Introduction

Obesity is associated with insulin resistance (IR)¹ and greatly increases the risk of developing type 2 diabetes in some individuals (1). IR is manifested by a failure of insulin to inhibit gluconeogenesis by the liver and a diminished glucose uptake in response to insulin by skeletal muscle and adipose tissue (1). A number of factors produced by adipocytes affect insulin sensitivity, including interleukin (IL) 6, which is mainly produced in resting individuals by white fat (2). Elevated IL-6 levels are associated with IR (3, 4, 5, 6) and predict the development of diabetes (7). Given that obesity increases both the level of IL-6 and the risk of diabetes, we hypothesized that interactions between genetic variants affecting IL-6 production and BMI may modify associations with IR or risk for diabetes.

IL-6 production is affected by the IL-6 gene promoter polymorphism (- 174 G/C, G = major allele) that shows tissue-specific effects on IL-6 production (8). The polymorphism has also been tested for association with diabetes. In a Native American and Spanish white population, the GG genotype was reported to increase risk of diabetes (9), whereas in a Finnish population, the GG genotype was associated with increased insulin sensitivity (and presumably lower diabetes risk) (10) and the CC genotype with diabetes in the context of the tumor necrosis factor- (G-308A) polymorphism (11). In a German population, the GG genotype was associated with diabetes in lean men (12). The variability in these associations could be explained by a previously unconsidered interaction between the - 174 IL-6 genotype and obesity on levels of IR and diabetes risk. In this report, we test this hypothesis in a large sample of unrelated individuals from a community-based sample.

Research Methods and Procedures

Subjects

The Framingham Heart Study (FHS) began in 1948 with a randomly selected cohort from the town of Framingham, MA. Initially, 5209 participants were enrolled. In 1971, 5124 offspring of the original cohort and the spouses of the offspring were added (13). We analyzed a group of 1526 unrelated individuals from the Offspring Study for whom complete phenotype and genotype data were available. This study was approved by the Boston University Institutional Review Board, and written informed consent was obtained from each subject.

Genotyping

Participants were genotyped for the - 174 polymorphism using a mass spectroscopy-based single nucleotide polymorphism detection assay (Sequenom, San Diego, CA). Polymerase chain reaction primers were 5'ACGTTGGATGAGCCTCAATGACGACCTAAG and 5'ACGTTGGATGGATTGTGCAATGTGACGTCC and the extension primer was 5'CCCCCTAGTTGTGTCTTGC. Genotyping was performed on those unrelated individuals for whom DNA was available. IL-6 - 174 genotypes were in Hardy Weinberg equilibrium (p = 0.2), and allele frequencies were similar to those reported in other Europid populations (9, 10).

Traits

Levels of fasting plasma glucose (FPG) and fasting insulin were measured at Examination Cycle 5 (1991 to 1995) in 2607 Offspring participants. Diabetes status at Exam 5 was defined as treatment for diabetes, having two or more FPG levels 7.0 mM at any of Exams 1 to 5 or a 2-hour post-challenge plasma glucose 11.1 mM (>200 mg/dL) administered at Exam 5 according to American Diabetes Association standards. In previous work, 99.2% of diabetes in the offspring was type 2 diabetes (14). Fasting insulin was measured in EDTA plasma as total immunoreactive insulin [ Coat-A-Count Insulin; Diagnostic Products Corporation, Los Angeles, CA; cross-reactivity with proinsulin at midcurve, 40% ; intra- and inter-assay coefficients of variation, 5.0% to 10.0% ; lower limit of sensitivity, 1.1 U/mL (8 pM)] . Serum insulin values were estimated from plasma insulin values using linear regression (15). Plasma glucose was measured with a hexokinase reagent kit (Agent glucose test; Abbott, South Pasadena, CA; duplicate assay intra-assay coefficients of variation, 2% to 3% ). IL-6 levels were measured by enzyme-linked immunosorbent assay (16). We used homeostasis model of assessment (HOMA)-IR as our primary surrogate measure of IR, where HOMA-IR = insulin (milliunits per liter) glucose (millimolar)/22.5, based on a homeostatic model validated in euglycemic-hyperinsulinemic clamp studies (17, 18, 19). We also estimated IR using the Gutt insulin sensitivity index (ISI): (ISI_0,120) = (m/mpg)/log(msi), where m = (75,000 + (FPG - 2-hour glucose) (milligrams per deciliter) 0.19 weight (kilograms))/120, mpg = (FPG + 2-hour glucose) (millimolar)/2, and msi = (FSI + 2-hour insulin) (microunits per liter)/2 (20).

