Association and Haplotype Analysis of the Insulin-Degrading Enzyme (IDE) Gene, a Strong Positional and Biological Candidate for Type 2 Diabetes Susceptibility
Abstract
The gene for insulin-degrading enzyme (IDE) represents a strong positional and biological candidate for type 2 diabetes susceptibility. IDE maps to chromosome 10q23.3, a region linked to diabetes in several populations; the rat homolog has been directly implicated in diabetes susceptibility; and known functions of IDE support an important role in glucose homeostasis. We sought evidence for association between IDE variation and diabetes by mutation screening, defining local haplotype structure, and genotyping variants delineating common haplotypic diversity. An initial case-control analysis (628 diabetic probands from multiplex sibships and 604 control subjects) found no haplotypic associations, although one variant (IDE2, −179T→C) showed modest association with diabetes (odds ratio [OR]1.25, P = 0.03). Linkage partitioning analyses failed to support this association, but provided borderline evidence for a different variant (IDE10, IVS20–405A→G) (P = 0.06). Neither variant was associated with diabetes when replication was sought in 377 early onset diabetic subjects and 825 control subjects, though combined analysis of all typed cohorts indicated a nominally significant effect at IDE2 (OR 1.21 [1.04–1.40], P = 0.013). In the absence of convincing support for this association from linkage partitioning or analyses of continuous measures of glycemia, we conclude that analysis of over 2,400 samples provides no compelling evidence that variation in IDE contributes to diabetes susceptibility in humans.
Linkage disequilibrium analysis within candidate genes represents the final common pathway for the identification of diabetes-susceptibility genes. Candidate selection is typically based on positional information from linkage studies in humans and rodent models and/or a perceived match between the gene’s function and the pathophysiology of the condition.
Several independent lines of evidence suggest a role for the gene encoding the insulin-degrading enzyme (IDE: LOCUSLINK reference 3416; EC reference 3.4.24.56) in type 2 diabetes pathogenesis, making this a promising candidate for analysis. First, IDE maps to a region, on chromosome 10q, showing evidence for linkage to type 2 diabetes in several populations. In a genomewide scan of 573 U.K. type 2 diabetic pedigrees, evidence for linkage (multipoint logarithm of odds [LOD] = 1.99) peaked near to D10S1765, only 4 Mb from IDE, and conditional analyses have suggested epistatic interaction with the well-replicated susceptibility locus on chromosome 1q (1). Linkage to 10q has also been reported in Finns (2), North American Europeans (3), Mexican Americans (4), and French (5), though the latter two map 40–50 cM telomeric and may reflect a distinct locus.
Second, the rat homolog of IDE is directly implicated in type 2 diabetes susceptibility. In the GK rat model (6,7), a locus for poststimulation glycemia was mapped to a region of rat chromosome 1 syntenic with human 10q23-26 (8,9) and subsequently localized to a 1-cM region around Ide (10). Two amino-acid substitutions in Ide were found in susceptible congenic strains, which, together, conferred postprandial hyperglycemia and reduced insulin degradation in isolated muscle cells, and other diabetes-related phenotypes (10).
Third, the known and presumed functions of IDE are consistent with a role in type 2 diabetes pathogenesis. IDE, a 110 kDa Zn2+-regulated metalloproteinase, displays a wide expression profile including all insulin-responsive tissues. Strong species conservation attests to important biological functions, and diverse cellular and metabolic roles have been proposed (11,12). Although some of these remain controversial (13), IDE seems to play a major role in the breakdown of insulin in insulin-responsive tissues (14), thereby influencing the extent of the cellular response to insulin (11). This association between IDE activity and insulin clearance and resistance is supported by in vitro and clinical studies (14–16). Furthermore, the substrate specificity of IDE coincides with peptides capable of amyloid formation (including both amylin and insulin), indicating that IDE may prevent accumulation of amyloidogenic peptides (12). Disruption of this scavenging function might thereby promote aggregation of the islet amyloid characteristic of type 2 diabetes and contribute to deficient β-cell capacity.
