Background

Emerging evidence indicates that the regulatory pathways underlying sexual differentiation result in phylogenetically widespread transcriptional sex-bias in the brain of organisms ranging from D. melanogaster to humans [1]. We earlier uncovered an evolutionary conserved signature of sex-biased gene expression in the cortex of catarrhine primates including humans [2]. These results suggested that sexual differences present in adult human brain are in part genetically controlled and not solely attributed to environmental differences between the sexes. Furthermore, we demonstrated that several genes on the Y-chromosome are expressed in many regions of prenatal human male brain [3], raising the possibility that expression of Y-linked genes may be partially responsible for sexual dimorphism during early development in the human brain. Moreover, sex-biased expression is partly conserved between mouse and human, suggesting that there is a common mechanism of transcriptional modulation of sex-bias operating across mammals.

The mechanisms by which sex differences in the brain are established and controlled during development are not understood in detail, but sex hormones clearly play major roles. However, sex hormones are not sufficient to explain all sex differences, and genes encoded in the sex chromosomes are also known to be important [4]. These genes include male-specific Y-encoded genes as well as female-biased X-encoded genes that escape the dosage compensating inactivation of one X-chromosome in females. In mouse, only a few genes that escape X-inactivation have been identified [5–9]. These genes are known to produce female-biased expression very early during development [10], but the molecular control mechanisms remain mostly unknown.

A genome-wide investigation of whole adult mouse brain and several peripheral tissues demonstrated that sexual dimorphism in gene expression is highly tissue specific [11]. However, sex-biased expression in distinct regions of the brain has not been previously investigated. Since the brain is a highly heterogeneous and functionally compartmentalized structure, it is of interest to explore whether gene expression in distinct regions of the brain is regulated differentially in males and females. Indeed, three-dimensional magnetic resonance microscopy in mice demonstrated sex-specific, post-puberty changes in brain structures [12]. Furthermore, sex-specific functional differences in various brain regions, particularly the amygdala and hypothalamus, have been observed in both human and rodent [13–15]. Given the implications of sex-biases for many neurological and psychiatric disorders [16], the importance of studying sexually dimorphic gene expression in the brain is increasingly being realized.

In the present study, we investigated male and female gene expression in distinct structures of the mouse brain using genome-wide microarray analysis. The study unravelled sub-structure-specific sexual gene expression dimorphism in the mouse brain. Our findings also include novel observations regarding the regulation of domains and genes esca** X-inactivation on the mouse X-chromosome.

Results

Sexually dimorphic gene expression in distinct brain regions

To investigate possible sub-structure-specific as well as potentially uniform sex-biased gene expression in the adult mouse brain, we assessed genome-wide RNA expression in striatum, neocortex, hippocampus and eye from male and female mice using oligonucleotide microarrays. Lung was included in the analysis to allow comparisons with a peripheral tissue. We reasoned that sex-biased transcription present both in brain and in a peripheral tissue would indicate wide-spread expression bias. This in turn would suggest a sex-biased control mechanism for these genes operating throughout the body and not specifically in the brain. Two microarray platforms were used: Illumina 6v1.1 in case of striatum and neocortex and Affymetrix M430v2.0 in case of hippocampus, eye and lung. A total number of 456 microarrays were included in the study, comprising a balanced number of male and female samples in each tissue (Table 1, Methods).

Table 1 Identification of sex-biased genes in five tissues
Full size table

Genes with sexually dimorphic RNA expression were identified in striatum (n = 173), neocortex (n = 19), hippocampus (n = 12), eye (n = 31) and lung (n = 160) using FDR ≤ 0.05 as significance criteria (Table 1). The female to male fold changes in expression were statistically significant but small (between 1.1 and 2-fold) in terms of mean and median values for most genes (Figure 1, Additional file 1). Only ** DNA. Developmental biology. 2008, 319 (2): 416-425. 10.1016/j.ydbio.2008.04.010." href="/article/10.1186/1471-2164-11-614#ref-CR25" id="ref-link-section-d28674529e1265">25] and several genes previously known to escape X-inactivation, including Kdm5c, Eif2s3x, Kdm6a (formerly termed Utx) and Ddx3x. All X-linked genes named above were female-biased. We also detected male-biased expression of their Y-linked paralogous genes: Kdm5d, Eif2s3y, Uty and Ddx3y. In addition to previously reported sex-biased genes, we identified five novel X-linked genes with up-regulation in female tissues: 2010308F09Rik, D330035K16Rik, 5530601H04Rik, 2610029G23Rik, and D930009K15Rik ("Novel X, Y-linked", Figure 2B). On the Y-chromosome, we identified two novel male-biased genes: 4921530F17Rik, up-regulated in male neocortex and striatum and C030026M15Rik, up-regulated in male eye and lung tissues. Among the genes encoded in autosomal chromosomes, we did not identify any transcript that was sex-biased in all brain structures (Figure 2B). Two autosomal genes were however sex-biased in both striatum and neocortex: Prl (Prolactin, chr13, female up-regulated), a gene that promotes lactation, and Arid1b (AT rich interactive domain 1B, chr17, male up-regulated), a chromatin-remodelling factor. 1700012B15Rik (unknown function, chr12, female up-regulated) was identified in both eye and lung.

