Repeated adaptive introgression at a gene under multiallelic balancing selection

Abstract

Recently diverged species typically have incomplete reproductive barriers, allowing introgression of genetic material from one species into the genomic background of the other. The role of natural selection in preventing or promoting introgression remains contentious. Because of genomic co-adaptation, some chromosomal fragments are expected to be selected against in the new background and resist introgression. In contrast, natural selection should favor introgression for alleles at genes evolving under multi-allelic balancing selection, such as the MHC in vertebrates, disease resistance, or self-incompatibility genes in plants. Here, we test the prediction that negative, frequency-dependent selection on alleles at the multi-allelic gene controlling pistil self-incompatibility specificity in two closely related species, Arabidopsis halleri and A. lyrata, caused introgression at this locus at a higher rate than the genomic background. Polymorphism at this gene is largely shared, and we have identified 18 pairs of S-alleles that are only slightly divergent between the two species. For these pairs of S-alleles, divergence at four-fold degenerate sites (K = 0.0193) is about four times lower than the genomic background (K = 0.0743). We demonstrate that this difference cannot be explained by differences in effective population size between the two types of loci. Rather, our data are most consistent with a five-fold increase of introgression rates for S-alleles as compared to the genomic background, making this study the first documented example of adaptive introgression facilitated by balancing selection. We suggest that this process plays an important role in the maintenance of high allelic diversity and divergence at the S-locus in flowering plant families. Because genes under balancing selection are expected to be among the last to stop introgressing, their comparison in closely related species provides a lower-bound estimate of the time since the species stopped forming fertile hybrids, thereby complementing the average portrait of divergence between species provided by genomic data.

Figure 1. Phylogeny of the 68 SRK sequences of A. lyrata and A. halleri.

The phylogeny was obtained by the neighbour-joining method on pairwise proportion of nucleotide divergence after Jukes-Cantor's correction. Brackets indicate interspecific pairs of sequences assumed to represent “trans-specifically shared S-alleles”, i.e. alleles assumed to have evolved from a single S-allele in the direct ancestor of A. lyrata and A. halleri.

Figure 2. Distribution of the number of pairwise nucleotide differences for SRK sequences in interspecific comparisons between A. halleri and A. lyrata.

Note the distinct peak of highly similar sequences observed. The vertical arrow represents the chosen threshold to define “trans-specifically shared” pairs of sequences (≤12 nucleotide differences). Note also that the two pairs of sequences with intermediate nucleotide differences (45 between AlSRK03 and AhSRK28, and 50 between AlSRK28 and AhSRK03) cannot represent trans-specifically shared S-alleles because they are not monophyletic (see Figure 1).

Figure 3. Divergence process between Arabidopsis lyrata, A. halleri and A. thaliana at unlinked genes (genomic background) and trans-specifically shared pairs of S-alleles.

Divergence times were taken from Koch et al. ,. θ_lyrata, θ_halleri and θ_A, refer to polymorphism (θ = 4Nμ· in A. lyrata, A. halleri and their common ancestor. As compared to unlinked genes, divergence between trans-specifically shared S-alleles is affected by two confounding factors: (1) lower effective population size than the genomic background reducing coalescence time in the ancestral species, and (2) expected higher introgression as represented by thicker dark grey arrows.

Figure 4. Predicted nucleotide divergence between A. halleri and A. lyrata for the genomic background (grey line), S-alleles with the same rate of introgression as the genomic background (dotted line) and S-alleles with 5-fold increased rate of introgression relative to the genomic background (black line).

10,000 coalescent simulations were performed for each case using maximum likelihood parameter estimates obtained under the “isolation with migration” model, except for the dotted line, where the 2.5% low ancestral population size estimate was used in order to be conservative. Observed nucleotide divergence for the genomic background and S-alleles are represented by grey and black stars on the x-axis, respectively.