Meta-genome-wide association studies identify a locus on chromosome 1 and multiple variants in the MHC region for serum C-peptide in type 1 diabetes

Aims/hypothesis The aim of this study was to identify genetic variants associated with beta cell function in type 1 diabetes, as measured by serum C-peptide levels, through meta-genome-wide association studies (meta-GWAS). Methods We performed a meta-GWAS to combine the results from five studies in type 1 diabetes with cross-sectionally measured stimulated, fasting or random C-peptide levels, including 3479 European participants. The p values across studies were combined, taking into account sample size and direction of effect. We also performed separate meta-GWAS for stimulated (n = 1303), fasting (n = 2019) and random (n = 1497) C-peptide levels. Results In the meta-GWAS for stimulated/fasting/random C-peptide levels, a SNP on chromosome 1, rs559047 (Chr1:238753916, T>A, minor allele frequency [MAF] 0.24–0.26), was associated with C-peptide (p = 4.13 × 10−8), meeting the genome-wide significance threshold (p < 5 × 10−8). In the same meta-GWAS, a locus in the MHC region (rs9260151) was close to the genome-wide significance threshold (Chr6:29911030, C>T, MAF 0.07–0.10, p = 8.43 × 10−8). In the stimulated C-peptide meta-GWAS, rs61211515 (Chr6:30100975, T/–, MAF 0.17–0.19) in the MHC region was associated with stimulated C-peptide (β [SE] = − 0.39 [0.07], p = 9.72 × 10−8). rs61211515 was also associated with the rate of stimulated C-peptide decline over time in a subset of individuals (n = 258) with annual repeated measures for up to 6 years (p = 0.02). In the meta-GWAS of random C-peptide, another MHC region, SNP rs3135002 (Chr6:32668439, C>A, MAF 0.02–0.06), was associated with C-peptide (p = 3.49 × 10−8). Conditional analyses suggested that the three identified variants in the MHC region were independent of each other. rs9260151 and rs3135002 have been associated with type 1 diabetes, whereas rs559047 and rs61211515 have not been associated with a risk of developing type 1 diabetes. Conclusions/interpretation We identified a locus on chromosome 1 and multiple variants in the MHC region, at least some of which were distinct from type 1 diabetes risk loci, that were associated with C-peptide, suggesting partly non-overlapping mechanisms for the development and progression of type 1 diabetes. These associations need to be validated in independent populations. Further investigations could provide insights into mechanisms of beta cell loss and opportunities to preserve beta cell function. Electronic supplementary material The online version of this article (10.1007/s00125-018-4555-9) contains peer-reviewed but unedited supplementary material, which is available to authorised users.

close to the genome-wide significance threshold (Chr6:29911030, C>T, MAF 0.07-0.10, p = 8.43 × 10 −8 ). In the stimulated Cpeptide meta-GWAS, rs61211515 (Chr6:30100975, T/-, MAF 0. 17-0.19) in the MHC region was associated with stimulated Cpeptide (β [SE] = − 0.39 [0.07], p = 9.72 × 10 −8 ). rs61211515 was also associated with the rate of stimulated C-peptide decline over time in a subset of individuals (n = 258) with annual repeated measures for up to 6 years (p = 0.02). In the meta-GWAS of random C-peptide, another MHC region, SNP rs3135002 (Chr6:32668439, C>A, MAF 0.02-0.06), was associated with Cpeptide (p = 3.49 × 10 −8 ). Conditional analyses suggested that the three identified variants in the MHC region were independent of each other. rs9260151 and rs3135002 have been associated with type 1 diabetes, whereas rs559047 and rs61211515 have not been associated with a risk of developing type 1 diabetes. Conclusions/interpretation We identified a locus on chromosome 1 and multiple variants in the MHC region, at least some of which were distinct from type 1 diabetes risk loci, that were associated with C-peptide, suggesting partly non-overlapping mechanisms for the development and progression of type 1 diabetes. These associations need to be validated in independent populations. Further investigations could provide insights into mechanisms of beta cell loss and opportunities to preserve beta cell function.
