EXTENSIVE INTROGRESSION OF MITOCHONDRIAL DNA RELATIVE TO NUCLEAR GENES IN THE DROSOPHILA YAKUBA SPECIES GROUP

Transcription

1 Evolution, 60(2), 2006, pp EXTENSIVE INTROGRESSION OF MITOCHONDRIAL DNA RELATIVE TO NUCLEAR GENES IN THE DROSOPHILA YAKUBA SPECIES GROUP DORIS BACHTROG, 1,2 KEVIN THORNTON, 2 ANDREW CLARK, 2 AND PETER ANDOLFATTO 1,3 1 Section of Ecology, Behavior and Evolution, University of California San Diego, La Jolla, California Department of Molecular Biology and Genetics, Cornell University, Ithaca, New York Abstract. Studies of gene flow between recently diverged species can illuminate the role of natural selection in the formation of new species. Drosophila santomea and D. yakuba are recently diverged, partially reproductively isolated species that continue to hybridize in the wild, and appear to be reproductively isolated from the more distantly related species D. teissieri. We examine patterns of nucleotide polymorphism and divergence in these three species at multiple X-linked, Y-linked, and mitochondrial markers. All three species harbor drastically reduced variability on the Y chromosome relative to the X, as expected for a nonrecombining chromosome subject to variation-reducing selection. The three species are generally well differentiated at the nuclear markers, with little evidence for recent introgression for either the X- or Y-linked genes. Based on the nuclear genes, we estimate that D. santomea and D. yakuba diverged about one-half million years ago and split from D. teissieri about one million years ago. In contrast to the pattern at nuclear loci, all three species share a very similar mtdna haplotype. We show that the mtdna must have recently introgressed across species boundaries in the D. yakuba subgroup and that its fixation was driven by either selection on the mitochondria itself or other cytoplasmic factors. These results demonstrate that different regions of the genome can have distinct evolutionary dynamics in the context of species formation. Although natural selection is usually thought of as accentuating divergence between species, our results imply that it can also act as a homogenizing force. Key words. Drosophila yakuba, hybrid zone, speciation, species divergence times. Genealogical analysis of multilocus datasets between closely related species provides the opportunity to distinguish between the allopatric, parapatric, and sympatric models of speciation. Under strict allopatry, all regions of the genome have a single divergence history (i.e., no gene exchange after the population split) and only vary in their coalescent time. In contrast, gene flow between diverging species under sympatric or parapatric speciation models allow nascent species to share genomic regions that have not yet diverged functionally, although genes or genomic regions that are well adapted in one species but not the other will tend to be eliminated by natural selection (Barton and Hewitt 1989). Thus, in the presence of gene flow, the genealogical history of the genome is expected to be a mosaic of different regions with disparate divergence times. In particular, genomic regions that have not evolved species-specific adaptations can be exchanged for a longer time between nascent species. The comparison of the genealogical history of several markers between recently formed (and maybe still hybridizing) species could therefore distinguish genomic regions associated with early reproductive isolation or species-specific adaptations from those lacking these associations. Indeed, several studies of gene flow between closely related species have found that some parts of the genome introgress more easily than others. For example, mitochondrial DNA (mtdna) tends to introgress much more readily than nuclear DNA (Ferris et al. 1983; Powell 1983; Sota and Vogler 2001; Doiron et al. 2002; Shaw 2002; Roca et al. 2005). The reasons for this are unclear, but may be due to the nature of mitochondrial genes (Coyne and Orr 2004). Most of these genes are housekeeping genes, which may be largely di- 3 Corresponding author: Division of Biological Sciences, University of California, San Diego, 9500 Gilman Drive, MC 0116 La Jolla, California 92093; pandolfatto@ucsd.edu The Society for the Study of Evolution. All rights reserved. Received June 21, Accepted December 14, vorced from external selective pressures. Thus, mtdna may be free of loci that contribute to hybrid unfitness or positive assortative mating, and most mitochondrial genes unlike nuclear genes may function fairly well in the genetic background of a related species (Coyne and Orr 2004). Therefore, mtdna might be free to introgress neutrally between species, whereas many regions of the nuclear genome may not. In this study, we compare the genealogical history of markers located on different parts of the genome in three species of the genus Drosophila: D. santomea, D. yakuba, and D. teissieri. This species group has only recently become of interest for speciation research, because D. santomea, an insular species endemic to the island of São Tomé off Westequatorial Africa, was only discovered a few years ago (Lachaise et al. 2000). Drosophila santomea is the closest relative of D. yakuba, which is widely distributed throughout the tropical sub-saharan African mainland (Lachaise et al. 2000). Both species coexist on São Tomé, but are segregated because they prefer different altitudes: D. santomea is present in mist rainforest habitats at higher elevation, whereas D. yakuba occupies the lower elevation habitats. At midelevation (between m) they form a potentially stable contact zone (Lachaise et al. 2000). Drosophila santomea and D. yakuba can form viable F 1 hybrids; hybrid females are fertile, whereas male hybrids are sterile (Lachaise et al. 2000). The rate of hybridization between D. yakuba D. santomea in this contact zone has been estimated to be about 1% (Lachaise et al. 2000), which is 100 times higher than the frequency of hybridization estimated in the wild for another species pair traditionally studied in speciation genetics, D. pseudoobscura and D. persimilis (Dobzhansky 1973). Drosophila teissieri, the outgroup for D. yakuba and D. santomea, is also widely distributed throughout the sub-saharan African mainlaind and seems reproductively isolated from D. santomea and D. yakuba, as hybridization attempts between D. teissieri and

