Common Haplotypes at the Adiponectin Receptor 1 (ADIPOR1) Locus Are Associated With Increased Risk of Coronary Artery Disease in Type 2 Diabetes

Two populations of Caucasian individuals with type 2 diabetes were studied, one from Boston (n = 411) and the other from San Giovanni Rotondo, Italy (n = 533). The two populations have been previously described (24). The study protocol and informed consent procedures were approved by the local research ethics committees. Each population included a group of case subjects with CAD and a group of control subjects without clinical evidence of CAD. In Boston, the CAD-positive case subjects, defined as subjects who had a stenosis >50% in at least one major coronary artery or their main branches, were recruited among type 2 diabetic patients who underwent cardiac catheterization at the Beth Israel Deaconess Medical Center. Control subjects were Joslin Clinic patients (the Joslin Clinic serves as the Beth Israel Deaconess Medical Center diabetes clinic), aged ≥55 years, who had diabetes for ≥5 years but had a negative cardiovascular history and a normal exercise treadmill test. The Italian population consisted of type 2 diabetic patients who attended the Endocrine Unit of the Institute “Casa Sollievo della Sofferenza ” in San Giovanni Rotondo, Italy. Case subjects were patients who had angiographic evidence of stenosis >50% in at least one major coronary artery or their main branches or who had acute myocardial infarction. Control subjects included diabetic patients without symptoms and with normal resting electrocardiogram and exercise treadmill test or with coronary stenosis (at angiography) ≤50%. Clinical features of case and control subjects from the two studies are shown in Table 1. The European diversity panel of the Joslin Genetics Core, consisting of healthy subjects from Italy (n = 192), Poland (n = 180), and the U.K. (n = 192), was also typed to evaluate population stratification as a plausible explanation of association findings.

Single nucleotide polymorphisms (SNPs) for the linkage disequilibrium study were typed in the Caucasian HapMap panel by means of PCR followed by single-base extension/fluorescence polarization (AycloPrime-FP SNP Detection System) using a Wallac VICTOR 2 Multilabel Plate Reader (Perkin-Elmer, Boston, MA). Haplotype-tagging polymorphisms were also typed in the study groups by single-base extension/fluorescence polarization, with the exception of rs7539542, which was typed by means of a TaqMan assay (Applied Biosystems, Foster City, CA) implemented on an ABI PRISM 7700 HT Sequence Detection System. Genotyping quality was tested by including six blinded duplicate samples in each 96-well assay. The average agreement rate of duplicate samples was >99%. Sequences of the primers and probes used for typing are available from the authors.

Genotype distributions were tested at each polymorphic locus for departure from Hardy-Weinberg equilibrium. Pairwise linkage disequilibrium coefficients (D′ and r²) were estimated and plotted using the Haploview software package (25). Linkage disequilibrium blocks were determined according to the CI method using the criteria of Gabriel et al. (26). Haplotype frequency and haplotype-tagging SNPs were determined by means of the algorithms implemented in the Haploview software (24), using 0.05 as the frequency threshold to define common haplotypes.

Allele frequencies were compared between case and control subjects within each population by χ² tests and in the two populations combined by the Mantel-Haenszel statistics. P values were adjusted for multiple comparisons by the Bonferroni correction, considering conservatively that six independent comparisons, corresponding to the six haplotype-tagging SNPs, were made. False discovery rates were calculated from the equation p × k/i, where p is the P value, k is the number of comparisons, and i is the P value ranking. The significance of between-population differences in the association with CAD was determined by means of the Breslow-Day test. For each SNP, the odds ratios (ORs) of CAD for heterozygotes and minor allele homozygotes compared with major allele homozygotes were estimated by logistic regression analysis using two different models: one including only the genotypes as predictors and the other also including known cardiovascular risk factors (age, sex, percent ideal body weight, smoking status, hypertension status, and HbA_1c [A1C]) as covariates. In the analysis of the two populations combined, an indicator variable for the study population (Boston versus Italy) was added to both models. Between-population differences in the CAD risk associated with each genotype were tested for significance by adding an interaction term (genotype times population) to the logistic regression models.

