The sterility and inviability of species hybrids is thought to evolve by the accumulation of genes that cause generally recessive, incompatible epistatic interactions between species. Most analyses of the loci involved in such hybrid incompatibilities have suffered from low genetic resolution. Here I present a fine-resolution genetic screen that allows systematic counting, mapping, and characterizing of a large number of hybrid incompatibility loci in a model genetic system. Using small autosomal deletions from D. melanogaster and a hybrid rescue mutation from D. simulans, I measured the viability of hybrid males that are simultaneously hemizygous for a small region of the D. simulans autosomal genome and hemizygous for the D. melanogaster X chromosome. These hybrid males are exposed to the full effects of any recessive-recessive epistatic incompatibilities present in these regions. A screen of ∼70% of the D. simulans autosomal genome reveals 20 hybrid-lethal and 20 hybrid-semilethal regions that are incompatible with the D. melanogaster X. In further crosses, I confirm the epistatic nature of hybrid lethality by showing that all of the incompatibilities are rescued when the D. melanogaster X is replaced with a D. simulans X. Combined with information from previous studies, these results show that the number of recessive incompatibilities is approximately eightfold larger than the number of dominant ones. Finally, I estimate that a total of ∼191 hybrid-lethal incompatibilities separate D. melanogaster and D. simulans, indicating extensive functional divergence between these species’ genomes.
THE last decade has seen important progress in the genetics of speciation. In particular, there is now broad agreement on three aspects of intrinsic postzygotic isolation, the sterility and inviability of species hybrids:
Hybrid fitness problems evolve gradually as incompatible epistatic interactions accumulate between species (Coyne and Orr 1989, 1997; Sasaet al. 1998; Presgraves 2002; Price and Bouvier 2002). The so-called Dobzhansky-Muller model posits that incompatibilities evolve by the substitution of advantageous or neutral mutations in one species, which, having never been tested in combination with those in other species, have harmful effects when brought together in hybrids (Dobzhansky 1937; Muller 1940, 1942; Orr 1995, 1996).
The alleles involved in these hybrid incompatibilities are thought to be, on average, partially recessive. This “dominance theory” neatly accounts for several phenomena, including Haldane’s rule (the observation that XY hybrids typically suffer more severe hybrid problems than do XX hybrids; Haldane 1922) and F2 hybrid breakdown (Muller 1942; Orr 1993; Turelli and Orr 1995, 2000).
Incompatibilities causing hybrid male sterility accumulate faster than those causing other types of hybrid fitness problems (Wu and Davis 1993; Hollocher and Wu 1996; Trueet al. 1996; Wuet al. 1996; Naveira and Maside 1998; Presgraves and Orr 1998; Wu and Hollocher 1998; Singh 1999). The “faster-male” theory posits that the rapid evolution of hybrid male sterility is caused by the faster divergence of male-specific fertility genes, perhaps driven by sexual selection, or by the inherent sensitivity of spermatogenesis to the genetic perturbations experienced by hybrids (Wu and Davis 1993).
Nevertheless, progress in the fine-genetic and molecular characterization of hybrid incompatibility loci has been slow. Despite much effort, few studies have succeeded in precisely and systematically counting, fine mapping, and, ultimately, identifying a large number of “speciation genes.” The reason is that the traits of interest, hybrid sterility and inviability, are by their very nature barriers to crossing and thus are refractory to standard genetic approaches. This problem is especially severe in one of our best model organisms, the fruit fly Drosophila melanogaster: All hybrids between D. melanogaster and its closest known relatives (D. simulans, D. mauritiana, and D. sechellia) are completely dead or sterile (Sturtevant 1920, 1929; Lachaiseet al. 1986). Consequently, nearly a century’s worth of accumulated genetic tools has not been brought to bear on the genetics of speciation, and genetic analyses of the D. melanogaster-D. simulans hybridization, despite an 80-year history, have suffered poor genetic resolution. The rough portrait that has emerged from this work is that a large number of hybrid male steriles but a modest number of hybrid lethals have accumulated between D. melanogaster and D. simulans (Muller and Pontecorvo 1940, 1942; Pontecorvo 1943a,b; Provine 1991; Sanchezet al. 1994; Coyneet al. 1998; Sawamuraet al. 2000; Sawamura 2000).
Two methods have been used to circumvent the problematic sterility and inviability of D. melanogaster species hybrids. First, the discovery of hybrid rescue mutations—alleles that restore the viability and fertility of normally unfit hybrids—raised the possibility of introgressing foreign genes into a D. melanogaster background, where they might be further analyzed (Daviset al. 1996). Unfortunately, hybrid female fertility rescue has proved extremely weak, and so far only two small introgressions of the D. simulans second chromosome have been studied (Sawamuraet al. 2000). Second, Coyne et al. (1998) used a battery of deficiencies (small chromosomal deletions) from D. melanogaster to uncover recessive D. simulans factors causing hybrid lethality in otherwise heterozygous F1 females. Their screen of ∼50% of the D. simulans genome revealed only a handful of hybrid lethals, none of which were unconditionally lethal. One reason why so few were detected may be that the screen was limited to incompatibilities involving a hemizygous (recessive) factor from D. simulans and a heterozygous (dominant) one from D. melanogaster (elsewhere in the genome). If most incompatibility alleles are recessive, as the dominance theory posits, then the most abundant class of hybrid incompatibility should be that in which the interacting loci are both homozygous or both hemizygous, i.e., genotypes normally produced only in F2 (or backcross) hybrids.
