Untangling the genetics of beta cell dysfunction and death in type 1 diabetes

Abstract

Background

Type 1 diabetes (T1D) is a complex multi-system disease which arises from both environmental and genetic factors, resulting in the destruction of insulin-producing pancreatic beta cells. Over the past two decades, human genetic studies have provided new insight into the etiology of T1D, including an appreciation for the role of beta cells in their own demise.

Scope of Review

Here, we outline models supported by human genetic data for the role of beta cell dysfunction and death in T1D. We highlight the importance of strong evidence linking T1D genetic associations to bona fide candidate genes for mechanistic and therapeutic consideration. To guide rigorous interpretation of genetic associations, we describe molecular profiling approaches, genomic resources, and disease models that may be used to construct variant-to-gene links and to investigate candidate genes and their role in T1D.

Major Conclusions

We profile advances in understanding the genetic causes of beta cell dysfunction and death at individual T1D risk loci. We discuss how genetic risk prediction models can be used to address disease heterogeneity. Further, we present areas where investment will be critical for the future use of genetics to address open questions in the development of new treatment and prevention strategies for T1D.

1. Introduction

Type 1 diabetes (T1D) is a complex autoimmune disease defined by progressive loss of insulin production due to beta cell death and loss of functional beta cell mass, and disease is triggered by environmental factors in genetically susceptible individuals. In European ancestry populations, at least half of the risk of developing T1D is driven by inherited factors [1,2]. Large-scale genetic association studies have identified over 140 variant associations across 93 genomic regions that affect T1D risk and provide insight into different disease mechanisms [[3], [4], [5], [6], [7], [8]]. While islet autoimmunity is a defining feature of T1D, loss of immune tolerance to islet autoantigens cannot fully account for disease onset. Emerging genetic support for non-immune factors contributing to T1D can complement our understanding of immune pathways influencing the disease. A subset of T1D-associated genetic variants has been shown to alter beta cell function, survival, or crosstalk with the immune system, demonstrating mechanisms through which beta cells can affect risk of developing T1D. Here, we review genetic risk factors for T1D, with a particular focus on those influencing beta cells, and discuss how human genetics can advance our understanding of beta cells in T1D pathogenesis and improve clinical care.

2. Pathophysiology of beta cells in T1D

In the 1980s, a model conceptualized T1D pathophysiology in six stages, beginning with inherited genetic susceptibility (Stage I), a triggering event (Stage II), and a period of active autoimmunity but normal glucose control (Stage III) [9]. In this model, progressive loss of beta cell mass due to autoimmune attack is the critical process leading to impaired insulin secretion (Stage IV), overt diabetes (Stage V), and eventually complete loss of detectable insulin secretion (Stage IV) [9]. Subsequent research has revealed that T1D pathogenesis is more heterogeneous and complex than initially appreciated. For example, no single causal agent has been identified as a trigger for islet autoimmunity, and different individuals may have distinct environmental triggers. The clinical presentation of T1D is also heterogeneous, including differences in age of onset [10] and residual insulin secretion [11]. In addition, the relationships between the emergence of islet autoantibodies (seroconversion), immune infiltration of islets (insulitis), beta cell destruction, and progression to diabetes are more nuanced than originally imagined. For example, only a subset of children with islet autoantibodies progress to beta cell destruction and T1D within ten years [12], and autoantibody type may influence disease progression [13]. As evidence supporting different endotypes of T1D accumulates, new models of T1D consisting of multiple distinct routes of progression and pathogenesis are increasingly being considered.

Multiple models have been proposed for how beta cells contribute to T1D pathogenesis. In the canonical model, beta cell death during T1D progression occurs primarily in the context of insulitis, where immune cells infiltrating the islets induce apoptosis in beta cells through cytotoxic T cell-mediated death or proinflammatory cytokines [[14], [15], [16], [17], [18]]. Whether beta cells are simply bystanders of immune attack or have a causal role in the pathogenesis of T1D, for example by triggering an immune attack or responding to immunological or environmental stress, has been widely debated [[19], [20], [21], [22]]. Here, we describe evidence supporting a causal role for beta cells in T1D.

2.1. Beta cells triggering autoimmunity

Islet autoantigen-specific T cells are equally frequent in the peripheral blood of unaffected and affected individuals and appearance of autoantibodies against islet antigens does not guarantee progression to T1D, suggesting that autoreactive T cell escape from central immune tolerance is not sufficient to initiate T1D [[23], [24], [25]]. The specific events that precede and precipitate immune infiltration of islets is an open question and may be heterogeneous across individuals and T1D subgroups. Beta cell damage, dysfunction, and stress have all been proposed as autoimmune triggers. Enhanced or aberrant antigen presentation by beta cells can initiate or exacerbate the autoimmune response [26,27]. Endoplasmic reticulum (ER) stress and the unfolded protein response (UPR) in beta cells can also contribute to autoimmune initiation [28]. However, a study of infants with monogenic diabetes triggering beta cell dysfunction and stress showed no evidence of enhanced islet autoantibody production [29]. While these results may not be representative of beta cell stress in older ages, they suggest additional factors, likely environmental, are required to initiate islet autoimmunity in T1D.

2.2. Beta cell vulnerability

Heterogeneity in beta cell resilience to immune attack or environmental stressors may be equally important in shaping disease progression. The beta cell fragility model suggests that certain individuals have beta cells that are less tolerant of immunological or metabolic stress, leading to increased cell death and risk of diabetes [30,31]. Both T1D and type 2 diabetes (T2D) are associated with variants near GLIS3 (9p24.2)^32,33, which encodes a transcription factor that regulates beta cell development [34]. Mice heterozygous for Glis3 display changes in genes regulating the UPR, consistent with a model of beta cell fragility in both forms of diabetes [30]. Other genetic associations shared between T1D and T2D may have similar functions.

Identifying environmental exposures that contribute to T1D incidence in a large fraction of patients has been challenging. Nonetheless, longitudinal monitoring of environmental exposures and islet autoimmunity biomarkers in high-risk individuals has offered clues about environmental contributors, with multiple studies linking chronic enterovirus infection in early childhood to T1D [35,36]. Beta cell response to environmental stressors or stimuli, such as viral infection or cytokines, may ultimately dictate whether an exposure will lead to the development of T1D. After cytokine exposure, beta cells increase expression of MHC class I along with a series of other peptides, including known beta cell autoantigens, misfolded insulin, different splice products, and even fusion peptides, with an enrichment in peptides originating from secretory granules [37]. The magnitude of these responses may determine whether an initial insult leads to chronic inflammation in the islet, which can exacerbate an immune response, leading to progression to later stages of T1D.

Beta cell crosstalk with the immune system has also been hypothesized to play a role in T1D progression [38]. Cytotoxic T cells are the most common immune cell in islets from recent onset T1D donors [39]. Meanwhile, HLA class I is hyper expressed in islets from recent onset donors, and islet-specific antigens, including neo-antigens, are presented for recognition by T cells. Cellular stress and remodeling of the beta cell microenvironment can alter the fidelity of processes such as mRNA and protein synthesis and processing, contributing to the generation of neo-antigens. How changes in the abundance and constitution of beta cell antigens, together with altered HLA class I expression, may help cause T1D pathogenesis is an area of active investigation. On the other hand, beta cells from individuals affected by T1D express molecules such as PD-L1, which inhibits invading cytotoxic T cells [40]. Supporting a role for PD-L1 in preventing T1D, nearly 3% of patients treated with PD-1-PD-L1 blockade in the context of metastatic cancers develop T1D [41] and individuals with inherited PD-L1 deficiency develop early onset T1D [42].

3. Linking T1D genetic associations to candidate genes

The complexity and heterogeneity of T1D pathogenesis is mirrored by the genetic basis of T1D. T1D is a highly polygenic disease where association studies have collectively identified over 90 distinct genomic regions affecting T1D risk. Genetic studies have implicated a range of cell types in the pancreas and other tissues as causal in disease, including T cells, antigen-presenting cells, beta cells, exocrine cells, and others. Human genetic studies can provide further insight into causal processes within these cell types driving T1D. However, causal genes and contexts have not yet been established at most T1D loci. While maps of T1D-associated regions are a valuable first step, GWAS signals must be interpreted in the context of local genetic architecture, genomic function, cellular processes, and overall physiology for their mechanistic and therapeutic value to be realized.

In this section, we review the basic principles of interpreting genetic associations, highlighting important considerations to avoid false conclusions, and outline resources that can help to contextualize T1D associations. Rigorous follow-up of T1D regions using these approaches can guide investigation of proposed models of T1D pathogenesis, including the role of beta cells in T1D etiology and progression. We note that strategies for following up on genetic association signals have been reviewed elsewhere [43], and refer readers to these sources for more detailed discussion of statistical methods and experimental approaches.

