Patterns of morphological diversification of mainland Anolis lizards from northwestern South America

Transcription

1 Zoological Journal of the Linnean Society, 2016, 176, With 11 figures. Patterns of morphological diversification of mainland Anolis lizards from northwestern South America RAFAEL A. MORENO-ARIAS* and MARTHA L. CALDERÓN-ESPINOSA Grupo de Biodiversidad y Sistemática Molecular, Instituto de Ciencias Naturales, Universidad Nacional de Colombia, Ciudad Universitaria, Bogotá D.C , Colombia Received 11 February 2015; revised 30 June 2015; accepted for publication 9 July 2015 Anolis lizards are one of the most diverse vertebrate genera and are the classic example of adaptive radiation and convergent evolution. Anoles exhibit great morphological diversity produced by the ecological opportunity to exploit several arboreal niches. Anole radiation in the Caribbean islands is well studied, but the mainland radiation is less understood. We used a large morphological data set and a molecular phylogeny to describe the morphological diversification of anoles from northwestern South America, a region with the highest anole diversity on a mainland. We describe morphological diversity as summarized by ten morphotypes, defined mainly by body size, limb proportions, and subdigital lamellae. We show that some morphotypes are limited to forested lowlands and others to Andean highlands; by contrast, Anolis assemblages from tropical rainforests are comprised of the same four morphotypes. We demonstrate that morphological diversification followed a pattern of adaptive radiation across a landscape of adaptive peaks. Our results are consistent with the most recent hypothesis of convergence stated for Caribbean radiation, and demonstrate convergence between mainland morphotypes and Caribbean ecomorphs, which suggests that common processes are driving both radiations The Linnean Society of London, Zoological Journal of the Linnean Society, 2016 doi: /zoj ADDITIONAL KEYWORDS: adaptive peaks adaptive radiation Colombia convergent evolution morphology. INTRODUCTION Lizards of the genus Anolis comprise one of the most diverse clades of vertebrates, with 390 described species (Uetz & Hošek, 2014). The geographical range of anoles includes the northern half of South America through Central America and into tropical Mexico and the southeastern USA, including every Caribbean island and some islands of the Pacific Ocean; anoles have also been introduced to southeastern Pacific islands (Losos, 2009). The Anolis clade is more phenotypically diverse than related clades and represents a classic example of adaptive radiation (Losos & Miles, 2002). Anolis radiated extensively in the Caribbean islands as well as in mainland Central and South America. The Caribbean island radiations have been widely studied and the adaptive basis of this diversification is well understood (Losos, *Corresponding author. rafamorearias@gmail.com ). In Cuba, Hispaniola, Puerto Rico, and Jamaica, anole lizards diversified into a set of species adapted morphologically to particular microhabitats, with these microhabitat specialists known as ecomorphs (Williams, 1972). Six ecomorphs are recognized and are named according their microhabitat use: crown-giant, grassbush, trunk, trunk-crown, trunk-ground, and twig (Rand & Williams, 1969; Williams, 1983). The Caribbean anole radiation is also significant because it shows independent evolution of these ecomorphs each ecomorph has evolved at least twice independently on each of the Greater Antilles, indicating convergent adaptation amongst islands (Williams, 1983; Losos et al., 1998; Mahler et al., 2013). In addition, the mechanistic basis of the relationship between morphology and ecology is well supported by ecomorphological and functional studies (Losos, 1990a, b, c; Losos & Irschick, 1996). Adaptive radiation in Anolis also occurred on the mainland, with more species distributed in Central and South

2 DIVERSIFICATION OF SOUTH 633 America than in the Caribbean. However, mainland anoles exhibit different patterns of radiation compared with those seen in the Caribbean, possibly because of differences in food resources, competitors, predators, and selective pressures (Irschick et al., 1997; Losos, 2009; Schaad & Poe, 2010). In general, mainland anoles occupy a different morphospace than Caribbean anoles, although most Caribbean ecomorphs (e.g. crowngiant, grass-bush, trunk-crow, trunk-ground, twig) are represented by a few mainland species (Irschick et al., 1997; Velasco & Herrel, 2007; Pinto et al., 2008; Schaad & Poe, 2010); however, in these cases, mainland species differ in their ecology from Caribbean ones (Irschick et al., 1997; Schaad & Poe, 2010). Additionally, some clades of mainland anoles have greater diversification rates than those in the Caribbean (Pinto et al., 2008). Therefore, the differences between mainland and Caribbean anoles suggest the possibility that other ecomorphs could occur on the mainland (Schaad & Poe, 2010). Colombia has the greatest diversity of Anolis, with 76 species (71 mainland restricted) of the 120 registered to northwestern South America Brazil, Colombia, Ecuador, Perú, and Venezuela (Uetz & Hošek, 2014). Therefore, we used anoles distributed in mainland Colombia as a model with which to analyse the morphological diversity of the mainland radiation in northwestern South America. We combined a large morphological data set of Colombian Anolis, multivariate analyses, and phylogenetic comparative methods with the following aims: (1) to characterize the morphological diversity, (2) to identify morphotypes and describe their geographical distribution and natural history features, (3) to test patterns of morphological evolution of Anolis from northwestern South America, and (4) to compare morphological similarity between mainland and Caribbean