2 SPECIES DIFFERENTIATION IN DROSOPHILA 293 the other two species generally fail (Lemeunier et al. 1986, Lachaise et al. 2000). Here, we survey patterns of DNA variability from multiple markers form mtdna and on the X and the Y chromosome in large samples from D. santomea, D. yakuba, and D. teissieri. In particular, by comparing the genealogical history of different genomic regions, we want to address the following questions: (1) Are all genomic regions between these species diverged similarly, or do some genomic regions show less differentiation than others? (2) Is there introgression of mtdna between species boundaries, as observed for other species and suggested by previous studies based on small samples (Monnerot et al. 1990; Lachaise et al. 2000; Cariou et al. 2001)? (3) And finally, what evolutionary forces have driven mtdna introgression across species boundaries in this group? Our analysis reveals that these species carry distinct alleles at nuclear markers, and diverged between 0.5 and one million years ago (Mya). Surprisingly, however, all three species share a similar mtdna haplotype. Selection, either on the mitochondria itself or on other cytoplasmic factors, has likely driven the recent introgression of mtdna across species boundaries. Thus, although selection may have maintained or contributed to divergence between these species for the X and the Y chromosome, selection has likely acted as a homogenizing force with respect to mtdna genealogies. MATERIALS AND METHODS Fly Lines Used A total of 16 isofemale lines (the descendents of a single wild collected female) of D. teissieri, 31 isofemale lines of D. santomea, and 41 isofemale lines of D. yakuba were used for this study. The D. yakuba lines originate from two different populations (Cameroon, collected by P. Andolfatto in 2002, and Gabon, collected by B. Ballard and S. Charlat in 2002), all D. santomea are from São Tomé (kindly provided by M. Long), and the D. teissieri lines are from several locations in Cameroon (collected by P. Andolfatto 2002), Gabon, and Congo (collected by B. Ballard and S. Charlat). One additional D. teissieri line of unknown origin was provided by the laboratory of C. Aquadro (Department of Molecular Biology and Genetics, Cornell University, Ithaca, NY). A summary of all fly strains used, including their geographic origin, is given in Supplementary Table 1 available online only at: Molecular analyses of multiple markers in D. santomea and sympatric and allopatric D. yakuba have shown that D. santomea is not more closely related to the sympatric D. yakuba than to any continental one, and no population structure was detected between allopatric mainland and sympatric island populations of D. yakuba (Cariou et al. 2001; Llopart et al. 2005). In addition, crosses between D. santomea and either sympatric or allopatric populations of D. yakuba show no differences in the extent of postzygotic isolation (Cariou et al. 2001). This suggests that the extant insular D. yakuba population on São Tomé results from a relatively recent secondary colonization by the mainland population. DNA Sequencing Genomic DNA was extracted from a single male of each line using the Puregene DNA extraction kit (Gentra Systems, Minneapolis, MN). Polymerase chain reaction (PCR) products were amplified as 500 bp 2000 bp fragments from genomic DNA using standard methods (Sambrook et al. 1989) and primer pairs given in Supplementary Table 2 available online only at: Primer pairs were designed using the draft D. yakuba genome sequence ( genome.wustl.edu/projects/yakuba/). PCR products were used as sequencing templates after treatment with the SAP/EXO reagent. Gene-specific internal primers and the original amplification primers were used for sequencing with the BigDye 3.0 cycle sequencing kit (Applied Biosystems, Foster City, CA) following the manufacturer s protocol. Sequences were run on an ABI 3730 automated sequencer, and each fragment was sequenced at least once on both strands. We obtained polymorphism data for 5019 bp of homologous Y-linked sequence (from five regions in the kl-2, kl-3, and ory genes) in 31 lines of D. santomea and 41 lines from D. yakuba. Most Y-linked primers were designed from noncoding DNA. Thus, only one primer pair successfully amplified a 472-bp fragment (of the 5 region of the kl-3 gene) in D. teissieri. For the mitochondria, we obtained sequence data for 1777 bp (from the COII and ND5 gene) in 25 lines of D. yakuba, 28 lines of D. santomea, and 15 lines of D. teissieri. The Y- linked and mtdna fragments were concatenated separately for the analyses. For the X chromosome, we obtained polymorphism data for six regions (a total of 4384 bp) in lines of D. yakuba, lines of D. santomea, and strains of D. teissieri. For locus- and species-specific estimates of variability see Supplementary Table 3, available online only at: DNA Polymorphism, Divergence, and Evolutionary Analyses Nucleotide sequences were edited using Sequencher 4.1 (Gene Codes, Ann Arbor, MI) and aligned by eye. Only silent sites (i.e., synonymous sites and noncoding DNA) were considered for the analysis, because these sites are less likely to be influenced by natural selection and therefore more useful in making demographic inferences of species divergence (Wang et al. 1997; Hey and Nielsen 2004). We excluded all sites with multiple nucleotide substitutions within a species. Tajima s D (Tajima 1989) was calculated based on the number of silent polymorphic sites and the average silent pairwise diversity. A maximum-likelihood version of the Hudson, Kreitman, Aguadé (HKA) test as implemented by Wright and Charlesworth (2004) was used to test for chromosome-specific departures from neutrality. We tested for population structure between species for the mtdna using the haplotype-based permutation test of Hudson et al. (1992a). To visualize locus-by-locus differentiation between species, we constructed simple neighbor-joining trees for each locus as implemented in PAUP 4.0b10 (Swofford 2002). No statistical inferences are drawn from these trees. For additional divergence analysis, homologous sequences for D. melanogaster were retrieved from Genbank. For the X chromosome and the mitochondria, mostly coding sequence was studied, which could be unambiguously aligned

