Evolution under strong balancing selection: how many codons determine specificity at the female self-incompatibility gene SRK in Brassicaceae?

Abstract

Background: Molecular lock-and-key systems are common among reproductive proteins, yet their evolution remains a major puzzle in evolutionary biology. In the Brassicaceae, the genes encoding self-incompatibility have been identified, but technical challenges currently prevent detailed analyses of the molecular interaction between the male and female components. In the present study, we investigate sequence polymorphism in the female specificity determinant SRK of Arabidopsis halleri from throughout Europe. Using a comparative approach based on published SRK sequences in A. lyrata and Brassica, we track the signature of frequency-dependent selection acting on these genes at the codon level. Using simulations, we evaluate power and accuracy of our approach and estimate the proportion of codon sites involved in the molecular interaction.

Results: We identified several members of the S-gene family, together with 22 putative S-haplotypes. Linkage to the S-locus and the presence of a kinase domain were formally demonstrated for four and six of these haplotypes, respectively, and sequence polymorphism was extremely high. Twenty-five codons showed signs of positive selection in at least one species, and clustered significantly (but not exclusively) within hypervariable regions. We checked that this clustering was not an artifact due to variation in evolution rate at synonymous sites. Simulations revealed that the analysis was highly accurate, thus providing a reliable set of candidates for future functional analyses, but with an overall power not higher than 60 %. Assuming similar power, we infer from our results that about 23% of all codons in the S-domain may actually be involved in recognition. Interestingly, while simulations demonstrated that this comparison remained reliable even at very high levels of divergence, codons identified in Brassica had higher posterior rates of non-synonymous to synonymous substitutions than codons identified in A. halleri or A. lyrata, possibly suggesting more intense selection in Brassica.

Conclusion: The signature of balancing selection can be identified reliably at the codon level even in cases of very high sequence divergence, provided that a sufficiently large set of sequences are analyzed. Altogether, our results indicate that a large proportion of codons may be involved in recognition and confirm the particular importance of hypervariable regions. The more intense signature of positive selection detected in Brassica suggests that allelic diversification in this genus was very recent, possibly following a recent demographic bottleneck.

Figure 1

Phylogeny of SRK alleles in A. halleri, A. lyrata and Brassica. A: 50% majority rule consensus phylogeny obtained independently for each taxon using MrBayes. B: Combined maximum likelihood phylogeny, used to simulate sequence evolution in the Evolver program. Note the scale difference between the single-species and the combined trees. Allele names have been abbreviated for clarity.

Figure 2

Sliding window analyses of amino acid diversity for A. halleri, A. lyrata and Brassica. The window was 5 amino acid wide and sled with a step of 1 amino acids. Horizontal black bars represent the three hypervariable regions HV2, HV3 and CVR.

Figure 3

Intron-exon structure of SRK and sites identified as evolving under positive selection in Brassica, A. halleri and A. lyrata. Sites 270–306, 328–343, and 413–423 correspond to hypervariable regions HV2, HV3, and CVR respectively. Codons identified by more than a single species are figured in bold.

Figure 4

Effect of sequence divergence on posterior ω. Posterior ω values were computed for cases where at least one site had a high (>0.95) probability of belonging to the ω >1 category. Results are averages over 5 independent replicates. Open circles represent situations where no site was identified in any of the five replicates. Despite a slight underestimation of posterior ω values (parametric value = 1.754; horizontal dashed line), there is no evidence that saturation of divergence alone could have caused the lower ω found in A. halleri and A. lyrata than in Brassica.

Figure 5

Effect of sequence divergence on power to detect sites evolving under positive selection. In the simulations, an average of 19 sites were assigned to evolve under positive selection (see text for simulation parameters). The level of sequence divergence estimated from the combined A. halleri, A. lyrata and Brassica sequences is 23.47 substitutions per codon. Results are averages over 5 independent replicates. The dashed vertical line represents the observed level of divergence estimated from the analysis of the 3 taxa combined.