Association of the Vitamin D Metabolism Gene CYP27B1 With Type 1 Diabetes

We studied a case-control collection comprising 7,854 patients with type 1 diabetes and 8,758 healthy control subjects from the U.K. The recruitment of these subjects and sample processing have been described elsewhere (15). We also studied CYP27B1 polymorphisms in a family collection including 2,774 type 1 diabetes families with one or two affected offspring (815 from the U.K. and Northern Ireland, 841 from Finland, 335 from the U.S., 360 from Norway, and 423 from Romania), providing 3,081 parent-child trio genotypes for CYP27B1 −1260 and 2,198 trio genotypes for CYP27B1 +2838. The collection of all DNA samples has been approved by relevant ethical committees. We obtained written informed consent from all participants.

In the CYP27B1 gene, we genotyped three single nucleotide polymorphisms (SNPs), CYP27B1 −1260C>A (rs10877012, located in the 5′ region) and CYP27B1 +2838T>C (rs4646536, located in intron 6), which were previously reported (11–14), and rs8176345, a synonymous SNP in exon 5 that we found by sequencing. We used HapMap data (16) to select tag SNPs that capture common variants in the CYP24A1 gene. Of the 111 HapMap SNPs located in the region (National Center for Biotechnology Information [NCBI] build 34, coordinates chromosome 20: 53,450,894.0.53,482,103), 54 SNPs had minor allele frequency (MAF) >0.05, and 16 were chosen as tag SNPs that capture association of other common variants with r² > 0.8. CYP24A1 SNPs were genotyped in up to 5,239 case and 5,539 control subjects (exact numbers for each SNP are shown in Table 3). Genotyping was done using TaqMan (Assay-by-design; Applied Biosystems, Warrington, U.K.; see the online appendix [available at http://dx.doi.org/10.2337/db07–0652]). All genotypes were scored by two researchers independently to minimize error. Genotypes of control subjects and parents did not deviate from Hardy-Weinberg equilibrium above that expected at random (P > 0.05).

Direct sequencing of nested PCR products from 32 healthy control subjects from the U.K. was performed using an Applied Biosystems 3700 capillary sequencer (Foster City, CA). Polymorphisms were identified using the Staden Package (http://www.mrc-lmb.cam.ac.uk/pubseq/) and mapped to the NCBI human genome build 35.

All statistical analyses were performed within Stata statistical package (http://www.stata.com), using additional Stata routines (http://www-gene.cimr.cam.ac.uk/clayton/software/). We analyzed case and control subjects using logistic regression models (17), adjusting for 12 broad geographical regions, to allow for geographic variation in allele frequencies across the U.K. (18). The families were analyzed using the transmission disequilibrium test (19) and conditional logistic regression (17). A score test was used to combine tests from family and case-control studies as described previously (15). We used htstep, htsearch, and haptag programs within Stata 8.2 to select tag SNPs in the CYP24A1 gene. For these SNPs, we performed a multilocus test using mlpop program in Stata 8.2, which tests for association between disease and the tag SNPs due to linkage disequilibrium with one or more causal variants in the region. It contrasts allele frequencies of a nonredundant set of tag SNPs between case and control subjects by use of Hotelling's t² test (20,21). We did not apply correction for multiple testing.

We found evidence that the promoter polymorphism CYP27B1 −1260 is associated with type 1 diabetes in both the case-control (P = 9.6 × 10⁻⁴; C/C genotype, odds ratio [OR] 1.22 [95% CI 1.10–1.36]; Table 1) and the family (P = 3.9 × 10⁻³; C/C genotype, relative risk [RR] 1.33 [95% CI 1.12–1.58]; Table 1) collections. Consequently, when we combined evidence from both collections, which are fully independent from each other, the combined test provided statistical support for an association between type 1 diabetes and CYP27B1 −1260 (P = 3.8 × 10⁻⁶ for the 2 degree of freedom [df] genotype model, see Table 1 legend). There was evidence of population heterogeneity in the parent allele frequencies of CYP27B1 −1260 (F_{3, 3486} = 3.44, P = 0.016) but no evidence for heterogeneity in the disease RR between populations above that expected at random (χ₆² = 3.11, P = 0.79). We found no evidence of regional heterogeneity in the control allele frequencies (F_{11, 7261} = 0.86, P = 0.58).

In contrast to other previously published studies (11–14), we found that intronic SNP CYP27B1 +2838 was also associated with type 1 diabetes in both collections. The major allele T was associated with increased type 1 diabetes risk in both the case-control (P = 0.010; T/T genotype, OR 1.20 [95% CI 1.07–1.36]; Table 2) and the family (P = 6.1 × 10⁻⁴; T/T genotype, RR 1.36 [1.11–1.67]; Table 2) collections. The combined P value was 8.5 × 10⁻⁵ (2-df genotype model, see Table 2 legend).

