Abstract
Mutation rate variation has the potential to bias evolutionary inference, particularly when rates become much higher than the mean. We first confirm prior work that inferred the existence of cryptic, site-specific rate variation on the basis of coincident polymorphisms-sites that are segregating in both humans and chimpanzees. Then we extend this observation to a longer evolutionary timescale by identifying sites of coincident substitutions using four species. From these data, we develop analytic theory to infer the variance and skewness of the distribution of mutation rates. Even excluding CpG dinucleotides, we find a relatively large coefficient of variation and positive skew, which suggests that, although most sites in the genome have mutation rates near the mean, the distribution contains a long right-hand tail with a small number of sites having high mutation rates. At least for primates, these quickly mutating sites are few enough that the infinite sites model in population genetics remains appropriate.