OBJECTIVE

More than 10% of patients with type 1 diabetes (T1D) do not have high-risk HLA-DR3 or -DR4 haplotypes with distinct clinical features, such as later onset and reduced insulin dependence. We aimed to identify genetic drivers of T1D in the absence of DR3/DR4 and improve prediction of T1D risk in these individuals.

RESEARCH DESIGN AND METHODS

We performed T1D association and fine-mapping analyses in 12,316 non-DR3/DR4 samples. Next, we performed heterogeneity tests to examine differences in T1D risk variants in individuals without versus those with DR3/DR4 haplotypes. We further assessed genome-wide differences in gene regulatory element and biological pathway enrichments between the non-DR3/DR4 and DR3/DR4 cohorts. Finally, we developed a genetic risk score (GRS) to predict T1D in individuals without DR3/DR4 and compared with an existing T1D GRS.

RESULTS

A total of 18 T1D risk variants in non-DR3/DR4 samples were identified. Risk variants at the MHC and multiple other loci genome wide had heterogeneity in effects on T1D dependent on DR3/DR4 status, and non-DR3/DR4 T1D had evidence for a greater polygenic burden. T1D-associated variants in non-DR3/DR4 were more enriched for regulatory elements and pathways involved in antigen presentation, innate immunity, and β-cells and depleted in T cells compared with DR3/DR4. A non-DR3/DR4 GRS outperformed an existing risk score GRS2 in discriminating non-DR3/DR4 T1D from no diabetes (area under the curve 0.867; P = 7.48 × 10−32) and type 2 diabetes (0.907; P = 4.94 × 10−44).

CONCLUSIONS

In total, we identified heterogeneity in T1D genetic risk dependent on high-risk HLA-DR3/DR4 haplotype, which uncovers disease mechanisms and enables more accurate prediction of T1D across the HLA background.

Type 1 diabetes (T1D) is an autoimmune disease characterized by the destruction of insulin-producing β-cells and which has complex etiology (1). T1D is highly polygenic with >90 known risk loci (2), and the largest genetic risk factors map to the MHC locus (3), which encodes cell surface receptors that present antigenic peptides to T cells (1). Haplotypes of the class II HLA-DR and -DQ genes DRB1*0301-DQA1*0501-DQB1*0201 (DR3) and DRB1*04:01/02/04/05/08-DQA1*03:01-DQB1*03:02/04 (DR4) (3) confer substantial risk of T1D and are detected in 90% of individuals of European ancestry (1). The DR3/DR4 haplotypes are thought to increase T1D risk by altering peptide binding and presentation to T cells (1).

The remaining 10% of individuals of European ancestry who develop T1D without DR3 or DR4 haplotypes have lower rates of T1D autoantibodies (4) and later onset of disease with lower insulin dependence, which can be misdiagnosed as type 2 diabetes (T2D) or latent autoimmune diabetes in adults (5–7), leading to mismanagement and complications (5). The preventive therapy teplizumab is most effective in at-risk individuals positive for DR4 and negative for DR3, and it is less established if it effectively delays onset in individuals without DR3/DR4 (8). Furthermore, existing genetic risk scores (GRSs) for T1D heavily weigh DR3/DR4 haplotypes due to their large effect, and T1D in individuals without DR3/DR4 may be poorly predicted by these scores. The PTPN22 locus has been shown to interact with DR3/DR4 status (9–11), suggesting that genetic heterogeneity outside the MHC locus exists in individuals without DR3/DR4. There is increasing evidence that adult-onset T1D, which has a lower rate of DR3/DR4 (12), is more prevalent than previously estimated (13), further underscoring the need to understand T1D in the absence of DR3/DR4.

In this study, we performed genome-wide association analyses of T1D in individuals without DR3/DR4 to understand genetic drivers of T1D in the absence of DR3 or DR4 haplotypes. We then used association data to evaluate genetic risk, biological pathways and cell types, and risk prediction in non-DR3/DR4 T1D.

Subjects and Genotype Imputation

We compiled genotype data from 10,100 individuals with T1D and 19,623 control individuals of European ancestry from publicly available cohorts (Supplementary Table 1). T1D case cohorts were matched to control cohorts based on country of origin and genotype array where possible, as previously described (2). We applied the Haplotype Research Consortium imputation preparation program (version 4.2.9; https://www.well.ox.ac.uk/∼wrayner/tools/) and used PLINK version 1.90 (14) to perform quality control prior to imputation to remove variants with a minor allele frequency <1%, with missing genotypes >5%, in violation of Hardy-Weinberg equilibrium (HWE) (P < 1 × 10−5 in the control cohort and P < 1 × 10−10 in the case cohort), and with a difference in allele frequency >0.2 compared with the Haplotype Research Consortium version r1.1 reference panel (15), as well as variants with allele ambiguity (2,14). Related individuals were removed based on identity by descent >0.2. We imputed genotypes for all samples into the Trans-Omics for Precision Medicine (TOPMed) version 2 and Michigan Multiethnic HLA reference panels (16,17). For the HLA reference panel, we note that classical HLA alleles and amino acids have binary encodings as A for absent or T for present (18). In genome-wide imputation, we removed variants with imputation accuracy r2 <0.3. In HLA imputation, we removed variants with imputation accuracy r2 <0.5 and SD in control allele frequency >0.055 across cohorts. Variants that passed quality control in all cohorts were tested for association.

Association Testing and Meta-Analysis

We categorized individuals based on DR3/DR4 status using two-field HLA alleles imputed from the Type 1 Diabetes Genetics Consortium (T1DGC) reference panel with SNP2HLA, which includes HLA-DQB1*02:02 not present in the Michigan HLA panel (19). DR3 was defined by HLA-DRB1*03:01-DQB1*02:01 and DR4 by HLA-DRB1*04:01/02/04/05/08-DQB1*03:02/04/02:02 (6). We excluded 23 individuals identified as without DR3/DR4 via SNP2HLA alleles but who had DR3 or DR4 tag single nucleotide polymorphisms (SNPs) (20,21). The non-DR3/DR4 group consisted of individuals without a DR3 or DR4 haplotype (DRX/DRX), and the DR3/DR4 group consisted of individuals with at least one DR3 or DR4 haplotype (DR3/DR4, DR3/DR3, DR4/DR4, DR3/DRX, DR4/DRX). In the non-DR3/DR4 group, 100 case and 300 control samples were removed prior to association analyses, and these samples were randomly selected from cohorts proportionate to the total cohort size. In both non-DR3/DR4 and DR3/DR4 groups, we tested variants for T1D association using Firth bias-corrected logistic regression in EPACTS (https://github.com/statgen/EPACTS). We tested variants with minor allele frequency >1% for the association, including covariates for sex and the first four genotype principal components (PCs). The PCs were included to account for population structure and were derived using PLINK with parameters –indep 50 52. Summary statistics were combined across cohorts in a fixed-effects inverse variance–weighted meta-analysis using METAL (22). The genomic inflation λ was 1.069 for non-DR3/DR4 and 1.098 for DR3/DR4. We used linkage disequilibrium (LD) score regression (LDSC) (23) to test for heritability excluding the MHC, using population prevalence of 1% and sample prevalence of 51% for DR3/DR4 and 10% for non-DR3/DR4.

