1. Introduction
In species with X Y or W Z sex determination, positive selection may be more effective at fixing favourable mutations that arise on the X or Z chromosome compared with the autosomes, because rare recessive or partially recessive mutations are fully expressed in the heterogametic sex, whereas in the homogametic sex they are mostly present as heterozygotes with wild-type alleles (Haldane, Reference Haldane1924). This can result in a higher rate of substitution of beneficial mutations at X-linked or Z-linked loci, provided that relevant conditions on dominance coefficients, dosage compensation, sex-specific mutation rates and initial frequencies of the mutations are met (Rice, Reference Rice1984; Charlesworth et al., Reference Charlesworth, Coyne and Barton1987; Orr & Betancourt, Reference Orr and Betancourt2001; Kirkpatrick & Hall, Reference Kirkpatrick and Hall2004; for a recent review, see Vicoso & Charlesworth, Reference Vicoso and Charlesworth2006). In contrast, under most conditions, recessive or partially recessive slightly deleterious mutations will experience a faster rate of substitution on the autosomes than on the X or Z chromosome, as a result of less effective selection against them (Charlesworth et al., Reference Charlesworth, Coyne and Barton1987; McVean & Charlesworth, Reference McVean and Charlesworth1999).
A major factor in determining whether X-linked loci will evolve faster or slower than autosomal loci is the fraction of mutations that are fixed by positive selection versus genetic drift (from now on, we will simply refer to X chromosomes, since the same results apply to Z chromosomes with a switch of gender). Recent studies of Drosophila suggest that 25–50% of divergent non-synonymous sites among related species were fixed by positive selection (Smith & Eyre-Walker, Reference Smith and Eyre-Walker2002; Bierne & Eyre-Walker, Reference Bierne and Eyre-Walker2004; Welch, Reference Welch2006; Andolfatto, Reference Andolfatto2007; Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007). Accordingly, we might expect to observe faster-X evolution, provided that new beneficial mutations are, on average, at least partially recessive. There is little relevant information on the levels of dominance of beneficial mutations, although indirect evidence from the genetics of species differences in highly selfing taxa of plants (Charlesworth, Reference Charlesworth1992), and from comparisons of rates of adaptive evolution of haploid and diploid laboratory populations of yeast (Zeyl et al., Reference Zeyl, Vanderford and Carter2003; Anderson et al., Reference Anderson, Sirjusingh and Ricker2004), is consistent with a predominance of at least partial recessivity of new, selectively favourable mutations.
The current availability of large DNA sequence datasets has facilitated comparative analyses of the rates of molecular evolution of the X chromosome and the autosomes. In particular, from between-species comparisons, we can estimate K a and K s, the rates of non-synonymous and synonymous divergence per nucleotide site, respectively. It is commonly assumed that K s mostly reflects nearly neutral or neutral evolution, and the ratio K a/K s is used to estimate the overall effect of selective forces. Neutral sequences evolve with K a/K s≈1, and negative (purifying) selection reduces this ratio, whereas recurrent positive selection increases it (Graur & Li, Reference Graur and Li2000). If positive selection is more effective on the X chromosome, we might thus expect X-linked sites to show higher K a/K s values than autosomal sites. This test can be combined with other evidence, such as the McDonald–Kreitman test (McDonald & Kreitman, Reference McDonald and Kreitman1991), to discriminate between alternative interpretations of differences in K a and K s between X chromosomes and autosomes, such as increased positive selection, relaxation of constraint and differences in mutation rates. Currently, the evidence for higher K a/K s on the Drosophila X chromosome is conflicting, with some studies finding evidence for adaptively driven faster-X effects (Thornton & Long, Reference Thornton and Long2002, Reference Thornton and Long2005; Clark et al., Reference Clark, Eisen, Smith, Bergman, Oliver and Markow2007; Singh et al., Reference Singh, Larracuente and Clark2007), and others failing to detect such effects (Betancourt et al., Reference Betancourt, Presgraves and Swanson2002; Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007; Connallon, Reference Connallon2007).
