Genetic Variants in the UCP2-UCP3 Gene Cluster and Risk of Diabetes in the Women's Health Initiative Observational Study

The WHI-OS is a longitudinal study designed to examine the association between clinical, socioeconomic, behavioral, and dietary risk factors with subsequent incidence of health outcomes, including cardiovascular disease (CVD) and diabetes. Details regarding the scientific rationale, eligibility, and other design aspects have previously been described (18). The study has been reviewed and approved by human subjects review committees at each participating institution, and signed informed consent was obtained from all women enrolled.

Incident diabetes cases were identified based on postbaseline self-report of first-time use of hypoglycemic medication (oral hypoglycemic agents or insulin) during a median follow-up period of 5.9 years. Approximately 82,069 subjects had no prior history of CVD and/or diabetes at baseline. Following the principle of risk-set sampling, for each new case, control subjects were selected randomly from women who remained free of CVD and/or diabetes at the time the case was identified during follow-up. Control subjects were matched to case subjects by age (±2.5 years), racial/ethnic group (Caucasian, African, Hispanic/Latino, and Asian/Pacific Islander), clinical center (geographic location), time of blood draw (±0.10 h), and length of follow-up. In our current study, each of the 968 Caucasian case subjects were randomly chosen and matched with one control subject. Of 616 incident cases among ethnic minority women, 366 case subjects were African American, 152 Hispanic, and 98 Asian/Pacific Islander. The 1:2 matching ratio was used for minorities to strengthen the power in these smaller sample sizes of cases (19).

We undertook a two-stage approach. The first stage consists of comprehensive common single nucleotide polymorphism (SNP) discovery by genotyping a total of 21 SNPs in 244 samples randomly selected from the WHI-OS source population. The second stage involved genotyping a total of 14 tag SNPs (tSNPs) in the entire case-control samples according to LD patterns defined during the first stage.

Briefly, we first surveyed all common genetic variants available from the National Center for Biotechnology Information dbSNP database. In total, an initial set of 21 SNPs were selected based on the following criteria: 1) functionality priority (nonsynonymous coding SNPs [cSNPs] and splicing-site SNPs were kept following the order of coding SNPs > splicing-site SNPs > 5′ UTR SNPs >3′ UTR SNPs > synonymous SNPs > intronic SNPs); 2) minor allele frequency (MAF) ≥5% in at least one ethnic group; and 3) relatively even spacing across the genomic region (20). The initial set of SNPs was genotyped using the high-throughput Illumina BeadArray platform (21) at Illumina (San Diego, CA) in a multiethnic panel of 244 women randomly chosen from the entire case-control sample (n = 61 for each group of Caucasian, African, Hispanic, and Asian-American subjects). A graphical depiction of the physical locations of these SNPs is shown in supplemental Fig. 1 (available in an online appendix at http://dx.doi.org/10.2337/db07-1269).

In the second stage, we selected tSNPs based on the LD patterns of those 21 SNPs. Pairwise LDs between SNPs were assessed using Lewontin's D′ statistic and the squared correlation statistic r². The Haploview program was used to calculate the LD coefficients and define haplotype blocks (22). We chose all common tSNPs with special focus on Caucasian and African-American samples, since Hispanic and Asian-American subjects only constitute relatively small proportions in either case or control groups. Using the r²-based Tagger program (22), tSNPs in African Americans were chosen by finding the minimum set of tSNPs with pairwise r² ≥0.80 and MAF ≥5%. We then added additional SNPs to reach a minimum set of tSNPs for Caucasian-American samples to ensure a sufficient and yet nonredundant parsimonious set of tSNPs. From the initial dense set of 21 SNPs, a total of 14 tSNPs were selected and genotyped in all case-control samples (Table 1). In the HapMap II dataset, the estimated coverage (r² ≥ 0.8) of the UCP2-UCP3 genetic variation by these 14 genotyped SNPs is 76% for Caucasians and 80% for Asians. The estimated coverage when r² ≥ 0.5 is 86% for Caucasians and 96% for Asians. The untaggable SNPs genotyped in HapMap dataset are located evenly across the whole 70-kb UCP2-UCP3 genome region. Genotyping was performed using the TaqMan allelic discrimination method in the molecular epidemiology laboratory (S.L., principal investigator) at the University of California, Los Angeles (19). A total of 138 randomly selected replicate samples were genotyped, and the consistency rate was 99% for each of the 14 tSNPs. The average genotyping dropout rate was 1.8% (from 1.3 to 2.5%).

