Résumé
Two current goals of speciation research are to understand whether genomic barriers to gene flow persist after contact between divergent lineages, and to identify the loci underlying such reproductive barriers. Locating barrier loci in empirical data is difficult due to the necessity of understanding the often complex demographic history of focal taxa and the heterogeneity in evolutionary forces across the genome. Furthermore, when identified, barrier loci are usually studied in isolation of one another, yet theoretical models predict an important role for coupling or epistasis in barrier persistence. In one classical model of epistasis, the Bateson-Dobzhansky-Muller (BDMI) model, incompatibilities are thought to be easily eroded by gene flow. However, this is rarely tested. Here we show empirically that BDMIs are persistent barriers when speciation happens with gene flow. We take advantage of the Formica hybrid wood ant system and recent methodological developments to (i) map barrier loci showing reduced migration during divergence between two species with the demographically explicit scan gIMble. This analysis used four whole-genome sequences from each parental species (Formica aquilonia and F. polyctena) and (ii) identify candidate BDMIs in naturally occurring hybrids through an imbalanced recombinant haplotype frequency analysis using the X(2) statistic. The X(2) method is related to linkage disequilibrium methods in that it looks for non-random associations between alleles, specifically missing haplotype combinations at loci of known ancestry. Here we used 386 whole-genome sequences from a single hybrid population in Finland. We find that a large number of both barriers and BDMIs are scattered across the genome, rather than a small number of clusters. Furthermore, candidate BDMIs partly overlap with the long-term barrier regions identified by gIMble. Intriguingly, candidate BDMIs show interconnectivity, and the number of pairwise interactions a BDMI has impacts how effectively it reduces gene flow: BDMIs with many pairwise interactions to other BDMIs tend to reduce gene flow more effectively than those with fewer interactions. Finally in regards to function, long-term barriers identified by gIMble arise outside regions of both gene coding sequences (CDS) and transposable elements. In contrast, regions where long-term barriers and BDMIs co-locate are significantly associated with introns. Our results draw attention to an underappreciated aspect of incompatibilities: the structure of the network in which they interact, which has evolutionary consequences. Furthermore, it suggests persistent BDMI loci may have specific molecular characteristics, for example acting as hub genes in molecular pathways. The overlap between candidate BDMIs and regions of no gene flow between species indicates that incompatibilities have persisted despite divergence with gene flow between the wood ant species. Finding that long-term BDMIs are significantly associated with intronic regions implies a potential role of alternative splicing or gene regulation in incompatibilities, rather than CDS divergence. Overall, our results highlight the underappreciated impact of multilocus BDMIs and the need to consider network connectivity of BDMIs in future work.