Pervasive Positive Selection on X-linked Ampliconic Genes in Primates

Abstract

Mammalian sex chromosomes harbor ampliconic gene families, which are multicopy genes with ≥97% sequence identity, predominantly expressed in testis tissue and essential for male fertility. The amplification of testis-specific genes is conserved across mammals, yet the specific gene families that expand show striking lineage-specific variation. Previous studies suggest a dynamic turnover with adaptive evolution for several of these families, but their analysis has been limited by the quality of reference genomes in repetitive regions. To characterize the molecular evolutionary processes of ampliconic gene families on both sex chromosomes, we analyzed telomere-to-telomere genome assemblies from eight primate species spanning 25 million years of evolution. We identified 53 X-linked and 19 Y-linked ampliconic gene families with dynamic copy number variation. Gene conversion through palindromic pairing and tandem arrays maintained high sequence similarity despite accumulating mutations. X-linked families maintained conserved chromosomal positions despite copy number changes, whereas Y-linked families showed frequent positional turnover. Strikingly, multiple X-linked families (GAGE, SSX, CSAG, and VCX) showed pervasive positive selection across the primate phylogeny and multiple (MAGEB, CT45, and HSFX) showed lineage-specific positive selection. Y-linked families predominantly evolve under purifying selection. Examining intraspecific copy number variation of the X-linked ampliconic families in chimpanzees, humans, and gorillas, we found variation among individuals but clear differences between species, with the largest families varying the most. These patterns could suggest that sperm competition, meiotic drive, or dosage-dependent selection drives the rapid, lineage-specific evolution of testis-expressed ampliconic genes in primates.

Significance.

Primate sex chromosomes contain multicopy gene families with almost identical gene copies that are essential for male fertility, yet why these families vary dramatically in copy number within and across species remains unclear. Using complete genome sequences from eight primate species, we document extensive copy number variation of these families on both sex chromosomes and reveal that several X chromosome gene families evolve under strong positive selection across all primates, whereas Y chromosome families predominantly show purifying selection. This comprehensive characterization of their molecular evolution enables investigation into the evolutionary forces, such as sperm competition, genetic conflicts, or dosage compensation, that drive the rapid, species-specific evolution of these male fertility genes and their role in fertility variation.

Introduction

Mammalian X and Y chromosomes evolved from a pair of autosomes approximately 170 million years ago (Cortez et al. 2014). Following the emergence of the sex-determining locus on the proto-Y chromosome, a series of inversions suppressed XY recombination across most of the chromosome length, excluding the pseudoautosomal regions (PARs). This resulted in great differentiation between the two chromosomes. In addition to the PAR and ancestral region (X-degenerate), the Y chromosome contains long ampliconic regions with extensive intrachromosomal homology. Ampliconic regions can be organized as tandem arrays as well as palindromes, long inverted repeats that undergo gene conversion, which counteracts the accumulation of deleterious mutations (Skov et al. 2017; Hallast et al. 2023; Rozen et al. 2003). Similarly, the X chromosome contains PAR(s), ancestral regions, and several palindromes (Miga et al. 2020; Makova et al. 2024). Within these regions, on both the X and the Y, lie ampliconic gene families with highly similar gene copies (≥97% sequence identity), and some families have expanded to high copy numbers. Most of these ampliconic genes are protein coding and expressed exclusively in testis tissue, suggesting specialized functions in male gametogenesis (Mueller et al. 2013).

This pattern of testis-specific gene amplification is widespread across species with sex chromosomes, including humans, mice, and cattle (Soh et al. 2014; Hughes et al. 2020; Zhou et al. 2023). Studies in model organisms have revealed distinct evolutionary mechanisms underlying ampliconic expansion. In Drosophila miranda, which acquired neo-sex chromosomes through chromosomal fusion, new ampliconic genes arose through dramatic tandem amplification, with two apparent evolutionary drivers: testis-specific, dosage-sensitive genes amplified on the Y chromosome (potentially to increase male fertility) and meiosis-related genes co-amplified on both the X and Y chromosomes (possibly reflecting genomic conflict over sex chromosome transmission) (Bachtrog et al. 2019). Similarly, the X- and Y-linked ampliconic genes Slx and Sly in mice exemplify co-amplification between sex chromosomes (Moretti et al. 2020; Arora and Dumont 2022; Kopania et al. 2022; Arlt et al. 2025). Functional studies have demonstrated the importance of ampliconic genes for male fertility. Knockout of several ampliconic gene families results in infertility in mice, including RHOXF (Borgmann et al. 2016), SPACA5 (Park et al. 2026), TCP11 (Castaneda et al. 2020; TCP11 in mice is on chr7 not X/Y), and DMRTC1 (Kawamata and Nishimori 2006). In humans, deletions of X- and Y-linked ampliconic regions are associated with male infertility and spermatogenic failure, including SSX (Liu et al. 2023); H2BW (Bai et al. 2025); genes in AZFc regions, such as DAZ, CDY, and BPY (Xu and Pang 2022); and genes in the AFZa region, including RMBY and TSPY (Krausz and Casamonti 2017).

Despite these conserved patterns of testis-specific amplification and functional importance for male fertility across species, the specific gene families that undergo amplification show striking lineage-specific characteristics. For example, human and mouse X-ampliconic genes show little overlap; only 31% of human X-ampliconic genes have mouse orthologs, whereas 95% of single-copy genes do. However, ampliconic genes in both mice and humans show specialization in spermatogenesis (Mueller et al. 2013). This suggests rapid lineage-specific turnover of these gene families despite their shared functions.

Most studies of ampliconic gene evolution have focused on the Y chromosome, due to their clinical role in male infertility. Recent studies of the Y chromosome have revealed substantial copy number variation (CNV) within and between primate species, further supporting rapid evolutionary turnover in ampliconic gene families (Tomaszkiewicz et al. 2016; Skov et al. 2017; Lucotte et al. 2018; Ye et al. 2018; Vegesna et al. 2020; Makova et al. 2024). In contrast, X chromosome ampliconic genes have received comparatively less attention, with studies primarily examining CNV or positive selection in individual gene families. These studies provided important insights into rapid evolution: X-linked ampliconic families, including GAGE (and its subfamilies PAGE and XAGE), CSAG, MAGE, and SSX presented signatures of positive selection in human–chimp comparisons (Stevenson et al. 2007; Liu et al. 2008), and SPANX has undergone rapid evolution and amplification in great apes (Kouprina et al. 2004). However, the reference genome assemblies available at the time contained gaps and collapsed repetitive regions in ampliconic sequences, limiting the resolution of these regions and preventing comprehensive analysis of sequence evolution and copy number dynamics for both sex chromosomes across multiple primate lineages.

Recent telomere-to-telomere (T2T) assemblies of sex chromosomes in humans (Miga et al. 2020; Rhie et al. 2023), great apes (Makova et al. 2024), and the crab-eating macaque (Zhang et al. 2025) have now resolved these previously intractable regions, enabling the complete identification and annotation of ampliconic gene families across primates for the first time. While initial analyses have already described CNV for Y chromosome families (Makova et al. 2024), the opportunity now exists for comprehensive analysis of molecular evolution across multiple species for ampliconic genes on both sex chromosomes. Here, we leverage these complete assemblies to examine the molecular evolution of X and Y ampliconic gene families across eight primate species, analyzing copy number dynamics of ampliconic clusters, chromosomal gene movement, and the evidence of positive selection.