MAF in the control samples was estimated for each ethnic group. The Hardy-Weinberg equilibrium test for each of the 14 tSNPs was performed using the χ² test (1 d.f.) (SAS version 9.1; SAS Institute, Cary, NC). We also tested for heterogeneity of genotype distributions across ethnicities by the χ² test (3 d.f.).

In both single-SNP and haplotype-based analyses, conditional multivariable logistic regression models were employed to estimate the odds ratios (ORs) (95% CI) for each genetic variant of type 2 diabetes risk with adjustments for multiple confounding variables including matching factors (age, clinical center, time of blood draw, and ethnicity), BMI, ln(fasting glucose), ln(fasting insulin), cigarette smoking (never, past, and current), alcohol intake (never, past, and current), postmenopausal hormone therapy (never, past, and current), type 2 diabetes family history (presence/absence), and total MET value (the energy expended by a person at rest; 1 MET = 1 kcal · kg⁻¹ body weight · h⁻¹ [23]) from recreational physical activity per week at baseline. In single-SNP analyses, each SNP was coded as additive, dominant, and recessive genetic models for estimations of ORs (95% CI) using conditional logistic regressions for each genetic model, respectively.

In haplotype-based analyses, haplotypes were estimated from the unphased genotype data using the HAPPY macro in SAS, version 9.1 (24,25), and the haplotype frequency estimates were then validated via HAPLOTYPER v2, a Bayesian haplotype inference algorithm (26). We reported haplotype frequencies for the haplotypes with frequency ≥2.5% in control subjects but pooled those rare haplotypes (<2.5%) together as one haplotype category, z_pooled. In all tests performed, the frequency of z_pooled was found to be <10% in either the control group or the case group, respectively. To account for the uncertainty of haplotype phasing, we used the expectation-substitution approach treating the expected haplotype scores under the additive model as observed covariates in the conditional logistic regression model (24). To increase the genomic coverage, we employed a sliding window (window width 4 SNPs) haplotype-based analysis. For each window, an omnibus likelihood ratio test was used, which was a χ² test (d.f. = number of haplotypes in a particular window − 1). The test was based on the difference of the logarithmic likelihood of two conditional logistic regression models: the reduced model, which does not contain the haplotype covariates, and the full model, which contains the haplotype covariates. As a priori, stratified analyses by overweight status (BMI >25 kg/m², overweight subgroup) were also performed.

To adjust for the single-point significance level for multiple testing with corrected type I error, we reported empirical P values based on global random permutation tests. We randomly permuted the case-control status of each subject, performed the same set of analyses (including single SNP analyses, sliding window analyses, stratified analyses by ethnic groups, and stratified analyses by overweight status), and record the minimal P value for each permutation dataset. The distribution of the minimal P values obtained from 10,000 permutation datasets was used to derive the empirical significance of the observed test statistic (P_permutation). The adjusted global-wide P values were determined as P_adjusted = P(P_observed = < P_permutation). All reported P values are from two-sided tests. To address the issue of undiagnosed diabetes at baseline and assess the robustness of our findings, we also conducted sensitivity analyses excluding case and control subjects who had a one-time measure of fasting glucose ≥126 mg/dl at baseline.

The characteristics of the 14 tSNPs are shown in Table 1. None of the tSNPs had genotype distributions deviating from Hardy-Weinberg equilibrium at P < 0.01 levels. The estimated MAFs in control subjects stratified by ethnicity were shown in Table 2. With the exceptions of UCP2 Ala54Val (rs660339) (P = 0.04), the genotype distributions of all tSNPs varied significantly across different ethnicities (Table 2). In particular, the MAFs in the African-American women differed significantly from those of all other ethnic groups for the majority of the tSNPs.

