3.2 Genetic variation in S-locus genes and their paralogs
Our results showed that, as expected due to S-locus hemizygosity, πS was lower in S-locus genes of thrums (0.0012 ± 0.0006 [mean ± SE]) than in their paralogs in pins and thrums (0.0034 ± 0.0008; Table 3A). Moreover, πS in homostyles was zero for both S-locus genes and their paralogs, except forGLOT and CYP734A51 , where πS was extremely low (0.0012 and 0.0008, respectively; Table 3B), thus supporting the prediction that the shift to predominant selfing should be associated with lower πS in homostyles than in heterostyles.
Furthermore, our results indicated that, on average, πNS values were higher for S-locus genes of thrums than for their paralogs in pins and thrums (1.01 ± 0.37 vs. 0.53 ± 0.29, respectively; Table 3A), implying lower purifying selection in S-locus genes. Additionally, πNS was lower in KFB1 than inKFBT NS = 0.23 and 1.83, respectively), but higher in CCM1 than inCCMTNS = 1.5 and 0.91, respectively; Table 3A), implying stronger and weaker purifying selection on the two paralogs than on their respective S-locus genes, respectively. Finally, we found that, within the S-locus, πNS was higher forCCMT , KFBT , andPUMTNS = 0.91, 1.83, and 10.36, respectively; Table 3A) than forCYPTNS = 0.28), whereas πNS inGLOT was not calculated due to the lack of variation at synonymous sites in this gene. In homostyles, πNS was zero for most S-locus genes and their paralogs, except for CYP734A51 , due to the lack variation at synonymous and non-synonymous sites (Table 3B).
3.3 S- locus genotypes in natural populations of P. vulgaris
Relative S-locus sequencing depth (RelS-locus depth ) allowed us to determine S-locus ploidy in the analyzed thrums and homostyles. Of the 37 thrums collected from six dimorphic and two trimorphic populations across our sampling range, 34 had a haploid S-locus (i.e., S/0) and three had a diploid S-locus (S/S; Figure 4). Of these three thrums, two were homozygous for the functional copy of CYPT(CYPT-1/CYPT-1 ; i.e., S/S) and belonged to one trimorphic and one dimorphic population each, respectively (Table 1), while one thrum was heterozygous and carried one functional and one disrupted copy of CYPT(CYPT-1/CYPT-2 ; i.e., S/S*). Of the 31 homostyles collected from two trimorphic (EN4-T and EN5-T) and one monomorphic population (EN6-M) in England, 10 (32%) had a haploid S-locus (S*/0) and 21 (68%) had a diploid S-locus (S*/S*) (Figure 4). Specifically, S*/0- homostyles represented 40% and 60% of homostyles in the trimorphic populations EN4-T and EN5-T, respectively, while all tested homostyles of the monomorphic population EN6-M had the S*/S*- genotype (Table 1; purple triangles in Figure 4).
We also calculated expected frequencies and observed frequencies of 0/0, S/0-, S*/0-, and S*/S*-genotypes under different assumptions for viability of S*/S*-genotypes compared to the other three genotypes and after different numbers of generations following the origin of homostyles (Table 4). The results of chi-squared tests showed that non-significant differences were found only in seven cases, of which four occurred in the monomorphic, homostylous population EN6-M and three in the two trimorphic populations EN5-T and EN4-T. Specifically, in EN6-M, observed frequencies matched expected frequencies at generations 30 and 40 under the assumption of equal and slightly lower viability (v = 1 and 0.9, respectively) for the S*/S*-homostyles. In EN5-T, observed frequencies matched expected frequencies at 20 generations under lower viability (v = 0.8) of S*/S*-homostyles. Finally, in EN4-T, observed frequencies matched expected genotypic frequencies at generations 20 and 30 under levels of viability for S*/S*-homostyles close or equal to those of the model in Figure 1C (i.e., v = 0.70 and 0.65, respectively).