Robust Hi-C Maps of Enhancer-Promoter Interactions Reveal the Function of Non-coding Genome in Neural Development and Diseases

Abstract

Genome-wide mapping of chromatin interactions at high resolution remains experimentally and computationally challenging. Here we used a low-input "easy Hi-C" protocol to map the 3D genome architecture in human neurogenesis and brain tissues and also demonstrated that a rigorous Hi-C bias-correction pipeline (HiCorr) can significantly improve the sensitivity and robustness of Hi-C loop identification at sub-TAD level, especially the enhancer-promoter (E-P) interactions. We used HiCorr to compare the high-resolution maps of chromatin interactions from 10 tissue or cell types with a focus on neurogenesis and brain tissues. We found that dynamic chromatin loops are better hallmarks for cellular differentiation than compartment switching. HiCorr allowed direct observation of cell-type- and differentiation-specific E-P aggregates spanning large neighborhoods, suggesting a mechanism that stabilizes enhancer contacts during development. Interestingly, we concluded that Hi-C loop outperforms eQTL in explaining neurological GWAS results, revealing a unique value of high-resolution 3D genome maps in elucidating the disease etiology.

Keywords: GWAS; Hi-C; HiCorr; bias correction; chromatin loop; eHi-C; enhancer-promoter interaction; neurogenesis; transcription regulation.

Figure 1.. Mapping 3D genome with eHi-C.

A, The scheme of eHi-C. B, Heatmaps show the contact matrices (Chr17) from Hi-C and eHi-C at 250kb resolution. The eigenvectors from Hi-C and eHi-C were very similar, leading to the same compartment A/B assignment. The comparison of eigenvectors between Hi-C and eHi-C in two other chromosomes are shown in the right panel. Histogram listed the r² values of all chromosomes when comparing eigenvectors between eHi-C and Hi-C data. C, Heatmaps of contact matrices from Hi-C and eHi-C at 40kb resolution. The top track is drawn using a published IMR90 Hi-C dataset with ~3 billion reads. A track of TAD structures is plotted in green. On the right is a scatter plot comparing the directionality indexes (DI). The +/− sign of DI is used to determine TAD boundary. Very few bins change their signs of DI, indicating consistent TAD boundaries between Hi-C and eHi-C. D. Heatmap showing the similarity between 5 Hi-C and 7 eHi-C datasets (including a low-depth IMR90 eHi-C dataset) at compartment level. The correlation coefficient is computed by comparing the correlation matrices from different samples.

Figure 2.. HiCorr improves the rigor of Hi-C bias-correction.

A, Chromatin loops contributes to cis but not trans Hi-C reads, leading to an elevated cis/trans visibility ratio. B, Scatter plot of all fragments in GM12878 Hi-C data showing a skew towards higher cis- than trans- visibility. C-D, Epigenetically marked regions and repeat elements have a higher cis/trans visibility ratio. E, Comparing the results of different visibility correction methods. The number in the lower left corner indicates color scale. For example, the color box of “2” in the ratio heatmaps indicates that any contact with O/E > 2 will be shown in dark red; contacts with 1< O/E < 2 will be in light red; white pixels in the heatmaps are O/E < 1. F. Comparison between HiCorr and ICE in capturing promoter-centered interactions from pcHi-C data in GM12878 cells. Note that for ICE curves, we performed ICE normalization followed by distance-correction. The promoter-center interactions from pcHi-C are divided into four groups based on distance (short- or long-range) and the type of interactions (promoter-promoter or promoter-other). The plots showing the number of recovered pcHi-C interactions when the same number of total loop pixels were called from HiCorr- or ICE-corrected contact heatmaps. Up to 500K total loop pixels were tested in these plots.

Figure 3.. Cell type-specific chromatin loops or enhancer aggregates.

