pared with the number of sites that was fixed between species for synonymous and nonsynonymous substitutions by using a Fisher’s exact test.
Sequences for Limenitis species were added to other L opsin sequences obtained from GenBank for which physiological data were also available. The resulting alignment was edited to retain only coding sequence, 263 aa in length, spanning transmembrane domains I–VI (Fig. 10.4). Phylogenetic relationships were reconstructed from nucleotide data by using maximum-likelihood and Bayesian methods, with the moth Manduca sexta L opsin sequence as an outgroup by using all three nucleotide positions. The optimal DNA substitution model for the maximum-likelihood phylogenetic analysis was determined by using a hierarchical likelihood ratio test in Modeltest (Posada and Crandall, 1998). A maximum-likelihood analysis was conducted in PHYML with TrN93 + I + G (invariant sites and gamma-distributed rates for sites) substitution model, and the reliability of the tree obtained was tested by bootstrapping with 500 replicates. Bayesian phylogenetic analyses were performed by using MrBayes 3.1 (Ronquist and Huelsenbeck, 2003). Because MrBayes 3.1 does not implement the TrN93 DNA substitution model, we used the next-less-complex (HKY85+I+G) and the next-more-complex (GTR+I+G) models in two separate analyses. Both models were run for 106 generations, with a sampling frequency of 102, using three heated and one cold chain and with a burnin of 2.5 × 103 trees.
The Bayesian and maximum-likelihood tree topology obtained was used to perform all selection tests. For the parallel change analysis, amino acid substitutions along each lineage in the opsin gene tree were reconstructed by using maximum parsimony in MacClade (Maddison and Maddison, 2005) and maximum likelihood in PAML (Yang, 1997). Parallel amino acid changes were detected along butterfly lineages that also displayed parallel phenotypic evolution in the L visual pigment λmax value. The statistical significance of these amino acid changes was tested by using the method (Zhang and Kumar, 1997) implemented in the program Converge.
For the branch-site test of selection, we used branch-site models (Yang and Nielsen, 2002; Zhang et al., 2005) that allow the dN/dS ratio (ω) to vary both among sites and among lineages because these models may be more likely to detect positive selection affecting only a few sites. Specifically, we used test 2 of Zhang et al. (2005) to construct a likelihood ratio test