Using improved, full-length genome assemblies of D. takahashii, D. kikkawai, D. ananassae, and D. bipectinata (Leung et al. 2023), we annotated all protein-coding genes on the heterochromatic Muller F element and manually identified their D. melanogaster orthologs (see Methods). We also annotated all protein-coding genes in a euchromatic region (∼2 Mb) near the base of Muller D, for use as a set of reference genes in each species (hereafter referred to as Muller D reference genes or Muller D reference region) (Table 1). The Muller D reference genes were selected based on their proximity to the pericentromeric heterochromatin (Leung et al. 2015). Although the majority of genes (ca. 95%) have been shown to remain on the same Muller element across species (Bhutkar et al. 2008), only 29 genes (16% to 30% of all reference genes, depending on species) are found in the reference regions of all four species, likely due to inversions and/or other chromosomal rearrangements.

We find that three genes on the F element in D. melanogaster are absent from the F elements of one or more species: in D. kikkawai, CG31999 has translocated to Muller B; in D. ananassae and D. bipectinata, Kif3C has translocated to Muller C. We were unable to find a RhoGAP102A homolog in the genomes of D. ananassae and D. bipectinata, suggesting that this gene has been lost entirely from the genomes of these two species (Table 2). In addition, we identified several genes that have recently moved onto the Muller F element in at least one of the four expanded F species studied here: these genes represent either translocations or duplication events from other Muller elements. We term these “wanderer genes” and have inferred their ancestral chromosome location via synteny analysis (Table 2). Overall, the vast majority of F element genes (74 of ∼80) are F-linked in all species studied here, including D. melanogaster. This result ensures that our comparisons of F element gene characteristics across species involve the same pool of genes.

To identify which F element features have expanded in size in these species, we compared sizes of various genic and intergenic features both within species (ie F element features versus features from the reference D element region) and between species, focusing specifically on the coding span (ie distance from start to stop codon, including introns) of each gene. In D. melanogaster, F element genes have larger coding spans compared to genes in the reference region on the D element (Leung et al. 2015). This same pattern was shown to be exaggerated in D. ananassae, indicating that the D. ananassae F element genes have expanded in size compared to their D. melanogaster orthologs (Leung et al. 2017). Here, we find that, similar to D. melanogaster, F element genes within each of the four species have larger coding spans and total intron lengths than their respective D element genes. The F element genes of D. takahashii, D. ananassae, and D. bipectinata also have larger coding spans and total intron lengths compared to the F element genes of D. melanogaster. In contrast, the median coding span and intron length of genes on the D. kikkawai F element are both similar to those of D. melanogaster (Fig. 2a and b). Further investigation on a gene-by-gene basis shows that, across the four expanded F elements, 34 intron-containing genes are either the same size compared to their D. melanogaster orthologs or have decreased in size in at least one species (Figure S1a and b). The majority of these (18 genes) are non-expanded only in D. kikkawai, which explains why the D. kikkawai F element coding spans and intron lengths are not significantly larger than those of D. melanogaster (Fig. 2a and b). Eleven other genes are non-expanded in D. kikkawai and at least one other species, four of which (Arf4, ATPsynbeta, RpS3A, and Zip102B) show either no size expansion or size contraction in all four species, compared to their D. melanogaster orthologs (Figure S1c). The five remaining non-expanded genes are distributed across the other three species (Figure S1c).

Comparing D. melanogaster F element genes to those on each of the four expanded F elements, there is no difference in the total size of coding sequence (Fig. 2c) or the number of coding exons (Fig. 2d). Finally, all four expanded F elements have larger intergenic regions compared to the D. melanogaster F element (Fig. 2e). In fact, intergenic expansion contributes disproportionately to total F element size expansion in all four species, explaining between 52% and 69% of the total expansion (Figure S2). These results suggest that F element expansion is accompanied by an expansion of intergenic sequence as well as an increase in coding span, driven by an increase in intron length.

Notably, several unique aspects of F element genes are present in all five species (D. melanogaster plus the four expanded F species). Specifically, the F element genes within all five species consistently have more coding sequence (ie encode larger proteins) (Fig. 2c), more coding exons (Fig. 2d), and more introns (Fig. 2f), compared to the D element reference genes. These measures do not vary significantly among species. Thus, although the ancestral gene content of the F element is biased in terms of exon number and protein size, these properties are not affected by F element size expansion.

