Dynamics of Adaptive Introgression from Archaic to Modern Humans

Recent evidence from the genomic variation of living people documents genetic contributions from archaic to later modern humans. This evidence of introgression contrasts with earlier findings from single loci that appeared to exclude archaic human genetic survival. The present evidence indicates that many “archaic” alleles may represent relicts of African archaics, and that some “archaic” variants both inside and outside of Africa have attained relatively high frequencies. Both observations may be surprising under the hypothesis that modern humans originated first in Africa and displaced archaic populations through expansion and drift. Here, we outline how natural selection may have enabled the uptake of introgressive alleles from archaic humans. Even if admixture or gene flow were minimal, the introgression of selected variants would have been highly probable. In contrast to neutral alleles, adaptive alleles may attain high frequencies after minimal genetic introgression. Adaptive introgression can therefore explain why some loci show evidence for some archaic human contribution even as others apparently exclude it. The dynamics of introgression also may explain the distribution of certain deep haplotype branches in Africa. Open questions remain, including the likelihood that archaic alleles retained their adaptive value on the genetic background of modern humans and the scope of functions influenced by adaptive introgression. How important is introgressive hybridization? I do not know. One point seems fairly certain: its importance is paradoxical. The more imperceptible introgression becomes, the greater is its biological significance. It may be of the greatest importance when by our present crude methods we can do no more than to demonstrate its existence. . . . Only by the exact comparisons of populations can we demonstrate the phenomenon, yet in such populations the raw material for evolution brought in by introgression must greatly exceed the new genes produced directly by mutation. The wider spread of a few genes (if it exists) might well be imperceptible even from a study of population averages, but it would be of tremendous biological import (Anderson 1949: 102). We inevitably reach the conclusion, therefore, that introgressive genotypes not only persist indefinitely, but that also, like polyploids, they can migrate far beyond the areas in which they originated, and can actually survive after the non-introgressed parental species has become extinct (Stebbins 1959: 241, emphasis added) InTRoduCTIon T anatomical and behavioral configuration of today’s humans emerged during the Late Pleistocene. Before this time, regional populations of “archaic” humans inhabited the core regions of Africa, Europe, and Asia. By 25,000 years ago, no archaic humans remained, and modern humans occupied Europe, Africa, Asia, and Australia. The evolutionary transformation to modernity began in Africa (Bräuer 1984; Stringer & Andrews 1988; Trinkaus 2005), but the early evolution of modern human anatomy and behavior involved subsequent changes both inside and outside Africa (Smith 1992; Klein 1995; D’Errico 2003). Specialists PaleoAnthropology 2006: 101−115. Copyright © 2006 PaleoAnthropology Society. All rights reserved. JoHn HAwKS Department of Anthropology, University of Wisconsin–Madison, 1180 Observatory Drive, 5240 Social Science Building, Madison, WI 53706, USA GrEGory CoCHrAn Department of Anthropology, University of Utah, 6708 Loftus, N.E., Albuquerque, NM 87109, USA disagree whether archaic and modern humans should be considered multiple species, subspecies, or evolving varieties of a single metapopulation. Regardless of their taxonomic rank, similarities in behavioral capacities and recurrent spatiotemporal contacts make it likely that there was at least some interbreeding between archaic and modern populations (wolpoff et al. 1984; Smith et al. 1989; Trinkaus 2005). The ultimate success of the modern human lineage was presumably a result of behavioral innovations in language, culture, or technology, all of which are implicated in the Late Pleistocene archaeological record with material evidence of symbolic culture (Chase 1999; Klein and Edgar 2002; Mellars 1989, 2005). But some late archaic humans, such as the European neandertals, left a record of comparable behavioral capabilities to early modern humans (D’Errico 2003; Zilhão 2006). Moreover, modern humans were in West Asia by 100,000 years ago (Stringer and Andrews 1988), in Australia by 50,000 years ago (Turney and Bird 2001), and in Europe by 36,000 years ago (Trinkaus et al. 2003), making it plausible that comparably-skilled archaic and modern humans were in contact for substantial time periods. These populations shared common ancestors during the Early to Middle Pleistocene—for example, the European neandertals appear to have shared genetic ancestry with modern humans between 300,000 and 700,000 years ago (Krings et al. 1999; Green et al. 2006). This time interval is very short for reproductive isolation to have evolved—for instance, no primates are known to have established postzygotic reproductive isolation during so short a time (Curnoe et al. 2006), and most mammalian sister taxa retain the ability to interbreed far longer (Holliday in press). 