A recent study identified a low-frequency variant at CCND2 associated with lower risk of type 2 diabetes, enhanced insulin response to a glucose challenge, higher height, and, paradoxically, higher BMI. We aimed to replicate the strength and effect size of these associations in independent samples and to assess the underlying mechanism. We genotyped the variant in 29,956 individuals and tested its association with type 2 diabetes and related traits. The low-frequency allele was associated with a lower risk of type 2 diabetes (OR 0.53; P = 2 × 10−13; 6,647 case vs. 12,645 control subjects), higher disposition index (β = 0.07 log10; P = 2 × 10−11; n = 13,028), and higher Matsuda index of insulin sensitivity (β = 0.02 log10; P = 5 × 10−3; n = 13,118) but not fasting proinsulin (β = 0.01 log10; P = 0.5; n = 6,985). The low frequency allele was associated with higher adult height (β = 1.38 cm; P = 6 × 10−9; n = 13,927), but the association of the variant with BMI (β = 0.36 kg/m2; P = 0.02; n = 24,807), estimated in four population-based samples, was less than in the original publication where the effect estimate was biased by analyzing case subjects with type 2 diabetes and control subjects without diabetes separately. Our study establishes that a low-frequency allele in CCND2 halves the risk of type 2 diabetes primarily through enhanced insulin secretion.

A recent study used whole-genome sequencing and imputation techniques to identify one of the first robust associations between a low-frequency variant (1.47% in Icelandic population) and type 2 diabetes (1). The effect of the G minor allele at rs76895963 was appreciably larger than that of known common variants (odds ratio [OR] 0.53) (1). The G allele was associated with lower fasting glucose levels and higher insulinogenic index, suggesting an effect on insulin secretion, but, paradoxically, was associated with higher BMI (0.56 kg/m2) (1).

Genetic associations need testing in independent studies to ensure associations are not false positive results and to establish an effect size less biased by winner’s curse (regression to the mean). Once replicated, it is then important to test the underlying physiological mechanisms.

The apparently paradoxical association between the diabetes protective allele and higher BMI needs further explanation. Genetic associations that are paradoxical to epidemiological correlations have been described before and provide excellent targets for further investigation of biological mechanisms (2,3). However, associations between known type 2 diabetes alleles and BMI can be biased by a form of “index event bias,” sometimes referred to as “truncation bias,” if data sets are restricted to case or control subjects. This form of bias has likely led to associations between the risk allele at TCF7L2 and lower BMI in case subjects because carriers of the risk allele do not need to be as overweight to develop diabetes (4).

We aimed to assess whether independent samples provide robust replication of the strength and effect sizes of the CCND2 associations and to investigate further the underlying mechanisms that result in a low-frequency allele reducing the risk of type 2 diabetes but increasing height and BMI. We genotyped the CCND2 variant in 29,956 individuals and tested its association with risk of type 2 diabetes and with measures of insulin sensitivity and insulin secretion.

We genotyped the low-frequency CCND2 variant (rs76895963) in 29,496 individuals of European origin. Study characteristics and genotyping details are in Table 1. Call rates in all samples exceeded 95%, with no evidence of departure from Hardy-Weinberg equilibrium (P > 0.05).

