Type 1 Diabetes Genetic Risk Contributes to Phenotypic Presentation in Monogenic Autoimmune Diabetes

Highly penetrant causative variants in autoimmune-related genes including AIRE, FOXP3, IL2RA, STAT3, LRBA, and TNFAIP3 lead to highly variable syndromes of monogenic autoimmunity, with some individuals presenting with monogenic autoimmune diabetes (1,2). Among individuals with these pathogenic variants, the prevalence of autoimmune diabetes is variable (ranging from 17% in LRBA to 71% in FOXP3) (1). Variability in phenotypic expression exists within families harboring the same causative variant, but the mechanisms are not understood (1).

While monogenic diseases typically result from highly penetrant rare variants (odds ratios [ORs] typically 100s to 1,000s) (3), polygenic diseases result from the cumulative impact of multiple common variants with moderate effect sizes interacting with environmental factors (4). Type 1 diabetes is a polygenic autoimmune disorder with 92 reported causative genetic loci (5) and genetic heritability (approximately >50%) largely attributable to HLA gene variants in the major histocompatibility region (6). HLA genes are critical for regulation and function of the immune system, showing strong association with multiple polygenic immune-mediated diseases (6,7). The penetrance of monogenic disease variants can be modified by polygenic risk variants, for example, in familial hypercholesteremia and Lynch syndrome (4,8).

We hypothesized that type 1 diabetes polygenic risk loci may contribute to phenotype variability in individuals with monogenic autoimmune diabetes. We tested this with a 67 single nucleotide polymorphism (SNP) type 1 diabetes genetic risk score (T1D-GRS) (9).

We calculated T1D-GRS in 62 individuals with monogenic autoimmune diabetes recruited by clinicians to the Exeter Genomics Laboratory for genetic testing through a dedicated referral form (https://www.diabetesgenes.org/download/217/?tmstv=1716890637). All individuals carried a pathogenic/likely pathogenic variant in one of six genes: AIRE, FOXP3, IL2RA, LRBA, STAT3, or TNFAIP3 (Table 1). These individuals either presented with diabetes very early in life (<52 weeks), had additional infancy-onset autoimmunity, or both.

We studied 285 individuals with type 1 diabetes. These individuals all had a clinical diagnosis of type 1 diabetes according to the World Health Organization diabetes diagnostic criteria (10), were diagnosed >6 months, and were without additional features (2) or family history suggesting monogenic disease.

Individuals diagnosed with type 1 diabetes very early in life have a slightly increased T1D-GRS (9). Because of the overlapping age of diagnosis in early-onset type 1 and monogenic autoimmune diabetes (Supplementary Fig. 1), we categorized individuals into either early-onset type 1 diabetes (diagnosed 6 months to <2 years of age, n = 247) or childhood-onset type 1 diabetes (n = 38, diagnosed 2–18 years of age) to investigate the discriminatory ability of T1D-GRS in those diagnosed at <2 years of age.

We used participants from the Exeter 10,000/Peninsula Research Bank project (EXTEND/PRB), an unselected population-based study of 11,074 individuals from the southwest U.K. (11). Individuals within EXTEND/PRB reporting type 1 diabetes were excluded (n = 249). After sample filtering, our study included 10,405 participants (Table 1).

We studied 180 individuals with nonautoimmune neonatal diabetes (NDM; diagnosed <6 months of age) recruited by clinicians to the Exeter Genomics Laboratory for genetic testing through a separate dedicated referral form for NDM (https://www.diabetesgenes.org/download/3564/?tmstv=1716890637). These individuals had pathogenic variants in either EIF2AK3, INS, ABCC8, or KCNJ11 (Table 1).

Individuals were genotyped using Illumina Global Screening Array using whole-blood DNA samples. Individuals with monogenic autoimmune diabetes, type 1 diabetes, and NDM were grouped together for genotyping. Standard data processing, quality control, and imputation were used (Supplementary Methods) (12–15).

