Reduced Diversity on the Neo-X

Natural selection can increase the frequency of a beneficial mutation in a population, thereby reducing neutral variation in the genomic region linked to the advantageous allele (i.e., a selective sweep [29,30]). Thus, one signature of directional selection at the DNA level is a reduction in neutral variation in genomic regions surrounding the targets of selection [29]. To test for increased rates of adaptation on the D. miranda neo-X chromosome, we surveyed DNA sequence polymorphism at 152 gene fragments across the neo-X and compare them to 112 gene fragments from the ancestral X chromosome (60 genes from X-L and 52 from X-R). Certain classes of genes tend to undergo increased rates of adaptive evolution in Drosophila, such as genes showing sex-biased expression or genes involved in some biological pathways [31,32]. Genes on both the neo-X and the ancestral X chromosome were selected randomly with regard to gene function or expression patterns, and no significant heterogeneity in gene count among gene ontology classes or patterns of sex-biased expression was detected between loci on the ancestral X chromosome and the neo-X (see Tables S1–S4).

Estimated levels of synonymous variation and synonymous divergence to an outgroup species are similar for genes on X-L and X-R (mean π_s = 0.51% vs. π_s = 0.71%; Wilcoxon two-sample test, p > 0.05 and mean K_s = 4.26% vs. K_s = 3.98% between D. miranda and D. pseudoobscura; Wilcoxon two-sample test, p > 0.05), consistent with observations that X-R has reached the typical properties of an evolved X chromosome (i.e., X-R appears fully dosage compensated in males and its gene content shows a similar deficiency of male-biased genes as X-L, the ancestral X chromosome; [12,17]). Table 1 summarizes average levels of synonymous diversity across the genomic regions studied. Synonymous site diversity is reduced by about 50% on the neo-X compared to the ancestral X (average π_s = 0.33% vs. π_s = 0.60%; Wilcoxon two-sample test, p < 3e-6), while levels of synonymous divergence to an outgroup species between the chromosomes are similar (K_s = 4.36% vs. K_s = 4.13% for the neo-X and the ancestral X chromosome between D. miranda and D. pseudoobscura; Wilcoxon two-sample test, p = 0.13), suggesting that the two chromosomes have similar mutation rates. Comparison of polymorphism and divergence levels at synonymous sites on the neo-X versus the ancestral X using a Hudson-Kreitman-Aguadé (HKA) test [30] confirms that the reduced diversity observed at neo-X genes is not attributable to a lower mutation rate (HKA test p < 10⁻⁴, see Table S6). Additionally, many more invariant loci are observed on the neo-X chromosome (23 vs. three genes, Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1.

Average Diversity Measures in Drosophila miranda across X-Linked and Neo-X Gene Fragments and Numbers of Loci Showing Evidence for Recent Adaptive Evolution

https://doi.org/10.1371/journal.pbio.1000082.t001

Nonequilibrium demography, such as recent population bottlenecks, or differences in life-history strategies between males and females can cause levels of diversity to differ between sex chromosomes and autosomes [25–27]. However, because the ancestral X chromosome and the neo-X chromosome show identical patterns of inheritance, demography and life history are expected to influence patterns of diversity on the X and the neo-X in a similar manner [25–27]. Note, however, that the neo-X chromosome was segregating as an autosome until the formation of the neo-sex chromosomes roughly 1 MY ago. This event was presumably associated with a modest decline in the population size of the neo-X (from 2N to 1.5N), but is sufficiently ancient (∼10N_e generations ago) to not leave signatures in current levels of population variation [33,34]. Thus, the ancestral X should serve as an adequate control for demographic effects on the neo-X, which suggests that natural selection is responsible for reduced levels of variability on the neo-X relative to the ancestral X. We employed several statistical approaches in order to quantify rates of adaptive evolution on the neo-X versus the ancestral X.

