Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Type 1 diabetes is an autoimmune disease that, in part, is genetically determined.Younger age at diagnosis of type 1 diabetes is associated with a steeper beta-cell decline.WHAT THIS STUDY ADDS Individuals who were compound heterozygous for the HLA-DR3/DR4 genotypes were on average 3 years younger at diagnosis compared with non-carriers.A single-nucleotide polymorphism near the HLA region rs76730244 was associated with age at diagnosis independently of the HLA-DR3/DR4 genotypes.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY These findings could provide insight into the pathological mechanisms associated with young onset age of type 1 diabetes.Introduction Type 1 diabetes is a complex disease characterized by the autoimmune destruction of insulin-producing beta cells in the pancreas. Although previously named “juvenile diabetes,” it is a heterogeneous disease with onset at any age. Type 1 diabetes is a disease with major genetic contributions, most strikingly the HLA genotypes, which are strongly associated with susceptibility to the disease. 1 Specifically, individuals who are compound heterozygote HLA-DR3/4 genotypes have an almost 50-fold increased risk to develop type 1 diabetes.2 In addition, at least 78 other genetic loci have been shown to be associated with type 1 diabetes incidence.3 4Approximately half of individuals with type 1 diabetes develop the disease before the age of 21 years.5–7 Only limited information is available about the factors that are associated with early disease onset. Earlier studies have suggested that family history, serum C-peptide, HLA genotypes, and other single-nucleotide polymorphisms (SNPs) are associated with age at diagnosis.8–13 Mainly the HLA-DR3/DR4 genotypes—already indicating higher susceptibility to develop diabetes—are also associated with a younger age at onset. Furthermore, variants in or near several other genes such as PTPRK/THEMIS (6q22.33) and PHF20L1 (8q24.22), as well as multiple loci on chromosome 17 (near IKZF3 and GRB7), have been associated with age at diagnosis.11–13There have been six studies to date that tested the association between genetic markers and age at diagnosis of type 1 diabetes. Four of these age at diagnosis studies have used, in part, overlapping cohorts in their analyses.12–15 Age at diabetes diagnosis was examined either as a quantitative trait11–13 15 or by comparing type 1 diabetes risk in categories of age at diagnosis; for example, under or over the age of 5 years12 and under the age of 7 years versus 13 years and over.14 In three studies, only candidate markers, previously associated with type 1 diabetes risk, were selected.12 15 16 The other studies reported results of a genome-wide association study (GWAS), analyzing markers across the genome for association with age at diagnosis.11–13 None of the previously studied cohorts were specifically recruited for genetic analysis of type 1 diabetes age at diagnosis.In this study, we aimed to evaluate which genetic factors are associated with age at diagnosis of type 1 diabetes. For this purpose, we performed a meta-analysis of genome-wide association data obtained from eight North American/Canadian and European cohorts. Additionally, we sought to test whether results from earlier reported polymorphisms could be replicated.Methods Study design We studied age at diagnosis of type 1 diabetes in participants from six Northern American (USA and Canada) cohorts, that is, Coronary Artery Calcification in type 1 diabetes, Diabetes Control Complications Trial/Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC), Epidemiology of Diabetes Complications (EDC), Genetics of Kidneys in Diabetes, Renin Angiotensin System Study, and Wisconsin Epidemiologic Study of Diabetic Retinopathy, and two cohorts from the Netherlands. 17–23 A detailed description of the cohorts can be found in the online supplemental information. We performed GWAS of age at diagnosis by cohort and subsequently meta-analyzed the results. Conditional GWAS was performed on the complex DR3/DR4 genotypes to identify independent (ie, not in linkage disequilibrium (LD)) SNPs. The HLA-imputed data were studied similarly. We studied the association of a type 1 diabetes Genetic Risk Score (GRS) with age at diagnosis and also analyzed the effects of individual type 1 diabetes loci on age at diagnosis.2 Finally, we attempted to replicate genetic polymorphisms that were reported earlier.12–16 The power of detection of a reported SNP association