Arboreality constrains morphological evolution but not species diversification in vipers

Abstract

An increase in ecological opportunities, either through changes in the environment or acquisition of new traits, is frequently associated with an increase in species and morphological diversification. However, it is possible that certain ecological settings might prevent lineages from diversifying. Arboreality evolved multiple times in vipers, making them ideal organisms for exploring how potentially new ecological opportunities affect their morphology and speciation regimes. Arboreal snakes are frequently suggested to have a very specialized morphology, and being too large, too small, too heavy, or having short tails might be challenging for them. Using trait-evolution models, we show that arboreal vipers are evolving towards intermediate body sizes, with longer tails and more slender bodies than terrestrial vipers. Arboreality strongly constrains body size and circumference evolution in vipers, while terrestrial lineages are evolving towards a broader range of morphological variants. Trait-dependent diversification models, however, suggest similar speciation rates between microhabitats. Thus, we show that arboreality might constrain morphological evolution but not necessarily affect the rates at which lineages generate new species.

1. Introduction

The exceptional diversity found in many lineages has long been associated with factors that promote an increase in ecological opportunities [1–4]. In its simplest form, ecological opportunities can be defined as ‘a wealth of evolutionary accessible resources little used by competing taxa’ [3]. The invasion of a new habitat, the emergence of a key innovation, and the extinction of competitors could all potentially lead to ecological opportunities (see [4]). Therefore, the exposure to ecological opportunities can trigger divergent selection and lead populations to adapt into several new niches, which ultimately promotes an increase in speciation and morphological diversification [5,6].

In fact, ecological opportunities are frequently suggested as the cause of adaptive radiations [4,7], which are usually characterized by a higher rate of speciation and morphological evolution. Classic adaptive radiations such as Darwin's finches [8] and Caribbean Anolis lizards [9] provide examples of such scenarios, where the invasion of previously unoccupied regions rapidly generated new species and morphotypes. Although adaptive radiations constitute remarkable examples of how shifts in ecological conditions can lead to a great increase in both the species and morphological diversification of a lineage, adaptive radiations are often considered exceptions in nature ([10] but see [7]).

Several processes that underlie the diversification patterns of different groups of organisms might not generate an impressive number of species or morphological variation (e.g. [11–13]). For example, the evolution of phenotypic innovations can change the performance of an organism in their environment but not necessarily promote an increase in diversification [13,14]. Certain ecological scenarios might impose strong selective pressures, and evolutionary forces that reduce trait variance, such as stabilizing selection (see [15]), might allow only lineages with a specific combination of phenotypes to persist favouring evolutionary stasis and niche conservatism [12,16].

Snakes are known to have radiated into a diverse array of habitats; therefore, they have faced different ecological conditions during their evolutionary history [17]. Indeed, recent findings suggest that microhabitat evolution underlies most of the morphological diversity in snakes [18,19]. Although diversification rates vary considerably among snake lineages [20], we do not know how this variation relates to distinct ecological scenarios and the associated ecological opportunities. Of particular interest is arboreality, which emerged multiple times during the evolutionary history of snakes [21]. Arboreal snakes are known to possess a suite of behavioural, morphological and physiological adaptations that allow elongated and limbless organisms to address the challenges of living in a structurally complex environment [21–23]. For example, heavier snakes would not be able to properly move among trees due to the fragility of tree branches, and snakes with shorter tails would lack the ability to hold branches, which would cause suboptimal locomotion in arboreal microhabitats [21–23]. Additionally, having larger body sizes in arboreal microhabitats would increase the gravitational disturbance in blood circulation, especially in elongated organisms such as snakes [21,23]. In contrast, being too small may also prevent snakes from spanning gaps between branches, which would also cause inefficient locomotion among trees [21]. In contrast to arboreal snakes, terrestrial snakes are known to have much more variable morphology ([17] but see [19]).

Vipers are ideal models for exploring how shifts in ecological conditions might affect both morphological and species diversification. Among 329 viper species, arboreality occurs in eight genera that are distributed on four different continents [24] (electronic supplementary material, figure S1 and table S1), which suggests arboreality evolved independently multiple times. Different than other snake groups, non-arboreal vipers do not comprise fossorial or aquatic species and are, for the most part, strictly terrestrial [25,26]. Moreover, some physiological adaptations to arboreality that could potentially overcome gravitational impacts on blood circulation (e.g. anterior heart position) apparently did not evolve among arboreal vipers (see [27]). Hence, this snake lineage is likely to be particularly constrained in morphological evolution, especially in terms of the morphological traits that are related to locomotion within trees and traits that could affect their blood circulation [28]. As a result, the evolution of arboreality in vipers might result in strong constraints on their morphology, which could potentially affect species diversification dynamics.

