Is the Mutation Rate Lower in Genomic Regions of Stronger Selective Constraints?

Abstract

A study of the plant Arabidopsis thaliana detected lower mutation rates in genomic regions where mutations are more likely to be deleterious, challenging the principle that mutagenesis is blind to its consequence. To examine the generality of this finding, we analyze large mutational data from baker's yeast and humans. The yeast data do not exhibit this trend, whereas the human data show an opposite trend that disappears upon the control of potential confounders. We find that the Arabidopsis study identified substantially more mutations than reported in the original data-generating studies and expected from Arabidopsis' mutation rate. These extra mutations are enriched in polynucleotide tracts and have relatively low sequencing qualities so are likely sequencing errors. Furthermore, the polynucleotide "mutations" can produce the purported mutational trend in Arabidopsis. Together, our results do not support lower mutagenesis of genomic regions of stronger selective constraints in the plant, fungal, and animal models examined.

Keywords: Arabidopsis; human; mutation; natural selection; yeast.

Fig. 1.

Relationships among AT content, density of polynucleotides, and d_N/d_S in Arabidopsis thaliana. (A) Pearson's correlation (r) and Spearman's correlation (ρ) between the coding region AT content and d_N/d_S across genes. Each dot represents a gene. The blue line is the linear regression. (B) Correlations between the number of poly(A) + poly(T) tracts per 1000 nucleotides (i.e., density) in the coding sequence (CDS) of a gene and its d_N/d_S. Each dot represents a gene. The blue line is the linear regression. (C) AT content in coding, intron, and intergenic regions, respectively. Errors are too small to present. (D) Mean density of poly(A) + poly(T) tracts in coding, intron, and intergenic regions, respectively. (E) Number of mutations per site in coding, intron, and intergenic regions, respectively, calculated using the sum of the three datasets in Monroe et al. (2022). (F) Correlations between the no. of mutations per site generated by simulation and d_N/d_S across genes. Each dot represents a gene. In the simulation, 70% of mutations are randomly distributed at non-polynucleotide sites across the genome while the remaining mutations are randomly distributed among poly(A) and poly(T) tracts. A similar Pearson's correlation is observed upon log transformations of the data. (G) Numbers of simulated mutations per site in coding, intron, and intergenic regions, respectively. In (D), (E), and (G), error bars represent 95% confidence intervals predicted by Poisson distributions.