Fine-Mapping Independent Signals

We identified 1-Mb regions around the lead variants of genome-wide significant loci. We performed conditional analysis at each locus using stepwise analysis iteratively including the most significant variant in the regression model and re-performing the meta-analysis until no significant variants remained. In conditional analyses, we used a locus-wide significance threshold of P < 1 × 10−5. We then performed Bayesian fine-mapping to create credible sets of likely causal variants for each signal (24). From the summary statistics, effect size, and SE were used to calculate the approximate Bayes factor (BF) for each variant in r2 >0.1 with the lead variant. We calculated the probability of association for each variant by dividing the BF by the total sum of BFs. We then created 99% credible sets by including variants in descending order of probability of association until the cumulative posterior probability was at least 99% (Supplementary Table 12).

Assay for Transposase-Accessible Chromatin With Sequencing Peak Calling

Assay for transposase-accessible chromatin with sequencing data for 20 immune cell types in resting and stimulated conditions were obtained from the Gene Expression Omnibus database (accession no. GSE118189) (25) and processed to generate peak coordinates. Reads were aligned using STAR (26) to hg19, and duplicate reads, reads mapping to blacklisted regions from ENCODE, and reads with mapping quality <30 were filtered. Peak calling was performed using MACS2 (27) on binary alignment map files further filtered for read pairs with insert size no larger than 140 bp (macs2 callpeak –nomodel –nolambda –keep-dup all –call-summits -f BAMPE -g hs –q 0.01), combining individual samples for each cell type and treatment, for a total of 40 distinct peak sets. We then used bedtools multiinter to obtain a consensus set of peaks and featureCounts (28) to obtain the peak counts in each cell type.

Genome-Wide Association Study Enrichment Analysis

For each group, we performed partitioned heritability LDSC to estimate genome-wide enrichment in immune cell–accessible chromatin (23,25). We used the summary statistics for each group excluding the MHC locus and formatted it for LDSC using the munge_sumstats.py script. We generated binary annotations from each accessible chromatin bed file and computed cell-specific LD enrichment scores for each risk cohort using the version 2.2 1000G baseline model. Additionally, we performed gene set enrichment analysis in the Gene Ontology, Kyoto Encyclopedia of Genes and Genomes, and Reactome pathways with the summary statistics for each group using MAGMA (29) with default parameters. We corrected for multiple tests in each group using the false discovery rate (FDR) and considered FDR <0.10 as significant and uncorrected P < 0.05 as nominally significant.

Heterogeneity Tests

We tested for differences in T1D effect between non-DR3/DR4 and DR3/DR4 using merged PLINK files of Michigan HLA and TOPMed imputed variants for all 29,723 samples. We tested for heterogeneity in marginal effects on T1D using Breslow-Day (BD) tests with the -bd flag in PLINK (14). We performed BD tests for lead variants at the six non-MHC loci identified in non-DR3/DR4 samples, as well as lead variants at 83 additional known T1D risk loci (2). At MHC and other loci with multiple signals, we tested for heterogeneity in effects on T1D conditional or other known variants. We generated regression models with PLINK using the -glm interaction firth flag, including sex, the first four genotype PCs, DR3/DR4 status, and additional variants as covariates, and evaluated the interaction with DR3/DR4 status. For the 12 MHC signals identified in non-DR3/DR4 samples, we conditioned on preceding lead variants from stepwise regression. For the IFIH1 and PTPN2 loci, we conditioned on lead variants for all other known signals at the locus (14). For the HLA locus, we conditioned on HLA-DRB1*03:01, HLA-DQB1*02:01, HLA-DRB1*04:01, HLA-DRB1*04:02, HLA-DRB1*04:04, HLA-DRB1*04:05, HLA-DRB1*04:08, HLA-DQB1*03:02, and HLA-DQB1*03:04 to examine heterogeneity in 40 known two-field HLA risk alleles independent of these DR3 and DR4 alleles. To obtain the conditional effect of each variant within non-DR3/DR4 or DR3/DR4, we performed logistic regression separately for each group using EPACTS, including sex, the first four genotype PCs, and the same variants from the interaction tests above. We performed additional association analyses of non-DR3/DR4, including T1D with age of onset >17 or <17 years, T1D age of onset directly, and T1D interaction with sex with the glm function in PLINK, including the first four genotype PCs and (for age of onset analyses) sex as covariates. We corrected for multiple tests, considering FDR <0.10 as significant and uncorrected P < 0.05 as nominally significant.

Generation and Statistical Analysis of the GRSs

We calculated GRS1 using SNP2HLA-imputed genotypes to define DR3 and DR4 tag SNPs and TOPMed variants for the remaining 28 signals (20). We calculated GRS2 by using TOPMed variants where possible (60 variants), Michigan HLA for rs116522341 and rs1281934, and the Michigan HLA proxies DQB1*06:02, B*18:01, DPB1*03:01, rs1611547, and rs114170382, for rs17843689, rs371250843, rs559242105, rs144530872, and rs149663102, respectively. For GRS2, we excluded individuals with ambiguous (>2) HLA-DR/DQ calls, in line with the published methods (30). For each of the signals identified in the non-DR3/DR4 group, we generated an 18-varaint GRS as the sum of all signals weighted by the β (B) for each effect allele (X) as follows (Supplementary Table 12):

We additionally leveraged 82 non-MHC T1D risk loci from a T1D genome-wide association study (GWAS) (2) weighted by effects in the non-DR3/DR4 GWAS to generate a 100-variant combined GRS. We determined the ability of GRS2, the 18-variant GRS, and the 100-variant GRS to distinguish non-DR3/DR4 T1D from either all control samples (DR3/DR4 + non-DR3/DR4) or non-DR3/DR4 control samples. We tested the ability of each GRS to discriminate T1D and nondiabetes using the area under the curve (AUC) of the receiver operating characteristic (ROC) statistics in pROC. We calculated the difference between GRS AUCs using the DeLong test and calculated the diagnostic criteria using the Youden index. We then generated percentiles for how many individuals in the non-DR3/DR4 T1D or control groups fall at various GRS thresholds. We calculated sensitivity at each GRS as true positive divided by (true positive plus false negative) and specificity as true negative divided by (true negative plus false positive). We further tested the ability to differentiate T1D from nondisease in the 100-variant non-DR3/DR4 T1D GRS using the independent test group of 100 case and 300 control samples removed prior to association analyses. We also compared the ability of each GRS to differentiate T1D from T2D using 1,999 individuals with T2D from the Wellcome Trust Case Control Consortium (WTCCC) study (31). After imputing the T2D genotypes into the TOPMed and Michigan HLA reference panels, we calculated the scores for each GRS as described above and derived ROC statistics comparing T1D with T2D. We additionally validated the performance of the 100-variant non-DR3/DR4 GRS to distinguish T1D from T2D using the 100-T1D case test group and all 1,999 T2D samples.