This congruence of positional and biological candicacy led us to evaluate the role of IDE variation in diabetes susceptibility in humans. Mutation screening of all exons, adjacent intronic sequence, and ∼600 bp of 5′ sequence in probands from 10q-linked pedigrees identified six confirmed variants (Table 1 and Fig. 1). For convenience, these are termed IDE single nucleotide polymorphisms (SNPs) 2, 3, 4, 5, 8, and 9 in accordance with previous nomenclature (17). We did not detect the uncommon IDE6 and IDE7 variants previously reported (17), and IDE1 and IDE10 mapped outside the sequence surveyed. None of the variants appeared to have strong functional credentials based on sequence context. A putative nonsynonymous exon 15 coding polymorphism reported in dbSNP (rs2229708) was not validated on sequencing or direct genotyping. Analysis of 348 European Collection of Cell Cultures (ECACC) control samples confirmed extensive linkage disequilibrium across the gene (all pairwise D′ values >0.60, Fig. 1). Six common haplotypes (HAP1 to HAP6) were identified, accounting for >97% of all chromosomes (Fig. 1). On the basis of allele frequencies (IDE5, IDE6, and IDE7: each <5%) and intermarker linkage disequilibrium patterns (IDE3 concordant with IDE1), SNPs 1, 2, 4, 8, 9, and 10 were selected for further genotyping.
All SNPs were in Hardy-Weinberg equilibrium in all data sets, except for IDE4 (P = 0.01) and IDE9 (P = 0.04), in the Diabetes in Families Study (DIF) control sample only. These samples were retyped with 100% concordance, and haplotype patterns provided no indication of mistyping (no novel haplotypes were generated). These modest departures from equilibrium are likely, therefore, to reflect stochastic variation and are not exceptional given the number of tests performed.
Allele and genotype frequencies for the Warren 2 proband (W2P) cases (n = 628) and control group (CON1) subjects (n = 604) are shown in Table 2 (see research design and methods for group characteristics). The only nominally significant association detected was between W2P and CON1 samples at IDE2 (genotypes P = 0.045; alleles P = 0.030; odds ratio [OR] for C allele 1.25 [95% CI 1.02–1.52]). There was no global difference in haplotype frequency distribution (Fig. 1) between the groups (P = 0.33). However, since HAP2 is defined by the C allele at IDE2, this haplotype was also more frequent in probands than CON1 control subjects (P = 0.03, uncorrected). Males and females did not differ in SNP or haplotype frequencies, and no additional associations were uncovered by sex-stratified analyses.
To determine whether associations might be more evident in families with apparent 10q linkage, we subdivided the W2P sample according to the family nonparametric linkage (NPL) score at the 10q locus. Such linkage partitioning methods have been applied in several recent type 2 diabetes studies (18–21). Here, we used linkage partitioning in a purely exploratory capacity, given uncertainties over the statistical properties of these methods when the evidence for linkage is not confirmed, as well as the possible consequences of partitioning bias when stratifying on proband genotype alone. No difference was evident between “linked” and “unlinked” probands in IDE haplotype frequencies (P = 0.86), IDE2 genotypes (P = 0.61), or IDE10 genotypes (P = 0.18). The association previously seen at IDE2, between W2P and CON1, was attenuated when only linked probands were considered (genotypes P = 0.10; alleles P = 0.08). There was no difference in IDE10 genotype frequencies between 10q-linked probands and CON1 control subjects (P = 0.34). However, the complementary analysis, testing whether IDE genotype partitioned the evidence for 10q linkage, was more encouraging. In families in which the proband carried the weakly associated IDE2 genotypes, CT and CC, the LOD score at D10S1765 increased from 1.72 to 1.96 (P = 0.09 for the increment). The same analysis based on IDE10 genotype (grouping GA and GG), yielded a LOD of 2.12 (P = 0.06).
Clarification of the role of IDE2 and IDE10 variants was sought by further genotyping in early-onset type 2 diabetic cases (YT2D, Warren 2 trios [W2T]) and a large control resource (CON2). Both young-onset case groups had similar clinical characteristics (Table 3), and genotypes were combined after appropriate tests of homogeneity. These replication cohorts have an average age-at-diagnosis some 15 years earlier than the sibpair probands; however, both rigorous exclusion of non-type 2 diabetes and the high parental prevalence of typical late-onset type 2 diabetes in both cohorts (22,23) argue against appreciable qualitative differences in the genetic basis of type 2 diabetes susceptibility between the initial and replication cohorts. Typing IDE2 and IDE10 effectively reduces IDE haplotypic diversity to three common haplotypes (Fig. 1). Genotype and allele counts are shown in Table 2. All data were in Hardy-Weinberg equilibrium. In these replication cohorts, we found no significant association of either SNP (IDE2: genotypes P = 0.14, alleles P = 0.17, OR for C allele 1.17[0.94–1.45]; IDE10: P > 0.80) or the 2-10 haplotype (P = 0.41), regardless of whether the data were stratified for sex. In the full Exeter Family Study (EFS) cohort (n = 841), there was no association between either SNP and fasting plasma glucose on combined or sex-stratified analyses.