Our analysis of F element genes suggests that repetitive DNA is a major contributor to F element expansion. To more fully characterize the F element repeat landscape in these species, we used Earl Grey (Baril et al. 2024) to generate de novo libraries of complex repeats (including TEs) and Tandem Repeats Finder (Benson 1999) to identify simple repeats in each species (see Methods). Roughly a quarter of the D. melanogaster F element in the iso-1 reference genome assembly is composed of TEs (Leung et al. 2017). In the other four species, we find that increased F element size generally correlates with increased TE density: D. kikkawai (1.7-fold expansion, 57% TE density), D. takahashii (2.4-fold expansion, 67% TE density), D. ananassae (14.4-fold expansion, 87% TE density), D. bipectinata (15.2-fold expansion, 85% TE density).

The repetitive portion of the F element is significantly larger in each expanded F species, compared to D. melanogaster (Fisher's Exact Test adjusted P < 0.05, Fig. 3a). This expansion of repetitive DNA is driven almost entirely by an increase in TE abundance, with most major TE classes/orders showing significant expansion in each species (Fig. 3a and Table S1). However, the degree of expansion of each TE class/order on the F element differs among species. For example, long terminal repeat (LTR) and long interspersed nuclear element (LINE) TEs show the largest expansion in D. kikkawai, D. bipectinata and D. ananassae, whereas DNA and helitron TEs show the largest expansion in D. takahashii (Table S1). In D. kikkawai, simple repeats show significant expansion in addition to TEs, but this is not the case in the other three species (Fig. 3a and Table S1). Comparing the fold-change in size in base pairs between each species and D. melanogaster reveals that TEs show the largest expansion in three of the four species (Fig. 3b). In D. kikkawai, simple repeats show the largest expansion, despite the fact that they occupy less of the F element compared to TEs. These findings are consistent with previous studies that have documented TEs and other repeats as playing a major role in chromosome (and genome) size expansion (Kidwell 2002; Leung et al. 2017; Gebert et al. 2025). ​​

Previous work has demonstrated that, across multiple Drosophila species, F element genes consistently show less biased usage of synonymous codons compared to genes on the other autosomes (Vicario et al. 2007; Leung et al. 2015, 2017). To assess how F element expansion affects the usage of synonymous codons, we characterized codon usage bias for F and D element genes using a version of the codon adaptation index (CAI) metric (Sharp and Li 1987) that is normalized to account for biases in nucleotide composition (Puigbò et al. 2008) (see Methods). Within each species, F element genes display a strong reduction in optimal codon usage compared to their respective D element genes (Fig. 4a), which is consistent with previous observations in D. ananassae (Leung et al. 2017). Furthermore, genes from the expanded F elements of three species (D. ananassae, D. bipectinata, and D. takahashii) have significantly reduced usage of preferred codons compared to the F element genes of D. melanogaster, with the ultra-long F elements of D. ananassae and D. bipectinata showing the largest reduction in codon usage bias (Fig. 4a). Surprisingly, the D. kikkawai F element genes show a significant increase in codon usage bias compared to D. melanogaster. This pattern is not solely due to the subset of D. kikkawai genes that have not expanded in size compared to D. melanogaster; in fact, these non-expanded genes have a lower median normalized CAI value compared to expanded genes, although the difference is not statistically significant (Figure S3). Overall, these results suggest that, with the exception of D. kikkawai (see Discussion), F element size expansion is associated with less efficient selection for preferred codons.

Recombination is reduced on the D. melanogaster F element, potentially due to its heterochromatic nature and/or centromere-mediated exclusion of crossing over (ie the “centromere effect” [Beadle 1932]) extending across its relatively short chromosome arm (Sandler and Szauter 1978; McMahan et al. 2013; Hartmann and Sekelsky 2017; Hatkevich et al. 2017; Riddle and Elgin 2018). It is unknown whether recombination occurs within the F elements of the four expanded F species. However, if the centromere effect is the main inhibitor of recombination on the D. melanogaster F element, the expanded F elements of these four species may be large enough to escape the centromere effect, at least at their distal ends (Riddle and Elgin 2018). Several previous studies have reported an association between recombination rate and codon usage bias in Drosophila (Kliman and Hey 1993; Marais et al. 2001; Campos et al. 2012; Charlesworth and Campos 2014). If recombination occurs on the F elements of these other species at a rate that increases with distance from the centromere, then genes located further from the centromere should show increased codon usage bias. Instead, there is no significant correlation between gene distance from the centromere and normalized CAI in any species (Fig. 4b). Rather, almost all F element genes show similarly low levels of codon usage bias regardless of their distance from the centromere (Fig. 4b). These results suggest that the reduced levels of recombination on the F element are more likely to be due to its packaging as a heterochromatic domain rather than any effect of the centromere.