species. MATERIAL AND METHODS MORPHOLOGICAL DATA To describe the morphological diversity we measured 778 male lizards of 59 species of Colombian mainland anoles (N = 1 77 specimens/species), representing 83.1% of Colombian mainland species and 52.2% of northwestern South American mainland species (Supporting Information Table S1). These consisted of 717 specimens from the Reptile Collection of Instituto de Ciencias Naturales (ICN-R), Universidad Nacional de Colombia, and 61 specimens measured in the field and subsequently released. We only measured adult male lizards because they exhibit the greatest morphological variation in anoles (Losos, 2009). We sexed individuals by the presence of a hemipenis, dewlap, or enlarged postanal scales, depending upon which characters occur in that species. For each individual we recorded the snout vent length (SVL), tail length only for specimens with no regenerated or broken tails (TLL), foreleg length (FLL), hindleg length (HLL), trunk length (LTR), head width (HW), head height (HH), and head length (HL) in mm. We used digital callipers with an accuracy of ±0.01 mm. We also counted the number of subdigital lamellae under the third and fourth phalanges of the fourth toe (LN). MORPHOLOGICAL DATA ANALYSES We calculated the mean value of each morphological trait per species. We log-transformed all data and corrected for the body size effect through regression analyses of SVL against each trait. The variables employed in subsequent statistical analyses corresponded to SVL and the residuals of each trait against SVL. To identify species with similar morphology, we performed a polythetic hierarchical divisive clustering method in an ordination space (Legendre & Legendre, 1998). This method is based on a principal components analysis, followed by a division of objects (species) into groups (morphotypes) according to whether their value for the first principal component was positive or negative. The same procedure was applied subsequently to each new group (Williams, 1976) until we obtained groups with at least two species with the same value for the first component. We evaluated the groups identified by previous analysis using a discriminant function analysis (DFA). We used the first significant functions to assess morphological differences amongst species and to classify them into morphotypes. Morphotypes were defined as the species group that showed the same sign in their punctuations to each significant function. Finally, we performed a second DFA to corroborate the morphotypes from the first DFA. The structure matrix from a DFA shows the most relevant variables for each function each function, so we used the matrix obtained in the last DFA to describe the morphotypes. We performed analyses of covariance with morphotypes as treatments and SVL as the covariate to find differences between morphotypes for each trait individually. It is important to mention that some species were poorly represented in our sample and so could be classified in other groups when more data are acquired. GEOGRAPHICAL DISTRIBUTION AND NATURAL HISTORY OF MORPHOTYPES To describe the distribution of morphotypes, we quantified the occurrence of each morphotype in ecogeographical regions (Fig. 1). These regions were defined by two criteria (1) geographical trans-andean regions (below 1000 m a.s.l. for western and eastern slopes of Central and Western Andes Cordillera and western slope of Eastern Andes Cordillera), cis-

3 634 R. A. MORENO-ARIAS AND M. L. CALDERO N-ESPINOSA Figure 1. Distribution of species richness and morphotype (MT) diversity across 13 ecogeographical regions, altitude range, and main regions based on life zones and predominant vegetation types for 59 Anolis species of Colombia. Bar size for each colour indicates number of species. ARF, Cis-Andean tropical rain forests in the Amazonas; CA, central Andes; CP, Caribbean Plains; EA, eastern Andes; LCV, Lower Cauca River Valley; LS, Cis-Andean savannahs in Los Llanos; MBC, Middle Biogeographical Chocó; MMV, Middle Magdalena River Valley; NBC, Northern Biogeographical Chocó; SBC, Southern Biogeographical Chocó; SNSM, Sierra Nevada de Santa Marta; UMV, Upper Magdalena River Valley; WA, western Andes. Andean regions (below 1000 m a.s.l. for eastern slopes of Eastern Andes Cordillera to Orinoco Llanos and Amazonas), and Andean regions (above 1000 m a.s.l. in the Andes system and Sierra Nevada de Santa Marta); (2) predominant life zone and vegetation types of Colombia (Rangel, 1997, 2004). This second criterion divides the country into tropical rain forest, tropical dry forest, cloud forests, Andean forest, scrubland, savannah, and Paramo. Integrating both criteria we identified 13 geographical regions for this study: trans-andean tropical rain forests in Northern Biogeographical Chocó (NBC), Middle Biogeographical Chocó (MBC), Southern Biogeographical Chocó (SBC), Lower Cauca River Valley (LCV), Middle Magdalena River Valley (MMV), TransAndean dry forests in Upper Magdalena River Valley (UMV), and Caribbean Plains (CP). We grouped Andean cloud forests (altitude range > 1000 < 2300 m a.s.l.), Andean forests and scrublands (Altitude range > m a.s.l.), and Paramo vegetation of each mountain range into western (WA), central (CA), and eastern Andes (EA), and Sierra Nevada de Santa Marta (SNSM). Cis-Andean savannahs in Los Llanos (LS), and Cis-Andean tropical rain forests in the Amazonas (ARF) were regions in eastern Andes (Fig. 1). The occurrence of species across ecogeographical regions was defined by collection and literature data. We corroborated identifications of all specimens and their collection locality, and complemented locality data with reports of Colombian inventories of reptiles (Lamar, 1987; Bernal-Carlo, 1991; Hernández-Ruz et al., 2001; Castaño-Mora et al., 2004; Carvajal-Cogollo & Urbina-Cardona, 2008; Moreno-Arias, Medina-Rangel & Castaño-Mora, 2008; Moreno-Arias et al., 2009; Medina-Rangel, 2011). In order to describe differences in species composition between ecogeographical regions we calculated Whittaker s diversity beta index (Whittaker, 1972). Microhabitat and vegetation strata

