On the number of species required to detect stabilizing selection in gene expression evolution

Abstract

Differentiating between genes that evolve under neutral drift versus stabilizing selection is one of the first explorations in comparative phylogenomic studies. Neutral drift and stabilizing selection of gene expression levels across a phylogeny are modelled using a Brownian motion and an Ornstein-Uhlenbeck process, respectively. Most empirical gene expression datasets contain only a few species, raising concerns about statistical power for model selection and parameter estimation. Furthermore, gene expression datasets often contain multiple samples per species, contrary to common macroevolutionary analyses of trait evolution. Here, we designed a targeted simulation study to evaluate the power to identify stabilizing selection, emphasizing the impact of the number of species, strength of selection and within-species variances. Our results show that the accuracy of selection strength estimation improves with increasing phylogeny size. Distinguishing weak selection, i.e., lower selection strength, from neutral drift is challenging, whereas varying levels of within-species variance have only minor effects on parameter estimation and model selection. Overall, our results suggest that we can have sufficient power to detect stabilizing selection with approximately 17 species, if the underlying strength of selection is sufficiently strong and if a relatively relaxed statistical significance threshold is adopted.