Detecting natural selection at the molecular level: a reexamination of some "classic" examples of adaptive evolution.

An important criterion used to detect adaptive evolution in DNA sequence data is omega(i) > 1, where omega(i) is the ratio of nonsynonymous to synonymous substitution rates in lineage i. However, the evaluation of multiple omega(i) within a phylogenetic tree can easily inflate the statistical type I error rate. We developed two rigorous methods of analysis that avoid this and other potential pitfalls. We applied these methods to four published examples of adaptive evolution. One case was strongly supported by our reanalysis (abalone sperm lysin), and one was weakly supported (baboon alpha-globin), but two examples (primate lysozyme and Antarctic fish beta-globin) did not show significant evidence of adaptive evolution. Our first method is a "bottom-up" hierarchical maximum likelihood approach, which (1) tests for significant heterogeneity in omega across the phylogeny, (2) locates its source using a sequence of planned comparisons, and (3) tests homogeneous groups of omega for omega > 1, using a modified level of significance that incorporates the pretesting. The second method is a "top-down" log-linear analysis based on estimates of nonsynonymous and synonymous substitutions in pairs of lineages. The log-linear test is applied to pairs of lineages joined at progressively deeper nodes. For each pair, the analysis simultaneously tests for adaptive evolution (omega > 1), a shift in natural selection (omega1 does not = omega2), and unequal evolution rate (the relative rate test). In both tests, we emphasized that the criterion omega1 not equal omega2 is an important additional indicator of a phylogenetic shift in the balance between natural selection and genetic drift between two related lineages.