Identifying the loci of speciation: the challenge beyond genome scans

In their review on the genomic landscape of speciation, Ravinet et al. (in press) highlight difficulties when attempting to draw inferences about speciation based on heterogeneous patterns of genome differentiation. These problems arise because various factors that are either causally unlinked or only spuriously associated with speciation can induce genomic heterogeneity and thus complicate the interpretation of genome scans. In the light of these difficulties, it is important to not restrict speciation research to genome scans but to view them as an additional tool that can be applied complementary to more traditional methodologies, for example field observations, experiments and crosses. In our opinion, interpreting the genomic landscape of speciation faces additional difficulties that were not addressed by Ravinet et al. (in press), mainly due to limitations in our understanding of whether genomic patterns caused by processes involved in speciation can be differentiated from those associated with adaptation within lineages. While the latter likely play an important role in many cases of speciation (Gavrilets, 2003; Schluter & Conte, 2009; Nosil, 2012), in other cases, they might not (Presgraves, 2010; Maheshwari & Barbash, 2011). We thus argue that reduced gene flow attributable to divergent adaptation, and potentially noticeable in genome scans, might contain limited information about speciation, as it is difficult to predict the future role of adaptive alleles in facilitating the evolution of reproductive isolation. This parallels the notion that current isolating barriers may be uninformative about their past importance during speciation (Coyne & Orr, 2004). The unpredictable future holds for any isolation barrier, but specifically so for those solely relying on environmental factors (e.g. Seehausen et al., 2008). We thus focus here on whether genome scans can identify barriers to gene flow that are potentially able to keep species separated in the long run and that may be less susceptible to or independent of environmental changes, for example intrinsic hybrid incompatibilities or mating preferences (Seehausen et al., 2014). While environmentally independent processes that potentially advance speciation (e.g. genomic conflict or assortative mating) can differ from processes involved in divergent adaptation between lineages, their genomic signatures may be difficult to distinguish in an empirical context. However, it is critical to understand the relative importance and interaction of different classes of barriers for the evolution of reproductive isolation and speciation. ‘Pleiotropy and (to a lesser extent) epistasis are key features of speciation. While they can also be important during anagenesis, they are not necessary for evolutionary change and may be largely irrelevant to many evolutionary phenomena […]. During speciation, however, the complex interactions between the genomes of two taxa guarantee that both the mathematical models and genetic analyses of speciation will differ from those used within species. This implies that speciation may show emergent properties not seen in traditional population-genetic models’. (Coyne & Orr, 2004, p. 56). To understand speciation, we thus need to go beyond studying genomic patterns of adaptation. This seems admittedly more difficult than scanning the genome for signatures of adaptation, for which a substantial number of statistical methods have been developed over recent years. We fully agree that identifying loci involved in adaptation is important, as adaptation (divergent or not) might often directly or indirectly contribute particularly to the early stages of speciation (Sobel et al., 2010; Butlin et al., 2012; Nosil, 2012; Seehausen et al., 2014), might trigger the evolution of intrinsic incompatibilities or mating preferences, or can help maintaining them (Gavrilets, 2003; Johnson & Porter, 2007; Sobel et al., 2010; Nosil & Flaxman, 2011; Bank et al., 2012), and because ecological differentiation is required for the coexistence of sympatric lineages (Weissing et al., 2011). However, most adaptive mutations likely do not contribute to further evolution of reproductive isolation, while other processes unrelated to adaptation can be important drivers of speciation. Indeed, in many cases, the evolution of hybrid incompatibilities appears to result from neutral mutational processes or genetic conflict between selfish genes and host genes (Presgraves, 2010; Maheshwari & Barbash, 2011). Even in cases where incompatibilities arise as by-products through pleiotropic effects of ecological adaptation, the particular adaptive traits that triggered the evolution of incompatibilities might have done so by chance. The specific role of these traits in adaptation is thus not necessarily informative about which traits are important to speciation. Ravinet et al. (in press) define ‘barrier loci as positions in the genome that contribute to a reduction in effective migration rate (me) relative to the expected rate’. This broad definition includes loci involved in intrinsic hybrid incompatibilities, loci accounting for differences in mating behaviour or loci under ecologically divergent selection. The latter can be unrelated to the evolution of intrinsic incompatibilities or differences in mating behaviour and might eventually be lost due to drift, gene flow or changes in ecological conditions, or replaced by universally advantageous alleles. In addition, those loci might not confer the often-assumed selective disadvantage in heterozygotes, for example if dominant allele expression evolves or if intermediate habitats are readily exploited. Further, ecological differentiation alone is usually not sufficient for speciation (Weissing et al., 2011). Intrinsic hybrid incompatibilities or differences in mating behaviour, however, can potentially result in more persistent barriers to gene flow and thus increase reproductive isolation rather than resulting in just (ephemeral) polymorphisms. It would