Here I present a new screen that modifies and combines these two approaches, taking advantage of both a hybrid rescue mutation from D. simulans and the genetic tools of D. melanogaster to test for the presence of these extreme recessive-recessive hybrid incompatibilities. In particular, I test the viability of F1 hybrid males hemizygous for a small region of the D. simulans autosomal genome and simultaneously hemizygous for the D. melanogaster X. These males are exposed to the full effects of any recessive-recessive X-autosome hybrid incompatibilities in the regions tested. Using the many deficiencies available in D. melanogaster, I systematically screened most of the D. simulans autosomal genome for such incompatibilities.
Deficiency mapping in hybrids offers several advantages over recombination mapping. First, large numbers of the relevant genotypes can be assayed in the F1 generation. Second, the existence and precise location of hybrid lethals in most cases can be easily confirmed using overlapping but independently derived deficiencies with defined cytological breakpoints on the polytene chromosome map (Bridges 1935). Third, the fitness effects of hybrid incompatibilities can be assessed without being confounded with recombination distance between the relevant locus and a genetic marker. Finally, and perhaps most important, the locations of hybrid lethals can be quickly narrowed to such small chromosome regions that the next step—identifying the particular genes causing hybrid lethality—becomes routine.
The present large-scale, fine-resolution screen reveals the existence of many new hybrid lethals between D. melanogaster and D. simulans. In later work, these hybrid lethals will be the subject of molecular study. Here, I use these data to address the following questions:
How many hybrid lethals separate D. melanogaster and D. simulans? This number is a proxy for functional divergence between the viability-essential components of the two species’ genomes.
Where are the hybrid lethals? Are they randomly distributed throughout the genome or clustered in particular regions?
Is hybrid lethality caused by epistasis between incompatible loci, as predicted by the Dobzhansky-Muller model?
Are most hybrid incompatibilities recessive, as predicted by the dominance theory? There are few direct tests of the dominance of incompatible alleles. Furthermore, comparing the number of recessive hybrid lethals discovered here with the number of dominant ones from other studies provides a quantitative test of the dominance theory.
What is the probability that any two divergent substitutions, one from D. melanogaster and one from D. simulans, are incompatible, causing hybrid lethality?
What is the distribution of fitness effects of hybrid-lethal incompatibilities? Do most have weak or strong effects on viability? How does this distribution compare with that for hybrid male sterility, which often appears to have a polygenic basis (Naveira and Maside 1998; Wu and Hollocher 1998)?
At what stage of development do hybrid lethals act? If genes acting early in development are more evolutionarily constrained than later-acting ones, and thus less diverged, then embryonic hybrid lethals should be rarer than postembryonic ones.
MATERIALS AND METHODS
A screen for X-autosome incompatibilities: The screening method is diagrammed in Figure 1A and represents a modification of the crossing design of Coyne et al. (1998). Using many D. melanogaster (mel) stocks, each heterozygous for a deficiency chromosome with known cytological breakpoints and a dominantly marked balancer chromosome (Lindsley and Zimm 1992), I crossed mel Deficiency/Balancer (Df/Bal) females to D. simulans (sim) males carrying the hybrid rescue mutation, Lethal hybrid rescue (Lhr). Lhr rescues normally dead F1 males from lethality that typically occurs at the larval-pupal transition (Watanabe 1979; Takamura and Watanabe 1980; Sawamuraet al. 1993). Half of the F1 hybrid progeny inherit the deficiency and half inherit the balancer. If hybrids of both sexes appear with both Df and Bal genotypes in roughly equal proportions, then no recessive hybrid lethal resides in the sim region uncovered by the deficiency. If, on the other hand, only Df hybrid males die, then a hybrid lethal resides in the sim autosomal region exposed by the deficiency. The fact that only Df hybrid males die, while their Df hybrid sisters do not, suggests either that the sim hybrid lethal is incompatible with a mostly recessive factor on the mel X (females are heterozygous for the X) or that the sim hybrid lethal is male specific. These possibilities are distinguished using follow-up crosses described below.
Table 1 shows the 193 deficiencies used. Together these deficiencies uncover ∼69-77% of 2L (chromosome 2, left arm), 49-59% of 2R, 72-77% of 3L, and 65-79% of 3R. I did not test the small-dot fourth chromosome of D. simulans as it is known to carry no hybrid lethals (Muller and Pontecorvo 1940, 1942; Pontecorvo 1943a; Orr 1992). All stocks are available through the Bloomington or Umeå Stock Centers (Bloomington Stock Center, http://flystocks.bio.edu/df-kit-infor.htm). Soichi Tanda kindly provided a stock with a newly extracted Df(2L)al. The most common dominant markers on chromosome 2 balancers were CurlyO and Glazed, and those on chromosome 3 balancers were Stubble, Serrate, Tubby, or a combination. When possible, deficiency stocks with a chromosome 3 balancer marked only with Tubby or Ultrabithorax were rebalanced over a different chromosome with a more easily scored marker (e.g., Stubble). Complete descriptions of all stocks, including the particular balancers used, are available upon request.