We next compared normalized CAI values for genes from each of the expanded F species to their orthologs in D. melanogaster. We observe significant positive correlations in CAI values of F element orthologs between D. melanogaster and each respective species, with the closest relatives (ie D. melanogaster and D. takahashii; D. ananassae and D. bipectinata) showing stronger correlations (Fig. 4c and d). Thus, despite the differences in codon usage bias between expanded F species, the relative differences in CAI among F element genes within each species are largely conserved.

Highly expressed genes show stronger codon usage bias, likely due to selection for translational efficiency and/or accuracy (Moriyama and Powell 1997; Plotkin and Kudla 2011; Frumkin et al. 2018). Therefore, the cross-species correlation of CAI among F element genes likely arises from conserved expression levels of genes across species (Figure S4 and Table S3). For example, ATPsynbeta and RpS3A are among the 5% most highly expressed F element genes in all four expanded F species and also show strong codon usage bias across all species, including D. melanogaster (Fig. 4b). This indicates that selection may still be responsible for differences in codon usage among F element genes in these species, but the overall efficiency of selection to drive higher levels of codon usage bias is reduced in species with expanded F elements, at least for the three species whose F elements show the largest expansion in size.

In the genome assemblies of D. melanogaster and the four expanded F species, the F element is represented by a single scaffold, allowing us to compare F element gene order across all five species (Fig. 5a). While the F elements of D. melanogaster and D. takahashii appear completely collinear, the F elements of the other three species are highly rearranged, both compared to each other and to D. melanogaster (Fig. 5a and b). The lack of synteny between D. ananassae and D. bipectinata is especially striking, given that these two species share a more recent common ancestor (divergence time of 8.7 Mya) compared to D. melanogaster and D. takahashii (divergence time of 13.2 Mya) (Suvorov et al. 2022). These results suggest that the large expansion of TEs and other repeats on the D. ananassae and D. bipectinata F elements may be causing a higher rate of chromosomal rearrangements on this chromosome.

To further investigate this discrepancy, we generated a whole genome alignment between these four species, which we used to calculate the size, in base pairs, of collinear (ie syntenic) alignment blocks between species pairs. On all other Muller elements, the mean synteny block size from the D. ananassae and D. bipectinata pairwise alignment is larger than that of D. melanogaster and D. takahashii, as expected given their more recent divergence time (Fig. 5c). However, the mean ananassae/bipectinata F element synteny block size is almost an order of magnitude smaller than that of melanogaster/takahashii (Fig. 5c). Synteny coverage, which represents the aligned fraction of each chromosome for a given species pair, is similar between both species pairs for the autosomal Muller elements B, C, and D (Fig. 5d). The ananassae/bipectinata X chromosome (Muller A) is metacentric in both species and contains a large region of pericentromeric heterochromatin (Leung et al. 2023), which likely explains the reduced synteny coverage in this species pair compared to melanogaster/takahashii. Most striking, however, is the F element, whose synteny coverage is dramatically reduced in the ananassae/bipectinata pairwise alignment compared to all other Muller elements across both the ananassae/bipectinata and melanogaster/takahashii pairwise alignments (Fig. 5d). These results suggest that, outside of coding regions, the ultra-long F elements of D. ananassae and D. bipectinata share very little sequence in common.

