Variance in genome-wide ancestry in admixed populations is predicted to be highest just after hybridization and decreases over time as recombination breaks down long ancestry tracts. When foreign ancestry is deleterious, selection during this initial period rapidly reduces the population's admixture proportion (Harris and Nielsen, 2016; Veller, 2019). This initial ‘fast’ period of purging lasts tens of generations (Veller, 2019), shifts ancestry genome-wide (Schumer et al., 2018; Matute et al., 2020), and begins to generate broad-scale differences in ancestry within and among chromosomes. Populations then enter a ‘slow’ period of purging, where selection on individual hybridization-derived haplotypes only subtly shifts genome-wide ancestry proportions. The shape and rate of this change in ancestry can vary from species to species (Veller, 2019), primarily as a function of the total recombination rate (see Principle 3). These predictions have been explored in a handful of systems to date (hominins, swordtails, Drosophila), but theory suggests they should be widespread. Similarly, research in this area has primarily focused on pulses of admixture but related patterns are expected for recent migrants in hybrid zone models (Sedghifar et al., 2015; Sedghifar et al., 2016).

Although the sources of selection on hybrids undoubtedly differ between species (Juric et al., 2016; Schumer et al., 2018; Brandvain et al., 2014), studies across diverse taxa have largely found that regions of the genome that are dense in coding or conserved elements tend to be particularly resistant to movement between species (Sankararaman et al., 2014; Teeter et al., 2008; Turissini and Matute, 2017; Martin et al., 2013; Calfee, 2021; Brandvain et al., 2014; Masly and Presgraves, 2007; Maxwell et al., 2018). In the case of conserved regulatory elements in humans, this pattern is stronger at enhancers that harbor derived mutations as opposed to ancestral variants (Telis et al., 2020). The consistency of the observation that introgression is depleted in functionally important regions implies that selection against minor parent ancestry generates barriers to introgression that are, in many cases, common, functionally broad, and polygenic (Calfee et al., 2020). These genome-scale observations echo classic work reporting depleted introgression on sex chromosomes (Martin et al., 2013; Payseur, 2004; Presgraves, 2008) and asymmetry in the effects of hybridization between the sexes (e.g., Orr, 1997; Schilthuizen et al., 2011; Turelli, 1998; Haldane’s rule), well-accepted rules in the speciation literature (Masly and Presgraves, 2007) that have been thoroughly reviewed elsewhere (Presgraves, 2008; Payseur and Rieseberg, 2016).

Selection acts to remove many introgressed haplotypes after hybridization. Because haplotypes derived from the minor parent species are longer in regions of the genome where recombination events are rare, minor parent haplotypes in low recombination rate regions are more likely to harbor variants that will be harmful in hybrids. This is conceptually similar to the reason why ancestry proportions shift drastically in the early generations after hybridization when ancestry tracts are long (i.e., Principle 1). Even after genome-wide admixture proportions have stabilized, theory predicts that minor-parent ancestry will be more fully removed from regions of the genome with low recombination rates (Wu, 2001; Nachman and Payseur, 2012). Data from diverse taxa, including swordtail fishes, humans, monkeyflowers, maize, and Heliconius butterflies (Schumer et al., 2018; Calfee, 2021; Edelman et al., 2019; Brandvain et al., 2014; Martin et al., 2019), support this theoretical prediction (but see Corbett-Detig and Nielsen, 2017). However, differing correlations between recombination rate and gene density can lead to local differences in minor parent ancestry, depending on where in the genome recombination primarily occurs. For example, in humans, recombination rates tend to be locally reduced near genes (Myers et al., 2005; Coop et al., 2008), resulting in a tendency to purge introgressed DNA near genes driven by both Principles 2 and 3, while in swordtail fishes and birds, recombination rates are elevated near genes (Baker et al., 2017; Singhal et al., 2015), pitting these rules against one another. In fact, the rapid evolution of the recombination landscape in some taxa (Auton et al., 2012; Baker et al., 2015) may be another factor contributing to variation in the landscape of introgression across species groups.