Results

Characterization of the Ampliconic Gene Family Architecture Across Primates

Using T2T assemblies, we identified 53 ampliconic gene families on the X chromosome and 19 on the Y chromosome across all species examined (Fig. 1a). These families were previously identified by Makova et al. (2024); however, how they group into broader multicopy families, and how many distinct ampliconic clusters exist within each family across species, has not previously been resolved. All included primate species share the same evolutionary strata (Zhou et al. 2023), yet exhibited clade- and lineage-specific expansions in gene family copy number (Fig. 1a and b). Among the X-linked families, 20 of the 53 were ampliconic in all eight species (Fig. 1a), whereas only 2 of 19 Y-linked families were ampliconic across species (Fig. 1b). The proportion of clade-specific differences in status was greater on the Y chromosome (13) than on the X chromosome (7) (P = 2.48 × 10⁻⁵, binomial test), reflecting the broad conservation of X-linked gene families in a single-copy, multicopy, or ampliconic state. Among the Y-linked families, five were present in more than half of the species analyzed (TSPY, CDY, DAZ, HSFY, and RBMY). We additionally detected species-specific expansions: Six X-linked and six Y-linked families were ampliconic in only a single species. Some of these families were entirely restricted to that one species: On the X chromosome, collagen alpha-1(IV) chain-like was found only in human; on the Y chromosome, KRT18Y was specific to bonobo, MTRNR2-like 17 to chimpanzee, centriole and centriolar satellite protein OFD1-like to gorilla, and TAF11L2 to siamang (Fig. 1a and b). To our knowledge, the species-specific amplification of MTRNR2-like 17 in chimpanzee and OFD1-like in gorilla has not previously been reported.

Fig. 1.

Evolution of ampliconic gene families on the X and Y chromosomes. a and b) Comparison of gene family across primate species (a, X chromosome; b, Y chromosome). Gene families vary in copy number across species, being ampliconic (≥97% identity) in some species and multicopy (50% to 97% identity) or single copy in others. Families are ordered by human chromosomal coordinates and grouped by evolutionary strata on the X chromosome. c) Proportion of gene families in each copy number category across species (left, X; right, Y). d and e) Ampliconic cluster structure and evolution (d, Y; e, X). Multicopy gene families are divided into ampliconic clusters: Gene copies sharing ≥97% identity with each other but <97% identity with other family members form separate clusters. Species-specific cluster names (eg PonAbe_PonPyg_cluster1) indicate clusters restricted to those species [species abbreviations: PanPan {bonobo}, PanTro {chimpanzee}, HomSap {human}, GorGor {gorilla}, PonAbe {Sumatran orangutan}, PonPyg {Bornean orangutan}, SymSyn {siamang}, MacFas {macaque}]. The circle size and numbers indicate the number of copies per cluster; absence indicates that there are no sufficiently similar copies. Expression annotations: filled blocks indicate expression in testis somatic cells (Sertoli/Leydig); gametogenesis timing shows expression relative to MSCI (pre-MSCI, post-MSCI, both, or unknown). Selection metrics: dN/dS values from PAML phylogenies (values < 1, indicating purifying selection; values ≥ 1, indicates possible positive selection); filled blocks in the site-test column indicate positively selected sites detected in an ampliconic cluster.

Many ampliconic gene families belong to broader multicopy gene families, defined as having copies with 50% to 97% sequence identity (compared to ≥97% for ampliconic genes). Here, we further resolve these multicopy families into discrete ampliconic clusters, defined by the most similar copies (≥97% identity) within them. For example, the MAGE family is a large multicopy family containing multiple distinct ampliconic clusters (Fig. 1e). This clustering pattern is consistent with the idea that some multicopy families represent divergent remnants of ancestral ampliconic expansions. In some cases, different species lineages have independent ampliconic clusters within the same multicopy family. For example, two orangutan species share a VCX ampliconic cluster (PonAbe-PonPyg-Cluster1; Fig. 1e) distinct from the VCX cluster found in other primates, indicating that orangutan VCX copies have diverged (below 97% identity) from other primate VCX copies to form a separate ampliconic cluster while still belonging to the broader VCX multicopy family.

Gene families with higher median copy numbers tend to exhibit greater variance in copy number across species on both the X and Y chromosomes (Fig. S1), as previously observed within humans and great apes (Lucotte et al. 2018; Vegesna et al. 2020). This pattern is consistent with the increased probability of copy number changes in larger gene families (Ghenu et al. 2016). Notably, cancer-testis families (13 on the X chromosome and 2 on the Y chromosome) were among the most variable families in copy number (Fig. S1). The X-linked cancer-testis families have been previously described as fast evolving (Stevenson et al. 2007). Compared with X chromosome families, Y chromosome families showed particularly high copy numbers and wider variation: TSPY reached 44 copies in human, CDY reached 20 copies in B. orangutan, and RBMY reached 26 copies in bonobo, an order or magnitude greater than most X chromosome families.

Some gene families showed consistent amplification across all species, while others displayed lineage-specific expansion patterns. Of the larger gene families, GAGE and MAGEA2 were consistently amplified across species, although bonobo possessed an additional GAGE ampliconic cluster, which was absent in all other species. SSX showed consistent amplification except for additional expansion in orangutans. In contrast, several gene families showed pronounced amplification in only one or a few species: SPANX, GAGE, VCX, and RMBY exhibited extreme amplification in bonobo, while CT45 and CT47 showed higher copy number in gorilla and human than in the other species. TSPY displayed the widest amplitude of variation, ranging from pronounced amplification in humans to intermediate levels in bonobo, chimpanzee, and orangutan and substantially lower copy numbers in gorilla, siamang, and macaque (Fig. 1d and e).

Across primates, we identified ampliconic gametolog pairs via reciprocal BLAST searches (≥30% sequence identity) and filtering for ampliconic status on both chromosomes. Ampliconic genes were significantly underrepresented among gametolog pairs on both the X and Y chromosomes across most primate species (significant for all species except siamang on the X and gorilla and bonobo on the Y; two-tailed Fisher's exact test; Table S4), indicating that gametolog pairs are less likely to acquire ampliconic status than nongametolog genes. This revealed two commonly co-amplified gene families across species: HSFX/Y and VCX/Y, although both HSFY and VCY were independently lost in some species (Fig. 1d and e). Several additional gametolog gene pairs showed amplification on one of the sex chromosomes: TSPY/X (Y ampliconic) and RBMY/X (Y ampliconic, X multicopy), TBL1X/Y (X ampliconic), and TMSB/TMSB4Y (X ampliconic). CSF2RA, which resides in the PAR, was present on both chromosomes but expanded only in siamang on the X.

Expression Timing of Ampliconic Genes

Most ampliconic gene families are expressed in the testis (Table S5; Fig. 1d and e). Expression pattern during spermatogenesis may reveal functional constraints and selective pressures: Genes expressed before meiotic sex chromosome inactivation (MSCI) must maintain function during early germ cell development, whereas genes expressed after MSCI escape silencing and may be subject to different selective forces, including potential meiotic drive. The expression of these genes in somatic testis cells (Sertoli and Leydig cells) versus germ cells may indicate distinct functional roles. Using single-cell RNA-seq data from human, chimpanzee, bonobo, and (rhesus) macaque (Murat et al. 2023; see Materials and Methods), we quantified expression across cell types and gametogenic stages. For germ cells, we categorized expression timing relative to MSCI: pre-MSCI expression (undifferentiated and different spermatogonia and leptotene through pachytene spermatocytes) or post-MSCI expression (round and elongated spermatids).