3 294 DORIS BACHTROG ET AL. FIG. 1. Model of species divergence used for inferences. An ancestral population splits into two species of sizes and f at time T. The model assumes no migration between species. with the homologous D. melanogaster sequence. For the Y chromosome, we investigated mainly noncoding and intron sequence (to increase the number of polymorphic sites). We could not unambiguously align these noncoding sequences between the D. yakuba group and D. melanogaster. Instead, we retrieved partial coding sequence from the kl-2, kl-3, and ory gene from the published D. yakuba genome sequence (a total of 4121 synonymous sites), and aligned them to their homologous genes in D. melanogaster to estimate the amount of sequence divergence on the Y chromosome. Average pairwise divergence was estimated using DnaSP 4.0 ( with a Jukes-Cantor correction for multiple hits. Estimating Divergence Times To estimate divergence times (T) between species, we developed a new approach based on summary likelihood using the HKA framework (Hudson et al. 1987). In pairwise comparisons of species, we assume that an ancestral population of neutral parameter splits into two populations at time T having neutral parameters and f, representing a reduced population size (f 1, see Fig. 1). We estimate for each locus in each species as n 1 1 1,j 1,j i 1 i S (1) where n is the sample size and S 1,j is the observed number of segregating sites in species 1 at locus j. 1j and 2j are estimated separately for species 1 and 2, respectively. We then estimate f as m m X 1,j 2,j m mtdna 1,j 2,j m Y 1,j 2,j m m f (2a) f (2b) f (2c) for the m segments surveyed on the X (m 6), the mtdna (m 2) and the Y (m 5), respectively. In the case of D. teissieri, only one of the five Y-linked regions could be surveyed. Because this segment lacked polymorphism in all three species, we set f Y 1 for comparisons involving D. teissieri. Following the HKA framework, we estimate T j for each locus j by T (D / ) 1 j 12,j 1,j (3) where D 12,j is the average pairwise divergence between species 1 and 2 at locus j. We transform this relative time into units of absolute time as T T /2 L abs,j j 1,j j j (4) where L j is the number of silent sites and j is the mutation rate per silent site per year for locus j estimated from per site synonymous site average pairwise divergence between D. melanogaster and the D. yakuba lineage. For Y chromosome comparisons involving D. teissieri, the region surveyed contained no polymorphisms. Thus, for these comparisons, we rearranged equation 4 as T (D )/2 L. abs,j xy,j 1,j j j (5) We assume per synonymous site per year (Li 1997) for loci on the X chromosome. Mutation rates for loci on the mtdna and the Y chromosome are adjusted based on Jukes-Cantor corrected average pairwise synonymous divergences for these chromosomes to D. melanogaster relative to the X chromosome (30.2, 35.6, and 36.7 per 100 silent sites for the X, Y, and mtdna, respectively). Various other codonbased models of divergence give very similar results (i.e., a slight excess of synonymous divergence at mtdna compared to X- or Y-chromosomal markers; data not shown), with the only notable exception being the maximum-likelihood method of Yang and Nielsen (2000). This method, which accounts for unequal transition and transversion rates and unequal base and codon frequencies, suggests an approximately fivefold higher synonymous divergence at mtdna compared to X or Y chromosome. There is a strong base-composition bias in mtdna in Drosophila (i.e., the AT-content at third codon positions is 93% in the regions investigated), which will cause other distance measure to underestimate the divergence at mtdna. However, assuming a lower mutation rate for mtdna is conservative for our purpose (i.e., we will overestimate the divergence time and confidence interval for mtdna between species). To estimate credibility intervals for T, we generated simulated genealogical samples that closely resemble the actual data using the program ms (Hudson 2002), which models the neutral coalescent process for a sample of individuals from a subdivided population. For each locus, we generate a sample from two populations with n 1 and n 2 individuals that split T abs years ago (over an interval of T abs from zero to three Mya in increments of 0.1), conditioning on 1j, f and T j. For each simulated dataset, we calculate T abs,j,simulated using equation 4 above and estimate the likelihood of T as Lik(T, f, T ) abs,j 1,j j Prob[(Tabs,j,simulated T abs,j,observed) ] (6) where T abs,j,observed is the absolute divergence time of locus j estimated from the real data, and is an arbitrary rejection criterion we set to 0.05 (1/2 of the increment size for T). For each value of T abs we generated 100,000 samples. We first used an approximate Bayesian method (Haddrill et al. 2005) to estimate the population recombination parameter, 4N e r, for X-linked loci (see Supplementary Fig. 1 available online at: Based on

4 SPECIES DIFFERENTIATION IN DROSOPHILA 295 TABLE 1. Levels of silent variability in Drosophila santomea, D. yakuba, and D. teissieri for X-linked, Y-linked, and mtdna markers. S gives the number of polymorphic sites, and gives the average pairwise divergence between alleles. Mean D is the mean Tajima s D (Tajima 1989). P-values of the observed mean D values, given in parentheses, were obtained by simulation of the neutral coalescent (see Methods); / (the number of recombination events per mutation, Hudson 1987) is assumed to be 15 for X-linked loci (see Supplementary Fig. 1). Species D. santomea D. yakuba D. teissieri Chrom. No. fragments surveyed Total length surveyed Silent sites S (Silent) (% Silent) Mean D X 6 4,384 1, (0.006) Y 5 5,019 4, (0.078) mtdna 2 1, (0.072) X 6 4,384 1, ( 10 4 ) Y 5 5,019 4, (0.086) mtdna 2 1, (0.034) X 6 4,384 1, ( 10 4 ) Y mtdna 2 1, (0.022) this estimate, we set 15 for simulations of X-linked loci. We set 0 for simulations of the mtdna and the Y chromosome. For