Patterns of Diversity and Linkage Disequilibrium Within the Cosmopolitan Inversion In(3R)Payne in Drosophila melanogaster Are Indicative of Coadaptation

Abstract

The cosmopolitan inversion In(3R)Payne in Drosophila melanogaster decreases in frequency with increasing distance from the equator on three continents, indicating it is subject to strong natural selection. We investigated patterns of genetic variation and linkage disequilibrium (LD) in 24 molecular markers located within and near In(3R)Payne to determine if different parts of the inversion responded to selection the same way. We found reduced variation in the markers we used compared to others distributed throughout the genome, consistent with the inversion having a relatively recent origin (<N_e generations). LD between markers and In(3R)Payne varied significantly among markers within the inversion, with regions of high association interspersed by regions of low association. Several factors indicate that these patterns were not due to demographic factors such as admixture and bottlenecks associated with colonization, but instead reflected strong epistatic selection. Furthermore, we found that nonadjacent regions with high association to the inversion contained markers with the strongest clinal patterns in allele frequency; in most cases, the level of clinal variation was beyond what could be explained by hitchhiking with In(3R)Payne, indicating that genes within these regions are targets of selection. Our results provide some support for the hypothesis that inversions persist in natural populations because they hold together favorable combinations of alleles that act together to facilitate adaptive shifts.

CHROMOSOMAL inversion polymorphisms and other mechanisms that influence the level of linkage disequilibrium among genes are widespread in organisms (Hoffmann et al. 2004). Inversion polymorphisms have been particularly well studied in species of the genus Drosophila, where both the standard and inverted forms of many chromosome arrangements occur in populations at an appreciable frequency (Krimbas and Powell 1992). Changes in inversion frequencies in space and time provide strong evidence that inversion polymorphisms are maintained by selection (e.g., Mettler et al. 1977; Stalker 1980). However, despite decades of research, the reasons why such polymorphisms persist in natural populations remain unclear. Crossing over within inversion loops in heterokaryotypes gives rise to nonfunctional or nonviable meiotic products, maintaining positive associations between alleles at loci even when these are not tightly linked. Under Dobzhansky's coadaptation hypothesis, it is thought that this leads to selection for combinations of alleles within chromosome arrangements that have a high fitness (Dobzhansky 1970; Krimbas and Powell 1992; Powell 1997), as well as to different combinations of alleles being present in both different chromosomal arrangements and the same chromosomal arrangement when isolated from separated populations (Schaeffer et al. 2003).

Molecular studies reveal that while there is little evidence of genetic exchange between chromosomal arrangements near breakpoints, the opposite pattern is observed in the middle of the inversion where unrestricted homologous pairing allows gene exchange by multiple crossing over and gene conversion (Hasson and Eanes 1996; Andolfatto et al. 2001). Nevertheless, despite the potential for gene exchange between chromosomal arrangements, many genetic markers located within or near inversions show significant associations with the inversion or with each other (Prakash and Lewontin 1968; Stalker 1976; Weeks et al. 2002; Munté et al. 2005). Such associations suggest that selection is holding together favorable combinations of alleles. However, unless the age of the inversion is known, it is difficult to discount the possibility that associations are only remnants of complete linkage disequilibrium (LD) within an inversion at the time of its formation (Aquadro et al. 1991). Stronger evidence of epistatic selection and coadaptation comes from a study by Schaeffer et al. (2003), who showed that in Drosophila pseudoobscura, high levels of LD persist within inverted regions, even for nonadjacent loci.

There is still no evidence linking patterns of LD within inversions with adaptive selection. If epistatic selection maintains associations among genes within an inversion, there should be evidence of selection on regions of the inversion in strong LD and away from inversion breakpoints. Here, we test for this pattern in In(3R)Payne, a common cosmopolitan inversion with stable frequency clines in natural populations of D. melanogaster. Parallel clines for this polymorphism are found on three continents, the inverted arrangement decreasing in frequency with increasing distance from the equator (Knibb 1982). These clines tend to be very steep; therefore, the latitudinal selection producing these patterns is likely to be particularly strong (Anderson et al. 2005).