Maximum likelihood estimates of haplotype frequencies in case and control subjects were derived using the expectation-maximization algorithm as implemented in the function HAPLO.EM of the Haplo Stats suite (27). The association between CAD and common (≥0.05) ADIPOR1 haplotypes was analyzed using the score statistics proposed by Schaid et al. (27) and implemented in the function HAPLO.SCORE of the Haplo Stats software. This method allows adjustment for covariates (in this case, age, sex, percent ideal body weight, smoking status, hypertension status, A1C, and study population) and provides a global test of association, as well as haplotype-specific tests. After testing for association with haplotypes, a combination of haplotypes (i.e., a diplotype) was assigned to each individual on the basis of the posterior probabilities of the different phases. The risk of CAD associated with each diplotype relative to CCG/CCG homozygotes was then estimated by logistic regression analysis as described above for individual SNPs. Forty-three subjects (4.6% of the total) were excluded from this analysis because none of the possible phases had a posterior probability >0.70.

Levels of ADIPOR1 mRNA were measured by quantitative real-time PCR in peripheral blood mononuclear cells from 60 nondiabetic employees of the Joslin Diabetes Center (mean age 33 ± 11 years, BMI 25.9 ± 3.8 kg/m²) and in adipose tissue biopsies from 28 obese subjects who underwent liposuction or gastric bypass surgery at Washington University (mean age 44 ± 10 years, BMI 45.2 ± 12 kg/m²). Peripheral blood mononuclear cells were isolated from peripheral blood by density gradient (Vacutainer CPT; Becton Dickinson, Franklin Lakes, NJ); subcutaneous abdominal adipose tissue was obtained by percutaneous needle aspiration. RNA was extracted using the RNeasy Kit (Qiagen, Valencia, CA). ADIPOR1 mRNA levels relative to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA levels were determined by quantitative RT-PCR using predeveloped TaqMan Gene Expression Assays from Applied Biosystems (Foster City, CA), according to the manufacturer’s instructions in an ABI PRISM 7700 HT Sequence Detection System. ADIPOR1/GAPDH mRNA ratios were obtained from the equation 2^−ΔCT, where ΔCT is the difference in threshold cycles between ADIPOR1 and GAPDH. At each polymorphic locus, the significance of mRNA level differences between genotypes was estimated by ANCOVA, using log-transformed values to account for the skewed distribution of the ADIPOR1/GAPDH ratios and age and sex as covariates.

AdipoR1 cDNA expression constructs carrying the two alleles of rs7539542 (pCMV-SPORT6-AdipoR1/G and pCMV-SPORT6-AdipoR1/C, respectively) were prepared by site-directed mutagenesis (QuickChange II XL kit; Stratagene, La Jolla, CA) starting from ADIPOR1 IMAGE clone no. 3878067 (GeneBank no. BC010743) (ATCC, Manassas, VA). HEK293 cells were cultured in 175-cm² culture flasks with DMEM/F12 containing 10% (vol/vol) FCS and 2 mmol/l glutamine at 37°C in a humidified atmosphere of 5% CO₂. Cells were plated in six-well plates, grown in complete medium for 48 h, and transfected with either pCMV-SPORT6 AdipoR1/C or pCMV-SPORT6 AdipoR1/G using FuGENE 6 Transfection Reagent (Roche Diagnostics, Monza, Italy) according to manufacturer’s instructions. Eighteen hours after transfection, actinomycin D (5 mg/ml) was added and total RNA extracted at different time points using the RNAeasy Quick kit (Qiagen, Milano, Italy). ADIPOR1 mRNA levels relative to GAPDH mRNA levels were determined by quantitative RT-PCR using a predeveloped TaqMan Gene Expression Assay from Applied Biosystems (Monza, Italy) according to the manufacturer’s instructions in an ABI PRISM 7700 HT Sequence Detection System.

The ADIPOR1 gene is located on chromosome 1q32 and includes eight exons encompassing 17.5 kb. To select markers that would comprehensively capture the variability at this locus, we typed the Caucasian (“CEU”) HapMap panel for 15 common SNPs (minor allele frequency ≥5%) covering the entire gene plus 5 kb on each side (a list of the SNPs and their positions is provided in the online appendix [available at http://diabetes.diabetesjournals.org]). Data were then integrated with those of 13 other SNPs typed by the HapMap initiative (HapMap public release no. 19) (28). Linkage disequilibrium analysis of the resulting set of 28 SNPs (1 kb average spacing) revealed two linkage disequilibrium blocks as defined according to the CI criteria proposed by Gabriel et al. (26). One block extends from the 5′ flanking region (−11,800 bp from the translation start site) to intron 4 (+5,843), and the other is at the 3′ end of the gene (Fig. 1A). One SNP, hCV37351, is between the two blocks and exhibits significant linkage disequilibrium with both (Fig. 1A). Based on this linkage disequilibrium structure, we selected six polymorphisms for association studies, three tagging the first block (rs2232853, rs12733285, and rs1342387) and three tagging the second (rs7539542, rs10920531, and rs4950894) (Fig. 1B).

