| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
General Internal Medicine and Clinical Epidemiology Units, General Medicine Division, Department of Medicine, Massachusetts General Hospital and Harvard Medical School (J.B.M.), Boston, Massachusetts 02114; Department of Biostatistics, Boston University School of Public Health (J.D., L.A.C.), Boston, Massachusetts 02118; Framingham Heart Study Genetics Laboratory, Department of Neurology, Boston University School of Medicine (A.G.H., C.L., P.W.F.W.), Boston, Massachusetts 02118; and Framingham Heart Study (P.W.F.W.), Framingham, Massachusetts 01702
Address all correspondence and requests for reprints to: Dr. James B. Meigs, General Medicine Division, Massachusetts General Hospital, 9th Floor, 50 Staniford Street, Boston, Massachusetts 02114. E-mail: jmeigs{at}partners.org.
 |
Abstract |
|---|
 Top Abstract IntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
|
Subjects and Methods |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
The FHS began in 1948 with a randomly selected cohort of 5209 men and women, aged 30–62 yr, residing in Framingham, MA. Essentially all participants were Caucasian and of mixed European ethnic descent. Participants were examined biennially for the interval development of cardiovascular diseases and their risk factors, including diabetes. For this analysis, the original cohort (the parents) was followed through 1993 (the 22nd examination) after a mean duration of 46 yr of follow-up, when the mean age of surviving parents was 83 yr.
In 1971, 5124 offspring of the original cohort and the spouses of the offspring, aged 12–58 yr at the first examination, were enrolled in the Framingham Offspring Study (11). Subjects were next examined 8 yr after the baseline exam and then every 4 yr from exam 2 through exam 7 (1999–2001). At each exam, subjects provided a medical history and underwent physical and fasting laboratory examination. Of the offspring, 1811 randomly selected unrelated individuals provided genotype data; of these, 883 had completed up to 26 yr of follow-up by the seventh offspring exam, had complete phenotypic data and information on diabetes status for at least one parent, and were included in the analysis of unrelated subjects. We also assessed a separate sample of 282 nuclear families with 116 genotyped parents (24 families with both parents genotyped, 68 families with only one parent genotyped, and 190 families with no parental genotype data). In these 282 nuclear families, there were 11 trios (one sibling and both genotyped parents), 168 sibling pairs, 72 sibling trios, and 31 sibships with four or more siblings, for a total of 698 offspring. Of these families, there were 125 informative nuclear families (at least one heterozygous parent or at least one sibling with a different genotype than the others) with 55 parents genotyped (16 families with both parents genotyped, 23 families with one parent genotyped, and 86 families with no parental data), including seven trios, 60 sibling pairs, 39 sibling trios, and 19 families with four siblings or more, for a total of 323 offspring. Every study subject provided written informed consent at every examination, including consent for genetic analyses, and the study was approved by the Boston University institutional review board.
Genotyping
We assessed the INS –23Hph1 polymorphism, which is in virtually complete (>99.5% concordance in Caucasian populations) linkage disequilibrium (LD) with the INS VNTR (4). Participants were genotyped for INS –23Hph1 using a mass spectroscopy-based single nucleotide polymorphism (SNP) detection assay (Sequenom, San Diego, CA) (12). Primers were supplied by Applied Biosystems, Inc. (Foster City, CA). PCR was performed at 50 C for 2 min and at 95 C for 10 min, then 45 cycles consisting of 95 C for 15 sec and 60 C for 1 min.