The LD (D′ was used as the pairwise LD metric) structures and haplotype blocks across the 70-kb UCP2-UCP3 gene cluster are shown in supplemental Fig. 2. Overall, four visually discernable blocks with slightly varying boundaries across different ethnicities were defined, including block one, which covers the DNAJB13 gene and the 5′ upstream region of the UCP2-UCP3 gene cluster; block 2, the DNAJB13-UCP2 intergenic region; block 3, UCP2 and 3′ UTR of UCP3; and block 4, the UCP3 region. The −866 G/A (rs659366) and the Ala55Val (rs660339) in UCP2 and the −55 C/T (rs1800849) and the Tyr99Tyr (rs1800006) in UCP3 were in high LD (D′ >0.95 and logarithm of odds [LOD] ≥2) for all four ethnic groups. However, the −866 G/A and Ala55Val polymorphisms of UCP2 were not in high LD (D′ <0.50 and LOD ≥2 for Caucasian, African, and Hispanic-American samples; D′ <0.80 and LOD ≥2 for the Asian-American sample) with either the −55 C/T or the Tyr99Tyr polymorphism of UCP3.

The association of each tSNP with type 2 diabetes risk in each ethnic group or in the combined samples was evaluated under the additive, dominant, and recessive genetic models. As shown in Table 3 (using the additive genetic model), nominally significant (i.e., P < 0.05) associations were found for rs2306820 (OR 1.8 [95% CI 1.1–3.1]), rs668514 (1.8 [1.1–3.0]), and Tyr99Tyr (rs1800006) (0.6 [0.4–0.9]) among African-American women. In the combined multiethnic sample, a nominally significant association was found between rs653263 and type 2 diabetes (0.7 [0.6–0.9]). However, none of these associations remained significant after adjustment for multiple testing. Similarly, no significant association was found under either the dominant or the recessive genetic model for any tSNP after further adjustment for multiple testing (data not shown). Leaving out baseline fasting glucose and insulin in multivariable models did not materially change these findings.

The sliding window (window width 4 SNPs) was employed to analyze haplotype-disease associations, which generated a total of 11 window frames for the 14 tSNPs. Figure 1 shows the most common haplotype and number of common haplotypes in each window. P values in Fig. 1 indicate the overall differences in haplotype frequencies from the omnibus likelihood ratio test. Two overlapping 4-tSNP windows (rs591758-rs668514-rs647126-rs1800006 [χ² = 24.14; 7 d.f.; nominal P = 0.00107; P_adjusted = 0.046] and rs668514, rs647126, rs1800006, and rs1800849 [χ² = 19.26; 5 d.f.; nominal P = 0.00171; P_adjusted = 0.063]) were found to be associated with a higher type 2 diabetes risk in Caucasian women. The latter was also found to be significantly associated with incident diabetes in the combined multiethnic sample (χ² = 20.19; 5 d.f.; nominal P = 0.00115; P_adjusted = 0.048). A marginal association was found between the haplotype set rs591758-rs668514-rs647126-rs1800006 and type 2 diabetes risk in Hispanic women (χ² = 13.47; 7 d.f.; nominal P = 0.0614; P_adjusted = 0.591).

This haplotype set (rs591758-rs668514-rs647126- rs1800006) was also found to be associated with an even higher risk of type 2 diabetes in overweight (BMI >25 kg/m²) Caucasian (χ² = 25.57; 7 d.f.; nominal P = 0.0006; and permutation P = 0.032) and Hispanic (χ² = 19.17; 7 d.f.; nominal P = 0.0077; permutation P = 0.174) women. Haplotype set rs591758-rs668514-rs647126-rs1800006 spans the UCP2-UCP3 intergenic and UCP3 regions. Using the most common haplotype, h1010, as the referent group, the haplotype-specific ORs for type 2 diabetes were 2.0 (95% CI 1.13–3.37) for haplotype h0001 specifically (19.5% in controls) in Caucasian women and 3.8 (1.44–9.93) in Caucasian overweight women. Without adjustment of baseline fasting glucose and insulin, the haplotype h0001–specific ORs for type 2 diabetes were 1.3 (1.01–1.75) in all Caucasian women and 3.3 (1.35–7.75) in Caucasian overweight women. The effect of this particular haplotype on risk of type 2 diabetes seems to be independent from baseline fasting glucose and insulin. We also did not find a significant interaction effect on type 2 diabetes risk between haplotype h0001 and BMI.