The association between these six SNPs and CAD was evaluated in two populations of Caucasian individuals with type 2 diabetes, one from Boston and the other from Italy. The two populations were similar with respect to age at examination, age at diagnosis and treatment of diabetes, and body weight (Table 1). Subjects from Boston had, on average, better glycemic control and a higher prevalence of former or current smokers than those from Italy (Table 1). In both populations, the prevalence of smoking was markedly higher in case than control subjects, consistent with the known powerful effect of this factor on cardiovascular risk. The two populations combined provided >80% power (α = 0.05) to detect genetic effects with allelic ORs ranging from 1.3 for rs1342387 to 1.4 for less frequent SNPs such as rs2232853 or rs7539542.

Genotype distributions were in Hardy-Weinberg equilibrium at all six loci. The three SNPs tagging the 5′ block had similar allele frequencies in case and control subjects in both populations (Table 2). By contrast, the three tags of the 3′ linkage disequilibrium block (rs7539542, rs10920531, and rs4950894) were all significantly associated with CAD in the Boston population, with nominal P values ranging from 0.001 to 0.01 (Table 2). In the Italian case and control subjects, the smaller differences in allele frequencies for these three polymorphisms were not statistically significant (Table 2) but neither were they significantly different from the allele frequency differences in the Boston population (P = 0.19, P = 0.14, and P = 0.21, respectively, for the difference in association in the two study populations [Breslow-Day test]). When the two populations were pooled and analyzed together using the Mantel-Haenszel statistics, all three loci were significantly associated with CAD, with nominal P values of 0.004 for rs7539542, 0.0009 for rs10920531, and 0.006 for rs4950894, and with Bonferroni-adjusted P values of 0.024, 0.005, and 0.036, respectively. The corresponding false discovery rates were 0.011, 0.005, and 0.012, respectively. For both rs7539542 and rs10920531, the allelic association was mostly due to an excess of minor allele homozygotes among case subjects. For rs10929531, this effect was significant in both populations (adjusted OR 3.8 [95% CI 1.9–7.6] and 1.8 [1.05–3.2] in Boston and Italy, respectively). The ORs were not significantly different between the two populations, and when the two studies were combined, minor allele homozygotes were estimated to have a more than twofold increase in the odds of CAD compared with major allele homozygotes (Table 3). A weak, nonsignificant effect was noted for heterozygotes, with common OR estimates of 1.2–1.3 (Table 3). In the case of rs4950894, both heterozygotes and minor allele homozygotes were significantly associated with protection from CAD, with adjusted ORs of 0.72 and 0.57, respectively (Table 3).

To assess whether population stratification might be responsible for the stronger association with CAD in Boston, we determined the frequencies of the risk alleles in a sample of healthy individuals from three European ethnic groups (Italian, n = 192; Polish, n = 180; and British, n = 192), reasoning that stratification was an unlikely explanation of our findings if allele frequencies varied little among European populations. As shown in online appendix Table 2, the allele frequencies of all three SNPs displaying association with CAD were not significantly different among European populations. At rs10920531 (the SNP showing the strongest association with CAD), the difference in allele frequencies between Boston case and control subjects was significantly larger than the largest difference in allele frequencies between European populations (0.114 vs. 0.038, P < 0.05). Furthermore, the frequency of the risk allele in Boston CAD case subjects was significantly higher than the highest allele frequency observed in the European populations (0.444 vs. 0.362, P < 0.025). Thus, population stratification was an unlikely explanation of our findings. We also consider the hypothesis that differences in clinical characteristics such as age, sex, A1C, or smoking history could be responsible for the different strength of the association in the two populations. However, we did not detect any significant interaction between this variables and SNPs (online appendix Table 3), indicating that this was also unlikely as an explanation of our findings.

Consistent with the individual SNP analyses, a significant association with CAD was observed for the haplotypes in the 3′ block but not for those in the 5′ block. Figure 2A shows the maximum likelihood estimates of the 3′ block haplotypes in the combined case and control subjects along with their score statistics for association with CAD as computed by the HAPLO.SCORE program (27). Haplotype distributions were significantly different in case and control subjects (global empirical P = 0.0002). Haplotype CCG, which exactly corresponds to the minor allele of rs4950894, was associated with protection from CAD (haplotype-specific empirical P = 0.00035), whereas haplotype GAA, exactly corresponding to the minor allele of rs7539542, was associated with predisposition (P = 0.021). A predisposing effect was also observed for haplotype CAA, although the P value did not reach significance (P = 0.11) due to the low frequency of this haplotype. When diplotypes were assigned to each subject, haplotypes GAA and CAA were significantly associated with an increased risk of CAD when they occurred as homozygotes, in combination with each other, or in combination with the neutral haplotype CCA (Fig. 2B). The ORs of CAD for these genotypes compared with CCG/CCG homozygotes ranged from 1.8 to 3.6. The ORs were lower and did not differ significantly from 1.0 when haplotypes GAA and CAA occurred as heterozygotes with the protective haplotype CCG (Fig. 2B).

