Pulling together or pulling apart: hybridization in theory and practice


Correspondence: Roger Butlin, Department of Animal and Plant Sciences, The University of Sheffield, Sheffield S10 2TN, UK.

Tel.: +44 114 2220097; fax:+44 114 2220002; e-mail: r.k.butlin@sheffield.ac.uk

Abbott et al. (2013) claimed that, ‘If hybridization is defined as reproduction between members of genetically distinct populations (Barton & Hewitt, 1985), producing offspring of mixed ancestry, then it occurs in almost all proposed processes of speciation’. An anonymous reviewer noted in response to this claim, ‘It really is not hyperbole to state that this statement alone makes this an important contribution … The conclusions of the authors are thus indicative of a major paradigm shift in evolutionary biology, away from the typological setting of the Neo-Darwinian synthesis’. The Commentaries assembled here suggest that any new paradigm recognising the near ubiquitous occurrence of hybridisation in animal and plant speciation is widely accepted. However, opinions vary much more widely as to its impact and potential role in the process of speciation.

Issues emerging from these commentaries include definitions of the nature of hybridisation (and, of course, species) and the likelihood of reinforcing selection and its consequences. However, the main debate concerns the broader significance of natural hybridisation and its potential role in biodiversity.

The mainly contrasting views of theoreticians versus empirical biologists are particularly striking, and thought-provoking. In a discussion on this point, Jim Mallet said, ‘The stance of theoreticians to the data on hybridization reminds me of Arthur Eddington's famous quip, “No experiment should be believed until it has been confirmed by theory.”’ It is hard to imagine that two forms may maintain genetic differences or undergo divergence while in contact because of the two-state rule suggested by Sewall Wright, where FST ≈ 1/(1 + 4 Nm). If Nm << 1 divergence can take place. If Nm >> 1 it will not. Nick Barton used to tell us in the 1980s that host races in insects, forms that exchange genes at a reasonable rate yet nevertheless diverge in ecological adaptations, should not exist. It is now clear that they do, although why they do is still a bit of a puzzle. Species of hybrid origin, both polyploid and homoploid, appear to be common in plants whatever the theory might say (see Servedio et al., 2013). Perhaps there is another, equally valid, version of Eddington's dictum: ‘No good theory should be believed until confirmed by adequate data’.

The definition of hybridization adopted by Abbott et al. (2013) focuses attention onto the offspring resulting from mixed matings. When divergent populations are in contact, there are many factors that potentially influence the probability of the formation of hybrid zygotes (Hochkirch, 2013; Rosenthal, 2013; Svensson, 2013) including phenology, mating signals and preferences, gamete competition, gamete recognition and so on. Regardless of whether viable zygotes are produced, these interactions can be costly and so impose selection on the underlying traits. Since this is direct selection, it may be more powerful than indirect selection resulting from reduced hybrid fitness (Shaw & Mendelson, 2013). The analogy with sexual selection is an important one: direct costs and benefits of mate preferences can have much stronger effects than indirect benefits due to offspring fitness (Svensson, 2013). For these very reasons, it is important to distinguish between the two sources of selection (as argued by Shaw & Mendelson, 2013) and, therefore, helpful to define hybridization in terms of offspring production. Offspring production is necessary for gene flow, although it does not guarantee it. So, it is important to distinguish production of hybrid offspring from mating interactions. Nevertheless, mating interactions, including those between reproductively isolated species that do not result in hybrid offspring, are a potent and potentially important source of selection shaping species interactions (Hoskin & Higgie, 2013; Svensson, 2013). Perhaps calling this ‘Reproductive interference’ is the best available term (Hochkirch, 2013) to distinguish it from more typical definitions of hybridization.