Each species cross was made by mass mating 15-20 mel Df/Bal females with 15-25 sim Lhr males. All crosses were done at 24° and flies were reared on standard cornmeal-yeast-agar medium. For each deficiency, I repeated the cross at least four times and report the pooled number of progeny. Contamination was detectable in several ways: consistency among replicate crosses, the segregation of dominant markers among progeny, the presence of diagnostic hybrid male genitalia, and the absolute sterility of all progeny. Although the cross shown in Figure 1A is in principle straightforward, some stocks showed strong sexual isolation, requiring many replicate crosses to obtain sufficient progeny.
Most crosses deviated from the expected 1:1:1:1 segregation ratio of Df females, Bal females, Df males, and Bal males, because Lhr rescue of hybrid males is incomplete. Hybrid males were significantly rarer than females in most crosses, with an average of 45% rescue. In a few crosses, no hybrid males were rescued (see also Granadinoet al. 1996); since these crosses are not informative, they are not reported here. For crosses that produced hybrid females and at least 10 Bal hybrid males, I estimated the relative viability of the deficiency by simply tabulating the ratio of Df:Bal progeny for each hybrid sex. I used χ2 tests to detect deviations from the expected 1:1 ratio of Df:Bal progeny within each sex and Fisher’s exact tests to detect differences in the ratio of Df:Bal progeny between sexes. I defined candidate X-autosome hybrid incompatibilities as those causing a significant deficit of Df hybrid males and a significantly smaller Df:Bal ratio in hybrid males than in hybrid females. I then classified hybrid incompatibilities by viability following the scheme of Crow and colleagues for deleterious mutations within species (reviewed in Simmons and Crow 1977; Crow and Simmons 1983): For regions causing significant reductions in Df hybrid male viability, those with viabilities ≤50% (i.e., a Df:Bal ratio <0.50 at least for one of two overlapping deficiencies) were considered hybrid “semilethals,” and those with viabilities of ≤10% were considered hybrid “lethals.”
Further investigation of hybrid-lethal deficiencies: I performed three further tests for most of the regions that caused hybrid lethality. First, to ensure that the lethality of deficiencies is hybrid specific, I confirmed the viability of deficiencies in nonhybrids by crossing Df/Bal mel females to wild type, Wolbachia-free Canton-S mel males. (Note, however, that the very existence of the deficiency stocks shows that they are not lethal when heterozygous within species.)
Second, when possible, I confirmed the presence of each hybrid lethal by testing at least one other overlapping deficiency. Such confirmation with multiple, independently derived deficiencies rules out the possibility that a putative hybrid lethal is an artifact of the genetic background of any particular stock. Those regions shown to be lethal by a single deficiency, but which could not be confirmed because overlapping deficiencies were unavailable, should be viewed as candidate hybrid-lethal regions.
Third, I tested whether the exposed sim factors caused lethality by epistasis with the mel X. To do this, I constructed mel attached-X stocks, each carrying a hybrid-lethal deficiency over a balancer, and crossed females from these stocks to sim Lhr males. As Figure 1B shows, the attached-X chromosome forces mother-to-daughter inheritance of the mel attached-X and father-to-son inheritance of the sim free-X. Hybrid males from this cross thus inherit a sim X rather than a mel X, while the rest of their genotype, including their cytoplasm, remains identical in species origin to the hybrid males from the original cross (compare F1 males in Figure 1, A and B). The important point is that if Df hybrid male lethality is caused by an incompatibility between a recessive sim autosomal factor and a recessive factor(s) on the mel X, then replacing the mel X with a sim X should rescue these males (as the sim X should be compatible with the sim autosomal factor). If, on the other hand, hybrid male lethality is caused by a male-specific incompatibility, replacing the X chromosome will not necessarily rescue hybrid males. The reason is that a male-specific incompatibility could involve an interaction between a recessive sim factor and a dominant mel factor anywhere in the D. melanogaster genome. Unfortunately, mel attached-X hybrid females from this cross, which would also have been informative, were not rescued in sufficient numbers (see results).
Lethal phase: For each incompatibility, I determined if hybrid lethality occurred at embryonic or postembryonic stages. If Df hybrid males suffer embryonic lethality, 25% of all embryos should die (i.e., 50% of all male embryos). To test if embryonic lethality was sufficient to account for the absent Df hybrid males, I calculated the ratio of the number of dead (necrotized) embryos divided by half the number of hybrid females: dead embryos/[(Df females + Bal females)/2]. If this ratio was ≥1, the deficiency was scored as embryonic lethal. If the ratio was ⊡1, the number of dead embryos could not plausibly explain the deficit of Df hybrid males, and the incompatibility was scored as postembryonic lethal. Establishing the exact phase of lethality for postembryonic hybrid lethals is, unfortunately, complicated by background mortality and the imperfect degree of overall hybrid male rescue by Lhr.
