Quantitative Trait Loci Near the Insulin-Degrading Enzyme (IDE) Gene Contribute to Variation in Plasma Insulin Levels

The gene encoding insulin-degrading enzyme (IDE) is located on chromosome 10q23-q24, within a region linked to type 2 diabetes and related quantitative traits (1–4 ). IDE is the major enzyme responsible for insulin proteolysis in vitro (5) and shares structural and functional homology with bacterial protease III, which may function in the termination of the insulin response (6,7). In mice, IDE hypofunction induced by IDE gene disruption leads to hyperinsulinemia (8). Furthermore, IDE activity in the diabetic Goto-Kakizaki (GK) rat is reduced by ∼30%, where polymorphism in IDE is likely to be the main contributing factor (9). Congruence of positional and functional data indicates that sequence variation in IDE may play a role in modifying insulin metabolism in human populations. Two recent genetic association studies investigated the IDE region in relation to type 2 diabetes. One produced significant evidence for effects on both type 2 diabetes and plasma glucose levels (10), whereas the other explored only case-control models and obtained no evidence of association (11). Neither of these studies, however, attempted to directly relate IDE variants to measures of insulin metabolism.

To evaluate the potential influence of genetic variation in IDE on insulin levels and correlated quantitative traits, a haplotype-tagging strategy was used (12) in a Swedish sample consisting of 321 impaired glucose tolerance (IGT) and 403 nondiabetic control subjects (Table 1). This study follows a recent report (13) in which we examined 26 polymorphic markers extending across IDE in relation to Alzheimer’s disease in a large Swedish population. These were used to define regional linkage disequilibrium (LD) structure, which highlighted a LD block spanning 276 kb around IDE. Three single nucleotide polymorphism (SNP) markers were identified as capable of delineating common haplotypes (>5%) in Swedes and were subsequently used in tests of association.

For the present study, 14 SNP markers extending locally across IDE were examined (Fig. 1), including previously identified tag markers. We began by genotyping these in the entire male sample to define LD structure and to identify common haplotypes. All 14 markers were found to be in Hardy-Weinberg equilibrium. Pairwise marker correlations suggested strong LD in the region, in agreement with earlier data (13). Common haplotypes (those >5% frequency), inferred with HAPLOTYPER using all 14 markers, were similar to those obtained based on a slightly modified marker set (13). LD estimates and inferred haplotypes for the entire sample are shown in Figs. 2 and 3, respectively. The definition of Patil et al. (14) (α = 0.8) was used to confirm that the studied region does represent an LD block. We also confirmed that a selection of three tag markers equivalent to those previously used (13) would be able to delineate common haplotypes and opted not to explore the use of alternative equally valid marker sets. We note that a variety of different approaches exist for defining optimal SNP sets (15), but we have not explored their use here. These and other issues may have relevance for attempts at replicating or refuting the present data, as LD structures can differ markedly between divergent populations (K.K. Kidd, A.J.B., personal communication).

For initial analyses, our intention was to reduce the number of exploratory tests by examining a critical set of phenotypes against a limited number of haplotypic genotypes (also referable to as diplotypes—for brevity, we have opted to use the latter term throughout this report). The principal criterion for diplotype construction is that they can be inferred with high probability and thus reflect with fidelity combinations of common haplotypes for each individual. In the presence of limited allelic heterogeneity, this can be advantageous over single marker tests, which can fail to detect effects due to insufficient or complex patterns of LD with pathogenic variants. In addition, testing many single markers requires a substantial degree of multiple testing. Our selection of quantitative phenotypes for analyses included six traits typically examined in relation to diabetes (16–18), and these are listed in Table 1. We note that studies (19,20) that have investigated IDE in relation to diabetes did not report efforts to relate genotypes to either insulin levels or obesity. Trait distributions were highly skewed or partially skewed for all six studied phenotypes, and thus only log-transformed data were modeled. Correlations between traits were extremely high, as was expected (data not shown).

Diplotypes were inferred using the three selected haplotype-tagging markers (Fig. 3) and tested against all six traits highlighted in Table 1 in men and women separately. Six of eight possible haplotypes are predicted using these markers, the sequences for which are H1-TCG, H2-TCA, H3-CCA, H4-TTA, H5-TTG, and H6-CCG. Posterior probability estimates were in excess of 0.80 for all phase calls that represent the various combinations of these haplotypes (a threshold that was set for retaining individuals for analyses). Significant and consistent effects upon traits were evident, with the notable exception of fasting glucose and 2-h glucose levels. Importantly, effects were evident exclusively in men, which are in agreement with results obtained by Karamohamed et al. (10). The results of these analyses are shown in Fig. 4, where data for the male sample are presented. For all analyses, we combined IGT individuals and nondiabetic control subjects and adjusted for any uniform phenotypic differences between them by including their group identity as a factor in ANOVA models. Diplotype by group interaction terms in second-order factorial ANOVA models were not significant. As the above analyses are based on probabilistically inferred data, we considered it important to also explore models that only entail the use of phase-known data. This can be accomplished by excluding individuals that are heterozygous at two or more sites (all individuals with only one heterozygote site are by definition phase known). To demonstrate this, an additional analysis was performed in men on the four associated traits shown in Fig. 4 (insulin, 2-h insulin, homeostasis model assessment of insulin resistance [HOMA-IR], and BMI). Results revealed association with all four of these, but significance was highest for BMI (F_7,220 = 3.5; P = 0.0014) (data for other comparisons not shown). For reference to Fig. 4, the groups that are excluded in this analysis are H1/H3, H2/H5, H3/H4, and H3/H5.