The drawback of this approach is that the X chromosome differs considerably in its gene content from the autosomes; for example, in Drosophila, male-biased genes are rarely found on the X (Parisi et al., Reference Parisi, Nuttall, Naiman, Bouffard, Malley, Andrews, Eastman and Oliver2003; Sturgill et al., Reference Sturgill, Zhang, Parisi and Oliver2007), and this could lead to systematic biases in the mean sex-specificity of selection coefficients of X-linked and autosomal mutations. Since the expected values of K a and K s for the X versus the autosomes are strongly affected by such sex-specificity (Rice, Reference Rice1984; Charlesworth et al., Reference Charlesworth, Coyne and Barton1987; Vicoso & Charlesworth, Reference Vicoso and Charlesworth2006), differences among chromosomes in these coefficients could mask an underlying faster-X effect, and this may be of concern in some of the previous studies. If genes with similar functions have similar selection coefficients, then focusing on gene groups with similar expected sex-specific effects could reveal faster-X evolution, as in the case of mammalian sperm proteins (Torgerson & Singh, Reference Torgerson and Singh2003) and Drosophila melanogaster sex-specific genes (Pröschel et al., Reference Pröschel, Zhang and Parsch2006; Baines et al., Reference Baines, Sawyer, Hartl and Parsch2008).
Another approach is to study the same group of genes in an autosomal and an X-linked context. The genus Drosophila is particularly favourable for this purpose, since its species vary both in the number and organization of their chromosomes. Muller (Reference Muller and Huxley1940) noted that genes linked in one species also appeared to be linked in others, and proposed that all karyotypic differences among Drosophila species could be explained by different combinations of the six basic chromosomal arms. The chromosomal arms involved have become known as Muller's elements A–F. Comparative analyses of Drosophila genomes have confirmed that, despite extensive within-arm rearrangements, only small fragments of DNA have been transposed between arms (Powell, Reference Powell1997; Richards et al., Reference Richards, Liu, Bettencourt, Hradecky, Letovsky, Nielsen, Thornton, Hubisz, Chen, Meisel, Couronne, Hua, Smith, Zhang, Liu, Bussemaker, van Batenburg, Howells, Scherer, Sodergren, Matthews, Crosby, Schroeder, Ortiz-Barrientos, Rives, Metzker, Muzny, Scott, Steffen, Wheeler, Worley, Havlak, Durbin, Egan, Gill, Hume, Morgan, Miner, Hamilton, Huang, Waldron, Verduzco, Clerc-Blankenburg, Dubchak, Noor, Anderson, White, Clark, Schaeffer, Gelbart, Weinstock and Gibbs2005; Clark et al., Reference Clark, Eisen, Smith, Bergman, Oliver and Markow2007).
A rearrangement that is particularly useful for investigating faster-X effects is found in species of the Drosophila pseudoobscura clade. Following the split from the D. melanogaster group and their sister Drosophila obscura clade, the D. pseudoobscura clade ancestor became fixed for a fusion between Muller's element D (the autosomal 3L arm of D. melanogaster) and the homologue of the D. melanogaster X chromosome (element A), to form, respectively, the R and L arms of the D. pseudoobscura X (Muller, Reference Muller and Huxley1940). The forces shaping the evolution of the X chromosome should, therefore, also be acting on this new R arm of the D. pseudoobscura X chromosome. It should be noted that the XR arm has already acquired dosage compensation, and transcription levels are therefore similar on both arms of the X chromosome (Abraham & Lucchesi, Reference Abraham and Lucchesi1974; Steinemann et al., Reference Steinemann, Steinemann and Turner1996).
This system was used to explore the question of faster-X evolution by Counterman et al. (Reference Counterman, Ortiz-Barrientos and Noor2004), who compared the K a/K s values of genes on element D for D. pseudoobscura/D. melanogaster with those for Drosophila simulans/D. melanogaster. As expected under faster-X evolution, they found a significant excess of 3L-XR genes with higher K a/K s for the D. pseudoobscura/D. melanogaster comparison relative to the D. simulans/D. melanogaster comparison. However, Musters et al. (Reference Musters, Huntley and Singh2006) found no difference between element D genes and autosomal genes with respect to K a using D. pseudoobscura/D. melanogaster genomewide data. Counterman et al. (Reference Counterman, Ortiz-Barrientos and Noor2004) also examined two different pairs of species, D. melanogaster/D. simulans and D. pseudoobscura/Drosophila miranda. D. miranda is a close relative of D. pseudoobscura and shares the new XR (Muller, Reference Muller and Huxley1940; Steinemann & Steinemann, Reference Steinemann and Steinemann1998). They found that the fraction of genes with higher K a/K s in the D. pseudoobscura/D. miranda pair than in D. melanogaster/D. simulans was larger for 3L-XR genes, although not significantly so. Thornton et al. (Reference Thornton, Bachtrog and Andolfatto2006) used a similar approach with a larger dataset, but found no evidence for faster-X evolution. The lack of statistically significant evidence for faster-X effects in this case could be due to the species pairs used, since they have very low levels of divergence (about 4% for D. pseudoobscura/D. miranda at synonymous sites in Bartolomé et al., Reference Bartolomé, Maside, Yi, Grant and Charlesworth2005). As such they may not be ideal for sequence comparisons, especially as some apparent inter-species differences may reflect polymorphisms within species (Bartolomé & Charlesworth, Reference Bartolomé and Charlesworth2006). Similarly, the evidence for faster-X evolution found in a whole genome comparison using D. melanogaster/D. simulans and D. pseudoobscura/Drosophila persimilis should be interpreted with caution, as D. persimilis is even closer to D. pseudoobscura than D. miranda (Singh et al., Reference Singh, Larracuente and Clark2007), so that effects of polymorphism levels may be confounded with true divergence.