was calculated using QUANTO software.24 We calculated the power of replication given the reported effect size and allele frequency from the discovery study and sample size in our cohort.SP110.1136/bmjdrc-2023-003877.supp1Supplementary dataGenotyping and imputation Genotyping was performed using different genotyping arrays ( online supplemental table 1). Quality control of the data was performed according to previously described guidelines.25 All subjects with genotyping call rates below 95% were removed. Ancestry was inferred using Kinship-based INference for Gwas (KING) V.2.2.9 software.26 Individuals who were not from inferred European descent were excluded. Outliers on autosomal heterozygosity (±2 SD from the mean) were removed. Relatedness was assessed using KING V.2.2.9 software and cryptically related individuals were excluded. Otherwise related individuals were retained in the datasets, and imputed SNPs were analyzed with specific software (described below), which is capable of adjusting for relatedness within cohorts. SNPs with call rates below 90% were removed. Data were imputed by cohort to the HRC reference panel V.1.1 on the Sanger imputation server,27 and the recommended EAGLE2 pipeline28 was used for phasing.GWAS and meta-GWAS Well-imputed SNPs with IMPUTE2 imputation quality (INFO) score>80% 29 were analyzed. GWAS was performed with GCTA V.1.92beta30 using the Mixed Linear Model Association (MLMA) option, in which a linear mixed model is analyzed whereby the model is corrected for a calculated genetic relationship model. The genetic relationship model corrects the mixed model for population stratification and possible relatedness between individuals within cohorts. The results were then meta-analyzed using METAL V.1.5.31 We used the Standard Error (STDERR) method, in which effect size estimates are weighted using the inverse of the corresponding SE. In addition, we performed a meta-analysis of the suggestive SNPs (p<5×10−5) from a recently published GWAS of age at diagnosis by Syreeni et al13 and our GWAS of age at diagnosis with no covariates in the model.13 Specific to this substudy, to allow meta-analysis using the STDERR method (METAL V.1.5), we natural log-transformed age at diagnosis in all cohorts.Major histocompatibility complex (MHC) imputation Classical HLA alleles, their amino acid sequences, and additional SNPs in the MHC region were imputed to the Type 1 Diabetes Genetic Consortium (T1DGC) reference dataset using SNP2HLA software ( http://software.broadinstitute.org/mpg/snp2hla/). The T1DGC dataset includes 5196 unrelated individuals of whom 182 have type 1 diabetes.32 The T1DGC panel contains four-digit classical HLA types (HLA-A, HLA-B, HLA-C, HLA-DPA1, HLA-DPB1, HLA-DQA1, HLA-DQB1, and HLA-DRB1) and 5868 SNPs (genotyped using the Illumina Immunochip).14 The total number of imputed variants in the MHC region was 8961.Statistical analysis We performed GWAS using multiple models. Model 1 was a straightforward genome-wide analysis of age at diagnosis without covariates. Model 2 was a GWAS including HLA-DR3/4 genotypes as categorical covariates with six categories (DR3/DR4, DR3/DR3, DR4/DR4, DR3/X, DR4/X, and X/X, where X=non-DR3/DR4). Model 3 comprised a conditional analysis, in which the genome-wide significantly associated top-SNP identified with model 2 was coded additively as a covariate.Specific analysis of the MHC imputed data was performed as follows: model 1 was an analysis of the association of MHC variants and haplotypes with age at diagnosis without covariates, while model 2 included a categorical covariate representing the DR3/DR4 genotypes. Using Bonferroni correction for multiple testing p<5×10−8 was considered statistically significant for GWAS.33 Manhattan plots were created using R V.3.1.034 and the qqman package.23 A region plot of the top-SNP from model 2 was created using locuszoom (http://locuszoom.sph.umich.edu/locuszoom/).35 Plots of the allele frequency of the identified locus in model 2 were created using RStudio V.1.2.503336 and the ggpubr package.37 Allele frequency was plotted as a percentage with 95% CI (calculated using two-tailed Binomial tests) across quantiles of age at diagnosis. European-ancestry individuals from the Lifelines population (n=36 339), genotyped using the Infinium Global Screening Array, were used as controls.38 Manhattan plots of the MHC analyses were created using HLA-TAPAS.39Testing of previously associated loci Previously identified loci for age at diagnosis were tested for association with