Super-enhancers in the control of cell identity and disease

Abstract

Super-enhancers are large clusters of transcriptional enhancers that drive expression of genes that define cell identity. Improved understanding of the roles that super-enhancers play in biology would be afforded by knowing the constellation of factors that constitute these domains and by identifying super-enhancers across the spectrum of human cell types. We describe here the population of transcription factors, cofactors, chromatin regulators, and transcription apparatus occupying super-enhancers in embryonic stem cells and evidence that super-enhancers are highly transcribed. We produce a catalog of super-enhancers in a broad range of human cell types and find that super-enhancers associate with genes that control and define the biology of these cells. Interestingly, disease-associated variation is especially enriched in the super-enhancers of disease-relevant cell types. Furthermore, we find that cancer cells generate super-enhancers at oncogenes and other genes important in tumor pathogenesis. Thus, super-enhancers play key roles in human cell identity in health and in disease.

Figure 1. Transcription factors at super-enhancers

A) Distribution of Med1 ChIP-Seq signal at enhancers reveals two classes of enhancers in ESCs. Enhancer regions are plotted in an increasing order based on their input-normalized Med1 ChIP-Seq signal. Super-enhancers are defined as the population of enhancers above the inflection point of the curve. Example super-enhancers are highlighted along with their respective ranks and their associated genes. B) ChIP-Seq binding profiles for the indicated transcription factors at the POLE4 and miR-290-295 loci in ESCs. Red dots indicate the median enrichment of all bound regions in the respective ChIP-Seq datasets, and are positioned at maximum 20% of the axis height. (rpm/bp: reads per million per base pair) C) Metagene representations of the mean ChIP-Seq signal for the indicated transcription factors across typical enhancers and super-enhancer domains. Metagenes are centered on the enhancer region, and the length of the enhancer reflects the difference in median lengths (703bp for typical enhancers, 8667bp for super-enhancers). Additional 3kb surrounding each enhancer region is also shown. D) Fold difference values of ChIP-Seq signal between typical enhancers and super-enhancers for the indicated transcription factors. Total signal indicates the mean ChIP-Seq signal (total reads) at typical enhancers and super-enhancers normalized to the mean value at typical enhancers. Density indicates the mean ChIP-Seq density at constituent enhancers (rpm/bp) of typical enhancers and super-enhancers normalized to the mean value at typical enhancers. Enhancer read % indicates the percentage of all reads mapped to enhancer regions that fall in the constituents of typical enhancer or super-enhancer regions. E) Metagene representations of the mean ChIP-Seq density for the indicated transcription factors across the constituent enhancers within typical enhancers and super-enhancers. Each metagene is centered on enhancer constituents. Additional 2.5kb surrounding the constituent enhancer regions is also shown. F) Table depicting transcription factor binding motifs enriched at constituent enhancers within super-enhancer regions, and associated p-values. The p-values of enrichment for the same motifs at typical enhancer constituents are listed in Supplemental Figure 1B. G) Revised model of the core transcriptional regulatory circuitry of ESCs. The model contains an interconnected autoregulatory loop consisting of transcription factors that meet three criteria: 1) their genes are driven by super-enhancers, 2) they co-occupy their own super-enhancers as well as those of the other transcription factor genes in the circuit, and 3) they play essential roles in regulation of ESC state and iPSC reprogramming. The layout of the circuit model was adapted from (Whyte et al., 2013). See also Figure S1.

Figure 2. RNA Polymerase II, co-factors and chromatin regulators at super-enhancers

