## Abstract

Spatial patterns of heritable phenotypic diversity reflect the relative roles of gene flow and selection in determining geographic variation within a species. We quantified color differentiation and genetic divergence among 20 populations of the red-eyed tree frog (Agalychnis callidryas) in lower Central America. Phylogenetic analyses revealed 5 well-supported mitochondrial DNA clades, and we infer from our phylogeny that geographic barriers have played a large role in structuring populations. Two phenotypic characters varied independently among isolated population groups: Flank coloration distinguished Caribbean from Pacific individuals, whereas leg coloration exhibited a more complex geographic pattern. We detected 3 generalized spatial patterns of genetic and phenotypic diversity: 1) phenotypic differentiation in the presence of historical connectivity, 2) phenotypic uniformity across genetically differentiated regions, and 3) codistribution of genetic and phenotypic characters. These patterns indicate that phenotypic diversification is highly regionalized and can result from spatial variation in localized adaptations, geographic isolation, genetic drift, and/or evolutionary stasis. Although the mode of selection underlying color variation was not the focal objective of this study, we discuss the possible roles of natural and sexual selection in mediating population differentiation. Our study underscores the fact that selection gradients vary across relatively small spatial scales, even in species that occupy relatively homogeneous environments.

The spatial distribution of divergent phenotypes provides insight into the relative roles of natural and sexual selection, gene flow, and vicariance in the diversification of a species (Grinnell 1924; Ford 1971; Slatkin 1985; Brown et al. 1996). Many taxa exhibit some phenotypic variation among individuals within populations, although the spatial distribution of genetic and phenotypic diversity varies substantially among species (Hoffman and Blouin 2000; Gray and McKinnon 2007). For some species, geographic clines in body size (Brown and Thorpe 1991; Storz 2002), behavior (Thompson 1990; Pröhl et al. 2006), color pattern (McDiarmid 1968; Stewart 1974; Hoffman and Blouin 2000; Woolbright and Stewart 2008), life-history traits (Dhondt et al. 1990), or ornamentation (Nevo 1973; Gray 1983; Brooks and Endler 2001a; Storz et al. 2001) result from selection acting on continuous traits across selective gradients (Endler 1973; Storz et al. 2001; Storz 2002), genetic drift, or both (Hoffman et al. 2006). In contrast, far fewer species are highly polytypic across their range, with significant phenotypic variation among but low variation within populations. In those cases, divergence usually results from isolation due to dispersal barriers that can also be reinforced by local adaptation, sexual selection, and genetic drift (Summers et al. 2003; Fuller et al. 2004; Maan et al. 2006; Boul et al. 2007). These 2 patterns of spatial variation are not exclusive; some species exhibit phenotypic divergence among regions (due to dispersal barriers) as well as clinal variation within regions (Endler 1973).

The interaction between gene flow and localized selection determines the extent of phenotypic differentiation among localities and the rate and direction of phenotypic change over time (Grinnell 1924; Slatkin 1985; Lenormand 2002). For example, high levels of gene flow, either due to dispersal capacity or the absence of isolating barriers, combined with limited localized selection result in widespread genetic and phenotypic homogeneity. In contrast, strong localized divergent selection and geographic isolation can result in highly structured populations (King and Lawson 1995; Boul et al. 2007). Congruence in spatial patterns of variation among multiple traits indicates that trait evolution has been mediated by similar historical microevolutionary processes. Alternatively, discordance between genetic and phenotypic characters could demonstrate evolutionary stasis for phenotypic traits despite restrictions in gene flow or from strong localized selection in the presence of gene flow (Endler 1973; Gray 1983; Dallimer et al. 2003; Hoekstra et al. 2005; Jordan et al. 2005; Pröhl et al. 2006; Rosenblum 2006).

The red-eyed tree frog, Agalychnis callidryas, is a common neotropical frog broadly distributed from Central Mexico to Colombia (Duellman 2001; Savage 2002); this species is polytypic (Savage and Heyer 1967; Duellman 2001; Robertson JM and Robertson AD 2008) and shows sufficient phenotypic differentiation to distinguish frogs from 5 biogeographic regions in Costa Rica (CR) and Panama (PA) with high accuracy (Robertson JM and Robertson AD 2008; Robertson et al. 2009). Unlike the green dorsum coloration in this species, leg and flank colors do not change with light intensity (Schliwa and Euteneuer 1983) or after adult coloration is acquired. Captive breeding of A. callidryas (Gomez-Mestre I, personal communication) indicates that color pattern has a strong heritable component. Thus, although further studies are necessary to rule out any environmental modulation of color or phenotypic plasticity, the available evidence indicates that color pattern is a reliable character for evolutionary studies.

Previous analyses revealed regional diversity in leg coloration for A. callidryas (Robertson JM and Robertson AD 2008); a subsequent study of gene flow patterns among these regions, using polymorphic microsatellite loci, indicated that populations were structured into 5 geographic regions with admixture among some, but not all, neighboring regions (Robertson et al. 2009). Here, we tested whether these 5 geographic regions are associated with historical barriers to gene flow in lower Central America by examining mitochondrial DNA (mtDNA) haplotype diversification and the distribution of 2 phenotypes (flank and leg coloration).

The 5 regions in this study have been shaped by the complex geological history of lower Central America (Holdridge 1947; Kohlmann et al. 2002; Savage 2002; Kohlmann et al. 2007a). Three of these regions occupy the wet Caribbean forest east of the Continental Divide (northeastern CR, southeastern CR, and Central PA), whereas 2 occupy the Pacific slopes of the Continental Divide (northwestern CR, southwestern CR; Figure 1). Three major landscape features coincide with the geographic borders between regions and are known to structure populations of other Central American taxa (Kohlmann et al. 2002). The exact nature of the other regional breaks is not well understood. Here, we discuss each of 5 putative barriers that we hypothesize contribute to geographic structure among red-eyed tree frog populations. 1) Major montane features include the Cordillera de Talamanca, and 3 smaller volcanic mountain ranges (Tilarán Mountain, Cordillera Central, and Cordillera Guanacaste) that lie along the Continental Divide and isolate Caribbean from Pacific populations. The uplift of the Cordillera de Talamanca occurred approximately 3–5.4 million years ago (Ma) (Coates and Obando 1996; Kohlmann et al. 2002) and has imposed a strong barrier to gene exchange between Pacific and Caribbean populations of terrestrial snakes (Zamudio and Greene 1997), frogs (Crawford 2003), beetles (Kohlmann et al. 2007a), and montane salamanders (García-París et al. 2000). This mountain range extends 400 km along the Central American Continental Divide and reaches its highest point of 3800 m at Cerro Chirripó (Coates and Obando 1996; Kohlmann et al. 2002). The wet tropical forest typical of lower elevations is replaced by cloud forest and dry Páramo above 3100 m (Kohlmann et al. 2002). These higher elevation habitats are inhospitable to A. callidryas and likely prohibit movement across the Cordillera de Talamanca. 2) The Osa Peninsula is well known for its high endemicity and unique distribution of plants and animals, supporting the geological hypothesis that the Osa Peninsula was an offshore island that drifted into the mainland of CR approximately 2 Ma (Kohlmann et al. 2002). 3) Prior to the formation of the Cordillera de Talamanca, the region of Bocas del Toro experienced a short-lived uplift, approximately 5–7 Ma. Multiple colonization events consistent with the uplift have been documented for other taxa (Zeh et al. 2003; Weigt et al. 2005). This region behaves as a geographic barrier for other anuran taxa (Crawford 2003; Summers et al. 2003; Wang et al. 2008) and has been referred to as the “Bocas Break” (Crawford et al. 2007). 4) Plate tectonic and microplate tectonic activity resulted in the isolation of dry forest in northwestern CR and wet forest in southwestern CR (Holdridge 1947; Kohlmann et al. 2002; Crawford et al. 2007). This divide is important for insect and plant taxa (Kohlmann et al. 2007a, 2007b), but the consequences for anuran population structure is unclear. We refer to this barrier as the Río Parrita because this river drains from the Cordillera de Talamanca into the Pacific Ocean and coincides with the geological history of Pacific coast forest. Although the river may not be a barrier, it marks the geographical location of historical barriers that occurred in that region. 5) The Caribbean lowland forest is primarily wet, lowland tropical forest in northeastern CR and transitions into a series of floodplain valleys that occupy the southeastern CR region (Valle de la Estrella, Valle de Talamanca, and Llanura de Santa Clara [Kohlmann et al. 2002, Figure 1]). The isolation of northeastern and southeastern CR is not well understood for terrestrial, lowland taxa but coincides with the geographic distribution of some anuran and beetle taxa (Kohlmann et al. 2002; Savage 2002; Hagemann and Pröhl 2007). Here, we refer to this putative barrier as the Caribbean Valley Complex.

Figure 1

Sampling localities for 20 populations of Agalychnis callidryas in five biogeographic regions. The Cordillera de Talamanca isolates the Pacific (open symbols) and Caribbean (dark symbols) versants of CR and PA. The shading of topological relief corresponds to elevation: dark gray (>1300 m), light gray (300 – 1299 m), white (0 – 300 m). Elevations higher than 1300 m exceed the physiological tolerance of A. callidryas. Dry forest habitat between Southwestern CR and Central PA prevents the occurrence of red-eyed tree frogs along the Pacific coast of PA. Putative biogeographic barriers tested in this study: Cordillera de Talamanca, Caribbean Valley Complex, Bocas del Toro, Río Parrita, and Osa Peninsula. Also featured are the Cordillera Central (CC), Tilarán mountains (T) and Cordillera Guanacaste (CG). Representative photographs of frogs from each region demonstrate spatial variation in leg coloration.

Figure 1

Sampling localities for 20 populations of Agalychnis callidryas in five biogeographic regions. The Cordillera de Talamanca isolates the Pacific (open symbols) and Caribbean (dark symbols) versants of CR and PA. The shading of topological relief corresponds to elevation: dark gray (>1300 m), light gray (300 – 1299 m), white (0 – 300 m). Elevations higher than 1300 m exceed the physiological tolerance of A. callidryas. Dry forest habitat between Southwestern CR and Central PA prevents the occurrence of red-eyed tree frogs along the Pacific coast of PA. Putative biogeographic barriers tested in this study: Cordillera de Talamanca, Caribbean Valley Complex, Bocas del Toro, Río Parrita, and Osa Peninsula. Also featured are the Cordillera Central (CC), Tilarán mountains (T) and Cordillera Guanacaste (CG). Representative photographs of frogs from each region demonstrate spatial variation in leg coloration.

In this study, we compared the distribution of variation in flank and leg color with that of genetic lineages across this region to test hypotheses about the role of selection and gene flow in patterns of diversification. Our sampling throughout CR and PA represents approximately 25% of the total geographic range of the species and encompasses all the known color variants (Duellman 2001). Our specific objectives were to test 1) the role of isolation due to geographic distance and/or geographic features in structuring genetic diversity, 2) whether regional color variation could also be explained by the same geographic factors, and 3) the hypothesis that spatial patterns of phenotypic and genetic diversity were concordant.

## Material and Methods

### Field Sampling

We quantified patterns of genetic and phenotypic variation among 20 red-eyed tree frog populations from 5 biogeographic regions in CR and PA (Figure 1). We expanded on an earlier study of phenotypic variation (Robertson JM and Robertson AD 2008) to include more populations, an additional measure of phenotypic diversity (flank coloration), and an analysis of mtDNA differentiation among the same individuals to serve as a comparative framework for the phenotypic data. We conducted field surveys during the breeding seasons (May–August) of 2003, 2004, and 2005. At each site, we captured 10–26 individuals and collected data on leg and flank coloration and pattern. We photographed every individual using a Nikon Coolpix 5700 against a black–white–gray card for color standardization. For each individual, we photographed the posterior surface of the thighs, left flank, and right flank of the body.