Table 1

Summary details and relevant characteristics of the studies

StudyN (N males/N females)Age (years)Type 2 diabetes case/control subjects (n/n)Fasting glucose (mmol/L)2-h OGTT (mmol/L)Matsuda indexInsulinogenic index (pmol/mmol)Disposition indexBMI (kg/m2)Height (cm)Genotyping
Avon Longitudinal Study of Parents and Children (ALSPAC) (66,597 (0/6,597) 28.1 (4.8) NA NA NA NA NA NA 22.9 (3.8) NA Fluorescence-based competitive allele-specific assay (KASPar) at LGC Genomics (Hoddesdon, U.K.) 
Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS) (1213,512 (7,078/6,434) 61.1 (10.6) 6,145/5,045 4.9 (0.7) NA NA NA NA 29.3 (5.8) 167.9 (9.7) Fluorescence-based competitive allele-specific assay (KASPar) at LGC Genomics 
Metabolic Syndrome In Men (METSIM) (58,102 (8,102/0) 57.2 (7.1) 1,602/6,500 5.7 (0.5) 6.1 (1.7) 6.9 (4.2) 131.3 (220.3) 739.2 (1,566) 26.8 (3.8) 176.0 (6.4) TaqMan Allelic Discrimination Assays (Applied Biosystems) 
Relationship between Insulin Sensitivity and Cardiovascular disease (RISC) (131,285 (574/711) 43.8 (8.3) NA 5.1 (0.6) 5.7 (1.5) 11.6 (6.1) 96.9 (82.6) 996.7 (990) 25.5 (4.1) 170.7 (9.4) Fluorescence-based competitive allele-specific assay (KASPar) at LGC Genomics 
StudyN (N males/N females)Age (years)Type 2 diabetes case/control subjects (n/n)Fasting glucose (mmol/L)2-h OGTT (mmol/L)Matsuda indexInsulinogenic index (pmol/mmol)Disposition indexBMI (kg/m2)Height (cm)Genotyping
Avon Longitudinal Study of Parents and Children (ALSPAC) (66,597 (0/6,597) 28.1 (4.8) NA NA NA NA NA NA 22.9 (3.8) NA Fluorescence-based competitive allele-specific assay (KASPar) at LGC Genomics (Hoddesdon, U.K.) 
Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS) (1213,512 (7,078/6,434) 61.1 (10.6) 6,145/5,045 4.9 (0.7) NA NA NA NA 29.3 (5.8) 167.9 (9.7) Fluorescence-based competitive allele-specific assay (KASPar) at LGC Genomics 
Metabolic Syndrome In Men (METSIM) (58,102 (8,102/0) 57.2 (7.1) 1,602/6,500 5.7 (0.5) 6.1 (1.7) 6.9 (4.2) 131.3 (220.3) 739.2 (1,566) 26.8 (3.8) 176.0 (6.4) TaqMan Allelic Discrimination Assays (Applied Biosystems) 
Relationship between Insulin Sensitivity and Cardiovascular disease (RISC) (131,285 (574/711) 43.8 (8.3) NA 5.1 (0.6) 5.7 (1.5) 11.6 (6.1) 96.9 (82.6) 996.7 (990) 25.5 (4.1) 170.7 (9.4) Fluorescence-based competitive allele-specific assay (KASPar) at LGC Genomics 

Data are mean (SD) unless otherwise indicated. NA, not available.

We tested the association of the low-frequency variant with risk of type 2 diabetes, diabetes-related intermediate traits (fasting glucose, 2-h oral glucose tolerance test [OGTT] glucose, Matsuda index of insulin sensitivity, insulinogenic index, disposition index of β-cell function, and proinsulin levels), BMI, fat percentage, and height.

We used Matsuda index as a surrogate index of peripheral insulin sensitivity, which is highly correlated (ρ = 0.7) with the gold standard measure of insulin resistance (euglycemic-hyperinsulinemic clamp [M value]) (5).

We calculated the following:

Matsuda index of insulin sensitivity

formula

Insulinogenic index

formula

where Ins0, Ins30, Ins120, Gluc0, Gluc30, and Gluc120 are insulin and glucose levels at 0, 30, and 120 min of the OGTT, respectively. We also calculated insulin disposition index as follows: Matsuda index of insulin sensitivity × insulinogenic index.

To provide additional statistical power to estimate the effect of the variant on BMI, we genotyped the variant in 6,597 female participants with prepregnancy BMI data from the Avon Longitudinal Study of Parents and Children (ALSPAC) (6). Ethics approval for the study was obtained from the ALSPAC Ethics and Law Committee and the local research ethics committees. The ALSPAC Web site contains details of all the data that are available through a fully searchable data dictionary (7).

To increase our statistical power to estimate the effect of the low-frequency variant on Matsuda index, insulinogenic index, and disposition index, we included 5,114 samples from the Inter99 study that was part of the original discovery (1,8). The Danish study was approved by the ethics committee of the Capital Region of Denmark.

