Adaptive introgression in animals: examples and comparison to new mutation and standing variation as sources of adaptive variation

Authors


Correspondence: Philip W. Hedrick,

E-mail: philip.hedrick@asu.edu

Abstract

Adaptive genetic variation has been thought to originate primarily from either new mutation or standing variation. Another potential source of adaptive variation is adaptive variants from other (donor) species that are introgressed into the (recipient) species, termed adaptive introgression. Here, the various attributes of these three potential sources of adaptive variation are compared. For example, the rate of adaptive change is generally thought to be faster from standing variation, slower from mutation and potentially intermediate from adaptive introgression. Additionally, the higher initial frequency of adaptive variation from standing variation and lower initial frequency from mutation might result in a higher probability of fixation of the adaptive variants for standing variation. Adaptive variation from introgression might have higher initial frequency than new adaptive mutations but lower than that from standing variation, again making the impact of adaptive introgression variation potentially intermediate. Adaptive introgressive variants might have multiple changes within a gene and affect multiple loci, an advantage also potentially found for adaptive standing variation but not for new adaptive mutants. The processes that might produce a common variant in two taxa, convergence, trans-species polymorphism from incomplete lineage sorting or from balancing selection and adaptive introgression, are also compared. Finally, potential examples of adaptive introgression in animals, including balancing selection for multiple alleles for major histocompatibility complex (MHC), S and csd genes, pesticide resistance in mice, black colour in wolves and white colour in coyotes, Neanderthal or Denisovan ancestry in humans, mimicry genes in Heliconius butterflies, beak traits in Darwin's finches, yellow skin in chickens and non-native ancestry in an endangered native salamander, are examined.

Introduction

The environment for most organisms constantly changes in some respects, making adaptive responses essential for continued persistence. In addition, the current environment for many organisms is also changing rapidly because of human-caused modifications such as climate change, increase in non-native species, decline in native species and impacts on other aspects of the environment. Generally, it is assumed that species that adapt to such environmental challenges change genetically, either from selection on existing (standing) genetic variation or by utilizing new mutants that provide a selective advantage (Barrett & Schluter 2008).

In addition, another potential source of adaptive genetic variation that has been documented, particularly in plants, is adaptive variation transmitted to the species by interbreeding or introgression with related taxa, a phenomenon now termed ‘adaptive introgression’ to distinguish it from introgression of nonadaptive (detrimental or neutral) genetic variation. To be more explicit, we should distinguish between ‘introgression that results in adaptive evolution’ and introgression that does not. Of course, it is implicit that both standing variation and new mutants used for adaptation are adaptive. Generally, the related taxa are different species (and below I will call them species or taxa), but as we will discuss below, they might be subspecies or even divergent populations of the same species as the marine and freshwater forms of threespine sticklebacks (Jones et al. 2012; as a result, in some cases, they will be called groups). Of course, when the taxa are different species or subspecies, the introduction of new variations from other taxa is generally called introgression because of the reproductive isolation present, while when the taxa are divergent populations of the same species, the source of adaptive genetic variation would be called gene flow.

Adaptive introgression has long been suggested as important by botanists with Anderson (1949) stating that ‘raw material brought in by introgression must greatly exceed the new genes produced directly by mutation’. In recent years, details of adaptive introgression have been documented for various traits in a number of plant species, such as between the sunflower species Helianthus annuus and H. debilis (Whitney et al. 2006, 2010), between the iris species Iris fulva and I. breviculis (Martin et al. 2006) and between the ragwort Senecio squalidus and the groundsel S. vulgaris (Kim et al. 2008). These examples in plants illustrate that adaptive introgression was important in adaptation to both the biotic and abiotic environments and in the recovery of ancestral traits lost in some taxa (Kim et al. 2008).

On the other hand, many zoologists (Mayr 1963; Dowling & Secor 1997) minimized the importance of hybridization in animals, and until recently, there have been few convincing examples of adaptive introgression in animals. The lack of examples of adaptive introgression in animals might be because the level of hybridization between animal species is often low. However, Grant & Grant (1992) suggested that nearly 10% of all bird species hybridize with a related species, and in addition, many amphibians (Wells 2007) and freshwater fishes (Scribner et al. 2001) hybridize. Low fitness found in many hybrid animals has also been thought to be a significant barrier to introgression (Barton 2001; Coyne & Orr 2004) although some hybrids show high fitness (Burke & Arnold 2001; Mallet 2005). One widely cited, early example of adaptive introgression in animals is that in the Australian fruit fly Bactrocera (formerly Dacus) tryoni. In this pest species, Lewontin & Birch (1966) suggested that the expansion of the geographical range of B. tryoni was the result of introgression from the closely related B. neohumeralis (Morrow et al. 2000; Clarke et al. 2010).