We assessed the association between LD and clinal variation in three steps, using 24 molecular markers located within or near In(3R)Payne and flies from eastern Australia. First, we examined genetic disequilibrium among the markers in the offspring of wild-caught D. melanogaster in a population where the inversion was present at an intermediate frequency. Next we tested for clinal patterns for the markers, to determine if loci on the same arrangement were differentiated among populations and to determine the impact of the inversion polymorphism on patterns of genetic variation in populations. Finally, we linked patterns of disequilibrium among the markers and the inversion to the clinal patterns, to test if markers within the inversion showing clinal patterns were also in strong disequilibrium with the inversion.

MATERIALS AND METHODS

Fly collections:

The collection of D. melanogaster from 32 sites along a 3000-km transect on the east coast of Australia was described previously in Kennington et al. (2003). Linkage disequilibrium between markers and In(3R)Payne was examined in 912 flies. These flies were the offspring of 400 field-collected females from a population at Coffs Harbour, New South Wales (30.27°S, 153.49°E); two to three offspring were sampled from each wild-caught female. By genotyping more than one offspring per female, the population size for examining patterns of linkage disequilibrium was boosted, and disequilibrium patterns still reflected those of the original population. We chose a population from Coffs Harbour to examine LD because it has an intermediate frequency of In(3R)Payne (Knibb et al. 1981; Anderson et al. 2005).

Genotyping:

DNA extraction from individual flies, PCR protocols, and allele scoring followed methods outlined in Gockel et al. (2001). For the latitudinal transect survey, 19 microsatellite loci and the 8-bp insertion/deletion polymorphism at the 5′ end of the Hsr-omega gene (McKechnie et al. 1998) were genotyped for each fly. Allele frequency data for an additional 4 microsatellite loci were taken from the Kennington et al. (2003) study, which was based on the same collection of flies.

For the analysis of LD within the midlatitude Coffs Harbour population, the same 24 markers analyzed for the latitudinal transect, plus two additional microsatellite loci (DMTRXIII2 and 3R20604755ta), which did not amplify well in the latitudinal survey, were genotyped. All markers were located on the right arm of chromosome 3, with most close to or within the breakpoints of In(3R)Payne. Primer sequences and information about microsatellite loci used can be found at http://www.ucl.ac.uk/biology/Goldstein/mList.htm and http://i122-server.vu-wien.ac.at.

The chromosomal arrangement of each fly was determined by genotyping a SNP polymorphism near the proximal breakpoint of In(3R)Payne, using the BI-PASA method described in Liu et al. (1997). Anderson et al. (2005) established that this marker was in complete linkage disequilibrium with In(3R)Payne, using single larvae from 108 isofemale lines from eight populations collected along the coast of eastern Australia. It was used in preference to the In(3R)Payne marker derived from the breakpoints described in Matzkin et al. (2005), because this marker cannot be used to distinguish inversion heterozygotes from inversion homozygotes, and because the association between this marker and the inversion has not been tested in flies from eastern Australia.

Data analysis:

Average heterozygosity was calculated for each site along the transect using the MICROSAT v1.4 software package (Minch et al. 1995). The relationship between latitude and average heterozygosity was assessed using simple linear regressions. Linear regression was used to assess latitudinal variation in the most common allele (MCA). The same approach was used to investigate the association between latitude and the frequency of the second-most common allele (SMCA) for each marker, in case the SMCA was influenced more strongly by latitudinal selection than the common allele. The first- and second-most common alleles should encompass most of the informative variability at each locus. Allele frequencies and average heterozygosities were angularly transformed prior to analysis, and all regressions were weighted by sample size. To test for differentiation between the standard and inverted chromosome arrangements, and to test whether markers within In(3R)Payne were differentiated among populations, we calculated pairwise F_ST values using the FSTAT software package (Goudet 1995). The significance of pairwise F_ST values was determined by permuting genotypes among populations 10,000 times, because this method does not rely on Hardy–Weinberg assumptions (Goudet et al. 1996).

