Gene Interactions and t h e Origin of Species NORMAN A . JOHNSON Many authors (e.g., Mayr 1991; Coyne 1992; Dennett 1995; O r r 1995) have noted that, despite the title of his book, Darwin (1859) did not solve the origin of species. That problem was solved in outline by Dobzhansky, Mayr, Muller, and the other architects of the modern synthesis. In contrast to Darwin, they saw biodiversity being broken into "a multitude of separate, discrete, distributions" (Dobzhansky 1937, p. 4), with species as the fundamental, discontinuous units of this diversity. This view of species is inherent in the biological species concept (BSC), wherein species are defined both by their ability to interbreed inter se and by their reproductive isolation from other such groups (Dobzhansky 1935; Mayr 1942). While many other species concepts have been put forth since (for a sample, see Table 1 of Coyne 1994), the BSC remains one of the most widely used (Coyne 1992, 1994; Orr 1995). Moreover, many evolutionary biologists who do not support the BSC still recognize the importance of reproductive isolation in the creation of biodiversity (e.g., Templeton 1989; Frost and Hillis 1990). In a recent attempt to forge a conciliation between the BSC and phylogenetic species concepts, Avise and Wollenberg (1997, p. 7754) argue that "to cleanse from species concepts all references to reproductive isolation would be to leave an unduly sterile epistemological foundation for the origin and maintenance of the biotic discontinuities so evident to Dobzhansky 60 years ago.' The reproductive barriers to gene flow are diverse but can be placed into two general categories according to when they occur: prezygotic barriers reduce the probability of the formation of hybrid zygotes, and postzygotic barriers reduce the fitness of those hybrid zygotes (Dobzhansky 1937; Mayr 1963; Futuyma 1998). In this chapter, I will discuss the relationship between genetic interactions and 198 Genetic Differentiation: From Populations to Species Gene Interactions and the Origin of Species 199 postzygotic reproductive isolation. With few exceptions, I will not discuss prezygotic reproductive isolation, as this subject is discussed in Meffert's chapter (chap. 11) in this volume. I will also not explicitly discuss hybrid zones; this subject is discussed by Gardner et al, (chap. 16) in this volume. The Dobzhansky-Muller Model Dobzhansky (1937) and Muller (1940, 1942) each recognized the importance of gene interaction in the evolution of postzygotic reproductive isolation (also called hybrid fitness reduction). In the allopatric divergence model of speciation, microevolutionary forces (mutation, random genetic drift, and natural selection) act upon geographically isolated populations. Without the homogenizing force of gene flow, these populations will diverge genetically. With increasing divergence, the genomes of the two nascent species become increasingly incompatible with each other. If the geographic barriers were removed to allow for the production of hybrids between diverged populations, the hybrids would possess combinations of genes that do not work well together (i.e., are incompatible). These genetic incompatibilities would result in the hybrids being (partially or completely) inviable or sterile and thus would produce reproductive isolation. E. Fitnesses of genotypes Figure 12.1 Figure 12.1 The Dobzhansky-Muller mode1 of evolution of postzygotic reproductive isolation. Assume that A and B represent different loci, each with two alleles (subscripted 0 and l), and that individuals with any combination of A l and Bl have greatly reduced fitness. (A) Stage I: the population is fixed for alleles An and Bo. ( B ) Stage 11: a geographic barrier splits the population and gene flow ceases. (C) Stage 111: the variant Al appears in the population on the left and the variant Bl appears in the population on the right. (D) Stage IV: the Al allele becomes fixed in the left population and Bl becomes fixed in the right population. (E) Hybrids from a cross between the left and right populations would be AoA,BoB, and would thus have reduced fitness. Note that neither population passes through an adaptive valley. (continued) Nonadditive interactions of variants at different loci (epistasis) are presumed to be the genetic basis for these incompatibilities (Fig. 12.1). In principle, the simpler case of single-locus underdominance could lead to incompatibility if populations could fix different homozygotes at each locus (hybrids would be heterozygotes and thus less fit). This underdominance model, however, is very unlikely due the exceedingly low probabilities of fixing different homozygotes in the different populations. In order for this to occur, one or both of the populations would have to go through an adaptive valley (where less-fit heterozygotes were transiently common). In contrast, reproductive isolation could occur without the need for either population to cross an adaptive valley if the incompatibility were caused by alleles of at least two loci that interact epistastically (Dobzhansky 1937; Mayr 1942; Walsh 1982; Barton 1989b; Orr 1991; Bateson in Orr 1996). Thus, in the allopatric divergence model, the evolution of hybrid incompatibility does not require the parental populations to transverse adaptive valleys, and reduces to the formation of incompatible sets of loci (Orr 1995). Although alternatives to this model have been proposed (e.g., White 1978; Diehl and Bush 1989), evolutionary biologists widely support the allopatric divergence model (Futuyma and Mayer 1980; Coyne 1992; Wu and Palopoli 1994; Futuyma Gene interactions and the Origin of Species 20 Box 12.1 (continued) Box 12.1 Allelic effects in negative and positive phenotypes MIelic variants at "hybrid incompatibility loci" have t w o distinct phenotypic effects: one ~ p o the n fitness and othertraits of hybrids (the negative ~henotype),and one upon the fitless and other traits of the pure species individuals that possess it (the positive phenotype) [see Fig. B12.1). If populations are diverging