Intrinsic incompatibilities evolving as a by‐product of divergent ecological selection: Considering them in empirical studies on divergence with gene flow

The possibility of intrinsic barriers to gene flow is often neglected in empirical research on local adaptation and speciation with gene flow, for example when interpreting patterns observed in genome scans. However, we draw attention to the fact that, even with gene flow, divergent ecological selection may generate intrinsic barriers involving both ecologically selected and other interacting loci. Mechanistically, the link between the two types of barriers may be generated by genes that have multiple functions (i.e., pleiotropy), and/or by gene interaction networks. Because most genes function in complex networks, and their evolution is not independent of other genes, changes evolving in response to ecological selection can generate intrinsic barriers as a by‐product. A crucial question is to what extent such by‐product barriers contribute to divergence and speciation—that is whether they stably reduce gene flow. We discuss under which conditions by‐product barriers may increase isolation. However, we also highlight that, depending on the conditions (e.g., the amount of gene flow and the strength of selection acting on the intrinsic vs. the ecological barrier component), the intrinsic incompatibility may actually destabilize barriers to gene flow. In practice, intrinsic barriers generated as a by‐product of divergent ecological selection may generate peaks in genome scans that cannot easily be interpreted. We argue that empirical studies on divergence with gene flow should consider the possibility of both ecological and intrinsic barriers. Future progress will likely come from work combining population genomic studies, experiments quantifying fitness and molecular studies on protein function and interactions.

