Genetic Risk Score of 46 Type 2 Diabetes Risk Variants Associates With Changes in Plasma Glucose and Estimates of Pancreatic β-Cell Function Over 5 Years of Follow-Up Open Access

The Inter99 study (clinical trial reg. no. NCT00289237, clinicaltrials.gov) is a population-based nonpharmacological intervention study for ischemic heart disease conducted at the Research Centre for Prevention and Health in Glostrup, Denmark (www.inter99.dk). A randomized sample of 13,016 individuals living in Copenhagen County (30–60 years) was drawn from the Civil Registration System and further prerandomized into high-intensity (90%) and low-intensity (10%) intervention groups. A total of 6,784 (52%) subjects attended the baseline health examination (median age 45 years). All participants received individual lifestyle counseling at the baseline examination, focused on habits of smoking, physical activity, dietary intake, and use of alcohol. The high-intensity group was in addition offered group-based lifestyle counseling if considered at high risk for ischemic heart disease. Follow-up examinations were conducted after 5 years with a participation rate of 66% (n = 4,511) (13–15).

A flowchart of the current study is shown in Fig. 1. Information on incident type 2 diabetes was collected from the Danish National Diabetes Register (16). Information included date of inclusion and criteria for inclusion. Data until 31 December 2010 were available. Information on death and emigration was obtained from the Danish Central Person Register until 31 December 2010.

At the baseline and 5-year follow-up examination, anthropometric and biochemical measures were obtained after an overnight fast. Weight (kilograms) and height (centimeters) were measured in light indoor clothes and without shoes. All participants without previously diagnosed diabetes were characterized by a standardized 75-g OGTT with plasma glucose and serum insulin measured at fasting and 30 and 120 min after the oral glucose load. Plasma glucose was analyzed using the hexokinase/glucose-6-phosphate dehydrogenase technique (Boehringer Mannheim, Germany), which has an intra-assay precision of coefficient of variation (CV) 1.1% and an interassay precision of CV 2.3%. Serum insulin (excluding des-31,32 and intact proinsulin) was measured using the AutoDELFIA insulin kit (Perkin-Elmer/Wallac, Turku, Finland), with an interassay precision of CV <6%. The conditions under which plasma glucose and serum insulin were measured were identical at baseline and at the 5-year follow-up examination.

Screen-detected diabetes was defined according to World Health Organization 1999 criteria (17). All lifestyle factors were estimated from self-reported questionnaire data as previously described (18–20). Written informed consent was obtained from all participants. The study was approved by the regional ethics committee of Copenhagen and is in accordance with the principles of the Declaration of Helsinki II.

Genotyping was performed with the Metabochip (21) using the Illumina HiScan (Illumina, San Diego, CA). Genotypes were called using the genotyping module (version 1.9.4) of GenomeStudio software (version 2011.1; Illumina), and custom cluster data generated from ∼6,000 Danish DNA samples were analyzed on the same Illumina HiScan. Quality control removed 1st- and 2nd-degree related individuals (n = 119), individuals with an extreme inbreeding coefficient (F > 0.1 or F < 0.1, n = 25), individuals with a low genotype call rate (call rate <90%, n = 30), individuals with mislabeled sex (n = 11), and individuals with a high discordance rate to previously genotyped single nucleotide polymorphisms (concordance <80%, n = 65), leaving 6,127 individuals who passed all quality-control criteria. The average call rate for all SNPs on the Metabochip was 99.0%. Of the previously identified type 2 diabetes risk variants, five were not present on the Metabochip and did not have perfect proxies, and thus, these were genotyped by KASPar (KBioscience, Hoddesdon, U.K.) with success rates >96% and error rates <0.5%. All variants obeyed Hardy-Weinberg equilibrium (P > 0.01), except HMGA2, rs1531343 (P = 0.0043). An overview of the 46 variants genotyped is provided in Supplementary Table 1. The 46 variants were chosen based on genome-wide significant (P < 5 × 10⁻⁸) association signals for the risk of type 2 diabetes identified or confirmed in European populations. Variants only found to be associated with type 2 diabetes in Asian populations were not included. The risk variant in DUSP is not included in the current study because of the location on the X-chromosome. Variants in or near FTO and MC4R were not included owing to a primary effect on obesity.