completely in allopatry, only pure species individuals (and not hybrids)will becreated ineachgeneration. Hence, evolutionaryforces(naturalselection, random genetic drift) will operate upon only the positive phenotype and not the negative phenotype. Incontrast.geneticstudiesofhybrid incompatibility, by necessity, beginwiththe negative phenotype (the hybrid effects). Forthe most part,thesestudies have not examined the positive phenotype. Thus, the evolutionary dynamics ofthe hybrid incompatibility loci arenotwell known.Thechallengeforspeciationgeneticresearcherswillbetounitethesetwo realms.JohnsonandWade's(1996)model.whichconsiderstheeffectsofasingleallelicsubstitution on purespeciesand hybrid genetic backgrounds, isthe first-steptoward aconceptual framework of this unification. This model, however, is still an oversimplification of the complexity involved (seetext). Conspecific (Positive) Phenotypes - Hybrid (Negative) Phenotypes Fig. 612.1 (continued) Figure B12.1 illustrates the challenges involved in the unification of these t w o realms. Here, genetic variants (middle panel) affect both the positive (pure species) phenotype (top) and the negative (hybrid) phenotype (bottom). In the figure, G l o and GI b are allelic variants at locus I, and G2a and G2b are allelic variants at locus 2. The traits C l and C a r e expressed in the positive (pure species) genetic background, and traits HI and H; are expressed in the negative (hybrid) genetic background. Relative t o a standard variant (not shown),Glagreatly increases(assh0wnby ++) bothCl andC2, whileitdecreasesHl. Likewise. GI b decreasesCl, while it greatly increases HI. Notethat for locus I,the relationship between effects on conspecifics and on hybrids is negative ( r <0, in Johnson and Wade's model). For locus 2, this relationship is positive: G2a greatly increasesboth C2and H2. Natural selection operates directly on Cl and C;,but only indirectly on H I and Hi. Selection for the increase of Cl may result in an increase in frequency of Gla. If so, t h e n H I should decrease. Another possibility is that selection for an increase of C j would result i n a decrease in frequency of Glb. If so, HI would decrease, and so would H2. Selection that alters C2 could result in changes in allelic frequency at either o r both loci, and could have effectsoneither o r both H I and Hi. Despite the apparent complexity of this figure, it greatly underrepresents t h e complexity of real systems. Many genes (often more than the t w o shown here) can affect a given trait (either pure species or hybrid). Furthermore, this figure does not depict conspecific epistasis. 1998). The controversy is not over whether speciation occurs via divergence in allopatry, but rather to what extent reproductive isolation arises via other means. It is also clear that, except for certain, quite restrictive conditions (Coyne 1974; Leibowitz 1994; Johnson and Wade 1995), postzygotic reproductive isolation does not arise as the result of direct selection (in contrast to the reinforcement model of prezygotic reproductive isolation; Dobzhansky 1940; Howard 1993; Kelly and Noor 1996), but only as a pleiotropic by-product of divergence. Microevolutionary forces that operate within species, coupled with geographical isolation, thus lead to the formation of intrinsic reproductive barriers and eventually to biological discontinuities. One can consider the observed hybrid fitness reduction to be a negative phenotypean epiphenomenon of the underlying hybrid fitness reduction (Box 12.1). The divergence itself is likely to be opportunistic. By this, I mean it is likely that many classes of "genes" (in the broad sense) will be involved, and the relative importance of different types of genetic systems that contribute to the divergence that underlies hybrid fitness reduction will probably vary across taxa (Johnson 1998). In the next two sections, I will present what we know about the genetic architecture of hybrid dysfunction. I will then discuss the consequences of the complexity of the genetic architecture of hybrid incompatibility for the Fisher-Wright debate, and in general. Next, Iconsider the plausibility that hybrid fitnessreduction is generated by the drift-induced fixation of intraspecific epistatically interacting variants. In the final section, I discuss what types of future research will be most beneficial to our understanding of the role that gene interaction plays in the evolution of reproductive isolation, and the consequences of such interactions. 202 Genetic Differentiation: From Populations to Species Detection of Gene Interactions that Underlie Hybrid Fitness Reduction Dobzhansky (1936) also pioneered the genetic analysis of hybrid traits. His method, the backcross analysis, consists of crossing individuals of one species marked with multiple genetic markers to individuals of a related species. If at least one sex of the F l hybrids is fertile, then the F l can be backcrossed to one or both of the parental species. The difficulty of examination of the genetic basis of the inability to produce viable, fertile hybrids by the production of fertile hybrids is at least partially alleviated by the large number of species in which one sex is (at least partially) fertile and the other sex is sterile. Across a wide variety of taxa, the vast majority of hybridizations with sex-biased asymmetry of hybrid fitness reduction follow Haldane's (1922) rule, where the heterogametic sex (male in Drosophila and mammals, female in birds and butterflies) is most severely affected. Various explanations for Haldane's rule are discussed in the reviews of Wu et al. (1996), Laurie (1997), and Orr (1997), and references therein. Genetic analyses that stop after only one or two generations of backcrossing are often limited in their ability to detect the locations of incompatibility genes, and the nature of the interactions, because the genotypes of the hybrids are too complex and ill-defined (Wu and Davis 1993; Wu and Palopoli 1994). With the introgression method, a refinement of the backcross