The genic view of the process of speciation

Abstract

The unit of adaptation is usually thought to be a gene or set of interacting genes, rather than the whole genome, and this may be true of species differentiation. Defining species on the basis of reproductive isolation (RI), on the other hand, is a concept best applied to the entire genome. The biological species concept (BSC; Mayr, 1963) stresses the isolation aspect of speciation on the basis of two fundamental genetic assumptions – the number of loci underlying species differentiation is large and the whole genome behaves as a cohesive, or coadapted genetic unit. Under these tenets, the exchange of any part of the genomes between diverging groups is thought to destroy their integrity. Hence, the maintenance of each species’ genome cohesiveness by isolating mechanisms has become the central concept of species. In contrast, the Darwinian view of speciation is about differential adaptation to different natural or sexual environments. RI is viewed as an important by product of differential adaptation and complete RI across the whole genome need not be considered as the most central criterion of speciation. The emphasis on natural and sexual selection thus makes the Darwinian view compatible with the modern genic concept of evolution. Genetic and molecular analyses of speciation in the last decade have yielded surprisingly strong support for the neo‐Darwinian view of extensive genetic differentiation and epistasis during speciation. However, the extent falls short of what BSC requires in order to achieve whole‐genome ‘cohesiveness’. Empirical observations suggest that the gene is the unit of species differentiation. Significantly, the genetic architecture underlying RI, the patterns of species hybridization and the molecular signature of speciation genes all appear to support the view that RI is one of the manifestations of differential adaptation, as Darwin (1859, Chap. 8) suggested. The nature of this adaptation may be as much the result of sexual selection as natural selection. In the light of studies since its early days, BSC may now need a major revision by shifting the emphasis from isolation at the level of whole genome to differential adaptation at the genic level. With this revision, BSC would in fact be close to Darwin’s original concept of speciation.

Introduction

What are species? This is undoubtedly one of the big questions in biology. What this perspective intends to do is to formulate the genetic process of differentiation between diverging populations of sexually reproducing organisms. Some, but not all, aspects of differentiation may be manifested as reproductive isolation (RI) (e.g. divergence in the spermatogenic programmes may lead to hybrid male sterility). If that process can be understood, it may then be possible to define the earliest stage in which species can be considered as formed. According to the biological species concept (BSC; Mayr, 1963), such a stage can be objectively and naturally delineated when genome‐wide RI is attained (defining RI on the basis of only a portion of the genome is logically incompatible with the central tenets of BSC; discussed later). This position is based on specific assumptions about the genetic basis of biological differentiation between species. If the assumptions are not fulfilled, species may not be as objectively definable as BSC prescribes. From this view point, our current understanding of the molecular and genetic basis of the process of speciation will be presented. The possible stages in which species may be delineated will be briefly outlined.

There are many operational species definitions for practical use (Coyne, 1994; Harrison, 1998; and other chapters in Howard & Berlocher, 1998; Jiggins & Mallet, 2000). The biological relevance of these definitions would be clearer in the evolutionary context. This perspective is not to compare them, or to add another one, but to address the underlying process of speciation. A detailed treatment of the species concept in the light of this process will be given elsewhere.

It has also been long recognized that the genetics of speciation may be quite distinct between plants and animals (Grant, 1971). Most of the views presented here are based on animal studies. Some may not be applicable to plants (e.g. level of genetic divergence between nascent species; Bradshaw et al., 1995; Doebley et al., 1997) whereas others may be common knowledge among plant evolutionists (Rieseberg et al., 1999; Rieseberg & Burke, 2001). In the animal literature, this perspective relies heavily on Drosophila studies. Although the general patterns of postmating isolation appear comparable across animal taxa (Wu & Davis, 1993; Wu & Palopoli, 1994; Orr, 1997; Presgraves & Orr, 1998), there is insufficient information on other characters. In fact, the sharp contrast in their genetic bases of sexual isolation between two Drosophila systems (Doi et al., 2001; Ting et al., 2001) serves as a sobering reminder on how little we know about the genetics of speciation. Finally, many of the conceptual issues addressed here have been reviewed recently from a different (Coyne & Orr, 1998) or similar (Schilthuizen, 2000) perspective.

Genes as the unit of species differentiation

That genes, or complexes of genes, are the units of evolution has been a widely adopted perspective in Biology. The study of speciation, however, is a conspicuous exception. Because the speciation process is still conceptualized at the level of individual (or whole genome), BSC which defines species as ‘groups of interbreeding natural populations that are reproductively isolated from other such groups’ (Mayr, 1963) remains the gold standard. To see the difference between genes and individuals, consider the simpler case of random‐mating Mendelian population, which is meaningful only at the genic level. For example, human populations may be panmictic at most loci and, indeed, mating is essentially random with respect to the ABO or MN bloodtype and most microsatellites. Nevertheless, mate choice in human is distinctly nonrandom at the individual level. Mating with respect to loci that govern aspects of human morphology, such as height, is likely nonrandom. It may be possible that, at a great majority of loci, mating is random but a very tiny fraction of the genome is more than sufficient to dictate highly nonrandom mating at the individual level.

Like random mating within population, differentiation between populations is more meaningful at the genic than individual level. Imagine two hypothetical populations of grasshoppers on the north and south‐facing slopes of a mountain range, which does not impede migration. Assume that the A, B and C alleles of three different loci are adapted to the northern slope and the a, b, and c alleles, to the southern slope. If local selection is sufficiently strong relative to gene flow, the two populations would differentiate at the three loci. Then, what about the rest of the genomes, which are equally fit in both habitats? A very low level of migration, down to some individuals per generation, is sufficient to prevent population differentiation (Crow & Kimura, 1970; Slatkin, 1987). Therefore, under most parapatric conditions, the genomes of the two populations would be mosaic in their extent of differentiation. Genomic regions very near the three loci of differential adaptation would be differentiated whereas the rest would not.

