An explained variance-based genetic risk score associated with gestational diabetes antecedent and with progression to pre-diabetes and type 2 diabetes: a cohort study

Authors

  • H Cormier,

    1. Department of Food Sciences and Nutrition, Laval University, Quebec City, QC, Canada
    2. Institute of Nutrition and Functional Foods (INAF), Laval University, Quebec City, QC, Canada
    Search for more papers by this author
    • These authors contributed equally to this work.
  • J Vigneault,

    1. Department of Food Sciences and Nutrition, Laval University, Quebec City, QC, Canada
    2. Institute of Nutrition and Functional Foods (INAF), Laval University, Quebec City, QC, Canada
    Search for more papers by this author
    • These authors contributed equally to this work.
  • V Garneau,

    1. Institute of Nutrition and Functional Foods (INAF), Laval University, Quebec City, QC, Canada
    Search for more papers by this author
  • A Tchernof,

    1. Department of Food Sciences and Nutrition, Laval University, Quebec City, QC, Canada
    2. Endocrinology and Nephrology, CHU de Quebec Research Center, Quebec City, QC, Canada
    Search for more papers by this author
  • M-C Vohl,

    1. Department of Food Sciences and Nutrition, Laval University, Quebec City, QC, Canada
    2. Institute of Nutrition and Functional Foods (INAF), Laval University, Quebec City, QC, Canada
    Search for more papers by this author
  • SJ Weisnagel,

    1. Department of Food Sciences and Nutrition, Laval University, Quebec City, QC, Canada
    2. Endocrinology and Nephrology, CHU de Quebec Research Center, Quebec City, QC, Canada
    Search for more papers by this author
  • J Robitaille

    Corresponding author
    1. Department of Food Sciences and Nutrition, Laval University, Quebec City, QC, Canada
    2. Institute of Nutrition and Functional Foods (INAF), Laval University, Quebec City, QC, Canada
    • Correspondence: Dr J Robitaille, Department of Food Sciences and Nutrition, Institute of Nutrition and Functional Foods (INAF), Laval University, Pavillon des services, 2440 Hochelaga Blvd, Room 2729-N, Quebec City, Quebec, Canada G1V 0A6. Email julie.robitaille@fsaa.ulaval.ca

    Search for more papers by this author

Abstract

Objective

To determine whether an explained-variance genetic risk score (GRS), with 36 single nucleotide polymorphisms (SNPs) previously associated with type 2 diabetes (T2D), is also associated with gestational diabetes mellitus (GDM), and with the progression to pre-diabetes and T2D among women with prior GDM.

Design

A cohort study.

Setting

Clinical investigation unit of Laval University, Quebec, Canada.

Population

A cohort of 214 women with prior GDM and 82 controls recruited between 2009 and 2012.

Methods

Associations between the GRS and GDM.

Main outcomes measures

GDM and prevalence of pre-diabetes and T2D.

Results

Women with prior GDM had a higher GRS compared with controls (38.6 ± 3.9, 95% CI 38.1–39.1, versus 37.4 ± 3.2, 95% CI 36.7–38.1; < 0.0001). In women with prior GDM, the explained-variance GRS was higher for pre-diabetic women compared with women who remained normoglucotolerant at testing (1.21 ± 0.18, 95% CI 1.18–1.23, versus 1.17 ± 0.15, 95% CI 1.13–1.20; < 0.0001). Similarly, women with T2D had a higher explained-variance GRS compared with women with prior GDM who remained normoglucotolerant (1.20 ± 0.18, 95% CI 1.14–1.25, versus 1.17 ± 0.17, 95% CI 1.13–1.20; < 0.0001). The predictive effects of the explained-variance GRS, age, and body mass index (BMI), or the additive effects of the three variables, were tested for pre-diabetes and T2D. We observed an area under the curve of 0.6269 (95% CI 0.5638–0.6901) for age and BMI, and adding the explained-variance GRS into the model increased the area to 0.6672 (95% CI 0.6064–0.7281) for the prediction of pre-diabetes.

Conclusions

An explained-variance GRS is associated with both GDM and progression to pre-diabetes and T2D in women with prior GDM.