A-B, The bias corrected Hi-C heatmaps at a GM12878-specific enhancer aggregate (A), and the transcription levels of the six genes in this region (B). C, Left: Browser tracks showing the GM12878 ChIP-seq data and the locations of guide RNAs for the enhancer inhibition with sgRNAs-CARGO (STAR Methods). Right: ChIP-qPCR results showing the loss of H3K27ac occupancy after inhibiting each of enhancers. D, The expression levels of every gene when the four enhancers indicated in (A and C) are repressed using CRISPRi; data are representative from > 3 independent experiments. Error bar: s.d. of 3 PCR replicates; * p < 0.05, ** p < 0.01 in t test. E, Architecture of HoxA gene cluster in H1, IMR90 and GM12878 cells. F, Expression of HoxA genes in these three cell types.

Figure 4.. Chromatin loops are hallmarks of neural differentiation and neural functions.

A, Venn diagram showing the overlap between chromatin interactions from hiPSCs, hNPCs and hNeurons. B, Distance distribution of chromatin loops in three cell types. C, Bar graph showing the percentage of chromatin interactions with various histone marks. D, Gene ontology terms for genes involved in top 3,000 chromatin loop pixels in each cell type ranked by ratio. E. Enrichment of neuron- or diabetes/obesity relevant GWAS SNPs at chromatin loops. ***p<0.001, binomial test. F, Compartment switching status of the hNPC- (upper) or hNeuron-specific (lower) loops. The four quadrants indicate the compartment-switching status after differentiation. Red dots: bins containing neural loops. All bins in the genome were plotted in the background as blue cloud. Number of red bins, total bins, and percentages are shown in each quadrant.

Figure 5.. Identifying E-P aggregates associated with neurogenesis.

A, An exemplary enhancer-promoter network with ~800 chromatin loops during neurogenesis. Neuron-specific network components can be identified as candidate neuronal enhancer aggregates. Genes in a few neural enhancer aggregates are listed on the right: red, upregulated in neural differentiation, green, downregulated. B, Formation of enhancer aggregate at the FOXG1 locus during neural differentiation. C, Summary of gene expression in neural enhancer aggregates. D, Classification of neural enhancer aggregates based on their dynamic gene expression during differentiation. E, H3K27ac occupancy at different categories of neural enhancer aggregates. F, Compare the strength (ratio) of loop pixels at the differentially expressed genes (DEGs). Top 500 DEGs were picked by comparing hNPC (left) or hNeuron (right) to hiPSC. ***, p < 0.001; **, p< 0.01 Wilcoxon rank-sum test.

Figure 6.. Chromatin loop outperforms eQTLs in explaining GWAS results.

A, Heatmap showing the chromatin loop predicted GWAS target genes, and their overlap with GTEx eQTL data. Highlighted: Tier 1 neural predictions supported by at least two neural Hi-C datasets. B, Distance distribution of predicted GWAS SNP-TSS pairs, based on whether they are supported by loop, eQTL, or both. C, We used neural loops to predict 1,096 target genes for brain GWAS SNPs, and compared their expression to eQTL predicted genes in 48 GTeX tissues. Tissue with red stars: neural loop-predicted genes have higher expression levels than eQTL-predicted genes. *p<1e-2,**p<1e-3,***p<1e-4,****p<1e-5 Wilcoxon rank sum test; Highlighted in yellow: 13 brain tissues. Numbers in parenthesis: the number of genes predicted with eQTL data in each tissue. D, Two GWAS loci examples for which neural loop and eQTL make conflicting predictions. E, GO terms enriched in loop or eQTL predicted target genes, when the two methods make conflicting predictions. F, The CACNA1C GWAS locus is associated with an hiPSC-specific CTCF loop. Highlighted are the three CTCF occupied regions and the CTCF motif directionality. G, Expression of CACNA1C during neurogenesis using RNA-seq data. H, CTCF deletion downregulates CACNA1C in hESC but not NPC. Data are representative from > 3 independent experiments. Error bar: s.d. of three independent experiments; * p < 0.05, ** p < 0.01 in t test.