Some X chromosome gene families are expressed during both pre-and post-MSCI phases (Fig. 1e), whereas Y chromosome families are expressed exclusively in either pre-MSCI or post-MSCI phases (Fig. 1d). Families with post-MSCI expression show minimal expression in testis somatic cells. Among the pre-MSCI expressed genes, for both the X and Y linked, genes are expressed at the highest levels in differentiated spermatogonia and leptotene spermatocytes, after which they are silenced. This same trend was observed for genes expressed pre- and post-MSCI, where most genes presented an expression peak in differentiated spermatogonia and leptotene spermatocytes and were reactivated in round spermatids after meiotic silencing. While the underlying expression data are available through Murat et al. (2023), the specific pattern we describe here had not previously been characterized across ampliconic sex chromosome families. This expression profile is broadly consistent with a coarser-resolution analysis in crab-eating macaque, which found that most X-linked multicopy and ampliconic genes are expressed premeiotically in spermatogonia, with a smaller subset expressed postmeiotically in spermatids. This pattern is reversed in mouse, where a greater proportion of X-linked multicopy genes are instead expressed postmeiotically (Mueller et al. 2008; Lau et al. 2020). Expression data are unavailable for a subset of families (ten for the Y linked and three for the X linked), because these families are not present in humans.

Gametolog gene pairs show divergent expression patterns. HSFY was expressed only post-MSCI in human and macaque but was not expressed in bonobo or chimpanzee (in which we also did not detect any gene copies), whereas its X-linked gametolog, HSFX, was expressed post-MSCI across all species. VCY is expressed in human both pre- and post-MSCI, while VCX is expressed throughout the pre- and post-MSCI phases in all species and at a higher level than VCY. This lower expression of VCY than VCX had been previously described in humans (Vegesna et al. 2019), as well as the pre-MSCI expression of VCX in preleptotene spermatocytes in humans and rhesus macaques (Shami et al. 2020). RBMY (Y-ampliconic) and RBMX (X multicopy) were both expressed pre-MSCI across species, with RBMY showing particularly high expression in bonobo and human, possibly corresponding to extreme amplification in bonobo (Fig. S2).

Synteny of Ampliconic Gene Families

The complete T2T genome assemblies now allow us to comprehensively examine how chromosomal organization and gene conversion shape ampliconic gene family evolution, an analysis not previously possible across as many families and species because of incomplete sex chromosome assemblies. We examined the genomic distribution of gene families by grouping genes into proximity blocks, as new copies typically arise through tandem duplication or within nearby palindromes (Figs. S5 and S6). When tracing the synteny of these blocks across species, we found that X-linked gene families have conserved chromosomal positions (Fig. 2). This agrees with the well-documented conservation of X chromosome synteny across placental mammals, thought to reflect selective constraints on the three-dimensional genome architecture needed for X chromosome inactivation, which would be disrupted by rearrangements (Brashear et al. 2021). Our finding that ampliconic gene family positions are similarly conserved across, despite CNV, suggests that this architectural constraint applies not only to single copy but also to multicopy X-linked genes. Only four X chromosome families showed distal gain or loss of gene copies while retaining their primary proximity blocks: TBL1X, SPANX, TCEAL8, and RPL36A-HNRNPH2 (Fig. S3). SPANX was a notable exception, having lost/not gained an entire block in macaque. Many gene families had at least some copies within palindromic regions (36/53 X-linked families). If palindromes play a functional role in maintaining these gene families, for instance, by facilitating gene conversion between copies, as described below, we would expect ampliconic genes to be preferentially located near palindromic sequences rather than distributed across the chromosome. Consistent with this, the distribution of X-linked gene families closely traced the locations of these palindromes, with genes significantly closer to palindromes than expected by chance in all seven species with palindrome annotations (permutation test, all P < 0.001; Fig. S4). We further found that the degree of clustering (observed/expected distance) was similar across species (clustering ratio = 0.36 ± 0.06, range: 0.30 to 0.47), with genes positioned approximately 2 to 3× closer to palindromes than expected under random distribution.

Fig. 2.

Gene family synteny across primate species. a) X chromosome. b) Y chromosome. Gene families on the X chromosome maintain conserved positions across species, whereas Y chromosome gene families show greater positional variation. Each gene family is represented by its average chromosomal position; families with multiple separated groups of gene copies are shown as multiple blocks. Gene family labels present in the bonobo are positioned on the left and on the right for families absent in the bonobo, with automatic grouping of nearby labels to prevent overlap.

In contrast, Y-linked gene families displayed marked positional turnover, including recurrent gain and loss events across lineages (Figs. 1b and 2; Fig. S6). This increased variability is consistent with the overall sequence class composition of the Y chromosome, which shows much greater variation across species than the X chromosome (Makova et al. 2024; Figs. S5 and S6). Similar to the patterns observed on the X chromosome, many Y-linked gene families were associated with palindromic regions (10/17 gene families). Y-linked gene copies also showed significant clustering near palindromes in most species (6/7 species, P < 0.05, not for the S. orangutan; Fig. S4), although with greater variability across the phylogeny (clustering ratio = 0.46 ± 0.23, range: 0.21 to 0.89) than the X chromosome. Some great apes showed the strongest clustering (gorilla, human, orangutan: ratios of 0.21 to 0.34, approximately 3 to 5× closer than expected), while the S. orangutan showed less concentration of gene families around palindromes (ratio = 0.89). This is likely because palindromes are spread across the whole chromosome in general in the S. orangutan (Fig. S6).

Gene Conversion Homogenizes Ampliconic Genes Within a Family

We subsequently tested whether gene conversion homogenizes sequences among ampliconic gene copies. Gene conversion is a well-established mechanism driving the concerted evolution in duplicated genomic regions generally and of sequence homogenization within tandem repeat arrays specifically (Gangloff et al. 1996; Vollger et al. 2023). The ampliconic families occur in two main strucutral contexts: within palindromes or in tandem repeats. We hypothesized that gene conversion homogenization would be strongest for copies located on opposite arms of palindromes, where the inverted repeat structure facilitates pairing and arm-to-arm gene conversion; a mechanism well documented on the Y chromosome and, by ways of GC-biased gene conversion (gBGC), in X chromosome palindromes across primates and mice (Fig. 3a) (Rozen et al. 2003; Skov et al. 2017; Bonito et al. 2021; Hallast et al. 2023; Makova et al. 2024; Pal et al. 2026). Both the X and Y chromosomes harbor multiple palindromes, and as described above, many gene families occur within or near these palindromes (Tables S2 and S3). Building on this prior evidence, we tested whether this pattern extended to the ampliconic gene families identified here, across both chromosomes and multiple primate species.

Fig. 3.

Gene conversion in ampliconic gene families. a) Schematic of palindrome structure promoting arm-to-arm pairing. b) Schematic of gene conversion homogenization in tandem arrays. c) Pairwise sequence differences between gene copies grouped by palindrome classification (top: X chromosome, 2.033 comparisons in 49 families; bottom: Y chromosome, 2.643 comparisons in 17 families, across seven species). Genes on opposite palindrome arms show the highest sequence similarity. The Y chromosome genes show greater overall similarity than the X chromosome genes regardless of palindrome status. d to f) Gene conversion signals in the X-linked GAGE gene family. d) GC3 content increases with copy number (135 genes across eight species). e) Pairwise sequence differences decrease with copy number (575 pairwise comparisons across eight species). f) Physically closer gene pairs show greater sequence similarity. Pairwise differences can appear slightly negative due to log(x + 0.1) transformation used for visualization.

We compared full gene sequences (including noncoding regions) for all pairwise combinations of gene copies within each ampliconic cluster per species. We then classified each gene pair based on whether genes co-occurred in palindromes and, if so, whether they were on the same or opposite palindrome arms. As expected, genes occurring on opposing arms of a palindrome show the highest sequence similarity (0.06% median difference on the X chromosome (CI: 0.027% to 0.107%) and 0.01% on the Y chromosome (CI: 0% to 0.059%) (Fig. 3c), consistent with recent findings of palindrome-associated homogenization among Y chromosome ampliconic gene families (Pal et al. 2026). Overall sequence differences were much lower on the Y chromosome than on the X chromosome (with the highest median difference of 0.56% on the Y chromosome vs. 7.70% as the highest on the X chromosome). Interestingly, gene copies that both did not occur in palindromes also showed low pairwise differences [0.15% for the Y {CI: 0.15% to 0.16%} and 0.49% for the X {CI: 0.43% to 0.61%}]. These copies were in tandem repeats, suggesting that gene conversion also operates within tandem arrays to some extent, consistent with the general role of gene conversion in homogenizing tandemly repeated sequence described above.