Recently, prominent adaptive introgression examples have been suggested for several animal species. The level of evidence for adaptive introgression varies in these examples, with some having convincing evidence, while in other ones, the evidence is not complete or is less compelling. In addition, adaptive introgression might be important in providing the genetic variation for the high rate of diversification documented in some adaptive radiations (Seehausen 2004; The Heliconius Genome Consortium 2012). Recently, Abbott et al. (2013) reviewed the general importance of hybridization in speciation and the potential influence of adaptive introgression on speciation (see also Barton 2013). Before discussing how adaptive introgression can be detected and some specific examples, let us first outline the broad differences and similarities between the three different potential sources of adaptive variation.

Sources of adaptive variation: new mutation, standing variation and adaptive introgression

One of the important unsolved questions in evolutionary biology is the source of adaptive genetic variation. Now we have the molecular tools to potentially identify whether specific adaptive variants are the result of new mutation, standing variation or adaptive introgression, and eventually we might be able to determine the relative importance of each of these sources of adaptive variation. How do the attributes of the three different sources of adaptive variation, mutation, standing variation and adaptive introgression, compare? Table 1 summarizes the differences and similarities between these sources with most supported evaluations possible for new mutation and standing variation (for further discussion and references, see Hermisson & Pennings 2005; Barrett & Schluter 2008; Chevin et al. 2010; Kawecki et al. 2012; Peter et al. 2012; Glemin & Ronfort 2013; Alcala et al. 2013) and only a very general assessment for the lesser known adaptive introgression. The amount of standing variation present in a particular species is a function of its past effective population size, and the probability of a favourable mutation is a function of the present population number (Hedrick 2011). Here, it is assumed that the adaptive variation from introgression is recent. If the introgression is ancient and/or long term, then adaptive variation from introgression might have characteristics more similar to standing variation.

Table 1. Comparison of the similarities and differences between new mutation, standing variation and adaptive introgression as sources of adaptive variation
 New mutationStanding variationAdaptive introgression
  1. For further discussion and references about the properties of new mutation and standing variation, see Hermisson & Pennings (2005); Barrett & Schluter (2008); Chevin et al. (2010); Kawecki et al. (2012); Peter et al. (2012); Glemin & Ronfort (2013); Alcala et al. (2013).

(a) Initial frequencyLowHigherLow
(b) Long waiting timeYesNoYes (or maybe)
(c) Mean selective advantageSometimes high but variableMay be lowSometimes high
(d) Negative pleiotropyYesNoNo
(e) Linked gene effectsYesLowYes
(f) RecessiveOftenLess likelyLess likely
(g) PretestedNoYesYes
(h) Multiple changesNoYesYes
(i) Multiple lociNoYesYes
(i) Molecular signatureYesLessYes
(j) Negative epistatic interactionNoNoYes
(k) Need hybridization with another speciesNoNoYes
(l) Need viable and reproductive F1 between speciesNoNoYes

Overall, adaptation from standing variation is thought to be faster and more certain, given that there is adaptive standing variation present (Hermisson & Pennings 2005; Barrett & Schluter 2008). On the other hand, many examples of adaptation to chemical pesticides and herbicides are thought to have originated from new mutations but many pests and weeds have large population sizes, facilitating the generation of and increase in these favourable mutants. Adaptive introgression appears to have been fast and effective in introducing adaptive variation in some plant species (Martin et al. 2006; Whitney et al. 2006, 2010; Kim et al. 2008) but it may be limited in animals because of lower hybridization frequency and lower fitness of hybrid F1s and backcrosses.

Two of the biggest differences between new mutation and standing variation are that adaptive mutants have both a much lower initial frequency and a longer waiting time until the variant is generated and present (Hermisson & Pennings 2005; Barrett & Schluter 2008). As we will discuss in Box 1, the initial frequency is a major determinant of the probability of fixation of a favourable variant. For adaptive introgression, the initial frequency might also be lower than that for standing variation and lower than that for new mutation. However, if introgression has occurred more than once, then the initial frequency could be higher. The waiting time also depends upon the amount of hybridization with the source (or donor) taxa. By definition, there is no waiting time for standing variation, while the expected waiting time for a new mutation is approximately 1/u generations (Kimura 1983) where u is the mutation rate per generation to a favourable mutant. This waiting time could be quite long if u is low but could be higher if there are n multiple loci that could mutate to a favourable allele; the expected waiting time is reduced to about n/u. If gene conversion (Chen et al. 2007; Klitz et al. 2012), which can generate new variants by transferring short sequences onto different haplotypic backgrounds and often occurs at a much higher rate than point mutation, is included as a source of adaptive variation from new mutation, then the waiting time for new mutation might be substantially reduced.

