We based our list of Acp genes on previous published lists (Wigby et al. 2020; Hurtado et al. 2022a, 2022b), and restricted our selection to genes with graded tissue specificity index τ ≥ 0.9, which was calculated for all genes from FlyAtlas2 RNA expression data (Yanai et al. 2004; Leader et al. 2017). Acp genes with τ ≥ 0.9 are genes with highly concentrated expression in the accessory glands, and very low expression elsewhere. A total of 155 Acps from the original list met these criteria (Supplementary Dataset 1). As a control group, a random sample of 155 genes taken from the genome, excluding the 155 Acps, was defined. Randomization was performed using the sample function in base R (R Core Team 2022). As a further control group, we analysed the top 50 genes with highly concentrated expression in the Malpighian tubules (τ ≥ 0.9).

Genomic sequences, PhyloP27way data and CAGE data were retrieved from UCSC Genome Browser using the UCSC Table Browser sequence retrieval tool (Kent et al. 2002; Karolchik 2004). A standardized sequence region of 2000 bases upstream of the annotated gene TSSs was exported for each gene to ensure inclusion of promoter regions. A region of 1000 nucleotides upstream and 300 nucleotides downstream of gene TSSs was collected from PhyloP27way data. Where genes had multiple TSSs, dominant transcripts were identified using available CAGE values proximal to annotated TSSs. To identify dominant TSSs while accounting for CAGE peak position inaccuracy, a nine-nucleotide sliding window score (f) was calculated using R version 4.4.2 for full gene lengths (R Core Team 2022). Annotated TSSs with the highest f score per gene were considered dominant and used as representative transcripts for analysis. Sequences were aligned with clustalW using the R package msa and manually refined using the MEGA11 package (Bodenhofer et al. 2015; Tamura et al. 2021).

To analyse promoter evolution hotspots, gene and promoter sequences were adjusted to span −1000/+300 nucleotides from the TSS of each gene. To assess the occurrence of sequence change hotspots within promoter sequences, we computed the hotspot accumulation score (d) using alignments derived from the trimming process, or alternatively, PhyloP data. These alignments were converted into “events” for each species when compared to D. melanogaster. Here, each nucleotide in the sequence was assigned a value of 0 if it represented a conserved sequence and 1 if it indicated a sequence change (either a base change or an indel event). Events were tallied for all changes, as well as separately for base changes and indels. The cumulative events were calculated for both concatenated gene groups and individual genes across all nucleotides. Subsequently, a sliding event (Se) score was computed from these tallies using a sliding window of five bases along the sequence. To determine the percentage of events surpassing the average control promoter Se score (d), a 350-nucleotide region upstream of the estimated TATA box region was examined. This involved dividing the total number of Se scores exceeding the average sliding event score of the control group (Se^C) by the total number of events within that region (N). To compute the promoter region scores (d^P) from PhyloP data, we substituted the total number of PhyloP (p) scores in the 350-nucleotide regions that were below the average of the control group promoter region (p^C) in place of the total number of Se events where Se was greater than Se^C as previously described. Significance was assessed using non-parametric chi-squared tests compared to the control group d score. Significance values where P ≤ 0.05 after Bonferroni correction were considered statistically significant.

To test for nonrandom accumulation of indels along the Acp extended gene regions between the five analysed Drosophila species (D. melanogaster, D. simulans, D. sechellia, D. yakuba, and D. erecta) we determined significant deviations from a uniform distribution of substitutions using an empirical cumulative distribution function (Civetta et al. 2016). The position of the indel event was defined as the 5′ site of the start of the indel in the alignment (Schaeffer 2002). The function (G) detects monotonic increases in substitutions (n) measured as the difference between the relative occurrence of a nucleotide change and its relative position in the alignment (Civetta et al. 2016). Whether differences between the values of the G function (ΔG) between substitutional events deviate from a random accumulation of changes is tested using Monte Carlo simulations to produce 100,000 samples of n events by sampling sites without replacement along the alignment (Civetta et al. 2016).