analysis, researchers have dramatically increased the resolution with which the genetic architecture of hybrid fitness reduction can be examined (e.g., Coyne and Charlesworth 1986; Wu et al. 1993; Wu and Palopoli 1994; see also True et al. 1996). This method allows for the production of a large number of genetically similar individuals, each with a small and defined portion of the genome from one species in the genetic background of another species. Various aspects of the phenotype (e.g., sperm motility, progeny production, mating ability) of these individuals then can be assayed. A major conclusion of the introgression studies in the Drosophila melanogaster species group is that dozens-if not h u n d r e d s ~ ogenes, f when placed in the genetic background of a closely related species, have severely deleterious effects on male fertility (Wu and Palopoli 1994; Palopoli et al. 1996; True et al. 1996). The large number of these so-called hybrid sterility genes demonstrates that, despite the lack of morphological differences between these species, substantial functional divergence has occurred (Wu and Palopoli 1994; Wu et al. 1996; Johnson and Kliman, submitted). The types ~.of interactions involved in hybrid fitness reduction are rather diverse, but some basic patterns are apparent. First, X-autosome interactions appear to be common (Dobzhansky 1936; Charlesworth et al. 1987; Coyne and Orr 1989a). It has been previously thought that the X chromosome has larger effects on hybrid fitness reduction than the autosomes (reviewed in Charlesworth et al. 1987; and Coyne and Orr 1989a), but recent studies in Drosophila show that the density of genes that contribute to hybrid sterility on the autosomes is comparable to that on the X chromosome (Hollocher and Wu 1996; True et al. 1996). Thus, the "large X effectnis probably an artifact of how the studies were performed. Analyses that stop Gene Interactions and the Origin of Species 203 after the first generation of backcrossing will typically make heterozygous substitutions of the autosomes, but hemizygous substitutions of the X chromosome (see Wu and Davis 1993 for further discussion). The Y chromosome sometimes, but not always, contributes to the sterility of hybrids between closely related species of Drosophila (see Johnson et al. 1993, and references within). Negative interactions between heterospecific X and Y chromosomes is a simple hypothesis to explain Haldane's rule (Coyne 1985). There are cases wherein hybrid dysfunction was thought to be due to X-Y interactions, but further investigation showed that this was probably not the case. For instance, the sterility of F l male hybrids between D. simulans and D. sechellia was more likely due to a combination of X-autosome and Y-autosome interactions than due to direct X-Y interactions (Johnson et al. 1992). Cointrogression studies, wherein both the Y chromosomes and portions of different autosomes are introgressed at the same time, provide strong direct evidence for Y-autosome interactions that cause hybrid male sterility between closely related species in both the D. repleta and D. virilis species groups (Pantazidis et al. 1993; Lamnissou et al. 1996). The cytoplasm of nearly every cell in most metazoans includes numerous mitochondria, each with their own genome. Divergence could lead to incompatibilities between the mitochondria1 genes (or other cytoplasmic factors) of one species and the nuclear genes of the other. The cytoplasmically located endosymbiont bacteria Wolbachia sp. has been implicated in intraspecific cytoplasmic incompatibility that occurs in many different orders of insects (Wade and Stevens 1985; Hoffman et al. 1986; Breeuwer and Werren 1990; Giordano et al. 1995). In many hybridization studies, interspecific replacement of the cytoplasm has no detectable effect on hybrid male sterility (e.g., Zeng and Singh 1993; Davis et al. 1994; Goulielmos and Zouros 1995). In other hybridizations, cytoplasmic interactions do play an important role in the reduction of hybrid fitness. For instance, both nuclear-cytoplasmic and nuclear-nuclear incompatibilities reduce the fitness of hybrids between two species of wasp in the genus Nasonia (Breeuwer and Werren 1995). Cage studies in Drosophila have shown instances of more subtle nuclear-cytoplasmic incompatibility, where the frequency of flies with heterospecific cytoplasm decreases (e.g., Hutter and Rand 1995). There are a few exceptions to Haldane's rule where, in male heterogametic taxa, female hybrids are more affected than their male counterparts. Many of these exceptional cases are believed to be due to incompatibilities between cytoplasmic (maternal) products and the X chromosome (Sawamura et al. 1993a, 1993b; Sawamura 1996; Wu et al. 1996; Laurie 1997). In the cross of XAXAfemales to XB males, the male hybrids receive their X chromosome and their cytoplasm from their mother, but females receive an additional X chromosome (XB) from their father. The combination of Xa and cytoplasm from species A may result in hybrid dysfunction, but only in females. Consistent with cytoplasmic-nuclear interactions, exceptions to Haldane's rule in Drosophila hybridizations occur more often for viability than for sterility, because maternal effects are likely to be less important in the latter case. Probably the best studied example o f nuclearcytoplasmic incompatibilities that lead to an exception of Haldane's rule is the cross between D. simulans females and D. melanogaster males (Sawamura et al. 1993a, 1993b). Gene Interactions and the Origin of Species 205 204 Genetic Differentiation: From Populationsto Species Epistasis Gets Complex While hybrid fitness reduction inherently requires epistatic interactions under the allopatric divergence model, it is still possible to consider reproductive isolation evolving via changes at a single genetic locus in each lineage. Most population genetics models of allopatric divergence make this assumption, and thus are essentially one-locus models (e.g., Orr 1995; Johnson and Wade 1996; Orr and Orr 1996). Johnson and Wade (1996) explicitly consider the effects of allelic substitutions in conspecific and hybrid genetic backgrounds. Their model essentially treats the epistasis involved in hybrid fitness reduction as a form of genotype-byenvironment interaction where the conspecific and hybrid genetic backgrounds are different environments for variants at a single genetic locus. Yet, the genetic architecture of hybrid fitness reduction often appears to be more complicated. In the past few years, it has been discovered that the genes that contribute to hybrid male sterility in the D. simulans/D. tnuuritiana hybridization appear to display complex epistasis (Cabot et al. 1994; Palopoli and Wu 1994; Perez and Wu 1995; Davis and Wu 1996; Masdie et al. 1998). That is, introgressions that sterilize males can be split into two introgressions, neither of which alone can sterilize males (Palopoli and Wu 1994). The question is now whether there are any sufficient genes for hybrid male sterility-ones that will cause sterility if introgressed into the genetic background of the other species without their requiring the introgression of other "helpern genes. Although Odysseus (Ods)locus first appeared to be a "major genen that causes sterility when introgressed from D. mauritiana into D. sechellia (Perez et al. 1993), finer-scale analysis revealed that Ods was not sufficient because introgressions that contain only Ods (without its helper factors) are semifertile (Perez and Wu 1995). At present, there is no compelling evidence for the existence of sufficient hybrid male sterility genes in Drosophila. While the best evidence for hybrid sterility that involves interactions of conspecific genes comes from the studies in the D.sirnulans clade (where the genetic resolution is highest), there is support for complex epistasis from other Drosophila species groups. For instance, fertility reduction in male hybrids between-species in the D. buzzatii group is due to many genes of small effect that interact to produce complex epistasis (Naveria and Fontadevila 1986, 1991; Naveria and Masdie 1998). In addition, at least three loci appear to interact to cause inviability of hybrids between D. buzzatii and D. koepferae (Carvajal et al. 1996). Complex Epistasis, the Wright-Fisher Debate, and the Evolution of Natural Populations The Issue Are population subdivision, linkage, and within-species epistasis important in the evolution of postzygotic reproductive isolation? This question is actually a microcosm of the debate between R. A. Fisher and Sewall Wright about how natural populations evolve (Brodie, chap. 1, this volume). While Fisher (1918, 1930) certainly believed that nonadditive genetic interactions could be commonplace, he generally regarded them as unimportant and similar to environmental variation in their effects. Instead, he focused on the average effect of an allelic substitution. In contrast, Wright (1931, 1932, 1977) believed that epistasis was universal a n d played a crucial role in evolution. In Wright's (1932, 1977) view, populations were often constrained at local, but not global, optima along the adaptive landscape, unable to scale higher peaks by mass selection. Wright proposed his shifting balance process as a mechanism in which these populations could reach higher peaks. This process requires assumptions about population structure, epistatic genetic variance, and linkage (Wright 1932, 1965, 1977). Wright's theory had tremendous influence on founder-effect speciation theory (e.g., Mayr 1963; Carson 1975; Templeton 1980a; Provine 1989). Discussions and debate about the shifting balance process, epistasis, and the nature of adaptive landscapes continue in the current evolutionary biology literature (e.g., Crow et al. 1990; Barton 1992; Wade 1992b; Kauffman 1993; Phillips 1993, 1996; Wagner et al. 1994; Whitlock et al. 1995; Michalakis and Slatkin 1996). The contrast between two recent perspectives in the journal Evolution (Coyne et al. 1997; Wade and Goodnight 1998) very clearly demonstrates that issues of the Fisher-Wright debate about the shifting-balance theory per se, and evolution in natural populations in general, are still contentious. The Introgression Approach What do the studies of the genetic architecture of hybrid fitness reduction tell us about the validity and importance of these peak shift models of speciation? What do they tell us about the evolutionary dynamics that act upon the positive phenotype and, specifically, whether epistasis is involved in this process. First, is epistasis necessarily involved in the negative phenotype? Although strong epistasis for the proportion of males with motile sperm is observed between conspecific genes, it is still possible that, on a different phenotypic scale, the effects of the genes could be additive (discussed in Palopoli and Wu 1994, and references therein). It is also possible that, at a biochemical level, the conspecific genes are not interacting with one another. Consider the case where two different loci ( A and B) from species 1 must be introgressed into species 2 to give sterility, or, in other words, neither A nor B are sufficient alone (Fig. 12.2). There are several possible models for the actual biochemical interactions, including the following two simple scenarios. In the "direct interaction model," A and Bcould both be interacting with each other and a locus or loci in the genetic background of species 2. Alternatively, A could be interacting with something in species 2's background (call it c) and B could be interacting with something else in the background (call it d ) . In this latter case, the "threshold model," sterility is a threshold-like character: one interaction cannot cause sterility, but two are sufficient (see also Naveira 1992; Naveira and Maside 1998). The finding that complex epistasis underlies hybrid sterility, while consistent with the Wrightian view of evolution, can still be consistent with a Fisherian view, as Palopoli and Wu (1994) discuss at length. The different types of complex epistasis discussed in the previous paragraph