We performed haplotype analyses for BMI in healthy control and diabetic case subjects for the diabetes-associated haplotype set rs591758-rs668514-rs647126-rs1800006. No significant association was found between haplotype and BMI in either control or case subjects. The diabetes-associated haplotype was found to be associated with overweight status (BMI >25 kg/m²) only in Hispanic-American diabetic case subjects (nominal P value = 0.005). We did not observe a significant association with obesity status (BMI >30 kg/m²).

To address the issue of undiagnosed diabetes at baseline and assess the robustness of our findings, we conducted sensitivity analyses excluding 630 case subjects who had fasting glucose ≥126 mg/dl at baseline. Significant associations with type 2 diabetes risk were found for the haplotype set rs591758- rs668514-rs647126-rs1800006 in Caucasian women (nominal P = 0.0013) and the overweight subgroup (nominal P = 0.0005). As a result of small sample size, marginally significant associations with incident diabetes were also found for this haplotype set in Hispanic women (nominal P = 0.0754) and the overweight subgroup (nominal P = 0.0322). Compared with those who carried the most common haplotype (h1010), Caucasian women with haplotype h0001 had a 1.9-fold (95% CI 1.10–3.38) higher risk of developing incident diabetes, and the risk increased to 3.3-fold (1.25–8.94) for Caucasian overweight women.

In this large prospective study of postmenopausal women with diverse ethnicity, we found that Caucasians who were carriers of haplotypes defined by rs591758, rs668514, rs647126, and rs1800006 that covered UCP3 and the UCP2-UCP3 intergenic region had a significant higher type 2 diabetes risk (nominal P = 0.00115, P_adjusted = 0.048), especially among those who were overweight (nominal P = 0.0006, permutation P = 0.032). Compared carriers of the most common haplotype (h1010), carriers of haplotype h0001 had a twofold (95% CI 1.13–3.37) higher risk of developing type 2 diabetes, and the risk increased to 3.8-fold (1.44–9.93) for those carriers who were also overweight. However, after adjustment for multiple comparisons, we did not observe any significant association in Hispanic, African, or Asian-American ethnicity.

Human UCP3 gene encodes a mitochondrial transmembrane carrier protein (3,27,28), and as an uncoupler of oxidative phosphorylation, UCP3 is thought to play an important role in maintenance of energy balance and body weight (29). UCP3 is predominantly expressed in skeletal muscle (27,30), a major tissue contributing to nonshivering thermogenesis in humans (28). Krook et al. (31) reported lower UCP3 mRNA levels in type 2 diabetic patients, whereas increased UCP3 mRNA levels were also observed by others (32). More recently, Vidal et al. (33) showed that obese type 2 diabetic patients had three- to fourfold higher UCP3 mRNA levels than obese control subjects. However, mRNA levels do not necessarily reflect protein content. Relatively little is known about the protein levels of UCP3 in humans, even though overexpression of UCP3 in transgenic mice resulted in increased glucose tolerance and reduced fasting plasma glucose levels (34).