Up to 3 individuals per population were preserved as vouchers and deposited at the Cornell University Museum of Vertebrates (CUMV: 14093, 14206–14211, 14228, 14231–14233) and the University of CR, San Jose, CA (accession numbers: 19100–19101, 19213). All photographs have been archived at the CUMV. Nonvouchered individuals were photographed, toe clipped for genetic material, and released at site of capture.

### Population Genetic Variation

We extracted genomic DNA from 125 individuals sampled for this study (Table 1). Toe clips or liver was digested in standard lysis buffer with proteinase K followed by purification using the Qiagen DNeasy Tissue Kit (Qiagen, Valencia, CA) following manufacturer's protocols. We amplified a fragment of the mtDNA including the partial 16S rRNA, the complete nicotinamide adenine dinucleotide plus hydrogen (NADH) dehydrogenase subunit 1, and the adjacent flanking tRNAMet (hereafter, we refer to this entire amplified sequence as ND1, for simplicity) using primers tmet-frog (5′-TTGGGGTATGGGCCCAAAAGCT-3′; Wiens et al. 2005) and a primer designed for A. callidryas (ACA-Int: 5′-ACGTGATCTGAGTTCAGACCG-3′). Polymerase chain reactions (PCRs) were performed in a total volume of 25 μl, each containing approximately 100 ng template DNA, 1× PCR buffer, 0.75 mM dNTPs, 1.5 mM MgCl2, 1 μM primer, and 0.625 U Taq polymerase (Roche Diagnostics, Switzerland). PCR conditions consisted of an initial 95 °C denaturation for 5 min; followed by 35 amplification cycles of denaturation at 94 °C for 1 min, annealing at 50 °C for 1 min, and extension at 72 °C for 1 min; and a final 5-min extension at 72 °C. We removed unincorporated oligonucleotides and dNTPs with 1 U each of exonuclease I and shrimp alkaline phosphatase with an incubation at 37 °C for 45 min and denaturation at 90 °C for 10 min. We performed cycle sequencing reactions with Big Dye terminator sequencing components according to manufacturer's protocol (Applied Biosystems, Perkin Elmer, Foster City, CA) using the same primers used for fragment amplification. Cycle sequencing reaction conditions were 25 cycles of 96 °C (30 s), 50 °C (15 s), and 60 °C (4 min). We sequenced gene fragments in both directions to resolve any base-calling ambiguities. Products were column purified to remove nonincorporated terminator dye using Sephadex G-50 and electrophoresed on an ABI 3100 Genetic Analyzer (Applied Biosystems, Foster City, CA). Electropherograms were checked by eye and fragments assembled into contiguous sequences using Sequencher 4.1 (GeneCodes, MI).

Table 1

Geographic sampling of phenotypic and mtDNA variation for populations of Agalychnis callidryas

 Region Province, Country Population GIS Color DNA No. Unique Haplotypes h Po π Northeast CR Heredia, CR La Selva 10.4327, −84.0080, 37 20 10 8 0.95 (0.059) 15 0.0034 (0.0021) Alajuela, CR San Ramon 10.2335, −84.5287, 638 14 4 2 0.50 (0.265) 6 0.0026 (0.0020) Guanacaste, CR Tilarán 10.5162, −84.9601, 637 22 6 2 0.73 (0.152) 7 0.0030 (0.0020) Heredia, CR Siquirres 10.0546, −83.551, 574 0 1 — — — — Heredia, CR Universidad de EARTH 10.2368, −83.567, 44 0 6 5 0.93 (0.121) 10 0.0032 (0.0021) Southeast CR/PA Limón, CR Manzanillo 9.6332, −82.6556, 2 26 8 7 0.96 (0.077) 32 0.0074 (0.0043) Limón, CR Cahuita 9.7189, −82.8143, 16 0 3 3 1.00 (0.272) 22 0.0127 (0.0098) Bocas del Toro, PA Almirante 9.1980, −82.3445, 13 10 2 2 1.00 (0.500) 1 0.0008 (0.0012) Bocas del Toro, PA Chiriqui Grande 8.9460, −82.1571, 21 21 6 5 0.93 (0.121) 29 0.0093 (0.0057) Central PA Panamá, PA Gamboa 9.1231, −79.6930, 51 22 6 5 0.93 (0.121) 13 0.0043 (0.0028) Coclé, PA El Copé 8.6299, −80.592, 792 22 7 5 0.85 (0.137) 12 0.0034 (0.0022) Coclé, PA El Valle 8.6299, −80.1159, 866 21 4 3 0.83 (0.222) 4 0.0018 (0.0015) Veraguas, PA Santa Fe 8.5070, −81.1141, 714 17 6 2 0.53 (0.172) 34 0.0157 (0.0094) Southwest CR Puntarenas, CR Sierpe 8.8892, −83.477, 17 19 11 6 0.80 (0.113) 7 0.0017 (0.0012) Puntarenas, CR Campo 8.6909, −83.5013, 35 16 5 4 0.90 (0.161) 5 0.0022 (0.0016) Puntarenas, CR Uvita 9.1235, −83.7011, 26 24 11 9 0.94 (0.065) 82 0.0345 (0.0183) Puntarenas, CR Pavones 8.4204, −83.1069, 37 20 7 4 0.71 (0.180) 4 0.0009 (0.0008) Northwest CR Puntarenas, CR Cabo Blanco 9.5805, −85.1246, 166 18 7 4 0.75 (0.139) 43 0.0097 (0.0056) Puntarenas, CR Carara 9.7256, −84.5313, 385 25 5 2 0.93 (0.121) 10 0.0026 (0.0019) Puntarenas, CR Playa Bandera 9.5188, −84.3774, 23 19 11 8 0.92 (0.066) 14 0.0038 (0.0023)
 Region Province, Country Population GIS Color DNA No. Unique Haplotypes h Po π Northeast CR Heredia, CR La Selva 10.4327, −84.0080, 37 20 10 8 0.95 (0.059) 15 0.0034 (0.0021) Alajuela, CR San Ramon 10.2335, −84.5287, 638 14 4 2 0.50 (0.265) 6 0.0026 (0.0020) Guanacaste, CR Tilarán 10.5162, −84.9601, 637 22 6 2 0.73 (0.152) 7 0.0030 (0.0020) Heredia, CR Siquirres 10.0546, −83.551, 574 0 1 — — — — Heredia, CR Universidad de EARTH 10.2368, −83.567, 44 0 6 5 0.93 (0.121) 10 0.0032 (0.0021) Southeast CR/PA Limón, CR Manzanillo 9.6332, −82.6556, 2 26 8 7 0.96 (0.077) 32 0.0074 (0.0043) Limón, CR Cahuita 9.7189, −82.8143, 16 0 3 3 1.00 (0.272) 22 0.0127 (0.0098) Bocas del Toro, PA Almirante 9.1980, −82.3445, 13 10 2 2 1.00 (0.500) 1 0.0008 (0.0012) Bocas del Toro, PA Chiriqui Grande 8.9460, −82.1571, 21 21 6 5 0.93 (0.121) 29 0.0093 (0.0057) Central PA Panamá, PA Gamboa 9.1231, −79.6930, 51 22 6 5 0.93 (0.121) 13 0.0043 (0.0028) Coclé, PA El Copé 8.6299, −80.592, 792 22 7 5 0.85 (0.137) 12 0.0034 (0.0022) Coclé, PA El Valle 8.6299, −80.1159, 866 21 4 3 0.83 (0.222) 4 0.0018 (0.0015) Veraguas, PA Santa Fe 8.5070, −81.1141, 714 17 6 2 0.53 (0.172) 34 0.0157 (0.0094) Southwest CR Puntarenas, CR Sierpe 8.8892, −83.477, 17 19 11 6 0.80 (0.113) 7 0.0017 (0.0012) Puntarenas, CR Campo 8.6909, −83.5013, 35 16 5 4 0.90 (0.161) 5 0.0022 (0.0016) Puntarenas, CR Uvita 9.1235, −83.7011, 26 24 11 9 0.94 (0.065) 82 0.0345 (0.0183) Puntarenas, CR Pavones 8.4204, −83.1069, 37 20 7 4 0.71 (0.180) 4 0.0009 (0.0008) Northwest CR Puntarenas, CR Cabo Blanco 9.5805, −85.1246, 166 18 7 4 0.75 (0.139) 43 0.0097 (0.0056) Puntarenas, CR Carara 9.7256, −84.5313, 385 25 5 2 0.93 (0.121) 10 0.0026 (0.0019) Puntarenas, CR Playa Bandera 9.5188, −84.3774, 23 19 11 8 0.92 (0.066) 14 0.0038 (0.0023)

Sampling information includes province, country, and population with GIS data (latitude, longitude, elevation [m]). The number of individuals sampled for phenotype (leg and flank color, color) and genetic variation (DNA). A summary of within-population diversity of NADH1 sequences: heterozygosity (h, with standard error SE), number of polymorphic sites (Po), nucleotide diversity (π with SE). Estimates from the population Siquirres were not calculated (—) due to small sample size, GIS, geographic information system.

We aligned ND1 sequences using ClustalW (Thompson et al. 1994) in the MegAlign 6.1.2 program of the Lasergene sequence analysis software (DNASTAR, Inc, Madison, WI). We conducted multiple alignments using the “slow/accurate” option and varied gap costs (4, 8, 10, 15) to identify possible regions of ambiguous homology that might bias phylogenetic analyses (Gatesy et al. 1993). We detected no gaps in our alignment and thus included the entire sequence length in subsequent analyses.

We estimated haplotype diversity (h) and nucleotide diversity (π) (Nei 1987) and the number of unique haplotypes in our sample using Arlequin 3.01 (Schneider et al. 2000). Overall genetic differentiation among regions was estimated using pairwise F statistics (Reynolds et al. 1983) (ϕST), and we compared estimates with a null distribution of no difference between regions to test for significance (α = 0.05) using 10 000 permutations in Arlequin.

We inferred a Bayesian phylogenetic tree using MrBayes v3.0 (Huelsenbeck and Ronquist 2001). The best fitting model of nucleotide substitution for our data was selected based on the Akaike information criterion as implemented in MrModelTest (Nylander 2004). Our Bayesian analyses consisted of 2 independent runs of 4 Markov chains, run for 10 000 000 iterations and with sampling every 1000th iteration. We applied default prior distributions in MrBayes with the exception of the α shape (exponential, mean = 1.0) and branch length parameters (exponential, mean = 0.1). We determined the appropriate number of burn-in samples and tested for stationarity of parameter values using trace plots in the software package TRACER (Rambaut and Drummond 2005). Removal of 10% of the initial samples provided ample burn-in in both analyses.

### Phenotypic Variation

In life, red-eyed tree frogs are green dorsally and have large red eyes and orange or violet front and hind feet. This species is one of few anuran taxa that exhibits relatively low levels of phenotypic variation within populations but high variation among populations (Hoffman and Blouin 2000; Savage 2002). To human observers, variation in leg and flank coloration among populations is obvious and includes hues of red–orange, yellow, blue, and violet.