Clinal variation was also assessed by II-values, a spatial autocorrelation statistic designed specifically for DNA data, calculated with the AIDA computer program (Bertorelle and Barbujani 1995). II-values vary between −1 and +1 and have an expectation close to zero when alleles are randomly distributed. Positive II-values indicate overall similarity between samples, and negative values indicate dissimilarity. Statistical significance of II-values was determined for each marker at each distance class by comparing observed values to confidence limits calculated from 2000 random permutations of the data. A marker exhibiting clinal variation is expected to show decreasing II-values from significantly positive to significantly negative, whereas a spatially random distribution results in a series of nonsignificant II-values at all distance classes and a decreasing set of II-values from significantly positive to nonsignificant at large distance classes is expected for isolation-by-distance, when genetic diversity reflects random drift and short-range gene flow (Barbujani 1987).

Linkage disequilibrium between genetic markers and In(3R)P in the Coffs Harbour population was quantified using the multiallelic version of Lewontin's D′-statistic (Lewontin 1964), . Alleles with a frequency <0.10 were pooled into a common class when calculating this statistic. Because this measure is biased upward as a function of sample size and heterozygosity, we corrected the D′_m-values using the permutation method developed by Devlin et al. (2001). This was done by permuting genotypes at each locus independently of genotypes at the molecular marker for In(3R)Payne and calculating the average D′_m from 500 simulations. A standardized D′_m was obtained by subtracting the mean of the simulated values from the observed value. Confidence limits were calculated by bootstrapping genotypes and recalculating the standardized D′_m-statistic 500 times. Because low levels of polymorphism sometimes led to spurious results (e.g., negative standardized D′_m-values) markers with low diversity were excluded from these analyses.

The significance of associations between markers and between markers and In(3R)Payne was assessed using exact tests performed with the GDA computer program (Lewis and Zaykin 2001). This method does not require information on the gametic phase and calculates the probability of a set of multilocus genotypes in a sample, using multinominal theory under the hypothesis of no association. Significance levels are then found by permutation procedures (Zaykin et al. 1995). To prevent within-locus disequilibrium affecting the significance of disequilibrium between markers, genotypes were preserved when performing shuffling tests.

RESULTS

Marker diversity:

Large differences in diversity were observed between loci. The number of alleles ranged from 2 to 19 and heterozygosities ranged from 0.02 to 0.88 (Table 1). Populations also differed in marker diversity. Average heterozygosity showed a distinct curvilinear relationship with latitude, peaking at intermediate latitudes and falling away on either side (Figure 1). This pattern contrasts with the pattern observed in 20 microsatellite markers randomly distributed on chromosomes 2 and 3, which showed a significant negative linear relationship of heterozygosity with latitude (Kennington et al. 2003).

TABLE 1.

Position, allele number, and heterozygosity of the markers used in this study

	Position
Markera	Genetic (cM)	Cytological	No. of allelesb	Heterozygosity
DROPROSA	51	86E6	11 (108–130)	0.73
DMTRXIII2c	55	88B1	3 (275–279)	0.49
DROTROPI2	56	88E13	6 (81–95)	0.62
AC006414	58	89A2	6 (186–196)	0.17
AC007647	59	89C1	8 (183–199)	0.51
DROABDB	59	89E4	12 (260–299)	0.69
3R1302339ga	60	90A1	7 (103–123)	0.67
DMEHAB	61	90B1	4 (352–368)	0.17
DMCP017G	61	90D1	7 (257–275)	0.71
AC009394	62	90F8	8 (212–242)	0.64
DRONANOS	66	91F7	19 (98–136)	0.86
3R15743903gt	67	92C1	12 (121–143)	0.63
3R16177365gt	68	92E8	6 (141–151)	0.53
AC009347	69	93A3	7 (215–237)	0.28
Hsr-omegad	71	93D4	2 (83–91)	0.44
DROHOXNK4	71	93D10	5 (117–138)	0.10
DMU25686	73	93F11	16 (130–160)	0.79
AC008193	77	94D2	18 (188–230)	0.85
DMPOINT1A	78	94E11	2 (123–126)	0.02
DMU1951	81	95C4	14 (176–214)	0.88
DMTF125	81	95C9	12 (276–324)	0.71
3R1963976ta	81	95C10	12 (102–138)	0.76
3R20604755ac	86	96B2	8 (96–114)	0.67
DROROUGH	92	97D6	7 (61–85)	0.37
3R23156893gt	93	97F11	8 (118–132)	0.72
DROLMALK	94	98A14	6 (156–171)	0.75

^aMarkers are published microsatellite loci unless otherwise specified.