Statistical Analysis

We computed descriptive statistics for the sample as means, percentages, and Pearson correlations. Analysis of associations between the - 174 IL-6 polymorphism and quantitative traits was accomplished by multiple linear regression methods using the SAS procedure PROC GLM to evaluate genotypic effects (SAS Institute Inc., Cary, NC). Models (separate for each sex) included dummy variables for genotype, a variable for BMI (as continuously distributed), first order interaction terms for BMI and genotype, and terms for age (including squared and cubed terms to allow for non-linearity), sex, smoking (number of cigarettes per day), physical activity, alcohol use, and estrogen use in women, assessed as previously described (21). We used logarithmic values to reduce heteroscedascity and robust covariance estimators (MAD.SAS and robust HB.SAS macros, http://www.ats.ucla.edu/stat/sas/webbooks/reg/chapter4/sasreg4.htm) to account for outliers, using reiterative median absolute deviation in the regression with a biweight (22). Formulas derived from this analysis were used to plot the predicted effects displayed in figures. An example of the calculation for the figures for the combined sample of men and women for HOMA-IR using the GG genotype as reference is: HOMA-IR = 1.82 + 0.186 sex - 1.30 (mean age) + 0.028 [ times[ (mean age)² - 0.00018 (mean age)³ - 0.102 (mean alcohol intake) + 0.0086 (mean number of cigarettes) - 0.89 (estrogen replacement therapy, 1 or 0) - 0.053 (mean physical activity index) + 0.96 (BMI) + 12.45 (GC genotype, 1 or 0) + 4.32 (CC genotype, 1 or 0) - 0.48 BMI (GC genotype, 1 or 0) - 0.167 BMI (CC genotype, 1 or 0).

We repeated this approach in a subsidiary analysis using waist circumference instead of BMI as the obesity metric (23). The odds of diabetes associated with genotype were assessed using logistic regression models employing the covariates listed above. Exponentiated model coefficients gave the odds ratio (OR) for the change in prevalence of diabetes per unit increase in BMI. We considered two-sided p values 0.05 to indicate statistical significance. Because of our focused hypothesis, no corrections were made for multiple testing.

Results

Characteristics of the 1525 study subjects are displayed in Table 1. Men were more obese and insulin resistant than women. As expected, BMI was a significant correlate of HOMA-IR (Figure 1), but we found no significant association between the IL-6 genotype and HOMA-IR [ GG, 6.69 1.02 HOMA-IR units (n = 562); GC, 6.55 1.02 HOMA-IR units (n = 740); CC, 6.78 1.04 HOMA-IR units (n = 223)] .

Figure 1.

Sex-specific analysis of unrelated men testing for the dependence of HOMA-IR on an interaction between the IL-6 genotype and BMI. (A) Predicted lines for each genotype from a weighted linear regression model using ln(HOMA-IR) as the dependent variable incorporating age (including squared and cubed terms to allow for non-linearity), smoking, physical activity, and alcohol use as covariates in addition to BMI and IL-6 genotype. (B-D) Plot of unadjusted data against the predicted line for each genotype. The circled points indicate those with a weighting of 0.5 or less. The dotted line indicates the 90th percentile for HOMA-IR.

Full figure and legend (71K)

Table 1. - Characteristics of 1525 unrelated Framingham Offspring Study subjects.

Full table

We tested the hypothesis that the effect of BMI on HOMA-IR depended on the - 174 genotype by incorporating first order interaction terms for BMI and the IL-6 genotype into models predicting HOMA-IR levels. For the pooled sample (p = 0.14) and women (p = 0.46), the interaction was not significant, indicating that the effect of BMI on HOMA-IR in women was the same regardless of - 174 genotype. Among men, the p value for the interaction term for IL-6 genotype and BMI predicting levels of HOMA-IR was highly significant (p < 0.0001, Figure 1). The effect for each genotype predicted in the interaction model for men is presented in Figure 1A, where HOMA-IR levels are plotted against BMI, and the lines predicted from regression models are shown. At BMI levels above 27 kg/m², men with the CC genotype had higher levels of HOMA-IR than men with G alleles, whereas at lower levels of BMI, men with the CC genotype had lower HOMA-IR levels than those with G alleles. The fits of these regression lines to the underlying unadjusted data are displayed in Figure 1, B to D. Outliers, with a weight < 0.5, are circled and comprise <3% of this sample. Omission of these outliers in subsidiary analyses did not affect the results.