4 DIVERSIFICATION OF SOUTH 635 were obtained from natural history literature for each species (Table S1). PHYLOGENETIC COMPARATIVE ANALYSIS We used the same traits as in the DFA above (SVL and residuals of traits vs. SVL) but in this case residuals were calculated using phylogenetic size correction (Revell, 2009) with the phyl.resid function of the phytools R package (Revell, 2012). We built a phylogeny of 28 species based on DNA sequences from GenBank (Table S2), utilizing sequences fragments from mitochondrial regions of nicotinamide adenine dinucleotide dehydrogenase subunit II (ND2, 1038 bp), five transfer-rna (trnatrp, trnaala, trnaasn, trnacys, trnatyr, 398 bp), and a small fragment of cytochrome c oxidase subunit I (COI, 30 bp). As consequence our inferences are about a gene tree instead of a species tree; however, several studies show that mitochondrial trees not only reconstruct nearly the phylogenetic history of anoles as trees builded with several data types (nuclear sequences and morphology) but also are adequate to make inferences about the evolution of other biological attributes (i.e. Castañeda & de Queiroz, 2011; Nicholson et al. 2012; Gamble et al., 2013). We aligned sequences using default values of gap costs from CLUSTAL X v. 2 (Larkin et al., 2007), and trnas were aligned based on secondary structure information from Kumazawa & Nishida (1993). We used three data sets: a data set without partitions, a data set partitioned by gene (three partitions: ND2, trna, and COI), and a data set partitioned by codon position for ND2 and gene identity for the other two fragments (three partitions: ND2 by codon, trna, and COI). We selected the model of evolution for each partition with Akaike s information criterion (AIC) using JModeltest 2 (Darriba et al., 2012), and performed phylogenetic analysis with MrBayes 3.2 (Ronquist et al., 2012). We verified convergence visually with TRACER v. 1.5 (Rambaut & Drummond, 2007) and evaluated each partition strategy with Bayes factors (BF). This method compares two models (two partition strategies) by their ratios of marginal likelihood (BF), then, if BF values larger than 10 indicate strong support of a model over other. Conditions for these analyses were as follows: four independent runs with four Markov chains per run, generations with a random starting tree, a temperature of 0.15, and trees sampled every 1000 generations. For all phylogenetic comparative analyses we used the strict consensus tree (Fig. 4) which was ultrametric transformed with the function chronos from the ape R package (Paradis, Claude & Strimmer, 2004). To explore if related species are more similar morphologically than expected by chance (phylogenetic signal: Blomberg & Garland, 2002), we calculated and tested Blomberg et al. s K statistic (Blomberg, Garland & Ives, 2003) for each continuous trait, using the function phylosig from the phytools R package (Revell, 2012). K statistic tests phylogenetic signal as the ratio of the mean of squared error from the data measured and the proposed phylogeny vs. the mean of squared error of data randomly permuted across the tips of the tree. Blomberg et al. s K values fluctuate from zero when phylogenetic independence exists in data, to one or more than one when data are distributed as expected under Brownian motion. As the significant values of phylogenetic signal also can be result from different evolutionary processes and rates (Revell, Harmon & Collar, 2008), we explored some evolutionary models that could explain the morphological diversification patterns in Colombian mainland anoles. We calculated the evolutionary scenarios with the surface R package and tested them with ΔAIC c Akaikes s information criterion, c: corrected to small sample size (Burnham & Anderson, 2004); the AIC c values were obtained using the approach suggested by Ingram & Mahler (2013). We tested the following models: 1. Brownian Motion (BM; Felsenstein, 1973), which indicates that the distribution of data follows a pattern related to the shared ancestry of species. In this case phenotypes change through time but the most closely related species are phenotypically more similar to each other than they are to others because of their shared history. 2. Early Burst (EB; Harmon et al., 2010) is a timedependent model that exponentially transforms evolutionary rates under the model R(t) = R(0) * exp (a*t), where R is the evolutionary rate, a is the rate change parameter, and t is time. We transformed the tree with a value of a = 3. This model tests if trait evolution has slowed over time from the root to the tips, by transforming the topology as if the evolutionary rate near the root is faster than near the tips of the tree. This model suggests that major changes in traits occur early in history and the phenotypic similarity between closest species is lower than expected by BM, which is a pattern that corresponds to adaptive radiations. 3. Ornstein Uhlenbeck (OU) Process. Stabilizing selection under evolution model based on the OU Process (Table S2), which exemplifies trait evolution towards an unique adaptive peak, where trait X changes according to two factors through time (t): a deterministic selection force (selection force), and a stochastic (random drift), according to the formula dx(t) = α[θ X(t)]dt σdb(t) where α is the selection force, θ is the adaptive optimum, and db is white noise. White noise indicates independent and normally distributed random variables, and σ is the intensity of the random fluctuations in the evolutionary process (Butler & King, 2004).