We imported photographs of each individual into Adobe Photoshop CS2 (Adobe Systems Incorporated) to correct for ambient light intensity and color by reference to the black–white–gray standard (QPcard 101) in the background of every photograph (Stevens et al. 2007). We quantified color as “hue” in the hue, saturation, and brightness realm because previous analyses of leg coloration confirmed that hue accurately represents variation when saturation values are high (McKenna et al. 1999; Robertson JM and Robertson AD 2008). Saturation is an index of the purity of a color; low saturation values correspond to “muddy” colors because they contain a mixture of all 3 primary colors, whereas highly saturated colors contain only 1 or 2 of 3 complementary colors. We used the Color Picker function in Adobe Photoshop CS2 to measure hue and saturation and conducted a homogeneity-of-variance test within populations to compare the variance in hue for individuals exhibiting high (>30%) and low (≤30%) saturation (implemented in JMP ver 7.0; SAS Institute Incorporated, 2006).

We measured leg and flank coloration of 14–26 frogs per population. Dominant leg colors of A. callidryas vary regionally; individuals from some populations are monochromatic (blue), others contain 2 dominant colors (blue and orange), whereas others contain a continuum of hues (e.g., reddish blue through greenish blue). To quantify color, we followed the protocol of Robertson JM and Robertson AD (2008) and imported the color-corrected photographs into ImageJ 1.37v (Rasband 2008) for downstream analyses. We selected the entire posterior surface of the leg in ImageJ to create a frequency histogram of the number of pixels of each hue. We then transformed the ImageJ hue data to the standard measure of hue with a range of 0°–360° and divided the color spectrum into 8 equal bins (each spanning 45°) named according to the central hue for each bin. For example, the standard hue definition of pure red is 0; therefore, the red bin spans 22.5° on each side of 0°. Final hue ranges for the 8 color bins were red (337.6–22.5), orange (22.6–67.5), yellow (67.6–112.5), green (112.6–157.5), light blue (157.6–202.5), dark blue (202.6–247.5), purple (247.6–292.5), and violet (292.5–337.5).

Our measurements of flank color differed from leg color because A. callidryas flanks have a series of disruptive, vertical stripes, precluding measurement of the entire flank region. However, flank coloration is nearly monochromatic, thereby justifying subsampling a representative patch of color between stripes at the midline of each frog. For both flank and leg measurements, we quantified the percent pixels in each of 8 hue bins, the average hue in each bin, and saturation of each selected color patch.

To test for population and regional differences in coloration, we used linear discriminant analysis implemented in JMP 7.0 (SAS 2007) to compare each individual with the group multivariate mean. The following color parameters were included in the model: the average hue (leg and flank) for each of 8 bins, the percentage of hue contained in each bin, and percent saturation. To quantify the number of individuals correctly assigned and those misclassified to source populations, we generated a classification matrix and tested the significance of individual assignments using a chi-square test. This method predicts assignment based on multivariate analysis of variance. Accurate assignment indicates that leg and flank coloration are highly diagnostic for the 5 regions examined.

### Matrix Regression Analyses

We tested the relative strength of multiple factors contributing to the geographic distribution of genetic and phenotypic diversity at 3 spatial scales: all populations combined, Caribbean populations only, and Pacific populations only. For all analyses, we implemented matrix regression analyses to identify the relative significance of multiple predictor variable, using a forward selection, 1–3 step process (Legendre et al. 1994). In step 1, we used individual pairwise Mantel tests to examine the association between the Y matrix (phylogenetic distance or phenotypic distance) and each X matrix (e.g., geographic distance, barrier, and phylogenetic distance for analyses of phenotype). If all X matrices significantly varied with the Y matrix in pairwise Mantel tests, then variables were included in subsequent partial Mantel tests (step 2). The variable with the highest R2 was listed first in each partial Mantel test. This examines the effect of the first X matrix while considering the variation due to the second matrix. For analyses that contained 3 variable X matrices, a third step was included, following the same methodology described for steps 1 and 2. At each step, the slope and probability values indicate the relative strength (and significance) of each variable in predicting genetic and phenotypic diversity patterns (Legendre et al. 1994). All tests were implemented in R ver 1.8.5 (R Core Development Team 2007). For each test, the standardized regression coefficients (std b), R2, and P values were generated by 9999 permutations.

Distance matrices for phenotype (leg and flank coloration) were derived as the Euclidean pairwise distance among individuals, constructed in R. Calculation of Euclidean distance was based on saturation, average hue, and the percentage of hue for each of 8 bins. For the genetic distance matrix, we calculated pairwise patristic distances from the Bayesian consensus tree using the program TreeEdit 1.0a10 (Rambaut and Charleston 2001). We used 2 measures of geographic distance: The first matrix contained the straight-line distance between all pairs of sites (proximitySTRAIGHT), and the second matrix reflected distances of likely dispersal paths based on our knowledge of the physiology and habitat requirements of the species (proximityAROUND). The second matrix accounted for the inhospitable habitats in the high elevation habitats of the Cordillera de Talamanca and the dry Pacific landscape located between southwestern CR and Central PA (Crawford et al. 2007); both these regions do not contain populations of red-eyed tree frogs. To test for the association between genetic and phenotypic diversity patterns with respect to 5 putative biogeographic regions, we created a “barrier matrix” using binary indicator variables (0,1) to designate whether populations occurred in the same (1) or different (0) regions with respect to each barrier. Three populations included in the mtDNA phylogeny (Universidad de EARTH, Siquirres, Cahuita) were excluded from matrix regression analyses because of insufficient data for coloration (Table 1).

## Results

### Population Genetic Variation

The mitochondrial fragment used for analyses was 1149 bp in length, including 118 bp of the 16S gene and 1031 bp of NADH1 (Genbank accession numbers: FJ489259–FJ489334). We sequenced this fragment for 125 A. callidryas and the outgroup taxon Agalychnis saltator and identified a total of 75 unique haplotypes with 178 variable sites, of which 142 were parsimony informative; no insertions/deletions were detected. Haplotype (h) and nucleotide (π) diversity varied among populations with high h (mean ± standard deviation [SD] = 0.846 ± 0.145, range = 50–100%) and relatively low π for most populations (mean ± SD = 0.006 ± 0.007, range = 0.0008–0.00345, Table 1). High haplotypic diversity and low nucleotide diversity generally indicate that populations are divergent and geographically subdivided (Grant and Bowen 1998). Our measures of π are similar to estimates for other sympatric anuran taxa sampled at the NADH2 mtDNA gene (Crawford 2003). We detected exceptionally high π for individuals sampled from Uvita (uvi), an order of magnitude greater than any other population of A. callidryas (Table 1) or other anuran taxa sampled in this geographic region (Crawford 2003). Pairwise ϕST values among regions were high and significant for all comparisons, ranging from 0.327 (northeastern–southeastern CR) to 0.863 (northwestern–northeastern CR, Table 2).

Table 2

Pairwise ϕST for 5 regions, representing 20 populations of Agalychnis callidryas in CR and PA

 Northeast CR Southeast CR Central PA Northwest CR Southwest CR Northeast CR 4.11 Southeast CR 0.327* 13.00 Central PA 0.704* 0.533* 13.01 Northwest CR 0.863* 0.755* 0.754* 6.80 Southwest CR 0.806* 0.737* 0.735* 0.781* 20.00
 Northeast CR Southeast CR Central PA Northwest CR Southwest CR Northeast CR 4.11 Southeast CR 0.327* 13.00 Central PA 0.704* 0.533* 13.01 Northwest CR 0.863* 0.755* 0.754* 6.80 Southwest CR 0.806* 0.737* 0.735* 0.781* 20.00

The diagonal elements (bolded) are the average pairwise nucleotide differences between haplotypes within each region.

*Significance P < 0.0001.

MrModelTest 3.7 selected the model general time reversible + I + Γ with unequal base frequencies (A = 0.3076, C = 0.22190, G = 0.11182, T = 0.35230; pinvar = 0.6311; and α = 2.1712). The Bayesian tree, including all haplotypes, showed an overall pattern of regional differentiation and well-supported regional clades (marginal posterior probabilities ranged from 89–99%); however, haplotypes from none of the regions formed a monophyletic group (Figure 2). The consensus tree showed an early divergence of the clade Pacific A (including only southwestern CR populations) relative to the other 4 regions (Figure 2). Within Pacific A, samples from the Osa Peninsula (Campo [cam]) were genetically distinct from other southwestern CR populations. Haplotypes from the remaining regions fell within 3 clades (Pacific B, Caribbean A, Caribbean B, Central Panama) united at their base by a polytomy (Figure 2). Pacific B included individuals not only from northwestern CR but also 4 from uvi. The other 7 uvi individuals were members of Pacific A (Figure 2). The clade Caribbean A contained individuals from 3 regions (northeastern CR, southeastern CR, northwestern CR), including individuals from both sides of the Cordillera de Talamanca. Caribbean B (sister to Caribbean A) contained individuals from southeastern CR and Central PA. The third major clade included haplotypes exclusively sampled from Central PA (Figure 2).

Figure 2

Bayesian consensus phylogram based on 1149 basepairs of the NADH1 mtDNA gene fragment for 125 red-eyed tree frog individuals. Values above branches are marginal posterior probabilities. Phylogram is rooted with the outgroup taxon, Agalychnis saltator. Bars are coded according to five geographic regions from Figure 1. Population and regions correspond to those listed in Table 1.

Figure 2

Bayesian consensus phylogram based on 1149 basepairs of the NADH1 mtDNA gene fragment for 125 red-eyed tree frog individuals. Values above branches are marginal posterior probabilities. Phylogram is rooted with the outgroup taxon, Agalychnis saltator. Bars are coded according to five geographic regions from Figure 1. Population and regions correspond to those listed in Table 1.

### Phenotypic Variation

Prior to color analyses, we conducted a homogeneity-of-variance test within populations to determine the range of saturation in which hue measurements would be usable. We expected high variance in hue measurements for individuals with “muddy” color patches because hue is determined solely by the colors that most contribute to coloration. For example, a muddy color patch may contain 33.5% red, 31.5% blue, and 35% yellow. In this case, the hue would be described as red/yellow (orange) with low saturation (i.e., high content of blue, the complement of orange). The color patch of another individual that closely resembles that same phenotype may be 34% red, 35% blue, and 31% yellow. This individual would be designated as red/blue (violet). Thus, slight changes in the relative contribution of any of the 3 colors will significantly skew the hue measurements for low saturation coloration. These values are obviously unreliable and should be excluded from hue-based population studies (McKenna et al. 1999).

We validated this expectation by performing a homogeneity-of-variance test on 3 populations. The other populations did not contain a sufficient number of individuals in both the low- and high-saturation category for statistical analyses. A significant result indicates that variance is not homogenous between high- and low-saturation groups. In 3 populations, the low-saturation group exhibited high variance (Almirante: F = 13.98, P = 0.011; Pavones: F = 33.51, P < 0.0001; El Copé: F = 12.24, P = 0.005), and thus, these individuals were excluded from further analyses. Visual inspection of how hue varies with respect to saturation within a population is evident for Gamboa (gam; Figure 3): Individuals that exhibit saturation levels less than 30% have hue measures that range from 0–270, whereas the few individuals with higher saturation levels exhibit a small range in hue (170–190). Based on this result, we subsequently measured and analyzed hue only for individuals with saturation levels >30%. The total number of excluded individuals was small (8.2% of all individuals; average number of individuals per region = 3.8). The percent of excluded individuals per population (mean ± SD = 21 ± 25%, range = 0–72%) varied such that few populations (n = 3) contained many excluded individuals, whereas most contained few.

