Genome-wide discovery of enhancer - promoter interactions in the human pancreas using an improved...

Enhancers are key drivers of tissue-specific gene expression and can contain variants associated with pancreatic diseases. Enhancer-target gene assignment remains challenging, with the Activity-By-Contact (ABC) model, integrating open-chromatin, histone modification and chromatin interaction data, consistently outperforming other approaches. Recently an advanced version, the generalized ABC (gABC) model was published, yet lacking a clear and unique promoter definition impairing interpretability. In pancreas the model has not yet been evaluated.

We applied both basal ABC and gABC to map gene-regulatory regions to their respective candidate target genes in pancreas datasets. Next, to balance high gABC performance and ABC interpretability, we implemented the novel canonical-transcript-based and adapted ABC (caABC) model using ENSEMBL canonical transcripts. We compared the performance of all three approaches to predict gene-regulatory regions overlapping with fine-mapped pancreatic expression quantitative trait loci (eQTLs) from GTEx (V8). At the eQTL-colocalized and fine-mapped chronic pancreatitis risk locus CTRC we exemplarily evaluated predicted enhancer-promoter interactions. Finally, we provide a genome-wide unified caABC dataset of pancreatic enhancers and regulated genes.

We demonstrate significantly improved performance of both gABC and caABC compared to ABC in the pancreas, with slightly better performance of gABC at the cost of impaired interpretability compared to caABC. At the chronic pancreatitis risk locus CTRC, caABC enhancer predictions separate fine-mapped risk-variants from high-LD non-fine-mapped variants.

We provide a genome-wide set of pancreas-specific enhancer regions and respective target genes. Our dataset will be helpful for the prioritization of regulatory disease-causing mutations in pancreatic tissue.