DISCUSSION

Our analysis of a large number of microsatellite loci in African and non-African D. melanogaster populations indicated that the level of variability differed between X chromosomes and autosomes. While X chromosomes were significantly more variable in African populations, non-African populations had higher heterozygosities for autosomal microsatellite loci. Similar results were recently obtained from a survey of published D. melanogaster sequences (Andolfatto 2001). The sequence analysis of 40 genes in non-African populations of D. simulans, a close relative of D. melanogaster, which also recently spread from Africa, also revealed lower levels of variability on the X chromosome (Begun and Whitley 2000). This consistency of various data sets strongly suggests a common underlying biological process. In the following we discuss processes that could, in principle, explain the observed distribution of variability.

Neutral explanations (demographic effects): Sex ratio: Male D. melanogaster have one X chromosome, while females carry two X chromosomes. Thus, the effective population size of X chromosomes is dependent on the variance of reproductive success of males and females. For African (Wuet al. 1995; Capyet al. 2000) and non-African (Korolet al. 2000) populations, mate selection has been reported. Furthermore, it has been proposed that due to mate selection X chromosomes can harbor ~90% (rather than 75%) of the autosomal variability (Nunney 1993; Charlesworth 2001). At least in European D. melanogaster populations, the effect could be countered by large variation in reproductive success among females (Bouletreau 1978). In summary, there is substantial evidence for variance in reproductive success in D. melanogaster, but the extent to which differential reproductive success skews the ratio of X chromosomal to autosomal variability is simply not known.

Our data indicate that, in non-African populations, levels of variability were no longer significantly different among chromosomes when a balanced sex ratio is assumed. For African X chromosomes we found that neither a balanced sex ratio nor a fivefold excess of females could explain the higher variability on African X chromosomes. On the basis of the available data, it is not possible to decide whether the sex ratio in African D. melanogaster populations is even more biased than we assumed. Nevertheless, a fivefold excess of females is more conservative than in previous reports, which suggested that X chromosomal variability should be 90% of the autosomal variability (Nunney 1993; Charlesworth 2001), which corresponds to a three fold reduction in male effective population size.

Founder effect: It is assumed that D. melanogaster started from Africa to colonize the rest of the world ~10,000 years ago (David and Capy 1988). Such founder effects are often associated with a loss of genetic variability. Because our data also indicated a loss of variability in the non-African populations, we were interested whether a bottleneck could explain the loss of variability. We pursued two different approaches to test for a bottle-neck in the non-African populations.

Our first test takes advantage of the well-investigated relationship between the number of alleles and heterozygosity. After a bottleneck, heterozygosity excess is expected at neutral loci (Cornuet and Luikart 1996). We used the three different test statistics, sign test, standardized difference test, and Wilcoxon test, all implemented in the Bottleneck 1.2 software (Cornuet and Luikart 1996). Irrespective of the mutation model used (strict stepwise or two-phase model with 30% nonstepwise mutations) we did not detect a significant heterozygosity excess in the non-African populations (P > 0.05).

The second test was based on the property of bottle-necks that they are a genome-wide phenomenon. Hence, if a founder event caused the reduction in variability, then all microsatellite loci located on the same chromosome should be affected to the same extent. To test this, we plotted the observed heterozygosity in non-African and African populations against recombination rate. If the reduction in variability were entirely due to a bottleneck, no correlation between variability and recombination rate should be detected. Interestingly, we observed a significant correlation between recombination rate and heterozygosity on the X chromosome only in non-African populations (non-Africa r = 0.49, P = 0.001; Africa r = 0.08, P = 0.64; Spearman rank correlation; Figure 2a). The same effect can also be seen for V_LC in non-African populations (r = 0.43, P = 0.005; Spearman rank correlation), where, in contrast, a marginally significant correlation is also observed in African populations (r = 0.31, P = 0.056; Spearman rank correlation). On the autosomes, no correlation could be detected (non-Africa r = −0.04, P = 0.69; Africa r = −0.01, P = 0.95; Spearman rank correlation; Figure 2b). This test shows that simple changes of population size cannot explain our data. Furthermore, even a reduction in the effective female population size would not have produced a correlation between variability and recombination rate. The correlation between recombination and levels of variability has been explained by selective sweeps of beneficial mutations (Begun and Aquadro 1992) or background selection (Charlesworthet al. 1993). Our observation that a significant correlation exists on the X chromosome suggests that selection has a greater effect on the non-African X chromosome than on autosomes, but provides no further insight whether the fixation of beneficial or the removal of deleterious mutations is determining the pattern of variability.

