Abstract
The hitchhiking model of population genetics predicts that an allele favored by Darwinian selection can replace haplotypes from the same locus previously established at a neutral mutation-drift equilibrium. This process, known as “selective sweep,” was studied by comparing molecular variation between the polymorphic In(2L)t inversion and the standard chromosome. Sequence variation was recorded at the Suppressor of Hairless (Su[H]) gene in an African population of Drosophila melanogaster. We found 47 nucleotide polymorphisms among 20 sequences of 1.2 kb. Neutrality tests were nonsignificant at the nucleotide level. However, these sites were strongly associated, because 290 out of 741 observed pairwise combinations between them were in significant linkage disequilibrium. We found only seven haplotypes, two occurring in the 9 In(2L)t chromosomes, and five in the 11 standard chromosomes, with no shared haplotype. Two haplotypes, one in each chromosome arrangement, made up two-thirds of the sample. This low haplotype diversity departed from neutrality in a haplotype test. This pattern supports a selective sweep hypothesis for the Su(H) chromosome region.
THE theory of genetic hitchhiking (Maynard-Smith and Haigh 1974) predicts that natural selection on favored genes can be revealed by its driving effect on allele frequencies at neighboring loci (Aguadéet al. 1989; Stephan and Langley 1989). Attempts to use this effect in genome-wide surveys of molecular polymorphism as evidence of Darwinian selection have failed due to a strong technical constraint. In highly recombining parts of the genome, the effect of such “selective sweeps” (Kaplanet al. 1989; Begun and Aquadro 1992) is limited to small regions and cannot be easily detected. Low recombining regions, such as Drosophila telomeres and pericentromeric regions, show an absence of variation that can be explained either by selective sweeps or by “background selection” (Charlesworthet al. 1993). The latter is the loss of diversity caused by the recurrent elimination of chromosomes bearing deleterious mutations. Comparing predictions of realistic quantitative models with empirical data shows that background selection can easily explain the contrast in polymorphism levels between “highly recombining” and “low recombining” regions at a genome-wide scale in Drosophila (Hudson and Kaplan 1995). Even though this global pattern conforms to neutral theory predictions, the burden of Darwinian selection undergone by genomes in natural conditions can still be substantial, and its study remains an essential objective of population genetics. Genetic hitchhiking can more easily be identified in highly recombining regions, where background selection would reduce segregating neutral variation, e.g., in Drosophila, by only ∼33% (Hudson and Kaplan 1995). In other words, any departure from neutrality at a specific locus could be ascribed to a selective sweep event, provided that we could compare the actual distribution of polymorphism to its expected value under the neutral mutation-drift equilibrium.
To answer this question, we studied the In(2L)t polymorphic inversion. The underlying rationale was that chromosome inversions can reveal selective sweeps because they strongly inhibit recombination between chromosomal types (see Ashburner 1989). They thus divide up a sample of chromosomes into two partially isolated subpopulations exchanging genetic variation through rare recombination events in inversion heterozygotes. A favorable mutation can appear and go to fixation in one of them. The other chromosomal type remains unaffected until the favored allele recombines into it. In the meantime, a strong contrast in variation pattern differentiates the two chromosomal types. This contrast is more easily detected in highly recombining regions, which have more preexisting segregating variation. This contrast can be expected to occur around inversion breakpoints (internally or externally) rather than in the middle of long inversions, where double crossover (Krimbas and Powell 1992) and gene conversion (Chovniket al. 1977) can take place.
