**2. Methods**

Each island was visited in July or August 2004, for 3–10 days. Because of permit restrictions, collection of samples for genetic analysis was restricted to only five individuals per species for all islands. In the case of *Dactyloa gorgonae*, collection was further limited to *N* = 2 by station officials on Isla Gorgona. Cryptic and arboreal niches allowed for collection of only one *D*. *chocorum* and one *N. parvauritus*, but *D. gorgonae*, *D. princeps*, and *N. medemi* were more common and useful for population-wide morphological data (see Supplementary Table S1 for specimen data). For both *N. townsendi* (Isla Cocos) and *D. agassizi* (Isla Malpelo) full populations were observed to assess habitat use and captured and measured when possible (see Morphology and Ecology section below). Additionally, to evaluate the taxonomic status of *N. parvauritus*, samples of *N. biporcatus* were collected throughout Central America (Mexico through Panama, Figure 3, Supplementary Table S2). We complied with all applicable Animal Care guidelines (CMU-IACUC # 10–02).

## *2.1. DNA Sequencing and Phylogenetic Reconstruction*

Tissue samples for each species were collected and preserved in 95% EtOH. Voucher specimens were fixed in formalin, stored in EtOH, and deposited with the Universidad de los Andes, Colombia. Qiagen DNEasy kits were used for DNA extraction. PCR amplification of several contiguous mitochondrial genes (ND2, tRNA*Trp*, tRNA*Ala*, tRNA*Asn*, tRNA*Cys*, tRNA*Tyr*, and partial CO1) used previously in the literature for phylogenetic reconstruction of mainland anoles [42,43] was conducted. Genes were amplified using Empirical Bioscience Mean Green Master Mix and the same cycling conditions and primers as Nicholson et al. [21]. PCR products were purified using Qiagen QIAquick purification kits, followed by sequencing at Michigan State University's (MSU, East Lansing, MI, USA) sequencing core. Sequence data were edited and aligned using MUSCLE, then combined with published sequences for individuals of the same species from mainland Central or South America and closely related species (Supplementary Table S1). Our newly collected data were combined with published data for 75 *Dactyloa* and *Norops* species plus 10 non-Dactyloid outgroup taxa (Table 1 and Supplementary Table S1) for a total of 1480 aligned base pairs of sequence data (Genbank vouchers in Supplementary Table S1).


**Table 1.** Mean dates and 95% confidence intervals for select nodes relevant to Pacific Island anoles, using multiple dating methods (fossil-fossil calibrations + mutation rate; rate = mutation rate alone).

Phylogenetic analyses were conducted to investigate the placement of taxa among established species of anoles for which comparative data were available. Phylogenetic reconstructions were conducted using BEAST2 for Bayesian analyses under a lognormal relaxed clock model and a Yule Speciation prior [63]. PARTITIONFINDER was used to select models of evolution as well as to examine the suitability of partitioning. In all cases each gene (including tRNAs) was entered as a potential partition for the first, second, and third codon positions. Branch lengths were unlinked, all models of evolution available in BEAST2 were tested, and a BIC information criterion and greedy algorithm were used. The PARTITIONFINDER analysis recommended a GTR + Γ model of evolution for the mtDNA segmen<sup>t</sup> without partitioning. The same alignment for each dating method was run for 20 million generations with 20% burn in removed via TreeAnnotator [63] after evaluating that the analysis had reached stationarity using Tracer v.1.6.0 [64].

We estimated the date of origin for each island species by analyzing our data using BEAST2. We used two approaches to estimating the age of divergence between each island lineage and its mainland relatives: (1) using the Macey et al. [65] calibration rate of 0.65 mutations per 100 bp per million years, used by several mtDNA studies of anoles and related taxa [66–71] and (2) implementing the same mutation rate in addition to three calibration dates for Iguanian fossils used by Prates et al. [37]

and sources within. We used this approach following Román-Palacios et al.'s [24] findings that implementing multiple dating strategies can avoid the pitfalls of trees based only on a single mutation rate [72].

From our phylogenetic reconstruction we compared divergence dates between the island and mainland lineages to the estimated timing of origin for each island; our null hypothesis was that each species' origin would post-date the emergence of each island but had no a priori expectations or hypotheses regarding when each species may have originated. For some species (*N. medemi*, *N. townsendi*) for which we had multiple samples within an island, we were also able to estimate the coalescence of species on each island, making the assumption that this date represents the younges<sup>t</sup> possible age of colonization.

## *2.2. Morphology and Ecology*

On each island, observations were made for all species during diurnal and nocturnal survey periods (except *N. agassizi*, as we were not allowed on Isla Malpelo at night). Individuals were captured with collapsible fishing poles fitted with string lassos in order to record population-wide morphological data. Measurements were taken on all captured individuals (SVL, mass, fore- and hind-limb length, head width, depth and length, number of lamellae rows) as well as recording ecological data (perch height, perch diameter, perch type; Supplementary Tables S3–S5).

Student's t-test was used to compare di fferences between the sexes for each species (although foreand hind-limb length, head width, depth and length, were all determined to covary with SVL and were omitted from the sexual dimorphism analysis). Additionally, ecological and morphological data for *N. townsendi* were collected at four di fferent locations on the island, including the highest point (Cerro Yglesias) and a small islet (Manuelita) that is disconnected from Isla Cocos. One-way ANOVA assuming unequal variance was implemented to test for di fferences among sites, which may sugges<sup>t</sup> separate populations. This was followed by a *t*-test for any variables that displayed a significant di fference (*p* < 0.05) to evaluate pairwise relationships between populations. To account for multiple tests in all cases, we applied a sequential Bonferroni correction. The same process was used to evaluate the di fferences in habitat use (perch height, perch diameter) among three of the Isla Gorgona species (*D. gorgonae*, *D. princeps*, *N. medemi*).
