Identified genetic loci for C-peptide and age at diagnosis (AAD) in individuals with type 1 diabetes (T1D) explain only a small proportion of their variation. Here, we aimed to perform large meta–genome-wide association studies (GWAS) of C-peptide and AAD in T1D and to identify the HLA allele/haplotypes associated with C-peptide and AAD. A total of 7,252 and 7,923 European individuals with T1D were included in C-peptide and AAD GWAS, respectively. HLA-DQB1*06:02, which is strongly protective against T1D, was associated with higher C-peptide. HLA-DQB1*03:02, HLA-DRB1*03:01, and HLA-A*24:02, which increase T1D risk, were independently associated with younger AAD. HLA-DR3-DR4 haplotype combination, the strongest T1D susceptibility factor, was associated with younger AAD. Outside the HLA region, rs115673528 on chromosome 5 (Chr5) (GABRG2) was associated with C-peptide, and an indel, rs111970692, on Chr15 within CTSH, a known T1D locus, was associated with AAD. Genetically predicted CTSH expression, methylation, and protein levels were associated with AAD. Mendelian randomization analysis suggested that higher levels of pro-cathepsin H reduced AAD. In conclusion, some HLA allele/haplotypes associated with T1D also contribute to variability of C-peptide and AAD. Outside HLA, T1D loci were generally not associated with C-peptide or AAD. CTSH could be a potential therapeutic target to delay development/progression of T1D.
Identified genetic loci for C-peptide and type 1 diabetes (T1D) age at diagnosis (AAD) explain only a small proportion of their variation.
We aimed to identify additional genetic loci associated with C-peptide and AAD.
Some HLA allele/haplotypes associated with T1D also contributed to variability of C-peptide and AAD, whereas outside the HLA region, T1D loci were mostly not associated with C-peptide or AAD. Genetic variation within CTSH can affect AAD.
There is still residual heritability of C-peptide and AAD outside of HLA that could benefit from larger meta–genome-wide association studies.
Introduction
C-peptide is produced equimolar with insulin following cleavage of proinsulin. It undergoes little first-pass clearance by the liver and has a much longer half-life than insulin. Unlike with insulin assays, exogenous insulin does not interfere with C-peptide assays. Therefore, C-peptide is a widely used marker of endogenous insulin production (1). Numerous studies (2–6), have shown that detectable C-peptide persists years after diagnosis in a substantial proportion of individuals with type 1 diabetes (T1D). Moreover, patients with higher C-peptide require lower insulin doses and have fewer hypoglycemic and diabetic ketoacidosis events (7–9). Younger age at diagnosis (AAD) has been associated with lower C-peptide for a given T1D duration (4). A better understanding of the genetic architecture underpinning the variability in AAD and C-peptide could provide insights into the pathways involved and facilitate identification of potential targets for preservation of β-cell function in T1D. Furthermore, genetic determinants of AAD and C-peptide may be relevant for prediction of T1D progression from stage 1 (presymptomatic with evidence for β-cell autoimmunity) to stage 3 (symptomatic) (10).
Single-nucleotide polymorphism (SNP) heritability of C-peptide has been estimated at 26%, but known T1D loci account for only 1% of the variation in C-peptide (4). We previously reported that a genetic risk score (GRS) for T1D based on HLA-DR3-DR4 is strongly associated with early AAD and lower C-peptide. We also reported significant SNP associations in the HLA and INS regions with C-peptide and confirmed a reported association of PTPN22. T1D and type 2 diabetes GRSs based on SNPs in the HLA region adjusted for the known T1D-associated serotypes are strongly associated with C-peptide but not AAD (4). In an independent meta–genome-wide association study (GWAS) of C-peptide, we identified a SNP on chromosome 1 (Chr1) (rs559047, not a known T1D locus) and multiple SNPs in the HLA region, excluding HLA-DR3-DR4. Only 5 loci of >50 known T1D loci at the time were nominally associated with C-peptide (HLA-A*24, IL27, PTPN2, INS, and DEXI) (11). T1D GRS and JAZF1 have been associated with area under the curve (AUC) C-peptide during oral glucose tolerance tests (OGTTs) in autoantibody-positive individuals prior to T1D development (12). TCF7L2 has also been associated with AUC C-peptide during OGTTs within individuals newly diagnosed with T1D (13). A study in a Chinese population reported that a T1D GRS based on eight SNPs associated with T1D in this population is associated with younger AAD and lower fasting C-peptide (14). Since identified loci for C-peptide explain only a very small proportion of variation, and the results are often conflicting (4,11), further exploration of the genetics of C-peptide in T1D is warranted. Similarly, there are limited data on the genetics of AAD with HLA and a locus near PTPRK/THEMIS being the only loci identified through GWAS (15). Here, we performed a large meta-GWAS of C-peptide and AAD in T1D that included cohorts from our earlier studies (4,11), with updated imputation both genome wide and for HLA.
Research Design and Methods
Study Design
We performed a meta-GWAS of C-peptide that including four T1D studies: the Scottish Diabetes Research Network Type 1 Bioresource (SDRNT1BIO) (4,16), Diabetes Control and Complications Trial (DCCT) (17,18), Coronary Artery Calcification in T1D (CACTI) (19,20), and Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) (21). A meta-GWAS of AAD was performed that comprised all studies included in the C-peptide meta-GWAS plus the Pittsburgh Epidemiology of Diabetes Complications (EDC) study (22). We also performed HLA imputation to investigate the association of HLA allele/haplotypes with C-peptide and AAD. In addition, we used individuals from Generation Scotland (GS) (23) as control individuals versus SDRNT1BIO case individuals to explore HLA allele/haplotype associations with T1D (Supplementary Tables 1–3).
New associations discovered in the meta-GWAS were followed up by mapping the lead trait-associated variants to the corresponding genes. The role of these identified genes was then investigated through integrative analysis of C-peptide and AAD GWAS data with molecular quantitative trait loci (QTL) for gene products. For each identified gene, we computed genetically predicted levels of gene expression, DNA methylation at CpG sites within the gene or its proximity, and the protein encoded by the gene. We then tested associations of these genotypic scores with C-peptide and AAD. The causal roles of identified genes were also investigated via two-sample Mendelian randomization (2SMR) analysis between the gene product (exposure) and either C-peptide or AAD (outcome).
Genotyping and Imputation
Illumina platforms (Illumina, San Diego, CA) were used for genotyping (Supplementary Table 4). Ungenotyped variants were imputed using Trans-Omics for Precision Medicine (TOPMed) reference panel r2 (GRCh38) (24).