We chose to carry out this study on the Suppressor of Hairless gene (Su(H)) after surveying length variation in several trinucleotide microsatellites from chromosome 2 (Michalakis and Veuille 1996) in natural populations showing inversion polymorphism (Bénassiet al. 1993; Veuilleet al. 1998). We found highly significant linkage disequilibrium between a microsatellite located in the Su(H) coding region and the In(2L)t chromosome polymorphism in a west African population. The Su(H) gene (Furukawaet al. 1991; Schweisguth and Posakony 1992) is a transcription factor that is involved in the Notch signalling pathway (Artavanis-Tsakonaset al. 1995). It presents an OPA-repeat, consisting of tandemly arranged glutamine codons (CAG or CAA). The relative positioning of Su(H) and In(2L)t is known. On the Drosophila melanogaster physical map, this gene maps to position 35B9-10 (Schweisguth 1995), which is between the proximal breakpoint of In(2L)t (in 34A8–9) and the centromere (in 40F). The Su(H) gene is not very far from the Adh locus (in 35B2), which shows a reduced recombination rate (5 × 10–4 recombination events per generation in females) with In(2L)t in heterokaryotypic females (Malpicaet al. 1987). This suggested that, although distant from the inversion (about one-hundredth of the Drosophila genome), the Su(H) gene could be maintained in linkage disequilibrium with the inversion for many generations. The difference in microsatellite allele frequency could indicate a selective sweep had occurred in one or another chromosomal arrangement. We therefore recorded sequence variation at this locus in this population, and found it to agree with the hypothesis of a recent selective sweep.
MATERIALS AND METHODS
A random sample of 85 isochromosomal lines for chromosome 2 was established by Benassi et al. (1993) from a natural population of D. melanogaster from the Ivory Coast (Lamto Ecological Station). In this population, the frequency of the In(2L)t inversion (0.62, SD = 0.05) is particularly high. We used a random sample of 47 lines for the microsatellite survey and a random subsample of 20 lines for the sequence survey. The proportion of In(2L)t and standard chromosomes in the two samples (25 vs. 22 for the microsatellite survey, respectively, and 9 vs. 11 for the sequence survey) did not significantly depart from their initial frequencies. The D. simulans sequence was obtained from a line bearing multiple recessive markers for chromosome 2 and therefore was expected to be largely homozygous for this chromosome at the molecular level, as previously observed for the Fbp2 gene (Benassiet al. 1999).
Su(H) microsatellite variation was studied according to Michalakis and Veuille (1996). To maximize genetic information in the sequence survey, we studied a region of the gene overlapping a long intron. We amplified a fragment including a 712-bp intron and 314 coding sites plus a varying number of indels and microsatellite repeats. Genomic DNA was amplified through standard PCR conditions between coordinates 142 and 1198 of the published Su(H) transcription unit (Furukawaet al. 1991), using primers AACCGTAGTTCGTAGG CAAT and GAACGCAGGCGATTGAACAG. They were sequenced in both orientations with these and four intervening primers (AGGGGTGAGCGGTTGGGGGATT, CTTCCGAA CAGATAAATGCA, TTGCTCAATTTGCGGGC, and GAAA GAAAATCTGAA). The reference sequence is 1056 bp long, while our data extend over 1196 bp to account for indels. The sequences are available from the GenBank database under accession nos. AF088255–AF088275.
Sequences were manually aligned. Their phylogeny was analyzed using MEGA (Kumaret al. 1993). Descriptive statistics were derived using the DNAsp2 program (Rozas and Rozas 1997). The effective numbers of synonymous and nonsynonymous sites were calculated using Nei and Gojobori's (1986) method.
The K-test and H-test of haplotype diversity were run according to Depaulis and Veuille (1998). These coalescentbased simulation tests assess whether the association of S polymorphic sites into K haplotypes in a sample of n sequences is in neutral mutation-drift equilibrium or not. The K-test is based on the observed haplotype number, and the H-test on the sample haplotype diversity. These tests were used under two models. The first is a no-recombination model, for which confidence intervals are available (Depaulis and Veuille 1998). This model is conservative when the number of observed haplotypes is lower than expected, because recombination increases its expectation. The second model takes into account the recombination rate at the studied locus. This rate can be estimated according to Hudson's (1987) method, which derives a value of Nr (where N is the diploid population size and r is the recombination rate per generation per nucleotide) from the variance of pairwise differences in natural populations under the assumption of neutrality. This estimate has a large variance. Hudson et al. (1994) compare this estimate to the value (r = 10–8/bp, yielding Nr = 10–2, assuming N = 106) directly derived from genetic experiments for highly recombining regions (Chovniket al. 1977). We obtained a lower value (Nr = 5.45 10–3) for Su(H), which is probably an underestimate.
