The total study population consisted of 2415 participants (1671 offspring; 744 controls). Baseline characteristics are shown in Table 2. The group of offspring was slightly older compared to their partners. The two groups were comparable with respect to measures of height, weight and body mass index, both crude and after adjustment for age and sex. Offspring from nonagenarian siblings had a lower prevalence of diabetes mellitus and hypertension as well as a slightly lower prevalence of myocardial infarction. In accordance with the lower prevalence of diabetes, offspring had lower glucose levels (P<0.001) and lower levels of insulin (P=0.006). After exclusion of all participants with diabetes, the association remained significant for glucose (P=0.001), but became non-significant for insulin (P=0.209).

We observed an increasing prevalence of diabetes mellitus with an increasing number of T2D risk alleles (P = 0.011, Table 3). We also tested the association between numbers of risk alleles and serum parameters of glucose metabolism in the total group. With increasing numbers of risk alleles, we found an increase in glucose (P= 0.016) but not insulin (P=0.450). We found that the number of risk alleles was not associated with body mass index (BMI), and repeating the analyses with adjustment for BMI did not materially change the results (data not shown). After exclusion of participants with DM, statistical significance was lost for the association of number of alleles with levels of glucose (P = 0.089). In the oral glucose tolerance test increasing number of alleles associated with increasing area under the curve for glucose (P = 0.018).

Next we tested the hypothesis that differences in allele frequencies of these SNPs could explain the observed difference in prevalence of DM and differences in glucose and insulin levels between offspring and controls (Table 4). For none of the SNPs, the allele frequency was significantly different between offspring from familial nonagenarians compared to their partners. Likewise, no differences were found in the mean number of T2D risk alleles between the groups of offspring and controls (14.5 vs. 14.5 respectively, P=0.977).

To assess whether the offspring of nonagenarian sibling pairs were more protected against the influences of the risk alleles than the controls, we compared the increase in glucose dependent on the number of risk alleles in the offspring and controls. When analyzing offspring and controls separately, levels of glucose increased with an increasing number of number of risk alleles in offspring (P=0.016) and also in controls, albeit not significantly (P= 0.369). The increase in the control group did not reach statistical significance, possibly because of the smaller size of the group. The increase in glucose levels dependent on number of type 2 diabetes risk alleles was not different between offspring and partners (P for interaction = 0.538). A similar finding was found for the area under the curve in the oral glucose tolerance test (Figure 1). In both offspring and controls a similar trend was seen, albeit not statistically significant (P = 0.093 and P = 0.159 respectively) due to small samples size. There was no significant interaction (P = 0.797).

Nonagenarian sibling pairs were included when aged older than 89 years for men and 91 years for women and having at least one sister or a brother fulfilling these age criteria, who was also willing to participate. Because proper controls are lacking at very high ages, the offspring of the nonagenarian siblings were asked to be included in the study as well. The partners thereof were included in the study to serve as a control group, representing the general population at an age comparable to the offspring. The total study population, excluding the nonagenarian siblings, consisted of 2415 participants (1671 offspring; 744 partners). The Medical Ethical Committee of the Leiden University Medical Centre approved the study and informed consent was obtained from all participants.

Blood samples were taken at baseline for extraction of DNA and the determination of non-fasted serum parameters. Glucose and insulin were available for 2337 and 2287 participants respectively. We collected additional information and biomaterials from the generation of offspring and partners, including self-reported information on life style and bodily measures. Body height and weight were obtained from 1670 participants. Body mass index was calculated from these data. Information on medical history was requested from the participants' general practitioners. In a subgroup of 234 offspring and their partners we performed an oral glucose tolerance test. We calculated the area under the curve for each individual as a measure of glucose tolerance [⁵].

Selection of Polymorphisms. We reviewed 266 GWAS that were published up to February 2009 (http://www.genome.gov/26525384), 10 of which reported on type-2 diabetes. These 10 GWAS reported 13 loci to be associated with type 2 diabetes (T2D) within at least two independent GWAS. To compile a set of disease risk alleles for each locus the most replicated SNP was selected and subsequently, in case of equal number of replications, the SNP with the lowest reported p-value. For two loci, one additional SNP was selected that was in low to moderate LD (r² < 0.80) with the most replicated SNP. Thus 15 SNPs were selected for analysis, covering 13 loci (see Table 1).

Genotyping. These 15 SNPs were genotyped using Sequenom iPLEX. The average genotype call rate for these SNPs was 96.9% and the average concordance rate was 99.7% among 128 duplicated control samples. Complete genotyping of all 15 SNPs succeeded in 2090 participants (87%). Because in complex human traits, single mutations are not expected to have large effects and are therefore hard to identify [²⁰], we calculated for each individual the total number of T2D risk alleles. Demographic and antropometric data, disease prevalence and levels of glucose and insulin were compared between these groups. Allele frequencies were comparable to those described in the literature and all genotype distributions were in Hardy-Weinberg equilibrium.

Continuous data were distributed normally except for levels of insulin, which was log transformed. Age, sex and disease prevalence frequencies are reported unadjusted. Means and 95% confidence intervals of antropometrics, glucose and insulin are reported adjusted for age and sex. Differences in disease prevalence between groups were assessed using logistic regression, adjusting for age and sex. Difference in antropometrics and levels of glucose and insulin were calculated using a linear regression model adjusting for age and sex. All p-values for differences between groups were adjusted for family relationships using robust standard errors, except for age and sex, which were calculated crude. Differences were considered significant when the p value was below 0.05. All statistical analyses were performed with STATA (version 10.0, USA) and SPSS (version 16.0, USA), and in all analyses we made use of robust standard errors to account for familial relationships among the offspring.