Computational protocol: The evolution of the platyrrhine talus: A comparative analysis of the phenetic affinities of the Miocene platyrrhines with their modern relatives

Similar protocols

Protocol publication

[…] An up-to-date platyrrhine phylogeny () was modified slightly in Mesquite v. 3.04 (), adjusting some species names to match those in the morphological dataset, adding some species (Ateles marginatus, Aotus infulatus, Chiropotes satanas, Mico melanurus, and Saguinus leucopus; , , , , ) by hand and removing species for which there were no talar data. The resulting phylogeny (; ) was used to perform all the described comparative analyses.Figure 2Figure 2 [...] It was necessary to establish if there was a significant association between talar morphology and locomotion to test whether talar morphology is a good proxy for locomotion. First the locomotor mode percentages (LMPs) (i.e., the percentage time a species spends performing a certain locomotor behavior) of 31 platyrrhine species were obtained from . This dataset compiled several sources from different publications, and considered five different locomotor behaviors: bridge/suspensory locomotion, arboreal quadrupedal walk, clamber/vertical climb, leap/drop/hop, and clawed locomotion. A PCA of the correlation matrix of the LMPs of the species used in the present study (n = 23) was carried out to see if main locomotion modes could be distinguished. The phylogenetic signal of the LMPs was estimated using a mathematical generalization of the K-statistic () appropriate for multivariate data (i.e., Kmult) (). The K-statistic varies between 0 (no phylogenetic signal in the data as in a star phylogeny) to 1 (data fit a Brownian motion (BM) model of evolution) or significantly more (species are more similar than expected under BM) (). Subsequently, both a standard partial least squares (PLS) and a phylogenetic PLS analysis were performed to examine the association between the LMPs and the shape variables of the species that were present in both datasets (). The standard PLS calculates the degree of covariation between the two datasets, while the phylogenetic PLS also accounts for phylogeny under a BM model of evolution (). Partial least squares has the advantage that it does not assume that one set of variables is dependent on the other, thus being a useful tool for assessing the relationship between sets of variables that might covary but for which there is no a priori directional relationship (). These results were expected to contribute to the understanding of the relationship between talar morphology and locomotion. In addition, the first two PCs of the PCA of the LMPs were used to estimate the ancestral states for internal nodes, first using maximum likelihood and then by interpolating the states along the branches of the tree according to in the R package ‘phytools’ (, ). In this way, we tried to reconstruct the ancestral locomotor condition of the NWM using published locomotion data. [...] Phylogenetic signal was estimated for talar shape, centroid size and body mass using the Kmult statistic (). To visualize the phylogenetic relationships in the morphospace, the phylogeny was projected onto the space identified by the first two PCs obtained from the covariance matrix of the average shapes of the analyzed taxa (). In addition, by using the squared-change parsimony approach of the ancestral body masses, centroid sizes and shapes () for the different nodes of the phylogeny were estimated. This approach was preferred because the squared-change parsimony reconstruction has maximum posterior probability under a BM evolutionary model (). Therefore, the ancestral reconstructions represent conservative hypotheses about the possible trait values of the actual ancestors.A multivariate phylogenetic generalized least square regression (PGLS) was used to evaluate the association between shape and some size measures (i.e., body mass and centroid size) to analyze the influence of allometry on talar shape. Even though talar centroid size and body size are highly correlated (R2 = 0.94; p-value < 0.001), two separate regressions were performed using these two size measures to provide a full picture. By modeling residual variation assuming a BM evolution mode, PGLS takes into account the expected absence of independence across taxa due to phylogenetic structuration, which is expected to affect the covariance in trait values (). The body mass data were gathered from the available literature (, ). As male and female body mass are highly correlated among the living platyrrhine species, average body mass was used in the analyses ().The first five PCs of the extant dataset (63.57% of explained variance) were used in the following comparative analyses based on the results obtained from a broken-stick model used to assess significance of variance (). This procedure was performed to reduce the number of variables, given that 40 taxa, each one represented by 30 3D landmarks, were analyzed.It was tested whether talar morphology exhibited shape convergence between some of the platyrrhine groups by using the SURFACE method implemented as the runSurface() function from the R package ‘surface’ (). This method fits a model of adaptive radiation in which lineages might experience shifts to adaptive peaks on a macro-evolutionary landscape without reference to a priori hypotheses specifying which lineages correspond to particular peaks (). Starting with an Ornstein-Uhlenbeck (OU) model in which all species are attracted to a single adaptive peak in trait space (), SURFACE uses a stepwise model selection process based on the finite-samples Akaike information criterion (AICc) to fit increasingly complex multi-peak models (). In the ‘forward phase’ a new peak shift is added to the branch of the phylogeny that most improves model fit across all traits, and shifts are added until none results in further improvement (i.e., ΔAICc < 2) (). Then in the ‘backward phase’ the method assesses whether the AICc score is improved further by collapsing regimes in different branches to shift toward shared adaptive peaks rather than requiring each to occupy a unique peak, to identify possible convergence (). This ‘backward phase’ proceeds step by step until no further improvement is achieved. The SURFACE method can thus survey several hundred OU models, obtaining a model with the highest absolute statistical support among those explored. Importantly, convergence is understood here as described by as evolution towards the same adaptive peak, therefore distinguishing between convergence occurring as a result of deterministic adaptation to specific ecological conditions and convergence occurring by chance under simple random-walk processes (). SURFACE does not consider the evolutionary correlations among variables, thus being unable to fit data in a multivariate way, therefore the model found by SURFACE was