We tested for the possibility of homogenization across gene copies within tandem repeats, using the GAGE family, which has high copy numbers and occurs primarily in tandem repeats rather than palindromes (Fig. S5; Tables S2 and S6). Three independent lines of evidence supported gene conversion in tandem arrays. First, median GC3 content of coding regions increased with copy number (slope = 0.003; P = 0.0048), consistent with gBGC (Rousselle et al. 2019; Fig. 3d). This pattern is expected because more copies provide a greater opportunity for gene conversion. Second, pairwise differences across whole gene sequences decreased with copy number (slope = −0.688; P < 0.001), indicating more homogenization with higher copy number (Fig. 3e). Third, gene copies in closer physical proximity showed higher sequence similarity than distant copies (slope = 2.48 × 10⁻⁵; P < 0.001) (Fig. 3f), consistent with proximity-dependent gene conversion across tandem repeats. A small number of individual gene copies, however, remained highly diverged from all neighbors regardless of physical distance, suggesting occasional loss of gene conversion activity at specific loci (Fig. 3e and f). This homogenization pattern resembles that previously reported for TSPY protein-coding copies on the Y chromosome, where recombination between palindrome arms and/or direct repeats maintains high sequence similarity (Bonito et al. 2021; Makova et al. 2024; Pal et al. 2026).

An alternative explanation for these patterns is that differences in duplication age, rather than gene conversion, underlie the observed sequence divergence. Similarly, the correlation between GC3 and copy number could reflect expression-associated codon usage rather than gBGC, since highly expressed genes tend to be more GC rich and are preferentially retained as duplicates (Collins et al. 2022). We did not directly control for expression level and cannot fully exclude these alternative explanations. However, we consider gene conversion more parsimonious: A systematic, directional increase in GC3 with copy number is a specific signature of gBGC rather than an expected outcome of duplication age or codon usage alone. Furthermore, as detailed below, copy number does not vary extensively within species, and tandem-array gene copies within an ampliconic cluster from the same individual tend to cluster together in sequence, rather than showing the divergence pattern expected under independent, age-driven accumulation of differences (Fig. 6; Figs. S8 to S10). A similar pattern was observed for Y ampliconic gene clustering across species: Genus-specific clustering was considered inconsistent with independent, lineage-specific retention of ancestral copies and was instead attributed to sequence homogenization via gene conversion (Pal et al. 2026). Together, these findings extend prior evidence of palindrome-associated and tandem-repeat gene conversion to a broader set of primate ampliconic gene families and demonstrate that this process operates on both the X and Y chromosomes, acting to keep ampliconic gene copies similar.

Fig. 6.

Intraspecific copy number variation in chimpanzee, human, and gorilla of X-linked ampliconic gene families under positively selection. Number of intact gene copies per family (columns) in each individual (rows). Copy numbers were estimated from long-read de novo assemblies per individual. Individual names are anonymized as “Species n.”

Pervasive Positive Selection in Gene Families on Multiple Levels

Positive Selection at the Species Level

To test for positive selection, we used PAML's maximum likelihood framework for estimating dN/dS rates (ω) across primate evolutionary history by integrating over all gene sequences in the context of a phylogeny (Yang 2007). Across species, we detected no signs of positive selection across the entire phylogeny for any of the Y chromosome families, which is consistent with previous findings (Makova et al. 2024; Fig. 1d). However, pairwise dN/dS estimates revealed lineage-specific selection patterns. DAZ showed elevated dN/dS values in comparisons between human and both bonobo and chimpanzee, and branch tests revealed significantly higher ω of the hominin clade (ω = ∼2 to 3 vs. ∼0.7), suggesting accelerated evolution on these lineages (Fig. 4b). RBMY showed elevated dN/dS (∼4) in comparisons between the chimpanzee and bonobo. All other Y chromosome families showed patterns consistent with purifying selection or neutral evolution (Table S7).

Fig. 4.

Species-specific positive selection in gene families. a) Pairwise dN/dS estimate across gene copies in ampliconic clusters. Between-species comparisons show dN/dS calculated between paralogous gene copies from different species; within-species comparisons show dN/dS calculated between different members of the same ampliconic cluster within one individual. SSX, GAGE, CSAG, and VCX show elevated evolutionary rates across multiple lineages, with no single species or clade driving the signal, suggesting pervasive positive selection across the phylogeny. Cells are marked where dN/dS is significantly >1; and seperately where dN/dS is significantly <1, both after bootstrapping. b) PAML branch test results. Omega (ω) values are shown for each branch of the gene tree, with foreground branches in red. LRTs were used to compare the null and branch models. PAML detected significant positive selection in the hominini clade (chimpanzee, bonobo, and human) for X-linked ampliconic clusters CT45 and HSFX4 and the Y-linked DAZ cluster. The ampliconic cluster MAGEB6 shows positive selection in the African great ape clade (chimpanzee, bonobo, human, and gorilla).

In contrast to the Y chromosome, several X-linked gene families showed evidence of positive selection across the species tree (Table S8). To assess whether these ω values were elevated compared with those of other X chromosome genes, we compared them to all single-copy X genes across the eight species, which had a median ω of 0.16 (IQR: 0.06 to 0.28). The gene families under positive selection were exclusively expressed in the testis, either before or after MSCI. A greater proportion of post-MSCI genes showed evidence of positive selection (8/16, 50%) than did pre-MSCI genes (9/36, 25%), although this difference was not statistically significant (P = 0.15, proportional binomial test) (Fig. 1e). This pattern is consistent with findings that expression evolution is fastest for postmeiotic cell types (Murat et al. 2023).

To identify which lineages or clades drove positive selection patterns, we estimated pairwise dN/dS values across species. Focusing on the largest ampliconic cluster within a gene family, five gene families (CSAG, GAGE, SSX, VCX, and MAGE) show higher dN/dS values across nearly all pairwise comparisons (Fig. 4a), suggesting pervasive rather than lineage-specific positive selection. We estimated higher dN/dS values for VCX genes in hominins, CSAG in gorilla and orangutans, and FLJ390960 (MAGED4) primarily in orangutan lineages. However, these patterns were not supported by branch tests (Table S9), likely due to short sequence lengths and uniformly elevated dN/dS values across the phylogeny, both of which reduce the power of branch-specific tests (Duchemin et al. 2023). Signals of positive selection on these gene families have been previously reported, although not across this broad range of species sampling (Stevenson et al. 2007; Liu et al. 2008). Within the MAGE family, both ampliconic and single-copy clusters showed evidence of positive selection, particularly in (African) great apes (Fig. 4b; Table S8). Several ampliconic clusters, particularly the rapidly evolving families CSAG, GAGE, SSX, and VCX, also showed elevated dN/dS values when comparing paralogous gene copies within a species (Fig. 4a; Table S8).

In addition to these gene families showing pervasive selection, we also detected evidence of positive selection in the hominini clade for CT45 and HSFX3/4, with ω estimates 3- to 4-fold higher along these branches, respectively (CT45: 3.79 to 7.38 in hominini vs. 1.10; HSFX3/4 4.12 to 5.67 in hominini vs. 1.24) (Fig. 4b). The function of CT45 is unknown, but interestingly, humans have twice as many copies as the chimpanzee and bonobo of CT45 (Fig. 1e). HSFX is predicted to function as a DNA transcription regulator of stress response genes (Hästbacka et al. 2025). The expression of HSFX3/4 is slightly greater in the hominini, than in macaque (Fig. S2). MAGEB6 exhibited positive selection in the African great ape clade, with ω estimates approximately 0.5 higher than those of all other branches (Fig. 4b).

