Computational protocol: Testing the island effect on phenotypic diversification: insights from the Hemidactylus geckos of the Socotra Archipelago

Similar protocols

Protocol publication

[…] Primers and conditions used for the amplification and sequencing of the different fragments followed methods described in ref. . DNA sequences were aligned using MAFFT v.6  with the options “maxiterate 1000” and “localpair”. Poorly aligned positions in the 12S marker were eliminated by means of G-blocks, using low stringency options. The final alignment consisted of 4,016 bp as follows: 379 bp of 12S; 1,137 bp of cytb; 402 bp of c-mos; 666 bp of mc1r; 1,024 bp of rag1; and 408 bp of rag2. Best fitting nucleotide substitution models were selected for each partition under the Akaike information criterion (AIC), using jModeltest v.0.1.1. The best models were GTR + I + G for 12s, cytb and mc1r, and TrN + G for c-mos, rag1 and rag2. Alignment gaps were treated as missing data and the nuclear gene sequences were not phased.After the alignment of our molecular dataset, phylogenetic analyses were performed using package BEAST v1.6.2  (see for more information on the BEAST analysis). A summary tree was calculated as the maximum clade credibility tree with median node heights and, to incorporate phylogenetic uncertainty into our comparative analyses, we resampled the posterior distribution of trees resulting from our BEAST analysis to obtain a sample of 1,500 trees that varied in topology and branch lengths.Ancestral state reconstructions were conducted using the function “make.simmap” from the R package “phytools”. This function essentially fits a continuous-time reversible Markov model (in our case allowing different transition rates between all states) and simulates plausible stochastic character histories along the tree using the most likely model in combination with the states assigned to the tips of the tree. [...] We used the function “phenogram” in the R package “phytools” to visualize the variation in size and head proportions across the summary tree. This function essentially projects the phylogeny into a space defined by the phenotype (on the y axis, including the values at the tips and the values reconstructed at the nodes) and time (on the x axis). Given the multivariate nature of the head proportions, we also visualized its variation by means of the “phylomorphospace”. In this case the tree is projected into a bivariate space represented by the species values and the reconstructed states at the nodes for each combination of two head variables. Both representations allow the identification of the major trends of phenotypic change across the tree, as well as the different magnitudes of variation (proportional to the vertical component of each branch in the “phenogram”, and proportional to the branch lengths in the “phylomorphospace”).To formally assess whether mainland and island species differed in morphology, we performed a phylogenetic ANOVA and MANOVA on size and head proportions respectively. Significance of the empirical F statistic (for ANOVA) and the Wilk’s lambda (for MANOVA) was assessed by means of null distributions based on 10,000 Brownian motion simulations. For body size, these simulations were based on a maximum likelihood (ML) estimate of the empirical rate parameter. For head proportions, simulations were based on the ML estimate of the evolutionary variance-covariance (vcv) matrix. Both analyses were performed in the R package “geiger” and were conducted on the summary tree and on the set of 1,500 trees. [...] We assessed the differences of disparity between mainland and island settings in two different ways. We first compared the range of body size and variation in head proportions (maximum - minimum value) existing in the two islands with 10,000 randomly assembled “pseudo-communities” of two and of five continental species, thus simulating the two species existing in Abd al Kuri and the five in Socotra. In this way we assessed whether any combination of species existing in the continent produced the range of phenotypic variation existing in the islands.Next, we compared island versus continental disparity incorporating the phylogeny in the analysis. To do this, we first defined disparity as the average squared Euclidean distance computed between the sizes and head proportions of all pairs of species coexisting in a given area. In this way we calculated the disparity in body size and head proportions in Socotra, Abd al Kuri and in the continent. We then measured the overlap between continental and island disparities by calculating the ratios between the disparities of Socotra and the continent, and between Abd al Kuri and the continent. These ratios were then compared with a null model consisting of 10,000 simulations in which body size and head proportions were stochastically simulated according to a Brownian motion model. Simulations were based on an empirical estimate of the rate parameter for body size, and on the estimated evolutionary variance-covariance matrix for head proportions.By comparing the empirical ratios with the simulated ratios according to the stochastic model, we assessed whether island disparities significantly departed from the mean continental disparity. This analysis was performed on the summary tree, and it was also replicated for each of the 1,500 trees in order to incorporate the phylogenetic uncertainty into our empirical and simulated disparity ratios. All the analyses of this section were performed using the “ape” and “geiger” packages in R. […]

Pipeline specifications

Software tools MAFFT, jModelTest, BEAST, SIMMAP, Phytools, GEIGER, APE
Application Phylogenetics