Chromosomal Redistribution of Male-Biased Genes in Mammalian Evolution with Two Bursts of Gene Gain on the X Chromosome

1In mammals and Drosophila, the X chromosome usually differs dramatically from autosomes since it is hemizygous in males. Sexual antagonism (beneficial for one sex, but deleterious for the other) enriches male-biased genes on the X chromosome, if alleles are generally recessive, and on the autosome if they are generally dominant –. On the other hand, inactivation of the X chromosome during spermatogenesis – drives the accumulation of male-biased genes on the autosomes where they can be expressed in the meiotic or post-meiotic phase –. These two processes can explain the gene traffic between the X and autosomes in Drosophila and mammals – as well as the excess of male-biased genes on the autosomes –.

2However, recent analyses of male-biased genes identified several X-linked genes that originated in the last 1–3 million years (myr) in Drosophila –. Whether or not these data implicate an effect of evolutionary time on the chromosomal location of male-biased genes remains unknown. In our investigation of how the various evolutionary forces impact the chromosomal distribution of sex-biased genes, we focused particularly on how the age of genes affects their chromosomal locations. By dating when genes arose in humans and mouse, we found male-biased genes were distributed at different locations in different phases of mammalian evolution: young male-biased genes are enriched in the X chromosome, but older male-biased genes favor autosomal locations. Interestingly, this redistribution seems to be associated with feminization of the X chromosome with more X-linked old genes expressed in ovaries.

3Besides the recent gene gain contributed by emergence of male-biased genes on the X chromosome, we found another burst of gene gain on X chromosome immediately after the divergence of opossum and eutherian mammals. Accelerated protein evolution and transcriptional evolution of X-linked genes reveal positive selection occurring in this period. These data support the recent notion, that our X chromosome originated in the therian ancestor instead of the common ancestor of all mammals.

4These two lines of findings significantly extend our knowledge of the origination and evolution of X chromosomes in mammals.

5We inferred the origination times of genes based on the presence and absence of orthologs in the vertebrate phylogeny and assigned 19,935 human and 21,122 mouse protein-coding genes into different evolutionary branches (Figure 1; Table S1, S2; Materials and Methods). We found that 1,828 human genes are primate-specific (branches 8 to 12 of Figure 1A) and 3,111 mouse genes are rodent-specific (branches 8 to 11 in Figure 1B) –. In subsequent analyses, except if specified elsewhere, we define them as young genes and the remaining as old genes.

6Compared to previous reports,, our method identified young genes more conservatively. For example, Church et al. identified up to 2,941 primate specific genes, considerably more than we found. Also, for the 67 human genes that intersect between our dataset and, we assigned 44 (66%) genes onto the same phylogenetic branch as they did. For the remaining 23 cases in conflict, we assigned 22 to older branches (Table S3) since we used a larger number of outgroup species.

7We tracked the relative gene abundance of individual chromosomes across 450 myr and identified two bursts of genes occurring on the X chromosome (Figure 2). One burst (branches 5–7) postdated the divergence of eutherian mammals (human or mouse) and marsupials (opossum) and the other occurred recently after the split of human and chimp and after the split of mouse and rat, respectively. For both peaks, the X chromosome contributes to 8%∼14% of genes, while it only accounts for 3% of genes in the first 300 myr of vertebrate evolution. In contrast, autosomes tend to vary less in their relative contribution to the whole genome (Figure S1). As the major contributor generating new genes, DNA-level duplication accounts for 73∼95% of genes of these two peaks. If we only use DNA-level duplicates, the pattern remains the same.

8Considering that many more genes arose in branch 5 compared to branch 6 or 7 (1,200∼1,400 versus 400∼500, Figure 1), the old peak seems to be best explained by the hypothesis that the X chromosome emerged in the therian ancestor and subsequently recruited many genes in an accelerated evolution of sex-related functions, as found with retrogene-based chromosomal movement studies. In contrast, the recent burst reveals a rapid addition of new genes into the mammalian X chromosome, which may be independent of major chromosomal changes.