Resequencing of all exons in 32 CAD-positive case subjects did not identify any missense polymorphism in linkage disequilibrium with rs7539542, rs10920531, or rs4950894. Thus, we hypothesized that the association with CAD may be mediated by an effect of polymorphisms on gene expression. To investigate this possibility, levels of ADIPOR1 mRNA were measured by quantitative real-time PCR in peripheral blood mononuclear cells from 60 nondiabetic individuals and in adipose tissue biopsies from 28 obese subjects. Both in mononuclear cells and adipose tissue, some of the genotypes associated with increased predisposition to CAD (i.e., minor allele homozygotes for rs7539542 or rs10920531, major allele homozygotes for rs4950894) had lower levels of ADIPOR1 mRNA (Fig. 3). In mononuclear cells, this pattern was significant at rs7539542, where G/G homozygotes had 30–40% lower ADIPOR1 mRNA levels than heterozygotes or C/C homozygotes (P = 0.001 and P = 0.014, respectively) (Fig. 3A). When the three SNPs were considered together as haplotypes, the CAD-predisposing GAA/GAA genotype was associated with significantly lower ADIPOR1 mRNA levels than the CAD-protecting CCG/CCG genotype (0.42 vs. 0.61, P = 0.007). In the adipose tissue biopsies, the difference in ADIPOR1 mRNA levels among genotypes was significant at rs10920531, where minor allele homozygotes had 35% lower expression than major allele homozygous (P = 0.013), with heterozygotes having intermediate values (Fig. 3B).

Since rs7539542 is placed in the 3′ untranslated region, we investigated whether the association of this SNP with CAD and with ADIPOR1 mRNA levels could be explained by an effect on mRNA stability. To this end, HEK293 cells were transfected with allelic forms of ADIPOR1 cDNA, and mRNA levels were measured at different time points after stopping transcription with actinomycin D. As shown in Fig. 4, the rates of mRNA decay were virtually identical for the two alleles, indicating that rs7539542 does not affect mRNA stability, at least in this cell type.

Our findings suggest that sequence variants in the 3′ region of the ADIPOR1 gene are significant determinants of cardiovascular risk in type 2 diabetes. This effect appears to be due to allelic differences in ADIPOR1 expression, which may influence the antiatherogenic effects of adiponectin on target tissues. These results confirm and extend the evidence implicating adiponectin as a physiological modulator of atherogenesis and point to variability in the ADIPOR1 gene as a key regulator of this effect.

Several features of these findings make the possibility of a false-positive result unlikely. First, the evidence of association with CAD was highly significant, with Bonferroni-adjusted P values ranging from 0.005 to 0.036 and false discovery rates ranging from 0.005 to 0.012. Second, there were no major differences in allelic frequencies among European populations that would explain these findings on the basis of population stratification. Third, the association with CAD, although with different strength, was observed in two distinct Caucasian populations recruited at different centers, one in Boston and the other in Italy. Fourth, some of the genotypes associated with CAD also showed an association with decreased mRNA levels in two different cell types, providing a plausible biological basis for the case-control results. On the other hand, our findings should be interpreted with caution. In the Italian study, the association with CAD was significant for only one SNP and only for the contrast between minor and major allele homozygotes. The fact that the ORs at this and other polymorphic loci were not significantly different in the two populations is reassuring, since it suggests that the estimates obtained in the two studies are different measures of the same genetic effect. Nonetheless, the results in the Italian population cannot be formally considered as a replication of those in Boston.

Some caveats also apply to the findings of the expression studies, since significance was reached in both cell types for only one of the three SNPs associated with CAD. It is also unknown whether the allelic differences that we observed in mRNA levels are mirrored by differences in the protein levels and, if so, whether a 30% reduction in the receptor levels is sufficient to determine tangible differences in adiponectin action. Thus, further studies are needed to unequivocally confirm the role of the ADIPOR1 gene as a modulator of CAD susceptibility in type 2 diabetes and precisely determine the magnitude of this effect.