Binary distinctions between concepts can be helpful to aid communication, but they may lead to oversimplification-as has been the case for the distinction between "sympatric" and "allopatric" speciation (Butlin, Galindo, & Grahame, 2008). Here, we discuss the overlap between "intrinsic" and "ecological" barriers to gene flow (see definitions in Box 1), aiming our article mainly at empiricists working on adaptive divergence and speciation with gene flow. We argue that, while ecological barriers are considered crucial in driving these processes, empirical studies in these fields often neglect the possibility of intrinsic barriers. The main reason for this is that purely intrinsic barriers are unlikely to evolve under continuous gene flow: they are selected against under many scenarios, and incompatible alleles cannot spread in areas where diverging populations frequently interbreed (Bank, B€ urger, & Hermisson, 2012).
However, there is a situation in which intrinsic barriers may evolve and be stable even under gene flow. That is the case when the very same loci that are under divergent ecological selection are also involved in intrinsic barriers (Baack, Melo, Rieseberg, & Ortiz-Barrientos, 2015;Bank et al., 2012;Dobzhansky, 1951;Gavrilets, 2000;Schluter & Conte, 2009). Early work, for example, by Dobzhansky (1951) already recognized that divergent selection on a locus can cause the evolution of intrinsic barriers as a by-product; this idea was further developed in later theoretical work (e.g., Bank et al., 2012;Barton, 2001;Chevin, Decorzent, & Lenormand, 2014).
While a large part of this work has focused on divergence in allopatry, there are clear indications that the same mechanism can generate barriers even under continuous gene flow (Baack et al., 2015;Gavrilets, 1999;Schluter & Conte, 2009;Slatkin, 1982)-the point we emphasize in this article.
First, divergent ecological selection on a locus may favour alleles that are incompatible with alleles at other loci in the other population, producing intrinsic isolation as a by-product (i.e., derivedderived incompatibility, Box 1). Second, adaptive changes at ecologically selected loci can generate new "genetic environments," enabling further changes at interacting loci in the same population. These cascading changes can themselves lead to intrinsic incompatibilities with the ancestral allele (i.e., ancestral-derived incompatibility, Box 1). If such patterns are common, intrinsic barriers may well contribute to primary ecological divergence and act as a pathway to "ecological speciation" (Rundle & Nosil, 2005). Then, the distinction between "ecological" and "intrinsic" barriers (Box 1) becomes blurred, as does the distinction between adaptive divergence and speciation.
The connection between divergent ecological selection and intrinsic barriers also has practical consequences for studies investigating the genomic basis of adaptive divergence and ecological speciation. Genomic scans for loci showing high differentiation between populations have become very popular and are commonly interpreted to reveal loci under divergent ecological selection (Beaumont & Balding, 2004;Nosil, Funk, & Ortiz-Barrientos, 2009). The logic behind this approach is that loci underlying local adaptation are able to resist gene flow and should therefore be the most differentiated genomic regions detected as high F ST peaks. However, there are various caveats, and high F ST values may not always indicate divergent selection (Cruickshank & Hahn, 2014;Wolf & Ellegren, 2017). The field of evolutionary biology is now moving towards increasing the reliability of outlier scans by controlling for confounding factors (e.g., Burri et al., 2015), using experiments to test whether outlier loci indeed respond to selection (Soria-Carrasco et al., 2014), and using genetic manipulation to establish the organismal role of outlier loci (Colosimo et al., 2005). Still, as pointed out by Bierne, Welch, Loire, Bonhomme, and David (2011), loci showing high differentiation between ecologically divergent populations may actually reflect intrinsic barriers trapped at the transition between two environments (i.e., ecotone) after secondary contact, rather than loci under ecological selection. Here, we emphasize a different aspect: even during primary divergence, intrinsic barriers may evolve and colocate with the ecotone because they are caused by, or interact with, the loci under divergent ecological selection.
To help bridge the gap between theoretical and empirical studies, we discuss how gene interaction networks and pleiotropy may facilitate the evolution of intrinsic barriers driven by ecological selection, ask under which conditions these barriers contribute to a reduction in gene flow and discuss consequences for genome scans and similar analyses. By outlining explicit mechanistic scenarios, we aim to facilitate the search of loci contributing to both ecological and intrinsic barriers from empirical data.
In much of this short article, we restrict ourselves to cases of two-locus scenarios for the sake of simplicity, but we discuss implications of more complex incompatibilities below. Single-locus barriers are not considered for reasons of space, but in principle they can evolve by the same mechanisms as multilocus ones (although they may be restricted to certain types of genes as they require repeated evolution at a single locus).
In all of our scenarios, at least one locus-denoted as locus A in population 1-is under divergent ecological selection. In addition, alleles at locus A are incompatible with alleles at locus B ( Figure 1).
Locus A is therefore influenced by both ecological and intrinsic selection pressures. In contrast, locus B is involved in an incompatibility with locus A, but not necessarily under divergent ecological selection ( Figure 1). There are two simple ways in which divergent selection can lead to intrinsic barriers in two-locus systems (Muller, 1942) (derived-derived and ancestral-derived); these are outlined in Box 1.
We emphasize that throughout this article, when we refer to "changes" at a locus, on the population level we mean a change in allele frequency at the locus in question. This does not necessarily imply fixation, but rather the emergence of a difference in allele frequencies between populations.

