Abstract
Mammalian genomes are biased towards GC bases at third codon positions, likely due to a GC-biased ancestral genome and the selectively neutral recombination-related process of GC-biased gene conversion. The unwanted transcript hypothesis posits that this high GC content at synonymous sites may be beneficial for protecting against spurious transcripts, particularly in species with low effective population sizes. Utilising a 240 placental mammal genome alignment and single-base resolution conservation scores, we interpret sequence conservation at mammalian four-fold degenerate sites in this context and find evidence in support of the unwanted transcript hypothesis, including a strong GC bias, high conservation at sites relating to exon splicing, less human genetic variation at conserved four-fold degenerate sites, and conservation of sites important for epigenetic regulation of developmental genes. Additionally, we show that high conservation of four-fold degenerate sites in essential developmental genes, including homeobox genes, likely relates to the low mutation rates experienced by these genes.
Similar content being viewed by others
Introduction
Evolutionary constraint, or limited sequence evolution over time due to strong purifying selection, is evident in functional regions of genomes. In protein-coding genes, the constraint is largely due to purifying selection acting on mutations that alter the amino acid sequence and disrupt protein function, while mutations at synonymous sites that have no effect on the amino acid specified were originally considered as evolving neutrally1. It is now well-established that selection can act on synonymous sites, particularly in fast-growing species2 and species with large effective population sizes (Ne), where efficient selection can optimise codon use to maximise translational velocity and accuracy3,4. However, it is not clear how pervasive constraint on synonymous sites is in the genomes of slower-growing and low-Ne species, such as humans and other mammals, where evidence for codon usage bias due to translational selection is lacking or absent5,6.
Several estimates of constraint at synonymous sites in mammals have been made; evidence for purifying selection acting on synonymous sites was found in 28% of mouse-rat orthologous genes7, and a genome-wide analysis estimated that ~25% and ~11% of fourfold degenerate (4d) sites are under constraint in hominids and murids respectively8. Evidence for selection on codons due to translational optimisation9, protein folding10,11, accurate splicing12,13, and mRNA stability14 is present in mammals15. In addition, overlapping regulatory features such as transcription factor (TF) binding sites16, binding sites of RNA-binding proteins17, or within upstream open-reading frames (uORFs)18, may bring secondary functions to 4d sites. Selection for CpG dinucleotides, the predominant site of DNA methylation in mammals19, in certain TFs and developmental genes also appears to lead to constraint at synonymous sites20.
Vertebrate genomes are generally GC-rich at third codon positions (‘GC3 content’ hereon), which is in contrast to a genomic background of low GC content due to mutational bias, and may reflect a GC-rich vertebrate ancestral genome21,22. Neutral processes are likely to have had a strong influence on the GC content at synonymous sites, such as GC-biased gene conversion (gBGC) where allelic conversions in heteroduplex complexes formed during meiotic recombination are biased towards GC alleles23,24,25. Indeed, gBGC can result in signatures that ‘mimic’ those due to selection26. For example, the abundant and conserved mammalian stop codon TGA (TAA in most other taxa) is selectively sub-optimal and likely a result of gBGC27, gBGC has likely led to the fixation of deleterious amino acid changes in primates28, and translocation of the mouse Fxy gene into a highly recombining region led to rapid increase in GC3 content due to gBGC29. Recently it has been shown that gene methylation affects local mutation rates30, and mutation rate heterogeneity can also result in different rates of sequence divergence throughout species’ genomes independently from selection. The pervasive impacts of these neutral processes can therefore lead to high sequence conservation among species and need to be taken into account when investigating signatures of selection on codon usage31.
Neutral and selective models of synonymous site evolution are not mutually exclusive and codon usage likely reflects a balance between selective and mutational pressures as well as drift5,31,32,33. Recently, the ‘unwanted transcript hypothesis’ (UTH) has been presented as a novel explanation for synonymous site GC bias and constraint in humans and other species with low Ne32. It proposes that selection on synonymous sites aids the retention of native, functional transcripts and the removal of spurious transcripts. Around 62% of the human genome is transcribed, yet only 1.22% encodes protein-coding exons34. Much of the transcription is likely spurious, generated from the integration of novel sequences into the genome such as from viruses and transposable elements (TEs), leading to costly transcripts whose removal is advantageous32. In species where Ne is low, unwanted transcripts may accumulate due to inefficient selection against their mildly deleterious effects35.
The UTH posits that much of the selection on synonymous sites relates to quality control of transcripts32. Synonymous sites can strengthen the signal that a transcript is native and differentiate it from spurious, non-functional transcripts. In the GC-poor genomes of mammals, native transcripts are generally GC-rich21, CpG poor (due to hypermutability of methylated CpGs), contain introns, and have short exons (80% are <200 bp36). According to the UTH, transcripts that are AT-rich, have high CpG content, are intronless, and contain long exons are therefore most likely to be spurious and targeted for removal. In agreement with this, it has been shown that higher GC content at synonymous sites of transgenes in mammals leads to increased protein levels related to increased nuclear export37,38,39, and intronless retrogenes tend to evolve higher GC content at 4d sites (GC4) than their intron-containing parent genes37. Selection on synonymous sites can therefore act to both reduce the production and increase the removal of spurious transcripts by ensuring accurate splicing and aiding their discrimination, respectively. Importantly, the UTH allows for a model whereby synonymous site GC bias has been driven by gBGC; if GC content is a signal of transcript nativity in mammals, then gBGC may preserve high GC content in native transcripts compared to spurious transcripts elsewhere in the genome32, with gBGC acting predominantly in constrained, gene-rich regions26,40 and opposing the GC → AT mutational bias.
In this study, we utilise an alignment of 240 placental mammal species genomes, produced as part of the Zoonomia project41, to comprehensively characterise synonymous site conservation in mammals. We identify shared synonymous sites from the alignment and use single-base resolution phyloP scores generated in Zoonomia42 to provide the most comprehensive characterisation and assessment of mammalian synonymous site conservation. We interpret our findings in the context of gBGC, mutation rate variation, and the UTH, advancing our understanding of the neutral and selective processes acting on synonymous sites in placental mammals.
Results
Conserved 4d sites are GC-biased
We identified all codons containing 4d sites in the human genome using Gencode annotation v.3943 and taking the canonical transcript per gene (n = 19,386), revealing a total of 5,278,470 4d sites. Many of these are unlikely to be 4d sites in other mammal genomes due to differences in gene content. We therefore used the Zoonomia alignment41 to identify codons where >85% of species were aligned and >95% of the aligned codons contained a 4d site. This resulted in a set of 2,621,118 4d sites distributed over 17,394 transcripts that formed the basis of our analysis (Supplementary Data 1). We extracted the base content of each 4d site within each aligned genome and used the single-base human-referenced phyloP scores generated in Zoonomia as estimates of conservation42. These scores range from −20 to 8.9, with negative scores indicating accelerated evolution, scores close to 0 indicating neutral evolution and positive scores indicating constrained evolution44. A significance threshold for constraint was previously established for this dataset using a 5% false discovery rate (FDR), where sites with a phyloP score ≥2.27 are under significant constraint42. While this 5% FDR threshold was previously used to identify functional sites likely under purifying selection (i.e. ‘constraint’), here when considering synonymous sites we prefer to use the term ‘conservation’ so as not to imply selection. Overall, 20.8% of 4d sites have phyloP scores above this threshold (Table 1). In comparison, 29.4% of threefold (3d), 36.6% of twofold degenerate (2d) and 74.1% of nondegenerate sites show significant conservation at this threshold. Synonymous sites with a phyloP score above the 5% FDR threshold have a mean of 230 species aligned (s.d. = 9.16) and a mean of 188.5 species sharing the same base (s.d. = 37.8). We observe 4906 4d sites (0.19% of all 4d sites) that are fixed for the same base across all 240 genomes (24.3% fixed for A, 21.7% fixed for T, 27.3% fixed for C, 26.7% fixed for G).
All but one of the mammal genomes considered have a GC4 content >50% (Fig. 1a), with only the hispid cotton rat genome (Sigmodon hispidus, Order: Rodentia) below 50% (49.5%). The common shrew (Sorex araneus, Order: Eulipotyphla) has the highest GC4 content at 64.3%, likely due to their large Ne and fast-evolving genomes45. When considering all 4d sites regardless of phyloP, GC content is only slightly higher than AT content in humans and across mammals (Table 1). However, over 80% of bases at 4d sites with phyloP ≥2.27 are G or C (Table 1), and we observe a positive correlation between GC4 content and conservation across the mammalian genomes (Fig. 1b–d; Pearson’s r = 0.28, P < 2.2 × 10−16). Base content at conserved 4d sites is generally skewed towards a specific base in all mammals (Fig. 1e, f). We also see that transcript GC4 content correlates strongly between the human genome and mammal genomes generally (Fig. S1A; Pearson’s r = 0.90, P < 2.2 × 10−16). To test whether 4d site conservation within transcripts relates to the GC4 content of transcripts, we compared the mean phyloP at 4d sites that are GC to the mean phyloP at 4d sites that are AT for each transcript. If 4d site conservation is largely driven by gBGC then we would not expect the two to correlate. However, we observe a strong positive correlation (Pearson’s r = 0.63, P < 2.2 × 10−16; Fig. 1g), suggesting that whilst the majority of conserved 4d sites are GC, conservation within transcripts is not determined solely by gBGC.
a Boxplots showing genome GC content at 4d sites (GC4) per mammalian Order (n = 240 placental mammal genome assemblies). All mammal genomes in the Zoonomia dataset show a GC4 bias except for the hispid cotton rat (Sigmodon hispidus, Order: Rodentia; GC4 = 49.5%). Boxes represent first and third quartiles with median line, whiskers extend ±1.5× IQR, outliers beyond whiskers are shown as points. b The number of each base at 4d sites across 240 placental mammal genomes, binned by phyloP score, where more positive scores indicate stronger conservation. Pie charts showing the proportion of each base seen at 4d sites across the mammal genomes for sites c phyloP <2.27 and d phyloP ≥2.27. At conserved 4d sites (phyloP ≥2.27), we observe a general bias towards C or G at GC sites (e) and a bias towards A or T at AT sites (f) among the mammal genomes. g Per-transcript mean phyloP at 4d sites that are GC positively correlates with mean phyloP at 4d sites that are AT (n = 17,394; two-tailed Pearson’s r = 0.63, P < 2.2 × 10−16). Blue trend line shows a linear regression using the ‘geom_smooth’ function in ggplot2 in R. Colour scale represents mean transcript 4d phyloP. Species silhouettes in (a) were obtained from phylopic.org, which are available under a creative commons licence. Source data are provided as a Source Data file.