Figure 2
figure 2

Identification of genes with significant sex-biased expression in multiple tissues. A: The Venn diagrams show the number of overlap** sex-biased genes in tissues within each array platform. i: Illumina 6v1.1 and ii: Affymetrix M430v2.0. The total number of sex-balanced genes in each tissue is given within brackets and the number of shared sex-biased genes in the tissues is given in the intersecting areas. All genes in the overlap** sets are named in panel B, shaded with the same colours as in the Venn diagrams. B: Genes that were sex-biased in more than one tissue in the microarray experiments are shown. Numbers denote mean fold changes, with positive signs denoting female up-regulation and negative signs denoting male up-regulation. The genes are divided into three categories: Known X, Y-linked: Genes on the sex chromosomes that were previously described as sex-biased, including ** DNA. Developmental biology. 2008, 319 (2): 416-425. 10.1016/j.ydbio.2008.04.010." href="/article/10.1186/1471-2164-11-614#ref-CR25" id="ref-link-section-d28674529e2178">25]. Of particular interest was that when we investigated the chromosomal localizations of the novel female-biased lncRNAs, we observed that all of these non-coding genes are positioned close to female-biased coding genes on the X-chromosome (Figure 3A). Moreover, a closer examination revealed that three out of four female-biased lncRNAs are proximally paired with female-biased coding genes that are known to escape X-inactivation (Figure 3B). Indeed, 2010308F09Rik is located 33 kbp upstream of Ddx3x and these two genes are encoded in a "head to head" orientation; D330035K16Rik is positioned within an intron of Eif2s3x and D930009K15Rik is located 2.6 kbp downstream of Kdm5c. Furthermore, the remaining female-biased non-coding gene, 5530601H04Rik, is co-localized with the protein-coding and female-biased gene 2610029G23Rik, and these genes are situated in a "head to head" arrangement separated by 9.7 kbp (Figure 3B).

Figure 3
figure 3

Female-biased coding and non-coding genes are clustered on the X-chromosome. A: Chromosomal locations (NCBI37/mm9 genome assembly) of the known and novel X-encoded female-biased genes presented in Figure 2B. Known coding X-inactivation escapees are marked with "*". B: Higher resolution map, showing that coding and non-coding female-biased genes are proximally paired on the X-chromosome.

Full size image

Quantitative RT-PCR experiments confirmed the female-biased expression of the above mentioned pairs of X-linked coding and non-coding genes in brain, eye and lung tissues (p ≤ 0.05, two-sided unequal variance t-test) (Figure 4). One exception was 2010308F09Rik, not confirmed as female-biased in lung, but evidently female-biased in brain and eye. Female to male fold-changes were in the same range as detected in the microarray experiments, and in the expected range for X-inactivation escapee genes [32]. An X-linked negative control for sex bias, Rps4x, which is inactivated on the silent mouse X-chromosome [33, 34] and therefore not expected to show female-biased expression, remained unchanged. A positive control for female-bias, ** (Figure 5A-B). Moreover, DNA signals for the female-biased cluster and the **. C: Quantitative data showing allelic expression of the novel escapee cluster in nuclei positive for ** X inactivation. Trends Genet. 2003, 19 (8): 432-438. 10.1016/S0168-9525(03)00177-X." href="/article/10.1186/1471-2164-11-614#ref-CR48" id="ref-link-section-d28674529e2593">48]. Our results suggest that clusters of genes that escape X-inactivation also exist in mice, although the size of these domains is apparently smaller than in humans. Escape mechanisms may be regulating both coding and non-coding genes collectively within confined domains in the mouse X-chromosome. It is possible that the female-biased lncRNAs escape X-inactivation as a side effect of being located closely to functional protein-coding escapee genes. However, an alternative possibility is that the lncRNAs themselves serve a function in the esca** domains, and this possibility will be discussed in the following section.