RESULTS
Association between the inversion and the Su(H) microsatellite: Four size variants differing by one repeat unit (one codon) were observed at the Su(H) microsatellite (Table 1). The alleles were 251, 254, 257, and 260 nucleotides long and were named after their size. The 251 allele was present on 24 of the 25 inverted chromosomes, but only on 4 of the 22 standard chromosomes. The standard arrangements were more variable, with four alleles and a larger sample heterozygosity (H = 0.615) than the other class (H = 0.076). The linkage disequilibrium between the microsatellite and the inversion was assessed using Fisher's exact test for multiple classes (Raymond and Rousset 1995) and was found to be highly significant (P < 10–4).
Linkage disequilibrium between the Su(H) microsatellite and the In(2L)t inversion
Sequence polymorphism: An alignment of polymorphisms is shown in Figure 1. We found nucleotide polymorphism at 47 sites, of which 17 were polymorphic only in standard, 6 only in In(2L)t, and the remaining 24 in both arrangements. These polymorphisms involved two changes out of 58.67 effective synonymous sites (π = 0.00924), no change out of 241.33 effective nonsynonymous sites, 42 substitutions out of 712 intron sites (π = 0.0206), and three intronic positions where both indels and substitutions occurred. The level of nucleotide variation, as estimated by π (Nei 1987) and θ (Watterson 1975) is shown in Table 2. These values were within the range of values observed for other genes located in highly recombining regions in this species, both for synonymous and noncoding variation (Charlesworthet al. 1995; Moriyama and Powell 1996). The level of variation did not depart from selective neutrality using Tajima's (1989) D test, Fu and Li's (1993) D tests, McDonald and Kreitman's (1991) test, or the HKA test (Hudsonet al. 1987) against 5′-Adhvariation, with D. simulans contributing as an outgroup (Table 3).
Molecular diversity at Su(H)
Linkage disequilibrium between nucleotide polymorphisms: Although levels of variation were thus unremarkable, nucleotide polymorphism was clumped into a small number of haplotypes, as shown in Figure 1. Only two haplotypes were found in In(2L)t and five in standard. The two arrangements shared no haplotype. Two haplotypes, one in the inversion and one in standard, made up two-thirds of the sample (13 out of 20 sequences). Linkage disequilibrium was studied by calculating the P value of Fisher's exact test in all pairwise associations of substitution polymorphisms. Of 741 possible pairwise combinations between informative sites, 290 were significant at the 0.05 level, of which 182 were significant at the 0.001 level. Bonferroni's correction was used to correct for multiple testing, and 35 tests remained significant at the 0.05 level. To our knowledge, this very high proportion by far exceeds the results of any similar study carried out on this species. The large number of tests that are significant at the 0.001 level results from the fact that an excess of chromosomes belonging to the same haplotypes repeatedly give the same result along the sequence. These observations rule out the null hypothesis of random association. They are in agreement with the structuring of basic nucleotide variation into haplotypes.
—Alignment of nucleotide changes at Su(H) in 20 D. melanogaster and one D. simulans sequences. The names of In(2L)t lines are in boldface; the reference sequence is a consensus. dot, same as in the consensus; dash, indel; star, synonymous change. The last column indicates the length of the fragment amplified for studying the microsatellite; repeats are multiples of 3 bp.