the six X-linked loci, we generate a joint likelihood surface for T abs as m abs,x abs,j 1,j j Lik(T ) Lik(T, f, T ) (7) Our estimate of T for the chromosome j is taken as the maximum of Lik(T abs,j ) and the approximate 95% confidence interval as two units of log likelihood. Code to implement the above calculations (STEbSL ver. 1) was written in Perl and C by P. Andolfatto and is available on request. Shared Polymorphism Arising by Homoplasy Assuming that all sites have an equal mutation rate, Clark (1997) showed that the expected number of shared polymorphisms arising by chance in a pair of species is simply the expectation of the hypergeometric distribution whose parameters are the number of sites surveyed, and the counts of segregating sites in the two species. Following this framework, we estimated the mean number of shared polymorphisms arising by homoplasy and their 99% confidence intervals. FIG. 2. Approximate Bayesian posterior distributions of the population mutation rate ( ) for Drosophila santomea, D. yakuba, and D. teissieri. RESULTS Species Variability We surveyed levels of polymorphism in a total of 5019 bp of Y-linked, 4384 bp of X-linked, and 1777 bp of mitochondrial DNA sequence in a sample of lines of D. santomea, lines of D. yakuba, and lines of D. teissieri (see Supplementary Table 1). Average pairwise silent (synonymous) site diversity ( ) for the X-linked markers is 0.63 % for D. santomea, 1.27 % for D. yakuba, and 2.27 % for D. teissieri (Table 1). Estimates of levels of silent diversity ( ) based on an approximate Bayesian method (Fig. 2) suggest that diversity in D. santomea (mode 0.77%, 95% CI %) is significantly lower than diversity in D. yakuba (mode 1.90%, 95% CI ), and D. teissieri (mode 2.49%, 95% CI ). Based on predictions of the neutral theory (Kimura 1983), differences in silent site diversity are expected to be proportional to differences in the effective population sizes. Thus, we infer that the effective population size of D. santomea is about half that of D. yakuba and onethird that of D. teissieri. These differences in levels of diversity between these species are consistent with expectations of their population size based on geographical distributions; D. santomea is endemic to an island, whereas D. yakuba and D. teissieri are distributed widely throughout the African continent. A high proportion of rare polymorphisms is apparent on all chromosomes in all three species, which is reflected in the strongly negative Tajima s D across all loci (a measure of the allele frequency spectrum; Tajima 1989; see Table 1 and Supplementary Table 1). This skew in the frequency spectrum of mutations is incompatible with the assumptions of a neutral population of constant size, and suggests that demographic and/or selective forces, such as population growth (Tajima 1989), purifying selection (Fu 1997; Gordo et al. 2002), bottlenecks (Fay and Wu 1999; Haddrill et al. 2005), or recurrent selective sweeps (Braverman et al. 1995) are affecting genome-wide patterns of variability in all three species. We did not find a pattern in the data that would allow us to easily distinguish among these models.

5 296 DORIS BACHTROG ET AL. TABLE 2. Hudson-Kreitman-Aguade tests on silent polymorphism and divergence at X-linked markers versus mtdna and Y chromosome, using maximum likelihood. The selection parameter k indicates how much diversity is elevated at the locus over neutral expectation given levels of divergence. Model ln L Likelihood-ratio statistic (df) P k (mtdna or Y) Drosophila santomea (outgroup: D. yakuba) Neutral (X and mtdna; k 1) (1) Selection on mtdna Neutral (X and Y; k 1) (1) Selection on Y chromosome D. yakuba (outgroup: D. santomea) Neutral (X and mtdna; k 1) (1) Selection on mtdna Neutral (X and Y; k 1) (1) Selection on Y chromosome D. teissieri (outgroup: D. yakuba) Neutral (X and mtdna; k 1) (1) Selection on mtdna Neutral (X and Y; k 1) a (1) Selection on Y chromosome a a Assuming one segregating site on the Y chromosome of D. teissieri. Chromosome Variability Both the Y chromosome and the mitochondria (mtdna) harbor reduced levels of polymorphism relative to the X chromosome in each species (Table 1). Silent divergence between the D. yakuba group and D. melanogaster at X-linked, Y- linked, and mtdna markers is comparable, suggesting that all three chromosomes have similar mutation rates (but see Methods for the possibility of a fivefold higher mutation rate of mtdna compared to nuclear markers). Lower levels of diversity are expected for mtdna and Y chromosomes because of their lower effective population size compared to the X (i.e., one-third in a neutral population with equal numbers of males and females due to differences in ploidy). In addition, nonrecombining genomes such as the Y and the mtdna are more likely to be susceptible to variation-reducing selection than genes on the recombining X chromosome (Gordo and Charlesworth 2001). In our data, the diversity reduction is most pronounced on the Y chromosome. Silent site diversity on the Y was found to be only 0.04% in both D. santomea and D. yakuba, whereas no polymorphisms were detected in a smaller fragment investigated in D. teissieri (Table 1). A statistical framework to test for heterogeneity in levels of polymorphism and divergence among unlinked markers is provided by the HKA-tests (Hudson et al. 1987). A maximumlikelihood version of the HKA test (Wright and Charlesworth 2004) reveals that variation is significantly reduced on the Y chromosome compared to the X in each species, given levels of divergence (Table 2). A similar reduction in levels of polymorphism has been detected on the Y chromosomes of other Drosophila species (Zurovcova and Eanes 1999; Bachtrog 2004), suggesting the action of linked selection on these chromosomes. In contrast, the pattern for the mtdna is quite different. Despite marked differences in levels of X-linked variability, mtdna diversity is equal in all three species, at 0.5% (Table 1). Given X-linked diversity, variability at mtdna is higher than expected in D. santomea ( 80% of X-linked variability), close to the neutral expectation in D. yakuba ( 39% of X-linked variability) and somewhat lower than expected