An important distinction can, therefore, be made between two classes of loci – those that directly affect differential adaptation and those that do not. Differential adaptation is a well defined form of divergence in which the alternative alleles have opposite fitness effects in the two populations (because the migration of these alleles to the other population is maladaptive, differential adaptation also implies restricted gene flow). It is plausible that, during the process of speciation, genes of differential adaptation would account for only a small fraction of the genome. I shall refer to them as ‘speciation genes’ and the rest as ‘marker loci’. As genes directly responsible for differential adaptation and RI are rarely known except in some cases (Lee & Vacquier, 1992; Lee et al., 1995; Metz & Palumbi, 1996; Ting et al., 1998; Wang et al., 1999), most of the genetic studies on species or population differentiation are based on the marker loci, including most allozymes, microsatellite and mitochondria DNA. Although the marker loci are informative about demographic events (Avise, 1994), the whole process of speciation depends primarily on the speciation genes (e.g. Wang et al., 1999; Ting et al., 2000).

Separating the two classes of genes, Fig. 1 outlines the very basic features of the process of speciation at the genic level. For simplicity, we may assume that the populations come into contact at the specified stage (although the specific mode does not really matter). In stage I, population differentiation has taken place at a small number of loci responsible for functional divergence. This could be in their mate recognition system as in the Zimbabwe – cosmopolitan behavioural races of Drosophila (Wu et al., 1995) or the mimicry colour patterns of the Heliconius butterflies (Mallet et al., 1998). Upon secondary contact, gene exchange would be extensive but, at the loci of functional divergence, exchange would be restricted. There is little disagreement that the populations belong to the same species because intermediate forms would be common in the absence of extrinsic barriers (for this stage to be considered as incipient speciation, it is implicitly assumed that there do not exist intermediate environments to which the intermediate forms are equally or better adapted; otherwise, it would be a simple case of population differentiation not germane to speciation). As generally recognized, a species may not necessarily be a homogeneous evolutionary unit (Butlin, 1998). For example, the two behavioural races of Drosophila melanogaster have diverged at more than 15 loci governing mating behaviours (Hollocher et al., 1997b; Ting et al., 2001) with mild fitness reduction in the hybrids (Alipaz, 2000).

In stage II, further differentiation has occurred and genic substitutions within each population could be coadapted in the physiological sense. As a result, intergroup crosses would yield some genotypic combinations that are hybrid inviable or sterile, as is the case of the Bogota vs. mainland populations of D. pseudoobscura (Prakash, 1972; Orr, 1989; Schaeffer & Miller, 1991; Wang et al., 1997). Alternatively, divergence may not have yielded hybrid incompatibility; instead, the ecological or behavioural divergence is so strong that hybrids are not as fit as the parents in their natural environments. The butterflies Heliconius himera × H. erato (Mallet et al., 1998), the Rhagoletis host races (Feder et al., 1997; Berlocher, 1998), the aphid races (Via et al., 2000), Darwin’s finches (Grant & Grant, 1998) and the sticklebacks (Schluter, 1998) may be such examples. It has been common to distinguish the nonadaptive hybrid inviability/sterility from the ecological/behavioural misfit (Rice & Hostert, 1993). This distinction is rather artificial because it presupposes the genic basis of, say, gametogenic divergence (which results in hybrid sterility) is nonadaptive. As discussed later, this presupposition has neither conceptual nor empirical basis. The other assertion that the physiological incompatibility is more genetically ‘hardwired’ than the ecological divergence has not been tested either.

In stage II, a very substantial portion, or even a great majority, of genomic regions would remain shared and undifferentiated. Populations at this stage are sometimes given the subspecies status to reflect the ambiguity in their species status. Will they diverge or fuse? It is possible that, without ecological perturbation, they may continue to diverge to become fully fledged species. However, upon massive hybridization, the subspecies or species may fuse and emerge as a new hybrid form, or hybrid swarm (Mayr, 1963, p. 521). Habitat destruction or land bridge formation, for example, can facilitate such massive hybridization. It would be most interesting to see whether recently evolved species groups such as the African cichlid flocks (Meyer, 1993) still retain the capacity to fuse.

In contrast, by stage III, the diverging groups have passed the point of no return and it is inconceivable that they will ever fuse (the projection of nonfusion does not entail ‘seeing the future’; a thorough understanding of the genetics of differentiation and RI may suffice). In complete sympatry in any habitat, they will either coexist or experience competitive exclusion. In parapatry, a narrow hybrid zone may be common. At this stage, the accumulation of ‘speciation genes’ has resulted in extensively divergent (between populations) and coadapted (within population) gene complexes. The groups are thus divergent in at least some aspects of reproductive biology, sexual behaviour, and/or morphology. Although there would be no disagreement over their species status, the genomes of these good species are by no means completely isolated. Gene‐sharing by introgression could persist for a long period of time and effectively retard or nullify species differentiation over some portion of the genome. Many closely related ‘good species’ in allopatry or sympatry are at this stage. Drosophila simulans and its two sibling species, D. mauritiana and D. sechellia, are one such example – premating isolation is incomplete and hybrid females in F1 and subsequent generations are often fertile (Coyne, 1984; Lachaise et al., 1988; Wu & Palopoli, 1994). The two species of Bombina toads that maintain a stable hybrid zone may be another example (Szymura & Barton, 1991). Stage II and III are where the concept of RI is ambiguous and has been inconsistently applied.