have different implications. Suppose 206 Genetic Differentiation: From Populationst o Species A. Complex Epistasis a v 6.Direct Interactions fertile C. Threshold Model Figure 12.2 (A) Different genetic interactions that underlie complex epistasis. Solid lines between loci represent conspecific interactions, dotted lines heterospecific interactions. (B) Direct interactions. A and B directly interact with c and d. (C) Threshold model. A interacts with c, B interacts with d. One interacting pair of genes is not sufficient to cause hybrid fitness reduction. that complex epistasis was often due to direct interactions. The fact that the negative phenotype of hybrid sterility requires biochemical interactions among conspecific loci makes it more likely that those combinations of genes were under positive selection. That is, the discovery of genes that directly interact in the negative phenotype implies that they would more likely directly interact in the positive phenotype. In this model, reproductive isolation would likely accrue from divergence that came about via a Wrightian process, not necessarily the shifting balance sensu stricto, but a process where population structure, gene interaction, and linkage were of importance (Wade and Goodnight 1998). In contrast, evidence for the threshold model, while not contrary to the Wrightian viewpoint, does not lend support to it either. Unfortunately, we do not have much direct evidence for either of these models. Palopoli and Wu (1994) present an indirect test. They argue that if epistasis were involved in the evolution of the positive phenotype, interacting hybrid male sterility factors should be linked (within a few centimorgan, cM) more often than random sets of loci. In practice, the introgression method is usually biased toward detection l i Wu 1994; Wu and Palopoli 1994). To of linked interacting factors ( ~ a l o ~ oand address this question, several studies have combined introgressions from different regions of the chromosome (> 10cM away) that singly do not cause sterility. The combinations, however, can result in sterility (Naveira 1992; Masdie et al. 1998; Y. Xu, M. F. Palopoli, and C.-I. Wu 1994, unpublished data). What is not known is whether the combinations of linked introgressions are more prone to result in sterility than combinations of similarly sized unlinked introgressions. Thus, we cannot yet infer, from the genetic architecture of the negative phenotype, whether epistasis is important in the positive phenotype. Gene Interactions and the Origin of Species 207 The Quantitative Genetic Approach What are the evolutionary processes responsible for the divergence that leads to hybrid fitness reduction (the positive phenotype), and what are the relative importances of each of these processes? It is difficult to address these questions, even in model systems like the D. simulans clade. Outside of these few model systems, the tools needed for fine-scale genetic mapping and molecular characterization are just not available. Given these limitations, what generalizations can we make about the evolution of postmating reproductive isolation in general, and, more specifically, about the role that epistasis may play? One complementary approach to the study of hybrid traits is the use o f quantitative genetic tools to characterize the underlying genetic variation and covariation within one species for traits that are expressed only in the hybrids produced by a cross with another species (Wade and Johnson 1994; Wade et al. 1994,1997, 1999). The characterization of variation is, in general, instrumental for the study of its adaptive significance,especially in organisms that do not lend themselves a s easily to fine-scale genetic analysis. While this approach has thus far been used only in hybridizations between two closely related species of flour beetles (Tribolium castaneum and T.freemani), it is quite general; it can be used with any pair of species that can produce hybrids and whose breeding can be controlled. Interspecific crosses of T. castaneum and T. freemani generally produce a large number of viable but sterile hybrid offspring (Wade and Johnson 1994). In the cross of T. castaneum males to T. freemani females, there are generally more female than male hybrids (owing to hybrid male inviability), and the surviving hybrid males often have morphological deformities (leg and antennal) (Wade and Johnson 1994; Wade et al. 1994, 1997, 1999). In contrast, morphological abnormalities are rare and sex ratios are near 1 : 1 in conspecific crosses, even under conditions of inbreeding and temperature extremes (Wade et al. 1997). Upon mating to the same reference strain of T.freemani, different naturally occurring and laboratory strains of T. castaneum produce hybrids that differ substantially in these hybrid traits. Moreover, by use of breeding designs (such as half-sib lineages), we have been able to detect abundant among-lineage, within-strain genetic variation (Wade et al. 1994, 1997, 1999). More complicated breeding designs can be used to detect epistatic genetic variance (M. J. Wade, N. A. Johnson, and J. Santiago-Blay 1999, unpublished manuscript). If groups of brothers are used as sires, first-cousin families-in addition to half-sib and full-sib families~canbe generated. With these three levels of grouping, it is possible to detect and estimate the additive x additive epistatic component of genetic variance for these hybrid traits (Falconer and McKay 1996). Significant additive x additive epistatic genetic variance for both traits (hybrid sex ratio and the proportion of hybrid males with deformities) segregates within a laboratory strain of T. castaneum (M. J. Wade, N. A. Johnson, and J. SantiagoBlay 1999, unpublished manuscript). What does epistatic genetic variation for these hybrid phenotypes imply for the evolution of the positive (pure species) phenotype? The question again i s one of continuity: "What does knowledge about the genetic architecture of the negative phenotype (hybrid fitness reduction) tell us about the evolution of