Neutrality tests in Su(H)
Haplotype tests: We tested the probability of observing k ≤ 7 haplotypes and a haplotype diversity of H = 0.76, given a sample of n = 20 sequences showing S = 44 diallelic polymorphisms using the K-test and the H-test (Depaulis and Veuille 1998). Results are shown in Table 4. The tests were significant in all cases. The most intuitive test is the K-test, which is based on the observed number of haplotypes. Its significance in the no-recombination test (P < 0.011) is remarkable because this test is conservative. A conservative estimate of the recombination rate was used in the recombination test, which was highly significant (P < 10–4). These tests assess whether or not the haplotypes originated under a neutral coalescent process. A lower than expected number of haplotypes indicates that an event of reduced variation has recently impoverished the haplotype diversity of the gene. The fact that the haplotype tests are significant even under a no-recombination model further means that this conclusion is true irrespective of the inhibitory effect of the inversion on recombination.
Phylogenetic analysis of the haplotypes: To illustrate the structuring of variation, we carried out a phylogenetic analysis of the haplotypes. This approach was validated by the fact that the minimum number of recombination events in the sample, as estimated after Hudson and Kaplan's (1985) method, was only one. A neighbor-joining tree is shown in Figure 2. The phylogenetic analysis provided the same topology under the un-weighted pair-group method using arithmetic averages (UPGMA) and maximum parsimony methods (data not shown) and was supported by high bootstrap values. The In(2L)t and standard chromosomes do not represent completely isolated lines. It was mentioned earlier that the two types of chromosomes shared 24 out of 47 polymorphic sites, suggesting a substantial genetic exchange of material between the two arrangements in the past. The phylogeny comprises a small cluster of four related sequences, consisting of two inverted and two standard haplotypes. This cluster substantially diverges from the other sequences, which make up a larger cluster. The latter involves the two major haplotypes (one inverted and one standard) and several intermediates belonging to standard. The only recombination event substantiated by the four-gamete rule (Hudson and Kaplan 1985) differentiates the two clusters, as is apparent in the alignment (Figure 1). There is no fixed difference between In(2L)t and standard.
Significance of haplotype tests in Su(H)
Neutrality tests within each chromosomal arrangement: We recorded linkage disequilibrium within each chromosomal arrangement. Linkage disequilibrium between the 31 polymorphisms from the two inverted haplotypes was significant (P = 0.028, Fisher's exact test). Among the 34 polymorphic sites from standard chromosomes, 185 of the 561 comparisons were significant (P < 0.05). They corresponded to three of the comparisons between the five haplotypes. This confirms the clustering of variation into a few combinations, as was already apparent from Figure 1. Neutrality tests were nonsignificant, except for Fu and Li's test, which was marginally significant among inversions (Table 3), meaning that singletons were not equally distributed in haplo-types from this subsample. However, because its nine sequences make up only two haplotypes, this result is not very informative. Haplotype tests deserve more interest, because they were significant on pooled data. If we apply these tests to each chromosome class separately, they are significant in all cases. This indicates that the significance was not due to a heterogeneity of the data caused by the inversion.
Molecular divergence between chromosomal arrangements: The genetic divergence between the two karyo-types can be assessed using a fixation index (Hudsonet al. 1992), the value of which was Fst = 0.252 (significance as estimated by permutations, P = 0.0170). This means that one-quarter of the distance between arrangements is due to the structuration into chromosomes. This is a considerable level of differentiation, as can be emphasized by noting that this value is higher than that observed between D. melanogaster populations from different continents. For nine polymorphic genes [excluding Su(H)] located at distant positions on the same chromosome, the average Fst was 0.105 in a group of European and African populations (Michalakis and Veuille 1996). This value was also higher than that observed between the two chromosome arrangements in Lamto for another gene, Adh. Four-cutter restriction site polymorphism at this locus (Veuilleet al. 1998) showed a relative divergence of Da = 0.63 vs. Dxy = 5.93 using Nei's (1987) distances, giving a ratio Gst = 0.106, which is equivalent to Hudson et al.'s (1992) Fst (Charlesworth 1998).