Diabetes-related intermediate traits were log10 transformed. We used age, sex (and age squared for height and BMI), and, if applicable, measures required to correct for genetic background, as covariates. We assumed an additive genetic model.

Analyses of glycemic traits, Matsuda index, insulinogenic index, and disposition index were performed in individuals without diabetes. For BMI, we limited analyses to studies most representative of the general population, with no or limited enrichment for or against type 2 diabetes. For the Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS), a diabetes case-control study, we randomly selected a subset of case subjects to include with all the control subjects such that the “population” consisted of 5% with type 2 diabetes and 95% control subjects. We also reanalyzed data from the population-based studies from the original study but without separating individuals with diabetes from individuals without and assessed the extent of enrichment for diabetes in the deCODE population-based study.

We performed fixed-effects inverse variance-weighted meta-analysis in R (9). Evidence of between-study heterogeneity was assessed using the Cochran Q test and the I2 statistic (10). The study complies with the Declaration of Helsinki.

The CCND2 Low-Frequency Allele Is Associated With a Lower Risk of Type 2 Diabetes

The frequency of rs76895963[G] was 1.97% in GoDARTS (Scottish), 2.15% in Metabolic Syndrome In Men (METSIM) (Finnish), 1.51% in Relationship between Insulin Sensitivity and Cardiovascular disease (RISC) (European wide), and 2.04% in ALSPAC (southwest U.K.). The rs76895963[G] allele was associated with a lower risk of type 2 diabetes with an effect size very similar to that described in the initial discovery study (OR unadjusted for BMI 0.53 [95% CI 0.45, 0.63], P = 2 × 10−13, OR adjusted for BMI 0.49 [0.40, 0.58], P = 2 × 10−14; 6,647 case vs. 12,645 control subjects) (Table 2 and Fig. 1). Meta-analysis with 12,939 case and 70,909 control subjects from the discovery studies revealed no evidence of heterogeneity of effect size across five studies (OR unadjusted for BMI 0.53 [0.48, 0.59], P = 1 × 10−30, OR adjusted for BMI 0.47 [0.42, 0.53], P = 1 × 10−35; 19,586 case vs. 83,554 control subjects; heterogeneity P = 0.9 for both unadjusted and adjusted model) (Table 2 and Fig. 1).

Table 2

Association of rs76895963 in CCND2 with type 2 diabetes, diabetes-related intermediate traits, and anthropometric traits