Codon-specific GC biases
The major allele in humans (identified from the genotypes of 138,922 individuals in the TOPMed dataset46) matches the most common base across the mammal genomes at 79.5% of all 4d sites, rising to 90.6% at conserved 4d sites. We generated base counts from the human genome across each of the codons containing 4d, 3d and 2d sites to assess whether GC bias is seen for all codons. Conserved 4d sites show a strong GC bias across all codons, with a minimum GC4 content of 59.5% (proline codons) and a maximum of 90.5% (valine codons; Fig. 2a). Differences in base proportions at 4d sites in the codons of each amino acid reflect context-specific restrictions on base content. For example, in codons where the second base is C (codons for alanine, proline, serine and threonine), 4d sites are rarely guanine, which would generate CpG sites. Indeed, 67% of conserved 4d sites in humans are G, and only 5% are C when the subsequent base is G. In glycine codons (GGN), 61% of the 4d sites are C, likely due to selection against GGG codons that can lead to the formation of G-quadruplex structures in mRNA47. However, if the subsequent base is G then 4d C content is low as this would generate CpG sites (38% A, 16% C, 32% G, 13% T). We also see strong biases for GC at conserved 3d and 2d sites in all codons except those for arginine. Conserved synonymous sites in all other codons are at least 72% C or G, with 74% C at conserved 3d sites in the codons for isoleucine (Fig. 2b, c). For sites with low conservation, we do not observe a general bias for higher GC content at synonymous sites (Fig. 2d–f).
Base content of significantly conserved a 4d (n = 544,503), b 3d (n = 141,931), and c 2d synonymous sites (n = 2,287,435) shows a strong GC bias for all amino acid codons in humans. Not significantly conserved d 4d (n = 2,076,615), e 3d (n = 340,674), and f 2d synonymous sites (n = 3,946,958) do not show such a bias. For conserved 4d sites, strong biases towards C or G bases largely reflect local base context, where sites that form CpGs are rare, as well as GGG codons that can lead to G-quadruplex structures in mRNA. Source data are provided as a Source Data file.
Higher GC4 content in single-exon genes and first exons
Mammalian transcripts typically contain introns. Under the UTH, native transcripts that are intronless should display high GC content at synonymous sites to aid their recognition as native transcripts despite lacking introns32. We tested this by comparing the proportion of 4d sites that are GC across the mammal genomes in single- (n = 1119) and multi-exon (n = 15,483) transcripts with >10 4d sites. Median GC4 in single-exon transcripts (0.67) is significantly greater than that in multi-exon transcripts (0.57; Wilcoxon rank-sum test, W = 6,264,319, P < 2.2 × 10−16; Fig. 3a). As first and last exons are rarely spliced out48 it may be important that they maintain higher GC content as they are present in most transcripts and, as such, are most exposed to transcriptional silencing and degradation mechanisms, as well as containing essential splice sites. For exons with >10 4d sites, GC4 in the first exons of multi-exon transcripts are significantly higher than in other exons, (Fig. 3a; median GC4 in first exons = 0.72 (n = 6070), other exons = 0.58 (n = 79,694); Wilcoxon rank-sum test, W = 157,517,032, P < 2.2 × 10−16). Last exons also show significantly higher GC4 content, but the difference is smaller (median GC4 in last exons = 0.62 (n = 8053), other exons = 0.58 (n = 78,830); Wilcoxon rank-sum test, W = 287,613,287, P < 2.2 × 10−16). We observe a significant negative correlation between exon length and GC4 content for single-exon genes (Pearson’s r = −0.18, d.f. = 1117, P = 1.83 × 10−9). This is predicted under the UTH as GC content is evidently most important towards the 5′ end for cytoplasmic localisation and translation of native transcripts37.
a GC4 content of single-exon genes (n = 1119) and the first (n = 6070) and last (n = 8053) exons of multi-exon genes are significantly greater than GC4 content of internal exons (two-sided Wilcoxon rank-sum tests, P < 2.2 × 10−16 in all cases). Violin plots show data density. b PhyloP is significantly higher at 4d sites within 3 bp of exon-intron boundaries (n = 81,221) compared to 4d sites elsewhere in exons (n = 2,539,897; ANOVA, F = 6561, d.f. = 2, P < 2.2 × 10−16). Red dotted line indicates neutrality. Boxplots in (a, b) show first and third quartiles with median line, whiskers extend ±1.5× IQR. Histograms of base counts across mammals at 4d sites at (c), the first (5′ exon edge, n = 1044) and d last (3′ exon edge, n = 12,049) positions in exons in phyloP bins show strong bias for G at conserved 4d sites. Logo plots for base content of 4d sites at (e), the first six positions (n = 29,871) and f last six positions (n = 19,792) in exons, where the probability reflects the counts of each base across the mammal genomes at 4d sites located at those positions. Grey diagram of exon puts the positions into the visual context of the exon/intron boundaries. Source data are provided as a Source Data file.
High conservation at splice sites
The UTH predicts high constraint at 4d sites involved in transcript splicing to avoid incorrect splicing that can lead to spurious transcripts. Our data revealed significantly higher phyloP at 4d sites positioned adjacent to splice sites compared to 4d sites elsewhere (Fig. 3b; n = 1044 (5′ exon boundary), 12,049 (3′ exon boundary), 2,608,026 (elsewhere in exons); mean phyloP = 1.82 (5′ boundary), 3.70 (3′ boundary), −0.17 (elsewhere); ANOVA, F = 6561, d.f. = 2, P < 2.2 × 10−16). A post hoc Tukey’s test showed that phyloP at exon 3′ boundaries is significantly greater than at 5′ boundaries (difference in mean phyloP = 1.88, P < 2.2 × 10−16). For conserved 4d sites, we see strong biases in base content at or near splice junctions, likely reflecting selection for effective splicing. For example, in mammals there is a strong bias for G at both 5′ and 3′ exon boundaries, with 79.1% and 93.6% G observed, respectively (Fig. 3c–f). This has been shown previously at first and last exonic positions in strong splice sites49,50. Expanding the analysis to encompass 4d sites from human alternative splice junctions that are not used in the canonical transcripts, we see generally lower phyloP than 4d sites at constitutive splice junctions (mean phyloP = 0.03, s.d. = 3.79), but 1267 of them show significant conservation and also show a strong GC bias (Table 1).
Human population variation at 4d sites
We used the TOPMed dataset of human genetic variation46 to assess levels of variation at 4d sites in humans. We see that a lower proportion of conserved sites are variable compared to nonconserved sites (20.9% versus 28.3%) and minor allele frequencies (MAF) are significantly lower at conserved sites (mean MAF = 2.5 × 10−4 and 6.9 × 10−4 at conserved and nonconserved sites respectively; two-sample t test, t = 31.45, d.f. = 1,465,717, P < 2.2 × 10−16). We calculated the proportion of 4d sites that were variable (MAF > 0) in phyloP bins and observed a significant negative correlation with phyloP (Fig. 4a). The relationship was more pronounced when considering the proportion of sites with more frequent variants (MAF > 0.001; Fig. 4b). These patterns are suggestive of stronger negative selection acting on conserved compared to other 4d sites. We also tested whether the proportions of variant types (i.e., major and minor allele combinations) at conserved sites were as expected if the variant type proportions are independent of phyloP. We found a significant difference between observed and expected variant type proportions (Pearson’s χ2 test; χ2 = 261, d.f. = 11, P < 2.2 × 10−16). Specifically, conserved 4d sites show a deficiency in variable A and T sites and an excess of variable C and G sites (Fig. 4c), with A → G sites having the most negative and C → T sites the most positive residuals (−6.56 and 4.50, respectively; Fig. 4d). An excess of GC → AT polymorphisms is known in humans and can be explained by gBGC24. In the context of our results, this suggests that some of the high conservation we observe at CG 4d sites in mammals may have resulted from gBGC rather than purifying selection, whereas a deficiency of variation at conserved AT sites is more suggestive of selection.
a The proportion of 4d sites that are variable (minor allele frequency >0; n = 716,661) based on single-nucleotide polymorphism data from TOPMed data freeze 8 was calculated in phyloP bins. For all four bases, the proportion of variable sites is significantly negatively correlated with mean phyloP in each bin (two-tailed Pearson’s r = −0.86, −0.91, −0.74, and −0.80, P = 0.012, 0.0048, 0.022, and 0.010 for A, T, C and G bases, respectively). b Same as (a) but considering only variants with a minor allele frequency >0.001 (n = 27,818). Again, all four bases show a significant negative correlation with mean phyloP in each bin (two-tailed Pearson’s r = −0.94, −0.92, −0.98, and −0.97, P = 0.0019, 0.0038,4.9 × 10−6, and 1.04 × 10−4 for A, T, C and G bases, respectively). c When considering only significantly conserved 4d sites (phyloP > = 2.27; n = 544,503), we observed a significant difference in the proportions of variant types compared to expected proportions based on total numbers of each variant type independent of phyloP (two-tailed Pearson’s Chi2 test; χ2 = 261, d.f. = 11, P < 2.2 × 10−16). d Residuals from the Chi2 test show that there is a deficiency in variable A and T sites and an excess of variable C and G sites amongst conserved 4d sites. MAF minor allele frequency. Source data are provided as a Source Data file.