^bMinimum and maximum allele sizes (base pairs) are in parentheses.

^cLocus scored in the midlatitude population only.

^dEight-base pair insertion/deletion polymorphism at the 5′ end of the Hsr-omega gene (McKechnie et al. 1998).

Figure 1.

Relationship between average heterozygosity (H) and latitude. Open circles are markers randomly distributed across chromosomes 2 and 3 (data from Kennington et al. 2003). Solid circles are the polymorphic markers used in this study.

Analysis of covariance with data set as a fixed effect and latitude as a covariate revealed that both the intercepts and slopes of the two data sets were significantly different (t = 7.45, P < 0.001 and t = −2.11, P = 0.039 for the intercepts and slopes, respectively). Heterozygosity was significantly higher and more strongly clinal in the randomly distributed markers. The difference between the latitudinal patterns was also evident from the fact that the addition of a quadratic term to the regression model significantly increased the proportion of variation explained by the model for the data from this study (F_[30,29] = 38.80, P < 0.001), but did not increase the variation explained by the model for the data based on the randomly distributed markers (F_[30,29] = 1.70, P = 0.203).

A paired t-test revealed that marker heterozygosities were significantly lower in the inversion compared to the standard chromosome arrangement (t = −2.62, P = 0.019), suggesting that historical processes had influenced levels of genetic variability in the inverted compared to the standard arrangement.

Genetic differentiation within and between chromosome arrangements:

Except for two markers with low diversity (DMEHAB and DMPOINT1A), all markers within In(3R)Payne showed significant population differentiation between the standard and inverted chromosomal arrangements (P < 0.001 in all cases). However, there was little evidence of differences in the same chromosome arrangement among populations. Of 66 pairwise tests for differentiation among populations with at least 10 individuals homozygous for In(3R)Payne, only 5 were significantly differentiated after Bonferroni correction.

Latitudinal variation:

With the exception of few markers located outside the inversion, three with low variability and two located near the middle of the inversion, the frequency of the MCA at each marker was significantly associated with latitude (Table 2). However, after correction for multiple comparisons, the number of significantly associated markers reduced to nine. These markers were all located within the inversion and were clustered in three broad regions: marker DROABDB (89E), between markers DMCP017G and DRONANOS (90D–91F), and between Hsr-omega and 3R196397ta (93D–95C).

TABLE 2.

The proportion of variation in allele frequencies explained by latitude and II-values at each distance class