Herein, we investigate if the evolution of arboreality in vipers was coupled with changes in morphological and species diversification regimes. We first explored the tempo and mode of the evolution of traits that are potentially associated with different microhabitats. Given the potential limitations that an arboreal lifestyle might impose on snake morphological evolution, we expected arboreal vipers to have a more constrained morphology compared to terrestrial ones. We then investigated how the evolution of arboreal habits might affect species diversification. We expected that lineages that switched from microhabitats with fewer constraints (terrestrial) to another microhabitat characterized by strong selective pressures (arboreal) would result in lower speciation rates, assuming there is a potential relationship between morphological evolution and speciation opportunities.

2. Material and methods

(a). Phylogenies

We used time-calibrated phylogenies of extant vipers that were randomly obtained from the posterior distribution of BEAST analyses performed elsewhere [29]. Phylogenetic reconstructions comprised six mitochondrial and five nuclear genes for 263 viper species. This represents the most complete dataset for vipers to date and includes 79% of all described species and all but one monotypic genus of vipers. Phylogenetic relationships and time of divergence were estimated using six fossils, an uncorrelated lognormal clock and a birth-death speciation model [29].

(b). Microhabitat and ancestral state reconstructions

We characterized the microhabitat of each species in the phylogeny using published literature and information gathered from viper experts (see electronic supplementary material, table S1). We categorized vipers as (i) arboreal, (ii) terrestrial and restricted to open habitats, (iii) terrestrial and restricted to forests, and (iv) terrestrial and occurring in both open and forested habitats. This type of subdivision is conservative from the perspective of our hypotheses, given that using a single terrestrial category would more easily suggest a more restricted morphology and lower speciation rates for arboreal species. We also ran analyses using a binary categorization (terrestrial versus arboreal vipers), which allowed us to further explore some methodological limitations (see electronic supplementary material).

We used stochastic character mapping (Simmap, [30]) to reconstruct the evolutionary history of microhabitat of vipers assigning categories to the internal branches of phylogenies. For each of the posterior trees, we set the root state to be sampled from the conditional scaled-likelihood distribution at the root and the transition-rate matrix (Q) to be estimated empirically [30]. We simulated 10 maps for each of the 100 posterior trees using a fixed transition-rate matrix that was estimated for each posterior tree. In total, we had 1000 simmaps for each microhabitat categorization scheme (i.e. four and two states). In general, our simmap reconstructions represented plausible scenarios for the evolution of microhabitat in vipers (see [22]). The resulting root states were estimated as terrestrial in 100% and 99% of the simmaps under the four-state (electronic supplementary material, figure S2) and two-state categories, respectively. The number of state transitions averaged 120.09 ± 95.31 and 17.87 ± 3.14 across simmaps (for four- and two-states respectively), with most transitions occurring among terrestrial categories; in contrast, transitions between terrestrial and arboreal microhabitats occurred much less frequently, and transitions from the arboreal microhabitat back to the terrestrial condition were even rarer (electronic supplementary material, figure S3). Overall, transition rates recovered in simmap reconstructions were similar to what was suggested by diversification analyses (see Results).

(c). Morphology

We used body size, mid-body circumference and tail length to explore if morphology evolved differently between arboreal and terrestrial vipers. We used snout-vent length as our measure of body size, which is considered to be a trait related to many physiological and ecological adaptations among different animals [31,32]. In snakes, body size is frequently related to different aspects of life history, including thermoregulation, fecundity and diet [17,21,32]. Mid-body circumference is considered a proxy for weight and slenderness, and together with tail length, mid-body circumference is suggested as a trait related to the evolution of arboreality in snakes [22,33,34]. We highlight that other phenotypic characters could also be constrained in arboreal snakes, but we chose to explore only the morphological aspects frequently suggested as possible adaptations to arboreality in snakes (e.g. [21–23]).