Exploring genome correlates with 4d site conservation and GC content
We tested for correlates of GC4 content and phyloP in mammals to further explore evidence for selection versus neutral, mechanistic processes acting on 4d sites, such as gBGC and transcription-coupled repair51. We observed a significant positive correlation between transcript GC4 content across mammals and local GC content (measured in 1 Mb windows across the human genome; Fig. S1B; Pearson’s r = 0.52, P < 2.2 × 10−16). The slope of the fitted linear model is 2.17, demonstrating that genes with high GC4 content show considerably higher GC content than that of their neighbouring sequence. This is likely explained by the fact that the GC content of non-coding regions will be reduced by TE insertions52, as well as a greater likelihood of recombination in conserved coding regions compared to non-coding regions40, leading to a greater effect of gBGC in coding compared to neighbouring non-coding sequences. We found a negative correlation between local GC content and transcript 4d site mean phyloP (Fig. S1C; Pearson’s r = −0.22, P < 2.2 × 10−16), showing that conservation of 4d sites is not associated with being in GC-rich genomic regions in the human genome. The GC4 content of transcripts also negatively correlates with mean 4d phyloP (Pearson’s r = −0.21, P < 2.2 × 10−16; Fig. S1D). In addition, for transcripts with a mean 4d phyloP above the significance threshold of 2.27, there is no relationship between GC4 content and mean phyloP (Pearson’s r = −0.005, P = 0.91; Fig. S1D), further suggesting that high conservation at 4d sites is not purely driven by gBGC. Transcript mean 4d site phyloP is positively correlated with phyloP at 0d sites (Pearson’s r = 0.42, P < 2.2 × 10−16), with a distinct elevation in mean 4d phyloP in the transcripts with the highest 0d phyloP (Fig. S1E). We also observed significant positive correlations between transcript mean 4d site phyloP and UTR mean phyloP (Pearson’s r = 0.47 and 0.53 for 5′ and 3′ UTRs respectively, P < 2.2 × 10−16 in both cases; Figs. S1F, G).
If high GC4 content at 4d sites is driven by transcription-coupled repair then it would be expected that GC4 content should relate to expression level, with highly expressed genes being subject to greater transcription-coupled repair51. To test this, we compared human gene expression data from multiple tissues and developmental stages53 to 4d site phyloP and GC4 content. We show that, in general, GC4 content negatively relates to expression levels, with lowly expressed genes showing higher GC4 content on average than highly expressed genes (Fig. S2; mean GC4 content of lowly and highly expressed genes = 0.59, and 0.54, respectively, ANOVA, d.f. = 2, F = 440.9, P < 2.2 × 10−16). In addition, lowly expressed genes enriched for highly conserved 4d sites have significantly higher GC4 content than those that are highly expressed (Fig. S2; mean GC4 content of lowly and highly expressed genes = 0.70, and 0.59 respectively, ANOVA, d.f. = 2, F = 829.3, P < 2.2 × 10−16). These findings suggest that high expression and related transcription-coupled repair cannot explain the high GC4 content we observe at conserved 4d sites.
Associations between synonymous site GC content, effective population size and genome transposable element content
The UTH predicts that greater GC content of synonymous sites is more important for species with low Ne, where inefficient selection can lead to a greater load of spurious transcripts32. We used mammalian historical Ne estimates generated as part of Zoonomia54 (n = 210 species) to test for a relationship between GC4 content and Ne. Mean Ne across the 210 species is 37,000 (s.d. = 39,000), with a ~1600-fold difference between the species with the highest Ne of 270,000 (prairie deer mouse, Peromyscus maniculatus, Order: Rodentia), and the lowest Ne of 162 (beluga whale, Delphinapterus leucas, Order: Cetartiodactyla). We do not see a significant correlation between Ne and GC4 generally (Pearson’s r = −0.13, P = 0.073; Fig. S3A). However, we do observe positive correlations in three Orders, namely Artiodactyla, Carnivora, and Primates (Fig. S3B). This fits the gBGC model, where a higher equilibrium GC content would be expected in larger populations55.
To assess whether Ne is related to GC content of highly conserved 4d sites, we reduced the data to a set of 4d sites that are represented in all 240 genomes (i.e., no missing data; n = 93,464 sites, mean phyloP = 3.84, s.d. = 1.32). We then estimated genetic distance among all species at these sites and generated a neighbour-joining tree (Fig. 5a). The tree recapitulates the expected topography of the placental mammal tree56, showing that divergence at these 4d sites is generally lineage-specific. GC content at these sites negatively correlates with historical Ne (Fig. 5b; Pearson’s r = −0.48, P = 5.3 × 10−13) as well as mean genetic distance to all other species (Fig. 5c; Pearson’s r = −0.87, P < 2.2 × 10−16). The Indochinese shrew (Crocidura indochinensis, Order: Eulipotyphla) is the most divergent species, having the lowest GC content as well as the third highest Ne in the dataset (~169,000). Contrastingly, the edible dormouse (Glis glis), a rodent with an exceptionally low Ne of 520 shows the highest GC content at these constrained 4d sites among rodents (74%), and cetaceans have amongst the highest GC content at these sites as well as the lowest historical effective population sizes. We also see a negative correlation between genome size and GC4 content at these 4d sites (Pearson’s r = −0.25, P = 8.9 × 10−5; Fig. S4A). This pattern is particularly pronounced when looking within certain groups, including Primates, Eulipotyphla, and Muridae rodents (Fig. S4B–D), and fits with a model of gBGC, where chromosome length is inversely related to recombination rate and, therefore, the effect of gBGC45.
a Neighbour-joining tree of 240 mammals based on genetic distances (sequence dissimilarity) calculated from 93,464 shared 4d sites. Significant negative correlations are seen between GC4 content at highly conserved sites and b historical Ne (two-tailed Pearson’s r = −0.48, P = 5.3 × 10−13) and c mean genetic distance of a species to all other species (two-tailed Pearson’s r = −0.87, P < 2.2 × 10−16). Indicated in (b), the edible dormouse (Glis glis, Order: Rodentia) has the lowest Ne and highest GC4 content of rodents, the grey whale (Eschrichtius robustus, Order: Artiodactyla) has the highest GC4 content of all species, and the Indochinese shrew (Crocidura indochinensis, Order: Eulipotyphla) has the greatest mean distance to all other species, the lowest GC content at conserved 4d sites, and the third highest Ne. Genome young TE (yTE) content (a proxy for unwanted transcript burden) negatively correlates with historical Ne (d two-tailed Pearson’s r = −0.20, P = 0.0034) and positively correlates with GC content at 4d sites (e two-tailed Pearson’s r = 0.16, P = 0.016). The two orders with the strongest correlations between yTE content and GC4 content are Rodentia (f n = 53, two-tailed Pearson’s r = 0.29, P = 0.035) and Chiroptera (g n = 30, two-tailed Pearson’s r = 0.40, P = 0.031). Trend lines in (b–g) were calculated using a linear model with standard error intervals around the mean shown in grey. Species silhouettes in (f) and (g) were obtained from phylopic.org, which are available under a creative commons licence. Source data are provided as a Source Data file.
Genome transposable element content may indicate (1) the efficiency of selection to purge unwanted/non-native sequences and (2) the burden of unwanted transcripts on a genome. A lower effective population size may result in higher genome TE content35 (although see ref. 57), which may subsequently lead to stronger selection on synonymous site GC content, as the problem of unwanted transcripts is elevated32. Despite this prediction of the UTH, we do not see a significant negative correlation between total genome TE content and Ne (Pearson’s r = −0.10, P = 0.14). However, when looking only at young TEs (based on low divergence from consensus sequences58), which may be more indicative of the burden of active TE-induced unwanted transcripts in a genome, the correlation is significant (Fig. 5d; Pearson’s r = −0.20, P = 0.0034). There is also a significant positive correlation between genome young TE content and GC4 content when considering all 4d sites (Fig. 5e; Pearson’s r = 0.16, P = 0.016). We also looked for associations within orders, as levels of young TE content vary considerably among and within mammalian orders58. Both Rodentia (Fig. 5f; n = 53, Pearson’s r = 0.29, P = 0.035) and Chiroptera (Fig. 5g; n = 30, Pearson’s r = 0.40, P = 0.031) show significant positive correlations between young TE genome content and GC4 content, as predicted under the UTH.
Distribution of conserved 4d sites among genes
Across the 17,394 transcripts within which we identified high-confidence 4d sites shared among mammals, 16,860 (96.9%) contain at least one and 12,257 transcripts (70.5%) have 10 or more conserved 4d sites. Excluding splice-related sites within 3 bp of exon boundaries, there are still 11,832 transcripts (68.0%) with 10 or more conserved 4d sites. The top 1% most highly conserved 4d sites (phyloP ≥7.13, n = 26,282, GC4 = 99%) are distributed over 7893 transcripts (45.4%), with 4475 transcripts (25.7%) having at least two and 503 transcripts (2.9%) at least 10 highly conserved 4d sites. The gene containing the topmost 1% highly conserved 4d sites is TNRC6B (Trinucleotide Repeat Containing Adaptor 6B) with 113 (13.1% of its 865 4d sites). Within this gene, 60% of its 4d sites are significantly conserved (n = 519, mean phyloP = 5.63, GC4 = 54.9%), with 72 sites that are fixed for the same base among all aligned genomes (GC4 = 61%). We assessed enrichment for conservation by comparing observed and expected rates of conserved 4d sites (Fig. 6a). There are 7197 transcripts (43.4%) with observed/expected rates greater than 1, demonstrating that 4d site conservation is not distributed evenly among transcripts. The top 95th percentile of conservation enrichment (observed/expected rate >1.98) contains 853 transcripts, and 166 transcripts are in the top 99th percentile (observed/expected rate >2.83; Supplementary Data 2). Gene ontology analysis revealed strong enrichment for functions relating to development and transcriptional regulation (Supplementary Data 3). The gene with the highest enrichment for 4d site conservation was IGIP, a single-exon gene coding for immunoglobulin A inducing protein (19/19 4d sites conserved, GC4 = 58%). Sixteen out of the 39 mammalian Homeobox (HOX) genes (41%) are within this top 99th percentile of 4d site conservation enrichment, and 24/39 (62%) are within the top 95th percentile.