Estimating Heritability of C-Peptide and AAD
We estimated heritability of C-peptide and AAD in SDRNT1BIO using pedigree- and SNP-based heritability simultaneously using genetic relatedness estimation through maximum likelihood (25) implemented in Genome-wide Complex Trait Analysis. To estimate the fraction of heritability attributable to variants outside HLA, we repeated the analysis while excluding all Chr6 variants between 28,277,797 and 33,648,354 base pairs (bp).
GWAS and Meta-GWAS
GWAS were performed using frequentist association tests and the additive model implemented in SNPTEST version 2.5 (https://www.well.ox.ac.uk/~gav/snptest/) (26,27), which performs an F test for a linear model for quantitative traits. Unrelated people of European ancestry, as determined by genetic principal components analysis (4,11,28,29), and SNPs (Chr1–22 and ChrX) with minor allele frequency (MAF) >0.01 and high imputation quality (INFO) >0.80 were included in the analysis.
C-peptide levels were first converted to pmol/L to ensure consistency across studies. Then, all values at the lower limit of detection were converted to 0. Next, all values were increased by 1 to avoid zeros that have undefined log10 and subsequently log10 transformed. In SDRNT1BIO, this transformation produced a near-normal distribution, and log10(C-peptide + 1) values were used in the GWAS model with sex, AAD, T1D duration, and the first three genetic principal components as covariates (29). In DCCT, CACTI, and WESDR, many C-peptide values were below the detection level, and log10(C-peptide + 1) transformation produced distributions that were still skewed. Linear regressions were performed with the transformed C-peptide as the outcome and sex, AAD, T1D duration, and cohort (in DCCT only) as predictors. The residuals from this model, which had close-to-normal distribution, were used as the outcome for GWAS. The effect sizes from these two pipelines (adding the covariates to the GWAS model or using the residuals with respect to these covariates as the GWAS outcome) are equivalent and can be used in meta-analysis.
AAD was square root transformed, and sex and the first three principal components (SDRNT1BIO only) were included as covariates in the model.
Meta-analyses were performed using METAL version 2011.03.25 (https://www.sph.umich.edu/csg/abecasis/Metal/index.html) (30) with STDERR method, which weights effect size estimates by the inverse of the corresponding SEs. Only SNPs present in both SDRNT1BIO and DCCT (the two largest cohorts) were included in the meta-GWAS.
HLA Imputation
Classical HLA alleles, amino acid (AA) changes, and SNPs in the HLA region were imputed using the four-digit multiethnic HLA reference panel version 1 on the Michigan imputation server (GRCh37), including 1,781 HLA alleles (31) in T1D cohorts. GS had been imputed using the G-group (two-digit) HLA panel. SDRNT1BIO was additionally imputed using the G-group panel to combine with GS for T1D risk analyses.
Phased haplotypes at DRB1-DQA1-DQB1 were used to derive three HLA class II DR-DQ haplotypes most strongly associated with T1D: the risk haplotypes DR3-DQ2 (DRB1*03:01-DQA1*05:01-DQB1*02:01) and DR4-DQ8 (DRB1*04-DQA1*03-DQB1*03:02) and the protective haplotype DR15-DQ6 (DRB1*15-DQA1*01-DQB1*06) (32–34). These haplotypes were defined using two-digit resolution for consistency across studies. Individuals were categorized into 10 groups based on whether they carry one or more of the DR3-DQ2, DR4-DQ8, and DR15-DQ6 haplotypes, with all other haplotypes denoted by X: DR3/X, DR4/X, DR15/X, DR3/DR3, DR3/DR4, DR3/DR15, DR4/DR4, DR4/DR15, DR15/DR15, and X/X. The X/X group, which denotes no copy of any of the three haplotypes, was used as the reference group in the association tests.
Associations of HLA alleles, DR-DQ haplotype combinations, AAs, and SNPs with C-peptide and AAD were tested in individual studies, and the results were combined through meta-analysis as described above. We also tested their association with T1D risk by comparing 4,958 case participants from SDRNT1BIO with 7,474 control individuals from GS.
Non-HLA T1D Risk Loci
Association of the Locus-Specific Scores for CTSHWith C-Peptide and AAD
To evaluate whether the genetic effect at cathepsin H (CTSH) is mediated through the products of this gene, we used the GENOSCORES platform (37) to compute genotypic scores reflecting the predicted level of genetically regulated expression, methylation, and protein level of CTSH and tested these for association with C-peptide and AAD. We used results from three large QTL studies based on >20,000 individuals: 1) GWAS of CTSH gene expression in blood from the eQTLGen consortium (38); 2) GWAS of CTSH methylation in blood from the Genetics of DNA Methylation Consortium (GoDMC) (39); and 3) GWAS of circulating levels of pro-cathepsin H protein (encoded by CTSH) in blood from deCODE (40).
For each GWAS, summary statistics were initially filtered to retain only SNPs associated at P < 1E-5 with the trait (CTSH expression, methylation, or protein level). The retained, trait-associated SNPs were grouped together in QTLs, where each QTL contained at least one SNP with P < 1E-7 and QTLs were separated by at least half a Mb. For each QTL, univariate regression coefficients for SNPs with P < 1E-5 were included in a locus-specific weights vector. This vector was multiplied by the inverse correlation matrix between SNP genotypes, computed in the European ancestry subset of the 1000 Genomes Project reference panel (41) to adjust for linkage disequilibrium (LD). Finally, a locus-specific genotypic score for each QTL was computed by multiplying the vector of adjusted weights by the matrix of SNP genotypes in SDRNT1BIO. This procedure approximates fitting all the SNPs in the QTL jointly in the original GWAS data set and obtaining the predicted values of this model for SDRNT1BIO individuals.
Scores were classified into cis, cis-x, and trans if the distance between the transcription site of CTSH and the start and end positions of the corresponding QTL (SNP positions at the two ends) was <50 kb, 50 kb to 5 Mb, and >5 Mb, respectively. Each score was then tested for an association with AAD and C-peptide using the same outcome transformation and covariates as in SDRNT1BIO GWAS. While association with cis scores for CTSH expression, methylation, and protein level might be driven by LD between the SNPs affecting the outcome and the SNPs affecting CTSH activity at that locus, scores at different loci are independent proxies of predicted activity of CTSH. Associations with consistent direction of effect from different QTLs would therefore support a causal role of CTSH for the outcome. To account for multiple testing, P values were adjusted using the Benjamini-Hochberg procedure to control the false discovery rate (FDR) as implemented using function p.adjust() in R. Associations with FDR-adjusted P < 0.05 are considered statistically significant.
