To investigate sex-biased gene expression in D. ananassae, we designed a species-specific microarray of PCR-amplified exon sequences from 148 genes (Additional file 1) whose orthologs had been previously classified as male-biased, female-biased or unbiased in D. melanogaster (see Methods). This set of genes was of particular interest because the vast majority (136 genes) had been studied at the population-genetic level in D. melanogaster and estimates of the rate of adaptive evolution in the melanogaster subgroup were already available 2 3 . After quality control, we obtained sufficient hybridization signal to reliably classify 60 genes (22 male-biased, 13 female-biased, and 25 unbiased). Of the remaining genes, we were able to classify 69 (21 male-biased, 10 female-biased, and 38 unbiased) using the D. ananassae whole-genome microarray data of Zhang et al. 11 . In total, 129 of the 148 genes (87%) were classified, with 43 male-biased, 23 female-biased, and 63 unbiased (Additional file 2).

To examine the conservation of sex-biased gene expression between species, we compared the above D. ananassae classifications to those previously determined for D. melanogaster. The majority of genes (59%) showed a conserved expression pattern, with the same bias in both species (Figure 1A). However, 37% of the genes were classified as sex-biased in only one species, and 4% showed the opposite sex-bias in the two species (Figure 1A). Overall, male-biased genes showed the greatest conservation: 72% of the genes with male-biased expression in D. ananassae also had male-biased expression in D. melanogaster, 65% of the genes with female-biased expression in D. ananassae also had female-biased expression in D. melanogaster, and 48% of the genes with unbiased expression in D. ananassae also had unbiased expression in D. melanogaster. Of the genes that were sex-biased in only one species, most (69%) were sex-biased in D. melanogaster, but unbiased in D. ananassae (Figure 1B). Three genes (CG13690, CG3024, CG4593) were male-biased in D. ananassae, but female-biased in D. melanogaster, while two genes (CG12684, CG7387) were female-biased in D. ananassae, but male-biased in D. melanogaster (Figure 1B). For CG13690, the molecular function is described as ribonuclease H activity and RNA binding. It is involved in the RNA metabolic process. The molecular function of CG3024 (torp4a) is described as ATP binding and unfolded protein binding. The gene is involved in protein folding. For CG7387, the function is given as unfolded protein binding and heat shock protein binding. It is also involved in protein folding. For the two remaining genes, no functions or biological processes are described. All described functions refer to what is known for D. melanogaster.

For analysis of evolutionary rates and tests for adaptive evolution, it is critical to have an appropriate outgroup species. Two recent molecular phylogenetic studies suggested that D. atripex and/or D. phaeopleura might serve as an appropriate outgroup to D. ananassae for these purposes 12 13 . However, these studies used a small number of loci that sometimes gave conflicting results. To further investigate the phylogenetic relationship of these species, we used the concatenated amino acid sequences of 12 genes that we sequenced in both D. atripex and D. phaeopleura (Additional file 3) and for which sequence data were available from D. ananassae, D. melanogaster, D. simulans, and D. pseudoobscura. The concatenated alignment of 3,446 amino acid positions was used to generate a phylogenetic tree by Bayesian inference (Figure 2). The tree topology was strongly supported (100% posterior clade probabilities for each node) and indicated that D. atripex and D. phaeopleura are more closely related to each other than either is to D. ananassae. Thus, both of these species can be used as an outgroup to D. ananassae. Furthermore, the divergence between D ananassae and D. atripex/D. phaeopleura is similar to the divergence between D. melanogaster and D. simulans, which facilitates the comparison of evolutionary patterns between the melanogaster and ananassae subgroups (see below).

We surveyed DNA sequence polymorphism in a sample of 12 isofemale lines from Bangkok, Thailand for a subset of 43 genes used in our microarray analysis (Additional file 3). This set included 17 male-biased genes, 12 female-biased genes, and 14 unbiased genes. Fourteen of these genes differed in their sex-bias classification between D. ananassae and D. melanogaster. In addition, we sequenced all genes in a single strain of either D. atripex or D. phaeopleura to use as an outgroup. For 13 genes we were able to get sequences from both D. atripex and D. phaeopleura, revealing that both exhibit similar levels of divergence to D. ananassae (on average, 20% at synonymous sites and 2% at nonsynonymous sites) and that the choice of outgroup does not bias estimates of interspecific divergence (Table 1).

