Diabetes is the most prevalent metabolic disorder worldwide, characterized by hyperglycemia. The two most common forms, type 1 and type 2 diabetes (T1D and T2D), although differing in their etiology, are both polygenic in nature. While strong genetic risk signals have been identified for T1D, particularly at the MHC locus, they are low penetrance in the population and the initial trigger for autoimmune attack is still largely unknown. While the protein and metabolic changes resulting from T1D and/or T2D have been widely studied, few have assessed whether these changes are causal in the development of diabetes. In this dissertation I first characterize transcriptional changes that occur during progression to T1D or T2D using single cell RNA sequencing (scRNAseq), identifying both cell types and pathways that are affected. While beta cells were the predominant cell type affected in both types of diabetes, the changes were largely immune-related in T1D (e.g. antigen processing and presentation) and largely related to metabolism in T2D (e.g. ATP synthesis coupled electron transport). In the second study, I use population-scale proteomics data to explore potential mechanisms at T1D risk loci as well as identify novel circulating protein biomarkers for T1D. Our results revealed four T1D risk loci (FUT2, CTRB2, CENPW and NOTCH2) that strongly colocalized to pancreatic secretions, with one enzyme in particular, CPA1, showing a consistent causal protective effect on T1D that precedes the onset of disease. In the third study, I move beyond genes and proteins and look at the genetic architecture underlying the plasma metabolome. Through genetic fine-mapping of metabolite quantitative trait loci (QTLs), we identified regions across the genome that regulate important metabolic pathways and by performing phenome-wide Mendelian randomization we discovered over 1,000 causal metabolite-trait pairs including increased biliverdin in gallstone formation and decreased arachidonate in pregnancy-induced hypertension. Together these results highlight how identification causal proteins and metabolites can lead to novel biomarkers for diabetes, can inform early disease prediction, and can provide new avenues for the development of therapeutics to prevent and treat diabetes.

Although genome-wide association studies (GWAS) have demonstrated the importance of the pancreas in diabetes risk, mechanistic insight remains challenging in part due to the non-coding nature of risk variants and the lack of cell type-resolved regulatory annotations. Beta cells within pancreatic islets are central to both forms of diabetes and are destroyed by autoimmune mechanisms in T1D or dysfunctional due to insulin resistance in T2D. However, the relevance of beta cell states or other pancreatic cell types to diabetes is relatively unknown. Here I present two studies, both of which use single cell chromatin accessibility to annotate cell type-specific genetic risk mechanisms of diabetes. In the first study, I use snATAC-seq of pancreatic islets to show that endocrine cell types are heterogenous at the gene regulation level, and beta cell states are differently enriched for T2D risk. I highlight a causal T2D variant (rs231361) at the KCNQ1 locus that was predicted to have state-specific effects on regulatory function, was located in a beta cell enhancer co-accessible with INS, and affected expression levels of INS in embryonic stem cell-derived beta cells. In the second study, I generate the largest-to-date GWAS of T1D in 520k samples to enable comprehensive fine mapping of T1D risk signals. I show that pancreatic exocrine cells are causal contributors of T1D through integration of fine mapping with a snATAC-seq reference map of pancreatic and immune cell types. I highlight a likely-causal T1D risk variant (rs7795896) at the CFTR locus that mapped within a ductal-specific enhancer co-accessible with the CFTR promoter, had allele-specific effects on enhancer activity and transcription factor binding in ductal cell models, and was associated with decreased CFTR expression in ductal cells. These two studies highlight the power of integrating comprehensive GWAS fine mapping with single cell epigenomics to understand how pancreatic cell types contribute to diabetes risk.

Obesity and type 2 diabetes (T2D) are complex diseases that are major public health concerns. Genetic variation contributes substantially to risk of developing obesity and T2D. Many previous studies on genetic factors influencing obesity (as measured by body mass index, or BMI) and T2D have focused on the effects of insulin resistance and glucose homeostasis in the pancreas and adipose tissue. A recent study demonstrated that genetic loci associated with BMI are enriched near genes expressed in the brain, but the effects of specific BMI variants on brain function is unknown. Furthermore, despite being a key regulator of energy intake as well as glucose homeostasis, the brain has not been a focus in the molecular genetics of T2D risk. In this work, we aimed to determine the effects of BMI and T2D variants on regulatory processes in the brain through computational analyses integrating genetic association data and brain epigenome data. We determined the enrichment of enhancers in various regions of the brain for BMI and T2D risk signals, identified BMI and T2D risk signals affecting gene expression in the brain, predicted the allelic effects of variants based on brain chromatin features, and prioritized specific BMI and T2D variants likely affecting brain regulation. These data together suggest that variants associated with BMI and T2D risk are broadly enriched for effects on regulatory processes in the brain. Together with pancreatic and adipose tissue, the brain should be considered when studying the molecular mechanisms of genetic variants affecting obesity and diabetes pathogenesis.