Finally, complete RI has evolved and the two gene pools would cease sharing alleles at any part of their genomes by means of breeding. At stage IV, premating isolation is often strong in nature and F1 hybrids of both sexes are usually inviable or sterile. When F1 or F2 are fertile, occasional introgressions are quickly eliminated by selection. A stage IV, genetic analysis is not possible except with some ‘trickeries’ (Muller & Pontecorvo, 1942; Orr, 1992; Davis et al., 1996; Coyne et al., 1998) and the level of functional divergence at the inception of this stage can be extremely high (Sawamura et al., 2000).

The genic view of species

In the light of the genic basis of the full process of speciation, when do we consider species formed? Would the delineation of species be like the demarcation between adult and child – there simply is not a clearcut transition between phases? At the extreme, one may insist on the strict application of BSC with complete RI across the whole genome; thus, only stage IV of Fig. 1 would fulfil the criteria. The concept of complete isolation does not distinguish loci governing traits of speciation from other loci. By this concept, it is dubious whether D. simulans and D. mauritiana could even be classified as species because of the continual gene flow in the past [Solignac & Monnerot, 1986; Ballard, 2000; Ting et al., 2000; were the two species in sympatry at present, gene flow would likely be much higher than estimated (Kliman et al., 2000)]. Given the overwhelming evidence of functional divergence (Coyne, 1984; Coyne & Charlesworth, 1989; Davis et al., 1994; Wu & Palopoli, 1994; Coyne, 1996; Davis & Wu, 1996; Hollocher & Wu, 1996; True et al., 1996, 1997; Zeng et al., 2000), declassifying them as good species seems farfetched.

The essence of any evolution‐based species concept should be as follows: speciation is the stage where the populations will not lose their divergence upon contact and, furthermore, will be able to continue to diverge. Thus, the very essence does not have to include RI. In other words, speciation is the stage where (i) the gene pools at loci of differential adaptation would not mix even when the extrinsic barriers to gene exchange are removed; and (ii) the spread of advantageous mutations are sometimes (or often, but not always) restricted to the population in which they originate. Because the rest of the genome may mix upon contact, the genomes of the diverging populations are expected to be mosaic in their extent of differentiation.

It needs to be emphasized that mere differential adaptation, including that along a cline, does not constitute species in this genic view. The two central criteria are: (i) not losing the differentiation, and (ii) being able to continue to diverge. Therefore, the production of heterozygotes and all sorts of hybrids, which can either transfer the genes of differential adaptation between populations or form a self‐sustained hybrid population with a mixture of adaptive characters, certainly represents the loss of differential adaptation. Most geographical races, or polytypic species, are in this category and belong to the same species. In this genic view, our ability to judge whether the divergence will be lost upon contact, and hence our ability to determine species status, is a function of our knowledge of the genetics, ecology and reproductive biology of the diverging populations in question. It seems futile to attempt to devise a generally applicable species definition that demands little prior knowledge of the genetics and ecology of the taxa.

The evolutionary trajectory (fusion vs. speciation) of populations of stage II in contact depends on how strongly genes are coadapted within each nascent species and how often the coadapted gene complexes are broken up. Both the genetic architecture of differentiation (the number of loci involved, the degree of fitness reduction associated with various gene combinations, etc.) and the extent of gene flow between nascent species are crucial. This is an area where theories may figure prominently in the future (Felsenstein, 1981; Barton & Hewitt, 1985; Charlesworth et al., 1987; Barton, 1992; Orr, 1995; Turelli & Orr, 1995; Gavrilets, 1997, 2000; Kondrashov et al., 1998; Dieckmann & Doebeli, 1999; Kondrashov & Kondrashov, 1999). The models of Charlesworth et al. (1987) and Gavrilets (1997, 1999, 2000) are especially relevant to the concept developed here. The genetic architecture of differential adaptation also determines the fate of new advantageous mutations. At stage II, some advantageous mutations may permeate through the population boundary, making no contribution to their further differentiation but others may interact positively with genes of the same population and negatively with those of the other population. Such mutations will spread in only one population, contributing to further adaptive divergence.

We may summarize the genic view of species as follows: species are groups that are differentially adapted and, upon contact, are not able to share genes controlling these adaptive characters, by direct exchanges or through intermediate hybrid populations. These groups may or may not be differentiated elsewhere in the genome.

BSC – the whole genome as a cohesive unit

BSC stresses the role of isolation (first geographical and extrinsic, then biological and intrinsic) in speciation. During the process, and at the conclusion, of speciation, the genomes of the diverging populations are entirely insulated from each other. RI as a whole‐genome concept has been amply underscored in the seminal writings on BSC (Dobzhansky, 1937; Mayr, 1963, Chap. 17). Recently, in order to reconcile the observations of gene flow between nascent or even good species, it has been suggested that RI may not have to be a whole‐genome phenomenon. However, if we allow more than a trivial portion of the genome to permeate through the boundary between nascent species, then RI would lose its logical robustness. For example, if two taxa are exchanging genes over 30% of their genomes, are they reproductively isolated? How about 60%? The whole‐genome cohesiveness of BSC in fact derives its logical consistency from the three tenets given below.

