Stochastic Character Mapping of Continuous Traits on Phylogenies

Living and fossilized organisms represent only a tiny fraction of Earth’s evolutionary history, motivating “ancestral state reconstruction” techniques that aim to infer the unobserved phenotypes of evolving lineages based on measurements of their relatives. Stochastic character mapping (“simmapping”) methods perform ancestral state reconstruction by randomly sampling maps (“simmaps”) of probable phenotypic evolutionary histories along phylogenies, allowing researchers to conveniently and flexibly analyze macroevolutionary patterns and processes while accounting for the inherent uncertainty of ancestral state estimates. Here, we introduce a flexible and efficient algorithm for simmapping continuous phenotypes evolving under Brownian Motion models, which we term continuous simmaps or “contsimmaps”, thereby generalizing existing simmapping methods which only work with discrete phenotypes. To demonstrate potential applications of contsimmaps, we develop a pipeline that uses contsimmaps to test for associations between rates of continuous trait evolution and continuously-varying factors (e.g., generation time, climatic niche)–a difficult statistical problem for which few methods are currently available. Through an extensive simulation study, we show that this novel pipeline can accurately and robustly infer factor-rate relationships from phylogenetic comparative data, albeit with low power under certain conditions. Lastly, we apply this pipeline to an empirical dataset, showing that rates of leaf and flower trait evolution are highly variable yet unrelated to height in a clade of eucalyptus trees spanning roughly 1 to 100 meters in maximum height. Ultimately, contsimmaps provide a valuable new tool for macroevolutionary biology by allowing researchers to more flexibly analyze the evolutionary dynamics of continuous phenotypes and test complex evolutionary hypotheses involving continuous variables.