We utilized publicly available RNA-seq data from D. melanogaster, D. simulans, and D. yakuba male whole flies (GSE28078) and from D. melanogaster and D. simulans hybrid male accessory glands (PRJNA913156) for our analyses (Graveley et al. 2011; Majane et al. 2024). Adapter sequences were removed from the raw reads using the Trimmomatic tool (v0.39) with default parameters (Bolger et al. 2014). Trimmed reads were aligned to the respective genome assemblies (dm6 for D. melanogaster, Prin_Dsim_3.1 for D. simulans, and droYak2 for D. yakuba) using the RNA-seq aligner RNA STAR (v2.7.10b) (Dobin et al. 2013). Gene counts were quantified using the featureCounts tool from the Subread package (v2.0.3) (Liao et al. 2013). Orthologous genes were identified using the Flybase ftp accession orthologue lookup table (Gramates et al. 2022). The initial analysis steps were performed using usegalaxy.eu (Community 2022). For the interspecies differential gene expression comparison, Transcript per Million (TPM) values were computed to mitigate potential influence of differences in gene lengths of orthologs in the different species. TPMs for orthologs in all three species were extracted using the TPMCalculator package (Vera Alvarez et al. 2019). Differential gene expression analysis was then performed using Limma and EdgeR packages implemented in iDEP (Ritchie et al. 2015; Ge et al. 2018; Chen et al. 2025). The same pipeline was applied to D. melanogaster/D. simulans hybrid expression data (n = 3) after aligning to a combined reference genome and then calling TPM for each species separately, resulting in three D. melanogaster and three D. simulans expression sets; reads with multiple alignments (indistinguishable between both species) and expression in less than 4 samples (avoidance of mono-allelic expression inherited from one species, i.e. the X chromosome genes inherited exclusively from the D. simulans parent) were removed.

Wild type D. melanogaster CantonS, D. simulans and D. sechellia Jallons (S-32) were used and kept on standard cornmeal-agar food (1% industrial-grade agar, 2.1% dried yeast, 8.6% dextrose, 9.7% cornmeal and 0.25% Nipagin, all in (w/v)) in a 12 h light:12 h dark cycle. To obtain hybrid males, D. melanogaster males were crossed to either D. simulans or D. sechellia virgin females.

Total RNA extraction was carried out using Tri-reagent from SIGMA according to the manufacturer's instructions. Reverse transcription was performed using Superscript II (Invitrogen) using an oligo-dT17 V primer as previously described (Haussmann et al. 2011). PCR was performed for 40 cycles with 1 μl of cDNA. Primers for analysis of hybrid gene expression were CG30486_F1 (GTGTCTGCCGGAAATAACGCATAT), CG30486_R1 (TCCGTATTGCTGCAGTAATCGTC), Obp56f_F1 (TCTCTGCTATCTGGCTCCA), Obp56f_R1 (GAAGTCTAATGCTTGGTAATGGACT), CG11598_F1 (GTCTCTATATTGACCTTGCCAAGGGA), CG11598_R1 (GCCAATGAATCCTGAGGTCCGTTTAGCA), CG15117_F1 (TTGCGGACTTTAAGACAGCACAAA), CG15117_R1 (CGGTAAACTTATTTACAGTGGCCGAATA), Rpl32_F1ms (CCGCCACCAGTCGGATCGATATGCTAAG), Rpl32_R1ms+62 (GCTATCCCAATCTCAGAAATGACAATTGAACTCGGCACTCGCACATCATTTTTTAACTAAAAGTCCGGTATATTAACGTTTACAAATGTG). Restriction enzymes (AcuI, BsaWI, MluI, NgoMIV, Sau3AI, or ScaI) were from New England Biolabs and DNA restriction digests were performed according to manufacturer's instructions. Experiments included at least two biological replicates. PCR product bands were quantified using Quantity One 1-D Analysis Software 4.6.8 (BioRad) according to the manufacturer's instructions. Band intensities were normalized to gene length for digested fragments. Hybrid-allele differential gene expression was calculated as the log₂ fold change for D. melanogaster relative to either D. simulans or D. sechellia.