Accordingly, we have used sequence data from the more distant species Drosophila affinis, with an average K s of about 23% from both D. pseudoobscura and D. miranda (Bartolomé et al., Reference Bartolomé, Maside, Yi, Grant and Charlesworth2005; Bartolomé & Charlesworth, Reference Bartolomé and Charlesworth2006), in order to obtain more reliable estimates of sequence divergence. We used D. melanogaster and Drosophila yakuba for the control comparison, since their mean synonymous divergence is similar to that for D. pseudoobscura and D. affinis (see Table 1 below).
Average rates of DNA sequence evolution for the species pairs D. pseudoobscura/D. affinis and D. melanogaster/D. yakuba
The average K a/K s is estimated from the ratio of the averages of K a and K s.
a S.E. values are given in parentheses.
2. Materials and Methods
(i) Gene selection
Our sample consisted of D. pseudoobscura, D. affinis, D. melanogaster and D. yakuba coding sequences for 69 3L-XR genes: 39 genes that are autosomal in both groups and 27 genes that are X-linked in both groups (from the X-XL chromosomal arm).
Carolina Bartolomé provided the coding sequence for the 39 autosomal D. affinis genes, for seven X-XL genes and for three 3L-XR genes (Bartolomé et al., Reference Bartolomé, Maside, Yi, Grant and Charlesworth2005; Bartolomé & Charlesworth, Reference Bartolomé and Charlesworth2006). The homologous sequences from other species were retrieved directly from the published genomes of D. melanogaster, D. yakuba and D. pseudoobscura with the NCBI Blast algorithm (http://www.ncbi.nlm.nih.gov/BLAST/).
For the other genes, D. melanogaster protein coding genes were downloaded from the FlyBase website (http://www.flybase.org). To minimize possible effects of close linkage to genes under selection, they were all chosen from regions of normal recombination in D. melanogaster (cytological region 3C3-15F3 for the X chromosome and 62A12-71A1 for 3L, as described by Charlesworth (Reference Charlesworth1996) ).
For each gene, we recovered all the corresponding mRNAs in the NCBI database with the Megablast algorithm (http://www.ncbi.nlm.nih.gov/BLAST/), and verified that they had a size between 1000 and 3500 base-pairs (bp), with at least 1000 bp without introns. We identified the D. yakuba homologue through the UCSC BLAT server (http://genome.ucsc.edu/cgi-bin/hgBlat?command=start) and the D. pseudoobscura homologue through the NCBI BLAST, and kept only genes whose location was syntenic for all three species.
(ii) DNA extraction
DNA was extracted from males of a D. affinis line originally collected from Nebraska (stock number 0141·2; Drosophila Species Resource Center) using a Qiagen DNA extraction kit (Qiagen, Crawley, West Sussex, UK).
(iii) DNA sequencing
Primers were designed with the DNAstar package and the Primer3 software (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi), using the D. pseudoobscura sequence to amplify 1000–1300 bp of the gene from D. affinis. Additional internal primers were designed for sequencing. Since the D. affinis sequences of the 39 autosomal genes we used were provided by Carolina Bartolomé (Bartolomé et al., Reference Bartolomé, Maside, Yi, Grant and Charlesworth2005; Bartolomé & Charlesworth, Reference Bartolomé and Charlesworth2006), all the genes we sequenced were on the D. pseudoobscura/D. affinis X chromosome (66 on the 3L-XR arm and 20 on the X-XL arm). PCR products were therefore directly sequenced on both strands using the BigDye (version 3) sequencing kit and run on an ABI 3730 Genetic Analyser (Applied Biosystems, Foster City, CA, USA) by the sequencing service of the School of Biological Sciences, University of Edinburgh. The sequences generated as part of this study have been deposited in the GenBank Data Libraries under accession numbers EU931120–EU931205.