the positive Gene Interactions and the Origin of Species 209 208 Genetic Differentiation: From Populationsto Species phenotype (the underlying divergence)?" These results would appear to support epistatic genetic variance playing an important role in the evolution of the positive phenotype. However, until we obtain a better theoretical understanding of the continuity of gene interaction across these two phenotypic realms, this conclusion cannot be considered definite. Epistasis and its Consequences Given that hybrid incompatibility requires nonadditive genetic interactions, which are often complex, what can we predict about the patterns of speciation? Orr's (1995) population-genetic model of the evolution of hybrid incompatibility predicts that postzygotic reproductive isolation between diverging populations should accumulate at a rate faster than linear with time (what he calls "snowballing"). Snowballing occurs because of the rapid rise in the number of combinations that could potentially reduce the fitness of hybrids as the number of genetic differences increases. Orr (1995) has also shown that it may be easier to evolve reproductive isolation with incompatibilities that involve three or more loci than those with just two (see also Cabot et al. 1994). At the initial stages of isolation, however, systems with only two incompatible loci should evolve reproductive isolation faster than those with three or more because of the required waiting time for mutation and fixation at each locus (N. Johnson 1999, unpublished results). Hence, the "snowballing" effect should be even more pronounced in incompatibility systems with a large number of loci. Coyne and Orr (1989b, 1997) and Sasa et al. (1998) have examined the relationship between genetic divergence and indices of reproductive isolation in Drosophila and frogs, respectively. In both the Drosophila and the frog data sets, the rates of acquisition of postzygotic reproductive isolation are linear, or even less than linear (not snowballing), with respect to genetic distance (as measured by Nei's D, which is believed to increase roughly linear with time). A plausible explanation for this apparent discrepancy is that the relationship between numbers of genetic incompatibilities and the indices of reproductive isolation used by Coyne and Orr (1989b, 1997) and Sasa et al. (1998) is asymptotic. For instance, there are many more incompatibility loci found in the D. simulans/D. mauritiana hybridization than in the D. simulansjD. sechellia hybridization (Johnson et al. 1993; Wu et al. 1993; Palopoli et al. 1996), but the postzygotic reproductive indices for these hybridizations are the same (0.5) (Coyne and Orr 1989b, 1997). Dobzhansky (1937) and Muller (1942) realized that their two-locus model of the evolution of reproductive isolation predicts asymmetry of the location of hybrid incompatibility loci in reciprocal crosses. That is, the genes from species A that are incompatible with the genetic background of species B should not be the same as the genes from species B that are incompatible with the genetic background of species A. Later researchers extended the theory and presented evidence for initial asymmetry followed by symmetrical sterility/inviability (Wu and Beckenbach 1983; Zouros 1986; Johnson et al. 1993; Zeng and Singh 1993). More recent data from high-resolution mapping also support asymmetry. Studies that introgress D. mauritiana X-chromosome segments into the genetic background I I i I of D. simulans map the Odysseus (Ods) hybrid male sterility factor to position 16D of the polytene chromosome (Perez et al. 1993; Perez and Wu 1995; Ting et al. 1998). Note that while Ods requires "helper factors" to yield complete sterility (Perex and Wu 1995), it can still be discretely mapped and characterized at the molecular level (Ting et al. 1998). Reciprocal introgressions-that is, introgression of the segments of D. simulans into the genetic background of D.mauritianareveal several factors that interact to cause sterility in hybrids (Palopoli and Wu 1994). None of these factors maps to position 16D (Palopoli and Wu 1994). Hybrid sterility factors are located at different positions in the reciprocal introgressiins. The genetic architecture of hybrid fitness reduction may also have implications for models of the evolution ofprezygotic reproductive isolation. In contrast to postzygotic reproductive isolation, prezygotic isolation may evolve because of the direct action of natural selection according to the reinforcement model (Dobzhansky 1940; Howard 1993). Although reinforcement was previously considered to be unimportant and perhaps even implausible (e.g., Butlin 1989), in the 1990s several lines of empirical evidence (Coyne and Orr 1989b, 1997; Howard 1993; Noor 1995; Butlin and Tregenza 1997; Saerte et al. 1997) have been found that support the importance of reinforcement. In addition, more recent theoretical models show that reinforcement is plausible under a fairly wide set of conditions (Liou a n d Price 1994; Kelly and Noor 1996; Servedio and Kirkpatrick 1997; Kirkpatrick and Servedio 1999). Two more recent models consider the feasibility of reinforcement, given different cases of the genetic architecture of hybrid fitness reduction (Kelly and Noor 1996; Kirkpatrick and Servedio 1999). Kelly and Noor (1996) find that, in cases where hybrid males are less fit than hybrid females (in accordance with Haldane's rule for male heterogametic taxa), reinforcement will be facilitated if X-autosome gene interactions underlie much of the hybrid fitness reduction. In the case where hybrid females are more adversely affected, X-autosome interactions that cause hybrid fitness reduction will impede reinforcement. Given the generality of ~ a l d a n e ' rule i (Wu et al. 1996; Laurie 1997; Orr 1997), we might expect reinforcement more often in male heterogametic taxa (e.g., flies and mammals) than in female heterogametic taxa (e.g., birds and butterflies) if X-autosome interactions are also generally responsible for much hybrid fitness reduction. For