In the current study, we found that overweight Caucasian women with the high-risk haplotype (h0001, C-C-G-C) had a 3.8-fold increased risk of type 2 diabetes. Exactly what specific variants in this region may account for this effect are not known. Haplotype h0001 is in high LD with both −866 A- and −55 T-alleles. This may indicate that previous significant finding for −866 G/A and −55 C/T polymorphisms may be due to some as yet unidentified variants covered by this diabetes-associated haplotype. A 45-bp I/D variant in the 3′UTR of UCP2 gene has been reported to be associated with BMI (15,17) and 24-h metabolic rate (35). This 45-bp I/D variant is not genotyped in the HapMap project or in our study. However, based on a study by Wang et al. (15) of a North European–ancestry Caucasian population, this 45-bp I/D polymorphism seems to be in high LD with the A55V polymorphism (D′ = 0.97; r² = 0.82) and with −866 G/A polymorphism (D′ = 0.75, r² = 0.55). Since haplotype h0001 is in high LD with −866 A- and −55 T-alleles, this 45-bp I/D variant may also be covered by the diabetes-associated haplotype set rs591758-rs668514-rs647126-rs1800006. In a 15-year follow-up study of healthy middle-aged Caucasian men, Gable et al. (13) found that those with both UCP2 −866 AA and UCP3 −55 TT genotypes had an OR of 4.2 (95% CI 1.70–10.37), and this association was particularly strong among men with a BMI >30 kg/m² (OR 19.2 [95% CI 5.6–63.7]). Taken together, these finding indicate that the adverse effect of these genetic variants on type 2 diabetes risk might be exacerbated by adiposity. One possible explanation for our findings is that being overweight could exacerbate the extent of insulin resistance as a consequence of an impairment of mitochondrial fat oxidation and accumulation of intramyocellular lipid due to reduced UCP3 protein expression by high-risk haplotypes (13). Recently, Schrauwen et al. (36) found that type 2 diabetic patients had a 50% lower UCP3 protein content compared with age-matched control subjects. UCP3 protein content was also found to be inversely correlated with plasma glucose and insulin levels (36). Thus, high levels of UCP3 may protect against the development of type 2 diabetes (36,37). This hypothesis is in accordance with the observation that mice overexpressing UCP3 were resistant to development of diet-induced diabetes (38). If, as observed in our study, the diabetes-associated haplotype UCP2-UCP3 is a true positive association, then this haplotype would be expected to be associated with decreased UCP3 protein levels in skeletal muscle. Thus, it would be warranted to assess the direct haplotypic effects on mRNA transcriptional efficiency of UCP3 and its protein activity in experimental settings.

Contributions of UCP2-UCP3 genetic variants to type 2 diabetes risk may differ significantly in different ethnicities. It has been suggested that population stratification may lead to false-positive results (39) because participants of different ethnicities/geographic locations might have different type 2 diabetes risk mainly because of their different environmental exposures or different allele frequencies for at-risk SNPs or haplotypes. To address any potential bias from population stratification, we carefully selected the control samples to be representative of the WHI-OS source population and also performed ethnicity-stratified analyses. However, in these ethnicity-stratified analyses, our study might be underpowered to detect associations of UCP2 −866 G/A (rs659366) and UCP3 −55 C/T (rs1800849) polymorphisms with type 2 diabetes.

Potential biases due to the inclusion of some women with undiagnosed diabetes need careful consideration. To avoid potential bias associated with the timing of outcome definition, we have adopted a well-established strategy: 1) utilizing standard and identical protocols to define case and control subjects; 2) excluding all of the prevalent case subjects from our original case-control sampling space; 3) matching each case-control pair on age, ethnicity, clinical center, time of blood draw, and follow-up time; 4) performing analyses by adjusting for obesity/insulin resistance indexes such as baseline fasting glucose, insulin levels, or HOMA-IR in the multivariable models; and 5) conducting secondary analyses after excluding those with fasting glucose ≥126 mg/dl or further excluding all cases occurring in the first year of follow-up. Therefore, we conducted sensitivity analyses by adjusting for baseline fasting glucose and insulin levels. It is possible that individuals with risk SNPs in UCP2–3 may have increased glucose or insulin concentrations already at baseline; thus, the absolute marginal effect of this gene on type 2 diabetes risk would be underestimated by the adjustment for baseline glucose and insulin levels in our model. However, we believe that adjusting for glucose and insulin would allow us to examine conditional effects of this gene on diabetes risk that may be independent of glycemica or insulinemia. We recognize that by adopting such a stringent criterion for diagnosis, we may pick up the more severe cases and therefore misclassify some less severe cases of diabetes into the nondiabetic groups. Nevertheless, our strategy serves to minimize false positive, which is a major threat to validity. The consistency between the results from our secondary analyses and our primary results speaks to the robustness of our findings.

Finally, our findings among postmenopausal women were likely to be generalizable only to women. Because our study included ethnically diverse women from 40 U.S. states, these findings may be generalizable to women of similar age and ethnically diverse background.

In summary, for each of the 14 tSNPs across the genomic region of the UCP2-UCP3 gene cluster, we did not observe significant effects on type 2 diabetes. However, Haplotype-based analyses suggest that a haplotype set defined by rs591758, rs668514, rs647126, and rs1800006 was significantly associated with type 2 diabetes risk in Caucasian women only, especially among those who were overweight. These data warrant confirmation in future prospective and experimental studies.

