Computational protocol: Environmental determinism, and not interspecific competition, drives morphological variability in Australasian warblers (Acanthizidae)

Similar protocols

Protocol publication

[…] We compared the fit of the best SURFACE model to three evolutionary models that lack deterministic convergence and to two models with a priori designation of selective regimes based on foraging niche categories and geographic distribution (region). Specifically, to each PC axis, we fitted the following models using maximum‐likelihood inference: (1) a BM model in which traits evolve following a random‐walk process and morphological disparity accumulates roughly linearly through time (due to randomly fluctuating selection or genetic drift) (Felsenstein, ); (2) an early‐burst (EB) or adaptive radiation model in which phenotypic change occurs rapidly after lineages enter available niches and decreases as niches are filled (Harmon et al., ; Simpson, ); (3) a single‐peak OU model (OU1) with one parameter for the variance of random‐walk (σ2) and strength of selection (α) toward a global optimum for all acanthizids (Butler & King, ); (4) a multi‐peak OU model (OUMregion) with separate random‐walk variances for each geographic region (Australia, New Guinea, Australia‐New Guinea, New Zealand, and Chatham Islands); (5) a multi‐peak OU model (OUMniche) with separate random‐walk variances for each one of the five foraging niche categories (“canopy,” “low trees,” “trunks,” “shrubs,” and “ground”) and one global selection parameter (α); and (6) a multi‐peak OU model (OUMSURFACE) with separate random‐walk variances for each one of the adaptive peaks identified using SURFACE (see Section ). To deal with phylogenetic uncertainty, the BM, EB, and OU1 models were run across a sample of 100 trees obtained from the posterior distribution of the Bayesian analysis. For the multi‐peak (OUM) models, we first built stochastic character‐mapped reconstructions (SIMMAP; Bollback, ) of (1) foraging niche categories, (2) regions, and (3) adaptive peaks estimated by SURFACE, for each of the 100 trees sampled from the posterior distribution, using phytools (Revell, ). Models were implemented using the R packages geiger (Harmon, Weir, Brock, Glor, & Challenger, ) and OUwie (Beaulieu & O'Meara, ) and compared by means of the sample size‐corrected Akaike's Information Criterion (AICc). [...] We used Mantel tests to determine whether more similar species are those that (1) are closely related; (2) are more geographically close; (3) do not exhibit range overlap; and/or (4) share climatic conditions. To this end, we first produced matrices representing the phylogenetic, morphological, climatic, and geographic distances between all pairs of species. The phylogenetic matrix represents the patristic distance between each pair of species in the phylogeny depicted in Marki et al. (). Patristic distances were obtained using the function cophenetic in the stats package (R Core Team ). For the morphological matrix, we computed the Euclidean distance for all pairwise comparisons between species in the space defined by the two PC axes (PC1 and PC2). To compute the matrix of geographic distances, we first obtained the distributional midpoint of each species from a presence–absence matrix using the R package letsR (Vilela & Villalobos, ). In addition, as information based on single location is not useful to quantify overlap (sympatry) between each pair of species, we also computed a range overlap matrix from our presence–absence matrix using the function “lets.overlap” in the R package letsR (Vilela & Villalobos, ). Range overlap represents the proportion of the smaller range that occurs within the larger range (Cheeser & Zink, ; Martin et al., ) so that values range from 0 (no overlap) to 1 (smaller range completely overlapped by the larger range). The climatic matrix was constructed as the matrix of Mahalanobis distances between species based on the two abiotic variables: mean annual temperature and annual precipitation. The level of correlation between matrices was assessed by means of Mantel tests with 9,999 random permutations as implemented in the ade4 library (Dray & Dufour, ). […]

Pipeline specifications

Software tools SIMMAP, Phytools, GEIGER, PATRISTIC, ADE4
Applications Miscellaneous, Phylogenetics