a We calculated enrichment for conserved 4d sites by comparing the observed to the expected rate of conserved sites within transcripts (n = 16,602 transcripts). Transcripts with ratios >1 show evidence for enrichment. The top 95th and 99th percentiles of enrichment are indicated (orange and blue). b Enrichment was also calculated for conserved 4d sites within CpGs for each transcript with at least five 4d sites within CpGs (n = 11,428 transcripts). c Enrichment of conserved 4d sites (excluding CpG sites) and conserved 4d CpG sites within transcripts positively correlates (two-tailed Pearson’s r = 0.42, P < 2.2 × 10−16). The correlation is even stronger when considering only the homeobox group of developmental transcription factors (n = 41 transcripts; two-tailed Pearson’s r = 0.78, P < 1.9 × 10−9). Smoothed trend lines were generated using the ‘geom_smooth’ function in ggplot2 with the ‘gam’ method with 95% confidence interval around mean values in cyan. d Enrichment of conserved 4d sites in the binding sites of human transcription factors, measured using ENCODE 3 transcription factor ChIP-seq peaks (n = 326 transcription factors). Binding sites of Polycomb group (PcG) transcription factors EZH2, SUZ12, and RNF2, as well as RNA Polymerase II Subunit A (POLR2A), show strong enrichment for conserved 4d sites. Dashed lines represent the top 95th (grey) and 99th (red) percentiles of enrichment. e Enrichment of conserved 4d sites positively correlates with overlap of PcG protein-binding sites in transcripts (n = 16,617 transcripts; two-tailed Pearson’s r = 0.24, P < 2.2 × 10−16). Genes are coloured by DNA methylation classes (from Mas-Ponte and Supek, 2024), where genes in the same class share similar local DNA hypomethylation profiles. The correlation in (e) is mostly driven by genes in methylation class c5. Genes in this class show hypomethylation along the whole length of the gene. f The correlation is strongest amongst Homeobox genes, nearly all of which are in methylation cluster c5 (n = 41 tranascripts; two-tailed Pearson’s r = 0.68, P = 7.3 × 10−7). The smoothed trend line was generated using the ‘geom_smooth’ function in ggplot2 with the ‘loess’ method with 95% confidence interval around mean values in grey. g Enrichment of conserved 4d sites within genes relates to hypomethylation, with genes in c5 having significantly greater enrichment of conserved 4d sites than genes in other methylation classes (n = 15,170 transcripts; ANOVA, d.f = 4, F = 162.4, P < 2.2 × 10−16). This gene class also contains 89.7% of genes enriched for PcG protein-binding sites (red). Boxes represent first and third quartiles with median line, whiskers extend ±1.5× IQR. Source data are provided as a Source Data file.
CpGs are generally depleted in mammalian genomes due to the hypermutability of methylated CpGs, except for CpG islands in the promoter regions of some genes59. We observe that 10% of all 4d sites form CpGs in the human genome. Conservation at 4d sites in human CpGs is significantly less than at other 4d sites (n = 265,111 and 2,357,007, mean phyloP = −3.98 and 0.28 at 4d sites in and not in CpGs respectively; two-sample t test, t = 441.23, d.f. = 293,603, P < 2.2 × 10−16). This likely reflects the hypermutability of CpG sites and transcripts with high CpG content are known to be suppressed by zinc-finger antiviral protein (ZAP)60. However, 14,195 4d sites in CpGs show significant conservation (2.6% of conserved sites), which likely relates to selection acting on epigenetic regulation and repression of spurious intra-gene body transcription of certain genes via methylation20,61.
We calculated the enrichment of conserved 4d CpG sites within transcripts and observed a subset of genes enriched for high 4d conservation in CpGs (Fig. 6b). The top 99th percentile (observed/expected ratio >9.00) contains 115 genes (Supplementary Data 2), 41 of which are also in the top 99th percentile of genes enriched for 4d conservation generally. Gene ontology enrichment analysis of the top 99th percentile of CpG-enriched genes revealed enrichment for similar functions to the general 4d site conservation set, including developmental processes and transcriptional regulation (Supplementary Data 4). The gene EPC2 (enhancer of polycomb homologue 2) had the highest enrichment with 8 out of 9 4d sites in CpGs conserved (enrichment score = 16.6). EPC2, a highly conserved Polycomb group (PcG) protein, is predicted to play an important role in the regulation of chromatin structure in eukaryotes62. The PcG family of proteins are involved in many cellular memory processes, including cell fate decisions during embryogenesis and maintaining the transcriptional repressive state of genes over successive cell generations via H3K27me3 histone modifications63,64,65. The gene CNOT2 (CCR4-NOT transcription complex subunit 2) also shows high enrichment for conserved CpG sites (enrichment score = 16.3). The CCR4-NOT complex interacts with TNRC6B, the gene with the most strongly conserved 4d sites, to regulate mRNA synthesis and degradation66. HOX genes also show high prevalence of conserved 4d CpG sites, with 14 in the top 99th percentile of enriched genes. For example, HOXC9 has 54 4d sites within CpGs, 40 of which are conserved.
Conservation at 4d sites relates to PcG binding and methylation
There is a positive correlation between enrichment for conserved CpGs at 4d sites and enrichment for conserved 4d sites (after removing CpG sites) in transcripts (Pearson’s r = 0.42, P < 2.2 × 10−16; Fig. 6c). This pattern is more pronounced when considering only HOX genes (Pearson’s r = 0.78, P = 1.9 × 10−9; Fig. 6c). This is expected under the UTH if selection on synonymous sites modulates RNA structure to physically ‘hide’ CpG dinucleotides in native transcripts from anti-CpG quality control measures such as ZAP32,60. In addition, it has recently been shown that within-gene mutation rate heterogeneity can be explained by methylation levels, with hypomethylated regions also being hypomutated30. In particular, genes enriched with the Polycomb silencing histone mark H3K27me3, many of which are developmental genes including HOX genes, are characterised by hypomethylation and hypomutation throughout their gene bodies30. This made us consider the possibility that mutation rate variation within and among genes may explain some of the variation in 4d site conservation, potentially explaining the positive correlations we see between enrichment for conserved 4d sites and conserved 4d CpG sites, as well as between phyloP at 4d GC sites and 4d AT sites within genes (Fig. 1g).
To assess a potential association between 4d site conservation and mutation rate variation, we first looked to see whether genes targeted by PcG transcriptional repressors (hallmarks of hypomethylated genes30,67,68) are enriched for conserved 4d sites. We measured overlap between 4d sites and ENCODE 3 TF ChIP-seq clusters69 representing 340 TFs from 129 cell types and calculated an enrichment score based on the number of overlaps between binding sites and conserved 4d sites for each TF. We found that the majority of TFs show no enrichment for binding to conserved 4d sites, with 60% of the TFs having an enrichment score below 0.01 and 95% an enrichment score below 0.59 (where 0 = no enrichment; Fig. 6D). However, four genes showed strong enrichment for conserved 4d sites in their TFBS ChIP signals, (99th percentile, enrichment score >2.1), three of which are PcG genes (EZH2, enrichment score = 4.75; SUZ12, enrichment score = 3.78; and RNF2, enrichment score = 2.36), as well as POLR2A (enrichment score = 2.24), revealing an association between PcG protein binding and 4d conservation. To test this further and identify genes where the association between PcG protein binding and 4d conservation is strongest, we then calculated the number of PcG-binding sites per conserved 4d sites for all genes and found a significant positive correlation with gene enrichment of conserved 4d sites (Pearson’s r = 0.24, P < 2.2 × 10−16; Fig. 6e). Despite this correlation, there is a subset of genes strongly enriched for conserved 4d sites that show no or very little binding by PcG proteins, demonstrating that high 4d site conservation is not only found in genes targeted by PcG proteins. Genes in the top 99th percentile of conserved 4d site-PcG protein-binding site overlap (n = 168 genes; Supplementary Data 5) are enriched for GO terms relating to transcriptional regulation and embryonic morphogenesis and development (Supplementary Data 6). Genes absent of binding sites for the three polycomb group proteins but enriched for conserved 4d sites (n = 109, Supplementary Data 7) are enriched for GO terms relating to the regulation of gene expression and metabolic processes (Supplementary Data 8).
We next considered a more comprehensive classification of gene methylation by using the five gene methylation classes (c1–c5) recently developed by Mas-Ponte & Supek30, where all genes within a class share similar methylation profiles. Briefly, genes in c1 have methylated promoters, low expression and a depletion in active transcription chromatin states, c2 genes are similar to c1 but show some promoter hypomethylation as well as active chromatin states, c3 and c4 are active, expressed genes with wider hypomethylation including the promoter and extending ~2 kb of the TSS. Genes in c5 have overall less methylated gene bodies across their whole length, overlap repressed TSS and enhancer states, and are enriched for the Polycomb silencing histone mark H3K27me330. Using these methylation classes, we saw that the majority of genes driving the positive correlation between gene enrichment of conserved 4d sites and the number of PcG-binding sites per conserved 4d site are hypomethylated genes in class c5 (Fig. 6e). For HOX genes, which are predominantly in the c5 methylation class, this positive correlation is even stronger (Pearson’s r = 0.68, P = 7.3 × 10-7); Fig. 6f). Interestingly, HOXB1, which does not contain any PcG protein-binding sites and is the only HOX gene in methylation class c1, has one of the lowest conserved 4d site enrichment scores among the HOX genes (score = 0.87).
We compared the conserved 4d site enrichment scores between genes in each of the five classes and found significantly greater enrichment for conserved 4d sites in gene classes with more extensive gene hypomethylation (increasing from c1 to c5; ANOVA, d.f. = 4, F = 162.4, P < 2.2 × 10−16; Fig. 6g). Genes in the c5 class, which have the most extensive hypomethylation, are more enriched for conserved 4d sites than all other classes (Tukey’s HSD, P = 5.5 × 10−10 in all four comparisons) and this class contains almost all (89.7%) of the genes enriched for PcG-binding sites.
