Analysis of Polymorphisms of the Interleukin-18 Gene in Type 1 Diabetes and Hardy-Weinberg Equilibrium Testing

Interleukin (IL)-18 is a strong candidate for association studies with type 1 diabetes, as prior biological information exists and a central role for this inflammatory cytokine in autoimmunity has been shown (1–6). Two single nucleotide polymorphisms (SNPs) in the gene promoter located at positions −607 (C→A) and −137 (G→C) are present in European American subjects at minor allele frequencies (MAFs) of 0.46 and 0.30, respectively (estimates obtained from the Innate Immunity Programs for Genomic Applications [IIPGA] database located at http://www.innateimmunity.net). These promoter SNPs have been associated with decreased IL18 expression (7). Allele C at −137 (G→C) has been shown experimentally to disrupt a confirmed H4TF-1–binding site, while nucleotide substitution at −607 (C→A) may disrupt a potential cAMP-responsive element–binding site (7). In an IL18 promoter transcription activity assay, Giedraitis et al. (7) demonstrated low promoter activity following stimulation for both the A and C alleles at positions −607 (C→A) and −137 (G→C) when present on the same haplotype. These results are suggestive of functionality, rendering these promoter SNPs attractive candidates in tests for genetic association with immune-mediated diseases.

These promoter SNPs have also been implicated as susceptibility loci for various diseases, including type 1 diabetes (8,9), rheumatoid arthritis (10,11), sarcoidosis (12), atopic eczema (13), adult-onset Still’s disease (14), and seasonal allergic rhinitis (15). At −137 (G→C), for example, the C allele showed evidence for a positive association with atopic eczema, seasonal allergic rhinitis, and type 1 diabetes (8,13,15). At −607 (C→A), the A allele and the A/A genotype were reported to be negatively associated with sarcoidosis in a Japanese sample and with rheumatoid arthritis in a Chinese sample (10,12). In type 1 diabetes, Kretowski et al. (8) genotyped 201 case and 194 unrelated control subjects from Poland and reported that the C allele at −137 (G→C) was associated with type 1 diabetes (odds ratio [OR] 1.6; P = 0.002). Kretowski et al. also obtained evidence for a genotypic effect at −137 (P = 0.0015), noting an increase, in case subjects, in the proportion of G/C (62.2% case and 48.5% control) and C/C (6.5% case and 3.0% control) genotypes. In addition, they reported that disease risk conferred by a given genotype at −137 (G→C) varied according to the genotype at −607 (C→A); the presence of a G/C or C/C genotype at −137 and either a C/C or C/A genotype at −607 was found to be positively associated with type 1 diabetes, whereas the presence of the A/A genotype at −607 with either a G/G or G/C genotype at −137 was not associated with disease. Ide et al. (9) also tested these promoter SNPs in a Japanese type 1 diabetic case-control collection (116 case and 114 control subjects), noting at −607 (P = 0.023) that the C/A genotype was present in a higher proportion of case compared with control subjects (69% in case and 57.9% in control subjects), whereas the proportion of case subjects with the A/A genotype was decreased (12.1% in case and 26.3% in control subjects). In addition, they found evidence (P = 0.006) of an increase in the frequency of the −607C/–137G haplotype in case subjects (87.9% in case and 73.7% in control subjects) (9). At −607, Kretowski et al. (8) also reported an increase in the proportion of the C/A genotype in case relative to control subjects (64.2% case and 56.7% control subjects) and a decrease in the proportion of the A/A genotype in case subjects (0% case and 5.7% control subjects). Thus, despite discrepant results at −137, data obtained by the two studies at −607 showed some consistency, a decreased proportion of case subjects carrying the A/A genotype.

Kretowski et al. (8) provided no Hardy-Weinberg equilibrium (HWE) information. Testing for deviations from HWE is an important quality control step in population genetic studies (16,17), and routine and appropriate testing of genotype data has been strongly advocated by many authors (18–21). Deviations from HWE can be very informative. In control subjects, this could indicate that one or more of the model assumptions have been violated or that a genotyping error has occurred. In case subjects, deviation from HWE, assuming sources of error have been eliminated, may indicate the association of a locus with disease.

Consequently, we tested the genotype distributions of Kretowski’s control subjects for deviations from HWE and found that both −607 and −137 were out of HWE due to an excess of heterozygotes (P = 0.0002 and P = 0.002, respectively). These deviations could be due to control subjects being sampled from different ethnic groups. Kretowski et al. does not provide any ethnicity information and therefore may not have controlled for this. Alternatively, such deviations could indicate a systematic genotyping error; Kretowski makes no mention of the quality control measures used, such as double scoring or duplicate typing. Arguably, heterozygous excess is most likely to occur as a result of genotyping error. Hypernormal control subjects could also cause deviation from HWE, which may indicate that the alternative hypothesis may be true, although the control subjects used by Kretowski et al. were normally selected, so this is unlikely to be the cause of the observed deviation from HWE.