in D. teissieri ( 20% of X-linked variability). If the mutation rate of mtdna is indeed fivefold higher than for nuclear genes (see Methods), mtdna diversity would be reduced in each species compared to X-linked genes. However, the most unusual feature of the mtdna is the significant lack of divergence between all species pairs, in comparisons to both X-linked and Y-linked markers (Table 2 and see below). Patterns of Differentiation between Species Figure 3 shows neighbor-joining trees for each of the markers investigated. The Y chromosome and the X-linked markers generally show clear differentiation between species (Fig. 3). For each of the six X-linked markers and the Y chromosome, all three species form monophyletic clusters. This is reflected in the molecular data by the large number of fixed differences between species, and very few or no shared polymorphisms at X- and Y-linked loci (Table 3). Shared polymorphic mutations between species could either result from shared ancestral polymorphism, introgression, or from homoplasy (i.e., multiple mutations at the same site). In fact, all of the shared polymorphism observed at the nuclear markers can be accounted for by homoplasy alone (see Table 3). In contrast, the mtdna shows a strikingly different pattern of differentiation. In the neighbor-joining tree for mtdna, none of the three species form a monophyletic group (Fig. 3). Although D. teissieri shows some degree of differentiation from the other two species, D. santomea and D. yakuba are intermingled in the tree. This is surprising, given the observed differentiation of these species at X- and Y-linked markers. There is one shared haplotype between D. teissieri and D. santomea, as well as between D. yakuba and D. santomea, but only one fixed nucleotide difference between D. yakuba and D. teissieri, and no fixed differences in the other two species comparisons (Table 3 and Supplementary Fig. 2 available online at: Contrary to the pattern observed at nuclear loci, homoplasy can-

6 SPECIES DIFFERENTIATION IN DROSOPHILA 297 FIG. 3. Neighbor-joining trees for X-linked regions, the Y chromosome, and the mitochondria. Sequence data for the nonrecombining Y chromosome and the mtdna were pooled for tree construction. Species are color-coded (red, Drosophila santomea; green, D. yakuba; and blue, D. teissieri).

7 298 DORIS BACHTROG ET AL. TABLE 3. Fixed differences (above diagonal) and shared polymorphism (below diagonal) between Drosophila yakuba, D. santomea, and D. teissieri for X-linked, Y-linked, and mitochondrial markers. Expected numbers of shared polymorphisms due to multiple hits and 99% confidence limits are given in parentheses. D. yakuba D. santomea D. teissieri X-chromosome D. yakuba D. santomea 4 (0.74; 0 4) 68 D. teissieri 4 (2.39; 0 6) 3 (1.00; 0 4) Y-chromosome D. yakuba a D. santomea 0 (0.01; 0 0) 13 a D. teissieri 0 (0) 0 (0) mtdna D. yakuba 0 1 D. santomea 3 (0.09; 0 1) 0 D. teissieri 3 (0.08; 0 1) 2 (0.07; 0 1) a In a smaller fragment (472 base pairs) studied. not account for the observed amount of shared polymorphism on the mitochondria (Table 3). Note that in the case of high levels of variability in substitution rates, as one finds for mtdna, the probability of homoplasy due to recurrent mutation at the hypermutable sites is increased. But homoplasy by this process is not an adequate explanation for the mtdna data, given the high overall mutation rate and low divergence at mtdna. Although D. yakuba and D. santomea appear randomly intermingled in the tree, a permutation test based on haplotype frequencies (Hudson et al. 1992b) reveals significant differentiation between them (P 0.001). We employed a new summary-likelihood approach, based on the HKA framework (Hudson et al. 1987), to infer divergence times between the three species at X-linked, Y- linked, and mtdna markers. The method estimates the most likely time of the species split (in years) based on two summaries of the data: the level of polymorphism within a species,, and the average pairwise divergence between species, D xy. For all three species comparisons, the estimated divergence times are very similar for the X- and Y-linked markers (Fig. 4 and Table 4). We estimate that D. yakuba and D. santomea diverged about 0.5 Mya whereas D. teissieri diverged from their ancestor about one Mya. In sharp contrast, the most likely divergence time for the mtdna is zero in each species comparison. Moreover, the confidence intervals for the mtdna time estimate do not overlap with divergence times inferred for the X and Y chromosomes (see Fig. 4 and Table 4). Note that the confidence intervals in divergence time estimated for the mtdna would be even narrower if a higher mutation rate of mtdna, as indicated by the Yang and Nielsen (2000) method, were assumed. This analysis strongly supports different evolutionary histories for nuclear and mtdna genomes of the three species. DISCUSSION Comparisons of genealogical histories of markers of closely related species provide a valuable tool for testing alternative hypotheses of species origin (Wang et al. 1997; Kliman et al. 2000; Machado et al. 2002; Hey and Nielsen 2004; FIG. 4. Relative log-likelihoods of divergence times (in millions of years) for mtdna X chromosome, and Y chromosome for (A) Drosophila yakuba vs. D. santomea, (B) D. teissieri vs. D. yakuba, and (C) D. tessieri vs. D. santomea. The dashed line indicates two log-likelihood units. Osada and Wu 2005; Won and Hey 2005; Won et al. 2005; Dopman et al. 2005). Genes involved in hybrid incompatibilities or species-specific adaptations should display higher levels of differentiation between species relative to other regions of the genome (Palopoli et al. 1996; Wu and Ting 2004). Thus, multilocus scans of the genome can yield information about the role of natural selection and continuing gene flow in the formation of new species and potentially lead to the identification of genomic regions underlying reproductive isolation. The D. yakuba subgroup is a promising model system for identifying such regions because these species are only recently diverged; D. santomea and D. yakuba are only partially reproductively isolated and form hybrids in nature (Lachaise et al. 2000).