Positive Selection at the Site Level

We next used PAML to search for evidence of positive selection at a subset of amino acids within the gene families (site-specific tests) (Yang 2007). No positive selection was detected at individual sites for any Y-linked gene family (Fig. 1d; Table S11). Among the X-linked gene families, in contrast, several showed evidence of positively selected sites despite exhibiting overall neutral or purifying selection across the species (Note S1; Table S10; Fig. S7). Two rapidly evolving X-linked families, GAGE and CT45A, each carried only one to three sporadic positively selected sites, which did not cluster in any particular protein domain or region (Table S10). Many ampliconic gene families, including three other rapidly evolving families (SSX, MAGEB6, and VCX), encode for primarily intrinsically disordered proteins (IDPs), which lack stable 3D structures (Fig. 5; Fig. S7). Because reliable 3D structural predictions are not available for IDPs, we instead mapped positively selected sites onto linear protein domain annotations for these three families. In all three rapidly evolving families, positively selected sites were generally dispersed across protein sequences. VCX showed a compact cluster of positively selected sites located outside its known VCX/Y domain (Fig. 5a). MAGEB6 displayed dispersed positive selected sites throughout the protein, with an apparent hotspot in the first helix of its MAGE domain (Fig. 5c). The SSX cluster showed a scattering of positively selected sites with a notable exception: The SSXRD domain contained no positively selected sites, while the KRAB domain accumulated eight (Fig. 5b). Disordered regions play key roles in cellular processes, including transcriptional control, cell signaling, and subcellular organization (Holehouse and Kragelund 2024), and are frequent targets of positive selection in humans and across the mammalian phylogeny, thought to be hotspots for genetic innovation (Afanasyeva et al. 2018).

Fig. 5.

Distribution of positively selected sites in rapidly evolving gene families. Protein domains are based on human amino acid sequences as predicted by InterPro (Blum et al. 2025). Positively selected sites were identified by PAML site tests. a) VCX shows a compact cluster of positively selected sites in the intrinsically disordered region outside the known VCX/Y domain. b) SSX shows positively selected sites dispersed across the protein sequence, with absence in the SSXRD domain and concentration in the KRAB domain. c) MAGEB6 shows positively selected sites dispersed throughout the protein, with a hotspot in the first helix of the MAGE domain.

Intraspecific CNV in Human, Chimpanzee, and Gorilla

Besides comparative genomics between different species, we investigated CNV within species for the X-linked gene families found to be under positive selection (CSAG, CT45, GAGE, HSFX4, MAGEB6, SSX, and VCX), using the long-read de novo assemblies of Porsborg et al. (2025) of human (n = 6), chimpanzee (n = 6), and gorilla (n = 3) individuals, all males. Copy number varied little among individuals of the same species, and the variation that did occur was concentrated in the largest families, consistent with the cross-species comparison (Fig. 6; Fig. S1) and with reports for Y-linked ampliconic families (Vegesna et al. 2020). MAGEB6 was the most stable with two copies in every individual of all three species, the only exception being the single copy in the chimpanzee reference. HSFX4 was likewise invariant at two copies in all human and chimpanzee individuals, with the exception of one chimpanzee having one copy; the family is absent from gorilla. VCX was stable within species but different markedly between them, with ∼11 copies in chimpanzee (range: 9 to 11 copies) against only two to four copies in human and a uniform four in gorilla. CT45 was invariant at four copies in all chimpanzees and near invariant in gorilla (range: three to four copies), varying only in human (range: four to eight copies), although less than previously reported in human populations (Lucotte et al. 2023).

GAGE and SSX were the most variable families, though not uniformly across species. GAGE ranged from 4 to 11 copies in chimpanzee and 9 to 21 in human, the human range reflecting the variation described among human populations (Lucotte et al. 2023), yet was completely stable at 14 copies in all gorillas. SSX varied from 10 to 15 copies in human and 10 to 12 in gorilla and was stable at eight copies in all chimpanzees except one individual carrying only two. SSX variation might be partly attributable to undercalling in copy number for some assemblies since it is distributed on more than one contig, but for the other six amplicons, we are confident in the CNV.

The gene trees clarify how these copies relate (Figs. S8 to S10). In most families, copies from different individuals interleave and resolve into a small number of divergent clusters, each containing roughly one copy per individual. This is clearest for MAGEB6 and HSFX4, where two clades separate, each containing a single copy from every individual. This is the signature of duplications at the root of the tree, with orthologous copies maintained in all species. VCX and SSX show a similar but weaker pattern, with some copies identical within an individual, reflecting gene conversion or recent duplication. GAGE is the clear exception: Copies from the same individual frequently group together (eg three chimpanzee two copies and two separate chimpanzee reference clusters), indicating gene conversion in the tandem array and concerted evolution homogenizing copies within an individual rather than preserving the duplication history.

We then calculated, for each individual, pN/pS among the gene copies within each ampliconic cluster (Table S12). These within-individual values are not directly comparable to the dN/dS estimates described above, which were computed on alignments of all species with complete deletion of gapped columns, but they are appropriate for comparisons within a species. Within a species, pN/pS was often broadly consistent across individuals irrespective of copy number, indicating that the paralog diversity of the reference genome is typical of its species. The reference itself was sometimes an outlier relative to this range: CT45 in human had pN/pS of ∼3.2–3.5 in five of the six individuals but only 1.18 in the reference, and chimpanzee CSAG gave pN/pS of ∼0.93 in individuals but 0 in the reference, where the two reference copies differ only synonymously. For CT45, positive selection was inferred from a branch test on the hominini branch, which contrasts divergence between species rather than variation among copies within an individual; the elevated within-individual pN/pS in several humans is therefore a separate observation and does not affect that conclusion. Second, copy number does not explain this variation: For human CT45, three individuals share seven copies, yet one (Human 4) has a pN/pS of 1.33 against 3.38 and 3.48 in the other two. HSFX4 copies are essentially identical within individuals (pN = 0 in chimpanzee), consistent with their near-zero branch lengths in the tree.

Discussion

Using complete T2T sex chromosome assemblies across eight primate species, we found pervasive positive selection on X-linked ampliconic genes, with several families showing elevated dN/dS ratios across the primate phylogeny. In striking contrast, Y-linked families predominantly evolved under purifying selection. Gene conversion operates through palindromic structures and tandem arrays to homogenize sequences within families, maintaining high sequence similarity despite accumulating mutations. CNV on X- and Y-linked families is extensive, with some families showing extreme lineage-specific amplifications, yet X-linked families maintain conserved chromosomal positions across species despite these dramatic copy number changes, while Y-linked families show frequent positional turnover. For the X-linked families under positive selection, intraspecific CNV was limited; copy number varied modestly among species within chimpanzee, human, and gorilla but differed between species, with the largest families varying most. The connection between gametogenesis-specific expression, positive selection, and extreme CNV of ampliconic families suggests strong selective pressures are acting on these genes. Several nonmutually exclusive mechanisms may explain the rapid evolution.