Body size, mid-body circumference and tail length were measured from adult specimens deposited in 12 scientific collections (electronic supplementary material). We also added literature records for 13 species that were not represented in the visited collections (electronic supplementary material, table S1). In total, we were able to obtain body size, tail length and mid-body circumference measurements for 94%, 92% and 86%, respectively, of the species included in the phylogeny (range of 1–26 individuals per species). For a description of how we chose individuals to measure and the potential biases related to our sampling size, see the electronic supplementary material.

We performed phylogenetic regressions of tail length and mid-body circumference on body size [35] using each posterior tree to provide phylogenetic information. We used the phylogenetic residuals to study the evolution of relative mid-body circumference (RCIRC) and relative tail length (RTL) in arboreal and terrestrial vipers by performing trait evolution analyses using each residuals + simmap combination.

(d). Trait evolution

We used Brownian motion (BM) and Ornstein–Uhlenbeck (OU) models of character evolution implemented in a model-fitting framework [36,37] to explore if the evolution of arboreality was associated with changes in the morphological diversification of vipers. The OU process is ideal for modelling changes in selection regimes, such as differences in the phenotypic optima (θ), phenotypic rate (σ²) or strength of selection (α) between microhabitats [38]. The θ describes hypothetical phenotypes towards which populations are evolving when under an OU process. The σ² measures the intensity of stochastic fluctuations in the evolutionary process, and α indicates the attraction towards θ and measures how fast a trait evolves to the optimum [15,36,39]. We expected that σ² would be smaller and/or α would be larger in arboreal microhabitats compared to terrestrial microhabitats, suggesting that arboreal vipers might be evolving under a distinct and possibly more constrained selection regime (e.g. [12]).

We separately fitted seven models to each morphological dataset (body size, RCIRC and RTL) and simmap trees (estimated under four-state and two-state microhabitat categories) using OUwie (v 1.45) [36]: (i) BM1, single-rate BM model, (ii) OU1, single optimum model, (iii) BMS, multi-rate BM model that allows σ² to differ among microhabitats, (iv) OUM, assumes distinct θ for each microhabitat while having a single σ² and α for all microhabitats, (v) OUMV, allows different σ² and θ for each microhabitat, (vi) OUMA, allows different α and θ for each microhabitat, and (vii) OUMVA, allows different σ², α, and θ for each microhabitat.

After performing OUwie analyses, we checked if the eigenvalues of the Hessian matrix were positive [36,37] and if the estimated θ values were biologically feasible (see electronic supplementary material, figures S4–S6). We excluded analyses that returned very unrealistic θ (e.g. those that were two orders of magnitude higher than the largest viper known in the dataset) or negative eigenvalues (see electronic supplementary material, tables S2–S7). We calculated the ΔAICc (the difference in AICc values of each model relative to the model with the lowest AICc value, where AICc is the Akaike information criterion corrected for small sample sizes) to investigate which model best explained trait evolution in each simmap tree. A model was considered better when the ΔAICc difference between the second and the best model was greater than two.

We performed additional analyses to test for model adequacy. We randomly selected simmaps used in OUwie analyses and simulated continuous traits using the corresponding empirical parameter values (θ, σ² and α) that were estimated by OUwie under the best model chosen for each simmap. We compared the resulting trait distribution to the empirical distribution to explore if simulated trait values were similar to empirical ones. We also explored if OUwie was able to recover the same trait evolution patterns as those suggested by our empirical data for arboreal lineages when using simulated trait values (details in the electronic supplementary material).

(e). Species diversification

To investigate if microhabitat evolution had an impact on species diversification, we used the Multiple State Speciation Extinction (MuSSE) model [40] fitted to 200 randomly chosen phylogenies. MuSSE models were implemented in a Bayesian Markov chain Monte Carlo (MCMC) framework that allowed phylogenetic and rate value uncertainties to be naturally incorporated in the analysis. We used the diversitree package [40] and a script that implemented the MCMC approach (https://github.com/dsilvestro/mcmc-diversitree), allowing all parameters to be independently estimated. We ran MuSSE for 5 000 000 steps to achieve convergence.

We combined posterior distributions of parameters into one distribution for each parameter (i.e. speciation and extinction in each microhabitat and transition rates among microhabitats). We calculated speciation minus extinction for each iteration to build the posterior distribution of the net diversification rates in each microhabitat. We calculated the differences between the estimated speciation rates for the arboreal and terrestrial categories in each MCMC step. By doing this, we built posterior distributions of differences and considered speciation rates of arboreal lineages to be smaller and distinct from the others when values were negative and 0 was not included within the 95% highest posterior density (HPD) of each distribution (we did the same for net diversification, extinction and transition rates).

