Our first goal was to build a comprehensive list of genes encoded on the mouse X-chromosome with sex-biased expression in somatic tissues. To provide sufficient statistical power to detect even small expression differences, we analyzed a large collection of oligonucleotide microarrays (n_arrays=728, Affymetrix Mouse 430v.2) hybridized with RNA samples obtained from six somatic tissues: kidney (n_arrays=152, with n_{female arrays}: n_{male arrays} equal to 99:53), liver (88, 44:44), lung (67, 33:34), striatum (46, 25:21), eye (176, 88:88) and hippocampus (199, 101:98). To identify differentially expressed genes, we applied Wilcoxon rank-sum test (results from parametric tests are broadly similar, see Methods). The result from this analysis is graphically presented in Figure 1, including data for X, Y and autosomal chromosomes. Table 1 gives an overview of the results obtained at the significance threshold p<0.001. Detailed data for all individual genes and tissues are found in Additional file 1 and Additional file 2.

The table presents the number (#) and the percentage (%) of significantly sex-biased assayed probes (p<0.001, two-sided Wilcoxon test) separated by chromosomal location. Abbreviations: Tot Total, F female-biased, M male-biased, N/A chromosomal location not available. Complete data for all individual probes is available in Additional file 1 and Additional file 2.

A first observation was that the overall percentage of sex-biased transcripts encoded on the X-chromosome was higher than the percentage of sex-biased transcripts encoded on the autosomes, and this observation was consistent in all tissues (See columns labelled “Autosomes Tot” and “X-chromosome Tot”, Table 1). Therefore, it can be concluded that the X-chromosome is a chromosomal location overrepresented for sex-biased genes, in agreement with previous results 12 . Also in line with previous investigations, we observed substantial tissue differences in the number of sex-biased genes (Table 1). This was the case when considering only the X-chromosome (ranging from 25 to 287 transcripts) as well as when considering autosomes (8–7825 transcripts), while the number of Y-linked genes remained nearly constant between the tissues analysed (12–17 transcripts).

We found that while sex-biased genes located on X are numerous, the magnitude of their sex differences was small, and this was valid for all somatic tissues investigated. For example, while 80 autosomal transcripts showed sex differences larger than 4-fold in the kidney, not a single transcript with such a large difference was found on the X-chromosome when Xist (X-inactive specific transcript, “Master regulator” of XCI) was excluded (Additional file 1). In the liver, 53 autosomal transcripts were sex-biased beyond 4-fold, while only a single male-biased X-linked gene (Alas2) showed differences beyond 4-fold (Xist excluded). These small differences are better illustrated in Figure 1, where only three crosses, all of them representing probes for parts of the Xist gene, resulted in female-bias greater than 4-fold. At the same time, the figure shows results for many autosomal transcripts with large sexual expression differences, particularly in the liver and kidney. Genes encoded by the Y-chromosome also resulted in large differences in all tissues as expected, since they are only expressed in males (Figure 1). The spatial distribution of sex-biased genes along the X-chromosome is illustrated in Figure 2A, where it can be observed that their distribution is widespread over the chromosome rather than concentrated on specific chromosomal segments, and does not significantly diverge from the overall gene density on the X.

To study the relative allocation of male- and female-biased genes, we compared their frequencies on the X-chromosome. We found significantly more transcripts with female-bias than with male-bias in all six somatic tissues examined (Figure 3A, comparison X_F vs. X_M, p<0.001). Next, we compared the X with autosomes, and found female-biased genes to be more abundant on the X in all tissues (Figure 3A, comparison X_F vs. A_F, p<0.001). More remarkable, in the most sexually dimorphic somatic tissues analysed: the kidney and liver, we found a significant depletion of male-biased genes on the X relative to the frequencies on autosomes (Figure 3A, comparison X_M vs. A_M, kidney: 25% relative depletion, p= 0.0032 liver: 27% relative depletion, p= 0.0099). These observations provide the first evidence of de-masculinisation of a mammalian X-chromosome in terms of X-gene expression in somatic tissues. The overall frequencies of male-biased genes in the remaining four tissues were too low to allow similar meaningful comparisons between X and autosomes in those tissues.