Introduction

Gestational diabetes mellitus (GDM) is defined as glucose intolerance with first onset or recognition during pregnancy, and is characterised by impaired insulin secretion and action.[1] About one woman out of five has a pregnancy complicated by GDM.[2, 3] GDM appears as a preview to important adverse metabolic complications in the years following delivery, such as cardiovascular diseases and type 2 diabetes (T2D).[2, 4, 5] Using population-based data, Feig et al.[2] observed that the prevalence of T2D after GDM was 13.1 and 18.9% at 5.2 and 9.0 years after delivery, respectively. A systematic review revealed that the cumulative incidence of T2D ranged from 2.6 to over 70.0% in studies that examined women from 6 weeks to 28 years postpartum.[6]

Gestational diabetes mellitus is recognised as a multifactorial disease in which it is likely that environment triggers interact with genetic variants.[7] Given that women with prior GDM are at an increased risk of developing T2D later in life,[5] and that women with a family history of T2D are at increased risk of GDM,[8] it is possible to hypothesise that GDM and T2D may share similar genetic susceptibilities and risk factors.[9-11] Genetic variants determining the risk of T2D may, therefore, also be associated with GDM.

New genetic variants associated with susceptibility to T2D have been discovered in recent genome-wide association studies (GWAS).[12, 13] Advances in genetic knowledge have demonstrated many new loci related with increased T2D risk.[14, 15] The Diabetes Prevention Program showed that a higher T2D genetic risk, estimated with a score from 34 single nuclear polymorphisms (SNPs) associated with T2D, was also predictive of the progression to diabetes in high-risk individuals.[16]

In the present study, genotype frequencies and effect sizes of the SNPs in the diabetogenic genes were compared in women with prior GDM and a control group of women without GDM history. Our objective was to determine whether a new explained-variance genetic risk score (GRS) with 36 SNPs related to T2D is associated with antecedent GDM, and whether this GRS is predictive of the subsequent development of pre-diabetes and T2D among women with prior GDM. We hypothesised that a higher explained-variance GRS, which includes T2D-related loci, is also associated with prior GDM and predicts pre-diabetes and T2D.

Methods

Participants and study design

The study population has been described previously.[17] Briefly, a total of 296 women were recruited through access to administrative data from the Régie de l'assurance maladie du Québec (RAMQ), the provincial health plan registry. Women living in the greater Quebec City area, aged ≥18 years, with a pregnancy between 2003 and 2010 were invited to participate. Women were excluded if they were pregnant at the time of the study or if they had type-I diabetes. A total of 3896 women were sent a letter. A total of 688 women contacted the research team, of whom 95 were not eligible (pregnant at the time of the study or had type-I diabetes) and 297 refused to participate. The study sample thus consisted of 214 women with prior GDM and 82 women without prior GDM (controls), recruited between October 2009 and August 2012. Women were tested at the Clinical Investigation Unit at the Institute of Nutrition and Functional Foods (INAF) of Laval University, Quebec, Canada. All subjects gave their written consent and ethical approval was obtained from the Laval University Ethics Committee. This trial was registered at clinicaltrials.gov as NCT01340924.

Data collection

Data have been carefully collected for each participant using standardised procedures and validated tools. Weight was measured to the nearest 0.1 kg on a calibrated balance, height was measured to the nearest millimeter with a stadiometer, and body mass index (BMI) was then calculated (kg/m2).

Oral glucose tolerance test

A 75-g, 2-hour, oral glucose tolerance test (OGTT) was performed in the morning after an overnight fast. Blood samples were collected at −15, 0, 15, 30, 60, 90, and 120 minutes for the measurement of plasma glucose and insulin levels. The interassay coefficient of variation for the OGTT was 1.0% for a basal glucose value set at 5.0 mmol/l. Plasma glucose was measured enzymatically,[18] whereas plasma insulin was measured by radioimmunoassay.[19] Impaired fasting glucose was characterised by a fasting glucose value of 6.1–7.0 mmol/l, and impaired glucose tolerance was characterised by a normal fasting glucose value (<6.1 mmol/l) and a 2-hour post-OGTT glucose value of 7.8–11.1 mmol/l.[20] The glycated haemoglobin (A1C) was measured using Cobas Integra 800 and standardised with the National Glycated Haemoglobin Standardization Program (Integra Inc., Roche, Switzerland). In accordance with the 2013 Canadian Diabetes Association latest guidelines, pre-diabetes was defined as impaired fasting glucose, and/or impaired glucose tolerance, and/or A1C (between 6.0 and 6.4%). T2D was defined as fasting plasma glucose ≥7.0 mmol/l, and/or 2-hour plasma glucose post-OGTT ≥ 11.1 mmol/l, and/or A1C ≥ 6.5%.[20]