therefore be advantageous if we could distinguish between barrier loci that contribute to reproductive isolation from those that might only confer ephemeral adaptation. However, it remains unclear whether making this distinction will be possible with genomic data alone. At least some nonadaptive substitutions that can drive the evolution of intrinsic hybrid incompatibilities, for example those of rapidly evolving selfish genes, are expected to affect the genome in a similar way as adaptive mutations (Presgraves, 2010; Maheshwari & Barbash, 2011). Although comprehensive theoretical work on speciation is available, few attempts have yet been made to assess models of speciation (e.g. Gavrilets, 2003, 2014; Bierne et al., 2011; Maheshwari & Barbash, 2011; Weissing et al., 2011; Bank et al., 2012) for their effects on genomewide differentiation and divergence. Lindtke & Buerkle (2015) studied several architectures of ecologically independent hybrid incompatibilities and their effects on genomic heterogeneity in secondary contact zones. Their results show that some architectures of reproductive isolation can lead to genomic outcomes very similar to those expected for loci under ecologically divergent selection, and that genomic heterogeneity can also arise when hybrid incompatibilities actually break down (i.e. are inefficient as barrier loci). This suggests that interpreting heterogeneous genome differentiation is not only difficult but can even be misleading for understanding speciation. Further, simulations by Flaxman et al. (2014) showed that with high levels of gene flow, many weakly divergently selected loci that induce genomewide linkage disequilibrium (LD) to pass a critical value can trigger a positive feedback process, causing rapid genomewide divergence. Identifying loci causal to this process is unlikely to succeed with genome scans. The build-up of LD is considered to be crucial to provide progress towards speciation in the face of gene flow (e.g. Felsenstein, 1981; Kirkpatrick & Ravigné, 2002; Barton & de Cara, 2009; Ortiz-Barrientos et al., 2016). Felsenstein (1981) found that this could be achieved when a locus causing assortative mating becomes associated with two other loci that are under divergent natural selection in two subpopulations. If recombination is too strong, speciation is prevented, as the association between the prezygotic isolation mechanism and adaptation is randomized (Felsenstein, 1981). To illustrate the problems involved with differentiating between genomic patterns of adaptation and processes considered important for speciation, we explored whether the evolution of assortative mating as described in Felsenstein's model will result in patterns of FST and LD that differ from those arising from multiple adaptive loci. In many cases, the genomic outcomes of models including or excluding assortative mating are practically indistinguishable (Fig. 1). Both the Felsenstein model and pure multilocus selection can build up LD (here measured as Zg; Storz & Kelly, 2008) and affect genomic heterogeneity in FST in similar ways. As might be expected from classical models of population genetics (e.g. Hedrick, 2017), even very strong assortative mating combined with moderate selection can be indistinguishable from genomic patterns resulting from random mating with very strong selection. This illustrates that it will be difficult to use conventionally applied genome scans to distinguish among genomic signatures from different classes of barriers to gene flow. In both cases, me was limited at barrier loci and sometimes also reduced at unlinked parts of the genome. However, only the first case, where a locus causing assortative mating becomes associated with two loci under divergent natural selection, has the potential to maintain reproductive isolation independent of changes in environment. We only briefly considered here one of several possible routes of speciation: the evolution of assortative mating. Various other models of speciation (e.g. Gavrilets, 2003, 2014) still need to be investigated regarding their effects on heterogeneity in genome differentiation, build-up of LD and genome divergence, and whether characteristic signatures can be identified that differ from adaptation that might be unrelated to speciation. However, it seems likely that different combinations of parameters could be picked for different models that would result in signatures that are nearly impossible to distinguish. To potentially distinguish loci that confer ecological adaptation from those involved in potentially more stable environmentally independent isolating barriers, investigating areas of sympatry and hybrid zones may be particularly useful as the confounding effects of ecological adaptation can be minimized. However, we believe that greatest progress towards understanding the evolution of barriers to gene flow that drive speciation will likely come from field observations, experiments and crosses. Unfortunately, by its very nature, the final stage of speciation is inaccessible to crossing experiments, as the loci responsible for complete intrinsic reproductive isolation make such approaches impossible. Genomics will help to identify organisms that are worth studying and will also facilitate determining candidate loci that can be investigated regarding their pleiotropic effects and epistasis and their putative contribution to speciation. However, genomics alone will not solve the problem of speciation. Thus, in addition to the complications in identifying barrier loci outlined by Ravinet et al. (in press) that arise from population history and the nonuniformity of genome evolution, we need to be aware of the limitations of genome scans that cannot be overcome even with sophisticated statistical methodology (although this should not keep us from trying to develop even more sophisticated methods that can possibly differentiate these types of loci). We are grateful to Mike Ritchie and Reto Burri for valuable comments on an earlier version of this article.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.