Computational protocol: The Evolution of Morphospace in Phytophagous Scarab Chafers: No Competition - No Divergence?

Similar protocols

Protocol publication

[…] Analyses of morphospace were implemented based on the Bayesian phylogenetic tree on the preferred alignment as a backbone. Morphospace was explored for the complete data set of all specimens and for five subsets that compare major sister clades. Comparisons between sister clades with low support values are omitted except those with relationships that are also well established in traditional morphology-based systematics. All calculations for the analysis of morphospace were made within the R statistics environment version 2.15 unless otherwise stated. Obtained linear measurements were log10-transformed to render more linear relations among variables and to obtain a similar dimension of variance , . Generally, the major component of variance in morphometric data sets of biological specimens is explained through size –. To avoid a strong bias of size over variation of shape, we employed approaches that separate size from shape information. In landmark-based geometric morphometrics, this is achieved using “two point registration” methods , , but for linear measurements there is still some debate regarding how to perform this separation , . Here, we employed the Burnaby Back Projection Method (BBPM) by projecting the log-transformed data on the isometric size vector and returning it to the original coordinate system , as implemented in an R-code provided by Blankers et al. . This method has the advantage of deriving a composite measure of size from all traits and considering shape as the projection onto the orthogonal space of this isometric vector. Data treated in this manner are subsequently referred to as size-corrected data (set). Correcting the data for size can strongly affect the results depending on the method used and must be considered well. Therefore, we compared the results from the BBPM with shape data derived from a linear regression (residuals) against overall body length , which was chosen to be representative of the beetles' body size. Because a high error is introduced to the total body length measure through the motility of the prothorax against the pterothorax, a proxy was used by calculating the logarithm of the sum of pronotal and elytral length (log(PL+EL)). The impact of size (percentage of variation that is explained by size alone) was assumed to be represented by the percentage of variation explained by the first principal component of the uncorrected data set.Patterns of morphometric covariation were analyzed with standard principal component analysis (PCAs; , ) on uncorrected and size-corrected data. Results were visualized with the help of the ade4 package . Additionally, the molecular phylogeny was projected onto the morphospace explained by PCs 1 and 2 using the function phylomorphospace in the R package phytools . The program therefore estimates the positions of the ancestral nodes using a maximum likelihood approach.Statistical evaluation of group differentiation in morphospace was done by MANOVA and linear discriminant analysis (LDA). To avoid confusion through noise introduced from measurement errors or minor unspecific variation –, we only used the principal components that explained 95% of total variation. Non-parametric MANOVA was performed for the complete data set and each sister clade subset in PAST 2.17 to test for significant differentiation between lineages. Sequential Bonferroni correction was applied. LDA was conducted on the same groupings to evaluate group discrimination by the reassignment probabilities which were evaluated by leave-one-out cross-validation using the MASS-package in R. Lineages represented only by a single species, i.e. Ablaberini, were included in the PCA but had to be excluded from LDA and MANOVA. [...] Inference of the potential influence of the food resource on morphospace variation was done by mapping feeding habits of each species onto morphospace. Details on feeding behavior were taken from the literature and were complemented with personal observations (). Coprophagous (COP) and saprophagous (SAP) species were represented by Aphodiinae/Scarabaeinae and Hybosoridae, respectively. Anthophilous (ANT) species exclusively forage on flowers, feeding on pollen and nectar, whereas herbivorous species (HERB) devour various plant materials, including foliage, twigs, and petals. Dynastinae species examined here are sap/fluid utilizers (SFU) feeding under ground on stems or roots in order to gain access to fluids from the wounds . Adults of Pachypus do not feed (NF).A correlation analysis between morphospace and feeding types was performed employing phylogenetic generalized least squares regression in the package caper using the pgls function . The assigned feeding types were used as independent variables and (standard) principal components explaining 95% of cumulative variation as dependent variables representing the morphospace. To improve the fit of the data to the tree, Pagel's branch length transformation variable λ (internal branch lengths are multiplied with λ) was set to be estimated by maximum likelihood. κ (each branch length is raised to the power κ, ) and δ (the node heights are raised to the power delta, ) were set at 1.A possible correlation between molecular and morphological distances between the specimens was estimated by Mantel-tests, performing 10,000 permutations of Pearson correlations with the R package vegan . The analysis was made for size-corrected data for all members of each feeding type separately. [...] Reduced correlation between molecular and multivariate morphometric distances is likely to indicate decoupling of molecular and morphological rates of evolution, with accelerated or decelerated rates of evolution in either of the traits, i.e. directed selection on morphospace evolution. Therefore, Mantel-tests with Pearson correlation were performed on distance matrices of patristic distances (calculated with the cophenetic function in the R-package ape from the molecular tree ) and Euclidean distances of the respective morphological data sets. To infer individual traits that underlay directed selection, i.e. that deviate from Brownian Motion, the descriptive K statistic of Blomberg et al. was calculated for every trait over the complete size-corrected data set using the R package phytools , . A K value greater than one implies that close relatives are more similar than expected under Brownian motion evolution .Branches in the phylogeny, where the molecular and the morphological distances between nodes deviate from each other, were detected by projecting both the uncorrected and the size-corrected data set on the constrained topology of the phylogenetic tree , . The branch lengths were inferred with the from the phytools package using Euclidean distance matrices of the respective data sets. Negative branches were set to zero. For both the size-corrected and the uncorrected data set, ratios of morphological and molecular branch lengths were calculated for each branch. Values above and below the 95% confidence interval of the ratios were considered as significantly different in their branch lengths, i.e. indicating an extraordinary decoupling of molecular and morphological rates and consequently directed selection at the respective ‘outlier’ branch. Because the lengths of internal branches and tips often largely differed, they were evaluated separately. (cf. )Additionally, we calculated standardized phylogenetic independent contrasts – in order to compare evolutionary rates of morphospace divergence between clades. For this objective, we used the multivariate approach introduced by McPeek et al. and applied it to both data sets. The method of McPeek et al. was implemented in R and the script is provided in . The ultrametric tree necessary for this approach was calculated based on the preferred alignment (2513 bp) of Ahrens and Vogler using PathD8 , with the root of an arbitrary age of one. Ancestral linear size measurements of traits possibly linked with presumptive key innovations were reconstructed with the function fastAnc in phytools using a Maximum Likelihood approach. […]

Pipeline specifications

Software tools Phytools, vegan, PATRISTIC, APE
Application Phylogenetics