Individuals with more than two missing genotypes were excluded (n = 146). For individuals with one (n = 588) or two (n = 109) missing genotypes, genotypes were imputed by assigning the most common genotype in Inter99 for the missing variants. A simple genetic risk score was constructed by summing up the number of risk alleles over 46 variants for each individual. The median risk score was 50, ranging from 31 to 66. A weighted genetic risk score was created as previously described (22) and by weighting each risk allele with the effect size (the natural log of the odds ratios) for risk of type 2 diabetes reported by the largest meta-analyses performed (1). Effect sizes used for weighting for the specific variants or the best proxies can be seen in Supplementary Table 1. All results are from analyses of the simple genetic risk score. The weighted genetic risk score was used to check whether weighting each allele would change the results obtained for the simple risk score.

Data were analyzed using the STATA statistical software (version 12.1; StataCorp, College Station, TX) and RGui, version 2.13.2 (http://www.r-project.org/). The Kaplan-Meier method was used to plot cumulative incidence of type 2 diabetes against age, and the log-rank trend test was used to test for differences between tertiles of the genetic risk score. Cox proportional hazards regression models were used to address the risk of incident type 2 diabetes. Individuals with self-reported diabetes at the baseline examination as well as individuals present in the Danish National Diabetes Registry before the baseline examination were excluded from the present analyses of incident type 2 diabetes. To automatically adjust for age, we used left truncation and age as a time scale. We tested the assumption of proportional hazards graphically by plotting log(cumulative hazard) as a function of age and with a test based on Schoenfeld residuals. We detected no major violations of the proportional hazards assumption.

Paired t tests were used to test for differences in quantitative traits at baseline and at follow-up. Linear regression was used to model the effect of the genetic risk score on 5-year changes in glycemic traits. Individuals with known diabetes at the baseline or at the follow-up examination were excluded from the present analyses of glycemic traits. Values of plasma glucose, serum insulin, and all indices were logarithmically transformed before analyses to obtain a normal distribution. Changes in lifestyle measures of physical activity, smoking, and diet were individually defined into three classes; healthier, unhealthier, or no change.

The fully adjusted regression model included the additive genetic risk score, baseline age, sex, baseline BMI, change in BMI, and baseline value of the trait analyzed.

Additive interactions between the genetic risk score and changes in BMI and changes in lifestyle factors (physical activity, diet, and smoking) were tested by including the interaction term “change in BMI × risk score” or “change in lifestyle factor × risk score,” where BMI change is a continuous variable and change of lifestyle is a three-class variable, into the fully adjusted regression model.

or

A total of 5,850 individuals had information on the genetic risk score and did not have known diabetes at the baseline examination. These individuals were followed with a median follow-up time of 11 years, and 327 individuals developed type 2 diabetes. The genetic risk score was strongly associated with incident diabetes (Fig. 2 and Table 1). The cumulative incidence of type 2 diabetes as a function of age increased with tertiles of the genetic risk score (log-rank trend test, P = 0.0004) (Fig. 2). Each additional allele increased the risk of type 2 diabetes with a multifactor-adjusted hazard ratio of 1.06 (95% CI 1.03–1.09, P = 1.1 × 10⁻⁴) (Table 1). To elucidate differential effects over age, we stratified the population into two age-groups; below or above 50 years of age at censoring. In a Cox regression model adjusted for age, sex, and BMI, the hazard ratio was 1.10 (95% CI 1.04–1.16) in individuals age <50 years, while it was 1.06 (95% CI 1.02–1.09) in individuals >50 years. No interaction with age-group was observed (P = 0.19).

To address the underlying phenotypes causing the increased risk of diabetes, we tested whether the genetic risk score had measureable effects on changes in quantitative glycemic traits during the 5 years of follow-up. A total of 3,727 had information on the genetic risk score and did not have known diabetes at the baseline or at the follow-up examination. On average, after 5 years the population had improved glycemic regulation (15). They experienced an average decrease in fasting, 30-min, and 120-min plasma glucose values, as well as in fasting serum insulin values. They were less insulin resistant and increased their disposition index, although there was an average weight gain in the population. A borderline improvement of corrected insulin response occurred, but no changes in the insulinogenic index were observed (Table 2).