State-dependent speciation extinction models have been criticized [41–43]. We performed posterior predictive simulations to evaluate how well the model could predict empirical data (electronic supplementary material) [44,45]. To properly address null model problems [43], we also investigated potential differences in speciation regimes using the recently developed Hidden State Speciation and Extinction models (HiSSE, [43]) (electronic supplementary material). Given that extinction rates can be poorly predicted using molecular phylogenies [46,47], caution should be taken when interpreting diversification rates. However, we expect diversification methods to be able to identify qualitative differences in the speciation rates.

3. Results

(a). Evolution of body size

Under the four-microhabitat categorization, we successfully fitted all trait evolution models in 187 of the 1000 analysed simmaps. Among these 187 simmaps, 173 (92%) suggested the most complex model, OUMVA, was the best model. Although most simmaps were not able to fit all models (returning negative eigenvalues or unrealistic optima), from all 1000 analysed simmaps, 89% suggested OU models with multiple θ and different α and/or σ² as the best models (electronic supplementary material, table S2). Looking at the fittings of each model separately (each line in electronic supplementary material, table S2), OUMVA was the best model in 93% of the simmaps to which it was fitted (311 of 335). Of the simmaps in which OUMV was fitted (992), 44% suggested that OUMV was the best model, but 31% suggested OUMVA. It is important to note that from the 992 simmaps that fitted OUMV, only 335 also fitted OUMVA, which might be an underestimation of the real proportion of simmaps that indicate OUMVA as the best model (electronic supplementary material, table S2). Among simmaps in which OUMA was fitted (469), 31% suggested it as the best model, whereas 37% suggested OUMVA. Similar to the OUMV fittings, from the 469 simmaps that fitted OUMA, only 187 also fitted OUMVA, which might be an underestimation of the real proportion of simmaps that indicate OUMVA as the best model (electronic supplementary material, table S2). Although not all models could be fitted, the results clearly suggest different θ associated with different variances, and this is achieved either by differences in α, σ² or both.

Parameter estimates show that a very similar scenario is depicted independently of the best model chosen (OUMVA, OUMV or OUMA) (figure 1; electronic supplementary material, figure S7). Body size θ are either larger or smaller in terrestrial vipers than in arboreal vipers (figure 1a,d,g). Arboreal microhabitats show higher α and smaller σ² than the terrestrial categories, suggesting that after a lineage becomes arboreal, body size evolves faster towards θ and with a smaller rate of stochastic fluctuation (figure 1b,c,e,f,h; electronic supplementary material, figure S7). It is important to note that the σ² estimated under OUMV comprise values with two different magnitudes (figure 1e,f), but each set of points represent estimates from a distinct sample of simmaps that comprise the same set among microhabitats. Thus, although this ‘bi-modality’ represents uncertainty among simmaps, both estimates suggest smaller σ² for arboreal microhabitats.

Figure 1.

Parameter estimates when OUMVA (a,b,c), OUMV (d,e,f), and OUMA (g,h) were chosen as the best models in the analyses of body size evolution (311, 436, 146 simmaps among the 335, 992, 469 simmaps each model was respectively able to be fitted to). Dots represent analyses performed in distinct simmaps. For better visualization, we excluded extreme values. Numbers on the top represent the proportion of values included. Because σ² estimates under the OUMV model span two orders of magnitude (panel e), we presented smaller values in panel f. Even at the smaller scale, the σ² estimates still vary, and extreme values are not shown. Proportions in panel f represent the proportions of values included relative to panel e, which in turn includes all σ² estimates under the OUMV model. Uncertainty in the parameter estimates (OUMVA model) is shown in electronic supplementary material, figure S10. (Online version in colour.)

Results were similar when using the two-microhabitat categorization; however, the successful fittings substantially increased (electronic supplementary material, table S3 and figure S8). In the case of the OUMVA fits, arboreal θ can be either higher or lower than terrestrial θ. Because arboreal snakes had intermediate θ in the four-category scheme, this pattern of optimum variation is not surprising and is somewhat affected by the binary categorization. More importantly, σ² and α estimates suggested a similar pattern to the four-microhabitat categorization, with arboreal lineages having higher α and lower σ² (electronic supplementary material, figure S8). Simulated trait values were similar to empirical ones and OUwie still recovered arboreal microhabitats with higher α and lower σ² when analysing simulated traits (electronic supplementary material, figures S9 and S10).