the situation wherein a n island population receives migrants from the mainland, Kirkpatrick and Sevedio (1999) show that the likelihood of reinforcement should increase as the number of loci that interact to cause hybrid incompatibility increases. This effect increases much less than linearly with the number of loci involved, such that an increase in the number of loci involved in hybrid incompatibility beyond about 10 has little further effect on increases of the likelihood of reinforcement (Kirkpatrick and Servedio 1999). Synthetic Lethals and Reproductive Isolation Synthetic lethals are genetic systems wherein variants at different loci cause lethality, but only in combination; individually, their effect on fitness is minimal (Dobzhansky 1946). Synthetic lethals and the more general case of synthetic 2 10 Genetic Differentiation: From Populations to Species deleterious loci (SDL) represent an extreme form of synergetic epistasis-a type of gene interaction that has important implications for the evolution of sex and inbreeding depression (Phillips and Johnson 1998; Phillips et al., chap. 2, this volume). While it was a fairly popular topic of study between approximately 1955 and 1975, the subject of synthetics in natural populations has been largely neglected in the more recent literature (but see Thompson 1986; Phillips and Johnson 1998). This neglect is unfortunate because an understanding of the role that variation in SDL systems plays in natural populations can help address the following question: "To what extent can hybrid dysfunction be generated during the separation of populations by variation already present in the original population, where there is no input of new mutations?" Muller (and the classical school) differed with Dobzhansky (and the balance school) on this issue. Their differences are highlighted in a passage from Lewontin (1974, pp. 27-28) who states [I]f populations are almost entirely homozygous [as held by the classical school], then speciation must await the occurrence of new mutations that may be advantageous in a new environment in which an isolated population is found.. . . Unlike the hypothesis of homozygosity, the balance hypothesis presumes that the genetic variation for speciation is always present, so that speciation awaits only the appropriate biogeographical and ecological events. To address how much hybrid fitness reduction can result from already-present variation, two factors are of interest. (1) "How much variation is likely to be present in SDL systems in mutation-selection balance?" (2) "To what extent can a given amount of variation in SDL systems lead to hybrid fitness reduction upon vicariance?" To address the first issue, Phillips and Johnson (1998) examined the expected frequencies of carriers at SDL at mutation-selection balance. They found that, in the case where only the double homozygotes (aabb individuals, where a and b are carrier alleles) are selected against, the frequency of SDL carriers can be rather high: with free recombination, approximately the quartic root of the mutation rate (u) divided by the selection coefficient (s), and even higher with tight-linkage. For instance, given u = l o 6 and s = 0.5, the expected carrier frequency is about 3.7%. If the extreme synergetic epistasis of SDLs is common, then it is likely that there will be much SDL variation within populations. Unfortunately, at this time, we know too little about the degree to which multilocus systems exhibit extreme synergetic epistasis to judge whether SDL variation should be common (Phillips and Johnson 1998). Despite the possibility of the existence of much standing variation in SDL systems, it appears unlikely that this variation can generate much reproductive isolation upon splitting of the original population and subsequent genetic drift in the subpopulations (N. A. Johnson 1999, unpublished results). Carrier alleles at alternative loci in an SDL system must be fixed in each population (i.e., one must become fixed for AAbb and the other fixed for aaBB). As each fixation event has a low probability (no greater than the frequency of the carrier allele prior to subdivision), the probability of carriers fixing at alternative loci will be very low. If after the vicariance event, however, the fitness matrices changed such that carrier Gene Interactions and the Origin of Species 2 1 I genotypes were advantageous, variation at SDL could be a major contributor to reproductive isolation in the early stages of speciation. Note that this scenario requires that assumptions are met about both the fitness parameters in new environments and the types of genetic interactions involved. How often such conditions will occur in nature is not known. Second, even if alternative alleles at SDL are fixed in different populations, the amount of reproductive isolation generated upon crossing the two populations is likely to be low. Given complete epistasis, alternative fixation of SDL carriers cannot result in fitness reduction in the Fl and will generate only some Fz fitness reduction (1116 of the Fz will be aabb with free recombination; fewer with linkage). If there were some selection against AaBb individuals (incomplete epistasis), more reproductive isolation would be generated by crossing populations fixed for alternative carriers. In that circumstance, however, the probability of fixation of alternative carriers would be dramatically reduced because (1) the expected frequency of carrier alleles at mutation-selection balance declines substantially with departures from complete epistasis (Phillips and Johnson 1998), and (2) there is more selection in operation against carriers after the populations split. Perspective and Prospectus T o use a single term, epistasis, to describe all gene interactions suggests we either do not care or do not know how to deal with the complexity that interlocus interactions bring. Different forms of gene interaction have different consequences, and we should move toward elucidating the nature of these consequences and which forms of interaction