Tet1 is required for maintaining 5hmC levels at defined genomic regions

We next mapped the 5hmC peaks to all annotated RefSeq genes in the mouse genome. We found that 5hmC was preferentially enriched in Tet1-bound genes as compared with Tet1-unbound genes (Fig. 2A), consistent with the known enzymatic activity of Tet1. Further analyses of 5hmC peaks within regions flanking transcriptional start sites (TSSs) of annotated genes indicated that 5hmC levels were high in genomic regions flanking CpG-rich proximal promoters and within Tet1-bound CpG-poor promoters (Supplemental Fig. S3A). Mapping of 5hmC peaks to Tet1-bound sites also supported the observation that 5hmC tended to be more enriched in genomic regions with medium levels of CpG density (Supplemental Fig. S3B,C). In fact, 5hmC appeared to be excluded from Tet1-binding sites with high CpG density (Supplemental Fig. S3C). The lack of high levels of 5hmC within these CpG-rich regions may possibly be explained by two nonmutually exclusive mechanisms: (1) Tet1 is capable of rapidly hydrolyzing 5mC into 5hmC, which in turn is converted to unmethylated cytosine by yet-to-be-identified downstream enzymes within CpG-rich regions. (2) High levels of trimethylated H3K4 (H3K4me3) within CpG-rich regions prevent efficient binding of the de novo DNA methyltransferase complex Dnmt3a2/Dnmt3b/Dnmt3l in mouse ES cells to these DNA sequences (Ooi et al. 2007), and thus inhibit accumulation of 5mC (Supplemental Fig. S3C).

Open in a separate window

Figure 2.

Tet1 is required for maintaining 5hmC levels at defined genomic regions in ES cells. (A) Distribution of 5hmC relative to all annotated genes in ES cells. Averaged 5hmC enrichment (measured by −log₁₀ peak P-value) in 200-base-pair (bp) bins upstream of/downstream from gene bodies or at 5% intervals within the gene body is shown along the transcription units from 5 kb upstream of TSSs to 5 kb downstream from the transcriptional end sites (TESs). Note that 5hmC levels are generally enriched in Tet1-bound genes as compared with Tet1-unbound genes. (B) Changes in 5hmC and 5mC enrichment (measured by −log₁₀ peak P-value) in response to Tet1 knockdown are shown for Tet1-bound regions associated with different genomic features (gene body, intergenic region, and promoter [2 kb flanking TSSs]). (C) Heat map representation of CpG islands and occupancy of Tet1, 5mC, and 5hmC in mouse ES cells at all Tet1-enriched regions (5 kb flanking the center of Tet1 peaks). The heat map is rank-ordered by CpG density of genomic regions within 500 bp flanking the center of Tet1 peaks. The enrichment of 5hmC and 5mC was determined by whole-genome tiling microarrays. Tet1-bound regions at gene bodies, promoters, and intergenic regions are shown separately. The enrichment of Tet1 binding was determined previously by ChIP-seq analyses (Wu et al. 2011). All average binding was measured by −log₁₀ (peak P-values) in 200-bp bins and are shown by color scale. The following color scales (white, no enrichment; blue, high enrichment) are used for 5hmC, 5mC, and Tet1, respectively: [0, 2], [0, 0.5], and [0, 50]. The presence of CpG islands is displayed in color (blue, present; white, absent). (D) Tet1 occupancy and changes in 5hmC/5mC levels are shown for a group of representative Tet1 targets (Pcdha gene cluster on chr18) in control (Con) and Tet1 knockdown (Tet1 KD) ES cells. Tet1 ChIP-seq data in control knockdown (Con KD) and Tet1 knockdown (Tet1 KD) are shown in read counts, with the Y-axis floor set to 0.2 read per million reads. 5hmC and 5mC levels are shown as log₂ ratios of immunoprecipitation/input (IP/input).

To further investigate the dependence of 5hmC distribution on Tet1 occupancy, we performed lentivirus-mediated knockdown of Tet1 in mouse ES cells (Ito et al. 2010). The results shown in Figure 2B demonstrate that knockdown of Tet1 resulted in reduced 5hmC levels at Tet1 regions throughout the genome that include Tet1-bound promoters, gene bodies, and intergenic regions. Many Tet1-binding sites, particularly those at CpG-rich proximal promoters, did not show a decrease in 5hmC level in the absence of Tet1 proteins (Fig. 2B,C; Supplemental Fig. S4A,B). This is likely due to a lack of 5hmC enrichment at these sites in wild-type mouse ES cells (Fig. 2C; Supplemental Fig. S3C). However, an increase in 5mC levels was still frequently observed within both promoter and nonpromoter Tet1-binding sites (Fig. 2B–D; Supplemental Fig. S4A), indicating that Tet1 has a role in maintaining a DNA hypomethylated state at these sites. Taken together, these results indicate that Tet1 is required for establishing a defined genomic pattern of 5hmC, and also for initiating an enzymatic cascade to maintain CpG-rich gene promoters in a DNA hypomethylated state.