Figure 3

Flank hue and saturation measures for red-eyed tree frog individuals sampled at 17 sites in CR and PA. Saturation ranges from 0 – 100 % and hue is based on the 0 – 360 color spectrum. The dotted line at 30 % represents the threshold saturation level for excluding individuals in the study of flank coloration.

Figure 3

Flank hue and saturation measures for red-eyed tree frog individuals sampled at 17 sites in CR and PA. Saturation ranges from 0 – 100 % and hue is based on the 0 – 360 color spectrum. The dotted line at 30 % represents the threshold saturation level for excluding individuals in the study of flank coloration.

Divergence in flank coloration was most evident between Pacific (orange) and Caribbean (blue) populations. Pacific populations had mostly orange legs, but a few individuals from the southwestern CR region exhibited some blue and green (Figure 4). In contrast, leg coloration among Caribbean regions varied from blue–violet (northeastern CR) to populations with unequal proportions of blue and orange (Figure 4). The average leg coloration among southeastern CR/PA individuals was approximately 65% orange and 35% blue; in contrast, Central PA individuals had a higher proportion of orange in the legs (ca. 90% orange/10% blue; Figure 5). Overall, these results indicate that leg coloration is more variable than flank coloration. The discriminant function analyses correctly classified 195 of 233 (83.6%) individuals to their region of origin based on leg and flank coloration alone. The number of misclassified individuals per population was very low (mean ± SD = 0.165 ± 0.659, range = 0–5). Therefore, these characters are diagnostic for the 5 regions ($χdf=162=595.45$, P < 0.0001). The largest misclassification occurred between northwestern and southwestern CR (52% of all misclassified individuals), reflecting the high similarity in coloration between these 2 regions (Figure 4).

Figure 4

The proportion of the leg and flank (area measured as percent pixels) that falls within eight color bins (red, orange, yellow, green, light blue, dark blue, purple, violet) for 125 red-eyed tree frogs included in the mtDNA analyses. Individuals are represented on the horizontal axes and the proportional leg and flank color for each individual are represented as a vertical histogram. The color frequency graphs underscore the low within-population variation and high regional variation characteristic of this species. The black, bold lines show the biogeographic barriers associated with flank coloration (Bocas del Toro and Osa Pennisula), leg coloration (Caribbean Valley Complex) and both flank and leg coloration (Cordillera de Talamanca). Five phylogenetic clades (light gray bars; Figure 2) show genetic admixture between neighboring regions and reveal the discordance between spatial patterns of genetic and phenotypic divergence. The clade Caribbean A includes one individual sampled from cab (dashed line).

Figure 4

The proportion of the leg and flank (area measured as percent pixels) that falls within eight color bins (red, orange, yellow, green, light blue, dark blue, purple, violet) for 125 red-eyed tree frogs included in the mtDNA analyses. Individuals are represented on the horizontal axes and the proportional leg and flank color for each individual are represented as a vertical histogram. The color frequency graphs underscore the low within-population variation and high regional variation characteristic of this species. The black, bold lines show the biogeographic barriers associated with flank coloration (Bocas del Toro and Osa Pennisula), leg coloration (Caribbean Valley Complex) and both flank and leg coloration (Cordillera de Talamanca). Five phylogenetic clades (light gray bars; Figure 2) show genetic admixture between neighboring regions and reveal the discordance between spatial patterns of genetic and phenotypic divergence. The clade Caribbean A includes one individual sampled from cab (dashed line).

Figure 5

Variation in the average (and standard error) percentage of leg (top) and flank (bottom) coloration measured as ‘orange’ (combined orange and red color bins) for 17 populations of Agalychnis callidryas. The greatest divergence in coloration occurred across the Cordillera de Talamanca.

Figure 5

Variation in the average (and standard error) percentage of leg (top) and flank (bottom) coloration measured as ‘orange’ (combined orange and red color bins) for 17 populations of Agalychnis callidryas. The greatest divergence in coloration occurred across the Cordillera de Talamanca.

### Matrix Regression Analyses

#### All Sampled Populations

Genetic diversity was largely structured by geographic regional barriers and by geographic distance (Table 3). At the broadest spatial scale, the geographic barrier between Pacific and Caribbean (Cordillera de Talamanca) was the largest determinant of genetic distance (std b = 0.435, P = 0.0001, Table 3). Accordingly, we also detected an association between genetic and geographic distance around the Cordillera de Talamanca (proximityAROUND) but not across the mountain range (proximitySTRAIGHT, Table 3).

Table 3

Forward selection matrix multiple regression of factors that predict the spatial variation in genetic diversity (A), leg coloration (B), and flank coloration (C) at 3 spatial scales: all populations, Caribbean populations, Pacific populations

 Region X matrix Step 1 Step 2 Step 3 std b P std b P std b P Interpretation Determinants of genetic diversity All proximityAROUND 0.398 0.0001 0.168 0.0002 Geographic distance around TAL proximitySTRAIGHT 0.142 0.0009 0.019 0.1844 Barrier (TAL) 0.533 0.0001 0.435> 0.0001 Barrier: Cordillera de Talamanca Caribbean KM 0.921 0.0001 0.878> 0.0001 0.757 0.0001 Geographic distance Barrier (CVC) 0.656 0.0001 0.058 0.0001 0.062 0.0001 Barrier: Limon Barrier (BDT) 0.762 0.0001 0.165 0.0001 0.167 0.0001 Barrier: Bocas del Toro Pacific KM 0.483 0.0001 0.035 0.0247 0.044 0.0087 Geographic distance Barrier (RIO) 0.717 0.0001 0.698> 0.0001 0.669 0.0001 Barrier: Rio Parrita Barrier (OSA) 0.249 0.0001 0.173 0.0001 0.174 0.0001 Barrier: Osa Peninsula Determinants of leg color All DNA 0.121 0.0003 −0.053 0.9904 KM 0.139 0.0002 −0.028 0.8231 Barrier (TAL) 0.303 0.0001 0.317> 0.0001 Barrier: Cordillera de Talamanca Caribbean DNA 0.259 0.0001 0.109 0.0001 −0.003 0.5930 KM 0.676 0.0001 0.307 0.0001 0.248 0.0001 Geographic distance Barrier (CVC) 0.756 0.0001 0.734> 0.0001 0.561 0.0001 Barrier: Caribbean Valley Complex Barrier (BDT) 0.515 0.0001 0.215 0.0001 0.072 0.0004 Barrier: Bocas del Toro Pacific DNA −0.005 0.4990 KM 0.075 0.0065 Geographic distance Barrier (RIO) −0.0110 0.6466 Barrier (OSA) −0.0448 0.9725 Determinants of flank color ALL DNA 0.3500 0.0001 0.029 0.0069 0.026 0.0133 DNA KM 0.347 0.0001 0.025 0.0142 0.021 0.0386 Geographic distance Barrier (TAL) 0.623 0.0001> 0.607 0.0001 0.598 0.0001 Barrier: Cordillera de Talamanca Caribbean DNA 0.099 0.0034 −0.025 0.2135 KM 0.076 0.0091 −0.071 0.9638 Barrier (CVC) −0.018 0.8031 Barrier (BDT) 0.140 0.0006 −0.125> 0.0025 Barrier: Bocas del Toro Pacific DNA 0.161 0.0010 0.078 0.0130 0.053 0.0560 DNA KM 0.089 0.0454 0.075 0.0510 0.050 0.1513 Barrier (RIO) 0.105 0.0120 0.068 0.0296 0.880 0.1656 Barrier (OSA) 0.346 0.0001> 0.326 0.0001 0.331 0.0001 Barrier: Osa Peninsula
 Region X matrix Step 1 Step 2 Step 3 std b P std b P std b P Interpretation Determinants of genetic diversity All proximityAROUND 0.398 0.0001 0.168 0.0002 Geographic distance around TAL proximitySTRAIGHT 0.142 0.0009 0.019 0.1844 Barrier (TAL) 0.533 0.0001 0.435> 0.0001 Barrier: Cordillera de Talamanca Caribbean KM 0.921 0.0001 0.878> 0.0001 0.757 0.0001 Geographic distance Barrier (CVC) 0.656 0.0001 0.058 0.0001 0.062 0.0001 Barrier: Limon Barrier (BDT) 0.762 0.0001 0.165 0.0001 0.167 0.0001 Barrier: Bocas del Toro Pacific KM 0.483 0.0001 0.035 0.0247 0.044 0.0087 Geographic distance Barrier (RIO) 0.717 0.0001 0.698> 0.0001 0.669 0.0001 Barrier: Rio Parrita Barrier (OSA) 0.249 0.0001 0.173 0.0001 0.174 0.0001 Barrier: Osa Peninsula Determinants of leg color All DNA 0.121 0.0003 −0.053 0.9904 KM 0.139 0.0002 −0.028 0.8231 Barrier (TAL) 0.303 0.0001 0.317> 0.0001 Barrier: Cordillera de Talamanca Caribbean DNA 0.259 0.0001 0.109 0.0001 −0.003 0.5930 KM 0.676 0.0001 0.307 0.0001 0.248 0.0001 Geographic distance Barrier (CVC) 0.756 0.0001 0.734> 0.0001 0.561 0.0001 Barrier: Caribbean Valley Complex Barrier (BDT) 0.515 0.0001 0.215 0.0001 0.072 0.0004 Barrier: Bocas del Toro Pacific DNA −0.005 0.4990 KM 0.075 0.0065 Geographic distance Barrier (RIO) −0.0110 0.6466 Barrier (OSA) −0.0448 0.9725 Determinants of flank color ALL DNA 0.3500 0.0001 0.029 0.0069 0.026 0.0133 DNA KM 0.347 0.0001 0.025 0.0142 0.021 0.0386 Geographic distance Barrier (TAL) 0.623 0.0001> 0.607 0.0001 0.598 0.0001 Barrier: Cordillera de Talamanca Caribbean DNA 0.099 0.0034 −0.025 0.2135 KM 0.076 0.0091 −0.071 0.9638 Barrier (CVC) −0.018 0.8031 Barrier (BDT) 0.140 0.0006 −0.125> 0.0025 Barrier: Bocas del Toro Pacific DNA 0.161 0.0010 0.078 0.0130 0.053 0.0560 DNA KM 0.089 0.0454 0.075 0.0510 0.050 0.1513 Barrier (RIO) 0.105 0.0120 0.068 0.0296 0.880 0.1656 Barrier (OSA) 0.346 0.0001> 0.326 0.0001 0.331 0.0001 Barrier: Osa Peninsula

Values reported are the standardized regression coefficients (std b) for each Mantel test with P values determined by 9999 random permutations. For step 2, > = variable with highest R2 in step 1. See text for details of forward selection methodology. KM = distance in kilometers, DNA = patristic distance based on consensus Bayesian phylogeny. Barriers examined include Cordillera de Talamanca (TAL), Caribbean Valley Complex (CVC), Bocas del Toro (BDT), Río Parrita (RIO), Osa Peninsula (OSA). In the case where multiple variables contributed to the model, the variable most strongly associated with the Y matrix is in bold face.

Leg coloration across all populations was strongly associated with a division between Pacific and Caribbean populations (std b = 0.317, P = 0.0001, Table 3). However, neither geographic nor genetic distance determined leg coloration at this broadest spatial scale (Table 3). Conversely, the distribution of flank coloration was associated with all factors examined (genetic distance, geographic distance, Cordillera de Talamanca) with the largest effect being the barrier dividing populations into Pacific and Caribbean versants (std b = 0.598, P = 0.0001, Table 3). At the broadest spatial scale, leg coloration was associated with the distribution of flank coloration (std b = 0.283, P = 0.0001).