DISCUSSION
Departure from neutrality in a highly polymorphic gene: The Su(H) gene shows a strong contrast between the distribution of polymorphisms at the nucleotide level and at the haplotype level. Nucleotide variation may be briefly described as a normal neutral polymorphism. It presents the composition expected for this species, both in the number of polymorphisms and in the proportion of rare and frequent variants. It therefore appears neutral under available neutrality tests. On the contrary, there is a drastic deficit of haplotypes. All tests lead to the conclusion that the number of haplo-types has recently been substantially reduced. A theoretical model by Barton (1998) shows that a sudden reduction of effective population size (through, e.g., a selective sweep or a bottleneck) splits preexisting neutral variation at a given locus into distantly related families of closely-related lineages. Haplotype distribution at Su(H) seems a remarkable illustration of this (see Figures 1 and 2).
—Neighbor-joining tree of the 20 Su(H) sequences, showing bootstrap values (percentage over 500 replicates) for each node. In(2L)t sequences are in boldface; the D. simulans sequence contributes as an out-group.
Comparing selective vs. demographic explanations: This effect could have been caused by a selective sweep or by a population bottleneck. The two processes have the same effect at the level of single genes, but the first applies to part of the genome, whereas the second applies to the whole of it. Molecular variation is known for other genes from the Lamto sample. Restriction site polymorphism has been observed in 85 chromosomes in Lamto for 2.4 kb of the Adh gene (Benassiet al. 1993). The observed haplotype diversity (H = 0.936) did not depart from that observed for populations from France (H = 0.89) and from Malawi (H = 0.99; Benassi and Veuille 1995). The genetic variability of these three populations has also been compared using a set of 10 polymorphic microsatellites spread over the whole length of chromosome 2 (Michalakis and Veuille 1996). The overall heterozygosity in Lamto (H = 0.421) was in the range observed for the other populations (H = 0.321–0.514).
Comparison with variation at the Acp26Aa and Acp26Ab loci (Aguadé 1998) is in agreement with a selective sweep at the Su(H) locus. These genes, like Su(H) and Adh, are located in a highly recombining region of the left arm of chromosome 2. They lie between positions 25D7 and 26A8-9, in the middle of the region covered by In(2L)t, and show no linkage disequilibrium with the inversion (Aguadé 1998). Sequence polymorphism at these loci has been recorded for 24 chromosomes drawn from Benassi et al.'s (1993) Lamto sample. Haplotype diversity was much higher than for Su(H). Their number of haplotypes (S) and haplotype diversity (H) were above their expectation under a recombination model (Depaulis and Veuille 1998) and were thus opposite to the result observed for Su(H). This remained true if considering only the 44 first polymorphic sites [the number found in Su(H)], where both Acp26A genes showed a number of haplotypes S = 21 and a sample haplotype diversity H = 0.968. We also computed the 0.05 limit intervals of S and H, using a recombination model (Depaulis and Veuille 1998) in combination with estimates of Nr obtained using Hudson's (1987) method. Variation at Acp26A genes appeared neutral (data not shown). They thus provide a negative control for Su(H), allowing us to exclude the hypothesis of a population bottleneck in the Lamto sample.
Relation of the inversion to the selective sweep: Our conclusion is thus that the haplotype pattern of Su(H) results from a selective sweep that affected both chromosome arrangements at a nearby locus. It is remarkable that we do not need to consider the two chromosome arrangements separately to come to this conclusion. Linkage disequilibrium and haplotype tests yield significant results on pooled data, even though they are also significant for each of the chromosome arrangements. This suggests that the same selective sweep event affected both arrangements. An outline of this process is presented in Figure 3. In a first step, an advantageous allele arises by mutation at an unidentified locus “U ” on chromosome 2. This new allele of U is linked to one of the Su(H) haplotypes and to one of the chromosome arrangements, In(2L)t or standard. If Su(H) is at a neutral mutation-drift equilibrium, most haplotypes will be different, as is observed in other D. melanogaster genes, and little linkage disequilibrium will be present between polymorphic sites. In a second step, the favored allele of U goes to fixation in the first chromosome arrangement. This causes a selective sweep at Su(H), increasing the frequency of the first haplotype associated with the favored allele. Because the recombination rate is not inhibited within chromosome arrangements, this allele is soon linked to several other Su(H) haplotypes, which thus survive the selective sweep event. In a third step, the favored allele recombines into the other chromosomal arrangement and links to another Su(H) haplotype. The selective sweep process continues in this arrangement, albeit involving different haplotypes. The strong differentiation observed between haplotypes thus conforms to predictions of a hitchhiking model with recombination where different alleles are affected in different populations (Slatkin and Wiehe 1998).