Trait/diseaseStudyEffect size95% CIPNI2 (%)Phet
Type 2 diabetes (BMI adjusted) (OR) Original study 0.46 0.40, 0.54 6 × 10−23 12,939 vs. 70,909†† 0.7 
 Current study 0.49 0.40, 0.58 2 × 10−14 6,647 vs. 12,645†† 0.6 
 Combined 0.47 0.42, 0.53 1 × 10−35 19,586 vs. 83,554†† 0.9 
Type 2 diabetes (BMI unadjusted) (OR) Original study 0.53 0.46, 0.61 8 × 10−19 12,939 vs. 70,909†† 0.7 
 Current study 0.53 0.45, 0.63 2 × 10−13 6,647 vs. 12,645†† 0.7 
 Combined 0.53 0.48, 0.59 1 × 10−30 19,586 vs. 83,554†† 0.9 
Fasting glucose (log) Original study −0.01 −0.02, −0.01 3 × 10−4 11,764 NA NA 
 Current study −0.02 −0.02, −0.01 5 × 10−5 11,739 0.9 
 Combined −0.01 −0.02, −0.01 9 × 10−8 23,503 0.8 
2-h OGTT (log) Original study −0.02 −0.04, 0.01 0.15 4,900 NA NA 
 Current study −0.05 −0.08, −0.03 6 × 10−5 8,261 0.6 
 Combined −0.04 −0.05, −0.02 1 × 10−4 13,161 44 0.2 
Matsuda index (log10) Current study 0.02 0.01, 0.03 5 × 10−3 13,118 0.8 
Disposition index (log10) Current study 0.07 0.05, 0.09 2 × 10−11 13,028 0.7 
Insulinogenic index (log10) Current and original study 0.05 0.03, 0.07 8 × 10−6 13,181 0.5 
Fasting proinsulin (log10)* Current study 0.01 −0.01, 0.02 0.5 6,985 NA NA 
30-min proinsulin (log10)* Current study −0.01 −0.02, 0.01 0.3 6,947 NA NA 
120-min proinsulin (log10)* Current study −0.01 −0.02, 0.00 0.1 6,978 NA NA 
BMI (kg/m2Population-based studies** 0.36 0.06, 0.65 0.02 24,807 0.8 
 Current study*** 0.05 −0.21, 0.30 0.7 22,464 0.7 
 Combined 0.25 0.08, 0.43 4 × 10−3 109,492 0.4 
Fat mass % (log10) Current study 0.00 −0.01, 0.01 0.5 6,979 NA NA 
Height (cm) Original study 1.16 0.83, 1.50 6 × 10−12 78,236 0.7 
 Current study 1.38 0.92, 1.84 6 × 10−9 13,927 0.6 
 Combined 1.24 0.97, 1.51 2 × 10−19 92,163 0.8 
Trait/diseaseStudyEffect size95% CIPNI2 (%)Phet
Type 2 diabetes (BMI adjusted) (OR) Original study 0.46 0.40, 0.54 6 × 10−23 12,939 vs. 70,909†† 0.7 
 Current study 0.49 0.40, 0.58 2 × 10−14 6,647 vs. 12,645†† 0.6 
 Combined 0.47 0.42, 0.53 1 × 10−35 19,586 vs. 83,554†† 0.9 
Type 2 diabetes (BMI unadjusted) (OR) Original study 0.53 0.46, 0.61 8 × 10−19 12,939 vs. 70,909†† 0.7 
 Current study 0.53 0.45, 0.63 2 × 10−13 6,647 vs. 12,645†† 0.7 
 Combined 0.53 0.48, 0.59 1 × 10−30 19,586 vs. 83,554†† 0.9 
Fasting glucose (log) Original study −0.01 −0.02, −0.01 3 × 10−4 11,764 NA NA 
 Current study −0.02 −0.02, −0.01 5 × 10−5 11,739 0.9 
 Combined −0.01 −0.02, −0.01 9 × 10−8 23,503 0.8 
2-h OGTT (log) Original study −0.02 −0.04, 0.01 0.15 4,900 NA NA 
 Current study −0.05 −0.08, −0.03 6 × 10−5 8,261 0.6 
 Combined −0.04 −0.05, −0.02 1 × 10−4 13,161 44 0.2 
Matsuda index (log10) Current study 0.02 0.01, 0.03 5 × 10−3 13,118 0.8 
Disposition index (log10) Current study 0.07 0.05, 0.09 2 × 10−11 13,028 0.7 
Insulinogenic index (log10) Current and original study 0.05 0.03, 0.07 8 × 10−6 13,181 0.5 
Fasting proinsulin (log10)* Current study 0.01 −0.01, 0.02 0.5 6,985 NA NA 
30-min proinsulin (log10)* Current study −0.01 −0.02, 0.01 0.3 6,947 NA NA 
120-min proinsulin (log10)* Current study −0.01 −0.02, 0.00 0.1 6,978 NA NA 
BMI (kg/m2Population-based studies** 0.36 0.06, 0.65 0.02 24,807 0.8 
 Current study*** 0.05 −0.21, 0.30 0.7 22,464 0.7 
 Combined 0.25 0.08, 0.43 4 × 10−3 109,492 0.4 
Fat mass % (log10) Current study 0.00 −0.01, 0.01 0.5 6,979 NA NA 
Height (cm) Original study 1.16 0.83, 1.50 6 × 10−12 78,236 0.7 
 Current study 1.38 0.92, 1.84 6 × 10−9 13,927 0.6 
 Combined 1.24 0.97, 1.51 2 × 10−19 92,163 0.8 

Analysis of diabetes-related intermediate traits and height reported in the table were performed in individuals without diabetes. NA, not applicable because data from only one study were available; Phet, heterogeneity P value.

*Values were adjusted for corresponding insulin measurements at the same time points during OGTT.