Having selected IDE on the basis of its positional and biological candidacy as a type 2 diabetes susceptibility gene, and undertaken a detailed survey of variation within the gene, we have been unable to compile any clear evidence for association between IDE variation and type 2 diabetes. This, despite the fact that our analyses utilized several large and well-characterized populations, and selection of the diabetic individuals—for early onset and positive family history—was designed to enhance detection of type 2 diabetes susceptibility alleles. The only SNP showing association in the initial case-control analysis (IDE2) was not associated in our replication set. However, it is worth noting that the combined OR for IDE2 across both initial and replication sets does achieve nominal significance (combined OR for allele C, 1.21 [1.04–1.40], P = 0.013). This result demands cautious interpretation, given allowance for the low (though unquantifiable) prior odds that even strong biological candidates, such as IDE, will harbor substantial susceptibility effects and for the lack of support from any of the other analyses performed. The only other variant emerging from the initial analyses, IDE10, showed some ability to partition the evidence for linkage, but showed no evidence of association with type 2 diabetes (combined OR for allele A, 1.01 [0.87–1.17], P = 0.91).
Our survey of IDE variation covered all exons, adjacent intronic sequence, and the promoter. None of the variants identified in this, or an earlier screen of U.K. subjects (17), has strong credentials for an impact on IDE structure and/or function. Clearly, our findings do not support the hypothesis that variation within IDE is exclusively responsible for the linkage signal detected on chromosome 10q. However, as with any association study, we cannot exclude the possibility that the variants typed (especially IDE2) have a lesser susceptibility role. The present study has, we estimate, >95% power to detect an association between IDE2 and type 2 diabetes under a multiplicative model with a genotypic relative risk of 1.2, at a threshold P value of 0.05. This falls to ∼60% under a more stringent significance threshold (P = 0.001) designed to reflect modest prior odds that the candidate selected plays a biologically significant susceptibility role. Furthermore, we cannot entirely exclude the possibility that IDE variants other than those studied have a more marked contribution to type 2 diabetes susceptibility. If so, such variants must lie outside the regions surveyed and display limited linkage disequilibrium with any of the typed variants. Current understanding of linkage disequilibrium patterns within human populations indicates that analyses, such as ours, which seek associations using SNPs capturing haplotypic diversity within the gene of interest, will have good power to detect associations with other, as yet unidentified, common variants within the same gene (24).
RESEARCH DESIGN AND METHODS
Subjects:
Clinical characteristics of the cohorts studied are provided in Table 3. Initial case-control analyses compared unrelated index probands (n = 628) from the Diabetes U.K. Warren 2 sibpair repository (“W2P”) (1) with control samples from two sources: 1) 348 random U.K. population samples (from CAMR, Salisbury, U.K.) (“ECACC”) and 2) 256 U.K. subjects ascertained for the Diabetes in Families study (“DIF”). All subjects in the latter were normoglycemic (fasting glucose <6.0mmol/l) and without family history of diabetes. When combined, these control samples are designated “CON1.” Replication analyses combined two collections of young-onset type 2 diabetes subjects with similar clinical characteristics: 1) offspring from parent-offspring trios (n = 142) ascertained for type 2 diabetes (“W2T”) (22) and 2) young-onset (<45 years) type 2 diabetic subjects (n = 235) (“YT2D”) (23). These cases were compared with 841 unrelated parents from a consecutive birth cohort (the EFS), 825 of whom were normoglycemic (“CON2”) (23). All case samples are therefore selected for inherited type 2 diabetes via early onset and/or positive family history. Non-type 2 diabetes was excluded using a combination of clinical, immunological, and genetic criteria (1,22). Confirmation of glycemic status in the control populations was limited to fasting plasma glucose measures for the DIF and EFS samples. All subjects are of exclusively British/Irish Caucasian origin (except for a small proportion with non-British but European ancestry in the W2T set).
Gene structure and mutation detection.
The genomic structure of IDE was determined through BLAST alignment of human mRNA sequence (M21188) with the finished genomic sequence of BAC clone AL356128. This confirmed IDE as having 25 exons distributed between positions 93.09 and 93.22 Mb on the NCBI30 assembly of chromosome 10. Variant detection was performed by direct sequencing in both directions (Big Dye Terminator chemistry; Perkin Elmer Applied Biosystems, Warrington, U.K.) on an ABI3700 capillary sequencer. We included all exons, >50 bp of flanking introns, and ∼600 bp of 5′ untranslated region (UTR) in 27 amplicons. Primer sequences are provided in online Table 1 (http://diabetes.diabetesjournals.org). In total, 11 individuals were sequenced. Eight were type 2 diabetic probands from sibships where all affected were identical-by-descent on both chromosomes at flanking microsatellites (D10S1765, D10S185). Because such probands, assuming 10q linkage, are enriched for 10q susceptibility variants, we estimate >90% power to detect disease-associated variants (assuming multiplicative models and genotype relative risk of 1.5) for allele frequencies >8.7%. The other three were unaffected siblings from the same families, doubly discordant for 10q haplotypes covering IDE, when compared with their affected sibs.
