Interlocus nonrandom association of polymorphisms in Drosophila chemoreceptor genes

Abstract

Some forms of multilocus selection with epistasis, such as truncation selection, can effectively reduce the mutation load [Kondrashov, A. S. (1988) Nature 336, 435–440]. Many quantitative characters, including complex genetic diseases, are likely to be subject to these types of selection. However, direct measurement of selection in natural populations is difficult and the effect of epistasis on within-species variations remains unclear. Epistatic interaction in the fitness effect can generate linkage disequilibrium (LD). Therefore, we may detect the action of natural selection from its amount and pattern. Here, we report a large number of interlocus nonrandom associations between polymorphisms in 98 Drosophila chemoreceptor genes. LD was examined in two fly samples collected at the same location, but in different seasons. The amount of LD was much larger in the spring sample than in the autumn one. The between-sample difference was much more striking for the replacement polymorphisms than for the silent polymorphisms. This difference between the replacement and silent polymorphisms could not be attributed to differences in the mean marker distances. We also found a significant excess of associations between one frequent and one less common allele for the replacement polymorphisms, but not for the silent polymorphisms. It is unlikely that a simple seasonal bottleneck could explain all these differences in the scale of LD between the samples and between the replacement and silent polymorphisms. Natural selection is suggested to play a significant role in shaping the pattern of LD observed in this study.

Some forms of multilocus selection with epistasis, such as truncation selection, can effectively reduce the mutation load (e.g., refs. 1–3). Many quantitative characters, including complex genetic diseases, are likely to be subject to these types of selection to some degree (4). Despite many theoretical analyses of multilocus selection, direct measurement of selection in natural populations remains difficult and the effect of epistasis on within-species variations is still unclear. Epistatic interaction in the fitness effect can generate linkage disequilibrium (LD). Therefore, we may detect the action of natural selection from its amount and pattern. During bottlenecks or fluctuations in selection pressure, the frequencies of slightly deleterious mutations may increase transiently (5) and may become sufficiently high in frequency to be subject to multilocus selection. Indeed, common risk alleles have high frequencies in human populations due to drift or selection (6).

LD studies of small regions of the genome do not have much power for assessing the various evolutionary forces acting on genetic variations due to the large stochastic variation in LD. Instead, the overall pattern of LD across a genome should provide sufficiently high power for such an assessment (7, 8). Comparisons among different classes of variations such as replacement and synonymous variations and analyses of LD distribution along chromosomes aid the assessment of evolutionary forces (9–11). Evaluation of the amount and pattern of LD in “independent” multiple populations is also important and useful for assessment (12). “Independence” is measured with the time of population separation. The criterion of independence should depend on the marker distances because the decay of LD depends on the amount of recombination. For closely linked sites, a long time of separation is required for independence. On the other hand, for loosely linked sites, the time for independence is relatively short. In this sense, it is easy to obtain independent population samples for analyses of long-distance LD.

Drosophila melanogaster has long been studied for both LD between allozymes and that between sequence variations in individual genes (13–17). The amount of LD in Drosophila decays quickly with distance between sites, and little LD is observed between sites separated by >1 kb, except for regions with very reduced crossover frequency (18). Most allozyme studies have failed to find significant LD, although Zapata et al. (19) recently reported strong long-distance LD among 15 allozyme loci on the third chromosome. On the other hand, previous Drosophila studies (20, 21) revealed the existence of recombination load, which is the reduction of fitness by recombination. Decreased fitness through recombination can occur when there are epistatic fitness interactions. In other words, adult organisms that have survived selection possess gene combinations that are superior to those generated by recombination. Thus, some natural variants are not randomly combined in an individual but are in LD. Limited locus choice in allozyme surveys could be one explanation for the failure to detect LD. Indeed, the studies are still limited to small numbers of genes, and the general patterns of LD in the Drosophila genome have not been evaluated on a genome-wide scale, despite the fact that D. melanogaster has the unique advantage of no male recombination for linkage studies. With the complete Drosophila genome sequence now available, we can measure LD between genes with common functions or in the same genetic pathways at the whole-genome level, and such comparisons should help us to assess the patterns and degrees of selection acting on natural variations in a multiple-gene context.