3.1. Prioritizing causal variants with genetic fine mapping

Genome-wide association studies (GWAS) provide robust and reproducible maps of genomic regions influencing a trait. However, genomic regions nominated by GWAS are broad (approximately 500 kilobases to 1 Megabase), containing hundreds to thousands of common genetic variants and harboring up to several dozen candidate genes. The low resolution of GWAS signals is due to genetic linkage, where genetic variants in close proximity are more often inherited together and thus alleles are correlated in the population (referred to as “linkage disequilibrium” (LD)). Specialized analyses, accounting for local LD patterns within each region, are required to distill broad association signals into sets of variants which may be causal for disease risk, called “credible sets.” The process of defining credible sets in GWAS regions is referred to as genetic fine mapping and is typically implemented using dedicated statistical algorithms for variable selection (Figure 1A). Many GWAS regions contain multiple causal variants, including over 30% of T1D risk loci, each potentially mediated by distinct molecular effects [6,7]. Modern genetic fine mapping algorithms attempt to define a credible set for each independent causal signal at a locus [44,45], with the underlying assumption that each independent signal, and thus credible set, contains a single causal variant.

Figure 1.

Dissecting a GWAS locus using (A) genetic fine mapping to define credible sets in a region; (B) molecular QTL mapping and colocalization with T1D association signals to nominate causal molecular mechanisms; (C) cell type-specific regulatory annotations and experimental systems to decode putative causal non-coding genomic regions. Created with Biorender.com.

Two recent T1D fine mapping analyses, using different fine mapping algorithms, defined highly concordant credible sets in many T1D regions [6,7], providing a starting point for mechanistic investigation. In some regions, T1D credible sets only partially overlapped between the two studies, reflecting some of the challenges inherent to genetic fine mapping and limitations of available T1D genetic data sets. The success of genetic fine mapping depends on several factors, including the LD structure in the region, the number of independent causal variants in the region, the effect sizes and allele frequencies of causal variants, the size and ancestral background of the study cohort, and the accuracy with which causal variants were assayed in the study (whether the variant was included on the genotyping array or imputed with high accuracy) [46].

The LD structure of a region is a primary determinant of how effective genetic fine mapping can be. If a causal variant is in high LD with many nearby variants, statistical fine mapping methods may not be able to prioritize causal variants over others based on genotype association patterns alone, resulting in large credible sets, sometimes containing thousands of candidate causal variants (e.g., T1D credible sets in the region encoding MEG3 and DLK1) [6,7]. In these regions, incorporating multiple ancestry groups into genetic studies, which have differing patterns of LD, can help improve fine mapping resolution [47]. To date, T1D studies of non-European ancestry groups are small and lack statistical power, limiting their utility for fine mapping [7,48,49]. However, future investment in diverse T1D cohorts could go a long way towards delineating causal variants in many T1D regions.

Along with LD structure, the number of independent causal variants in a region is critical to the robustness of credible sets. Generally, credible sets are most robust in regions where only a single signal is identified (e.g., T1D credible sets in the region encoding GLIS3) [6,7]. Fine mapping algorithms can struggle to confidently define credible sets in especially complex regions with several independent causal variants, particularly when causal variants are in partial LD with each other (e.g., T1D credible sets in the regions encoding IL2RA, CTLA4, and UBASH3A) [6,7]. Credible sets in these regions may be more sensitive to technical artifacts or modeling assumptions and are more likely to change as additional data become available [47].

In summary, genetic fine mapping is a useful tool for refining broad GWAS signals into tractable credible sets for experimental follow up. Existing T1D credible sets can be integrated with molecular data to nominate causal cell types, regulatory elements, and genes using approaches we discuss in the following sections. At the same time, it is important to keep locus-specific factors in mind when interpreting credible sets. In particular, delineating credible sets is challenging in regions with extended LD or multiple partially correlated signals. Credible sets should be interpreted as sets of candidate causal variants prioritized using available genetic data, with the understanding that they may change as additional data become available, particularly from diverse cohorts.

3.2. Prioritizing causal cell types using regulatory annotations

Most common disease-associated variants are in non-coding regions of the genome, likely affecting regulatory elements that govern gene expression across diverse cellular contexts (e.g., enhancers and promoters, stimulatory and basal conditions) [50,51]. Targeted efforts to annotate regulatory elements in pancreas cell types have provided more refined maps of islet cell type-specific regulatory activity (Table 1). More recently, single cell epigenomics has been useful for profiling regulatory elements active in specific cell types within a heterogeneous tissue, such as the pancreas, and has enhanced the definition of regulatory elements in each islet cell type. Integrating regulatory maps with genetic association data can help indicate which tissues and cell types are broadly involved in T1D risk. Functional enrichment analyses are used to determine whether trait-associated variants preferentially overlap regulatory elements active in a given tissue or cell type [[52], [53], [54], [55], [56], [57], [58], [59], [60]]. T1D-associated variants are enriched in immune cell regulatory elements [5], most prominently in T cells, as well as regulatory elements active in islets, particularly those specific to beta cells [6,7,61]. One study identified enrichment specifically in cytokine-induced regulatory elements in islets [62], suggesting that T1D risk in beta cells and other islet cell types may act in response to cytokine signaling. Collectively, functional enrichment analyses support that T1D risk is affected by genetic effects on immune cells and beta cells, as well as non-endocrine pancreatic cell types such as exocrine acinar and ductal cells [6,7,62,63].

Table 1.

Resources for mapping gene expression and cis regulatory element activity in human islets.

Year	First Author	Last Author	Journal	Tissue	N_donors	N_cells/nuclei	Cohort	Modalities	DOI	Memo
2019	Ramos-Rodriguez	Pasquali	Nature Genetics	human islet	5	NA		RNA-seq, ATAC-seq, H3K27ac ChIP-seq, and DNA methylation (EPIC array)	https://doi.org/10.1038/s41588-019-0524-6
2021	Varshney	Parker	Diabetes	human islet	57	NA	FUSION	Cap analysis of gene expression (CAGE)	https://doi.org/10.2337/db20-1087
2021	Chiou	Gaulton	Nature Genetics	human islet	3	15,298	in-house	snATAC-seq	https://doi.org/10.1038/s41588-021-00823-0
2021	Chiou	Gaulton	Nature	human islet + immune cells	11	1,31,554	in-house	snATAC-seq	https://doi.org/10.1038/s41586-021-03552-w
2022	O Sean	Sneddon	Molecular Metabolism	fetal pancreas (8-20w gestation)	8	33,206	University of Washington Birth Defects Research Laboratory (BDRL) and Advanced Bioscience Resources, Inc.	scRNA-seq and snATAC-seq	https://doi.org/10.1016/j.molmet.2023.101735
2022	Albanus	Parker	biorxiv	human islet	11	1,31,554	in-house	scRNA-seq and snATAC-seq	https://doi.org/10.1101/2022.11.12.516291
2022	Fasolino	Vahedi	Nature Metabolism	human islet	24	80,000	HPAP	scRNA-seq, CyTOF, and Imaging mass cytometry (IMC)	https://doi.org/10.1038/s42255-022-00531-x	∗RNA only, no cis regulatory maps
2023	Elgamal	Gaulton	Diabetes	human islet	65	1,92,203	HPAP	scRNA-seq	https://doi.org/10.2337/db23-0130	∗RNA only, no cis regulatory maps

3.3. Prioritizing candidate genes and regulatory mechanisms

Given tissues and cell types broadly enriched for T1D-associated variants, the next challenge is to determine mechanisms of action at specific T1D associations, including the affected gene(s). Both genome-wide and targeted methods can be implemented to this end.

A molecular quantitative trait locus (QTL) is a genetic variant that affects a quantitative molecular trait, such as gene expression (“eQTL”), splice isoform expression (“splice QTL”), protein expression (“pQTL”), or chromatin accessibility (“caQTL”) (Figure 1B) [64]. Collaborative efforts, such as the Genotype-Tissue Expression (GTEx) project [65], have generated eQTL maps across diverse cell types and contexts. Human islet QTL studies provide resources for investigating islet-centric mechanisms of GWAS risk variants (Table 2). At present, most QTL maps, including those in pancreatic islets, are tissue-level analyses based on bulk RNA-sequencing, obscuring the effects of individual cell types. Emerging QTL studies based on single cell data are revealing cell type-specific QTLs [66,67].

Table 2.

Molecular quantitative trait loci (molQTL) analyses in human islets.