Results were similar when we used waist circumference (risk per inch change) instead of BMI as the obesity metric. In an analysis otherwise identical to that shown in Figure 1, among men, the risk factor-adjusted p value for the waist circumference IL-6 genotype interaction term was 0.003; among women, the p value was 0.06, and in the pooled sample, the p value was 0.08. Results were also similar when we used the Gutt ISI (20) instead of HOMA-IR as the surrogate measure of IR; again, the BMI IL-6 genotype interaction was significant (p = 0.006) in men but not in women (p = 0.63).

Because an increase in HOMA-IR values is a function of elevated FPG or fasting insulin, we tested for effect of the - 174 IL-6 genotype on these trait values. For the entire sample, there was no statistical difference in the mean FPG associated with each genotype (GG, 5.3 0.06 mM; GC, 5.3 0.06 mM; CC, 5.4 0.06 mM) nor for the mean fasting insulin associated with each genotype (GG, 4.7 0.2 pM; GC, 4.6 0.2 pM; CC, 4.7 0.02 pM). Likewise, levels of FPG and fasting insulin were similar across genotypes for men and women separately (data not shown). We then tested whether levels of FPG or fasting insulin were predicted by an interaction between - 174 IL-6 genotype and BMI. In sex-specific analyses of men, we found a significant interaction with BMI on the effect of IL-6 genotype for both fasting insulin (p < 0.001, Figure 2A) and for FPG (p < 0.016, Figure 2B). These results suggest that interaction by the - 174 IL-6 polymorphism and BMI on levels of HOMA-IR reflects similar underlying effects on levels of FPG and fasting insulin. Among women, we found no significant interaction between - 174 IL-6 genotype and BMI for FPG (p = 0.54) or fasting insulin (p = 0.57).

Figure 2.

Sex-specific analysis of unrelated men testing for the dependence of glycemic trait measures on an interaction between the IL-6 genotype and BMI. (A) Fasting serum insulin. (B) Fasting plasma glucose.

Full figure and legend (71K)

We then examined the hypothesis that the - 174 IL-6 genotype, BMI, or interactions among them contributed to prevalence of diabetes in men. The prevalence of diabetes among men for each genotype tended to increase with an increase in the number of C alleles (8.20% for the GG genotype, 9.57% for the GC genotype, and 9.92% for the CC genotype), although this trend was not statistically significant. Analysis of the interaction between the IL-6 genotype and BMI revealed that the CC genotype was associated with a 30% greater prevalence for each unit increase in BMI [ OR, 1.30 per unit increase in BMI; 95% confidence interval (CI), 1.11 to 1.50] . In contrast, the prevalence of diabetes for GG (OR, 1.10 per unit increase in BMI; 95% CI, 0.98 to 1.22) and GC genotypes (OR per unit increase in BMI, 1.05; 95% CI, 0.97 to 1.14) was less affected by an increase in BMI. Collectively, it appears that much of the increased prevalence in obesity-associated diabetes in men occurs within the - 174 IL-6 CC genotype group.

Finally, we examined whether the effect of the - 174 IL-6 genotype was mediated through plasma IL-6 levels. Although BMI was significantly and positively associated with IL-6 levels (p < 0.0001), the - 174 IL-6 genotype was not (p = 0.84) as previously reported (12), suggesting that the effect we observed of the - 174 IL-6 genotype on IR was not directly explained by systemic IL-6 levels. We also tested whether systemic IL-6 levels and the - 174 IL-6 genotype were independent predictors of IR. We found that BMI (p < 0.0001), log(IL-6 level) (p = 0.0002), - 174 IL-6 genotype (p = 0.018), and - 174 IL-6 genotype by BMI interaction (p = 0.009) were all independent predictors of log (HOMA-IR) when incorporated in the same model. Taken with our other results, these findings suggest that the - 174 IL-6 genotype-BMI interaction influences HOMA-IR in a manner distinct from any influence that IL-6 levels may have on risk of diabetes itself.