In D. melanogaster, 60 olfactory receptor genes (Or) and 60 gustatory receptor genes (Gr) have been identified by searches of the GenBank database (22). These genes are likely to encode seven transmembrane-domain receptors that bind and recognize odor and taste molecules. It is hypothesized that each receptor recognizes multiple odorants or taste molecules, and that one molecule is recognized by multiple receptors (23). However, little is known about the functional relationships among the receptors (but see also ref. 24).

In this study, we mainly used two fly samples collected at the same location, but in different seasons (autumn and spring), and evaluated the amount and pattern of interlocus nonrandom associations between polymorphisms in 98 Drosophila chemoreceptor genes distributed throughout the genome. We found large differences in the scale of LD between the two samples and between the replacement and silent polymorphisms.

Materials and Methods

Sampling. Male and female flies were collected in Kyoto (in the western part of the mainland of Japan) with banana bait traps in October 2001 and in late May through early June 2002. Each male was crossed separately to the highly inbred A135 strain originating from Aomori, Japan (25) and to the inbred attached X chromosome strain, TT-35 [C(1)RM, y w^a/y w]. Genomic DNA was prepared individually from each parental male and F₁ female progeny from crosses with the A135 strain (Af progeny). DNA was also extracted from two or three F₁ male progeny from crosses with TT-35 (Xm progeny) en masse. The DNAs of the Af progeny were used as templates for autosome PCR typing in the heterozygous condition, and those of the Xm progeny were used for X chromosome PCR typing in hemizygotes.

To distinguish homologous autosomes, we first searched for heterozygous markers in each parental male (see Table 3, which is published as supporting information on the PNAS web site). Each Af progeny was typed for the heterozygous marker, and both chromosomes of each parental male were selected from an Af sample. Thus, we established DNA samples of 178 wild-caught males collected in 2001 (KY01au) and 205 males collected in 2002 (KY02sp), allowing analysis of 383 X, 766 second, and 766 third chromosomes.

For LD analyses of X chromosomes, we used four additional fly samples from the following collections: two samples from Kyoto collected in October 2002 (KY02au, n = 186 males) and in late May through early June 2003 (KY03sp, n = 191 males); and two samples from Iriomote Island in Okinawa Prefecture (one of the southernmost islands of Japan, ≈1,500 km from Kyushu island, which is the southernmost island of the four largest islands of Japan), collected in late November through early December 2001 (IR01; n = 192 males) and April 2003 (IR03; n = 191 males). DNA was extracted from the Xm progeny with the exception of the KY02au sample, where DNA was isolated directly from the wild-caught males.

Typing Markers and Methods. From our survey of DNA sequence variations in candidate genes for olfactory and gustatory receptors (A.K. & T.T-S., unpublished data), we constructed 11 X, 64 second-, and 39 third-chromosome single-nucleotide polymorphisms (SNPs) and length-variation markers (see Table 4, which is published as supporting information on the PNAS web site). Although Robertson et al. (22) did not include the Gr65a gene in the gustatory receptor gene family, two markers at Gr65a were included in this study. Heterozygosity at the Gr8a replacement site was very low, and therefore this site was only typed for the KY01au and KY02sp samples. With the exception of a small number of restriction fragment-length polymorphisms, PCR typing was performed by allele-specific oligonucleotide hybridization (26). The allele-specific oligonucleotides used as probes are also listed in Table 4.

In addition, we used DNAs extracted from the original wild-caught males for PCR genotyping of two inversions, In(2L)t and In(3L)P. The PCR primers described by Andolfatto et al. (27) were used for In(2L)t typing. For In(3L)P, we designed a set of three PCR primers to amplify different fragments from standard and inversion chromosomes (Table 4).

All of the genotyping data are shown in Tables 5–7, which are published as supporting information on the PNAS web site.

Missing Data and Typing Errors. In autosomal typing, failure to amplify genes originating from wild-caught males would result in typing errors. Therefore, the PCR primers were tested before use in typing, with 15 different DNA samples, seven from inbred strains originating from six different locations (Japan, Africa, Australia, Brazil, Taiwan, and Malaysia) and eight from Kyoto strains. When PCR amplification was successful for all of these samples, the PCR primer sets were used for the following typing study. The only exception was the marker Or98b. We could not amplify an Or98b fragment from one inbred strain from Africa with the primers listed in Table 4. This result may be due to a large deletion because four additional primer sets within the same region also failed to amplify fragments, and thus the frequency of the deletion typed at Or98b may be underestimated.