Fine-mapping hybrid lethals: Table 1 gives the results for all tested deficiencies. The distributions of viability effects for the two sexes are shown in Figure 2. The most striking result is that while very few deficiencies uncover lethality in hybrid females, many uncover lethality in hybrid males (Figure 2): The distribution of viability for Df hybrid females is unimodal, with most crosses showing normal viability, i.e., roughly equal representation of Df and Bal females. In contrast, the distribution of viability for Df hybrid males is bimodal, with one peak at normal viability and a second near complete lethality. Rare “escaper” males bearing hybrid-lethal deficiencies are typically weak and sometimes show developmental anomalies (e.g., malformed abdominal tergites, bristles, wings, and eyes). In all, 40 nonoverlapping autosomal regions in D. simulans cause significant lethality when uncovered in Df hybrid males. Recessive hybrid lethals are thus common in the sim autosomal genome.
Of these regions, 20 are hybrid-lethal and 20 are hybrid-semilethal incompatibilities (Table 1). These regions appear to be distributed randomly among the two major autosomes (χ2 = 3.333, d.f. = 1, P = 0.068) and the four autosomal arms, with 14 on 2L, 14 on 2R, 6 on 3L, and 6 on 3R (χ2 = 3.195, d.f. = 3, P = 0.363; Figure 3). I performed further analyses on 20 hybrid lethals and 3 hybrid semilethals (these semilethals were nearly classifiable as lethal; i.e., the ratio of Df:Bal hybrid males was ≤0.12 for all 3). The lethal effects of these deficiencies are hybrid specific because none caused strong haploinsufficiency when heterozygous in pure species individuals (Table 2). Of the 23 hybrid lethals and semilethals, I confirmed 18 (78%) with overlapping deficiencies that yielded consistent results (Table 1). Each of these regions thus contains at least one hybrid incompatibility factor.
It is important to note, however, that more hybrid lethals probably exist than the above numbers suggest because the available deficiencies from D. melanogaster cumulatively uncover only 64-73% of the autosomal genome. Correcting for this incomplete coverage yields an estimate closer to ∼27 hybrid lethals and ∼27 hybrid semilethals. These remain minimum estimates as each region could harbor more than one hybrid lethal. However, the fact that most deficiencies fail to uncover hybrid lethality suggests that hybrid lethals are sparsely distributed. Therefore, either most hybrid-lethal regions harbor a single hybrid-lethal locus or hybrid lethals are tightly clustered. If the former is true, as seems most plausible, it follows that many incompatibilities have major effects on hybrid viability.
The recessivity of the incompatible factors on the mel X can be illustrated in three ways. First, if factors on the mel X were completely dominant (and the incompatibilities are not sex specific), Df hybrid female viability would be correlated with Df hybrid male viability, with slope ≈1. (The slope of this relationship can be thought of as a proxy for the mean dominance coefficient of incompatible factors on the mel X.) Instead, while there is a weak positive correlation between Df hybrid female and male viability, the slope of the least-squares relationship is b = 0.103—a value ⊡1.0—consistent with the general recessivity of incompatible factors on the mel X (Figure 4; Spearman’s r = 0.326, d.f. = 163, P < 0.0001; statistics use only crosses producing ≥10 Bal hybrid males). Second, of 16 regions that cause significant lethality in Df hybrid females, 9 cause significantly stronger lethality in Df hybrid males (e.g., Table 1, line 4). Third, and most convincing, 30 of 40 deficiencies that caused significant lethality in Df hybrid males have no discernable effects in Df hybrid females. These findings strongly suggest that the incompatible factors on the mel X are also overwhelmingly close to completely recessive.
These results are consistent with those of Yagyu and Yamamoto (1996), who independently performed similar crosses using 15 deficiencies on chromosome 2.
Are hybrid lethals epistatic? The hybrid-lethal and semilethal sim autosomal factors detected above could be either incompatible with factors on the mel X or male-specific and incompatible with a dominant mel factor anywhere in the genome. To distinguish these possibilities, I used attached-X crosses shown in Figure 1B to test the viability of Df hybrid males that are genotypically and cytoplasmically identical to those above but possess a sim X rather than a mel X. These hybrid males are hemizygous for the sim X and hemizygous for a hybridlethal sim autosomal factor. If the lethal sim autosomal factors are incompatible with the mel X, changing the species origin of the X should rescue hybrid lethality. If, on the other hand, the sim autosomal factors cause male-specific hybrid lethality, changing the species origin of the X may not rescue hybrid lethality.
As Table 3 shows, 18 of the 18 hybrid-lethal deficiencies tested become viable when hybrid males are given a sim X instead of a mel X. (For two hybrid-lethal deficiencies, the attached-X stocks proved too weak to maintain and could not be tested.) Thus, the hybrid lethality of 18 of 18 sim autosomal regions depends on the species origin of the X, confirming that these are true hybrid lethals caused by incompatible epistatic interactions.