We performed T1D association analyses in individuals without high-risk DR3 or DR4 haplotypes (Fig. 1). We obtained genotype data for 29,723 individuals with T1D and control individuals from five European ancestry cohorts (Supplementary Table 1) and imputed genotypes into 308 million variants in the TOPMed version 2 and 56,310 variants in Michigan HLA reference panels (32,33). We partitioned individuals into 1,292 with T1D and 11,424 control subjects without a DR3 or DR4 haplotype (non-DR3/DR4 group) and 8,808 individuals with T1D and 8,199 control subjects with at least one DR3 or DR4 haplotype (DR3/DR4 group) (Supplementary Table 1). We then tested for T1D association in the non-DR3/DR4 group, as well as in the DR3/DR4 group for comparison. Seven loci, including PTPN22, INS, IFIH1, PTPN2, IKZF4, OSTN, and MHC, had a genome-wide significant (P < 5 × 10−8) T1D association in non-DR3/DR4 (Fig. 1B and Supplementary Table 2). We identified 12 additional signals at locus-wide significance (P < 1 × 10−5) at the MHC locus through conditional association analysis (Research Design and Methods, Fig. 1D, and Supplementary Table 3). At each signal, we then generated a set of variants (termed credible set) most likely causal for T1D (Fig. 1C and E and Supplementary Table 4).

Figure 1

Genetic discovery in non-DR3/DR4 T1D. A: Overview of genetic discovery and risk prediction for T1D in individuals without DR3/DR4. B: Genome-wide T1D association (log10P values from meta-analysis of n = 12,316 samples) in individuals without DR3/DR4 haplotypes. Known T1D loci are blue, and novel loci are purple. All loci are labeled with the nearest gene. C: Number of candidate causal variants in fine-mapped non-MHC T1D signals. D: T1D association at the MHC locus (log10P values from marginal association in meta-analysis of n = 12,316 samples). Known signals are blue, and novel signals are purple. The location of class I and II HLA genes are shown on the bottom. E: Number of candidate causal variants in fine-mapped MHC signals. CS, credible set; DCCT, Diabetes Control and Complications Trial; GENIE, Genetics of Nephropathy, an International Effort; GoKinD, Genetics of Kidneys in Diabetes.

Figure 1

Genetic discovery in non-DR3/DR4 T1D. A: Overview of genetic discovery and risk prediction for T1D in individuals without DR3/DR4. B: Genome-wide T1D association (log10P values from meta-analysis of n = 12,316 samples) in individuals without DR3/DR4 haplotypes. Known T1D loci are blue, and novel loci are purple. All loci are labeled with the nearest gene. C: Number of candidate causal variants in fine-mapped non-MHC T1D signals. D: T1D association at the MHC locus (log10P values from marginal association in meta-analysis of n = 12,316 samples). Known signals are blue, and novel signals are purple. The location of class I and II HLA genes are shown on the bottom. E: Number of candidate causal variants in fine-mapped MHC signals. CS, credible set; DCCT, Diabetes Control and Complications Trial; GENIE, Genetics of Nephropathy, an International Effort; GoKinD, Genetics of Kidneys in Diabetes.

Close modal

We compared the T1D effects of risk variants at these seven loci between the non-DR3/DR4 and DR3/DR4 groups using a BD test and considered tests significant at FDR <0.10 (Research Design and Methods). Of the six non-MHC loci, five are known to affect T1D risk, while OSTN was previously unreported (Fig. 2A). At the novel OSTN locus, the lead variant had significant (FDR <0.10) heterogeneity in effect on T1D between the non-DR3/DR4 and DR3/DR4 groups (b = 0.4871 vs. 0.0281; FDR = 3.86 × 10−4) (Fig. 2B). We also observed significantly stronger effects in the non-DR3/DR4 group for lead variants at the IFIHI (b = −0.35 vs. −0.091; FDR = 2.33 × 10−4) and PTPN2 (b = −0.37 vs. −0.21, FDR = 0.0356) loci (Fig. 2C), both of which are implicated in β-cell function (34) (Supplementary Table 5). We further examined heterogeneity at 83 additional T1D loci (2), and although no further loci had significant heterogeneity, we observed more nominal evidence (P < 0.05) at nine loci. Among these, five, including PRR15L, RAD51B, PRF1, PRKD2, and 6q27, had larger effects in the non-DR3/DR4 group, including several implicated in immune cells and β-cells. By comparison, the four loci, including 14q32, IL2RA, IL2, and CD69, with smaller effects in the non-DR3/DR4 group largely affected T-cell function (Supplementary Fig. 1 and Supplementary Table 6).

Figure 2

Heterogeneity of genetic and biological mechanisms in non-DR3/DR4 and DR3/DR4 T1D. A: Number of known and novel signals with heterogeneity in effect on T1D depending on DR3/DR4 background. B: Locus plots of T1D association at the novel OSTN locus in DR3/DR4 (left) and non-DR3/DR4 (right). C: Non-MHC loci with genome-wide significant T1D association in non-DR3/DR4 association analyses with effect sizes and interaction analysis significance for difference in effect. D: Enrichment scores for DR3/DR4 and non-DR3/DR4 in stimulated and unstimulated immune cell regulatory elements. Dark green points are significant at FDR <0.10 in non-DR3/DR4 and DR3/DR4, while light blue points are significant by FDR in only DR3/DR4. E: Biological pathway enrichment in non-DR3/DR4 T1D. F: Biological pathway enrichment in DR3/DR4 T1D. G: T1D effect sizes of class I and II HLA alleles in each risk group after conditioning on nine DR3 and DR4 risk alleles. H: T1D case and control group frequencies of known HLA alleles in non-DR3/DR4 and DR3/DR4. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; Mem, memory; PDC, plasmacytoid dendritic cell; S, stimulated; TF, transcription factor; Tfh, T follicular helper cell; Th1, T helper 1 cell; Treg, regulatory T cell; U, unstimulated.

Figure 2

Heterogeneity of genetic and biological mechanisms in non-DR3/DR4 and DR3/DR4 T1D. A: Number of known and novel signals with heterogeneity in effect on T1D depending on DR3/DR4 background. B: Locus plots of T1D association at the novel OSTN locus in DR3/DR4 (left) and non-DR3/DR4 (right). C: Non-MHC loci with genome-wide significant T1D association in non-DR3/DR4 association analyses with effect sizes and interaction analysis significance for difference in effect. D: Enrichment scores for DR3/DR4 and non-DR3/DR4 in stimulated and unstimulated immune cell regulatory elements. Dark green points are significant at FDR <0.10 in non-DR3/DR4 and DR3/DR4, while light blue points are significant by FDR in only DR3/DR4. E: Biological pathway enrichment in non-DR3/DR4 T1D. F: Biological pathway enrichment in DR3/DR4 T1D. G: T1D effect sizes of class I and II HLA alleles in each risk group after conditioning on nine DR3 and DR4 risk alleles. H: T1D case and control group frequencies of known HLA alleles in non-DR3/DR4 and DR3/DR4. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; Mem, memory; PDC, plasmacytoid dendritic cell; S, stimulated; TF, transcription factor; Tfh, T follicular helper cell; Th1, T helper 1 cell; Treg, regulatory T cell; U, unstimulated.