ASSOCIATION BETWEEN ECOLOGICAL AND INTRINSIC BARRIERS
Why should ecological barrier loci also be involved in intrinsic incompatibilities? One reason is that ecologically selected loci may change more rapidly than neutrally evolving ones. The more interdependencies between genes there are, the more likely it is that ecologically driven selection generates incompatibilities as a by-product. Indeed, no gene is an island; instead, genes are dependent on other genes and regulatory sequences through networks and feedback cascades (Phillips, 2008;Wright, 1968). They are also frequently pleiotropic and are likely to serve multiple functions depending on when and where they are expressed. Below we detail how, due to these interdependencies, intrinsic barriers may be caused by divergent ecological selection. We do not focus on their stability with gene flow yet (this topic is dealt with in the later sections).
We define a pleiotropic gene as a locus for which there is a high probability that an allelic substitution will have effects on more than one trait. A "trait" in this sense can be either morphological, behavioural or biochemical. For example, nucleotide substitution in a pleiotropic gene could lead to changes in both colour and size. At a molecular level, the product of a gene could be involved in several different molecular functions or could perform different tasks when expressed in different tissues (Paaby & Rockman, 2013).
Gene interactions occur whenever the products of different genes are part of the same functional network. This can happen via direct interaction, where two gene products, for example, form a protein complex, or one protein catalyses a conformational change in another. Because genes usually interact with several partners and are part of interaction cascades, the number of indirect interactions is arguably much larger.
Pleiotropy and gene interactions are related: on average, the more functions a gene has, the more traits it affects (pleiotropy) and the more other loci it interacts with (gene interaction). However, they can also act independently, that is, one can occur without the other for a specific gene.
Pleiotropy and gene interactions can lead to derived-derived incompatibilities. If a pleiotropic locus experiences nucleotide changes due to ecological selection on one trait, this can have effects on other traits or the molecular functions this locus underlies ( Fig. S1A). Any such change in a trait or function in population 1 might be incompatible with another change in population 2. The more pleiotropic the locus is, the more traits or functions can potentially be altered, and the higher the risk of an adaptive mutation producing an incompatibility between populations as a side effect.
Similarly, the more interactions a gene product is involved in, the higher the chance for an incompatibility with a gene product in another population (Fig. S1B).
Pleiotropy and gene interactions are also likely to cause ancestralderived incompatibilities. Any change in population 1 might lead to follow-up changes at other loci within the same population (Pavlicev & Wagner, 2012), resulting in co-adaptation of genes (Fig. S1C). For example, adaptive change in the gene product of locus A can enable an adaptive conformational change in the physically interacting gene product of locus B (Fig. S1D). As another example, adaptive changes in a pleiotropic gene might cause negative pleiotropic side effects, and follow-up changes at other genes in the same population might compensate for these, again leading to co-adaptation (Lehner, 2011;Pavlicev & Wagner, 2012). The more pleiotropic effects or gene interactions there are, the more possibilities for changes at other loci emerge that would not have been possible in the previous background.
These considerations demonstrate that with pleiotropy and gene interactions, ecological selection may often cause intrinsic barriers.
So, how common are pleiotropy and gene interactions according to empirical studies? Few studies (reviewed in Paaby & Rockman, 2013) have systematically tested for genomewide pleiotropy by reverse genetics, that is, by mutating single genes one by one and measuring the effects. Alternatively, pleiotropy can be measured by QTL studies. These approaches employed in yeast (Dudley, Janse, Tanay, Shamir, & Church, 2005), nematodes (Wang, Liao, & Zhang, 2010), mice (Wagner et al., 2008) and sticklebacks (Albert et al., 2008) F I G U R E 1 Evolution of incompatibilities as a by-product of divergent ecological selection. Shown are two hypothetical loci, A and B, situated on two different chromosomes and each with two alleles. A mutation from allele a to A indicates a locus under divergent ecological selection. Mutation from b to B can arise by divergent ecological selection or is selected for other reasons (e.g., compensating pleiotropic side effects of other mutations). Green arrow indicates divergent ecological selection between alleles of the same locus, and red arrow indicates intrinsic incompatibility between loci. Panel A: derived-derived incompatibility; panel B: derived-ancestral incompatibility NEWS AND VIEWS | 3095 suggest that an average gene affects three to seven traits (Paaby & Rockman, 2013). Comprehensive screens and functional information are likely to be available for only a handful of genes, but an indication of pleiotropy is the number of splice variants a gene has. Generally speaking, the Drosophila genome contains three times more proteins than there are genes (Nei, 2013, p. 115), suggesting that on average a single gene produces three functional variants. As an extreme example, Dscam, a gene encoding a membrane protein and involved in development of Drosophila, has 24 exons and theoretically would be able to produce 38 016 different types of proteins (Nei, 2013, pp. 127-128). Of course, the crucial question is whether all these splice variants are functional, and whether they serve the same or different functions. In any case, pleiotropy is probably widespread in the genome.