First, sperm competition intensity varies with mating systems and could influence ampliconic gene evolution since their expression is primarily in the testis. Species with multi-male mating systems (polyandry and polygynandry) experience greater sperm competition, as evidenced by increased relative testis size and increased meiotic index (Skakkebæk et al. 2022). The elevated evolutionary rates or increased copy numbers in some ampliconic families could reflect adaptation to species-specific sperm competition regimes. Consistent with this hypothesis, bonobo (the species experiencing the highest degree of sperm competition among those examined) shows extreme copy number expansion in several families, including GAGE, SPANX, VCX, and RBMY. The functional significance of such CNV is supported by findings in humans that RBMY copy number correlates with sperm count and motility (Yan et al. 2017), suggesting direct phenotypic consequences of ampliconic gene dosage and therefore possibly selection in bonobo. Conversely, gorillas, whose polygynous, single male mating system reduces sperm competition, show relaxed purifying selection on male reproductive genes and reduced spermatogenic efficacy (Bowman et al. 2026), and correspondingly lack the expansions seen in bonobo. However, the relationship between copy number and sperm competition is not universal, nor does high copy number appear universally advantageous. Some families show extreme amplification in species with lower sperm competition: TSPY reached 44 copies in human (low sperm competition species) and CDY 20 copies in B. orangutan (intermediate sperm competition species). In humans, selection acts against nonreference copy numbers for many ampliconic families, suggesting possible optimal copy number ranges (Lucotte et al. 2023). TSPY is a notable exception, as it shows high copy numbers across all species, with most copies occurring outside palindromes, contrasting with the typical palindrome-associated amplification pattern. Together, these findings suggest that additional selective forces beyond sperm competition are at play.

Second, meiotic drive may contribute to ampliconic gene evolution, where co-amplified X and Y genes act as drivers and suppressors in genomic conflict over sex chromosome transmission, which could contribute to ampliconic gene evolution. Previous studies in mice and cattle have shown co-amplification of gametolog gene families on the X and Y chromosomes (Soh et al. 2014; Hughes et al. 2020). In mice, the knockout of either Slx (X-linked) or Sly (Y-linked) in mice causes transmission distortion and non-Mendelian sex chromosome segregation (Moretti et al. 2020; Arora and Dumont 2022; Kopania et al. 2022; Arlt et al. 2025). This co-amplification reflects an evolutionary arms race, where amplification of a driver selects for either suppression or counterdrive. We confirmed co-amplification in two gametolog pairs (VCX/Y and HSFX/Y), consistent with a similar type of potential meiotic drive. If meiotic drive operates in primates, we would expect stronger effects in species with lower sperm competition, where relaxed selection on sperm quality permits transmission bias. The expression patterns might support this prediction: VCY is expressed in human, and HSFY is expressed in human and macaque (low/intermediate sperm competition), but both not in bonobo and chimpanzee (high sperm competition). However, their X-linked counterparts, VCX and HSFX, are expressed similarly across all species. The expression timing of the gene pairs differs: HSFX/Y is expressed post-MSCI in round and elongated spermatids, whereas VCX/Y is expressed pre-MSCI in leptotene spermatocytes. Notably, loss of HSFY is associated with male infertility in humans (Shinka et al. 2004), suggesting functional importance despite species-specific expression patterns.

Meiotic drive need not involve gametolog pairs directly. In Drosophila, selfish distorters exploit a meiotic checkpoint gene to cause segregation distortion by eliminating competing sperm (Ridges et al. 2026). Similar mechanisms could operate in primates, with different gene families competing through amplification, although such effects are difficult to detect.

A third mechanism for possible amplification is dosage compensation, which has been documented in mice and Drosophila (Lucchesi and Kuroda 2015) and may involve primate ampliconic genes. Dosage compensation could provide an alternative explanation for the VCX/Y and HSFX/Y patterns we observed. Rather than representing meiotic drive through genomic conflict, continued expression of VCX and HSFX across all species could compensate for the loss of VCY and HSFY expression (Vegesna et al. 2019). This interpretation is consistent with our finding that VCX and HSFX maintain expression across all species despite variation in their Y-counterpart expression. Beyond compensation, another possible mechanism is dosage regulation, where ampliconic genes are expressed at lower levels than their nonampliconic homologs. In great apes, some Y ampliconic gene families maintain stable expression levels despite substantial CNV (Vegesna et al. 2020). In humans, specifically, most Y ampliconic gene families show expression levels lower than their non-Y homologs, indicating this dosage regulation (Vegesna et al. 2019). This downregulated expression could reflect subfunctionalization (eg testis-specific expression that benefits germline development) or relaxed selection. The latter is supported by the finding that Y ampliconic genes show weaker purifying selection (higher dN/dS) than nonampliconic Y genes (Betrán et al. 2012). This relaxed constraint, combined with gene conversion, could facilitate rapid functional differentiation (Vegesna et al. 2019; Makova et al. 2024).

These mechanisms, possibly driving ampliconic gene evolution, sperm competition, meiotic drive, and dosage compensation/regulation, are not mutually exclusive and likely interact. Distinguishing among these alternatives will require additional data and approaches. Our expression analysis used data from four species, and our intraspecific analysis tree, and resolving how these mechanisms combine will need broader sampling still. Intraspecific CNV is substantial for at least several ampliconic families. In humans, the X-linked families GAGE, CT45, and SPANX show CNV, as do the Y-linked families TSPY, DAZ, BPY, CDY, and HSFY, often associated with palindrome duplications, and such families are more highly expressed in pachytene spermatocytes (Lucotte et al. 2018). Using finer-resolution cell-stage data, we found that families with high copy numbers and inter- and intraspecific CNV (GAGE, SSX, VCX, MAGE, CT45, and CSAG) are predominantly expressed in differentiated spermatogonia and leptotene spermatocytes (pre-MSCI), and these are also families that show positive selection across primates. Overall, families expressed post-MSCI show weaker interspecific CNV but still carry signals of selection. Several next steps would bring us closer to disentangling these mechanisms. First, denser sampling, more individuals per species and more species, would characterize variation across the primate phylogeny. Complementing this, as long-read sequencing becomes more available, sequencing parent offspring trios could capture duplication and gene-conversion events as they arise, giving insights the mutational process of ampliconic gene families. Beyond sequence alone, expression and functional data across spermatogenesis would help connect copy number and selection to the stage-specific roles of different gene families. Together, integrating CNV, positive selection, and expression patterns, across both inter- and intraspecific scales, will move us toward understanding which evolutionary mechanisms drive the strange dynamics of ampliconic gene families.

Materials and Methods

Detection of Ampliconic Gene Families

To study ampliconic genes, we obtained T2T reference genomes from NCBI for six great ape species [human {Homo sapiens}, bonobo {Pan paniscus}, chimpanzee {Pan troglodytes}, western lowland gorilla {Gorilla gorilla}, B. orangutan {Pongo pygmaeus}, and S. orangutan {Pongo abelii}], one lesser ape [siamang gibbon {Symphalangus syndactylus}], and one monkey [crab-eating macaque {Macaca fascicularis}] (Miga et al. 2020; Rhie et al. 2023; Makova et al. 2024; Zhang et al. 2025). Multicopy gene families were identified using the pipeline from Makova et al. (2024), based on the longest isoform protein sequences for all X and Y genes. Gene families were classified as multicopy when ≥2 sequences showed >50% identity over >35% of their length with an e-value of ≤0.001. To identify ampliconic gene families, we applied a ≥97% identity threshold (Table S1). A gene family was considered ampliconic if at least one species contained ≥2 copies with ≥97% identity, based on a natural breakpoint in the pairwise sequence identity distribution for Y multicopy genes (Makova et al. 2024). For each family, we compiled copy number counts, sequence class assignments across the chromosomes (as classified by Makova et al. 2024; Tables S2 and S3) for gene copy locations, and all pairwise identities of gene copies within each species.

To identify X–Y paralogous gene pairs (gametologs), we performed reciprocal BLASTp from BLAST+ v2.16.0+ searches between all X and Y chromosome protein sequences (Camacho et al. 2009). Gametolog pairs were retained if they showed ≥30% sequence identity over ≥50% of the alignment length with an e-value of ≤0.001. Gene pairs were further classified based on their amplification status on each chromosome using the ampliconic classifications defined below to identify co-amplification of X and Y gene families. To test whether ampliconic status was associated with gametolog status, we performed a two-tailed Fisher's exact test for each species, separately for the X and Y chromosomes. For each chromosome, genes were classified by whether they had a gametolog pair on the other sex chromosome and whether they were ampliconic, and the association between these two classifications was tested using a 2 × 2 contingency table.