Discussion

We found that the IL-6 - 174 genotype modifies the association of BMI with surrogate measures of IR in men in a community-based cohort study, where homozygotes for the minor C allele were the most insulin resistant and had the greatest prevalence of diabetes. These data demonstrate a gene-environment interaction that may help explain how obesity increases the risk of IR and diabetes in some, but not all, individuals (1).

The effects of interaction between the - 174 genotype and BMI were observed only in men in this sample. By their nature, interaction outcomes will depend on the magnitude of the quantitative variables involved. Thus, it is reasonable that the sex-specific differences we observed arose from a narrower range of differences in underlying trait values among women compared with men. Men tended to be heavier, consume more alcohol, be more active, and have a higher level of fasting insulin and glucose and a higher prevalence of diabetes than women (Table 1). Sex-specific effects may also arise from differences in physiology between men and women (8). The - 174 region of the IL-6 promoter is subject to negative regulation by androgens and estrogens, although these effects do not involve a direct interaction of their cognate receptors with the IL-6 promoter (24, 25). The effects seem to be mediated through other transcription factors such as the CCAAT/enhancer-binding protein and SMAD4 interaction (9). Sex-specific effects of these enhancer binding proteins have been reported in some developmental systems (25). In women, these differences in regulation could limit our ability to detect the gene-obesity interaction, similar to those observed in men.

It is possible that depending on the sample chosen, the sex-specific composition and the distribution of BMI could confound the analysis, leading to the finding of different associations between IL-6 genotype and insulin sensitivity. The finding in Native American and Spanish white populations (9) of increased diabetic risk with the GG genotype contrasts with the increased insulin sensitivity in our data and that previously reported in a Finnish population (10). The different outcomes may, in part, be due to the remarkable differences in BMI among the study populations (for example, in the Finnish Study, the average BMI of GG genotypes is reported as 25.3 4.7 kg/m²; in our study, it is 28.32 4.17 kg/m² for men, whereas in the Pima population, the average BMI is 37.9 16.6 kg/m²). It is also possible that the Pima study was confounded by population admixture because full-heritage Native Americans have only the G allele at the - 174 IL-6 locus. Indeed, in a Pima Indian study, less conclusive results were obtained when the association between the GG genotype and diabetes was analyzed using transmission disequilibrium tests (9). Interestingly, in a German Study of elderly individuals (12), the increased risk of diabetes due to the G allele relative to the C allele was in leaner subjects (BMI < 28.7 kg/m²) in the region where we find HOMA-IR to be slightly greater for the GG genotype relative to the CC genotype (Figure 1A).

The mechanism of increased IR in this study cannot be addressed with available data. It is likely that increased BMI leads to increased IL-6 plasma levels and that this, in turn, would lead to increased IR (3, 4, 5, 6). However, we and others (12) find that systemic IL-6 levels are not predicted by the - 174 IL-6 genotype; rather, IL-6 levels and the - 174 IL-6 genotype are both independent predictors of HOMA-IR. A number of caveats apply to interpreting these results. We do not currently know whether average, peak, or inflammatory levels of serum IL-6 are most relevant to the development of IR and diabetes or whether each of these variables contributes independently to risk (26). It is likely that IR and risk of diabetes are influenced by a number of different genetic elements in inflammatory pathways, of which the - 174 IL-6 genotype is only one.

Our results suggest that there is an interaction among BMI and the - 174 IL-6 promoter polymorphism on metabolic risk that is independent of IL-6 levels and associated correlates of increased BMI. This situation might arise if IL-6 concentration in adipose tissue is an important variable causing IR, reflecting an autocrine rather than endocrine mechanism (26). In some cases, IL-6 may act indirectly through tissue-specific mediators that produce only local effects on glucose uptake (12). It is also possible that the activity of the - 174 IL-6 promoter polymorphism differs in particular compartments of adipose tissue. For example, subcutaneous fat may be a more important determinant of systemic IL-6 levels than visceral fat, yet be less affected by the - 174 IL-6 genotype than those pathways in visceral fat important to the development of diabetes. In any case, in our analysis, the - 174 IL-6 variant modified the association of both BMI and waist circumference with IR, whereas plasma IL-6 levels were associated with IR and diabetes risk independent of adiposity measure and IL-6 genotype.