More Recent Selective Sweeps on the Neo-X

Recent positive selection results not only in a local reduction of variation in the genomic region surrounding the target of selection, but also in a skew in the frequency distribution of mutations surrounding the target of selection (i.e., the hitchhiking effect; [29,35]). In particular, recent adaptive evolution in the genome results in an excess of both low- and high-frequency mutations relative to neutral expectations [33,36,37]. Such approaches to detect selection using population variability data cannot be applied to invariant loci, but the 129/152 polymorphic neo-X–linked and 109/112 polymorphic X-linked loci can be examined.

A composite likelihood ratio (CLR) test [38] that utilizes these population variation patterns to detect recent adaptations reveals that a greater proportion of genes located on the neo-X chromosome relative to the ancestral X reject a model of neutral sequence evolution in favor of a genetic hitchhiking model at the 5% significance level (19/129 testable neo-X genes, 15%, vs. 3/109 testable X-linked genes, 3%, Table 1; p = 0.0013, Fisher exact test). Further, the distributions of CLR test p-values for the neo-X and X are significantly different from one another (p = 2e⁻⁹, Kolmogorov-Smirnov test). Although the CLR test is not robust to some demographic scenarios [34], the ancestral X chromosome functions as an internal control to account for such effects (see above). Thus, we detect about 5-fold more adaptive events on the newly formed neo-X compared to the ancestral X. To further evaluate evidence in support of more adaptation on the neo-X, we also applied a goodness-of-fit test (GOF test) [34] to genes rejecting the CLR test. This statistic was proposed to assess the fit of data to a selective sweep model, in order to identify loci with significant CLR tests that may be explained by demographic effects. Only one of the three genes that rejected the CLR test on the ancestral X chromosome were consistent with a selective sweep model using the GOF test, whereas 13 of the 19 neo-X genes rejecting the CLR test were consistent with a recent selective sweep (Table 1).

Adaptive evolution also leaves characteristic signatures in patterns of linkage disequilibrium (LD), with reduced LD across the target of selection, and increased LD in genomic regions flanking the target [39–41]. The ω_max statistic [39] was used to identify loci under selection based on these patterns of LD. Again we detect more selection on the neo-X chromosome (7/129 testable genes, 5%, reject neutrality at the 5% significance level) compared to the ancestral X (1/109 testable genes, 1%, reject neutrality at the 5% significance level, Table 1; p = 0.07 Fisher exact test). The genes identified as targets of recent adaptive evolution using the ω_max statistics are a subset of those identified with the CLR+GOF test. Thus, several locus-by-locus tests identify a much larger fraction of genes having undergone recent adaptive evolution on the neo-X chromosome compared to the ancestral X.

Note that the locus-by-locus tests for selection are not corrected for multiple testing, because our main interest lies in quantifying the relative excess of statistical tests rejecting neutrality on the neo-X relative to the ancestral X, and not the absolute number of significant rejections. A similar excess of adaptive evolution of neo-X–linked genes relative to the ancestral X is found if we use the false discovery rate to account for multiple testing (see Table S7). In addition, the above locus-by-locus tests of selection assume that the genomic regions surveyed are unlinked. Indeed, the genes surveyed on the neo-X and the ancestral X chromosome appear mostly independent from each other, with levels of LD being similarly low between loci (unpublished data). Also, the genomic regions we identified as having undergone recent selection on the neo-X show little evidence of clustering (the median distance between loci using the D. pseudoobscura genome sequence as a guide is 0.97 Mb for the 19 significant loci identified using the CLR test, and only two regions rejecting this test are adjacent to each other). In addition, many of the gene fragments studied here were mapped previously by in situ hybridization experiments, and found to be scattered along the polytene chromosomes of D. miranda [42–44]. Thus, most selective sweeps identified on the neo-X represent independent events.