(iv) Estimation of Ka and Ks
All sequences were translated and virtual protein sequences were aligned with the European Bioinformatics Institute ClustalW interface (http://www.ebi.ac.uk/Tools/clustalw/index.html). The resulting alignment was used to align the DNA sequences with Tranalign (http://phytophthora.vbi.vt.edu/cgi-bin/emboss.pl?_action=input&_app=tranalign), which aligns coding DNA according to a protein alignment. The K s and K a were calculated using Nei & Gojobori's (Reference Nei and Gojobori1986) model of substitution (Nei & Gojobori, Reference Nei and Gojobori1986), implemented in DnaSP version 4.50 (Rozas et al., Reference Rozas, Sánchez-DelBarrio, Messeguer and Rozas2003; http://www.ub.es/dnasp/), with the Jukes–Cantor correction for multiple hits. Since several models of substitution can lead to artifactual biases in K s when there are differences in codon usage bias (Bierne & Eyre-Walker, Reference Bierne and Eyre-Walker2003), we also analysed the data using the Goldman & Yang (Reference Goldman and Yang1994) model of substitution (using the PAML software package: http://abacus.gene.ucl.ac.uk/software/paml.html). The results from tests using the Goldman & Young (1994) d N and d S measures are given in the Supplementary Material.
(v) Codon usage
The alignments obtained for the K a and K s analyses were used to estimate the frequency of optimal codons, Fop, using CodonW (http://codonw.sourceforge.net/). We used the D. melanogaster table of preferred codons (Shields et al., Reference Shields, Sharp, Higgins and Wright1988), as patterns of codon usage have been shown to be highly conserved between Drosophila species (Powell & Moriyama, Reference Powell and Moriyama1997; Clark et al., Reference Clark, Eisen, Smith, Bergman, Oliver and Markow2007), and we needed to have the same set of codons for the different species under comparison.
(vi) Statistical analysis
Descriptive statistics and statistical tests were carried out using the StatView software (version 4.5, http://www.statview.com).
3. Results
(i) Within-clade comparisons
We obtained and aligned sequences for 69 3L-XR and 66 non-3L-XR (27 X-XL and 39 autosomal) genes for the species pairs D. pseudoobscura/D. affinis and D. melanogaster/D. yakuba (see Materials and Methods section). The average K a, K s and K a/K s values are shown in Table 1 and Fig. 1 (the values for individual genes are given in the Supplementary Material). Overall, the mean values seem to support the faster-X hypothesis: both K a and K a/K s values are higher for X-linked chromosomal arms than autosomal arms in both the D. pseudoobscura and D. melanogaster groups.
Mean K a, K s and K a/K s for 3L-XR, X-XL and autosomal genes in the two clades.
It should, however, be noted that, while the higher K a and K a/K s of the 3L-XR genes in the D. pseudoobscura group is in apparent agreement with the faster-X hypothesis, these genes also exhibit particularly high K a and K a/K s values in the D. melanogaster group when compared with the rest of the autosomes (Table 2 and Fig. 1). This is likely to be caused by a sampling bias, as most of the autosomal genes were chosen from long, well-studied genes (Bartolomé et al., Reference Bartolomé, Maside, Yi, Grant and Charlesworth2005). In contrast, the genes that we selected for sequencing (most of the 3L-XRsample) correspond to small, unnamed (mostly unstudied) transcripts with no known function. Genes with no annotated function have been shown to be less constrained than genes with known functions (Clark et al., Reference Clark, Eisen, Smith, Bergman, Oliver and Markow2007). Consistent with this, pairwise Mann–Whitney tests (Table 2) show that the autosomal sample has significantly lower K a values than the 3L-XR sample, and a lower K a/K s in the D. pseudoobscura/D. affinis comparison. This should, however, not affect the comparison of rates of evolution on the same chromosomal arm between the two clades, since we have the same set of genes in all four species.
Significance values for comparisons of different chromosomal arms within species pairs
Significant P-values are shown in boldface.