9Based on human body index data (GSE7307, Materials and Methods) and mouse tissue profiling data at the NCBI GEO database, we identified genes with sex-biased expression (Materials and Methods). As shown in Figure 3, both human and mouse demonstrate a similar pattern regarding the proportion of male-biased genes and the age of the branch in which they arose.

10For younger branches (less than 50 myr), male-biased genes are enriched in the X chromosome compared to autosomes (∼50% versus ∼30%, Chi-square test p<0.05), which might be driven by fixation of recessive male-beneficial alleles under sexual antagonism. This pattern decreases for genes originating in earlier branches. Male-biased genes older than 300 myr are overrepresented on the autosomes (∼30% versus ∼15%, p = 1×10−9). This pattern was independently supported by an Affymetrix exon array panel with larger coverage of new genes (Figure S2). Thus, the recent peak observed in Figure 2 could be attributed to a burst of male-biased genes on X chromosome younger than 50 myr. Figure 3 also demonstrates that the X chromosome consists of a similar or even higher proportion of male-biased genes compared to autosomes from 90 myr ago (branch 7) to 130 myr ago (branch 5). Thus, many of the genes gained in the first, older peak may also have male-biased expression.

11Notably, the proportion of female-biased genes on branch 5 was greater on the X chromosome compared to autosome (39% versus 20% in Table 1). In contrast, for branches 6 and 7, the proportion of female-biased genes is around 15% for both the X chromosome and autosomes (Table S4). Again, this suggests that the newly originated X chromosome was subjected to enhanced positive selection and recruited an excess of both male- and female-biased genes.

12The earlier peak in Figure 2 indicates the mammalian X chromosome emerged before the divergence of eutherian and marsupial. Thus, the nascent X chromosome changed remarkably, gaining an excessive number of genes. If this scenario is true, those preexisting genes on the ancestral X chromosome might have accumulated many evolutionary changes during this period (branch 5), as did genes linked to the neo-X chromosome in Drosophila. That means we would expect these ancient genes on the X chromosome to show signatures of positive selection. To test this scenario, we investigated the evolutionary path of ancient genes shared by vertebrates by comparing the ratio between non-synonymous substitution rate and synonymous substitution rate (Ka/Ks) (Materials and Methods). In other words, we compared the Ka/Ks of X- and autosomal-linked old genes in separate evolutionary periods. Across evolution of 450 myr, the X chromosome did not show significantly higher Ka/Ks except in branch 5 (Table 2), which strongly corroborates the hypothesis that the X chromosome did not acquire sex-chromosome status until this period.

13We extended this analysis to genes gained since branch 5. We directly estimated the proportion of replacement substitutions (α) based on polymorphism and divergence data in and a maximum-likelihood method implemented in the DoEF package. As shown in Table S5, young genes generally show higher α compared to old genes, and X-linked male-biased genes show the highest α, 0.501. This pattern shows that positive selection instead of neutrality drives the evolution of X-linked genes arising since branch 5, especially those with male-biased expression.

14However, positive selection of nucleotide substitutions can only suggest that initial fixation may also be driven by positive selection. More direct evidence comes from copy number polymorphism (CNP) data in Drosophila, which showed that the X chromosome is subject to stronger purifying selection than autosomes. In human, it was also noted that the X chromosome shows a paucity of CNPs. Together with bursts of adaptive fixations occurred on the neo-X of Drosophila, it is likely that positive selection instead of drift accounts for two bursts of genes on the X chromosome.