Year	First_Author	Last Author	Journal	Tissue	N	Cohort	Modalities	DOI
2018	Khetan	Stitzel	Diabetes	human islet	19	in-house	ATAC-seq	https://doi.org/10.2337/db18-0393
2020	Vinuela	McCarthy	Nature Communications	human islet	420	InsPIRE	RNA-seq	https://doi.org/10.1038/s41467-020-18581-8
2021	Alonso	Torrents	Cell Reports	human islet	404	TIGER	RNA-seq	https://doi.org/10.1016/j.celrep.2021.109807
2022	Atla	Ferrer	Genome Biology	human islet	399	4 EGA and GEO datasets	RNA-seq	https://doi.org/10.1186/s13059-022-02757-0
2022	Taylor	Taylor	PNAS	human islet	63	in-house	smRNA-seq	https://doi.org/10.1101/2022.04.21.489048
2023	Nguyen	Frazer	Nature Communications	iPSC-derived pancreatic progenitor cells (fetal pancreas)	107	iPSCORE	RNA-seq	https://doi.org/10.1038/s41467-023-42560-4

One approach to generating mechanistic hypotheses at GWAS loci is to integrate candidate causal disease variants with QTL maps using colocalization analysis [68], which formally tests whether genetic associations for two traits may be driven by a shared causal variant (Figure 1B). Like genetic fine mapping, QTL colocalization analysis is performed with dedicated statistical analysis tools [69,70]. Moreover, the same factors that affect genetic fine mapping influence colocalization analyses, including LD structure and the number of independent associations in a locus [64,68,69]. Based on available resources, less than half of GWAS signals colocalize to known eQTLs, and some models indicate that eQTL studies would require very large sample sizes to explain most GWAS associations [71]. These conclusions are informed by existing eQTL studies, which lack cell type- and context-dependent expression measurements. As QTL study sample sizes increase and more diverse cell state contexts and populations are profiled, more eQTL-GWAS colocalizations will be discovered.

High throughput chromatin conformation capture assays can be used to generate three dimensional chromatin maps, aiding in our understanding of the spatial organization of enhancers and other non-coding regions of the genome [[72], [73], [74]]. Using this information, it is possible to annotate target genes by their physical contact with enhancers and other regulatory elements [75]. Further, disease risk information can be overlaid onto these enhancer-gene maps in disease-relevant tissues [72]. While this approach has been used to identify T2D target genes not yet supported by eQTL evidence, such as GLIS3 and INS [[76], [77], [78]], this approach has not yet been systematically applied to non-coding T1D risk variants. Other methods have been developed to link non-coding regions to target genes in the absence of physical interaction data. Cicero, which looks at co-accessibility of regulatory elements in single cell chromatin accessibility data, identified SOCS1 as a potential target gene in cytokine-treated beta cells, and newer methods have been developed that use paired RNA- and ATAC-seq (multiome) data [[79], [80], [81], [82]]. Together, these methods can bridge the gap when annotating regulatory elements and their target genes and can aid in identifying putative causal genes at T1D risk loci, especially in non-coding regions of the genome.

Moving forward, scalable molecular perturbation approaches, including CRISPR-based editing and inhibition/activation screens and massively parallel reporter assays (MPRAs), will provide orthogonal insight into non-coding regulatory mechanisms in disease relevant tissues (Figure 1C) [83]. We expect that combining perturbation-informed regulatory predictions with well-powered QTL maps generated using single cell approaches in T1D-relevant contexts will help annotate molecular mechanisms for many more T1D-associated regions.

4. Models to study candidate T1D genes

Understanding how candidate genes contribute to disease in the broader context of cellular- and tissue-level function is required to realize the therapeutic potential of T1D genetics. Studies of monogenic forms of diabetes have provided vital insights into the importance of disease genes and the associated pathways that lead to beta cell dysfunction or failure. For example, the majority of cases of Wolfram syndrome are caused by autosomal recessive mutations in the Wolfram syndrome 1 (WFS1) gene, which encodes wolframin [84,85]. Wolframin deficiency results in altered calcium signaling [86], impaired GSIS [86,87], and ER stress [88] in beta cells. Studies of WFS1 deficiency may inform mechanisms contributing to beta cell stress or dysfunction and could provide insight into disease processes relevant to T1D, such as beta cell fragility or how ER stress might precipitate autoimmunity [84].

However, the study of monogenic forms of diabetes is unable to holistically model the complex interplay of multiple pathways and tissues that act in concert to cause T1D. The development of T1D models that mimic natural disease etiology and progression has also proved challenging. Heterogeneity in disease course, interactions between multiple tissues and cell types, and contributions of diverse environmental triggers have all been difficult to replicate in animal or in vitro models. Despite these limitations, rodent models, primary human islets, and appropriately chosen cell lines have all still provided tremendous insight into the roles of beta cells and candidate genes in T1D pathophysiology. Rodent models of T1D have been extensively reviewed elsewhere [89,90]. Below, we describe models of T1D with special relevance to the field of T1D genetics, with an emphasis on human models.

4.1. Rodent models

Mouse and rat models of T1D have been used to study disease progression and for preclinical development of disease-modifying therapies (Figure 2A). Here we focus on models of virally-induced and spontaneous diabetes, as well as humanized models in which human tissue is engrafted into immunodeficient mice to study autoimmunity.

Figure 2.

Models used to study candidate T1D susceptibility genes. (A) Mouse, cell line, human, and stem cell-derived models and their applications to study T1D genetics and disease processes. (B) In vitro stressors to model disease processes in the context of diverse genetic backgrounds. GSIS, glucose-stimulated insulin secretion; ER, endoplasmic reticulum; hESCs, human embryonic stem cells; iPSCs, induced pluripotent stem cells; SC-beta cells, stem-cell derived beta-like cells. Created with Biorender.com.

Transgenic mice expressing lymphocytic choriomeningitis virus (LCMV) nucleoprotein (NP) or glycoprotein (GP) antigen under control of the rat insulin promoter (RIP) are a common virus-induced diabetes model [91,92]. These mice express LCMV-NP/GP as “self” antigen on beta cells. Naïve RIP-NP/GP mice do not develop diabetes spontaneously, but infection with LCMV causes LCMV-specific T cells to recognize the GP/NP-expressing beta cells, resulting in beta cell destruction and development of diabetes within 1–2 weeks [91,92]. This model was designed to investigate the roles of viral infection and loss of peripheral tolerance in T1D development [91,92], and it provides the advantage of a defined autoantigen with readily available antigen-specific T cell receptor transgenic mice, as well as the ability to control the timing of diabetes development. A limitation of the RIP-LCMV model is that it does not model the complexity of human T1D in which a variety of genetic and environmental conditions contribute to development of disease.

The most widely used rodent model of spontaneous T1D is the non-obese diabetic (NOD) mouse [93]. In NOD mice, insulitis begins around 3–4 weeks of age, and overt diabetes typically presents between 12 and 14 weeks of age, although incidence and age of onset vary by colony [94]. NOD mouse autoimmune diabetes shares some genetic risk factors with human T1D, including major contributions from MHC class II alleles [95]. NOD mice express a distinct I-A^g7 allele, which contains a polymorphism at position 57 of the I-A β chain [96,97]. The same polymorphisms in the human ortholog, HLA-DQ β57, are associated with T1D risk in humans [98]. Outcrossing NOD mice with other strains identified dozens of additional loci, termed insulin-dependent diabetes (Idd) loci, underlying diabetes in NOD mice [99]. While causal genes remain unknown in most Idd loci, several contain orthologs for human T1D genes, including CTLA4 and IL2 [100]. Genetic variants within the Idd9 locus have also been shown to modulate beta cell susceptibility to autoimmune attack [101]. Although the causal genes at this locus remain unknown and may not be orthologous to human T1D genes, understanding the genetic basis of diabetes in NOD mice has potential to illuminate disease processes that may be present in human T1D. Studies of T1D genetics in the NOD mouse have been extensively reviewed elsewhere [99,101]. Because of the similarities to human disease, pre-clinical studies of T1D-modifying therapies are frequently performed in NOD mice. These therapies have yielded largely disappointing results in clinical trials [102,103], although the success of the anti-CD3 monoclonal antibody Teplizumab (Tzield), which first showed promise as a T1D therapy in NOD mice, supports their value in drug development [104,105].

Humanized mouse models engrafted with functional human tissue have been developed to better mimic human disease processes in mice [106,107]. Immunodeficient recipient mouse strains have been primarily developed on an NOD background [106]. In particular, NOD-scid IL2rγnull (NSG) mice lack mature lymphocytes and NK cells and have been widely used for engraftment of human islets, peripheral blood mononuclear cells (PBMCs), and hematopoietic stem cells (HSC) [107,108]. By engrafting tissues from T1D donors with diverse genetic backgrounds or with targeted genetic modifications, these models can be used to study how different genetic backgrounds contribute to T1D development [106].

4.2. Cell lines

In addition to in vivo models of T1D, several cell lines have been used to probe the role of T1D risk genes in beta cells. Glucose-responsive beta cell lines allow for rapid genetic modification and assessment of glucose-stimulated insulin secretion (GSIS) and cell survival in response to various environmental stimuli. Both rodent and human cell lines have been developed for in vitro studies of beta cell biology and dysfunction (Figure 2A).