Type 1 Diabetes (T1D) is a heterogenous polygenic disease characterized by beta cell loss and insulin deficiency. Genome wide association studies (GWAS) have been conducted to identify genetic risk associated with T1D, but the totality of genetic risk which can lead to a spectrum of phenotypes has not been fully examined. In this thesis, I present three studies that identified novel T1D associated signals and generated genetic risk scores for specific applications. In the first study, I performed a fine-mapped genetic association study in the largest European cohort to date, identifying 19 novel signals across the genome. I leveraged 199 associated T1D variants into a machine learning model to generate the most robust T1D genetic risk score to date. Additionally, cluster analysis revealed distinct genetic patterns associated with phenotypic features such as age of onset and renal complications. In the second study, I performed association analyses and heterogeneity tests to examine differences between individuals with and without high-risk HLA-DR3/DR4 haplotypes in T1D. The non-DR3/DR4 individuals exhibited stronger risk in an HLA class I factor, HLA-B*39:06, and increased polygenic risk across the genome including a novel risk signal, OSTN. I used these new effect sizes to generate a genetic risk score that better identifies those with T1D in the non-DR3/DR4 population and outperforms existing models. In the third study, I created an additive genetic risk score to better target individuals of differing ancestry by using European and African American associated variants. I examined the effect of genetic risk in a multi-ancestry single cell dataset and identified changes in high-risk individuals. I highlight differential changes in pancreatic cell types associated with higher T1D risk including GLIS3-AS1 in beta cells. This thesis demonstrates how identifying and characterizing genetic risk in T1D can create novel risk scores that more accurately capture the diverse risk profiles of the disease and enhance understanding of its underlying biological mechanisms.

Type 1 diabetes (T1D) is caused by the autoimmune destruction of the insulin producing pancreatic beta cells, leading to lifelong dependence on insulin therapy. By the time T1D is diagnosed nearly 90% of beta cells have been destroyed and identifying dysregulated processes prior to T1D diagnosis is needed to improve our understanding of T1D pathogenesis as well as facilitate early detection and intervention. Autoantibodies (AAB) against islet-specific proteins are often observed in T1D, and single and multiple AAB positivity (AAB+) can be used as biomarkers for stages of T1D progression. As an autoimmune disorder there is a bidirectional miscommunication between pancreatic cells and immune cells. Understanding inter- and intra-cell type signaling within the pancreas is critical in understanding T1D pathogenesis. In this dissertation, we use multimodal single cell profiling using single nuclei RNA sequencing, single nuclei ATAC sequencing, and paired single cell multiomic sequencing to identify gene regulatory changes in specific cell types throughout T1D progression in the pancreas. In chapter 1, we identified changes in abundance, gene expression and regulation changes in pancreatic cell types including beta cells occurring in recently diagnosed individuals as well as prior to T1D diagnosis in non-diabetic T1D AAB+ individuals. In chapter 2, we combined a large, multimodal map of pancreatic cell type gene regulation with a multi-ancestry genetic risk score for T1D risk loci which revealed changes in non-diabetic single AAB+ donors in individuals at high genetic risk for developing T1D. Together these aims provide insight into how cell types in the pancreas are dysregulated during T1D progression that provide new avenues for understanding disease mechanisms and potential areas for therapeutic intervention.

Diabetes mellitus, a metabolic disease of great importance to public health, is defined by hyperglycemia and impaired insulin production. The underlying disruptions to genomic regulation that lead to disease pathogenesis are not well characterized at the tissue and cell type level. Diabetes is a complex disease with a large heritable component, and most genetic variants affecting diabetes risk map to non-coding sequence and likely affect genomic regulation in specific cell types. It is therefore critical to study the relationships between genetic risk variants and genome regulation at the cellular level.