8 SPECIES DIFFERENTIATION IN DROSOPHILA 299 TABLE 4. Inference of population parameters and divergence times under an allopatric speciation model with no migration a. Comparison ˆ ˆf ˆT b X Y mtdna (95% CI) c D. yakuba D. santomea D. teissieri D. yakuba D. teissieri D. santomea D. yakuba D. santomea D. teissieri D. yakuba d D. teissieri D. santomea d D. yakuba D. santomea e D. teissieri D. yakuba D. teissieri D. santomea a The model is essentially that of Hudson et al. (1987) except that the ancestral population is assumed to be the same size as the larger of the two populations (instead of the average); the population mutation parameter is defined as for the first species in the comparison and f for the second species (see Fig. 1). b Divergence time in Mya, estimated using equation 4 in Methods. c The approximate 95% credibility interval on T by simulation to the nearest 0.05 (see Methods). d Because diversity is zero in both species, we set f 1 and T was estimated using equation 5 (see Methods). e Equation 4 yielded a negative value which we interpret as not different than zero under the assumed model. Differentiation at Nuclear Loci Observations from the wild have suggested a rate of hybridization between D. yakuba and D. santomea of about 1% on the island of São Tomé (Lachaise et al. 2000). None of our D. yakuba lines is sampled from the contact zone, so we cannot directly address the question of ongoing gene flow between D. santomea and D. yakuba in the context of the hybrid zone. Instead, our genealogical approach should allow us to detect historic introgression between these species (or backcrossed alleles from the D. yakuba population in São Tomé to mainland Africa). Despite the formation of hybrids in nature between D. yakuba and D. santomea, the nuclear loci that we surveyed do not provide strong evidence for continuing gene flow between these species. The differentiation between species at X- and Y-linked markers is most easily visualized as each species forming monophyletic clusters in neighbor-joining trees (Fig. 3). The estimates of speciation time based on the X-linked loci (jointly) and the Y chromosome under the simple allopatric model suggests that D. yakuba and D. santomea diverged about 0.5 Mya, and D. teissieri split from a yakuba santomea ancestor about one Mya (Table 4). However, these estimates could be significantly downwardly biased if there has been historical gene flow between species (Osada and Wu 2005). Because only female hybrids are fertile (Lachaise et al. 2000) in accordance with Haldane s rule (Coyne and Orr 2004) little or no gene flow is expected for the Y chromosome. However, backcrosses of fertile hybrid females could result in considerable introgression of X-linked genes. To test for significant heterogeneity in speciation times between D. yakuba and D. santomea inferred from individual X-linked loci, we performed a likelihood ratio test, comparing the likelihood of six locus-specific speciation times to the null hypothesis of one speciation time for all loci. We detected no significant heterogeneity among loci assuming no recombination ( 2 ln(lik) 10.4, df 5, P 0.07). However, when allowing / 15, consistent with recombination rates estimated from the data (see Supplementary Fig. 1), we do detect significant heterogeneity among loci ( 2 ln(lik) 17.8, df 5, P 0.003). Thus, although the overall pattern for X-linked genes suggests more differentiation than for the mtdna, specific X-linked loci may be experiencing more introgression than others. Further quantification of this apparently heterogeneous pattern across the X chromosome must await data from more loci and better estimates of recombination rates. Sex chromosomes may not be representative of the rest of the genome because they are known to play a large role in hybrid sterility in Drosophila and other species (Coyne and Orr 2004). Thus, sex chromosomes may move through hybrid zones much less readily than do autosomes. One reason is that recessive Muller-Dobzhansky incompatibilities between the X-linked and autosomal loci are revealed in males because they are hemizygous for the X chromosome (Coyne et al. 2004). In contrast, recessive incompatibilities among autosomal loci are shielded from negative selection by their homolog in male hybrids. In support of this hypothesis, there appears to be extensive gene flow among autosomal but not sex-linked genes in hybridizing species of Ficedula flycatchers (Saetre et al. 2003). Genetic analyses of hybrid sterility in D. yakuba and D. santomea have shown that hybrid male sterility is caused by at least three genes on the X chromosome and at least one on the Y, with the cytoplasm and large sections of the autosomes having no effect (Coyne et al. 2004). This is in accordance with our finding of higher levels of differentiation at X- and Y-linked genes relative to the mtdna. It will be of interest to investigate levels of introgression at autosomal genes in D. yakuba and D. santomea, which may reveal a different picture from X- or Y-linked loci. Lack of Differentiation at mtdna In contrast to the pattern observed at nuclear markers, mtdna shows very little differentiation between the three species investigated (see also Lachaise et al. 2000). Mitochondrial phylogenies provide no resolution of the three species, and our population genetic analysis shows there is only