age at diagnosis in the meta-analysis of age at diagnosis with no covariates in the model (model 1). In addition, a type 1 diabetes GRS 2 was tested for association with age at diagnosis. The type 1 diabetes GRS consists of 28 SNPs from across the genome, including three SNPs in the HLA region and, in addition, a score for the six HLA-DR3/DR4 genotype categories. Linear regression models were performed to test the association of the GRS with age at diagnosis in a mega-analysis of all cohorts combined. The GRS was tested using the total sum of the score, as well as divided into three subcategories: DR3/DR4 genotype categories, the sum of the HLA variants, and the sum of non-HLA variants. All previously identified non-HLA SNPs for type 1 diabetes risk, resulting from the latest type 1 diabetes study by Robertson et al,3 were tested for association with age at diagnosis in model 1. A plot of the tested SNPs was created by plotting the age at diagnosis beta against the SNP’s log-odds ratio for type 1 diabetes risk (EUR case–control ancestry cluster) using RStudio V.1.2.503336 and the ggplot2 package.40Results Table 1 depicts the main characteristics of the included cohorts. The mean age at diagnosis varied between cohorts, from 7.9±4.0 (SD) years in EDC up to 21.2±8.1 years in DCCT/EDIC. Distributions of age at diagnosis by cohort are visualized in online supplemental figure 1. Overall mean age at diagnosis for all cohorts combined was 15.5±9.4 years. Mean GRS was similar among cohorts, as was the distribution of DR3/DR4 genotypes (table 1).Table 1Baseline characteristics of study populationsAge at diagnosis, mean (SD)T1D GRS, mean (SD)DR3/DR4,N (%)DR4/DR4,N (%)DR3/DR3,N (%)DR4/X,N (%)DR3/X,N (%)X/X,N (%)CACTI, n=54213.1 (8.0)16.1 (1.5)159 (29%)39 (7%)38 (7%)156 (29%)91 (17%)59 (11%)EDC, n=4247.9 (4.0)16.0 (1.5)126 (30%)12 (3%)29 (7%)123 (29%)82 (19%)52 (12%)DCCT/EDIC, n=132021.2 (8.1)15.9 (1.5)338 (26%)90 (7%)114 (9%)335 (25%)225 (17%)218 (17%)GoKind, n=171512.4 (7.0)16.0 (1.6)542 (32%)101 (6%)175 (10%)369 (22%)285 (17%)243 (14%)DT1B, n=54915.1 (11.4)16.2 (1.4)174 (32%)33 (6%)57 (10%)133 (24%)94 (17%)58 (11%)Nijmegen, n=50920.8 (12.4)16.1 (1.5)139 (27%)31 (6%)42 (8%)147 (29%)90 (18%)60 (12%)RASS, n=22118.6 (10.2)15.9 (1.6)58 (26%)11 (5%)22 (10%)52 (24%)47 (21%)31 (14%)WESDR, n=63013.9 (7.5)16.0 (1.7)203 (32%)40 (6%)39 (6%)168 (27%)89 (14%)91 (14%)All, n=591015.5 (9.4)16.0 (1.7)1739 (29%)357 (6%)516 (9%)1483 (25%)1003 (17%)812 (14%)CACTI, Coronary Artery Calcification in type 1 diabetes; DCCT/EDIC, Diabetes Control Complications Trial/Epidemiology of Diabetes Interventions and Complications; DT1B, Dutch Type 1 diabetes Biomarker Study; EDC, Epidemiology of Diabetes Complications; GoKind, Genetics of Kidneys in Diabetes; RASS, Renin Angiotensin System Study; T1D GRS, Type 1 diabetes Genetic Risk Score; WESDR, Wisconsin Epidemiologic Study of Diabetic Retinopathy; X, non-DR3/DR4.Genome-wide significant signals Meta-analyses of model 1 showed association of the C allele of rs2856721, A>C B (SE) = 1.19 (0.18), p=3.3×10 −11, effect-allele frequency (EAF)=0.16 in model 1 (online supplemental figures 2 and 3). This SNP is in the MHC region near HLA-DQB1, and it is present on seven out of eight haplotypes of the MHC Haplotype Project.41In the meta-analysis of model 2, age at diagnosis adjusted for the DR3/DR4 categories, rs2856721 (C allele) was nominally significantly associated with age at diagnosis B(SE)= 0.39 (0.18), p=0.03. With this model, we identified another locus in the MHC region with the most significant association being rs76730244, C>T B (SE) = −1.24 (0.21), p=4.9×10−9, EAF=0.12 (online supplemental table 2). The results by study (prior to meta-analysis), including the INFO score, are depicted in online supplemental table 2. This SNP is in the intergenic region of HLA-A/HLA-G (figure 1, online supplemental figures 4 and 5). The minor allele frequency of the T allele of rs76730244 by quartiles of age at diagnosis is depicted in figure 2. Meta-analysis of model 3, which was age at diagnosis adjusted for the DR3/DR4 genotypes and rs76730244 (coded additively), did not reveal any additional genome-wide significant signals (online supplemental figures 6 and 7).Figure 1Manhattan plot of age at diagnosis in meta-analysis with DR3/DR4 genotype categories included as covariate (model 2). λgc: 1.0108 (Q–Q plot visualized in online supplemental figure 3. Variants are plotted on the x-axis according to their position on each chromosome (Hg19) with the −log10(p value) of the association