Estimating Correlated Rates of Trait Evolution with Uncertainty.

L J Harmon,D S Caetano,Mark Holder

doi:10.1093/sysbio/syy067

L J Harmon, D S Caetano + Show 1 more

Open Access

https://doi.org/10.1093/sysbio/syy067

Copy DOI

Abstract

Correlated evolution among traits, which can happen due to genetic constraints, ontogeny, and selection, can have an important impact on the trajectory of phenotypic evolution. For example, shifts in the pattern of evolutionary integration may allow the exploration of novel regions of the morphospace by lineages. Here, we use phylogenetic trees to study the pace of evolution of several traits and their pattern of evolutionary correlation across clades and over time. We use regimes mapped to the branches of the phylogeny to test for shifts in evolutionary integration while incorporating the uncertainty related to trait evolution and ancestral regimes with joint estimation of all parameters of the model using Bayesian Markov chain Monte Carlo. We implemented the use of summary statistics to test for regime shifts based on a series of attributes of the model that can be directly relevant to biological hypotheses. In addition, we extend Felsenstein's pruning algorithm to the case of multivariate Brownian motion models with multiple rate regimes. We performed extensive simulations to explore the performance of the method under a series of scenarios. Finally, we provide two test cases; the evolution of a novel buccal morphology in fishes of the family Centrarchidae and a shift in the trajectory of evolution of traits during the radiation of anole lizards to and from the Caribbean islands. [Anolis; Centrarchidae; comparative methods; evolutionary integration; evolutionary rates; modularity; pruning algorithm.].

Highlights

Correlated evolution among traits, known as evolutionary integration, is ubiquitous across the tree of life and can have an important impact on the trajectory of phenotypic evolution (Olson and Miller, 1958; Klingenberg and Marugan-Lobon, 2013; Armbruster et al, 2014; Klin[32] genberg, 2014; Goswami et al, 2014, 2015; Melo et al, 2016)
Our implementation allows for multiple regime configurations and/or phylogenetic trees to be incorporated in the Markov chain Monte Carlo (MCMC) chain, integrating the uncertainty associated with ancestral state estimates and phylogenetic reconstruction to the analysis
The evolutionary correlation statistics correctly showed low proportion of support for a change in the evolutionary correlation with 3 traits (Table 1) whereas results with 6 traits only reflected the true parameters with trees of large size (Table 2). When changes in the pattern of integration were simulated in the data, results with 3 traits followed the same pattern as the shifts in rates of evolution—power increased with relative

Summary

Introduction

Correlated evolution among traits, known as evolutionary integration, is ubiquitous across the tree of life and can have an important impact on the trajectory of phenotypic evolution (Olson and Miller, 1958; Klingenberg and Marugan-Lobon, 2013; Armbruster et al, 2014; Klin[32] genberg, 2014; Goswami et al, 2014, 2015; Melo et al, 2016). The evolutionary rate matrix is ideal for studying 93 patterns of evolutionary integration because it allows for simultaneous estimate of the individ[94] ual rates of evolution of each trait as well as the evolutionary covariance between each pair of traits It is a flexible model, since any number of evolutionary rate matrix regimes can be fitted to the same phylogenetic tree (Revell and Collar, 2009). 119 Recently, Adams (2014b) described a method to estimate the rate of evolution under Brow[120] nian motion of traits defined by several dimensions (high-dimensional data), even when the number of trait dimensions exceeds the number of lineages in the phylogeny This method was extended to a plethora of variations based on the same general framework (Adams and Felice, 2014; Adams, 2014a; Denton and Adams, 2015; Adams and Collyer, 2017, see Goolsby (2016) for a different implementation). We provide results from extensive simulations showing that our approach has good performance under diverse scenarios of correlated evolution

Methods

Results

Discussion

Conclusion