Dobzhansky–Muller hybrid incompatibilities (DMIs) occur when mutations that have arisen in each parental species’ genome interact negatively in hybrids (Dobzhansky, 1982). DMIs are the best documented and best understood mechanism of selection on hybrids. Indeed, the search for DMIs predates the recognition of the ubiquity of hybridization (Barbash, 2010). In addition to incompatible substitutions that arise in directly interacting proteins, DMIs can take the form of reciprocal losses following gene duplication or modifications in co-evolving regulatory elements, among other mechanisms (Mack and Nachman, 2017; Dion-Côté and Barbash, 2017; Nguyen et al., 2017; Brucker and Bordenstein, 2013). The DMI model has also been explored in the context of developmental pathways (commonly referred to as ‘developmental systems drift’), where compensatory changes in gene interaction networks can lead to divergent molecular pathways that are incompatible in hybrids, despite resulting in the same phenotypes in both parent species (Schiffman and Ralph, 2018; Tulchinsky et al., 2014; Porter and Johnson, 2002). The loci involved in DMIs identified to date are functionally diverse (Presgraves, 2010a; Maheshwari and Barbash, 2011; Atanasov et al., 2018; Dujon and Louis, 2017), but existing theory and data have hinted at broader evolutionary forces that drive the emergence of hybrid incompatibilities.

DMIs are largely expected to locally restrict gene flow by preventing introgression at the incompatible loci and regions linked to them (Teeter et al., 2008; Zuellig and Sweigart, 2018; Macholán et al., 2008; Macholán et al., 2008; Carling and Brumfield, 2009), but can also favor the adaptive introgression of pairs of compatible alleles (Bank et al., 2012). Thus, the genomic location of DMIs and the forces that drive their evolution will directly impact where introgression can occur in the genome. One well-established example is the observation that introgression is reduced on the sex chromosomes, presumably because DMIs are overrepresented on the sex chromosomes, due to factors such as faster X evolution, meiotic drive, and the importance of X chromosome genes in male fertility (Trier et al., 2014; Presgraves, 2008; Maheshwari and Barbash, 2011; Qvarnström and Bailey, 2009; Storchová et al., 2010). Beyond sex chromosomes, certain genes appear to be repeatedly involved in hybrid incompatibilities (Figure 1; Trier et al., 2014; Atanasov et al., 2018; Barr and Fishman, 2010; Chase, 2007; Schartl, 2008; Arévalo, 2020; Smagulova et al., 2016; Davies et al., 2016). While some of this overrepresentation may reflect sampling biases (Chase, 2007), as DMIs are characterized across more species it will become increasingly possible to test the hypothesis that certain genes act as ‘hotspots’ for the formation of hybrid incompatibilities. Looking forward, unanswered questions about the number of DMIs that distinguish recently diverged species, the strength of selection acting on them (Schumer et al., 2014), and the rate at which they evolve (Presgraves, 2010b) will be crucial in distinguishing signatures of selection against DMIs from other forms of selection on hybrids.

Historically researchers have considered selection against introgression to reflect interactions between diverged genomes. However, processes occurring within populations can also generate barriers (or thoroughfares) to introgression (Juric et al., 2016; Harris and Nielsen, 2016; Bierne et al., 2002). In other words, selection on introgressed ancestry might reflect the unconditional deleterious effect of a mutation, rather than its poor interaction with other sites in the genome (as seen in DMIs). Such mildly deleterious alleles will preferentially reach fixation in populations with weaker purifying selection, such as those with smaller effective population sizes. With sufficient time, a large number of weakly deleterious mutations can accumulate within a species, which would generate a strong selective force after hybridization with a species that harbors fewer such mutations. In such cases, hybridization can introduce deleterious mutations at much higher frequencies in the recipient population than expected from mutation-selection balance in the species with higher historical effective population size. Although each mutation is weakly selected against individually, in aggregate these mutations strongly reduce hybrid fitness relative to populations with fewer deleterious mutations, because they are linked to the same haplotypes after hybridization. Interestingly, this prediction holds even if the census population size of the admixed population is relatively small (Schumer et al., 2018). In the case of a pulse model of admixture, after genome-wide admixture proportions have equilibrated, selection against specific deleterious sites may still drive long-term ancestry purging (see Effects of hybrid demography). Empirical studies support this prediction, showing that ancestry from the species with less effective purifying selection can be depleted over many generations of selection, particularly in coding and conserved non-coding regions of the genome (Juric et al., 2016; Harris and Nielsen, 2016).

In contrast to other models, which predict widespread selection against minor parent ancestry, the additive hybridization load model predicts selection for directional introgression from the species that harbors fewer deleterious mutations. Alternatively, if deleterious mutations are recessive, theory and some empirical data predict that selection will favor an excess of foreign ancestry (Box 2; Harris and Nielsen, 2016; Kim, 2017), especially in the species with lower historical population sizes. In principle, selection against hybridization load could produce patterns that are distinguishable from other models of selection on hybrids because these weakly deleterious mutations are expected to be broadly distributed throughout the genome and fall within a particular range of selection coefficients (Box 1; Juric et al., 2016; Harris and Nielsen, 2016).