A more formal approach to take account of multiple testing and linkage when comparing rates of evolution on the ancestral and the newly formed X chromosome is to consider all loci simultaneously rather than testing them individually. To this end, we employed a composite likelihood method (the maximized composite likelihood surface test [MCLS test]) [45] for detecting positive selection, which uses a similar likelihood framework as the CLR test. As opposed to the CLR or GOF tests, however, which use a specific population genetic model as the null (the equilibrium neutral model or a selective sweep model, respectively), the MCLS statistic derives the null model from the empirical data itself. In this way, the test seeks to identify loci that show unusual patterns of variation relative to the other loci (the “background loci” [45]) in the genomic screen. To determine significance of the test statistic, it is still necessary to simulate data based on an explicit model [45]. In our implementation, we use the background allele frequency distribution (the background site-frequency spectrum) obtained from the ancestral X, to test for selection on both the ancestral X and the neo-X chromosome. Again, we find evidence for many more genomic regions having undergone recent positive selection on the neo-X (13 nonoverlapping regions) compared to the ancestral X (three genomic regions, Figure 2; p = 0.035, Fisher exact test). The distributions of the minimum p-value windows are significantly different for loci surveyed on the neo-X and the ancestral X (p = 0.022, Kolmogorov-Smirnov test). All 13 neo-X regions identified as targets of recent adaptive evolution using the MCLS approach correspond to loci identified as positively selected using the locus-by-locus CLR+GOF test (above).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. The Maximized Composite Likelihood Surface Calculated for Genes from the Ancestral X and the Neo-X Chromosome

The horizontal line indicates the 5% cutoff values (LR_crit) as determined separately for the X (LR_crit = 4.5) and the neo-X (LR_crit = 2.7) by simulation under a neutral equilibrium model. More significant peaks are identified on the neo-X, suggesting that more recent selective sweeps have occurred on this chromosome.

https://doi.org/10.1371/journal.pbio.1000082.g002

Could other systematic biases—such as differences in overall recombination rates or levels of variability between sampled loci—result in differential power to detect selection on the two chromosomes? Specifically, selective sweeps are more easily identified in low-recombining regions due to the increased effects of hitchhiking [34], and statistical tests have reduced power to detect selection if levels of variability are low [34]. Recombination rates do not appear to systematically differ between loci on the neo-X and the ancestral X chromosome (average levels of LD, as measured by Wall's B or Q [46], are not significantly different within loci on the two chromosomes; p > 0.2, Kolmogorov-Smirnov test). However, the neo-X chromosome has significantly reduced levels of variability and more invariant genes relative to the X (Table 1), as expected under a model of recurrent selection. Methods based on the site-frequency spectrum (CLR, GOF, and MCLS) and LD (ω_max) have reduced power to detect selection when levels of variability are low (and cannot be applied to invariant loci), implying less power to detect selective events on the neo-X chromosome. This suggests that the difference in rates of adaptive evolution between the neo-X and the ancestral X is likely underestimated. Nevertheless, utilizing many different features of our data, we consistently estimate that the fraction of genes having undergone recent adaptive evolution on the newly formed X chromosome increases 5–10-fold relative to background levels of adaptation on the ancestral X.

Increased Rates of Adaptation on the Neo-X

To what extent are rates of adaptation accelerated on the neo-X? To evaluate the difference in rates of adaptation on the neo-X versus the ancestral X using all loci (including invariant ones), we estimate parameters of a recurrent selection model for each chromosome. Under a model of recurrent adaptation, the average reduction in levels of variability depends on the rate at which adaptive substitutions occur (2N_eλ) and their average effect on fitness (s) [47]. We use a recently developed approximate Bayesian approach [48] to estimate these parameters. This approach uses multiple summary statistics of the population variation data (see Materials and Methods), to obtain maximum a posteriori (MAP) estimates of both 2N_eλ and s. Figure 3 shows the marginal and joint posterior distributions of the rate and the strength of selection inferred for loci on the ancestral X and the neo-X. The MAP estimate of the rate of adaptive substitutions again is roughly 10-fold higher for neo-X–linked genes compared to genes that are located on the ancestral X (Figure 3). Interestingly, we also estimate the strength of selection to be an order of magnitude higher for genes on the neo-X chromosome (Figure 3). Importantly, MAP estimates of the strength and the rate of sweeps for the neo-X fall outside of the 95% credibility intervals for the estimates on the X. Thus, not only does adaptive evolution appear to be more frequent on a newly formed X chromosome, but also the selective benefit of mutations arising may be larger for genes that have only recently become X-linked relative to genes that have been evolving under a stable chromosomal configuration. This may be expected if genes on a young X chromosome are further away from their optimum fitness [49], since they experience a new genomic environment (i.e., they used to segregate as an autosome but have only recently become X-linked).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Approximate Bayesian Estimation of the Rate of Adaptive Substitutions (2N_eλ) and Their Average Effect on Fitness (s) for Genes from the Ancestral X and the Neo-X Chromosome