#### Caribbean Populations Only

Genetic diversification of Caribbean populations was strongly associated with geographic distance and weakly (but significantly) associated with the 2 putative geographic barriers, the Caribbean Valley Complex (between northeastern CR and southeastern CR) and Bocas del Toro (between southeastern CR and PA, Table 3). Leg coloration was associated with geographic distance (std b = 0.248, P = 0.0001, Table 3) and both barriers but not with genetic distance (std b = −0.003, P = 0.593, Table 3). In contrast, only the Bocas del Toro barrier significantly structured flank coloration (std b = −0.125, P = 0.0025, Table 3). Flank and leg coloration were weakly associated across all Caribbean populations (std b = 0.069, P = 0.0143).

#### Pacific Populations Only

Geographic factors also structured genetic diversity on the Pacific versant. The barrier Río Parrita (dividing northwestern and southwestern CR) was the strongest predictor (std b = −0.669, P = 0.0001, Table 3). Leg coloration did not vary with respect to either geographic barrier but was weakly associated with geographic distance (std b = −0.075, P = 0.0065, Table 3). Flank coloration was strongly associated with whether populations were located on the Osa Peninsula or mainland (std b = −0.331, P = 0.0001, Table 3), weakly associated with genetic diversity (std b = −0.0130, P = 0.0001, Table 3), and did not vary with geographic distance (Table 3). Flank and leg coloration were associated across all Pacific populations (std b = 0.142, P = 0.0046).

## Discussion

Our objectives were to quantify genetic and phenotypic diversity in red-eyed tree frogs and examine the extent to which their spatial distribution might be concordant at multiple geographic scales and across putative biogeographic barriers. Although flank and leg coloration were associated at all spatial scales of this study (albeit only weakly for Caribbean and Pacific populations analyzed separately), leg coloration was more variable than flank coloration. In addition, the geographic and genetic determinants of diversity were not equal for these 2 traits, indicating that different evolutionary mechanisms mediate the spatial distribution of composite body coloration. At some scales, variation in leg and flank coloration was associated with mtDNA diversity, implicating the underlying roles of historical and geological factors in population structure. However, we detected multiple departures from this pattern, demonstrating that limitations to historical connectivity alone cannot fully explain regional diversification in coloration. A direct test for selection on color was not the primary objective of our study; nonetheless, our analyses reveal contexts in which there is opportunity for selection. We discuss the interplay among evolutionary and geographic processes with respect to the 3 general patterns of diversity detected in this study: codistribution of genetic and phenotypic diversity across historical barriers, phenotypic similarity across genetically isolated groups, and phenotypic differentiation in the presence of historical connectivity.

### Historical Barriers and Codistributed Patterns of Phenotypic and Genetic Diversity

The codistribution of genetic and phenotypic diversity patterns across geographic barriers indicates that similar processes have shaped aspects of population diversity and that geographic isolation has likely facilitated differentiation. Genetic diversification in A. callidryas populations was associated with levels of phenotypic differentiation with respect to 2 biogeographic barriers: the Cordillera de Talamanca and Bocas del Toro.

Matrix regression analyses, pairwise ϕST values, and the multivariate discriminant function analyses based on coloration all indicate isolation of Caribbean and Pacific populations with respect to the Cordillera de Talamanca. However, our phylogenetic analyses revealed a single Pacific individual with a haplotype nested within the clade Caribbean A (cab1; Figure 2), indicating either incomplete lineage sorting of mtDNA haplotypes or that historical gene flow connected Pacific and Caribbean populations. The observation that flank-stripe patterns were not clearly distinguishable between northwestern and northeastern CR populations, a trait that otherwise exhibited regional variation, led Savage and Heyer (1967) to hypothesize that mountain passes at these elevations (ca. 900 m) may have historically facilitated migration through dispersal corridors under more lenient climatic and habitat conditions. Three volcanic mountain ranges isolated from the northern edge of the Cordillera de Talamanca (Cordillera de Tilarán, Cordillera Central, and Cordillera Guanacaste) are younger and lower in elevation than the peak height of the Cordillera de Talamanca, possibly facilitating dispersal across the Continental Divide, as observed in other anuran taxa (Wang et al. 2008). Our previous analyses of population genetic structure using microsatellite markers revealed that gene flow on more recent time scales was highly restricted across the Continental Divide (Robertson et al. 2009), indicating that phenotypic divergence has likely resulted from divergence in geographic isolation. Increased population sampling could elucidate the genetic nature of a potential contact zone across the Continental Divide.

The greatest divergence in coloration was detected between northwestern and northeastern CR despite close geographic proximity (56 km) and relatively recent time since divergence (Figure 2). Leg and flank color differentiation was greater between these populations than between 2 of the most geographically distant sites in our study, Gamboa “gam” in Central PA and Pavones (“pav” in Southwest CR), populations that are 800 km apart when considering the distance around the Cordillera de Talamanca (Figure 5). Combined, data from this mtDNA study and prior findings of restricted nuclear gene flow (Robertson et al. 2009) indicate that phenotypic diversity on either side of the Continental Divide is likely due to genetic drift and continued geographic isolation over evolutionary time scales.

Historical limitations to gene flow between populations sampled from southeastern CR and Central PA (either side of Bocas del Toro) contributed to the spatial diversification of flank and leg coloration (Figure 4). However, the effect of this putative barrier was small in matrix regression analyses (compared with the effect of geographic distance) and is consistent with the relatively subtle differences in gene and color differentiation compared with the divergence observed across the Cordillera de Talamanca. Thus, the Bocas del Toro barrier could be more permeable to historical gene flow, as evidenced by mtDNA admixture of southeastern CR and central PA populations and low to moderate levels of phenotypic differentiation. Multiple colonization events of montane populations in this region have been documented for other taxa (Crawford 2003; Zeh et al. 2003; Weigt et al. 2005), and the permeable nature of this barrier is consistent with findings for other anurans (Crawford et al. 2007). The admixture revealed for A. callidryas populations could be due to incomplete lineage sorting reflecting multiple vicariant events in the Bocas del Toro region. Alternatively, the geographic proximity of these admixed populations could arise from secondary contact of previously isolated populations. Recent gene flow estimates, based on nuclear microsatellite loci, revealed substantial gene flow and admixture of genetic demes across this region (Robertson et al. 2009) suggesting that our phylogenetic analyses tracked historical processes of temporal isolation but that more recent gene flow connects populations spanning this region.

### Phenotypic Uniformity in Structured Populations

We detected limited phenotypic variation among individuals sampled from genetically differentiated populations along the Pacific versant. The striking similarity among most individuals could be explained by evolutionary stasis (Rutherford 2000; Wellborn and Broughton 2008) or stabilizing selection, although studies are required to differentiate between these mechanisms. The genetic structure of Pacific populations into 2 clades (northwestern and southwestern CR; Figure 2) was consistent with a priori expectations of historical isolation due to plate tectonic activity in CR (Kohlmann et al. 2002). Genetic isolation could be due to these historical geological factors, temporal differences in rainfall and reproductive activity (Grinnell 1914, 1924), and/or dispersal limitations imposed by one of several large riverine barriers that drain from the Cordillera de Talamanca into the Pacific Ocean. Independent of the exact nature of the barrier, the matrix regression analyses and phylogenetic analyses corroborate a geographic division between northern and southern populations (ϕST = 0.781), with a potential contact zone centered between Playa Bandera and uvi. Our estimates of π were exceptionally high for individuals sampled at uvi (Table 1), possibly indicating that this population has a long evolutionary history and experienced secondary contact with divergent lineages (Grant and Bowen 1998). Alternatively, admixed clades in mitochondrial gene genealogies can result from incomplete lineage sorting or historical connectivity (Wright 1937; Slatkin 1985).

A second Pacific barrier indicated population isolation on the Osa Peninsula relative to the mainland: The phylogenetic tree and matrix regression analyses showed that the single population from the Osa Peninsula (cam) forms a deeply divergent monophyletic clade within a larger southwestern CR clade. Genetic isolation of Osa Peninsula populations has also been detected for other vertebrates, including frogs (Crawford 2003; Crawford et al. 2007) and snakes (Zamudio and Greene 1997) suggesting that Osa Peninsula populations have a history of genetic isolation. Despite this historical isolation, we found little evidence of phenotypic divergence in leg coloration, thus indicating evolutionary stasis for a single phenotype in the Osa Peninsula and adjacent mainland populations or that contemporary gene flow is sufficiently high to result in phenotypic homogeneity of Pacific populations. Estimates of recent gene flow based on nuclear markers show little genetic differentiation between cam and other southwestern CR populations suggesting the later for leg coloration (Robertson et al. 2009). These results favor the hypothesis that more recent gene flow explains uniformity of leg color, but not flank color, between mainland and Osa Peninsula populations.

### Phenotypic Divergence in the Presence of Gene Flow

Phenotypic divergence of red-eyed tree frog populations could not be explained by historical isolation of regions that occupy lowland Caribbean forest. Leg coloration varied among populations, ranging from entirely blue legs in the north to mixed blue/orange coloration in the south that did not coincide with genetic diversification of these same populations (Figure 4). Phylogenetic analyses revealed historical cohesion between 2 phenotypically divergent regions, northeastern and southeastern CR (Figure 2), indicating that color differentiation could be mediated by strong localized selection sufficient to counteract the homogenizing effects of historical connectivity. More recent gene flow estimates revealed that northeastern and southeastern CR populations were not isolated but exhibited some genetic admixture (Robertson et al. 2009). Our data thus indicate that a geographic break structures phenotype but not genotype for Caribbean populations. The nature of this break is not well understood for terrestrial taxa, but it does coincide with the northern/southern edges of geographic ranges in other Central American beetle taxa (Kohlmann et al. 2002); delimits color morphs of the lowland frog, Oophaga (=Dendrobates) pumilio (Hagemann and Pröhl 2007); and appears to structure red-eyed tree frog populations, as well. Combined, mitochondrial and nuclear markers (Robertson et al. 2009) provide evidence that color pattern differentiation cannot be solely explained by geographic isolation, indicating the probable role of diversifying selection in the maintenance of color variation. Alternatively, rapid colonization of the Caribbean lowland forest, followed by intermittent geographic barriers to gene flow, could result in a signature of modest gene flow with the accumulation of phenotypic differentiation (Reeves and Bermingham 2006). Under either scenario, differences could be exaggerated by strong selection against maladapted phenotypes (Nosil and Crespi 2004; Nosil et al. 2005; Rosenblum et al. 2007).

### Mode of Diversification

Our data indicate that the high degree of phenotypic regionalization in the red-eyed tree frog reflects a number of evolutionary processes and that these processes vary across spatial scales in this study. The distribution of genetic or phenotypic diversity often correlates with landscape history (Pröhl et al. 2006) and microhabitat differences (Thorpe and Baez 1993; García-París et al. 2000). Although microhabitat features of A. callidryas populations have yet to be rigorously quantified, factors that could mediate strong localized selection are not obvious. For example, sampling sites varied in habitat type, elevation, or percent canopy cover within but not among geographic regions (Table 1; Jeanne Robertson, personal observation). Divergent color morphs raised in captivity maintain their respective coloration despite identical feeding regimes (Ivan Gomez-Mestre, personal observation); thus, although regional diet has not been quantified for this species, it is unlikely that differences in prey base among regions determine coloration for this species. The one environmental factor that we know to vary regionally and may contribute to the patterns of genetic and phenotypic distribution is rainfall. Many anurans, including A. callidryas, rely on rainfall for reproduction. Because rainfall patterns vary across regions in CR and PA (Holdridge 1947; Kohlmann et al. 2002), these differences might reinforce spatial isolation due to asynchronous reproduction. In addition, subtle regional differences in abiotic features, selection regimes (sexual selection or predator pressures), or both could potentially mediate spatial patterns of phenotypic divergence (Brooks and Endler 2001b; Price 2006; Gosden and Svensson 2008).