2SMR Analysis of CTSH Effect on C-Peptide and AAD
For MR analysis, we followed guidelines (42) using TwoSampleMR package (43). Instruments for the exposure (circulating pro-cathepsin H) were selected using summary statistics from the deCODE study (40). Protein QTLs for pro-cathepsin H were LD clumped using PLINK 2.0 (44) to select independent instruments. In total, eight protein-associated lead SNPs (P < 1E-10) were selected (three cis and five trans). The SDRNT1BIO GWAS of AAD and C-peptide were used to extract summary statistics for the two outcomes. 2SMR analysis was performed using inverse variance–weighted, simple median, and simple mode methods (43).
Data and Resource Availability
DCCT clinical data are available from https://repository.niddk.nih.gov/studies/edic/, and genetic data are available from https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000086.v3.p1. Applicants who wish to access SDRNT1BIO data can submit an initial request by emailing [email protected]. The summary statistics for C-peptide and AAD meta-GWAS are available from Zenodo (https://doi.org/10.5281/zenodo.8319943, https://doi.org/10.5281/zenodo.8320208).
Results
C-Peptide
Heritability
SNP heritability of C-peptide attributable to variants outside HLA was estimated at 12.9%, approximately one-half of the full SNP heritability of 26%.
Meta-GWAS
C-peptide meta-GWAS included 7,252 participants and 8,150,645 SNPs (Supplementary Table 5 and Supplementary Figs. 1 and 2). The genome-wide significance threshold (P < 5E-8) was attained for 18 SNPs in HLA-DR-DQ (top SNP: rs9271349; Chr6:32,616,053, A > G; MAF = 0.01–0.02; β [SE] = 0.57 [0.08]; P = 1.62E-12) and a single SNP in intron 1 of γ-aminobutyric acid type A receptor subunit gamma2 (GABRG2) (rs115673528; Chr5:162,087,705, A > G; MAF = 0.01–0.02; β [SE] = 0.50 [0.09]; P = 1.16E-8) (Fig. 1 and Supplementary File CP.xlsx, Sheet A). The signal at rs115673528 appeared to mostly come from CACTI, where it was associated with C-peptide at genome-wide significance (P = 1.43E-9). The direction of effect was consistent with that in the other studies but was only nominally significant in SDRNT1BIO (P = 0.04) and not significant in DCCT (P = 0.14) and WESDR (P = 0.08). As expected, on excluding CACTI from the meta-analysis, rs115673528 was no longer genome-wide significant (β [SE] = 0.31 [0.10]; P = 2.64E-3). SNP rs9271349 was associated with younger AAD (β [SE] = −0.60 [0.07]; P = 3.03E-16), whereas rs115673528 was not associated with AAD (P = 0.24).
C-Peptide meta-GWAS. A: Manhattan plot. B: Q-Q plot. C: HLA locus (TOPMed imputation). D: GABRG2 locus. E: HLA region (HLA imputation). F: HLA region conditional to HLA-DQB1*06:02. The plots in A–D were made using LocusZoom (https://my.locuszoom.org/). The LD is based on the European population from the 1000 Genomes LD panel derived from deep whole-genome sequencing (50). The coordinates are based on GRCh38. The plots in E and F were made using HLAManhattan in HLA-TAPAS (https://github.com/immunogenomics/HLA-TAPAS) (31). HLA haplotypes, AA changes, and SNPs/indels are in red, yellow, and gray, respectively. The coordinates are based on GRCh37.
C-Peptide meta-GWAS. A: Manhattan plot. B: Q-Q plot. C: HLA locus (TOPMed imputation). D: GABRG2 locus. E: HLA region (HLA imputation). F: HLA region conditional to HLA-DQB1*06:02. The plots in A–D were made using LocusZoom (https://my.locuszoom.org/). The LD is based on the European population from the 1000 Genomes LD panel derived from deep whole-genome sequencing (50). The coordinates are based on GRCh38. The plots in E and F were made using HLAManhattan in HLA-TAPAS (https://github.com/immunogenomics/HLA-TAPAS) (31). HLA haplotypes, AA changes, and SNPs/indels are in red, yellow, and gray, respectively. The coordinates are based on GRCh37.
HLA Imputation
HLA-DQB1*06:02:01:01 (β = 0.90, P = 1.18E-16) and HLA-DRB1*15:01:01:01 (β = 0.70, P = 1.68E-12) alleles were associated with higher C-peptide (Table 1, Fig. 1, and Supplementary File CP.xlsx, Sheet B). Furthermore, after including HLA-DQB1*06:02 in the model, no other allele was genome-wide significant, with HLA-A*24 being the top-associated allele (β [SE] = −0.18 [0.03]; P = 3.00E-7; frequency = 0.10) (Supplementary File CP.xlsx, Sheet E).