These attached-X crosses were also expected to produce hybrid females homozygous for the mel X (and thus genotypically identical to hybrid males from the original screen; compare female hybrids in Figure 1B to male hybrids in Figure 1A). If produced in sufficient number, we could further test the sex specificity of each hybrid incompatibility: If hybrid lethals are not sex specific, Df hybrid females from attached-X crosses should also be inviable. This outcome appears to hold in three cases (Table 3, lines 2, 4, and 13), but in general I obtained too few hybrid females to draw meaningful conclusions.
Lethal phase: I determined the lethal phase of 18 hybrid incompatibilities. Only one causes embryonic lethality (see Table 4). I compared this distribution to that for lethal phases of mutations within species. In D. melanogaster, 60.5% of lethal P insertions cause embryonic lethality (Toroket al. 1993; Deaket al. 1997). There are thus significantly fewer embryonic lethal incompatibilities in hybrids than expected from the within-species data (1 embryonic vs. 17 postembryonic hybrid-lethal incompatibilities; 2515 embryonic vs. 1640 postembryonic within-species lethal mutations; Fisher’s exact P < 0.0001). Thus hybrid-lethal incompatibilities more often afflict later, postembryonic stages of development.
This study yields three main results. First (and perhaps most surprising), many hybrid-lethal incompatibilities have evolved between D. melanogaster and D. simulans: The ∼70% of the D. simulans autosomal genome examined harbors at least 20 factors that cause complete or near complete lethality in the presence of the D. melanogaster X. Taking into account certain corrections (see below), this value implies that the genome-wide number of recessive-recessive lethal incompatibilities is ∼169. Second, the incompatible genes on both the autosomes and the X are nearly completely recessive and, as shown below, these recessive incompatibilities vastly outnumber dominant ones. These results provide strong support for the dominance theory. Third, all of the lethal hybrid incompatibilities tested involve an epistatic interaction with the X chromosome; i.e., 18 of 18 are rescued when the mel X is replaced by a compatible sim X, as expected under the the Dobzhansky-Muller model.
While these data confirm the ubiquity of recessive, epistatic hybrid incompatibilities, they also allow us to go further. These data can be used to estimate four quantities: (1) the total number of hybrid lethals separating D. melanogaster and D. simulans; (2) the relative rates of evolution of dominant vs. recessive incompatibilities; (3) the probability that two randomly chosen divergent substitutions (one from each species) are incompatible, causing hybrid lethality; and (4) the fraction of viability-essential genes that have diverged to such an extent that they are no longer functionally compatible with alleles at interacting loci from the other species.
Dominance and the total number of hybrid lethals: By including data from previous genetic analyses, we can estimate the total number of hybrid lethals that have evolved between D. melanogaster and D. simulans. I make the simplifying assumption that hybrid lethality involves pairs of incompatible loci. Hybrid incompatibilities can then be classified into dominant-dominant, recessive-dominant, and recessive-recessive types (or, following Turelli and Orr 2000, H0, H1, and H2 incompatibilities, respectively, where the subscripts 0, 1, and 2 indicate the number of homozygous or hemizygous loci involved). More than two loci could be involved in a given incompatibility—so-called complex epistasis—but this should not seriously affect our estimates as long as the three incompatibility types (H0, H1, and H2) are equally prone to such complex interactions.
H0 incompatibilities: No H0 incompatibilities separate D. melanogaster and D. simulans. The cross of D. melanogaster females with D. simulans males produces perfectly viable hybrid females that are heterozygous at every locus in the genome (Sturtevant 1920, 1929; but see Barbashet al. 2000 for temperature effects on viability of hybrid females).
H1 incompatibilities: There are ∼22 H1 incompatibilities between D. melanogaster and D. simulans. This value comes from two kinds of data. The first involves studies of hybrid rescue mutations. There is now good evidence that such mutations are rare “compatible” alleles at normally incompatible loci (Hutteret al. 1990; Sawamura and Yamamoto 1997; Barbashet al. 2000; Orr and Irving 2000). We can therefore infer two H1 incompatibilities from the existence of two known pairs of complementary hybrid rescue mutations (Hutteret al. 1990; Sawamuraet al. 1993; Orr and Presgraves 2000). The second comes from Coyne et al.’s (1998) deficiency screen for H1 incompatibilities. This screen uncovered five hybrid-lethal regions (this lethality was not unconditional, however, as some became viable at permissive temperatures or using different stocks). Coyne et al.’s number requires two corrections as only 50% of the D. simulans genome was screened, and the reciprocal experiment in which D. melanogaster regions are made hemizygous could not be done as no deficiencies are available in D. simulans. Correcting for these two considerations yields an estimate of ∼20 hybrid-lethal regions. Thus ∼22 H1 incompatibilities separate D. melanogaster and D. simulans. This number is obviously rough, but as shown below, it differs qualitatively from that for H2 incompatibilities.