To evaluate whether biallelic expression of genes previously known to escape XCI could explain the excess of female-biased transcripts observed on X, we compared the frequencies of male- and female-biased X-genes (Figure 3B, left panel) with the frequencies of sex-biased genes excluding previously known mouse escapees (Figure 3B, right panel). As expected, the exclusion of escapees reduced the female to male imbalance on the X, however, a substantial excess of female-biased genes remained. This suggests the existence of novel tissue-specific escapees in our gene sets, and/or other sex-skewed mechanisms exist that regulate the expression of the remaining female-biased genes.

We have previously identified small female-biased gene clusters on the mouse X-chromosome, including female-biased long non-coding RNAs (lncRNAs) and protein coding genes that escape XCI, raising the possibility that such lncRNAs might be involved in the epigenetic regulation of escape domains 22 . Using the more extensive datasets included in the current study, we searched for additional female-biased clusters or additional genes in previously found clusters. To do this, we performed a second analysis in which the significance cut-off was relaxed from p<0.001 to p<0.01, increasing the sensitivity to detect sex-biased genes. At the same time, we only selected genes with consistent bias in more than one tissue, decreasing the expected number of false positives. The result from this analysis is shown in Table 2, containing 94 genes (and 8 additional non-annotated intronic sequences) with female-bias, and only 33 genes (and 2 intronic sequences) with male-bias. The spatial distribution of these genes on the X-chromosome is shown in Figure 2B. Female-biased genes located in close proximity to each other (defined as less than 50 kb) are marked “a f” on the right side of Table 2. The table then shows 6 clusters, and each of them, with the exception of cluster “e”, containing an lncRNA. One of the female-biased clusters “f” deserves special attention because it includes several novel non-coding genes. In addition to the known escapees Kdm5c (coding) and D930009K15Rik (non-coding), three additional female-biased 3’-located non-coding genes were herein identified in this cluster, including 2810454L23Rik, 1441020_at (intergenic) and 2900056M20Rik.

The female-biased (A) and male-biased (B) genes with the same direction of sex-bias in more than one tissue are listed. The mean fold-changes for significant differences are given for each tissue. Fold-changes in Italic font denote the significance level p<0.01, and fold-changes in bold font denote the significance level p<0.001. The letters “a-f”, to the right in the table, indicate female-biased gene clusters as described in the main text. The additional notation used in the table is as follows. i: intronic transcript. ^: the exact probe mapping coordinates was not specified in the array annotation file, and gene start/stop locations are therefore given according to the NCBI37/mm9 genome assembly. ^&: these genes showed opposing sex-bias for different probes. *: known XCI escapee gene.

We sought to investigate whether any of the newly identified female-biased genes might escape XCI. Candidates for this analysis were selected from Table 2, since these genes exhibit bias in at least two tissues, and therefore the evidence for possible escape from inactivation is stronger than for genes biased only in one tissue. Indeed, every gene in Table 2 that were female-biased consistently in all tissue analysed, have been previously recognised as escapees 22 23 24 25 . New selected candidates might escape XCI in a tissue-specific manner. As a first step, to identify novel escapee genes, we here investigated possible biallelic expression of five candidate genes in primary female mouse embryonic fibroblasts derived from skin. We performed RNA-FISH using a probe for Xist RNA in parallel with a probe for either one of the five candidate genes (4732460I02Rik/Rbmx ^24/25 , Rbm4 ^70*, Tmem164 ⁷³ , Tmem29 ⁷⁴ and Ctps2 ⁸⁷ ; superscripted numbers corresponding to Table 2) and a positive control for escape: Kdm5c ⁷⁶ . Kdm5c showed biallelic signals in 25% of the cells (13 of 52 counted, Figure 4). One of the candidates, Tmem29, was confirmed to partially escape XCI, showing biallelic signals in 12% of the cells (10 of 82 counted, Figure 4). None of the other candidates indicated biallelic expression in any of the examined cells, suggesting that they are X-inactivated in mouse embryonic skin fibroblasts. The possibility remains that they might escape inactivation in other cell types.