Estimation is based on 10⁶ draws from the prior (s ∼ Uniform (1.0E−06, 1.0) and 2N_eλ ∼ Uniform (1.0E−07, 1.0E−01)).

(Top) The joint posterior distributions for the X and neo-X. The dotted lines correspond to MAP estimates, and darker regions indicate greater posterior density.

(Bottom) The marginal posterior distributions for the strength and rate of sweeps. The marginal posterior distribution for the X is indicated in grey, and for the neo-X in black. Both the rate and the strength of selection are inferred to be an order of magnitude higher for neo-X–linked genes.

https://doi.org/10.1371/journal.pbio.1000082.g003

Conclusions

Using a variety of approaches, we consistently infer a higher rate of adaptive evolution on the neo-X chromosome relative to the ancestral X. Systematic differences between the chromosomes, such as biases in the types of genes studied or overall differences in rates of recombination, appear unlikely to account for elevated rates of adaptation on the neo-X. Instead, increased adaptation is likely related to the recent formation of the X chromosome, and as such, there are a priori reasons to expect rapid adaptive changes to respond to novel selective pressures created by Y degeneration and sex-biased transmission. Specifically, old X chromosomes have evolved dosage compensation mechanisms [12], and genes with male-biased expression are underrepresented on the Drosophila X [17], resulting from selective gene extinction and gene movement off the X [17]. In addition, hemizygosity of X-linked genes in males increases the efficacy of natural selection acting on recessive mutations (faster-X evolution [14]). Expression profiling and comparative sequence analysis have shown both the acquisition of dosage compensation and demasculinization of gene content on chromosome element X-R in D. pseudoobscura (Muller's element D), a neo-X chromosome that formed approximately 10 MY ago [12,17]. We observe no evidence for elevated rates of adaptation on X-R relative to the older X-L (Muller's element A) in D. miranda, which is consistent with the functional data that X chromosomes evolve dosage compensation and demasculinization of their gene content in a relatively short evolutionary time period (i.e., within 10 MY). In contrast, the much younger neo-X chromosome in D. miranda (Muller's element C) appears still in transition from being an ordinary autosome to an X chromosome and exhibits chromosomal patterns intermediate between the autosomes and the ancestral X. In particular, the neo-X chromosome of D. miranda has evolved partial dosage compensation [11,12]. Here, we have shown that this evolutionary transition from an autosome to an X chromosome has been accompanied by a tremendous acceleration in rates of adaptive evolution, which presumably reflects the ongoing acquisition of dosage compensation on the neo-X of D. miranda [10] and possibly the demasculinization of its gene content. Thus, contrary to being a passive player in sex chromosome evolution, we demonstrate that the X chromosome has a very active role. We propose that a newly formed X chromosome actively responds to both its female-biased transmission and to Y degeneration, through bursts of adaptive evolution.

Survey of coding regions.

A total of 112 X-linked and 152 neo-X–linked gene fragments were surveyed in this study with a sample size of 12–18 D. miranda alleles (mean sample size 16). Genes were selected randomly with respect to gene function and expression patterns (see Tables S1–S4). The D. pseudoobscura genome sequence (release 2.0, http://flybase.org/) was used to provide estimates of divergence. Details of PCR primers are available from the authors on request. Information about the individual loci surveyed and the geographic origin of the D. miranda strains investigated can be found in Tables S1–S5. Many of the gene fragments studied were mapped previously to the polytene chromosomes of D. miranda using in situ hybridization, and found to be scattered along the ancestral X chromosome and the neo-X [42–44].