5 636 R. A. MORENO-ARIAS AND M. L. CALDERÓN-ESPINOSA 4. Surface Regimes (SURF; Table S2). This is a method to test convergent phenotypic evolution based on OU under the adaptive landscape framework. The method, known as surface (Ingram & Mahler, 2013), adds regimes of phenotypic shifts step by step and calculates their respective likelihoods after comparing all calculated regimes according to their AIC values to choose the best regime shifts of phenotypic shifts (forward phase). Surface searches which regimes can be collapsed, again with maximum likelihood, and step by step calculates convergent regimes a posteriori. Finally, surface finds the best convergent regime of phenotypic shifts by comparing the AIC values of the calculated convergent regimes (backward phase). 5. Morphotype Regimes (MORPH fwd). We assigned a corresponding morphotype to each species and the most probable morphotype to each node. To do this we estimated an ancestral reconstruction of morphotype states with a maximum likelihood approach. To test morphotype regimes we used the ace function from the ape R package (Paradis et al., 2004). We tested the MORPH fwd performance model with the startingmodel function of the surface R package (Ingram & Mahler, 2013). This model tested if morphological evolution followed stabilizing selection under ten adaptive peaks. 6. Morphotype Regimes Collapsed (MORPH bwd). We collapsed all MORPH fwd regimes with the surfacebackward function of the surface R package (Ingram & Mahler, 2013). This model tested if the morphological evolution occurred toward less of the morphotypes defined by MORPHfwd. COMPARISONS BETWEEN MAINLAND MORPHOTYPES AND CARIBBEAN ECOMORPHS To compare anole species morphologically, we used SVL and size-corrected values of TLL, FLL, HLL, and LN. Morphological data for Caribbean anoles were taken from Losos (1990a) and Losos (1992). To detect morphological differences between mainland morphotypes and Caribbean ecomorphs, we used two approaches. First, we conducted a DFA without phylogenetic correction with our 59 species and 27 Caribbean anoles (which represented six Caribbean ecomorphs), using morphotype as a grouping variable to classify Caribbean species into the mainland morphotypes. Then, we analysed the data with a phylogenetic correction. To do this, we calculated a phylogenetic size correction with the phyl.resid function (Revell, 2009) using the topology of Gamble et al. (2013), which we pruned and kept the species for which morphological data were available (54 species). Finally, we carried out a surfacebackward analysis for these data and this topology to find convergent regimes in 54 species classified according to their morphotype or ecomorph. RESULTS MORPHOLOGICAL DIVERSITY We obtained 21 species groups using polythetic hierarchical divisive clustering that were collapsed by the DFA into ten groups that represent ten morphotypes (Fig. 2). The most diverse morphotypes were morphotype 1 (MT1) and MT3, with eight species each, and the least diverse were MT7 and MT9 with four and three species, respectively. The remaining morphotypes contained between five and seven species. DFA allowed the differentiation of morphotypes based on four significant discriminant functions (DF1 ΛWilks = 0.001, P < 0.01; DF2 ΛWilks = 0.08, P < 0.01; DF3 ΛWilks = 0.069, P < 0.01; DF4 ΛWilks = 0.220, P < 0.01). The first three DFs accounted for 92.1% of the morphological variation of morphotypes and were correlated with SVL, relative hindleg length, and relative number of lamellae, respectively (Fig. 2). The fourth function accounted for 4.4% of variation and was correlated with tail length. Morphotypes differed in SVL (F = 45.35, P < 0.05). Small-sized anole species belonged to MT1 to 5 and generally had a mean SVL of < 60 mm. Large-sized anole species belonged to MT6, 7, and 8, and had a mean SVL of between 60 and 90 mm. Giant anole species belonged to MT9 and 10 with a mean SVL of > 90 mm (Table 1). We found that tail length was also different amongst morphotypes (F = 3.88, P = 0.01). Most species had tails between 1.5 and two times their SVL. Short-tailed anoles belonged to MT4 and generally had a tail less than 1.5 times their SVL. Long-tailed anoles belonged to MT1, 6, 8, 9, and 10, and exhibited a tail twice as long as their SVL (Table S1). Morphotypes also showed differences in their relative foreleg length (F = P < 0.01). Most Colombian mainland anoles had forelegs of between 35 and 40% of their SVL in length. Short foreleg anoles belonged to MT4 and generally exhibited forelegs of less than 35% of their SVL in length. Anoles with long forelegs belonged to MT2, 3, 5, 6, and 10, and possessed forelimbs of more than 40% of their SVL in length (Table 1). Relative hindleg length differed amongst morphotypes (F = 26.39, P < 0.01). Most Colombian mainland anoles had legs between 60 and 80% of their SVL in length. Short-hindleg anoles belonged to MT4, and had hindlegs that were less than 50% of their SVL in length. Long-hindleg anoles belonged to MT2, 5, 6, and 10, and had hindlegs that were more than 80% of their SVL in length (Table S1). Morphotypes also differed in relative lamellae number (F = 9.37, P < 0.01). Small species with few lamellae belonged to MT1, 2, and 5, and possessed fewer than 18 lamellae. Large or giant anoles with few lamellae (< 22) belonged to MT6, 8, and 9 (Table S1). We did not find any significant differences amongst morphotypes in

6 DIVERSIFICATION OF SOUTH 637 Figure 2. Discriminant functions describing ten anole morphotypes (MTs) based on 59 species of Colombian Anolis. Blue: MT1, green: MT2; black: MT3, purple: MT4; yellow: MT5; red: MT6; cyan: MT7; grey: MT8, orange: MT9, white: MT10. Table 1. Comparisons of Anolis beta diversity amongst ecogeographical regions MBC SBC LCV MMV ARF UMV CP LS WA CA EA SNSM NBC MBC SBC LCV MMV ARF UMV CP LS WA CA EA 1.00 Values > 0.6 are in bold.arf, Cis-Andean tropical rain forests in the Amazonas; CA, central Andes; CP, Caribbean Plains; EA, eastern Andes; LCV, Lower Cauca River Valley; LS, Cis-Andean savannahs in Los Llanos; MBC, Middle Biogeographical Chocó; MMV, Middle Magdalena River Valley; NBC, Northern Biogeographical Chocó; SBC, Southern Biogeographical Chocó; SNSM, Sierra Nevada de Santa Marta; UMV, Upper Magdalena River Valley; WA, western Andes. trunk length relative to SVL (F = 1.59, P = 0.143) or head dimensions relative to SVL (width: F = 0.66, P = 0.74; height: F = 1.07, P = 0.41; length: F = 0.658, P = 0.74). GEOGRAPHICAL DISTRIBUTION AND NATURAL HISTORY OF MORPHOTYPES We found seven morphotypes shared between the lowlands and Andean regions, two morphotypes (MT2 and 5) that were unique to the lowlands, and one (MT4) that was exclusive to montane regions (Fig. 1). When comparing lowland regions, only MT7 was exclusive to cis-andean tropical rain forests. MT10 was only observed in trans-andean rain forests and cloud forests of WA; the trans-andean dry forests did not have any unique morphotypes, whereas MT9 was shared by CP and Chocó forests. In montane ecogeographical regions, the Andean and cloud forests of WA, CA, and EA shared