Our results indicate that the increase in prevalence of diabetes associated with obesity is modified by the - 174 IL-6 genotype. Thus, increasing weight induces greater IR in men with a - 174 IL-6 CC genotype than others. In this context, a G allele is protective. A recently published study using a nested case-control design based on 188 type 2 diabetes cases supports our findings (27). For individuals with a BMI greater than or equal to 28 kg/m², Möhlig et al. have reported (27) that the CC genotype conferred a 5-fold increased risk of developing disease. Both studies underscore the importance of gene-environment interactions in complex diseases such as diabetes. Although it remains possible that underlying population stratification could have influenced our results, such effects are not likely to be a major source of bias in population-based studies, including the FHS ((28;) Wilk JB, Manning AK, Dupuis J, Cupples LA, unpublished data). To the extent that obesity represents an environmental influence on phenotypic expression of diabetes susceptibility genes, our findings provide evidence of a gene-environment interaction on risk of IR and diabetes. In men, it may be that those with the - 174 IL-6 CC genotype may especially benefit from targeted weight loss regimens to improve metabolic risk. Clinical trials and additional investigations in other populations are warranted to confirm our findings.

Notes

¹ Nonstandard abbreviations: IR, insulin resistance; IL, interleukin; FHS, Framingham Heart Study; FPG, fasting plasma glucose; HOMA, homeostasis model assessment; ISI, insulin sensitivity index; OR, odds ratio; CI, confidence interval.