The first tenet (that speciation is a whole‐genome phenomenon) rests on two strong assumptions about the genetic architecture of speciation, namely the number of genes involved is large and the genetic changes within each species are strongly coadapted, forming a cohesive unit (Dobzhansky, 1937; Mayr, 1963). In an influential address, Mayr (1959) challenged the genic view of species differentiation and labelled it as ‘beanbag genetics’. The genic view of population genetics mainly deals with simple genetic changes and ignores genic interactions whereas, Mayr contended, genic interaction as the norm of the genetics of speciation. This is where the high‐resolution genetic analyses of the last decade become most relevant (see Wu & Hollocher, 1998, for a review).

The second tenet, that complete RI and cessation of gene flow between diverging populations is the prerequisite for speciation, is a logical extension of the first tenet. Coadaptation of genes across the whole genome would certainly preclude the sharing of random fragments of genomes between species. Conversely, exchanging genomic fragments would also jeopardize the integrity of the diverging genomes. A genic concept of speciation, such as the one outlined in Fig. 1, would be incompatible with the two tenets. Indeed, Mayr (1963, Chap. 6) devoted a whole chapter on the rarity of breakdown of isolating mechanisms in nature, which he suggested is the prima facie evidence for genomic cohesion.

The third tenet, implicit in BSC, pertains to the stage at which speciation is considered complete. As the process of speciation is continuous, one might expect demarcating the stages ‘before and after’ speciation subjective, as Darwin (1859) noted. BSC is able to prescribe a transition in this process because it assumes that the whole‐genome coadaptation, and hence RI, evolves rapidly once it gets started. In this view, there exists a natural transition, which characterizes speciation.

Conflicts (gene vs. genome and adaptation vs. isolation) and consequences

Although BSC assumes a highly coadaptive genetic architecture, which led directly to the conception of whole‐genome isolation, RI was always considered as a by product of differential adaptation by Mayr (1963, p. 548–554). The process of speciation since Darwin (1859) has been about differential adaptation to different environments or diverging mating systems (Darwin, 1875). Although Darwin (1859, Chap. 8) did discuss RI and recognized its role in preventing diverging populations from becoming homogenized, he viewed RI as a secondary phenomenon. In BSC, because whole genome coadaptation and genome‐wide isolation are treated as two sides of the same coin, it is only necessary to study either, but not both, in order to understand speciation. Indeed, RI eventually became the dominant side and the thinking of adaptation was largely eclipsed as BSC subtly evolved. This imbalance has brought in several popular, albeit somewhat illogical, concepts about speciation.

(1) Non‐adaptive ‘causes’ of speciation – In a popular interpretation of BSC, the ‘causes of speciation’ denotes the event that completes RI. ‘Speciation genes’ thus refer to those that cause RI and are sometimes assumed to belong to a special class of genes unrelated to normal adaptation. On the other hand, if RI is viewed as a by product of differential adaptation as portrayed in Fig. 1, there is really no ‘causes’ of speciation; speciation genes are simply genes that determine aspects of differential adaptation and whose by product happens to be fitness reduction in the hybrids.

There are indeed a number of special genetic events whose primary phenotype appears to be RI. These events are often nonadaptive for the host’s genome (or at least of no known adaptive value). The invasion of cytoplasmic symbionts causing hybrid incompatibility (Werren, 1998) is the primary example. When a population of host acquires a certain cytoplasmic symbiont such as the bacteria Wolbachia, crosses between males of this population and females of an uninfected one yield inviable progeny. If this second population later acquires a different symbiont of cytoplasmic incompatibility, then RI will be complete between them at the time of invasion of the second symbiont. Mayr’s lack of enthusiasm for such an event is entirely consistent with his emphasis on differential adaptation (Schilthuizen, 2000). Transposable element invasion and meiotic drive divergence may be other such nonadaptive examples (Hurst & Pomiankowski, 1991; Hurst & Schilthuizen, 1998). Chromosomal changes, including translocations, inversions and polyploidization, have also occupied a large section in the speciation literature (White, 1954). The interest has been their potential in causing problems in meiosis in interspecific hybrids and, hence, postmating RI. Whether these changes may or may not be advantageous is not known and was not the primary reason for the interest. Indeed, it is now feasible to create lines in model organisms that are reproductively isolated from the rest of the species by transforming them with a toxin gene and its repressor. The transformed lines are normal but hybridization with the untransformed lines will result in hybrid inviability because of the derepression of the toxin gene. Will we then have created a new species? The suggestion seems absurd in the sense of biological differentiation but does not contradict the isolation concept.

In the Darwinian view, the genetic events discussed above are special cases because they do not result in differential adaptation to different environments or sexual systems. In BSC, they are an important category; in fact, they may be the only known causative genetic events of speciation.

(2) Does selection or drift drive speciation? In the view of differential adaptation, speciation is driven by the same forces that drive adaptation to changing environments or mating structure. It is essential to define the adaptive characters and the underlying genes. In the case of postmating isolation, the challenge is most revealing. As hybrid inviability and sterility are negative by products, what could be the normal function and selective advantage of the underlying genic changes? How does RI arise as a correlate of the genetic divergence? These questions can only be approached after genes have been identified.

In the last two decades, the formulation of the question of postmating isolation is strikingly different as it follows the isolation concept. The possible normal function and selective advantage are not considered, the only selection being that against the inviable or sterile genotypes. Most efforts have thus been directed towards the question: how can mutation and genetic drift drive the underlying genetic changes without being retarded by evolving into the unfit genotypes? To state it simply, in the total genotypic space, two pure species occupy different adaptive peaks and the space in between are filled with unfit genotypes, representing F1, F2 breakdown, etc. How did evolution proceed from one adaptive peak to the other? There are many solutions to this problem (Dobzhansky, 1937; Muller, 1940; Nei et al., 1983; Cabot et al., 1994; Orr, 1995; Turelli & Orr, 1995, 2000; Gavrilets, 1997). The purpose of this perspective is not to review them but to point out that few incorporate adaptive advantage into the model.