Close modal

We next determined whether individuals with T1D in the absence of HLA-DR3/DR4 may carry a greater burden of polygenic risk. We determined whether known T1D risk variants broadly had stronger effects in the non-DR3/DR4 group. A higher proportion (60.2%) of variants at 88 known T1D loci had a stronger effect in non-DR3/DR4, although this was not significant (binomial P = 0.069). We also identified a small increase in heritability for T1D in non-DR3/DR4 compared with DR3/DR4 (h2 [SE] 0.2846 [0.0795] vs. 0.2693 [0.0428]). Finally, there was a significant increase in the ability to distinguish T1D from control in non-DR3/DR4 compared with DR3/DR4 (AUC 0.752 vs. 0.733; P = 0.0238) using a risk score derived from only non-MHC variants (2). The ability to distinguish T1D from control in non-DR3/DR4 was further improved when using effects derived from the non-DR3/DR4 T1D GWAS (AUC 0.768; P = 1.66 × 10−5). Overall, these results support a modest increase in polygenic contribution to T1D risk in the absence of high-risk DR3/DR4 haplotypes.

We next investigated whether variants within the MHC locus contribute to heterogeneity between non-DR3/DR4 and DR3/DR4 (Research Design and Methods). Of the 12 MHC signals identified in non-DR3/DR4 T1D, 8 exhibited significant heterogeneity (FDR <0.10) (Fig. 2A and Supplementary Table 7). Most signals were linked to known T1D alleles, except for two apparent novel signals at HLA-DRB1*09:01-DQA1*03:01-DQB1*03:03 and B*44:02 (Fig. 2A and Supplementary Table 8). Among a larger set of 40 known T1D MHC alleles, 22 two-field class I and class II alleles had significant heterogeneity (FDR <0.10) (Research Design and Methods, Fig. 2D and E, and Supplementary Table 9). In class I, HLA-B*39:06 had increased risk in non-DR3/DR4 (FDR = 0.027), HLA-A*01:01 was more protective in non-DR3/DR4 (FDR = 0.015), and HLA-B*18:01 had opposed effects in non-DR3/DR4 and DR3/DR4 (FDR = 6.31 × 10−6) (Fig. 2D and E). In class II, DRB1*15:01-DQB1*06:02 was more protective in DR3/DR4 (DRB1*15:01 FDR = 1.91 × 10−7, DQB1*6:02 FDR = 1.51 × 10−6) (Fig. 2D and E), DRB1*13:01-DQB1*06:03 was more protective in non-DR3/DR4 (FDR = 0.027) (Fig. 2A, D, and E), and DPB1*03:01 had increased risk in DR3/DR4 (FDR = 0.085) (Fig. 2D and E). Together this reveals numerous MHC alleles with heterogeneous effects on T1D in a non-DR3/DR4 background.

Individuals without DR3/DR4 haplotypes have typically later onset of T1D; thus, we examined whether heterogeneity in T1D risk between non-DR3/DR4 and DR3/DR4 was consistent across age of onset. As all T1D ages of onset were <40 years, we were unable to examine non-DR3/DR4 heterogeneity in older-onset T1D. We compared the effects of non-DR3/DR4 signals in each cohort, including WTCCC and T1DGC where all case individuals had an age of onset of <17 years (Supplementary Table 1). Among loci significant in non-DR3/DR4 that also exhibited heterogeneity, we observed no difference in T1D effects in WTCCC and T1DGC compared with the other cohorts with an age of onset up to 40 years (Cochran Q >0.05) (Supplementary Table 2 and Supplementary Fig. 2). Similarly, none of these loci showed an association with T1D stratified by age of onset of 17 years or T1D age of onset directly (all P > 0.05). Signals with heterogeneity in non-DR3/DR4 were largely distinct from those with an age-dependent association from a previous study (6), and HLA-B*39:06 had a larger effect in non-DR3/DR4 yet was associated with younger onset (Supplementary Fig. 1). As sex also impacts T1D, we determined whether the T1D association in non-DR3/DR4 interacted with sex and found no significant evidence (all P > 0.05). This supports that heterogeneity in non-DR3/DR4 is distinct from T1D age-of-onset– and sex-dependent effects.

We next determined whether there were broader differences in T1D association patterns between non-DR3/DR4 and DR3/DR4. We first tested whether T1D risk variants showed different enrichment for gene regulatory elements in specific immune cell types (23,25). In non-DR3/DR4, there was significant enrichment (FDR <0.10) of T1D variants in regulatory elements for B cells, mature natural killer (NK) cells, and unstimulated regulatory T cells (Fig. 2F and Supplementary Table 10). By comparison, in DR3/DR4, T1D variants were most enriched (FDR <0.10) in regulatory elements for T-cell populations (Supplementary Table 10). We next determined whether T1D risk variants were enriched for genes involved in different biological pathways (Fig. 2G and Hand Supplementary Table 11). T1D in DR3/DR4 had significant enrichment (FDR <0.10) for T-cell–related pathways, including IL-2 signaling, and T-cell differentiation, activation, and proliferation (Fig. 2H). In non-DR3/DR4, the strongest enrichments, although not significant after FDR correction, were for regulation of lipid storage, macrophage-stimulating factor, stress-induced apoptotic signaling, and complement pathways (Fig. 2G). Collectively, this suggests a greater antigen-presenting, innate immune, and β-cell contribution and lower T-cell contribution to T1D in non-DR3/DR4.

Finally, we evaluated the ability of GRSs to predict T1D in individuals without DR3/DR4. More than 35% of individuals without DR3/DR4 T1D were below the 5th percentile for T1D in the published GRS2 (30), and only 7.5% were above the 50th percentile used to classify T1D with high specificity. In line with this, GRS2 had less ability to distinguish individuals without DR3/DR4 T1D from all control subjects (AUC 0.764) (Fig. 3A and E) compared with all individuals with T1D (AUC 0.927) (30). We generated a non-DR3/DR4–specific T1D GRS using the 18 risk signals in our non-DR3/DR4 GWAS (Research Design and Methods and Supplementary Table 12). This GRS improved discrimination of individuals without DR3/DR4 T1D from all control subjects compared with GRS2 (AUC 0.835; P = 6.30 × 10−15) (Fig. 3A, E, and F). We next generated a larger 100-variant non-DR3/DR4 T1D GRS that included 82 additional T1D risk loci (Research Design and Methods and Supplementary Table 12), which further discriminated T1D from control (AUC 0.867; P = 7.48 × 10−32) (Fig. 3B, E, and G). When subsets of individuals with T1D and control subjects were compared with individuals without DR3/DR4, the 100-variant non-DR3/DR4 GRS improved discrimination of T1D over both the 18-variant GRS (AUC 0.894; P = 5.70 × 10−26) (Fig. 3C, F, and G) and GRS2 (AUC 0.878; P = 0.026) (Fig. 3D, E, and G). We further confirmed the ability of the 100-variant non-DR3/DR4 GRS to predict T1D using a test set of 100 T1D and 300 control non-DR3/DR4 samples excluded from association tests (AUC 0.887) (Supplementary Fig. 3A and B).