Ecological selection is a potentially important but poorly understood source of selection on hybrids. This is in part because less is known about both the genetic architecture of ecological adaptation and the ways in which ecological traits can become decoupled in hybrids. Moreover, this source of selection is sensitive to the environments in which hybrids find themselves.

Hybrids may express ecological traits that are intermediate to those of their parent populations (e.g., Hermansen et al., 2011; Hayden et al., 2011) or express ‘phenotypically mismatched’ traits (Thompson, 2019a and Box 2). In such cases, ecological selection will disfavor hybrids (Delmore and Irwin, 2014; Linn et al., 2003; Gow et al., 2007; Melo et al., 2014; Soria-Carrasco et al., 2014; Nosil et al., 2005; Scordato et al., 2020), unless hybrids exist in an intermediate ecological niche or an environment favoring these mismatched phenotypes (Schluter, 2000; Stelkens et al., 2009; Rieseberg et al., 1999; Selz and Seehausen, 2019; Hessenauer et al., 2020). Like the DMI and hybridization load models, ecological selection is predicted to result in biased ancestry around functionally relevant genomic regions, though the expected direction of bias depends on the environment (Box 1).

What patterns of ancestry can indicate the presence of ecological selection on hybrids? The answer to this question depends largely on the architecture of ecological traits (Thibert-Plante and Hendry, 2009; Zhang et al., 2020; Comeault, 2018; Lindtke et al., 2013). While it is straightforward to make predictions about the outcome of ecological selection on hybrids when the trait in question is controlled by a handful of genes, we know less about ancestry shifts after hybridization in ecologically relevant traits with a highly polygenic basis. Theory has explored how traits with a polygenic genetic architecture respond to different types of selection within a species (Barghi et al., 2020; Hayward and Sella, 2019), but these models do not capture the increased trait variance and ancestry linkage disequilibrium expected in hybrids (see next section).

Our discussion of ecological selection on hybrids above ignores ‘transgressive’ segregation – where hybrid trait values fall outside of the range of phenotypes observed in either parent (Rieseberg et al., 1999). We discuss the possible interaction of ecological selection and transgressive segregation in Box 2.

Given that populations evolve independently before admixture, hybridization has the potential to decouple suites of co-adapted alleles originally linked within the parental species. In hybrids, selection on polygenic traits has been frequently modeled using Fisher’s geometric model (Fisher, 1930), a simple mathematical description of the distance of an individual from its phenotypic optimum, that predicts many of the dynamics of selection against hybrids (Simon et al., 2018; Schiffman and Ralph, 2018; Tulchinsky et al., 2014; Porter and Johnson, 2002; Tenaillon, 2014; Schneemann, 2019; Yamaguchi and Otto, 2020; Fraïsse et al., 2016). We note that because of its generality Fisher’s geometric model has also been used to model selection on DMIs among other phenomena, but focus on its application to polygenic traits here.

In a Fisherian model of polygenic adaptation, individual fitness in the parental species can be described as a function of distance from a phenotypic optimum in quantitative trait space, and isolated populations maintain their respective optima through the independent fixation of sets of trait-increasing and trait-decreasing alleles (Tenaillon, 2014). Crucially, given enough time, the sets of loci underlying the trait and the sign of their phenotypic effects are likely to differ across populations, even between populations with identical phenotypic optima (Slatkin and Lande, 1994; Barton, 2001). In hybrids, recombination decouples these sets of parental alleles. This can result in hybrid phenotypes that fall outside of the phenotypic optima of either parental species, reducing fitness through a phenomenon known as segregation load (Barton, 1989; Slatkin and Lande, 1994). More precisely, when parental alleles are mixed into different genetic backgrounds, hybrids can show greater variance in a trait than observed in either of the parental species (Figure 2). If the trait is also under stabilizing selection in hybrids, this increased variance could drive purging of minor parent ancestry over time. Notably, these predictions should hold when parental species are adapting to similar (Melo, 2019) or distinct (Walter et al., 2020) phenotypic optima (Figure 2), and when genotypic effects are non-additive (Schneemann, 2019; Fierst and Hansen, 2010).