(ii) Lower Ks for X-linked genes
Further examination of Fig. 1 shows that the most striking pattern is the smaller K s for the 3L-XR genes in the D. pseudoobscura clade compared with the D. melanogaster clade, whereas other chromosomal arms show no significant differences in K s. A similar phenomenon was described in a human–chimpanzee comparison by Lu & Wu (Reference Lu and Wu2005), who found that X-linked genes had significantly lower K s values than autosomal genes. However, this is not seen for X-linked genes in comparisons among members of the D. melanogaster clade in our data, nor in a whole-genome analysis (Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007). Since the corresponding K a values do not differ significantly among the two clades, this comparison indicates that the faster-X effect suggested by the K a/K s results is in fact caused by a lower K s for XR genes, not by faster non-synonymous site evolution. This could be a result of more effective selection to maintain codon usage bias in X-linked genes (McVean & Charlesworth, Reference McVean and Charlesworth1999); we examine this possibility below.
(iii) Comparisons of codon usage
We evaluated the frequency of optimal codons (Fop), a measure of codon usage bias (see the Materials and Methods section), for all genes in the sample (Table 3). Although X-XL genes have the highest levels of codon bias in each species, 3L-XR genes have similar levels of Fop to the autosomes in D. pseudoobscura and D. affinis. This might simply reflect sampling bias, since our D. melanogaster and D. yakuba 3L genes have lower levels of Fop than other autosomal genes, suggesting that direct comparisons between different chromosomal arms are, once again, unreliable. A more interesting result comes from comparisons between the same chromosomal arm in the two clades. D. melanogaster is known to have experienced a reduction in codon usage bias, thought to be due to a reduction in effective population size resulting in less efficient selection on this lineage (Akashi, Reference Akashi1995, Reference Akashi1996). We find, in agreement with previous studies, that D. melanogaster has significantly reduced levels of codon usage for all the chromosomes compared with D. yakuba (not shown). We therefore used D. pseudoobscura/D. yakuba to compare the Fop values in the two clades (use of D. affinis/D. yakuba yields similar results). While Fop values are similar in the two clades for our control genes (Table 3), they are significantly higher for XR in the D. pseudoobscura/D. affinis pair than for 3L in D. yakuba (P<0·001), consistent with the hypothesis that selection to maintain optimal codon usage is more efficient when loci are X-linked than when they are autosomal. These results will be considered further in the Discussion section.
Average values of Fop for 3L-XR, X-XL and autosomal genes
Boldface values indicate X-linked genes. Since D. melanogaster has significantly reduced levels of codon usage for all the chromosomes compared with D. yakuba (not shown), we used D. yakuba/D. pseudoobscura to compare the Fop values in the two clades (using D. yakuba/D. affinis yields similar results). The P-values were obtained using Wilcoxon Signed Rank tests.
a S.E. values are given in parentheses.
(iv) Pairwise comparisons
In order to test whether the behaviour of the 3L-XR genes is different from that of the control genes, we also examined the proportion of genes that were evolving faster (genes that have higher K a or K a/K s) in the D. pseudoobscura/D. affinis pair than in D. melanogaster/D. yakuba. In the absence of a faster-X effect, this value will be similar for 3L-XR and non-3L-XR genes. If, on the other hand, there is faster-X evolution, this proportion will be higher for 3L-XR genes. The values are presented in Table 4. Unlike previous observations (Counterman et al., Reference Counterman, Ortiz-Barrientos and Noor2004), there is no detectable faster-X effect for non-synonymous changes; despite the fact that the proportion of genes with higher K a/K s in the D. pseudoobscura group is slightly higher for 3L-XR genes, this effect is not significant, and the opposite is observed for the K a values. Once again, a lower K s for 3L-XR genes in D. pseudoobscura/D. affinis is the only significant pattern.
Proportions of genes with higher rates of evolution in the D. pseudoobscura/D. affinis pair compared with D. melanogaster/D. yakuba
The P-values were obtained using Fisher's exact tests.
(v) Is there a faster-X effect in faster evolving genes?
Various factors could have contributed to this failure to detect a faster-X effect for non-synonymous mutations (see the Introduction section). If a high K a/K s at least partially reflects a prevalence of positive over purifying selection, then we might expect genes with high K a/K s when autosomal to experience faster-X evolution when they become X-linked. Genes with low K a/K s, on the other hand, might experience slower-X evolution due to more efficient purifying selection (Charlesworth et al., Reference Charlesworth, Coyne and Barton1987; McVean & Charlesworth, Reference McVean and Charlesworth1999). To test for this, we divided our sample into fast, medium and slow evolving genes in the following way: we ordered our 3L-XR genes according to their K a/K s in the D. melanogaster/D. yakuba pair, and classified the first 23 as slow-evolving genes, the next 23 as medium-, and the last 23 as fast-evolving genes. We then repeated the analysis for these three classes, using as controls the non-3L-XR genes that had K a/K s values in D. melanogaster/D. yakuba in the same range as the 3L-XR genes. The resulting sample contains 39 fast-evolving genes (23 3L-XR and 16 non-3L-XR), 46 medium (23 3L-XR and 23 non-3L-XR) and 50 slow evolving genes (23 3L-XR and 27 non-3L-XR). The proportion of genes with higher K a, K s and K a/K s in the D. pseudoobscura/D. affinis pair is shown in Fig. 2 (the numbers of genes involved is given in Table 5).
