Trait macroevolution in the presence of covariates

Abstract

Statistical characterisations of traits evolving on phylogenies combine the contributions of unique and shared influences on those traits, potentially confusing the interpretation of historical events of macroevolution. The Fabric model, introduced in 2022, identifies historical events of directional shifts in traits (e.g. becoming larger/smaller, faster/slower over evolutionary time) and of changes in macroevolutionary 'evolvability' or the realised historical ability of a trait to explore its trait-space. Here we extend the model to accommodate situations in which the trait is correlated with one or more covarying traits. The Fabric-regression model identifies a unique component of variance in the trait that is free of influences from correlated traits, while simultaneously estimating directional and evolvability effects. We show in a dataset of 1504 Mammalian species that inferences about historical directional shifts in brain size and in its evolvability, having accounted for body size, differ qualitatively from inferences about brain size alone, including finding many new effects not visible in the whole trait. A class of fundamental macroevolutionary questions awaits testing on the variation uniquely attributable to traits, and the ability to accommodate statistically one or more covariates opens the possibility of bringing the formal methods of causal inference to phylogenetic-comparative studies.

Fig. 1. Outcomes of simple PGLS and Fabric-regression models.

a Observed brain size vs predicted brain size as derived from two models: magenta dots are predictions from a phylogenetic least squares regression model with linear and quadratic components (phylogenetically corrected R² ~ 0.86); yellow dots are predictions derived from the Fabric-regression model including directional and variance scaling effects and mostly fall closer to the regression line (phylogenetically corrected R² ~ 0.91); b residuals from the two models in (a) showing the narrower range of the residuals from the Fabric-regression model.

Fig. 2. Phylogenies showing placement of directional and evolvability effects.

a The position of directional effects in whole brain (red dots) and independent-brain (blue dots); b the position of changes to evolvability in whole (red squares) and independent brain size (blue squares). Black symbols in both panels indicate branches in which an effect occurred in the whole-brain and in the independent-brain.