Figure 3

Genetic risk prediction in individuals with T1D and no DR3/DR4. ROC curves showing the ability of GRSs to differentiate T1D from nondiabetes in non-DR3/DR4 cohort and corresponding violin plots. The AUCs are shown for each GRS, and the P values comparing predictive ability of the GRSs are calculated using the DeLong test. A: GRS2 compared with the 18-variant non-DR3/DR4 GRS in the non-DR3/DR4 T1D group vs. all (non-DR3/DR4 and DR3/DR4) control subjects. B: GRS2 compared with the 100-variant non-DR3/DR4 T1D GRS in non-DR3/DR4 T1D group vs. all control subjects. C: The 18-variant non-DR3/DR4 T1D GRS compared with the 100-variant non-DR3/DR4 T1D GRS in non-DR3/DR4 T1D and control groups. D: GRS2 subset of the non-DR3/DR4 T1D and control groups compared with the 100-variant non-DR3/DR4 T1D GRS in the non-DR3/DR4 T1D and control groups. EG: Violin plots for scores in all control subjects, control subjects without DR3/DR4, and individuals with T1D are depicted for GRS2 (E), 18-variant non-DR3/DR4 T1D GRS (F), and 100-variant non-DR3/DR4 T1D GRS (G). The number of samples differs between GRSs due to ambiguous HLA-DR/DQ alleles leading to several sample exclusions in GRS2. var, variant.

Figure 3

Genetic risk prediction in individuals with T1D and no DR3/DR4. ROC curves showing the ability of GRSs to differentiate T1D from nondiabetes in non-DR3/DR4 cohort and corresponding violin plots. The AUCs are shown for each GRS, and the P values comparing predictive ability of the GRSs are calculated using the DeLong test. A: GRS2 compared with the 18-variant non-DR3/DR4 GRS in the non-DR3/DR4 T1D group vs. all (non-DR3/DR4 and DR3/DR4) control subjects. B: GRS2 compared with the 100-variant non-DR3/DR4 T1D GRS in non-DR3/DR4 T1D group vs. all control subjects. C: The 18-variant non-DR3/DR4 T1D GRS compared with the 100-variant non-DR3/DR4 T1D GRS in non-DR3/DR4 T1D and control groups. D: GRS2 subset of the non-DR3/DR4 T1D and control groups compared with the 100-variant non-DR3/DR4 T1D GRS in the non-DR3/DR4 T1D and control groups. EG: Violin plots for scores in all control subjects, control subjects without DR3/DR4, and individuals with T1D are depicted for GRS2 (E), 18-variant non-DR3/DR4 T1D GRS (F), and 100-variant non-DR3/DR4 T1D GRS (G). The number of samples differs between GRSs due to ambiguous HLA-DR/DQ alleles leading to several sample exclusions in GRS2. var, variant.

Close modal

Since individuals with T1D and no DR3/DR4 tend to have a later onset and a lower dependence on insulin therapy, leading to a potential misdiagnosis of T2D (5,6,12), we next evaluated the ability of GRS to differentiate T1D from T2D (Supplementary Fig. 4). In GRS2 there was limited ability to distinguish non-DR3/DR4 T1D from all T2D (AUC 0.759) (Supplementary Fig. 4A and E). In contrast, our non-DR3/DR4 GRS strongly discriminated non-DR3/DR4 T1D from T2D compared with GRS2 (18-variant AUC 0.890 [P = 1.51 × 10−32]; 100-variant AUC 0.907 [P = 4.94 × 10−44]) (Supplementary Fig. 4A, B, and EG). When subsets of individuals with T1D and T2D were compared with individuals without DR3/DR4, the 100-variant non-DR3/DR4 GRS again improved on both the 18-variant GRS (AUC 0.920 and 0.905, respectively; P = 0.0775) (Supplementary Fig. 4C, F, and G) and GRS2 (AUC 0.876; P = 1.89 × 10−6) (Supplementary Fig. 4D, E, and G). The predictions were consistent when using a test set of 100 T1D samples (AUC 0.906) (Supplementary Fig. 3C and D). The ability to distinguish non-DR3/DR4 T1D from T2D appeared largely driven by the improved estimation of effects at the MHC locus (T1D vs. T2D HLA AUC 0.885), as the extensive HLA combinations in GRS2 still do not enable differentiation of non-DR3/DR4 T1D from T2D (T1D vs. T2D GRS2 HLA AUC 0.698).

We finally determined the diagnostic value of a GRS for T1D in a non-DR3/DR4 background. Individuals with T1D and no DR3/DR4 on the published GRS2 scale do not meet minimum requirements for diagnostic feasibility (maximum Youden index <0.5). When GRS2 is compared with only non-DR3/DR4 and rescaled (Table 1), the 50th percentile of T1D had a specificity of 95.4%, and the score had improved diagnostic ability (maximum Youden index = 0.609). The 18-variant non-DR3/DR4 GRS at the 50th percentile for T1D had a specificity of 95.2% and a maximum Youden index of 0.607. Furthermore, the 100-variant non-DR3/DR4 GRS had improved specificity (96.9%) at the 50th percentile for T1D and a higher maximum Youden index of 0.644. In addition, while GRS2 had a limited ability to distinguish T1D from T2D (maximum Youden index = 0.414), the 100-variant non-DR3/DR4 GRS had diagnostic value (maximum Youden index >0.5) in differentiating individuals with T1D and no DR3/DR4 T1D from individuals with T2D as well as all control subjects. Overall, our results provide a GRS that can likely be adopted in a clinical setting to distinguish non-DR3/DR4 T1D from both nondiabetes and T2D.

Table 1

Sensitivity, specificity, diagnostic value, and scales for each T1D GRS

T1D centileNondisease centileSensitivity (%)Specificity (%)Youden index
Non-DR3/DR4–only GRS2      
 8.86 47.6 95.00 47.63 0.4263 
 10.33 15 73.7 85.00 73.72 0.5872 
 11.20 25 85.3 75.00 85.35 0.6035 
 11.77 35 91.3 65.00 91.27 0.5627 
 12.40 50 95.4 50.00 95.41 0.4541 
 13.39 75 98.9 25.00 98.85 0.2385 
 14.85 95 99.9 5.000 99.95 0.0978 
 11.23** 25.2 86.0 74.8 86.0 0.6086 
18-variant non-DR3/DR4 GRS      
 9.067 42.0 94.97 41.95 0.3691 
 10.59 15 72.5 84.98 72.60 0.5758 
 11.36 25 84.5 75.00 84.53 0.5953 
 11.85 35 90.3 65.02 90.26 0.5527 
 12.50 50 95.2 50.00 95.20 0.4519 
 13.55 75 98.8 25.00 98.78 0.2377 
 15.08 95 99.9 5.034 99.87 0.04908 
 11.05** 19.4 80.1 80.6 80.1 0.6068 
100-variant non-DR3/DR4 GRS      
 21.25 46.9 94.97 46.90 0.4187 
 23.06 15 78.7 84.98 78.68 0.6366 
 23.84 25 88.4 75.00 88.41 0.6341 
 24.43 35 93.6 65.02 93.56 0.5858 
 25.11 50 96.9 50.00 96.90 0.4689 
 26.28 75 99.3 25.00 99.33 0.2433 
 27.82 95 99.9 5.034 99.96 0.0450 
 23.21** 16.4 80.8 83.6 80.8 0.6439 
T1D centileNondisease centileSensitivity (%)Specificity (%)Youden index
Non-DR3/DR4–only GRS2      
 8.86 47.6 95.00 47.63 0.4263 
 10.33 15 73.7 85.00 73.72 0.5872 
 11.20 25 85.3 75.00 85.35 0.6035 
 11.77 35 91.3 65.00 91.27 0.5627 
 12.40 50 95.4 50.00 95.41 0.4541 
 13.39 75 98.9 25.00 98.85 0.2385 
 14.85 95 99.9 5.000 99.95 0.0978 
 11.23** 25.2 86.0 74.8 86.0 0.6086 
18-variant non-DR3/DR4 GRS      
 9.067 42.0 94.97 41.95 0.3691 
 10.59 15 72.5 84.98 72.60 0.5758 
 11.36 25 84.5 75.00 84.53 0.5953 
 11.85 35 90.3 65.02 90.26 0.5527 
 12.50 50 95.2 50.00 95.20 0.4519 
 13.55 75 98.8 25.00 98.78 0.2377 
 15.08 95 99.9 5.034 99.87 0.04908 
 11.05** 19.4 80.1 80.6 80.1 0.6068 
100-variant non-DR3/DR4 GRS      
 21.25 46.9 94.97 46.90 0.4187 
 23.06 15 78.7 84.98 78.68 0.6366 
 23.84 25 88.4 75.00 88.41 0.6341 
 24.43 35 93.6 65.02 93.56 0.5858 
 25.11 50 96.9 50.00 96.90 0.4689 
 26.28 75 99.3 25.00 99.33 0.2433 
 27.82 95 99.9 5.034 99.96 0.0450 
 23.21** 16.4 80.8 83.6 80.8 0.6439 