In addition, the advantageous effect of a new mutant might be larger, the negative pleiotropy effects greater, the linked gene effects stronger and likelihood of being recessive higher (for an examination of the level of dominance of new mutations, see Manna et al. 2011), than for standing variants. These effects would be likely because mutations are thought to be new and unfiltered (not pretested) by subsequent selective and recombinational events (Barrett & Schluter 2008). As a result, the initial net advantageous effect of a new mutant might be large in some cases or it might be reduced in other cases because of pleiotropy, linked gene effects or recessivity. Adaptive introgression variants could be new and their net effect higher than for mutational variants with sometimes high selective advantage, less pleiotropy, less strong linked gene effects and less recessivity. However, for an adaptive introgressed variant, the positive effect needs to be larger than the aggregate negative effects due to associated genes.

In the initial hybrid F1, the adaptive introgressive variant should be expressed, that is, it should not be completely recessive and should potentially have an advantageous effect on fitness. In the F1, all of chromosomes will be heterozygous in ancestry between the donor and recipient species, not just the chromosome with the adaptive variant so that the presence of outbreeding depression (low fitness) in this generation might greatly reduce the subsequent success of the adaptive variant.

Standing and adaptive introgression variants might include accumulated multiple changes in a variant, something that is very unlikely in a new mutant, and they may include changes at multiple different loci. Important for distinguishing the source of the adaptive variants, new mutants and introgressive variants generally should give the molecular signal of a selective sweep, while standing variants are likely to only provide the weaker signal of a soft selective sweep (Hermisson & Pennings 2005). Adaptive variants with larger effects might be incorporated more quickly and provide a stronger signal but variants with smaller effects are probably important in adaptation. Adaptive introgressive variants need to be introduced from the donor species into the recipient species by hybridization, and the F1 offspring from these crosses need to be both viable and reproductive (and either the F2 or backcrosses viable and reproductive). The present frequency of a potential adaptive variant would also depend upon the number of generations since it was first introduced. The variant might have a negative epistatic interaction in this new genetic background of the recipient species. In addition, the variant must also be adaptive in the environment of recipient species, which may be different from that of the donor species.

One of the biggest differences between standing variation and new mutation is the probability of fixation of favourable variants because of differences in initial frequency (Hermisson & Pennings 2005; Barrett & Schluter 2008). Box 1 examines this difference and suggests that adaptive variants introduced by adaptive introgression may be low in initial frequency if hybridization is infrequent or higher if hybridization events are more frequent and successful in producing fit F1s and backcrosses.

Box 1. Effect of the initial frequency on the probability of fixation

A major difference between adaptation from mutation and standing variation is that the initial frequency of a new mutant is 1/(2N) where N is the population size, while the frequency of an allele already segregating in the population could be substantially higher. Assume that there is selection s favouring an advantageous homozygote and that heterozygotes have a fitness exactly in between the two homozygotes (additivity). The probability of fixation of the favourable allele u(p), given that its initial frequency is p, is approximately (Kimura & Ohta 1971)

display math

To illustrate the difference between mutation and standing variation, assume that the initial frequency for an advantageous allele is = 1/(2N) for mutation and that 10/(2N) or 100/(2N) represent two levels of the initial frequency when there is standing variation. If = 10 000, then the initial frequencies are 0.00005, 0.0005 and 0.005, respectively. Assume that 2Ns ranges from 0 to 2000, which would be the range if = 10 000 and s ranges from 0 to 0.1. The figure gives the probability of fixation of the favourable allele, which for mutation = 1/(2N) ranges from 0 to 0.095 when = 0.1. On the other hand, if there is standing variation and the initial frequency is higher, the probability of fixation is much higher. For example, with = 10/(2N) or 100/(2N) and = 0.1, the probabilities of fixation are 0.632 and 1, respectively. For a comparison of the probability of fixation for standing variation, given the adaptive variant was already segregating under neutrality or mutation–selection balance when the environment that conferred advantageous selection began, see Hermisson & Pennings (2005).image_n/mec12415-gra-0001.png

The initial frequency of adaptive variants arising from introgression might have a higher initial frequency than those from mutation but lower than those from standing variation. This intermediate level might occur because several copies of the same adaptive allele could be introduced from one or more hybridization events. Or, F1 individuals from the cross of two related species may have higher fitness (heterosis) than the parental genotypes because some accumulated recessive detrimental variants are temporarily masked resulting in a higher effective initial frequency.

Distinguishing the processes of adaptive introgression, trans-species polymorphism from incomplete lineage sorting or balancing selection and convergence

When the same sequence or variant is found in different species that overall are substantially diverged from each other, there are several possible evolutionary explanations for this occurrence (Vekemans 2010). Different evolutionary histories that might explain a shared polymorphism between two taxa (groups) are shown in Fig. 1 (the ancestral variant or sequence is labelled A1 and the derived variant is labelled A2). First, convergence or parallel evolution (Fig. 1a) might occur if the two different taxa each generated a new adaptive A2 variant that arose separately by mutation and increased in frequency.

Figure 1.