The cell lines Min6 [109] and INS-1 [110] were generated from mouse and rat insulinomas, respectively. Min6 and INS-1 cells are glucose-responsive to physiologically relevant glucose concentrations [[109], [110], [111]], and INS-1 GSIS has been further enhanced in the INS-1 832/13 subclone stably transfected with a human insulin expression vector [112]. Min6 and INS-1 (and its derivatives) have been widely used for in vitro studies of beta cell function and survival in response to genetic alterations and environmental stressors. An additional murine cell line, NIT-1, was developed from a beta cell adenoma originating from an NOD mouse and displays modest GSIS [113,114]. NIT-1 cells are especially useful for studying the relevance of beta cell gene expression in the development of T1D, as NIT-1 can be rapidly genetically engineered and transplanted into NOD mice. Recently, a genome–wide CRISPR screen of NIT-1 cells transplanted into NOD mice identified Rnls as a modifier of ER stress and beta cell survival [115], and RNLS maps to a T1D risk locus in humans [116].

The development of the first functional human beta cell line was a long-anticipated breakthrough. In the early 2000s, EndoC-βH1 cells were developed through targeted oncogenesis of human fetal pancreatic tissue [117]. Critically, EndoC-βH1 cells were glucose-responsive and expressed common beta cell markers [117]. Recent generations of EndoC-βH display improved functionality and maturity. EndoCβH3 harbor Cre-excisable immortalization factors to generate cells that can be expanded and then induced to quiescence to better mimic mature human beta cells [118,119]. EndoC-βH5 cells show a nearly 10-fold increase in insulin release in response to glucose and susceptibility to proinflammatory cytokines similar to human islets [120]. EndoC-βH are genetically tractable, allowing researchers to investigate the effects of genetic manipulation on cell function, stress, and survival within a human genetic background. Nonetheless, some studies underline potential limitations of EndoC-βH cells as a model of beta cells in T1D. Karyotypic abnormalities have been reported in EndoC-βH cells, suggesting experiments with these lines should be designed and interpreted with caution [121]. Additionally, EndoC-βH cells express minimal nitric oxide synthase in response to cytokine exposure [122,123]. Ductal cells have been shown to produce nitric oxide synthase in response to cytokines and may be a source of nitric oxide synthase in primary human islets [124,125]. These results highlight the importance of confirming results in primary islets, where beta cells interact with other potentially relevant cell types.

4.3. iPSC-derived beta-like cells

The development of mature beta cells from human embryonic stem cells (hESCs) and induced pluripotent stem cells (iPSCs) for transplantation into T1D patients has been pursued as a potential T1D cure. The first protocols for generating stem-cell derived beta-like (SC-beta) cells in vitro were published in 2014 [126,127], and the SC-beta cells produced displayed reduced insulin secretion and transcriptional signatures similar to fetal beta cells [126,127]. Since then, differentiation protocols have been developed that yield SC-beta cells approaching the functionality of primary human islets [128,129]. Importantly, current differentiation protocols yield cell preparations containing all endocrine cell types (alpha, beta, delta, gamma, epsilon) that can assemble into islet-like organoids (SC-islets), permitting studies of SC-beta cells in an environment more similar to native islets [130].

In addition to their therapeutic potential, SC-beta cells allow for crucial studies of T1D genetics. iPSCs can be generated from healthy patients or those with T1D and genetically modified to study the effects of risk variants in diverse backgrounds (Figure 2A) [[131], [132], [133]]. iPSC models also enable perturbation experiments, multi-omic analysis, and functional studies (e.g., secretion assays) to be performed on SC-beta cells derived from individual patients, an undertaking that has historically been challenging with limited primary islet tissue. iPSCs can be transplanted into immunodeficient SCID or NOD/SCID mice to study the in vivo role of genetic variation on T1D development. Importantly, iPSCs-derived islet-like cells respond to proinflammatory cytokines similarly to primary human islets [134,135] and are amenable to co-culture with immune cells from the same individual to investigate crosstalk between immune and beta cells. Co-culture of iPSC-derived beta-like cells with PBMCs in vitro has demonstrated that thapsigargin-induced ER stress in SC-beta cells activates co-cultured autologous T cells, further supporting the relevance of beta cell stress to T1D pathogenesis and highlighting the value of iPSC-derived models [136].

Although hESCs and iPSCs are valuable tools for studying beta cell development, inefficient differentiations limit their utility. Current differentiation protocols yield SC-beta cells that remain functionally and transcriptionally immature [[137], [138], [139]], and SC-islets contain cell types not present in mature human islets, most notably enterochromaffin-like cells and polyhormonal cells [139,140]. On the other hand, hESC- and iPSC-derived islet cells respond to proinflammatory cytokines similarly to adult human islets [134,135], and may thus represent an interesting experimental model to study responses to inflammation in early life, a period when beta cells are not yet fully mature but may be already exposed - in some individuals – to the early stages of insulitis.

Procurement of immune cells for autologous co-culture studies presents a challenge for modeling T1D autoimmune processes. Immune cells must be either differentiated from iPSCs or collected from donors, timing blood draws for PBMCs with cell differentiations. A more in-depth discussion of modeling T1D processes using SC-derived models is provided in a previous Human Islet Research Network review [141].

4.4. Primary human tissue

Using primary cadaveric human islets to study T1D genetics and beta cell dysfunction has high translational potential. However, access to primary human islet tissue is limited, particularly from donors with diabetes, and heterogeneity between donors creates variability in experimental results [142]. Experimental work in primary human islets has been limited by low transfection efficiency of cells within intact islets. However, recent advances in protocols for pseudo-islet generation allow for efficient transduction of dissociated cells prior to re-aggregation into functional pseudo-islets [143,144]. This approach makes CRISPR-mediated genome editing possible in human pseudo-islets [144]. Genetically engineered pseudo-islets can be used to investigate the impact of genetic modifications on beta cell function and survival in the context of functional human islet architecture (Figure 2A). Since beta cells work in coordination with each other and the other cell types in the islet [[145], [146], [147]], these models will likely offer insights about genetic mechanisms in T1D that would be inaccessible using isolated beta cell models.

In addition to primary human islets, live pancreatic slices from nondiabetic and diabetic individuals are emerging as a powerful tool to study islet function and morphology in the context of the surrounding exocrine pancreas (Figure 2A). Unlike isolated islets, pancreas sections preserve the surrounding islet microenvironment and information on islet localization within the organ, retaining cell-to-cell interactions and tissue compartments [148]. Live pancreatic slices have been used to investigate beta cell mass in T1D patients [149] and islet capillary function in the context of diabetes [149,150]. Pancreas slices may be valuable for understanding genetic variant effects in the context of different islet immune niches or pancreatic anatomy in autoantibody positive or T1D patients. The use of live pancreatic slices for studies of diabetes pathogenesis has been recently reviewed elsewhere [151].

4.5. In vitro stressors

During T1D progression, beta cells are exposed to proinflammatory cytokines and display evidence of heightened ER stress [[152], [153], [154]]. These conditions can be modeled in vitro in primary islets, pseudo-islets, cell lines, or SC-islets (Figure 2B).

Culturing cells with the proinflammatory cytokines interleukin-1 beta (IL-1β), interferon gamma (IFN-γ), and tumor necrosis factor alpha (TNF-α), which are secreted by islet-infiltrating immune cells during progression of T1D, is one of the most common models of beta cell inflammatory stress [154]. Exposure to interferon alpha (IFN-α), which promotes upregulation of MHC class I and ER stress and mediates beta cell death in the presence of IL-1β, has also been used to model early inflammatory processes in T1D [155,156]. Cytokine cocktails induce transcriptional programs in beta cells that are similar to those seen in beta cells from T1D donors, supporting the relevance of this model to human disease [157].

Proinflammatory cytokines in the islet may be secondary to viral infection or other environmental stressors. Viral triggers of islet autoimmunity have been modeled by infecting islets with viruses implicated in T1D, such as Coxsackievirus, or by mimicking viral infection using double-stranded RNA [158,159]. Thapsigargin, a sarco/endoplasmic reticulum Ca²⁺ ATPase inhibitor, and tunicamycin, which inhibits N-linked glycosylation of proteins, have been used as in vitro models of ER stress [115,160]. Other stimuli have been used to model additional aspects of T1D progression, such as hyperglycemia, hypoxia, and oxidative stress.

Given the complex genetic basis of T1D, a subset of T1D-associated variants likely affect disease by modifying beta cell responses to diabetogenic conditions. However, enormous heterogeneity between individuals in environmental exposures makes such gene-by-environment interaction effects challenging to detect in epidemiological studies. The controlled experimental conditions provided by in vitro systems can increase power to detect modifying effects of genetic variants on beta cell response to exposures. In particular, in vitro models of environmental stimuli can be combined with genetic modification and beta cell function or survival assays to investigate genetic effects on beta cell sensitivity to known or hypothesized environmental causes of T1D.