H2 incompatibilities: The total number of H2 incompatibilities is large. We can estimate the genome-wide number by extrapolating from the number of X-autosome H2 incompatibilities (see results). The X is roughly equivalent in size and gene content to one of the five major chromosome arms (i.e., X, 2L, 2R, 3L, and 3R). Holding the X effectively homozygous (hemizygous) and screening the rest of the genome with deficiencies yielded ∼27 lethal H2 incompatibilities. If we could repeat this screen, successively holding the autosomal arms 2L, 2R, 3L, and 3R homozygous for D. melanogaster while scanning the rest of the D. simulans genome for lethal incompatibilities, we would expect to uncover another ∼27 H2 incompatibilities for each arm. We therefore expect ∼135 hybrid lethals genome-wide. Using similar logic to correct for intra-arm H2 incompatibilities brings the total number of H2 incompatibilities to ∼169. (Note that extrapolating from the number of X-autosome incompatibilities to the number genome-wide assumes that X-linked and autosomal loci diverge at similar rates; Betancourt et al.’s (2002) survey of divergence at >250 genes from D. melanogaster and D. simulans supports this assumption.)
Two conclusions follow from these calculations. First, summing across H0, H1, and H2 incompatibility types gives an estimate of the total number of hybrid-lethal incompatibilities separating D. melanogaster and D. simulans: 191. (This does not include the hybrid semilethals found here; including them nearly doubles the estimate.) Given the conclusion from previous analyses that the number of hybrid lethals between these two species is small, this number comes as a surprise and reveals an unexpected degree of functional divergence.
These numbers also provide strong quantitative confirmation of the dominance theory: D. melanogaster and D. simulans are separated by no H0 incompatibilities, by ∼22 H1 incompatibilities, and by ∼169 H2 incompatibilities. The number of hybrid lethals thus jumps nearly an order of magnitude as we increase the number of homozygous loci involved, leaving little doubt that most hybrid lethals are recessive. Why hybrid incompatibilities tend to be recessive remains a mystery. Although the similar effects of loss-of-function mutations within species and incompatibilities in hybrids have led to speculation that the latter mimic the former (Stebbins 1958; Orr 1993; Turelli and Orr 1995), several recent lines of evidence now appear inconsistent with this interpretation (Barbashet al. 2000; Orr and Irving 2000; Orr and Presgraves 2000). But regardless of why most incompatibilities act as recessives, the present results leave little doubt that they do.
Epistasis and hybrid lethality: Using the above estimate of the total number of hybrid-lethal incompatibilities, along with a molecular estimate of the total number of divergent substitutions between D. melanogaster and D. simulans, we can estimate (to an order of magnitude) another important quantity from speciation genetics theory: the probability, p, that any two randomly chosen divergent sites (one from one species, one from the other) are incompatible in hybrids, causing (for our purposes) complete lethality. As Orr and Turelli (2001) show, the expected number of hybrid incompatibilities, I, between pairs of genes is where k is the genome-wide rate of substitution and t is the time since speciation, so that 2kt is the total number of substitutions separating the genomes of two species. By substituting estimates of I and kt and rearranging, we can solve for p. To ensure that our estimate of p for hybrid lethality is directly comparable to the estimate of p for hybrid male sterility calculated by Orr and Turelli (2001), I use the same data sources for k and t and follow their calculation exactly. (See appendix for fuller details and discussion of the calculation.) The total number of hybrid lethals is, from above, I ≈ 191. D. melanogaster and D. simulans have accumulated 2kt ≈ 156,000 nonsynonymous substitutions since they diverged t = 2.5 MYA (Hey and Kliman 1993; Li 1997). Solving the above equation thus gives p ≈ 1.6 × 10-8. Two nonsynonymous substitutions chosen randomly (one from each species’ genome) will therefore cause complete hybrid lethality ∼10-8 of the time. As discussed in the appendix, this value appears to be an order of magnitude smaller than the value for hybrid male sterility estimated by Orr and Turelli (2001).
Interestingly, p can be thought of as a crude index of the ruggedness of the molecular landscape. To see this, consider the extreme cases: When p = 0, negative epistasis is absent, all substitutions are compatible, and hybrid genotypes never fall into fitness valleys; when p = 1, however, negative epistasis is complete, all divergent substitutions after the first are incompatible, and all hybrid genotypes fall into fitness valleys. Thus, the exceedingly small value of p suggests that the molecular landscape is reasonably smooth: Substitutions that have never “seen” each other in their evolutionary histories are almost always compatible (or at least not lethal in combination).
Functional divergence: Finally, we can estimate the fraction of viability-essential genes that have diverged to the extent that they are no longer functionally compatible. If hybrid-lethal incompatibilities involve pairs of loci, as assumed above, it follows from the Dobzhansky-Muller model that both loci must have diverged (Orr 1995). The above estimate of 191 hybrid-lethal incompatibilities thus implies that at least 382 loci have experienced significant functional divergence since the D. melanogaster-D. simulans split, i.e., ∼11% of all viability-essential genes (or 382/3600; see Spradlinget al. 1999). This level of divergence at a class of genes that we might a priori expect to be relatively conserved (i.e., those essential for viability) seems remarkable.