BOX 1 Definitions
Ecological barrier loci An ecological (and extrinsic) barrier occurs when a locus is under divergent ecological selection and reduces gene flow between populations. This means ecological selection favours one allele in population 1 and another allele in population 2, leading to selection against unfit immigrants and/or the formation of hybrids that are maladapted in both environments (Nosil, 2012;Schluter, 2000). A purely ecological barrier is always environment-dependent and would not function as a barrier in a homogeneous environment (e.g., under standardized laboratory conditions).

Intrinsic barrier loci
We define intrinsic barriers (i.e., incompatibilities) as those where interactions between alleles result in lowered fitness of individuals carrying their combination. Such barriers may either involve alleles at the same or at different loci; we here focus on the latter and will not discuss the former for simplicity. Purely intrinsic barriers are environment independent, meaning they result in a lowered fitness of hybrids or recombinants in any relevant environment and under standardized laboratory conditions. Purely intrinsic barriers may evolve by drift, usually in allopatry (Turelli, Barton, & Coyne, 2001 and references therein). Alternatively, the incompatible alleles may each be favoured by selection that is uniform across environments in population 1 and 2 (e.g., global temperature increase), but if the alleles are combined within the same genotype they are incompatible. For example, population 1 may adapt by evolving allele A at locus A and population 2 adapts by evolving allele B at locus B, but when brought together allele A and B are incompatible with each other (i.e., mutation order speciation; Mani & Clarke, 1990).

Loci involved in both intrinsic and ecological barriers
Our focus is on loci that are involved in both intrinsic and ecological barriers. The ecological barrier occurs because the locus is under divergent ecological selection. The intrinsic effect results from the interaction of one or more alleles at the locus with one or more alleles at other interacting loci that cause reduced fitness of hybrids (i.e., intrinsic incompatibilities) (Figure 1). Loci involved in both intrinsic and ecological barriers will show evidence of divergent selection in the field, as well as reproductive isolation in a standardized laboratory environment.

Derived-derived incompatibility
Derived-derived incompatibility can evolve when two interacting loci change in each of two diverging populations, that is, an ancestral genotype aabb evolves into AAbb in population 1 and aaBB in population 2 ( Figure 1A). When the derived alleles (A and B) at the two loci are combined within the same individual (e.g., AaBb), they may be incompatible with each other (Dobzhansky 1936(Dobzhansky , 1951Muller, 1942) leading to derived-derived incompatibility and intrinsic barrier (Orr, 1995).

Ancestral-derived incompatibility
Ancestral-derived incompatibility may emerge if divergent selection drives change in population 1 (e.g., the replacement of the ancestral genotype aabb by AAbb), and this enables a change in locus B in the same population, leading to AABB genotypes. Because the B allele only works when the A allele is present, combining the B allele with the ancestral allele a in hybrids (e.g., aaBB) generates ancestral-derived incompatibility leading into intrinsic barrier (Orr, 1995) (Figure 1B).
A possible example of ecological selection driving evolution of intrinsic barriers because of pleiotropy comes from studies of hybrid necrosis in plants (Bomblies, 2010;Chae et al., 2014). One particular locus, DM2, shows signatures of diversifying selection at the sequence level. Selection is likely driven by pathogen pressure as the locus is involved in pathogen recognition (Chae et al., 2014). Interestingly, DM2 interacts with at least five different loci causing necrosis and problems in hybrids, suggesting natural selection from parasites generates incompatibilities between DM2 and loci interacting with it (Bomblies, 2010;Chae et al., 2014).
Molecular biology has also shown that interactions of gene products are ubiquitous. A study testing 1000 genes in yeast found that the number of confirmed interactions per gene varied from 1 to 146, with an average of 34 interactions per gene (Tong et al., 2004).
Another study on gene essentiality identified 44 genes that are needed for the viability of the Sigma1278b isolate of S. cerevisiae but not for the standard S288c strain. Remarkably, genetic analysis revealed that in the majority of tested cases the differences in essentiality were influenced by at least four different loci in the genome suggesting complex multigenic interactions (Dowell et al., 2010). Even the classical case of DMI in Drosophila is genetically complex, where hybrid lethality of Nup160 depends on one or more unknown additional factors in the autosomal background (Tang & Presgraves, 2015).
Interactions between different loci in the genome do not only arise because of gene-gene interactions; they arise also between regulatory sequences like transcription factors, microRNAs, siRNAs and their target regions. Loci that regulate gene expression, called eQTL, generally appear to affect a small number of gene expression traits, but typically a handful of eQTL hotspots affect abundances of hundreds to thousands of transcripts (Paaby & Rockman, 2013). Taking also regulatory variation into consideration, the number of interactions per locus is further increased and indeed, these types of complex interactions have been suggested to contribute to sterility of hybrids in the house mouse (Turner, White, Tautz, & Payseur, 2014).