References

1. Reaven, GM. (2003) Importance of identifying the overweight patient who will benefit the most by losing weight. Ann Intern Med. 138: 420–423.
2. Mora, S., Pessin, JE. (2002) An adipocentric view of signaling and intracellular trafficking. Diabetes Metab Res Rev. 18: 345–356. | Article | PubMed | ChemPort |
3. Esposito, K., Pontillo, A., Di Palo, C., et al (2003) Effect of weight loss and lifestyle changes on vascular inflammatory markers in obese women: a randomized trial. JAMA 289: 1799–1804. | Article | PubMed | ISI | ChemPort |
4. Fernandez-Real, J. M., Broch, M., Vendrell, J., et al (2000) Interleukin-6 gene polymorphism and insulin sensitivity. Diabetes 49: 517–520. | PubMed | ISI | ChemPort |
5. Senn, J. J., Klover, P. J., Nowak, I. A., Mooney, RA. (2002) Interleukin-6 induces cellular insulin resistance in hepatocytes. Diabetes 51: 3391–3399. | PubMed | ChemPort |
6. Vozarova, B., Weyer, C., Hanson, K., Tataranni, P. A., Bogardus, C., Pratley, RE. (2001) Circulating interleukin-6 in relation to adiposity, insulin action, and insulin secretion. Obes Res. 9: 414–417. | PubMed | ISI | ChemPort |
7. Pradhan, A. D., Manson, J. E., Rifai, N., Buring, J. E., Ridker, PM. (2001) C-reactive protein, interleukin 6, and risk of developing type 2 diabetes mellitus. JAMA 286: 327–334. | Article | PubMed | ISI | ChemPort |
8. Terry, C. F., Loukaci, V., Green, FR. (2000) Cooperative influence of genetic polymorphisms on interleukin 6 transcriptional regulation. J Biol Chem. 275: 18138–18144. | Article | PubMed | ISI | ChemPort |
9. Vozarova, B., Fernandez-Real, J. M., Knowler, W. C., et al (2003) The interleukin-6 (- 174) G/C promoter polymorphism is associated with type-2 diabetes mellitus in Native Americans and Caucasians. Hum Genet. 112: 409–413.
10. Kubaszek, A., Pihlajamaki, J., Punnonen, K., Karhapaa, P., Vauhkonen, I., Laakso, M. (2003) The C-174G promoter polymorphism of the IL-6 gene affects energy expenditure and insulin sensitivity. Diabetes 52: 558–561. | Article | PubMed | ISI | ChemPort |
11. Kubaszek, A., Pihlajamaki, J., Komarovski, V., et al (2003) Promoter polymorphisms of the TNF-alpha (G-308A) and IL-6 (C-174G) genes predict the conversion from impaired glucose tolerance to type 2 diabetes: the Finnish Diabetes Prevention Study. Diabetes 52: 1872–1876. | Article | PubMed | ChemPort |
12. Illig, T., Bongardt, F., Schopfer, A., et al (2004) Significant association of the interleukin-6 gene polymorphisms C-174G and A-598G with type 2 diabetes. J Clin Endocrinol Metab. 89: 5053–5058.
13. Kannel, W. B., Feinleib, M., McNamara, P. M., Garrison, R. J., Castelli, WP. (1979) An investigation of coronary heart disease in families: The Framingham Offspring Study. Am J Epidemiol. 110: 281–290. | PubMed | ISI | ChemPort |
14. Meigs, J. B., Cupples, L. A., Wilson, PW. (2000) Parental transmission of type 2 diabetes: the Framingham Offspring Study. Diabetes 49: 2201–2207.
15. Meigs, J. B., Nathan, D. M., Wilson, P. W., Cupples, L. A., Singer, DE. (1998) Metabolic risk factors worsen continuously across the spectrum of nondiabetic glucose tolerance: the Framingham Offspring Study. Ann Intern Med. 128: 524–533.
16. Vasan, R. S., Sullivan, L. M., Roubenoff, R., et al (2003) Inflammatory markers and risk of heart failure in elderly subjects without prior myocardial infarction: the Framingham Heart Study. Circulation 107: 1486–1491. | Article | PubMed | ISI | ChemPort |
17. Bonora, E., Targher, G., Alberiche, M., et al (2000) Homeostasis model assessment closely mirrors the glucose clamp technique in the assessment of insulin sensitivity: studies in subjects with various degrees of glucose tolerance and insulin sensitivity. Diabetes Care 23: 57–63. | Article | PubMed | ISI | ChemPort |
18. Matthews, D. R., Hosker, J. P., Rudenski, A. S., Naylor, B. A., Treacher, D. F., Turner, RC. (1985) Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 28: 412–419. | Article | PubMed | ISI | ChemPort |
19. McAuley, K. A., Williams, S. M., Mann, J. I., et al (2001) Diagnosing insulin resistance in the general population. Diabetes Care 24: 460–464. | Article | PubMed | ISI | ChemPort |
20. Gutt, M., Davis, C. L., Spitzer, S. B., et al (2000) Validation of the insulin sensitivity index (ISI(0,120)): comparison with other measures. Diabetes Res Clin Pract. 47: 177–184. | Article | PubMed | ChemPort |
21. Meigs, J. B., Panhuysen, C. I., Myers, R. H., Wilson, P. W., Cupples, LA. (2002) A genome-wide scan for loci linked to plasma levels of glucose and HbA(1c) in a community-based sample of Caucasian pedigrees: The Framingham Offspring Study. Diabetes 51: 833–840. | Article | PubMed | ChemPort |
22. Rousseeuw, P. J., Leroy, AM. (1987) Robust Regression and Outlier Detection Wiley New York.
23. Hanley, A. J., Williams, K., Gonzalez, C., et al (2003) Prediction of type 2 diabetes using simple measures of insulin resistance: combined results from the San Antonio Heart Study, the Mexico City Diabetes Study, and the Insulin Resistance Atherosclerosis Study. Diabetes 52: 463–469. | Article | PubMed | ISI | ChemPort |
24. Manolagas, S. C., Kousteni, S., Jilka, RL. (2002) Sex steroids and bone. Recent Prog Horm Res. 57: 385–409. | Article | PubMed | ISI | ChemPort |
25. Wu, L., Wu, Y., Gathings, B., et al (2003) Smad4 as a transcription corepressor for estrogen receptor alpha. J Biol Chem. 278: 15192–15200. | Article | PubMed | ISI | ChemPort |
26. Papanicolaou, D. A., Vgontzas, AN. (2000) Interleukin-6: the endocrine cytokine. J Clin Endocrinol Metab. 85: 1331–1333. | PubMed |
27. Möhlig, M., Boeing, H., Spranger, J., et al (2004) Body mass index and C-174G interleukin-6 promoter polymorphism interact inpredicting type 2 diabetes. J Clin Endocrinol Metab. 89: 1885–1890. | Article | PubMed | ChemPort |
28. Cardon, L. R., Palmer, LJ. (2003) Population stratification and spurious allelic association. Lancet 361: 598–604. | Article | PubMed | ISI |

Acknowledgments

This work was supported by the National Heart, Lung and Blood Institute's FHS (Contract N01-HC-25195), NIH Grants HL64753 and HL076784 (Emelia J. Benjamin, Principal Investigator), and by a grant from the American Diabetes Association (A. Herbert, Principal Investigator). J.B.M. is supported by an American Diabetes Association Career Development Award. We thank the staff of the FHS Genetics Laboratory for their careful work.