All genes interact with other genes, and their additive effects and epistatic interactions affect an organism's phenotype and fitness. Recent theoretical and empirical work has advanced our understanding of the role of multi-locus interactions in speciation. However, relating different models to one another and to empirical observations is challenging. This review focuses on multi-locus interactions that lead to reproductive isolation (RI) through reduced hybrid fitness. We first review theoretical approaches and show how recent work incorporating a mechanistic understanding of multi-locus interactions recapitulates earlier models, but also makes novel predictions concerning the build-up of RI. These include high variance in the build-up rate of RI among taxa, the emergence of strong incompatibilities producing localized barriers to introgression, and an effect of population size on the build-up of RI. We then review recent experimental approaches to detect multi-locus interactions underlying RI using genomic data. We argue that future studies would benefit from overlapping methods like ancestry disequilibrium scans, genome scans of differentiation and analyses of hybrid gene expression. Finally, we highlight a need for further overlap between theoretical and empirical work, and approaches that predict what kind of patterns multi-locus interactions resulting in incompatibilities will leave in genome-wide polymorphism data. This article is part of the theme issue ‘Towards the completion of speciation: the evolution of reproductive isolation beyond the first barriers’.

New species are formed when populations become reproductively isolated, that is they no longer interbreed or exchange genetic material (Mayr, 1970). However, until complete reproductive isolation has evolved, gene flow may still occur between the diverging lineages. This genetic exchange is usually localized in the genome, with some regions resisting gene flow better than others. Such regions resistant to gene flow are thought to harbour "barrier loci" that can drive the divergence between lineages despite the homogenizing effect of gene flow (Ravinet et al., 2017). Barrier loci could contribute to reproductive isolation by allowing differential adaptation between diverging lineages (Ravinet et al., 2017). Alternatively, barrier loci could consist of genes incompatible between the diverging lineages (Dobzhansky-Muller incompatibilities, DMIs) (Dobzhansky, 1936; Muller, 1942) owing to drift or divergent selection (Kulmuni & Westram, 2017). Several recent studies have searched for barrier loci using genome scans and identified genomic islands of differentiation using various measures of genetic divergence (e.g., Martin

16While speciation underlies novel biodiversity, it is poorly understood how natural 17 selection shapes genomes during speciation. Selection is assumed to act against gene 18 flow at barrier loci, promoting reproductive isolation and speciation. However, 19 evidence for gene flow and selection is often indirect. Here we utilize haplodiploidy 20 to identify candidate barrier loci in hybrids between two wood ant species and 21 integrate survival analysis to directly measure if natural selection is acting at 22 candidate barrier loci. We find multiple candidate barrier loci but surprisingly, 23 proportion of them show leakage between samples collected ten years apart, natural 24 selection favoring leakage in the latest sample. Barrier leakage and natural selection 25 for introgressed alleles could be due to environment-dependent selection, 26 emphasizing the need to consider temporal variation in natural selection in future 27 speciation work. Integrating data on survival allows us to move beyond genome 28 scans, demonstrating natural selection acting on hybrid genomes in real-time. 29 30 New species are formed when populations become reproductively isolated, i.e. they 68 do not interbreed and exchange genetic material [1]. However, until complete 69 reproductive isolation has evolved gene flow may still occur between the diverging 70 lineages. This genetic exchange is usually localized in the genome, with some regions 71 of the genome resisting gene flow better than others. Regions resistant to gene flow 72 are thought to harbor "barrier loci" that can drive the divergence between lineages 73 despite the homogenizing effect of gene flow [2]. Barrier loci could contribute to 74 reproductive isolation by allowing differential adaptation between diverging lineages 75 [2]. Alternatively, barrier loci could consist of genes incompatible between the 76 diverging lineages (Dobzhansky-Muller incompatibilities [3,4], DMIs) driven by drift 77 or divergent selection [5]. 78 79 Several recent studies have searched for barrier loci using genome scans and 80 identified genomic islands of differentiation using various measures of genetic 81 differentiation [e.g. refs 6-8]. The underlying assumption in these genome scans is 82 that the most differentiated genomic regions between hybridizing lineages are those 83 that are resistant to gene flow and hence drive reproductive isolation. Genetic 84 differentiation however is an indirect measure of gene flow and several problems with 85 genome scans have been discussed [e.g. refs 9, 10]. The main problem lies in the fact 86 that several other evolutionary forces (such as low recombination rate or drift) can 87 bring about genomic regions of seemingly high differentiation between populations. 88Thus, we need more studies that test if natural selection is promoting speciation and 89 halting gene flow at these genomic regions in natural populations. 90 91 Here we use ants as a model system to discover genomic regions of divergence (i.e. 92 putative barrier loci) and to test for natural se...

Hybridisation and gene flow can have both deleterious and adaptive consequences for natural populations and species. To better understand the extent and consequences of hybridisation in nature, information on naturally hybridising non-model organisms is required, including characterising the structure and extent of natural hybrid zones. Here we study natural populations of five keystone mound-building wood ant (Formica rufa group) species across Finland. No genomic studies across the species group exist and the extent of hybridisation and genomic differentiation in sympatry is unknown. Combining genome-wide and morphological data, we show that Formica rufa, F. aquilonia, F. lugubris, and F. pratensis form distinct gene pools in Finland. We demonstrate more extensive hybridisation than previously thought between all five species and reveal a mosaic hybrid zone between F. aquilonia, F. rufa and F. polyctena. We show that hybrids between these climatically differently adapted species occupy warmer habitats than the cold-adapted parent F. aquilonia. This suggests hybrids occupy a different microclimatic niche compared to the locally abundant parent. We propose that wood ant hybridisation may increase with a warming climate, and warm winters, in particular, may provide a competitive advantage for the hybrids over F. aquilonia in the future. In summary, our results demonstrate how extensive hybridisation may help persistence in a changing climate. Additionally, they provide an example on how mosaic hybrid zones can have significant ecological and evolutionary consequences because of their large extent and independent hybrid populations that face both ecological and intrinsic selection pressures.

Hybridisation and gene flow can have both deleterious and adaptive consequences for natural populations and species. To better understand the extent of hybridisation in nature and the balance between its beneficial and deleterious outcomes in a changing environment, information on naturally hybridising nonmodel organisms is needed. This requires the characterisation of the structure and extent of natural hybrid zones. Here, we study natural populations of five keystone mound‐building wood ant species in the Formica rufa group across Finland. No genomic studies across the species group exist, and the extent of hybridisation and genomic differentiation in sympatry is unknown. Combining genome‐wide and morphological data, we demonstrate more extensive hybridisation than was previously detected between all five species in Finland. Specifically, we reveal a mosaic hybrid zone between Formica aquilonia, F. rufa and F. polyctena, comprising further generation hybrid populations. Despite this, we find that F. rufa, F. aquilonia, F. lugubris and F. pratensis form distinct gene pools in Finland. We also find that hybrids occupy warmer microhabitats than the nonadmixed populations of cold‐adapted F. aquilonia, and suggest that warm winters and springs, in particular, may benefit hybrids over F. aquilonia, the most abundant F. rufa group species in Finland. In summary, our results indicate that extensive hybridisation may create adaptive potential that could promote wood ant persistence in a changing climate. Additionally, they highlight the potentially significant ecological and evolutionary consequences of extensive mosaic hybrid zones, within which independent hybrid populations face an array of ecological and intrinsic selection pressures.