Not knowing what the normal functions of these genes may be, some authors have focused on genetic drift (e.g. Nei et al., 1983; Gavrilets, 1997) whereas others attempt to address the effect of the fixation of speciation genes on patterns of RI and bypass modelling the process of fixation itself (e.g. Turelli & Orr, 1995, 2000). Charlesworth et al. (1987) presented a pioneering effort at modelling RI as a by product of adaptive evolution by contrasting the evolutionary dynamics of X‐linked vs. autosomal genes. Recently, Gavrilets (1999, 2000) and Gavrilets et al. (2000) have started developing explicit models incorporating differential adaptation, which speeds up speciation well beyond what mutation‐drift can sustain. We shall return to the molecular evidence in the end.

That genetic drift has played a big role in modelling RI is more by default and expediency than by reality. It is somewhat ironic that, even silent substitutions are increasingly believed to be non‐neutral (Li, 1987; Akashi, 1997), changes that can cause severe fitness reduction in some genetic backgrounds are widely suggested to be driven by genetic drift (that the modelling is a legacy of Dobzhansky’s (1937, 1970) genetic analysis on speciation is doubly ironic). Note that the introduction of genetic drift into these population genetic models of RI is for very different reasons than those of Mayr (1963) or Carson & Templeton (1984) in their proposal of genetic revolution through founder populations, the latter being about evolution from one to another local adaptive peak.

The quest to explain the origin of RI may be analogous to the attempt at explaining the evolution of the Major Histocompatibility Complex (MHC), which was discovered as molecules for graft rejection in transplantation surgery. It would not have been possible to understand the evolution of graft rejection without first knowing about MHC and its adaptive function in immune surveillance. Graft rejection, like postmating isolation, is a by product and its origin cannot be understood on its own.

(3) Gene flow during speciation – in the strict isolation concept, any degree of gene flow is perceived to be disruptive of genome cohesiveness and capable of reversing the process of speciation. The studies of hybrid zones (Barton & Hewitt, 1985; Harrison, 1990, 1993; Szymura & Barton, 1991; Arnold & Emms, 1998; Butlin, 1998; Mallet et al., 1998) and sympatric speciation (Bush, 1975, 1998; Schliewen et al., 1994; Feder et al., 1997; Berlocher, 1998; Dieckmann & Doebeli, 1999; Kondrashov & Kondrashov, 1999), in conjunction with the genetic analysis of RI, may be gradually changing that view (Schilthuizen, 2000). Gene flow will be discussed further.

Resolution of conflicts

As BSC evolved in the years to account for new observations, two tiers of conflicts have built up within the concept itself. These conflicts are now in serious need of resolution. The first conflict is whether to view the gene or the genome as the unit of speciation; in other words, whether each genetic substitution would interact with future changes everywhere in the genome. The second conflict is whether to view speciation primarily as the process of differential adaptation or the evolution of RI. In its original formulation, BSC was a 100% genomic concept. It also viewed RI as a by product of differential adaptation. In the recent past, the whole genome concept could not be abided by and, by expediency, RI has usually been studied without reference to adaptation.

The attempts to resolve these conflicts bring us to the fine‐mapping and cloning of genes of RI. Whether the gene or genome is the unit of speciation can only be answered when the speciation history of genes of RI can be contrasted with those of the rest of the genome (Hilton et al., 1994; Palopoli et al., 1996; Wang et al., 1997; Ting et al., 2000). More importantly, if RI is a by product of differential adaptation, the underlying genes should reveal the process and mechanisms of adaptation. This is especially true for genes of hybrid male sterility which is the manifestation of divergence in the male reproductive system, and its evolution is probably driven by sexual selection (Eberhard, 1985; Coulthart & Singh, 1988; Wu & Davis, 1993; Wu et al., 1996; Presgraves & Orr, 1998; Ting et al., 1998; Wu & Hollocher, 1998).

In this perspective, I shall rely mainly on the genetic analysis among the four species in the D. melanogaster clade but the conclusion does not depend only on these studies. In Fig. 2, three levels of divergence are illustrated. The lowest one is between the two behavioural races, Z and M, of D. melanogaster, which are at stage I of Fig. 1. The next level is between D. simulans and D. mauritiana, which are at stage III and represent the most extensively analysed pair of species for the genetics of RI. Both are sibling species of D. melanogaster which provides the genetic information necessary for the analysis of their hybridization. Drosophila simulans is cosmopolitan in distribution and D. mauritiana is endemic to the island of Mauritius. There is not report of D. simulans on Mauritius but this species is present abundantly on the nearby island of Reunion. F1 hybrid females do not suffer much in viability or fecundity and are inseminated by males of the pure species with little discrimination. F1 males are sterile. The third level is between D. melanogaster and D. simulans, which produce only sterile or inviable hybrids and are in stage IV.

The first conflict: is the gene or genome the unit of speciation?

The conflict centres around the two tenets of BSC: (i) the genomes between species are extensively divergent; (ii) little genetic exchange is possible during incipient speciation. Surprisingly (i) is probably right but (ii) is not, as described below.

