Evolutionary genomics of inversions in Drosophila pseudoobscura: Evidence for epistasis

Results

Nucleotide Heterozygosity and Divergence. Nucleotide heterozygosity can be estimated either from the number of pairwise differences π (40–42), or from the number of segregating sites Θ (43). The heterozygosity estimate based on pairwise differences is sensitive to different selective and demographic processes. The estimates of silent site heterozygosity within each gene arrangement, among all gene arrangements, and the average divergence between D. pseudoobscura and D. miranda are shown in Table 1. Heterozygosity estimates across all gene arrangements were consistent with the maximum estimate of Θ within inversions based on coalescent simulations.

Heterozygosity was estimated within inversions to determine whether nucleotide diversity is reduced, indicating recent adaptive selection on single inversions or, if genetic variation is elevated, indicating genetic exchange and/or balancing selection. Some loci individually showed a significant excess or deficiency of heterozygosity (Θ) relative to the estimate of Θ from the total sample (Table 1). None of the loci showed an excess or deficiency of polymorphism within the AR arrangement. PP had an excess of heterozygosity based on segregating sites in the F6 locus. The en locus had an excess of polymorphism within the ST arrangement. The en, vg, eve, F6, and EcR loci had an excess of diversity in the CH gene arrangement. The TL arrangement had a deficiency of variation in eve and EcR, whereas the en, vg, Amy 1, and mef 2 loci had an excess of nucleotide diversity.

Estimates of divergence between D. miranda and any of the five D. pseudoobscura gene arrangements were not significantly different for any of the genes except vg. At vg, the divergence of D. miranda to D. pseudoobscura PP alleles was significantly less than the divergence of D. miranda to the four other D. pseudoobscura inversion types.

Tests of Selective Neutrality. We used two approaches to test variation at the eight loci for departures from an equilibrium neutral model. The Tajima (44) D test statistic is the difference of two heterozygosity estimates, one based on pairwise differences (π) and the other on segregating sites (Θ). If D is negative, there is a significant excess of rare variants, suggesting purifying selection or population expansion. Alternatively, if D is positive, there is an excess of intermediate frequency variants, indicating balancing selection or recent population contraction or bottle-neck. Most loci (73.7%) within gene arrangements had a negative Tajima's D, indicating an excess of rare variants, but only two cases had a significant negative D (vg with CH and mef 2 within ST; see Estimates of Tajima's D Within Gene Arrangements in Supporting Text, and Table 4, which are published as supporting information on the PNAS web site). Estimates of Tajima's D across all gene arrangements also tended to show an excess of rare variants in all gene regions, but only exu 1 and EcR had a significant excess of rare variants (Table 1).

We used the Hudson, Kreitman, and Aguadé (6) test to determine whether the polymorphism to divergence ratio is similar between a test locus and a neutrally evolving locus. If the ratio is greater in the test locus, then balancing selection is implicated. On the other hand, if the ratio is lower for the test locus, then directional selection is suggested. Silent polymorphism within D. pseudoobscura and divergence between D. pseudoobscura and D. miranda across all gene arrangements were used for the test. Silent variation at the alcohol dehydrogenase locus on the fourth chromosome of D. pseudoobscura (45) was used as the neutral locus for comparisons with each of the eight third chromosomal loci. The eight loci failed to reject a neutral model in HKA tests when variation was estimated for the pooled chromosome types. We also performed the HKA test on the eight loci within each inversion. The vg and mef 2 loci within AR, the exu 1 and Amy 1 locus within PP, the exu 1 and eve locus within ST, and the eve and EcR loci within TL each showed a significant deficiency of polymorphism relative to divergence (see Tables 5–10, which are published as supporting information on the PNAS web site, for all HKA results).

Genetic Differentiation Within and Among Inversions. Coadaptation predicts that genes will be different between inversions and within inversions sampled from different populations. We used a random permutation method to test the eight third chromosomal genes for significant differentiation among gene arrangements (46). Genes within inverted regions showed significant differentiation among gene arrangements, whereas nearly all genes outside inverted regions failed to reject genetic similarity (Table 2). There were two exceptions. exu 1 is outside the region inverted between ST and AR, but it is significantly differentiated. F6 is within the inverted segments between CH and TL, but it failed to show significant genetic differentiation. Only genes within inverted regions had fixed differences among gene arrangements, whereas no fixed differences were found in genes outside inverted segments. Thus, the first prediction of coadaptation is met.

The AR and ST arrangements were tested for significant genetic differentiation among populations. We failed to find evidence for significant differentiation among the four populations sampled within the AR gene arrangement or between the two populations within the ST arrangement with the random permutation test (46). The neutral migration parameter Nm was estimated from F_st in the AR and ST arrangements, where N is the effective population size and m is the migration rate (47). When Nm is ≥1, gene flow is sufficient to prevent genetic differentiation at all loci in the genome (48). The Nm value estimated for the eight concatenated genes in the ST and AR gene arrangements was 4.5 and 9.5 migrants per generation, respectively. Thus, the second prediction of coadaptation is not met.