Keywords C-peptide . Genome-wide association study . Insulin-secreting cells . Single nucleotide polymorphism . Type

Introduction
Type 1 diabetes develops when pancreatic beta cells fail to produce sufficient insulin to maintain euglycaemia [1]. Cpeptide measurement after a standard meal (stimulated Cpeptide) is the primary way to evaluate insulin secretion in type 1 diabetes [2,3]. However, fasting/random C-peptide measurements are less demanding and feasible to obtain in large population samples. The alpha and beta chains of the proinsulin molecule are joined by C-peptide, which is removed during insulin secretion. Therefore, C-peptide is co-secreted with insulin in an equimolar ratio. Unlike insulin, C-peptide undergoes little first-pass clearance by the liver, has a much longer half-life and is not interfered with by exogenous insulin [2,3].
Type 1 diabetes is a heterogeneous disease that can occur at any age. C-peptide levels vary at diagnosis. Early-onset type 1 diabetes (age <2-5 years) is associated with rapid beta cell loss, the presence of high-risk MHC type 1 diabetes genetic susceptibility and high titres of autoantibodies compared with late-onset disease [4]. A large proportion of people with type 1 diabetes have detectable C-peptide years after diagnosis, indicating that some beta cells are still functional [2,[5][6][7][8]. Sensitive assays have detected stimulated C-peptide in 73-100% of people with type 1 diabetes for ≥30 years [7,8].
C-peptide preservation has been reported to be associated with favourable metabolic and clinical outcomes in type 1 diabetes. Individuals with preserved C-peptide levels maintain better long-term glycaemic control despite a lower exogenous insulin dose per body weight [3]. They are also at lower risk for both hypoglycaemia and diabetic complications, which are major health problems [3,[9][10][11][12]. The association of C-peptide with long-term diabetic complications (i.e. retinopathy and nephropathy) remains significant after adjusting for long-term glycaemic control, suggesting that at least part of the effect of C-peptide on diabetic complications is independent of its effect on glycaemic control (or, alternatively, that frequency of blood glucose/HbA 1c measurements is suboptimal) [3,9]. However, association of C-peptide levels with long-term diabetic complications has not been observed in other studies [13].
There have been no family/twin studies of C-peptide levels in type 1 diabetes. In the Diabetes Control and Complications Trial (DCCT), investigations of detectable stimulated Cpeptide levels in participants with type 1 diabetes (n = 98) and their first-degree affected relatives (n = 109) could not reach conclusions because of an insufficient sample size (see electronic supplementary material [ESM] Tables 1, 2). To date, more than 50 type 1 diabetes risk loci have been identified. The majority are thought to be involved in the immune system, and many are shared with other autoimmune diseases [14]. However, there have been no genome-wide association studies (GWAS) of C-peptide in type 1 diabetes (www.ebi.ac. uk/gwas, accessed 1 July, 2017) and candidate geneassociation studies have investigated a limited number of genes in small populations with no replication in independent studies [15]. It is clear from other diseases that factors for the risk of a disease and its progression can be independent [16]. Here, we aimed to identify genetic loci associated with residual beta cell function in type 1 diabetes as measured by Cpeptide through meta-analysis of GWAS.

Study design
We performed meta-GWAS of stimulated, fasting and random (non-fasting) C-peptide separately in European (confirmed by population structure analysis) participants from the following five type 1 diabetes studies ( Fig. 1 In the analysis of stimulated C-peptide, DCCT participants were divided into three groups (a primary cohort and a secondary cohort with diabetes duration of 1-5 and 5-15 years), since the range of stimulated C-peptide levels differed among the three groups based on the inclusion/exclusion criteria. In contrast, because all DCCT participants had fasting C-peptide levels of ≤0.2 pmol/ml, the fasting C-peptide analysis was performed with all participants as a single group ('all').
Another meta-analysis was performed to combine the GWAS results for stimulated (DCCT), fasting (CACTI and EDC) and random (Joslin 50-Year Medalist and WESDR) C-peptide (Fig. 1). We specifically examined the associations of known type 1 diabetes loci and type 1 diabetes genetic risk scores (GRS) with C-peptide levels [14,17]. The associations of the identified loci and type 1 diabetes loci/GRS were tested with stimulated C-peptide change over time in a subset of DCCT participants with repeated annual measures for up to 6 years (n = 258).