There is no doubt that haplotypes introduced by hybridization can confer an adaptive advantage (Hedrick, 2013; Racimo et al., 2015; Song et al., 2011; Heliconius Genome Consortium, 2012; Norris et al., 2015; Feurtey et al., 2019; Corcoran et al., 2016). Along with hybrid incompatibilities, adaptive introgression is among the best studied consequences of hybridization and has been thoroughly discussed in many previous reviews (Marques et al., 2019; Hedrick, 2013; Racimo et al., 2015; Suarez-Gonzalez et al., 2018; Mallet, 2005). Here, we briefly summarize what is known about the impact of adaptive introgression on ancestry in the genome. Intuitively, adaptive introgression increases minor parent ancestry locally around the adaptive allele. The footprint of these peaks in minor parent ancestry is expected to be positively correlated with the selection coefficient in the absence of negative selection on linked sites (see below). However, adaptive haplotypes can actually be shorter in certain demographic scenarios, since they persist longer in hybrid populations, leaving more time for them to be broken up by recombination (Sedghifar et al., 2016; Shchur et al., 2020).

Recent work has also highlighted how adaptive introgression can generate unique signatures of ancestry heterozygosity at neighboring sites (Setter et al., 2019). As an adaptive haplotype sweeps to fixation, it drives minor parent ancestry to intermediate frequency in large flanking regions, creating a ‘volcano-shaped’ signature of genetic diversity approaching the site under positive selection (Setter et al., 2019). This shape is generated because recombination events that shorten the haplotype but still contain the adaptive allele are retained during the sweep, resulting in a concentration of ancestry heterozygosity at the edges of a fixed adaptive haplotype (Setter et al., 2019; Moest et al., 2020).

Given broad selection against minor parent ancestry, adaptive introgression is often occurring against a background of genome-wide purging (Calfee, 2021; Edelman et al., 2019; Turner and Harr, 2014). In some cases, this has made adaptively introgressed haplotypes easy to identify empirically since they form peaks of high minor parent ancestry against a background of low minor parent ancestry (Sankararaman et al., 2014; Vernot and Akey, 2014). This also means that several factors will impact the probability that a globally adaptive allele will introgress between species. These include the locations of potentially adaptive alleles relative to deleterious neighbors, the relative selection coefficients on adaptive and deleterious sites, and other features of genome organization (Veller, 2019). Although there is little empirical or theoretical work in this area to date, some predictions can be made from first principles. For example, an adaptive haplotype in a region of the genome with a very low recombination rate would have a lower probability of introgressing than a haplotype with the same advantage in a higher recombination rate region (Principle 3).

These factors highlight that beyond distinguishing between sources of selection on hybrids (Box 1), another difficult hurdle is characterizing how they may interact. Although research to date has largely focused on each mechanism in isolation, most hybridization events likely involve the interplay between several modes of selection. As discussed above, in the admixture event between humans and Neanderthals, both hybridization load and adaptive introgression have shaped Neanderthal ancestry in modern human genomes (Sankararaman et al., 2014; Juric et al., 2016; Harris and Nielsen, 2016). This combination of positive and negative selection on hybridization-derived haplotypes can generate interference, especially in the early generations following hybridization when long haplotypes of each ancestry type are common (Figure 3). Simulations hint that it may be possible to disentangle different signals of selection on hybrids using local ancestry variation (Figure 3) or changes in ancestry over time (Figure 4; Uecker et al., 2015), presenting exciting opportunities for future work.

The mechanisms discussed above likely represent an incomplete picture of the breadth of forms of selection on hybrids. For example, weak but pervasive epistatic interactions (e.g., of interacting genes in pathways) could select for similar shifts in ancestry as expected from selection on polygenic traits, but whether such weak epistatic interactions are common is unknown. It is also important to also note that, to some extent, the distinctions made above between different sources of selection can be arbitrary and not biologically meaningful. In some cases, two or more of the selective frameworks may be simultaneously applied – such as a DMI caused by genes underlying an ecologically relevant trait (Arnegard et al., 2014; Thompson, 2019a).

In other cases, multiple selective frameworks may best describe different aspects of the same empirical case. For example, in the case of hybrid gene regulation two frameworks of selection may be applied to the same genes. Often under stabilizing selection within the parental species, it is common for cis- and trans-acting regulatory factors to show evidence of compensatory evolution within species (Mack and Nachman, 2017; Mack et al., 2016; Sánchez-Ramírez et al., 2021). As a result, mismatches in these interacting factors in hybrids can lead to dramatic under- or overexpression of the genes they regulate (Figure 5). We speculate that this type of misexpression could result in two forms of selection on hybrids. Large-effect expression aberrations would be selected against as a DMI, via selection acting against heterospecific allelic combinations at cis- or trans-acting loci. For example, allelic combinations that reduce or eliminate expression of a given gene (Figure 5) can lead to strong selection on this non-functional genotype combination. After the misexpression is resolved, additional smaller effect variants from the two parental species may still have an impact on variance in expression (e.g., Figure 2B). Changes in ancestry at these variants would then be driven by stabilizing selection on the overall expression of the gene.