In the fully adjusted model, each additional allele in the genetic risk score associated with a relative increase in plasma glucose at fasting and during an OGTT over the 5-year follow-up period (Table 3). Moreover, the genetic risk score associated with relative decrements in insulinogenic index, corrected insulin response, and disposition indices—all surrogate measures of pancreatic β-cell function. No associations with changes in values of serum insulin or indices of insulin sensitivity were observed (Table 3). Similar results were seen when changes in lifestyle factors were included in the regression model (data not shown). Although the genetic risk score associated with increases in plasma glucose per allele, the average individual still had levels lower over the 5-year period. Consequently, individuals at increasingly higher genetic risk improved their plasma glucose values less compared with individuals with a lower genetic risk.

When subgroups of upper versus lower decile of genetic risk were compared, the difference evolving over the 5-year period between these two groups was 2.3% for fasting plasma glucose, which on average corresponds to ∼0.14 mmol/L. With these results extrapolated to a follow-up time of 20 years (representing an average year span in Inter99 from 45 to 65 years of age) and with linearity assumed in the genetic effects over the 20 years, this would correspond to a difference between the groups of ∼0.54 mmol/L for fasting plasma glucose. For comparison, the effect of a one-unit BMI increase over 5 years is associated with a 1% (∼0.05 mmol/L) relative increase in fasting plasma glucose in this population.

To understand the effect of adding more risk variants to the genetic risk score, we calculated the effect of a genetic risk score comprising 15 variants overlapping the study by Lyssenko et al. (3). When we included only these 15 variants in the genetic risk score, the effect on changes in fasting plasma glucose was 0.22% (95% CI 0.11–0.32, P = 1 × 10⁻⁴ per allele) compared with a per allele effect of 0.18% (95% CI 0.12–0.24, P = 9 × 10⁻⁹) when including 46 variants in the genetic risk score. Based on the estimated effect sizes and the number of variants included in two genetic risk scores, a theoretical maximum difference between individuals carrying no risk alleles and individuals carrying the maximum number of risk alleles (30 or 92) was calculated for changes in fasting plasma glucose. This corresponded to a theoretical maximum change of ∼7% for the risk score comprising 15 variants and ∼17% for the updated score comprising 46 variants.

For exploration of whether the effect of the genetic risk score on changes in glycemic traits was affected by changes in BMI or lifestyle factors, we tested for potential interactions. Since the genetic risk score had the strongest statistical association with fasting plasma glucose, potential interactions were only assessed for this outcome to achieve the highest statistical power.

We found an interaction between the genetic risk score and BMI changes over 5 years in relation to changes in fasting plasma glucose (P = 0.004). An increased BMI over the 5-year period resulted in a larger effect of the genetic risk score on changes in fasting plasma glucose (Fig. 3). Individuals who gained weight and who were in the highest tertile of genetic risk on average increased their unadjusted fasting plasma glucose values by 0.19%, whereas individuals who lost weight (or were weight stabile) and belonged to the lowest tertile of genetic risk on average decreased their unadjusted fasting plasma glucose by −3.3% over 5 years (Fig. 3A). When dividing the population into quartiles of BMI change, we observed a stepwise increasing effect of the genetic risk score with higher quartiles of BMI change in the fully adjusted model (Fig. 3B).

No interactions between the genetic risk score or changes in smoking, physical activity, or diet were observed (P = 0.20–0.44). Throughout all analyses, similar results were obtained for the weighted genetic risk score (data not shown).

In the current study, we have performed prospective analyses of a genetic risk score comprising 46 genetic risk variants for type 2 diabetes in the Danish Inter99 population of middle-aged to elderly people without accounting for single variant associations or gene-gene interactions.

In line with previous studies, we found that an updated genetic risk score was strongly associated with incident type 2 diabetes (3–8), although effect sizes were modest. What is new, we found that in the Inter99 population the genetic risk score had measureable effects on changes in plasma glucose and estimates of pancreatic β-cell function during 5 years of follow-up. Of interest, we report an interaction suggesting that individuals increasing their BMI are more susceptible to the effect of the genetic risk.

Our reported hazard ratio of 1.06 per allele resides in the lower end of the spectrum previously reported (3–8). One explanation may be that Inter99 is a lifestyle intervention study and it has been suggested that lifestyle intervention may attenuate the effect of genetic risk. This has been seen particularly for the variant in TCF7L2 in the Diabetes Prevention Program (23) and Finnish Diabetes Prevention Study (24) and was also suggested when a genetic risk score of 34 variants was investigated (6). Longer follow-up times have also been suggested to increase the effect of genetic variants. Moreover, it has recently been observed that the effects of genetic variants are stronger with younger age (<50 years) (5). When stratifying our study population into age-groups below or above 50 years, we observed a higher hazard ratio in the younger group, thus suggesting that also in Inter99 the genetic risk of type 2 diabetes may be slightly stronger with younger age onset; however, there was no significant interaction.