15As we noted before, enrichment of young male-biased genes on the X declines for those originating in earlier evolutionary branches. Using expression data from mouse spermatogenesis, we compared different age groups to investigate which force underlies such a demasculinization process (Table 3). As previous studies such as found, old genes are expressed more in the pre-meiosis stage (spermatogonia) but are silent from meiosis (pachytene spermatocyte) to post-meiosis (round spermatid). In terms of whole testes, however, old X-linked genes are underrepresented (Table 3). New genes show a distinct pattern: while often expressed in spermatogonia, they are not silent in meiosis. Moreover, a much greater proportion of new genes on the X are expressed in the post-meiosis stage compared to genes on the autosome (70% versus 27%, Chi-square test p = 5×10−10). This is consistent with a previous observation of X-linked postmeiotic multicopy genes, the vast majority of which we found were very young (Materials and Methods). Such a pattern suggests that the young X-linked genes are not affected by MSCI. An independent microarray dataset of mouse spermatogenesis confirms high expression of X-linked young genes in spermatid (Figure S3). In addition, we note that the customized array by Khil et al. was comprised mainly of old and conserved genes, with only 1.7% of the set being young genes. In contrast, the Affymetrix array data we used covered 14,923 Ensembl genes, 3.9% of which are young genes.

16This striking contrast between young and old genes suggests that MSCI plays an important role in determining the age-dependent chromosomal distribution of male-biased genes. In order to investigate how this contrast occurred in such a short time, we analyzed four major cell types including sertoli cells, spermatogonium, spermatocyte, and spermatid between mouse and rat. We used the Euclidean distance of relative abundance (RA) to measure how orthologous genes have diverged in their expression (Materials and Methods). Consistent with a previous comparison of human and chimpanzee, the testis expression of genes on the X chromosome diverge more between rat and mouse than genes on autosomes (Wilcoxon rank sum test p = 4×10−6, Figure 4). Furthermore, X-linked young genes show significantly higher divergences, compared to all other three groups (p<0.05).

17While we found that expression in various spermatogenesis stages is generally conserved with only about 3% divergence (Figure S4), X-linked young genes show the largest expression divergence in spermatid. Specifically, after the split of mouse and rat 37 myr ago, young X-linked genes show 6.9% divergence in spermatid, which is much higher than the genomic average for spermatid, 3.3% (Wilcoxon rank sum test p = 0.002). This increased divergence suggests that, although these genes seem to escape MSCI and preferentially transcribe in post-meiosis, the expression profile is not conserved. It remains unknown whether these genes get up-regulated or down-regulated in one species. But if the latter case were true, it indicates that the high post-meiotic expression would be silenced by MSCI in later evolution. This could also explain how the different pattern between young and old genes in Table 3 is achieved.

18We investigated the distribution of female-biased genes on chromosomes and its correlation with gene ages. Interestingly, female-biased genes are distributed in a pattern symmetrical to male-biased genes (Figure S5 versus Figure 3): the old X-linked genes are more often female-biased, while young genes are not.

19We characterized ovary expression of genes using the Affymetrix mouse exon array panel data. Consistent with Figure S5, ovary expression also depends on the age of the gene's origination. Specifically, young autosomal genes show significantly higher expression in ovaries than young X-linked genes (Wilcoxon rank sum test p = 5×10−12, Figure 5). However, old X-linked genes generally show higher expression in ovaries (p = 5×10−7). Thus, as gene age increases, this expressional excess of autosomal genes reverses and older X-linked genes show significantly higher expression in ovaries.

20It can be argued that such an age-dependent pattern of expression is not a specific property of ovary evolution and other organs might also show a similar pattern. To test this possibility, we investigated gene expression in the major organs: brain, heart, kidney, liver, lung, muscle, spleen, and thymus. All these tissues, except for brain, showed a significant excess of expression for new genes (branch≥5) on autosomes compared to that of X-linked genes (Wilcoxon rank sum test p<0.01, Figure S5). However, for old genes (branch≤4), they are evenly distributed (p>0.05).