To investigate whether sex-biased X-linked genes are associated with a specific functionality that is overrepresented, we performed a Gene Ontology classification 26 27 . To do this, we contrasted the proportion of sex-biased genes in each functional category with the proportion of the same functional category among all genes in the genome (Additional file 3 A) or with all genes on the X-chromosome (Additional file 3 B). These analyses revealed three groups with significant over-representation (p<0.05 after Bonferroni correction) in the genome-wide comparison, namely IPR006911: Protein of unknown function DUF634 (p=3E-06; overrepresented among female-biased genes in kidney), GO:0031090 Organelle membrane (p=0.009; overrepresented among male-biased genes in kidney), and perhaps most interesting because of possible sex differences in synaptic plasticity: mmu04720: Longterm potentiation (p=0.04; among female-biased genes in striatum).

It was earlier shown that genes on the X-chromosome with female-biased function (defined by high expression in female reproductive tissues) are mainly evolutionarily old genes, while genes with male-biased function (defined by high expression in male reproductive tissues) are enriched among more recently formed genes (<50 myr) 28 . To investigate whether similar conclusions hold for X-genes with sex-biased expression in non-reproductive tissues, we determined the age of the X-linked genes included in our dataset. To do this, we used gene age data obtained recently by comparative genomic analysis for the presence or absence of orthologs in a vertebrate phylogeny 28 , and we calculated the proportion of female- and male-biased genes among the genes assigned to each branch in the phylogeny. We found that the proportion of genes with female-biased expression in kidney and liver was higher among evolutionarily old genes, preceding the radiation of placental mammals, than among younger genes (Figure 5). The difference in proportion of female-biased genes older than 100 myr (branch 0–5) versus younger genes (<100 myr, branch 6–11) was significant both in kidney (p=1.6E-6, two-sided Fischer’s exact test) and liver (p=1.9E-4). A modestly significant age-related decrease in male-biased genes was observed in kidney (p=0.032), while no significant difference was found in liver (p=0.64). Similar analyses of the other somatic tissues included in our study were not possible due to the low number of sex-biased genes detected in those tissues.

We have used a large sample set of oligonucleotide microarrays to identify with high-resolution transcripts encoded on the X-chromosome that are expressed in a sex-biased manner in somatic tissues. We observed a skewed allocation of sex-biased somatic transcripts encoded on the X-chromosome: namely an abundance of female-biased transcripts and a paucity of male-biased transcripts. A previous large genome-wide analysis in mice, including four somatic tissues (liver, adipose, muscle and brain) also found over-representation of female-biased genes on the X 12 . However, under-representation of male-biased genes was not reported, possibly due to the lower probe content of their arrays. Our study might thus provide the first empirical evidence of de-masculinisation of a mammalian X-chromosome in terms of X-gene expression in somatic tissues. Our investigation of somatic tissues complements earlier analyses of sex-biased gene expression on the X-chromosome that investigated expression in reproduction-related tissues such as testis, ovary and placenta 29 . Khil et al. demonstrated that genes enriched in mouse testis are under-represented on the X-chromosome, likely as a consequence of meiotic sex chromosome inactivation (MSCI), and at the same time, genes with high expression in ovary and placenta were shown to be enriched 29 . We here demonstrated a similarly skewed allocation of male- and female-biased X-genes in somatic tissues. However, a crucial difference is that the observed paucity of male-biased genes in somatic tissues is unlikely to be explained by MSCI, since this process does not occur in these tissues. The results are relevant in terms of interpreting the Rice hypothesis 9 . This hypothesis states that any recessive allele on the X that gives males a reproductive advantage is immediately available for positive selection since males carry a single X-chromosome. Accordingly, an excess of male-beneficial genes should accumulate on the X. This may be the case for genes specifically expressed in spermatogonia 30 , but seemingly not for genes enriched in whole testis 29 or for genes expressed in somatic tissues, as indicated in our current study. Therefore, it is most plausible that different selection pressures operate on X-genes active in different tissues, and that the final result in terms of the evolution of gene composition in the X (masculinisation versus feminisation) is a compromise between different evolutionary forces acting on reproductive and somatic tissues, as has been hypothesised both for rodents 8 and similarly for Drosophila 31 . Our data gives support to this hypothesis.