SNP genotyping.
Selected IDE SNPs were genotyped using a combination of approaches including restriction fragment-length polymorphism detection, artificial induced restriction sites, and tetra-primer amplification refractory mutation methods. Assays of the first two types included obligate restriction sites as internal digest controls. For assay details, see online Table 2 (http://diabetes.diabetesjournals.org).
Statistical methods.
Genotype and allele frequency distributions were compared by standard contingency table methods with exact probability estimation, using STATXACT (Cytel, Cambridge, MA). Genotype comparisons used the Kruskal-Wallis statistic given the ordinal categorization. Separate data sets (e.g., the W2T and YT2D sets) were combined for contingency table analyses only after appropriate homogeneity testing. Haplotype patterns were estimated by maximum-likelihood methods (SNPHAP, www-gene.cimr.cam.ac.uk/clayton/software/) and haplotype frequency distributions compared by likelihood-ratio testing. Significance was determined by permutation (10,000 replicates). Measures of linkage disequilibrium between SNPs were derived using the PM (Permutation and Model-free Analysis) program (25). To assess the evidence for linkage partitioning (18–21), two complementary approaches were adopted. First, W2P probands were subdivided into “linked” (positive family NPL score at IDE) and “unlinked,” and variant frequency comparisons were repeated. Second, the evidence for linkage at 10q was recomputed (using GENEHUNTER-PLUS [1]) after conditioning on the proband’s IDE genotype. The significance of changes in the linkage statistic was determined by permutation (10,000 replicates). The power calculations described assume, for simplicity, that all cases are ascertained from affected sibpairs (to incorporate the ascertainment selection) and that all control samples are population based.
Genomic position, linkage disequilibrium relationships, and haplotype frequencies of variants within the IDE gene. The depiction of the IDE gene displays the exon distribution (not to scale, see distances provided above) and genomic location of the SNPs examined in this study. The middle panel details linkage disequilibrium parameters (D′ above the diagonal, P value below), calculated in 348 random U.K. samples (the ECACC sample) for the six SNPs genotyped. The lower panel lists the ten most frequent haplotypes estimated from the typed data sets and their frequencies in the CON1 control subjects (made up of the ECACC and DIF samples) and the W2P case set.
IDE variants relevant to this study
Genotype and allele frequencies at IDE haplotype-tag (ht) SNPs
Clinical characteristics of the populations studied
Acknowledgments
This work was supported by Diabetes U.K. (which also supported ascertainment of many of the samples collected), the European Union (GIFT consortium, QLG2-CT-1999-00546), and the South and West National Health Service Research Directorate.
T.M.F. is a career scientist of the South and West National Health Service Research Directorate, K.R.O. is a Diabetes U.K. Clinical Training Fellow, and A.T.H. is a Wellcome Trust Career Leave Research Fellow.
We thank all those who participated in these collections, and the many health professionals who contributed. We thank Dr. Sian Ellard for laboratory organization and Diane Jarvis for DNA extraction; Dr. Lesley Jones (Cardiff) for sharing IDE variant data; and Prof. Paul Burton and colleagues (Leicester) for valuable statistical insights.
Footnotes
-
Address correspondence and reprint requests to Prof. Mark McCarthy, Robert Turner Professor of Diabetes, Oxford Centre for Diabetes, Endocrinology and Metabolism, Churchill Site, Old Road, Headington, Oxford OX3 7LJ, U.K. E-mail: mark.mccarthy{at}drl.ox.ac.uk.
Received for publication 5 November 2002 and accepted in revised form 10 February 2003.
Additional information for this article can be found in an online appendix at http://diabetes.diabetesjournals.org.
CON, control; DIF, Diabetes in Families Study; ECACC, European Collection of Cell Cultures; HAP, haplotype; IDE, insulin-degrading enzyme; LOD, logarithm of odds; NCBI, National Center for Biotechnology Information; NPL, nonparametric linkage; OR, odds ratio; SNP, single nucleotide polymorphism; UTR, untranslated region; W2P, Warren 2 probands; W2T, Warren 2 trios; YT2D, young type 2 diabetes.
- DIABETES
REFERENCES
In this Issue