21The brain shows a unique pattern. Young genes (branch>7) are relatively abundant on autosomes (p = 0.001, Figure S5), but old genes (branch≤7) are overrepresented on the X chromosome (p≤0.01). This is consistent with previous findings that X chromosome is enriched with genes expressed in brain,. Notably, different from ovaries, enrichment in the brain did not show clear age dependence, since genes originating from branches 5 to 7 presented the most significant excess (Figure S6).

22The coincidence that the X chromosome is enriched with both ovary-expressed and brain-expressed genes occurring in branch 5 (Table 1; Figure S5) motivated us to perform more thorough transcriptional profiling to get a more complete picture of how genes from this evolutionary period are transcribed.

23We investigated mouse exon atlas data (GSE15998) to ask whether X-linked genes are more frequently expressed in the tissue of interest across different age groups. We clustered tissues by the proportion of X-linked genes expressed versus the proportion of autosomal genes expressed and identified three major groups: nervous system, testes, and all other tissues (Figure 6). Remarkably, the X-linked genes originating in branch 5 are transcriptionally permissive with a larger proportion of them expressed in many tissues compared to autosomal genes. This excess is most pronounced for brain samples.

24Consistently, human data revealed that a greater proportion of X-linked genes emerging on branch 5 are expressed more widely than autosomal genes originating in this period, which is strongest for the brain (Figure S7). Since human and mouse share a similar pattern, parsimony suggests this striking transcriptional pattern of branch 5 derived genes is ancestral. Notably, none of these genes show sex bias in human brain profiling data, which suggests they might be important for both sexes.

25We have described evolutionary patterns of protein-coding genes, which could be driven by natural selection in various forms like sexual antagonism or MSCI. If, however, such a pattern is a product of some mutational bias of gene origination, we would not detect similar evolutionary patterns in non-coding RNA genes, such as X-linked miRNAs. Therefore, we investigated the chromosomal distribution of miRNA genes annotated in miRBase and found that miRNA duplicates are distributed in a pattern similar to that observed for protein-coding genes (Table S6). Specifically, both human and mouse show significant miRNA gene gain in branches 5 to 7 compared to the proportion of all miRNA genes (18∼22% versus 10∼13%, Fisher's Exact Test p<0.05). Moreover, they also show an excess for the youngest branch. Although it is not significant for the human data due to small sample size, it is for mouse (p = 0.02).

26Like protein-coding genes, a larger proportion of X-linked miRNAs originating in branch 5 are transcribed in nine tissues (statistically significant for six of them) surveyed on Agilent chip compared to autosomal genes (Table S7; Materials and Methods). Moreover, semi-quantitative PCR data of X-linked miRNAs in 12 tissues show 9 out of 13 (69%) young genes are expressed higher in testes than at least six non-testis tissues. However, this percentage drops to 23% for old X-linked genes (9 out of 39, Fisher's Exact Test p = 0.005). Consistent with protein-coding genes, these data also show that old genes have moderate or high expression in ovaries and the young genes show only trace levels of expression (Wilcoxon rank sum test p = 0.01).

27The age-dependent locations and expression profiles of miRNAs support that it is evolutionary forces, rather than some mutation bias intrinsic to a certain type of gene, which account for the dynamics of X-linked gene evolution.

28It is known that the X chromosome can be divided into five evolutionary strata because of step-wise repression of recombination –. The X-conserved region (XCR) consists of the oldest strata 1 and 2, while the X-added region (XAR) includes younger strata 3 that is shared by primates and rodents, and much younger 4 and 5 that were derived within primates,. Since sexual antagonism or other sex related forces like the faster-X process (see Discussion) depends on hemizygosity of the X chromosome in male, we expect the accordance between bursts of gene gain with the formation of corresponding strata. If these forces shape the evolution of gene content on the X chromosome, we should find that X-linked genes originating at a given time period should accumulate only in the strata already formed at that time. In other words, we should find a correlation between the ages of genes and the strata in which they are located. Consistent with these predictions, Figure 7 shows that the older strata 1 to 3 are associated with relatively older genes, while strata 4 or 5 are enriched with younger genes (one sided Fisher's Exact Test p = 0.03). This finding parallels the temporal correspondence between the occurrence of strata and the out-of-X retrogene traffic.