5. T1D-associated regions influencing beta cell function

In the sections above, we described tools for nominating T1D candidate genes based on genetic evidence and models for investigating these candidates in cellular-, tissue-, and organismal contexts. To date, there are few, if any, studies which have conclusively mapped a T1D-associated region to a causal variant and gene acting in beta cells and further demonstrated a cellular/organismal function leading to T1D. Therefore, at present, we only have a partial understanding of how beta cells contribute to T1D risk. In this section, we review T1D-associated regions and candidate genes with genetic evidence supporting a role in beta cells. We also note that there is a large body of literature that has experimentally assessed the function of candidate genes at T1D-associated loci in beta cells, but many of these genes have not yet been linked directly to T1D risk variants. To help demonstrate the current gaps in knowledge at these and other loci, we include here a handful of strong candidate genes that are predicted to affect beta cell function but which have varying degrees of evidence linking them to T1D credible variants (Table 3, Figure 3).

Table 3.

Evidence linking T1D credible variants from recent fine mapping studies to candidate genes.

Genomic region	Number of independent T1D association signals in the region	Candidate genes	Evidence linking credible variants to candidate genes
			Protein-coding credible variant(s)	Islet QTL colocalization	Chromatin interaction with promoter (PCHi-C or Hi-C)
11p15.5	3 or 4a	INS	No	No	Yes (PMID: 36070683)
6p21.3	Manyb	HLA	Yes	NA	NA
18p11.21	3	PTPN2	No	No	No
9p24.2	1	GLIS3	No	No	Yes (PMID: 31253982; PMID: 31064983; PMID: 36070683)
16p13.13	2	CLEC16A	No	No	No
		DEXI	No	No	Yes (PMID: 21989056)
		SOCS1	No	No	Yes (PMID: 36778047)
2q24.2	4	IFIH1	Yes	No	No
14q32.2	2	DLK1	No	No	No
		MEG3	No	Yes (PMID: 36109769)	No
15q25.1	1	CTSH	Yes	No	No
19p13.2	2	TYK2	Yes	No	No

^aStudies report three6 or four7 independent signals in this region.

^bRecent T1D fine mapping studies excluded the HLA region, but previous work indicates several independent associations.

Figure 3.

A model of candidate gene contributions to beta cell destruction in T1D. Loss of GLIS3 or DLK1 may impede beta cell differentiation or enhance post-natal beta cell apoptosis, possibly potentiating beta cell fragility in the setting of autoimmunity. In mature beta cells, viral dsRNA signaling through MDA5 (encoded by IFIH1) may induce cytokine and chemokine release, contributing to immune cell recruitment and cytokine signaling in beta cells. Proinflammatory cytokine signaling in beta cells is modulated by TYK2, DEXI, PTPN2, and SOCS1, resulting in downstream transcription factor activation, ER stress (enhanced by reductions in GLIS3), and accumulation of damaged organelles (potentiated by loss of CLEC16A), which together may contribute to beta cell apoptosis (attenuated by overexpression of CTSH) in the setting of proinflammatory cytokine release. In addition to apoptosis, beta cell dysfunction and reductions in glucose-stimulated insulin secretion may be mediated by GLIS3 and CLEC16A. Finally, environmental stressors induce expression of HLA class I and components of HLA class II in beta cells, likely contributing to T cell recognition of beta cells. During T cell development, tolerance to insulin autoantigens is mediated by INS expression in thymic epithelial cells. Autoreactive T cells that avoid clonal deletion in the thymus can be activated by islet autoantigens presented on MHC, resulting in autoimmune destruction of beta cells. Created with Biorender.com.

5.1. INS

Genetic variation in the region encoding insulin is the largest genetic determinant of T1D susceptibility outside of the MHC. A highly polymorphic variable number tandem repeat (VNTR) in the promoter region of the insulin gene (INS) was identified in the 1980s [161] and confirmed by long-read sequencing to consist of a 14 base pair sequence that repeats up to 200 times [162]. Observed INS VNTR alleles have been grouped into three classes (I, II, and III), where class I alleles have the fewest repeats and class III alleles have the most. The longer class III alleles confer dominant protection against T1D [163,164], potentially by promoting negative selection of autoreactive T cells specific for insulin-derived peptides [165,166]. This hypothesis is supported by evidence showing the protective alleles are correlated with higher insulin expression in the human thymus [166,167].

Fine mapping of the INS region has identified multiple independent associations with T1D [6,7], suggesting that reducing INS region haplotypes to three broad VNTR classes may obscure additional mechanisms at this locus. Based on existing resources, none of the genetic variants associated with T1D affect basal INS expression or splicing in beta cells; however, as discussed in Section 3, QTL maps of human islet cell types are based on expression profiling of aggregate islet tissue under basal conditions from a limited number of donors. Integration of T1D fine mapping with molecular data points to rs4929965 as a candidate causal variant which maps to a beta cell-specific distal regulatory element that contacts the INS promoter [168], indicating that it may affect INS expression in beta cells in the right context. The T1D association tagged by rs4929965 also influences risk of T2D but has opposite effects on the two diseases [33]. Well-powered, islet cell type-specific QTL maps across diverse contexts, for example using in vitro stressors, will likely reveal new regulatory mechanisms of T1D- and T2D-associated variants near INS.

5.2. HLA

The most substantial genetic determinants of T1D risk are the human leukocyte antigen (HLA) class II genes (HLA-DRB1, -DQA1, -DQB1, -DPA1, and -DPB1), which encode components of major histocompatibility complex (MHC) class II molecules. In particular, the haplotypes DRB1∗03:01-DQA1∗05:01-DQB1∗02:01 (“DR3”) and DRB1∗04:01/02/03/05-DQA1∗03:01-DQB1∗03:02 (“DR4”) are strong predictors of T1D risk [169] and have already been used to prioritize high-risk individuals for longitudinal prospective studies of T1D etiology [170]. Additional association signals are also seen in genes encoding MHC class I molecules (HLA-A, -B, and -C) [171]. Both MHC I and MHC II molecules present peptide antigens for recognition by T cells, an essential step in T cell-mediated adaptive immunity. HLA variants mediating T1D risk are concentrated in the peptide binding pockets of MHC molecules [171] where they are suspected to influence binding and presentation of self-antigen.

Increased expression of HLA class I genes is observed in insulin-containing islets from recent onset T1D patients [26] and may enhance beta cell destruction by cytotoxic CD8⁺ T cells. Increased islet HLA class I expression can precede insulitis, and autoantigen presentation by beta cells on MHC I molecules may contribute to T1D etiology [172,173]. MHC II complexes are typically expressed by professional antigen presenting cells (APCs), such as dendritic cells, macrophages, or B cells. However, there is some evidence of ectopic MHC II expression within the islet [[174], [175], [176]]. Studies on the relationship between T1D-associated alleles and MHC I and II expression in beta cells, however, have been limited [177].

5.3. PTPN2

At the 18p11 locus, fine mapping identified three independent associations in intronic regions of protein tyrosine phosphatase non-receptor type 2 (PTPN2) gene [6]. PTPN2 has been implicated in regulating beta cell responses to proinflammatory stress. Expression of PTPN2 in primary human islets and rodent beta cell lines was found to increase after exposure to proinflammatory cytokines [178]. Another tyrosine phosphatase, PTPN22, also strongly associated with T1D, showed no change in expression in beta cells in response to cytokines [178]. Further work showed that PTPN2 regulates IFN-γ signaling and modulates ER stress after cytokine exposure [160]. Additionally, PTPN2 was found to modulate the deleterious effects of TNF on human beta cells via regulation of JNK activity [179]. PTPN2 has also been shown to affect beta cell survival after cytokine exposure in a genome-wide CRISPR loss-of-function screen in EndoCβH1 cells [80]. Knocking out PTPN2 in stem-cell-derived beta-like cells led to increased HLA class I expression and consequently increased recognition by autoreactive T cells [180]. These studies suggest PTPN2 may modulate beta cell apoptosis, by dephosphorylating downstream targets of cytokine signaling, or protect beta cells from immune recognition. While these studies demonstrate a role for PTPN2 in beta cell function and survival, we note that none thus far have formally linked altered PTPN2 activity in beta cells to T1D-associated variants directly and that other evidence indicates a role of PTPN2 in both adaptive and innate immune systems.