(b). Evolution of mid-body circumference

Using the four-microhabitat categorization, we successfully fitted all models in 359 of the 1000 analysed simmaps. From these, 122 (33%) suggested OUMVA was the best model, 41 (11%) suggested OUMA, 18 (5%) suggested OUM, one suggested OUMV, and 177 did not suggest any best model. Different from body size analyses, ties between models (i.e. ΔAICc less than two) frequently occurred. However, when considering all 1000 simmaps, models with varying α (OUMA and OUMVA) were among the best models in at least 75% of the simmaps (including OUMA ties) (electronic supplementary material, table S4). The model with varying θ but without differences in α/σ² (OUM) was among the best models in 31% of the simmaps (OUM ties included). Parameter estimates suggest that arboreal lineages are evolving toward smaller RCIRC θ and higher α (figure 2a,b; electronic supplementary material, figure S11), suggesting that after a lineage becomes arboreal, their RCIRC evolves faster towards the optima compared to terrestrial microhabitats. The results were similar when using two-microhabitat categorization, but successful fittings and uncertainty in choosing a best model both increased (electronic supplementary material, table S5 and figure S12). Simulated trait values were similar to empirical values, and OUwie still recovered arboreal microhabitats with lower θ and higher α when analysing simulated traits (electronic supplementary material, figures S13 and S14).

Figure 2.

Parameter estimates when OUMVA was chosen as the best model in the analyses of relative mid-body circumference (a,b,c) and relative tail length evolution (d,e,f) (190 and 383 simmaps among the 467 and 652 simmaps OUMVA was able to be fitted to, respectively). Dots represent analyses performed in distinct simmaps. For better visualization, we excluded extreme values. Numbers on the top represent the proportion of values included. Uncertainty in the parameter estimates (OUMVA model) is shown in electronic supplementary material, figure S14. (Online version in colour.)

(c). Evolution of tail length

Using the four-microhabitat categorization, we successfully fitted all models in 307 of the 1000 analysed simmaps. From these 307, 136 (44%) suggested OUMVA was the best model, 54 (17%) suggested OUMA, 13 (4%) suggested OUM, one suggested OUMV was the best model, and 103 simmaps (34%) were left without a best model. Similar to RCIRC analyses, ties between models (i.e. ΔAICc less than two) frequently occurred. When considering all 1000 simmaps, models with varying α (OUMA and OUMVA) were among the best models in at least 66% of the simmaps (OUMVA ties included) (electronic supplementary material, table S6). OUM was among the best models in 41% of the simmaps (OUM ties included). Although parameter estimates suggest that arboreal lineages are evolving towards longer RTL, α and σ² estimates seem to differ only within terrestrial categories (figure 2d–f; electronic supplementary material, figure S15). The results were similar when the two-microhabitat scheme was used, at least in terms of the parameter estimates under the OUMA and OUM models, and the number of successful fittings increased (electronic supplementary material, table S7 and figure S16). Parameter estimates under OUMVA, however, suggest lower α and higher σ² for RTL among arboreal lineages. Simulated trait values were similar to empirical values and OUwie still recovered arboreal microhabitats with higher θ but no substantial differences in α and σ² (electronic supplementary material, figures S13 and S14).

(d). Species diversification

Speciation, extinction and net diversification rates did not differ among microhabitats (figure 3a,b; electronic supplementary material, figure S17). Transition rates were similar to those estimated in simmap reconstructions (figure 3c and electronic supplementary material, figure S3) and predominantly occurred between terrestrial microhabitats, while transitions to arboreal microhabitats were mostly limited to terrestrial lineages restricted to forests (figure 3c; electronic supplementary material, figure S18). Transition estimates suggest that once a lineage becomes arboreal, it will rarely become terrestrial again. Our posterior predictive simulations indicate that MuSSE estimates represent a plausible evolutionary scenario (electronic supplementary material, figure S19). Additionally, HiSSE analyses support our findings and indicate that there was no difference in speciation or extinction rates between arboreal and terrestrial microhabitats (electronic supplementary material, table S8).