Studies
DCCT Individuals aged 13-39 years with type 1 diabetes and fasting serum C-peptide levels ≤0.2 pmol/ml were recruited between 1983 and 1989. The primary cohort included participants with 1-5 years of diabetes, no pre-existing retinopathy, normal albuminuria (<40 mg/24 h) and stimulated C-peptide ≤0.5 pmol/ml. The secondary cohort included participants with 1-15 years of diabetes, pre-existing mild retinopathy, albumin excretion rate <200 mg/24 h and stimulated Cpeptide ≤0.5 and 0.2 pmol/ml for participants with diabetes duration 1-5 and 5-15 years, respectively [ESM Table 3]. Participants were randomly assigned to receive intensive or conventional insulin treatment and were monitored until 1993 [18]. C-peptide levels were measured annually for up to 6 years in those with stimulated C-peptide >0.2 pmol/ml at eligibility testing [19,20].
CACTI CACTI was an observational, population-based study of individuals with type 1 diabetes of duration >10 years at enrolment. Participants were recruited from Denver, CO, USA, between March 2000 and April 2002 [21]. Fasting serum Cpeptide was measured during 2014-2016 [22] ESM Table 3].
EDC This study recruited a cohort of individuals with incident, childhood-onset (<17 years of age) type 1 diabetes who were diagnosed or seen within 1 year of diagnosis  at the Children's Hospital of Pittsburgh. The first participant assessments occurred in 1986-1988 [23]. Serum C-peptide was measured at follow-up year 25 [ESM Table 3]. Participants were advised to fast for ≥8 h, but 17% (n = 26) did not. Of the participants who failed to fast, only 15% (n = 4) had detectable C-peptide (unpublished data, T. J. Orchard and T. Costacou).
Joslin 50-year Medalist study This was a cross-sectional study. All participants provided their original medical record or three forms of documentation of ≥50 years of insulin-dependent diabetes and also resided in the USA at the time of participation [24]. Random serum C-peptide levels were measured and validated at the Northwest Lipid Research Laboratory at the University of Washington [ESM Table 3] [13].
WESDR People with diabetes who received primary care in an 11-county area in southern Wisconsin from 1979 to 1980 were recruited and subsequently followed for diabetic complications [25,26]. The participants were all younger-onset individuals who were diagnosed before 32 years of age and treated with insulin. Random serum C-peptide levels were measured at years 4 and 10 of follow-up [ESM Table 3] [27], with the most recent results used for analysis.
Genotype dosage data from Illumina 1M BeadArrays were used to analyse stimulated C-peptide in the DCCT, as approximately 3 million more SNPs were imputed with good quality (INFO ≥0.8); and genotype dosage data from HumanCoreExome BeadArrays were used for the remaining analyses in all studies. SNPs with minor allele frequency (MAF) >0.01 and high imputation quality (INFO >0.8 and R 2 > 0.5 from IMPUTE2 and Minimac, respectively; see ESM Methods for software details) were included in the analysis.

Fig. 1
Overview of the C-peptide cross-sectional data analysis and the main findings. The dashed arrows show the studies included in the meta-GWAS of stimulated, fasting and random C-peptide. HCE, human core exome GWAS and meta-GWAS of cross-sectional C-peptide SNPs were tested for association with C-peptide (natural log transformed) under an additive genetic model using Tobit models to appropriately model the large number of participants with values at the lower limit of detection (ESM Table 3) [30]. The Tobit model beta coefficient can be interpreted as the relationship between the genotype and the uncensored latent (unobserved) C-peptide variable. VGLM models from the VGAM library v0.9-7 were fitted using R v3.1.0 [31]. Sex, age at diagnosis and diabetes duration were included as covariates in the model. For fasting C-peptide analysis in the DCCT, the cohort (primary vs secondary) was also included as a covariate in the model.
Meta-GWAS analysis of stimulated C-peptide was performed using METAL v1.5 (www.sph.umich.edu/csg/ abecasis/Metal/index.html) with the STDERR method, which weights effect-size estimates by the inverse of the corresponding SEs. However, because C-peptide was measured using different assays in the different studies, and effect-size estimates and standard errors were in different units, the other meta-GWAS analyses used the SAMPLESIZE method, which converts the SNP association test p values to Z scores using the direction of effect, and combines Z scores across studies weighting by the study-specific sample size. Between-study heterogeneity was tested using Cochran's Q statistic. I 2 , which depends on Q, is used to quantity the level of heterogeneity among studies [32].