**Results from population studies with no apparent enrichment for or against type 2 diabetes, including three studies from the original publication (Iranian TLGS study, Danish Inter99 study, and Danish Health2006 study) and ALSPAC.

***To avoid index event bias or truncation bias, we used our population-based studies. (See 2research design and methods and 8discussion.)

†For insulinogenic index, we give the meta-analysis results including data presented in the original article from the Inter99 study. For Matsuda index and disposition index, we give the meta-analysis results including a new analysis of the Inter99 study not previously presented.

††Number of cases vs. controls.

Figure 1

Forest plot of the association between the CCND2 rs76895963 low-frequency allele and type 2 diabetes (unadjusted for BMI) in discovery and replication studies. The dashed line indicates null effect. The top, middle, and bottom diamonds represent the effect size (center of diamond) and 95% CIs (horizontal ends) from the discovery studies, replication studies, and overall meta-analysis, respectively.

Figure 1

Forest plot of the association between the CCND2 rs76895963 low-frequency allele and type 2 diabetes (unadjusted for BMI) in discovery and replication studies. The dashed line indicates null effect. The top, middle, and bottom diamonds represent the effect size (center of diamond) and 95% CIs (horizontal ends) from the discovery studies, replication studies, and overall meta-analysis, respectively.

The CCND2 Low-Frequency Allele Is Associated With Higher Insulin Secretion

The G minor allele was associated with lower fasting glucose (β = −0.02 log [95% CI −0.02, −0.01], P = 5 × 10−5; n = 11,739) and lower 2-h OGTT glucose (β = −0.05 log [−0.08, −0.03], P = 6 × 10−5; n = 8,261). The combined meta-analysis estimated −0.01 log [−0.02, −0.01] lower fasting glucose levels (P = 9 × 10−8; n = 23,503) and −0.04 log [−0.05, −0.02] lower 2-h OGTT glucose levels (P = 1 × 10−4; n = 13,161) per copy of type 2 diabetes protective allele (Table 2).

The type 2 diabetes protective allele was associated with improved ability to secrete insulin in response to a glucose challenge test: higher insulinogenic index (β = 0.06 log10 [95% CI 0.03, 0.09], P = 1 × 10−4, n = 8,067; and β = 0.05 log10 [0.03, 0.07], P = 8 × 10−6, n = 13,181 including the original study) (Table 2). The low-frequency allele was associated with higher disposition index (β = 0.08 log10 [0.05, 0.11], P = 1 × 10−7; n = 8,050). Disposition index was not presented in the original study, but we analyzed the Danish Inter99 study and meta-analyzed with METSIM and RISC, which provided an effect of 0.07 log10 [0.05, 0.09] with higher disposition index (P = 2 × 10−11; n = 13,028) (Table 2).

The G allele was not associated with any measures of proinsulin levels adjusted for corresponding insulin levels at the same time points during OGTT (Table 2).

The analysis of the CCND2 low-frequency allele and the Matsuda index in METSIM and RISC produced a borderline result (β = 0.03 log10 [95% CI 0, 0.05], P = 0.05; n = 8,134). A meta-analysis of all 13,118 individuals without diabetes from METSIM, RISC, and Danish Inter99 resulted in a small association with Matsuda index (β = 0.02 log10 [0.01, 0.03], P = 5 × 10−3) (Table 2).

The Effect Size of the CCND2 Low-Frequency Allele With Height Is Consistent With the Original Study

The G minor allele was associated with higher adult height (β = 1.38 cm [95% CI 0.92, 1.84], P = 6 × 10−9; n = 13,927) (Table 2). The combined meta-analysis including data from the original study estimated 1.24 cm [0.97, 1.51] higher height per copy of the type 2 diabetes protective allele (P = 2 × 10−19; n = 92,163) (Table 2).