‘Proportion of genes with higher K a, K s and K a/K s in the D. pseudoobscura/D. affinis pair’. P-A stands for D. pseudoobscura-D. affinis and M-Y for D. melanogaster-D. yakuba.
Numbers of slow-, medium- and fast-evolving genes with different Ka, Ks and Ka/Ks values in the two clades
P-values were obtained using a Fisher's exact test.
Significant P-values are shown in boldface.
Although our sample size is now drastically reduced, the proportion of genes evolving faster in the D. pseudoobscura/D. affinis pair is behaving in the predicted direction, but the results are not statistically significant. For fast-evolving genes, 61% of the 3L-XR loci have higher K a/K s in D. pseudoobscura/D. affinis, compared with 44% of the non-3L-XR loci, consistent with the faster-X hypothesis. Slow-evolving genes, on the other hand, do indeed show a moderate ‘slow-X’ effect: 70% of 3L-XR genes have higher K a/K s in the D. pseudoobscura group, versus 78% for the control genes. This trend is partially caused by differences in K s (3L-XR fast-evolving genes have the lowest K s in the D. pseudoobscura group), but the values for K a are also consistent with the faster-X hypothesis. For fast-evolving genes, the 3L-XR arm has a higher proportion of genes with higher K a in the D. pseudoobscura group than the non-3L-XR sample. For slow evolving genes, we observe the opposite. None of these differences are, however, statistically significant. It should also be noted that, when we use the values of d N and d S obtained with PAML (Supplementary Material), we obtain similar patterns for K a and K s, but not for d N/d S: in this case, there is a faster-X effect for all classes of gene, although it is closer to significance for fast-evolving genes.
4. Discussion
(i) Is selection more efficient on the X chromosome?
We have found that genes located on Muller's element D, which is X-linked in the D. pseudoobscura clade but autosomal in the D. melanogaster clade, have a significantly lower K s when they are X-linked than when they are autosomal (Fig. 1), whereas other genes do not differ in K s between the two clades. It is possible that this effect could be accounted for by a difference in mutation rate between X and autosomes. A higher mutation rate in males than in females leads to a lower rate of neutral evolution for the X chromosome, because the X is transmitted by females two-thirds of the time, whereas autosomes are transmitted by females only one-half of the time (Miyata et al., Reference Miyata, Hayashida, Kuma, Mitsuyasu and Yasunaga1987; Vicoso & Charlesworth, Reference Vicoso and Charlesworth2006). In mammals and some lineages of Drosophila, there is evidence for this effect (Ebersberger et al., Reference Ebersberger, Metzler, Schwarz and Paabo2002; Singh et al., Reference Singh, Larracuente and Clark2007), but no statistically significant evidence supporting it has been found in X-autosome comparisons in D. melanogaster, D. yakuba or D. pseudoobscura (Bauer & Aquadro, Reference Bauer and Aquadro1997; Richards et al., Reference Richards, Liu, Bettencourt, Hradecky, Letovsky, Nielsen, Thornton, Hubisz, Chen, Meisel, Couronne, Hua, Smith, Zhang, Liu, Bussemaker, van Batenburg, Howells, Scherer, Sodergren, Matthews, Crosby, Schroeder, Ortiz-Barrientos, Rives, Metzker, Muzny, Scott, Steffen, Wheeler, Worley, Havlak, Durbin, Egan, Gill, Hume, Morgan, Miner, Hamilton, Huang, Waldron, Verduzco, Clerc-Blankenburg, Dubchak, Noor, Anderson, White, Clark, Schaeffer, Gelbart, Weinstock and Gibbs2005; Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007; Singh et al., Reference Singh, Larracuente and Clark2007). While a higher rate of substitution of silent mutations has been found on the neo-Y chromosome of D. miranda compared with the neo-X, consistent with male-driven evolution (Bachtrog, Reference Bachtrog2008), this can be accounted for by the fixation of ancestral polymorphisms on the neo-Y, caused by its greatly reduced effective population size (Bartolomé & Charlesworth, Reference Bartolomé and Charlesworth2006). It therefore seems unlikely that a male–female mutation rate difference could account for our observations on K s.