** Score with the maximum Youden index.

In the absence of high-risk DR3/DR4 haplotypes, we observed heterogeneity in T1D risk for class I and class II MHC alleles as well as other loci genome wide. Genes at loci with larger effects in non-DR3/DR4, such as PTPN2, IFIH1, PRR15L, and RAD51B, are implicated in inflammatory signaling (34,35) and survival (36) in immune cells and β-cells. We also identified a novel T1D locus in non-DR3/DR4 near OSTN, which has been linked to diabetic cardiomyopathy (37). Conversely, loci such as IL2RA, IL2, and CD69 with a reduced effect in non-DR3/DR4 impact T-cell function and activation (38). Genomic annotations and molecular pathways related to B cells, NK cells, and β-cells were also more enriched in non-DR3/DR4 T1D, and T cells less enriched, compared with DR3/DR4. These results suggest that mechanisms of T1D in a non-DR3/DR4 background may depend more on inflammation and β-cell dysfunction than T-cell activation (5–7). Given modest evidence for a higher polygenic burden in non-DR3/DR4 T1D, these individuals may require additional polygenic risk to develop T1D.

Based on previous GRSs, many individuals with T1D and no DR3/DR4 would not have been predicted to develop T1D. The GRS reported here enables accurate discrimination of T1D from nondiabetes and T2D in individuals without DR3/DR4. To improve accessibility, variants in the non-DR3/DR4 GRS are all derived from publicly available imputation panels. While the subset of individuals that develop T1D without DR3/DR4 exhibit later onset and lower insulin dependence, they have highly similar genetic risk profiles to DR3/DR4 T1D and not T2D. Accurate prediction of T1D in non-DR3/DR4 will help avoid misdiagnosis of T2D and prevent ketoacidosis and future complications. While the non-DR3/DR4 T1D background has generally later onset, it can occur at any age, and our results suggest that the effects of variants on T1D risk in a non-DR3/DR4 background are consistent across age of onset. Furthermore, our GRS may have value in discriminating T1D in populations in which DR3/DR4 is uncommon.

The ability to distinguish at-risk individuals is also critical for determining eligibility for clinical trials and therapies. The preventive therapy teplizumab modulates T-cell activity and is most efficacious in individuals with HLA-DR4 (8), and several clinical trials aiming to preserve β-cell function at onset preferentially recruit individuals with DR3/DR4 (39,40). Given evidence for differences in disease mechanisms, particularly a less prominent contribution from T cells, alternate therapies may be needed to prevent T1D in individuals without DR3/DR4. In addition, as autoantibodies are seen at lower rates in non-DR3/DR4 T1D, additional biomarkers are needed for this group (4).

There are several notable limitations of our study. First, as the non-DR3/DR4 GRS was tested in samples from the same cohorts, which are not fully independent, the accuracy of the GRS requires validation in other cohorts. Second, all individuals with T1D in this study were younger than 40 years, and as many T1D diagnoses occur after age 40, similar studies are needed to establish genetic heterogeneity in older-onset T1D. In addition, several cohorts consisted of individuals with T1D selected for diabetes complications, which may result in biases in the case population. Finally, larger and more ancestrally diverse sample sizes in association analyses will help establish the full extent of heterogeneity in T1D genetic risk due to HLA background, including within previously defined endotypes.

Our findings are in line with a growing body of literature (6,41) supporting T1D as a heterogeneous disease consisting of subtypes with distinct pathophysiological features. More broadly, stratifying by high-risk genetic background may be an effective strategy to discover genetic and mechanistic heterogeneity in other complex diseases.

This article contains supplementary material online at https://doi.org/10.2337/figshare.27325125.

J.C. is currently affiliated with Internal Medicine Research Unit, Pfizer Research and Development, Pfizer, Cambridge, MA.

Acknowledgments. The Diabetes Control and Complications Trial (DCCT) and its follow-up, the Epidemiology of Diabetes Interventions and Complications (EDIC) study, were conducted by the DCCT/EDIC Research Group and supported by National Institutes of Health (NIH) grants and contracts and by the General Clinical Research Center Program, National Center for Research Resources. The data from the DCCT/EDIC study were supplied by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Central Repositories.

The Genetics of Nephropathy, an International Effort (GENIE), study was conducted by the GENIE Investigators and supported by NIDDK. The data from the GENIE study reported here were supplied by the GENIE Investigators from the Broad Institute of MIT and Harvard, Queens University Belfast, and the University of Dublin.

The Genetics of Kidneys in Diabetes (GoKinD) study was conducted by the GoKinD Investigators and supported by the JDRF, the Centers for Disease Control and Prevention, and the Special Statutory Funding Program for Type 1 Diabetes Research administered by NIDDK. The data (and samples) from the GoKinD study were supplied by the NIDDK Central Repositories. This manuscript was not prepared in collaboration with investigators of the GoKinD study and does not necessarily reflect the opinions or views of the GoKinD study, the NIDDK Central Repositories, or the NIDDK.

This research uses resources provided by the Type 1 Diabetes Genetics Consortium (T1DGC), a collaborative clinical study sponsored by the NIDDK, National Institute of Allergy and Infectious Diseases, National Human Genome Research Institute (NHGRI), National Institute of Child Health and Human Development, and JDRF and supported by grant U01 DK062418. The U.K. case series collection was additionally funded by JDRF and the Wellcome Trust and the National Institute for Health Research Cambridge Biomedical Centre at the Cambridge Institute for Medical Research, U.K., which is in receipt of Wellcome Trust Strategic Award 079895. The data from the T1DGC study were supplied by database of Genotypes and Phenotypes (dbGaP). This manuscript was not prepared in collaboration with investigators of the T1DGC study and does not necessarily reflect the opinions or views of the T1DGC study or the study sponsors.

This study makes use of data generated by the Wellcome Trust Case Control Consortium (WTCCC). A full list of the investigators who contributed to the generation of the data is available from https://www.wtccc.org.uk. Funding for the project was provided by the Wellcome Trust under award 076113.

