Functional divergence caused by ancient positive selection of a Drosophila hybrid incompatibility locus

Abstract

Interspecific hybrid lethality and sterility are a consequence of divergent evolution between species and serve to maintain the discrete identities of species. The evolution of hybrid incompatibilities has been described in widely accepted models by Dobzhansky and Muller where lineage-specific functional divergence is the essential characteristic of hybrid incompatibility genes. Experimentally tractable models are required to identify and test candidate hybrid incompatibility genes. Several Drosophila melanogaster genes involved in hybrid incompatibility have been identified but none has yet been shown to have functionally diverged in accordance with the Dobzhansky-Muller model. By introducing transgenic copies of the X-linked Hybrid male rescue (Hmr) gene into D. melanogaster from its sibling species D. simulans and D. mauritiana, we demonstrate that Hmr has functionally diverged to cause F1 hybrid incompatibility between these species. Consistent with the Dobzhansky-Muller model, we find that Hmr has diverged extensively in the D. melanogaster lineage, but we also find extensive divergence in the sibling-species lineage. Together, these findings implicate over 13% of the amino acids encoded by Hmr as candidates for causing hybrid incompatibility. The exceptional level of divergence at Hmr cannot be explained by neutral processes because we use phylogenetic methods and population genetic analyses to show that the elevated amino-acid divergence in both lineages is due to positive selection in the distant past-at least one million generations ago. Our findings suggest that multiple substitutions driven by natural selection may be a general phenomenon required to generate hybrid incompatibility alleles.

Figure 1. Models of Hybrid Incompatibility

(A) D-M model of HI evolution. We diagram here an X–autosome incompatibility; for simplicity only haploid genotypes are shown. This model can be easily extended to include more complex multilocus interactions. (1) The ancestral species is fixed for the X-linked allele x and the autosomal allele a. (2) As the two species independently diverge, one becomes fixed for allele X at the first locus and the other for allele A at the second locus. (3) HI is caused by the interaction between these derived alleles, X and A. (4) This interaction may cause misregulation of downstream effector genes (d1, d2, and d3), which in turn causes the HI phenotype. (B) Hmr is an HI gene. Hmr_mel has evolved in the D. melanogaster lineage and interacts to cause HI with A_sib, an allele of a hypothesized autosomal gene that has evolved in the sibling-species lineage. Mutations in Hmr_mel allow hybrid viability by eliminating the activity of this incompatibility allele. (C) Hmr is a downstream effector gene. Here, two unknown genes cause HI by misregulating Hmr. Mutations in Hmr allow hybrid viability by acting as downstream suppressors of the HI alleles. (D) Model with Hmr and gene A extensively diverging (see Discussion). Both Hmr and gene A coevolve with many changes along both lineages. HI could be caused by interactions between derived alleles or between a derived and an ancestral allele. All models to identify the codons in Hmr responsible for functional divergence have two constraints: first, that Hmr and gene A must be fully compatible with each other in each lineage, and second, that candidate codons must differ between Hmr_sib from all three sibling species and Hmr_mel.

Figure 2. Structure and Expression of Sibling-Species Hmr⁺ Transgenes

(A) Diagram of Hmr⁺ transgenic constructs. The Hmr gene structure is shown, with the rightward arrow indicating the predicted translation start site. Sibling-species constructs used in this study are shown, together with D. melanogaster constructs previously shown (Barbash et al. 2003) to be Hmr⁺. (B) Restriction map of RT-PCR fragments spanning part of exons 3 and 4, showing diagnostic restriction site polymorphisms found in the transgenic alleles and the stocks used to assay them. The D. melanogaster map corresponds to both the Hmr¹ and In(1)AB rescue alleles, as well as all D. melanogaster alleles from our population sample. (C) RT-PCR products from interspecific hybrids. Hybrids were from the crosses described in Table 1. RNA was collected from 48- to 72-h-old larvae and 2- to 4-d-old adult males. Note that larval samples contain RNA from males and females, half of whom carry the sibling-species Hmr⁺ transgene. The portion of the PCR product derived from the transgenes is that digested by XbaI or HpaI for the D. simulans and D. mauritiana transgenes, respectively. M, 100-bp ladder marker; G, undigested PCR from an Hmr genomic clone (this product contains a 59-bp intron); cD, undigested PCR from an Hmr cDNA clone. The following are all RT-PCR products: U, undigested; C, ClaI-digested; H, HindIII-digested; X, XbaI-digested; C/X, ClaI- and XbaI-digested; Hp, HpaI-digested; C/Hp, ClaI- and HpaI-digested; –, control containing no reverse transcriptase. Note that undigested lanes (U) contain half the amount of DNA as digested samples.