To determine type 1 diabetes risk allele contribution to monogenic autoimmune diabetes, we used the 67 SNP T1D-GRS (9). We used previously defined DQA1 and DQB1 allele combinations tagging DR-DQ haplotypes to determine HLA status (9). T1D-GRSs were generated as previously described (9).

We further separated T1D-GRS into the 18 class II HLA haplotype interactions, 16 nonclass II HLA loci, and 32 non-HLA loci components.

Receiver operating characteristic (ROC) curves were used to assess the discriminatory power of the T1D-GRS model. A logistic regression model was used to test for associations between T1D-GRS (predictor) and disease (type 1 diabetes vs. no type 1 diabetes) as a binary outcome variable (one or zero). Genetic ancestry effects were controlled for by adjusting for the first five genetic principal components within the logistic regression model by cohort (control cohort: 96% White European; monogenic autoimmune diabetes: 29%; nonautoimmune NDM: 36%). Cochran Q test was used to test for evidence of heterogeneity in meta-analysis of T1D-GRS in different phenotypes. Statistical analyses were performed in R.

The monogenic autoimmune diabetes and NDM control cohorts are within the Genetic Beta Cell Research Bank, which has overarching ethical approval from the North Wales Research Ethics Committee, Wrexham, U.K. (Integrated Research Application System [IRAS] project ID 231760). Individuals with early-onset type 1 diabetes were recruited to the Extremely Early-Onset Type 1 Diabetes (EXE-T1D) study, which has ethical approval from Derby Research Ethics Committee, Derby, U.K. (IRAS project ID 228082). The EXTEND/PRB study has ethical approval from South West-Cornwall & Plymouth NHS Research Ethics Committee, Bristol, U.K. (reference 14/SW/1089; 5-year extension following initial approval: reference 09/H0106/75).

The code to generate scores can be found on an open GitHub repository (https://github.com/USF-HII/hla-prs-toolkit). All nonclinical data analyzed during this study are included in this published article (and its Supplementary Material). Clinical and genotype data can be used to identify individuals and are therefore available only through collaboration to experienced teams working on approved studies examining the mechanisms, cause, diagnosis, and treatment of diabetes and other β-cell disorders. Requests for collaboration will be considered by a steering committee after an application to the Genetic Beta Cell Research Bank (https://www.diabetesgenes.org/current-research/genetic-beta-cell-research-bank/). Contact by e-mail should be directed to M.B.J. ([email protected]). All requests for access to data will be responded to within 14 days.

Individuals with monogenic autoimmune diabetes were diagnosed with diabetes at a median age of 6 weeks (interquartile range [IQR] 2–26), which was similar to the nonautoimmune NDM group (median age = 9 weeks, IQR 5–15.25, P = 0.8) (Table 1 and Supplementary Fig. 1). For individuals in the type 1 diabetes group, the age of diagnosis was higher (median age = 58.4 weeks, IQR 45–89.4) compared with monogenic autoimmune diabetes (P = 1.2 × 10⁻²⁶). There was an overlap in age of diagnosis between those with monogenic autoimmune diabetes and those with the earliest diagnoses of early-onset type 1 diabetes (Supplementary Fig. 1).

Individuals with monogenic autoimmune diabetes had a higher T1D-GRS compared with those with nonautoimmune NDM (mean GRS = 11.3, 95% CI = 10.8–11.9 vs. mean GRS = 9.8, 95% CI = 9.5–10.1, P = 1.7 × 10⁻⁵) and the control group (mean GRS = 10.3, 95% CI = 10.2–10.3, P = 7.5 × 10⁻⁶) (Fig. 1) but lower than those with type 1 diabetes (mean GRS = 14.9, 95% CI = 14.7–15.1; P = 3.3 × 10⁻²¹). T1D-GRS of individuals with nonautoimmune NDM was similar to the control group (P = 0.93). We observed similar results in an ancestry-specific sensitivity analysis split by European versus non-European genetic ancestry (Supplementary Fig. 2).