A) ChIP-Seq binding profiles for RNA Polymerase II (RNAPII) and the indicated transcriptional co-factors and chromatin regulators at the POLE4 and miR-290-295 loci in ESCs. OSN denotes the merged ChIP-Seq binding profiles of the transcription factors Oct4, Sox2 and Nanog, and serves as a reference. Red dots indicate the median enrichment of all bound regions in the respective ChIP-Seq datasets, and are positioned at maximum 20% of the axis height. (rpm/bp: reads per million per base pair) B) Metagene representations of the mean ChIP-Seq signal for RNAPII and the indicated transcriptional co-factors and chromatin regulators across typical enhancers and super-enhancer domains. Metagenes are centered on the enhancer region, and the length of the enhancer reflects the difference in median lengths (703bp for typical enhancers, 8667bp for super-enhancers). Additional 3kb surrounding each enhancer region is also shown. C) Fold difference values of ChIP-Seq signal between typical enhancers and super-enhancers for RNAPII and the indicated transcriptional co-factors and chromatin regulators, and RNA-Seq. Total signal indicates the mean ChIP-Seq signal (total reads) at typical enhancers and super-enhancers normalized to the mean value at typical enhancers. Density indicates the mean ChIP-Seq density at constituent enhancers (rpm/bp) of typical enhancers and super-enhancers normalized to the mean value at typical enhancers. Enhancer read % indicates the percentage of all reads mapped to enhancer regions that fall in the constituents of typical enhancer or super-enhancer regions. Reads mapped to exons were removed for the RNA-Seq analysis. D) Metagene representations of the mean ChIP-Seq density for RNAPII and the indicated transcriptional co-factors and chromatin regulators across the constituent enhancers within typical enhancers and super-enhancers. Each metagene is centered on enhancer constituents. Additional 2.5kb surrounding the constituent enhancer regions is also shown. E) Model showing RNAPII, transcriptional co-factors and chromatin regulators that are found in ESC super-enhancers. The indicated proteins are responsible for diverse enhancer-related functions, such as enhancer looping, gene activation, nucleosome remodeling and histone modification. See also Figure S2.

Figure 3. Super-enhancers and candidate master transcription factors in many cell types

A) Heatmap showing the classification of super-enhancer domains across 26 human cell and tissue types. Color scale reflects the density of H3K27ac signal at the super-enhancer regions. B) Gene Ontology terms for super-enhancer-associated genes in 13 human cell and tissue types with corresponding p-values. C) Candidate master transcription factors identified in 6 cell types. All of these transcription factors were previously demonstrated to play key roles in the biology of the respective cell type or facilitate reprogramming to the respective cell type. See also Figure S3.

Figure 4. Disease-associated DNA sequence variation in super-enhancers

A) Catalogue of single nucleotide polymorphisms (SNP) linked to phenotypic traits and diseases in genome wide association studies (GWAS). (left) Pie chart showing percentage of SNPs associated with the highlighted classes of traits and diseases. (middle) Distribution of trait-associated SNPs in coding and non-coding regions of the genome. (right) Location of all non-coding trait-associated SNPs relative to all enhancers identified in 86 human cell and tissue samples. X-axis reflects binned distances of each SNP to the center of the nearest enhancer. SNPs located within enhancers are assigned to the 0 bin. B) Radar plots showing the density of trait-associated non-coding SNPs linked to the highlighted traits and diseases, in the super-enhancer domains identified in 12 human cell and tissue types. The center of the plot is 0, and a colored dot on the respective axis indicates the SNP density (SNP/10MB sequence) in the super-enhancer domains of each cell and tissue type. Lines connecting the density values to the origin of the plot are added to improve visualization. See also Figure S4.