To further investigate the sequence evolution of the D. ananassae and D. bipectinata F elements, we sought to analyze additional relatives of these two species. D. ananassae and D. bipectinata are members of the ananassae species group, which contains 21 species divided among three species complexes: ananassae, ercepeae, and bipectinata, plus D. varians, which is related to ercepeae group species but excluded from the complex (Matsuda et al. 2009; Signor et al. 2013). We used publicly available long-read genome assemblies (Kim et al. 2024) for nine additional species from within the ananassae group: three additional species each from the ananassae and bipectinata complexes, two species from the ercepeae complex, and D. varians (Fig. 6a). Using gene content to identify F-linked contigs, we estimate that the size of the F element across these nine species ranges from 8 Mb to 22 Mb. These sizes are likely underestimated because the F elements in each assembly are composed of multiple contigs, whereas our size estimates include only contigs containing at least one F element gene. Thus, any F-linked contigs containing only noncoding sequence were not counted towards the total size. Nevertheless, these estimates suggest that all species within the ananassae species group have highly expanded F elements compared to D. melanogaster, which is consistent with previous cytological evidence (Matsuda et al. 2009).

To compare F element sequence conservation across this group in more detail, we created a whole genome alignment from the four species studied here, plus the nine additional ananassae group species and D. melanogaster (see Methods). We selected three focal species: D. ananassae, D. bipectinata, and D. melanogaster, and compared each in turn to the 13 other species in the whole genome alignment. We calculated the percentage of aligned bases in 20 kb windows across the entire F element of each focal species, as well as across their Muller D reference region (Fig. 6b).

For the Muller D element, each focal species shows the highest alignment rate with its closest relative, as expected, and many blocks of aligned sequence are conserved across all 14 species (Fig. 6b). The D. melanogaster F element shows a similar pattern, although there are fewer conserved alignment blocks (Fig. 6b). Strikingly, the D. ananassae and D. bipectinata F elements appear to align only to the F elements of their two closest relatives (D. ananassae: D. anomalata and D. pandora; D. bipectinata: D. parabipectinata and D. malerkotliana) (Fig. 6a and b). This result is consistent with our previous observation that the D. ananassae and D. bipectinata expanded F elements share little sequence in common (Fig. 5c and d). The sequences that do align are highly enriched for coding sequences of genes. For example, ∼56% of D. ananassae F element coding sequence is aligned to the D. bipectinata F element, whereas only ∼1% of F element non-coding sequence aligns (Fig. 6c). There are two possible explanations for these results: (i) the large F elements carried by D. ananassae and D. bipectinata expanded in size independently, or (ii) the large F elements are evolving so rapidly that their non-coding sequence has completely turned over since their divergence from a common ancestor (see Discussion).

To further investigate the evolution of the noncoding portion of the F element, we assessed conservation of repeat content between all ananassae group species plus D. kikkawai, D. takahashii, and D. melanogaster. We used F-specific repeat libraries from each species to mask the assembled F element sequence in each of the other species. Two species whose F elements are populated by the same repeat families will show similar fractions of masked sequence regardless of which of the two repeat libraries are used. These reciprocal repeat masking analyses show that the F elements within each of the ananassae group species complexes (ie ananassae, bipectinata, and ercepeae) have similar repeat content. D. varians appears to have a subset of distinct repeats that are not found in its closest relatives, D. vallismaia and D. ercepeae, supporting its exclusion from the ercepeae species complex (Fig. 7a). Similarly, members of the ananassae species complex share a subset of repeats that are missing from members of the bipectinata species complex, and vice versa.

To more directly visualize the turnover of F-linked repeats across the ananassae group species, we focused on the F element gene zfh2 and independently annotated all D. ananassae and D. bipectinata repeats in the introns of this gene across 11 ananassae group species. The repeat content of orthologous introns appears to be entirely distinct between species that diverged more than 2 to 3 Mya (Figs. 7b and 6a). We repeated this analysis on six additional F element genes (Tdg, Pur-alpha, ND-49, Ephrin, CG1909, and Rad23) and obtained similar results, suggesting that distinct intronic repeat content is not unique to zfh2 but is likely a common feature across all or most F element genes in the ananassae group (Figure S5). Thus, F element expansion (or repeat turnover) has occurred independently in four different lineages within the ananassae group: the ercepeae, bipectinata, and ananassae species complexes, as well as in D. varians.

Genome size expansion has previously been suggested to be associated with population bottlenecks or other reductions in effective population size (eg Lefébure et al. 2017). We used publicly available population resequencing data for D. melanogaster, D. kikkawai, D. ananassae, and D. bipectinata (there are no publicly available population data for D. takahashii) to estimate the effective population size (Ne) of both the F element and the autosomes (see Methods). We find that the F elements of all three expanded F species show much smaller Ne compared to the F element of D. melanogaster (Fig. 8a). The D. kikkawai and D. bipectinata autosomal Ne is also smaller than that of D. melanogaster, whereas the autosomal Ne of D. ananassae is larger than that of D. melanogaster, consistent with previous work (Grath et al. 2009).