			II-values for distance class (km)
Locus			0	0.1–500	501–1000	1001–1500	1501–2000	2001–2500	>2500
DROPROSA	0.10	0.03	1.6*	0.4	0.0	−0.7*	−0.4	−0.2	0.6*
DROTROPI2	0.20**	0.14*	1.8**	0.3	−0.6*	0.0	0.0	−1.4***	1.2**
AC006414	0.37***	0.08	2.5***	0.5**	0.1	0.4	−0.2	−1.1***	−4.2***
AC007647	0.08	0.25**	2.8***	2.0***	−0.2	−0.7*	−1.9***	−0.7*	0.0
DROABDB	0.78***	0.63***	16.9***	12.0***	7.3***	0.1	−6.3***	−18.3***	−33.7***
3R1302339ga	0.14*	0.45***	4.5***	2.3***	0.6***	0.0	−2.7***	−2.8***	−1.4***
DMEHAB	0.01	0.10	1.3**	0.0	−0.1	−0.1	−0.6	0.3	−0.6
DMCP017G	0.28**	0.70***	6.0***	4.0***	1.6***	−0.5	−1.5***	−3.7***	−10.9***
AC009394	0.62***	0.63***	3.4***	2. ***	1.6***	−0.2	−1.2***	−3.8***	−6.8***
DRONANOS	0.31**	0.46***	1.8***	1.0***	0.5***	−0.5*	−0.4	−1.5***	−2.7***
3R15743903gt	0.04	0.16*	6.5***	1.1***	−0.8**	−0.1	−2.1***	−0.3	1.3***
3R16177365gt	0.01	0.00	0.9	1.3***	−0.3	−1.8***	−0.4	1.0**	−0.1
AC009347	0.14*	0.45***	2.3**	0.9**	−1.0**	−0.3	−0.4	0.8*	−1.6*
Hsr-omega	0.83***	0.83***	13.6***	11.5***	6.3***	1.1***	−6.4***	−16.2***	−31.5***
DROHOXNK4	0.03	0.03	0.6	0.3*	−0.4	0.1	0.0	−1.3***	1.1**
DMU25686	0.71***	0.60***	10.6***	6.4***	2.4***	−0.6*	−3.8***	−10.6***	−8.1***
AC008193	0.57***	0.56***	12.3***	7.5***	3.9***	−0.2	−3.5***	−10.8***	−20.1***
DMPOINT1A	0.01	0.04	2.3*	0.7*	−1.2***	0.5*	−0.3	−0.9*	0.8
DMU1951	0.23**	0.48***	3.0***	1.3***	0.6***	−0.1	−0.7**	−2.4***	−3.9***
DMTF125	0.78***	0.57***	22.2***	15.2***	8.5***	1.2***	−7.9***	−18.9***	−41.6***
3R19639736ta	0.41***	0.47***	7.9***	5.5***	2.6***	−0.1	−3.3***	−6.7***	−13.9***
DROROUGH	0.20*	0.15*	1.4*	0.2	−0.2	0.1	−0.1	−0.7*	−0.8
3R23156893gt	0.01	0.14*	5.0***	2.9***	1.5***	0.6**	−1.9***	−5.3***	−8.4***
DROLMALK	0.03	0.15*	4.7***	2.6***	0.6***	−3.0***	−2.6***	−0.4	2.5***

The middle section is composed of markers located within In(3R)Payne. is the proportion of variation in the most common allele frequency explained by latitude. is the proportion of variation in the second-most common allele frequency explained by latitude. II-values are ×100. *P < 0.05; **P < 0.01; ***P < 0.001.

The frequency of the SMCA was also significantly associated with latitude in the majority of markers (Table 2). Twelve markers were significantly associated with latitude after correction for multiple comparisons. Interestingly, most of these markers were significantly associated with latitude when the most common allele was used, but 4 were not, illustrating that clinal patterns in microsatellite loci can be missed if only the most common allele is considered. The 4 additional markers showing clinal patterns were located within or near each of the three broad regions of high association with latitude identified with the most common allele frequency. Spatial autocorrelation analysis (which uses all alleles) also revealed that markers within these regions had strong clinal patterns, as evidenced by their decreasing II-values with increasing distance classes (Table 2). A few markers within the three broad regions were not associated with latitude and did not have clinal autocorrelation profiles, but this was attributed to their low diversity.

As expected, the frequency of In(3R)Payne was significantly associated with latitude (F_[1,30] = 81.23, P < 0.001). To test if the clinal variation in the significantly associated markers was explained mostly by variation in In(3R)Payne frequency, we compared a regression model, with MCA frequency as the response variable (or the SMCA frequency if it was more strongly associated with latitude) and In(3R)Payne frequency and latitude as explanatory variables, to a simplified model with only In(3R)Payne frequency as the explanatory variable, using analysis of variance. We found that for five of these markers (DROABDB, DMCP017G, Hsr-omega, DMU25686, and DMTF125), removing latitude from the regression model caused a significant increase in deviance (Table 3), indicating that it explained a significant proportion of the variation in allele frequency beyond that explained by In(3R)Payne. The order in which the explanatory variables were added to the regression model did not change the total amount of variation explained and hence did not affect the outcome of these analyses.

TABLE 3.

