PROJECT SUMMARY/ABSTRACT
Pancreatic beta cells secrete insulin in order to maintain blood glucose homeostasis. Insulin secretion is tightly
regulated by glucose and modulated by numerous environmental signals, including other nutrients, hormones,
and inflammatory cytokines. Exposure of beta cells to environmental signals affects gene regulatory programs
within hours, and these signal-dependent changes serve to adapt insulin secretion to changes in organismal
states. Genetic variants associated with measures of insulin secretion are strongly enriched in genomic elements
active in beta cells, and many of these variants are also associated with risk of diabetes. Beta cells therefore
possess characteristics that make them an ideal cellular model for studying signal-dependent gene regulatory
processes relevant to human health and disease. However, the specific genomic programs that drive signal-
induced state changes in beta cells remain poorly characterized. Recent advances in the development of human
pluripotent stem cell (hPSC)-derived multi-cellular islet organoid models by us and others provide a genetically
tractable beta cell model for linking genomic activity to cellular phenotypes. Our group has further pioneered the
development of numerous single cell assays, including chromatin accessibility, ultra-high-throughput paired
chromatin accessibility and gene expression, and paired 3D chromatin interactions and DNA methylation;
methods that we have successfully applied to both primary human islets and hPSC-islet organoids. We have
further developed machine learning and network-based approaches for variant interpretation including from
single cell RNA and epigenetic data. In this proposal we will develop novel gene regulatory network (GRN)
models to predict network-level relationships among genomic elements, genes, and phenotypes derived from
single cell multiomic maps charting signal- and time-dependent changes in hPSC-islet organoids. In Sections
B and C we will measure genomic element activity, gene expression, and insulin secretion in hPSC-islet
organoids exposed to ten different secretory signals each across four time points using paired single nucleus
accessible chromatin and gene expression and paired single cell DNA methylation and 3D chromatin architecture
assays. In Section D we will generate a GRN from these data, use machine learning to infer element-gene and
element-phenotype relationships and use the trained models to refine the GRN. From the resulting GRN we will
predict the effects of genetic variants in specific genomic elements on target gene expression, gene network
activity, and cellular phenotype. In Section E we will validate and refine models by using medium-scale CRISPR
interference of genomic elements individually and in combination as well as allele-specific gene editing of
selected glucose-associated variants in hPSC-islet organoids and measuring gene expression changes in cis
and trans. Together, the results, data, and methods from this project using a model of a cell type which both
rapidly responds to environmental signals and has a quantifiable phenotypic output will be widely applicable to
the community studying the dynamics of genomic regulation.