| Moving towards more complex and realistic scenarios
In summary, we predict that the potential for ecological selection causing intrinsic barriers as a by-product is enormous because both pleiotropy and gene interaction are common. In fact, the simple twolocus scenarios described above are probably often an oversimplification, and intrinsic and ecological barriers can also be indirectly

| Pleiotropic and connected genes can evolve under positive selection
Several authors have suggested that mutations in highly pleiotropic or interconnected genes are likely to have deleterious consequences (Fisher, 1930;Orr, 2000;Stern & Orgogozo, 2009). For this reason, they are less likely to respond to positive selection, being rather highly conserved. If this is true, they are less likely to evolve differences between closely related species and to serve as intrinsic incompatibilities, contradicting our above hypothesis.
However, ecological adaptation frequently requires changes in several traits; especially under gene flow, these changes may evolve more easily by mutations in a single pleiotropic locus compared to mutations in several independently segregating loci. This is because in the latter case the favourable allele combination is broken down every generation by recombination (Smadja & Butlin, 2011). Therefore, highly pleiotropic loci might be more effective in generating adaptive divergence, while at the same time being especially likely to generate intrinsic barriers as a by-product. Another argument for the involvement of highly connected or multifunctional loci in adaptive divergence is simply that there might be no other option. Adaptive changes may occur as long as their positive effect outweighs these negative side effects.
Empirical evidence for positive selection and fast evolution in highly connected genes is mixed. In humans, long-term positive selection is less likely in highly connected genes (i.e., genes that have multiple interaction partners) compared to genes with fewer connections (Luisi et al., 2015). In contrast, recent positive selection was more likely to target genes with higher centralities (i.e., highly connected) during human evolution (Luisi et al., 2015). shown to experience bouts of recent selection between different honey bee races (Kent, Issa, Bunting, & Zayed, 2011), and chemosensory genes, which bind a wide range of chemicals, have been suggested to play a role in local adaptation to different host plants in pea aphids (Eyres et al., 2016;Smadja et al., 2012).

| TH E R OLE OF DIFFERENT TYPES OF BARRIER LOCI IN D IVERGENCE AND SPECIATION
In the previous sections, we have discussed mechanisms why the emergence of intrinsic barriers as a by-product of divergent ecological selection may be common. Both the ecologically selected loci themselves and other interacting loci may be involved. However, we have so far only shown that adaptive mutations should often generate incompati- bilities. An important question is whether these mutations, first usually present only in a single individual, will rise in frequency, specifically when there is gene flow between the diverging populations. Under which conditions does the additional intrinsic barrier drive the system closer to speciation, compared to a purely ecological barrier?
To understand the role of barrier loci influenced by both intrinsic incompatibility and ecological selection in reducing gene flow, we will first briefly look at the fate and stability of purely ecological and purely intrinsic barrier loci separately. Intrinsic barriers evolving as a by-product of divergent ecological selection will be affected by the dynamics of both of these.