(i) Extensive genetic divergence between species and coadaptation within species

In Table 1, I summarize the extent of genetic divergence underlying different traits of speciation and at various levels of speciation. All these are sibling species. There are three salient features about the studies summarized in Table 1. First, the number of genes involved in speciation is substantial, even between sibling species. Secondly, the extent of gene interaction underlying RI was also unexpected (not shown in Table 1; see Palopoli & Wu, 1994; Cabot et al., 1994). In fact, there is no real evidence in the literature for the simplest genetic model of postmating isolation, the so‐called Muller‐Dobzhansky model in which only one gene from each species interacts to cause RI (Palopoli & Wu, 1994; Wu & Palopoli, 1994; Perez & Wu, 1995; Turelli & Orr, 1995, 2000; Wu & Hollocher, 1998). Conspecific and heterospecific genic interactions are both indispensable for hybrid incompatibility, a feature important in later discussions on introgression. Thirdly, different speciation traits have evolved at highly disparate rates, notably the difference between hybrid male sterility and hybrid inviability/female sterility. This is an indication that RI is not driven only by mutation and genetic drift; each trait must be experiencing different selective pressure.

The above results have indeed been surprising to ‘beanbag’ geneticists. Mayr’s assertion that species represent thoroughly reconstituted genomes is indeed closer to the experimental evidence than many geneticists’ belief that speciation is largely the result of some major changes (e.g, Goldschmidt, 1940; Nei, 1975; Raff & Kaufman, 1983; Brakefield et al., 1996). However, this triumph is insufficient to support the tenets of BSC that whole‐genome RI is absolutely central to the process of speciation.

(ii) Genetic exchanges during incipient speciation – speciation genes vs. the rest of the genome

Imagine the extreme case where all loci freely recombine. It is obvious that genes not directly involved in differential adaptation and/or RI can permeate through the nascent species boundary without impediment. Even if we assume that as many as 10% of the genes in the genome are involved in speciation at the incipient stage, the bulk of the genome should still resume to sharing a common gene pool once the diverging populations come into contact. Of course, genes exist in linkage groups. Hence the key to the question of the cohesiveness of the entire genome is linkage and recombination.

Consider stage II of Fig. 1 where alleles at loci A–E are differentially adapted between the two diverging populations. When allele A is introgressed from population 1 to population 2, it will be gradually eliminated because of reduced fitness. Whether alleles at other neutral loci co‐introgressed with A will be able to persist in population 2 depends on their being dissociated from allele A before elimination. Therefore, selection intensity against the introgressed alleles and recombination distance are both important.

Let the migration rate into population 2 each generation be m. Because of selection against speciation genes from population 1, their effective migration rate would be extremely low. Other genes linked to them will have a reduced effective migration rate, m′. In the simplest case of one speciation gene and one neutral locus that recombine with the rate r (Barton & Bengtsson, 1986; N. Takahata, pers. comm.),

Note that as long as four Nm′ > 1, where N is the effective size of population 2, divergence between populations at the neutral locus would not be able to continue (Crow & Kimura, 1970; Slatkin, 1987). Therefore, unless selection against the introgression of speciation genes is very strong, only a small fraction of the genome near them would be affected. The question then is how strong such selection is, and how large the region affected would be.

Selection against genes associated with RI depends on the underlying genetic interactions and is generally much weaker than might have been conjectured. A single gene upon introgression into another species rarely, if ever, induces hybrid sterility or inviability in animals (Wu & Palopoli, 1994 and references cited in Table 1). As a general rule, a specific constellation of genes are jointly needed to give rise to hybrid incompatibility. For example, if the A and B allele of Fig. 1 would jointly cause 90% hybrid sterility, A or B by itself might reduce fertility by <1%. This is the evidence for strong epistasis among conspecific genes cited earlier. In other words, many genes that are part of hybrid incompatibility interactions can in fact be individually introduced into another species when the constellation is broken by recombination. Although each may not be fixed upon introgression, it is eliminated slowly, allowing other cointrogressed genes to be dissociated from it.

In summary, all three parameters of eqn (1) have to assume rather extreme values (strong selection against introgressed genes, little migration and low recombination) to make the whole genome ‘congeal’ during speciation. Such theoretical arguments rely on many assumptions and we shall turn to empirical observations below.

From Fig. 1, one would expect that, at the incipient stages of speciation, the diverging genomes are mosaic with respect to the genealogies of different gene regions. This can often be indirectly inferred. For example, loci pertaining to the phenotypic characters of human racial groups are likely to be strongly geographically patterned whereas most loci unrelated to such characters exhibit little differentiation among racial groups (Nei & Roychoudhury, 1993; Barbujani et al., 1997). The behavioural races of D. melanogaster also show strong incongruities between the phenotypic and genic differentiation. Although central–southern African lines can be distinguished from the cosmopolitan ones by their mating preferences (Wu et al., 1995; Hollocher et al., 1997a) such differences are largely determined by the autosomes (Hollocher et al., 1997b), variations at most autosomal loci overlap extensively among populations (Tsaur et al., 1998; Andolfatto, 2001). Endler (1977) reviewed many cases of strong spatial differentiation in spite of gene flow. Recent Drosophila studies suggest that the divergence in phenotypic characters can occur within a few kilometers (Capy et al., 2000) or even shorter distance (Korol et al., 2000), apparently in the presence of substantial gene flow. Lewontin & Krakauer (1973) have earlier suggested that different degrees of population differentiation among random loci may indicate local adaptation for the more highly differentiated ones. This approach has been used by Wang et al. (1997).