Comparison of regression models with allele frequency as the response variable and In(3R)Payne frequency and latitude as explanatory variables, to a simplified model with only In(3R)Payne frequency as the explanatory variable using analysis of variance

Marker	Cytological position	F[29,30]	P
AC006414a	89A2	2.17	0.151
DROABDBa	89E4	11.23	0.002
3R1302339gab	90A1	0.02	0.892
DMCP017Gb	90D1	17.47	<0.001
AC009394b	90F8	2.42	0.13
DRONANOSb	91F7	0.05	0.825
AC009347b	93A3	2.22	0.147
Hsr-omegaa	93D4	19.61	<0.001
DMU25686a	93F11	6.39	0.017
AC008193a	94D2	2.05	0.163
DMU1951b	95C4	1.32	0.260
DMTF125a	95C9	11.65	0.002
3R19639736tab	95C10	1.89	0.180

^aMost common allele frequency used as response variable.

^bSecond-most common allele frequency used as response variable.

Linkage disequilibrium:

Exact tests revealed significant levels of LD between the inversion and all markers, except for two with low diversity situated inside the inversion breakpoints and two situated outside the inverted region (P < 0.05 after Bonferroni correction in all cases). Of the 325 pairwise comparisons between markers, 229 were significantly associated at the P = 0.05 level. When analyses were limited to individuals homozygous for the standard chromosome arrangement, 157 comparisons had P-values <0.05, and there were 165 comparisons with P-values <0.05 when only individuals homozygous for the inverted arrangement were considered.

As expected, the level of LD between markers and the inversion tended to be higher for markers within the inversion than for those outside it (Figure 2). Two exceptions to this pattern were the two markers located immediately outside the proximal and distal breakpoints, which had significantly higher standardized D′_m-values (nonoverlapping confidence limits) than several markers located within In(3R)Payne (Figure 2). Linkage disequilibrium also varied significantly among markers located within the inversion (Figure 2). Markers with the highest standardized D′_m-values were DROABDB, located just inside the proximal breakpoint, and Hsr-omega and DMTF125, located toward the middle of the inversion. Standardized D′_m-values were significantly lower for the remaining markers and tended to be of a similar magnitude to those for the two markers located just outside the inversion breakpoints. Exceptions to this pattern were a region of low LD near the middle of the inversion and minor peaks of high LD on the proximal side of the region of low association (Figure 2). Notably, regions with higher than average LD within In(3R)Payne had the strongest clinal polymorphisms. This relationship was confirmed by the significant association between D′_m- and R²-values derived from regressions with latitude, using the most common allele frequency for loci within the inversion (Spearman's rank correlation, r_s = 0.883, P < 0.001). This relationship was also evident when the R²-values were derived from regressions using the second most common allele frequency and when the highest R²-value for each locus (best-fitting allele frequency) was used, demonstrating the robustness of this relationship (r_s = 0.760, P = 0.001 and r_s = 0.875, P < 0.001 for the second-most common and best-fitting allele frequencies, respectively).

Figure 2.

Linkage disequilibrium between marker loci and In(3R)P. D′_m is the standardized multiallelic version of Lewontin's D′-statistic (Lewontin 1964). Open circles are markers located outside In(3R)Payne. Solid circles are markers located inside In(3R)Payne. The shaded area depicts the genetic region spanned by In(3R)Payne.

To determine the consistency of these patterns and whether they changed when the two chromosomal arrangements were analyzed separately, we calculated standardized D′_m-values between each marker and the centrally located insertion/deletion polymorphism in the Hsr-omega gene in the two homokaryotypes for each chromosome arrangement. In both chromosome arrangements, D′_m-values tended to increase with increasing distance from the Hsr-omega gene marker (as they became closer to the inversion breakpoints) but then fell away to low values for comparisons involving markers located well outside of the inverted region (>5 Mb from the Hsr-omega gene) (Figure 3). Markers inside the inverted region had significantly higher D′_m-values than those situated >1 cM (∼700 kb) outside the inversion breakpoints in both chromosome arrangements (t = 3.22, P = 0.005 and t = −5.62, P < 0.001 for the standard and inverted arrangements, respectively). This again demonstrates that LD was typically higher among markers within the inverted region, that unlinked regions within the inverted region had higher than expected LD, and that these regions are interspersed with regions of lower LD.

Figure 3.

Linkage disequilibrium between marker loci and an insertion/deletion polymorphism in the Hsr-omega gene situated near the middle of In(3R)Payne. Open circles indicate LD calculated from individuals homozygous for the standard chromosome arrangement, and solid circles indicate LD calculated from individuals homozygous for the inverted arrangement.