HLA haplotypes associated with C-peptide or AAD
Haplotype . | bp (HG19) . | Freq . | Effect . | SE . | P . | Direction* . | Het I2 . | Het P . |
---|---|---|---|---|---|---|---|---|
C-peptide | ||||||||
HLA-DQB1*06:02:01:01 | 32,627,324 | 0.01 | 0.90 | 0.11 | 1.18E-16 | + + + + | 79.00 | 2.52E-03 |
HLA-DRB1*15:01:01:01 | 32,546,786 | 0.01 | 0.70 | 0.10 | 1.68E-12 | + + + + | 85.10 | 1.55E-04 |
AAD | ||||||||
HLA-DQB1*03:02:01:01 | 32,627,261 | 0.29 | −0.26 | 0.03 | 1.76E-24 | − − − − − | 84.20 | 4.30E-05 |
HLA-DQA1*02:01:01:01 | 32,605,195 | 0.07 | 0.33 | 0.04 | 3.81E-14 | + − + + + | 86.50 | 5.62E-06 |
HLA-DRB1*15:01:01:01 | 32,546,786 | 0.02 | 0.66 | 0.09 | 6.07E-14 | + + + + + | 62.10 | 3.20E-02 |
HLA-DQB1*06:02:01:01 | 32,627,324 | 0.01 | 0.71 | 0.09 | 8.74E-14 | + + − ? + | 79.90 | 1.91E-03 |
HLA-DRB1*07:01:01:01 | 32,546,630 | 0.07 | 0.32 | 0.04 | 3.56E-13 | + − + + + | 83.70 | 6.19E-05 |
HLA-DQA1*03:01:01 | 32,605,198 | 0.37 | −0.17 | 0.02 | 1.54E-12 | − − − + − | 63.80 | 2.59E-02 |
HLA-DQB1*03:01:01:01 | 32,627,258 | 0.09 | 0.26 | 0.04 | 1.01E-11 | + + − + + | 58.50 | 4.68E-02 |
HLA-DQA1*01:02:01:01 | 32,605,187 | 0.07 | 0.27 | 0.04 | 3.41E-11 | + + + − + | 84.40 | 3.74E-05 |
HLA-A*24:02:01:01 | 29,910,454 | 0.09 | −0.22 | 0.04 | 5.18E-10 | − − − − − | 14.80 | 3.20E-01 |
HLA-DRB1*04 | 32,546,587 | 0.33 | −0.15 | 0.03 | 1.14E-09 | − − − − − | 51.20 | 8.45E-02 |
AAD conditional to HLA-DQB1*03:02 | ||||||||
HLA-DRB1*03:01:01:01 | 32,546,570 | 0.32 | −0.18 | 0.02 | 6.73E-15 | − − − + − | 65.80 | 1.99E-02 |
AAD conditional to HLA-DQB1*03:02 and HLA-DRB1*03:01 | ||||||||
HLA_A*24:02:01:01 | 29,910,454 | 0.09 | −0.27 | 0.04 | 3.08E-14 | − − − − − | 6.56 | 1.61E-01 |
Haplotype . | bp (HG19) . | Freq . | Effect . | SE . | P . | Direction* . | Het I2 . | Het P . |
---|---|---|---|---|---|---|---|---|
C-peptide | ||||||||
HLA-DQB1*06:02:01:01 | 32,627,324 | 0.01 | 0.90 | 0.11 | 1.18E-16 | + + + + | 79.00 | 2.52E-03 |
HLA-DRB1*15:01:01:01 | 32,546,786 | 0.01 | 0.70 | 0.10 | 1.68E-12 | + + + + | 85.10 | 1.55E-04 |
AAD | ||||||||
HLA-DQB1*03:02:01:01 | 32,627,261 | 0.29 | −0.26 | 0.03 | 1.76E-24 | − − − − − | 84.20 | 4.30E-05 |
HLA-DQA1*02:01:01:01 | 32,605,195 | 0.07 | 0.33 | 0.04 | 3.81E-14 | + − + + + | 86.50 | 5.62E-06 |
HLA-DRB1*15:01:01:01 | 32,546,786 | 0.02 | 0.66 | 0.09 | 6.07E-14 | + + + + + | 62.10 | 3.20E-02 |
HLA-DQB1*06:02:01:01 | 32,627,324 | 0.01 | 0.71 | 0.09 | 8.74E-14 | + + − ? + | 79.90 | 1.91E-03 |
HLA-DRB1*07:01:01:01 | 32,546,630 | 0.07 | 0.32 | 0.04 | 3.56E-13 | + − + + + | 83.70 | 6.19E-05 |
HLA-DQA1*03:01:01 | 32,605,198 | 0.37 | −0.17 | 0.02 | 1.54E-12 | − − − + − | 63.80 | 2.59E-02 |
HLA-DQB1*03:01:01:01 | 32,627,258 | 0.09 | 0.26 | 0.04 | 1.01E-11 | + + − + + | 58.50 | 4.68E-02 |
HLA-DQA1*01:02:01:01 | 32,605,187 | 0.07 | 0.27 | 0.04 | 3.41E-11 | + + + − + | 84.40 | 3.74E-05 |
HLA-A*24:02:01:01 | 29,910,454 | 0.09 | −0.22 | 0.04 | 5.18E-10 | − − − − − | 14.80 | 3.20E-01 |
HLA-DRB1*04 | 32,546,587 | 0.33 | −0.15 | 0.03 | 1.14E-09 | − − − − − | 51.20 | 8.45E-02 |
AAD conditional to HLA-DQB1*03:02 | ||||||||
HLA-DRB1*03:01:01:01 | 32,546,570 | 0.32 | −0.18 | 0.02 | 6.73E-15 | − − − + − | 65.80 | 1.99E-02 |
AAD conditional to HLA-DQB1*03:02 and HLA-DRB1*03:01 | ||||||||
HLA_A*24:02:01:01 | 29,910,454 | 0.09 | −0.27 | 0.04 | 3.08E-14 | − − − − − | 6.56 | 1.61E-01 |
Freq, frequency; Het, heterogeneity.
*The order of studies for C-peptide analysis is SDRNT1BIO, DCCT, CACTI, and WESDR, and for AAD, the order is SDRNT1BIO, DCCT, CACTI, EDC, and WESDR.
None of the DR-DQ haplotype combinations was associated with C-peptide at the genome-wide significance threshold. However, haplotype combinations DR15/X, DR15/DR4, and DR15/DR15 were all nominally associated with higher C-peptide (Table 2 and Supplementary Tables 6 and 7).