The diverging genomes that are mosaic in their gene genealogies can be observed at the species level as well. For example, Darwin’s finches have been exchanging genes between species whose divergent morphological characters are stable (see Grant & Grant, 1996, 1998). Although the sharing of DNA variation at random loci apparently did not reverse the divergent evolution in phenotype among species, the strict adherence to the whole‐genome isolation concept has made some authors question the species status of Darwin’s finches (Freeland & Boag, 1999). Other examples may include the benthic and limnetic species of sympatric sticklebacks (Schluter, 1998; Rundle et al., 2000) and the Rhagoletis species group (Berlocher, 2000).

The above examples are indirect inferences because the genes underlying the phenotypic divergence have not been identified. Direct evidence can only be obtained with the prior identification of the speciation gene, followed by a population genetic analysis near it. The Ods gene of hybrid male sterility between D. mauritiana and D. simulans is one such example. In the coding region of Ods, gene genealogies sort cleanly by species and D. simulans is unambiguously clustered with D. mauritiana, with D. sechellia as an outgroup (Ting et al., 2000). This is not true for most other genes that have been surveyed so far (Caccone et al., 1996) suggesting that the Ods speciation gene indeed has a distinct genealogical history. What is most interesting is that the DNA pattern merely 2 kb away shows a genealogical history like the rest of the genome. That a speciation gene would leave such a small footprint of influence on the evolutionary dynamics of the genome is commensurate with theoretical considerations (Maynard Smith & Haigh, 1974; Fay & Wu, 2000) as well as other empirical observations. For example, one of the key loci affecting the domestication of maize bears the signature of artificial selection in the regulatory region but not the coding region (Wang et al., 1999). In a recent study aimed at identifying the gene responsible for the pheromonal difference between D. melanogaster populations, Takahashi et al. (2001) were able to show strong differentiation between races at a desaturase locus. Again, this differentiation is restricted to a region of only a few kb.

A rigorous proof of the genic nature of speciation shown in Fig. 1 demands three pieces of evidence: (i) the genome is mosaic with respect to the speciation history of each gene; (ii) variations at speciation loci are not shared between diverging species whereas other variations may permeate through the nascent species boundary; (iii) species divergence at the speciation loci tends to be more ancient than at other loci, inferrable from silent substitutions of these genes. Without (iii), the pattern of (ii), i.e. the sharing of variation at random loci, can be accounted for by ancestral polymorphisms, whereas the lack of sharing at speciation loci can be explained by selective sweep. The third piece of evidence has rarely been demonstrated and is only partially evident between D. mauritiana and D. simulans (Ting et al., 2000, and references therein).

Another type of observation germane to the genic view of speciation is the natural mixing of species genomes upon secondary contact. In plants, such mixing has been shown to have the potential of creating new species, or recombinational speciation (Rieseberg et al., 1995). Whether this has happened in animals can only be speculated (Grant & Grant, 1998) although the creation of dog breeds and other domesticated animals may provide a clue (Darwin, 1859). Most of the documented genome mixing in animals is the introgression of genetic materials between species. Although some of the introgressions may be advantageous, for example, the t‐complex between Mus (Hammer et al., 1989) or a chromosomal inversion between Anopheles species (Besansky et al., 1994), the main question is whether the bulk of the neutral variations can remain shared. The large body of literature on hybrid zones provides some of the best evidence of gene sharing across race or species boundary (Barton & Hewitt, 1985; Harrison, 1990, 1993; Shaw et al., 1990; Szymura & Barton, 1991; Arnold & Emms, 1998; Butlin, 1998; Mallet et al., 1998). Nevertheless, the true extent of introgression can be difficult to gauge. When there is no divergence at any locus in the entire study range, it is often conservatively attributed to the lack of divergence before secondary contact although the homogenization after the contact could produce the same pattern. Narrow hybrid zones may also tend to be cases of low introgression whereas very broad hybrid zones may be of less interest to investigators. In some studies, even allozymes themselves were suspected to be under local selection (Szymura & Barton, 1991).

In summary, the diverging genomes during (or even after) speciation can be quite ‘porous’ with respect to gene flow at nonspeciation loci.

The second conflict: is RI the byproduct of differential adaptation?

In the Darwinian view, RI is a by product of differential adaptation, whereas BSC sees the former as the prerequisite for the latter. To resolve the conflict over such a fundamental concept, we may choose either of the two approaches. First, we can analyse the genetics of adaptive differences between species or races and then study the possible roles the underlying genes may play in RI (Jones, 1998; Macnair & Gardner, 1998). The approach is of great value in its own right but the proportion of adaptive differences that have pleiotropic effects on the fitness of hybrids remains largely unknown.

The second approach, which is more direct, is to clone genes that are responsible for RI and determine whether and how the changes were driven by adaptive evolution. If we view hybrid incompatibility as an adaptive valley, flanked by the adaptive peaks of the two pure species, the focus is on how the adaptive peak of the common ancestor split (Wu & Hollocher, 1998). What forces drive the two peaks apart? How many genetic events does it take to separate them far enough to create a fitness valley? How many are adaptive changes and how many are random events?

A strong indication that the evolution of RI may be a by product of either natural or sexual selection is the relative rate of evolution towards hybrid male sterility vs. inviability. Hybrid male sterility reflects the rapid differentiation in the genetics of spermatogenesis and hybrid inviability reflects the differentiation in everything else. This may be the most interesting facet of Haldane’s rule of RI (Wu et al., 1996; Presgraves & Orr, 1998). If RI is decoupled from differential adaptation and is primarily driven by mutation, one would expect the evolution of hybrid inviability to outpace hybrid male sterility by 10:1, in terms of the number of loci involved [this ratio in Drosophila is known on the basis of the mutagenic potentials of inviability vs. male sterility (see Wu & Davis, 1993)]. However, the realized evolutionary patterns are overwhelmingly in the opposite direction with a ratio skewed further from 1:10, hence a discrepancy of two orders of magnitude (hybrid female sterility is unremarkable in its rate of evolution). This has been summarized in Table 1 which contrasts the number of loci contributing to different traits of speciation. Why do the genes for sperm‐making evolve so rapidly? A plausible suggestion is sexual selection (Coulthart & Singh, 1988; Wu et al., 1996; Presgraves & Orr, 1998; Ting et al., 1998).