Such ‘priority’ effects of selection on hybrids, with rapid purging of interactions in response to strong selective pressures and slower purging associated with weaker selective pressures, are reminiscent of the fast versus slow purging of ancestry tracts after initial hybridization (Principle 1). While in many cases there is no strong line between the mechanisms of selection discussed in this section, we propose that this approach of considering phases of selection on hybrids may be a fruitful way of understanding the complexity of several intertwined selective forces acting on hybrid genomes.

Biologists have long been fascinated with the question of whether evolution is predictable (Blount et al., 2018). A key unanswered question is the extent to which we can predict outcomes of hybridization within and between pairs of species. At a broad scale, some predictions can be made due to the interplay between selection and features of genomic organization such as recombination rate and the locations of coding and conserved basepairs, which appear to have consistent effects on ancestry in many species (e.g., Principles 2 and 3). Moving beyond these broad-scale features, there are good reasons to expect that replicated hybridization events between the same species will lead to predictable outcomes at the genomic level. In repeated hybridization events between the same species, the same genetic interactions and selective forces are predicted to drive concordant changes in ancestry along the genome. Indeed, this has been observed in experimental hybrid populations, natural hybrid populations (Schumer et al., 2018; Chaturvedi et al., 2020; Matute et al., 2020; Calfee et al., 2020; Kovach et al., 2016; Bay et al., 2019; Hufford et al., 2013; Riquet et al., 2019; Smukowski Heil et al., 2019), and in replicated cline studies (Westram et al., 2021; but see Nolte et al., 2009).

While it seems sensible to expect that replicated hybridization events should lead to similar patterns of local ancestry, recent work has suggested that in some cases we may expect more repeatability across taxa than predicted by classic evolutionary theory (Orr, 1995). Studies in Arabidopsis and Xiphophorus have repeatedly uncovered some of the same genes underlying hybrid incompatibilities (Figure 1; Chae et al., 2014; Alcázar et al., 2014; Powell et al., 2020; Schartl et al., 2010), and certain genetic interactions, such as cytonuclear incompatibilities, are common across the tree of life (Chase, 2007; Hill, 2017; Bar-Yaacov et al., 2015; Beresford et al., 2017; Bolnick et al., 2008; Burton and Barreto, 2012; He et al., 2020; McDaniel et al., 2007; Giordano et al., 2018). These results suggest that some types of genetic interactions are more prone to breaking down in hybrids, perhaps due to their function, the rate at which they accumulate substitutions, or their position in a gene network. Whether incompatibilities frequently evolve in the same genes or pathways has important implications for whether we expect regions resistant to introgression to be shared across species.

Similarly, comparisons of dynamics of adaptive introgression across species have identified possible hotspots in terms of gene families that appear to confer advantages when introgressed. This process could also generate fine-scale repeatability in local ancestry. The best documented examples include immune-related genes (Sankararaman et al., 2014; Enard and Petrov, 2018; Fijarczyk et al., 2018), pigmentation genes (Moest et al., 2020; Vickrey et al., 2018; Giska et al., 2019; Jones et al., 2018; Rotival and Quintana-Murci, 2020; Stryjewski and Sorenson, 2017; Wallbank et al., 2016), and genes that underlie resistance (Song et al., 2011; Lindtke et al., 2013; Norris et al., 2015; Bechsgaard et al., 2017; Whitney et al., 2006; Ford et al., 2015), but many other functional categories of genes or selfish elements could have similar behavior (Staubach et al., 2012; Meiklejohn et al., 2018).

Compared to incompatibilities and adaptive introgression, we know much less about how other forms of selection on hybrids might lead to predictable outcomes at the local scale. Although this has not been directly studied, selection against hybridization load could lead to partially predictable outcomes across replicated hybridization events. Regions of the genome with lower local N_e (i.e., due to variation in the effects of background selection; McVicker et al., 2009; Elyashiv et al., 2016) should accumulate more weakly deleterious mutations within populations and thus be more likely to be purged after hybridization. Additionally, gene-dense regions provide a larger target for functionally relevant mutations to occur and may therefore experience stronger selection in the early generations after hybridization when ancestry tracts are long.

For other mechanisms of selection, we expect much lower predictability across systems. For example, if species have independently adapted to distinct ecological conditions, we would not expect the genetic architecture of such traits to be shared except in rare cases (e.g., Jones et al., 2012). Without selection on the same underlying regions of the genome, any repeatability in local ancestry patterns in hybrids should not exceed what is expected due to broad-scale features such as gene density (Principles 2 and 3).