are most prevalent. (Phillips 1998, p. 1169) With respect to the role of epistasis, postzygotic reproductive isolation is different from other areas of evolutionary genetics. While gene interaction may or may not be of general importance for the evolution of natural populations, it need not be. Adaptation can work perfectly well in an additive world. In contrast, postzygotic reproductive isolation requires gene interaction, at least between heterospecific loci. Because of the ubiquity and necessity of gene interaction in postzygotic reproductive isolation, researchers in this field need to pay special attention to Patrick Phillips' words quoted above. Simple documentiation of the regions of the chromosome (or even the specific genetic loci) that interact to cause hybrid fitness reduction does not allow us to easily understand the processes by which reproductive isolation has accumulated between the diverging species. We expect genetic interaction to be responsible for the reduction of fitness in hybrids, regardless of the evolutionary processes and forces responsible for the divergence that has hybrid fitness reduction a s a byproduct. Further, just to show that the interactions are complex provides little information about whether gene interaction was responsible for the divergence, and what types of gene interactions were involved. Different kinds of gene interaction 212 Genetic Differentiation: From Populations to Species could be consistent with the data shown in the complex epistasis studies, and the implications of different types of complex epistasis could be quite different. We need to continue to use the introgression approach (where feasible) to better characterize the exact nature of what has been called complex epistasis. At the very least, studies such as those of Naveira (1992) and Masdie et al. (1998). wherein introgressions from different parts of the genome are combined, need to be performed more systematically. Unfortunately, these fine-scale genetic dissections are currently feasible in only a few model systems, such as the Drosophila simulans clade, and require an enormous amount of time and resources. We also need more empirical work on speciation genetics outside of model systems. While highresolution genetic mapping studies are impossible at the moment, and probably will remain so for the foreseeable future, quantitative-genetic approaches such as the one outlined above, in connection with developing theory, can be very useful. Despite some recent progress in this area (e.g., Orr 1995; Johnson and Wade 1996;Orr and Orr 1996), neither population- nor quantitative-genetic theory is well integrated with speciation genetics. We need to establish a conceptual framework in order to understand how the negative phenotype seen in the hybrid genetic background is affected by the evolutionary forces that act upon the positive phenotype in the pure species genetic background. Johnson and Wade's (1996) theoretical consideration of the allelic effects at a single locus under two different types of genetic backgrounds (conspecific and hybrid) is an important first-step in the formation of this conceptual foundation. Yet this study, too, is an oversimplification as it considered only single-locus changes in one species. We need to know the expected signature of different processes that operate upon the conspecific phenotype that would appear by observation of the hybrid phenotype. Computer simulations of the divergence of genetic pathways in allopatric populations provide one promising approach to address this issue (N. A. Johnson and A. H. Porter 1999, unpublished data). Acknowledgments My thoughts about the role of gene interactions in evolutionary biology and speciation, in general, have been shaped to a great extent from my countless discussions with Mike Palopoli, Patrick Phillips, Mike Wade, and Chung-I. Wu. For this, I am very thankful. Discussions with Mohamed Noor about reinforcement have been also very useful. Sally Otto, Patrick Phillips, Adam Porter, the editors, and anonymous reviewers provided many helpful suggestions that improved the readability of this chapter. Epistasis as a Genetic Constraint within Populations and an Accelerant of Adaptive Divergence amongThem MICHAEL J . W A D E Genetic Architecture: Complexity, Additivity, and Epistasis The "genetic architecture" of a phenotype is its characterization in terms of the direct-effects of genes and environments, as well as the suite of genetic and environmental interactions that affect it. In principle, with controlled breeding experiments, genes can be identified by patterns of Mendelian segregation associated with function or phenotypic value, and then mapped to a position in a genome relative to other segregating factors. Associations between genotypic and phenotypic value can be decomposed (Table 13.1) into. additive (a and b), dominance (DAand DB), additive-by-dominance (KAaBand KABb),dominance-byepistatic effects. Environmental dominance (JAaBb),and additive-by-additive (IAB) factors generally cannot be so easily defined, distinguished from one another, or mapped colinearly with respect to other environmental factors. Major environmental factors that affect phenotypes in nature can sometimes be identified by the use of geographic or temporal correlations, but, in the absence of experimental controls, it is rare that a single environmental factor can be causally isolated with respect to its phenotypic effects. After an environmental factor has been identified as being important, its distribution in nature as experienced by the organism is rarely known, so that most experimental studies must study environmental factors as fixed, rather than random, effects. The characterization of genotype-byenvironment interactions (G x E) suffers from many of these same limitations. In this chapter, I will discuss the relationship between gene-by-gene (g x g or intragenornic epistatic), genotype-by-genotype (G x G), and the more familiar genotype-by-environment interactions (G x E). (In keeping with the usage elsewhere