An integrated resource for genome-wide identification and analysis of human tissue-specific differentially methylated regions (tDMRs)

Abstract

We report a novel resource (methylation profiles of DNA, or mPod) for human genome-wide tissue-specific DNA methylation profiles. mPod consists of three fully integrated parts, genome-wide DNA methylation reference profiles of 13 normal somatic tissues, placenta, sperm, and an immortalized cell line, a visualization tool that has been integrated with the Ensembl genome browser and a new algorithm for the analysis of immunoprecipitation-based DNA methylation profiles. We demonstrate the utility of our resource by identifying the first comprehensive genome-wide set of tissue-specific differentially methylated regions (tDMRs) that may play a role in cellular identity and the regulation of tissue-specific genome function. We also discuss the implications of our findings with respect to the regulatory potential of regions with varied CpG density, gene expression, transcription factor motifs, gene ontology, and correlation with other epigenetic marks such as histone modifications.

Figure 1.

(A) Schematic description of Batman. The left panel shows five hypothetical genomic regions of varying CpG densities and number of methylated CpG sites (filled and empty circles represent methylated and unmethylated CpG sites, respectively). As MeDIP enrichment is proportional to the number of methylated CpG sites, the normalized enrichment ratios of these five hypothetical regions, shown in the second panel, will not accurately reflect the absolute methylation levels at the genomic region of interest (ROI). Batman is based on the observation that the log-ratio MeDIP signal of methylated DNA scales linearly with the number of methylated CpG sites in a sequence. We use a Bayesian deconvolution strategy, taking into account the estimated distribution of DNA fragment lengths, to find the most likely configurations of methylated and unmethylated CpGs in a sequence that explains the observed MeDIP signals. This allows estimation of absolute methylation levels at the ROI. Yellow, green, and blue represent unmethylated, intermediately methylated, and methylated regions, respectively. Batman is described in detail in Down et al. (2008). (B) Integration of Batman-called methylation values into the Ensembl genome browser—screenshot of the data integrated into the Ensembl genome browser (www.ensembl.org). The web display uses a color gradient to show the Batman methylation score for each of the probes in the ROI. The color represents the value of the probe on a sliding scale from 20 (bright yellow) to 80 (dark blue). Probes with Batman values of less than 20 or greater than 80 are colored with the maximum and minimum shades to increase the contrast in the display. Each tissue-type is configured as a dedicated DAS source, allowing the user to select any possible subset of tissues for viewing. Users clicking on a probe will see a small pop-up window, which displays the exact chromosome position of the probe and the Batman score.

Figure 2.

Analysis of somatic DNA methylation profiles. (A) Distribution of data with respect to CpG_o/e for the different genome feature categories. The data were operationally categorized into unmethylated (<40%), intermediate (40%–60%), and methylated (>60%). Within the nonpromoter categories, we focused predominantly on CpG islands as annotated in the Ensembl genome browser (NCBI_36). Therefore, the average CpG_o/e within the nonpromoter categories in our data set was higher than the genome average (refer to Table 1). However, because probes could not be chosen for all nonpromoter CpG islands, we also randomly selected some CpG-poor nonpromoter regions, and hence, a bimodality of CpG_o/e is also observed in some of the nonpromoter categories. Methylation data used in these plots are from whole blood. (B) Comparison of promoter DNA methylation with gene expression across a range of promoter CpG_o/e. Whole-blood DNA methylation data (only ROIs overlapping the TSS were used) was correlated with whole-blood genome-wide expression profiles obtained from the GNF SymAtlas database (Su et al. 2004). There were insufficient data for intermediately methylated promoters in the CpG_o/e = 1.2 category, and methylated promoters in the CpG_o/e ≥ 1 categories. The color code is the same as in A, and error bars represent 95% confidence intervals. (C) Gene expression levels for ICAM3 were obtained from a public database (Su et al. 2004). Expression values represent average difference values computed by Affymetrix software. These values are proportional to mRNA content in the sample. (D) Correlation of DNA methlyation with H3K4me3, H2A.Z, RNA PolII, and CTCF enrichment. DNA methylation data (500 bp ROIs) from our study were correlated with genome-wide enrichment profiles for 20 histone lysine and arginine methylations, H2A.Z, RNA PolII, and CTCF generated by Barski et al. (2007) using Illumina 1G sequencing technology. The remaining 19 comparisons are presented in the Supplementary section. The X-axes represent CpG_o/e (there were insufficient data to stratify by CpG_o/e in the nonpromoter categories), the Y-axes DNA methylation levels, and the grayscale represents the average tag count for the histone modification or protein indicated. The exon and intron categories were combined into a single “genic” category. Hatched regions indicate that insufficient data were available.

Figure 3.

Analysis of tissue-specific differentially methylated regions (tDMRs). (A) The top half of each panel shows the DNA methylation profiles observed in sperm. In the bottom half of each panel, bars above the line represent the proportion of ROIs in each CpGo/e category that display <40% methylation in sperm, but >60% methylation in all somatic tissues (gray) or >60% methylation in one or more (but not all) somatic tissues (red, i.e., these are somatic tDMRs). Bars below the line (i.e., negative values) represent the proportion of ROIs in each CpGo/e category that display >60% methylation in sperm, but <40% methylation in all somatic tissues (gray) or <40% methylation in one or more (but not all) somatic tissues (red, i.e., these are also somatic tDMRs). (B) Comparison of tissue-specific DNA methylation and gene expression. (Left) Comparison of promoter DNA methylation (only ROIs overlapping the TSS were used) and gene expresson between whole blood and uterus. Gene expression data are from GNF SymAtlas database (Su et al. 2004). Insufficient data were available for CpG_o/e > 0.8. Yellow bars represent genomic regions that display <40% methylation in whole blood and >60% methylation in uterus. Blue bars represent genomic regions that display >60% methylation in whole blood and <40% methylation in uterus. (Right) Comparison of gene-body methylation with gene expression of the associated gene. Intronic and exonic methylation data were combined for this analysis. All ROIs in these categories were used, not just CpG-dense ROIs, but we did not stratify by CpG_o/e since there were insufficient numbers of exonic/intronic CpG-poor ROIs. Both figures show 95% confidence interval. The color code is the same as at left (promoter). Pairwise comparisons of other tissues showed similar results (data not shown). (C) Known transcription factor motif analysis of tDMR promoters. We used the JASPAR database (http://jaspar.genereg.net) to search for over-represented transcription factor binding sites in tDMRs. (+) tDMRs that are hypermethylated in the tissue of interest; (−) tDMRs that are hypomethylated in the tissue of interest.