Possible roles for long non-coding RNAs in domains that escape X-inactivation

The mechanisms that control escape from silencing on the inactive X-chromosome remain cryptic. Our observation of four chromosomal domains, each containing a female-biased coding and a non-coding gene, raises the possibility that lncRNA are themselves involved in the regulation of X-chromosome domains that escape inactivation. Although the functions of most lncRNAs remain enigmatic, there is increasing evidence suggesting roles in epigenetic regulation [30, 31, 49]. Recent reports showed that expression of lncRNAs can remodel local chromatin and promote transcriptional activation of closely located genes [50]. For example, the expression of fbp1 in yeast was shown to be mediated by expression of 5' located lncRNA transcripts, and accompanied by progressive opening of proximal chromatin [51]. Another example is the activation of yeast pho5, where the transcription of an intergenic, 3' located, antisense lncRNA contributed positively to chromatin plasticity and nucleosome disassembly [52]. Expression of lncRNAs in domains that escape X-inactivation might regulate the expression of neighbouring protein-coding genes by similar mechanisms, involving relaxation of nucleosome complexes, increased accessibility for transcriptional machineries and recruitment of transcription factors that activate transcription in the domain.

Other mechanisms of lncRNA control in X-inactivation esca** domains are possible. For example, lncRNAs may mediate movement of escapee domains to transcriptionally active compartments in the nucleus. Similar lncRNA-dependent movements were previously described, although towards transcriptionally silent compartments [53, 54]. Another described role for lncRNAs involved the establishment of chromosomal boundaries between transcriptionally active and inactive domains [55, 56]. By analogy, it is possible that transcription of lncRNA in esca** domains mediates the formation of chromatin boundaries to restrict the spread of heterochromatin from neighbouring silent regions in the X-chromosome, thereby ensuring escape of the co-localized protein-coding genes.

The proposed hypotheses could be tested in the future by mutation or silencing of the lncRNAs identified in our study.

X-linked lncRNAs are known to be involved in multiple steps of control of the binary transcriptional state of the X-chromosomes in mammalian females, including initiation, establishment and maintenance of X-chromosome silencing [57, 58]. The current report raised the possibility that lncRNAs might act as gene expression modifiers in domains that escape X-inactivation. Unlike proteins and small RNAs, lncRNAs can remain tethered to the site of transcription, and therefore uniquely direct allelic regulation [57]. This property could favour the choice of lncRNAs as regulators of sexually antagonistic genes on the X-chromosome during evolution. Selection for cis-acting modifiers of expression of sexually antagonistic genes on the X-chromosome is certainly predicted [59]. lncRNAs as local expression modifiers in domains that escape X-inactivation may very well prove to be a novel example of such cis-acting elements in the X-chromosome.

Sex chromosomes have evolved from autosomes independently many times in animals [60]. However, the main dilemma of dosage compensation and regulation of sexually antagonistic genes has been alike for each sex chromosome system [61]. It is possible that lncRNAs may have evolved as a solution for allele-specific regulation of confined domains in several types of sex chromosomes. The fast sequence evolution of non-coding RNAs [62] could make these molecules suitable for regulation of sexually antagonistic genes. To evaluate this possibility, the roles of lncRNAs in allele-specific regulation of genes encoded in the sex chromosomes should be studied not only in mouse but also in species with other sex chromosome systems.

Conclusions

Our data indicate that the degree of sex-biased gene expression varies between distinct brain structures in adult mouse brain. In particular, striatum showed wider transcriptional sex-bias than neocortex both in terms of number of sex-biased genes and chromosomal distribution of these genes. More interestingly, we identified female-biased lncRNA genes clustered on the X-chromosome with protein-coding genes that escape X-inactivation. These clusters are free of repressive histone marks, supporting the view that female-biased lncRNAs are biallelically expressed in domains that escape X-inactivation. Based on our observations, we propose that these lncRNAs might modulate the epigenetic state of nearby coding genes.