A comparison of convergence (a), trans-species polymorphism or incomplete lineage sorting by chance (b) or balancing selection (c) and adaptive introgression (d). Allele A1 is the ancestral variant present except for balancing selection (c) where both alleles A1 and A2 are assumed to be ancestral variants. The crosshatches and subsequent red lines indicate mutation from A1 to A2, and the broken horizontal line indicates introgression from one taxa to another. Note that the timescale of incomplete lineage sorting for balancing selection (c) is much longer than that by chance (b).

Second, an ancestral group might have two (or more) different variants, and these are retained in both descendant taxa, sometimes called trans-species polymorphism, either because of incomplete lineage sorting (Fig. 1b) by chance (ancestral origin of the variant maintained in both species) or because these variants are adaptive and selectively maintained as a polymorphism (Fig. 1c). That is, either the species might have split so recently that even for neutral genes, the fixation process has not had time for completion or because balancing selection maintained these allele lineages beyond the species split. The phylogenetic pattern for incomplete lineage sorting in these two cases is similar but when with balancing selection, there is a greatly extended timescale compared with neutrality (Takahata 1990). Balancing selection for multiple variants for polymorphic genes, such as major histocompatibility complex (MHC) loci in vertebrates, self-incompatibility (S) loci in plants and complementary sex determination (csd) loci in hymenoptera, is thought to result in long-term trans-species polymorphism.

Finally, a favourable mutant might have occurred in one (donor) group and then was introduced into the second (recipient) group by introgression (Fig. 1d) so that both contemporary groups have the same favourable variant. Introgression implies that the variant has originated from another group. This is only true for this last option because for convergence, the variants arose independently in the two groups and for trans-species polymorphism, due to either incomplete lineage sorting or balancing selection, the adaptive variants are in different groups because they were there from the ancestral group, not due to introgression.

Because it is not generally possible to observe such changes over time, it is often difficult to differentiate between these explanations but Table 2 compares the possible differences in the evolutionary history for convergence, trans-species polymorphism from either incomplete lineage sorting or balancing selection and adaptive introgression. One potential approach to distinguish them would be to examine ‘ancient’ samples from earlier generations if they are available. In this case, the adaptive variant would be expected to be present in both groups if it has resulted from convergence before the time of the sample or trans-species polymorphism but would be missing from the ancient sample of the recipient group if it originated recently from adaptive introgression.

Table 2. Comparison of the explanations that might result in the same variant being present in two different related taxa
 ConvergenceTrans-species polymorphismAdaptive introgression
Incomplete lineage sortingBalancing selection
(a) Find adaptive variant in ‘ancient’ samples from both taxaYesYesYesNo
(b) Linked sequence indicative of origin in other speciesNoNoNoYes
(c) Linked sequences same or similarNoYesNoNo
(d) Variant adaptiveYesNoYesYes

If the sequence closely linked to the adaptive variant is indicative of another species, then this should indicate that the variant is the result of adaptive introgression. The linked sequence (haplotype) is expected to be more similar for trans-species polymorphism from incomplete lineage sorting than for the other explanations, mainly because of more recent common ancestry. Finally, the variant should be adaptive for all the options except incomplete lineage sorting for neutral loci. With these general comparisons, it appears possible to distinguish between these explanations with the exception of convergence and trans-species polymorphism due to balancing selection. In this comparison, in general, the divergence time of the trans-species polymorphism variants might be significantly greater than that of the convergent variants.

Examples of adaptive introgression

What are the important indicators, or fundamental evidence, necessary to document adaptive introgression (Rieseberg 2011)? First, the identification of the donor species, genes (sequences) and traits in this species is basic. Second, the presence of these genes and/or traits introgressed from the donor species into the recipient species is documented. Third, the adaptive significance or fitness effects of the variants in the recipient species should be identified with potentially an understanding of the basis of these fitness differences. For example, the fitness effect of a variant might differ between the donor and recipient species. Finally, the reconstruction of the molecular evolutionary history of the genes (sequences) in the donor and recipient species and of linked sequences is important. Of course, providing all this information in a given case can be quite difficult so that inference from other data is often used. For example, a high frequency of a variant from another species in the recipient species has been suggested, without any direct fitness estimation or information, as a potential indicator of its adaptive advantage. More specifically, differential introgression among loci may identify genes that are adaptively introgressed (Beaumont & Balding 2004; Gosset & Bierne 2012), a topic that has been extensively considered in the hybrid zone literature (Payseur et al. 2004; Gompert & Buerkle 2009; Nolte et al. 2009; Payseur 2010).

Balancing selection for multiple alleles: MHC, S and csd

When there is balancing selection for multiple alleles and subdivision of the population, substantial introgression between groups is predicted because of the strong selective advantage of rare alleles, compared with neutral loci (Schierup et al. 2000). However, loci with balancing selection for multiple alleles also retain ancestral variation much longer than loci with neutral variation (Takahata 1990). In other words, distinguishing between the introgression from related taxa and retention of ancestral variation is generally difficult for loci under balancing selection.