An alternative hypothesis is that two selective sweep events, one on each chromosomal arrangement, occurred independently. This would cause a balanced polymorphism pattern between them. This explanation was put forward by Kirby and Stephan (1996) for small sequences of DNA, where the absence of recombination is due to the small genetic distance. It is known as the “traffic hypothesis.” This explanation is probably less parsimonious than the former in the case of Su(H).
—Hypothetical scheme of selective sweep in Su(H). Step 1: the Su(H) locus is in neutral mutation-drift equilibrium in two partially divergent chromosomal arrangements (A) inverted and (B) standard (or the letters could be reversed); a favored mutation at another locus appears in the same chromosome as haplotype 1. Step 2: a complete selective sweep at the selected locus occurs in (A) resulting in loss of variation at Su(H), for the A subsample; the loss is partial, due to intra-arrangement recombination. Step 3: a recombination between chromosome arrangements imports the favored mutation into haplotype 6, and the selective sweep event continues in B. Former haplotypes are numbered on the left and surviving haplotypes on the right.
This schema is based on simple hypotheses. In addition, an implicit observation is that selection did not sweep away the inversion polymorphism. The focus of this study is not to evaluate the individual contribution of genes to this phenomenon. We cannot say, from this study, to what extent individual genes can affect an inversion frequency. We can only rule out the hypothesis of a balanced polymorphism at Su(H). The possible role of inversions in maintaining genetic polymorphism under balanced selection was put forward by Wright and Dobzhansky (1946) for D. pseudoobscura and was an influential model in the development of population genetics (Lewontinet al. 1981). Data obtained for another D. melanogaster inversion, In(3L)Payne, showed no significant departure from neutral equilibrium (Hasson and Eanes 1996) in spite of patterns sometimes suggesting selective sweeps. Similar patterns were observed in Drosophila species from the obscura group, which also present inversion polymorphisms (Rozas and Aguadé 1990, 1993; Babcock and Anderson 1996; Popadicet al. 1995). Recombination is not completely inhibited between Su(H) and In(2L)t in heterokaryotypic females. According to Strobeck (1983), even a small recombination rate is sufficient to generate random associations between anciently coexisting polymorphic sites. An old balanced polymorphism could not therefore have caused the pattern observed in Su(H), first because the two chromosomal arrangements are very divergent at this locus, and second because haplotypic variation is depleted within each arrangement.
Our study indicates a way to observe selective sweeps in genomes showing many inversion polymorphisms, as in D. melanogaster (Lemeunier and Aulard 1992). In the case of In(2L)t, the Su(H) locus is located outside the inversion, at the substantial distance of one chromosome division. The effect of inversions on variation can thus extend far from inversion breakpoints. Future research should tell whether the observations made in Su(H) are characteristic of only this gene or can be replicated for other genes associated with this inversion.
Acknowledgments
We thank Anne Turbé for contributing to sequencing as a DEUG student, Michèle Huet for technical assistance, and Matthew Cobb for comments on the manuscript. This work was supported by Centre National de la Recherche Scientifique, Université Pierre-et-Marie Curie, and École Normale Supérieure.
Footnotes
-
Communicating editor: W. Stephan
- Received September 4, 1998.
- Accepted April 8, 1999.
- Copyright © 1999 by the Genetics Society of America