Enrichment of conserved 4d CpG sites relates to gene expression
The UTH predicts that highly expressed genes may show higher intra-gene body methylation to suppress the generation of unwanted transcripts. To address this, we looked at whether genes enriched for conserved 4d CpGs—the main sites of methylation—show greater expression than other genes. Mean gene expression levels from Cardoso-Moreira et al.53, of genes enriched for CpG conservation are significantly higher than in other genes, both within and across the seven organs in the dataset (Fig. S5; mean RPKM expression across all organs = 11.6 and 24.3, s.d. = 258 and 81.5 for genes lacking CpG enrichment and genes in the 99th percentile of CpG conservation enrichment respectively; ANOVA, d.f. = 3, F = 24.44, P = 8.28 × 10−16; Tukey’s HSD P = 2.45 × 10−12). Conserved CpG-enriched genes are also transcribed in significantly more cells in the mouse organogenesis cell atlas70 (Fig. S6; mean cell counts = 69,915 and 178,020 for genes lacking CpG enrichment and conserved CpG-enriched genes respectively; ANOVA, d.f. = 2, F = 195.9, P = <2.2 × 10−16; Tukey’s HSD P = 7.65 × 10−9). This supports the predictions of the UTH, where intra-gene body methylation in highly expressed genes in regions of open chromatin can act to suppress the generation of unwanted transcripts61 and suggests selection acting on these CpG sites.
Conservation at 4d sites relate to overlapping regulatory features
We performed a general linear model between 4d site phyloP and 11 potential constraining features, as well as other potential correlates (background GC content, synonymous site mutation rate; Table 2) for all 4d sites with a phyloP ≥0 (n = 1,663,871) to test how much of the observed conservation can be explained by these features. Sites with negative phyloP scores were excluded to test only for predictors of sequence conservation and not acceleration. The model explained 9.7% of the variation in phyloP and all but one variable (overlap with microRNA (miRNA) binding sites, P = 0.20) significantly explained some of the variance at P < 0.05. An insignificant effect of miRNA binding is not surprising given that it has previously been shown that only a very low proportion of synonymous site conservation is attributable to miRNA pairing71. The variable that explained most of the variance, with a relative importance of 85.5%, was GC proportion across mammals. Distance to the exon boundary (2.7% relative importance) and the background GC content (2.1% relative importance) were the next most important variables in the model. The remaining ten variables had a combined relative importance of 10.1%, suggesting that a significant but limited amount of the conservation observed as synonymous sites across mammals can be explained by being within conserved regulatory features such as transcription factor binding sites, candidate cis-regulatory elements (cCREs), RNA-binding protein (RBP) binding sites, or exonic splicing enhancers (ESEs).
Among conserved sites, we used a similar general linear model to test whether feature overlap related to gene expression, potentially revealing differences in the selection pressures acting on synonymous sites in lowly and highly expressed genes. Whilst the model was significant (Supplementary Data 9), it had an r2 of only 0.038, showing that very little of the variation in feature overlap relates to gene expression (Fig. S7A). Synonymous sites in lowly expressed genes have a much lower overlap with RBP-binding sites (relative importance of 97.5% in the model) compared to those in more highly expressed genes (Fig. S7B). However, very little of the conservation at 4d sites is explained by overlap with RBP-binding sites (Table 2) and this result likely reflects that binding of RBPs, which are involved in transcript processing17, relates to expression levels independent of constraint. The model therefore did not reveal any evidence that selection pressures acting on 4d sites differ between highly and lowly expressed genes.
Discussion
We have carried out a comprehensive characterisation of conservation at synonymous sites in mammalian genomes using the Zoonomia 240-species genome alignment and single-base resolution phyloP scores42, providing the most complete picture to date of synonymous site conservation over the 100 million years of placental mammal evolution. All species except the hispid cotton rat (Sigmodon hispidus) show a GC bias at 4d sites across their genomes. We show that conservation at 4d sites strongly relates to GC content, with the vast majority (~80%) of highly conserved 4d sites in mammals being G or C. We demonstrate that the GC4 content of human transcripts correlates strongly with that across mammals generally, suggesting that similar processes have shaped synonymous site base content across mammals. Whilst high sequence similarity across species is often used to imply purifying selection and evolutionary constraint42,72, neutral processes such as gBGC and mutation rate variation can also lead to high sequence conservation. In this discussion, we will consider the evidence we find for selection acting on 4d sites against neutral processes, and place our findings in light of the recently presented UTH.
Codon usage bias is a ubiquitous phenomenon in nature and mammals share similar codon usage frequencies with a bias towards GC at codon third positions73,74. One hypothesis proposed to explain this is that the ancestral vertebrate genome was GC-rich, which would have created a prevalence of GC-ending codons early in vertebrate evolution21. Substantial increases in synonymous site GC content in certain mammalian lineages over the last 100 million years have been related to differences in the strength of gBGC among species45. As the impact or extent of gBGC in a genome is the product of the effective population size by the recombination rate by the repair bias26, larger effective population sizes and smaller chromosomes are predictors of a higher equilibrium GC content and a faster increase in GC content under the gBGC model45,55. Our findings fit with these expectations, where we see variable GC4 content among the mammal genomes that relates positively with historical Ne in some orders, and negatively with genome size (as a proxy for chromosome size) overall.
Our findings also support the gBGC model in that we observe a positive correlation between GC content of 4d sites and neighbouring non-coding sequences. Many genes have GC4 contents that are considerably higher than their neighbouring sequences and 4d site conservation does not relate to local GC content, which may be more suggestive of selection than gBGC. However, non-coding neighbouring sequences are more vulnerable to insertions by AT-rich TEs, which over time can result in a differential between the GC content of coding and surrounding non-coding sequences52. In addition, we see an excess of mutations at conserved CG 4d sites in humans, which is suggestive of a lack of purifying selection on these sites and fits with the gBGC model24.
Whilst our findings suggest that high ancestral GC content and the gBGC model can explain much of the GC4 content observed in mammals, in agreement with previous studies24,26,31,52, there are several observations we make that cannot be so easily explained by gBGC. In our dataset, there are negative correlations between 4d site conservation and transcript GC4 as well as local GC content. This suggests that conservation is lower in regions more recently affected by gBGC, likely due to the fact that recombination landscapes vary greatly across the genomes of different species, with recombination hotspots evolving rapidly75. For example, hotspot conservation is low between humans and chimpanzees76 and local recombination rates correlate only very weakly between strains of mice77. We show that 4d site conservation is strongest in genes whose proteins are under high constraint in mammals and therefore have experienced strong purifying selection throughout mammalian evolution. Repeated exposure to gBGC throughout the histories of synonymous sites in these ancient genes could then lead to elevated and conserved GC4 content40. However, whilst this may be true for some genes, we show that transcript GC4 content is not a good predictor of 4d site conservation, where we observe a whole spectrum of GC4 proportions among the genes with high conservation at their 4d sites. In addition, we observe a positive correlation between 4d phyloP at AT and GC sites in transcripts, which would not be expected if conservation was driven by gBGC. An explanation of transcription-coupled repair leading to higher GC4 content is also not supported by our results, as genes with high GC4 content and conservation actually show lower levels of expression than other genes.
We find several lines of evidence that conservation at synonymous sites may relate to unwanted transcript mitigation, lending support to the UTH. The strong GC bias observed at conserved 4d sites may help ‘flag’ transcripts as native32. Higher GC content of 4d sites in single-exon genes and first exons of multi-exon genes, a pattern previously shown in humans37 and shown here to be true across mammals, is also predicted by the UTH to differentiate native from unwanted transcripts. We also show strong evidence for high constraint at 4d sites located at splice sites, essential for ensuring the production of accurate transcripts37,78,79,80. Whilst 4d sites at exon boundaries are generally highly conserved, they explain <3% of the overall observed 4d conservation. We find evidence that secondary functions, such as 4d sites being located within TFBSs, uORFs, RBP-binding sites, and ESEs, can explain only a limited amount of the observed 4d site conservation.
A recent analysis of selection against point mutations in the human genome found no evidence for ultraselection (complete absence of point mutations) at 4d sites but that 39% experience weak purifying selection79. Our analysis of human genetic variation at 4d sites also provides evidence of purifying selection acting on some sites. Human variation at 4d sites correlates negatively with phyloP, and the 21% of 4d sites identified as under significant conservation show significantly less variation than other 4d sites, both in terms of the proportion of sites containing variants and the allele frequencies at variable sites. Whilst there is an excess of variants at conserved GC 4d sites, which is expected under the gBGC model24, there is a deficiency of variants at conserved AT 4d sites, suggestive of purifying selection. Variants at synonymous sites have also been shown to be rare close to exon ends, likely due to disruption of ESEs81.
The UTH predicts that spurious transcription should be a greater issue for species with low Ne as selection is not efficient enough to remove weakly deleterious mutations that lead to spurious transcripts32. Whilst we did not see a relationship between GC4 content and Ne across all species, and a positive correlation within some orders as predicted by the gBGC model, we did find that species with lower Ne generally have higher GC content at 4d sites present across all 240 genomes. Under the UTH, this may be interpreted as a higher constraint on GC4 content at these sites in mammals with lower Ne to mitigate spurious transcription32. In addition, it may reflect faster evolutionary rates in species with high Ne and more efficient selection on codon use, although this is not widely observed in mammals45,82. We also saw that GC4 content positively correlates with the young TE content of a genome, which can be a source of spurious transcription, with elevated GC4 content in species with a potentially greater unwanted transcript load. This is despite the fact that young TE content correlates positively with genome size, which has a negative association with GC4 content. This is most strongly observed in bats and rodents, where variation in young TE content is greatest. Selection for error-mitigating properties in the error-prone genomes of species with small Ne has been previously suggested as an explanation for the observation of an association between Ne and intronic content and splice site usage across species83. Our findings support this hypothesis further, where higher GC4 content may act to limit the impact of spurious transcription and improve the detection of native transcripts amongst profuse unwanted transcripts32.
Recently, it has been shown that mutation rate heterogeneity at the sub-gene scale in humans is caused by methylation variability, with hypomethylated regions having low mutation rates30. Reduced mutation rates in certain genes and gene regions will lead to elevated conservation over time, which can be reflected in synonymous site divergence. Such mutation rate heterogeneity may partly explain the positive correlation we see between phyloP at GC and AT 4d sites within transcripts, as well as the negative correlation between phyloP and genetic variability in humans. It is known that certain genes and gene regions show strong hypomethylation (so-called ‘DNA methylation valleys’84), many of which are developmental genes associated with the Polycomb repressive mark H3K27me330,67. Mas-Ponte and Supek have now demonstrated that these genes experience lower mutation rates due to their hypomethylation30. We have shown that many genes enriched for conserved 4d sites and CpG sites are also hypomethylated developmental genes targeted by PcG proteins. These genes encode amongst the most conserved mammalian proteins under strong evolutionary constraint42, including HOX genes whose highly orchestrated expression regulates embryo development along the anterior-posterior axis85. We suggest that the high 4d site conservation in these genes likely relates to the lower mutation rates they have experienced throughout mammalian evolution compared to other genes.