test on the y-axis. The red line indicates the threshold for genome-wide significance (p=5×10−8) and the blue line indicates the suggestive loci (p=1×10−5).Figure 2Minor allele frequency of the T allele of rs76730244 by age at diagnosis quartiles.Analysis of the MHC region Analysis of the MHC region resulted in one significantly associated locus near HLA-DQA1 in model 1 (online supplemental figure 8A). None of the classical HLA alleles or amino acid changes reached the genome-wide significance threshold in model 2, which was age at diagnosis adjusted for the DR3/DR4 genotype categories (online supplemental figure 8B). Notably, rs76730244 was not imputed in the HLA imputation, rs422343 was the most nearby in the HLA-imputed analyses, and it is 77 bases upstream from rs76730244. These SNPs were not in LD (EUR populations, D′=0.32 and R2=0.002).42Association of a GRS for type 1 diabetes The most common DR3/DR4 genotype category was the compound heterozygote combination of DR3/DR4, n=1739, 29%. In total, 14% of the individuals (n=812) did not carry the DR3 nor DR4 genotype (X/X), n=812 ( table 1). Individuals with X/X genotypes were significantly older at diagnosis than individuals carrying any other DR3/DR4 genotype combination (table 2 and online supplemental figure 9). All subcategories of the GRS (DR3/DR4 categories, HLA SNPs, and non-HLA SNPs) were separately and combined nominally significantly associated with age at diagnosis, with and without study indicator in the linear regression model (table 2).Table 2Regression analysis of Genetic Risk Score for age at diagnosis in all cohortsLinear model for age at diagnosis including study indicator as covariateLinear model for age at diagnosis not including study indicator as covariateBetaSEP valueBetaSEP value(Intercept)27.241.89<2×10−1629.992.07<2×10−16Study CACTIRef.––– DT1B2.040.515.3×10−5––– EDC−5.220.54<2×10−16––– DCCT/EDIC7.920.43<2×10−16––– GoKind−0.790.410.06––– NM7.680.51<2×10−16––– RASS5.380.678.3×10−16––– WESDR0.800.490.10–––DR3-DR4 genotypes X/XRef.Ref. DR3/X−0.950.390.02−1.160.440.008 DR4/X−1.590.371.3×10−5−1.750.411.8×10−5 DR4/DR4−1.240.530.02−1.080.590.07 DR3/DR3−1.180.470.01−1.390.530.008 DR3/DR4−3.140.36<2×10−16−3.660.40<2×10−16GRS HLA variants−0.770.210.0002−0.950.234.3×10−5GRS non-HLA variants−1.060.161.8×10−11−0.880.185.3×10−7CACTI, Coronary Artery Calcification in type 1 diabetes; DCCT/EDIC, Diabetes Control Complications Trial/Epidemiology of Diabetes Interventions and Complications; DT1B, Dutch Type 1 diabetes Biomarker Study; EDC, Epidemiology of Diabetes Complications; GoKind, Genetics of Kidneys in Diabetes; GRS, Genetic Risk Score; NM, type 1 diabetes participants treated at the Radboud Medical Center in Nijmegen; RASS, Renin Angiotensin System Study; Ref, reference; WESDR, Wisconsin Epidemiologic Study of Diabetic Retinopathy.Association of published type 1 diabetes SNPs with age at diagnosis Of the 78 non-HLA SNPs for type 1 diabetes risk as reported by Robertson et al,3 9 were associated with age at diagnosis of type 1 diabetes in model 1 with nominal significance (p<0.05, figure 3 and online supplemental table 3). All significant SNPs were consistent in effect; the allele associated with higher type 1 diabetes risk was associated with younger age at onset and vice versa. Alleles associated with lower type 1 diabetes risk were associated with older age at onset.Figure 3Association of 78 non-HLA single-nucleotide polymorphisms for type 1 diabetes risk with age at diagnosis (meta-analysis model 1, no covariates).Meta-analysis of suggestive SNPs Meta-analysis of the suggestive SNPs from the recently published study by Syreeni et al’s paper13 and our meta-analysis of natural log-transformed age at diagnosis showed no genome-wide significant signals (online supplemental table 4). Online supplemental figure 10A–D depicts the Manhattan and Q–Q plots of the GWAS of the raw and natural log-transformed age at diagnosis with no covariates in the model. The most significant association was rs2941522 on chromosome 17 near GRB7, which reached borderline genome-wide significance (p=6.4×10−8). This signal was already genome-wide significantly associated by Syreeni et al.13 and reached borderline nominal significance with a similar direction of effect in our original meta-analysis (p=0.055, online supplemental table 5). Power for replication of this signal at significance of p=0.05 (two sided) was 0.98. Manhattan plots and Q–Q plots of the p values of linear and natural log-transformed age at diagnosis meta-GWAS