Disentangling the selective environment controlling the evolution of color is difficult because several different processes could underlie divergent phenotypic expression, including natural (Hairston 1979; Endler 1980) and sexual selection (Endler 1983; West-Eberhard 1983; Panhuis et al. 2001; Masta and Maddison 2002; Maan et al. 2004). Natural selection for habitat background matching is the most common form of selection documented to date, including for rodents (Hoekstra et al. 2004; Hoekstra et al. 2005), lizards (Thorpe and Baez 1993; Thorpe 2002; Rosenblum 2006; Rosenblum et al. 2007; Stuart-Fox et al. 2007), frogs (Pyburn 1961; Nevo 1973; Stewart 1974; Hoffman and Blouin 2000; Woolbright and Stewart 2008), insects (Kettlewell and Conn 1977; Sandoval and Nosil 2005; Nosil et al. 2006), and snakes (King and Lawson 1995). Background matching is unlikely to mediate leg and flank color divergence in A. callidryas because there are no observable sharp environmental gradients that coincide with breaks among regions. In addition, the flank and leg color patches measured in this study are not cryptic. At rest, red-eyed tree frogs lie still with their brightly colored flanks and limbs tucked underneath their body; in this position, only the green dorsum is visible and perfectly matches the leaves they sit on (Schwalm et al. 1977; Emerson et al. 1990). In contrast, whereas active at dusk and throughout the night they sit upright and expose their colorful limbs and flanks. The conspicuous nature of this display suggests that color pattern functions as a visual signal to conspecifics and/or predators at night when this species is active.

Natural selection contributes to the evolution of conspicuous coloration in other anurans (Summers and Clough 2001; Siddiqi et al. 2004; Reynolds and Fitzpatrick 2007; Rudh et al. 2007). Bright, aposematic coloration is an effective signal to deter predators and is observed for many, usually toxic, dendrobatid species (Summers and Clough 2001; Siddiqi et al. 2004). Natural selection for conspicuous coloration could evolve through predator interactions if the contrasting coloration of A. callidryas functions to warn predators of the noxious skin peptides common in this and other phyllomedusine frogs (Sazima 1974; Mignogna et al. 1997; Conlon et al. 2007). Geographic variation in skin peptide composition covaries with differences in coloration in the Australian tree frog, Litoria rubella (Steinborner et al. 1996); therefore, color differences could reflect geographic regionalization in skin peptide profiles in A. callidryas as well.

Sexual selection can rapidly promote population divergence through female mate choice (West-Eberhard 1983; Masta and Maddison 2002; Maan et al. 2004; Siddiqi et al. 2004; Summers et al. 2004, 2003). Female preferences may vary under different environmental conditions, which in turn will promote population divergence (Maan et al. 2004; Maan, Hopfker et al. 2006; Gray and McKinnon 2007). For the polytypic frog, Dendrobates (=Oophaga) pumilio, mate choice experiments demonstrated that sexual selection mediates intraspecific population divergence in coloration (Summers et al. 1999; Reynolds and Fitzpatrick 2007). More recently, mate preference experiments with O. pumilio found that female choice is based on aposematic coloration, indicating the potential for colorful visual signals to evolve through an interaction of natural and sexual selection (Maan and Cummings 2008). If dendrobatid frogs serve as a generalized model of color evolution in aposematic frogs, then we would predict that bright, contrasting coloration evolved as a response to natural predators in A. callidryas and that sexual selection through female choice also mediated divergence among populations. For A. callidryas populations, divergent hues must be sufficiently perceived and distinguishable when this crepuscular species is active for coloration to serve a social function and evolve through sexual selection (Hailman and Jaeger 1974; Lythgoe and Patridge 1991; Endler 1992). Visual signaling functions for conspecific communication in many nocturnal anurans, indicating that social cues are transmitted and received, even in low light conditions (McDiarmid and Adler 1974; Buchanan 1994; Haddad and Giaretta 1999; Hartmann et al. 2005; Cummings et al. 2008). Sexual selection provides the most compelling initial hypothesis of selection to examine with future behavioral studies.

The potential combined effects of genetic drift and geographic isolation on differentiation among populations is very high (Slatkin 1985; Orr 1998; Gavrilets 2003; Hoffman et al. 2006). Previous studies have shown that phenotypic variation in behavior (Castellano et al. 2002; Phillips and Johnston 2008), color pattern (Bittner and King 2003; Hoffman et al. 2006; Ohmer et al. 2009), and body size (Knopp et al. 2007) can evolve, in part, through nonselective regimes, and these processes possibly contribute to phenotypic diversification among populations of red-eyed tree frogs. Regional fixation from an ancestral, polymorphic state, or linkage to other traits evolving through selection, could also account for the patterns we see in A. callidryas. However, the high levels of within-population haplotypic diversity (Table 1) and substantial variation at microsatellite loci (Robertson et al. 2009) indicate that genetic drift has not been the primary process shaping divergence among these populations.

### Color Evolution in Red-Eyed Tree Frogs

The genetic control of coloration is poorly understood in most anurans. However, our analyses lend support to the hypothesis that regional variation in A. callidryas is due to differential fixation of preexisting variation rather than the evolution of novel coloration. Our phylogenetic analyses revealed that Southwestern CR diverged first and is sister to all other regions. This clade contained individuals with predominantly orange legs and orange flanks; however, the 2 southernmost populations in this region were the most variable of all Pacific populations (Figure 5). The presence (albeit limited) of blue coloration in these 2 populations informs us that the full range of coloration occurred deep in the phylogeny and is present across most populations, including populations that are historically isolated from other populations.

Geographic barriers have clearly contributed to isolation of populations or groups of populations in the red-eyed tree frog, and the long-term nature of some of these barriers resulted in genetic and phenotypic divergence. However, not all barriers are absolute, and in some cases, populations will evolve under selection-gene flow equilibrium. Our phylogenetic analyses showed admixture among most neighboring regions, corroborating that both geographic barriers and localized selection contribute to color divergence among A. callidryas populations. We found that the relative roles of selection and gene flow likely differ among biogeographic regions: Genetic isolation and divergent coloration appear to be strongest across the Continental Divide, moderate among Caribbean populations, and weakest among Pacific populations. Our study underscores the fact that selection gradients vary across relatively small spatial scales, even in species that occupy a relatively wide range of environments or habitats.

## Funding

National Science Foundation (NSF) (Doctoral Dissertation Enhancement Grant DEB-0506043); Organization for Tropical Studies; Smithsonian Tropical Research Institute; Tinker Foundation; Einaudi Center; Department of Ecology and Evolutionary Biology at Cornell University (to J.R.); NSF (Population Biology Grant DEB-0542848 to K.Z.).

We thank the staff of the Smithsonian Tropical Research Institute, especially M. Leone and O. Arosemena and the staff of the Organization for Tropical Studies for logistical support during this project. The Ministry of the Environment (Autoridad Nacional del Ambiente) in PA and the Ministry of the Environment and Energy in CR-granted research, collecting, and export permits. We thank La Selva Biological Station, Universidad de EARTH, and the Cabo Blanco Nature Preserve for facilitating our research. C. Dardia (CUMV) and G. Chaves (Museo de Zoología, Universidad de Costa Rica, San Jose, CA) assisted with curation of voucher specimens. We are especially grateful to L. Davis, J. Touchon, A. Vega, C. Vriesendorp, and M. Williams for field assistance; L. Harmon, J. Richmond, A. Robertson, G. Hooker, and F. Vermeylen of the Cornell Statistical Consulting Unit, for help with analyses; and K. Jenkins and J. Sigala for the map. This work benefited from resources of the Computational Biology Service Unit at Cornell University, a facility partially funded by Microsoft Corporation. We thank F. Bolaños, B. Kohlmann, L. Chan, A. Dhondt, K. Galbreath, H. Greene, A. Vega, C. Vriesendorp, B. LeGender, members of the Zamudio Laboratory and 2 anonymous reviewers for discussions about biogeography, selection, and color and for improvements to earlier versions of the manuscript.