DNA methylation is an important mechanism regulating the expression of developmental genes, such as HOX genes, located within Polycomb-associated DNA methylation valleys (reviewed in ref. 86). Sites of methylation in these genes, including CpG sites, are therefore likely to be under strong selection to ensure their accurate epigenetic regulation in time and space20. We show that many developmental genes are enriched for conserved CpG sites, despite the hypermutability of methylated CpGs leading to low CpG content of native transcripts generally. An enrichment of conserved CpG sites in the exons of HOX and other developmental genes leading to intragenic methylation20 could serve to protect gene bodies from rogue entry of RNA polymerase II, which can lead to spurious expression61,87. Irregular methylation of these genes, often related to PcG gene misregulation and altered H3K27me3 deposition88, has been associated with developmental disorders68,89,90, cancer86,88 and ageing91,92,93. A potential association between synonymous site constraint and ageing is intriguing; CpG methylation at or near HOX genes is a strong predictor of longevity in mammals91 and changes within the mammalian chromatin landscape associate with aberrant transcription initiation inside genes during senescence and ageing87. Higher conservation of 4d sites in the developmental genes of larger, longer-lived mammals suggested in our data may reflect higher constraint on epigenetic regulatory features, including intragenic methylation. Whether this can facilitate longer life by reducing the effects of spurious transcription makes for an intriguing future line of research.
In conclusion, we show that synonymous site conservation in mammals is extensive and has likely arisen through a combination of several non-mutually exclusive processes, including gBGC, mutation rate variation, and selection. Highly constrained, integral genes involved in developmental processes and transcriptional regulation are enriched for conserved 4d sites, likely resulting from a combination of experiencing low mutation rates and strong selection on 4d sites relating to tight epigenetic regulation. We present some evidence in support of the prediction of the unwanted transcript hypothesis that species with small Ne and high spurious transcript burden have greater GC4 content, which may help to differentiate native from unwanted, spurious transcripts.
Methods
Genome alignment and phyloP scores
We used the 241-way multi-genome Cactus94 alignment of placental mammal genomes and human-referenced (Hg38) phyloP scores generated as part of the Zoonomia project42. PhyloP scores were generated across all autosomes using the PHAST package44,95 with inputs of the Cactus alignment and ancestral repeats as a model of neutrally evolving sequences. PhyloP scores were generated at single-base resolution and ranged in value from −20 to 8.9. Positions showing significant conservation were identified by converting phyloP values to q values using a FDR correction. Any positions with positive phyloP scores and a q ≤ 0.05 were considered to be significantly conserved. This gave a significance threshold of phyloP ≥2.27 at 0.05 FDR. Further details can be found in the original paper42. PhyloP scores are available on the UCSC genome browser.
Transcript selection and identification of mammalian 4d sites
Gene annotations for Hg38 were obtained from GENCODE release 3943. As most genes contain multiple transcripts, we chose an approach to select one representative transcript per protein-coding gene to simplify analyses. We selected representative transcripts using the Matched annotation from NCBI and EMBL-EBI set of representative transcripts (MANE Select)96. If unavailable here, then we selected the canonical transcript used by gnomAD97 or BUSCO98. We used BEDtools v.2.30.0 intersect99 to extract phyloP scores for the CDS of each transcript from the single-base phyloP score files. The degeneracy of each site within CDS was assessed by looking up each codon in a DNA codon table using a custom Perl script, revealing a total of 5,278,470 4d sites. Many of these are unlikely to be 4d sites in other mammal genomes due to, for example, differences in the gene content among genomes. We therefore extracted alignments for all codons from the Zoonomia alignment and identified codons where >85% of species were aligned and >95% of the aligned codons contained a 4d site (i.e. each 4d site is present in at least 194 species). This resulted in a set of 2,621,118 4d sites distributed over 17,394 transcripts that formed the basis of our analysis. Base content per 4d site per species were then extracted from the alignment and summarised using custom Perl scripts. Whether a fourfold degenerate site forms a CpG site was determined from the human reference genome using a custom Perl script which identified all 4d Cs followed by a G and all 4d Gs preceded by a C. Exon edge sites, exon counts per transcript, and single- exon genes were identified from the GENCODE annotation. GC content in 1 Mb windows was measured across the human reference genome (Hg38) using a custom Perl script with a bed file as output. We then used the ‘intersect’ tool of BEDTools v.2.30.099 between this file and a bed file of transcripts to obtain a local GC percentage for each transcript.
Human population variation at 4d sites
We used the NHLBI Trans-omics for precision medicine (TOPMed) data freeze 8 (https://topmed.nhlbi.nih.gov/) containing whole genome sequencing of 138,922 individuals to assess human genetic variation at 4d sites. We used the ‘intersect’ tool of BEDTools v.2.30.0 to intersect the TOPMed VCF file and our 4d site bed file. At each 4d site with variants, we ascertained whether or not the allele in the human reference genome (Hg38) matched the major allele in the human population using a custom Perl script. This allowed us to identify variants that were specific to Hg38 and ensure that we were considering the most common allele when looking at base content of human 4d sites.
Effective population sizes, transposable elements, and divergence at 4d sites
Estimates of historical Ne for 210 of the Zoonomia species were obtained from Wilder et al.54, where they used PSMC100 to estimate Ne based on heterozygous positions in each genome. Data on genome transposable element content was obtained from Osmanski et al.58, where they annotated the TE content of all Zoonomia genomes including identifying young insertions as TEs with sequences with Kimura two-parameter genetic distances <4% compared to their consensus58. Genome size was taken as the total assembly size of each genome in the alignment. To analyse divergence at highly conserved 4d sites we first identified a set of sites where all 240 species were aligned, giving 93,464 4d sites with no missing data. We then created concatenated pseudosequences in fasta format for each species from the alignment using a custom perl script. A single fasta format file containing these pseudosequences for all species was then read into R using the package Biostrings (https://doi.org/10.18129/B9.bioc.Biostrings). We then calculated a genetic distance matrix from the sequences using the R package DECIPHER (https://doi.org/10.18129/B9.bioc.DECIPHER), where each value in the matrix is the dissimilarity between two sequences. A neighbour-joining tree was then generated based on these distances using the Ape package in R. The tree was visualised using ggtree in R (https://doi.org/10.18129/B9.bioc.ggtree).
Conservation enrichment
For each transcript, we calculated the observed versus expected proportion of conserved 4d sites, conserved 4d sites in CpGs, and TFBS overlap with conserved 4d sites. In each case, the expected proportion was calculated by 1/(total number of events/number of conserved events) (i.e. total 4d sites/total conserved 4d sites, total 4d CpG sites/total conserved 4d CpG sites, total TFBS-4d site overlaps/total TFBS-conserved 4d site overlaps). For each transcript, the observed rate was calculated by 1/(total events/conserved events). We then divided the observed by the expected proportion to get an enrichment score. We used the quantile function in R to identify highly enriched genes in each case, taking the top 95th and 99th quantiles. Gene ontology enrichment analysis was run on gene sets enriched for 4d conserved, 4d CpG conserved, and with conserved 4d sites enriched in TF binding sites using PANTHER101 with a Bonferroni correction for multiple testing. To assess whether gene expression relates to conservation, we made use of three gene expression datasets: the GTEx gene expression data for 54 human tissues102 (https://gtexportal.org/home/), the Evo-Devo mammalian organs dataset53 (https://apps.kaessmannlab.org/evodevoapp/), and single-cell expression data from the mouse organogenesis cell atlas70 (https://oncoscape.v3.sttrcancer.org/atlas.gs.washington.edu.mouse.rna/landing). We tested for associations between mean RPKM expression scores and levels of conservation- and conserved CpG enrichment using analysis of variance tests, as well as for associations between conserved CpG enrichment and the number of cells expressed in using the mouse organogenesis cell atlas.
Functional annotation datasets
Various datasets were used to look for relationships between 4d site conservation and functional annotations. All human cCREs were obtained from ENCODE SCREEN. Per-gene synonymous site mutation rates were obtained from supplementary dataset 11 in the gnomAD paper97. miRNA targets were obtained from TargetScanHuman release 8.0 (https://www.targetscan.org/vert_80/) and converted from Hg19 to Hg38 coordinates using the UCSC liftover tool (https://genome.ucsc.edu/cgi-bin/hgLiftOver). uORF coordinates were retrieved from http://github.gersteinlab.org/uORFs/18 and converted to Hg38 using liftover. RNA-binding protein-binding sites were obtained from ENCODE eCLIP signals detailed in ref. 103 and available from https://www.encodeproject.org/. Transcription factor binding sites were obtained from ENCODE 3 TF ChIP-seq datasets here: https://hgdownload.soe.ucsc.edu/goldenPath/hg38/encRegTfbsClustered/. Assignment of human genes to five distinct methylation classes was obtained from Mas-Ponte & Supek30. Hg38 coordinates of constrained TFBSs were provided by the authors of42. For identifying putative exonic splicing enhancers we downloaded the 238 hexamers identified as candidate ESEs in humans104 from http://hollywood.mit.edu/burgelab/rescue-ese/ as well as the INT2 intersect set of 316 ESE motifs from Cáceres and Hurst (2013) and removed redundancy from the two sets. We then used a custom Perl script to check whether each 4d site was inside one of the hexamers to define whether a 4d site was inside or outside ESEs. We included only ESEs if they were within 70 bp of exon ends where they are most likely to be involved in splicing and therefore under constraint13. Indeed, we do see higher 4d site conservation in ESE motifs within 70 bp of exon ends compared to elsewhere in an exon (mean (sd) phyloP = 0.26 (3.40) and −0.19 (3.58) for 4d sites in ESEs within 70 bp of exon ends and elsewhere respectively; Welch two-sample t test, t = −29.68, d.f. = 130937, P < 2.2 × 10−16).