Some loci will, by chance, deviate from HWE, and both −607 and −137 could be examples of these. However, if these loci were perfectly correct and only deviated from HWE by chance, then it becomes important to use the most appropriate test of association. When deviations from HWE have been observed and they cannot be attributed to genotyping error or to selection or nonrandom mating and are in fact due to some unknown factor, then the Pearson χ² test applied to allele counts is not appropriate, and a test such as the Armitage trend test applied to counts of genotypes should be used (22–25) in order to reduce the chance of false-positive associations (24).

Therefore, we sought to assess the validity of these reported associations at −607 and −137, including evaluation of the possibility that these SNPs show natural deviation from HWE, in a large sample collection consisting of a U.K. Caucasian case-control collection (1,560 case and 1,715 control subjects typed at −607 and 4,323 case and 4,610 control subjects typed at −137) as well as a type 1 diabetic Caucasian collection comprised of families of European ancestry from four countries, including the U.S., the U.K., Norway, and Romania (1,347 families tested at −137 and 1,356 families tested at −607). At the −137 (G→C) SNP, we had 94% power to detect a susceptibility allele (type 1 error, α = 0.01) with an odds ratio of 1.15 and present at an MAF of 26% in our case-control collection (4,300 case and 4,600 control subjects). We had over 99% power to detect an effect size of 1.3. Similarly, we had 98% power in 1,465 families to detect a susceptibility allele (type 1 error, α = 0.01) conferring a relative risk (RR) of 1.3 and present at an MAF of 0.26 and >45% power to detect a genetic risk factor present at a comparable MAF and conferring an RR of 1.15. At the −607 (C→A) SNP, our power increased, reflecting the increase in allele frequency for this SNP (MAF 0.40), giving 57% power to detect (type 1 error, α = 0.01), in 1,600 case and 1,700 control subjects, a genetic risk factor with an OR of 1.15 and 99.6% power to detect a genetic risk factor with an OR of 1.3.

Genotypes at both loci were in HWE in both our control subjects and unaffected parents (P > 0.05). No evidence for association was obtained in the case-control collection either at −607 (OR 1.06 [95% CI 0.95–1.19], P = 0.27) or at −137 (1.00 [0.93–1.07], P = 0.94; Table 1). Likewise, association analysis in the family collection revealed no evidence of increased transmission of the minor allele to probands either at −607 (1.00 [0.91–1.09], P = 0.96) or at −137 (0.96 [0.87–1.05], P = 0.37; Table 2). We assumed a multiplicative model for both −607 and −137 in both the case-control collection and the families because this model was not shown to differ significantly from the full genotype model, which assumes no specific mode of inheritance. Importantly, the frequency of the −607 A/A genotypes did not differ between case and control subjects (15.1 and 14.1%, respectively) nor between affected offspring and pseudocontrol subjects (15.6 and 15.1%, respectively; Tables 1 and 2) and similarly for the C/A genotype. Our results are consistent with results from a recent study involving a small Northern Irish case-control collection (433 case and 426 control subjects) and 285 Northern Irish parent-offspring trios, 246 of which overlap with the U.K. component of our family collection (26).

Both Kretowski et al. (8) and Ide et al. (9) reported evidence for an associated haplotype. Ide et al. observed an increased frequency of the −607C/–137G haplotype in case subjects, while Kretowski et al. observed an increased frequency of the −607C/–137C haplotype in case subjects, a haplotype not found in the Japanese study. We tested for an associated haplotype in both the families and the case-control collection but found no evidence for an associated haplotype (online appendix Table 1 [available from http://diabetes.diabetesjournals.org]). Hence, no further subgroup analysis was undertaken, including a test for gene-gene interaction between an IL18 haplotype and the susceptible CTLA4 +49G/G genotype as that performed by Ide et al. (9).

We sought to evaluate, independently of previously reported associations with promoter SNPs, the possibility that common variation elsewhere in IL18 was associated with disease susceptibility. Therefore, we undertook a further study using tag SNPs to test for regional associations across IL18 in our U.K. type 1 diabetic case-control collection (1,561 case and 1,721 control subjects). IL18 has five exons and has been mapped to chromosome 11q22 (27). Resequencing data from a panel of 23 Caucasian individuals was extracted from the IIPGA “SeattleSNPs” database (as previously described). Thirty-eight SNPs with allele frequencies >5% were used to select eight tag SNPs with a minimum allelic R² of 83.6% for the untyped SNPs. Allele frequencies and predicted R² values for the untyped SNPs are tabulated (online appendix Table 2). Genotype data were in HWE for the control subjects (P > 0.05). A multilocus test for an association of the whole gene region was performed (28), and no evidence for association was found (P = 0.11).

We conclude that common allelic variation in IL18 is unlikely to contribute substantially (OR >1.15) to type 1 diabetes in the populations tested and recommend routine application of tests for HWE in population-based studies for genetic association.