To analyse promoter evolution of Acps we first generated a list of 155 bona fide genes predominantly expressed in male accessory glands from annotated gene expression data (Yanai et al. 2004; Leader et al. 2017; Wigby et al. 2020; Hurtado et al. 2022a, 2022b) (Supplementary Dataset 1). Although some Acps show low-level expression outside the accessory glands (Thayer et al. 2024), we restricted our analysis to genes with a tissue-specificity index τ ≥ 0.9 in male accessory glands, a commonly used threshold for identifying tissue-specific genes (Yanai et al. 2004). We then used PhyloP27way data, a conservation score based on 27 insect species at each nucleotide position, to generate comparative metagene plots of regions 1000 nucleotides upstream and 300 nucleotides downstream of transcription start sites (TSS) of Acp genes and compared them to all other genes in the D. melanogaster genome (Fig. 1a). In addition, we analysed an additional control group comprised of genes near-exclusively expressed in the Malpighian tubules (τ ≥ 0.9), chosen as this tissue shares features with the accessory glands: both are adult secretory tissues that produce and release proteins, and both are primarily composed of two cell types (Dow et al. 2021). Quantification of the metagene plot values and analysis of mutation hotpots between the Acp genes, Malpighian tubule genes, and all non-Acp genes revealed that Acps accumulate mutations more rapidly in both their coding regions and promoters compared to the controls (Fig. 1, a to c). In addition, accumulation of mutations was also observed in 5` untranslated regions (UTRs) of Acps, in contrast to the 5` UTRs of NPC genes and piRNA factors (McQuarrie et al. 2023, 2024). Because Acp genes are generally small and compact genes, regulatory elements could possibly be in 5` UTRs (Supplementary Dataset 1, Fig. 1, a to c).

To understand whether promoter regions evolve rapidly between closely related Drosophila species, we focused on five species in the melanogaster subgroup (D. melanogaster, D. simulans, D. sechellia, D. yakuba, and D. erecta) (Fig. 1d). We quantified accumulation of sequence changes in promoters for insertions and deletions (indels), and base changes in Acps and compared them to an equal number of random sampled genes (Fig. 1, e to h). Our analysis confirmed that, compared to control genes, promoters of Acps evolve rapidly among closely related species by accumulating both indels and base changes.

To determine whether changes in promoter regions of Acp genes correlate with divergent gene expression, we examined differential gene expression in adult males between three pairs of closely related species in the melanogaster subgroup (D. melanogaster with D. simulans, D. melanogaster with D. yakuba, and D. simulans with D. yakuba). In this analysis we found that expression of Acp genes diverges more compared to all non-Acp genes and the random sample genes for the D. melanogaster vs D. simulans comparison (Fig. 2a). As a negative control we analysed highly conserved ribosomal component gene expression which showed significantly less divergence compared to the Acp genes for all comparisons (Fig. 2a). While the frequency of significantly diverging Acps is greatest between closely related species, the distribution of expression changes is more skewed in comparisons involving the more distantly related D. yakuba (Fig. 2b).

To analyse whether Acps with rapidly evolving promoters were also differentially expressed between species, we divided Acp and control genes into three conservation score groups based on the calculated d score (0 to 33, >33 to 66, >66 to 100 d) for each gene and plotted the percentage of genes with significant expression changes (>1 or <−1 log₂ fold change) for each group (Fig. 2, c to e, Supplementary Dataset 1) (McQuarrie et al. 2023). The random sample control group generally displayed the highest differences in expression for the middle conservation score group (>33 to 66 d) (Fig. 2, c to e). However, Acps were enriched for significant expression changes in the faster evolving group (>66 to 100 d) across all species comparisons (Fig. 2, c to e). Of note, this trend was also observed in the other groups (0 to 33 and >33 to 66 d) for the species comparisons to D. yakuba, but not between D. melanogaster and D. simulans (Fig. 2, c to e).