9 300 DORIS BACHTROG ET AL. subtle differentiation between the species. Several processes could explain the discordance in levels of divergence for the mtdna versus nuclear markers, particularly in comparisons involving D. teissieri. First, an extremely low substitution rate of mtdna in the D. yakuba group and/or incomplete sorting of ancestral haplotypes could result in low levels of sequence divergence among the three species investigated. Silent site divergence for the regions investigated between the D. yakuba group and D. melanogaster (see Methods) is about 37% for mtdna. This estimate is slightly higher than the estimate of 30% silent site divergence for X-linked genes and 36% for Y-linked genes. Moreover, the Yang and Nielsen (2000) method of estimating divergence suggests a fivefold higher substitution rate for the mtdna relative to nuclear genes. Thus, the mtdna, if anything, appears to have a higher mutation rate than nuclear genes. In addition, levels of mtdna diversity within species are not particularly unusual relative to X-linked markers. Finally, the lower effective population size of mtdna implies that ancestral polymorphism will be lost more quickly by genetic drift at mtdna relative to X-linked genes. Thus, the retention of ancestral haplotypes of mtdna (i.e., incomplete lineage sorting) is unlikely to explain the observed sharing of mtdna haplotypes among the species in the D. yakuba group, when X-linked and Y- linked loci show few shared polymorphisms. An alternative to shared ancestral polymorphism is that recent introgression of mtdna due to interspecific hybridization has led to the observed pattern. In the laboratory, fertile female hybrids are observed between crosses of D. yakuba and D. santomea (Lachaise et al. 2000), whereas hybridization attempts between D. teissieri and the other two species generally fail (Lemeunier et al. 1986; Lachaise et al. 2000). Thus, although hybridizations between D. teissieri and D. yakuba/d. santomea are difficult, it is possible that some strains of D. teissieri can produce viable offspring with other strains of D. yakuba or D. santomea in nature. Indeed, strainspecific variation in the levels of reproductive isolation has been observed in crosses involving D. melanogaster and D. simulans (Hutter and Ashburner 1987). Once introgressed, foreign mtdna can replace the resident mtdna in a species either through genetic drift or natural selection. Genetic drift would lead to patterns of within-species variability resembling a neutral genealogy, whereas directional selection (on either the mitochondria itself or other cytoplasmic factors) would distort patterns of within-species variability towards a star-like genealogy (Hudson 1990; Simonsen et al. 1995). Silent site polymorphisms at mtdna are skewed significantly towards low-frequency variants in all three species (Table 1), consistent with directional selection driving the introgression and fixation of mtdna across species boundaries. However, the X- and Y-linked markers also reveal a general excess of low-frequency variants in all three species investigated. This general distortion in the expected frequency spectrum makes it difficult to distinguish mtdna-specific selection from effects of processes that have a genome-wide impact on variability patterns (such as demography or pervasive selection). However, because hybridization is rare or almost absent between D. teissieri and D. yakuba/d. santomea, mtdna that may be occasionally introgressed between species would almost certainly be lost by genetic drift if it were selectively neutral. Takahata and Slatkin (1984) showed that, if hybrids are viable, the time required for an introgressed mitochondrial genotype to become common is of the order of the inverse of the immigration rate. Although the rate of hybridization between D. yakuba and D. santomea may be as high as 1% in the contact zone, the hybrid zone is narrow relative to the geographic distribution of the two species (particularly for D. yakuba). Thus, the overall hybrid production between these two species is low. The lack of viable hybrids observed between D. teissieri and the other two species implies that the immigration rate of mtdna between them must be even lower. Thus, it would take an extremely long time to introgress and replace the mtdna between D. teissieri and D. yakuba/d. santomea by neutral drift alone. Moreover, only a small amount of selection against immigrant mtdna is sufficient to prevent its establishment (Takahata and Slatkin 1984). These considerations strongly suggest that positive selection is the most likely cause of the displacement of mtdna at least between D. teissieri and the sister species pair D. yakuba/d. santomea. Target of Selection Driving mtdna Introgression Although it appears that selection has driven the introgression of mtdna across species boundaries, we cannot distinguish whether the target of selection is the mitochondria itself, or whether selection operated on other maternally inherited factors. Introgression of distinct mtdna haplotypes as a result of fitness differences has been demonstrated experimentally in Drosophila using microinjection studies. In heteroplasmic D. melanogaster lines that contained mtdna injected from D. mauritiana, the foreign mtdna most often completely replaced the resident D. melanogaster mtdna (Niki et al. 1989). The D. mauritiana strain used in that study had the same mtdna haplotype that has introgressed in nature from D. simulans into D. mauritiana (Ballard 2000). However, direct selection on mitochondria across species boundaries will be counterbalanced by cytonuclear interactions between the mtdna and the nuclear genome. In fact, epistatic interactions between genes encoded on the mtdna and in the nuclear genome were observed in backcross hybrids between D. simulans and D. mauritiana (Sackton et al. 2003). Our finding of introgression of mtdna across species boundaries in the D. yakuba group imply that these cytonuclear interactions (if present) must be weaker than the selective pressure driving mtdna introgression. Selection can also change allele frequencies at a locus not responsible for fitness differences (i.e., the hitchhiking effect). Because mtdna in Drosophila is inherited maternally, any other maternally inherited factor can influence the evolution of mtdna. For instance, Wolbachia are maternally inherited bacteria that manipulate the reproduction of their hosts in ways that enhance their own transmission (O Neill and Karr 1990). One such mechanism is cytoplasmic incompatibility. Crosses involving infected females produce a normal number of offspring, while cytoplasmic incompatibility is expressed in crosses between infected males and uninfected females (O Neill and Karr 1990). This can lead to an increase of Wolbachia frequency in natural populations.