Standard PCR procedures were used to amplify each region from genomic DNA from single male flies, using X or neo-X specific primers. PCR products were cleaned using Exonuclease I and Shrimp Alkaline Phosphatase, and sequenced on both strands with the original PCR primers and internal sequencing primers if necessary, using Big-Dye (Version 3; Applied Biosystems). Sequence reactions were cleaned with sephadex plates (Edge Biosystems) and run on an ABI 3730 capillary sequencer. Chromatograms were edited and assembled using Sequencher (Gene Codes) software and multiple sequence alignments were generated using MUSCLE (http://www.drive5.com/muscle/) with protein-alignment–assisted adjustments to preserve reading frames. Exon–intron boundaries were determined from the D. pseudoobscura genome sequence annotation (release 2.0). Sequences have been deposited in GenBank under accession numbers FN252903–FN256223.

Polymorphism and divergence analysis.

A library of Perl scripts was used to calculate the estimated number of synonymous sites, average pairwise diversity (π) and average pairwise divergence to D. pseudoobscura (K). A Jukes-Cantor correction was used to correct π and K for multiple hits. Insertion–deletion polymorphisms and polymorphic sites overlapping alignment gaps were excluded from the analysis. To compare polymorphism and divergence on the neo-X and the ancestral X, we implemented a multilocus two-class HKA test [30]. The 262 loci were pooled into two classes (i.e., X-linked vs. neo-X–linked). The p-values for this test are based on 10,000 replicate simulations of the standard neutral coalescent model using the program ms [50] with global (i.e., speciation time, T) and locus-specific parameters (i.e., sample size, n, and population mutation rate, θ).

Locus-by-locus tests for selection.

Several statistical tests to identify recent adaptive evolution were applied to genes on the X versus the neo-X chromosome. The CLR test [38] uses the spatial distribution of mutation frequencies in a genomic region and levels of variability among a population sample of DNA sequences to test for evidence of a selective sweep. This method compares the ratio of the composite likelihood of the data under the standard neutral model of constant population size, neutral evolution, and random mating, L_N (Data) to the composite likelihood of the data under the model of a selective sweep, L_S(α̂,X̂|Data), where α is the maximum likelihood estimate (MLE) of 2Ns, and X is the MLE of the location of the beneficial mutation. The CLR test statistic employed is . The null distribution of Λ_KS is obtained for each region by applying the CLR test to datasets obtained from simulation under the standard neutral model (using the program ms [50]) with the observed region length (L) and θ. The recombination rate ρ per site is set at 8.8 × 10⁻⁸ per site per generation [28]. For each locus, 1,000 neutral replicates were simulated using locus-specific parameters in order to assess significance. A complete users manual, as well as all necessary code, can be found at: http://www.yuseobkim.net/YuseobPrograms.html. The neutral model is rejected at level γ (5% used here) when the observed Λ_KS is greater than the 100(1−γ) percentile of the null distribution.

The CLR test is sensitive to deviations from the assumptions of the standard neutral model, with population substructure and recent bottlenecks leading to a high false-positive rate [34]. However, demography is not a concern for our study as we only compare patterns on the neo-X to the ancestral X, which serves as an internal control for the inflated variances of the test statistics, associated with demographic effects. The ancestral X chromosome is an ideal control for that purpose, since it has the same population size and shows identical patterns of inheritance as the neo-X, and thus is influenced by demography and life history in a similar manner. To assesses the fit of individual loci to a selective sweep model, we also employed a GOF test that contrasts the null hypothesis H₀ that the data are drawn from a selection model as simulated by the CLR test to the alternative hypothesis H_A that the data are not drawn from such a model [34]. A composite likelihood scheme is used to approximate the probability of the data given the null, P(Data|H₀), to the probability of the data given the alternative, P(Data|H_A), on the basis of the site frequency spectrum of mutations. Simulations (using the program ssw [38]) under the null hypothesis are used to find the critical value of the composite likelihood ratio GOF statistic for each region, with locus-specific (maximum likelihood) estimates of S, L, α, and X. Note that in this instance, the null model is a selective sweep as this test is employed conditional on rejecting the CLR test [34]. The program for calculating this statistic is available for download at http://www.yuseobkim.net/YuseobPrograms.html.