A number of reports have lately addressed the predictive ability of genetic risk scores in the progression to type 2 diabetes and collectively conclude that genetic risk assessment at best modestly improves prediction and reclassification over conventional risk factors such as age, BMI, and family history (3–8). Some of these reports have longer follow-up times and include more individuals than our study, and thus, it was not in the scope of the current study to investigate this further.

It was not known whether an updated genetic risk score of 46 type 2 diabetes risk variants would affect changes in glycemic traits during 5 years of follow-up or how the genetic risk may interact with changes in BMI and lifestyle factors. A previous study in the Botnia cohort including 2,444 Scandinavian individuals evaluated the effect of a genetic risk score, comprising 16 type 2 diabetes risk variants, on changes in OGTT-derived glycemic measures during 8 years of follow-up. The authors found that this genetic risk score associated with decreased insulin secretion and disposition index, whereas it did not affect changes in insulin sensitivity over time (3). In line with the results from the Botnia study, we found that despite a 5-year lifestyle intervention a genetic risk score comprising 46 risk variants associated with 5-year changes in plasma glucose and measures of β-cell function, but it did not associate with changes in measures of insulin sensitivity. In accordance with our results, the majority of variants included in the risk score have in cross-sectional studies been suggested to modulate the risk of type 2 diabetes through β-cell dysfunction and not through insulin resistance (rev. in 25).

The estimated per-allele effect size was slightly lower for the genetic risk score based on all 46 variants compared with a risk score based on only 15 variants from the study by Lyssenko et al. (3), while the theoretical difference in change in fasting plasma glucose between individuals carrying no risk alleles and maximal number of risk alleles was notably larger for the risk score based on all 46 variants. Together, these results suggest that inclusion of more risk variants with relatively lower individual effect sizes observed at the cross-sectional level does not infer uncertainty in the statistical models but strengthens the association between a genetic risk score and changes in fasting plasma glucose.

A recent study using the Whitehall II population reported that a genetic risk score of six variants associating with 2-h glucose values had a stronger impact on levels of 2-h glucose with increasing age, while a genetic risk score of 16 fasting plasma glucose variants had constant effects on fasting plasma glucose over the age span investigated (11). In the current study, we did not find differential effects of the genetic risk score on changes in fasting or 2-h plasma glucose among different age-groups, and we did not demonstrate any interactions with age (data not shown). Of note, the Whitehall II study (11) evaluated variants specifically associated with fasting or 2-h glucose values in nondiabetic individuals, which are only partially overlapping those variants found to increase risk of type 2 diabetes investigated in the current study.

We observed an interaction between the genetic risk score and changes in BMI, suggesting that individuals increasing their BMI may be more susceptible to the effects of type 2 diabetes risk variants. From a different point of view, this suggests that weight loss may attenuate the effect of the genetic risk, since the genetic risk score had a smaller effect in the group of individuals who decreased their BMI. Even though a large part of the general population will benefit from weight loss, these results suggest that it may be a particularly fruitful intervention for those at increased genetic risk of type 2 diabetes. In line with this, a recent study reported that intensive lifestyle intervention was effective in preventing type 2 diabetes at any genetic risk but with suggestive evidence that it may be more effective in the group with highest genetic risk (6).

The influence of obesity on genetic risk of type 2 diabetes has been debated. A large genome-wide association study stratifying lean and obese cases found that the majority of previously validated type 2 diabetes risk variants had a higher risk estimate for type 2 diabetes among lean case compared with obese case subjects (26). In contrast, a smaller study suggested that a genetic risk score had a stronger effect among obese participants (22), and it has also been reported that the effect of a genetic risk score on risk of impaired glucose tolerance was not evident in lean or insulin-sensitive persons (27). These reported interactions clearly illustrate the diversity underlying genetic risk of complex type 2 diabetes, and further well-powered studies investigating potential interactions are highly warranted.

We were not able to show any interactions with changes in lifestyle factors, which may be due to a modest statistical power to exploit the potential modifying impact of these factors. It is, however, likely that a large part of lifestyle changes may be “summarized” in the BMI change, thus explaining why BMI change may be a statistically more powerful factor to include in the model.