Outside of the D. melanogaster clade where detailed genetic analysis is not feasible, we may rely on the patterns of species hybridization. In general, the number of species pairs yielding hybrid male sterility is much larger than those yielding inviability or hybrid female sterility. For Drosophila, the observed numbers are 199:14:3 (for male sterility:male inviability:female sterility) and, for mammals, 25:0: 0 (Wu & Davis, 1993). Even in mosquitoes whose males remain homogametic and where the recessivity of genes would confound the rapid evolution of males’ reproductive characters, a careful analysis has nonetheless revealed the rapid evolution of hybrid male sterility (Presgraves & Orr, 1998).

These observations suggest that the biology of males’ reproductive systems diverges much faster than other systems (see also Andersson, 1994; Endler & Basolo, 1998). This is an extension of Eberhard (1985) treatment of the rapid evolution of animals’ male genitalia. Sexual selection operates forcefully on males’ reproductive functions, of which male genitalia is a conspicuous example. Divergence in spermatogenic programmes should fall within the same framework and the manifestation of such divergence is hybrids’ inability to make sperm.

Recent molecular studies have strongly supported the view of sexual selection driving the rapid evolution of male reproductive genes. The most interesting example is the protamine genes of primates. Protamine is a main component of the DNA‐protein complex in the nucleus of sperm and are functionally analogous to the histones in the somatic cells. Whereas histones are the slowest evolving proteins in the mammalian genome, protamine genes are the fastest evolving ones (Wyckoff et al., 2000). There are many other examples from the studies of mammals (Sutton & Wilkinson, 1997), marine invertebrates (Lee & Vacquier, 1992; Metz & Palumbi, 1996) and Drosophila (Civetta & Singh, 1998; Nurminsky et al., 1998; Ting et al., 1998; Tsaur et al., 1998). Fast evolution, of course, is not equivalent to positive selection. Detailed molecular evolutionary and population genetic analyses are usually necessary to distinguish between the relaxation of negative selection and the operation of positive selection (Fay & Wu, 2000; Wyckoff et al., 2000).

In summary, the discussions above link the evolution of RI with the action of positive (sexual) selection, but only indirectly. Direct evidence would have to show that genes of RI themselves have been under the influence of directional selection. Lysin and bindin on the sperm of abalone and sea urchin, respectively, are two such examples. Both function in sperm–egg interactions and both have been shown to evolve rapidly because of positive selection (Lee et al., 1995; Metz & Palumbi, 1996). Interestingly, the evolution of the receptor of lysin on the egg membrane does not show a comparable, elevated rate of amino acid substitution (Swanson & Vacquier, 1998). This observation thus supports the hypothesis of sexual selection over the classical lock‐and‐key interpretation (Eberhard, 1985). The other example is the Ods gene of speciation (Ting et al., 1998), which has already been discussed in the context of species’ genic history (Ting et al., 2000). Ods is a major component of hybrid male sterility between D. mauritiana and D. simulans (Perez & Wu, 1995) and belongs to the homeodomain family of regulatory genes. Significantly, whereas the homeodomain is generally highly conserved, the homeodomain of Ods has evolved faster than even the neighbouring neutral DNAs between D. mauritiana and D. simulans. The molecular signature indisputably suggests positive selection driving this rapid evolution. There are many outstanding questions concerning the evolution of RI and differential adaptation. Molecular characterization of speciation genes may offer the promise of studying speciation at a deeper conceptual and mechanistic level.

Conclusion

The process of speciation is gene‐based but RI is fundamentally a genomic concept. Speciation defined by the criteria of RI, as does BSC, would be inconsistent with the process of speciation itself. During the years, BSC has also decoupled the analysis of RI from the study of differential adaptation, contrary to what was originally conceived (Mayr, 1963). Empirical evidence has increasingly suggested the need to revise BSC by redirecting the focus from RI of the whole genomes to differential adaptation at the genic level. In the revised (or updated) BSC, RI would be treated as divergence in the underlying reproductive (spermatogenic or oogenic), developmental (e.g. embryogenic) or behavioural (sexual) traits. For example, the divergence in the reproductive characters would lead to hybrid sterility. This divergence can be used to delineate species even when RI remains quite incomplete. The revision preserves the essential character of BSC that species are divergent and incompatible entities (but considers the incompatibility as part of divergence, and at the genic level). A more thorough treatment of the practice and implications of this revision of BSC may be a worthy subject in the near future.

Acknowledgments

I thank John Endler, Nick Barton, Roger Butlin, Jim Mallet and Brian Charlesworth for raising difficult conceptual issues confronted by the genic view of speciation. I am grateful to them, Loren Rieseberg, Ian Boussy, Stewart Berlocher, Leigh Van Valen, Michael Kohn, Tony Greenberg, Josh Shapiro, Menno Schilthuizen, Sergey Gavrilets, Dolph Schluter, Norman Johnson and two reviewers for valuable comments on the first draft (or part of it). I also thank these people and members of the internet speciation discussion group for sharing their views on the subject of speciation. This paper is supported by grants from NSF and NIH.

References