Conserved mechanisms that shape ancestry after hybridization can also point to cases where we predict to see differences between species. We recently found differences in the extent to which introgressed haplotypes were retained in coding regions in the genomes of swordtail fishes and humans, likely due to differences in the underlying recombination maps (Schumer et al., 2018). Both species share a strong positive correlation between introgression and the local recombination rate. However, recombination is concentrated in promoters and other functional regions in swordtail fishes (Baker et al., 2017), and tends to occur away from such regions in humans (Myers et al., 2005; Coop et al., 2008). This results in distinct patterns of local ancestry, with swordtail fishes retaining more minor parent ancestry than humans in and around genes (presumably due to differing outcomes of the action of Principles 2 and 3 in the two species groups). As data from more diverse systems accumulates, comparative analyses of patterns of introgression as a function of these features of genome structure, combined with theoretical analyses, will further develop our understanding of how selection acts in admixed genomes.

Similarly, as discussed in Principle 1, the speed of initial purging of minor parent ancestry is sensitive to the aggregate recombination rate, which differs widely between species (Veller, 2019). This is because the aggregate recombination rate is strongly influenced by the total number of chromosomes and whether recombination occurs in both sexes – properties that vary widely across the tree of life (Veller, 2019; Stapley et al., 2017). Notably, these factors together may be important in explaining the variation in admixture proportions observed in the genomes of different species that are known to commonly hybridize, from cases where retention of minor parent ancestry after hybridization is rare, such as Drosophila (Coyne and Orr, 2004), to those where extensive introgression is common, such as swordtail fishes (Cui, 2013).

As is the case in non-admixed populations, we expect that certain features of genome evolution will be sensitive to the demographic history of hybrid populations themselves. The importance of demography has long been appreciated in the theoretical literature on tension zones (though less frequently incorporated into empirical analyses), where dynamics of dispersal and population density at the contact zone (Pierce et al., 2017; Larson et al., 2019) play a key role in the outcomes of hybridization and interpretation of cline analysis (Payseur, 2004; Barton, 2001; Barton and Hewitt, 1989; Bierne et al., 2011). The impacts of demography on hybridization have been less thoroughly explored in the context of pulses of admixture. However, there are multiple reasons to predict that pulses of hybridization may coincide with strong bottlenecks since they are often driven by ecological disturbance (Fisher et al., 2006; Chunco, 2014) and because selection on hybrids can be so strong that it essentially drives population collapse (Shorter et al., 2017).

Intuitively, the long-term size of hybrid populations and the proportion of parental genetic diversity retained in hybrids should have important impacts on genome evolution. In many cases, selection on hybrids will be strong enough to overcome the effects of genetic drift, even in small populations, especially in early generation hybrids when many selected sites are linked. Over long time periods, however, populations with a small effective size will be less efficient at purging weakly deleterious variants that occur in short ancestry tracts.

Another important consideration is the number of parental individuals from each species that contributed to a hybridization event, which will shape the raw material on which selection can act. We recently mapped the genetic basis of a hybrid melanoma that develops from a tail pigmentation spot in swordtail fishes. Notably, this tail pigmentation spot is polymorphic in one of the parental species (~30% frequency; Powell et al., 2020). Presumably due to differences in the founding parental populations, some hybrid populations have both a high frequency of the tail spot and of melanoma, whereas others have a low frequency of both (Powell et al., 2020). Though just one example, this highlights how the genetic contribution of the parental species can be an important element influencing how selection will act within hybrid populations. Studies in other systems such as Drosophila, Mimulus, Mus, and Caenorhabditis elegans have identified polymorphic hybrid incompatibilities, suggesting that these founder dynamics could have important impacts on hybrid populations (Zuellig and Sweigart, 2018; Corbett-Detig et al., 2013; Cutter, 2012; Larson et al., 2018; Turner et al., 2012; Good, 2008).

These factors will also impact the repeatability and predictability of genome evolution after hybridization. Distinct demographic histories in hybrid populations or variation in genetic contributions from the parental species could drive differences in local ancestry between populations by shaping features such as the distribution of ancestry tract lengths and the efficacy of selection. These factors could in turn impact the inferences researchers make about the extent of repeatability of local ancestry patterns in studies of replicate hybrid populations (Schumer et al., 2017). Whether demographic differences will substantially limit repeatability in local ancestry in replicate hybrid populations will depend in part on the strength of selection relative to demographic forces such as genetic drift. Simulations matching the inferred demographic histories of independently formed hybrid populations can help researchers begin to tease apart differences due to distinct sources of selection versus demography.