The importance of our results can also be discussed in relationship to studies of gene composition of the sex chromosomes of other species. Indeed, the paucity of male-biased genes in both reproductive and somatic tissues of the mouse is similar to that reported in Drosophila melanogaster 10 and Caenohabditis elegans 11 . Interestingly, while the overall pattern is the same in many species, the particular genes are not. Moreover, the collections of genes biased in reproductive tissues are different from those found in somatic tissues from the same species 32 33 . Nonetheless, the similarity in our results from the mouse X-chromosome and in the evolutionarily unrelated X-chromosomes in distant species such as Drosophila and Caenohabditis elegans, indicate the generality of at least some selective forces acting on X-chromosomes.

The analysis of genes with sex-bias in more than one tissue revealed clues to the potential mechanisms controlling sex-bias of X-encoded genes as well as to whether such mechanisms operate in a ubiquitous versus tissue-specific fashion. Interestingly, while several of the genes listed in Table 2were female-biased in all six tissues, none of the male-biased X-linked transcripts were consistently biased in four or more tissues. Indeed, only two transcripts, Lamp2 and Astx, were male-biased in more than two tissues. This was in stark contrast to the widespread expression differences of several female-biased transcripts, with 25 unique genes female-biased in more than two tissues (Table 2). These observations are in agreement with the conclusion that the mechanisms controlling female-bias versus male-bias are different. It is worth noting that all transcripts that were female-biased in all tissues were encoded by genes known to escape XCI 22 23 24 25 , namely Ddx3x, Kdm6a, Eif2s3x, Xist, 5530601H04Rik, 2610029G23Rik, Kdm5c, and D930009K15Rik. Our microarray analysis of somatic tissues, together with previous global analysis of escape from XCI performed on fibroblasts in vitro 23 , indicate that most of the ubiquitously expressed coding escapee genes in mouse have now been identified. On the other hand, the fact that most of the genes included in Table 2were female-biased in a tissue-specific manner, might suggest that escape from XCI is conveyed by tissue-specific mechanisms, at least for some genes.

Our RNA-FISH experiments confirmed that Tmem29 escapes XCI, adding yet another gene to the 15 or so escapees previously known in mouse 34 . Tmem29 escaped inactivation in 12% of the fibroblasts counted, while Kdm5c presented biallelic expression in 25% of the cells. This deviation suggests dissimilarity in control mechanisms for biallelism for these two genes, but we cannot fully rule out that differences in probe composition contributed to the divergence observed.

Genes escaping X-inactivation 35 and even the Xi heterochromatin itself 36 may underlie significant phenotypic differences between the sexes. It should however be pointed out that female mice that are monosomic for X are fertile and show a mild phenotype compared to women with Turner Syndrome 37 .

Our current investigation extended our previous discovery of female-biased gene clusters on the mouse X-chromosome 22 . Here we investigate in more detail the cluster including Kdm5c, because it contains the largest number of female-biased non-coding genes. X-inactive repressive histone marks, such as histone variant macroH2A1 (macroH2A1) and tri-methylation at lysine 27 of histone 3 (H3K27me3), are known to be depleted in escape domains 23 24 . It was previously shown by us and others that Kdm5c and D930009K15Rik are located in a chromosomal region virtually depleted of these repressive marks, and that both escape XCI 22 23 24 25 . Novel in the current study is the identification of three additional female-biased non-coding genes in this cluster, located downstream of Kdm5c and D930009K15Rik, apparently in the transition region of macroH2A1 and H3K27me3 enrichment (Figure 6A). This pattern is consistent with a model for the Kdm5c escape domain, in which Kdm5c and D930009K15Rik are located in a transcriptionally active chromosomal domain while the subsequent non-coding genes are located in the transition region between active and silenced domains (Figure 6B). We note that one of these non-coding genes, 2900056M20Rik, was in a previous study classified as X inactivated 38 , perhaps indicating tissue variability in escape among genes in the transition region. The coding genes located upstream (Iqsec2) and downstream (Tspyl2) of the female-biased Kdm5c cluster have previously been shown to be X-inactivated 39 , and therefore the limits of the Kdm5c escape domain are established. Whether the non-coding RNAs encoded in this escaping domain serve any function in the regulation of escape, the preservation of the boundary between active/inactive domains, or the silencing of neighbouring inactive genes, remains to be explored.