Methods

Identification of sex-biased genes

Microarray data: Microarray collections from discrete tissues were contributed by GeneNetwork (GN) (University of Tennessee). Details of the processing of tissues and the array data are described earlier [63–66], and summarised on the GN homepage http://www.genenetwork.org. In brief, Illumina Mouse-6v1.1 arrays (striatum and neocortex) were normalized with rank invariant normalization and Affymetrix M430v2 arrays (hippocampus and eye) were normalized with robust multichip average (RMA). Signals were log transformed and standardized to a mean of 8 and a standard deviation of 2. RMA normalization was also applied to the lung data (Affymetrix M430v2). Array data can be accessed via GN http://www.genenetwork.org. GN Striatum data set: "HQF BXD Striatum Illumina Mouse-6.1 November 2007 Rank Invariant Data Set", GN Accession: GN152. GN Neocortex data set: "HQF BXD Neocortex ILM6v1.1 (Feb08) RankInv", GN Accession: GN157. GN Hippocampus data set: "Hippocampus Consortium M430v2 (June06) RMA", GN Accession: GN110. GN Eye data set: "Hamilton Eye Institute Mouse Eye M430v2 Data Set (Sept08) RMA", GN Accession: GN207. GN Lung data set: "HZI Lung M430v2 (Apr08) RMA", GN Accesion: GN160. The analysis was restricted to include only C57BL/6 × DBA/2 mice. Prior to any further analysis, the data collection was screened for arrays in which sex had been potentially reversed in the databases, since such misclassifications would distort and reduce power of the sex-specific analysis. This was done by graphing the log2 intensity values of two **s t probes in each array and tissue in scatter plots. Male and female arrays in which **st expression fell in the expected quadrant of the opposite sex were excluded from the analysis (Additional file 3). The data sets were subsequently balanced to include an equal number of arrays from each sex. Identification of sex-biased genes: Sex-biased genes were identified in each tissue using the Wilcoxon Mann-Whitney test followed by Benjamini-Hochberg correction of the p-values (TIGR MeV v.4.3 package [67]). The nonparametric Wilcoxon test avoids possible violations of assumptions of normal distribution of signals in each particular probe-set, while the Benjamini-Hochberg correction hinders inflation of false positives in the resulting gene lists, which may otherwise be a caveat in microarray approaches. The criterion for differential expression was a Benjamini-Hochberg adjusted p-value lower or equal to 0.05. Both raw and adjusted p-values for each significant probe are listed in Additional file 1. Heatmaps were generated in TIGR MeV v.4.3 [67]. Gene ontology and disease association analysis, presented in Additional file 2, was performed using Ingenuity Pathway Analysis Software v8.6 (Ingenuity Systems). FDR of the observed co-localization of female-biased genes on the X-chromosome was calculated by considering **st, Jpx and the coding escapee genes (coding genes in Figure 3) as fixed and repeatedly drawing four random X-linked probes from the arrays (since four lncRNAs were identified). FDR was calculated as the number of times four drawn probes were located at a maximum of 50kb distance up-/down-stream of the fixed genes divided by the number of permutations (104).