Statistical analysis
All statistical analysis was carried out in R v.4.3.1 “Beagle Scouts”. We used the ‘cor.test’ function to calculate Pearson’s r for correlation analyses. We performed two-tailed t tests (two groups) or ANOVA (>2 groups) to test for differences between means using the ‘t.test’ and ‘aov’ functions, respectively. Post hoc Tukey’s honestly significant difference tests were used to identify significant differences following ANOVA. A Chi2 test was used to test for differences in observed versus expected proportions of variant types at conserved 4d sites using the ‘chisq.test’ function. We used the ‘glm’ function to perform a general linear model to test for associations between 4d site phyloP and multiple factors that may influence sequence conservation. For all conserved 4d sites, we used a GLM to test whether differences in gene expression (calculated as mean expression over all tissues and developmental stages from the Cardoso-Moreira et al.53, dataset) relate to 4d site overlap with regulatory features.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
The Cactus 241-way alignment and phyloP constraint scores are publicly available at https://cglgenomics.ucsc.edu/data/cactus/ and at https://genome.ucsc.edu/cgi-bin/hgTrackUi?db=hg38&g=cons241way. Mammalian 4d sites that formed the basis of the analysis are provided in supplementary data file 1 in Hg38-referenced bed file format. An R data object containing all data and associated scripts used to analyse the data is available from figshare via the following https://doi.org/10.6084/m9.figshare.26318410. Source data are provided with this paper.
Code availability
Scripts detailing all of the analyses are available from figshare via the following https://doi.org/10.6084/m9.figshare.26318410.
References
Kimura, M. Preponderance of synonymous changes as evidence for the neutral theory of molecular evolution. Nature 267, 275–276 (1977).
Sharp, P. M., Emery, L. R. & Zeng, K. Forces that influence the evolution of codon bias. Philos. Trans. R. Soc. B Biol. Sci. 365, 1203–1212 (2010).
Ikemura, T. Codon usage and tRNA content in unicellular and multicellular organisms. Mol. Biol. Evol. 2, 13–34 (1985).
Sharp, P. M. et al. DNA sequence evolution: the sounds of silence. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 349, 241–247 (1997).
Duret, L. Evolution of synonymous codon usage in metazoans. Curr. Opin. Genet. Dev. 12, 640–649 (2002).
Kanaya, S., Yamada, Y., Kinouchi, M., Kudo, Y. & Ikemura, T. Codon usage and tRNA genes in eukaryotes: correlation of codon usage diversity with translation efficiency and with CG-dinucleotide usage as assessed by multivariate analysis. J. Mol. Evol. 53, 290–298 (2001).
Resch, A. M. et al. Widespread positive selection in synonymous sites of mammalian genes. Mol. Biol. Evol. 24, 1821–1831 (2007).
Eőry, L., Halligan, D. L. & Keightley, P. D. Distributions of selectively constrained sites and deleterious mutation rates in the hominid and murid genomes. Mol. Biol. Evol. 27, 177–192 (2010).
Dhindsa, R. S., Copeland, B. R., Mustoe, A. M. & Goldstein, D. B. Natural selection shapes codon usage in the human genome. Am. J. Hum. Genet. 107, 83–95 (2020).
Buhr, F. et al. Synonymous codons direct cotranslational folding toward different protein conformations. Mol. Cell 61, 341–351 (2016).
Walsh, I. M., Bowman, M. A., Soto Santarriaga, I. F., Rodriguez, A. & Clark, P. L. Synonymous codon substitutions perturb cotranslational protein folding in vivo and impair cell fitness. Proc. Natl. Acad. Sci. USA 117, 3528–3534 (2020).
Parmley, J. L., Chamary, J. V. & Hurst, L. D. Evidence for purifying selection against synonymous mutations in mammalian exonic splicing enhancers. Mol. Biol. Evol. 23, 301–309 (2006).
Savisaar, R. & Hurst, L. D. Both maintenance and avoidance of RNA-binding protein interactions constrain coding sequence evolution. Mol. Biol. Evol. 34, 1110–1126 (2017).
Chamary, J. & Hurst, L. D. Evidence for selection on synonymous mutations affecting stability of mRNA secondary structure in mammals. Genome Biol. 6, R75 (2005).
Chamary, J. V., Parmley, J. L. & Hurst, L. D. Hearing silence: non-neutral evolution at synonymous sites in mammals. Nat. Rev. Genet. 7, 98–108 (2006).
Stergachis, A. B. et al. Exonic transcription factor binding directs codon choice and affects protein evolution. Science 342, 1367–1372 (2013).
Hentze, M. W., Castello, A., Schwarzl, T. & Preiss, T. A brave new world of RNA-binding proteins. Nat. Rev. Mol. Cell Biol. 19, 327–341 (2018).
McGillivray, P. et al. A comprehensive catalog of predicted functional upstream open reading frames in humans. Nucleic Acids Res. 46, 3326–3338 (2018).
Jones, P. A. Functions of DNA methylation: islands, start sites, gene bodies and beyond. Nat. Rev. Genet. 13, 484–492 (2012).
Branciamore, S., Chen, Z.-X., Riggs, A. D. & Rodin, S. N. CpG island clusters and pro-epigenetic selection for CpGs in protein-coding exons of HOX and other transcription factors. Proc. Natl. Acad. Sci. USA 107, 15485–15490 (2010).
Nabiyouni, M., Prakash, A. & Fedorov, A. Vertebrate codon bias indicates a highly GC-rich ancestral genome. Gene 519, 113–119 (2013).
Figuet, E., Ballenghien, M., Romiguier, J. & Galtier, N. Biased gene conversion and GC-content evolution in the coding sequences of reptiles and vertebrates. Genome Biol. Evol. 7, 240–250 (2015).
Lamb, B. C. The properties of meiotic gene conversion important in its effects on evolution. Heredity 53, 113–138 (1984).
Galtier, N., Piganeau, G., Mouchiroud, D. & Duret, L. GC-Content evolution in mammalian genomes: the biased gene conversion hypothesis. Genetics 159, 907–911 (2001).
Brown, T. C. & Jiricny, J. Repair of base–base mismatches in simian and human cells. Genome 31, 578–583 (1989).
Duret, L. & Galtier, N. Biased gene conversion and the evolution of mammalian genomic landscapes. Annu. Rev. Genomics Hum. Genet. 10, 285–311 (2009).
Ho, A. T. & Hurst, L. D. Unusual mammalian usage of TGA stop codons reveals that sequence conservation need not imply purifying selection. PLoS Biol. 20, e3001588 (2022).
Galtier, N., Duret, L., Glémin, S. & Ranwez, V. GC-biased gene conversion promotes the fixation of deleterious amino acid changes in primates. Trends Genet. 25, 1–5 (2009).
Montoya-Burgos, J. I., Boursot, P. & Galtier, N. Recombination explains isochores in mammalian genomes. Trends Genet. 19, 128–130 (2003).
Mas-Ponte, D. & Supek, F. Mutation rate heterogeneity at the sub-gene scale due to local DNA hypomethylation. Nucleic Acids Res. 52, 4393–4408 (2024).
Galtier, N. et al. Codon usage bias in animals: disentangling the effects of natural selection, effective population size, and GC-biased gene conversion. Mol. Biol. Evol. 35, 1092–1103 (2018).
Radrizzani, S., Kudla, G., Izsvák, Z. & Hurst, L. D. Selection on synonymous sites: the unwanted transcript hypothesis. Nat. Rev. Genet. https://doi.org/10.1038/s41576-023-00686-7 (2024).
Bulmer, M. The selection-mutation-drift theory of synonymous codon usage. Genetics 129, 897–907 (1991).
Dunham, I. et al. An integrated encyclopedia of DNA elements in the human genome. Nature 489, 57–74 (2012).
Lynch, M. & Conery, J. S. The ORIGINS OF GENOME COmplexity. Science 302, 1401–1404 (2003).
Sakharkar, M. K., Chow, V. T. K. & Kangueane, P. Distributions of exons and introns in the human genome. Silico Biol. 4, 387–393 (2004).
Mordstein, C. et al. Codon usage and splicing jointly influence mRNA localization. Cell Syst. 10, 351–362.e8 (2020).
Zuckerman, B., Ron, M., Mikl, M., Segal, E. & Ulitsky, I. Gene architecture and sequence composition underpin selective dependency of nuclear export of long RNAs on NXF1 and the TREX complex. Mol. Cell 79, 251–267.e6 (2020).
Kudla, G., Lipinski, L., Caffin, F., Helwak, A. & Zylicz, M. High guanine and cytosine content increases mRNA levels in mammalian cells. PLoS Biol. 4, e180 (2006).
Birdsell, J. A. Integrating genomics, bioinformatics, and classical genetics to study the effects of recombination on genome evolution. Mol. Biol. Evol. 19, 1181–1197 (2002).
Genereux, D. P. et al. A comparative genomics multitool for scientific discovery and conservation. Nature 587, 240–245 (2020).
Christmas, M. J. et al. Evolutionary constraint and innovation across hundreds of placental mammals. Science 380, eabn3943 (2023).
Harrow, J. et al. GENCODE: the reference human genome annotation for The ENCODE Project. Genome Res. 22, 1760–1774 (2012).
Siepel, A., Pollard, K. S. & Haussler, D. New methods for detecting lineage-specific selection. In Research in Computational Molecular Biology (eds. Apostolico, A. et al.) 190–205 (Springer, 2006).
Romiguier, J., Ranwez, V., Douzery, E. J. P. & Galtier, N. Contrasting GC-content dynamics across 33 mammalian genomes: relationship with life-history traits and chromosome sizes. Genome Res. 20, 1001–1009 (2010).
Taliun, D. et al. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program. Nature 590, 290–299 (2021).
Mirihana Arachchilage, G., Hetti Arachchilage, M., Venkataraman, A., Piontkivska, H. & Basu, S. Stable G-quadruplex enabling sequences are selected against by the context-dependent codon bias. Gene 696, 149–161 (2019).
Park, E., Pan, Z., Zhang, Z., Lin, L. & Xing, Y. The expanding landscape of alternative splicing variation in human populations. Am. J. Hum. Genet. 102, 11–26 (2018).