Diabetes phenotypes
We defined parental diabetes as treatment with insulin or oral hypoglycemic medications or a casual plasma glucose level of 11.1 mmol/liter or more at any examination, or a plasma glucose level of 11.1 mmol/liter or more 1 h after a 50-g oral glucose tolerance test administered at cohort examination 10. Fasting plasma glucose levels were never assessed in the parents. We defined offspring diabetes as treatment with insulin or oral hypoglycemic medications or a fasting plasma glucose level of 7.0 mmol/liter or more at two or more of seven examinations (13). The age of onset of offspring diabetes was defined as the age at the examination at which diabetes was first diagnosed. For offspring who presented with diabetes at the first examination, chart review was performed to confirm the diagnosis of type 2 diabetes and to accurately establish the age at onset for use in incidence analyses. More than 99% of diabetes in the offspring is type 2 diabetes (10). Fasting plasma glucose was measured at each exam with a hexokinase reagent kit (A-gent glucose test, Abbott Laboratories, Inc., South Pasadena, CA). Glucose assays were performed in duplicate; assay coefficients of variation (CVs) were less than 3%. For each subject we also calculated the mean fasting plasma glucose measured at each exam to estimate the 26-yr, time-averaged glucose level. Fasting plasma insulin was measured at exam 7, with a specific insulin level having essentially no cross-reactivity to insulin split-products (Linco Research, Inc., St. Charles, MO); assay CVs were less than 6.8%. We used the homeostasis model as a surrogate measure of insulin resistance, where homeostasis model assessment of insulin resistance = (fasting insulin x fasting glucose)/22.5 (14). Hemoglobin A1c was measured by HPLC, with a mean ± SD level among nondiabetic subjects of 5.22 ± 0.6%, and assay CVs were less than 2.5%. At each exam, height and weight were measured, and body mass index (BMI) was expressed in kilograms of body weight divided by height in meters squared.
Statistical analysis
We conducted analyses of the association of INS –23Hph1 A/T and incident diabetes in the unrelated sample using the software R (www.r-project.org), and considered P < 0.05 to indicate statistical significance. Hazard ratios for incident diabetes associated with the –23Hph1 SNP or parental diabetes were calculated with proportional hazards regression models, using the age of onset of diabetes and the age at the last exam attended as the time failure and censoring variables, respectively. Outcomes included all cases of offspring diabetes through exam 7, and nested regression models included terms for age in years at baseline (exam 1), offspring sex, BMI measured at baseline (exam 1), –23Hph1 AT and TT genotypes, maternal diabetes, paternal diabetes, and sex by maternal diabetes and sex by paternal diabetes first-order interactions. We used similarly parameterized linear regression models to estimate associations of diabetes-related quantitative traits with INS –23Hph1 genotypes. We used the hazard ratio from unadjusted models and the –23 SNP allele frequencies to calculate the population-attributable risk percentage (PAR%) associated with risk genotypes as: PAR% = (prevalencegenotype x (hazard ratiogenotype – 1)) /(prevalencegenotype x (hazard ratiogenotype – 1) + 1) x 100.
We inferred parental diabetes for 229 parents with this phenotype missing by assigning the sample-based probability of diabetes among spouses to parents with missing information. In other words, rather than having an affectation status of 0 or 1, some parents were assigned a fractional affectation status, reflecting their probability of having diabetes based on spouse prevalence data, where the prevalence of diabetes in husbands of women with diabetes was 17.1% and among women without diabetes was 11.6%; the prevalence of diabetes in wives of men with diabetes was 15.2%, and that among men without diabetes was 10.2%. Inference was used only in cases where one parent was missing phenotype information; no inference was used if both parents were missing phenotypic information. Of fathers, 745 had known and 138 had inferred phenotypes; of mothers, 792 had known and 91 had inferred phenotypes. Analyses conducted excluding parents with inferred phenotypes gave similar results as analyses including these subjects; therefore, only results of the latter analyses are shown.
We also applied the family-based association test implemented in FBAT to the diabetes phenotype (15, 16). This approach tests for the association between the phenotype and excess transmission of a specific allele from parent to offspring. Under the null hypothesis of no association, alleles are passed from parent to offspring according to Mendelian law. However, if an allele increases disease susceptibility or the value of a quantitative trait of interest, one would expect to see an excess number of the risk allele in affected individuals or in individuals with high trait values. By conditioning on the parental genotypes when available or the minimally sufficient statistic when parental genotypes are not available, the test is not affected by population admixture. We used the FBAT option –o to compute an offset value to minimize the variance and increase analytic power (17). To adjust for covariates, logistic regression was applied, and the deviance residuals from the model were analyzed as continuous traits with FBAT (18).