We noted that in all population samples that we studied, including control subjects from U.K. and parents of the patients from U.K. and Northern Ireland, Norway, or Romania, there is almost perfect linkage disequilibrium between SNPs CYP27B1 −1260 and +2838 with D′ = 1.0 and r² = 0.99 (we obtained lower P values for CYP27B1 −1260 because more samples were genotyped for this SNP than for +2838). To verify genotyping of CYP27B1 −1260 and +2838, we directly sequenced 376 case subjects and 533 control subjects and found complete concordance in the results. This raised the possibility that in the German and Polish population samples studied previously (11–14), there may have been genotyping error in the analysis of CYP27B1 −1260 polymorphism. Therefore, in Cambridge, we re-genotyped 120 DNA samples from 36 type 1 diabetes families from the original German laboratory for the two SNPs, obtaining only 88.2% concordance between the two genotype datasets for CYP27B1 −1260, and this problem was compounded by evidence of data handling errors. Contrary to previous analyses (11,12,14), in these German samples, we found the most perfect linkage disequilibrium between the two SNPs (CYP27B1 −1260 and +2838 SNPs: D′ = 1.00 and r² = 0.99) as we report here for all other populations studied, indicative of past genotyping and data analysis errors.

We then resequenced 8 kb of the CYP27B1 gene, including all exons, introns, and 2 kb 5′ and 3′ of the gene, using DNA samples of 32 healthy subjects from U.K. to test for the presence of an obvious candidate for a causal variant, such as an amino acid–changing polymorphism or a splice mutation. We discovered two novel rare SNPs with MAFs <0.01, one in the promoter at position −1138 and one in the 3′-untranslated region (ss67078180 and ss67078183, respectively; http://www.ncbi.nlm.nih.gov/SNP/). We did not genotype these SNPs because even large samples that we studied here were underpowered to demonstrate association of such rare variants. We also found a synonymous SNP rs8176345 in exon 5 with MAF = 0.03 that was not in linkage disequilibrium with the common CYP27B1 SNPs at positions −1260 and +2838 (r² = 0.06 and 0.06, respectively). We genotyped rs8176345 in a subset of the case-control collection comprising 3,040 type 1 diabetic patients and 3,349 control subjects but obtained no evidence of an association (P = 0.23; OR 0.87 [95% CI 0.71–1.09]). We also identified a common promoter SNP rs3782130 at position −1074 with MAF = 0.33. Because we were unable to develop a working high throughput genotyping assay for this SNP, we sequenced it in 376 case subjects and 533 control subjects and found that it was also in almost perfect linkage disequilibrium with SNPs at positions −1260 and +2838 (r² = 0.99 and 0.97, respectively).

We performed case-only gene-gene interaction tests (15) between known type 1 diabetes susceptibility loci and CYP27B1 −1260. We did not undertake the same analyses for CYP27B1 −2838 because these SNPs are in almost perfect linkage disequilibrium. There was no consistent evidence of an interaction (that is the deviation from a multiplicative model) for the joint effects of CYP27B1 −1260 and INS rs689 (−23HphI), PTPN22 rs2476601 (Arg620Trp), or CTLA4 rs3087243 (Supplementary Table 1). However, there was some evidence for an interaction with HLA-DRB1 (Supplementary Table 1), but when we analyzed CYP27B1 −1260 stratified by specific HLA-DRB1 genotypes, we found that risk ratios were not consistent between the case-control and family collections (Supplementary Table 2). Therefore, we conclude that in conferring risk of type 1 diabetes CYP27B1 does not interact with the previously known disease genes. We conducted a similar interaction analysis for CYP27B1 −1260 and seven VDR SNPs (FokI, ApaI, BsmI, TaqI, rs2544043, rs12721366, and rs4303288) (9,10). However, we found no evidence of an interaction (Supplementary Table 1). We also tested CYP27B1 −1260 for age-at-diagnosis and sex effects in a case-only analysis but did not find evidence for these (Supplementary Table 1) or for parent-of-origin effect in the affected families (P = 0.76).

In the case-control collection, we tested 16 tag SNPs that capture association of the common variants that were present in the CYP24A1 gene in HapMap (Table 3). A multilocus test revealed no evidence of association between CYP24A1 polymorphisms and type 1 diabetes (P = 0.23). Therefore, we did not undertake follow-up genotyping of any of the CYP24A1 polymorphisms in additional case and control subjects or families.

The present study provides the first evidence of association between CYP27B1 polymorphisms and type 1 diabetes in a fully validated analysis. Our results in the present report indicate what appears to have been technical and analytical errors in the previous studies (11–14). Nevertheless, these initial reports did contribute to our motivation to carry out the current analysis of CYP27B1 in type 1 diabetes.