Figure 3.

(a) Posterior distributions of speciation rates for arboreal and terrestrial lineages. (b) Posterior distributions of differences between speciation rates in arboreal and terrestrial lineages. Bars represent the 95% HPD interval, and dots represent the medians. (c) Medians of transition rates between lineages from distinct microhabitat categories, as estimated by MuSSE. Transition rates less than or equal to 0.01 are not shown. (Online version in colour.)

4. Discussion

Being elongated and limbless, snakes face several challenges in arboreal microhabitats, such as moving through a discontinuous physical environment and experiencing gravitational effects on blood circulation [21,23,33]. As a consequence, arboreal microhabitats demand a much more specialized morphology than do terrestrial environments [17,21], and might prevent morphological diversification. Herein, we found evidence for distinct morphological evolutionary regimes among microhabitats in vipers, with arboreal microhabitats imposing stronger limits on body size and RCIRC evolution.

Ecological conditions that constrain morphology have been suggested before. The evolution of piscivory in fishes [12] and arboreality in birds [16] and some lizards [48], seem to limit morphological evolution. Additionally, different microhabitats are associated with distinct optimal head shapes in pythons and boas [19], and morphological convergence across distantly related snake lineages probably reflect their microhabitat use more than their dietary habits [18]. Our results corroborate the idea that arboreality might be especially challenging and impose strong selective pressures in certain lineages. The idea of strong selection in arboreal microhabitats is especially intuitive when interpreting our results in terms of phylogenetic half-lives (or ‘t_1/2’, ln(2)/α), which measure the time an ancestral phenotype takes to evolve halfway towards the optimum after a lineage enters a new selective regime (e.g. a microhabitat) [38,49]. Arboreal microhabitats comprise much shorter t_1/2 (8.91 millions of years, my) for body size than do terrestrial microhabitats (e.g. 49.42 my). More importantly, t_1/2 for body size in arboreal vipers is shorter than the stem age of most arboreal lineages (approx. 25–35 my, [29]), suggesting that the time frame for morphological selection in arboreal microhabitats is within the time frame depicted for the evolution of arboreality.

The relationship between the arboreal lifestyle and the evolution of snake body size has been explored before using correlational methods [22,23,33]. Using phylogenetic contrasts, Martins et al. [22] and Pizzatto et al. [33] did not find any significant effect of arboreality on body size in pitvipers and boids respectively. Sheehy et al. [23] studied several snake lineages and found arboreality to be associated with a decrease in body size (relative to tail length). Here, we found that arboreal vipers are evolving towards an intermediate body size under stronger selection than that experienced by terrestrial vipers.

Although it is intuitive to think of arboreal snakes as having intermediate body sizes, the optimal body size suggested here probably represents a specific case for arboreal vipers rather than the rule for arboreal snakes. In fact, boids comprise very large arboreal species (e.g. Corallus caninus) [34], and some arboreal elapids grow to more than 2 m in snout-vent length [50]. Contrary to several arboreal lineages, arboreal vipers appear to lack a more anterior-positioned heart, which might impose limits on their maximum body size [27,28]. Moreover, vipers are primarily sit-and-wait predators that move less frequently than active-foragers [17]. Thus, arboreal vipers would not have to cross gaps as frequently as other arboreal active-forager snakes and may not face selective pressures that would cause a larger body size (and/or a more anteriorly positioned heart [27]) to evolve. Hence, an interaction between environmental selective pressures (i.e. arboreality) and phylogenetic conservatism (i.e. hunting strategy and a more posterior heart position) might have led to morphological ‘solutions’ to an arboreal lifestyle that differed from other snake lineages.

Mid-body circumference is a proxy for mass and slenderness in snakes (e.g. [51]). For a given body size, arboreal snakes tend to become slenderer, which decreases their overall weight. This morphological modification allows arboreal snakes to move across fragile branches and confers a branch-like appearance that provides better camouflage in the arboreal environment [21]. Even among the stout boids, arboreal lineages tend to have substantial decreases in their relative circumference [33]. However, some life-history traits are strongly associated with stouter bodies (i.e. a diet based on bulky prey such as small mammals, as opposed to slender and small lizards) [51,52]. This suggests that the degree of decrease in circumference, and consequently, the distance to an arboreal optimum, should differ in arboreal snakes varying in these life-history aspects.