Escape from XCI is not the only mechanism that leads to female-biased expression of X-linked genes. For example, some of the female-biased genes may be controlled by regulators acting in trans, such as hormonally regulated factors. If this is so, the abundance of female-biased genes on X may correlate with an overrepresentation of the corresponding transcription factor binding sites in X-gene promoters. Conversely, it is conceivable that X-gene transcriptional down-regulating factors are enriched in males, and binding motifs for these factors are then similarly expected to be enriched on the X. Sex-bias of some of the identified genes may also reflect sex differences in cell composition within the tissues. Another possible mechanism for sex-biased gene expression is genomic imprinting of X-chromosome genes 41 42 . Indeed, widespread sex-specific parent-of-origin-dependent allelic expression was reported recently in the brain of rodents in several studies 43 44 45 46 . This may also mediate or resolve sexually antagonistic effects 47 . If any of the mechanisms described above would affect one sex but not the other, the dosage compensation achieved by random X-inactivation would be altered, resulting in sex-biased expression in the particular tissue in which the mechanism operates.

Multiple studies have demonstrated that different evolutionary processes and patterns are observed on X-chromosome and autosomal chromosomes 48 . For example, a recent study showed that two bursts of gene gain on the X-chromosome occurred during the evolution of mammals 28 . Moreover, it was suggested that the X-chromosome recently acquired a burst of young male-biased genes (operating in testis), which is consistent with a fixation of recessive male-beneficial alleles by sexual antagonism 28 . Our results provide a different story in terms of the evolution of sex-biased genes acting in somatic tissues. While both studies indicate that female-biased genes are old, the results for male-biased genes differ between our study of somatic tissues and the study of gonadal expression. We found a moderately significant abundance of evolutionarily old genes also among male-biased genes, at least in the kidney. These combined results demonstrate that considerations on the potential effect of sexual antagonism during the evolution of X-chromosome will require a concerted analysis of the selective forces operating both in gonads and in somatic tissues.

We used array data (Affymetrix Mouse 430v.2) from GeneNetwork (University of Tennessee), available at: http://www.genenetwork.org. Datasets: Kidney: “Mouse kidney M340v2 Male (Aug06) RMA” (GN240), “Mouse kidney M340v2 Female (Aug06) RMA” (GN239); Liver: “GenEx BXD Sal Liver Affy M430 2.0 (Feb11) RMA Males” (GN311), "GenEx BXD Sal Liver Affy M430 2.0 (Feb11) RMA Females" (GN312); Lung: “HZI LungM430v2 (Apr08) RMA” (GN160); Striatum: “HBP Rosen Striatum M340V2 (Apr05) RMA” (GN69); Eye: “Hamilton Eye Institute Mouse Eye M430v2 Data Set (Sept08) RMA” (GN207); Hippocampus: “Hippocampus Consortium M340v2 (Jun06) RMA” (GN110). The number of biological specimens was 2–6 individuals per array, as specified for each dataset in the GeneNetwork database. The arrays were normalised with Robust Multichip Average (RMA), and intensities were log-transformed and standardised to a mean of 8 and a standard deviation of 2.