Clustering of Ampliconic Gene Families

Nucleotide coding sequences were extracted for each gene in an ampliconic gene family. Isoforms were selected via the following priority hierarchy: RefSeq select tag if present; otherwise, no isoform tag; isoform X1; isoform 1 or a; or the lowest isoform number or alphabetically first isoform. This hierarchy was necessary because isoform naming was inconsistent within and across annotation files, and we wanted to use the first recorded isoform for each gene to remain consistent. It should be noted that transcript isoform diversity was not included in this study, even though this exists for at least some Y-linked ampliconic gene families (Tomaszkiewicz et al. 2023). We used BLASTn from BLAST+ v2.16.0+ to compare all sequences within an ampliconic family across species against a BLAST database built from the extracted sequences (Camacho et al. 2009). The genes were clustered into ampliconic clusters using transitive clustering (if genes A and B shared sequence identity and genes B and C shared sequence identity, genes A, B, and C formed a single cluster). Genes from different species belonged to the same ampliconic cluster if they showed ≥95% identity over 80% of their sequence length. For the macaque, we used a more lenient threshold of ≥90% identity over 80% of sequence length to account for its ∼5% divergence from the other species. Overall identity thresholds were lower than protein-based clustering because we used nucleotide sequences, and isoform length variation could affect clustering results. Individual gene copies in one species, forming its own cluster, were excluded from ampliconic cluster overview because they could not be compared.

Identifying Gene Family Movement Across the Chromosome

To assess gene family synteny across species, we first plotted all gene locations and orientations on the X and Y chromosomes for each species, colored by sequence class (Makova et al. 2024). For this overview visualization, we did not subset genes into ampliconic clusters to maintain clarity and reduce visual complexity. To visualize gene family movements across species, we created chromosome synteny plots using D3.js v6 (Bostock et al. 2011). Gene families were clustered based on genomic proximity using a 500 kb tolerance threshold, grouping nearby gene copies into single positional markers. For each gene family in each species, we calculated the average genomic position of clustered copies. When a gene family had multiple separated copy groups across a chromosome, each group was represented as a distinct cluster. Gene families were tracked across consecutive species in the phylogeny to visualize positional changes. For families with equal cluster numbers between adjacent species, we drew direct one-to-one connections. For families with unequal cluster numbers, we used nearest-neighbor matching for multicluster families (X chromosome: H2AB, MAGEB, PABPC, TMSB15B, and SPANX; Y chromosome: VCY, TSPY, RBMY, HSFY, DAZ, CDY, and BPY) to minimize visual complexity, connecting each cluster to its nearest genomic neighbor in the adjacent species. Other families with unequal cluster numbers were connected using all possible pairwise combinations to show potential splits or mergers. Centromeric regions were defined based on satellite annotations in the sequence classes, and chromosomes were drawn to scale based on their actual lengths (Makova et al. 2024). Gene family labels were positioned on the left for families present in bonobo and on the right for families absent in bonobo, with automatic grouping of nearby labels to prevent overlap.

Clustering of genes near palindromes was assessed using permutation tests. For each species, we calculated the distance from each gene midpoint to the nearest palindrome boundary. To generate a null distribution, we randomly placed genes across the chromosome (uniform distribution from 0 to maximum chromosome coordinate) and calculated mean distances to palindromes (10,000 permutations). Statistical significance was assessed using permutation tests (P < 0.05). Clustering strength was quantified as the ratio of observed to expected mean distances.

Species-Specific Positive Selection Tests

Pairwise Positive Selection Tests

When at least one species had more than one gene in an ampliconic cluster, we performed codon-based alignment using MACSE v2.07 (-prog alignSequences); clusters containing only single genes per species were excluded from further analysis. Alignments were refined (-prog refineAlignment) and cleaned (-prog exportAlignment -codonForInternalStop NNN -codonForInternalFS ----charForRemainingFS ---) to enable downstream analyses (Ranwez et al. 2018). Pairwise synonymous (dS) and nonsynonymous (dN) nucleotide substitution rates were calculated using the modified Nei–Gojobori method with complete deletion of gaps in MEGA v11.0.13 (Kumar et al. 2012; Tamura et al. 2021). These values were averaged across all pairwise gene comparisons within each ampliconic cluster for every two-species comparison, yielding dN and dS estimates for all gene copies in each cluster across all species pairs. Some comparisons resulted in infinite dN/dS values when no synonymous substitutions were present, reflecting insufficient sequence divergence for reliable rate estimation. To test for significance, each ampliconic cluster alignment was bootstrapped 10,000 times by codon. For each bootstrap replicate, dN and dS were recalculated. We interpreted dN/dS as significantly greater than 1 (positive selection) if ≤0.05% of bootstrapped dN/dS values were below 1, and significantly less than 1 (purifying selection) if ≤0.05% of bootstrapped values were above 1.

Lineage-Specific Branch Tests

Pairwise species comparisons of dN/dS values revealed that certain species showed positive selection more frequently than others across comparisons. Based on these patterns, we identified candidate species or clades that might be driving positive selection in each ampliconic cluster's phylogeny. To test these hypotheses, we performed branch tests using PAML CODEML v.4.10.9 (Yang 2007), designating the candidate species or clades as foreground branches to test for lineage-specific positive selection. For all ampliconic gene clusters on the X and Y chromosomes, we used the multisequence codon alignments generated previously and randomly selected one gene per species from each alignment. This random selection was necessary because CNV between species within multisequence alignments could bias selection tests. Based on the selected genes, we pruned the unrooted species tree (((((bonobo, chimpanzee), human), gorilla), (S. orangutan, B. orangutan)), siamang, macaque)) to include only species present in each alignment, as some species lacked sufficiently similar copies to be included in certain ampliconic clusters. An unrooted tree was used because the branch models are time reversible and do not assume a molecular clock, since all tests were done on clades that were not part of the root. All gaps were removed from alignments before running CODEML, and cleandata = 0 was used to preserve information in the data as recommended by Álvarez-Carretero et al. (2023). Alignments were converted from FASTA to PHYLIP format using FASTAtoPHYL.pl (Álvarez-Carretero et al. 2023). Using each alignment and pruned species tree, we ran PAML CODEML under the M0 (one-ratio) model and a branch model allowing ω to vary between foreground and background branches. We compared models using likelihood ratio tests (LRTs), where the test statistic is defined as 2Δℓ = 2(ℓ1 − ℓ0), with ℓ0 and ℓ1 representing log-likelihoods under the null and alternative models, respectively. The LRT statistic follows a χ² distribution with degrees of freedom equal to the difference in free parameters between models. To assess robustness, we repeated this process six times for each ampliconic cluster with different randomly selected gene copies representing each species. This number was chosen to match the highest observed median cluster copy number (6.5 copies within one ampliconic cluster per species), ensuring that we could sample the majority of available gene copies and capture within-cluster sequence diversity.