7 638 R. A. MORENO-ARIAS AND M. L. CALDERÓN-ESPINOSA four morphotypes; the unique morphotype in the Paramos was MT4, and LS and SNSM were the regions with least morphotype diversity (Fig. 1). Trans- Andean rain forests contained species that represent 90% of morphotypes and the highest species diversity. Cis-Andean rain forest contained only 50% of morphotypes, represented by nine species (Fig. 1). Dry regions like CP and LS possessed 60% of morphotypes. In Andean regions, WA possessed the highest diversity of species whereas EA had the highest morphotype diversity (70%) (Fig. 1). Mean beta diversity values amongst ecogeographical regions was high (0.85), with areas very similar in species composition (0.18) to fully dissimilar (1.0). Lowland regions with the lowest values (< 0.4) of beta diversity were NBC-MBC, MBC-SBC, LCV-CP-MMV, and LS-ARF. However, amongst those four groups, mean beta diversity was high (> 0.60). Montane regions were highly dissimilar, with beta diversity values above 0.70 (Table 1). Anoles living at the lowest stratum, which use ground and herbaceous strata mostly in leaf litter and bushes, belonged to MT2; species of MT5 also used the ground but reached the upper understorey stratum and used mostly trunk microhabitats (Table S1). Species of morphotypes that used mostly understorey stratum were from MT1, 4, and 6; these preferred bushes, and some species used grasses or trunks. Species of both MT3 and 8 inhabit strata from the understory to the canopy, but species of MT3 use trunks whereas anoles of MT8 prefer bushes. Anoles living in the upper levels of the habitat belong to MT7 and 10, and use trunks to reach the canopy stratum (Table S1). Anoles of MT9 do not share strata and microhabitat use between them; for example, Anolis fraseri and Anoles mirus inhabit bushes in the understorey, but A. fraseri can reach the canopy and also uses trunks. By contrast, Anolis onca is more terrestrial and prefers ground and herbaceous strata; however, it has been observed on trunks of small trees and wooden structures of human buildings (M. L. Calderón-Espinosa, pers. observ.; Table S1). MORPHOTYPE EVOLUTION In the phylogenetic analysis we found that the data set without partitions ( natural logarithm of Likelihood lnl ) was preferred over the other data sets (gen partition lnl and codon partition lnl ). The Bayes factor of the codon partition vs. no partition data sets was 395.5, and that for the gen partition vs. no partition data sets was 11.03, indicating support for the data set without partitions. The inferred topology (Fig. 3) was fully resolved and showed two main clades with high support, representing the Dactyloa and Norops clades; the main groups within Dactyloa correspond to the Phenacosaurus, Western, and latifrons clades sensu Castañeda & de Queiroz (2011). Phylogenetic signal was detected for most traits (SVL K = 1.05 P = 0.001, TLL K = 0.97 P = 0.001, FLL K = 1.13 P = 0.001, HLL K = 1.22 P = 0.001, TRL K = 0.75 P = 0.038, LN K = 1.01 P = 0.001). Only three traits showed no phylogenetic signal (HW K = 0.72 P = 0.085, HH K = 0.48 P = 0.780, HL K = 0.43 P = 0.911). Evolution of morphological diversity in mainland anoles seems to have followed four convergent regimes across ten shifts, as suggested by the most strongly supported model MORPHbwd. Two of the four regimes matched with MT4 and 8 and remain regimes are associated with the main clades Norops and Dactyloa (Fig. 4). The other multipeak scenarios (SURF and MORPHfwd) had no support (Fig. 4). Other scenarios also had no support and did not describe the morphological diversification of Anolis of northwestern South America: OU (lnl = 1139, Δ AIC c = 94.2), BM (lnl = 1132, Δ AIC c = 62.4), and EB (lnl = , Δ AIC c = 125). COMPARISON BETWEEN MAINLAND MORPHOTYPES AND CARIBBEAN ECOMORPHS Four first DFs based on data without phylogenetic correction accounted for 96% of the morphological variation (DF1: ΛWilks = P < 0.01; DF2: ΛWilks = 0.02 P < 0.01; DF3: ΛWilks = 0.1 P < 0.01; DF4: ΛWilks = P < 0.01). The first and second DFs correlated with SVL and relative hindleg length, respectively (Fig. 5). The grass-bush species were classified into MT1 and 5. The trunk-crown species were classified into MT7 and 3. The trunk-ground species classified into MT5 and 3. Trunk anole species classified into MT3. The twig anole and crown-giant anole species were classified into MT4 and 10, respectively. Analysis of data with surface identified six regimes, collapsing MT10 with crown-giant, twig species with MT4, trunk-ground with MT5, trunk-crown species with MT7, and MT8 species in MT6. In addition, MT1, 2, 3, grassbush, and trunk species were combined into a single regime (Fig. 5). DISCUSSION Mainland anoles represent almost 60% of Anolis species (Losos, 2009) and our evaluated species represent nearly 30% of the mainland radiation. Nonetheless, our results essentially represent the full range of phenotypic diversity across the main geographical regions of northwestern South America. The main morphological patterns of anoles identified here are not expected to change, although those species for which little data are available might be classified into other morphotypes when more data are acquired. In general, the morphological diversity of Colombian mainland anoles can