There appear to be relative few genes where variation is maintained by heterozygote advantage selection at multiple alleles (Hedrick 2012; and balancing selection in general), with the primary examples being variation at MHC genes in vertebrates, S genes in plants and csd genes in Hymenoptera. Variants at both S and csd genes have intrinsic properties, mating system and sex determination, respectively, that would have the same function in both related recipient species and the donor species with no new specific adaptation to a new environment necessary. Similarly, MHC variants, which provide resistance to infectious diseases, might meet the same selective challenges in related species, given that related species are challenged with the same, or similar, infectious diseases.

Castric et al. (2008) examined very similar pairs of S alleles in Arabidopsis halleri and A. lyrata, species that diverged about 2 million years ago. They concluded that these alleles were adaptively introgressed between the species and that selection resulted in a fivefold increase in introgression for S alleles compared with the rate for other genes. On the other hand, Klein et al. (1998, 2007) concluded that similar MHC variants in different species indicate retained ancestral polymorphism from balancing selection, rather than convergent evolution. Vilà et al. (2005) suggested that introgression from wild ancestors was important in determining the extent of MHC variation in domestic animals, and, as discussed below, Abi-Rached et al. (2011) concluded that adaptive introgression from archaic hominins resulted in a MHC allele in modern humans.

Overall, the potential for adaptive introgression for loci under balancing selection for multiple alleles is substantial. Even though this appears to be a limited category of genes, the potential for important adaptive evolution is present. Obviously, the transfer of increased resistance to infectious diseases from adapted MHC variants might be critical. For example, resistance to the chytrid fungus from MHC variants in a leopard frog species (Savage & Zamudio 2011) might provide similar resistance in other related frog species. Or, the coyote (Canis latrans) has much more MHC variation than the endangered red wolf (C. rufus; Hedrick et al. 2002) with which it hybridizes and could provide the red wolf with disease resistance variants.

Pesticide resistance in mice

Mice and rats developed resistance to the anticoagulant rodenticide warfarin within a decade of its use that began in the 1950s, and this resistance is widespread in contemporary populations (Rost et al. 2009; Pelz et al. 2012). Song et al. (2011) demonstrated that the origin of some of the resistant variants in the western European house mouse (Mus musculus domesticus) is from the Algerian mouse (Mus spretus). These two species are thought to have separated between 1.5 and 3 million years ago, were allopatric over most of their range before the introduction of agriculture and demonstrate substantial reproductive isolation, and there is no evidence of their past hybridization although there are areas of recent contact in parts of Africa and Europe.

Song et al. (2011) identified a 20-megabase segment in M. m. domesticus containing the warfarin resistance gene vkorc1 (vitamin K epoxide reductase subcomponent 1) that was introgressed from M. spretus. Most of these variants differ from M. m. domesticus alleles by four nonsynonymous substitutions in the coding region of vkorc1, indicating fast evolutionary change under positive selection. In Spain, 27 of 29 M. m. domesticus sampled had M. spretus vkorc1 variants, and even in Germany, far from any hybrid zones, 16 of 50 M. m. domesticus mice had M. spretus vkorc1 variants. Modelling selective sweeps at vkorc1, Song et al. (2011) estimated that the increase in the M. spretus variant was quite recent.

In addition, they examined the fitness of M. m. domesticus from Germany, which were homozygous for a M. m. domesticus variant (vkorc1dom/vkorc1dom) or a M. spretus variant vkorc1spr/vkorc1spr when given no-choice diets with three different anticoagulant rodenticides (Table 3). The survival of M. m. domesticus mice with the M. spretus allele was much higher for all three rodenticides, and the average relative fitness of vkorc1dom/vkorc1dom mice based on these survival values was only 12% that of vkorc1spr/vkorc1spr mice, demonstrating the strong selective advantage of the introgressed M. spretus variant in M. m. domesticus.

Table 3. The proportion of Mus musculus mice from northern Germany surviving (S) when they are homozygous for a dom (domesticus) or a spr (spretus) variant at the vkorc1 gene and are given three different rodenticides (Song et al. 2011). These survival values are standardized to give the relative fitnesses (w) of the two genotypes
GenotypeRodenticide 1Rodenticide 2Rodenticide 3Mean
S w S w S w S w
vkorc1dom/vkorc1dom0.080.100.150.160.000.000.080.12
vkorc1spr/vkorc1spr0.8010.9110.2010.641

Interestingly, M. spretus is highly resistant to rodenticides, even though they have not been used extensively on this species. Song et al. (2011) suggested that this rodenticide resistance is a pleiotropic effect of an adaptation for living on a granivorous diet that is deficient in vitamin K. Also, even though in hybrids between M. m. domesticus and M. spretus all males and some females are sterile, the dominance of resistance alleles (Pelz et al. 2005) appears to provide a fitness advantage to F1 mice where rodenticides are used.