102 • PaleoAnthropology 2006 Anatomical evidence from early modern humans also suggests intermixture between archaic and modern populations. At the peripheries of human occupation in Europe and Southeast Asia, early modern humans continued to exhibit some traits that had been common in preceding archaic populations (wolpoff et al. 2001, 1984; Duarte et al. 1999; Trinkaus 2005). For the most part these traits were initially present at only low frequencies, which declined over time toward the present, at least in Europe (Frayer 1993, 1998). From this decline in frequencies, it seems probable that the alleles underlying archaic human morphological patterns were not adaptive in the modern human population. Likewise, extensive sampling of modern humans for mitochondrial and Y chromosome haplotypes has shown no evidence of ancient lineages such as might have existed within archaic human populations (Serre et al. 2004; Currat and Excoffier 2004; weaver and roseman 2005). These data may be consistent with a hypothesis of some gene flow from archaic to modern populations (Smith et al. 1989; Trinkaus 2005), but the amount of such gene flow was evidently slight. Recent genetic reports have demonstrated that living people retain alleles from multiple archaic populations (Garrigan et al. 2005a, b; Hardy et al. 2005; Plagnol and wall 2006; Hayakawa et al. 2006; Evans et al. 2006). Plagnol and wall (2006) found that the pattern of linkage disequilibrium among SnPs in the human genome was inconsistent with an unstructured ancient population, and estimated that five percent of genetic variation in Europe and in West Africa originated in archaic humans such as the neandertals. Two facts about this possible admixture are surprising from a paleoanthropological perspective. First, evidence for archaic ancestry is nearly as strong in Africa as in Europe, also confirmed by at least one single-locus study (Garrigan et al. 2005b). Second, at least some of the apparent archaic variants have been found at high frequencies in living populations (Garrigan et al. 2005a, b; Hardy et al. 2005). These observations seem inconsistent with the hypothesis that an initially low level of genetic contribution from archaic humans declined over time. In particular, they conflict with evidence that previously suggested near-total genetic replacement of archaic humans (Serre et al. 2004; Currat and Excoffier 2004; Vigilant et al. 1991; Takahata et al. 2001). Preliminary reports suggest that the neandertal genome also included an excess of human-derived single nucleotide polymorphisms (Green et al. 2006). natural selection on introgressive variants from archaic humans can explain these data. Although the morphological pattern of archaic humans has disappeared, their long existence may have led to the persistence or appearance of alleles that did not occur in early modern humans. Some of these alleles may have been globally adaptive even outside the archaic human populations in which they originated. others may have generated purely local advantages. In both instances, adaptive introgression is the most credible way for alleles from declining populations of archaic humans to survive and reach high frequencies today. ADAPTIVE INTROGRESSION Genetic introgression, or “introgressive hybridization,” is classically taken as the introduction of alleles from one species into another species through hybridization (Anderson and Hubricht 1938). However, because species and subspecies boundaries are often imprecisely known, or fuzzy in nature, naturalists often adopt a more permissive definition that encompasses gene flow between subspecies, races, or varieties in addition to species (Rieseberg and Wendel 1993). An ecological theory of introgression emerged during the 1930’s and 1940’s (Anderson and Hubricht 1938; Anderson 1949; Heiser 1949), centered around the observation of introgressive hybridization in sunflowers, iris, and domesticated crops. Introgression has been demonstrated by several different methods for different species, including morphological traits, molecular markers, cytogenetic characters, and karyotypes (Jarvis and Hodgkin 1999). Traditionally, hybridization and introgression have been considered unimportant in the evolution of animal species (Mallet 2005). The lack of interest in introgression mainly stems from the observation that interspecific hybrids often display reduced fitness or sterility (Mayr 1963), an observation that can be extended to plants as well (Mayr 1992). At first glance, if F1 hybrids fail to thrive then genetic exchanges appear questionable. But even though reduced hybrid fitness may tend to limit gene flow between populations, it does not prevent relatively high levels of adaptive introgression (Arnold 1997; Arnold et al. 1999). This is because any allele introduced recurrently into a population will succeed or fail based on the strength of selection upon it. This insight and molecular assays of multiple genes have caused a resurgence of interest in hybridization and introgression in mammals. For example, a survey of 13 X-linked loci found evidence for adaptive introgression across a hybrid zone between Mus domesticus and Mus musculus (Payseur et al. 2004). Introgressive hybridization often increases between populations when ecological conditions change or are disturbed. In contemporary organisms, such change often results from human disturbance or deliberate introductions (rhymer and Simberloff 1996). Some of the best known instances involve mallards and endemic ducks (Mank et al. 2004; rhymer and Simberloff 1996), red and sika