Differential selective forces are operating in African and non-African populations: Any study attempting to explain the impact of selection on autosomal and X chromosomal genes has to consider the effect of dominance. While the dynamics of dominant mutations will not differ substantially among X chromosomes and autosomes, recessive mutations will be selected more efficiently when located on the X chromosome. Recently, Begun and Whitley (2000) reviewed the consequences of dominance effects for levels of variability under various selection regimes. For the remainder of the text, however, we assume that mutations are recessive.

Evidence for multiple beneficial mutations on non-African X chromosomes: The genetic consequences of exposure to a novel habitat are well investigated in simple organisms, such as Escherichia coli and yeast. Habitat shifts were shown to increase the fixation rate of beneficial mutations, leading to a higher fitness in the new environment (Novick and Szilard 1950; Lenski and Travisano 1994; Imhof and Schlötterer 2001). Similarly, the colonization of more temperate environments outside of Africa has undoubtedly resulted in an increased potential for the fixation of beneficial mutations in D. melanogaster. In principle, two different scenarios have to be distinguished. First, the beneficial mutations are novel and occurred during the colonization event. Second, the beneficial mutations were already present in the founder populations, but only with the habitat shift did these mutations become advantageous. While, for the first scenario, lower X-linked variability could be assumed (Charlesworthet al. 1987), the predictions are less clear for the second model. Orr and Betancourt (2001) considered a sudden environmental change, which rendered previously deleterious alleles (in mutation-selection equilibrium) beneficial (Orr and Betancourt 2001). Under this model, the authors showed that a stronger reduction in variability on the X chromosome may or may not be expected if adaptive mutations involve previously deleterious alleles.

Until now, literally nothing has been known about mutations that are beneficial in a novel environment and the characterization of those mutations is among the most exciting challenges to evolutionary biologists. Despite lacking a clear prediction on the nature of beneficial mutations, we tested whether the pattern of variability is consistent with beneficial mutations reducing X chromosomal variation in non-African populations. Because microsatellite loci are neutral markers, reduced microsatellite variability at a given locus is most likely caused by linkage to a selected site. Thus, the prediction is that some microsatellites on the X chromosome will be linked to a beneficial mutation, while others are not linked. Given that many beneficial mutations are required to reduce the variability on the X chromosome, we expect a reduced variability at several microsatellite loci. Consequently, microsatellite variability should show large variation among loci. Because autosomes did not experience a strong reduction in variability, fewer microsatellite loci are expected to be associated with a beneficial mutation. Hence, the variation in variability among loci is expected to be less pronounced among autosomal loci than among X chromosomal loci. To test this we calculated the variance of RV_LC (RV_LC = V_nA/V_Afr) values for X chromosomal and autosomal loci. To obtain a confidence interval for this variance we generated 100 pseudoreplicas by bootstrapping the RV values for each group of loci. Figure 3 shows the mean of the variance of RV_LC and its 95% confidence interval. This value reflects the variation of RV_LC values on X chromosomes and autosomes. Consistent with a higher impact of selective sweeps on the X chromosome, we found a larger variance of RV_LC values on the X chromosome. Nevertheless, this result does not indicate whether the beneficial mutations occurring on the X chromosome were already segregating in the African population or are of recent origin. However, it should be noted that our data do not suggest that a larger number of beneficial mutations occurred on the X chromosome than on autosomes. Due only to the larger hitchhiking effect of beneficial mutations on the X chromosome is their presence more obvious. We assume that a similar number of beneficial mutations (per gene) also occurred on the autosomes, but a much smaller genomic region has hitchhiked with them.