5.4. GLIS3

The 9p24.2 region is one of a few loci associated with both T1D and T2D risk, and formal colocalization analysis supports a shared causal variant with the same direction of effect for both traits [32,33]. T1D fine mapping defined a single causal signal in this locus, with credible variants mapping to an approximately 14 kilobase region intronic to GLIS3 [6,7]. Pancreatic islet chromatin interaction maps indicate the credible set in this locus interacts with multiple genes in the region, including RFX3, RFX3-AS1, and GLIS3, but deletion of the putative causal enhancer only affected expression of GLIS3 [77]. The GLIS3 gene encodes a GLI-similar Kruppel-like Zinc finger transcription factor that regulates pancreatic beta cell development [34], and mutations of GLIS3 cause a form of neonatal diabetes [181]. Chromatin immunoprecipitation studies in rodent models have shown that GLIS3 interacts directly with the Ins2 promoter, as well as with PDX1, MAFA, and NEUROD1 to regulate activity at the insulin promoter [182]. Mouse Glis3-deficient models display hyperglycemia and shortened lifespan. Additionally, Glis3 heterozygous mice had changes in genes regulating the UPR, leading to downstream beta cell stress and supporting the shared beta cell fragility model of T1D and T2D [30]. CRISPR deletion of GLIS3 during embryonic stem cell differentiation into pancreatic beta cells led to decreased representation of INS-positive differentiated cells [183,184]. GLIS3 may also play a role in beta cell survival, as indicated by in vitro studies that assessed beta cell apoptosis in response to proinflammatory cytokines or glucolipotoxicity [185,186]. Given its role in T1D, T2D, and monogenic diabetes, pathways regulated by GLIS3 may represent a therapeutic opportunity in the treatment of multiple forms of diabetes [31].

5.5. CLEC16A/DEXI/SOCS1

The 16p13 locus is a gene-rich region harboring multiple independent T1D associations [6,7] and several potential candidate genes [187], including C-lectin domain containing 16 A (CLEC16A), dexamethasone-induced transcript (DEXI), suppressor of cytokine signaling 1 (SOCS1), and MHC class II transactivator (CIITA). Fine mapping defined two T1D credible sets in the region. The primary T1D signal at 16p13 maps to a single CLEC16A intron [6,7]. CLEC16A encodes an E3 ubiquitin ligase essential for mitophagy (selective autophagy of damaged mitochondria) [[188], [189], [190]]. CLEC16A deficiency in rodent and human islets leads to impaired beta cell function and reduced beta cell survival following exposure to proinflammatory cytokines and inflammatory insults [191]. T1D credible variants in CLEC16A are associated with reduced insulin secretion and decreased expression of CLEC16A in human beta cells [188], though there has been no formal colocalization between T1D GWAS signals and islet eQTL for CLEC16A. In immune cells, T1D credible variants in CLEC16A are associated with DEXI expression and overlap a regulatory element that contacts the DEXI promoter [192]. DEXI modulates the type I IFN/STAT pathway in beta cell lines and primary human islets [193]. Modulation of DEXI in NOD mice, however, did not affect the development of T1D [193,194].

The secondary T1D signal at 16p13 spans multiple introns of recQ mediated genome instability 2 (RMS2) [6,7]. T1D credible variants in RMS2 overlap a cytokine-responsive regulatory element which is thought to regulate cytokine-dependent expression of SOCS1 in beta cells [80]. In a genome-wide CRISPR screen in EndoCβH1 cells, SOCS1 promoted cytokine-mediated beta cell survival and affects beta cell survival in human and animal models by dampening the inflammatory response [80]. Finally, although not linked specifically to T1D-associated variants, CIITA, a transcriptional regulator of MHC class II gene expression, represents a fourth potential candidate gene in the 16p13 locus [195,196].

This T1D locus illustrates the complexity of interpreting disease associations in regions with multiple compelling candidate genes. Functional validation of variant-to-gene links in disease-relevant models will be vital to teasing apart the true causal mechanisms underlying T1D association in this region.

5.6. IFIH1

Interferon-induced helicase 1 (IFIH1) encodes the cytoplasmic viral RNA detector melanoma differentiation-associated protein 5 (MDA5), which is vital for antiviral signaling [[197], [198], [199]]. T1D fine mapping implicates multiple low-frequency variants altering the MDA5 protein [7], the most common of which is rs1990760, an A946T missense mutation in the carboxy terminal domain (CTD) of MDA5 [197]. The A946T variant, which increases risk for T1D, causes increased cytokine production and gene expression in human PBMCs [200] and associates with stronger interferon response to Coxsackievirus B (CVB) in human islets [201]. Reduced expression of MDA5 or defects in the MDA5 helicase 1 domain on the NOD background reduced incidence of CVB-associated diabetes in part due to reductions in type 1 IFNs [202,203], but complete deletion of MDA5 led to an accelerated onset of diabetes in NOD mice following CVB exposure [203]. Taken together, human and mouse evidence indicate the MDA5-mediated antiviral response is likely involved in T1D etiology and suggest therapeutic potential for tuning these responses. However, whether MDA5 contributions to T1D are mediated primarily by its activity within islets, immune cells, or both remains an open question.

5.7. DLK1/MEG3

Human genetic studies support paternally inherited risk for T1D in the imprinted region of chromosome 14q32, which contains the genes maternally expressed 3 (MEG3) and delta-like homolog 1 (DLK1) [204]. T1D fine mapping indicates two independent association signals in 14q32 [6]. One T1D credible set was refined to a single candidate variant, rs56994090 [6,7] and colocalized with an islet splice-QTL for the lncRNA MEG3 [205]. The other credible set contains a variant, rs3783355, that overlaps a beta cell-specific regulatory element [63] and showed allelic bias in islet transcription factor ChIP-seq data [206]. Together, these data support rs56994090 and rs3783355 as candidate causal variants for T1D potentially acting through two distinct regulatory mechanisms or genes within the islet. MEG3 is a maternally expressed long non-coding RNA, whose expression is downregulated in islets of T2D donors [207]. The paternally imprinted DLK1 encodes a delta-like non-canonical Notch ligand, which is broadly expressed in rodents during development and later restricted to pancreatic beta cells, pituitary somatotroph cells, bone marrow, adrenal gland, and gonadal tissues [[208], [209], [210]]. Conditional loss of Dlk1 in mouse beta cells did not affect islet size, number, or architecture up to 6 weeks after birth [210]. However, mice bearing transgenic overexpression of Dlk1 within beta cells displayed increased islet mass and insulin secretion [211]. Studies in isogenic hESCs revealed that loss of DLK1 and disruption of DLK1 regulatory regions led to increased beta cell apoptosis [63].

5.8. CTSH

T1D fine mapping analyses identified a single credible set of 4 or 5 variants in the chromosome 15q25.1 region, including a nonsynonymous variant in the cathepsin H (CTSH) gene [6,7], which colocalized with a whole-blood eQTL for the same gene [212]. Earlier work suggested that T1D-associated variants may also influence CTSH expression in pancreas [213,214], however, formal colocalization of these effects has not been evaluated using credible sets or eQTL resources. CTSH encodes a lysosomal cysteine protease, which is ubiquitously expressed and vital for degradation of specific cargo delivered to lysosomes [215]. Cathepsins have been broadly implicated in immune cell function, as MHC class II molecules present antigens derived following lysosomal processing, as well as autophagy [215,216]. Impairments in beta cell macroautophagy and lysosome function have been observed in T1D [217]. CTSH expression is suppressed by cytokine exposure in both rodent and human beta cells, and overexpression of CTSH protected beta cells against cytokine-mediated apoptosis in part through decreased JNK and p38 signaling and reduced expression of the proapoptotic factors Bim, DP5, and c-Myc [214,218]. These beneficial effects of CTSH appear to be mediated through regulation of the small GTPase Rac2, as Rac2 deficiency abolishes the protective effects of CTSH on beta cell survival [219]. Further, CTSH knockout mice display reduced islet insulin content. Together with observations with CLEC16A, involvement of the CTSH locus suggests organellar quality control in beta cells may be important in T1D.

5.9. TYK2

Tyrosine kinase 2 (TYK2) encodes a non-receptor Janus kinase critical for type I IFN signaling that is broadly expressed among immune cell types and beta cells. Fine mapping suggests two independent nonsynonymous variants in TYK2 (rs34536443 (P1104A) and rs12720356 (I684S)) offer protection against T1D [6,7]. Peripheral immune cells from individuals bearing the P1104A variant had significantly reduced STAT1/3 phosphorylation, a readout of TYK2/Janus kinase activity, following exposure to type I IFN across all immune cell subsets [220]. In mouse beta cells, complete Tyk2-deficiency accelerated diabetes induction following exposure to a diabetogenic form of encephalomyocarditis virus that was specifically dependent on beta cell Tyk2 loss [221]. This may be due to the importance of TYK2 in beta cell development, as TYK2 knockout human iPSCs also had an impaired emergence of endocrine precursors [222]. Alternatively, knockdown of TYK2 appeared to be protective against experimental forms of beta cell damage in EndoCβH1 cells or human islets following exposure to the viral dsRNA mimic PIC or in iPSC-derived islets following exposure to IFN-α [155,[222], [223], [224]]. The importance of the protective P1104A kinase domain mutant has not been directly studied in human or rodent beta cells. However, use of a TYK2 pharmacologic inhibitor, which stabilizes the TYK2 pseudokinase domain and has been reported to have similar effects as the P1104A variant, led to reduced IFN-α-mediated upregulation of MHC class I in iPSC-derived human islets and reduced T cell cytotoxicity in co-culture assays [222,225]. These studies suggest that partial TYK2 deficiency or pharmacological recapitulation of effects of the TYK2 P1104A variant, as opposed to complete TYK2 loss of function, may have beneficial effects on beta cells in the prevention of T1D by inhibiting type 1 IFN signaling and the consequent upregulation of MHC class I and chemokine production to recruit cytotoxic T cells.