Next, we further molecularly analysed Acp genes with significant expression changes in promoter conservation in more detail. We selected three genes with consistent differences across pairwise species comparisons but with different levels of promoter evolution between D. melanogaster, D. simulans, and D. yakuba, (Obp56f, CG30486, and CG15117) (Fig. 3a, Supplementary Dataset 1). Analysis of sequence change hotspot accumulation in the promoters of these genes revealed substitution hotspots localized to promoters upstream of the TSSs for all three genes (Fig. 3, b to d).

Interspecies comparisons of differential gene expression cannot rule out the effect of differences in the genomic background or resulting changes to the cellular environment that might have arisen during speciation. To test whether changes in the cellular environment of Acp genes can alter expression, we generated hybrids, with a mixed cellular environment, between D. melanogaster and D. simulans and between D. melanogaster and D. sechellia. In these hybrids, accessory glands are known to develop normally in contrast to the gonads (Wong et al. 2010), thus ruling out possibly allometric effects due to tissue atrophy (Sztepanacz and Houle 2021).

To determine expression levels of CG30486, Obp56f, and CG15117 genes with high promoter evolution in hybrids from each allele, we identified single nucleotide polymorphisms (SNPs) that are part of restriction enzyme cut sites. We then designed an RT-PCR amplicon around these SNPs such that PCR fragments of different sizes for each species were obtained after a restriction digest and quantified the levels of PCR products in hybrids (Fig. 4, c to h). In addition, we analysed CG11598 because its promoter contains a hotspot for sequence changes (Supplementary Fig. 1), while interspecies expression differences are moderate between the analysed species, and as a control for low promoter and expression changes, we used Rpl32 (Fig. 4, a and b).

In this quantitative analysis of the two parent alleles in hybrids we expected that the changes in expression would reflect the changes in promoters. For CG30486, the D. simulans and D. sechellia alleles were dominant over the D. melanogaster allele (Fig. 4, c and d). Unexpectedly, our analysis of the fastest-evolving promoter genes, Obp56f and CG11598, revealed that the D. melanogaster allele was dominant in D. melanogaster × D. simulans hybrids, but that the D. sechelia allele was dominant in D. melanogaster × D. sechelia hybrids (Fig. 4, e to h). For CG15117 and the control RpL32, no dominant allele is evident (Fig. 4, a, b, i and j).

To expand this analysis to all Acp genes, we analysed RNA-seq data from D. melanogaster × D. simulans hybrids. In this analysis, Acp genes showed an increased frequency of expression divergence compared to all control groups (Fig. 5a). As seen with the initial species comparisons (Fig. 2b), fold change distribution was not significantly different between hybrid alleles (Fig. 5b). To check whether Acps with rapidly evolving promoters were enriched for significant expression changes between hybrid alleles, we separated genes into three conservation score groups based on d (0 to 33, >33 to 66, >66 to 100 d). Here, the percentage of genes with >1 or <−1 log₂ fold changes were quantified for each group (Fig. 5c). Acps in the faster evolving group (>66 to 100 d) and slowest evolving group (0 to 33 d) were enriched for significant expression changes (Fig. 5c). Next, we analysed hybrid allele expression changes for the five genes tested in Fig. 4 (Rpl32, CG30486, Obp56f, CG11598, and CG15117). The direction of expression divergence was consistent for Obp56f and CG15117, but differed for the other genes likely reflecting the different Drosophila strains used (Canton S was used in this study, while the inbred line RAL-517 was used by Majane et al. (2024)) or technical sensitivity between methods (Table 1, Supplementary Dataset 1) (Majane et al. 2024).

Taken together, our data show that promoter evolution drives changes in expression, because species-specific patterns of expression are either maintained in hybrids or dominated by one allele through cis- or trans-regulatory changes. Hence, our data implicate promoter evolution as a main source for fast changes in gene expression.