8 DIVERSIFICATION OF SOUTH 639 Figure 3. Consensus Bayesian tree for 28 species of mainland Colombian Anolis, their main clades, and their geographical distribution. Values at nodes indicate posterior probability (PP); black circles at nodes indicate PP = 1.0. Coloured ovals at tips identify morphotypes (MTs) as follows: blue: MT1, green: MT2; black: MT3, purple: MT4; yellow: MT5; red: MT6; cyan: MT7; grey: MT8, orange: MT9, white: MT10. Filled bars indicate geographical distribution: montane (white), cis-andean (grey), trans-andean (black), and wide distribution (red). be summarized into ten morphotypes defined mainly by their body size and secondarily by their body shape, defined here as proportions of limbs, tail, and lamellae number (as a proxy of the toe length) to SVL. The pattern of morphological variation found here resembles the Anolis assemblage from the Middle Biogeographical Chocó (Velasco & Herrel, 2007) and the West Indies anoles (Losos, 1992), where morphological variation is explained primarily by body size and secondarily by body shape. This finding is not surprising given that these traits are important for structuring anole communities (Schoener, 1968; Williams, 1972; Roughgarden, 1974). Differences in body size and shape promote coexistence because morphology determines greatly the manner in which species use the resources within a habitat (Garland & Losos, 1994). In this way, body size relates to food type and perch height whereas limb proportion relates to perch type and foraging behaviour (Losos, 1990b, c, 1992). In addition, all of these traits together not only restrict the performance of species to the use of specific structural habitats, but also restrict

9 640 R. A. MORENO-ARIAS AND M. L. CALDERO N-ESPINOSA Figure 4. Morphotype Regime (MORPHfwd), Morphoype Regime Collapsed (MORPHbwd) and Surface Regime (SURF) hypothesis performance of morphological regime shifts for 28 species of Anolis of northwestern South America. Colours indicate morphotypes (MTs) as follows: blue: MT1, green: MT2; black: MT3, purple: MT4; yellow: MT5; red: MT6; cyan: MT7; grey: MT8, orange: MT9, light grey: MT10. interspecific relationships with other anole species. Our results reinforce the idea that morphological variation is a relevant factor in structuring anole communities in northwestern South America. The morphological diversity of Anolis from mainland Colombia seems to be higher than in the West Indies anoles when only the six typical ecomorphs described for the islands are considered. However, many other species considered as unique ecomorphs, such as false chameleons, terrestrial, semiaquatic, rock-wall dwelling and cave ecomorphs, occur throughout the West Indies (Losos, 2009). Recently, Mahler et al. (2013) found 15 adaptive peaks (morphotypes) occurring in the West Indies, including unique ecomorphs. Direct comparison between our findings and Mahler et al. s (2013) study is not possible owing to differences in the morphological and phylogenetic data. Our findings are based on species and four to nine traits, whereas Mahler et al. (2013) used a phylogeny of 100 species and 11 traits; however, then we identified higher morphological diversity with less morphological and phylogenetic data. It is possible that we are underestimating the real morphological variation of the anole mainland radiation, to include only 59 species. However if we combine our data (small phylogeny with nine traits and large

10 Figure 5. Morphological comparison between species of northwestern South America and the Caribbean. In the DFA, the triangles are Caribbean ecomorphs. Colours indicate morphotypes (MTs) as follows: blue: MT1, green: MT2; black: MT3, purple: MT4; yellow: MT5; red: MT6; cyan: MT7; grey: MT8, orange: MT9, light grey: MT10. Ecomorphs are labelled as CG, crown-giant; GB, grass-bush; TC, trunk-crown; TG, trunk-ground; TR, trunk-ground; and TW, twig Caribbean ecomorphs. Blue colour (A) in backward chronograph includes MT1, MT2, MT3, GB, and TR. DIVERSIFICATION OF SOUTH 641

11 642 R. A. MORENO-ARIAS AND M. L. CALDERÓN-ESPINOSA phylogeny with five traits) some interesting issues emerge: first, we identified a multipeak adaptive model as the most probable scenario to explain the morphological variation in mainland Anolis; second, our analyses supported the convergent nature of the morphological diversity of Anolis of northwestern South America; and third, some morphotypes match the Caribbean ecomorphs and there are at least two morphological regimes unique to the mainland. MORPHOLOGICAL COMPOSITION OF ANOLIS ASSEMBLAGES OF NORTHWESTERN SOUTH AMERICA The morphological composition of the Anolis mainland assemblages showed an interesting pattern of shared morphotypes amongst sites with similar vegetation types and altitudes despite the sites having different species compositions, as reflected in the high beta diversity values. Morphological Anolis assemblages can be grouped according to elevation and vegetation as follows: anoles of lowland forest, montane forest, grassland and savannah, and Paramos. Lowland forests include dry forests with under 1800 mm of precipitation and deciduous vegetation in the dry season, and rain forests with precipitation above 1800 mm and evergreen vegetation. Despite climatic differences between these lowland forests, both types offer multiple vegetation strata that are favourable to Anolis, such as canopy, understorey, herbaceous layer, and leaf litter. As anoles are mostly arboreal lizards and given what we know about their adaptive evolution, we expect these environments to support high species and morphotype diversity, but not all do so. For example, Anolis assemblages from the dry forests of the upper Magdalena River show low species diversity. Despite differences in species diversity amongst localities, each anole assemblage of lowland trans- or cis-andean forests was composed of a minimum of four morphotypes: MT1, 2, 3, and 5, each one associated with one of the main forest strata. Some peculiarities in morphotype composition of forest lowland assemblages emerge. In localities from SBC forest inhabits Anolis fasciatus, a species of MT6 that is mostly montane. The occurrence of montane morphotypes in lowland forests, especially in the Chocó forest, could be explained by the lower limit of montane forests on the western slope of the Andes than on the central and eastern slopes. Alternatively, Amazonian assemblages also include a unique morphotype (MT7) composed of species of the Dactyloa Eastern Clade with cis-andean distributions, whose origin is probably Amazonian (Castañeda & de Queiroz, 2011). The next group with high morphological diversity corresponds to montane forest assemblages and some studies of Colombian montane anoles indicate that at least three species are observed per locality (Hernández-Ruz et al., 2001; Caicedo et al., 2006; Molina-Zuluaga & Gutiérrez-Cárdenas, 2007; Moreno-Arias et al., 2009). Cloud and Andean forests offer the same vegetal strata as lowland forests; species in these environments correspond to arboreal anoles and include morphotypes of canopy (MT4 and 10), understory anoles (MT3 and 6), and understorey and herbaceous anoles (MT1). However, unlike lowland forests, species of morphotypes that use leaf litter are not found in cloud and Andean forests. Paramos only occur above 3200 m a.s.l. and these are environments dominated by tall grasses in the herbaceous stratum and by