In hemizygous typing of X chromosome sites, we experienced PCR amplification or probe hybridization failures, which could cause errors in autosomal heterozygous typing. Failures occurred with 4 of 11 X chromosome markers. However, the numbers of missing data were only 1 of 1,334 tests for two markers, 2 of 1,334 for one marker, and 13 of 1,334 for one marker. The average failure frequency was ≈0.1%, and it is unlikely that autosomal typing errors affected our results.

During inversion typing, individuals were first examined with sets of three PCR primers that amplify both inversion and standard chromosomes. Single fragments of different sizes were expected to be amplified for inversion and standard chromosome homozygotes. However, typing errors could occur with this inversion typing because the relatively high levels of variation among standard chromosomes could cause binding failures of the PCR primers. As a result, inversion heterozygotes would be scored as inversion homozygotes. We thus reexamined the homozygosity of In(2L)t for samples typed as inversion homozygotes by PCR using an additional primer pair for standard chromosome amplification (Table 4). None of the reexamined homozygotes yielded amplification products with this primer set. On the other hand, such errors were unlikely for the standard chromosome homozygotes because of the low levels of DNA sequence variation, especially at breakpoint-junction regions, in the In(2L)t (27) and In(3L)P chromosomes (28).

Cytological Survey for Inversions. Isofemale lines were established from females collected at the same time as KY01au and KY02sp males. F₂ and F₃ progeny larvae (a single larva per line) were examined cytologically for inversions.

ld. LD between polymorphisms was tested statistically by two-tailed Fisher's exact test. We performed 55 tests for the X, 2,080 tests for the second, and 780 tests for the third chromosome. The critical value for the Bonferroni correction is then 5%/(55 + 2,080 + 780) = 0.0017%. This was a conservative value because a fraction of the tests could not generate significant results with Fisher's exact test, due to the low allele frequency and the direction of the association between the polymorphisms (29). The level of LD was also measured by the squared correlation coefficient (r²) and D′ (30). Because an inversion is usually tightly linked with many markers located in the inversion itself or in adjacent regions, all analyses of LD between 32 second-chromosome markers from Gr21a (21E2) through Gr39b (39D3) and between 19 third-chromosome markers from Or63a (63A2) through Or74a (74A4) were performed without the In(2L)t and In(3L)P chromosomes, respectively. In the analyses of LD direction, we excluded 10 second-chromosome and 4 third-chromosome markers, because the less common alleles differed between KY01au and KY02sp or because they had 0.4 or larger frequencies in both samples. In each sample, we further excluded markers where the less common alleles appeared only once.

Recombination Frequency. Recombination frequency between markers was calculated by using the Kosambi formula (31) and the standard genetic map (32). We considered all possible gene arrangement karyotypes of seven inversions found in the cytological survey, assuming linkage equilibrium between inversions on different chromosome arms and Hardy–Weinberg equilibrium for overlapping inversions. We estimated the inversion frequencies from the cytological survey of the isofemale lines and assumed no recombination in the inverted regions in inversion heterozygotes. The recombination frequency was multiplied by two-thirds for the X chromosome and by one-half for the autosomes because of the lack of male recombination.