In previous sections, we discussed the uncertainties surrounding how hybrid incompatibilities arise and the degree to which we expect incompatibilities to arise repeatedly (Figure 1), either in the same genes (Powell et al., 2020; Schartl et al., 2010) or in the same regions of the genome (Schumer et al., 2018; Brandvain et al., 2014; Martin et al., 2019). Such repeatability in the evolution of hybrid incompatibilities could undermine assumptions of the snowball effect, which posits that because newly arising mutations in one species can interact with any derived mutations in the second species, the number of incompatibilities between two species should increase exponentially over time (Orr and Turelli, 2001). Data consistent with this phenomenon has been documented in several systems (Matute et al., 2010; Moyle and Nakazato, 2010; Wang et al., 2015). However, if the mutations that cause DMIs are enriched in the same genes or genomic regions, the rate of this accumulation should slow (Kalirad and Azevedo, 2017; Maya-Lastra and Eaton, 2021). Similar predictions emerge from theoretical studies of gene regulatory network evolution, where the likelihood of a gene’s involvement in DMIs is directly related to the density of the gene network (Yang and Scarpino, 2020; Satokangas, 2020). Systematic differences in gene network connectivity between species could drive differences in the distribution of DMIs across the tree of life (Cutter and Bundus, 2020). Though limited by the experimental and statistical challenges inherent in identifying DMIs, both evidence for DMI ‘hotspots’ and a slowed snowball effect should be detectable from empirical data in experiments with sufficient power.

The differences in genetic architecture assumed by each model of selection on hybrids is one promising route to inferring their role in shaping local ancestry after hybridization. Selection on DMIs is generally thought to be stronger and less polygenic than hybridization load models (Box 1; but empirical evidence is lacking, see Fierst and Hansen, 2010 for an exploration of polygenic epistatic selection). Higher levels of polygenicity will increase the proportion of neutral basepairs that are linked to sites that are deleterious in hybrids (Figure 4). Moreover, under a DMI model ancestral and transitional genotypes can be favored by selection, which will actually act to increase minor parent ancestry in some regions of the genome (Figure 4—figure supplement 1). Together these factors will lead to greater purging of minor parent ancestry over time under polygenic models of selection against minor parent ancestry, as opposed to classic DMI models (Figure 4). Comparing the predictions of these different architectures of selection on hybrids using modeling or simulations could serve as a powerful tool to allow researchers to distinguish between them, at least on a genome-wide scale (as in Juric et al., 2016).

Studies of selection in contemporary hybridizing populations offer another route to merge pattern and process, and to tease apart forms of selection acting in admixed populations. For example, Chen et al., 2019; Chen et al., 2016 and Fitzpatrick, 2019 studied weakly differentiated populations and found that genome-wide selection broadly favored ancestry derived from migrants in small populations, consistent with the idea that in small populations foreign ancestry can be favored to lighten the genetic load. In contrast, we recently found little evidence that hybridization load shapes genome-wide ancestry in hybrid swordtail populations formed between species with substantial differences in historical effective population size (Schumer et al., 2018). While these studies used genomic tracking in natural populations, other researchers have leveraged laboratory crosses and systematically varied environmental conditions to explore how ecological selection shapes genome evolution (Smukowski Heil et al., 2019). Combining such observational and manipulative approaches with comparisons across diverse species may reveal the relative importance of the forces shaping evolution after hybridization along the speciation continuum.

Several studies have highlighted the key role of local and global recombination rates in mediating the retention and purging of minor parent ancestry in the presence of selection on hybrids. It is also possible that hybrids will have different recombination landscapes than their parental species due to features such as structural differences between species that suppress recombination (Kirkpatrick, 2010), dysfunction in the recombination process (Davies et al., 2016), recombination modifiers (Brand et al., 2018), or different hotspot usage in hybrids compared to the parental species (Li et al., 2019). While historically the difficulty of constructing accurate recombination maps for non-model species would have made understanding these complexities infeasible, the dropping costs of sequencing in combination with new methods for directly mapping double-strand breaks in meiotic cells (Canela et al., 2016) may soon put the answers to these questions within reach for a number of hybridizing species.

Examples of hybridization across the tree of life poise the field for a broader analysis of what genetic and biological features are associated with variation in rates of introgression. For one, theory predicts that species with fewer chromosomes will undergo faster and stronger purging of minor parent ancestry in their genomes, due to a low aggregate recombination rate (Veller, 2019, e.g., in species such as Arabidopsis, Drosophila, mosquitoes). In addition to empirical analyses to address key theoretical predictions, the wealth of newly available data opens up a large number of possible studies of underexplored features of organismal biology that could influence retention of minor parent ancestry after hybridization, which we discuss briefly here.

