Mutations in clusters and showers

Several years ago, mutants arising spontaneously in a mutation-reporter sequence in a mouse were noted to contain more mutants bearing two or more mutations than was predicted by the mutation frequency and the assumption of a random distribution of mutations among mutants (1). Many of these overloaded mutants contained two mutations that were moderately close to each other. In this issue of PNAS, Wang et al. (2) describe the results of a robust sequencing effort to examine the same mutants over many kilobases outside of the original mutation reporter, and they find that some of the mutants contain even more mutations farther out in genetic space.

Those who write about mutation usually either explicitly state, or else imply, that it is a random process. However, although mutations are poorly predictable, they have already been observed to be nonrandom in one important way: from the earliest decades of mutation research (the 1920s through the 1940s), mutation rates were observed to vary greatly across different eukaryotic genes. In the middle of this period, when physicists were beginning to speculate about the fundamental properties of hereditary material, target theory was adapted from atomic physics to estimate gene sizes by using the assumed randomness of mutagenic radiation “hits” (3), and the results were modeled with the Poisson distribution. Use of the Poisson to model mutagenesis persists to the present and underlies such key methods as the Luria–Delbrück fluctuation test, one of the gold standards for estimating mutation rates (4).

In the 1950s, a prepared mind and an obscure comment in a thesis led Seymour Benzer to develop the powerful T4 rII system to estimate the fundamental parameters of the gene on a scale approaching that of DNA base pairs (5). One of his most enduring findings was the discovery of hugely variable site-specific …

*E-mail: drake{at}niehs.nih.gov