Gender is predicted by DNA methylation patterns in the mouse liver
We generated snapshots of genome-wide cytosine methylation patterns with reduced representation bisulphite sequencing (RRBS). RRBS captures only ~1% of the mammalian genome, but the captured fraction is highly enriched for functional regions such as CpG islands, shores, and other gene regulatory elements. Here we used an enhanced RRBS protocol to produce methylomes from the livers of six female and six male isogenic C57BL/6 J mice. Overall, data quality was very high, as were inter-sample correlations (Additional file 1: Figure S1). For all analyses we considered only those CpGs that were present at >10× coverage in all 12 samples. To combine statistical evidence of neighbouring CpGs, we calculated DNA methylation levels in tiles of 100 bp across the genome (resulting in 820,388 tiles covered across all 12 samples).
Unsupervised hierarchical clustering separated samples by gender, even when data from sex chromosomes was removed (Fig. 1a). Principal components analysis (PCA) also showed a distinct spatial clustering of samples by gender irrespective of whether sex chromosomes were included (Fig. 1b) or not (Fig. 1c). We identified regions of difference between genders using methylKit [10]. We found 1093 tiles that were differentially methylated (q-value < 0.01 and ≥ 25% methylation difference) between males and females (Fig. 1d; Additional file 2: Table S1). Despite the huge enrichment for CpG islands in our RRBS libraries, the majority of differentially methylated tiles (DMTs) were found outside of CpG islands, in both genic and intergenic regions (Fig. 1e). Around a third of the DMTs overlapped with ENCODE liver H3K4me1 and H3K27Ac peaks, a highly significant enrichment (both p < 1 × 10−4). This is consistent with many of the intergenic DMTs residing in active enhancers.
We chose a subset of ten DMTs to validate by COBRA [11]. All but one of the ten loci showed a difference in methylation levels between males and females as predicted by the RRBS signal (Additional file 3: Figure S2). This experimental validation indicates that our RRBS and informatics strategy detects gender-specific differences with high confidence.
Gender DMTs in the liver can be testosterone independent
We next reanalysed data from Reizel et al.. [9] (GEO Accession GSE60012), who previously identified 160 gender DMTs in the adult mouse liver. Using our bioinformatics pipeline with their liver dataset, we identified 83 autosomal gender DMTs, of which 48 (~58%) overlapped with the gender DMTs from our data (Fig. 2a, Additional file 2: Table S2). The difference in the number of DMTs identified can be at least partially attributed to the difference in dataset size (our liver dataset contained 820,388 tiles, and Reizel et al’s contained 167,462 tiles). Like Reizel and colleagues, we found that the majority of gender DMTs in the liver could be attributed to hypermethylation in females relative to males; this was true across all autosomes (Fig. 2b). Interestingly, the gender DMTs identified in the liver by Reizel et al. were absent from the liver of males who were castrated when young, and could be reconstituted in castrated males with exogenous testosterone administration [9]. This implies that gender DMTs in the liver are testosterone-dependent. We extended these observations by performing an unbiased comparison of the Reizel et al. females with the castrated males, using our own informatics pipeline. In doing so we were able to identify 228 gender DMTs, despite the absence of testosterone in the males (Additional file 2: Table S3). These testosterone-independent gender DMTs were also heavily skewed towards hypermethylation in females (Fig. 2c). Only four of these DMTs are also found in our set of 1093 DMTs (Fig. 2d), although almost all the tiles in the castrated male dataset were also represented in our dataset. This suggests that there are additional factors beyond testosterone that are able to specify gender-specific methylation patterns.
Heart and brain also harbour gender DMTs
The robust gender differences observed in the mouse liver in this study and others, along with reports of gender bias in human non-liver tissues such as peripheral blood leukocytes, prompted us to ask whether the gender DMTs might be conserved across tissues. Given that some DMTs in liver appear to be sex hormone-independent, it is possible that these differences in methylation arise in the germline, or very early in development. To address this possibility we interrogated the methylomes of brain and heart from three of the males and three of the females used for liver analysis. We chose these tissues as they derive from different germ layer origins (heart, mesoderm; brain, ectoderm; liver, endoderm). Unlike the liver methylome, analysis of the brain methylome by hierarchical clustering or PCA did not separate the genders, even when sex chromosomes were included (Fig. 3a); likewise in the heart we observed no distinct clustering, although the genders separate very slightly on the first principal component (PC1; Fig. 3b). Despite the lack of unsupervised clustering by gender, both heart and brain harboured gender DMTs. Applying the same stringent parameters as used for the liver analysis, we identified 957 autosomal DMTs in the brain (Fig. 3c; Additional file 2: Table S4), and 145 in the heart (Fig. 3d; Additional file 2: Table S5). While the brain gender DMTs exhibited, like the liver, a clear bias towards hypermethylation in females (Fig. 3f), this bias was absent from the gender DMTs in the heart (Fig. 3g). The gender DMTs in both tissues were again mostly outside CpG islands, in intronic and intergenic regions (Fig. 3e). Like the liver DMTs, the brain DMTs were significantly associated with enhancer regions (p < 1 × 10−4), albeit with less fraction of overlap (~5% of DMTs); but the heart DMTs were not (p = 0.74).
Gender DMTs are tissue specific
We then asked whether there was any overlap of the gender DMTs we identified among tissues of distinct germ layer origin that might suggest that they were inborn. Despite the large number of DMTs identified across the three tissues in total we found almost no overlap (Fig. 4a). Only two liver gender DMTs were common to the brain, and no liver DMT was common to the heart (although 11 of the 145 heart DMTs were also differentially methylated in the brain). We determined functional pathways associated with the DMTs in each tissue; in most cases, significantly enriched pathways were functionally related to the relevant tissue (Fig. 4b). This may not be surprising given that the DMTs tend to be over-represented in enhancer regions. PCA of all of our RRBS data across all tissues confirmed that methylation patterns overall are tightly associated with tissue type (Fig. 4c); it is interesting to note that methylation patterns in the brain are much more variable among isogenic individuals than the other tissues examined.
While this analysis suggests that gender DMTs do not arise in the germline, we considered whether gender DMTs might be common to tissues of the same germ layer origin (i.e. set early in development). We identified DMTs from additional mesodermal tissues studied by Reizel et al. (skeletal muscle and spleen from adult mice) to compare with our mesodermal (heart) tissue. We identified 354 and 48 gender DMTs in muscle and spleen respectively (Additional file 2: Tables S6 and S7), but these DMTs were exclusive to both each other and to the heart (Fig. 4d); this is despite almost all tiles from the muscle and spleen datasets being represented in our heart dataset. Taken together, these comparisons indicate that differential cytosine methylation between genders is almost entirely tissue-specific.