Differentially expressed (DE) genes were identified in each tissue using the Wilcoxon Mann-Whitney test (R v2.6.2 Environment for Statistical Computing http://www.r-project.org, wilcox.test(); standard approximation of p-values) at p<0.001 or p<0.01 (not adjusted for multiple testing), as described in the results section. Non-parametric (Wilcoxon) and parametric (t-test) tests showed similar results in number of DE probes. Significant DE X-linked probes using Wilcoxon (or t-test) were as follows for each tissue. Kidney: 287 (290), Liver: 176 (142), Lung: 48 (45), Striatum: 38 (28), Eye: 38 (35), Hippocampus: 25 (25). The Wilcoxon test was chosen to avoid violations of the assumption of normal distribution of probe intensities in each gene. The results for all probes (using both statistics) are available in Additional file 1 and Additional file 2. Since body weight and hormonal state were not controlled, the regulatory origin of any differential expression may be the compound effects of chromosomal constitution, hormonal differences and metabolic states between the sexes. The selection criterion for probes in Table 2was a consistent (defined as a significant deviation in the same direction) female-bias or male-bias (p<0.01) in more than one tissue. SignalMap v1.9.0.03 (NimbleGen) was used to visualise the distribution of sex-biased genes shown in Figure 2. The statistical significance of the observed over-/under-representation of chromosomal location for sex-biased genes was determined with a two-sided Fisher’s exact test and the significance criterion p<0.01. For the analysis shown in Figure 3B (X’), we excluded the following known escapees: Xist, Shroom4, Mid1, Kdm6a, Kdm5c, Jpx, Ftx, Eif2s3x, Ddx3x, Car5b, Bgn, BC022960, D930009K15Rik, 6720401G13Rik, 5530601H04Rik, 2610029G23Rik, and 1810030O07Rik 22 23 24 25 49 50 . The female mouse liver H3K27me3 23 and macroH2A1 24 data used in Figure 6 were downloaded from the Gene Expression Omnibus ( http://www.ncbi.nlm.nih.gov/geo, Accession: GSE20617: GSM517918, GSE18963: GSM469459), and SignalMap v1.9.0.03 (NimbleGen) was used to visualize enrichment. For the gene age analysis, branch assignment of sex-biased genes was set according to Zhang et al. (Additional file 2) 28 , matched by Ensembl ID, and the phylogeny in the upper section of Figure 5 was depicted accordingly (Zhang et al. 2010, page 2). The lower part of Figure 5 shows only those branches that were represented by at least 10 probes on the Mouse 430v.2 array. An apparent decrease in the proportion of sex-biased genes was observed around the split between branch 5 and 6. A two-sided Fisher’s exact test was applied to determine the statistical significance of the differences in the proportion of sex-biased genes assigned to branches before or after this split. GO analysis was performed in DAVID Bioinformatics Resources 6.7, NIAID/NIH 26 27 , using two reference sets: all X-chromosome genes in the Mouse 430v.2 array, or all genes included in the array, independent of chromosomal location, as specified in Additional file 3 A-E.

Mouse embryonic fibroblasts (MEFs) derived from skin were cultured in Minimum Essential Medium Eagle (Sigma) supplemented with 2% L-glutamine, 10% fetal bovine serum and 1.5% penicillin/streptomycin (Invitrogen). Upon reaching confluence, MEFs were subcultured 1:4 in Culturewell MultiWell cell culture system (Molecular Probes) and grown until reaching 80% confluence. Cultures were maintained at 37°C and 5% CO₂. Cells were fixed by 3% paraformaldehyde for 15 min at room temperature, followed by permeabilization in 0.5% TritonX-100 in PBS with 10 mM Ribonucleoside Vanadyl Complex (New England Biolabs) for 5 min. Fixed cells were stored in 70% ethanol at −20°C until usage. Probes (Xist: RP23-84A16, Tmem29: RP23-84G23, Tmem164: RP23-239E14, Kdm5c: RP23-202H16, Ctps2: RP23-106K4, Rbm41: RP23-381C12, 4732460I02Rik/Rbmx: RP23-44G23) were labeled with ChromaTide Alexa Fluor 488-5-dUTP (Molecular Probes) or Cy3-dCTP (Amersham) PA53021, using the BioPrime Array CGH Genomic Labeling system (Invitrogen). Hybridisation with labelled DNA (40 ng/μl) and mouse Cot1 DNA (100 ng/μl) (Invitrogen) was performed overnight at 38°C in 2×SSC, 50% formamide and 12% dextran sulfate, with 10 mM Ribonucleoside Vanadyl Complex. Cells were washed with 2×SSC and 50% formamide (3x5 min; 40°C) and 2×SSC (3x5 min; 40°C). Slides were mounted with Vectashield (Vector Labs). Cell imaging was carried out using a confocal microscope (Zeiss LSM 510 META, Göttingen, Germany). The images presented in Figure 4 were uniformly processed and merged in Adobe Photoshop CS3 (Adobe).