Testing for the heterogeneity of the SNP effect between stimulated and fasting C-peptide To investigate whether the SNP associations differed between stimulated and fasting C-peptide in the DCCT, we used the Tobit model implementation in the Zelig v 4.2-1 library in R v3.1.0 [31] to account for within-individual correlations of the two C-peptide measurements. Sex, age at diagnosis, diabetes duration, cohort (primary vs secondary) and an indicator for fasting/stimulated C-peptide were included as covariates in the models. Heterogeneity was tested using an interaction between the fasting/stimulated indicator and SNP.

Type 1 diabetes loci
We investigated two sets of type 1 diabetes loci derived from Oram et al [17] and Onengut-Gumuscu et al [14]. The study of Oram et al included robustly associated genetic variants in both MHC and non-MHC regions from published studies (n = 30 SNPs) [33][34][35][36], whereas Onengut-Gumuscu et al included only non-MHC regions and was based on Immunochip data in >9000 participants with type 1 diabetes and >12,000 non-diabetic control participants with multiple independent variants in some loci (n = 51 SNPs) (see ESM Methods for details).
We stratified SNPs according to the presence (n = 70 SNPs) or absence of a prior association with type 1 diabetes, and performed false discovery rate (FDR) control separately for each stratum using the stratified (S)FDR method [37,38].
Each of the classical HLA alleles, their amino acid sequences and SNPs in the MHC region were tested for associations with C-peptide, and combined through meta-analysis (as described above).

Longitudinal stimulated C-peptide in the DCCT
Associations of identified SNPs with repeated measures of stimulated C-peptide (natural log transformed) were tested under an additive genetic model using linear mixed models (random slope and random intercept) and first-order autoregressive AR(1) correlation structure in R v3.1.0 [31]. SNP, sex, age at diagnosis, diabetes duration at baseline, treatment group (intensive vs conventional), cohort (primary vs secondary), an indicator for baseline (vs later assessment), time (DCCT follow-up year), (centred time) 2 ; and interactions between time and treatment, treatment and baseline, and SNP and time were included in the model. Differences among SNP genotypes in the rate of C-peptide decline over time were investigated by hypothesis testing of SNP and time interaction. 1q53 In the stimulated/fasting/random C-peptide meta-GWAS, rs559047 on 1q53 (Chr1: 238753916, T>A) exceeded the genome-wide significance threshold (p < 5 × 10 −8 )( Fig. 1; ESM Figs 3, 4). The A allele was associated with lower C-peptide (p = 4.13 × 10 −8 ). The direction of the rs559047 association with C-peptide was consistent in all studies (Table 1, Fig. 2).

Meta-GWAS
At DCCT eligibility, fasting and stimulated C-peptide levels were highly positively correlated (Spearman correlation coefficient 0.88, p < 0.0001). rs559047 was associated with both stimulated and fasting C-peptide at DCCT eligibility, but there was a significant difference in the association between stimulated and fasting C-peptide (p = 0.010), with a larger effect on stimulated compared with fasting C-peptide (−0.28 vs −0.13). The association of rs559047 with stimulated C-peptide remained significant after including fasting Cpeptide in the model but not vice versa, suggesting that the rs559047 association with fasting C-peptide is indirect and due to its association with stimulated C-peptide (Table 2).
rs559047 was not associated with the rate of decline in stimulated C-peptide over time using longitudinal DCCT data. Characteristics of the participants included in the DCCT longitudinal data analysis are summarised in ESM Table 6.
