DNA methylation of the first exon is tightly linked to transcriptional silencing

Abstract

Tissue specific patterns of methylated cytosine residues vary with age, can be altered by environmental factors, and are often abnormal in human disease yet the cellular consequences of DNA methylation are incompletely understood. Although the bodies of highly expressed genes are often extensively methylated in plants, the relationship between intragenic methylation and expression is less clear in mammalian cells. We performed genome-wide analyses of DNA methylation and gene expression to determine how the pattern of intragenic methylation correlates with transcription and to assess the relationship between methylation of exonic and intronic portions of the gene body. We found that dense exonic methylation is far more common than previously recognized or expected statistically, yet first exons are relatively spared compared to more downstream exons and introns. Dense methylation surrounding the transcription start site (TSS) is uncoupled from methylation within more downstream regions suggesting that there are at least two classes of intragenic methylation. Whereas methylation surrounding the TSS is tightly linked to transcriptional silencing, methylation of more downstream regions is unassociated with the magnitude of gene expression. Notably, we found that DNA methylation downstream of the TSS, in the region of the first exon, is much more tightly linked to transcriptional silencing than is methylation in the upstream promoter region. These data provide direct evidence that DNA methylation is interpreted dissimilarly in different regions of the gene body and suggest that first exon methylation blocks transcript initiation, or vice versa. Our data also show that once initiated, downstream methylation is not a significant impediment to polymerase extension. Thus, the consequences of most intragenic DNA methylation must extend beyond the modulation of transcription magnitude.Sequencing data and gene expression microarray data have been submitted to the GEO online database (accession number SRA012081.1). Supporting information including expanded methods and ten additional figures in support of the manuscript is provided.

Figure 1. Overview of the STAMP assay and representative data.

(A)The steps involved in the preparation of a methylated DNA library for massively parallel sequencing is illustrated schematically. 1) Genomic DNA is purified from cells. 2) The DNA is then randomly sheared by sonication. 3) Fragmented DNA containing methyl-CpGs, indicated as red spheres, is purified using His-MBD beads. 4) Bound DNA is purified and ligated to SOLiD sequencing adapters. The resulting DNA library is subsequently sequenced and mapped to the genome of interest. These sequence tags inform subsequent analysis of DNA methylation patterns. (B) Sequence tag maps and methylation profile is shown at the CDKN2B locus for the AML cell line M091. Upper, red vertical bars and lower, green vertical bars represent individual sequence tags mapping to the sense and antisense strands, respectively. The dashed red and green lines represent the methylation signal for the top strand and bottom strand, respectively. The black line represents the composite STAMP signal. A light blue box surrounds each densely methylated element (DME). CpG Islands are shown as green boxes below the plot and the gene body is indicated schematically. The location of the PCR amplicon used for bisulfite qPCR is indicated by a blue box. (C) STAMP analysis at the GAPDH locus, as described for panel (B), shows no methylation at this locus. STAMP analysis at the CDKN2B and GAPDH loci was confirmed by (D) Bisulfite qPCR (Methylight) (see Supporting Information S1, Table 1). In this panel, the fraction of total DNA present (assessed using methylation-insensitive primers) that is detected as methylated is shown. (E) STAMP analyses of biological replicate cultures of the AML cell line is shown in the left panel scattergram. A STAMP signal (log scale) for the replicates was calculated at 15,000 randomly selected loci. The red and grey dashed lines represent unchanged and two-fold changed signal The right panel compares one of the replicates to sequence tags obtained from unenriched DNA from the same cell line and demonstrates that the high replicate correlation depends upon His-MBD enrichment.

Figure 2. Patterns of gene cassette methylation in T cells.

Each gene cassette element was classified as unmethylated (lowest 10% methylation quantile) or methylated (top 90% methylation quantile). The odds ratio (log2 transformed) indicates the likelihood of an element being methylated if the gene body (A), promoter (B), TSS (C), first exon (D), any intron (E), any internal exon (F), last exon (G) or TTS (H) is methylated. Odds ratios, calculated using Fisher's exact test for count data, represent a conditional maximum likelihood estimate quantifying the strength of the correlation between methylation of each gene cassette element. The odds ratio for the autocorrelation of each element is infinite and represented by grey boxes. Representative data is shown for human blood T cells but the pattern is the same in granulocytes and AML cells.

Figure 3. Genomic distribution of Densely-Methylated Elements (DMEs).

