“Evolution of the mutation spectrum across a mammalian phylogeny”

Abstract

Little is known about how the spectrum and etiology of germline mutagenesis might vary among mammalian species. To shed light on this mystery, we quantify variation in mutational sequence context biases using polymorphism data from thirteen species of mice, apes, bears, wolves, and cetaceans. After normalizing the mutation spectrum for reference genome accessibility and k-mer content, we use the Mantel test to deduce that mutation spectrum divergence is highly correlated with genetic divergence between species, whereas life history traits like reproductive age are weaker predictors of mutation spectrum divergence. Potential bioinformatic confounders are only weakly related to a small set of mutation spectrum features. We find that clocklike mutational signatures previously inferred from human cancers cannot explain the phylogenetic signal exhibited by the mammalian mutation spectrum, despite the ability of these clocklike signatures to fit each species' 3-mer spectrum with high cosine similarity. In contrast, parental aging signatures inferred from human de novo mutation data appear to explain much of the mutation spectrum's phylogenetic signal when fit to non-context-dependent mutation spectrum data in combination with a novel mutational signature. We posit that future models purporting to explain the etiology of mammalian mutagenesis need to capture the fact that more closely related species have more similar mutation spectra; a model that fits each marginal spectrum with high cosine similarity is not guaranteed to capture this hierarchy of mutation spectrum variation among species.

The evolution and convergence of mutation spectra across mammals

“…We highlight how the median of ratios method, introduced for normalising RNA-seq data, is also an effective approach for addressing the issue of normalising mutation spectra, so that they can be compared across species that exhibit different mutation counts. An alternate approach adopted in a recent study to correct for this is to downsample the variants to match the species with the lowest diversity 13 . However, as highlighted in this previous study, this method has the potential disadvantage of effectively adding noise to the estimates, and can lead to mutation counts of 0 following downsampling, effectively removing signals.…”

“…Some HAL files are publicly available, such as the 241 mammals from the zoonomia project 31 , or can be generated de novo using the progressive cactus alignment tool 32 . Although other studies of mutation spectra have used different approaches to determine ancestral alleles, such as the est-sfs approach 13,33 , an important disadvantage of such approaches is that they rely on variant allele frequencies, that are in turn dependent on the availability of good quality population genomics data for the focal group. In contrast, the approach implemented in nSPECTRa only requires the availability of reference genomes, that are increasingly available for most species, and can characterise ancestral alleles across both polymorphic and fixed sites in the focal species.…”

“…This effect is known to be skewed towards increasing the frequency of C:G base pairs at the expense of A:T base pairs. However, recent studies have indicated that BGC is not likely to be a major driver of the observed differences in mutation spectra observed between individuals 13 .…”