Given the positive correlation between recombination and Ne in Drosophila (Gossmann et al. 2011), we next sought to determine whether the F elements of these species differ in terms of signatures of recombination. We calculated the mean physical distance at which linkage disequilibrium (LD) decays to ¾ of its maximum value across the F element and across the autosomes of each species (Fig. 8b). Recombination reduces LD between adjacent alleles; therefore, as the rate of recombination increases, LD will decay to background levels across a shorter physical distance. The LD decay results mirror our estimates of Ne: for the autosomes, D. ananassae has both the largest Ne and the fastest rate of LD decay, while D. kikkawai has the smallest Ne and the slowest rate of decay in LD (Fig. 8a and b). For the F element, D. melanogaster has both the largest Ne and the fastest rate of LD decay, while D. ananassae has the smallest Ne and the slowest rate of decay in LD (Fig. 8a and b). The ratio of autosomal versus F element LD decay for each species recapitulates the degree of F element size expansion across species: D. kikkawai shows a slightly larger ratio compared to D. melanogaster, while the ratios for D. ananassae and D. bipectinata are much larger (Fig. 8c). Population genetic analyses have previously suggested that recombination does occur on the F element in Drosophila, even in D. melanogaster (Wang et al. 2002; Arguello et al. 2010). Our results support this previous work and also raise the possibility that a sharp reduction in recombination, specifically on the F element, may explain its large expansion in size in D. ananassae and D. bipectinata.

Strain information and approaches used to construct the genome assemblies for D. takahashii, D. kikkawai, D. ananassae, and D. bipectinata have been previously described (Leung et al. 2023). The annotations for all genes on the F element scaffold (and Muller D reference region) of each species were hand-curated by undergraduate student participants in the Genomics Education Partnership (GEP) (Shaffer et al. 2010) with guidance from their faculty mentors at their home institutions. GEP student annotators identified D. melanogaster orthologs for each gene via protein BLAST searches conducted on the FlyBase website (Öztürk-Çolak et al. 2024).

Annotators then used a mirror of the UCSC Genome Browser to examine different lines of experimental (eg RNA-Seq) and computational evidence (eg sequence similarity to D. melanogaster protein sequences, gene predictions, splice junction predictions) to construct gene models across the four species. The paired-end RNA-Seq data from adult females, adult males, and mixed embryos used in the gene annotation process were produced by the modENCODE project (Graveley et al. 2011); individual NCBI BioProject accessions are listed in Table S5. The evidence tracks derived from these RNA-Seq datasets included RNA-Seq read coverage, splice junction predictions from RegTools (Cotto et al. 2023), and assembled transcripts from StringTie2 (Trapnell et al. 2010; Kovaka et al. 2019) and Oases (Schulz et al. 2012).

Each gene annotation project was completed by at least two GEP student annotators working independently. Reconcilers are undergraduate students, or recent graduates, with extensive training in gene annotation. Working with supervision from GEP faculty and staff, they reconciled these gene annotations using Apollo 2 (Dunn et al. 2019) to create the gene models analyzed in this study. In some cases, such as when genes were missing two independent student annotations, reconcilers performed initial gene annotation in addition to reconciliation. The Annotation Instruction Sheet (available on the GEP web site at https://thegep.org/projects/felement/) describes the protocols for handling cases where the annotations were ambiguous because of insufficient evidence. Additional curricula (eg walkthroughs, presentations) that describe the GEP annotation protocol are available under the “Curriculum” section of the GEP web site (https://thegep.org/curriculum/).