Positive selection results in strong LD flanking the target of selection, and reduced LD across the target [40,41]. We also employ patterns of LD to test for selection at individual loci using the ω test [39]. The ω-statistic, which is defined as divides the S polymorphic sites in the dataset into two groups, one from the first to the lth polymorphic site from the left and the other from the (l + 1)th to the last site (l = 2, … , S − 2), where L and R represent the left and right set of polymorphic sites, and r_ij² is the squared correlation coefficient between the ith and jth sites. Thus, ω increases with increasing LD within each of the two groups and decreasing LD between the two groups (i.e., the larger the value of the statistic, the more “sweep-like” the underlying pattern). For a locus, the value of l that maximizes ω (ω_max) is found. Singletons were excluded prior to calculation. The null distribution of ω for each genomic region is obtained from simulation under the standard neutral model (using the program ms [50]) with fixed θ and L. As above, we set ρ = 8.8 × 10⁻⁸ per site per generation. The program for calculating this statistic is available for download at: http://www.molpopgen.org/software/libsequence.html.

Multilocus tests for selection.

We used a multilocus test for selection based on a modification of the CLR test proposed by Nielsen et al. [45], called the MCLS test, to compare rates of adaptation on the ancestral X and the neo-X chromosome. Rather than using the standard neutral model to define the test statistic, this multilocus approach uses the background site frequency spectrum obtained from the data itself to test for selection. We use the site frequency spectrum of the ancestral X chromosome of D. miranda, to test for selection on both the ancestral X and the neo-X. This approach is not model independent, however, as coalescent simulations under the standard neutral model, using the exact configuration of the empirical dataset (with each locus having a unique θ and L), are necessary to determine the significance of the test statistic. This method takes polymorphism data and creates a grid of locations over the given region, and compares the maximum composite likelihood ratio of the hypothesis of a sweep with the null hypothesis of no sweep for each location. The parametric approach used is described by equation 6 of [45]. The gridsize parameter was set at 10⁴. As above, we set ρ = 8.8 × 10⁻⁸ per site per generation. We document the minimum p-value window for each surveyed locus. The program for calculating this statistic is available for download at: http://fisher.berkeley.edu/cteg/software.html.

To estimate selection parameters under a recurrent hitchhiking model, we use the approximate Bayesian approach of Jensen et al. [48]. The level of reduction in variation due to recurrent selection depends on the joint parameter 2N_esλ [47]. Both the rate, λ, and the fitness effect, s, of recurrent selection are estimated based upon their relationship with the means and standard deviations of common polymorphism summary statistics (the mean average pairwise diversity [π], the number of segregating sites [S], θ_H, and ZnS; see [48]). Calculating these summary statistics from the observed data, and from simulated data with parameters drawn from uniform priors, we implement the regression approach of Beaumont et al. [51] which fits a local-linear regression of simulated parameter values to simulated summary statistics, and substitutes the observed statistics into a regression equation. The prior distributions used were s ∼ Uniform (1.0E−06, 1.0) and 2N_eλ ∼ Uniform (1.0E−07, 1.0E−01), and the tolerance, ε = 0.001. Estimation is based on 10⁶ draws from the prior using the recurrent selective-sweep coalescent simulation machinery described in ref. [48]. We set ρ = 8.8 × 10⁻⁸ per site per generation and N_e = 568,851 [28]. For inferences on selection parameters assuming an exponential distribution of λ and s see Figure S1. A complete users manual, as well as all necessary code for estimating λ and s, can be found at: http://www.molpopgen.org/software/JensenThorntonAndolfatto2008/.

Sequences have been deposited in GenBank under accession numbers FN252903–FN256223.