This distinction is very relevant to the issue of reinforcement. Abbott et al. (2013) said little about the likelihood of the process of reinforcement (it was striking how divergent were the opinions expressed at the workshop), yet the topic has attracted much attention from commentators. Hoskin & Higgie (2013) emphasise the existence of ‘solid empirical examples’ and argue that reinforcement is likely to be more common than these examples suggest because it is so difficult to demonstrate. In contrast, Barton (2013) suggests that reinforcement ‘may be so rare as to make little overall contribution’ to speciation and has been demonstrated mainly in cases with widespread overlap, indicative of strong isolation, rather than in hybrid zones. Shaw & Mendelson (2013) conclude that the ‘jury is still out’. This seems to be the safest position, especially where the question is whether reinforcement makes a common and substantial contribution to speciation and, if so, in what circumstances it does so (Servedio et al., 2013). Unfortunately, part of the problem here is again one of definitions. Many authors use a very wide definition of reinforcement (Servedio & Noor, 2003; Servedio et al., 2013) encompassing both direct selection (reproductive interference) and indirect selection (hybridization) (cf. Shaw & Mendelson, 2013). Note that reproductive interference can occur in the whole range from cases with hybrid fitness equal to parental fitness through to those where hybrid fitness is zero or hybrids are never formed. Reinforcement may occur after secondary contact or primary divergence (‘adaptive speciation’, Servedio et al., 2013). Reinforcing selection can occur but result in only small changes in mating pattern and so have little impact on gene flow (Bímová et al., 2011; Barton, 2013). It will be hard to obtain a clear picture of the contributions of these various processes to speciation while they are all lumped under a single heading of ‘reinforcement’. From the point of view of ‘cascade reinforcement’ or ‘RCD speciation’ (see Hoskin & Higgie, 2013), the distinction between direct and indirect selection is not crucial. As Servedio et al. (2013) argue, this mode of speciation is a special case of local adaptation: special because selection acts specifically on traits that have a high probability of contributing to reproductive isolation. This example of the broader significance of interspecies interactions for speciation deserves further empirical study; like reinforcement itself, there are some apparently convincing empirical examples (Pfennig & Pfennig, 2009; Hoskin & Higgie, 2013) despite theoretical scepticism (Servedio et al., 2013).

In hybrid zones, reinforcement is not the only type of modification possible. Schilthuizen & Lammers (2013) emphasise selection favouring modifiers that reduce hybrid fitness costs. However, this form of modification is subject to the same difficulties as reinforcement (Sanderson, 1989). Another, perhaps more likely, form of amelioration is the regeneration of compatible (probably ancestral) allele combinations by recombination (Gavrilets, 1997; Shuker et al., 2005). Whether they are related to the modification of hybrid fitness or not, rare alleles that appear in hybrid zones are fascinating. They are an example of the novelty that can be generated by hybridization, which is also emphasised and expanded by Soltis (2013). The rare alleles may represent new (perhaps recombinant) alleles (Baird, 2006) and, in addition, new phenotypes can be generated from new combinations of existing alleles. Transgressive genetic variation is common (Rieseberg et al., 2003). This argues against the views expressed by Barton (2013) and Servedio et al. (2013) that introgression adds no more variation than would be expected from a single large population and is unlikely to generate divergence. Where two populations have adapted to different environments (A and B), combinations of alleles can be generated in hybrids that are not advantageous in either A or B but are favoured by selection in another environment C. Particularly where multiple alleles are involved, this can make colonisation of C, and subsequent divergence, possible where it is very unlikely due to mutation alone. This is the scenario that underlies many suggested cases of adaptive introgression and hybrid speciation. Given the practical difficulties of inferring population histories (Strasburg & Rieseberg, 2013), it is too early to say how commonly such events occur. Certainly allopolyploid origins are common, and they are commonly associated with colonisation of new environments, but the steps are hard to tease apart and it may be even harder to do so for homoploid events.

Barton (2013) and Seehausen (2013) both emphasise that reinforcement is distinct from coupling, in the broad sense, between components of reproductive isolation. Coupling can involve any combination of reproductive barriers, from any of the four major domains: pre and postzygotic, intrinsic and extrinsic (Fig. 1). Seehausen (2013) proposes a mechanism, facilitated by hybridization, in which any structure imposed by environmentally-dependent selection can capture other barrier loci, including those loci subject only to endogenous selection. This is essentially the idea developed by Bierne et al. (2011), coupling in the more restricted sense where it is dependent on linkage disequilibrium due to mixing of distinct populations (Barton & de Cara, 2009; Barton, 2013) and whose foundations go back to Felsenstein (1981). Abbott et al. (2013) developed the idea in a spatial context, because this can make the argument easier to follow, but ecologically, rather than spatially-imposed structure can capture other barrier loci, whatever the form of the barrier they create. In strict sympatry, frequency dependence is required to maintain polymorphism (Barton & de Cara, 2009) but there is likely to be a component of spatial separation of habitats in most systems, as there is for the cichlid fish studied by Seehausen and colleagues (Seehausen et al., 2008).

Figure 1.

Coupling can bring together the effects of barriers of any type. Four major domains of reproductive isolation are defined by the prezygotic–postzygotic and endogenous–exogenous axes but some forms of isolation do not fall neatly into these domains.