Comparison of the divergence between D. ananassae and D. atripex/D. phaeopleura with the divergence between D. melanogaster and D. simulans for the different sex-bias classes of genes suggests that the accelerated rate of evolution of male-biased genes is limited to the melanogaster subgroup (Table 1). When comparing D. melanogaster and D. simulans, values of d _N and d _N/d _S are consistently higher for male-biased genes than for female-biased or unbiased genes, which is not the case for comparisons of D. ananassae to either D. atripex or D. phaeopleura (Table 1).

However, there is no significant difference in d _N /d _S between male-biased genes on the different lineages (Mann-Whitney test, P > 0.05). It should be noted, however, that the number and identity of the genes are not constant among the three comparisons.

Levels of DNA sequence polymorphism in the Bangkok population of D. ananassae are comparable to levels in an ancestral African population of D. melanogaster (Table 2). On average, synonymous polymorphism is slightly higher in D. ananassae than in D. melanogaster, suggesting that the former has a slightly larger effective population size (N _e). This is consistent with the Bangkok population representing an ancestral population of D. ananassae 8 . Taken together, levels of polymorphism and divergence suggest that our D. ananassae population and outgroup species are appropriate for comparison to the African D. melanogaster population with D. simulans as an outgroup.

We also investigated levels of synonymous polymorphism on the X chromosome relative to the autosomes in both D. ananassae and D. melanogaster. This provides an estimate of the effective population size of males and females in each species. The ratio is expected to be 3/4 if there is an equal number of breeding males and females in the population. If there is sexual selection acting on males, the male effective population size will be reduced and the X:autosome polymorphism ratio will be increased above 3/4. After standardizing polymorphism by divergence to correct for possible differences in mutation rate between the X and the autosomes, we find that the X:autosome polymorphism ratio is greater than 3/4 for both species (Table 3). To test the significance of this, we multiplied the X chromosomal values by 4/3 and compared them to the autosomal values with a Mann-Whitney test. For D. melanogaster, the difference was significant (P = 0.04), indicating that the original ratio was significantly greater than 3/4. For D. ananassae, an X:autosome polymorphism ratio of 3/4 could not be rejected (P = 0.13). There was no significant difference in the X:autosome polymorphism ratio between D. ananassae and D. melanogaster (P = 0.42).

To evaluate the type of selection operating on individual genes, we applied single-locus McDonald-Kreitman (MK) tests 14 to the 43 genes for which we had polymorphism data from D. ananassae and divergence data from either D. atripex or D. phaeopleura. In total, six genes (14%) gave a significant MK test result (Additional file 3). Half of the significant genes departed from neutrality in the direction of positive selection (i.e. a relative excess of nonsynonymous divergence) (Table 4), while the other half departed from neutrality in a pattern consistent with either balancing or weak purifying selection (i.e. a relative excess of nonsynonymous polymorphism). These are likely to be cases of weak purifying selection, as average values of Tajima's D 15 are significantly negative at nonsynonymous sites (Table 5), which suggests that many segregating nonsynonymous polymorphisms are slightly deleterious.

A comparison of our MK test results to those from D. melanogaster/D. simulans 2 3 revealed that, in all but one case (CG10750), the individual genes showing significant evidence for positive selection (i.e. those with MK-test P-values less than 0.05) differed between the two lineages (Table 4). Most genes significant for positive selection were male-biased in at least one of the species. However, the proportion of positively-selected male-biased genes in D. ananassae (1/17 = 6%) was not as high as in previous studies of D. melanogaster where 7/33 (21%) of autosomal male-biased 3 , and 7/17 (41%) of X-linked male-biased genes 2 gave significant MK tests for positive selection. Of the three genes with significant MK tests for positive selection in D. ananassae, one (CG6980) showed conserved male-biased expression between D. ananassae and D. melanogaster, while two (CG14717 and CG10750) were male-biased in D. melanogaster, but unbiased in D. ananassae (Table 4).

To further investigate expression class-dependent patterns of evolution, we constructed 12 groups of genes (Table 6). The first nine classes were partitioned with regard to sex-bias in D. ananassae and D. melanogaster. This allowed us to compare genes with conserved sex-bias between the two species with genes showing sex-bias in only one species or genes showing a reversal of sex-bias between species. The final three classes were partitioned solely on expression in D. ananassae, without regard to expression in D. melanogaster. Application of the MK test to the summed values of polymorphism and divergence within each group 16 revealed a significant departure from neutrality in the direction of positive selection for genes with conserved male-biased expression and for genes with male-biased expression private to D. melanogaster, whereas the female-biased and unbiased genes did not differ significantly from the neutral expectation (Table 6). The genes with male-biased gene expression private to D. ananassae also did not differ significantly from the neutral expectation.