Espeletia spp. and Ericaceae in the shrub stratum. Only anole species of MT4 are found in these mountain-top environments. Savannah environments are grasslands with scarce forest vegetation, which includes only riparian forest, palm forests, and matas de monte (isolated arboreal formations inside savannahs). Despite the sparse vegetation, this environment also includes all strata mentioned previously, yet only understorey species like MT1 and leaf-litter species of MT5 are observed in these environments. Savannah is probably the extreme lowland environment for anoles because precipitation is very high in the rainy season and very low in the dry season, thus maintaining hydric stress throughout the year. In addition, forest vegetation is a limiting resource that is only available in narrow riparian forest and small and isolated matas de monte. In the highlands, Paramos are extreme environments, not only because of the altitude, which is a constraint for reptiles, but also because of the extreme temperature change across the day, and because Paramos also suffer hydric stress, especially during the dry season. Drastic reductions in numbers of Anolis species and morphological diversity are associated with increases in elevation. Recent evidence has shown that environmental filters, such as elevation and climate, are strong determinants of the structure of local Anolis ecomorph communities because they affect environmental niches that functionally restrict some ecomorphs, and hence produce a patchy distribution of sets of coexisting ecomorphs (Wollenberg, Veith & Lötters, 2014). In general terms, this pattern can be used to describe differences in the lowland and highland anole morphotype compositions across northwestern South America. An idea to be tested is whether or not species of MT2 and 5 exhibit low thermoregulation performance in highlands compared with lowlands. This could explain the absence of forest terrestrial anoles in montane forests given that the lowest strata in montane forests are less warm owing to the altitude and the constant presence of clouds. As we mentioned above, because highland environments are climatically extreme, we would expect that

12 DIVERSIFICATION OF SOUTH 643 species display adaptations to these challenging conditions; indeed, some highland species have larger body scales and perform better under hydric stress than lowland anoles. In addition, differences in the vegetation structure in highlands compared with lowlands may relate to the absence of canopy anoles. However, it is interesting that it is highland-specific anoles (MT4) that exhibit the most extreme morphology, shortest arms and legs, of all the anole morphotypes. They preferentially use twigs and very narrow surfaces and exhibit very slow movements (owing to their very shorts limbs). Therefore, environmental filters could be explain these features, and we hypothesize that predator interactions (biotic filters) could play a more relevant role, and that very slow movements are probably the best strategy to avoid predators that respond to fast movements, such as birds. These ideas are reinforced when taking into account that virtually all highland diurnal predators are birds, and that arboreal diurnal snakes (other typical predators of anoles in lowlands) are only represented by Chironius monticola. ECOLOGY OF ANOLIS ASSEMBLAGES OF NORTHWESTERN SOUTH AMERICA Several studies have revealed relationships amongst morphology, ecology, and behaviour (Miles & Ricklefs, 1984; Pianka, 1986). In Anolis, some morphological attributes translate into performance capability, and these capabilities constrain the ecology of anoles (Losos, 1990a). Anole lizards are an example of ecomorphological relationships. Williams (1972) coined ecomorph as a term to define groups of anole species with similar morphology and habitat use but without close phylogenetic relationships. Later, Losos (1990a, c) demonstrated quantitatively the links amongst morphological, ecological, and behavioural traits in Greater Antilles anoles. Owing to the evolution of subdigital lamellae, anoles are mostly arboreal but several species also show diverse microhabitat uses, and are semiarboreal, terrestrial, or semiaquatic (Losos, 2009). Colombian anoles are not an exception and also use a wide variety of microhabitats; six morphotypes are arboreal, and the remaining four are semi-arboreal. Semiaquatic species also occur but they cannot be classified into a particular morphotype. Although the ecology of many mainland anole species is unknown, the matches between morphotypes and ecomorphs found here permit us to infer some associations between our morphological groups and their predominant habitat use. Giant species of MT10 and some of MT7 use mostly tree trunks in the upper strata and canopy. Anolis frenatus and Anolis punctatus seem ecologically similar to the crown-giant ecomorph from the Greater Antilles (Irschick et al., 1997), but A. punctatus is not enough large to be a giant anole and it also matches with the trunk-crown ecomorph. The absence of MT10 in Amazonas can be explained because giant anole evolution was constrained by the presence of larger canopy lizards, such as Plica plica and Uracentron flaviceps, which are large lizards that use trunks of large trees (Vitt, Zani & Avila-Pires, 1997; R. A. Moreno-Arias pers. observ. 2014). These lizards may have restricted the evolution of anoles towards giant forms owing to niche incumbency or a priority effect. In addition, several species of MT8 use shrubs or tree trunks and branches in the canopy and understorey strata, but species of MT6 are more common in the understorey to herbaceous strata. These two morphotypes apparently do not have a Caribbean ecomorph counterpart, although Anolis biporcatus and Anolis chocorum are recognized as crown-giants (Castro-Herrera, 1988; Losos, 2009). Small anoles of MT3 live in the lowlands, and use mostly tree trunks from the understorey to the canopy (Table S1). Whereas Anolis anchicayae, Anolis chloris, Anolis ortonii and Anolis peraccae resemble the trunkcrown ecomorph, Anolis pentaprion was previously recognized as twig anole (Irschick et al., 1997; Losos, 2009). For the latter two species, recent ecological data (R. A. Moreno-Arias, unpubl. data) plus their lichenose colour pattern indicate that these species resemble the Caribbean trunk ecomorph more than the trunkcrown ecomorph. Small and tiny anoles of MT4 are found from midmountain to Paramos regions, and use branches and twigs from the understorey to the canopy (Table S1). This group shows a greater similarity to a Caribbean ecomorph than