The impact of spatial structure in the habitat is an important issue. It depends critically on the scale of dispersal. Barton (2013) emphasises the fact that the barrier to the flow of neutral alleles across a sharp hybrid zone, such as that between Bombina bombina and B. variegata, can be relatively weak: being equivalent to only a modest spatial separation. Relatively minor physical barriers can trap clines in such species. However, in species with more extensive dispersal, major environmental discontinuities may be necessary to trap clines and the importance of the first genetic barriers to be formed may be greater. This may be why hybrid zones are clustered differently for marine species with planktonic dispersal compared with species with more restricted dispersal (Pelc et al., 2009; Bierne et al., 2011). In general, historical contingency can have major effects on the barriers we see today, a very important issue raised by Seehausen (2013). This was the reason why Abbott et al. (2013) emphasised the importance of studying replicated hybrid zones when possible. Sometimes this has resulted in unanticipated outcomes, such as an inverted relationship with environmental variation (e.g. the distributions of Mytilus edulis and M. trossulus show opposite relationships with salinity in Europe and the Canadian Maritimes, Bierne et al., 2011) or alternative coupling with habitat (e.g. association between soil profile and chromosomal race distribution in Australian morabine grasshoppers, Jackson et al., 2012). This is not simply a question of which alleles are available to respond to selection: a wide range of other historical effects may be important, particularly range changes (Abbott et al., 2013; Barton, 2013) or extinction–recolonization processes (Bierne et al., 2011).

Feder et al. (2013) describe the interactions among barriers in terms of ‘divergence’ and ‘genomic’ hitchhiking (DH and GH, respectively). Using the terminology of hitchhiking is perhaps unfortunate as it suggests only a transient effect of spreading alleles, whereas the issue is more one of persistent barrier effects (see Harrison, 2012 for a discussion of the importance of language in speciation research). Also, the DH/GH distinction cuts across a continuum at the limit of physical linkage whereas recombination typically reaches 0.5 within chromosomes and coupling can extend to unlinked loci, particularly where linkage disequilibrium is generated by dispersal or ecological structure. Feder et al. (2013) try to divide the continuum of speciation (emphasised by Hochkirch, 2013) into four phases which are defined by transitions from direct selection only to indirect selection on tightly linked regions and then to genome-wide effects. They accept that these phases have fuzzy boundaries. We question the value of this distinction into descriptive phases. Selection on any barrier locus influences differentiation at that locus and throughout the genome, depending on the strength of selection and the level of linkage disequilibrium, which decays through recombination. This continuous effect is well captured by the established idea of a ‘coupling coefficient’ (Barton, 1983; Baird, 1995; Abbott et al., 2013) incorporating the effects of both selection and recombination. Furthermore, a plethora of barrier loci can couple together efficiently, well before the barrier to neutral gene flow becomes genome-wide (e.g. most hybrid zones studied to date), while genome-wide barriers can be produced by a few genes of strong effect. The former may be persistent while barriers of the latter type are not necessarily stable in the long term (e.g. colour vision in cichlids). There is no necessary sequence from DH to GH to speciation.

The common framework for selection driven speciation presented by Feder et al. (2013) is in striking contrast to the highly non-parallel opportunistic process envisaged by Björklund (2013) in which every hybridisation event is unique, with unpredictable outcome. Hybridization undoubtedly does have highly variable outcomes (Svensson, 2013), with much of the variation dependent on factors other than hybrid fitness (Rosenthal, 2013). That some of this variation is among individuals within interacting populations is a critical point, underlying important consequences such as the Stam effect (Stam, 1983). Learning may contribute to this variation and Svensson's (2013) point that it is more likely to be influenced by direct than by indirect fitness effects is a valuable insight. Sætre (2013) argues that hybridization generates much variation that is not captured (one might almost say that it is wilfully ignored) by focusing on the species as the unit of diversity. This complexity means that there is much to do, but also different ways of seeing the task ahead. It may be that the outcome of particular population interactions will never be predictable but there remains the potential to find general patterns and this should be the focus of research on hybridization, as it should be for speciation more generally (Butlin et al., 2012).


The contributors to the target review (Abbott et al., 2013) hold many and varying views on hybridization and speciation, which led to fascinating discussions during the Gregynog FroSpects workshop. It is a pleasure to see these discussions continued in the stimulating set of commentaries published here, particularly where they pick up on points raised at the meeting but left out of the paper for reasons of space. In the review, we attempted to reach a consensus that all contributors could accept. We (R.K.B. and M.G.R.) take responsibility for this article. It does not necessarily reflect the views of all workshop participants but we are very grateful to those participants who provided insightful reactions to the commentaries and so helped in the formulation of this response, particularly Nicolas Bierne and Stuart Baird.