## References

Bittner
TD
King
RB
Gene flow and melanism in garter snakes revisited: a comparison of molecular markers and island vs. coalescent models
Biol J Linn Soc
,
2003
, vol.
79
(pg.
389
-
399
)
Boul
KE
Funk
WC
Darst
CR
Cannatella
DC
Ryan
MJ
Sexual selection drives speciation in an Amazonian frog
Proc R Soc Lond Ser B Biol Sci
,
2007
, vol.
274
(pg.
399
-
406
)
Brooks
R
Endler
JA
Direct and indirect sexual selection and quantitative genetics of male traits in guppies (Poecilia reticulata)
Evolution
,
2001
, vol.
55
(pg.
1002
-
1015
)
Brooks
R
Endler
JA
Female guppies agree to differ: phenotypic and genetic variation in mate-choice behavior and the consequences for sexual selection
Evolution
,
2001
, vol.
55
(pg.
1644
-
1655
)
Brown
JH
Stevens
GC
Kaufman
DM
The geographic range: size, shape, boundaries, and internal structure
Annu Rev Ecol Syst
,
1996
, vol.
27
(pg.
597
-
623
)
Brown
RP
Thorpe
RS
Within-island microgeographic variation in body dimensions and scalation of the skink Chalcides sexlineatus, with testing of causal hypotheses
Biol J Linn Soc
,
1991
, vol.
44
(pg.
47
-
64
)
Buchanan
BW
Sexual dimorphism in Hyla squirella, chromatic and pattern variation between the sexes
Copeia
,
1994
(pg.
797
-
802
)
Castellano
S
Tontini
L
Giacoma
C
Lattes
A
Balletto
E
The evolution of release and advertisement calls in green toads (Bufo viridis complex)
Biol J Linn Soc
,
2002
, vol.
77
(pg.
379
-
391
)
Coates
AC
Obando
JA
Jackson
BC
Budd
AF
Coates
AG
The geological evaluation of the Central American Isthmus
Evolution and environment in tropical America
,
1996
Chicago
University of Chicago Press
(pg.
21
-
56
)
Conlon
JM
Woodhams
DC
Raza
H
Coquet
L
Leprince
J
Jouenne
T
Vaudry
H
Rollins-Smith
LA
Peptides with differential cytolytic activity from skin secretions of the lemur leaf frog Hylomantislemur (Hylidae: Phyllomedusinae)
Toxicon
,
2007
, vol.
50
(pg.
498
-
506
)
Crawford
AJ
Huge populations and old species of Costa Rican and Panamanian dirt frogs inferred from mitochondrial and nuclear gene sequences
Mol Ecol
,
2003
, vol.
12
(pg.
2525
-
2540
)
Crawford
AJ
Bermingham
E
Polania
C
The role of tropical dry forest as a long-term barrier to dispersal: a comparative phylogeographical analysis of dry forest tolerant and intolerant frogs
Mol Ecol
,
2007
, vol.
16
(pg.
4789
-
4807
)
Cummings
ME
Bernal
XE
Reynaga
R
Rand
AS
Ryan
MJ
Visual sensitivity to a conspicuous male cue varies by reproductive state in Physalaemus pustulosus females
J Exp Biol
,
2008
, vol.
211
(pg.
1203
-
1210
)
Dallimer
M
Jones
PJ
Pemberton
JM
Cheke
RA
Lack of genetic and plumage differentiation in the red-billed quelea Quelea quelea across a migratory divide in southern Africa
Mol Ecol
,
2003
, vol.
12
(pg.
345
-
353
)
Dhondt
AA
F
Matthysen
E
Kempenaers
B
Nonadaptive clutch sizes in tits
Nature
,
1990
, vol.
348
(pg.
723
-
725
)
Duellman
WE
The hylid frogs of Middle America
,
2001
Ithaca (NY)
Society for the Study of Amphibians and Reptiles
Emerson
SB
Cooper
TA
Ehleringer
JR
Convergence in reflectance spectra among treefrogs
Funct Ecol
,
1990
, vol.
4
(pg.
47
-
51
)
Endler
JA
Gene flow and population differentiation
Science
,
1973
, vol.
179
(pg.
243
-
250
)
Endler
JA
Natural selection on color patterns in Poecilia reticulata
Evolution
,
1980
, vol.
34
(pg.
76
-
91
)
Endler
JA
Natural and sexual selection on color patterns in poeciliid fishes
Environ Biol Fishes
,
1983
, vol.
9
(pg.
173
-
190
)
Endler
JA
Signals, signal conditions, and the direction of evolution
Am Nat
,
1992
, vol.
139
(pg.
S125
-
S153
)
Ford
EB
Ecological genetics
,
1971
London
Chapman & Hall
Fuller
RC
Carleton
KL
JM
TC
Travis
J
Population variation in opsin expression in the bluefin killifish, Lucania goodei: a real-time PCR study
J Comp Physiol A Neuroethol Sens Neural Behav Physiol
,
2004
, vol.
190
(pg.
147
-
154
)
García-París
M
Good
DA
Parra-Olea
G
Wake
DB
Biodiversity of Costa Rican salamanders: implications of high levels of genetic differentiation and phylogeographic structure for species formation
Proc Natl Acad Sci USA
,
2000
, vol.
97
(pg.
1640
-
1647
)
Gatesy
J
DeSalle
R
Wheeler
W
Alignment-ambiguous nucleotide sites and the exclusion of systematic data
Mol Phylogenet Evol
,
1993
, vol.
2
(pg.
152
-
157
)
Gavrilets
S
Perspective: models of speciation: what have we learned in 40 years?
Evolution
,
2003
, vol.
57
(pg.
2197
-
2215
)
Gosden
TP
Svensson
EI
Spatial and temporal dynamics in a sexual selection mosaic
Evolution
,
2008
, vol.
62
(pg.
845
-
856
)
Grant
WS
Bowen
BW
Shallow population histories in deep evolutionary lineages of marine fishes: insights from sardines and anchovies and lessons for conservation
J Hered
,
1998
, vol.
89
(pg.
415
-
426
)
Gray
RH
Seasonal, annual and geographic variation in color morph frequencies of the Cricket Frog, Acris crepitans, in Illinois
Copeia
,
1983
, vol.
(2)
(pg.
300
-
311
)
Gray
SM
McKinnon
JS
Linking color polymorphism maintenance and speciation
Trends Ecol Evol
,
2007
, vol.
22
(pg.
71
-
79
)
Grinnell
J
Barriers to distribution as regards birds and mammals
Am Nat
,
1914
, vol.
48
(pg.
248
-
254
)
Grinnell
J
Geography and evolution
Ecology
,
1924
, vol.
5
(pg.
225
-
229
)
CFB
Giaretta
AA
Visual and acoustic communication in the Brazilian torrent frog, Hylodes asper (Anura: Leptodactylidae)
Herpetologica
,
1999
, vol.
55
(pg.
324
-
333
)
Hagemann
S
Pröhl
H
Mitochondrial paraphyly in a polymorphic poison frog species (Dendrobatidae; D. pumilio)
Mol Phylogenet Evol
,
2007
, vol.
45
(pg.
740
-
747
)
Hailman
JP
Jaeger
RG
Phototactic responses to spectrally dominant stimuli and use of color vision by adult anuran amphibians—comparative survey
Anim Behav
,
1974
, vol.
22
(pg.
757
-
795
)
Hairston
NG
Adaptive significance of color polymorphism in 2 Species of Diaptomus (Copepoda)
Limnol Oceanogr
,
1979
, vol.
24
(pg.
15
-
37
)
Hartmann
MT
Giasson
LOM
Hartmann
PA
CFB
Visual communication in Brazilian species of anurans from the Atlantic forest
J Nat Hist
,
2005
, vol.
39
(pg.
1675
-
1685
)
Hoekstra
HE
Drumm
KE
Nachman
MW
Ecological genetics of adaptive color polymorphism in pocket mice: geographic variation in selected and neutral genes
Evolution
,
2004
, vol.
58
(pg.
1329
-
1341
)
Hoekstra
HE
Krenz
JG
Nachman
MW
Local adaptation in the rock pocket mouse (Chaetodipus intermedius): natural selection and phylogenetic history of populations
Heredity
,
2005
, vol.
94
(pg.
217
-
228
)
Hoffman
EA
Blouin
MS
A review of colour and pattern polymorphisms in anurans
Biol J Linn Soc
,
2000
, vol.
70
(pg.
633
-
665
)
Hoffman
EA
Schueler
FW
Jones
AG
Blouin
MS
An analysis of selection on a colour polymorphism in the northern leopard frog
Mol Ecol
,
2006
, vol.
15
(pg.
2627
-
2641
)
Holdridge
LR
Determination of world plant formations from simple climatic data
Science
,
1947
, vol.
105
(pg.
367
-
368
)
Huelsenbeck
JP
Ronquist
F
MRBAYES: bayesian inference of phylogenetic trees
Bioinformatics
,
2001
, vol.
17
(pg.
754
-
755
)
Jordan
MA
Snell
HL
Snell
HM
Jordan
WC
Phenotypic divergence despite high levels of gene flow in Galapagos lava lizards (Microlophus albemarlensis)
Mol Ecol
,
2005
, vol.
14
(pg.
859
-
867
)
Kettlewell
HBD
Conn
DLT
Further background choice experiments on cryptic Lepidoptera
J Zool
,
1977
, vol.
181
(pg.
371
-
376
)
King
RB
Lawson
R
Color pattern variation in Lake Erie water snakes—the role of gene flow
Evolution
,
1995
, vol.
49
(pg.
885
-
896
)
Knopp
T
Cano
JM
Crochet
PA
Merila
J
Contrasting levels of variation in neutral and quantitative genetic loci on island populations of moor frogs (Rana arvalis)
Conserv Genet
,
2007
, vol.
8
(pg.
45
-
56
)
Kohlmann
B
Solis
A
Elle
O
Soto
X
Russo
R
Biodiversity, conservation, and hotspot atlas of Costa Rica: a dung beetle perspective (Coleoptera: Scarabaeidae: Scarabaeinae)
Zootaxa
,
2007
(pg.
1
-
34
)
Kohlmann
B
Solis
A
Elle
O
Soto
X
Russo
R
Biodiversity, conservation, and hotspot atlas of Costa Rica: a dung beetle perspective (Coleoptera: Scarabaeidae: Scarabaeinae)
2007

Zootaxa. 68–68
Kohlmann
B
Wilkinson
J
Lulla
K
Costa Rica desde el espacio/Costa Rica from space
,
2002
San Jose (CA)
Fundacíon Neotrópica
Legendre
P
Lapointe
FJ
Casgrain
P
Modeling brain evolution from behavior—a permutational regression approach
Evolution
,
1994
, vol.
48
(pg.
1487
-
1499
)
Lenormand
T
Gene flow and the limits to natural selection
Trends Ecol Evol
,
2002
, vol.
17
(pg.
183
-
189
)
Lythgoe
JN
Patridge
JC
The modelling of optimal visual pigments of dichromatic teleosts in green coastal waters
Vision Res
,
1991
, vol.
31
(pg.
361
-
371
)
Maan
ME
Cummings
ME
Female preferences for aposematic signal components in a polymorphic poison frog
Evolution
,
2008
, vol.
62
(pg.
2334
-
2345
)
Maan
ME
Hofker
KD
van Alphen
JJM
Seehausen
O
Sensory drive in cichlid speciation
Am Nat
,
2006
, vol.
167
(pg.
947
-
954
)
Maan
ME
Seehausen
O
Soderberg
L
Johnson
L
Ripmeester
EAP
Mrosso
HDJ
Taylor
MI
van Dooren
TJM
van Alphen
JJM
Intraspecific sexual selection on a speciation trait, male coloration, in the Lake Victoria cichlid Pundamilia nyererei
Proc R Soc Lond Ser B Biol Sci
,
2004
, vol.
271
(pg.
2445
-
2452
)
Maan
ME
van der Spoel
M
Jimenez
PQ
van Alphen
JJM
Seehausen
O
Fitness correlates of male coloration in a Lake Victoria cichlid fish
Behav Ecol
,
2006
, vol.
17
(pg.
691
-
699
)
Masta
SE
WP
Sexual selection driving diversification in jumping spiders
Proc Natl Acad Sci USA
,
2002
, vol.
99
(pg.
4442
-
4447
)
McDiarmid
RW
Populational variation in the frog genus Phyrnohyas Fitzinger in Middle America
Los Angel Cty Mus Contrib Sci
,
1968
, vol.
134
(pg.
1
-
25
)
McDiarmid
RW
K
Notes on territorial and vocal behavior of neotropical frogs of the genus Centrolenella
Herpetologica
,
1974
, vol.
30
(pg.
75
-
78
)
McKenna
SJ
Raja
Y
Gong
S
Tracking colour objects using adaptive mixture models
Image Vis Comput
,
1999
, vol.
17
(pg.
225
-
231
)
Mignogna
G
Severini
C
Erspamer
GF
Siciliano
R
Kreil
G
Barra
D
Tachykinins and other biologically active peptides from the skin of the Costa Rican phyllomedusid frog Agalychnis callidryas
Peptides
,
1997
, vol.
18
(pg.
367
-
374
)
Nei
M
Molecular evolutionary genetics
1987
New York
Columbia University Press
Nevo
E
Adaptive color polymorphism in cricket frogs
Evolution
,
1973
, vol.
27
(pg.
353
-
367
)
Nosil
P
Crespi
BJ
Does gene flow constrain adaptive divergence or vice versa? A test using ecomorphology and sexual isolation in Timema cristinae walking-sticks
Evolution
,
2004
, vol.
58
(pg.
102
-
112
)
Nosil
P
Sandoval
CP
Crespi
BJ
The evolution of host preference in allopatric vs. parapatric populations of Timema cristinae walking-sticks
J Evol Biol
,
2006
, vol.
19
(pg.
929
-
942
)
Nosil
P
Vines
TH
Funk
DJ
Perspective: reproductive isolation caused by natural selection against immigrants from divergent habitats
Evolution
,
2005
, vol.
59
(pg.
705
-
719
)
Nylander
J
MrModeltest
,
2004
Uppsala
University of Uppsala
Ohmer
ME
Robertson
JM
Zamudio
KR
Discordance in body size, color pattern, and advertisement call across genetically distinct regions of a Neotropical anuran (Dendropsophus ebraccatus)
Biol J Linn Soc
,
2009
, vol.
97
(pg.
298
-
313
)
Orr
HA
Testing natural selection vs. genetic drift in phenotypic evolution using quantitative trait locus data
Genetics
,
1998
, vol.
149
(pg.
2099
-
2104
)
Panhuis
TM
Butlin
R
Zuk
M
Tregenza
T
Sexual selection and speciation
Trends Ecol Evol
,
2001
, vol.
16
(pg.
364
-
371
)
Phillips
CT
Johnston
CE
Geographical divergence of acoustic signals in Cyprinella galactura, the whitetail shiner (Cyprinidae)
Anim Behav
,
2008
, vol.
75
(pg.
617
-
626
)
Price
TD
Phenotypic plasticity, sexual selection and the evolution of colour patterns
J Exp Biol
,
2006
, vol.
209
(pg.
2368
-
2376
)
Pröhl
H
Koshy
RA
Mueller
U
Rand
AS
Ryan
MJ
Geographic variation of genetic and behavioral traits in northern and southern Tungara frogs
Evolution
,
2006
, vol.
60
(pg.
1669
-
1679
)
Pyburn
WF
Blair
WF
The inheritance and distribution of vertebral stripe color in the cricket frog, Acris crepitans
Vertebrate speciation
,
1961
Austin, TX: University of Texas Press
(pg.
235
-
261
)
R Core Development Team
R: A language and environment for statistical computing [Internet]
,
2007
Vienna (Austria)
R Foundation for Statistical Computing.