(A) The fraction of the genomic DME span overlapping the indicated UCSC genome browser annotation tracks is shown for normal human T cells (blue bars), granulocytes (green bars) and an AML-derived cell line (M091) that was treated with decitabine (orange bars) or left untreated (pink bars). Gene, Promoter, TSS, TTS, Exon, Intron, Exon-5′, Exon-3′, and Exon-In represent the entire gene body, the region −1000 bp upstream of the TSS, the 500 bp surrounding the TSS or TTS, all exons, all introns, and the first, last or middle exons, respectively. Also annotated are CGIs, the most conserved genomic elements (Consd), various repeat classes and a group of random genomic loci comprising 10% of the genome (Random). The sum of the DME fractions is greater than one because DMEs may hit more than one annotation due to overlap of some genomic annotations. (B) The proportion of the genome allocated to each annotation (left panel) is compared to the proportional size of DMEs within each annotation for T cells (right panel) and for the AML-derived cell line (lower panel). For clarity, gene bodies are excluded. (C) The fraction of the annotation span overlapping DME is shown for normal human T cells (blue bars), granulocytes (green bars) and an AML-derived cell line that was treated with decitabine (orange bars) or left untreated (pink bars), as described for (A). (D) The log-odds ratio for the extent of DME overlap compared to that expected from the relative genomic span of the annotation is shown as described for (A).

Figure 4. Sequence characteristics of Densely-Methylated Elements (DMEs).

(A) Density plot of GC fraction (fG+fC) vs CpGoe, the observed/expected CG fraction, fCG/(fC*fG) for a 200 bp window surrounding each CpG dinucleotide in the genome. (B) The GC fraction vs. CpGoe is plotted for each annotated CGI in the genome. CGIs are partially defined by GC>0.5 and CpGoe>0.6. (C) The sequence characteristics of DMEs are plotted. DMEs are enriched for regions with moderate CpGoe. (D) The distribution of DME lengths is shown along with dashed red lines representing the 5^th (260 bp) and 95^th (2140 bp) percentiles. The median length is 590 bp.

Figure 5. STAMP signal at the TSS and TTS for refSeq genes (refGene).

Composite density plots reveal the pattern of STAMP methylation surrounding all TSS (A) or TTS (B) in M091 cells. STAMP signal was calculated for each TSS or TTS flanked by ∼15 kb. Data are shown for both untreated cells (green line) and decitabine-treated cells (red line). (C) A heatmap representing the STAMP signal surrounding the TSS is shown for genes with a profile highly correlated to the composite density (A). Rows ordered by the location of mode and blue level is proportional to STAMP methylation signal. Each row represents an individual gene and columns represent distance from TSS as indicated. (D) A STAMP signal heatmap was generated for TTS as described in panel (C). Genes with a profile most similar to the composite density (B) are shown with rows ordered by STAMP signal. (E) Heatmap genes with poor correlation to the composite TSS density, as described for panel (B). (F) The correlation of each refGene to the composite TSS profile is plotted against the STAMP signal density (signal per bp) in the 1 kb surrounding the TSS. This plot demonstrates that refGenes with high signal near the TSS have a methylation pattern that is highly correlated to the composite profile shown in (A).

Figure 6. Correlation of transcript expression with the pattern of intragenic methylation in M091 cells.

(A) The composite density plot of methylation surrounding the TSS is shown for all transcripts (solid black line) and for transcripts that are in various expression quantiles (dashed colored lines). Methylation for transcripts in the lowest 10% expression quantile (red) is significantly higher than for genes that are within the next 15% expression (orange). Transcripts with higher expression have even less methylation. (B) The ratio of first exon to promoter methylation is shown for transcripts in the lowest 10% (red), 10–25% (orange), 25–50% (green), 50–75% (cyan), 75–90% (blue) and 90–100% (magenta) expression quantiles. (C) Intragenic STAMP methylation is shown for transcripts in each of 10 expression quantiles (lowest 10% to 100%). Box plots of methylation within the promoter (white), first exon (green), any intron (grey), internal exons (cyan) and the last exon (blue) are shown as indicated. (D) Schematic of the promoter and intragenic elements is shown with the color code utilized in (C) and (E). (E) The odds ratio (log2 transformed), calculated using Fisher's exact test for count data, shows the likelihood of a transcript being expressed (greater than the lowest 10% expression quantile) if it is methylated (top 90% methylation quantiles) within the indicated component of the gene cassette.

Figure 7. Decitabine induced hypomethylation of the first exonic region is associated with transcriptional activation.

(A) Profiles of the composite methylation density surrounding the TSS of transcripts upregulated by decitabine are shown before (red) and after (green) decitabine treatment of M091 cells. For comparison, the composite methylation profile of all genes is shown as in Fig. 5A (grey). Also show is the change in methylation seen for transcripts that are upregulated (blue dashed line) or unchanged (orange dashed line) following decitabine treatment. (B) Profiles of the composite methylation density surrounding the TSS of transcripts downregulated by decitabine are shown before (red) and after (green) decitabine treatment. For comparison, the composite methylation profile of all genes is shown as in Fig. 5A (grey). Also show is the change in methylation seen for transcripts that are downregulated (blue dashed line) or unchanged (orange dashed line) following decitabine treatment.