Fig 3. The effect of selfing on the fixation of compensatory mutations depends on the dominance coefficient in the double heterozygote (k_c), and the recombination rate(r).

(A) Schematic representations of the fitness landscapes for compensatory mutations with k_c = 0 (left) or k_c = 1 (right). A₁ and B₁ are the ancestral alleles. A₂ and B₂ are the derived alleles. (B) Time to fixation of a pair compensatory mutations with k_c = 0 (left) or k_c = 1 (right). The lines correspond to the analytical approximations (Eq 14) for r = 0 (black), r = 0.001 (dark grey), r = 0.01 (grey) and r = 0.1 (light grey). Dots show the corrected mean times to fixation of a pair of compensatory mutations from our two-locus simulations. We corrected the raw means because the threshold made our data right censored (i.e., missing estimates above 10⁹ generations), which we accounted for by first estimating the full distribution by fitting gamma distributions on our simulation outputs using the fitdistriplus R package [55], from which we estimated the corrected mean times to fixation. The dashed horizontal lines indicate the generation threshold after which simulations stop. N = 1, 000, μ = 10⁻⁵, h_c = 0.5, s_c = 0.01. 1, 000 iterations. (C) Population fitness (black dots—right y axis) and the frequencies of the 10 possible genotypes on the two loci fitness landscapes (solid lines—left y axis) over the last 4200 generations preceding the fixation of the pair of compensatory mutations. The line colours match the genotype colours on the fitness landscapes. N = 1, 000, μ = 10⁻⁵, h_c = 0.5, s_c = 0.01. 100 iterations. See S3 and S4 Figs for the visualisation of additional parameter combinations.