translated into the ‘mvMORPH’ package and tested along diverse alternative hypotheses in order to test if the SURFACE model was also the best adaptive explanation for the evolution of talar shape.It has been suggested that the talus has been shaped through habitat utilization within specific contexts – both locomotor and ecological – therefore being associated with the adaptive radiation suggested for platyrrhine evolution (). Using the platyrrhine phylogeny and talar shape and size data a series of evolutionary models were tested for congruence with the actual morphological data (). Model selection analyses were performed with the ‘mvMORPH’ package for R (), which allowed fitting several evolutionary models to trait data and a phylogeny in a multivariate framework. For each model, the relative fit was assessed using the AICc (). Several models were assessed, with BM as the simplest, while more complex models included early burst (EB) () as well as several adaptive OU models (). Under BM, trait evolution is simulated as a random walk through trait space, and phenotypic difference between sister taxa is expected to grow proportional to the sum of branch lengths between them (). Support for a BM model suggests that morphological disparity is uniformly increasing over time. In the EB model, the rates of Brownian evolution decays exponentially with time, thus representing niche-filling scenarios (). Support for the EB model suggests that most of the morphological disparity present in extant NWM was partitioned early in their evolutionary history and therefore provides weight to the LLH (). The OU model describes trait evolution under stabilizing selection, whereby there is attraction to a selective optimum; the strength of attraction to this selective optimum (i.e., the strength of selection) is measured using the α parameter (). Several OU models were constructed () to test if adaptive evolution could explain talar shape diversification. Each one of the proposed models represents an alternative biological hypothesis regarding the possible factors that might have influenced the adaptive landscape for platyrrhines. These models were based on different adaptive evolution hypotheses and ecological niches suggested for platyrrhine species (, , , , , ). Many of the analyzed models were derived and adapted from the work of , , however due to the fact that these models were generated to analyze different traits (i.e., brain shape and body mass), only those that were more general were applied, while others were not considered. In addition, other models specifically designed for talar morphology were generated.The first multi-peak model contained three separate optima that corresponded to the three platyrrhine families (OU-Clade), while the second was based on data concerning diet composition (OU-Diet Composition) and also had three optima (i.e., average annual percentages of plant parts and insects in the diets of platyrrhine genera) (). This diet model was considered because access to different diets requires differences in both locomotion and postural repertoire (). The third (OU-Locomotion A) was defined according to main locomotion categories and had three optima (clamber/suspensory, leaper/clawed and arboreal-quadrupedalism) (). Another locomotor model (OU-Locomotion B) similar to the previous one was tested, however in this one, only Callimico, Callithrix and Cebuella were considered within the leaper/clawed category, while the rest of the callitrichines were classified as arboreal quadrupeds based on the fact that they exhibited higher percentages of arboreal quadrupedal walking (). Additionally a third locomotor model (OU-Locomotion C) was designed by combining the OU-Locomotion A and the convergence result obtained from the SURFACE method; this model had four optima representing the three locomotor categories already mentioned, as well as one adaptive peak representing the convergence result found by SURFACE.Following , a multidimensional niche model was defined (OU-Multidimensional Niche) with five optima that combined diet and locomotion information (). Two other models were generated based on the main canopy level occupied by the different species analyzed. The first one (OU-Canopy A) had three different optima (understory, middle and upper), while the second (OU-Canopy B) had four optima, which were the same as the three previous ones, but included an additional optimum for Aotus, which has been observed occupying all canopy levels with relative frequency (). The canopy level classifications were performed using the data available in the Animal Diversity Web (ADW) of the University of Michigan (http://animaldiversity.org/) and . Different canopy levels are differentially structured, thus requiring different locomotor behaviors, therefore it was expected that these differences might impact on talar morphology.It is relevant to bear in mind that these different evolutionary models are generated to help in the understanding of possible underlying evolutionary processes, but they do not necessarily represent complete explanations (i.e., model selection is not an end in itself but a helpful approach in contributing to reasoning about the evolutionary mechanisms that might explain the observed variation in the analyzed traits) (). The different OU models based on different biological criteria were tested and their relative fit was assessed using AICc scores. In this manner, a measure of the relative explanatory power of each hypothesis (ΔAICc) was obtained. In addition to the OU models based on biological criteria, a single-peak OU model was also tested (if supported, that would suggest that there is a single, optimal talar shape for all of the platyrrhines), as well as a model representing the result obtained from the SURFACE method.A mean relative disparity-through-time (DTT) plot of the temporal pattern of change in relative talar shape disparity along the platyrrhine phylogeny was calculated using the first five PCs obtained from the shape PCA and also for centroid size (). Disparity was measured as D=∑(di)/n−1 where di is the pairwise Euclidean distance between species and n is the number of species. First, disparity was calculated for the entire platyrrhine clade, and then for each sub-clade. Disparity of each sub-clade was standardized by dividing it by the disparity of the entire clade (relative disparity sensu ). Such analyses allow comparison of the observed pattern of intra-clade versus among-clade disparity through time with a BM expectation. Therefore, high relative disparity values are a sign of extensive within-clade diversification and among-clade overlap, whereas values near 0 might imply that variation is mostly partitioned among clades (). The ‘geiger’ package for R () was used to generate DTT plots. […]

Pipeline specifications

Software tools Mesquite, PHYSIG, Phytools, GEIGER
Application Phylogenetics