Association of DR-DQ haplotype combinations with C-peptide, AAD, and T1D risk
Haplotype combination . | Frequency . | T1D risk . | C-peptide . | AAD . | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SDRNT1BIO . | GS . | β . | SE . | P . | β . | SE . | P . | Direction* . | Het I2 . | Het P . | β . | SE . | P . | Direction* . | Het I2 . | Het P . | |
DR3/X | 0.186 | 0.164 | 1.53 | 0.09 | 2.E–70 | −0/08 | 0.06 | 0.16 | + + − − | 66.90 | 0.03 | −0.23 | 0.06 | 6.96E-05 | − − + − − | 43.10 | 0.13 |
DR3/DR3 | 0.130 | 0.028 | 3.25 | 0.13 | 2.E–129 | −0.04 | 0.07 | 0.58 | + + + − | 49.20 | 0.12 | −0.25 | 0.07 | 1.14E-04 | − − − − + | 6.40 | 0.37 |
DR3/DR4 | 0.270 | 0.058 | 2.92 | 0.09 | 1.E–216 | −0.01 | 0.05 | 0.92 | + + − − | 62.90 | 4.43E-02 | −0.52 | 0.05 | 7.68E-22 | − − − − − | 82.10 | 1.69E-04 |
DR4/X | 0.229 | 0.174 | 1.36 | 0.07 | 2.E–77 | −0.08 | 0.05 | 0.12 | − + − − | 58.30 | 0.07 | −0.29 | 0.05 | 1.67E-07 | − − − − − | 44.40 | 0.13 |
DR4/DR4 | 0.079 | 0.028 | 2.11 | 0.11 | 9.E–85 | −0.02 | 0.07 | 0.74 | + − + − | 22.50 | 0.28 | −0.25 | 0.07 | 3.43E-04 | − − + − − | 0.00 | 0.49 |
DR3/DR15 | 0.007 | 0.052 | −0.48 | 0.2 | 2.E–02 | 0.34 | 0.23 | 0.13 | + + − + | 35.80 | 0.20 | 0.39 | 0.20 | 0.05 | + + 0 + ? | 19.20 | 0.29 |
DR4/DR15 | 0.009 | 0.058 | −0.66 | 0.18 | 2.E–04 | 0.48 | 0.16 | 2.00E-03 | + + + + | 40.70 | 0.17 | 0.13 | 0.15 | 0.40 | + + − − + | 0.00 | 0.76 |
DR15/X | 0.011 | 0.163 | −1.42 | 0.15 | 4.E–20 | 0.84 | 0.16 | 3.95E-07 | + + − + | 73.40 | 1.04E-02 | 0.53 | 0.15 | 2.56E-04 | + + − + ? | 80.50 | 1.49E-03 |
DR15/DR15 | 0.001 | 0.029 | −1.97 | 0.46 | 2.E–05 | 2.27 | 0.63 | 3.02E-04 | + ? + ? | 0.00 | 0.83 | 0.25 | 0.46 | 0.58 | + ? + ? ? | 0.00 | 0.57 |
X/X | 0.079 | 0.246 | Reference | Reference | Reference |
Haplotype combination . | Frequency . | T1D risk . | C-peptide . | AAD . | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SDRNT1BIO . | GS . | β . | SE . | P . | β . | SE . | P . | Direction* . | Het I2 . | Het P . | β . | SE . | P . | Direction* . | Het I2 . | Het P . | |
DR3/X | 0.186 | 0.164 | 1.53 | 0.09 | 2.E–70 | −0/08 | 0.06 | 0.16 | + + − − | 66.90 | 0.03 | −0.23 | 0.06 | 6.96E-05 | − − + − − | 43.10 | 0.13 |
DR3/DR3 | 0.130 | 0.028 | 3.25 | 0.13 | 2.E–129 | −0.04 | 0.07 | 0.58 | + + + − | 49.20 | 0.12 | −0.25 | 0.07 | 1.14E-04 | − − − − + | 6.40 | 0.37 |
DR3/DR4 | 0.270 | 0.058 | 2.92 | 0.09 | 1.E–216 | −0.01 | 0.05 | 0.92 | + + − − | 62.90 | 4.43E-02 | −0.52 | 0.05 | 7.68E-22 | − − − − − | 82.10 | 1.69E-04 |
DR4/X | 0.229 | 0.174 | 1.36 | 0.07 | 2.E–77 | −0.08 | 0.05 | 0.12 | − + − − | 58.30 | 0.07 | −0.29 | 0.05 | 1.67E-07 | − − − − − | 44.40 | 0.13 |
DR4/DR4 | 0.079 | 0.028 | 2.11 | 0.11 | 9.E–85 | −0.02 | 0.07 | 0.74 | + − + − | 22.50 | 0.28 | −0.25 | 0.07 | 3.43E-04 | − − + − − | 0.00 | 0.49 |
DR3/DR15 | 0.007 | 0.052 | −0.48 | 0.2 | 2.E–02 | 0.34 | 0.23 | 0.13 | + + − + | 35.80 | 0.20 | 0.39 | 0.20 | 0.05 | + + 0 + ? | 19.20 | 0.29 |
DR4/DR15 | 0.009 | 0.058 | −0.66 | 0.18 | 2.E–04 | 0.48 | 0.16 | 2.00E-03 | + + + + | 40.70 | 0.17 | 0.13 | 0.15 | 0.40 | + + − − + | 0.00 | 0.76 |
DR15/X | 0.011 | 0.163 | −1.42 | 0.15 | 4.E–20 | 0.84 | 0.16 | 3.95E-07 | + + − + | 73.40 | 1.04E-02 | 0.53 | 0.15 | 2.56E-04 | + + − + ? | 80.50 | 1.49E-03 |
DR15/DR15 | 0.001 | 0.029 | −1.97 | 0.46 | 2.E–05 | 2.27 | 0.63 | 3.02E-04 | + ? + ? | 0.00 | 0.83 | 0.25 | 0.46 | 0.58 | + ? + ? ? | 0.00 | 0.57 |
X/X | 0.079 | 0.246 | Reference | Reference | Reference |
Het, heterogeneity.
*The order of studies for the C-peptide analysis is SDRNT1BIO, DCCT, CACTI, and WESDR, and for AAD, the order is SDRNT1BIO, DCCT, CACTI, WESDR, and EDC.
Five AA changes and 94 SNPs in DR-DQ plus an AA change within HLA-A (AA_A_307_29912374_exon5_M) were associated with C-peptide (P < 5E-8) (Supplementary File CP.xlsx, Sheets C and D). All HLA alleles, AA changes (except for AA_A_307_29912374_exon5_M), and SNPs/indels associated with C-peptide were also associated with T1D and AAD in the expected directions: Those associated with higher C-peptide were associated with lower T1D risk and older AAD, whereas those associated with lower C-peptide were associated with higher T1D risk and younger AAD (Supplementary Figs. 3 and 4). AA_A_307_29912374_exon5_M, which was associated with lower C-peptide, was not associated with T1D risk (P = 0.24). However, it was nominally associated with younger AAD (β [SE] = −0.10 [0.04]; P = 4.49E-3).
Non-HLA T1D Loci
Only INS (rs689; Chr11:2,160,994, T > A) was associated with C-peptide. The A allele of rs689, which is protective against T1D, was associated with higher C-peptide (β [SE] = 0.14 [0.03]; P = 6.34E-7) (Supplementary Table 8).
AAD
Heritability
SNP heritability of AAD outside of HLA was estimated at 10.6%, approximately one-third of the full heritability of 33.5%.
