CpG island structure and trithorax/polycomb chromatin domains in human cells

Abstract

TrxG and PcG complexes play key roles in the epigenetic regulation of development through H3K4me3 and H3K27me3 modification at specific sites throughout the human genome, but how these sites are selected is poorly understood. We find that in pluripotent cells, clustered CpG-islands at genes predict occupancy of H3K4me3 and H3K27me3, and these "bivalent" chromatin domains precisely span the boundaries of CpG-island clusters. These relationships are specific to pluripotent stem cells and are not retained at H3K4me3 and H3K27me3 sites unique to differentiated cells. We show that putative transcripts from clustered CpG-islands predict stem-loop structures characteristic of those bound by PcG complexes, consistent with the possibility that RNA facilitates PcG recruitment or maintenance at these sites. These studies suggest that CpG-island structure plays a fundamental role in establishing developmentally important chromatin structures in the pluripotent genome, and a subordinate role in establishing TrxG/PcG chromatin structure at sites unique to differentiated cells.

Figure 1. Relationship between CpG islands and H3K4me3 modified nucleosomes in pluripotent stem cells

Co-occurrence plot showing H3K4me3 occupancy is highly coincident with CpG islands in human ES cell line WIBR2. The local CG dinucleotide density and H3K4me3 ChIP-Seq density were tabulated across the genome and are presented in a heatmap where each spot in Figure 1 shows the percent of the genome that has a H3K4me3 occupancy level (y-axis) for a given local CG dinucleotide density (x-axis) (Described in Supplemental Information). The top of the scale is 20% and the bottom of the scale is 0%.

Figure 2. CpG island structure predicts the genomic occupancy of H3K4me3 and H3K27me3 modified nucleosomes in pluripotent stem cells

A. Genes were categorized by the number of CpG islands associated with their transcription start site. The genes encoding the homeobox transcription factor sine oculis homeobox homolog 2 (SIX2), RNA polymerase I polypeptide B (POLR1B), and sodium channel, voltage-gated, type I alpha (SCN1A), all located on chromosome 2 are shown as examples. H3K27me3 and H3K4me3 ChIP-Seq density in the hES line WIBR2, local CG dinucleotide density, and CpG islands are shown. B. The portion of genes with zero, one, two, three, four, or more than four (five+) CpG islands that are occupied by H3K4me3 (left) and H3K27me3 (right) modified nucleosomes in WIBR2 pluripotent stem cells is shown.

Figure 3. Relationships between H3K4me3 and H3K27me3 modified nucleosomes and CpG islands in differentiated cells

A. (Left) For the set of H3K4me3 occupied genes that are common to pluripotent and differentiated cells, the percentage of genes that are associated with at least one CpG island was calculated. Pluripotent cell lines including ES (WIBR1, WIBR2, WIBR7) and iPS cells (hiPSC_17, hiPSC_21, hiPSC M23F), and differentiated cell lines, Primary CD4+ T cells, IMR90 fetal lung fibroblasts, T-cell lymphoma cells, CD24+ and CD44+ mammary cells, and HCC1954 breast cancer cells, are indicated at bottom. (Middle) For the set of H3K4me3 occupied genes that are specific to pluripotent cells, the percentage of genes that are associated with at least one CpG island was calculated. (Right) For the set of H3K4me3 occupied genes that are specific to differentiated cells, the percentage of genes that are associated with at least one CpG island was calculated. B. (Left) For the set of H3K27me3 occupied genes that are common to pluripotent and differentiated cells, the percentage of genes that are associated with at least one CpG island was calculated. Cell lines are indicated as in (A). (Middle) For the set of H3K27me3 occupied genes that are specific to pluripotent cells, the percentage of genes that are associated with at least one CpG island was calculated. (Right) For the set of H3K27me3 occupied genes that are specific to differentiated cells, the percentage of genes that are associated with at least one CpG island was calculated. See Supplemental information and Table S3 for genes bound in each cell type.

Figure 4. RNA transcripts from H3K27me3 domains have a high probability of GC hairpin structures

A. The dual RepA stem-loop RNA hairpin that is known to bind PRC2 is shown [21; 33]. B. The number of RNA hairpin hits around the TSS (+/−5kb) for genes within zero, one, two, and three or more CpG islands shows a bias for more hits in genes with multiple CpG islands (see Supplemental Information for the definition of an RNA hairpin hit). The relative fold enrichment versus all genes is shown for each CpG class. C. The number of RNA hairpin hits for genes with one CpG island with or without H3K27me3 occupancy and for genes with more than one CpG island with or without H3K27me3 occupancy is shown. Data are represented as in (b).