The Effect Size of the CCND2 Low-Frequency Allele With BMI Is Lower Than Reported in the Original Study

The original report found an association between the low-frequency CCND2 allele and higher BMI (0.56 kg/m2) analyzed separately in individuals with or without type 2 diabetes resulting in spurious associations resulting from index event biases. To further test the BMI association, we first showed that individuals from the deCODE study with both CCND2 genotype and BMI available had a slight excess for diabetic cases (Supplementary Fig. 1). We showed that this type of enrichment in population studies results in a bias toward an association between the protective allele and lower BMI because the diabetes case subjects tend to be heavier and carry less protective alleles than individuals without diabetes (Supplementary Fig. 2). We thus decided to focus our analysis of the BMI association to the four population studies with no apparent enrichment for or against type 2 diabetes, i.e., three studies from the original publication (the Iranian Tehran Lipid and Glucose Study [TLGS] [n = 8,658], the Danish Inter99 study [n = 6,228], and the Danish Health2006 study [n = 3,324]) and ALSPAC (n = 6,597). This resulted in a smaller effect of the variant on BMI than reported in the original publication (β = 0.36 kg/m2 [95% CI 0.06, 0.65], P = 0.02; n = 24,807) (Table 2). We observed no heterogeneity between the effect estimates between these four population-based studies (heterogeneity P = 0.8) (Table 2). This analysis represented our least biased estimate of the effect size. When we combined results from the four studies included in the original publication (reanalyzed including both type 2 diabetes case and control subjects) with results from four studies added in this article (with GoDARTS individuals sampled so as to include only 5% diabetes case subjects), the association with BMI was present but with lower effect size (β = 0.25 kg/m2 [0.08, 0.43], P = 4 × 10−3; n = 109,492) (Table 2 and Fig. 2) and with some evidence of heterogeneity (P = 0.03).

Figure 2

Forest plot of the association between the CCND2 rs76895963 low-frequency allele and BMI including eight studies with no, or limited, ascertainment or enrichment for or against type 2 diabetes. These eight studies included four from the original article, including deCODE individuals and a sample of GoDARTS individuals made to consist of 5% diabetes case subjects. The dashed line indicates null effect. The top, middle, and bottom diamonds represent the effect size (center of diamond) and 95% CIs (horizontal ends) from the discovery studies, replication studies, and overall meta-analysis, respectively.

Figure 2

Forest plot of the association between the CCND2 rs76895963 low-frequency allele and BMI including eight studies with no, or limited, ascertainment or enrichment for or against type 2 diabetes. These eight studies included four from the original article, including deCODE individuals and a sample of GoDARTS individuals made to consist of 5% diabetes case subjects. The dashed line indicates null effect. The top, middle, and bottom diamonds represent the effect size (center of diamond) and 95% CIs (horizontal ends) from the discovery studies, replication studies, and overall meta-analysis, respectively.

We found no association with fat mass percentage (β = 0.00 [95% CI −0.01, 0.01], P = 0.5; n = 6,979 individuals without diabetes) (Table 2).

Our study provides robust replication of the relatively large protective effect of a low-frequency variant at CCND2 against risk of type 2 diabetes and its association with improved insulin secretion and higher height. The estimate of the effect size on risk of type 2 diabetes in our study was very close to that of the discovery studies and therefore confirms an unbiased estimate of the effect size: carriers of the low-frequency allele are at approximately half the risk of type 2 diabetes compared with noncarriers. Our results, together with data from the original study, provide very strong evidence of the mechanism of diabetes protection. The associations with improved disposition index and insulinogenic index but smaller effects with the Matsuda index, in up to 13,118 individuals, show that the protective diabetes effect operates primarily through a mechanism of relatively favorable insulin secretory response to a glucose challenge and to lower blood glucose more effectively than noncarriers. The effect is unlikely to act through improved insulin processing, as we saw no association with proinsulin levels in the METSIM study, despite previous observations of associations between the TCF7L2 and other diabetes risk alleles in this study (2).

Our data suggest that the association between the CCND2 protective variant and higher BMI is lower than that previously reported. A reanalysis of previous data, together with new data, provided evidence of an association between the CCND2 protective allele and higher BMI, but we observed a smaller effect and heterogeneity between studies. Determining the true biological effect of the variant on BMI was very difficult because of “index event” bias. The “index event” in this case was a classification of normoglycemia; therefore, people carrying a type 2 diabetes protective allele remain normoglycemic at higher BMIs. Similar such likely biases have been observed between strong diabetes risk alleles and BMI where the risk allele at TCF7L2 was associated with lower BMI in individuals with type 2 diabetes because individuals carrying a risk allele will develop diabetes at lower BMIs than noncarriers on average (4,11). Index event bias means it is extremely difficult to determine whether diabetes risk alleles have biological effects on BMI.