The pattern of different vkorc1 variants increasing in different areas (Pelz et al. 2005) suggested that adaptation at this locus utilized new mutations. However, the high speed of adaptation suggested that the different variants were already present in low frequency and that adaption was from standing variation (Hedrick 2006). Now with the identification of variants in a number of areas from M. spretus, it appears that adaptive introgression is important in this situation and only with further detailed molecular examination will the relative contributions of mutation, standing variation and adaptive introgression be understood.

Black colour in wolves and white colour in coyotes

Related species of canids often interbreed, and in particular dogs, wolves and coyotes are generally not reproductively isolated, providing an opportunity for adaptive introgression. For example, Anderson et al. (2009) examined the molecular basis of black colour in wolves (Canis lupus), which is found in high frequency in some North American populations. They found that the dominant allele KB at the beta-defensin protein locus (known as K or CBD103) that resulted in black colour in wolves was the same allele as found in many breeds of dogs. As a result, they suggested that the KB allele was introgressed from dogs into wolves and that the dark coloration was adaptive in forested areas. They consequently concluded that ‘our results imply that variants that appear under domestication can be viable in the wild and enrich the genetic legacy of natural populations’. Subsequently, Coulson et al. (2011) estimated in the Yellowstone wolf population, about half of which are black wolves, that there was a heterozygote advantage at this locus. They estimated that the relative mean lifetime reproductive success of grey wild-type homozygotes (kyky), black heterozygotes (KBky) and black homozygotes (KBKB) was 0.779, 1.0 and 0.013, respectively.

Although this example provides both detailed molecular data from the putative donor and recipient taxa and estimates of fitness in the recipient wolf population, there are still some important unanswered questions. First, although Anderson et al. (2009) suggested that there is a correlation between wolf colour and habitat, there are a number of counter-examples with wolves of light colour in forest populations and black wolves on tundra (Mech et al. 1998). Second, the largest selective difference identified by Coulson et al. (2011) was between black heterozygotes and black homozygotes, which have apparently indistinguishable colour. In other words, this large difference does not appear to be associated with the hypothesis of colour and habitat matching but with ‘some other function of the gene, perhaps via its role in cellular immunity’ (Coulson et al. 2011). Third, there is no evidence of the near lethality of the black homozygotes (1.3% the fitness of black heterozygotes) estimated in the Yellowstone populations in the many dog breeds that are fixed for this variant.

Fourth, the case for the dog origin of the KB allele would be greatly strengthened if there were some evidence from ancient DNA that the variant was in North American dogs thousands of years ago or if there were historical data indicating dark coloured dogs among early Amerindians. Finally, the age of the KB allele was estimated to be 47 000 years ago (95% confidence limits 13 000–121 000), which overlaps with the traditional estimated date of dog domestication (15 000–40 000 bp; Vilà et al. 1997; Savolainen et al. 2002). However, this does not exclude a shared ancestral polymorphism for these variants in wolves and dogs, and there is evidence of multiple domestications of dogs, some of which could be more recent (Axelsson et al. 2013).

Since the 1980s, coyotes (Canis latrans) have expanded their range into eastern Canada, and since 2003, white coyotes (six of <6000 animals) have been observed in Newfoundland and Labrador. Brockerville et al. (2013) found that a recessive allele at the Mc1r locus was responsible for the white colour and that this allele is the same as found in some dog breeds, such as the golden and Labrador retrievers. In fact, Brockerville et al. (2013) reported that in 2001, a male golden retriever and a female coyote paired and probably produced hybrid progeny. Although there appears to be evidence of recent introgression of this dog allele into coyotes, the frequency of the variant is still low, and there is no evidence yet of the adaptive significance of the allele in coyotes.

Neanderthal or Denisovan ancestry in humans

One of the most provocative findings from human molecular genomics is that 1–4% of the modern Eurasian genome originated from Neanderthals (Green et al. 2010) and that 4–6% of the modern Melanesian genome originated from Denisovans (Reich et al. 2010; for a review of paleopopulation genetics, see Wall & Slatkin 2012). Given this level of hybridization, then some variants from Neanderthals and Denisovans might be expected to have substantial frequencies in humans by chance history although there are several variants from these hominins that have been suggested as present in modern humans because of adaptive introgression. In contrast, Eriksson & Manica (2012) suggested that ancient population structure with no hybridization might provide a similar signal to hybridization.

Abi-Rached et al. (2011) suggested that a substantial proportion of human leukocyte antigen (HLA) (human MHC) variation in modern humans is descended from introgression from archaic hominins. In particular, the variant B*73, which is similar to HLA variants in chimpanzees, is found in high frequency in modern humans in western Asia but not elsewhere. Examining HLA in a Denisovan, Abi-Rached et al. (2011) found that she had a HLA-C*15 allele that is generally found in linkage disequilibrium with B*73. As a result, they concluded that humans in western Asia inherited this variant from the Denisovans, and it subsequently increased in frequency because of a selective advantage and is an example of adaptive introgression. However, the B*73 allele was not actually found in the Denisovan, and, as we discussed above, HLA variation shows trans-specific polymorphism due to balancing selection.