DISCUSSION

Our data show there were strong associations among markers within and near In(3R)Payne and between these markers and the inversion itself. Strong associations were found both in markers situated near the breakpoints and toward the middle of the inversion, where the potential for gene exchange between chromosome arrangements was higher (Hasson and Eanes 1996). Our data also show that regions of high association within the inversion were interspersed by regions of low association, indicating that genetic exchange occurs between chromosome arrangements and also suggesting that linkage disequilibrium within In(3R)Payne is being maintained by epistatic selection. A similar conclusion was reached by Schaeffer et al. (2003), who found high levels of LD among nonadjacent loci on the inversion-rich third chromosome in D. pseudoobscura.

The Schaeffer et al. (2003) study also found significant reductions in nucleotide diversity at some loci compared with interspecies divergence, suggesting that they were situated near inversion breakpoints or targets of directional selection. In this study, we found that markers within In(3R)Payne in strong LD with the inversion were those exhibiting the strongest clinal variation, resulting in a strong correlation between clinal patterns and disequilibrium. Thus, our data show that there is a direct link between patterns of LD consistent with epistatic selection and regions of the inversion that are under selection, as would be expected if the inversion was holding together favorable combinations of alleles. Furthermore, because several markers within the inversion show clinal patterns that cannot be explained by hitchhiking with the inversion (Table 3), our data suggest that genes within the inversion near these markers are the target of natural selection.

An alternate explanation for patterns of LD we observed is that they have arisen as a consequence of demographic factors such as admixture and bottlenecks associated with colonization of the Australian continent. However, there are several lines of evidence that argue against a demographic explanation. First, markers well outside the inverted region have significantly lower levels of LD than those inside it, regardless of whether LD was measured between markers and the inversion or between markers within each chromosomal arrangement. If demographic factors had been important, markers inside and outside the inversion would be expected to show similar patterns. Second, the most recent survey of a random set of microsatellites suggests that Australian populations of D. melanogaster were derived from a single population (Kennington et al. 2003). Furthermore, implementation of a model-based clustering method (Pritchard et al. 2000) using these data showed that there was no admixture within the population we used for measuring LD (i.e., the number of populations was estimated to be one). Finally, markers within In(3R)Payne showed no association with clinally varying traits other than body size (Weeks et al. 2002), whereas associations with a range of traits might be expected if demographic factors had influenced the whole genome.

Although molecular data suggest that LD extends only relatively short distances (<5 kb) in D. melanogaster (Long et al. 1998; Langley et al. 2000), LD between pairs of protein loci appears to be relatively abundant. In a metaanalysis of LD studies on Drosophila, Zapata and Alvarez (1992) found that 28% of pairs of protein loci were in significant LD and the average effective frequency of recombination was 9 cM. More recently, Zapata et al. (2002) found that 22% of pairwise comparisons among 15 widely distributed protein loci along the third chromosome were in significant LD in a natural population of D. melanogaster. Interestingly, they found that most instances of significant LD involved functionally related protein loci, suggesting that they were due to epistatic interactions. Given that the average effective recombination rate between functionally related loci was relatively large (9 cM), they further suggested that epistatic interactions must be very strong along the chromosome. By comparison, we found that 70% of the pairwise comparisons among markers were in significant LD and the average distance between markers was 14.3 cM. This again illustrates the difference in LD between loci inside and outside of inversions, even in cases where the loci outside of the inversion are under epistatic selection. High levels of LD over large distances with inversions (∼4 Mb) have also been reported recently in D. subobscura (Munté et al. 2005).

The In(3R)Payne inversion in Australia shows a strong latitudinal cline, increasing from very low frequencies in the south to near fixation in the north (Knibb et al. 1981; Anderson et al. 2005). On the basis of collections made in 2002 and 2004, Anderson et al. (2005) found that the cline in this inversion was particularly steep and had altered when compared to the cytologically defined cline described by Knibb et al. (1981) based on flies collected in 1979. The cline described here based on collections made in 2000 supports the pattern found in Anderson et al. (2005). The slope and intercept of the linear regression model in our data (slope = −0.027, intercept = 1.703) were similar to those based on collections made in 2002 and 2004 (slopes = −0.028 and −0.036, intercepts = 1.694 and 1.847 for 2002 and 2004 collections), but differed from those based on the 1979 collection (slope = −0.050, intercept = 2.057).