MHC region Multiple variants in the MHC region were associated with different C-peptide measures. In the stimulated/ fasting/random C-peptide meta-GWAS, two loci in the MHC region, rs3135002 (Chr6:32668439) and rs9260151 (Chr6:29911030), approximately 2.8 Mb apart from each other, were close to genome-wide significance. A single nucleotide deletion in the MHC region, rs61211515 (Chr6:30100975), had the lowest p value in the stimulated C-peptide meta-GWAS ( Fig. 1; ESM Figs 3, 4). Detailed results for these three loci follow. 15], p = 0.077), the effect was in the same direction; further, there was no significant difference between the fasting and stimulated Cpeptide associations (p = 0.46). rs3135002 was not associated with fasting C-peptide in the CACTI or EDC studies, and nor was it associated with the rate of stimulated C-peptide decline over time in the DCCT (p = 0.62) (Table 3, Fig. 3). rs9260151 rs9260151 (C>T) is an intronic SNP within HLA-A. Its T allele was associated with higher C-peptide with a p value close to the genome-wide significance threshold in the stimulated/fasting/random C-peptide meta-GWAS (p = 8.43 × 10 −8 ). The direction of the effect was consistent in all studies (Table 3, Fig. 3). rs9260151 was associated with both stimulated and fasting C-peptide at DCCT eligibility but the effects were significantly different (p = 0.024), with a larger effect on stimulated compared with fasting C-peptide (0.34 vs 0.20). Its association with stimulated C-peptide remained significant after including fasting C-peptide in the model, but not vice versa (Table 2). rs9260151 was not associated with differences in the rate of decline in stimulated C-peptide over time using longitudinal DCCT data.  Table 7. rs61211515 is a mononucleotide T repeat (n = 12 Ts in RefSeq (https://www.ncbi.nlm.nih.gov/refseq/, accessed 1 July, 2017)). An insertion allele at the same locus (T/TT) was not associated with stimulated, fasting or random Cpeptide; and comparing five genotype categories with a homozygous reference, only categories with deletion had significant effects (ESM Tables 8, 9).
Independence of the MHC signals rs9260151, rs3135002 and rs61211515 are in linkage disequilibrium (LD) (ESM Table 10). However, putting them in the same model, all three remained significantly associated with stimulated and fasting C-peptide at DCCT eligibility (ESM Table 11). The likelihood ratio test comparing a model with no SNPs (only covariates in the model and stimulated C-peptide as the outcome) to a model with all three SNPs (rs61211515, rs9260151, rs3135002 and covariates) was strongly significant (p = 5.67 × 10 −10 ). Alleles are non-effect allele > effect allele Associations were tested using QLIM procedure in SAS v9.4 (SAS, Cary, NC, USA), and are based on extracted genotypes from dosage data according to the best guess using GTOOL v0.7.5 (www.well. ox.ac.uk/~cfreeman/software/gwas/gtool.html) adjusting for sex, age at diagnosis, type 1 diabetes duration and cohort (primary vs secondary) β, β coefficient; F, fasting; S, stimulated   (a) Stimulated C-peptide meta-GWAS, which includes the DCCT primary cohort, the secondary cohort with diabetes duration 1-5 years and the secondary cohort with diabetes duration 5-15 years. Since rs61211515 (T/-) is an insertion/deletion (in/del) second-to-top hit, rs2074477, in high LD with rs61211515 (r 2 0.89; D′ 0.98) was used by LocusZoom for LD measures. (b) Fasting C-peptide meta-GWAS, which includes the DCCT, CACTI and EDC studies. (c) Random C-peptide meta-GWAS, which includes the Joslin 50-Year Medalist study and WESDR. (d) Stimulated/fasting/random C-peptide meta-GWAS, which includes stimulated C-peptide results from the DCCT primary cohort, the secondary cohort with diabetes duration 1-5 years and the secondary cohort with diabetes duration 5-15 years; fasting C-peptide results from the CACTI and EDC studies; and random C-peptide results from the Joslin 50-Year Medalist study and WESDR. Left y-axis: SNP p values; right y-axis: estimated recombination rates; x-axis: SNP genomic position (Genome Reference Consortium Human Build 37 [GRCh37]/hg19); the LD measures are based on the 1000 Genomes November 2014 EUR population. The plot was made using LocusZoom (http://locuszoom. sph.umich.edu/locuszoom/) [44]. Chr, chromosome; Rec, recombination MHC imputation A total of 424 classical HLA alleles were imputed: 279 were polymorphic, and of them 252 had good imputation quality (R 2 > 0.5) in the DCCT. Similarly, 1136/1276 amino acid changes were polymorphic, and 1106 met imputation quality criteria. A total of 7144/7261 SNPs were polymorphic and 7066 were imputed well (ESM Table 12). None of the HLA classical alleles or amino acid changes reached the genome-wide significant threshold in the meta-GWAS.
Association of type 1 diabetes loci and C-peptide  Table 13.
The SFDR method estimated the proportion of null SNPs (π 0 ) at 0.91 in the prior type 1 diabetes stratum vs 0.99 in the stratum of the remaining SNPs. Setting an FDR control level of 20% yielded four test rejections for prior type 1 diabetes SNPs and no test rejections in the remaining SNPs. The FDR q value estimates were 0.001 for both rs1264813 and rs151234, and 0.17 for both rs12971201 and rs689. SFDR ranked these four SNPs above rs61211515, the top associated SNP in the stimulated C-peptide meta-GWAS.
