How well do evolutionary trees describe genetic relationships among populations?

Abstract

Bifurcating evolutionary trees are commonly used to describe genetic relationships between populations, but may not be appropriate for populations that did not evolve in a hierarchical manner. The degree to which bifurcating trees distort genetic relationships between populations can be quantified with R(2), the proportion the variation in a matrix of genetic distances between populations that is explained by a tree. Computer simulations were used to measure how well the unweighted pair group method with arithmetic mean (UPGMA) and neighbor-joining (NJ) trees depicted population structure for three evolutionary models: a hierarchical model of population fragmentation, a linear stepping-stone model of gene flow and a two-dimensional stepping-stone model of gene flow. These simulations showed that the UPGMA did an excellent job of describing population structure when populations had a bifurcating history of fragmentation, but severely distorted genetic relationships for the linear and two-dimensional stepping-stone models. The NJ algorithm worked well in a broader range of evolutionary histories, including the linear stepping-stone model. A computer program for performing the calculations described in this study is available for download at www.montana.edu/kalinowski.