In addition, we implemented the FBAT proposed by Li and Fan (19) to test for association between the INS polymorphism and diabetes age of onset. In this approach, the association of excess transmission of a specific allele from parent to offspring with diabetes age of onset is tested using a Cox proportional hazard model. The genotype of each individual is decomposed into two orthogonal components: the average number of minor alleles in the pedigree and the excess number of minor alleles that an individual carries; both components are then incorporated into the model, where the regression coefficient for the excess number of minor alleles provides an estimate of the association between diabetes age of onset and the polymorphism that is valid in the presence of population admixture. Exponentiation of the regression coefficient from this model provides a hazard ratio that is interpreted as the increased hazard per unit of excess T alleles transmitted over what would be expected by chance transmission. In addition, the approach allows for the incorporation of covariates into the model. Correlation between siblings is taken into account using a robust variance estimate (20).
|
Results |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
. INS –23Hph1 alleles were in Hardy-Weinberg equilibrium, and the frequency of the –23HpH1 TT genotype was 8.0%, similar to the approximately 8% INS VNTR class III/III genotype frequency reported in other Caucasian samples (4, 21). About half the subjects were women, and the distributions of sex, age, and BMI were similar across –23Hph1 genotypes. About 13% of offspring had a father and 13% a mother with diabetes; this proportion was similar across genotypes for fathers, but tended to increase with an increasing number of –23Hph1 T alleles for mothers (P = 0.06). Offspring subjects were followed over the entire course of the FHS to date, contributing 53,031 cumulative lifetime years of observation and 92 incident cases of type 2 diabetes, giving a mean lifetime rate until an average of 61 yr of age (range, 33–88 yr) of 1.7 cases/1,000/yr.
|
and Fig. 1). AT heterozygotes were not at increased risk of diabetes. Subjects with maternal or paternal diabetes had about a 2-fold increased age- and sex-adjusted hazard ratio for diabetes (model 2; P 
0.01). Obesity may be a confounder or may be in the causal pathway linking the INS VNTR to diabetes; after additional adjustment for BMI, the –23Hph1 TT genotype was associated with a smaller 71% (P = 0.09) increased hazard of diabetes. We then adjusted for parental diabetes status to test the hypothesis that maternal diabetes confers an imprinting effect on INS VNTR. If this were so, we expected to see a weakening of both the A/T association and the paternal diabetes association with offspring diabetes, because fathers are hypothesized to be the source of the functional class III alleles, so including paternal diabetes and –23HpH1 A/T in the same model should introduce collinearity that would weaken the parental and genetic effects. Likewise, we expected to see maintenance of the maternal association, because maternal class III alleles are hypothesized to be inactivated. Instead, we observed a maintenance of the paternal effect and a weakening of the maternal effect (model 4; Table 2) while the homozygote TT effect remained only slightly weakened, conferring a borderline significant (P = 0.06) 80% increased hazard ratio for offspring diabetes.
|
|
) suggested a stronger association of –23Hph1 TT with incident diabetes among women than men. Consequently, we computed sex-specific hazard ratios, finding that women with –23Hph1 TT had an approximately 4-fold increased relative hazard of diabetes (model 4w; P = 0.001) compared with women with AA, but men with –23Hph1 TT were not at increased risk of diabetes (model 4m). Despite these observed differences, the statistical significance of the term for offspring sex by TT genotype interaction was borderline at P = 0.07 in age-, sex-, BMI-, and parental diabetes-adjusted models.
In these sex-stratified models (models 4w and 4m; Table 2), men with maternal diabetes and women with paternal diabetes were at increased risk for diabetes, suggesting offspring sex by maternal or offspring sex by paternal diabetes interactions. Consequently, in an attempt to obtain the most accurate estimate of the hypothesized additive effect of –23HpH1 A/T on diabetes risk, we fit a model with terms for age, sex, BMI, maternal diabetes, paternal diabetes, and offspring sex by maternal diabetes and offspring sex by paternal diabetes interaction terms. P values for these two interaction terms were 0.02 and 0.04, respectively, and the –23Hph1 TT genotype was associated with a significantly increased hazard ratio for incident diabetes [hazard ratio, 1.89; 95% confidence interval (CI), 1.01–3.52; P = 0.045]. Based on a hazard ratio of 1.89 and a genotype frequency of 0.08, the population-attributable risk percentage associated with the –23Hph1 TT genotype was 6.6%.