We also used a multi-locus version of the MK test to estimate α, the fraction of amino acid replacements between species that were fixed by positive selection 17 . For this, we divided the genes into six different groups. First, we considered male-, female-, and unbiased genes that had conserved sex-biased expression between D. melanogaster and D. ananassae. Second, we considered the genes according only to their bias in D. ananassae. We could not consider all 12 groups of genes mentioned above, because some of the groups contained too few genes to allow maximum likelihood analysis (e.g. there were only two genes with male-biased expression private to D. ananassae). Because the segregation of slightly deleterious nonsynonymous mutations can lead to a downward bias in the estimate of α, we excluded all polymorphisms (both synonymous and nonsynonymous) segregating at frequency ≤ 15% from our analysis 18 .

For the genes with conserved sex-biased expression between species, we observed a significant signal of positive selection only for those with conserved male-biased expression, although in all cases the mean estimate of a was greater than zero (Figure 3). However, when we considered the genes only by their sex-bias classification in D. ananassae we found evidence for adaptive evolution in all three classes of genes, with higher α estimates and more significant departures from neutrality detected for female-biased and unbiased genes than for male-biased genes (Figure 3). Thus, the increased rate of adaptive evolution seen for male-biased genes appears to be limited to those with conserved male-biased expression between the two species. For D. ananassae, estimates of α were 43%, 60%, and 53% for male-biased, female-biased, and unbiased genes, respectively.

To analyze sex-biased gene expression in D. ananassae, we designed a species-specific PCR-amplicon microarray. We began with a set of 148 genes that had previously been studied in D. melanogaster 2 3 and used the available D. ananassae genome sequence (Assembly August 2005; http://genome.ucsc.edu/ 29 ) to identify their orthologs and design PCR primers that amplify exonic sequences (mean length = 458 bp). A complete list of genes and PCR primers is provided in (Additional file 1). PCR products were purified using genPURE PCR 96 well filter plates (Genetix) and spotted on UltraGAPS slides (Corning) using a GeneMachines OmniGrid Accent microarrayer. D. ananassae genomic DNA (gDNA) was also spotted as a control. Eight replicates of each gene probe were spotted per array, with each replicate in a different subarray. 12-14 control gDNA probes were spotted per subarray.

For hybridization, we extracted total RNA from four- to five-day old males and females using Trizol reagent (Invitrogen) and the manufacturer's protocol. Two inbred strains of D. ananassae from Kota Kinabalu, Borneo were used for RNA extraction 8 . Reverse transcription was performed using 25 μg of total RNA and an anchored oligo(dT) primer. cDNA was labelled with fluorescent dyes (Alexa Fluor 555 and 647) using the amino-allyl labelling system (Invitrogen). Labelled male and female cDNA was competitively hybridized to the arrays for 20 hours at 42 C. Arrays were scanned using a Genetix aQuire 2-laser microarray scanner and the Genetix Qscan software. In total, we performed 12 replicate hybridizations. Six of these were biological replicates, i.e. from different RNA extractions performed at different times, while the other six were "dye-swap" replicates where we exchanged the fluorescent dyes used to label male and female cDNA within each biological replicate.

Prior to statistical analysis, normalization of the two dye channels was performed. Since our arrays contained only a subset of the genome with a non-random distribution of sex-biased genes, we used the gDNA spots as controls for normalization. The signal for each spot was calculated by subtracting the median local background from the mean spot intensity. If this result was negative for a given channel, that channel was assigned a value of 0.5 according to the MINIMUM approach of LIMMA 30 . For each replicate subarray, the red/green ratio was calculated for each of the gDNA control spots and the mean of these ratios was taken as the normalization factor. The raw red value of each spot was then multiplied by this factor to obtain the normalized red value used to calculate the corrected red/green ratio. Note that this is a local normalization, with a different normalization factor for each replicate subarray. As a quality control measure, only spots with a mean intensity at least 20% above the mean local background in at least one of the channels were used. For any one array, we required that at least half of the replicate spots per gene (in this case four) displayed adequate signal as defined above. The median red/green ratio (after normalization) of these replicate spots was used for statistical analysis using the BAGEL software 31 32 . To determine the P-value cut-off used to define significant sex-biased genes, we performed randomizations of the BAGEL input file to estimate the false-discovery rate (FDR) for the dataset. For classification of sex-biased genes, we used a P-value cut-off of 0.01, which corresponds to a FDR of 10%. All new microarray data have been deposited in NCBI's Gene Expression Omnibus 33 and are accessible through GEO Series accession number GSE19096 http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE19096.