LD Among Loci. Strong epistatic selection will maintain nonrandom associations or LD among genetic variants. Pairs of segregating sites across all gene arrangements were tested for significant LD. A total of 7,140 of the possible 31,347 pairwise comparisons of the 238 segregating sites were capable of rejecting the null hypothesis of no association with Fisher's exact test (49). Over one-fourth or 65 segregating sites showed a significant association with at least one other polymorphic site within or between loci. A total of 302 of 7,140 valid comparisons, or 4.2%, were in significant LD (Fig. 2) when the sequential Bonferroni correction was applied to account for multiple tests. This fraction of sites in significant LD is greater than that in alcohol dehydrogenase on the fourth chromosome (0.2%, P < 0.0001; ref. 50) or among five genes across the X chromosome (0.5%, P < 0.0001; ref. 51), but is less than expected assuming no recombination among the inversions in coalescent simulations (95% confidence interval, 20.3–79.8%).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Significant LD among pairs of segregating sites in eight gene regions from the third chromosome of D. pseudoobscura with a sequential Bonferroni correction (39). The heavy lines delineate intra- and interlocus comparisons. Comparisons of each site with the five gene arrangements are shown at the bottom.

We partitioned the tests for significant LD into intra- and interlocus site comparisons. A total of 12.9% of intralocus site comparisons were significantly associated, while 2.8% of interlocus site comparisons were in significant LD (intralocus, 131 significant of 1,016 valid comparisons; interlocus, 171 significant of 6,124 valid comparisons; Fig. 2). We measured the strength of LD with r², the square of the correlation of gene frequencies (52). The average value of r² for intralocus comparisons was 0.670, whereas r² for interlocus comparisons was 0.515. Interlocus percentages of significant LD were greatest when nucleotide sites were within inverted regions. exu 1 locus did not exhibit any significant interlocus LD, although exu 1 is within the inverted regions of most gene arrangements.

Interlocus LD did not always decrease with distance. Amy 1 and eve are separated by an average of four subsections on the cytological map, yet 1.6% of the site comparisons between these loci are in significant LD, which is less than the mean percentage of significant LD among loci. In contrast, 15.8 subsections on the cytological map separate vg and mef 2, yet 8.6% of the site comparisons are in significant LD, the maximum observed among loci (Fig. 2; and see Analysis of Interlocus LD in Supporting Text, and Fig. 4, which are published as supporting information on the PNAS web site).

Segregating sites within gene arrangements were tested for significant LD. A total of 16 of 1,932 valid comparisons (0.8%) within the AR arrangement were in significant LD, a proportion that is much less than what was seen across all gene arrangements. The four other gene arrangements failed to show any significant associations, but these results should be viewed with caution because the power to detect significant associations is low given the small sample sizes of PP, ST, CH, and TL (6–18).

We tested the 238 segregating sites for significant LD with the AR, PP, ST, CH, and TL arrangements (Fig. 2). A total of 375 of the possible 1,260 comparisons could reject the null hypothesis of no association with Fisher's exact test (37). Of the 375 comparisons, 58 sites, or 15.5%, showed a significant association with at least one of the gene arrangements (Fig. 2). Associations of nucleotide sites with gene arrangements were greatest for genes that had the highest probability of being within inverted segments. The en locus is only within one inverted segment (PP), and 4.4% of the en sites showed significant LD with the PP arrangement. Alternatively, Amy 1 is within the inverted region of all arrangements, and 21.2% of the Amy 1 sites showed nonrandom associations with at least one of five gene arrangements (Fig. 2).

The D. miranda sequences were used to infer the ancestral or derived states of segregating sites that were in significant LD. Our naïve expectation was that all significant linkage disequilibria would result from an excess of repulsion versus coupling phase gametes, i.e., the frequency of ancestral–derived and derived–ancestral gametes would be greater than ancestral–ancestral and derived–derived gametes because new mutations would not tend to co-occur on the same chromosome. The ancestral and derived states could be inferred for both sites in 278 of 302 significant LD comparisons. Pairs of sites in significant LD tended to be in coupling rather than repulsion phase (180 coupling phase gametes of 302 total tests, or 59.6%). The coalescent simulations, however, showed that the observed fraction of coupling phase gametes was close to the mean (63.8%) and well within the 95% confidence interval (50.1–95.2%). This result is likely to be due to the co-occurrence of derived mutations on the genealogical branches near the origin of common inversion types. The frequency of coupling gametes was significantly greater for intralocus comparisons (76.3%) than for interlocus comparisons (46.8%) (Fisher's exact test, P < 0.0001). Coalescent simulations show that these fractions would be expected to be equal, indicating that recombination has reduced coupling-phase gametes in interlocus comparisons.

Discussion

The Nature of Coadaptation in the D. pseudoobscura Gene Arrangements. The coadaptation model (10) makes two predictions about genetic variation within and between the D. pseudoobscura gene arrangements. The first prediction, that genetic loci will be differentiated among gene arrangements, was supported by our data (Table 2). Genes near breakpoints or within inverted regions are differentiated most likely because reduced genetic exchange leads to an accumulation of nucleotide differences among inversions (53). The only exception is the F6 region, which lies within the inverted segments of CH and TL, yet is not differentiated. Genetic exchange in regions outside of inverted regions in the en and EcR loci has probably prevented genetic differentiation.