To confirm findings in the unrelated sample and to assess whether the diabetes –TT genotype association was a function of population admixture, we examined the association of the –23Hph1 A/T polymorphism with type 2 diabetes in a separate sample of 282 nuclear families using a generalized transmission disequilibrium test (FBAT) to test for association (Table 3) (16). The z-score for the INS –23Hph1 T allele was positive and statistically significant (P 0.04), indicating that the T allele was associated with increased risk of incident diabetes relative to the A allele. Additional adjustment for BMI at baseline weakened the association (P = 0.09). FBAT treats diabetes as a prevalent present/absent phenotype. We also assessed the incidence of diabetes using age of onset as the time failure variable and tested for association between the INS polymorphism and diabetes using an implementation of the LD-based proportional hazards model and robust score test proposed by Li and Fan (19). The hazard ratios from crude, age- and sex-adjusted, and age-, sex-, and BMI-adjusted tests of the diabetes-T allele association were positive and significant (P 0.03; Table 3). Thus, two different analytic strategies in the family samples support an association of the –23Hph1 T allele and increased risk of type 2 diabetes that was not attributable to population admixture.
|
|
|
Discussion |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
It is possible that associations of diabetes with the TT genotype could be due to a variant other than the INS VNTR, for instance, with the –23Hph1 itself or with another variant in LD with –23Hph1. Another mechanism for the apparent diabetogenic effect of the INS VNTR could be LD with the IGF-II gene, which lies less than 5 kb downstream from INS, is a known mediator of fetal development, and is suspected to be associated with adult diseases, including type 2 diabetes (27, 30). In Pima Indians, the –23Hph1 T allele was associated with lower birth weight in simple association tests as well as in analyses that accounted for population admixture (8). In the analysis by Ong et al. (5), the effect of the class III/III genotype was substantially strengthened (from an 
40% to an 400% increased risk relative to class I alleles) when birth weight and early postnatal growth were accounted for in the analysis. Birth weight data are not available in the FHS, so we were unable to assess these effects in our analyses.
In other studies, imprinting and parent of origin effects have appeared to be important modulators of INS expression and associated risk of type 2 diabetes. An analysis of British trios found an excess risk for type 2 diabetes associated with class III alleles transmitted from heterozygous (class I/III) fathers (6). Father-specific transmission effects have also been reported in Pima Indians (9). In a prior analysis of FHS participants, we found an elevated risk of diabetes in offspring of diabetic mothers that was equivalent to risk in offspring of diabetic fathers; however, most of the maternal-associated risk occurred in mothers who had diabetes during child-bearing years (10). If some part of the risk of diabetes associated with maternal diabetes arises from fetal exposure (31), then one explanation for our previous findings was maternal imprinting on one or more genetic loci. However, the present analysis does not provide support for the imprinting hypothesis. Apart from the finding that controlling for paternal diabetes did not strongly attenuate –23HpH1 A/T effects, our data suggest that increased risk was associated only with TT homozygosity (corresponding to the class III/III genotype). Ong et al. (5) also observed this homozygous TT effect; this finding argues against the imprinting hypothesis, where disease is expected to occur when the expression of one parental allele is suppressed. Finally, our unrelated data were too sparse to obtain a stable estimate of effects of parental diabetes by –23Hph1 A/T genotype interactions, and in the family sample there were too few trios to identify the parental source of excess T alleles in affected offspring. Thus, our data do not provide support for the hypothesis that maternal imprinting on or near INS is involved in diabetes transmission.
There are additional limitations of our study. The observations of increased diabetes risk only in offspring women associated with –23Hph1 TT, increased diabetes risk in offspring women associated only with paternal diabetes, and increased diabetes risk in offspring men associated only with maternal diabetes were not expected a priori, because we did not observe significant offspring sex by parental diabetes interactions in our prior study of parental diabetes transmission in this population (10). Discussion of probable causes or mechanisms of these effects would be purely speculative, and the findings must be considered as hypothesis-generating observations to be confirmed by other data. However, the observation of increased diabetes risk in offspring women associated only with paternal diabetes may suggest the hypothesis that an X-chromosome locus modulates the effect of the INS VNTR on diabetes risk or that some other hormone-dependent pathway may modulate the effect. Finally, our analysis is based on a relatively small number of diabetes cases. That we found an association in the FHS sample when other studies have not always found an association suggests that our estimate of relative hazards (and, consequently, of population attributable risk) associated with the –23Hph1 TT genotype may be spuriously high. However, our finding of significant excess allele sharing in the family-based sample supports the conclusion that variation at –23Hph1 has at least a small, positive effect on the risk of incident type 2 diabetes.