Figure 5. Examples of disease-associated SNPs in super-enhancers

A) (top left) Radar plots showing the density of non-coding SNPs linked to Alzheimer’s disease (AD) in the super-enhancer domains and typical enhancers identified in 12 human cell and tissue types. The center of the plot is 0, and a colored dot on the respective axis indicates the SNP density (SNP/10MB sequence) in the super-enhancer domains or typical enhancers of each cell and tissue type. Lines connecting the density values to the origin of the plot are added to improve visualization. (top right) Distribution of non-coding SNPs linked to AD in the typical enhancers and super-enhancers of brain tissue. (bottom left) List of genes associated with AD SNP-containing super-enhancers in brain tissue. (bottom right) ChIP-Seq binding profile for H3K27ac at the BIN1 locus in brain tissue. The positions of AD-SNPs are highlighted as red bars, and the super-enhancers are highlighted as black bars above the binding profile. Indel rs59335482 (a three base pair insertion) is also highlighted. (rpm/bp: reads per million per base pair) B) (top left) Radar plots showing the density of non-coding SNPs linked to type 1 diabetes (T1D) in the super-enhancer domains and typical enhancers identified in 12 human cell and tissue types. The center of the plot is 0, and a colored dot on the respective axis indicates the SNP density (SNP/10MB sequence) in the super-enhancer domains or typical enhancers of each cell and tissue type. Lines connecting the density values to the origin of the plot are added to improve visualization. (top right) Distribution of non-coding SNPs linked to T1D in the typical enhancers and super-enhancers of Th cells. (bottom left) List of genes associated with T1D SNP-containing super-enhancers in Th cells. (bottom right) ChIP-Seq binding profile for H3K27ac at the IL2RA locus in Th cells. The positions of T1D SNPs are highlighted as red bars, and the super-enhancers are highlighted as black bars above the binding profile. C) (top left) Radar plots showing the density of non-coding SNPs linked to systemic lupus erythematosus (SLE) in the super-enhancer domains and typical enhancers identified in 12 human cell and tissue types. The center of the plot is 0, and a colored dot on the respective axis indicates the SNP density (SNP/10MB sequence) in the super-enhancer domains or typical enhancers of each cell and tissue type. Lines connecting the density values to the origin of the plot are added to improve visualization. (top right) Distribution of non-coding SNPs linked to SLE in the typical enhancers and super-enhancers of B cells. (bottom left) List of genes associated with SLE SNP-containing super-enhancers in B cells. (bottom right) ChIP-Seq binding profile for H3K27ac at the HLA-DRB1 and HLA-DQA1 loci in B cells. The positions of SLE SNPs are highlighted as red bars, and the super-enhancers are highlighted as black bars above the binding profile. See also Figure S5.

Figure 6. Super-enhancers in cancer

A) Selected genes associated with super-enhancers in the indicated cancers. Blue box indicates the gene being associated with a super-enhancer in the respective cancer. CML stands for chronic myelogenous leukemia. B) Cancer cells acquire super-enhancers. ChIP-Seq binding profiles for H3K27ac are shown at the gene desert surrounding MYC in pancreatic cancer, T cell leukemia, colorectal cancer, and healthy counterparts. In colorectal cancer several regions in the 1MB window upstream of MYC were shown to interact with the MYC gene in colorectal cancer (Ahmadiyeh et al., 2010; Pomerantz et al., 2009). (rpm/bp: reads per million per base pair) C) Chromosomal translocation, overexpression of transcription factors and focal amplification may contribute to super-enhancer formation in cancer. Displayed are ChIP-Seq binding profiles for H3K27ac and indicated transcription factors at the gene desert surrounding MYC in the indicated cancers. (top) A translocation event places the MYC gene proximal to an inserted IgH super-enhancer in multiple myeloma. (middle) Tal1 binding is observed at a distal super-enhancer in T cell leukemia. (bottom) Large H3K27ac domains are observed at the site of focal amplification in lung cancer. The red bars below the binding profiles indicate the genomic positions of focal amplification in six different samples, two of which (SM09-019T and SM09-11T1) are primary patient samples (Iwakawa et al., 2013). D) Tumor-specific super-enhancers associate with hallmark cancer genes in colorectal cancer. (top) Diagram of the ten hallmarks of cancer adapted from Hanahan and Weinberg, 2011. Genes associated with super-enhancers in colorectal cancer but not in healthy colon samples were assigned to hallmark categories based on their functions and their previous implication in tumorigenesis. Prominent genes that associate with tumor-specific super-enhancers are highlighted at each cancer hallmark. (bottom) Distribution of H3K27ac signal across enhancers identified in colorectal cancer. Uneven distribution of signal allows the identification of 387 super-enhancers. Prominent genes associated with super-enhancers in colorectal cancer but not in healthy colon are highlighted with their respective super-enhancer ranks and cancer hallmarks they were assigned to. E) Super-enhancers acquired by cancer cells associate with hallmark genes. Each cancer hallmark was assigned a Gene Ontology term, and the number of genes that are associated with acquired super-enhancers and are included in that GO term is displayed for each cancer. Asterisk denotes statistical significance above genomic expectation (hypergeometric test, p<0.05). See also Figure S6.