In summary, we replicated the diabetes and height growth effects of the low-frequency variant at CCND2 in 29,956 individuals. Our best estimate of the effect of the variant on BMI suggests that the effect is smaller than reported in the original publication owing to index event bias. Further studies are needed to establish the size of the BMI association. Our data, together with the original finding, show a mechanism through improved insulin secretion, which results in lower fasting glucose levels, lower 2-h OGTT glucose levels, and a lower risk of type 2 diabetes. Combining all data including 19,586 type 2 diabetes case and 83,554 control subjects from the original study and our study provides evidence that carrying this variant reduces the risk of type 2 diabetes by ∼50% relative to noncarriers.

Acknowledgments. The authors thank the European Research Council, Wellcome Trust, and University of Exeter Medical School. The authors are grateful to all the participants in GoDARTS, the general practitioners, the Scottish School of Primary Care for their help in recruiting the participants, and the whole team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses. The authors acknowledge the support of the Health Informatics Centre, University of Dundee, for managing and supplying the anonymized data and NHS Tayside, the original data owner. The authors are extremely grateful to all the families who took part in ALSPAC, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.

Funding. H.Y. and T.M.F. are funded by the European Research Council award (323195 and SZ-50371). R.M.F. is a Sir Henry Wellcome Postdoctoral Fellow (Wellcome Trust grant 085541/Z/08/Z). D.A.L. is funded by SP/07 1008/24066. A.T.H. is a Wellcome Trust Senior Investigator and a National Institute of Health Research senior investigator. M.I.M. is a Wellcome Trust Senior Investigator (Wellcome Trust: 090532 and 098381; Medical Research Council [MRC]: G0601261.) The Wellcome Trust provides support for GoDARTS (awards 072960/z/03/z and 099177/z12/z). METSIM was supported by grants from the Academy of Finland, University of Eastern Finland, and Sigrid Juselius Foundation. The U.K. MRC and the Wellcome Trust (grant 092731) and the University of Bristol provide core support for ALSPAC. The Danish substudy was supported by research grants from the Novo Nordisk Foundation Center for Basic Metabolic Research, an independent research center at the University of Copenhagen partially funded by an unrestricted donation from the Novo Nordisk Foundation (www.metabol.ku.dk); the Lundbeck Foundation (www.lucamp.org); the Danish Agency for Science, Technology and Innovation; the PhD School of Molecular Metabolism; University of Southern Denmark; and the Copenhagen Graduate School of Health and Medical Sciences. The Inter99 study was initiated by T.J. (principal investigator), K. Borch-Johnsen (co-principal investigator, Steno Diabetes Centre and Holbæk Hospital), H. Ibsen (Holbæk Hospital), and T.F. Thomsen (Research Centre for Prevention and Health, Glostrup, Denmark). The steering committee is comprised of the former two and C. Pisinger (Research Centre for Prevention and Health, Glostrup University Hospita)l. The Inter99 project was financially supported by research grants from the Danish Research Council, the Danish Centre for Health Technology Assessment, Research Foundation of Copenhagen County, Ministry of Internal Affairs and Health, the Danish Heart Foundation, the Danish Pharmaceutical Association, the Augustinus Foundation, the Ib Henriksen Foundation, and the Becket Foundation.

The funders had no influence on study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Duality of Interest. The Inter99 project was also funded by Novo Nordisk. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. H.Y. and T.M.F. conceived and designed the study. H.Y. analyzed data. R.M.F., S.M.R., D.A.L., and G.D.S. (with ALSPAC); E.F., A.M., and M.W. (with RISC); T.J., T.H., and O.P. (with Inter99); A.D.M. and C.N.A.P. (with GoDARTS); and M.L. (with METSIM) performed genotyping and phenotyping. H.Y. (with RISC, GoDARTS, and ALSPAC), A.S., and J.V. (with METSIM) performed statistical analysis. R.M.F., D.F.G., G.T., U.T., M.I.M., C.N.A.P., and M.L. commented on the manuscript. A.S., R.M.F., J.V., M.N.W., W.X., A.R.W., E.F., A.M., S.M.R., D.A.L., G.D.S., T.J., T.H., O.P., V.S., D.F.G., G.T., U.T., K.S., A.T.H., M.W., A.D.M., M.I.M., C.N.A.P., and M.L. agreed with the manuscript results and conclusions. T.M.F. 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.