| Purely ecological barrier loci
In contrast to many purely intrinsic barriers, ecological barriers may be stable even with gene flow and recombination because environmental heterogeneity continuously favours divergence (Fitzpatrick, Gerberich, Kronenberger, Angeloni, & Funk, 2015;Garant, Forde, & Hendry, 2006;Kawecki & Ebert, 2004). Stable barriers may evolve and be maintained as long as selection is strong enough to overcome the counteracting effect of gene flow (Haldane, 1930).

| Purely intrinsic barrier loci
Purely intrinsic barriers can evolve either by genetic drift or under uniform selection (see Box 1); this will affect their dynamics, but in both cases, they are unlikely to increase in frequency where gene flow is high.
Intrinsic barriers arising under drift alone are likely to be maintained only in allopatric phases. This is because incompatible alleles without any selective benefits tend to be removed by selection under gene flow (Bank et al., 2012). Such barriers are therefore less likely to contribute to divergence and speciation with gene flow compared to barriers maintained by selection (but see the effects of additional factors below).
Intrinsic barriers may be more stable if different, but generally beneficial, mutations (e.g., alleles that confer adaptation to increasing global temperatures) become fixed in diverging populations ("mutation order speciation"; Mani & Clarke, 1990;Schluter, 2009). However, this process is also counteracted by gene flow because in this case, adaptive alleles will often spread between populations before incompatible alleles can establish (Nosil & Flaxman, 2011).

| Intrinsic barriers as a by-product of divergent ecological selection
Intrinsic barrier loci evolved as a by-product of divergent ecological The evolutionary fate of intrinsic barriers that evolved as a byproduct of ecological selection has partly been explored in the theoretical literature (e.g., Agrawal, Feder, & Nosil, 2011;Bank et al., 2012;Gavrilets, 2000;Slatkin, 1982) and depends on various parameters of the system: the strength of divergent ecological selection on locus A, the strength of divergent ecological selection on loci interacting with locus A and the epistatic interactions between loci. In the following, we discuss a two-locus system for simplicity and restrict ourselves to a discussion of general principles. We note that more complex incompatibilities may show different dynamics, which we cannot explore here. In addition, physical linkage between ecological barrier loci and interacting loci could work towards maintaining ecologically driven incompatibilities. We will assume unlinked loci here, but emphasize that linkage needs to be considered in empirical work and in a deeper exploration of the topic in general. Another potentially important factor that we ignore here are dominance effects.
In general, if both of the interacting loci are affected by divergent ecological selection, as well as being involved in the intrinsic barrier, this increases the barrier stability as divergence at both loci is favoured by environmental selection (Agrawal et al., 2011). Figure 3A shows an example where both loci are under divergent ecological selection, and involved in an intrinsic incompatibility that is not extremely strong. In this case, the barrier will be maintained under gene flow, and both the intrinsic and the extrinsic components will contribute to the overall reproductive isolation (Slatkin, 1982). adaptation. However, in areas far away from the contact, local adaptation via incompatible alleles is possible (Gavrilets, 1997).
The involvement of locus A in an intrinsic incompatibility can even lead to the loss of both the ecological and the intrinsic barriers (Agrawal et al., 2011), potentially across the species' range. Consider the example in Figure 3C. Here, locus B has evolved an allele that is universally adaptive (e.g., that confers adaptation to generally rising temperatures-that is, a trait that would be adaptive in both populations), but this universally adaptive allele (B) is strongly incompatible with allele A. In this case, allele B will spread across both populations and "drag" the compatible allele a with it. Consequently, both the ecologically driven and the intrinsic barriers are lost ( Figure 3C).
Of There is already much evidence for the ubiquity of pleiotropy and gene interaction networks (see above). However, we need specific estimates for loci likely to become involved in ecological divergence. Estimating selection coefficients of new mutations and mapping interactions between loci becomes more and more feasible (e.g., Gerke, Lorenz, & Cohen, 2009;Wang et al., 2010), but is still an endeavour when done on nonmodel organisms.
Empirical work does point towards association between ecological and intrinsic barriers (Schluter & Conte, 2009). For example, multiple studies exposing Drosophila populations to different selective conditions produced intrinsic postzygotic isolation in addition to local adaptation (reviewed in Rice & Hostert, 1993). Such observations are potentially explained by ecologically selected loci generating intrinsic isolation as a by-product (Dobzhansky, 1951;Rundle & Nosil, 2005). Further evidence comes from yeast, where populations grown in two distinct environments for 500 generations evolved intrinsic postzygotic isolation affecting growth rate and meiosis (Dettman, Sirjusingh, Kohn, & Anderson, 2007), and where it has been shown that environmental selection can generate strong incompatibilities at the same loci as a by-product (Anderson et al., 2010). Fascinating work on plants demonstrates that loci putatively adapting to local parasite pressures are also involved in incompatibilities, generating hybrid necrosis (Bomblies, 2010;Chae et al., 2014).
In sticklebacks, a functional mismatch in traits involved in niche use reduces the performance of F2 hybrids beyond that of additive genetic effects, suggesting epistatic interactions between genes underlying niche differentiation (Arnegard et al., 2014). This functional mismatch might lead to hybrid incompatibilities that are analogous to those underlying intrinsic reproductive isolation but depend on the ecological context (Arnegard et al., 2014 Schluter, 1999); the question is then whether intrinsic barriers map to the same loci as ecological barriers.
Another approach relies on detailed annotation of candidate genes, followed by computational approaches to explore whether they are likely to be pleiotropic or part of a common interaction network (e.g., genemania.org). If interdependence between candidate loci can be