Taking into account prior epidemiological and experimental links between vitamin D and type 1 diabetes (3–8,22–27) and the association between CYP27B1 and type 1 diabetes that we established here, we suggest that common inherited variation in the CYP27B1 gene affects vitamin D metabolism and is an etiological factor that predisposes type 1 diabetes. Rare CYP27B1 mutations that completely inactivate 1α-hydroxylase are known to cause vitamin D–dependent rickets type I (OMIM [Online Mendelian Inheritance in Man] no. 264700), characterized by low concentrations of 1α,25(OH)₂D (28,29). We hypothesize that the presence of the CYP27B1 −1260 C allele or another variant in linkage disequilibrium with it (such as two that we have studied here, CYP27B1 +2838 in intron 6 and rs3782130 in the 5′ region) reduces the level of the active 1α-hydroxylase and conversion of 25(OH)Dto 1α,25(OH)₂D, leading to increased predisposition to type 1 diabetes. Recently, preliminary data have suggested that type 1 diabetic patients carrying at CYP27B1 −1260 risk genotype CC had lower CYP27B1 mRNA levels in the peripheral blood mononuclear cells compared with healthy control subjects with the CC genotype (30). Functional roles of the CYP27B1 polymorphisms should be investigated in further experiments, evaluating their effects on 1α-hydroxylase activity and 1α,25(OH)₂D concentration, in particular, in the immune cells, such as dendritic cells and monocytes, that underpin immune responses (31,32).

Given our evidence that variation in the CYP27B1 gene etiologically contributes to type 1 diabetes risk, other genes that control vitamin D metabolism are also biologically plausible candidates, and studies of their association with type 1 diabetes are required. Here, we investigated the CYP24A1 gene that encodes vitamin D 24-hydroxylase, an enzyme that inactivates 1α,25(OH)₂D, and found no evidence of association. Studies of CYP27A1 or CYP2R1 that encode vitamin D 25-hydroxylases and of the vitamin D–binding protein gene (33,34) are also needed.

In the immune system, 1α,25(OH)₂D has been shown to suppress production of the interleukin (IL)-12, IL-2, tumor necrosis factor-α, and γ-interferon cytokines; to activate expression of transforming growth factor-β1 and IL-4 cytokines, thereby inhibiting Th1-type responses; and to induce regulatory T-cells (35). It can also regulate differentiation and maturation of dendritic cells critical in induction of T-cell–mediated immune responses (36). These immunomodulatory effects may explain the reported protective effects of vitamin D in type 1 diabetes (37). In the animal models, 1α,25(OH)₂D₃ and its analogs have been effective in prevention of autoimmune diabetes (23–27) and of other autoimmune diseases (38–42). Epidemiological studies in humans also indicate that intake of vitamin D and its high circulating levels are associated with a lower risk of rheumatoid arthritis, multiple sclerosis, and systemic lupus erythematosus (43–45). Genetic studies reported association of the CYP27B1 polymorphisms with Addison's disease, Hashimoto's thyroiditis, and Graves’ disease (12,13), but these results await confirmation. The possibility that CYP27B1 and 1α,25(OH)₂D may be involved in multiple autoimmune diseases suggests that effects of vitamin D on type 1 diabetes involve immune regulation, but this does not rule out additional effects, such as protection of pancreatic β-cells and their functions.

Our study indicates that genetic variation in the vitamin D metabolism is an etiological factor in type 1 diabetes. This evidence justifies further experiments investigating molecular and cellular actions of vitamin D and mechanisms of its protective effect in type 1 diabetes. Epidemiological studies indicate that vitamin D supplementation in early childhood may reduce type 1 diabetes risk (6–8). Given that vitamin D insufficiency is more common among children and young adults than was previously thought (46), its correction may be a viable approach to prevent type 1 diabetes or delay its development.

This work has received support from the Wellcome Trust and the Juvenile Diabetes Research Foundation International. E.H. is a Department of Health (U.K.) Public Health Career Scientist. E.R.-L. has received support from the European Foundation for the Study of Diabetes (EFSD). K.B. has received support from the EFSD. S.N. is a Diabetes Research and Wellness Foundation Non-Clinical Fellow.

We acknowledge the participation of all of the patients, control subjects, and family members. We thank the Human Biological Data Interchange and Diabetes U.K. for the U.S. and U.K. multiplex families, respectively; the Norwegian Study Group for Childhood Diabetes, D. Undlien, and K. Rønningen for the collection of the Norwegian families; C. Guja and C. Ionescu-Tirgoviste for the collection of the Romanian families; T. Siegmund for the collection of the German families; and B. Widmer for the collection of the British type 1 diabetic patients. We acknowledge use of DNA from the British 1958 Birth Cohort collection (R. Jones, S. Ring, W. McArdle, and M. Pembrey; funded by Medical Research Council Grant G0000934 and Wellcome Trust Grant 068545/Z/02), and we thank D. Strachan and P. Burton for their help. We thank S. Nutland the help in preparing DNA samples and C. Power for the advice on this manuscript.