Life history traits may play an important role in variation in introgression across the tree of life (Nouhaud et al., 2020). For example, the extent of selfing or asexual reproduction impacts the genetic diversity of the parent populations, their genetic load, and the frequency with which recombination reshuffles parental haplotypes, and therefore can shape the extent and direction of introgression (Pickup et al., 2019; Hamlin et al., 2020). Similarly, some data suggest that systems with facultative asexual reproduction can retain larger minor parent contributions (Langdon et al., 2019; Marcet-Houben and Gabaldón, 2015; Gallone et al., 2019), and tolerance of genome duplication and aneuploidy will interplay with retention or loss of parental genomic material (Steensels et al., 2021).

Variation in the structure and function of the genome between species may also play a key role. Decades of work have established an important role for inversions in locally restricting (or promoting; Edelman et al., 2019; Jay et al., 2018; Todesco et al., 2020) gene flow (Noor et al., 2001; Wellenreuther and Bernatchez, 2018; Kirkpatrick and Barton, 2006; Faria et al., 2019). Beyond inversions, the frequency and activity of transposable elements in the genome is a classic mediator of selection against hybrids, but mixed evidence for its generality necessitates broader study (Smukowski Heil et al., 2021; Castillo and Moyle, 2020; Dion-Côté et al., 2014). Gene expression (or misexpression) that is specific to life cycle stage or tissue type could lead to temporal or tissue-specific fitness effects in hybrids. Notably, recent work has demonstrated that there is weaker selection against Neanderthal ancestry in enhancers that are tissue specific in modern humans (Telis et al., 2020). This highlights the potential for such context dependence, which would certainly vary across species groups (e.g., fungi versus plants and animals), and shape how admixed genomes are exposed to the varied forms of selection discussed above. Other features of genome organization that differ widely across species groups such as the presence of micro-chromosomes, polyploidy, and recombination mechanism will all be rich areas to study in this regard (Singhal et al., 2015; Nouhaud et al., 2020; Marcet-Houben and Gabaldón, 2015; D'Angiolo et al., 2020; Bozdag et al., 2021; Lamichhaney et al., 2020; Elgvin et al., 2017; Soltis, 2004; Lafon-Placette et al., 2018; Backström et al., 2008).

The hybridization-mediated transfer of parental alleles that increase fitness in the recipient population.

A more general term than hybridization that encompasses all gene flow between distinct populations, which may or may not be diverged enough to be considered species.

Statistical association between haplotypes of the same ancestry that can be caused by physical linkage of sites, selection, or population structure; in the case of linkage disequilibrium due to physical linkage, ancestry LD extends over a much greater physical distances than is typical for non-admixed populations.

Selection driven by the fitness of an organism’s traits in the context of its environment.

A general model of selection where fitness is determined by distance from a phenotypic optimum, which has been applied in the hybridization literature to describe selection on polygenic traits (either stabilizing or directional; Figure 2), ecological selection on hybrids, and hybrid incompatibilities.

The number, effect size, and location in the genome of loci contributing to a phenotype.

A physically contiguous tract of DNA inherited from a single parent unbroken by recombination.

Mutations that arise in interacting genes after two lineages diverge such that when individuals from these populations hybridize a previously ‘untested’ combination of alleles reduces hybrid viability or fertility.

The burden of mildly deleterious mutations that preferentially accumulated in the parental lineage with less effective selection, leading to reduced fitness of hybrids that harbor more of that species’ genome and selection against ancestry derived from that species. Fitness in hybrids is not reduced relative to the parental species with lower historical effective population size.

Transfer of a region of the genome between species due to hybridization.

The species that contributed a majority of the genome of an admixed population.

The species that contributed a minority of the genome of an admixed population.

A trait where phenotypic variation is explained by the combined effects of many, sometimes thousands, of variants spread throughout the genome.

The decrease in average fitness of hybrids expected due to the disruption of co-adapted sets of alleles inherited from the parental species that are broken apart by recombination and independent assortment.

Selection driven by mate choice and competition for mates.

The faster-than-linear increase in the number of DMIs with increasing numbers of substitutions between two species that is predicted by evolutionary theory.

Two groups of organisms where hybrids between them have reduced viability or fertility. This can range from moderate impacts on viability or fertility to complete inviability or infertility.

Spatial zone where hybrids form between the geographic regions occupied by two parental species.

A stable zone where hybrids are found in a narrow geographical region as a result of balance between ongoing dispersal of individuals from parental populations and strong selection against hybrids.