Mendez et al. (2012a,b) have identified two other genes, STAT2 and OAS1, both part of the immune system but unlinked to HLA, that they suggested as candidates for introgression into modern humans from archaic hominins. The STAT2 haplotype N shares a recent common ancestry with a Neanderthal sequence and is found in high frequency in modern Melanesian populations. This frequency pattern does not appear consistent with neutrality, and they suggested that the STAT2 haplotype N was adaptively introgressed into modern humans. The OAS1 gene has a divergent haplotype in modern Melanesians that is closely related to a Denisovan sequence but the haplotypes are consistent with neutral introgression (Mendez et al. 2013). Although both the HLA-B*73 and STAT2 haplotype N high frequencies are consistent with adaptive introgression, a fitness advantage has not been directly demonstrated for either variant.

Mimicry genes in Heliconius butterflies

The neotropical butterfly genus Heliconius has been long studied in ecology, behaviour and speciation because of its diverse wing colour patterns and chemical defences (Turner 1981; Jiggins et al. 2001). Sympatric Heliconius species often have similar warning wing colour patterns as predator protection (they are unpalatable to vertebrate predators) and they serve as Müllerian mimics to each other. Their similarity in appearance enables different Heliconius species to share the cost of educating predators of their unpalatability. In addition, closely related species often have quite divergent wing colour patterns in different areas but this is again consistent with similarity to other sympatric Müllerian mimics. It appears that the rapid radiation of Heliconius species is closely tied to these mimetic wing colour patterns.

Several loci have been identified that control wing colour pattern in Heliconius and recently molecular examination has identified the primary genomic regions involved (Reed et al. 2011; Nadeau et al. 2012). The Heliconius Genome Consortium (2012) concluded that three comimic Heliconius species have exchanged genes in two genomic regions that control wing colour pattern essential for this mimicry, and Pardo-Diaz et al. (2012) examined the direction of introgression for one of these regions and found it inconsistent with neutral expectations. In other words, there appears to be strong evidence of adaptive introgression of protective warning wing colour variation between closely related Heliconius species. Brower (2013) suggested that mimetic patterns in Heliconius have evolved without apparent hybridization although his claims appear inconsistent with the bulk of research indicating introgression (for a detailed discussion, see Eratosignis 2013a,b).

Beak traits in Darwin's finches

Long-term research into Darwin's finches provides one of best studied examples of the evolution and ecology of a closely related group of species (Grant & Grant 1989, 2008; Grant 1999). In particular, Grant & Grant (2008) documented bidirectional introgression in beak morphology between the Darwin's ground finches, Geospiza fortis and G. scandens, on the small island Daphne Major in the Galápagos archipelago. They showed that hybridization between the species occurred at a low rate (<2%) over years, resulted in increased genetic and phenotypic variation in beak morphology, and have generally increased fitness by improving feeding.

Grant and Grant (2010) estimated that the impact of hybridization between these two species on Daphne Major on the amount of genetic variation was larger than immigration from conspecific individuals from Santa Cruz Island, the nearest source of conspecifics of the two species. In fact, this significant level of hybridization might explain the relative genetic homogeneity (Freeland & Boag 1999) observed over the phenotypically diverse Darwin's ground finches.

Identification of the genes that determine beak morphology differences and other phenotypic divergence in Darwin's finches might potentially make discerning the relationships and origins of these loci in different species possible. However, if the effects resulting from particular genes are small and the history of exchange between species has been frequent and bidirectional, then understanding these sources might be difficult.

Yellow skin in chickens

Darwin (1868) concluded that domestic chickens were descended from red junglefowl (Gallus gallus), and molecular studies have supported his view. However, yellow skin is a common phenotype in domestic chickens, and the regulatory, recessive variant causing yellow skin at BCDO2, which encodes beta-carotene dioxygenase 2, is not found in red junglefowl. Eriksson et al. (2008) provided strong support that the yellow skin variant comes from another species, the grey junglefowl (Gallus sonneratii).

These two species do not hybridize in the wild (Johnsgard 1999), and a cross in captivity between grey junglefowl cocks and red junglefowl hens produced mostly sterile progeny (Morejohn 1968). However, hybridization between grey junglefowl and domestic chickens has been reported (Johnsgard 1999), suggesting that introgression of the yellow skin region occurred after domestication. It appears that introgression of the yellow skin phenotype was favoured by human selection for colour because Eriksson et al. (2008) were not able to identify a quantitative trait association with BCDO2.

