No Evidence for Association of OAS1 With Type 1 Diabetes in Unaffected Siblings or Type 1 Diabetic Cases

To date, four loci have been identified with convincing and reproducible statistical support that predispose to type 1 diabetes development: the HLA class II genes (1), the insulin gene on chromosome 11p15 (2,3), the CTLA4 locus on 2q33 (4,5), and PTPN22 (6,7). Most recently, strong evidence for association of the IL2RA/CD25 locus with type 1 diabetes has been reported (P = 1.3 × 10⁻¹⁰) (8), but this finding requires independent replication. All of these genes are involved in T-cell activation, homeostasis, and repertoire formation.

Field et al. (9) recently reported that the A allele of an A/G splice-site single nucleotide polymorphism (SNP) (rs10774671) in the OAS1 gene, encoding 2′5′-oligoadenylate synthetase, had a type 1 diabetes protective effect in 401 unaffected siblings but no effect in 835 affected siblings. The 401 unaffected and 835 affected siblings were from 574 families comprising 83 Danish Families, 206 Danish diabetes-discordant sibpairs, 156 Canadian families, and 128 U.S. families from the Human Biological Data Interchange (9). Field et al. compared the A/G splice-site SNP genotype frequencies between affected and unaffected siblings using a χ² test and found differences between the genotype frequencies of diabetic and unaffected siblings (derived: χ²₂ = 8.31, P = 0.016; reported: GG+GA [high-risk] versus AA [low-risk] χ²₁ = 8.05, one-sided P = 0.0023). However, as the statistical test of Field et al. was invalid in that interdependence between sibling genotypes was ignored and the P value is not very small, this difference between diabetic and healthy siblings may be a false-positive. A more important caveat regarding the interpretation of their data arises from their results using the AFBAC (affected-family based controls) method (10) to compare the frequencies of transmitted and untransmitted alleles from parents to 368 diabetic siblings (one per family) and to 198 unaffected siblings; they found no evidence of an association between type 1 diabetes and the A/G splice-site SNP in the affected siblings (P = 0.27) but did find evidence of an increased frequency of the A allele in unaffected siblings (A allele frequency 0.71 in transmitted vs. 0.61 in nontransmitted alleles; P = 0.003) (9). Consequently, Field et al. concluded that the A allele was having a protective effect in the unaffected siblings. Our interpretation of this published data, given the biological and genetic implausibility of obtaining an allelic association with low penetrance in unaffected siblings with absence of significant deviation in transmission to diabetic siblings, was that the reported increase in A allele frequency in the unaffected siblings was a chance event.

To further investigate the potential role of the OAS1 A/G splice-site SNP in type 1 diabetes, we genotyped it in 455 U.K., 327 U.S. (which included ∼128 families previously analyzed by Field et al. [9]), 360 Norwegian, and 410 Romanian multiplex and simplex families, providing 1,913 parent-affected sibling trio genotypes and 856 parent-unaffected sibling trio genotypes. In addition, we genotyped 4,287 type 1 diabetic cases and 4,735 geographically matched control subjects from the U.K. (11).

We found no evidence for a protective effect in unaffected siblings: we observed random transmission of the A allele to unaffected sibs, using the transmission/disequilbrium test (TDT) (12) (369 transmissions vs. 375 nontransmissions, 49.6% transmission; P = 0.83) (Table 1). We found, consistent with Field et al. (9), no evidence for an association with type 1 diabetes in the family collections from four populations (Table 1) (relative risk [RR] for allele G = 1.00 [95% CI 0.91–1.10]; P = 0.96). In addition, we found no evidence for association between the A/G splice-site SNP and type 1 diabetes in the case-control collection (odds ratio for allele G = 0.99 [0.93–1.06]; P = 0.83) (Table 2). We found virtually identical genotype frequencies in the diabetic offspring and pseudo-control subjects (13) for the “high-risk GG+GA genotype” (56.7 and 56.7%, respectively), as well as in the case-control collection (58.5 and 58.9%, respectively) (Table 2).

More recently, Tessier et al. (14) reported marginally significant evidence for an association between type 1 diabetes and the OAS1 splice-site SNP, using the TDT in 784 families of mixed European descent, mainly French Canadian (RR for the G allele = 1.18 [95% CI 1.02–1.37]; P = 0.033). They did not test transmission to unaffected siblings. Tessier et al. also genotyped two nonsynonymous SNPs (nsSNPs) in OAS1, rs3741981 (C>T) and rs3177979 (A>G), obtaining an RR for allele C of 1.19 (1.03–1.37; P = 0.021) and an RR for allele G of 1.19 (1.03–1.38; P = 0.026), respectively. On examining haplotypes for the three SNPs, Tessier et al. found that transmission from C-A-A/T-A-A to diabetic siblings was overtransmitted in favor of C-A-A (P = 0.009). Consequently, they suggested that rs3741981 was the sole functional variant accounting for the genetic effect (14). They reported that maternal transmissions to affected offspring were increased for the three SNPs compared with transmissions from fathers.

We also genotyped the two nsSNPs from Tessier et al. (14) and found no evidence of association between rs3741981 or rs3177979 and type 1 diabetes in the family or case-control collections (supplementary Tables 1 and 2 [online appendix, available at http://diabetes.diabetesjournals.org]). We also tested for an association with the three–OAS1 SNP haplotype (rs10774671-rs3741981-rs3177979) and found no evidence of association (P = 0.28 and 0.86 for the case-control and family collections, respectively). The linkage disequilibrium between the SNPs was consistent with that reported by Tessier et al. (14); in control subjects, the D′ was 1.00 for each pair of SNPs, and the r² was 0.74 between rs10774671 and rs3741981 and 0.98 between rs10774671 and rs3177979. Finally, despite no evidence of primary association, we tested for increased maternal or paternal transmission to affected siblings (one affected sibling per family) for the three OAS1 SNPs but could not replicate the findings of Tessier et al. (14) (parent-of-origin test χ²₁ = 2.86, P = 0.091).

The RR observed by Tessier et al. (14) was 1.18 for the G allele for the A/G splice-site SNP and 1.19 for both rs3741981 and rs3177979, minor alleles C and G, respectively. Using our family and case-control collections, we had 94% power in the family collections and 100% in the case-control collection to detect such effect sizes at the 5% significance level.

The inconsistencies between the two previous studies are further highlighted by the reply from Field et al. (15) to Tessier et al., in which Field et al. reanalyzed their own data but failed to find a haplotypic association or evidence for OAS1 parental sex-specific effects on diabetes susceptibility, indicating that the results of Tessier et al. were probably false-positives; a conclusion consistent with the results found in the present report. Previously, we had reported evidence for an association of the LRP5 gene in type 1 diabetes, for which most of the statistical support came from a lack of transmission of a haplotype to unaffected siblings (16). This sort of effect, which can be described as a protective effect in unaffected siblings, is difficult to explain biologically, especially in the absence of any effect in affected siblings and when the effect size or penetrance of the potential susceptibility gene under analysis is very low. Hence, we subsequently analyzed much larger sample sizes, and the undertransmission of LRP5 gene region alleles to unaffected siblings disappeared, as one would expect (17). Nevertheless, it remains a possibility that there is an effect at or near OAS1 in type 1 diabetes. However, if there is, then it must be very small and/or susceptible to population-specific effects, requiring replication in extremely large population-specific studies.