The association of rs1264813 (HLA-A*24) with stimulated C-peptide at DCCT eligibility did not remain significant when rs61211515 (the top GWAS-associated SNP in this region) was included in the model (p = 0.45). Similar results were observed with imputed HLA-A*24; it was associated with lower stimulated C-peptide at DCCT eligibility (β [SE] = −0.39 [0.09], p = 1.50 × 10 −5 ), but this association became insignificant after adjustment for rs61211515 (p = 0.09) (ESM Table 14). rs1264813 (HLA-A*24) was also nominally associated with fasting Cpeptide at DCCT eligibility (β [SE] = −0.19 [0.07], p = 0.013) but its association with stimulated C-peptide was significantly stronger (p = 0.011). The rs1264813 association with stimulated C-peptide remained significant when fasting C-peptide was included in the model but not vice versa (Table 2). rs1264813 was also associated with an increased rate of decline in stimu-  Table 15).
HLA-DQB1 position 57, HLA-DRB1 position 13 or HLA-DRB1 position 71 amino acid variants, which have been reported to independently drive type 1 diabetes risk and explain >90% of the type 1 diabetes-HLA association [41], were not associated with stimulated C-peptide at DCCT eligibility.

Discussion
By combining GWAS data from five type 1 diabetes studies including approximately 1300 individuals with stimulated and approximately 3500 individuals with stimulated, fasting or random C-peptide data, we identified association with a locus on chromosome 1 (rs559047) and multiple independent variants in the MHC region.
rs559047 is an intergenic SNP located 700 kb from the nearest gene (ZP4). It is not a significant cis-expression quantitative trait locus (cis-eQTL) for nearby genes in any tissue, including the pancreas (www.gtexportal.org, accessed 14 July 2017) or pancreatic islets (http://theparkerlab.org/tools/isleteqtl, accessed 14 July 2017). However, the cis-eQTLs were produced using a limited number of samples (149 and 112 for pancreas and islets, respectively) from non-diabetic individuals. rs559047 has not been associated with the risk of type 1 diabetes (ESM Table 16), and variants in this region have not previously been associated with type 1 diabetes related phenotypes.
We identified three SNPs (rs9260151, rs61211515 and rs3135002) in the MHC region associated with C-peptide levels. Of these, rs9260151 and rs3135002 have been associated with risk of type 1 diabetes (ESM Table 16). rs9260151 is an intronic SNP within HLA-A, while rs61211515 is located approximately 187 kb downstream of HLA-A. Both SNPs appear to be primarily associated with stimulated C-peptide although rs9260151 was identified in the meta-GWAS of stimulated/fasting/random C-peptide, likely due to the larger available sample size. HLA-A*24 is a known risk variant for type 1 diabetes [34,42]. However, after including rs9260151, rs61211515 and HLA-A*24 (either imputed or tagging SNP) in the model, rs9260151 and rs61211515 both remained significantly associated with C-peptide, indicating that they are independent signals and distinct from HLA-A*24. The association between HLA-A*24 and C-peptide was no longer significant, suggesting that this association is due to LD between HLA-A*24 and rs9260151 and/or rs61211515 (ESM Table 10). Similarly, the association of rs9260151 with type 1 diabetes could be explained by its LD with HLA-A*24. rs61211515 and HLA-A*24, which were associated with lower C-peptide levels, were also associated with an accelerated decline in C-peptide levels over time, providing more evidence that this association is a true finding. HLA-A*24 has previously been associated with undetectable Cpeptide in a Japanese type 1 diabetes population sample [15] and with more rapid beta cell destruction post islet transplantation [43]. The third SNP, rs3135002, is intergenic between HLA-DQB1 and DQA2, 2.5 Mb from rs9260151 and rs61211515. Its association with C-peptide remained significant when both rs9260151 and rs61211515 were included in the model, indicating that it is independent. DR3 and DR4-DQ8 were not associated with C-peptide and the borderline association observed between DR3/DR4-DQ8 compound heterozygotes and Cpeptide did not withstand multiple testing correction, suggesting that this effect is also distinct from the DR3/DR4-DQ8 haplotype, consistent with previous findings [15]. The association of rs3135002 with type 1 diabetes could be explained by the extremely large effect of DR3/DR4-DQ8 on type 1 diabetes and the LD between rs3135002 and both DR3 (D′ = 1, r 2 = 0.012) and DR4-DQ8 (D′ = 0.87, r 2 = 0.010). The rs3135002 association was mainly with fasting/random C-peptide rather than with stimulated C-peptide. No significant interactions were observed between the three SNPs in the MHC affecting stimulated/fasting C-peptide (data not shown), and none of them was a significant cis-eQTL for the nearby genes in any tissue including the pancreas and pancreatic islets (http://www.gtexportal.org, accessed July 14, 2017; http://theparkerlab.org/tools/isleteqtl, accessed July 14, 2017).