Vipers are considered stout-bodied, which is often related to a diet that includes bulky prey, a diet commonly described for viper species [51,53]. In contrast, several non-viperid arboreal snakes such as Oxybelis aeneus and Uromacer oxyrhynchus, which frequently eat lizards, are comparatively slenderer [54]. The conflicting selective pressures that shape a diet based on large prey and those that shape an arboreal lifestyle could result in arboreal vipers having a smaller optimal circumference compared to other vipers but in a higher optimal circumference compared to other arboreal snakes (e.g. those relying on other prey types or sizes). This constraint might also help to explain the intermediary body size optima found for arboreal vipers and the strong selective pressure inferred for both body size and RCIRC on arboreal vipers. To increase their body sizes, arboreal vipers would have to substantially decrease their circumference to avoid being too heavy to properly move across branches. Diet adaptations, however, would prevent any further decreases in circumference. However, similar to the effect on body size, a sit-and-wait foraging strategy might also underlie differences in RCIRC among arboreal vipers compared to other arboreal snakes. Because arboreal vipers do not frequently move across branches, a substantial decrease in their circumference would not be required to overcome the selective pressures imposed by the arboreal microhabitat.

The higher optima for RTL in arboreal lineages agree with what was previously suggested for vipers [22] and arboreal snakes in general [23,33]. A longer tail helps an arboreal snake properly move across tree branches and generates the force required to bridge gaps [21,23]. Indeed, tail tissues in arboreal snakes have a specific configuration that decreases blood pooling when vertical positions are assumed [23] and allows for tail elongation. In contrast, an increase in tail length might not confer any functional advantages when a snake is moving on the ground ([55] but see [56]).

Because morphology usually reflects the ecology of organisms [57], and because selection in different niches is known to underlie species diversification in several groups [8,9], speciation is frequently coupled with morphological changes (see [6]). Thus, an ecological regime that imposes strong constraints on morphological evolution might limit speciation opportunities in a given lineage. Although trait evolution analyses suggest that the evolution of body size and RCIRC were more constrained in arboreal lineages, this limited adaptive landscape does not translate into different speciation regimes between microhabitats.

One possible interpretation is that mechanisms underlying species diversification in vipers are not coupled with morphological changes, and both terrestrial and arboreal vipers speciate in similar ways. Under this scenario, selective pressures imposed on morphology, even if constraining it, may not affect the probability of speciation among microhabitats, and this would result in similar speciation rates. A morphological-speciation decoupling has also been shown for New World pitvipers [58], elapids [59], several fossil lineages [60,61], and might be widespread across extant lineages other than snakes ([62] but see [63]). It is also possible that speciation is more related to traits other than the morphological aspects associated with microhabitat use that were measured in our study. For example, speciation could be more related to traits associated with sexual selection, and, in fact, differential recognition with a subsequent preference for different chemical signals is suggested to play a significant role during the speciation process in organisms, such as snakes [64,65], that more heavily rely on chemical senses than vision [66,67]. Lastly, morphological differentiation could still occur during speciation in terrestrial vipers, but arboreal snakes would be more prone to another speciation mechanism, such as geographical isolation (e.g. vicariance).

Overall, our results add to the existing evidence that certain ecological conditions can strongly constrain morphological evolution (e.g. [12]). The arboreal microhabitat represents one such condition that imposes strong selective pressures and results in constrained morphology in arboreal lineages across different taxa (e.g. [16,48], present study). On the other hand, the constrained evolutionary trajectory in morphology of arboreal lineages did not affect speciation rates. Macroecological spatially oriented studies, and phylogeographic and microevolutionary approaches are all promising to help disentangle the processes underlying the diversity found in distinct ecological regimes.

Data accessibility

Datasets are part of the electronic supplementary material.

Authors' contributions

L.R.V.A., T.B.Q. and M.M. planned the study. L.R.V.A. and M.M. collected the data. L.R.V.A. and G.B. performed the analyses. L.R.V.A. and T.B.Q. wrote the first draft of the manuscript. All authors substantially contributed to revision.

Competing interests

We have no competing interests.

Funding

L.R.V.A., M.M., G.B. and T.B.Q. were funded by São Paulo Research Foundation (FAPESP) (grants nos 2012/02038-2, 2011/50206-9, 2014/03621-9, #2012/04072-3). L.R.V.A. received a grant from The Field Museum of Natural History.