For the first chapter of my dissertation, I annotated 17 cell types using single nucleus ATAC-seq (snATAC-seq) profiling in peripheral blood mononuclear cells. I identified putative cis-regulatory elements (cREs) in each immune cell type-resolved chromatin profile, from which we identified 6,901 immune cell type chromatin accessibility QTLs (caQTLs), many of which had immune cell type-specific allelic effects. We linked cell type caQTLs to putative target genes and annotated 622 candidate causal variants for multiple immune-related traits, including type 1 diabetes. A type 1 diabetes-associated variant at the 6q15 locus, rs72928038, was a caQTL in naïve CD4+ T cells, and was linked to BACH2, implicating this gene in type 1 diabetes by altering T cell genomic regulation. We then validated the allelic effects of this variant on regulatory activity in Jurkat cells.

For the second chapter of my dissertation, I studied the effects of hypoxia on genomic regulation within primary pancreatic islet cells and an in vitro pancreatic beta cell model using a chemical mimic of hypoxia, cobalt chloride. During the pathogenesis of diabetes, pancreatic islets are exposed to hypoxic conditions resulting from hyperglycemia, but the tissue- and cell type-specific response to this is poorly understood. I identified major changes in genomic regulation both at the tissue level, using bulk RNA-seq and ATAC-seq data, and at the cell type level, using single cell multiome data. Changes to genomic regulation included significant up-regulation of key hypoxia-related elements (in both gene expression and chromatin accessibility of transcription factor binding motifs) and down-regulation of hormone secretion and islet identity elements across many cell types, as well as the beta cell model. I also identified cell type-specific responses, such as the up-regulation of the unfolded protein response and JAK-STAT signaling in beta cells. I linked chromatin accessibility changes in the beta cell model to known variants that have been associated with diabetes risk. Hypoxia-responsive cREs were significantly enriched for diabetes-associated genetic variants, and we identified several loci that may mediate diabetes risk through altered hypoxia response.

Together, the results from these two chapters reveal insight into the genomic regulation of cell types in key diabetes-relevant tissues and identified putative mechanisms of how diabetes risk variants impact disease by altering cell type-specific genomic regulation.

The two most common forms of diabetes are type 1 diabetes (T1D) which is an autoimmune disorder and type 2 diabetes (T2D) which is a metabolic disorder. Large-scale genome-wide association studies (GWAS) have identified hundreds of loci associated with disease risk, but determining the molecular mechanisms of these loci is not trivial. One major challenge in interpreting GWAS loci is that there is extensive linkage disequilibrium in the human genome where many variants will show association through linkage with the causal variant but not be causal themselves. A second challenge is that most GWAS loci map to non-coding regions of the genome and have no immediately obvious function or affected gene. Together these challenges motivate research to fine-map causal variants at diabetes risk loci and leverage epigenomic and functional genomic data to determine the mechanisms of fine-mapped variants.

In my work I developed strategies and created resources for fine-mapping diabetes risk signals identified in GWAS and determining their molecular mechanisms. In the first chapter, we identify genetic risk shared by T1D and T2D. We then fine-map causal variants at specific shared loci and perform molecular characterization of candidate causal variants at the shared risk loci GLIS3 and CTRB1/2 in pancreatic islets. In the second chapter, we use ATAC-seq and RNA-seq on dexamethasone-treated and untreated pancreatic islets to generate a map of glucocorticoid- responsive islet chromatin sites and gene expression, as well as genetic variants that interact with glucocorticoid signaling to affect islet regulation. We identify enrichment of T2D-associated variants in glucocorticoid-responsive islet chromatin and characterize a fine-mapped T2D risk variant with glucocorticoid-dependent effects on islet accessible chromatin and SIX2/3 expression. Finally, in the third chapter we develop a novel framework for allelic imbalance mapping using ChIP-seq and ATAC-seq data. We quantify the allelic effects of variants on epigenomic sequencing data in islets and liver cells and demonstrate that these effects can help predict likely causal variants for expression QTLs and T2D risk loci. At the HMG20A locus, we identify a fine-mapped T2D risk variant with allelic imbalance in pancreatic islet accessible chromatin and validate allelic effects on pancreatic islets.

Article highlights