Obesity is believed to contribute to the risk of type 2 diabetes through several mechanisms including insulin resistance, inflammation, and lipotoxicity of the pancreatic β-cells and other organs (28). It is not known which factors may explain the interaction with the genetic risk observed in the current study; yet, a plausible explanation may be that a high genetic risk of type 2 diabetes renders the pancreatic β-cells more susceptible to an obesogenic environment.

The major strengths of the current study are as follows: 1) the longitudinal follow-up in a large well-phenotyped homogenous cohort, 2) investigations of an updated list of type 2 diabetes risk variants genotyped with good genotyping quality and state-of-the-art quality control, and 3) the repeated examinations including anthropometric measurements, OGTTs, and extensive characterization of lifestyle, which provide a wide collection of glycemic traits and relevant modifying factors.

Since Inter99 is a lifestyle intervention study, it is not known whether our results are directly transferrable to other populations, and further evidence is highly warranted. It has been described previously that in the Inter99 study, the intensity of the lifestyle intervention (high vs. low intensity) did not have any differential effect on 5-year changes in fasting plasma glucose (15), and we obtained similar results when adjusting for prerandomized intervention groups in the current study (data not shown). Interestingly, the group of individuals attending the follow-up examination overall improved their glycemic regulation, although their body weight increased. This seems to contrast with results in previous studies. Nonetheless, the group of individuals belonging to the highest tertile of genetic risk and who also gained weight did increase their values of plasma glucose over 5 years (fasting plasma glucose, 0.02 mmol/L; 2-h glucose, 0.14 mmol/L). The general glycemic improvement may be caused by enrollment in the Inter99 intervention study itself, and attendees for follow-up may also be those most motivated for improvement of their lifestyle (15). An alternative explanation may be that unidentified bias has skewed the distribution from the baseline to 5-year examination. This will, however, not affect the validity of our results, as the genotypic distribution can be assumed to be random over such potential bias, and our results are in line with those observed in the study by Lyssenko et al. (3).

In conclusion, a genetic risk score comprising 46 type 2 diabetes risk variants associated strongly with incident type 2 diabetes, changes in plasma glucose, and estimates of pancreatic β-cell function over a 5-year period in the Inter99 population of middle-aged to elderly Danish people. Individuals who increased their BMI were more susceptible to the cumulative impact of risk variants on fasting plasma glucose. This suggests that especially individuals at high genetic risk may benefit from weight loss intervention; however, this hypothesis remains to be tested.

The Inter99 study was initiated by T.J. (principal investigator [PI]) (Research Centre for Prevention and Health), Knut Borch-Johnsen (co-PI) (Steno Diabetes Center A/S), Troels Thomsen (Research Centre for Prevention and Health), and Hans Ibsen (Division of Cardiology, Holbæk University Hospital, Holbæk, Denmark). The present steering group comprises T.J., Knut Borch-Johnsen, and Charlotta Pisinger (Research Centre for Prevention and Health).

This project was funded by the Lundbeck Foundation and produced by the Lundbeck Foundation Centre for Applied Medical Genomics in Personalised Disease Prediction, Prevention and Care (LuCamp) (www.lucamp.org). The Novo Nordisk Foundation Center for Basic Metabolic Research is an independent Research Center at the University of Copenhagen partially funded by an unrestricted donation from the Novo Nordisk Foundation (www.metabol.ku.dk). Further funding came from the Danish Council for Independent Research (Medical Sciences).

No potential conflicts of interest relevant to this article were reported.

E.A.A. drafted the manuscript, conceived and designed the experiments, analyzed data, revised the manuscript, and contributed to discussions and interpretations. K.H.A. and C.H.S. conceived and designed the experiments, analyzed data, revised the manuscript, and contributed to discussions and interpretations. R.R.-M., T.S., J.M.J., and M.N.H. performed the genotyping experiments and the quality control, revised the manuscript, and contributed to discussions and interpretations. T.J., T.H., and O.P. conceived and designed the experiments, revised the manuscript, and contributed to discussions and interpretations. E.A.A. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

The authors thank A. Forman, T. Lorentzen, B. Andersen, M. Andersen, and G. Klavsen for technical assistance and A. Nielsen, P. Sandbeck, G. Lademann, and M.M.H. Kristensen for management assistance (all of Novo Nordisk Foundation Center for Basic Metabolic Research).