The second prediction, that genetic loci within individual gene arrangements would be differentiated among populations, was not supported by the eight loci examined in this study (Table 2). Two pieces of evidence suggest that gene flow promotes the free exchange of gene arrangements among populations and limits genetic differentiation within populations. First, there was no evidence of differentiation of AR chromosomes among the four populations, and second, LD levels were low within the AR chromosome. If each population had unique AR haplotypes that were adapted to each microhabitat, then one would expect significant genetic differentiation and high LD levels. Migration parameter estimates from third chromosomal loci within the AR arrangement are consistent with previous gene flow estimates for genes on the X, second, and fourth chromosome that suggest extensive gene flow in this species (51, 54, 55).

It is possible that gene arrangements are differentiated among geographic populations, but we could not detect these differences for biological or statistical reasons. The possible reasons are that the fitness effects associated with coadaptation are due to many loci, each with small effect, the regions examined in this study are not at or near a target of epistatic selection, or the statistical tests of differentiation and LD had low power. We increased the power of the differentiation test by concatenating loci within inverted regions, which increases the number of segregating sites, but still failed to find significant genetic differences among the population. Studies of nucleotide diversity in more gene regions within inversions will be necessary to find targets of coadaptation.

Evidence for Selection in Heterogeneous Environments. The inversions may have arisen in D. pseudoobscura as the species expanded its range and adapted to the diverse environments of western North America (56). Recombination modifiers that raise the level of genetic exchange will increase in a species that is adapting to a heterogeneous environment and when beneficial mutations are in repulsion phase. On the other hand, modifiers that lower recombination rates will increase in populations if beneficial mutations are in coupling phase (57).

D. pseudoobscura appears to have recently expanded its range and rapidly increased its population size (45, 51, 58). Allelic genealogies in an expanding population are expected to be more star-like, with many new mutations occurring on terminal branches resulting in negative Tajima's D values. Tajima's D is negative for all regions sampled on the third chromosome of D. pseudoobscura, providing further support for a range expansion and rapid population growth in D. pseudoobscura.

The nucleotide data from this study lead us to propose that D. pseudoobscura has acclimated to microhabitats in its current geographic range and that suppressors of recombination have emerged to maintain the positive epistatic gene combinations within the boundaries of the inversion breakpoints. The strongest evidence for epistatic selection in our data set is based on the low frequency of nonrandom associations among sites of closely linked loci (Amy 1 and eve) and the high fraction of nonrandom associations among sites in loosely linked loci (vg and mef 2). Studies of more gene regions are needed to determine the LD landscape of the third chromosome and the regions that show potential epistatic interactions.

Selection in heterogeneous environments does not require that gene arrangement diversity be maintained through over-dominance, but that karyotypic heterogeneity is maintained through an average heterosis over diverse microhabitats (59). The population structure of D. pseudoobscura is consistent with a selection in heterogeneous environments model because gene flow is sufficient to homogenize gene frequencies among populations (51, 54, 55). Thus, the selection that alters the frequencies of the third chromosome gene arrangements must be quite strong to counteract the homogenizing effects of migration. The basis for the selection may be related to the climatic shifts among western North American habitats because the gene arrangement frequencies shift with transitions among the major physiographic provinces that comprise D. pseudoobscura's geographic range (2). In this case, the average heterosis predicted by the Levene model has not elevated polymorphism at any of the D. pseudoobscura third chromosomal loci, indicating that the inversion polymorphism may be relatively young.

The study of eight gene regions on the third chromosome of D. pseudoobscura should be viewed as a preliminary glimpse into the pattern and organization of nucleotide diversity of these gene arrangements. The observation of low levels of genetic exchange among the gene arrangements refutes the hypothesis that each inversion type is effectively one superallele where all variation is in absolute LD, and demonstrates that we should be able to identify local gene regions subject to selection. The molecular basis of selection is unknown, but selection could operate either on gene expression or allelic differences among chromosomes. Gene expression differences can be assayed rapidly with microarray experiments, but choosing the loci in which to study allelic differences may be more problematic because individual inversions in D. pseudoobscura may have between 399 and 1,543 genes.

The data presented in this study support a model of coadaptation where genes along particular gene arrangements are maintained by epistatic selection. We have not discovered any genes within gene arrangements that support the local adaptation component of this model. Our study suggests that polymorphism and recombination levels within inversion are sufficient to detect local adaptation if it exists. It will be important to repeat Dobzhansky's population cage experiments with known strains, which can be characterized with mapping and functional studies if classical coadaptation is detected. The availability of the complete D. pseudoobscura genome sequence will greatly facilitate future experiments on the third chromosomal gene arrangements because it will be possible to define which genes are within inversion breakpoints and to characterize mapped selective targets with polymorphism or functional analyses.