5hmC is enriched in both Tet1-activated and Tet1-repressed genes

Previous genome-wide analyses have shown that distribution of DNA methylation in the genome may also be regulated by histone modifications (Ooi et al. 2007; Schlesinger et al. 2007; Mohn et al. 2008). For example, H3K4me3 at proximal promoters generally regulates DNA methylation levels in a negative fashion (Weber et al. 2007; Meissner et al. 2008), whereas H3K36me3 within actively transcribed gene bodies is positively correlated with the presence of high levels of DNA methylation (Ball et al. 2009). To further investigate the relationship between Tet1, 5hmC, and major histone modifications at Tet1-bound genes, we cross-referenced the 5hmC profile with published genome-wide occupancy of major histone modifications. As Tet1 can bind to both actively transcribed genes and Polycomb repression complex 2 (PRC2)-repressed developmental regulators (positive for Ezh2 and H3K27me3) (Wu et al. 2011), we analyzed these two groups of Tet1-bound genes separately (Fig. 3A). This analysis revealed that 5hmC was relatively more enriched at intragenic regions (Fig. 3B,C), particularly at the 3′ end of the gene body for actively transcribed Tet1-only targets (e.g., Rest in Fig. 3C), similar to the transcription elongation mark H3K36me3 (Mikkelsen et al. 2007). In contrast, enrichment of 5hmC was more prominent at extended promoter regions—including both upstream of and downstream from TSSs (Fig. 3B,C)—of Tet1/PRC2-cobound targets (e.g., Lhx2 in Fig. 3C). Thus, 5hmC enrichment at Tet1-bound genes may contribute to maintenance of both the transcriptionally active and inactive chromatin states by functionally interacting with distinct histone modifications and their associated proteins.

Open in a separate window

Figure 3.

5hmC is enriched in both repressed (bivalent, Tet1/PRC2-cobound) and actively transcribed (Tet1-only) genes. (A) Heat map representation of genomic regions with enriched 5hmC, binding profile of Tet1 and Ezh2, and major histone modifications (H3K4me3, H3K27me3, and H3K36me3) (Mikkelsen et al. 2007) in mouse ES cells at all Tet1 target genes (5 kb flanking TSSs). The heat map is rank-ordered from genes with the highest H3K27me3 enrichment to no H3K27me3 within 5-kb genomic regions flanking TSSs. The enrichment of 5hmC and 5mC was determined by whole-genome tiling microarrays. The enrichment of Tet1, H3K4me3, H3K27me3, and H3K36me3 binding was determined previously by ChIP-seq analyses (Mikkelsen et al. 2007; Wu et al. 2011). All average binding was measured by −log₁₀ (peak P-values) in 200-bp bins and is shown by color scale. The following color scales (white, no enrichment; blue, high enrichment) are used for 5hmC/5mC, Tet1/H3K27me3/H3K36me3, and H3K4me3, respectively: [0, 2], [0, 50], and [0, 100]. The presence of CpG islands is displayed in color (blue, present; white, absent). (B) Average distribution profiles of 5hmC enrichment are shown for Tet1/PRC2-cobound targets, Tet1-only targets, and nontargets. Averaged expression levels of these three groups of genes are shown in the bottom panels (measured by log₂ values of expression microarray signals). (C) Shown are profiles of Tet1 (Wu et al. 2011), 5hmC, Ezh2 (Ku et al. 2008), RNA polymerase II (Seila et al. 2008), and major histone modification (Mikkelsen et al. 2007) occupancy at two representative Tet1 targets: a Tet1/PRC2-cobound target (Lhx2), and a Tet1-only target (Rest promoter). ChIP-seq data in mouse ES cells are shown in read counts, with the Y-axis floor set to 0.2 read per million reads.

Relationship between 5hmC distribution and chromatin occupancy of pluripotency-related transcription factors and other genomic features

The fact that DNA methylation can affect the binding of many DNA-binding proteins to their target sequences raises the possibility that 5hmC may also be involved in regulating the protein–DNA interactions. To investigate this potential relationship in mouse ES cells, we mapped 5hmC microarray signals to previously determined binding sites of a set of proteins important for pluripotency (e.g., Nanog, Sox2, and Oct4) (Chen et al. 2008). In contrast to a general depletion of 5mC at DNA–protein interaction sites, we observed a relative enrichment of 5hmC toward the site of most DNA-binding proteins (Supplemental Fig. S5). Previous analysis of DNA methylation in human ES cells using whole-genome bisulfite sequencing suggests that 5mC in a non-CpG context, but not CpG DNA methylation, is greatly depleted from binding sites of transcription factors related to pluripotency (Lister et al. 2009). Since bisulfite treatment cannot discriminate 5mC from 5hmC (Huang et al. 2010; Jin et al. 2010), bisulfite sequencing may overestimate the 5mC levels at these binding sites. Indeed, specific antibody-based immunoprecipitation analysis of 5mC and 5hmC in mouse ES cells indicated that 5mC was generally depleted from DNA–protein interaction sites, whereas 5hmC was relatively enriched at these sites (Supplemental Fig. S5).