The other possibility is that there is more effective selection to maintain codon usage on X-linked genes than on autosomal genes. Although synonymous substitutions are often treated as effectively neutral, there is ample evidence in Drosophila that synonymous codons are used in genes at different frequencies, because of selection for ‘preferred’ codons with higher efficiency or accuracy of translation (Powell & Moriyama, Reference Powell and Moriyama1997; Clark et al., Reference Clark, Eisen, Smith, Bergman, Oliver and Markow2007). McVean & Charlesworth (Reference McVean and Charlesworth1999) investigated the expected influence of X-linkage on codon usage bias, under the Li–Bulmer model of selection, genetic drift and reversible mutation between preferred and unpreferred codons (Li, Reference Li1987; Bulmer, Reference Bulmer1991). If unpreferred codons are, on average, recessive or partially recessive in their effect on fitness, they will be selected out of the population more efficiently when they are on the X, leading to higher codon usage bias (McVean & Charlesworth, Reference McVean and Charlesworth1999). As we discuss in the next section, another (not mutually exclusive) possibility is that differences in effective population size between the X chromosome and the autosomes could be increasing the level of codon usage bias on the X.
Singh et al. (Reference Singh, Davis and Petrov2005) estimated codon bias levels in D. melanogaster, D. pseudoobscura and Caenorhabditis elegans and found that these were higher on the X chromosome than on the autosomes in all three species. They excluded other factors that are correlated with codon usage bias, such as gene expression, gene length, recombination rate, gene density and protein evolution as possible causes for the X-autosome difference, suggesting that more efficient selection on the hemizygous male X is the main cause of increased codon usage bias on the X. They also compared D. pseudoobscura with D. melanogaster, and found a significant increase in codon bias for XR genes compared with their autosomal counterparts, consistent with the results in Table 3. More recently, the analysis of 12 Drosophila genomes has confirmed that the X chromosome has consistently higher levels of codon usage bias than the autosomes (Clark et al., Reference Clark, Eisen, Smith, Bergman, Oliver and Markow2007; Singh et al., Reference Singh, Larracuente and Clark2007).
This evidence suggests at first sight that the reduced K s we have detected for genes on XR in the D. pseudoobscura clade, compared with the same genes on 3L in the D. melanogaster clade, reflects more effective selection to maintain codon usage bias, as a result of these genes becoming X-linked. A puzzle that arises from this observation, though, is why K s for the ancestral X is, if anything, slightly higher (but not significantly so) than for the autosomes in the D. melanogaster/D. yakuba comparison in both our data and in the genomewide comparison of the sequenced members of the D. melanogaster group (Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007); in particular, looking at divergence along the branch leading to D. melanogaster from D. simulans, K s for the X chromosome is approximately 1·1 times larger than for the autosomes (Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007).
One possible explanation is that codon usage in the D. melanogaster group is not in equilibrium, perhaps because of a historical reduction in population size, for which there is support from the genome sequence comparisons, especially for the D. melanogaster branch of the phylogeny (Akashi, Reference Akashi1995, Reference Akashi1996; Akashi et al., Reference Akashi, Ko, Piao, John, Goel, Lin and Vitins2006; Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007; Nielsen et al., Reference Nielsen, Bauer DuMont, Hubisz and Aquadro2007). Takano-Shimizu (Reference Takano-Shimizu1999) showed that a reduction in effective population size is expected to result in a transient increase in the substitution rate above the equilibrium level, which is largest for genes with high levels of codon usage bias, since these depart the most from their final equilibrium. If the X chromosome has higher codon usage bias than the autosomes, X-linked genes would therefore show a higher transient substitution rate. The difference between XR and 3L that we have detected may, therefore, be primarily caused by an inflated rate of synonymous substitutions in the D. melanogaster group, with K s for XR in the D. pseudoobscura clade reflecting a level that is closer to the equilibrium value.
(ii) The effective population sizes of the X chromosome and the autosomes
The ratio of Fop between the X and major autosomes in Drosophila is of the order of 1·1 (Table 3 and Singh et al., Reference Singh, Davis and Petrov2005). This is much larger than the maximum value predicted by McVean & Charlesworth (Reference McVean and Charlesworth1999), who assumed that the effective population size N e for the X (N eX) is three-quarters of that for the autosomes (N eA). An even larger effect can, however, arise under a wide range of conditions, if the effective population size for the X chromosome is much bigger than this, as selection is less efficient on chromosomes with small effective population sizes.