29Our analyses demonstrated that the X chromosome evolved dramatically on both the sequence and expression levels after the split of eutherian mammal and marsupials. Specifically, the X chromosome showed a burst of gene gain during this time, and many of these genes quickly invaded the transcriptional network of various tissues, especially the brain. Furthermore, genes predating the birth of the X showed rapid protein-level evolution. A straightforward interpretation is that the newborn mammalian X was subjected to strong positive selection similar to the neo-X chromosome in Drosophila. Moreover, the X-linked genes arising in branch 5 seem to have played important roles, as shown by their broad expression. Their transcription pattern suggests that the early evolution of placental mammals was associated with rapid changes in the brain. Furthermore, analysis of gene ontology showed that many of these genes mainly played regulatory roles in transcription and metabolism (Table S8). Thus, regulatory change contributed by gene gain on the X chromosome was extensively involved in the initial evolution of eutherian mammals. The fact that this peak ranges between branches 5 and 7 suggests remodeling of incipient X chromosome might take about 90 myr (−160∼−70 myr, Figure 2), which is consistent with one report based on retrogene movement. However, the selective pressures driving this dramatic change in branch 5 appear to be smaller in subsequent branches (Table 2).

30Our analyses reveal chromosomal redistribution of X-linked male-biased genes. Sexual antagonism may contribute to the initial fixation of X-linked recessive alleles as described previously,. The faster-X hypothesis was initially proposed to fix more mutations on the X chromosome only if they are recessive and beneficial. Recently, it was observed that this force was most pronounced for male-biased genes. This suggests that the faster-X process could also be involved in the emergence of young X-linked male-biased genes, as the hypothesized sexual antagonism might.

31These young X-linked male-biased genes could be later silenced by MSCI as suggested by Table 3, Figure 4, and Figure S4. At least two processes could be involved in this switch. First, we found a statistically significant excess of male-biased retrogenes generated in the X→A movement process and X-enrichment of the female-biased parental genes for both human and mouse (Table S9). Thus, the demasculinization and feminization of the X chromosome could be coupled in retrogene traffic. Moreover, our RA analysis (Figure S6) extends the out-of-the-testes hypothesis to non-retroposed new genes. We found that new genes generally acquire transcription in more tissues during evolution although they are initially enriched in testes. With increasing MSCI and expanding expression breadth, X-linked male-biased genes might become unbiased or even female-biased as Figure S6 shows.

32If new strata on the X chromosome represent regions that did not develop recombination repression until recently, the genes encoded in these regions will often escape MSCI. Thus, it is expected that the X-linked male-biased genes more likely escape MSCI when located on young strata or pseudoautosomal regions (PARs). However, out of 13 young male-biased genes in humans, the relatively young strata 4 and 5 encode only one (Table S10), which does not significantly differ much from the expected number based on its genomic size. How then did the remaining 12 genes, those situated on older strata, escape from MSCI?

33It was proposed that the excess of inverted repeats (IRs) encoded by human and mouse X chromosome could protect genes contained by these IRs from MSCI. IRs suppress MSCI through formation of cruciforms or other unusual chromatin structures. Moreover, cancer/testis (CT) genes that are often expressed in normal testes and in cancerous tissues frequently overlap with IRs. Given that X-linked CT genes underwent recent expansion, it is not surprising that some of them could form highly homologous IRs. In fact, 8 out of 13 young X-linked male-biased genes are CTs (Table S10). Thus, the high IR abundance on the mammalian X chromosome might be one reason that these genes can be transcribed in meiosis or postmeiosis.