We next analyzed a set of genomic features defined by histone modifications or sequence-specific DNA-binding proteins, including H3K4me3-enriched promoter regions, methylated H3K4 (H3K4me1)-enriched enhancers, Ctcf-marked insulators, and H3K36me3-enriched transcribed intragenic regions (Mikkelsen et al. 2007; Chen et al. 2008; Meissner et al. 2008). Except for H3K36me3-enriched regions, 5mC was, in general, depleted from these genomic features (Supplemental Fig. S5), consistent with the notion that DNA methylation negatively regulates most DNA–protein interactions. In contrast, average signal profiles of 5hmC showed a relative enrichment at promoters, enhancers, transcribed regions, and insulators (Supplemental Fig. S5). Notably, the general enrichment of 5hmC at H3K4me3 peaks indicates that the absence of 5hmC at a subset of CpG-rich, H3K4me3-enriched proximal promoters is probably due to the existence of additional regulatory factors of Tet1 activity or proteins capable of rapidly converting 5hmC into unmethylated cytosine at these sites. Furthermore, the observed enrichment of 5hmC and concomitant depletion of 5mC at enhancer or insulator sequences may therefore contribute to maintaining a more accessible chromatin structure for binding of enhancer proteins (e.g., p300) and Ctcf to these sites. Enrichment of both 5hmC and 5mC at actively transcribed regions marked by high levels of H3K36me3 suggests a transcriptional link between these two marks (Supplemental Fig. S5).

Whole-genome tiling microarray analysis

For whole-genome DNA tiling microarray analysis of 5hmC distribution, immunoprecipitated and input DNA was prepared from both control and Tet1-depleted ES cells and amplified using a whole-genome amplification kit (Sigma). Probe labeling, amplification, hydridization, data extraction, and analysis were performed as described previously (Wu et al. 2011).

For identification of probes associated with significant levels of 5hmC, a nonparametric one-sided Kolmogorov-Smirno (KS) test was used. Briefly, from the scaled log₂ ratio data, a fixed-length window (750 bp) was placed around each consecutive probe, and the one-sided KS test was applied to determine whether the probes were drawn from a significantly more positive distribution of intensity log ratios than those in the rest of the array. The resulting score for each probe was the [−log₁₀ P-value] from the windowed KS test around that probe. Peak data files were generated from the P-value data files using NimbleScan version 2.5. Peaks within 500 bp of each other were merged. For calculating the absolute 5mC levels in control knockdown and Tet1 knoockdown ES cells, the MEDME program (Pelizzola et al. 2008) was used to correct the nonlinear relationship between microarray signals and genomic CpG density. To visualize 5mC and 5hmC distributions in the Cisgenome browser (Ji et al. 2008), probe-level smoothing (log₂ ratios of probes within 1 kb are averaged) was performed for each probe. To calculate the peak distribution, averaged 5hmC or 5mC enrichment (measured by [−log₁₀ peak P-value]) was binned to 200-bp intervals within genomic regions 5 kb upstream of and downstream from TSSs or transcriptional end sites (TESs) of annotated RefSeq genes. Heat maps were generated and visualized using Cluster3 and Java Treeview, respectively. 5mC and 5hmC whole-genome tiling microarray data have been deposited in the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under accession number GSE26833 (Wu et al. 2011) and GSE27613, respectively.

Constructs and antibodies

All of the constructs and antibodies used in this study were described previously (Ito et al. 2010; Wu et al. 2011).

ChIP-seq, gene expression profiling, and data analysis

ChIP-seq experiments and data analysis for Tet1 were described previously (Wu et al. 2011). ChIP-seq data sets of H3K4me3, H3K27me3, H3K36me3 (Mikkelsen et al. 2007), Ezh2 (Ku et al. 2008), and RNA polymerase II (Seila et al. 2008) were obtained from previous publications and reanalyzed in MACS using identical parameters (except statistical cutoff was set to P-value < 10⁻⁵). ChIP-seq sequencing read counts for each ChIP-seq experiments were binned into 400-bp windows at 100-bp steps along the genome and visualized in the Cisgenome browser (Ji et al. 2008). To assign ChIP-seq enriched regions to genes, RefSeq genes were downloaded from the UCSC Table Browser (May 2010). For all data sets, genes with enriched regions within 5 kb of their TSSs were called bound. Gene expression profiling analysis of control and Tet1-depleted mouse ES cells was carried out using the Affymetrix GeneChip Mouse Genome 430 2.0 array. Tet1 ChIP-seq and gene expression microarray data have been deposited in the Gene Expression Omnibus under accession number GSE26833 (Wu et al. 2011).

We thank Brian Abraham and Iouri Chepelev for help with data transfer, Jinzhao Wang for FACS sorting, and Susan Wu for critical reading of the manuscript. This work was supported by NIH grant GM68804 (to Y.Z.), and support to the Division of Intramural Research Program of National Heart, Lung, and Blood Institute from the NIH (to K.Z). S.I. is a research fellow of the Japan Society for the Promotion of Science. Y.Z. is an Investigator of the Howard Hughes Medical Institute.