Funding support for the Genome-Wide Association of Schizophrenia Study was provided by National Institute of Mental Health grants R01 MH67257, R01 MH59588, R01 MH59571, R01 MH59565, R01 MH59587, R01 MH60870, R01 MH59566, R01 MH59586, R01 MH61675, R01 MH60879, R01 MH81800, U01 MH46276, U01 MH46289 U01 MH46318, U01 MH79469, and U01 MH79470, and the genotyping of samples was provided through the Genetic Association Information Network. The data sets used for the analyses described in this manuscript were obtained from dbGaP (https://www.ncbi.nlm.nih.gov/gap) through dbGaP accession no. phs000021.v3.p2. Samples and associated phenotype data for the Genome-Wide Association of Schizophrenia Study were provided by the Molecular Genetics of Schizophrenia Collaboration (principal investigator Pablo V. Gejman, Evanston Northwestern Healthcare and Northwestern University).

Support for the collection of the data for Philadelphia Neurodevelopment Cohort was provided by grant RC2MH089983 awarded to Raquel Gur and RC2MH089924 awarded to Hakon Hakonarson. Subjects were recruited and genotyped through the Center for Applied Genomics (CAG) at The Children’s Hospital in Philadelphia (CHOP). Phenotypic data collection occurred at the CAG/CHOP and at the Brain Behavior Laboratory, University of Pennsylvania.

Support for the Electronic Medical Records and Genomics (eMERGE) Network is as follows. The Group Health Cooperative/University of Washington receives funding support for the Alzheimer’s Disease Patient Registry (ADPR), and the Adult Changes in Thought (ACT) study was provided by National Institute on Aging grant U01 AG006781 (Eric B. Larson, principal investigator). A gift from the 3M Corporation was used to expand the ACT cohort. DNA aliquots sufficient for genome-wide association studies from ADPR participants with probable Alzheimer disease who had been enrolled in Genetic Differences in Alzheimer’s Cases and Controls (grant R01AG007584 to Walter Kukull, principal investigator) and obtained under that grant were made available to the eMERGE Network without charge. Funding support for genotyping, which was performed at Johns Hopkins University, was provided by NIH grant U01HG004438. Genome-wide association analyses were supported through NHGRI cooperative agreement U01HG004610 (Eric B. Larson, principal investigator). The Mayo Clinic provided samples and associated genotype and phenotype data used in this study. Funding support for the Mayo Clinic was provided through NHGRI cooperative agreement U0IHG004599 and by National Heart, Lung, and Blood Institute grant HL75794. Funding support for genotyping, which was performed at the Broad Institute, was provided by NIH grant U01HG004424. The Marshfield Clinic Research Foundation received funding support for the Personalized Medicine Research Project (PMRP) through NHGRI cooperative agreement U01HG004608, with additional funding from the National Institute for General Medical Sciences. The samples used for PMRP analyses were obtained with funding from Marshfield Clinic, Health Resources Service Administration Office of Rural Health Policy grant D1A RH00025, and Wisconsin Department of Commerce Technology Development Fund contract TDF FYO10718. Funding support for genotyping, which was performed at Johns Hopkins University, was provided by NIH grant U01HG004438. Northwestern University provided samples and data used in this study through the NUgene Project (https://www.cgm.northwestern.edu/cores/nugene/index.html). Funding support for the NUgene Project was provided by the Northwestern University’s Center for Genetic Medicine, Northwestern University, and Northwestern Memorial Hospital. Assistance with phenotype harmonization was provided by the eMERGE Coordinating Center (grant U01HG04603). This study was funded through the NIH NHGRI eMERGE Network (grant U01HG004609). Funding support for genotyping, which was performed at the Broad Institute, was provided by NIH grant U01HG004424.

Funding. This work was supported by NIH grants DK122607 and DK120429 to K.J.G.

Duality of Interest. K.J.G. has done consulting for Genentech, received honoraria from Pfizer, and is a shareholder of Neurocrine Biosciences. J.C. is an employee and shareholder of Pfizer. P.B. was previously an employee and shareholder of Shoreline Biosciences. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. C.M. performed genetic association, genomic enrichment, and genetic risk score analyses. C.M. and K.J.G designed the study and wrote the manuscript. J.C. and P.B. contributed to the genetic association analyses. R.E. contributed to the genomic enrichment analyses. A.M.L. and R.A.O. contributed to genetic risk score analyses. C.M. and K.J.G. are the guarantors of this work and, as such, had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 83rd Scientific Sessions of the American Diabetes Association, San Diego, CA, 23–26 June 2023.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Steven E. Kahn and Emily K. Sims.