34Furthermore, out of 12 genes encoded by PARs and covered by unique probes (Table S11), there is only one (8%) male-biased gene, PPP2R3B, which is shared by human and mouse. Thus, different from our intuition, PARs do not harbor an excess of male-biased genes compared to the remaining strata (18%) and to autosomes (24%). Albeit of small sample size, this observation suggests that sex-related forces like sexual antagonism or faster-X process account for the observed excess of young X-linked male-biased genes.

35There are only limited number of genes with unique probes on strata 4 (five) and 5 (eight). For the remaining strata, stratum 3 is enriched with male-biased genes, which is much higher than stratum 1 (27% versus 17%, one-sided Fisher's Exact Test p = 0.02) and stratum 2 (27% versus 15%, p = 0.03). This pattern suggests that stratum 3 recruits more young male-biased genes and there was not enough evolutionary time to be feminized as occurred in the oldest strata 1 and 2.

36As shown in Figure 2, the emergence of young male-biased genes peaks in recent evolution of human and mouse. However, this peak started 30 myr ago (before the divergence of mouse and rat) in the rodent lineage, while the peak appeared in the last 5 myr in human lineage.

37This difference is consistent with the fact that the mouse X encodes more young male-biased genes than the human X. Specifically, male-biased genes account for 52% and 74% of X-linked young genes in human and mouse, respectively (Figure 3; one sided Fisher's Exact Test, p = 0.07). Exon array data are similar (Figure S2; 45% versus 76%, one sided Fisher's Exact Test, p = 2×10−8). Origination of significantly more male-biased young genes suggests that stronger positive selection acts on rodents and could explain why the recent peak of gene gain (Figure 2) began earlier in the mouse lineage than in the human.

38We downloaded Ensembl release 51 (November, 2008) as the basic gene dataset for our analyses. We used MySQL V5.0.45 to organize the data, BioPerl and BioEnsembl to fold the pipeline, and R V2.8.0 to perform all statistical analyses.

39We developed a genome-alignment based pipeline to infer the origination time of a given genomic region by modifying a previous gene-alignment based method. We analyzed UCSC netted chained file for human (hg18) and mouse (mm9) to verify whether a given human/mouse locus has a reciprocal syntenic alignment in the outgroup genome such as chimpanzee, rat, chicken, and so on. In other words, we investigated whether a best-to-best match could be found between human/mouse loci and outgroup loci regardless of chromosomal linkage. In this way, we can identify orthologous genes; even those with different chromosomal location due to fusions or translocations such as those found in XAR region will be identified as well. Then, in order to handle occasional sequencing gaps, we scanned multiple outgroups and assigned this locus to a specific branch by following a parsimony rule. Compared to the previous method, our strategy is independent of gene annotation of outgroups and robust with gene translocation. Thus, we generated a more stringent young gene dataset (as described in the Result section). And, as Figure S8 shows, we have not assigned most genes encoded by XAR as young genes simply because this region changed the linkage by fusing to X chromosome. Conversely, several genes originated in branch 5 are located in strata 1 and 2 that are not XAR (Figure S8), also supporting that our pipeline is robust with gene translocations.

40Notably, for regions without reliable synteny, our method might not work. This situation would be most pronounced for telomeres, which tend to be repetitive and prone to recombine and thus have very limited synteny. For example, we dated 17 genes situated on PARs of the X chromosome (Table S11). For three genes encoded by PAR2, repeats contribute less than 16% of the gene loci based on UCSC annotation. Accordingly, our age assignments for these three genes are always consistent with those inferred by tree reconstruction provided by Ensembl. In contrast, for 14 genes linked with PAR1, repeats are prevalent with a median contribution of 55% to the gene loci. In this case, our results are consistent for only three out of nine cases with Ensembl age information.