Results and Discussion

Markers. From a sequence survey, we generated 114 diallelic typing markers (49 Or and 65 Gr): 11 markers for 10 X chromosome loci, 64 markers for 52 second-chromosome loci, and 39 markers for 36 third-chromosome loci. The markers included two single-nucleotide-length variations (Gr2a and Gr39aC) and four electrophoretically resolvable deletions (Or98b, Gr36b, Gr39aA, and Gr85a). The insertion at Gr2a caused a frameshift, and three markers (Or98b, Gr39aA, and Gr85a) were long deletions that spanned coding regions. The remaining markers were all SNPs comprising 78 replacement polymorphisms (35 Or and 43 Gr) and 30 silent polymorphisms (13 Or and 17 Gr). The replacement polymorphisms included four nonsense mutations (Gr22b, Gr36c, Gr47a, and Gr58a). The frequencies of eight putative loss-of-function polymorphisms (Or98b, Gr2a, Gr22b, Gr36c, Gr39aA, Gr47a, Gr58a, and Gr85a) ranged from 1% for the Gr22b nonsense mutation to 39% for the Gr39aA deletion. Even the Gr2a frameshift mutation on the X chromosome was detected at a frequency of 30%. For these genes, estimates of the nucleotide diversity (π) at replacement sites were ≈30% of those at synonymous sites, except for two genes, Gr39aA and Gr85a (A.K. and T.T-S, unpublished data). At Gr39aA, we did not find any nucleotide changes in the coding region except for the deletion, and only Gr85a showed a high ratio (1.4) of π at replacement sites to that at synonymous sites. In addition, at least, three genes (Gr2a, Gr47a, and Gr58) are known to be expressed in the labellum or labral organ (33, 34). Selection on these genes must be very weak, but does not appear to be completely absent. Low efficacy of selection may be expected due to redundant function of the chemoreceptors (23, 24).

Amount of LD. The LD among the 11 X chromosome markers in KY01au and KY02sp samples is shown in Fig. 1. Significant LDs were observed for 3 and 7 of 55 pairs in the KY01au and the KY02sp samples, respectively, at the 1% level. One significant LD was found in each sample, even with the Bonferroni correction at the 0.0017% level. Only the Gr10b-Gr10a and Gr10a-Or10a marker pairs showed significant LD at the 1% level in both samples. We then studied LD among 10 X chromosome markers (excluding one of the two Gr8a markers) in four additional samples (Fig. 1). Most of the marker pairs were not in significant LD in any sample (25 marker pairs) or were significant at the 1% level in only one sample (16 marker pairs). However, the Gr10b-Gr10a and Gr10a-Or10a pairs showed significant LD at the 1% level in five and four of the six samples (including KY01au and KY02sp), respectively. The Fisher's exact test probability for the Gr10b-Gr10a pair was also only low in the nonsignificant sample, KY02au (P = 0.02). In addition, the linkage phase in this marker pair was in the same direction in all six samples. The Gr10b, Gr10a, and Or10a markers were all located close to each other. The distances between the Gr10b and Gr10a markers and between Gr10a and Or10a were 1,859 and 925 bp, respectively. The significant LDs in these pairs were probably due to the close linkages.

Fig. 1.

LD between 11 X chromosome variations in six samples (from top to bottom): KY01au (Kyoto 2001 autumn; n = 178 X chromosomes), KY02sp (Kyoto 2002 spring; n = 205), KY02au (Kyoto 2002 autumn; n = 186), KY03sp (Kyoto 2003 spring; n = 191), IR01 (Iriomote 2001; n = 192), and IR03 (Iriomote 2003; n = 191). One of two Gr8a variations was examined only in the KY01au and KY02sp samples. The matrix indicates the statistical significance determined by Fisher's exact test. Black boxes, P < 0.000017 [= 5%/(55 + 2,080 + 780), a critical value for Bonferroni correction]; checked boxes, P < 0.01; light gray boxes, P < 0.05; white boxes, P ≥ 0.05.

The average r² values, excluding the three pairs between the three closely located markers (Gr10b, Gr10a, and Or10a), were 0.0077 in the KY01au sample, 0.0170 in the KY02sp sample, 0.0065 in the KY02au sample, 0.0179 in the KY03sp sample, 0.0071 in the IR01 sample, and 0.0090 in the IR03 sample. The amount of LD was always greater in spring than in autumn in the Kyoto samples. In contrast, the r² values of the Iriomote autumn (IR01) and spring (IR03) samples were similar. Iriomote Island is located at 24° north latitude, and the average temperature is 18°C even in January and February. Therefore, a severe reduction in the population size during winter is less likely in this population.

The fly population on Iriomote Island seems to be isolated to some degree from the populations on the mainland and other southern islands of Japan (35). Assuming that the population is isolated and at equilibrium and that 27 generations occurred in 17 months, we estimated the effective size of the Iriomote population to be ≈2,200 (36). This estimate was smaller than, but comparable with, previous estimates of 4,000–5,000 individuals on the basis of the lethal allelism rate (37). With our estimate and the average recombination frequency of 0.12, the expected r² value would be 0.0058 (38). If the population size is 5,000, the expected r² becomes 0.0053. The observed r² values were 1.2–1.3 (0.0071 in the IR01 sample) and 1.6–1.7 (0.0090 in the IR03 sample) times larger than these expected values. This finding may reflect a nonequilibrium state of the population, the action of selection, or both.

