A Test for Faster X Evolution in Drosophila

Charlesworth, Coyne, and Barton (1987) showed that, under certain conditions, loci on the X chromosome are expected to have higher rates of adaptive evolution than those on the autosomes. In particular, if the average beneficial mutation is both new (i.e., not a previously deleterious allele segregating at mutation-selection balance; see Orr and Betancourt 2001 ) and at least partially recessive (mean h < 0.5), X-linked loci will evolve faster (Charlesworth, Coyne, and Barton 1987 ). (This calculation assumes that the distribution of selection coefficients does not systematically differ between X-linked and autosomal loci, as seems reasonable.) Although X-linked loci, having smaller population sizes than do autosomal loci, experience fewer beneficial mutations per generation, a greater fraction of these mutations is fixed by hemizygous selection in the XY sex (we assume hereafter that males are the XY sex). When h = 0.5, the two forces cancel each other, and X-linked and autosomal loci evolve at the same rate. When h < 0.5, X-linked loci evolve faster. The resulting “faster X” effect is further enhanced for genes with male-specific expression. The faster X theory has been proposed as a possible explanation of a number of biological phenomena, such as the disproportionately large effect of the X chromosome on hybrid sterility and inviability (Coyne and Orr 1989 ; Orr and Coyne 1989 ) and on morphological and behavioral species differences, especially in Lepidoptera (Sperling 1994 ; Prowell 1998 ; Reinhold 1998 ). More recently, faster X evolution was suggested as a possible explanation of an important finding in molecular population genetics. In a survey of 40 loci, Begun and Whitley (2000) showed that loci on the X chromosome of Drosophila simulans harbor about half the neutral polymorphism (adjusted for mutation rate) of loci on an autosome. This study and others like it (e.g., Andolfatto 2001 ; Kauer et al. 2002 ) are important because they may provide a means to distinguish between the recurrent hitchhiking (Maynard Smith and Haigh 1974 ; Kaplan, Hudson, and Langley 1989 ; Wiehe and Stephan 1993 ) and background selection models (Charlesworth, Morgan, and Charlesworth 1993 ; Charlesworth 1994 ). Although both models can explain the correlation between recombination and neutral variation (Begun and Aquadro 1992 ; Nachman 1997 ; Nachman et al. 1998 ; Stephan and Langley 1998 ), they make different predictions about patterns of variation on the X versus the autosomes. The background selection model predicts a relative excess of neutral polymorphism on the X (Aquadro, Begun, and Kindahl 1994 ; Begun and Whitley 2000 ). The reason is that the effects of background selection—the reduction of levels of linked neutral polymorphism caused by the recurrent production and elimination of deleterious mutations—are strongest when deleterious mutations reach appreciable frequencies before ultimately being removed from a population. But because hemizygous selection eliminates recessive deleterious mutations quickly, the X has a larger proportion of mutation-free chromosomes than do the autosomes and, thus, greater neutral polymorphism (Aquadro, Begun, and Kindahl 1994 ; Begun and Whitley 2000 ). The recurrent hitchhiking model, on the other hand, might predict a relative dearth of neutral polymorphism on the X for two reasons. First, there may be reduced opportunity for recombination on the X during a selective sweep because X-linked alleles fix more quickly: when h = 0.5, transit times on the X are approximately three fourths those on the autosomes (Avery 1984 ), and this disparity is even greater for recessive favorable mutations. Second, if the conditions of Charlesworth, Coyne, and Barton (1987) for faster X evolution hold in natural populations (i.e., mean h < 0.5 for new beneficial mutations), there will be more selective sweeps per locus per generation on the X than those on the autosomes. Here, we use DNA divergence data between D. melanogaster and D. simulans to test for faster X evolution. This test of the faster X theory also provides a useful glimpse of the distribution of dominance coefficients of new beneficial mutations (see Charlesworth 1992 ; Bourguet et al. 1997 ; Orr and Betancourt 2001 ). The data consist of 254 published sequences, many of which are from a male-specific expressed sequence tag (EST) screen (Swanson et al. 2001 ), with average lengths of 749.6 ± 71.1 bp and 512.7 ± 32.9 bp (x ± 1 SE; SE refers to standard error) for X-linked and autosomal loci, respectively. We calculated maximum-likelihood estimates of silent- and amino acid–site divergence using PAML (Yang 2000 ) and used map positions from Flybase (http://flybase.bio.indiana.edu/) to determine whether genes are X linked or autosomal. (All data are included in an online supplementary table at the MBE website: www.molbiolevol.org.) Because Charlesworth, Coyne, and Barton (1987) showed that the faster X effect is greater for loci with male-specific fitness effects, we also test the effect of sex-limited expression on X versus autosomal substitution rates. Below, we assume that selection acts equally on both sexes at loci expressed in both sexes and that selection acts only on males at loci with male-limited expression. We obtained expression data from Flybase and assumed that genes encoding accessory gland proteins (Acps) and genes from the male-specific EST screen reported by Swanson et al. (2001) for which expression data were otherwise unavailable were male-limited in expression. (We have too few female-specific genes to perform a test using only these.) Unfortunately, many expression data are of somewhat limited resolution. For