Some SNP associations for C-peptide in type 1 diabetes are different from those for type 1 diabetes. Two of the identified variants (rs559047 on chromosome 1 and rs61211515 in the MHC region) have not been previously associated with type 1 diabetes; while rs9260151 and rs3135002 in the MHC region have been associated with type 1 diabetes. The variants associated with C-peptide but not with type 1 diabetes may affect gene expression, and fine-mapping and functional studies are required to determine their effects. However, SFDR analysis suggests that the proportion of type 1 diabetes loci associated with C-peptide is greater than for the rest of the genome; four out of 70 type 1 diabetes loci (HLA-A*24, IL27, INS and PTPN2) that were nominally associated with lower C-peptide levels ranked higher than rs61211515, the top associated SNP in the stimulated C-peptide meta-GWAS. Nevertheless, the composition and magnitude of SNP effects for type 1 diabetes development vs progression could differ, explaining why there is no significant association between type 1 diabetes GRS and C-peptide.
Some limitations include that only fasting/random C-peptide measurements were available for about 60% of individuals, whereas stimulated C-peptide is more appropriate to evaluate insulin secretion in type 1 diabetes. In addition, individuals from different cohorts were quite different in terms of their age at diagnosis, diabetes duration and other inclusion/exclusion criteria, all of which may influence C-peptide levels. C-peptide can also be measured with different methods with different lower limits of detection, and these were not standardised among the studies. The mean age at diagnosis ranged from 8.3 years in the EDC study to 21.2 years in the DCCT, and the mean type 1 diabetes duration ranged from 5.6 years in the DCCT to 54.7 years in the Joslin 50-Year Medalist study. Participants from the DCCT, the only study with stimulated C-peptide data, were highly selected as they were diagnosed with type 1 diabetes later in life and had diabetes for a shorter period of time. Therefore, the DCCT had a larger proportion of participants with detectable Cpeptide compared with the other studies. This heterogeneity among the cohorts might have reduced the power of the analysis and contributed to non-replication at specific loci.
In conclusion, we have identified a locus on chromosome 1 and multiple variants in the MHC region that are associated with C-peptide levels. However, these associations need to be validated in independent populations. Further investigations could eventually provide insight into mechanisms of beta cell loss and opportunities to preserve beta cell function.
contributions from Mr. T. Beatson (Beatson Foundation, Phoenix, AZ, USA) and the many participants. Support for CACTI was provided by National Heart, Lung, and Blood Institute grants R01 HL113029, HL61753, HL79611, and HL113029, DERC Clinical Investigation Core P30 DK57516 and JDRF grant 17-2013-313. The study was performed at the Adult CTRC at University of Colorado Denver, CO, USA supported by NIH-M01-RR00051 and CTSA Grant UL1 TR001082, at the Barbara Davis Center for Childhood Diabetes and at the Colorado Heart Imaging Center (Denver, CO, USA). JKS-B was supported by an ADA Career Development Award (7-13-CD-50).
The WESDR was supported by grant R01EY016379 and NIH R01EY03083 from the National Eye Institute, NIH, and an unrestricted grant from Research to Prevent Blindness (New York, NY, USA).
The EDC study was supported by NIH grant R01-DK34818 and the Rossi Memorial Fund.
Funding for genotyping by Illumina 1M in DCCT was provided by NIH R01DK-077510 and genotyping by HumanCoreExome in all studies was provided by JDRF grant 17-2013-9.
DR was partly supported by a fellowship award from the Canadian Institutes of Health Research Strategic Training for Advanced Genetic Epidemiology (STAGE).
The study sponsor was not involved in the design of the study; the collection, analysis, and interpretation of data; writing the report; or the decision to submit the report for publication.