariance matrix to insure positive semidefiniteness. We used the Kenwood-Roger approximation for degrees of freedom for models with fixed effects, and report type III Wald F tests for these effects (Littell, Milliken, Stroup, Wolfinger, & Schabenberger, 2006).

| RESULTS
We found significant multivariate sexual dimorphism in morphol-   Table 2). Amongtaxa variation in sexual dimorphism, however, measured as the first eigenvector of the covariance matrix of canonical coefficients defining the sexes, s max, was strongly aligned with γ max (vector correlation r v = −0.843; 95% confidence interval for r v = −0.606 to −0.87; Figures 5   and 6). Nonoverlapping confidence intervals indicate that amongspecies and among-sex correlations with γ max were also significantly different from each other, and the vector correlation between s max and d max indicate the association between total variation among species and variation in sexual dimorphism is weak (vector correlation r v = 0.276; 95% confidence interval for r v = 0.009-0.44; Figure 6).
These geometric comparisons of covariance matrices support the patterns illustrated in Figures 2 and 3; variation among species primarily reflects variation in absolute trait size (Figure 2), while divergence between the sexes is most strongly associated with variation in head shape independent of body size (Figure 3). Although theoretical models of sympatric speciation, under some conditions, can predict an outcome of orthogonal divergence between the sexes and species (Cooper et al., 2011) similar to our results, speciation in Notophthalmus has likely involved a large allopatric component despite evidence of gene flow between nascent species (Gabor & Nice, 2004;Takahashi et al., 2014;although N. perstriatus is sympatric with and the sister species of N. viridescens, too little is known about F I G U R E 5 Divergence between the sexes correlates with within-subspecies disruptive selection across Notophthalmus. The direction of maximum divergence between the sexes, s max (green dashed arrows) defined as the first eigenvector of the covariance matrix S of canonical coefficients from taxon-specific canonical discriminant analyses on the sexes, correlates strongly with the γ max (black arrows), the direction of maximum disruptive selection measured in N. v. viridescens. The same qualitative conclusions were obtained with the mean vector of canonical coefficients (solid green arrows). In A and B, points are standardized canonical coefficients for each taxon, illustrating S in two dimensions. In C and D, points are taxon sex-specific ( (Takahashi & Parris, 2008;Takahashi, Takahashi, & Parris, 2010;Takahashi et al., 2011). Second, the correspondence between withinspecies estimates of SA disruptive natural selection and divergence in sexual dimorphism across species indicates that sexual dimorphism in head shape in Notophthalmus may in part represent the outcome of within-pond resource competition. Taken together, our work suggests that the evolution of sexual dimorphism could play a key role in ecological speciation in Notophthalmus by resolving sexual conflict arising from SA natural selection and reducing intraspecific resource competition associated with life in aquatic environments. This interpretation is consistent with the idea that the evolution of sexual dimorphism may play an important role in ecological speciation by facilitating adaption and thus population persistence.

Our results and interpretation of morphological evolution in
Notophthalmus are also broadly consistent with recent ideas that suggest sexual antagonism and its resolution may go hand in hand with local adaptation (Connallon, 2015), environmental change (Connallon & Hall, 2016) be generally unlikely to expect a constraining effect of character displacement between the sexes on ecological speciation. This suggestion begs more empirical tests of competition's role in the evolution of sexual dimorphism in other study systems.
Subspecies of N. viridescens represent ecologically distinct groups that have arisen rapidly in the last 10,000 years via niche divergence during range expansion from glacial refugia (Takahashi et al., 2014), yet little is known about where these subspecies lie on the speciation continuum. Although there is some evidence of assortative mating by body size, consistent with our finding of a large size component to among-species and subspecies divergence, prezygotic reproductive isolation appears to be incomplete between some subspecies (Takahashi et al., 2010). Although lack of complete speciation in this clade does not effect the interpretation of our data, as subspecies do represent ecologically-and phenotypically diverged entities (Takahashi et al., 2014), future work examining the extent of pre-and postzygotic reproductive isolation in N. viridescens, experimental assessment of divergent selection and variation in sexual dimorphism across subspecies ranges would be informative.
We are (partially) aquatic predators that share the same feeding apparatus and inhabit standing water bodies (Petranka, 1998) that would be expected to share similar distributions of aquatic prey. Second, theory suggests that under conditions of a relatively constant resource distribution, frequency-dependent competition creates stable fitness minima that maintains pervasive disruptive selection (Abrams, Matsuda, & Haranda, 1993;Rueffler et al., 2006).
Our work also adds to a small but growing number of empirical studies that have related properties of the adaptive landscape, as inferred from within-population selective surfaces, to macroevolution-  Schluter, 2000;Wagner, Harmon, & Seehausen, 2012), an alternative (but not mutually exclusive) cause of phenotypic divergence between the sexes, which is viewed as an important component of adaptive radiation by creating and maintaining reproductive barriers between nascent species. In contrast, ecological sexual dimorphism is typically viewed as a constraint on adaptive radiation. Yet empirical examples of a constraining effect of sexual dimorphism on adaptive radiation are limited and lack explicit evidence for character displacement between the sexes (Butler et al., 2007;Schoener, 1977). We have shown that sexual dimorphism driven by ecological character displacement, and ecological speciation, can occur together during adaptive radiation in different combinations of traits. Thus, speciation and the evolution of ecological sexual dimorphism need not be strange bedfellows.

ACKNOWLEDGMENTS
We thank the herpetology divisions at the Carnegie, Smithsonian, and American Museums of Natural history for access to their collections, and David Punzalan for discussion of analyses and results. Funding was provided by grants from NSERC and the Canada Research Chairs program to L.R.