The mechanisms through which variants at the ADIPOR1 locus could influence cardiovascular risk are only hypothetical at this time. A recent study in nondiabetic Mexican Americans has described a strong, positive correlation between ADIPOR1 expression levels in skeletal muscle and insulin sensitivity as determined by the glucose clamp (29). Thus, the association between ADIPOR1 polymorphisms and increased cardiovascular risk might be mediated by an increase in insulin resistance secondary to the decreased ADIPOR1 expression associated with the polymorphisms. In this regard, a recent report from Germany (22) has described an association between insulin resistance, as determined by oral glucose tolerance tests and the glucose clamp, and ADIPOR1 polymorphisms in the 5′ part of the gene. Unfortunately, the 3′ flanking region of the gene, for which we detected the association with CAD, was not investigated in that report.

ADIPOR1 polymorphisms may also affect cardiovascular risk through mechanisms that are independent of insulin sensitivity. In cultured cells, adiponectin suppresses monocyte adhesion to the endothelium, inhibits smooth muscle cell proliferation, and reduces foam cell formation (9–11). Administration of recombinant adenovirus-expressing human adiponectin to apolipoprotein E–deficient animals causes a 30% reduction in the formation of atherosclerotic lesions in the absence of any effects on metabolic traits (30,31). In humans, high adiponectin levels have a protective effect on cardiovascular events that is independent of other cardiovascular risk factors and systemic inflammation markers (32,33). The decrease in ADIPOR1 expression associated with the ADIPOR1 SNPs in the 3′ block may impair such direct antiatherogenic actions of adiponectin. The fact that the association between ADIPOR1 SNPs and gene expression was detected in circulating monocytes, the precursors of foam cells in atherosclerotic plaques, makes this hypothesis especially attractive.

While the effect of ADIPOR1 variability on CAD risk appears to be mediated by differences in gene expression, the identities of the sequence variants that are responsible for this effect are unknown. The observation that the four haplotypes fall into more than two risk classes suggests the involvement of multiple polymorphisms, either the tagging SNPs or variants in linkage disequilibrium with them, interacting with each other. Of the three tagging SNPs, rs7539542 was especially attractive as one of the possible functional variants because of its location in the 3′ untranslated region, a region that plays a pivotal role in the control of gene expression by binding proteins that regulate mRNA processing, translation, or degradation (34,35). However, our experiments in HEK 293 cells, failing to show an impact of this SNP on ADIPOR1 mRNA stability, do not support the hypothesis of a functional role of this polymorphism. While these findings should be interpreted with caution because of the tissue-specific nature of the regulation of mRNA stability, it is likely that the functional variants are placed elsewhere, for instance in currently unknown 3′ flanking regulatory elements.

Some limitations of our study should be acknowledged. The two linkage disequilibrium blocks on which the tagging SNP selection was based did not include 5 kb of the gene, corresponding to exons 5, 6, 7, and part of 8. Thus, genetic effects of variants in this region might have been missed. On the other hand, the one SNP that was typed in this region was in strong linkage disequilibrium (r² > 0.9) with SNPs in the first block, suggesting that variants in this interval might have been nonetheless captured by the haplotype-tagging SNPs. Another limitation is that control subjects might have included some individuals with significant CAD as a result of the occurrence of silent ischemia in diabetes and the less-than-perfect negative predictive value of the exercise treadmill test (36). Since such misclassification, if any, is expected to be random (that is, independent from the genotype), this might have biased the results toward the null hypothesis. Finally, it must be emphasized that this was a cross-sectional study, which might be susceptible to bias due to possible effects of genetic variants on survival. Thus, our findings must be confirmed in prospective studies before ADIPOR1 polymorphisms can be considered as predictors of CAD in type 2 diabetes.

This research was supported by National Institutes of Health Grants HL73168 and HL71981 (to A.D.), DK60837 (Diabetes Genome Anatomy Project), DK36836 (Genetics Core of the Diabetes and Endocrinology Research Center at the Joslin Diabetes Center), DK56341 (Clinical Nutrition Research Unit of Washington University), a Grant-in-Aid from the American Heart Association (to A.D.), a grant from the Donald W. Reynolds Foundation (to A.D.), and by Italian Ministry of Health Grants RC0303ED19 and RC0201ED02 (to V.T.). W.M. was supported by a mentor-based postdoctoral fellowship from the American Diabetes Association.

We are grateful to David Poznik for computational help and Maya Becker, Aviva Bashan, David Nolan, and Jill Duffy for assistance with the recruitment of study subjects. We acknowledge the invaluable contribution by the individuals who participated in this study.