Meta-GWAS
AAD meta-GWAS included 7,923 participants and 8,154,710 SNPs (Supplementary Table 9 and Supplementary Figs. 5 and 6). A total of 1,721 SNPs in HLA spanning Chr6:29,851,598–32,976,202 (top SNP: rs1794269; Chr6:32706117, T > C; MAF = 0.20–0.26; β [SE] = 0.28 [0.02]; P = 2.86E-37) and an indel on Chr15 within a four-nucleotide (TGTT) tandem repeat rs111970692 (Chr15:78943251, CTGTT>C; MAF = 0.05–0.08; β [SE] = 0.21 [0.04]; P = 3.77E-8) within CTSH were associated with AAD (Fig. 2 and Supplementary File AAD.xlsx, Sheet A). rs1794269 (P = 0.25) and rs111970692 (P = 0.75) were not associated with C-peptide.
AAD meta-GWAS. A: Manhattan plot. B: Q-Q plot. C: CTSH locus. D: HLA-DR-DQ locus. E: HLA-A locus. F: HLA region (HLA imputation). G: HLA region conditional to HLA-DQB1*03:02. H: HLA region conditional to HLA-DQB1*03:02 and HLA-DRB1*03:01. I: HLA region conditional to HLA-DQB1*03:02, HLA-DQB1*03:02, and HLA-A*24:02:01:01. The plots in A–E were made using LocusZoom (https://my.locuszoom.org/). The LD is based on the European population from 1000 Genomes LD panel derived from deep whole-genome sequencing (50). The coordinates are based on GRCh38. The plots in F and G were made using HLAManhattan in HLA-TAPAS (https://github.com/immunogenomics/HLA-TAPAS) (31). HLA haplotypes, AA changes/indels are in red, yellow, and gray, respectively. The coordinates are based on GRCh37.
AAD meta-GWAS. A: Manhattan plot. B: Q-Q plot. C: CTSH locus. D: HLA-DR-DQ locus. E: HLA-A locus. F: HLA region (HLA imputation). G: HLA region conditional to HLA-DQB1*03:02. H: HLA region conditional to HLA-DQB1*03:02 and HLA-DRB1*03:01. I: HLA region conditional to HLA-DQB1*03:02, HLA-DQB1*03:02, and HLA-A*24:02:01:01. The plots in A–E were made using LocusZoom (https://my.locuszoom.org/). The LD is based on the European population from 1000 Genomes LD panel derived from deep whole-genome sequencing (50). The coordinates are based on GRCh38. The plots in F and G were made using HLAManhattan in HLA-TAPAS (https://github.com/immunogenomics/HLA-TAPAS) (31). HLA haplotypes, AA changes/indels are in red, yellow, and gray, respectively. The coordinates are based on GRCh37.
HLA Imputation
DQB1*03:02:01:01 (β = −0.26; P = 1.76E-24), DQB1*06:02:01:01, DQA1*02:01:01:01, DRB1*15:01:01:01, DRB1*07:01:01:01, DQA1*03:01:01, DQB1*03:01:01:01, DQA1*01:02:01:01, DRB1*04, and A24*02:01:01 alleles, as well as 1,233 AA changes and 4,196 SNP/indels, were associated with AAD (Table 1, Fig. 2, and Supplementary File AAD.xlsx, Sheets B, C, and D). All these HLA alleles, AA changes, and SNPs except for eight were also associated T1D risk in the expected direction; those associated with older AAD were associated with a lower T1D risk and vice versa (Supplementary Fig. 7).
Furthermore, after including HLA-DQB1*03:02, the top-associated HLA allele, in the model, seven alleles retained association with AAD, exceeding genome-wide significance, with HLA-DRB1*03:01:01:01 becoming the top-associated allele (β = −0.18; P = 6.73E-15) (Table 1 and Supplementary File AAD.xlsx, Sheet E). After including HLA-DQB1*03:02 and DRB1*03:01 in the model, only HLA-A*24:02:01:01 reached the genome-wide significance threshold (β = −0.27; P = 3.08E-14) (Table 1 and Supplementary File AAD.xlsx, Sheet F). Further adding HLA-A*24:02:01:01 in the model, no other HLA allele reached the genome-wide significance threshold (Supplementary File AAD.xlsx, Sheet G).
Among DR-DQ haplotype combinations, DR3/DR4, which was strongly associated with T1D risk, was associated with younger AAD (β = −0.52; P = 7.68E-22). DR3/X, DR4/X, DR3/DR3, and DR4/DR4 were all associated with younger AAD, whereas DR15/X, which was strongly protective against T1D, was associated with older AAD at a nominal level (Table 2and Supplementary Tables 7 and 10).
Non-HLA T1D Loci
Three loci, including rs2816313 (Chr1:192,570,207, G > A; RGS1), rs61839660 (Chr10:6,052,734, C > T; IL2RA), and rs34593439 (Chr15:78,942,615, G > A, CTSH), were associated with AAD. rs2816313 increases T1D risk and was associated with younger AAD (β [SE] = −0.10 [0.02]; P = 1.53E-5), whereas both rs61839660 and rs34593439 are protective against T1D and were associated with older AAD (β = 0.19 [SE 0.04; P = 2.00E-6] and 0.20 [SE 0.04; P = 2.14E-7], respectively) (Supplementary Table 8).
Association of Locus-Specific Genotypic Scores for CTSH With AAD
We computed 20 locus-specific genotypic scores for CTSH traits: 6 for methylation (cis scores for 6 methylation probes mapped to CTSH), 3 for expression in whole blood (1 cis and 2 trans scores), and 11 for pro-cathepsin H protein in plasma (1 cis and 10 trans scores for SOMAmer SL000346 in SomaScan version 4) (Supplementary Table 11).
In total, seven scores were significantly associated with AAD (FDR P < 0.05), including five cis scores for CTSH methylation, the cis score for CTSH whole blood expression, and the cis score and one trans score for pro-cathepsin H (Table 3). Of the nine remaining trans scores computed for pro-cathepsin H, six had a consistent direction of effect with higher pro-cathepsin H associated with younger AAD (Supplementary Table 11). None of the scores was associated with C-peptide.