example, some of the male-specific ESTs are not perfectly male limited—at least 39 of the 171 genes from the screen are reported elsewhere to show at least some expression in both sexes. Therefore, given the inherent limitations of these data, the classification of gene expression used in this study is necessarily imperfect. Because the tests below depend mainly on the differences between X-linked and autosomal loci, and less on differences in sex-limited expression, the crudeness of this classification scheme is unlikely to compromise the results. To compare rates of evolution, we first compared the rates of sequence divergence for X-linked and autosomal loci using all loci in our data set. Importantly, we can rule out any difference in mutational input because we find no difference in the rate of (presumably nearly neutral) silent-site substitution (dS) between the X and autosomes (table 1 ). This finding also provides further evidence that there is no sex-biased mutation rate in Drosophila, confirming the findings of an earlier study that used a smaller sample of loci (Bauer and Aquadro 1997 ). Thus, any X-autosome difference in rates of amino acid substitution should reflect a difference in the efficacy of natural selection. But, as table 1 shows, there is no difference in the rate of amino acid substitution between X-linked and autosomal loci (using dN or dN/dS). Even at loci with male-limited expression, for which theory predicts the faster X effect to be slightly stronger, we find no evidence for faster X evolution (table 1 ). In fact, it appears, if anything, that autosomal loci may evolve slightly faster (though not significantly so). It is important, however, to note that these data are not a random sample of genes: the EST screen from which most of our genes are derived was designed to enrich for male-specific transcripts, many of which are rapidly evolving putative Acps (Swanson et al. 2001 ). Because disproportionately few Acps are found on the X (Swanson et al. 2001 ), analyses that include Acps may be biased against detecting faster X evolution. We therefore removed all putative Acps from our data set and repeated the analyses on the remaining 185 genes. Using this reduced data set, we find that the rates of evolution of X-linked and autosomal loci remain similar and do not differ significantly (table 2 ). We therefore conclude that there is no evidence for faster X evolution, at least in the present data set. Our tests, however, are limited and come with several caveats. First, because we eliminated many rapidly evolving genes (the Acps) in the second part of our analysis, much of the remaining amino acid divergence could, in principle, be neutral. If so, failure to find an X-autosome difference is not surprising because neutral divergence should accumulate at equal rates on the X and the autosomes. This scenario, however, seems unlikely. A substantial fraction of amino acid substitutions in Drosophila appear to be adaptive, even for genes that individually do not show statistical evidence for positive selection (Smith and Eyre-Walker 2002 ). Second, we may have simply missed the relevant loci. For instance, our survey includes only coding sequences; the results may differ for cis-regulatory elements. Thus, we cannot rule out that a comparison of all potential X-linked targets of selection against all autosomal targets would reveal a faster X effect. Third, a recent analysis comparing the divergence of duplicate genes on the X versus autosomes gives some indication of a faster X effect (Thornton and Long 2002 ). The disparity between the results of Thornton and Long (2002) and those presented in this study may reflect interesting differences in the distribution of dominance effects between duplicate and single-copy genes. Specifically, beneficial mutations arising in duplicate genes may be more recessive than those in single-copy genes, as seems plausible. If the absence of faster X evolution holds in larger data sets, several implications follow. For one, depressed polymorphism on the D. simulans X may still be caused by increased hitchhiking effects at X-linked loci but because of reduced transit times (Avery 1984 ; Begun and Whitley 2000 ) rather than higher substitution rates. Alternatively, other, non–selection-based explanations such as demographic effects might account for the pattern (J. Wall, P. Andolfatto, M. Przeworski, personal communication). Most important, the fact that X-linked and autosomal loci evolve at the same rate means that either most replacement substitutions are neutral, not beneficial, or the average beneficial allele, if new, is not a partially recessive mutation but is likely close to additive (h ≈ 0.5). We suspect that the second interpretation is more plausible and that there must be other explanations for the large X effect. Wolfgang Stephan, Reviewing Editor Keywords: adaptation faster X hitchhiking X chromosome Address for correspondence and reprints: Andrea J. Betancourt, Department of Biology, University of Rochester, Rochester, New York 14627. aabt@mail.rochester.edu Table 1 Comparison of Evolutionary Rates for X versus Autosomal Loci Table 1 Comparison of Evolutionary Rates for X versus Autosomal Loci Table 2 Comparison of Evolutionary Rates for X versus Autosomal Loci, Excluding Putative Acps Table 2 Comparison of Evolutionary Rates for X versus Autosomal Loci, Excluding Putative Acps We thank P. Andolfatto, J. Bollback, J. Huelsenbeck, A. Orr, W. Stephan, and two anonymous reviewers for helpful comments and discussion. This work was supported by funds from NIH grant GM526738 and the David and Lucile Packard Foundation to A. Orr, and by Caspari Fellowships and Messersmith Fellowships to A.J.B and D.C.P.