LD for the second and third chromosomes is illustrated in Fig. 2. Excluding the In(2L)t and In(3L)P inversions, the numbers of marker pairs in significant LD in both samples were 6 (6/2,080 = 0.3%) on the second chromosome and 9 (9/780 = 1.2%) on the third chromosome at the 0.0017% level. There were 31 (31/2,080 = 1.5%) and 19 (19/780 = 2.4%) significant pairs at the 0.5% level on the second and third chromosome, respectively (see Table 8, which is published as supporting information on the PNAS web site). These results indicate that LD occurred more frequently than expected by chance. LD for 51 of these 52 marker pairs (including two on the X chromosome) had the same sign in the KY01au and KY02sp samples. This finding is in sharp contrast to the result for the marker pairs that were not significant in one or both samples. In ≈42% of such pairs, the direction of LD differed between the two samples. This high frequency of changes in LD direction suggests that these two samples can be taken as nearly independent samples for the LD analyses and that the consistent results in the above 51 pairs are not simply coincidence.

Fig. 2.

LD between 65 second-chromosome variations (A) and 40 third-chromosome variations (B) in the KY01au (n = 356 chromosomes) and KY02sp (n = 410) samples. We included PCR genotyping results for two inversions, In(2L)t and In(3L)P. LD between 32 markers located in the In(2L)t inversion and adjacent regions and between 19 markers in and around In(3L)P were studied without the inversion chromosomes. Significant LDs were observed for 126 marker pairs in the KY01au sample and 276 marker pairs in the KY02sp sample at the 0.5% level and for 45 marker pairs in the KY01au sample and 93 marker pairs in the KY02sp sample at the 0.0017% level. Black boxes, P < 0.000017 in both samples; checked boxes, P < 0.005 in both samples; light gray boxes, P < 0.05 in both samples; white boxes, P ≥ 0.05 in at least one sample.

Significant LDs were found in loosely linked marker pairs, although they were more often observed between very closely linked markers. For example, we found very high r² values for the Or42a-Or46aA (replacement site), Or42a-Or46aB, and Or42a-Or47b pairs (0.16, 0.15, and 0.03 in the KY01au sample, and 0.55, 0.52, and 0.09 in the Ky02sp sample, respectively), despite the lack of inversion in this region and high recombination frequency (2%).

The average r² values for the second and third chromosomes are summarized in Table 1. The results again indicated that there was a greater degree of LD in the KY02sp sample than in the KY01au sample. Importantly, the amount of LD was more markedly different between the two samples for the replacement polymorphisms than for the silent polymorphisms. The bootstrap values (965/1,000 = 97% for the second chromosome and 979/1,000 = 98% for the third chromosome) supported that the average r² value was greater for the replacement polymorphisms than for the silent polymorphisms in the KY02sp sample, but not in the KY01au sample (showing bootstrap values of 54% for the second and 56% for the third chromosome). The ratios of r² values in the spring samples to those in the autumn samples for the X chromosomal markers were also larger for the replacement polymorphisms than for the silent polymorphisms in both comparisons of Ky02sp vs. Ky01au samples and Ky03sp vs. Ky02au samples (excluding three pairs between Gr10b, Gr10a, and Or10a, data not shown), although the number of marker pairs was very small. The effects of demographic history on LD may depend on marker distances and allele frequencies. For example, a bottleneck effect persists for longer for closely linked marker pairs than for loosely linked ones. However, the average recombination frequency was even lower for the silent polymorphisms than for the replacement polymorphisms (Table 1). Therefore, marker distances alone could not explain the observed results. Natural selection is suggested to play a role in determining the amount of LD detected in this study.

Table 1. Average r² values and marker distances.