Association of genotypic risk scores with AAD
Trait . | Chr . | Start (HG38) . | End (HG38) . | Type . | Effect . | P . |
---|---|---|---|---|---|---|
Methylation* (probe location in HG38) | ||||||
cg17270013 (Chr15:78,944,497) | 15 | 78,185,519 | 79,036,756 | cis | −0.10 | 4.61E-06 |
cg07448499 (Chr15:78,944,815) | 15 | 78,185,519 | 79,036,756 | cis | −0.08 | 5.62E-05 |
cg17922215 (Chr15:78,944,803) | 15 | 78,388,660 | 79,012,268 | cis | −0.08 | 9.20E-05 |
cg20059407 (Chr15: 78,941,976) | 15 | 78,796,769 | 78,965,955 | cis | −0.07 | 3.99E-04 |
cg25744700 (Chr15:78,944,875) | 15 | 77,955,110 | 79,019,093 | cis | −0.07 | 6.89E-04 |
Expression† | ||||||
Whole blood (eQTLGen) | 15 | 77,974,313 | 79,068,033 | cis | −0.08 | 2.93E-04 |
Protein levels‡ | ||||||
Pro-cathepsin H | 6 | 31,483,699 | 33,225,090 | trans | −2.01 | 4.44E-05 |
Pro-cathepsin H | 15 | 67,936,125 | 88,137,453 | cis | −0.07 | 6.41E-03 |
Trait . | Chr . | Start (HG38) . | End (HG38) . | Type . | Effect . | P . |
---|---|---|---|---|---|---|
Methylation* (probe location in HG38) | ||||||
cg17270013 (Chr15:78,944,497) | 15 | 78,185,519 | 79,036,756 | cis | −0.10 | 4.61E-06 |
cg07448499 (Chr15:78,944,815) | 15 | 78,185,519 | 79,036,756 | cis | −0.08 | 5.62E-05 |
cg17922215 (Chr15:78,944,803) | 15 | 78,388,660 | 79,012,268 | cis | −0.08 | 9.20E-05 |
cg20059407 (Chr15: 78,941,976) | 15 | 78,796,769 | 78,965,955 | cis | −0.07 | 3.99E-04 |
cg25744700 (Chr15:78,944,875) | 15 | 77,955,110 | 79,019,093 | cis | −0.07 | 6.89E-04 |
Expression† | ||||||
Whole blood (eQTLGen) | 15 | 77,974,313 | 79,068,033 | cis | −0.08 | 2.93E-04 |
Protein levels‡ | ||||||
Pro-cathepsin H | 6 | 31,483,699 | 33,225,090 | trans | −2.01 | 4.44E-05 |
Pro-cathepsin H | 15 | 67,936,125 | 88,137,453 | cis | −0.07 | 6.41E-03 |
*Methylation probes on HumanMethylation450 and EPIC arrays corresponding to CpG island in CTSH locus: Chr15:78944698–78945229.
†RNA sequencing and arrays meta-analysis for eQTLGen.
‡Pro-cathepsin H protein levels measured in plasma with SomaScan version 3.
MR Analysis of Pro-Cathepsin H Effect on AAD
Significant association of two independent scores (one cis and one trans) for levels of pro-cathepsin H with AAD suggested a causal role of this protein. We further investigated this via 2SMR analysis. We identified eight SNPs as instruments for pro-cathepsin H from the protein QTL summary statistics (40) (Supplementary Table 12). Although MR analysis of pro-cathepsin H effect on AAD was not statistically significant, all three methods agreed and suggested that higher levels of pro-cathepsin H reduce AAD (Supplementary Table 13 and Supplementary Fig. 8). MR analysis did not support a causal effect of pro-cathepsin H on C-peptide (Supplementary Table 14 and Supplementary Fig. 9).
Discussion
We conducted a large GWAS of C-peptide and AAD in individuals with T1D from European ancestry using TOPMed and HLA imputation. Multiple SNPs in HLA and rs115673528 within GABRG2 were associated with C-peptide. HLA imputation and conditional analysis showed that HLA-DQB1*06:02 is the only HLA allele independently associated with C-peptide. HLA-DQB1*06:02, which is protective against T1D, was associated with higher C-peptide. DR4/DR15, DR15/X, and DR15/DR15 haplotype combinations, which are protective against T1D, were nominally associated with higher C-peptide. The GABRG2 signal was mainly coming from CACTI, where it was associated with C-peptide at the genome-wide significance threshold. The direction of effect was consistent in the other studies, but the effect sizes were smaller and the association only nominally significant in SDRNT1BIO with the largest sample size. Due to winner’s curse (45), it is very likely that the effect size was overestimated in CACTI, and considering low frequency of the SNP (MAF ∼1–2%), there was not enough statistical power to detect it in the other studies with smaller sample sizes. There was also significant heterogeneity across cohorts. Since our primary analysis here was the meta-GWAS where the signal reached genome-wide significance, this finding requires validation in independent cohorts. GABRG2 is not a known T1D locus and has not been associated with C-peptide previously.
We also identified many SNPs in HLA and an indel rs111970692 within CTSH associated with AAD. HLA imputation and conditional analysis revealed three independent HLA alleles associated with AAD: HLA-DQB1*03:02, HLA-DRB1*03:01, and HLA-A*24:02. All three alleles increase T1D risk and were associated with younger AAD. DR3/DR4 haplotype combination, which is a strong risk factor for T1D, was associated with younger AAD. DR3/X, DR3/DR3, DR4/X, and DR4/DR4, which increase T1D risk, were nominally associated with younger AAD, whereas DR15/X, which is protective against T1D, was nominally associated with older AAD. CTSH is a known T1D locus (36,46) and has been associated with AAD, although not at genome-wide significance (47).