Two lines of evidence suggest that the genes regulating the earliest phases of development in the two species remain largely compatible. First, most hybrid genotypes survive embryonic phases of development only to die later (Carvajalet al. 1996). Second, my deficiency screen, as well as that of Coyne et al. (1998), could have detected any maternal-zygotic, embryonic lethal incompatibilities involving the D. melanogaster cytoplasm and hemizygous D. simulans autosomal factors. But very few of our incompatibilities can possibly fall into this category. Instead, most hybrid incompatibilities occur between zygotic genes acting at postembryonic stages. Since viability-essential genes are most often mutable to embryonic lethality within species, the paucity of embryonic vs. postembryonic hybrid lethals suggests that: (i) There are greater functional constraints (on gene sequence and/or expression) at early acting genes; (ii) most physiological and ecological adaptation in Drosophila occurs by divergence in postembryonic phases of development; or (iii) more genes are simultaneously active and thus prone to incompatible interactions during later stages of development. The latter possibility seems unlikely to account for the strong preponderance of postembryonic hybrid lethality as most genes exhibiting developmentally modulated expression during the Drosophila life cycle are expressed at some point during embryogenesis (>88%; Arbeitmanet al. 2002).
Caveats: Using deficiencies and a hybrid rescue mutation to detect X-autosome incompatibilities makes two key assumptions. The first is that the effects of D. simulans autosomal regions when hemizygous are equivalent to when they are homozygous. If this assumption is incorrect, then instead of uncovering “hybrid lethals,” the screen might simply uncover regions that are haploinsufficient. (Hemizygosity of the D. melanogaster X in males is equivalent to homozygosity because of dosage compensation.) Three facts, however, militate against this possibility (see also Coyneet al. 1998). First, although deficiencies can uncover recessive lethality within species, they are not inherently lethal as heterozygotes within species (Table 2).
Second, one might argue that deficiencies cause haploinsufficiency, but only in hybrids. Two observations rule out this possibility: deficiencies that cause lethality in hybrid males do not, in most cases, harm Df hybrid females; more important, even hybrid males that carry a hybrid-lethal deficiency become viable when carrying a D. simulans, rather than a D. melanogaster, X. Thus, the lethality of particular deficiencies depends on species genotype at background loci (i.e., epistasis), as expected for hybrid incompatibilities.
Third, there are now two examples in which factors from D. simulans are known to behave identically in species hybrids, whether homozygous or hemizygous:
Muller and Pontecorvo (1940, 1942) discovered that the fourth chromosome of D. simulans causes hybrid male sterility when homozygous on an otherwise D. melanogaster genetic background. Orr (1992) fine mapped the region of the D. simulans fourth responsible using D. melanogaster deficiencies, thus confirming that the recessive, incompatible D. simulans factor causes sterility when homozygous or hemizygous.
Sawamura et al. (2000) used a mutation that weakly rescues hybrid female fertility to introgress two small regions from D. simulans 2L into an otherwise D. melanogaster background. According to my results, these regions harbor three factors that each cause hybrid lethality when hemizygous in a genetic background containing a hemizygous D. melanogaster X and heterozygous autosomes (Table 4, lines 1, 3, 4). Sawamura (2000) found that, in an identical genetic background, these regions also cause hybrid lethality when homozygous.
Thus, both Orr’s and Sawamura’s results show that recessive hybrid incompatibilities behave identically when homozygous or hemizygous.
The second assumption of the deficiency analysis is that the 40 hybrid incompatibilities detected are independent of hybrid male rescue by Lhr. The primary cross (Figure 1A) was intended to rescue hybrid males from one incompatibility (involving the incompatible, wild-type allele Lhrsim) and to then expose them to other potential incompatibilities. It is formally possible, however, that some of these hybrid lethals are actually suppressors of Lhr rescue. If true, these suppressors would still be of interest as interactors of a known hybrid incompatibility locus (Lhr). But this Lhr-suppressor scenario seems unlikely for several reasons. First, it seems unlikely that since the D. melanogaster-D. simulans split ∼2.5 MYA, the only genetic pathways to evolve incompatibilities necessarily all involve Lhr. Second, such putative Lhr suppressors would necessarily be recessive D. simulans factors that suppress the rescue effects of a particular rare allele, Lhr. It is hard to believe that so many different recessive, conspecific loci are capable of such a trick. Third, Lhr rescues hybrid male lethality that normally occurs at the larval-pupal transition. But at least one hybrid lethal does not affect this phase of development, causing embryonic lethality instead (Table 4, line 7), and thus cannot be explained by Lhr suppression. Casual observations further suggest that more detailed study of the postembryonic lethal phases will reveal other cases acting later than the larval-pupal transition.
Conclusions: This study shows that most hybrid incompatibilities are recessive and epistatic. The most surprising finding is the extraordinary degree of functional divergence that has occurred between D. melanogaster and D. simulans during the last 2.5 MY. Such extensive cryptic divergence would seem to confirm Muller’s suggestion that “(t)wo groups of organisms which are not ordinarily allowed to cross with one another will thus automatically become increasingly immiscible, and their genic, chemical paths of evolution will diverge more and more... even in cases where their evolution is, from the phenotypic standpoint, strikingly parallel” (Muller 1939, p. 278).