6. Genetic support for T1D heterogeneity

T1D is a heterogeneous disease marked by variation in several traits, including (but not limited to) age of onset [10], first autoantibody present [226], rate of autoantibody spreading and types of subsequent autoantibodies [227], immune infiltration of islets [228], residual insulin secretion [11,229], and susceptibility to secondary complications [230]. Variation across these features is non-random. For example, earlier onset disease is associated with lower residual insulin secretion [11], hyperimmune islets [228], and faster disease progression [231]. Meanwhile, the first-appearing autoantibody distinguishes two patterns of genetic and environmental exposures [226]. This apparent clustering of traits suggests that T1D can potentially be divided into multiple endotypes, each with a distinct mechanistic underpinning that could be addressed by an appropriately matched therapeutic strategy [230]. An alternative hypothesis is that T1D development for each individual is determined by different combinations of multiple causal pathways. This concept - termed the “palette” model - was first proposed in the context of type 2 diabetes risk [232]. Regardless of whether T1D can be broken into discrete endotypes or represents a composite of effects on multiple causal pathways, recognizing patterns of T1D heterogeneity and the causal processes underlying them may facilitate tailored treatments and improved outcomes for patients. For instance, individuals diagnosed after 13 years of age have reduced B cell infiltration and higher retention of beta cell mass [228], suggesting that beta cell dysfunction, rather than beta cell death, may play a more prominent role in older patients and therapies restoring beta cell function may be more effective in this group. Here, we discuss existing genetic support for heterogeneous T1D etiology and pathophysiology and opportunities for using genetics to further dissect T1D heterogeneity.

6.1. Genetic underpinnings of T1D heterogeneity

Individual T1D loci are known to correlate with features of disease etiology and progression. High-risk children with HLA-DR4 haplotypes tend to develop insulin autoantibodies (IAA) as the first-appearing autoantibody within the first two years of life [[233], [234], [235]]. In contrast, HLA-DR3 haplotypes are associated with glutamic acid decarboxylase antibody (GAD) as the first-appearing autoantibody with seroconversion occurring between two to five years of age [233,234]. Multiple T1D risk alleles have been associated with earlier T1D onset [236], and genetic risk factors had a larger effect on T1D risk in younger individuals [212]. However, this does not necessarily imply that young onset T1D is more ‘genetic’ (i.e., heritable) than older onset disease. Most genetic discovery for T1D has focused on pediatric (<16 years of age) cohorts of European ancestry, and, therefore, established T1D risk variants will be enriched in T1D cases from these age and ancestry groups. Recent work compared T1D genetic risk prediction in self-reported Hispanic, Black, and White individuals in the Search for Diabetes in Youth (SEARCH) study [237]. This work highlighted the importance of including a larger number of HLA variants more common in non-European populations to capture T1D risk across diverse ancestries and showed a variable distribution of T1D genetic risk scores across self-reported ethnicities. Expanding studies of T1D to a more diverse patient population, including individuals with later disease onset, lower-risk HLA haplotypes, and non-European ancestries, will likely reveal new pathways contributing to disease in these groups [48,238]. Additionally, genetic studies of T1D within putative endotypes could provide insight into where their etiologic mechanisms diverge.

6.2. Genetic prediction models to address T1D heterogeneity

T1D is distinct among common complex diseases as known genetic risk factors explain the majority of disease heritability. Genetic risk scores (GRS) for T1D can distinguish high-risk individuals with area under the curve (AUC) in independent validation cohorts of >0.9 in Europeans and >0.8 in other ancestry groups [48,[239], [240], [241]]. Given the strong performance of existing T1D GRS, they will be useful for prioritizing high-risk individuals for monitoring and enrollment in early intervention trials. However, there may also be an opportunity to use genetic risk prediction models to dissect T1D heterogeneity, either as a marker for T1D endotypes or to partition genetic effects into causal pathways.

Most existing biomarkers supporting T1D endotypes are intractable in the general population. For example, establishing the first-appearing autoantibody requires longitudinal autoantibody testing prior to overt symptoms. Similarly, the immune cell composition of pancreatic islets cannot currently be evaluated in living patients. In contrast, genetically-derived T1D endotype scores could be assessed at birth to prioritize high-risk children for longitudinal monitoring, or at the time of T1D diagnosis to inform therapeutic decision-making. In a related approach, T1D GRS has already been used to discriminate T1D from other forms of diabetes [[242], [243], [244]].

Ultimately, T1D may not reduce to a fixed number of discrete endotypes. The genetic complexity of T1D hints that its etiology for most individuals is a blend of causal pathways, similar to other complex diseases [232,245]. As new molecular resources are developed to map genetic associations to causal genes in islet and immune cell types, one may eventually use genome-wide profiles to estimate the relative contribution of relevant pathways to the disease process in individual patients. This approach has been pioneered in recent T2D studies, which partitioned genetic risk into etiological pathways and demonstrated heterogeneity across individuals in terms of which pathways were the predominant factor underlying disease risk [246,247]. For T1D, one may envision a genetic score for each of several contributing pathways (e.g. innate viral response, beta cell stress response, or beta cell antigen processing and presentation). Whether through discrete endotypes or cumulative effects of causal pathways, the manifestations of T1D heterogeneity (e.g., variation in beta cell survival and severity of insulitis [228]) are likely to be, in part, determined by the relative contribution of immune and beta cell-intrinsic processes. Genetics will be critical to discerning the relative contribution of these causal processes to disease burden in the population and to the disease process in individual patients.

7. Proposed areas for future focused effort in T1D genetics

Substantial progress has been made towards understanding the genetic basis of T1D through concerted efforts to recruit T1D cohorts for genetic studies [248]. However, there is still a large gap in translating genetic discoveries into therapeutic opportunities for prevention and treatment. Looking forward, large population biobanks pairing whole genome sequencing (WGS) with deep phenotyping will empower a new era of discovery, including investigation of rare variation, T1D-related traits in healthy individuals, and “phenome-wide” analyses. These expanded genetic studies will be invaluable in contextualizing T1D genetic associations and understanding their effects on beta cell function. In addition to larger genetic association studies, orthogonal efforts will be needed for a complete understanding of the diverse mechanisms contributing to T1D. Here, we highlight three areas critical to a holistic view of T1D genetics and for translating genetic discovery into knowledge. Focused investment in building these resources may reveal novel therapeutic opportunities for T1D.

7.1. Diverse T1D GWAS

T1D prevalence is expected to increase by 46–78% by 2040 in most parts of the world, and more than 100% in the Middle East and Africa [249]. In the US, T1D incidence is highest in non-Hispanic whites but rising fastest among minority populations, who also have worse clinical outcomes [250]. Meanwhile, later-onset T1D is more common than previously thought [251]. Since nearly all T1D association studies have been performed on pediatric European ancestry cases, genetic studies of T1D in both non-European ancestry and later-onset cohorts are urgently needed.

Recruitment of larger, diverse T1D association cohorts will improve fine mapping of causal variants, uncover new risk loci [252], and improve genetic risk prediction for T1D. Importantly, focused recruitment efforts will be required, as there is low prevalence of T1D in population-based biobanks, particularly in non-Eurocentric groups. For example, one of the largest population biobanks to date, the UK Biobank, contains genetic and health information from more than 500,000 participants. However, over 90% of the UK Biobank samples were collected from Eurocentric populations [253]. While the UK Biobank has fueled valuable insights about the genetics of T1D [10,254,255], it contains fewer than 1,500 individuals with T1D total and fewer than 100 T1D-affected individuals of non-European ancestry, limiting its utility for understanding T1D across diverse groups.

Limited genetic studies in non-European populations highlight the importance of targeted recruitment of minority groups in T1D genetic studies. Genotyping of individuals from 22 Arab countries revealed that HLA haplotypes may have different effects on T1D risk depending on ancestry [256]. For example, the DRB1∗0401-DQB1∗0302 haplotype is protective amongst Lebanese patients [257] but confers increased susceptibility to Italian and Bahraini populations [257,258]. Different directions of effect across populations could reflect LD patterns (i.e., unmeasured causal variants residing on different haplotypes in the two populations) or environmental modifiers altering the effect of a causal variant on disease risk. In both cases, heterogeneity in effects will diminish the effectiveness of existing T1D risk prediction tools in non-Eurocentric populations. A European GRS [259] using 30 SNPs performed poorly in African-ancestry individuals compared to an African-ancestry GRS using only 7 SNPs [48]. Subsequent application of these GRS models to an independent cohort confirmed the need for more diverse T1D cohorts to improve risk prediction [260]. Moreover, if pathways contributing to disease pathogenesis are heterogeneous across age and ancestry groups, we may struggle to detect or differentiate causal mechanisms which are more prominent in the poorly represented groups. In summary, failure to address the knowledge gap between the genetics of T1D in pediatric-onset European ancestry groups and the genetics of T1D in all other populations may lead to exacerbated health disparities and missed opportunities.