In an attempt to refine the possible position of pathogenic variant(s) in the region, single marker tests were conducted on the phenotypes that provided evidence of association in diplotype models. In considering multiple testing issues, we noted that the strongest individual finding in HOMA-IR (P = 0.0001) affords 500 individual tests before strict multiple testing correction renders this finding insignificant (given α = 0.05). These tests, using all 14 markers, were performed only in the male sample because diplotype analyses gave us no a priori reason to suspect that latent associations for single markers would be present in the female set. These results are shown in Fig. 5, where consistent effects of rs2251101 and rs2249960 can be seen upon traits. The common allele at rs2251101 and the rare allele of rs2249960 were associated with lower trait levels. Marker rs2251101 and rs1887922 were tested against fasting glucose levels because the latter was specifically reported on by Karamohamed et al. (10), and these two markers are in extremely high LD (r² > 0.5) (Fig. 2). Neither marker showed evidence of association, but we acknowledge that the present study has less power than the aforementioned one. The result for rs2251101 upon multiple phenotypes nonetheless supports the likely presence of a primary trait-modifying sequence in the 3′ region of IDE, as has been speculated (10). The effect of rs1887922 on 2-h insulin levels was significant, however, albeit considerably attenuated in comparison with rs2251101. To further explore the effects of rs2251101, a final model was fitted in which rare-allele homozygotes and heterozygotes were combined and compared with common allele homozygotes. The results of this were as follows: F_1,366 = 9.1, P = 0.0028; F_1,363 = 18.4, P = 0.000023; F_1,365 = 7.0, P = 0.0085; and F_1,365 = 11.4, P = 0.0008; for fasting insulin, 2-h insulin, HOMA-IR, and BMI, respectively. Although throughout this study tests were conducted on the combined IGT and normal samples (and a correction applied for possible differences between them), it was considered important to confirm that an effect exists in both populations in at least one analysis. Using the above model and focusing on the rs2251101 marker and the 2-h insulin trait, significant effects were observed for both the IGT (F_1,155 = 13.0, P = 0.0004) and normal (F_1,206 = 6.5, P = 0.011) groups; in both cases the common allele homozygote group had the lowest trait levels.

In summary, we provide strong evidence that sequence variation within or near IDE contributes to variability in measures of insulin metabolism. There are several compelling features to these results. First, both our study and the study by Karamohamed et al. (10) suggest that a primary functional variant occurs in the 3′ region of IDE. In addition, the marker that previously exhibited maximum evidence of association, rs1887922 (10), is in very strong LD with the marker (rs2251101) with the greatest effect in the present study. A true pathogenic variant is likely to be in high LD with both of these (and indeed could be one of these markers). Second, both of these studies suggest effects either exclusively or predominantly in male subjects. Numerous studies have detected genetic effects exclusively in men (20,21), and sex specificity may be a major confounding factor in genetic analyses (22). Third, the marker that was in strongest association with quantitative traits related to Alzheimer’s disease in our recent study (13) was rs2251101 (from among 26 markers). Importantly, however, the present analyses suggest that there may be more than one functionally relevant polymorphic site in or near IDE. This is evident in diplotype models, where no single haplotype appears responsible for either elevating or reducing trait levels and where a number of possible interactions are also apparent. Evidence of allelic heterogeneity and interaction will in general make fine mapping of the responsible variants a daunting task (23), and this will need to be considered in future studies investigating the IDE region.

We have not directly addressed the question of whether variation in IDE influences the risk for type 2 diabetes. The two studies that have tested this relationship have produced either no or modest evidence of association, despite examining very large clinical samples. Our study provides no evidence of association with fasting glucose levels (the primary phenotype used to distinguish diabetic patients). Rather, the present data likely capture the fundamental influence of IDE upon circulating insulin levels, whereby a genetically determined deficit in enzyme activity may lead to hyperinsulinemia. This could manifest itself as a modulator role in type 2 diabetes, but might be more relevant to other phenotypes, such as coronary heart disease (24) and primary hypertension (25). We believe that continued replication efforts in both normal and morbid populations, as well as functional studies, are now highly merited and will be needed to elucidate the molecular basis and clinical relevance of these observed phenotypic effects.