with no covariates are visualized in online supplemental figure 10.Discussion We performed a meta-GWAS of age at diagnosis in almost 6000 individuals with type 1 diabetes from six Northern American/Canadian and two European cohorts, and confirmed that HLA-DR3/DR4 genotypes, which are strongly associated with type 1 diabetes risk, are associated with younger age at diagnosis. Compound heterozygotes for the HLA-DR3/DR4 genotype had the youngest age at diagnosis. In addition, we identified one independent locus in the MHC region and replicated six SNPs (with nominal significance) that were reported in previous studies of age at diagnosis (online supplemental table 5). Perhaps not surprisingly, multiple genomic regions associated with type 1 diabetes risk were also associated with age at diagnosis.We identified that rs2856721 was associated with age at diagnosis (p=3.3×10−11) in the meta-analysis of model 1, in which no covariates were included. Rs2856721 is an SNP in the MHC region near HLA-DQB1 and MTCO3P1. The SNP effect was attenuated (p=0.03) when the HLA-DR3/DR4 genotype categories were included as covariates in the meta-analysis of model 2, indicating that this variant in the MHC region is not an independent association for age at diagnosis. In the meta-analysis of model 2, another SNP in the HLA region, rs76730244, was associated with age at diagnosis (p=4.9×10−9). The fact that this locus was genome-wide significantly associated with age at diagnosis in a model that included the HLA-DR3/DR4 genotype categories suggests that this is an independent locus for age at diagnosis. In the meta-analyses of the MHC-imputed data, we did not identify additional variants within this region that were associated with age at diagnosis independently of HLA-DR3/DR4 categories. It should be noted, however, that rs76730244 was not imputed in the HLA-specific analysis.We replicated six SNPs for age at diagnosis identified in previous papers with nominal significance (online supplemental table 5). We meta-analyzed our results with the suggestive SNPs (p<5×10−5) from the most recent study of age at diagnosis by Syreeni et al.13 In order to perform this analysis using the STDERR approach in METAL, we natural log-transformed age at diagnosis in our cohorts. In this meta-analysis, we did not identify genome-wide significant signals. rs2941522 (C>T) at 17q12 was genome-wide significantly associated with age at diagnosis by Syreeni et al,13 and including our data in meta-analysis slightly attenuated the SNP effect (p=6.8×10−8). The direction of effect, however, was similar in our data and the SNP reached borderline nominal significance in our meta-analysis of age at diagnosis of type 1 diabetes on model 1 (no covariates), with a p value of 0.055 (online supplemental table 5). The fact that we were unable to replicate this locus at 17q12 with nominal significance despite having sufficient power may be explained by several factors. First of all, Syreeni et al13 included individuals with type 1 diabetes from three different cohorts. The largest one was the FinnDiane study, in which type 1 diabetes was defined as a diabetes onset before the age of 40 years, and insulin treatment started within 1 year from diagnosis. The other cohorts were UK GRID and Sardinia, which included individuals with diabetes mostly diagnosed in childhood. The mean age at diagnosis varied between 7 and 13 years among these three cohorts. In our current study, we included individuals from Northern America/Canada and the Netherlands, whose mean age at diagnosis varied from 8 to 21 years, with an overall mean of 15 years. Second, the locus on chromosome 17q12 may have a different magnitude of effect in different populations (European vs American/European). Third, as a result of the phenomenon called the “winners curse,” there may be an overestimation of the effect size in the discovery study and, as a result, our study may actually be underpowered for replication.43In our analysis of SNPs that were previously associated with type 1 diabetes risk, we found that, in addition to HLA-DR3/DR4 genotype categories, multiple loci outside of the HLA region were also associated with age at diagnosis. Loci in or near FASLG, IFIH1, CTLA4, ASCC2/LIF, CCR5, RBPJ, PLGRKT, IL2RA, RNLS, GSDMB, PRKD2 were associated with age at onset. For all these loci, a higher risk for type 1 diabetes corresponded to a younger age at onset. Notably, INS was not associated with age at diagnosis, while this SNP showed strong association with type 1 diabetes risk