				Average r2 values (No. of marker pairs)
	Data	Recombination frequency*	Frequency†	KY01au	KY02sp
Second chromosome	All	0.129 (0.06)		0.0048 (1,924)	0.0081 (1,898)
	R	0.132 (0.05)	0.21	0.0049 (1,163)	0.0084 (1,162)
	S	0.117 (0.10)	0.24	0.0046 (90)	0.0059 (84)
Third chromosome	All	0.118 (0.05)		0.0044 (715)	0.0064 (715)
	R	0.121 (0.04)	0.26	0.0047 (340)	0.0071 (340)
	S	0.098 (0.05)	0.30	0.0041 (63)	0.0045 (63)

R and S indicate replacement and silent polymorphisms, respectively. Very closely linked marker pairs are excluded from the analysis. See Materials and Methods for calculation of recombination frequency.

^*Average recombination frequency between markers. Fraction of marker pairs of 1% or less recombination frequency is given in parentheses.

^†Average frequency of less common alleles.

Direction of LD. The LD parameter D can be positive or negative and the direction of LD was tested with the sign test (29). Alleles at all marker sites were assigned so that the LD sign was negative when there was an excess of chromosomes with one frequent and one less common allele (39). Markers were excluded when the frequency of the less common allele was 0.4 or greater. The numbers of positive (including linkage equilibrium) and negative-phase disequilibria did not differ significantly from the expected numbers for the whole data sets in both samples, but did differ for the replacement polymorphisms in the KY02sp sample (Table 2). Even when only the markers where the frequency of the less common allele was 0.2 or less were analyzed, a similar outcome was obtained for the replacement polymorphisms in KY02sp (observed positive or zero/negative = 107:198; expected = 128:177; G′ = 6.11, P < 0.02), despite pronounced asymmetry of the expected numbers (29). The replacement polymorphisms in the KY01au sample and the silent polymorphisms in both samples did not show significant deviations from the expected values.

Table 2. Numbers of positive- and negative-phase LDs.

		No. of observed/expected pairs
Sample	Data	Positive and zero	Negative
KY01au	All	934/900	952/986
	R	521/516	568/573
	S	49/51	54/52
KY02sp	All	913/933	1,039/1,019
	R*	489/528	628/589
	S	59/55	53/57

R and S indicate replacement and silent polymorphisms, respectively. In(2L)t, In(3L)P, and 14 markers were excluded from the analysis (see Materials and Methods).

^*G′ = 5.57, P < 0.02 in G test for goodness of fit with df = 1.

The contrast in the direction of LD between the two samples can be seen in the distributions of D′ (Fig. 3). D′ for the replacement polymorphisms was more strongly negatively biased in the KY02sp sample than in the KY01au sample. The average D′ values were –0.155 in the KY01au sample and –0.188 in the KY02sp sample, and a significant difference was found in the D′ distributions between the two samples (G′ = 26.0, df = 7, P < 0.001). This tendency was not observed for the silent polymorphisms. The average D′ values for the silent polymorphisms were –0.064 in the KY01au sample and –0.048 in the KY02sp sample.

Fig. 3.

Distributions of D′ for the replacement (R) and silent (S) polymorphisms. White and black bars show the results of KY01au and KY02sp samples, respectively.

Because of the nonindependence of pairwise tests, we also performed bootstrap and permutation tests for the replacement polymorphisms on the second chromosome to evaluate the excess of the negative-phase LD in the KY02sp sample. Bootstrap analysis showed 100% probability (500/500) that the numbers of negative-phase LD in the samples are greater than the expected numbers. In permutation tests, the probability that the number of negative-phase LD is equal to or greater than the observed numbers was 0.9% (45/5,000). These results supported the excess of the negative-phase LD in the KY02sp sample.

In summary, a less common replacement polymorphism at one locus tended to be coupled with a common replacement polymorphism at another locus in the KY02sp sample. This tendency was not observed for the silent polymorphisms. Together with the larger amount of LD for the replacement polymorphisms, this finding suggests the action of selection in shaping the scale of LD. Langley and Crow (39) showed that negative-phase LD is produced when the fitness of the double homozygotes is a constant fraction (<1) of the product of the two-component single homozygotes. They further suggested that LD tends to be negative due to the common synergistic interaction among variations, which may be seen from the decline in fitness due to inbreeding. Our results may indicate that synergy exists between replacement polymorphisms at the chemoreceptor genes.