In summary, our analysis of separate unrelated and family-based samples from the population-based FHS support the hypothesis that the INS –23Hph1 TT genotype (corresponding to the INS class III/III VNTR) is a genetic risk factor for incident development of type 2 diabetes in Caucasians. The INS VNTR appears to be a modest genetic risk factor, with the –23Hph1 TT genotype accounting for about 6% of diabetes prevalence in the FHS population.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
First Published Online November 23, 2004
Abbreviations: BMI, Body mass index; CI, confidence interval; CV, coefficient of variation; FBAT, family-based association test; LD, linkage disequilibrium; SNP, single nucleotide polymorphism; VNTR, variable number tandem repeat.
Received June 24, 2004.
Accepted November 10, 2004.
|
References |
|---|
|
TopAbstractIntroductionSubjects and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  E. Liu, J. B. Meigs, A. G. Pittas, N. M. McKeown, C. D. Economos, S. L. Booth, and P. F. Jacques Plasma 25-Hydroxyvitamin D Is Associated with Markers of the Insulin Resistant Phenotype in Nondiabetic Adults J. Nutr., February 1, 2009; 139(2): 329 - 334. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M.-F. Hivert, A. K. Manning, J. B. McAteer, J. C. Florez, J. Dupuis, C. S. Fox, C. J. O'Donnell, L. A. Cupples, and J. B. Meigs Common Variants in the Adiponectin Gene (ADIPOQ) Associated With Plasma Adiponectin Levels, Type 2 Diabetes, and Diabetes-Related Quantitative Traits: The Framingham Offspring Study Diabetes, December 1, 2008; 57(12): 3353 - 3359. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. B. Meigs, P. Shrader, L. M. Sullivan, J. B. McAteer, C. S. Fox, J. Dupuis, A. K. Manning, J. C. Florez, P. W.F. Wilson, R. B. D'Agostino Sr., et al. Genotype Score in Addition to Common Risk Factors for Prediction of Type 2 Diabetes N. Engl. J. Med., November 20, 2008; 359(21): 2208 - 2219. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. J. Gunter, R. B. Hayes, N. Chatterjee, M. Yeager, R. Welch, R. E. Schoen, L. Yakochi, A. Schatzkin, and U. Peters Insulin Resistance-Related Genes and Advanced Left-Sided Colorectal Adenoma Cancer Epidemiol. Biomarkers Prev., April 1, 2007; 16(4): 703 - 708. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. Landmann, F. Geller, J. Schilling, S. Rudloff, E. Foeller-Gaudier, and L. Gortner Absence of the Wild-type Allele (192 Base Pairs) of a Polymorphism in the Promoter Region of the IGF-I Gene but Not a Polymorphism in the Insulin Gene Variable Number of Tandem Repeat Locus Is Associated With Accelerated Weight Gain in Infancy Pediatrics, December 1, 2006; 118(6): 2374 - 2379. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. Bougneres Genotypic and phenotypic complexity at the insulin variable number of tandem repeats locus. J. Clin. Endocrinol. Metab., November 1, 2006; 91(11): 4246 - 4249. [Full Text] [PDF]
|
|
|
|
|
|
N. Santoro, G. Cirillo, A. Amato, C. Luongo, P. Raimondo, A. D'Aniello, L. Perrone, and E. Miraglia del Giudice Insulin Gene Variable Number of Tandem Repeats (INS VNTR) Genotype and Metabolic Syndrome in Childhood Obesity J. Clin. Endocrinol. Metab., November 1, 2006; 91(11): 4641 - 4644. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. Tuomi Type 1 and Type 2 Diabetes: What Do They Have in Common? Diabetes, December 1, 2005; 54(suppl_2): S40 - S45. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. L. Fur, D. Fradin, P. Boileau, and P. Bougneres Association of Kir6.2 and INS VNTR variants with glucose homeostasis in young obese Physiol Genomics, August 11, 2005; 22(3): 398 - 401. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Endocrinology | Endocrine Reviews | J. Clin. End. & Metab. |
| Molecular Endocrinology | Recent Prog. Horm. Res. | All Endocrine Journals |