• Download figure
• Open in new tab
• Download powerpoint

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Heterozygosity of non-African and African populations plotted against recombination rate [recombination events/(kilobase × generation)]. Crosses and the dashed line represent African, and dots and the solid line non-African, populations. To account for the absence of recombination in males, we corrected the recombination rates of autosomes and X chromosomes, assuming a 1:1 sex ratio. (A) A significant correlation between heterozygosity and recombination rate was observed for non-African X chromosomal loci (non-Africa r = 0.49, P = 0.001; Africa r = 0.08, P = 0.64; Spearman rank correlation). (B) On the autosomes no correlation could be detected (non-Africa r = −0.04, P = 0.69; Africa r = −0.01, P = 0.95; Spearman rank correlation).

The pattern of variability in African populations: One further important result of our survey is that in African populations X chromosomal microsatellites have a significantly higher variability than autosomes. Recently, Andolfatto (2001), who obtained a similar pattern in a survey of published sequence data, proposed that autosomal variability in African populations could be depressed by a recent increase in frequency of chromosomal inversions in these populations. In fact several inversions have been found to form clines with higher frequencies at low latitudes near the equator (Lemeunier and Aulard 1992). Navarro et al. (2000) found that variability levels within inversions could be reduced as long as the inversion has not reached equilibrium (Navarroet al. 2000). Experimental data, however, indicate that a reduction of variability is confined to the proximity of the inversion breakpoints. No reduced variability was detected in the middle of inversions (Hasson and Eanes 1996; Aguadé 1998, 1999; Andolfattoet al. 1999; Bénassiet al. 1999; Depauliset al. 1999). Furthermore, an interesting finding of Andolfatto (2001) was that D. simulans, a sibling species of D. melanogaster, which shows very few inversion polymorphisms, showed no elevated X-linked variability in African populations.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Mean variance of 100 resamples of RV_LC (RV_LC = V_LCnA/V_LCAfr) values for X chromosomes and autosomes. The error bars indicate the 95% confidence interval of the mean RV_LC.

Given the possible implications of the data, we tested for an influence of common cosmopolitan inversions, for which clines with high frequencies in African populations were described. Because most of the lines used were not kept alive, we did not determine the inversion status of each line. Thus, the inversion frequencies in our African samples are not known. However, if these inversions have affected our variability estimates, this should be recognizable even without the knowledge of the inversion status of each of the lines analyzed. We split the autosomal microsatellites into loci, which are located outside and within genomic regions harboring these inversions. No loci were found to map to the inversion breakpoints. Our microsatellite set contained only three of the five inversion locations known to reach high frequencies in Africa. The chromosomal locations of the inversions, as well as the microsatellite variabilities within and outside these regions, are given in Table 5. In contrast to the expectations for recently swept inversions, we did not detect any evidence for a reduction in variability of microsatellites located within the chromosomal regions to be possibly an inversion. This observation holds irrespective of whether we analyzed microsatellites within each inversion individually or all three inversions jointly and compared them to all microsatellites located outside of the inversion regions (Table 5).

While the separate analysis of microsatellite loci within inversions and outside inversions seems not to support the hypothesis of Andolfatto (2001), we note that our test is rather crude and restricted to the impact of common cosmopolitan inversions. However, an alternative explanation of the data may be that the pattern of variability in African populations is consistent with background selection. According to the background selection model, X chromosomes should harbor more neutral variation than autosomes (Charlesworthet al. 1987). Recessive deleterious alleles are purged more efficiently from X chromosomes and, thus, fewer neutral alleles are removed together with the deleterious alleles.