Quantitative RT-PCR

Mouse tissues: Tissues were collected from female and male C57BL/6 × DBA/2 mice, and these samples were independent from the individuals used in generating the microarray data. Animals had been housed at Uppsala Biomedical Center, Sweden, in agreement with animal research ethical regulations (Swedish ethical committee permit: c79/9). Dissected brain (rostral part, including striatum, neocortex and parts of hippocampus, excluding olfactory bulbs), eye and lung tissues were snap-frozen on dry ice and subsequently kept at -80°C. RNA and cDNA: RNA was extracted using Trizol (Invitrogen) according to the manufacturer's instructions. To inspect RNA quality, 28 S and 18 S ribosomal bands were examined by gel electrophoresis, and optical 260/280, 260/230 ratios were measured (ND-1000, NanoDrop Technologies) ranging 2.0-2.2. RNA was reversely transcribed to cDNA using a Dynamo cDNA synthesis kit F-470L (Finnzymes) and the following reagents: 0.95 μg total RNA, 15 ng/μl random hexamers, 10 U/μl M-MuLV RNase H- reverse transcriptase, 1 × RT buffer, nuclease-free water, in a total reaction volume of 20 μl. Incubations were performed in a PTC-100 Peltier Thermal Cycler (MJ Research): 25°C; 10 min, 37°C; 45 min, 85°C; 5 min. cDNA samples were subsequently diluted 1:10 in double distilled water and kept at -20°C until usage. Primers: Transcript-specific primers (Additional file 4) were designed on opposite sides of exon/exon junctions using Primer3 [68]. Since D330035K16Rik and D930009K15Rik lack exon/intron structures, intra-exonic primers were used for these transcripts. Multiple primer pairs were tested for each transcript, and primers with the highest PCR efficiencies were selected for the experiments. qPCR: Quantitative PCRs were performed in a ABI Prism 7000 Sequence Detection System (Applied Biosystems). Reactions contained 0.3 μM of each primer, 1 × Power SYBR Geen Master Mix (Applied Biosystems), 4 μl cDNA sample and distilled water in a total reaction volume of 30 μl. Thermal cycles were: 50°C; 2 min, 95°C; 10 min, 40 cycles: 95°C; 15 s, 60°C; 1 min. To ensure that single PCR products of intended lengths were amplified, a melting program was executed subsequent to the quantifications and PCR products were separated by gel electrophoresis. To control for potential signals from amplification of genomic DNA (gDNA) during quantification, minus RT controls were employed (reverse transcription without M-MuLV RNase H- reverse transcriptase). gDNA signals were consistently undetectable or negligible. Data analysis: Background-subtracted expression values (copy numbers) were determined relative to a standard curve (cDNA dilution series) in 7000 SDS v.1.2.3 (Applied Biosystems). Expression values in each sample were normalized to the geometric mean of the endogenous expression of Actb and Gapdh. The criterion for differential expression was p ≤ 0.05, two-tailed unequal variance t-test (R v2.6.2 Environment for Statistical Computing http://www.r-project.org, t.test(), Arguments: alternative = "two.sided", var.equal = FALSE).

Sequential RNA - DNA FISH

Cells: Adult mouse fibroblast cultures were prepared as previously described [69] with minor modifications. Briefly, female mice were decapitated and the tails were cut and washed in PBS. After removing superficial dermis, remaining tails were cut into 2-3 mm pieces and placed into gelatin-coated 6 well plates (Costar) containing 1 ml medium in each well. Culture medium was composed of Dulbecco's Modified Eagle Medium (DMEM) containing 4.5 g/l glucose, 10% fetal bovine serum and 0.5% penicillin/streptomycin (Gibco). Tails were incubated at 37°C and 5% CO2 for 5 days, during which fibroblasts migrated out of the explants. Tissue pieces were then removed and cells were cultured in fresh medium until they reached confluence. RNA - DNA FISH: Mouse fibroblasts were passaged and grown on Culturewell™ MultiWell cell culture system (Molecular Probes) for 36 hours, and then 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. Sonicated BAC DNA (5530601H04Rik and 2610029G23Rik: RP23-149J5, **st: RP23-84A16) was labelled with Green-dUTP or Orange-dUTP (Abbott Molecular) using the BioPrime Array CGH Genomic Labeling system (Invitrogen). Hybridization with labeled DNA (10 ng/μl) and mouse Cot1 DNA (100 ng/μl) (Invitrogen) was performed overnight at 37°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 (15 min; 40°C) and 2 × SSC (15 min; 40°C). DNA was counterstained and cells were mounted in Vectashield (Vector Labs). Cell imaging and generation of optical sections in 3D were carried out on Leica DMI 3000B. After image acquisition, the slide with cells were washed two times by 4 × SSC with 0.05% Tween20 (10 min; 40°C), followed by treatment with 10 ng/μl RNase A (QIAGEN) at 37°C for 1 h. Cells were then denatured in 2 × SSC and 50% formamide at 80°C for 40 min. DNA FISH hybridizations were carried out with labelled DNA (10 ng/μl) and mouse Cot1 DNA (100 ng/μl) (Invitrogen) in 2 × SSC, 50% formamide and 12% dextran sulfate overnight at 37°C, followed by the same washing and mounting steps as described for RNA FISH. DNA FISH images, of cells with recorded RNA FISH signals, were acquired in the same way. The images presented in Figure 5 were uniformly processed and merged in Adobe Photoshop CS3 (Adobe).

H3K27me3 ChIP-chip data

H3K27me3 data from female and male adult mouse liver [43] was downloaded from NCBI Gene Expression Omnibus (GEO accession: GSE20617, data sets: GSM517918_female_liver _H3K27me3_peaks, GSM517917_male_liver_H3K27me3_peaks). LiftOver (UCSC Genome Browser) and SignalMap v1.9.0.03 (NimbleGen) was used to visualize H3K27me3 enrichment shown in Figure 6.