For this study, we used the D. kikkawai, D. takahashii, D. ananassae, and D. bipectinata genome assemblies described in Leung et al (2023), each of which has Muller F assembled as a single scaffold. We then performed additional analyses to confirm that the Muller F scaffolds are oriented correctly within each assembly. The Muller F element in the D. melanogaster reference genome assembly is oriented such that the centromere is at the beginning of the scaffold and the telomere is at the end, including a head-to-tail array of HTT telomeric retrotransposons. The D. takahashii Muller F is almost perfectly collinear with the D. melanogaster F element, and we therefore oriented this scaffold based on synteny. One end of the D. kikkawai F element contains a ∼95 kb array of tandem repeats composed of two large DNA satellites (640 bp and 1.9 kb, respectively), which potentially represents centromeric or pericentromeric sequence. We were unable to identify telomeric retrotransposons at the other end of the chromosome scaffold; however, we noted the presence of the genes Mitf, Arf4, CG33521, PIP4K, pho, and Kif3C near this end, all of which are also near the telomere in the D. melanogaster Muller F scaffold. We therefore oriented the scaffold such that the satellite array is at the beginning and the genes with telomeric orthologs in D. melanogaster are at the end.

The F elements of D. ananassae and D. bipectinata are metacentric. The two chromosome arms were previously oriented based on Hi-C contact maps (Leung et al. 2023). We used a combination of synteny, putative centromeric tandem repeats, and telomeric retrotransposons to determine whether the arms were oriented correctly. At the beginning of the D. ananassae Muller F scaffold, we identified orthologs of Arf4, Mitf, PIP4K and CG33521. We also found the ortholog of D. melanogaster yellow-h, which is near the telomere in D. kikkawai, and the orthologs of D. melanogaster Dyrk3 and Cadps, which are near the telomere of the D. melanogaster F element. Interestingly, the other end of the D. ananassae F element scaffold contains the genes Zyx, ATPsynbeta, PlexA, toy, CaMKII, and apolpp, which are located in the subtelomeric regions of either the D. melanogaster or D. kikkawai F elements, or both. Together, these results suggest that both arms of the D. ananassae F element were indeed oriented correctly during Hi-C scaffolding and raise the possibility that there is selection to maintain the telomeric location of a subset of genes on the F element. Most Drosophila species have telomere-specialized Jockey retrotransposons at their chromosome ends (Villasante et al. 2007; Saint-Leandre et al. 2019; Son et al. 2025). We identified a head-to-tail array of Jockey clade retrotransposons at the beginning of the D. ananassae F scaffold, which likely represent telomeric retrotransposons, as well as a ∼100 kb array of simple tandem repeats in the middle of the scaffold, which may mark the location of the centromere.

We then compared synteny between D. ananassae and D. bipectinata F element scaffolds, which suggested that the two chromosome arms of D. bipectinata were previously oriented incorrectly. In the middle of the D. bipectinata F element scaffold is a large array of tandem repeats that may mark the location of the centromere. There are gaps on either side of the array where the chromosome arms were scaffolded together. Orthologs of the telomeric D. melanogaster genes Cadps, Dyrk3, and PIP4K were near this putative centromere in the previous version of the assembly (Leung et al. 2023); however, in the new version of the assembly used here, we have inverted this arm, placing these genes at the telomere. This arrangement places Rad23 and pho near one telomeric end, and bt, Cadps, Dyrk3, and PIP4K near the opposite end, adjacent to the telomeric TE array.

Reciprocal repeat masking was performed using RepeatMasker along with the full Earl Grey repeat library from each species shown in Fig. 7 to mask the F element sequences of all other species shown in Fig. 7. The percentage of masked sequence was calculated by dividing the number of masked base pairs by the total number of F element base pairs.

For the various calculations regarding coding span, coding exons, introns, and intergenic regions, we identified in each species the transcript ortholog corresponding to the principal splice isoform as defined in the APPRIS database (Rodriguez et al. 2022) using the BDGP6 assembly (Hoskins et al. 2015) and Ensembl release 110 annotation (Harrison et al. 2024). For any gene with more than one splice isoform marked principal, the longest splice isoform was selected as the representative transcript for that gene. Two-way non-parametric analysis of variance tests were performed using the ARTool package (Wobbrock et al. 2011) with default options, with chromosome and species as independent variables. Post hoc pairwise comparisons were performed using the ART-C (Aligned Rank Transform for Contrasts) procedure (Elkin et al. 2021) implemented in the art.con() function, with P-values adjusted for multiple comparisons using the Holm method. For Fig. 3, noncoding sequences were identified as regions that do not overlap annotated coding sequences or repeats identified by RepeatMasker and Tandem Repeats Finder, respectively. Note that the “intergenic” category includes the untranslated regions of F and D element genes because these regions were not annotated during the manual curation process. Genes were considered to be non-expanded if their coding span was less than 1.1-fold the length of their ortholog in D. melanogaster.