The patterns of gene diversity seen here indicate that the inversion influences genetic variation. Unlike the clinal pattern of heterozygosity observed in a random group of microsatellite loci (Kennington et al. 2003), there is an increase in heterozygosity in populations from intermediate latitudes, compared to a decrease at both high and low latitudes. This pattern was evident in regions adjacent to the inversion and well within the inversion, even for markers away from breakpoints. A high level of genetic variability in low-latitude populations has previously been interpreted as evidence that D. melanogaster invaded Australia from the north and then migrated southward (Kennington et al. 2003). The data presented here indicate that the impact of such population processes on genetic variation is countered by the In(3R)P inversion polymorphism. This reiterates the caution required in assuming that microsatellite patterns are neutral descriptors of population processes.

The lower level of genetic variation within the inversion compared to the standard arrangement is consistent with nucleotide sequence data indicating that In(3R)Payne has a relatively recent origin (Matzkin et al. 2005) and with the claim made by Andolfatto et al. (2001) that most inversion polymorphisms are not ancient and have ages on the order of N_e generations. In ancient inversion polymorphisms, levels of genetic variability can be similar or even higher in the chromosomal region of the inversion because the inverted and standard arrangements can represent separate gene pools (at least near the breakpoints) (Navarro et al. 2000; Andolfatto et al. 2001). We found that LD extended well into the inversion and was also detectable outside the breakpoints. This also argues against an ancient inversion. On the basis of the absence of fixed differences in the sequence variation between and within the standard and inverted sequences around the breakpoints of In(3R)Payne, Matzkin et al. (2005) further suggested that this inversion was relatively recent, although the confidence intervals for its age were large at between 0.33N_e and 4.11N_e.

The data show that chromosome arrangements are differentiated. Under Dobzhansky's model of coadaptation, alleles are adapted to act together within inversions as well as between different gene arrangements (Dobzhansky 1970). This leads to high fitness for inversion heterozygotes, at least when the heterozygotes are derived from inversion arrangements from the same population. However, In(3R)Payne frequencies range from near zero to near fixation, and such a steep cline is not expected under heterozygote advantage. It seems more likely that the inversion polymorphism is maintained because it allows combinations of alleles to be held together along the cline. As with chromosomal arrangements on the third chromosome in D. pseudoobscura, In(3R)Payne might be maintained by selection in heterogeneous environments rather than by overdominance. This could explain the lack of genetic differentiation within chromosome arrangements among populations and is consistent with high levels of gene flow along the cline (Kennington et al. 2003).

Studies investigating geographic variation in mosquitoes also show much lower differentiation among populations with the same chromosomal arrangement than between different chromosomal rearrangements from the same population (e.g., Lehmann et al. 2003; Tripet et al. 2005). These results suggest that gene flow between populations within chromosome arrangements overrides local gene exchange between arrangements, illustrating that they are effective mechanisms for maintaining combinations of alleles. Recent models have suggested that it is this role of suppressing recombination within and near inverted regions in heterozygotes, and not the fitness cost of inversion heterozygosity, that explains the potential importance of inversions in the speciation process (Rieseberg 2001; Hey 2003).

The consistency in genomic regions identified from the clinal analysis and the linkage disequilibrium analysis suggests that genes within the inversion influence quantitative traits that vary clinally. There are a number of candidate traits (Weeks et al. 2002; Anderson et al. 2003), but a particularly strong candidate is body size, which has already been associated with markers located within In(3R)Payne. Moreover, QTL mapping of strains derived from cline ends in both Australia and South America implicates this region (Gockel et al. 2002; Calboli et al. 2003). Combinations of alleles may be required to generate morphological variation along the cline, and inversions facilitate the maintenance of these allele combinations in populations. Chromosomal polymorphisms like In(3R)Payne are likely to be important as a way of allowing combinations of alleles that act together to persist and influence adaptive shifts.