An important aspect of the many hybrid incompatibilities identified here should not be overlooked: Their fine-scale resolution in a model genetic organism will greatly facilitate their routine molecular characterization. The 20 hybrid lethals and 20 hybrid semilethals discovered each reside in just a few cytological divisions. It is conceptually simple (although labor intensive) to move from identifying blocks of candidate genes using deficiency complementation tests to identifying the relevant genes using single-locus complementation tests. Establishing the molecular identity, function, and evolutionary history of a large collection of speciation genes is certain to reveal new patterns and to answer some long-standing questions in evolution: What are the normal functions of “speciation genes” within species? Are most functionally relevant substitutions concentrated in coding or regulatory regions? Is natural selection the primary force causing the fixation of divergent substitutions? Preliminary work in several of the hybrid-lethal regions identified here suggests that the molecular characterization of the genes responsible should be possible.
APPENDIX: CALCULATION OF p
To obtain 2kt, the total number of replacement changes separating D. melanogaster and D. simulans, I use Li’s (1997) estimate of the rate of nonsynonymous substitution in Drosophila from 30 loci, 1.91 × 10-9/site/year. To get the genome-wide rate of substitution, I take into account the number of relevant sites in the genome: 13,600 genes/genome × 1800 bp/gene × ∼2/3 are nonsynonymous sites = 1.632 × 107. Multiplying gives the number of replacement changes per genome per year: k = 0.0312. Thus the total number of replacement changes that have accumulated between D. melanogaster and D. simulans during the last 2.5 MY is 2kt ≈ 156,000. Solving the equation gives p ≈ 1.6 × 10-8.
We can check the plausibility of this estimate of p for hybrid lethality using data from another species pair. In particular, given p for hybrid lethality from above and the number of genome-wide substitutions separating a younger species pair, D. mauritiana and D. simulans, we can ask: How many hybrid lethals should we expect to find in a genetic analysis of D. mauritiana-D. simulans hybrids? (This assumes, of course, that p itself has not evolved substantially over the last few million years, as seems reasonable.) We already know the number of hybrid lethals between D. mauritiana and D. simulans to be ∼5-10 from the work of True et al. (1996) and so can compare it to the one predicted by the Orr-Turelli equation. To get the predicted number, we use p = 1.6 × 10-8 from above and the genome-wide number of nonsynonymous substitutions between D. mauritiana and D. simulans, 2kt = 48,000 (see Orr and Turelli 2001). Solving for I, the predicted number of hybrid lethals is 18. Thus, given nothing more than p for hybrid lethality, as estimated from the D. melanogaster-D. simulans hybridization and a crude estimate of the number of substitutions, the equation comes remarkably close to predicting the true number of hybrid lethals.
The estimate of p for hybrid lethality is an order of magnitude smaller than the one estimated for hybrid male sterility using genetic data from D. simulans and D. mauritiana, p ≈ 1.04 × 10-7 (Orr and Turelli 2001). This discrepancy could be explained by either of the two putative causes of faster-male evolution (Wu and Davis 1993; Wuet al. 1996): (i) Spermatogenesis might be inherently more sensitive than viability to the incompatibilities experienced by hybrids, and thus a greater fraction of interspecific gene interactions cause hybrid male sterility; or (ii) our calculations have not taken into account that male-specific genes evolve faster, on average, than viability-essential genes (Begunet al. 2000; Swansonet al. 2001; Betancourtet al. 2002). As defined here and in Orr and Turelli (2001), p—the probability that any two divergent replacement changes (one from each species) from any gene in the genome are incompatible—ignores that not all genes interact with each other, that not all genes are mutable to sterility, that not all genes are mutable to lethality, that not all incompatibilities are protein-protein interactions, etc. In the absence of such perfect knowledge, it is convenient to estimate p as the genome-wide probability of incompatibility, averaging over all genes. But since male-specific genes evolve more rapidly than others (and, in particular, faster than the 30 loci used to estimate k from Li 1997 above), p for hybrid male sterility will be overestimated. Consequently, p for hybrid lethality and p for hybrid sterility are likely closer than they appear. The important point, however, is that both values are exceedingly small.
I thank Shawn Ciecko for expert technical assistance during the early phases of this work. I am also grateful to Andrea Betancourt, Andy Clark, Jerry Coyne, Allen Orr, and Jack Werren for helpful discussions and advice on the project. I also thank Andrea Betancourt, Audrey Chang, Jerry Coyne, J. P. Masly, Mohamed Noor, two anonymous reviewers, and especially Allen Orr for comments that greatly improved the manuscript. This work was supported by funds from an Ernst Caspari Fellowship and a George and Agnes M. Messersmith Fellowship to the author and by funds from the National Institutes of Health (GM-526738) and David and Lucile Packard Foundation to Allen Orr.
Communicating editor: M. A. F. Noor
- Received September 20, 2002.
- Accepted December 3, 2002.
- Copyright © 2003 by the Genetics Society of America