Multiscale selection in spatially structured populations

Abstract

The spatial structure of natural populations is key to many of their evolutionary processes. Formal theories analysing the interplay between natural selection and spatial structure have mostly focused on populations divided into distinct, non-overlapping groups. Most populations, however, are not structured in this way, but rather (self-)organise into dynamic patterns unfolding at various spatial scales. Here, we present a mathematical framework that quantifies how patterns and processes at different spatial scales contribute to natural selection in such populations. To that end, we define the Local Selection Differential (LSD): a measure of the selection acting on a trait within a given local environment. Based on the LSD, natural selection in a population can be decomposed into two parts: the contribution of local selection, acting within local environments, and the contribution of interlocal selection, acting among them. Varying the size of the local environments subsequently allows one to measure the contribution of each length scale. To illustrate the use of this new multiscale selection framework, we apply it to two simulation models of the evolution of traits known to be affected by spatial population structure: altruism and pathogen transmissibility. In both models, the spatial decomposition of selection reveals that local and interlocal selection can have opposite signs, thus providing a mathematically rigorous underpinning to intuitive explanations of how processes at different spatial scales may compete. It furthermore identifies which length scales--and hence which patterns--are relevant for natural selection. The multiscale selection framework can thus be used to address complex questions on evolution in spatially structured populations.