Inversions. The inversions In(2L)t and In(3L)P were typed by PCR amplification of the breakpoints. The frequencies of In(2L)t were 20.8% in the KY01au sample and 19.0% in the KY02sp sample, and those of In(3L)P were 5.1% in the KY01au sample and 2.2% in the KY02sp sample. In(2L)t showed strong LD with most markers in and around the inversion. Fifteen markers were in significant LD with the inversion in both samples at the 0.0017% level and 19 markers at the 0.5% level. In contrast to the findings of associations between nucleotide polymorphisms excluding In(2L)t and In(3L)P, the average r² value of In(2L)t with the silent polymorphisms in and around the inversion was larger than that of In(2L)t with the linked replacement polymorphisms. No difference was found in the average r² values between the two samples (0.270 in the KY01au sample and 0.210 in the KY02 sample for the silent polymorphisms and 0.105 in the KY01au sample and 0.104 in the KY02sp sample for the replacement polymorphisms). The amount of LD between In(3L)P and markers in and around the inversion was lower than that between In(2L)t and surrounding markers. Only 4 of the 19 In(3L)P marker pairs were in significant LD in both samples at the 0.5% significance level (none at the 0.0017% level). The average r² values, 0.032 (0.033 for silent and 0.031 for replacement polymorphisms) in the KY01au sample and 0.023 (0.013 for silent and 0.031 for replacement polymorphisms) in the KY02sp sample, were much smaller than those for In(2L)t (0.146 in the KY01au sample and 0.131 in the KY02sp sample). In(3L)P may therefore be evolutionally older than In(2L)t. In addition, the numbers of the observed positive (including zero) and negative-phase disequilibria between the inversions and linked replacement markers did not differ from the expected numbers in both samples (observed positive or zero/negative = 13:14 and expected = 12:15 in the KY01au sample; observed = 15:12 and expected = 12:15 in the KY02sp sample).

The inversions could have affected the scale of LD on the standard chromosomes through recombination and gene conversion. However, the above results indicate that this effect alone could not explain the differences in the scale of LD on the standard chromosomes between the replacement and silent polymorphisms and between the samples.

From the cytological survey of isofemale lines established from females collected at the same time as the typed males (KY01au and KY02sp), we found five other polymorphic inversions, In(2R)NS, In(2R)51B-55E, In(3R)P, In(3R)C, and In(3R)Mo. The frequencies of In(2R)NS (breakpoints: 52A2-B1 and 56F9–13) and In(2R)51B-55E (51B and 55E) were 8.6% and 7.2%, respectively. None of the genes examined in this study is included in In(2R)51B-55E and only Or56a (56E2) is in In(2R)NS (32). Other genes close to the breakpoints of In(2R)NS are Gr57a at 57B1, Gr58c at 58A4, and Gr58b and Gr58a at 58B1. Only one pair of markers (Or56a and one of two Gr57a markers) of the 21 marker pairs for these five genes was in significant LD in both samples. In short, although we could not type In(2R)NS and In(2R)51B-55E, these inversions appeared to have only negligible effects on the LD analyses. Three third-chromosome inversions, In(3R)P (89C-D and 96A), In(3R)C (92D1-E1 and 100F2–3), and In(3R)Mo (93D and 98F2–3); were detected at frequencies of 12.4%, 9.5%, and 5.7%, respectively. Eleven genes are included in the inversions. These overlapping inversions may have caused a high incidence of significant LD between markers located in this region (Fig. 2).

Conclusions. We found a large number of interlocus nonrandom associations between polymorphisms in 98 Drosophila chemoreceptor genes. The amount of LD was much larger in the spring Kyoto samples than in the autumn Kyoto samples. This between-sample difference was much more striking for the replacement polymorphisms than for the silent polymorphisms. In the spring sample, we also found a significant excess of associations between one frequent and one less common allele for the replacement polymorphisms but not for the silent polymorphisms. It is unlikely that a simple bottleneck could explain all these differences in the scale of LD between the samples and between the replacement and silent polymorphisms. Natural selection is suggested to play a significant role in shaping the pattern of LD through epistatic effects between replacement polymorphisms in the chemoreceptor genes, but confirmation of our results by using additional samples is necessary. The observed differences in the scale of LD warrant further investigation.