Most of the 86 known T1D loci outside HLA (35,36) were not associated with C-peptide or AAD. Only INS was associated with C-peptide, and three loci (RGS1, IL2RA, and CTSH) were associated with AAD. Although this could partly be due to a lack of statistical power, it suggests that outside HLA, loci affecting C-peptide or AAD do not generally overlap with T1D loci. This was consistent with previous findings that T1D GRS is associated with random C-peptide but mainly driven by HLA and is nonsignificant when excluded (6). rs111970692 (MAF = 0.05–0.08, INFO ≥0.92 in all five studies) is located within a tandem repeat (TGTT, reference has five repeats) in intron 1 of CTSH, causing deletion of one repeat. The deletion was associated with older AAD and is in LD (r2 = 0.06, D′ = 1 based on the CEU population in 1000 Genomes [GRCh38 high coverage]) with the A allele of rs34593439 (G > A, A frequency = 0.11) 537 bp away, which is protective against T1D (36). Allele rs111970692 is also in LD (r2 = 0.06, D′ = 1) with the T allele of rs2289702 (Chr15:78944951, C > T, Gly12Arg in NM_004390.5, T frequency = 0.10), a missense variant within CTSH. rs2289702 was associated with AAD but below genome-wide significance (β = −0.196; P = 3.12E-7; INFO ≥0.93 in all five studies). CTSH encodes two minor histocompatibility antigen epitopes: HLA-A*3101 and HLA-A*3303. The immunogenicity of both epitopes depends on the Arg residue at position 12, and its substitution with Gly completely eliminates their binding to their corresponding HLA molecules (48). However, neither HLA-A*3101 (β = 0.04; P = 0.74) nor HLA-A*3303 (β = −0.01; P = 0.74) was associated with AAD, and there was no significant interaction between HLA-A*3101 (β = 0.31; P = 0.28) or HLA-A*3303 (β = 0.05; P = 0.94) and rs2289702 affecting AAD as tested in the SDRNT1BIO cohort. The cis scores for genetically predicted CTSH expression in blood, DNA methylation at CTSH, and circulating pro-cathepsin H protein (encoded by CTSH) were associated with younger AAD, suggesting that detected genetic effects at CTSH might affect AAD by altering CTSH products. This was further supported by a genotypic score combining trans effects of SNPs at the HLA region (Chr6:31,483,699–33,225,090) on pro-cathepsin H levels, which was also associated with younger AAD. MR analysis using three cis-SNPs and five trans-SNPs as instruments did not detect a significant causal effect of pro-cathepsin H on AAD. However, the five trans-SNPs have very small effects on pro-cathepsin H and may not satisfy the MR relevance assumption (strong instruments). Furthermore, since there were only eight instruments in total, statistical methods that attempt to account for pleiotropy become less reliable, as there is not enough information to differentiate between valid instruments and instruments that violate the MR exclusion restriction assumption (pleiotropic effects). These limitations reduce the power of the MR analysis to detect a significant causal effect, and thus, findings from this analysis should be considered together with other lines of evidence in support of a causal role for CTSH on AAD.
The previously reported rs559047 C-peptide GWAS locus (11) did not reach the genome-wide significance threshold in the current meta-GWAS. Both rs864745 (JAZF1) and rs4506565 (TCF7L2) previously associated with AUC C-peptide during OGTT in autoantibody-positive individuals (12) and individuals newly diagnosed with T1D (13) were not associated with C-peptide (P = 0.02 [opposite direction] and 0.39, respectively) in the current analysis. rs72975913 (15) and rs2941522 (49), which have been associated with AAD, were both nominally associated with AAD in the current analysis, with a consistent direction of effect (β = −0.09 [SE 0.03; P = 9.73E-4] and −0.04 [SE 0.02; P = 0.049], respectively).
Our analyses have some limitations. The participants included in the cohorts we used may not be representative of the general population of subjects with T1D as extensive inclusion/exclusion criteria were applied in some cohorts. Many T1D genetic studies included only individuals diagnosed at <18 years of age to reduce the chance of including people with type 2 diabetes and for logistical reasons, such as recruitment through pediatric hospitals (27). Therefore, AAD is generally older in cohorts included in the current analyses compared with similar studies performed earlier (15,47,49). C-peptide was measured with different assays with different sensitivity (i.e., lower limit of detection) and in different conditions (i.e., stimulated, fasting, random) in different cohorts.
In conclusion, our analyses show that some T1D HLA allele/haplotypes affect C-peptide or AAD, whereas outside the HLA, T1D risk loci do not generally affect C-peptide or AAD. Genetic variations within CTSH are associated with AAD and with DNA methylation, gene expression, and circulating protein levels. This locus could potentially be a therapeutic target for personalized medicine to delay development of T1D. Identified loci for C-peptide and AAD still explain a small proportion of variation, and larger studies are required to identify additional loci.
This article contains supplementary material online at https://doi.org/10.2337/figshare.27733341.
Article Information
Acknowledgments. The authors thank the fieldworkers, supervisors, and collaborators from the NHS Health Boards across Scotland for facilitating the SDRNT1BIO and GS research studies in their facilities. The authors also thank the DCCT/EDIC research group for contributing its data and all participants from these studies.
Funding. This study was supported by Medical Research Council grant MR/T032340/1 and Canadian Institute of Health Research grant UCD-170583. The establishment of the SDRNT1BIO was supported by Chief Scientist Office of the Scottish Government Health Directorates grant ETM/47, by Diabetes UK grant 10/0004010), and by in-kind contributions from the Scottish Diabetes Research Network. Genotyping was supported by JDRF grant 17-2013-7. GS received core support from Chief Scientist Office of the Scottish Government Health Directorates grant CZD/16/6 and Scottish Funding Council grant HR03006. The development of the GENOSCORES platform was supported by Academy of Medical Sciences Springboard Award SBF006/1109, supported in turn by the Wellcome Trust; the UK Government Department of Business, Energy and Industrial Strategy; the British Heart Foundation; and Diabetes UK. C.H. was supported by MRC Human Genetics Unit programme grant U. MC_UU_00007/10. The CACTI study was funded by National Institutes of Health grants P30 DK057516, R01 HL113029, R01 HL079611, UL1 TR002535, P30 DK116073, M01 RR000051, and R01 HL061753. EDC is funded by National Institutes of Health grant R01-DK034818 and by the Rossi Memorial Fund.
Duality of Interest. H.M.C. receives honoraria from Novo Nordisk; is a member of advisory boards for Novo Nordisk and Bayer AG; receives or has recently received research funding from Diabetes UK, JDRF, IQVIA, and the European Union Commission; and P.M.M. holds shares in Roche Pharmaceuticals and Bayer AG. No other conflicts of interest relevant to this article were reported.
Author Contributions. D.R. contributed to the formal analysis, data curation, and visualization and drafted the manuscript. A.S., A.I., and D.L. contributed to the formal analysis, data curation, and visualization. S.J.M. contributed to the data curation and provision of software and resources. C.H., B.E.K.K., K.E.L., G.L.K., M.R., T.C., and R.G.M. contributed to the data curation and provision of resources. All authors made important contributions to the manuscript revision and approved the final version of the manuscript for publication. S.B.B., P.M.M., A.D.P., and H.M.C. contributed to the conceptualization of the study, methodology, and funding acquisition. P.M.M., A.D.P., and H.M.C. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in oral form at the 58th Annual Meeting of European Association for the Study of Diabetes, Stockholm, Sweden, 20–23 September 2022, and in poster form at the American Society of Human Genetics Annual Meeting, Los Angeles, CA, 25–29 October 2022.