Computational protocol: Study of morphological variation of northern Neotropical Ariidae reveals conservatism despite macrohabitat transitions

Similar protocols

Protocol publication

[…] To investigate the influence of habitat and phylogeny on skull shape variation we performed analyses that take the phylogenetic relatedness into account. The following analyses were carried out on a subset of the shape dataset (Additional file : Table S1) that contains only species that are present in the available phylogeny. To this end, we first calculated mean shapes per species using the mshape function. The phylogenetic tree (Fig. b) was taken from Stange et al. []. The tree is derived from a multispecies-coalescent analysis based on single-nucleotide polymorphisms and internal node calibration based on fossils instead of the common biogeographic calibration point, the final closure of the Panamanian Isthmus. It was read in using read.nexus (ape).Blomberg’s K [] is an estimator that assesses the strength of phylogenetic signal in any quantitative variable. Phylogenetic signal in this context is the association of phenotypic similarity derived from Procrustes coordinates to phylogenetic relatedness among the taxa under study and is determined by the generalized version of K for multivariate data []. The estimation of K is implemented in the physignal function, which was run on the averaged species shape data with 1000 random permutations for significance testing. K is the ratio of the observed trait variance and the expected trait variance as predicted under Brownian motion. K has an expected value of 1 under Brownian motion (strong phylogenetic signal), a K < 1 implies higher shape divergence of taxa, and a K > 1 implies more shape similarity of closely related taxa than expected by a Brownian motion model of trait evolution. K = 0 resembles the null hypothesis, stating that there is no phylogenetic signal in the shape data and that closely related taxa are not more similar to each other than distantly related taxa [].To test whether the distinctiveness of habitat-specific shapes holds true also after taking the phylogenetic dependence of the taxa into account, we performed a phylogenetic ANOVA on the shape data (Procrustes coordinates). The shape data were analysed applying a generalized least squares approach [, ], as implemented in the procD.pgls function. The significance of differences among groups was tested in a permutation test based on residual randomization [] (RRPP = TRUE) with 999 random permutations.We assessed phenotypic convergence, hypothesizing that species from the same habitat group would ‘converge’ towards similar shapes. First, for an initial visual inspection, we produced a phylomorphospace plot in PC1-PC2 shape space. The phylogeny was projected on the mean species shape scores of the tip data and the reconstructed ancestral states derived from maximum likelihood analysis using the plotGMPhyloMorphoSpace function. Second, following the argumentation by Zelditch et al. [] we tested for convergence in the full shape space instead of using principal components, as the latter do not exhibit independent rates of adaptation and diffusion. We chose to apply the ‘C-metrics’ [] as these are also applicable to multi-dimensional shape data opposed to SURFACE [] which is only suitable for multivariate data []. We therefore follow the procedure proposed by Zelditch et al. [] to first compute a tanglegram using the cophylo function from phytools [], comparing the phylogeny and the phenogram. The phenogram is a UPGMA tree computed from Procrustes distances from species mean configurations. Lines are drawn between the phylogeny and the phenogram connecting identical tips. Convergence is indicated by crossing lines in the tanglegram and those instances are chosen to be analysed with the ‘C-metrics’. We calculate C1 to C4 by using the calcConv function as provided in the supporting information of Zelditch et al. []. This code has been adapted to perform calculations based on distances in the full shape space instead of being based on principal components. C1 measures the distance in shape space of two species as a proportion of the maximum distance the lineages have experienced. C2 is based on the same distance measures as C1 but it is measured on an absolute scale in contrast to being relative to the maximum phenotypic distance. C1 and C2 are comparable within datasets but not between them. C3 and C4 are based on standardising C2 for the total amount of evolutionary change leading from the most recent common ancestor (MRCA) to both tips, and standardising by the total amount of evolutionary change along all lineages descended from the MRCA of the two focal tips, respectively, which allows comparison between data sets [, ]. We do not analyse C5 here (the frequency of convergence) as the dataset does not contain more instances of convergence than variables. […]

Pipeline specifications

Software tools APE, PHYSIG, Phytools
Application Phylogenetics
Organisms Danio rerio