Computational protocol: Mutualistic Interactions Drive Ecological Niche Convergence in a Diverse Butterfly Community

Similar protocols

Protocol publication

[…] We generated a phylogeny of the 58 ithomiine species (24 genera) of the community using Bayesian Inference under the uncorrelated lognormal relaxed clock model implemented in BEAST [], using a 2,296 bp mitochondrial region spanning the CoI, tRNA leu and CoII genes, as well as a 1,215 bp fragment of Elongation Factor 1a, a nuclear gene () []. Each region followed a GTR+Γ model of substitution, and two MCMC chains were run for 200 million generations (sampling every 1,000 generation, 10% burn-in). Two species had two subspecies each with different color patterns in the community, which were considered as separate taxa in both cases (). To increase the accuracy of the tree topology we added published sequences of genera that were not represented in the community [,], but these additional taxa were pruned from the tree prior to the analyses presented below. Six rare ithomiine species were also excluded from the analyses because of missing ecological data (). [...] All analyses on the phylogeny were performed on the maximum clade credibility tree with average branch lengths, computed by TreeAnnotator [] from the 360,000 trees retained. To account for phylogenetic uncertainty we collapsed all nodes that had a posterior probability less than 0.90. Analyses were conducted on all 54 taxa for which sufficient ecological data was recorded. All statistical analyses were performed with R [], using packages APE 2.1–3 [], Geiger 1.2–06 [], Vegan 1.11–0 [], and Cluster 1.11.10 [].To investigate ecological segregation of mimicry complexes we performed a MANOVA on all variables and an ANOVA for each individual variable. These analyses were performed on the entire data set to test whether mimicry complexes are ecologically segregated (regardless of species identity), and on species' averages to test whether co-mimetic species are more similar than species picked at random (regardless of species abundance). The significance of ANOVA/MANOVA test statistics was assessed by permuting the observed ecological data (10,000 permutations).Correlations between phylogenetic and ecological distances among all species, among co-mimics and among non-co-mimics were computed and their significance was tested by permuting the observed ecological data among species (10,000 permutations, one-tailed test for positive correlation). We performed a partial Mantel test for the correlation between ecological and mimicry distances while controlling for phylogeny. We did this by first regressing ecological and mimicry distances against phylogenetic distances and then assessing the significance of correlations between residual ecological and mimicry distances by permutation of the residual distances among species pairs (logistic regression for mimicry distances; 10,000 permutations, one-tailed test).Tests for ecological convergence and divergence for each microhabitat variable and for the global microhabitat were performed in two additional ways: (1) simulated character evolution, which incorporates phylogeny into the expected distribution of type I error, and (2) regression-based methods, which remove the phylogenetic component of variation in ecological characters. In the first way, we calculated the average ecological distance among co-mimics standardized by the total tree ecological distance (tests for ecological convergence among co-mimics), or the average distance among all non-co-mimics standardized by the total tree ecological distance (tests for ecological divergence between mimicry complexes). Character evolution was then simulated on the phylogeny assuming gradual and speciational character evolution (10,000 simulations), and the actual value of the parameters of interest compared with the distribution of the same parameter generated by the simulations to obtain p-values. As we wanted to test whether scaled ecological distances were smaller than expected among co-mimics and greater than expected among non-co-mimics, one-tailed tests were used in each case. In the second way, ecological distances were regressed against phylogenetic distances using all species. We then permuted these residual ecological distances among species pairs to determine whether the observed residuals were more negative or positive then expected by chance for co-mimics and non-co-mimics, respectively (10,000 permutations).Niche complementarity was investigated by calculating the correlation coefficient of distances along one ecological variable with distances along another variable, for all pairs of variables. The significance of correlation coefficients was tested in two ways. First, as above, the actual value was compared with the distribution of the same parameter generated from 10,000 replicate simulations of character evolution under both gradual and speciational models. Second, a partial Mantel test was carried out for each pair of ecological variables, in which distances along each of the two variables were regressed against phylogenetic distances. We assessed the significance of correlations between residuals for each pair of ecological variables by permutation of the residual distances among species pairs. The tests were one-tailed (negative correlations) and the significance of each test was based on 10,000 permutations.We controlled for multiple tests using the false discovery rate procedure [], a powerful alternative to the Bonferroni correction that seeks to minimize both type I and type II errors, with the allowed proportion of false positives set at 0.05. Unless stated otherwise, all tests presented in the text, tables and figures are significant after this correction. […]

Pipeline specifications

Software tools BEAST, APE, GEIGER, vegan
Application Phylogenetics