Positive Site Selection Tests

To detect positive selection at codon sites beyond overall and species-specific ampliconic gene cluster estimates, we performed site tests using PAML CODEML v.4.10.9 (Yang 2007). Following the same approach as for branch tests, we randomly selected one gene per species from each multisequence codon alignment, cleaned the alignment of gaps, converted it to the phylip format and pruned the species tree accordingly, resulting in unrooted trees. To assess robustness, we repeated this process six times for each ampliconic cluster with different randomly selected gene copies representing each species, matching the approach used for branch tests. Using each alignment and pruned species tree, we ran PAML CODEML under various site models that allow ω to vary across codons: the homogeneous M0 (one-ratio) model and the site-heterogeneous models M1a (nearly neutral), M2a (positive selection), M7 (beta), and M8 (beta and ω). We compared nested models (M0 vs. M1a; M1a vs. M2a; M7 vs. M8) using LRTs, as described above. This approach identified codons under positive selection. We excluded clusters with only one to three positively selected sites, as these likely represented false positives or alignment errors. We also excluded positively selected sites from clusters containing only two species, as the random sampling approach would base these results on only two sequences, limiting reliability. For clusters with positively selected sites, we mapped the sites to the corresponding positions in a human gene from that ampliconic cluster. We predicted protein domains using InterPro (Blum et al. 2025) and visualized the distribution of positively selected sites along the protein to assess for clustering patterns in specific protein domains.

Concerted Evolution of Multicopy Gene Families

We performed four tests to assess the occurrence of gene conversion across copies within gene families. First, we calculated median GC3 content for coding sequences per gene family in each species and correlated these values with copy number. GC3 was calculated on the codon alignment made previously for dN/dS estimations. We expected gene families with higher copy numbers to show increased GC3 content due to gBGC because of increased opportunity for gene conversion (Galtier et al. 2001; Rousselle et al. 2019). The significance of each correlation slope was assessed by permuting the median GC3 versus copy number relationship across gene families. Second, we calculated pairwise sequence differences between all gene pairs within each ampliconic cluster per species. We used whole gene sequences (5'UTR, CDS exons, 3' untranslated regions and introns) rather than coding regions alone to increase power, as gene conversion likely affects entire sequences. Whole gene sequences were retrieved per family per species and aligned using MAFFT v7.525 with 1,000 iterations (Katoh and Standley 2013). After removing all gaps, we calculated pairwise differences and correlated them with palindrome status: both genes on opposite palindrome arms, both on the same arm, in different palindromes, one in a palindrome and one not, or both outside palindromes. Palindrome classifications were based on sequence classes from Makova et al. (2024). We expected genes on opposite palindrome arms to show greater similarity due to gene conversion. Third, we correlated pairwise differences within gene families with copy number, expecting larger families to show fewer pairwise differences due to increased opportunity for gene conversion. Fourth, we compared pairwise differences with both absolute genomic distance (base pairs calculated from midpoint to midpoint of the two genes). We expected gene conversion to decrease with distance, as it occurs more frequently in tandem repeat arrays, as demonstrated for TSPY and RBMY arrays on the Y chromosome (Hallast et al. 2023). The significance of these relationships was also assessed by permutation testing to account for nonindependence of pairwise comparisons.

Expression Analysis During Spermatogenesis

Single-nucleus and single-cell RNA-seq data analyzed in this study were obtained from Meritxell Riera et al. (personal communcation), including newly generated chimpanzee testicular snRNA-seq data from an individual housed at Copenhagen Zoo and previously published datasets for human, chimpanzee, bonobo, and rhesus macaque (Murat et al. 2023). Reads were aligned to the human reference genome (T2T-CHM13v2.0) using STAR/STARsolo, with UMI quantification allowing multimapping reads resolved by expectation–maximization (Nurk et al. 2022). Comparative analyses were restricted to 1:1 orthologous protein-coding genes and long noncoding RNAs identified by reciprocal best BLAST hits. Low-quality nuclei/cells, predicted doublets, ambient RNA contamination, and empty droplets were removed prior to downstream analysis. Data were normalized with Sctransform and integrated across species using scVI to generate a shared latent space (Lopez et al. 2018; Hafemeister and Satija 2019; Choudhary and Satija 2022). Cells were clustered with Leiden, and spermatogenic trajectories were inferred by diffusion pseudotime analysis, computed separately for diploid and haploid stages and rescaled for cross-species comparison. Cells were binned along X- and Y-bearing spermatid pseudotime trajectories, and log-normalized gene expression was modeled across pseudotime using a spline-based hierarchical Bayesian framework in PyMC (Abril-Pla et al. 2023). Expression was decomposed into shared (across-species) and species-specific components, and genes were classified as expressed or species specific based on posterior credible intervals and log-fold change thresholds. All downstream analyses in the present study were performed using the processed, normalized, integrated, and pseudotime-resolved data as provided by the original publication.

Intraspecific Variation in Human, Chimpanzees, and Gorillas From Long-Read De Novo Assemblies

To examine X-linked ampliconic clusters showing signs of positive selection (in the branch test or pairwise species comparison: CSAG, CT45, GAGE, HSFX, MAGEB6, SSX, and VCX), we used long-read de novo assemblies from male individuals from Porsborg et al. (2025). We included humans (n = 6), chimpanzees (n = 6), and gorillas (n = 3) and compared the copies of each family across individuals within species. As the assemblies were contig level rather than chromosome resolved, we first identified the contigs of X-chromosomal origin by mapping each haploid assembly against its species' reference X and Y chromosomes with wfmash v0.24.2-0 (Guarracino et al. 2021) (approximate mapping, minimum identity 95%, minimum block length 20 kb). A contig was retained as X linked if its total alignment to the X chromosome was ≥20 kb, covered ≥30% of the contig length, and exceeded its total alignment to the Y, excluding Y-linked gametolog. All subsequent searches were restricted to these contigs. For each ampliconic cluster, we extracted the CDS of the longest isoform of every member locus and searched them against each individual's X-linked contigs with BLAT v.35 (Kent 2002), retaining alignments with ≥50% query coverage and ≥97% identity. Overlapping hits from different loci were merged into single copies. In tandem arrays, where copies are nearly identical, BLAT can wrongly align a single query across several adjacent array copies, producing one alignment that spans the whole array. When these are merged, this collapses the entire array into a single copy and undercounts it. We therefore discarded any hit whose genomic span exceeded three times the family's median single-copy span, so that these array-spanning alignments did not merge separate copies together.

For each copy, we recovered its coding sequence by aligning the cluster reference proteins to the copy region with miniprot v0.13-r248 (Li 2023), taking the best-matching alignment; copies were flagged intact when a protein aligned as a complete, frameshift-free ORF. This yielded, per individual and family, a copy number and each copy's coding sequence, with the reference genome included as an additional individual. Codon-based alignments of the intact copies were built with MACSE v2.07 (-prog alignSequences), refined (-prog refineAlignment) and cleaned (-prog exportAlignment -codonForInternalStop NNN -codonForInternalFS ---charForRemainingFS ---) (Ranwez et al. 2018). From these, we counted synonymous and nonsynonymous differences by the Nei–Gojobori method between all copy pairs within each individual, summed per individual and divided by the number of aligned synonymous and nonsynonymous sites to give per-site rates (pS, pN). To visualize relationships within and between individuals, we inferred a maximum-likelihood tree for each family and species from the codon alignment with IQ-TREE3 v3.1.1 (Hoang et al. 2018; Wong et al. 2026), with the best-fitting substitution model selected automatically by ModelFinder (-m MFP) and branch support assessed with 1000 ultrafast bootstrap replicates (-B 1000) (Kalyaanamoorthy et al. 2017; Hoang et al. 2018).

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Data Availability

All genome assemblies analyzed during this study are publicly available from NCBI: human (GCF_009914755.1), chimpanzee (GCF_028858775.2), bonobo (GCF_029289425.2), gorilla (GCF_029281585.2), Bornean orangutan (GCF_028885625.2), Sumatran orangutan (GCF_028885655.2), siamang (GCF_028878055.3), and crab-eating macaque (GCF_037993035.2). De novo genome assemblies used for intraspecific variation were obtained from Porsborg et al. (2025). Code used in the manuscript is available at GitHub: https://github.com/emmadiepeveen/Ampliconic-Evolution-Primates