Estimating the parameters of background selection and selective sweeps in Drosophila in the presence of gene conversion

Abstract

We used whole-genome resequencing data from a population of Drosophila melanogaster to investigate the causes of the negative correlation between the within-population synonymous nucleotide site diversity (π_S ) of a gene and its degree of divergence from related species at nonsynonymous nucleotide sites (K_A ). By using the estimated distributions of mutational effects on fitness at nonsynonymous and UTR sites, we predicted the effects of background selection at sites within a gene on π_S and found that these could account for only part of the observed correlation between π_S and K_A We developed a model of the effects of selective sweeps that included gene conversion as well as crossing over. We used this model to estimate the average strength of selection on positively selected mutations in coding sequences and in UTRs, as well as the proportions of new mutations that are selectively advantageous. Genes with high levels of selective constraint on nonsynonymous sites were found to have lower strengths of positive selection and lower proportions of advantageous mutations than genes with low levels of constraint. Overall, background selection and selective sweeps within a typical gene reduce its synonymous diversity to ∼75% of its value in the absence of selection, with larger reductions for genes with high K_A Gene conversion has a major effect on the estimates of the parameters of positive selection, such that the estimated strength of selection on favorable mutations is greatly reduced if it is ignored.

Keywords: Drosophila melanogaster; background selection; gene conversion; selective sweeps; sequence diversity.

Fig. 1.

The plot of synonymous diversity (π_S) for genes in a Rwandan population of D. melanogaster against their nonsynonymous divergence from D. yakuba (K_A); ρ is the Spearman rank correlation coefficient. The green line is the least-squares linear regression (the dashed lines are its 95% CIs).

Fig. 2.

This plots the theoretical values of mean E (percent) against values of mean ω_na (percent) for the standard model of a single gene with five exons of 100 codons each; a gamma distribution of selection coefficients with β = 0.3 was assumed, with γ_c = 5. For the results obtained by the summation method (red and blue solid lines), the exons were separated by four introns of 100 bp. For the results obtained from the integral model (black and green dashed lines), a continuous stretch of coding sequence was assumed. The green and blue lines show the net BGS effects arising from both NS and UTR sites; the black and red lines show the effects for NS sites alone. Two-thirds of coding sites were assumed to result in NS mutations. The rate of crossing over per base pair was 1 × 10⁻⁸, and the mutation rate was 4.5 × 10⁻⁹ per base pair. The gene conversion parameters for the low gene conversion case (A) were g_c = 1 × 10⁻⁸ and d_g = 440; for the high gene conversion case (B), g_c = 5 × 10⁻⁸ and d_g = 500. No large effect mutations were allowed.

Fig. 3.

The black diamonds are the observed values of −ln(π_S) for each bin of K_A values for autosomes, corrected for the correlation between π_S and K_S as described in Materials and Methods, Primary Data Analyses. The circles are the theoretical values of mean E for each bin, obtained by the integral model of BGS, assuming a single gene with 500 NS sites. The crosses are the predicted values of −ln(π_S) for each bin, given by the combined BGS and SSW models at NS and UTR sites. Red and blue correspond to the low and high gene conversion rates used in Fig. 2. The mutation rate and crossing-over parameters are as in Fig. 2, except that large effect mutations constitute 15% of all mutations, with a selection coefficient against heterozygotes of 0.044.

Fig. 4.

The black circles are the predicted values of π_S relative to its expected value in the absence of hitchhiking; the red diamonds are the estimates of γ_a (multiplied by 10⁻³); the blue crosses are the estimates of p_a (multiplied by 10³). The other parameters are as in Fig. 3, assuming effects of BGS and SSWs at both NS and UTR sites.