Sibley, C. R., Blazquez, L. & Ule, J. Lessons from non-canonical splicing. Nat. Rev. Genet. 17, 407–421 (2016).
Mishra, A. et al. Intron exon boundary junctions in human genome have in-built unique structural and energetic signals. Nucleic Acids Res. 49, 2674–2683 (2021).
Hanawalt, P. C. & Spivak, G. Transcription-coupled DNA repair: two decades of progress and surprises. Nat. Rev. Mol. Cell Biol. 9, 958–970 (2008).
Duret, L. & Hurst, L. D. The elevated GC content at exonic third sites is not evidence against neutralist models of isochore evolution. Mol. Biol. Evol. 18, 757–762 (2001).
Cardoso-Moreira, M. et al. Gene expression across mammalian organ development. Nature 571, 505–509 (2019).
Wilder, A. P. et al. The contribution of historical processes to contemporary extinction risk in placental mammals. Science 380, eabn5856 (2023).
Duret, L. & Arndt, P. F. The impact of recombination on nucleotide substitutions in the human genome. PLoS Genet. 4, e1000071 (2008).
Foley, N. M. et al. A genomic timescale for placental mammal evolution. Science 380, eabl8189 (2023).
Marino, A., Debaecker, G., Fiston-Lavier, A.-S., Haudry, A. & Nabholz, B. Effective population size does not explain long-term variation in genome size and transposable element content in animals. Preprint at https://doi.org/10.1101/2024.02.26.582137 (2024).
Osmanski, A. B. et al. Insights into mammalian TE diversity through the curation of 248 genome assemblies. Science 380, eabn1430 (2023).
Gardiner-Garden, M. & Frommer, M. CpG Islands in vertebrate genomes. J. Mol. Biol. 196, 261–282 (1987).
Takata, M. A. et al. CG dinucleotide suppression enables antiviral defence targeting non-self RNA. Nature 550, 124–127 (2017).
Neri, F. et al. Intragenic DNA methylation prevents spurious transcription initiation. Nature 543, 72–77 (2017).
Stankunas, K. et al. The enhancer of polycomb gene of Drosophila encodes a chromatin protein conserved in yeast and mammals. Dev. Camb. Engl. 125, 4055–4066 (1998).
Whitcomb, S. J., Basu, A., Allis, C. D. & Bernstein, E. Polycomb Group proteins: an evolutionary perspective. Trends Genet. 23, 494–502 (2007).
Kerppola, T. K. Polycomb group complexes—many combinations, many functions. Trends Cell Biol. 19, 692–704 (2009).
Di Croce, L. & Helin, K. Transcriptional regulation by Polycomb group proteins. Nat. Struct. Mol. Biol. 20, 1147–1155 (2013).
Inada, T. & Makino, S. Novel roles of the multi-functional CCR4-NOT complex in post-transcriptional regulation. Front. Genet. 5, 135 (2014).
Jeong, M. et al. Large conserved domains of low DNA methylation maintained by Dnmt3a. Nat. Genet. 46, 17–23 (2014).
Heyn, P. et al. Gain-of-function DNMT3A mutations cause microcephalic dwarfism and hypermethylation of Polycomb-regulated regions. Nat. Genet. 51, 96–105 (2019).
Davis, C. A. et al. The Encyclopedia of DNA elements (ENCODE): data portal update. Nucleic Acids Res. 46, D794–D801 (2018).
Cao, J. et al. The single-cell transcriptional landscape of mammalian organogenesis. Nature 566, 496–502 (2019).
Hurst, L. D. Preliminary assessment of the impact of microRNA-mediated regulation on coding sequence evolution in mammals. J. Mol. Evol. 63, 174–182 (2006).
Pollard, K. S., Hubisz, M. J., Rosenbloom, K. R. & Siepel, A. Detection of nonneutral substitution rates on mammalian phylogenies. Genome Res. 20, 110–121 (2010).
Subramanian, K., Payne, B., Feyertag, F. & Alvarez-Ponce, D. The codon statistics database: a database of codon usage bias. Mol. Biol. Evol. 39, msac157 (2022).
Nakamura, Y., Gojobori, T. & Ikemura, T. Codon usage tabulated from international DNA sequence databases: status for the year 2000. Nucleic Acids Res. 28, 292 (2000).
Paigen, K. & Petkov, P. Mammalian recombination hot spots: properties, control and evolution. Nat. Rev. Genet. 11, 221–233 (2010).
Ptak, S. E. et al. Fine-scale recombination patterns differ between chimpanzees and humans. Nat. Genet. 37, 429–434 (2005).
Paigen, K. et al. The recombinational anatomy of a mouse chromosome. PLoS Genet. 4, e1000119 (2008).
Savisaar, R. & Hurst, L. D. Exonic splice regulation imposes strong selection at synonymous sites. Genome Res. 28, 1442–1454 (2018).
Dukler, N., Mughal, M. R., Ramani, R., Huang, Y.-F. & Siepel, A. Extreme purifying selection against point mutations in the human genome. Nat. Commun. 13, 4312 (2022).
Warnecke, T., Weber, C. C. & Hurst, L. D. Why there is more to protein evolution than protein function: splicing, nucleosomes and dual-coding sequence. Biochem. Soc. Trans. 37, 756–761 (2009).
Cáceres, E. F. & Hurst, L. D. The evolution, impact and properties of exonic splice enhancers. Genome Biol. 14, R143 (2013).
Kessler, M. D. & Dean, M. D. Effective population size does not predict codon usage bias in mammals. Ecol. Evol. 4, 3887–3900 (2014).
Wu, X. & Hurst, L. D. Why selection might be stronger when populations are small: intron size and density predict within and between-species usage of exonic splice associated cis-motifs. Mol. Biol. Evol. 32, 1847–1861 (2015).
Xie, W. et al. Epigenomic analysis of multilineage differentiation of human embryonic stem cells. Cell 153, 1134–1148 (2013).
Parker, H. J. Mammalian Embryo: Hox Genes. Encyclopedia of Life Sciences (Wiley, 2025).
Hu, X. et al. DNA methylation of HOX genes and its clinical implications in cancer. Exp. Mol. Pathol. 134, 104871 (2023).
Sen, P. et al. Spurious intragenic transcription is a feature of mammalian cellular senescence and tissue aging. Nat. Aging 3, 402–417 (2023).
Piunti, A. & Shilatifard, A. The roles of Polycomb repressive complexes in mammalian development and cancer. Nat. Rev. Mol. Cell Biol. 22, 326–345 (2021).
Liu, P. et al. Altered DNA methylation pattern reveals epigenetic regulation of Hox genes in thoracic aortic dissection and serves as a biomarker in disease diagnosis. Clin. Epigenet. 13, 124 (2021).
Terranova, R., Agherbi, H., Boned, A., Meresse, S. & Djabali, M. Histone and DNA methylation defects at Hox genes in mice expressing a SET domain-truncated form of Mll. Proc. Natl. Acad. Sci. USA 103, 6629–6634 (2006).
Haghani, A. et al. DNA methylation networks underlying mammalian traits. Science 381, eabq5693 (2023).
Turner, D. C. et al. DNA methylation across the genome in aged human skeletal muscle tissue and muscle-derived cells: the role of HOX genes and physical activity. Sci. Rep. 10, 15360 (2020).
Siebold, A. P. et al. Polycomb repressive complex 2 and trithorax modulate Drosophila longevity and stress resistance. Proc. Natl. Acad. Sci. USA 107, 169–174 (2010).
Armstrong, J. et al. Progressive Cactus is a multiple-genome aligner for the thousand-genome era. Nature 587, 246–251 (2020).
Hubisz, M. J., Pollard, K. S. & Siepel, A. PHAST and RPHAST: phylogenetic analysis with space/time models. Brief. Bioinform. 12, 41–51 (2011).
Morales, J. et al. A joint NCBI and EMBL-EBI transcript set for clinical genomics and research. Nature 604, 310–315 (2022).
Karczewski, K. J. et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434–443 (2020).
Simão, F. A., Waterhouse, R. M., Ioannidis, P., Kriventseva, E. V. & Zdobnov, E. M. BUSCO: assessing genome assembly and annotation completeness with single-copy orthologs. Bioinformatics 31, 3210–3212 (2015).
Quinlan, A. R. BEDTools: the Swiss-Army tool for genome feature analysis. Curr. Protoc. Bioinforma. 47, 11.12.1–11.12.34 (2014).
Li, H. & Durbin, R. Inference of human population history from whole genome sequence of a single individual. Nature 475, 493–496 (2011).
Mi, H. et al. Protocol update for large-scale genome and gene function analysis with PANTHER classification system (v.14.0). Nat. Protoc. 14, 703–721 (2019).
Lonsdale, J. et al. The Genotype-Tissue Expression (GTEx) project. Nat. Genet. 45, 580–585 (2013).
Van Nostrand, E. L. et al. A large-scale binding and functional map of human RNA-binding proteins. Nature 583, 711–719 (2020).
Fairbrother, W. G., Yeh, R.-F., Sharp, P. A. & Burge, C. B. Predictive identification of exonic splicing enhancers in human genes. Science 297, 1007–1013 (2002).
Acknowledgements
Computations and data handling were enabled by resources in projects NAISS 2023/5-90 and NAISS 2023/6-193 provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through grant agreement no. 2022-06725. K.L.T. is a Distinguished professor funded by the Swedish Research Council.
Funding
Open access funding provided by Uppsala University.
Author information
Authors and Affiliations
Contributions
K.L.T. and M.J.C. conceived the project; M.J.C. and M.X.D. carried out the bioinformatics; M.J.C. performed the analyses; M.X.D., J.R.S.M., S.V.K. and K.L.T. contributed to interpretation; M.J.C. wrote the first draft and all authors edited the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Communications thanks Jonathan Romiguier and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Source data
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Christmas, M.J., Dong, M.X., Meadows, J.R.S. et al. Interpreting mammalian synonymous site conservation in light of the unwanted transcript hypothesis. Nat Commun 16, 2007 (2025). https://doi.org/10.1038/s41467-025-57179-w
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s41467-025-57179-w
