Background

The human genome contains approximately 20,000 protein-coding genes, but the non-coding regions that regulate their expression account for 98% of the genome. Enhancers are key regulatory elements dispersed within these non-coding regions, orchestrating the spatio-temporal regulation of target gene expression. The challenge lies in systematically identifying which enhancers regulate which genes.

Previous studies have estimated enhancer-gene connections in individual cell lines or a limited number of tissues, but a comprehensive map encompassing hundreds of cell types at the whole-genome scale has been lacking. The ENCODE (Encyclopedia of DNA Elements) project has conducted over 16,000 genomic experiments over two decades, building a catalog of gene regulatory elements. However, a systematically organized encyclopedia of functional connections between enhancers and their target genes, categorized by cell type, remained a gap.

Key Findings

The ENCODE4 consortium has completed an encyclopedia mapping over 92 million enhancer-gene regulatory interactions across more than 352 cell types and tissues. This represents a roughly seven-fold expansion compared to the 13 million reported in a 2023 preprint, enabled by the integration of an expanded registry of candidate cis-regulatory elements (cCREs) of 2.37 million and additional biosample data.

The research team developed a new predictive model called ENCODE-rE2G (ENCODE-regulatory Element to Gene). This model goes beyond the existing ABC (Activity-by-Contact) model, employing a logistic regression-based supervised learning approach that utilizes 13 features derived from DNase-seq data. A key finding is that, in addition to enhancer activity and the frequency of three-dimensional enhancer-promoter contacts, promoter class and enhancer-enhancer synergy are crucial determinants of enhancer-promoter communication.

Model validation utilized 10,411 element-gene pairs measured in CRISPR interference experiments, over 30,000 precisely mapped eQTLs, and 569 precisely mapped GWAS variants associated with causal genes. ENCODE-rE2G achieved an average recall of 78% in ChIA-PET and HiChIP contact benchmarks, surpassing the 74% of the ABC model, demonstrating superior predictive performance. A four-feature model combining promoter characteristics with the ABC score further improved the AUPRC by 0.044.

Significance and Prospects

This encyclopedia systematically links non-coding variants to target genes for 94 GWAS traits, potentially resolving a major bottleneck in the interpretation of disease-associated genetic variants. Analyses of 197 non-coding credible sets and tissue enrichment for 76 traits have already been performed.

However, limitations exist. The current map relies heavily on data from bulk tissues, and validation at single-cell resolution is still in its early stages. Furthermore, additional research is needed to capture the dynamic changes in enhancer-gene connections, i.e., the rewiring that occurs during development or in disease states. The predictive performance of the ENCODE-rE2G model also varies depending on the cell type, so improving accuracy in rare cell types remains a challenge.