10 SPECIES DIFFERENTIATION IN DROSOPHILA 301 Because Wolbachia spreads by maternal cytoplasmic transmission, mtdna will hitchhike along as well. As a result, a Wolbachia invasion will be accompanied by an increase in frequency of a single mtdna haplotype (Turelli et al. 1992). Interspecific horizontal transfer of Wolbachia has been observed in wasps (Huigens et al. 2004), and could potentially explain the observed patterns of variation at mtdna in the D. yakuba species group. A single, recent Wolbachia invasion transmitted through all three species of the D. yakuba species group could account for the low level of divergence of mtdna in these species. Indeed, changes in mtdna variants have been found associated with the spread of a Wolbachia infection in a D. simulans population (Turelli et al. 1992): all infected flies share the same mtdna haplotype, whereas uninfected flies are polymorphic. However, none of the Wolbachia strains of D. yakuba investigated so far has been found to cause cytoplasmic incompatibility, either in the lab or in nature (Charlat et al. 2004). Wolbachia could also provide direct fitness effects to its host, which could promote the spread of a Wolbachia infection in the absence of cytoplasmic incompatibility. Indeed, enhanced survival and/or fecundity effects associated with Wolbachia have been observed in four strains of D. melanogaster, while another strain responded positively to Wolbachia removal (Fry et al. 2004). Nevertheless, none of these three species in the D. yakuba species group was found to be fixed for a specific Wolbachia strain. The frequency of Wolbachia infections appears to be 10 30% in D. yakuba, 30% in D. santomea, and 95% in D. teissieri (Lachaise et al. 2000; Charlat et al. 2004). Wolbachia could have been lost secondarily; however, a study in natural D. yakuba populations suggests that transmission of maternal Wolbachia is almost perfect (Charlat et al. 2004). This finding argues against a role of Wolbachia-induced incompatibility or fitness effects as the cause for the rapid spread of one mtdna type through the three Drosophila species. An alternative scenario is that all three species could have faced a burden of Wolbachia load prior to a single mtdna mutation that rescued them from Wolbachia-induced incompatibility. In this case there would be a direct selective advantage for the variant, not accompanying the spread of any Wolbachia strain. With rare hybridization, this mtdna variant might have spread across the three species. Yet another possibility is that another, as yet undiscovered, endosymbiont may cause similar Wolbachia-like effects. It will therefore be of interest to investigate whether the species of the D. yakuba subgroup carry such organisms. MtDNA has been found to obscure species boundaries in other taxa as well. Indeed, several examples of mitochondrial introgession unaccompanied by nuclear introgression have been reported. For example, mtdna of D. pseudoobscura and D. persimilis are very similar in sympatric populations while completely divergent in allopatry (Powell 1983). Other examples of introgession of mtdna in the absence of nuclear gene flow have been observed in carabid beetles (Sota and Vogler 2001), Hawaiian crickets (Shaw 2002), mice (Ferris et al. 1983), elephants (Roca et al. 2005), and trout (Doiron et al. 2002). However, in all these studies hybrids between species are (at least partially) viable and fertile, which means that genetic drift alone can be responsible for the introgression of a mitochondrial haplotype across species boundaries (Takahata and Slatkin 1984). In contrast, D. teissieri and its sister species are very strongly reproductively isolated. Thus, positive selection has to be invoked to explain the introgression and complete replacement of mtdna across D. teissieri and D. yakuba/d. santomea. These findings demonstrate that distinct regions of the genome can have very different evolutionary dynamics in the process of species formation. Although natural selection is usually thought of as accentuating divergence between species, our results imply that it can also act as a homogenizing force in the case of the mtdna of these three species. ACKNOWLEDGMENTS We thank S. Charlat and M. Long for providing fly lines. We thank D. Hoyle, the Wildlife Conservation Society (WCS), and the Cameroon Ministry of Environment and Forests (MINEF) for making new collections of Drosophila yakuba and D. tessieri possible. All new sequences surveyed in this study have been submitted to Genbank under accession numbers DQ DQ This work was funded in part by a Biotechnology and Biological Sciences Research Council Grant to PA and a National Institute of Health Grant GM64590 to AC. DB is supported by a Postdoctoral Fellowship from the Austrian Academy of Sciences. KT is supported by an A. P. Sloan Postdoctoral Fellowship. PA is supported by an A. P. Sloan Fellowship in Molecular and Computational Biology. LITERATURE CITED Bachtrog, D Evidence that positive selection drives Y-chromosome degeneration in Drosophila miranda. Nat. Genet. 36: Ballard, J When one is not enough: introgression of mitochondrial DNA in Drosophila. Mol. Biol. Evol. 17: Barton, N., and G. Hewitt Adaptation, speciation and hybrid zones. Nature 341: Braverman, J., R. Hudson, N. Kaplan, C. Langley, and W. Stephan The hitchhiking effect on the site frequency spectrum of DNA polymorphisms. Genetics 140: Cariou, M., J. Silvain, V. Daubin, J. Da Lage, and D. Lachaise Divergence between Drosophila santomea and allopatric or sympatric populations of D. yakuba using paralogous amylase genes and migration scenarios along the Cameroon volcanic line. Mol. Ecol. 10: Charlat, S., J. Ballard, and H. Mercot What maintains noncytoplasmic incompatibility inducing Wolbachia in their hosts: a case study from a natural Drosophila yakuba population. J. Evol. Biol. 17: Clark, A. G Neutral behavior of shared polymorphism. Proc. Natl. Acad. Sci. USA 94: Coyne, J. A., and H. A. Orr Speciation. Sinauer Associates, Sunderland, MA. Coyne, J., S. Elwyn, S. Kim, and A. Llopart Genetic studies of two sister species in the Drosophila melanogaster subgroup, D. yakuba and D. santomea. Genet. Res. 84: Dobzhansky, T Is there gene exchange between Drosophila pseudoobscura and Drosophila persimilis in their natural habitats? Am. Nat. 107: Doiron, S., L. Bernatchez, and P. Blier A comparative mitogenomic analysis of the potential adaptive value of Arctic charr mtdna introgression in brook charr populations (Salvelinus fontinalis Mitchill). Mol. Biol. Evol. 19: Dopman, E. B., L. Perez, S. M. Bogdanowicz, and R. G. Harrison Consequences of reproductive barrieres for genealogical