| PRACTICAL IMPLICATIONS
If ecological selection frequently causes intrinsic barriers as a by-product, we need to keep this in mind when analysing genome scans and similar data. As already mentioned, there are other challenges with outlier scans that need to be taken into account. A correct inference of selected loci is a prerequisite for the considerations below.
First, a locus may be under divergent ecological selection and act as an intrinsic incompatibility as well (e.g., locus A in Figure 3A).
Such a locus will show evidence of selection in outlier scans and may be associated with a divergently selected adaptive trait-but it is not clear what proportion of the reduction in gene flow is due to the ecological barrier effect. Ignoring this could lead to an overestimation of the strength of ecological selection acting on the locus.
Second, outlier loci may only indirectly be associated with divergent selection (see Section "2.1"). Imagine a locus under divergent ecological selection that is part of a gene interaction network (Figure 2). Then, ecological selection on one locus within the network can drive further changes at other interacting loci. Then, these secondary changes can cause intrinsic incompatibilities with derived alleles of the network from another population. Because alleles within the network are dependent on each other, linkage disequilibrium emerges among interacting loci and they will appear as independent outlier peaks in a genome scan ( Figure 2). Importantly, some of these peaks are created by intrinsic barriers and are not the direct target of ecological selection. This idea has been developed in the "Selection Pleiotropy Compensation" model of adaptive evolution, which suggests that most adaptive signatures detected in genome scans could be the result of selection on a pleiotropic loci followed by compensatory changes, rather than of progressive character adaptations (Pavlicev & Wagner, 2012).
Interestingly, outlier loci might also be produced by changes that are universally adaptive, but which are confined to one genetic background: A universally adaptive allele will spread in the population where it is compatible with the genetic background, but cannot enter the second population due to incompatibilities, thereby producing an outlier signal.

| CONCLUSIONS
We conclude that the emergence of intrinsic barriers as a by-product of divergent ecological selection may be common, as suggested by the ubiquity of molecular interactions and pleiotropy. However, their role in speciation with gene flow is unclear. Depending on the conditions, the intrinsic barrier effect that evolved as a side effect of ecological selection may either strengthen or weaken the overall barrier to gene flow. More research is needed to estimate the relative importance of these effects; they may well differ between study systems and traits, depending, for example, on the nature and genetic architecture of ecologically selected traits. We need to consider the interdependencies between ecological selection and intrinsic incompatibilities in empirical studies of local adaptation and ecological speciation in order to correctly interpret the results of outlier scans and similar approaches.