Species limits and phylogeography
The species status of both lava lizards and geckos from Floreana and surrounding islets is strongly supported by morphological evidence (Van Denburgh, 1912; van Denburgh & Slevin, 1913; but see Discussion). To further test that populations from Floreana and surrounding islets represent the same species, we examined species limits of both M. grayii and P. baurii with phylogenetic analyses and genetic distances. We assume that monophyly and genetic distances relative to closely related species are strong criteria for delimiting species. Sequences were assembled and aligned in Geneious v9.1.8 (http://www.geneious.com/) under default settings for MAFFT (Algorithm: Auto, Scoring Matrix: 200PAM/k=2, Gap open penalty: 1.53, Offset value: 0.123; Katoh & Standley, 2013). Alignments were verified by translating sequences into amino acids to check for misplaced stop codons. After partitioning each dataset (i.e., Microlophus andPhyllodactylus ) by codon position, we conducted maximum likelihood analyses using RAxML-HPC2 v.8.2.9 with a heuristic search that included 100 random addition replicates and 1000 thorough bootstrap pseudo-replications. We applied the GTR+GAMMA model to all partitions. Analyses were performed using the CIPRES platform (Miller, Pfeiffer, & Schwartz, 2010). The phylogenetic trees were visualized and edited using FigTree v1.4.2 (Rambaut, 2014). In addition, for each dataset we calculated uncorrected genetic distances in MEGA 7 (Kumar, Stecher, & Tamura, 2016) to address genetic differentiation within M. grayii , P. baurii , and their corresponding radiations.
The relationship among haplotypes from Floreana, Champion and Gardner was evaluated for both lizard species by inferring TCS networks (Templeton, Crandall, & Sing, 1992) in PopART (Leigh & Bryant, 2015). This method has been used widely with nucleotide sequence data to infer population level genealogies when divergences are low (Georgiadis et al., 1994; Gerber & Templeton, 1996; VilĂ  et al., 1999). Because haplotype networks can be misleading in the presence of missing data (Joly, Stevens, & van Vuuren, 2007), two datasets were analyzed for each species: one including all sampled sequences, and the other removing those sequences containing significantly more undefined states than others.