This work was supported in part by National Institutes of Health Grants R01 DK062290, R01 DK066401, and R01 HG002518. The WHI-OS is funded by the National Heart, Lung and Blood Institute, U.S. Department of Health and Human Services.

We thank Dr. JoAnn Manson of Brigham and Women's Hospital and Dr. David Altshuler of Massachusetts General Hospital, both of whom provided advice in the design stage of this study. We also thank the dedicated and committed participants and the entire staff of the WHI-OS. The following list of WHI investigators participated in this study: Program office, National Heart, Lung, and Blood Institute (Bethesda, MD): Elizabeth Nabel, Jacques Rossouw, Shari Ludlam, Linda Pottern, Joan McGowan, Leslie Ford, and Nancy Geller. Clinical Coordinating Centers: Fred Hutchinson Cancer Research Center (Seattle, WA), Ross Prentice, Garnet Anderson, Andrea LaCroix, Charles L. Kooperberg, Ruth E. Patterson, and Anne McTiernan; Wake Forest University School of Medicine (Winston-Salem, NC), Sally Shumaker; Medical Research Labs (Highland Heights, KY), Evan Stein; University of California at San Francisco (San Francisco, CA), Steven Cummings. Clinical Centers: Albert Einstein College of Medicine (Bronx, NY), Sylvia Wassertheil-Smoller; Baylor College of Medicine (Houston, TX), Jennifer Hays; Brigham and Women's Hospital, Harvard Medical School (Boston, MA), JoAnn Manson; Brown University (Providence, RI), Annlouise R. Assaf; Emory University (Atlanta, GA), Lawrence Phillips; Fred Hutchinson Cancer Research Center (Seattle, WA), Shirley Beresford; George Washington University Medical Center (Washington, DC), Judith Hsia; Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center (Torrance, CA), Rowan Chlebowski; Kaiser Permanente Center for Health Research (Portland, OR), Evelyn Whitlock; Kaiser Permanente Division of Research (Oakland, CA), Bette Caan; Medical College of Wisconsin (Milwaukee, WI), Jane Morley Kotchen; MedStar Research Institute/Howard University (Washington, DC), Barbara V. Howard; Northwestern University (Chicago/Evanston, IL), Linda Van Horn; Rush Medical Center (Chicago, IL), Henry Black; Stanford Prevention Research Center (Stanford, CA), Marcia L. Stefanick; State University of New York at Stony Brook (Stony Brook, NY), Dorothy Lane; Ohio State University (Columbus, OH), Rebecca Jackson; University of Alabama at Birmingham (Birmingham, AL), Cora E. Lewis; University of Arizona (Tuscon/Phoenix, AZ), Tamsen Bassford; University of Buffalo (Buffalo, NY), Jean Wactawski-Wende; University of California at Davis (Sacramento, CA), John Robbins; University of California at Irvine (Irvine, CA), F. Allan Hubbell; University of California at Los Angeles (Los Angeles, CA), Howard Judd; University of California at San Diego (San Diego/La Jolla/Chula Vista, CA), Robert D. Langer; University of Cincinnati (Cincinnati, OH), Margery Gass; University of Florida (Gainesville/Jacksonville, FL), Marian Limacher; University of Hawaii (Honolulu, HI), David Curb; University of Iowa (Iowa City/Davenport, IA), Robert Wallace; University of Massachusetts/Fallon Clinic (Worcester, MA), Judith Ockene; University of Medicine and Dentistry of New Jersey (Newark, NJ), Norman Lasser; University of Miami (Miami, FL), Mary Jo O'Sullivan; University of Minnesota (Minneapolis, MN) Karen Margolis; University of Nevada (Reno, NV), Robert Brunner; University of North Carolina (Chapel Hill, NC), Gerardo Heiss; University of Pittsburgh (Pittsburgh, PA), Lewis Kuller; University of Tennessee (Memphis, TN), Karen C. Johnson; University of Texas Health Science Center (San Antonio, TX), Robert Brzyski; University of Wisconsin (Madison, WI), Gloria E. Sarto; Wake Forest University School of Medicine (Winston-Salem, NC), Denise Bonds; and Wayne State University of Medicine/Hutzel Hospital (Detroit, MI), Susan Hendrix.