1.
Steinthorsdottir
V
,
Thorleifsson
G
,
Sulem
P
, et al
.
Identification of low-frequency and rare sequence variants associated with elevated or reduced risk of type 2 diabetes
.
Nat Genet
2014
;
46
:
294
298
[PubMed]
2.
Strawbridge
RJ
,
Dupuis
J
,
Prokopenko
I
, et al.;
DIAGRAM Consortium
;
GIANT Consortium
;
MuTHER Consortium
;
CARDIoGRAM Consortium
;
C4D Consortium
.
Genome-wide association identifies nine common variants associated with fasting proinsulin levels and provides new insights into the pathophysiology of type 2 diabetes
.
Diabetes
2011
;
60
:
2624
2634
[PubMed]
3.
Frayling
TM
,
Hattersley
AT
.
Physiology helps GWAS take a step closer to mechanism
.
Diabetes
2014
;
63
:
1836
1837
[PubMed]
4.
Stolerman
ES
,
Manning
AK
,
McAteer
JB
, et al
.
TCF7L2 variants are associated with increased proinsulin/insulin ratios but not obesity traits in the Framingham Heart Study
.
Diabetologia
2009
;
52
:
614
620
[PubMed]
5.
Stancáková
A
,
Javorský
M
,
Kuulasmaa
T
,
Haffner
SM
,
Kuusisto
J
,
Laakso
M
.
Changes in insulin sensitivity and insulin release in relation to glycemia and glucose tolerance in 6,414 Finnish men
.
Diabetes
2009
;
58
:
1212
1221
[PubMed]
6.
Boyd
A
,
Golding
J
,
Macleod
J
, et al
.
Cohort Profile: the ‘children of the 90s’—the index offspring of the Avon Longitudinal Study of Parents and Children
.
Int J Epidemiol
2013
;
42
:
111
127
[PubMed]
7.
University of Bristol. Avon Longitudinal Study of Parents and Children [Internet],
2015
. Available from http://www.bris.ac.uk/alspac/researchers/data-access/data-dictionary/.
8.
Jørgensen T, Borch-Johnsen K, Thomsen TF, Ibsen H, Glumer C, Pisinger C. A randomized non-pharmacological intervention study for prevention of ischaemic heart disease: baseline results Inter99. Eur J Cardiovasc Prev Rehabil
2003
;
10
:
377
386
9.
Viechtbauer
W
.
Conducting meta-analyses in R with the metafor package
.
J Stat Softw
2010
;
36
:
1
48
10.
Higgins
JP
,
Thompson
SG
.
Quantifying heterogeneity in a meta-analysis
.
Stat Med
2002
;
21
:
1539
1558
[PubMed]
11.
Grant
SF
,
Thorleifsson
G
,
Reynisdottir
I
, et al
.
Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes
.
Nat Genet
2006
;
38
:
320
323
[PubMed]
12.
Doney
AS
,
Fischer
B
,
Leese
G
,
Morris
AD
,
Palmer
CN
.
Cardiovascular risk in type 2 diabetes is associated with variation at the PPARG locus: a Go-DARTS study
.
Arterioscler Thromb Vasc Biol
2004
;
24
:
2403
2407
[PubMed]
13.
Hills
SA
,
Balkau
B
,
Coppack
SW
, et al.;
EGIR-RISC Study Group
.
The EGIR-RISC STUDY (The European Group for the Study of Insulin Resistance: Relationship between Insulin Sensitivity and Cardiovascular disease risk): I. Methodology and objectives
.
Diabetologia
2004
;
47
:
566
570
[PubMed]

Supplementary data