Demographic effects (such as a female-biased sex ratio), increased recombination on the X chromosome, and increased variance of male reproductive success have all been shown be potential causes of increased N eX/N eA. One way to assess if N e for the X chromosome is higher or lower than the autosomal N e is to compare neutral or nearly neutral polymorphism levels at X-linked and autosomal sites, since neutral polymorphism levels are proportional to the effective population size (π, the pairwise average diversity, is equal to 4N eμ, where μ is the neutral mutation rate). Although they are not strictly neutral, synonymous sites are commonly used to approximate neutral polymorphism levels in Drosophila. Several studies have found that, in African populations of D. melanogaster, X-linked synonymous sites have higher levels of polymorphism than autosomal sites (Andolfatto, Reference Andolfatto2001; Kauer et al., Reference Kauer, Zangerl, Dieringer and Schlotterer2002; Mousset & Derome, Reference Mousset and Derome2004), suggesting that the effective population of the X is indeed higher than the expected three-fourth of the autosomal effective population size (Hutter et al., Reference Hutter, Li, Beisswanger, De Lorenzo and Stephan2007). This is likely to be contributing to the patterns of evolution we observe on the X chromosome, as it has a more powerful effect on the ratio of equilibrium codon usages for the X and autosomes than recessivity of deleterious mutational effects (Singh et al., Reference Singh, Davis and Petrov2005).
(iii) Is there a faster-X effect for non-synonymous sites?
Our data show a faster rate of non-synonymous site evolution for the ancestral X chromosome compared with the autosomes in both species comparisons (Tables 1 and 2), as has also been found in a genomewide comparison of the sequenced members of the D. melanogaster group (Begun et al., Reference Begun, Holloway, Stevens, Hillier, Poh, Hahn, Nista, Jones, Kern, Dewey, Pachter, Myers and Langley2007). For the reasons given in the first part of the Results section, however, this apparent support for faster-X evolution should be treated with some caution. The more critical test is to compare K a for 3L-XR between the two species pairs, and we found no significant effect; the significant difference in K a/K s was entirely due to the difference in K s (Fig. 1), as discussed above. In fast-evolving genes, however, there was a suggestion of higher K a for XR-linked loci, and the opposite was observed for slow-evolving genes, but neither of these patterns was significant (Table 5 and Fig. 2).
Although they are not conclusive, these results are of interest in view of the contradictory results obtained by previous studies on faster-X evolution, as they suggest that such an effect might only be observed for genes that are under particularly strong positive selection and/or relaxed negative selection. In fact, all the studies that previously detected faster-X evolution in Drosophila were in some way biased towards fast-evolving genes. For example, Counterman et al. (Reference Counterman, Ortiz-Barrientos and Noor2004) obtained part of their sample from a male-specific expressed sequence tag (EST) screen (Swanson et al., Reference Swanson, Clark, Waldrip-Dail, Wolfner and Aquadro2001). Male-specific genes are not only expected to show an enhanced faster-X evolution, but it has also been claimed that they have a faster evolution than non-sex-biased genes in Drosophila, possibly as a consequence of increased positive selection (Zhang et al., Reference Zhang, Hambuch and Parsch2004). Consistent with this, several studies of male-biased or male reproductive genes have detected faster rates of evolution on the X (Torgerson & Singh, Reference Torgerson and Singh2003; Wang & Zhang, Reference Wang and Zhang2004; Pröschel et al., Reference Pröschel, Zhang and Parsch2006; Baines et al., Reference Baines, Sawyer, Hartl and Parsch2008). Thornton et al. (Reference Thornton, Bachtrog and Andolfatto2006), while following a similar approach to that of Counterman et al. (Reference Counterman, Ortiz-Barrientos and Noor2004), chose their genes randomly, and observed no faster-X effect. This suggests that further studies on faster-X evolution should focus on fast-evolving genes, and that some of the discrepancies among different studies described in the Introduction section may arise from the use of different types of genes.
B.V. was supported by the Portuguese Foundation for Science and Technology, P.R.H. is supported by the Natural Environmental Research Council (UK) and B.C. was supported by the Royal Society (UK). We thank Dan Halligan for assistance with PAML. We also thank two anonymous referees for their comments on the manuscript.