1.
Redondo
MJ
,
Steck
AK
,
Pugliese
A
.
Genetics of type 1 diabetes
.
Pediatr Diabetes
2018
;
19
:
346
353
2.
Chiou
J
,
Geusz
RJ
,
Okino
M-L
, et al
.
Interpreting type 1 diabetes risk with genetics and single-cell epigenomics
.
Nature
2021
;
594
:
398
402
3.
Noble
JA
,
Valdes
AM
.
Genetics of the HLA region in the prediction of type 1 diabetes
.
Curr Diab Rep
2011
;
11
:
533
542
4.
Singh
S
,
Usha
,
Singh
G
,
Agrawal
NK
,
Singh
RG
,
Kumar
SB
.
Prevalence of autoantibodies and HLA DR, DQ in type 1 diabetes mellitus
.
J Clin Diagn Res
2016
;
10
:
EC09
EC13
5.
Leslie
RD
,
Evans-Molina
C
,
Freund-Brown
J
, et al
.
Adult-onset type 1 diabetes: current understanding and challenges
.
Diabetes Care
2021
;
44
:
2449
2456
6.
Inshaw
JRJ
,
Cutler
AJ
,
Crouch
DJM
,
Wicker
LS
,
Todd
JA
.
Genetic variants predisposing most strongly to type 1 diabetes diagnosed under age 7 years lie near candidate genes that function in the immune system and in pancreatic β-cells
.
Diabetes Care
2020
;
43
:
169
177
7.
Barker
A
,
Lauria
A
,
Schloot
N
, et al
.
Age-dependent decline of β-cell function in type 1 diabetes after diagnosis: a multi-centre longitudinal study
.
Diabetes Obes Metab
2014
;
16
:
262
267
8.
Herold
KC
,
Bundy
BN
,
Long
SA
, et al.;
Type 1 Diabetes TrialNet Study Group
.
An anti-CD3 antibody, teplizumab, in relatives at risk for type 1 diabetes
.
N Engl J Med
2019
;
381
:
603
613
9.
Hermann
R
,
Lipponen
K
,
Kiviniemi
M
, et al
.
Lymphoid tyrosine phosphatase (LYP/PTPN22) Arg620Trp variant regulates insulin autoimmunity and progression to type 1 diabetes
.
Diabetologia
2006
;
49
:
1198
1208
10.
Smyth
DJ
,
Cooper
JD
,
Howson
JMM
, et al
.
PTPN22 Trp620 explains the association of chromosome 1p13 with type 1 diabetes and shows a statistical interaction with HLA class II genotypes
.
Diabetes
2008
;
57
:
1730
1737
11.
Bjørnvold
M
,
Undlien
DE
,
Joner
G
, et al
.
Joint effects of HLA, INS, PTPN22 and CTLA4 genes on the risk of type 1 diabetes
.
Diabetologia
2008
;
51
:
589
596
12.
Howson
JMM
,
Rosinger
S
,
Smyth
DJ
,
Boehm
BO
,
Todd
JA
,
ADBW-END Study Group
.
Genetic analysis of adult-onset autoimmune diabetes
.
Diabetes
2011
;
60
:
2645
2653
13.
Harding
JL
,
Wander
PL
,
Zhang
X
, et al
.
The incidence of adult-onset type 1 diabetes: a systematic review from 32 countries and regions
.
Diabetes Care
2022
;
45
:
994
1006
14.
Purcell
S
,
Neale
B
,
Todd-Brown
K
, et al
.
PLINK: a tool set for whole-genome association and population-based linkage analyses
.
Am J Hum Genet
2007
;
81
:
559
575
15.
McCarthy
S
,
Das
S
,
Kretzschmar
W
, et al.;
Haplotype Reference Consortium
.
A reference panel of 64,976 haplotypes for genotype imputation
.
Nat Genet
2016
;
48
:
1279
1283
16.
Das
S
,
Forer
L
,
Schönherr
S
, et al
.
Next-generation genotype imputation service and methods
.
Nat Genet
2016
;
48
:
1284
1287
17.
Fuchsberger
C
,
Abecasis
GR
,
Hinds
DA
.
minimac2: Faster genotype imputation
.
Bioinformatics
2015
;
31
:
782
784
18.
Michigan Imputation Server. Accessed 15 December 2023
. Available from https://imputationserver.readthedocs.io/en/latest/pipeline/
19.
Jia
X
,
Han
B
,
Onengut-Gumuscu
S
, et al
.
Imputing amino acid polymorphisms in human leukocyte antigens
.
PLoS One
2013
;
8
:
e64683
20.
Oram
RA
,
Patel
K
,
Hill
A
, et al
.
A type 1 diabetes genetic risk score can aid discrimination between type 1 and type 2 diabetes in young adults
.
Diabetes Care
2016
;
39
:
337
344
21.
Nguyen
C
,
Varney
MD
,
Harrison
LC
,
Morahan
G
.
Definition of high-risk type 1 diabetes HLA-DR and HLA-DQ types using only three single nucleotide polymorphisms
.
Diabetes
2013
;
62
:
2135
2140
22.
Willer
CJ
,
Li
Y
,
Abecasis
GR
.
METAL: fast and efficient meta-analysis of genomewide association scans
.
Bioinformatics
2010
;
26
:
2190
2191
23.
Finucane
HK
,
Bulik-Sullivan
B
,
Gusev
A
, et al.;
RACI Consortium
.
Partitioning heritability by functional annotation using genome-wide association summary statistics
.
Nat Genet
2015
;
47
:
1228
1235
24.
Wakefield
J
.
Bayes factors for genome-wide association studies: comparison with P-values
.
Genet Epidemiol
2009
;
33
:
79
86
25.
Calderon
D
,
Nguyen
MLT
,
Mezger
A
, et al
.
Landscape of stimulation-responsive chromatin across diverse human immune cells
.
Nat Genet
2019
;
51
:
1494
1505
26.
Dobin
A
,
Davis
CA
,
Schlesinger
F
, et al
.
STAR: ultrafast universal RNA-seq aligner
.
Bioinformatics
2013
;
29
:
15
21
27.
Zhang
Y
,
Liu
T
,
Meyer
CA
, et al
.
Model-based analysis of ChIP-Seq (MACS)
.
Genome Biol
2008
;
9
:
R137
28.
Liao
Y
,
Smyth
GK
,
Shi
W
.
featureCounts: an efficient general purpose program for assigning sequence reads to genomic features
.
Bioinformatics
2014
;
30
:
923
930
29.
de Leeuw
CA
,
Mooij
JM
,
Heskes
T
,
Posthuma
D
.
MAGMA: generalized gene-set analysis of GWAS Data
.
PLoS Comput Biol
2015
;
11
:
e1004219
30.
Sharp
SA
,
Rich
SS
,
Wood
AR
, et al
.
Development and standardization of an improved type 1 diabetes genetic risk score for use in newborn screening and incident diagnosis
.
Diabetes Care
2019
;
42
:
200
207
31.
Wellcome Trust Case Control Consortium
.
Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls
.
Nature
2007
;
447
:
661
678
32.
Taliun
D
,
Harris
DN
,
Kessler
MD
, et al.;
NHLBI Trans-Omics for Precision Medicine (TOPMed) Consortium
.
Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program
.
Nature
2021
;
590
:
290
299
33.
Luo
Y
,
Kanai
M
,
Choi
W
, et al.;
NHLBI Trans-Omics for Precision Medicine (TOPMed) Consortium
.
A high-resolution HLA reference panel capturing global population diversity enables multi-ancestry fine-mapping in HIV host response
.
Nat Genet
2021
;
53
:
1504
1516
34.
Colli
ML
,
Moore
F
,
Gurzov
EN
,
Ortis
F
,
Eizirik
DL
.
MDA5 and PTPN2, two candidate genes for type 1 diabetes, modify pancreatic β-cell responses to the viral by-product double-stranded RNA
.
Hum Mol Genet
2010
;
19
:
135
146
35.
Trivedi
P
,
Graham
KL
,
Krishnamurthy
B
, et al
.
Perforin facilitates beta cell killing and regulates autoreactive CD8+ T-cell responses to antigen in mouse models of type 1 diabetes
.
Immunol Cell Biol
2016
;
94
:
334
341
36.
Benaglio
P
,
Zhu
H
,
Okino
M-L
, et al
.
Type 1 diabetes risk genes mediate pancreatic beta cell survival in response to proinflammatory cytokines
.
Cell Genom
2022
;
2
:
100214
37.
Zhang
X
,
Hu
C
,
Yuan
X-P
, et al
.
Osteocrin, a novel myokine, prevents diabetic cardiomyopathy via restoring proteasomal activity
.
Cell Death Dis
2021
;
12
:
624
612
38.
Garg
G
,
Tyler
JR
,
Yang
JHM
, et al
.
Type 1 diabetes-associated IL2RA variation lowers IL-2 signaling and contributes to diminished CD4+CD25+ regulatory T cell function
.
J Immunol
2012
;
188
:
4644
4653
39.
Van Rampelbergh
J
,
Achenbach
P
,
Leslie
RD
, et al
.
First-in-human, double-blind, randomized phase 1b study of peptide immunotherapy IMCY-0098 in new-onset type 1 diabetes
.
BMC Med
2023
;
21
:
190
40.
Nowak
C
,
Lind
M
,
Sumnik
Z
, et al
.
Intralymphatic GAD-Alum (Diamyd) improves glycemic control in type 1 diabetes with HLA DR3-DQ2
.
J Clin Endocrinol Metab
2022
;
107
:
2644
2651
41.
Leete
P
,
Oram
RA
,
McDonald
TJ
, et al.;
TIGI Study Team
.
Studies of insulin and proinsulin in pancreas and serum support the existence of aetiopathological endotypes of type 1 diabetes associated with age at diagnosis
.
Diabetologia
2020
;
63
:
1258
1267
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.