deer (Goodman et al. 1999), mule and whitetail deer (Cathey et al. 1998), dogs and coyotes (Adams et al. 2003), coyotes and grey wolves (Lehman et al. 1991), tilapia (Gregg et al. 1998), and brown trout (Marzano et al. 2003; Almodóvar et al. 2001). In extreme cases, hybridization and introgression can result in the merger of formerly separate species or the formation of new species (rhymer and Simberloff 1996; Dowling and Secor 1997). The effects of historic introgression are also sometimes seen in species with no evidence of current hybridization; this may be a consequence of past ecological changes, changes in species ranges, or expansion from glacial refugia. Examples include coyotes (Lehman et al. 1991), willow (Hardig et al. 2000), water flea (Taylor et al. 2005), lake trout (wilson and Bernatchez 1998), European newts (Babik et al. 2005), and Japanese land snails Dynamics of Adaptive Introgression• 103 (Shimuzu and Ueshima 2000). Because hybridizing species share a large proportion of their genetic background, a new allele that is adaptive in one species may retain its selective advantage after introgressing into another (Anderson 1949; Lewontin and Birch 1966; Arnold 2004b). Such adaptive introgression has emerged as an important mechanism for the introduction of adaptive variation (Arnold 2004b; Rieseberg et al. 2004). Many domesticated species originated through hybridization of wild populations; others show evidence of substantial adaptive introgression from wild populations after their origins (Jarvis and Hodgkin 1999; Bruford et al. 2003), a topic discussed further below. Evidence for the introgression of adaptive alleles in wild populations was once rare, but has increased in recent years because molecular techniques allow easier tests of selection. Long-distance, or “dispersed” introgression involving discrete portions of the genome is a sign that positive selection favors an introgressive allele. Selection may be confirmed by field studies that show that an introgressive allele has observable effects on survival or reproduction. For example, introgression in natural populations of plants has often been noted to spread biotic resistance traits, as in sunflowers (whitney et al. 2006) and lodgepole and jack pines (Wu et al. 1996). Introgression in Louisiana iris has introduced shade tolerance in local populations (Arnold 2004b), while xeric tolerance apparently spread from Utah cliffrose to bitterbrush (Stutz and Thomas 1964). Examples of adaptive introgression in animals include damselflies (sex-specific color morphs) (Sánchez-Guillén et al. 2005), Anopheles mosquitoes (pyrethroid resistance) (Weill et al. 2000), Lutzomyia (mating song) (Bauzer et al.2002), cichlid fishes (adaptive radiation under influence of hybridization) (Streelman et al. 2004), mountain and European hares (mtDnA related to climate) (Melo-Ferreira et al. 2005; Thulin et al. 2006), lake trout (mtDnA from arctic charr) (wilson and Bernatchez 1998), and trypanosomes (drug tolerance) (Machado and Ayala 2001). It is interesting that several examples of adaptive introgression involve mitochondrial genomes, although such cases are probably highly represented because of the widespread use of mtDnA as a population marker. In many other instances, the function of an introgressive allele may not be known, but selection can be inferred from its present molecular variation. In his monograph on introgressive hybridization, Anderson (1949) concluded with two observations. First, for species in contact with close relatives, introgression might be a greater source of new adaptive variation than new mutations. And second, phenotypically imperceptible introgression may be a more important source of adaptive variation than distinct hybrid zones. Anderson did not formulate these conclusions in terms of population genetics, but a consideration of the relevant theory confirms their general validity. First, the importance of introgression relative to mutation emerges from the high chance of fixation of introgressive variants. The probability of fixation of an adaptive dominant allele introduced as a single copy is 2s, where s is the selection coefficient applying to homozygotes (Haldane 1927). This probability applies to any single copy of an adaptive allele, whether it is introduced by mutation or hybridization. But a new adaptive mutation generally occurs initially as a single copy, unless the mutation rate is very high. In contrast, along even a very thin hybrid zone many interbreeding events between two populations will occur. Each of these hybrids may carry adaptive alleles from both populations, and each backcross into a source population provides the opportunity for each of these adaptive alleles to spread to fixation with probability 2s. At this likelihood, it takes relatively few such hybrids to ensure the ultimate fixation of such an introgressive allele, as calculated below. Because of the high chance of fixation with recurrent interbreeding, hybridizing species should share a large number of adaptive alleles—probably most of those that retain their selective advantage on the cross-species genetic background. A set of such hybridizing species will effectively pool the adaptive potential of any single one of them, providing a larger source of adaptive variants than mutation alone. Moreover, an adaptive allele from another species may differ by several mutations from the allele it replaces, and some of the intermediate steps