For the polymorphism survey, we used 12 inbred strains of D. ananassae from Bangkok, Thailand. Like the Kota Kinabalu strains used for the microarray experiments, these strains were included in a previous population genetic survey of intronic loci and found to be in the ancestral range of D. ananassae 8 . Thus, we expect little differentiation between these populations. We chose the Bangkok population for the polymorphism survey because a larger number of strains were available.

One strain each of D. atripex and D. phaeopleura (kindly provided by M. Schug) were used as outgroups. We used the D. ananassae genome (Assembly August 2005; http://genome.ucsc.edu/ 34 ) to design PCR primers flanking the coding sequence of 43 target genes. These genes were chosen to represent different categories of sex-biased gene expression on the basis of the microarray data described above. Following PCR, the amplified products were purified with ExoSAP-IT (USB, Cleveland, OH) and sequenced on both strands using BigDye version 1.1 chemistry and a 3730 automated sequencer (Applied Biosystems, Foster City, CA). The PCR primers were also used as sequencing primers in the polymorphism survey. A complete list of primers is provided in Additional file 5. For some genes, we were unable to get successful PCR or DNA sequence from all 12 D. ananassae strains (Additional file 3). The average number of strains sequenced per gene was 11 and we required at least eight sequences for a gene to be included in our analysis. All new DNA sequences have been submitted to the GenBank/EMBL databases under accession numbers FN546265-FN546780.

Sequences were edited using DNAstar (Madison, WI) and multiple alignments were calculated using MUSCLE 35 http://www.ebi.ac.uk/Tools/muscle/. Polymorphism and divergence statistics were calculated using DnaSP 4.5 36 . For McDonald-Kreitman (MK) table data we used the number of segregating mutations instead of the number of segregating sites as some genes had sites with three segregating variants. For divergence, we considered only sites with fixed differences between all D. ananassae lines and a single D. atripex (or D. phaeopleura) sequence. Multilocus Tajima's D tests were performed using the HKA program http://lifesci.rutgers.edu/~heylab/heylabsoftware.htm. To calculate the fraction of positively selected amino acid substitutions, a, with the method of 17 , we used the DoFE program http://www.lifesci.sussex.ac.uk/home/Adam_Eyre-Walker/Website/Software.html. For MK tests and a calculations, the D. atripex sequence was used as the outgroup whenever available. When the D. atripex sequence was not available, D. phaeopleura was used as the outgroup.

For 13 genes we were able to get sequences from both D. atripex and D. phaeopleura [Additional file 3]. In addition, we downloaded the amino acid sequences of D. melanogaster, D. simulans, D. ananassae, and D. pseudoobscura for the entire set of genes analyzed from Flybase http://flybase.org/. For one gene (CG18418), we were unable to get the orthologous sequence from D. pseudoobscura. This left us with 12 genes that could be aligned across all species (CG2577, CG3004, CG3024, CG4593, CG6459, CG7840, CG8277, CG9135, CG10853, CG11379, CG15336, and CG15717). Amino acid sequences of these 12 genes were concatenated for phylogenetic reconstruction Bayesian inference (BI), as implemented in MrBayes version 3.1.1 37 , was used to reconstruct the phylogeny with D. pseudoobscura set as the outgroup. For the BI analysis, two distinct 10,000 generation runs were conducted (three incrementally heated chains with model jumping between fixed-rate amino acid models were used and trees were saved to a file every 10 generations). Identical topologies were recovered from both runs. A burn-in period of 2,500 generations was determined graphically by plotting likelihood values for each sample. The results were presented in the form of a 50% majority-rule consensus tree in which trees corresponding to the burn-in period were discarded. Support for the nodes was given by posterior probability estimates of clades. Using model jumping, only the Jones (JTT) model 38 contributed to the final result with 100% posterior probability. This was also supported by the program ProtTest version 2.0 39 , which determines the best-fit substitution model for amino acid data under a likelihood framework. Within this framework, the Akaike information criterion 40 selected the JTT 38 model including the proportion of invariable sites and the gamma distribution of rate variation among sites (JTT + I + G) as the best-fitting model with an Akaike weight of 0.54. The second-best model was JTT+G with an Akaike weight of 0.46. Trees were visualized with TreeView 41 .