We created de novo repeat libraries using Earl Grey (Baril et al. 2024) and used RepeatMasker (www.repeatmasker.org) to identify repeat locations within each genome assembly. We used Tandem Repeats Finder (Benson 1999) to identify simple repeats in each assembly. For each expanded F species, the F-specific repeat library was constructed by extracting the subset of repeat consensus sequences from the species-specific Earl Grey repeat library that have one or more matches on the F element. To annotate F element genes in additional ananassae group species, we used LiftOn (Chao et al. 2025) with options “-copies -sc 0.95” and our hand-curated D. ananassae F element gene annotations as a reference.

To calculate the CAI for each species, we used the top 10% of genes in each species with the lowest effective number of codons (ENC) as the reference gene set, similar to Vicario et al. (2007). To identify these genes, we used the NCBI RefSeq annotations for each species and the longest isoform per gene. We used the codon-bias package (Diament 2025) to calculate ENC and to export the reference gene set for each species. We used the options within the codon-bias package that allow for calculating ENC in a way that corrects for both background nucleotide composition (Novembre 2002) and sequence length (Sun et al. 2013). Biased nucleotide and amino acid composition can distort the CAI measure. We therefore used the approach described in Puigbò et al. (2008), implemented in the codonuse package (https://github.com/s-andrews/codonuse), to calculate the expected CAI (eCAI) for each gene, which is the upper limit of CAI when codon usage bias results only from mutational bias (Puigbò et al. 2008). The codonuse package calculates the normalized CAI for each gene by dividing its true CAI by its eCAI. Thus, selection for preferred codons is likely to be contributing to codon usage bias in genes with normalized CAI values larger than one, whereas the codon usage bias of genes with normalized CAI values less than one can be explained mainly by mutational bias (Puigbò et al. 2008).

To infer the ancestral chromosome location of each gene in Table 2, we used a combination of prior literature (Carvalho and Clark 2013), synteny, and parsimony. In cases where the ancestral chromosome was ambiguous based on the five species in Table 2, we identified the chromosome locations of homologs in D. miranda and D. virilis.

We used NGenomeSyn (He et al. 2023 ) to create the F element synteny maps shown in Fig. 5a and GRIMM (Tesler 2002) to infer the minimum number of rearrangement events between the F elements of species pairs in Fig. 5b.

We used progressive cactus (Armstrong et al. 2020) to create whole genome alignments and the HAL utilities halSynteny (Krasheninnikova et al. 2020) to identify synteny blocks and calculate synteny coverage and halAlignmentDepth to calculate the fraction of aligned bases between species (Hickey et al. 2013).

We obtained Illumina population resequencing data for D. melanogaster (Sprengelmeyer et al. 2020), D. kikkawai (Li et al. 2022), D. ananassae (Choi et al. 2015), and D. bipectinata (Li et al. 2022). We used bowtie2 (Langmead and Salzberg 2012) to align Illumina reads to their corresponding reference genome, sorted read alignments and marked duplicates with sambamba (Tarasov et al. 2015), and used freebayes (Garrison and Marth 2012) to identify nucleotide variants. We filtered freebayes variants to include only biallelic sites using vcffilter from vcflib (Garrison et al. 2022).

We used SMC++ (Terhorst et al. 2017) and the filtered freebayes variants to infer population size histories for each species. We then calculated the harmonic mean of effective population size estimates from generations 3,600 to 36,000 (corresponding to 300 - 3,000 years ago, assuming 12 generations per year), as in Bergeron et al. (2023).

To examine the decay of linkage disequilibrium, we used vcftools (Danecek et al. 2011) to calculate LD as the squared correlation coefficient between genotypes for all pairs of nucleotide variants within 1 Mb of each other on the F element and within 500 kb of each other on the other autosomes. We then calculated the mean LD for 1,000 bp bins (F element) and 100 bp bins (autosomes) and calculated the bin size at which LD decreases to ¾ of its maximum value. We note that, although it is more common to calculate LD decay as the distance at which LD decays to half of its maximum value (eg Kim et al. 2007), we use ¾ here because F element LD plateaus in some species before reaching that point.