7.2. Biobanks linking human islets to genetic variation

Over the past several decades, biobanking efforts of islets from human donors have expanded rapidly [261]. Biobanks store patient tissue samples linked to clinical records and are collected in a standardized manner reducing variability in tissue collection, storage, and processing. There are several biobanks relevant to the role of the pancreas in T1D. The Network for Pancreatic Organ Donors with Diabetes (nPOD) is a major T1D-specific biobank in the US, having collected tissue from nearly 200 donors with T1D, as well as 60 individuals positive for T1D autoantibodies. In addition to pancreas tissue samples, nPOD also manages the T1D exchange biobank, housing biospecimens and clinical data on over 1,000 patients with T1D. The Human Pancreas Analysis Program (HPAP) is another biobank organized and funded through the Human Islet Research Network and the National Institute of Diabetes, Digestive, and Kidney diseases (NIDDK) [262]. HPAP specializes in the procurement of whole pancreata from T1D, T1D autoantibody positive, and T2D donors along with matched non-diabetic controls [263]. Samples collected in the HPAP are subject to a diverse series of genetic, genomic, cellular, and tissue-based assays, which are made publicly available with the goal of understanding beta cell loss in T1D and T2D. The Integrated Islet Distribution Program (IIDP) is a major provider of human pancreatic islets and related tissue samples primarily from individuals without diabetes, but also includes samples from organ donors with T2D [264]. The IIDP has phenotyping and genotyping cores and offers extensive clinical, medical, and islet characteristic data to investigators [265]. Finally, the Alberta Diabetes Institute (ADI) IsletCore is another international biobank providing human pancreatic islets and other associated tissues (spleen, lymph, adipose, etc.). Notably, the ADI IsletCore provides data for several genomic assays including bulk and single cell RNA-sequencing along with electrophysiological (Patch-seq) and metabolic (Seahorse, GSIS) data [266,267]. Collectively, these efforts to bank human islet and pancreas samples and make data publicly available to researchers will help improve our understanding of the genetic regulation of these T1D-relevant tissues.

Tissue biobanks present an incredible resource for understanding genetic variant effects on beta cells and intermediate phenotypes leading to T1D, including clinical variables, hormone secretion, and molecular profiles. Existing biobanks and islet resources can also be leveraged to link intermediate features to each other (e.g., correlate molecular features to ex vivo perifusion assay readouts), to explore environmental effects on beta cells, or to study interactions between the immune system and beta cells. However, as with GWAS studies, improved recruitment of diverse patients and democratization of future and current data will be necessary to maximize the impact of biobank resources.

7.3. Expanded variant-to-gene maps in human islets

Resources for investigating genetic effects on molecular phenotypes such as gene expression within human islets are currently limited in terms of sample size, omics modalities, and quality. The largest human islet eQTL studies include only a few hundred donors and used bulk RNA-sequencing data of islets from heterogeneous sources. Incorporating molecular phenotyping of human islets into biobank and consortium efforts will provide a hugely improved resource for molecular QTL analysis. High quality molecular profiling of specific human islet cell types, for example using bulk assays of sorted cells or single cell multi-omic assays, will enable linking genetic variation to molecular features in native contexts. Furthermore, generating molecular maps using beta cells from autoantibody positive or T1D donors could link genetic variants to molecular phenotypes at specific stages of disease progression. Similarly, maps generated from islets following exposure to ex vivo stimuli could yield insight into variants that impact beta cell response to environmental stressors.

More complete molecular maps will help to close the gap between genetic discovery and molecular processes underlying T1D and help to delineate which loci affect beta cell function, survival, and crosstalk with the immune system. Specific risk alleles can be further validated in cell-based systems, such as EndoCβH and iPSCs, which are genetically tractable, renewable sources of human beta cells. Using genetically modified EndoCβH and iPSCs to identify the effects of variants on beta cell function or survival is an important area of focus to mechanistically link genetic variants to beta cell dysfunction. Finally, resolving the cell type and context of T1D associations may help provide molecular explanations for observed T1D heterogeneity and inform our understanding of potential T1D endotypes [268].

8. Concluding remarks

In the last two decades, since the advent of large-scale genetic association studies, new mechanistic frontiers have been reached that provide novel in-roads into understanding T1D pathophysiology. It is now clear that T1D is a multi-system disease where beta cells, immune cells, exocrine cells, and other cell types in the pancreas, as well as other tissues such as thymus and lymph nodes, likely play an etiological role [6,268]. Continued human genetic studies of T1D will be crucial to expand our understanding of how beta cells contribute to T1D risk. Applying stringent criteria to link risk variants to genes at T1D loci will produce more robust insights and therapeutic targets. Human genetics can also support personalized medicine approaches, including matching genetically supported T1D endotypes or causal pathways to appropriate therapies. Finally, the development of next generation models of T1D, including humanized mouse models of T1D and isogenic cell systems to study immune and beta cell crosstalk, will allow investigation of T1D risk loci and their functional effects across multiple cell types [106,131,[269], [270], [271]]. Translating risk loci into mechanistic insight will ultimately help unlock novel therapies to treat or prevent T1D.

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Availability of data and materials

Not applicable.

Funding

This work was written using resources and/or funding provided by the NIDDK-supported Human Islet Research Network (HIRN, HIRN, RRID: SCR_014393; https://hirnetwork.org), including U24DK104162 to J.S.K., U01DK127786 to D.L.E., UC4DK104162 to S.D., U01DK127747 to P.A. and S.A.S., and U01 DK127777 to S.C.J.P. and S.C. C.C.R. is supported by DK007245. B.A.H.K. acknowledges support from the NIH (T32 GM145304, T32 AI007413, F31 DK138544). P.A. is supported by NIH U01DK127747 and NIH R01 DK48280. S.D. is supported by the NIH (R01DK120523), a Human Islet Research Network New Investigator Award (via NIDDK UC4DK104162), Wanek Family Foundation to Cure Type 1 Diabetes, and City of Hope ARMDRI Pilot Awards. D.L.E. acknowledges the support of grants from JDRF International (3-SRA-2022-1201-S-B [1] and 3-SRA-2022-1201-S-B [2]); theNational Institutes of Health Human Islet Research Network Consortium on Beta Cell Death & Survival from Pancreatic β-Cell Gene Networks to Therapy (HIRN-CBDS) (grant U01 DK127786); and the National Institutes of Health NIDDK grants RO1DK126444 and RO1DK133881-01. J.S.K. is supported by U24DK104162. S.C.J.P. and S.C. are supported by DK127777. K.J.G. is supported by NIH grants DK138512, OD036440, HG012059, DK105554, the LL Hillblom Foundation, and the Foundation for the NIH. S.A.S. acknowledges support from the JDRF (COE-2019-861, SRA-2023-1392), the NIH (R01 DK108921, R01 DK135032, R01 DK135268, R01 DK136671, R01 DK127270, U01 DK127747, P30 DK020572), the Department of Veterans Affairs (I01 BX004444), the Brehm family, and the Anthony family.

CRediT authorship contribution statement

Catherine C. Robertson: Writing – review & editing, Writing – original draft, Conceptualization. Ruth M. Elgamal: Writing – review & editing, Writing – original draft, Visualization, Conceptualization. Belle A. Henry-Kanarek: Writing – review & editing, Writing – original draft, Visualization, Conceptualization. Peter Arvan: Writing – review & editing, Conceptualization. Shuibing Chen: Writing – review & editing, Conceptualization. Sangeeta Dhawan: Writing – review & editing, Conceptualization. Decio L. Eizirik: Writing – review & editing, Conceptualization. John S. Kaddis: Writing – review & editing, Conceptualization. Golnaz Vahedi: Conceptualization. Stephen C.J. Parker: Writing – review & editing, Writing – original draft, Conceptualization. Kyle J. Gaulton: Writing – review & editing, Writing – original draft, Conceptualization. Scott A. Soleimanpour: Writing – review & editing, Writing – original draft, Conceptualization.

Declaration of competing interest

KJG has done consulting for Genentech, received honoraria from Pfizer, and holds stock in Neurocrine biosciences. S.C. is the co-founders of OncoBeat, LLC. S.A.S has received grant funding from Ono Pharmaceutical Co., Ltd. and is a consultant for Novo Nordisk. DLE is a member of the Scientific Advisory Board of InSphero AG.

Data availability

No data were used for the research described in the article.