may not have been adaptive by themselves in the host species. In effect, introgression may allow species to cross an adaptive valley without the fitness cost of intermediate alleles. In this way, introgression can permit adaptations that might never occur by new mutation in a single population. Anderson’s (1949) second concluding observation was that the fitness importance of morphologically distinct hybrid swarms is likely to be relatively limited. This result may also be derived from population genetic principles. recognizable parental morphotypes depend on many coadapted genes. Even if some mixture of these genes were advantageous, selection upon such coadapted phenotypes is far less effective than on single alleles (Eswaran 2002). A single introgressive allele may have a more limited phenotypic effect, but selection will be much more effective. For animal species, many morphological characters carry the additional burden of being involved in mate recognition— some may even impede interbreeding at the hybrid zone through reinforcement (Howard 1993). The adaptive value of introgressive alleles may frequently be cryptic, such as resistance to disease or parasites, changes in metabolic or sensory systems, or alterations in developmental schedules. The genetic structure of phenotypes help to explain Anderson’s “paradoxical” nature of introgression, in which “the more imperceptible introgression becomes, the greater is its biological significance” (Anderson 1949: p. 102). INTROGRESSION fROm ARChAIC humANS Adaptive introgression can explain one of the most important problems in the origin of modern humans. Early modern humans not only retained many of the characters of archaic humans within Africa (Smith 1992), they also retained features of archaic European and Asian populations (Frayer 1993; Frayer et al. 1994; Hawks et al. 2000; wolpoff et al. 104 • PaleoAnthropology 2006 2001; Duarte et al. 1999). Such features provide evidence of intermixture among these populations. But the largest single-locus genetic samples from living people appear to preclude admixture from archaic Europeans or Asians (Manderscheid and rogers 1996; Currat and Excoffier 2004; weaver and roseman 2005; Serre et al. 2004). neither of these sources of evidence suggests that interbreeding was necessarily very common between archaic and modern humans, and the proportion of archaic traits did decline over time where it can be observed (Frayer 1993). yet, some genetic loci show evidence of ancient population structure, such as would be expected from an archaic human ancestry (Templeton 2005; Garrigan et al. 2005b, a; Zietkiewicz et al. 2003; Hayakawa et al. 2006; Hardy et al. 2005; Evans et al. 2006), and the presumed “archaic” alleles are sometimes at high frequencies. Introgression of alleles from archaic populations has been proposed as the most probable explanation for these observations. As a note of skepticism, balancing selection on low-recombination regions or genetic inversions remains possible for some of these loci, but has been excluded in at least the case of MCPH1 (Evans et al. 2006). Here, we investigate the conditions under which introgression is credible, and do not pursue the details of any single gene that might preserve such evidence. The survival of neutral genetic variants from archaic humans is not a likely explanation for high-frequency alleles. The last archaic human populations existed only around 30,000 years ago, a time during which modern humans expanded from an initially low population size to much higher numbers (Stiner et al. 2000; Harpending et al. 1998). It is true that neutral alleles are unlikely to be lost from an expanding population; even alleles introduced at a very low initial frequency should have had a reasonable chance of surviving into the present-day human population (Manderscheid and rogers 1996). But the probability of fixation of any neutral alleles introduced during this time frame is essentially zero, and indeed a neutral allele should remain very near the low frequency at which it was introduced. For most genetic samples available today, in particular the HapMap and other large-scale genomic surveys (The International HapMap Consortium 2005), the low frequency of neutral introgressive alleles should make them invisible to ascertainment. neutral introgression is therefore a poor explanation for many of the observations from modern human genetic variation. However, a denser sampling of human genetic variation in the future may pick up a higher proportion of such neutral introgressive variants, which might be recognizable on the basis of their sequence divergence (Wall 2000). A positively selected allele behaves according to very different rules from a neutral allele. whereas the fixation probability for a single copy of a neutral allele is 1/2n, the corresponding probability for a selected dominant allele is 2s. Moreover, exponential population expansion, which approximates the modern human demographic history, actually increases the probability of fixation by double the intrinsic growth rate (otto and whitlock 1997). A relatively small number of interbreeding events will greatly increase the chance of fixation of adaptive introgressive variants. Haldane (1927) arrived at the fixation probability for a selected allele by generalizing from the probability of extinction of all copies of an adaptive allele in each generation following the introduction; we can use a similar