does any other group. In the same vein as other studies (Schaad & Poe, 2010), our results also identified MT4 species as twig anoles. Species of MT1 use mostly trunks and branches of small trees, shrubs, and herbs of the understorey or herbaceous strata, but rarely use the ground (Table S1). Only two species (Anolis auratus and Anolis fuscoauratus) have been recognized by other authors as grass-bush anoles (Irschick et al., 1997; Schaad & Poe, 2010). Species of MT2 inhabit the ground, leaf litter, and log falls, and sometimes use shrubs and herbs (Table S1). Schaad & Poe (2010) identified Anolis granuliceps as a member of the grass-bush ecomorph despite it was the species morphologically furthest from members of the grass-bush group. Irschick et al. (1997) showed that Anolis trachyderma is the morphological centroid of the trunk-ground morphotype, but its nearest neighbour was a grass-bush species. Matching a Caribbean ecomorph for the MT2 species is complicated, and requires more quantitative data of their ecology; however, we suggest that species of MT2 correspond

13 644 R. A. MORENO-ARIAS AND M. L. CALDERÓN-ESPINOSA to a unique mainland ecomorph owing to their greater preference for the ground compared with the other morphotypes. MT5 anoles use tree trunks, shrubs, and ground in nearly the same proportions (Table S1). Anolis scypheus grouped within the trunk-ground ecomorph (Irschick et al., 1997), and the ecology of the remaining species suggests that they are also trunk-ground anoles. Large anoles of MT9, despite their shared morphological attributes, show disparate microhabitat uses. Anolis fraseri uses tree trunks in the canopy like species of MT10; A. mirus is similar to species of MT6 that also use branches in the understorey, whereas A. onca is a typical anole of xerophytic vegetation near beaches that uses bushes, shrubs, and the ground, similar to species of MT1. It is interesting that MT5 anoles possess the most robust bodies and some authors have observed very aggressive behaviour in A. fraseri and A. mirus (F. Castro-Herrera & S. Ayala, unpubl. data), which leaves the relationship between body robustness and aggressive behaviour occurring in this clade as an open hypothesis to test. Semiaquatic anoles do not constitute a morphologically distinctive group on the basis of the characters used here. Instead, three species (Anolis lynchi, Anolis poecilopus, and Anolis rivalis) were assigned to MT5 and the other two species (Anolis macrolepis and Anolis maculigula) to MT1 and MT8, respectively. All semiaquatic species use vegetation of herbaceous and understorey strata surrounding streams and pools, but also use other microhabitats. Anolis macrolepis is also found on the ground, whereas A. maculigula, A. poecilopus, and A. rivalis use rocks and boulders in streams and pools (Williams, 1984; Miyata, 1985; F. Castro-Herrera & S. Ayala, unpubl. data). Therefore, semiaquatic anoles support the idea that the semiaquatic habit can be acquired through a variety of body plans. Another possibility is that other traits not evaluated here, such as tail shape (related to swimming performance), could be a link amongst species exhibiting a semiaquatic ecology, as suggested by Leal, Knox & Losos (2002). Studies on performance and morphology will give some clues to defining semiaquatic anoles as an ecomorph type. EVOLUTION OF ANOLIS MORPHOLOGICAL DIVERSITY OF NORTHWESTERN SOUTH AMERICA A phylogenetic signal was present in the data analysed here for six traits (SVL, TLL, FLL, HLL, TRL and LN). Most morphological traits followed a trajectory influenced by the phylogenetic history of species. Nevertheless, Revell et al. (2008) showed that, in addition to genetic drift, high values of phylogenetic signal occur when evolution is not entirely neutral, such as in stabilizing selection at a constantly weak strength, highly fluctuating natural selection, or when the evolutionary rate or changes in adaptive shifts decrease through time. Evolution of morphological traits in anoles shows different intensities of phylogenetic signal; variables representing limbs and size features can have low or high Blomberg et al. K values, but always with significant values for phylogenetic signal (Pinto et al., 2008; Hertz et al., 2013). In addition, convergent and deterministic evolution of morphology and ecology, bounded by ecological limits, is recognized as the main pattern of evolution of anole Caribbean radiation (Losos et al., 1998; Mahler et al., 2010, 2013). This pattern was the most frequently observed in our data, as most traits exhibited high phylogenetic signal without fitting neutral models, suggesting a deterministic morphological pattern of evolution for mainland anoles in northwestern South America. Other traits, like head dimensions, which showed low Blomberg et al. K values, exhibited phylogenetic correlation. Low values for HW could be a result of punctuated divergent selection. Low values for HH and HL could be explained by a strong stabilizing selection or bounded phenotypic evolution under high mutation rates, as some theoretical models suggest for evolutionary processes that show low phylogenetic signal (Revell et al., 2008). In relation to the tempo of evolution of morphological traits, the neutral model BM had high support compared with the variable rate of evolution model EB, suggesting that phylogeny best predicts the patterns and rates of evolution for these particular traits; this supports the stabilizing selection scenarios, which were more supported than the BM or EB, for explaining morphological evolution. Although we cannot infer the magnitude of the fluctuation of selection across the phylogeny to make inferences of where the main changes occur, the links between (1) constant rates of evolution based on very poor support of EB (AIC c = 57.6) compared with BM, (2) strength of selection towards a particular phenotypic peak, and (3) high values of phylogenetic signal in most traits, suggest that the main morphological shifts are concentrated near the root of the tree. It is likely that major morphological shifts in northwestern South America mainland anoles occurred early in their history. Therefore, the species to fill the same morphological peak increased their morphological covariance, with respect to species of other morphological peaks, more than expected by neutral evolution, which explains the high values of phylogenetic signal observed in most traits. Thus, the observed pattern of morphological evolution of mainland anoles from northwestern South America is similar to the niche occupancy (Revell et al., 2008) and niche