Glittering gold and the quest for Isla de Muerta

The interpretation of patterns of divergence between related species across the genome is complex, and many of these complications are clearly reviewed in the article by Ravinet et al. (2017). Indeed, such a review is timely and clarifies much of the confusion about what can, and cannot, be inferred from such genome scans. Although the authors state that ‘We argue one of the principal aims of the field is to identify the barrier loci involved in limiting gene flow’, they also show how genome data can be used to investigate the genetic architecture of speciation without identifying individual barrier loci. We will argue that in many cases, the identification of barrier loci based on genomic data alone will be extremely challenging or impossible, and that it can be equally valuable to characterize general features of species barriers, such as their architecture and evolution through time. We caution against an obsession with the identification of functional loci that might distract researchers from addressing the many other exciting questions in speciation that do not require such specific information. Perhaps the first and most obvious reason why identification of barrier loci might be impossible is if the barriers themselves between species are highly polygenic. For example, it has long been known that a highly polygenic architecture, with many incompatibility loci of infinitesimal effect, can lead to a strong barrier to gene flow and ‘genome-wide congealing’ between hybridizing species (Barton & Bengtsson, 1986). In a hypothetical species pair isolated in this way, there would undoubtedly be considerable heterogeneity in FST across the genome. This could be due to random clustering of minor effect incompatibility loci, or due to variation in recombination rate, gene density, demographic stochasticity and all the other reasons outlined in the review article, but the search for particular barrier loci would be fruitless. As acknowledged by Ravinet et al., a similar argument can be made for loci involved in adaptation more generally: in speciation as in adaptation, it is likely to be the case that ‘All that is gold does not glitter’ (Rockman, 2012). Nonetheless, in many (perhaps most) species pairs, there are likely to be at least some barrier loci of major effect. In Heliconius butterflies, there is good evidence for major ‘speciation genes’. Wing pattern differences between species are controlled by a few large-effect loci and lead to strong preand post-mating isolation (Naisbit et al., 2003; Jiggins, 2008). Selection against hybrid butterflies is so strong that we can measure it in the field with simple experiments involving model butterflies (Merrill et al., 2012). These wing pattern loci were identified using traditional genetic approaches, and their genetic basis is now well understood. Nonetheless, when the genomes of strongly reproductively isolated species, such as Heliconius melpomene and Heliconius cydno, are compared, wing patterning loci do not stand out against the genomic background, despite ample evidence for gene flow between the species (Martin et al., 2013; Seehausen et al., 2014). FST across the genome is highly heterogeneous, with many ‘peaks’, which include but are not limited to wing pattern loci. Similar patterns are seen using dXY or other measures such as Fd which is a more unbiased measure of migration (Martin et al., 2015). As suggested by Ravinet et al., in the future, we can do better by accounting for genome-wide variation in patterns of recombination and selection, but we anticipate that some major effect barrier genes will remain indistinguishable in genome scans. In many cases, therefore, we will need some additional evidence to find barrier loci. These barriers may be like the Isla de Muerta in the Pirates of the Caribbean, which Jack Sparrow claimed ‘cannot be found except by those who already know where it is’. In contrast, between more closely related taxa, there are examples in which genome scans have identified highly divergent regions that contain functional ‘barrier genes’. These include hooded vs. carrion crows, freshwater vs. marine sticklebacks and divergent subspecies of Heliconius (Jones et al., 2012; Nadeau et al., 2012; Poelstra et al., 2014). A genome scan across many subspecies of the Heliconius erato radiation identified multiple narrow divergent regions that differentiate wing pattern races, representing modular regulatory loci around known wing patterning genes (Belleghem et al., 2017). All of these comparisons are between geographic forms in which background FST is close to zero, such that hybridizing populations are virtually panmictic across much of the genome. Another case in which a low background FST is combined with a relatively small number of divergent peaks are the sympatric cichlid fish in Lake Massoko (Malinsky et al., 2015) – an unusual case of sympatric species showing such a pattern. The search for barrier genes is therefore easier when taxa are very closely related and rates of hybridization are high. A further problem is the technical challenge of identifying narrow outliers against a noisy genomic signal. For example, Cruickshank and Hahn examined our data from Peruvian Heliconius wing pattern races, and did not find any peaks of divergence (dXY), suggesting little Correspondence: Chris D. Jiggins, Department of Zoology, University of Cambridge, Cambridge CB2 3EJ, UK. Tel.: +44 1 223 269 021; fax: +44 1 223 336 676; e-mail: c.jiggins@zoo.cam.ac.uk