41We slightly modified the previous pipeline,– and classified young genes as DNA-level duplicates, RNA-level duplicates (retrogenes), and de novo genes. Briefly, we performed all-against-all BLASTP search for human and mouse proteins. It was reported previously that retrogenes can recruit other neighboring genome regions with introns after being retroposed. Thus, in order to define a new gene as retrogene, we requested that in the aligned region between the most similar paralog (candidate parental gene) and child genes, the former contain at least one intron and the latter to be intronless. Otherwise, it will be classified as DNA-level duplicates. Notably, if there is no hit with BLAST evalue cutoff 10−6 found and no annotated paralog by Ensembl, the gene will be defined as de novo.

42In order to avoid non-specific probes and to cover more recently annotated genes, we used the customized array annotation files (released on November, 2008) downloaded from University of Michigan, HGU133Plus2_Hs_ENSG (Affymetrix Human 133 plus 2) and Mouse4302_Mm_ENSG (Affymetrix Mouse Genome 430 2.0 Array) for human and mouse, respectively. For exon array analysis, we used HuEx-1_0-st-v2,U-Ensembl49,G-Affy.cdf and MoEx-1_0-st-v1,U-Ensembl50,G-Affy,EP.cdf generated by Aroma.affymetrix team. Thus, we excluded some candidate young genes that were too similar to their paralogs and did not have specific probes.

43Based on R and Bioconductor platform, we used RMA to normalize and generate gene-level intensity for 3′ gene array and Aroma.affymetrix to normalize and summarize gene-level signal for exon arrays. We used MAS5 to call expressional presence and absence for 3′ gene array. In case of exon array, we used Affymetrix dabg (detection above background) algorithm to generate chip specific background signal and then compared gene-level signal to this background with Wilcoxon rank sum one-tail test. Considering multiple-testing issues, we converted all p values to q values using the qvalue package. The q value of 0.01 was used as the cutoff. For Agilent miRNA array, we used “gIsGeneDetected” column generated by Agilent Feature Extraction software to define presence or absence calls. We required a gene to be present in all replicates to be considered a presence and a gene to be absent in all replicates to be considered an absence. We removed all ambiguous cases from the final statistics.

44We used the LIMMA package to call expressional difference, with a false discovery rate corrected p of 0.05 used as the cutoff. Although we compared testis and ovary, we used the term “male-bias” or “female-bias” rather than “testis-bias” or “ovary-bias.” The reason is that these two datasets are nearly equivalent. A previous study showed that the proportion of germline male-biased genes is much higher than that of somatic male-biased genes (20% versus 2%).

45For meta-analyses of mouse and rat spermatogenic data, we followed the concept of RA and euclidean distance (d) to measure the between-species expression divergence. Specifically, we defined RA as the proportion of expression intensity of one tissue out of all tissues and d as the sum of the square of RA difference for all tissues between mouse and rat, i.e.,.

46We mapped 20 out of 33 representative genes in to our gene age data using unique NCBI gene names. Remarkably, 16 (80%) are rodent-specific, with 11 of them originating after the mouse and rat split. We note here that this dataset does not overlap with what we described in Table 3, since Table 3 only presents genes with unique probes, which 19 of these 20 genes do not have.

47We downloaded the vertebrate-wide 44-way coding sequence alignment from UCSC. UCSC known genes mapping to multiple Ensembl genes were discarded. For Ensembl genes mapping to multiple UCSC known genes, we retained only one UCSC gene with the longest coding region. Then, considering that low quality assembly often causes unreliable estimation of Ka/Ks, we extracted 17 species with relatively better quality (Figure 1) and then removed all in-frame stop codons or gaps in the alignment. According to our age dating information, taxa conflicting with the age were removed. Based on the species tree (Figure 1), we estimated Ka/Ks for each branch using free ratio model in PAML.

48We downloaded Gene Ontology (GO) annotations for Ensembl V51. We used the program analyze.pl V1.9 of TermFinder package to identify those significant terms for new genes, with multiple test corrected p of 0.05 as the cutoff and the whole genome as the background. Herein, TermFinder was updated to V0.83, which corrected a mistake in calculating false discovery rate.