ROC area under the curve (AUC) analysis indicated that, despite T1D-GRS having greater discriminative power (P = 0.01) between early-onset type 1 diabetes versus NDM (ROC AUC 0.95, 95% CI = 0.94–0.97) compared with early-onset type 1 diabetes versus monogenic autoimmune diabetes (ROC AUC 0.90, 95% CI = 0.86–0.94), it remained strongly discriminative in both cohorts (Supplementary Fig. 3). Using a cutoff of below the 25th T1D-GRS centile of early-onset type 1 diabetes identifies ∼87% of monogenic autoimmune diabetes (Supplementary Fig. 4).

As the HLA risk alleles make up ∼50% of T1D-GRS, we next assessed whether the HLA and non-HLA components of the score contributed differently to monogenic autoimmune diabetes phenotype. Individuals with monogenic autoimmune diabetes had a higher class II HLA GRS (mean class II HLA GRS = 0.36, 95% CI = −7.5 × 10⁻⁵ to 0.74) compared with nonautoimmune NDM (mean class II HLA GRS = −0.69, 95% CI = −0.95 to −0.43; P = 2.8 × 10⁻⁴) and the control group (mean class II HLA GRS = −0.56, 95% CI = −0.59 to −0.52; P = 8.2 × 10⁻⁵) (Fig. 2), but a much lower class II HLA GRS compared with type 1 diabetes (mean GRS = 2.25, 95% CI = 2.07–2.42; P = 1.1 × 10⁻¹³). Individuals with monogenic autoimmune diabetes had a similar non-HLA GRS (mean non-HLA GRS = 3.4, 95% CI = 3.1–3.6) compared with nonautoimmune NDM (mean non-HLA GRS = 3.3, 95% CI = 3.2–3.5; P = 0.95) and the control group (mean non-HLA GRS = 3.5, 95% CI = 3.5–3.5; P = 0.51) (Fig. 2), but this was again lower than the non-HLA GRS for type 1 diabetes (mean non-HLA GRS = 4.3, 95% CI = 4.17–4.35; P = 2.6 × 10⁻¹⁰).

Using imputed HLA variants, we determined that the higher class II HLA haplotype genetic risk was driven by a higher frequency of the HLA DRB1*03:01-DQB1*02:01-DQA1*05:01 (DR3-DQ2) haplotype in monogenic autoimmune diabetes. DR3-DQ2 had increased contribution to disease risk in monogenic autoimmune diabetes compared with nonautoimmune NDM (OR = 1.19; P = 5.4 × 10⁻⁴) and control participants (OR = 1.16; P = 1.1 × 10⁻³) but lower compared with type 1 diabetes (OR = 0.87; P = 2.1 × 10⁻³) (Fig. 3 and Supplementary Tables 1 and 2).

We have shown that genetic loci associated with type 1 diabetes genetic susceptibility play a role in the development of monogenic autoimmune diabetes. We found a higher T1D-GRS in 62 individuals with monogenic autoimmune diabetes compared with 10,405 individuals without diabetes and 180 individuals with monogenic nonautoimmune NDM. Class II HLA, specifically the high-risk type 1 diabetes DR3-DQ2 haplotype, drove the higher GRS within monogenic autoimmune diabetes.

Our data support the hypothesis that background polygenic risk, particularly HLA-mediated risk, contributes to diabetes phenotype development in monogenic autoimmunity. We observed higher T1D-GRSs in monogenic autoimmune diabetes compared with the control groups, although still markedly lower than in type 1 diabetes. This is consistent with other monogenic diseases showing modification with polygenic risk, including hereditary breast cancer and cystic fibrosis (4,8).