Available from: URL http://www.R-project.org
Rambaut
A
Charleston
M
TreeEdit, phylogenetic tree editor. Version 1.0a10
,
2001

[Internet] [cited 2002]. Available from: URL http://tree.bio.ed.ac.uk/software/treeedit/
Rambaut
A
Drummond
AJ
Tracer
,
2005

[Internet] [cited 2008]. Available from: URL http://tree.bio.ed.ac.uk/software/tracer/
Rasband
W
ImajeJ
,
2008
Bethesda (MD)
U.S. National Institutes of Health

[Internet] [cited 2009 June]. Available from: URL http://rsbweb.nih.gov/ij/
Reeves
RG
Bermingham
E
Colonization, population expansion, and lineage turnover: phylogeography of Mesoamerican characiform fish
Biol J Linn Soc
,
2006
, vol.
88
(pg.
235
-
255
)
Reynolds
J
Weir
BS
Cockerham
CC
Estimation of the co-ancestry coefficient—basis for a short term genetic distance
Genetics
,
1983
, vol.
105
(pg.
767
-
779
)
Reynolds
RG
Fitzpatrick
BM
Assortative mating in poison-dart frogs based on an ecologically important trait
Evolution
,
2007
, vol.
61
(pg.
2253
-
2259
)
Robertson
JM
Duryea
MC
Zamudio
KR
Discordant patterns of evolutionary differentiation in two Neotropical treefrogs
Mol Ecol
,
2009
, vol.
18
(pg.
1375
-
1395
)
Robertson
JM
Robertson
Spatial and temporal patterns of phenotypic variation in a neotropical frog
J Biogeogr
,
2008
, vol.
35
(pg.
830
-
843
)
Rosenblum
EB
Convergent evolution and divergent selection: lizards at the White Sands ecotone
Am Nat
,
2006
, vol.
167
(pg.
1
-
15
)
Rosenblum
EB
Hickerson
MJ
Moritz
C
A multilocus perspective on colonization accompanied by selection and gene flow
Evolution
,
2007
, vol.
61
(pg.
2971
-
2985
)
Rudh
A
Rogell
B
Hoglund
J
Non-gradual variation in colour morphs of the strawberry poison frog Dendrobates pumilio: genetic and geographical isolation suggest a role for selection in maintaining polymorphism
Mol Ecol
,
2007
, vol.
16
(pg.
4284
-
4294
)
Rutherford
SL
From genotype to phenotype: buffering mechanisms and the storage of genetic information
Bioessays
,
2000
, vol.
22
(pg.
1095
-
1105
)
Sandoval
CP
Nosil
P
Counteracting selective regimes and host preference evolution in ecotypes of two species of walking-sticks
Evolution
,
2005
, vol.
59
(pg.
2405
-
2413
)
SAS
JMP Version 7
,
2007
Cary (NC)
SAS Institute Inc
Savage
JM
The Amphibians and Reptiles of Costa Rica: a herptofauna between two continents, between two seas
,
2002
Chicago
The University of Chicago Press
Savage
JM
Heyer
WR
Variation and distribution in the tree-frog genus Phyllomedusa in Costa Rica, Central America
Beitr Neotrop Fauna
,
1967
, vol.
5
(pg.
111
-
131
)
Sazima
I
Experimental predation on the leaf-frog Phyllomedusa rohdei by the water snake Liophis miliaris
J Herpetol
,
1974
, vol.
8
(pg.
376
-
378
)
Schliwa
M
Euteneuer
U
Comparative ultrastructure and physiology of chromatophores, with emphasis on changes associated with intracellular-transport
Am Zool
,
1983
, vol.
23
(pg.
479
-
494
)
Schneider
S
Roessli
D
Excoffier
L
ARLEQUIN. A software for population genetics data analysis
,
2000
Geneva, Switzerland
University of Geneva
Schwalm
PA
Starrett
PH
McDiarmid
RW
Infrared reflectance in leaf-sitting Neotropical frogs
Science
,
1977
, vol.
196
(pg.
1225
-
1227
)
Siddiqi
A
Cronin
TW
Loew
ER
Vorobyev
M
Summers
K
Interspecific and intraspecific views of color signals in the strawberry poison frog Dendrobates pumilio
J Exp Biol
,
2004
, vol.
207
(pg.
2471
-
2485
)
Slatkin
M
Gene flow in natural populations
Annu Rev Ecol. Syst
,
1985
, vol.
16
(pg.
393
-
430
)
Steinborner
ST
Wabnitz
PA
Waugh
RJ
Bowie
JH
Gao
CW
Tyler
MJ
Wallace
JC
The structures of new peptides from the Australian red tree frog ‘Litoria rubella’. The skin peptide profile as a probe for the study of evolutionary trends of amphibians
Aust J Chem
,
1996
, vol.
49
(pg.
955
-
963
)
Stevens
M
Parraga
CA
Cuthill
IC
Partridge
JC
Troscianko
TS
Using digital photography to study animal coloration
Biol J Linn Soc
,
2007
, vol.
90
(pg.
211
-
237
)
Stewart
MM
Parallel pattern polymorphism in Genus Phrynobatrachus (Amphibia-Ranidae)
Copeia
,
1974
, vol.
(4)
(pg.
823
-
832
)
Storz
JF
Contrasting patterns of divergence in quantitative traits and neutral DNA markers: analysis of clinal variation
Mol Ecol
,
2002
, vol.
11
(pg.
2537
-
2551
)
Storz
JF
Balasingh
J
Bhat
HR
Nathan
PT
Doss
DPS
Prakash
AA
Kunz
TH
Clinal variation in body size and sexual dimorphism in an Indian fruit bat, Cynopterus sphinx (Chiroptera: Pteropodidae)
Biol J Linn Soc
,
2001
, vol.
72
(pg.
17
-
31
)
Stuart-Fox
D
Moussalli
A
Whiting
MJ
Natural selection on social signals: signal efficacy and the evolution of chameleon display coloration
Am Nat
,
2007
, vol.
170
(pg.
916
-
930
)
Summers
K
Clough
ME
The evolution of coloration and toxicity in the poison frog family (Dendrobatidae)
Proc Natl Acad Sci USA
,
2001
, vol.
98
(pg.
6227
-
6232
)
Summers
K
Cronin
TW
Kennedy
T
Variation in spectral reflectance among populations of Dendrobates pumilio, the strawberry poison frog, in the Bocas del Toro Archipelago, Panama
J Biogeogr
,
2003
, vol.
30
(pg.
35
-
53
)
Summers
K
Cronin
TW
Kennedy
T
Cross-breeding of distinct color morphs of the strawberry poison frog (Dendrobates pumilio) from the Bocas del Toro Archipelago, Panama
J Herpetol
,
2004
, vol.
38
(pg.
1
-
8
)
Summers
K
Symula
R
Clough
M
Cronin
T
Visual mate choice in poison frogs
Proc R Soc Lond Ser B Biol Sci
,
1999
, vol.
266
(pg.
2141
-
2145
)
Thompson
DB
Different spatial scales of adaptation in the climbing behavior of Peromyscus maniculatus—geographic variation, natural selection, and gene flow
Evolution
,
1990
, vol.
44
(pg.
952
-
965
)
Thompson
JD
Higgins
DG
Gibson
TJ
CLUSTAL W: improving the sensitivity of progressive multiple sequence alignments through sequence weighting, position specific gap penalties and weight matrix choice
Nucl Acids Res
,
1994
, vol.
22
(pg.
4673
-
4680
)
Thorpe
RS
Analysis of color spectra in comparative evolutionary studies: molecular phylogeny and habitat adaptation in the St. Vincent anole (Anolis trinitatis)
Syst Biol
,
2002
, vol.
51
(pg.
554
-
569
)
Thorpe
RS
Baez
M
Geographic variation in scalation of the lizard Gallotia stehlini within the island of Gran Canaria
Biol J Linn Soc
,
1993
, vol.
48
(pg.
75
-
87
)
Wang
IJ
Crawford
AJ
Bermingham
E
Phylogeography of the Pygmy Rain Frog (Pristimantis ridens) across the lowland wet forests of isthmian Central America
Mol Phylogenet Evol
,
2008
, vol.
47
(pg.
992
-
1004
)
Weigt
LA
Crawford
AJ
Rand
AS
Ryan
MJ
Biogeography of the tungara frog, Physalaemus pustulosus: a molecular perspective
Mol Ecol
,
2005
, vol.
14
(pg.
3857
-
3876
)
Wellborn
GA
Broughton
RE
Diversification on an ecologically constrained adaptive landscape
Mol Ecol
,
2008
, vol.
17
(pg.
2927
-
2936
)
West-Eberhard
MJ
Sexual selection, social competition, and speciation
Q Rev Biol
,
1983
, vol.
58
(pg.
155
-
183
)
Wiens
JJ
Fetzner
JW
Parkinson
CL
Reeder
TW
Hylid frog phylogeny and sampling strategies for speciose clades
Syst Biol
,
2005
, vol.
54
(pg.
719
-
748
)
Woolbright
LL
Stewart
MM
Spatial and temporal variation in color pattern morphology in the tropical frog, Eleutherodactylus coqui
Copeia
,
2008
, vol.
2008
(pg.
431
-
437
)
Wright
S
The distribution of gene frequencies in populations
Proc Natl Acad Sci USA
,
1937
, vol.
23
(pg.
307
-
320
)
Zamudio
KR
Greene
HW
Phylogeography of the bushmaster (Lachesis muta: Viperidae): implications for neotropical biogeography, systematics, and conservation
Biol J Linn Soc
,
1997
, vol.
62
(pg.
421
-
442
)
Zeh
JA
Zeh
DW
Bonilla
MM
Phylogeography of the harlequin beetle-riding pseudoscorpion and the rise of the Isthmus of Panama
Mol Ecol
,
2003
, vol.
12
(pg.
2759
-
2769
)

## Author notes

Corresponding Editor: Robert C. Fleischer