Figure 3. Maximum-Likelihood Estimates of Hmr Divergence among Drosophila Lineages

Estimates of the number of changes per nonsynonymous site (D_N) are shown above each lineage, and the number of changes per synonymous site (D_S) are shown below each lineage, calculated separately for each branch. The coding region of Hmr is 1,390 to 1,427 amino acids long in the five species. D_N/D_S ratios differ significantly among branches as tested by the methods of Nielsen and Yang (1998). A model where all D_N/D_S ratios were free to vary along all branches (Model 2 [M2]) fit the data better than a model with a fixed D_N/D_S ratio for all branches (M1) (2Δ| = 308, p < 0.0001, chi-square distribution), as did a model where D_N/D_S ratios for the D. melanogaster lineage and the lineage leading to the sibling species differed from the rest of the tree (local clock, 2Δ| = 292, p < 0.0001). This suggests that most of the heterogeneity in the D_N/D_S ratio among branches of the phylogeny is due to an elevated ratio for the lineages leading from the ancestor of D. melanogaster and the sibling species. The tree is unrooted and we assume a trifurcation among the sibling species.

Figure 4. Sliding Window Analysis of Hmr Divergence and Polymorphism

Calculations were made with a window size of 150 nucleotides and a step size of 50 nucleotides. Nucleotide position 1 on the x-axis is the start of the coding sequence, and the last position is the stop codon. The dashed line indicates where the ratio is one. Arrows at the top indicate the positions of codons identified as being under positive selection in Figure 5. Exon boundaries are indicated below the x-axis with horizontal bars. A repeatability analysis (Smith and Hurst 1998) revealed that polymorphism ratios for each window were not correlated (p = 0.43) with divergence ratios between D. melanogaster and D. simulans.

Figure 5. Phylogenetic Analysis of Positive Selection at Individual Codons

Site-specific codon Model 8 (M8) in PAML was used to identify codons under selection. This model, which considers a discrete distribution of D_N/Ds values plus a “selection category”—D_N/D_S values greater than one—fit the data better than a neutral model (M1) (2Δ| = 25.43, p < 0.001). Codons listed are those with p-values from posterior distributions greater than 0.5. The positions of these codons are also shown in Figure 4. Because D. melanogaster is incompatible with all three of its sibling species, we expect that Hmr codons involved in HI must be different between D. melanogaster Hmr and all three sibling-species alleles. Codons that fit a model of incompatibility between Hmr_mel and A_sib are shaded blue, and those that fit a model of incompatibility between Hmr_anc and A_sib are shaded yellow (see Discussion). The five remaining codons (unshaded) are identical between D. melanogaster and at least one of the sibling alleles and are thus excluded from both models.

Figure 6. Posterior Distribution of Time (in Generations) Since the Most Recent Selective Sweep at Hmr for D. melanogaster

Population size (N) is assumed to be 1 × 10⁶ for D. melanogaster. Samples were generated from the joint posterior distribution of five parameters of a selective sweep model assuming a selection event occurred sometime in the past at the Hmr locus, and from three summaries of polymorphism, including the number of segregating sites (54), Tajima's D (−0.61), the population recombination rate (4Nr = 43, where N = population size and r = per gene recombination rate; McVean et al. 2002) and the number of haplotypes (11). The data is least consistent with a selective sweep in the recent past and is most consistent with selective sweeps occurring more than N generations ago. If there are ten generations per year, this suggests that the last selective sweep occurred at least 100,000 years ago. Data for D. simulans are not shown, as the population structure for the D. simulans Wolfskill populations we sampled would inflate estimates through increasing marginal frequencies of segregating sites (Wall et al. 2002).