process to consider the probability that several copies of an allele resulting from introgression would be lost without being fixed. This probability of total loss of n copies decreases according to If the modern human population expanded at a rate of 1 percent per generation, then an introgressive allele with s = 0.01 (i.e., a 1 percent fitness advantage) would have a 95 percent probability of fixation in modern humans, with only 74 archaic-modern matings. For an allele with a 5 percent fitness advantage, the corresponding number of events would be only 24. After their introduction into the modern human population, adaptive introgressive variants would have rapidly increased in frequency. Today, such variants may therefore occur at high frequencies, especially in their region of origin. Modern humans had dispersed throughout the world by 30,000 years ago. This means that any alleles that introgressed from archaic human populations must have been introduced by approximately 1200 generations ago. Within a panmictic population, the number of generations to fixation of an additive advantageous allele is given by Crow and Kimura (1970): As the allele approaches fixation, the rate of change in frequency is lower, so that the average time to fixation overrepresents the time during which an allele may remain ascertainable in relatively small samples of a population. For the present version of the HapMap (The International HapMap Consortium 2005), the maximal ascertainment frequency of long-range haplotypes is less than 80 percent (wang et al. 2006; Voight et al. 2006). Different genetic surveys with larger samples may have higher maximal ascertainment frequencies. Here we consider the case in which all variants with frequencies of less than 99 percent will be ascertained. At this level of ascertainment, most variants with s < 0.015 would remain segregating in present-day genetic samples. It is notable that this class of variants not only is weakly selected, it also represents those with the lowest chance of fixation. Alleles with a stronger chance of fixation (i.e., larger selection coefficients) will in general be more likely to have already approached fixation. For instance, a gene with a 4 percent advantage would approach fixation in only around 400 generations, or around 10,000 years. These are illustrative values, and there are some reasons to be conservative about the rate and time to fixation of such introgressive variants. For example, if the initial frequencies were lower (because the global population was larger), then the time to fixation will be higher. A larger ln pt/qt = ln p0/q0 + st / 2 Pr[loss of n copies] = [1 2(s + r)]n Dynamics of Adaptive Introgression• 105 amount of archaic-modern interbreeding would reduce the time to fixation, because it would increase the initial frequency of the introgressive allele. In any case, the alleles most likely to be fixed will come to fixation the fastest, and a substantial number of such archaic alleles may already be near fixation in humans. hOw ImPORTANT wAS INTROGRESSION? The emergence of modern humans was a rapid evolutionary event that involved many genetic changes in a population that became increasingly dispersed over time. A flush of adaptive alleles from archaic human populations, which were already geographically dispersed, may have accelerated this evolutionary change. To the extent of their genetic differentiation, archaic human populations would potentially have had different adaptive genetic variants. But no good estimate of the genetic differentiation of archaic human populations is available, and other genetic parameters, such as their effective population sizes, are also uncertain. Without such knowledge, we cannot make an accurate theoretical estimate of the true importance of adaptive introgression in the emergence of modern humans. However, we can propose some demographic and genetic parameters, and examine the way that variation in such assumptions would affect the expected amount of adaptive introgression. In particular, it is worthwhile to consider whether introgression may have been a larger or smaller source of adaptive variants compared to new mutations in the modern human population. To examine this question, we will assume the following simplified demographic parameters: A population divergence between modern and archaic humans 235,000 years ago. This time is consistent with the divergence of neandertal mtDnA from recent samples, but a “population divergence” itself may overstate the amount of actual genetic differentiation. Equal effective population sizes in the modern and combined archaic populations. Evidence from living populations indicates a larger longterm effective size in Africa than Eurasia (Relethford 1998), but it is not clear whether this value applies to archaic populations, which would have included some populations within Africa. Reproductive contact between modern and archaic populations 35,000 years ago. This assumption greatly simplifies the true pattern of interaction, which was heterogeneous across time and space. An equal rate of adaptive mutations in modern and archaic lineages, μ per genome per year. It is plausible that environmental changes may have altered the number of new adaptive variants (by rendering some previously deleterious changes adaptive), but we have no data whatsoever on this point. Equal fixation probabilities for mutations in both the modern and archaic populations. Again, en1.