and was significantly associated with age at diagnosis and residual C-peptide secretion in individuals with type 1 diabetes in other studies.11 44 The fact that this SNP is associated with both traits suggests a potential shared mechanism between age at diagnosis and residual C-peptide secretion.We chose to analyze age at diabetes diagnosis as a continuous variable because the transformation of the data did not make the distribution more normally distributed. Only for the purpose of comparing our results and meta-analyzing them with those of Syreeni et al,13 we also analyzed natural log-transformed data. Although there were some differences at the individual SNP level, no major differences between these approaches were observed (online supplemental figure 5). Other investigators have used different statistical approaches for analysis of the same trait. For instance, Inshaw et al compared age at onset <7 years old and >13 years old.14 While some loci may have effects that are relatively homogeneous across age at diagnosis groups, other loci may have effects only in certain age groups. As a result, there may not be a single optimally powerful approach to identify loci associated with age at diagnosis. Opting for a quantitative trait versus a categorical data approach can be a difficult analytic decision and also raises the issue of multiple testing.Strengths and limitations We studied multiple well-characterized cohorts of patients with type 1 diabetes and, overall, meta-analyzed age at diagnosis of type 1 diabetes in 5910 individuals who were not previously included in other studies of the same trait. In addition, we performed a conditional analysis adjusting for HLA-DR3/DR4 genotype categories, which are known to drastically increase the risk of type 1 diabetes and are associated with age at diagnosis. This is a robust approach that allows identification of independent signals and led to the discovery of an independent locus for age at diagnosis in the HLA region.All participating cohorts included individuals with type 1 diabetes; however, the inclusion criteria and the criteria to define type 1 diabetes varied among cohorts. This may have direct and indirect effects on the age at diagnosis in specific cohorts. As an example, the DCCT/EDIC cohort only included individuals 13–39 years of age with 1–15 years of type 1 duration at baseline and deficient C-peptide secretion. Although we are aware that this is not a random evaluation of people with type 1 diabetes, it should be noted that 75% of the participants had their type 1 diabetes diagnosed before the age of 22. Recently, we reported that residual insulin secretion, at a level higher than used in DCCT/EDIC, can be present in people with type 1 diabetes.23 EDC only included individuals with type 1 diabetes diagnosed before the age of 17 years. As in approximately half of modern cases of type 1 diabetes, diabetes is diagnosed after the age of 21 years,5–7 which reflects only a subset of individuals within the heterogeneous spectrum of type 1 diabetes. Moreover, measurement of autoantibodies against GAD and other constituents, as well as calculation of GRS, may better define the subtype of type 1 diabetes but do not strictly differentiate between type 1 diabetes and latent autoimmune diabetes in adults (LADA). Although there is an overlap between type 1 diabetes and LADA, with the latter often being considered a slow-developing subtype of type 1 diabetes, there have been reported genetic differences between both types.45 To have a harmonized dataset with similar methodology, genetic data from all cohorts were imputed with HRC V.1.1. As a result of legal issues, we could not use TOPMed imputation for the European cohorts, although this imputation method is available and may result in more and better-quality imputed SNPs.46In conclusion, we confirmed multiple variants for type 1 diabetes, and previously identified SNPs for age at diagnosis were also associated with age at diagnosis of type 1 diabetes in our study. Importantly, we identified rs76730244 C>T in the HLA region, associated with younger age at diagnosis of type 1 diabetes, which was independent of the HLA-DR3/DR4 genotype categories. Since type 1 diabetes is a heterogeneous disease with a wide heterogeneity in age at diagnosis and unknown “triggers,” more research, including large cohorts of participants with type 1 diabetes focusing specifically on the different endotypes, may unravel novel mechanistic insights and help improve type 1 diabetes treatment strategies.