Non-native ancestry in an endangered native salamander

The California tiger salamander (Ambystoma tigrinum) is threatened, mainly because of habitat loss and fragmentation. Now hybridization with the non-native barred tiger salamander (A. t. mavortium), introduced in the 1940s and 1950s from Texas because their large aquatic larvae are used as fish bait, is an additional threat. The two taxa were previously geographically isolated and are thought to have diverged more than 3 million years ago (Fitzpatrick et al. 2010). Some hybrids between the two taxa have high fitness both in the wild and in laboratory experiments (Fitzpatrick & Shaffer 2007).

Fitzpatrick et al. (2010) found that populations near the introduction area were variable for both native and introduced variants at 68 SNP loci. Most markers (65 of 68) became variable only for the native variants within about 25 km of the introduction area but the other three, unlinked SNP loci showed very similar patterns to each other and had non-native variants in high frequency up to 100 km away from the introduction area. Using computer simulation, Fitzpatrick et al. (2010) concluded that natural selection favoured the non-native variants at these three loci. To further understand this situation, it would be useful to know the genes involved in the putative fitness advantage of the non-native alleles and the actual type of fitness advantage they confer in the native population are important. It would also be valuable to determine the basis of the concordance of the distribution of these three favourable variants and to know whether it is what is expected given the time period or whether there is some barrier to further expansion.

Conclusions

Introgression of adaptive variants from other species is widely recognized as significant in plants but appears to be underappreciated in animals. The present extent and quality of evidence for adaptive introgression in animals varies widely in the different examples discussed above. In some situations, such as pesticide resistance in mice and mimicry genes in Heliconius butterflies, the evidence appears solid and compelling, while in others such as black colour in wolves and HLA Denisovan ancestry in humans, it appears as a possible interpretation of the data and more support for adaptive introgression is needed. As more systems are examined in detail, only then will the relative importance of mutation, standing variation and adaptive introgression for the origin of adaptive variation become clear. Without detailed molecular examination, adaptive introgression might be unrecognized and adaptive responses might be attributed to either mutation or standing variation.

A general genomic approach that promises to detect potential adaptive introgressive variants has been applied to the house mouse (Mus musculus; Staubach et al. 2012). Using high-density SNPs, genomic regions that both demonstrated positive selection and, using two populations from each of two divergent subspecies, M. m. domesticus and M. m. musculus, introgression, were identified. This approach detected at least nine genomic regions that appeared to have experienced positive selection, about 10% of the genome that appeared to have been introgressed, and at least two genomic regions with both positive selection and introgression, both of which included candidate genes. Interestingly, the region in M. m. domesticus involved in anticoagulant resistance (Song et al. 2011) introgressed from M. spretus was not apparently identified, perhaps because both samples of M. m domesticus had high frequencies of the introgressed region.

The origin of adaptive variation from introgression potentially has some of the advantages of standing variation and other benefits. First, adaptive introgressive variants might be complex variants with multiple changes within a gene (or at multiple linked genes) without negative pleiotropic effects because they have already been tried out adaptively in another species. Second, if there is a new adaptive challenge for the species, then there may not be any standing variation available for adaptation but another taxon that has already adapted to this environment might have adaptive genetic variation that could be used via introgression. Third, adaptive introgression might be particularly effective for regaining traits that have been lost over time in the recipient species and that have already functioned in a common ancestor of the recipient and donor species (Rieseberg 2009). Fourth, a variant or trait present in a domestic animal might be introgressed into the wild species from which it was derived or vice versa. This could provide resistance to a disease lost in one of the species, for example, a wild, endangered species might gain resistance from its domestic relative although introgression of domestic animal ancestry into wild relatives is generally more of a problem than a positive phenomenon (Randi 2008).

For clarification, genetic rescue, which results in an increase in fitness due to a decrease in frequency of detrimental alleles (Tallmon et al. 2004; Hedrick & Fredrickson 2010), is a somewhat different phenomenon than adaptive introgression. Genetic rescue occurs in a population with low fitness because of past inbreeding or genetic drift has resulted in low fitness (Johnson et al. 2010), while adaptive introgression occurs in a population that has a new environmental challenge. Although in theory, these two scenarios can be differentiated, but in practice, it may be difficult to tell whether low fitness is due to inbreeding depression or because the population is not adapted to a new environment. For example, the genomic sweep documented by Adams et al. (2011) in wolves and the rapid increase in migrant ancestry in inbred butterflies (Saccheri & Brakefield 2002) are thought to be the result of low fitness due to inbreeding or genetic drift but increase in adaptation might also be of significance.

Acknowledgements

I appreciate the support of the Ullman Distinguished Professor for this research, Xavier Vekemans and Tom Dowling for drawing Fig. 1 and comments by Peter Grant, Rosemary Grant, Mike Hammer, Michael Kohn, Jim Mallet, Loren Rieseberg, Dolph Schluter, Xavier Vekemans and two anonymous reviewers on the topic or an earlier version of the manuscript.

P.H. is a population and conservation geneticist.

Ancillary