The higher T1D-GRS observed was driven by the class II HLA DR3-DQ2, one of the highest type 1 diabetes risk haplotypes (OR = 3.54) (6,9). The observed interaction between HLA alleles and monogenic autoimmune variants suggests overlapping pathophysiological pathways, underscoring the involvement of adaptive (auto)immunity. Previous studies have shown that APS-1 (or APECED, caused by pathogenic variants in AIRE) phenotypes are impacted by HLA haplotypes (16), including a protective effect from DRB1*15:01-DQB1*06:02 (DR15-DQ6.2) against diabetes. As all our cases had diabetes, we were unable to assess this protective effect. Our results show that specific HLA haplotypes have the same directional effect (i.e., risk for DR3-DQ2 and protection for DR15-DQ6.2) in monogenic autoimmune diabetes and polygenic type 1 diabetes (9,16). Nevertheless, in monogenic autoimmunity, the overwhelming disease risk comes from the highly penetrant causative monogenic gene variant (3,17).

Genetic risk scores are being adopted in genetic testing pathways to guide testing and interpret results (1,18). The overlap in disease age of onset between polygenic early-onset type 1 diabetes and monogenic autoimmune diabetes highlights the utility of GRSs in discerning phenotypes. As there is evidence of interaction, higher thresholds should be used to select individuals for testing. A T1D-GRS cutoff at the 25th centile of early-onset type 1 diabetes could be used to identify ∼87% of monogenic autoimmune diabetes cases while excluding 75% of polygenic type 1 diabetes.

This study is the largest to date in monogenic autoimmune diabetes and therefore has greater power to detect a role for polygenic risk than previous studies (2). The near doubling of sample size (n = 62 vs. n = 37) and the increased discrimination using the 67 SNP score, as opposed to a 10 SNP score, is the likely reason this study detected a novel contribution of T1D-GRS to monogenic autoimmune diabetes. Despite this, we did not have statistical power to demonstrate evidence of GRS differences in genetic ancestry-stratified analyses within our cohort; however, we observed similar results in European versus non-European genetic ancestries. Future work in a larger and more diverse cohort is needed.

In conclusion, we have shown that polygenic risk variants, particularly HLA class II, likely contribute to clinical presentation in monogenic autoimmune diabetes.

Acknowledgments. The authors thank the patients, their relatives, and their physicians for participating in this study. This study was supported by the National Institute for Health and Care Research Exeter Biomedical Research Centre and the National Institute for Health and Care Research Exeter Clinical Research Facility.

The views expressed are those of the authors and not necessarily those of the National Institute for Health and Care Research or the Department of Health and Social Care.

Funding. This study was funded by The Leona M. and Harry B. Helmsley Charitable Trust (grants 2016PG-T1D049, 2018PG-T1D049, 2103–05059, and G-2404-06858). M.B.J. is a Diabetes UK and Breakthrough T1D (formerly Juvenile Diabetes Research Foundation) RD Lawrence Fellow (23/0006516). E.D.F. is a Diabetes UK RD Lawrence Fellow (19/005971) and the recipient of a European Foundation for the Study of Diabetes/Novo Nordisk Foundation Future Leaders Award (NNF23SA0087432). R.A.O. is a Diabetes UK Harry Keen Fellow (16/0005529). R.A.O. and M.N.W. had a U.K. Medical Research Council Confidence in Concept grant to develop a type 1 diabetes GRS biochip with Randox and have ongoing research funding from Randox R&D. W.A.H. is supported by U.S. National Institutes of Health grant 5U01DK128847 and by an unrestricted research grant from Sanofi U.S. A.M.L. is funded by a PhD studentship from Randox Laboratories Ltd.

Duality of Interest. R.A.O. has served as a consultant for Sanofi Pharmaceuticals, ProventionBio, and Janssen Pharmaceuticals. R.A.O. and M.N.W. had a U.K. Medical Research Council confidence-in-concept award to develop a type 1 diabetes Genetic Risk Score biochip with Randox R & D and have ongoing research funding from Randox Laboratories Ltd. A.M.L. is funded by a PhD studentship from Randox Laboratories Ltd. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.M.L. wrote the manuscript and researched the data. G.H. and H.D.G. contributed to the methods and reviewed the manuscript. E.D.F., W.A.H., B.O.R., M.N.W., and R.A.O. reviewed/edited the manuscript and contributed to the discussion. M.B.J. wrote the manuscript. M.B.J. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.