An atlas of active enhancers across human cell types and tissues

Abstract

Enhancers control the correct temporal and cell-type-specific activation of gene expression in multicellular eukaryotes. Knowing their properties, regulatory activity and targets is crucial to understand the regulation of differentiation and homeostasis. Here we use the FANTOM5 panel of samples, covering the majority of human tissues and cell types, to produce an atlas of active, in vivo-transcribed enhancers. We show that enhancers share properties with CpG-poor messenger RNA promoters but produce bidirectional, exosome-sensitive, relatively short unspliced RNAs, the generation of which is strongly related to enhancer activity. The atlas is used to compare regulatory programs between different cells at unprecedented depth, to identify disease-associated regulatory single nucleotide polymorphisms, and to classify cell-type-specific and ubiquitous enhancers. We further explore the utility of enhancer redundancy, which explains gene expression strength rather than expression patterns. The online FANTOM5 enhancer atlas represents a unique resource for studies on cell-type-specific enhancers and gene regulation.

Figure 1. Bidirectional capped RNAs is a signature feature of active enhancers

a, Enhancers identified by co-occurrence of H3K27ac and H3K4me1 ChIP-seq data, centered on P300 binding sites, in HeLa cells were overlaid with HeLa CAGE data (unique positions of CAGE tag 5’ ends, smoothed by a 5 bp window), revealing a bidirectional transcription pattern. Horizontal axis shows the ± 500 bp region around enhancer midpoints. b, Density plot illustrating the difference in directionality of transcription according to FANTOM5 pooled CAGE tags mapped within ± 300 bp of 22,486 TSSs of RefSeq protein-coding genes and center positions of 10,138 HeLa enhancers defined as above. c, Success rates of in vitro enhancer assays in HeLa cells. Vertical axis shows the fraction of active enhancers (success defined by Student's t-test, P<0.05 vs. random regions; also see Supplementary Figure 9). Numbers of successful assays are shown on the respective bar. See main text for details.

Figure 2. Features distinguishing enhancer TSSs from mRNA TSSs

a, Densities of the genomic and processed RNA lengths of transcripts starting from enhancer TSSs and mRNA TSSs using assembled RNA-seq reads from 13 pooled FANTOM5 libraries. b, Frequencies of RNA processing motifs (5’ splice motif (5'sS, left panel) and the transcription termination site hexamer (TTS, right panel)) around enhancer and mRNA TSSs. Vertical axis shows the average number of predicted sites per bp within a certain window size from the TSS (horizontal axis) in which the motif search was done. Dashed lines indicate expected hit density from random genomic background. The window always starts at the gene or enhancer CAGE summits and expands in the sense direction. c, Average nucleotide frequencies (top panel) and DNase I cleavage patterns (lower panel) of enhancer CAGE peaks (arrow at +1 indicates position of the main enhancer CAGE peaks; direction of transcription goes left to right) reveal distinct cleavage patterns at sequences resembling the INR and TATA elements. d, De novo motif enrichment analyses around enhancers and non-enhancer FANTOM5 CAGE-defined TSSs (CAGE TSSs matching annotated TSSs are referred to as “promoters”), contingent on CGI overlap. Top enriched/depleted motifs are shown along with their best-known motif match name. Enrichment vs. random background is presented as a heat map. e, Vertical axis shows average HeLa CAGE expression fold change vs. control at enhancers and RefSeq TSSs after exosome depletion. Horizontal axis shows position relative to the TSS or the center of the enhancer. Translucent colors indicate the 95% confidence interval of the mean.

Figure 3. CAGE expression identifies cell type-specific enhancer usage

a, Relationship between CAGE and histone modifications in blood cells. Rows represent CAGE-defined enhancers that are ordered based on hierarchical clustering of CAGE expression. Columns for the CAGE tags (pink) represent the expression intensity for three biological replicates. DNase I hypersensitivity and H3K27ac and H3K4me1 ChIP-seq signals ± 1kb around the enhancer midpoints are shown in green, blue and orange, respectively. b, Mean signal of DNase-seq as well as ChiP-seq for H3K27ac and H3K4me1 (vertical axis) per cell type (rows) in ± 1kb regions (horizontal axis) around enhancer midpoints, for enhancers with blood cell type-specific CAGE expression (columns). c, Dendrogram resulting from agglomerative hierarchical clustering of tissue samples based on their enhancer expression: each leaf of the tree represents one CAGE tissue sample (for a labeled tree and the corresponding results on primary cell samples, see Supplementary Figs 18 and 19). Sub-trees dominated by one tissue/organ type or morphology are highlighted. Some of the enhancers responsible for the fetal-specific subgroup in the larger brain subtree are validated in vivo (Fig. 4).

Figure 4. In vivo validation in zebrafish of tissue-specific enhancers

Validations of in vivo activity of CAGE-defined human enhancers CRE1-3 in zebrafish embryos at long-pec stage. Each panel shows, from left to right: i) representative YFP and brightfield images of embryos injected with the human enhancer gata2 promoter reporter gene construct. Muscle (mu) and yolk syncytial layer (ysl) activities are background expression coming from the gata2 promoter-containing reporter construct. All images are lateral, head to the left. ii) YFP zoom-ins and iii) CAGE expression in TPM in human tissues/cell types for the enhancer. Note the correspondence between zebrafish and human enhancer usage/expression. Supplementary Figure 20 shows UCSC browser images of each selected enhancer. a, CRE1, ~230kb upstream of the MEFC2 gene, drives highly robust expression in the brain (brain) and neural tube (nt). Right panel gives zoom-in overlay image showing expression in the forebrain (fb), midbrain (mid), hindbrain (hin) and spinal cord (sp). b, CRE2, 5kb upstream of the POU3F2 gene, is active in the floor plate (fp). c, CRE3, 10kb upstream of the SOX7 gene TSS, shows specific expression in the vasculature (including intersegmental vessels (iv), dorsal vein (dv) and dorsal aorta (da)).

Figure 5. Enhancer usage and specificity in groups of cells

The upper panel gives the number of detected enhancers per million CAGE tags within each group (facet) of related cell type libraries. The expression specificity of the enhancers is shown as a heat map in the panel below. Colors show the fraction of expressed enhancers in each facet (columns) that are in each specificity range (rows). For corresponding plots on organ/tissue facets and genes, see Supplementary Figure 21.

Figure 6. Linking enhancers to TSSs and disease-associated SNPs

a, The proportional contribution (See Methods) of the 10 most proximal enhancers within 500kb of a TSS in a model explaining gene expression variance (vertical axis) as a function of enhancer expression. X axis indicates the position of the enhancer relative to the TSS: 1 the closest, etc. Bars indicate interquartile ranges and dots medians. b, Relationship between the number of highly correlated (‘redundant’) enhancers per locus (horizontal axis) and the maximal expression (TPM) of the associated TSS in the same model over all CAGE libraries (vertical axis). c, GWAS SNP sets preferentially overrepresented within enhancers, exons and mRNA promoters. The horizontal axis gives enrichment odds ratios. The vertical axis shows GWAS traits or diseases. d, Diseases with GWAS associated SNPs over-represented in enhancers of certain expression facets. The horizontal axis gives the odds ratio as in panel C, broken up by expression facets: each point represents the odds ratio of GWAS SNP enrichment for a disease (vertical axis) in a specific expression facet. Summary annotations of point clouds are shown. Also see Supplementary Figure 31.