The homeostasis model assessment for insulin sensitivity (HOMA-IS) index was calculated using the following formula: (fasting insulin concentration × fasting glucose concentration)/22.5.[21] The Matsuda index was calculated as follows: 10 000/√[fasting glucose concentration × fasting insulin concentration × (mean glucose × mean insulin)].[22] The area under the curves for glucose and insulin excursions after the glucose load were calculated using the trapezoidal rule.[23] The insulin secretion index was estimated by the ratio between the area under the curve for insulin and the area under the curve for glucose.[24] The insulinogenic index for insulin secretion [(insulin at 30 minutes−fasting insulin)/(glucose at 30 minutes−fasting glucose)] was also calculated.[25, 26]

Selection of risk alleles

The GRS was assessed based on the 34 SNPs confirmed to be associated with T2D in a previous study by Hivert et al.[16] on a population of European ancestry. We added two SNPs, rs2237895 (KCNQ1) and rs12255372 (TCF7L2), in the new explained-variance GRS model, which were previously found to be associated with GDM.[11, 27, 28] As shown in Table 1, to construct the explained-variance GRS model, selected SNPs (rs2237895, rs12255372, rs7903146, rs10811661, rs7754840, rs1801282, rs1111875, rs1470579, rs5219, rs13266634, rs7578597, rs1552224, rs10923931, rs11708067, rs10010131, rs4747969, rs7578326, rs13292136, rs864745, rs757210, rs1531343, rs4607103, rs7961581, rs10830963, rs243021, rs231362, rs4457053, rs340874, rs917793, rs972283, rs8042680, rs7957197, rs2191349, rs780094, rs896854, and rs11634397) were genotyped using validated primers and TaqMan probes (Life Technologies Corporation, Burlington, ON, Canada).[29] DNA was mixed with TaqMan Genotyping Master Mix with a gene-specific primer and with probe mixture (pre-developed TaqMan SNP Genotyping Assays; Life Technologies Corporation) in a final volume of 10 μl. Genotypes were determined using a StepOnePlus Real-Time PCR System, and analysed using stepone 2.1 (Life Technologies Corporation). To assess accuracy, we duplicated the genotyping of all SNPs for 5% of the study participants.

Table 1. Type 2 diabetes risk loci
GenesSNPsAlleles (risk/other)MAFaORbStudyOverall study A, risk allele/a, other allele
AA n (%)Aa n (%)aa n (%)
  1. DIAGRAM, diabetes genetics replication and meta-analysis; MAGIC, meta-analysis of glucose and insulin-related traits consortium.

  2. a

    MAF: minor allele frequency, derived from the ALLELE procedure in sas genetics 9.3.

  3. b

    Odds ratio reported in previous literature (column Study).

TCF7L2 rs7903146T/CT = 0.401.37DIAGRAM41 (13.8)154 (51.9)102 (34.3)
CDKN2A/2B rs10811661T/CC = 0.181.26DIAGRAM202 (68.0)84 (28.3)11 (03.7)
CDKAL1 rs7754840C/GC = 0.351.25DIAGRAM33 (11.1)139 (46.8)125 (42.1)
PPARG rs1801282C/GG = 0.101.18DIAGRAM241 (81.1)52 (17.5)4 (1.4)
HHEX rs1111875C/TT = 0.371.17DIAGRAM113 (38.1)146 (49.2)38 (12.8)
IGF2BP2 rs1470579C/AC = 0.341.17DIAGRAM36 (12.1)129 (43.4)132 (44.4)
KCNJ11 rs5219T/CT = 0.401.16DIAGRAM37 (12.5)162 (54.6)98 (33.0)
SLC30A8 rs13266634C/TT = 0.251.15DIAGRAM166 (55.9)112 (37.7)19 (6.4)
THADA rs7578597A/GG = 0.131.15DIAGRAM241 (81.1)51 (17.2)5 (1.7)
CENTD2 rs1552224A/CC = 0.131.14DIAGRAM+226 (76.1)65 (21.9)6 (2.0)
NOTCH2 rs10923931T/GT = 0.111.13DIAGRAM5 (1.7)56 (18.9)236 (79.5)
ADCY5 rs11708067A/GG = 0.171.12MAGIC207 (69.7)80 (26.9)10 (3.4)
WFS1 rs10010131G/AA = 0.431.11DIAGRAM103 (34.7)131 (44.1)63 (21.2)
CDC123 rs4747969C/TC = 0.191.11DIAGRAM9 (3.0)97 (32.7)191 (64.3)
IRS1 rs7578326A/GG = 0.361.11DIAGRAM+123 (41.4)137 (46.1)37 (12.5)
CHCHD9 rs13292136C/TT = 0.061.11DIAGRAM+262 (88.2)32 (10.8)3 (1.0)
JAZF1 rs864745G/AA = 0.461.10DIAGRAM80 (26.9)151 (50.8)66 (22.2)
HNF1B rs757210T/CT = 0.421.10DIAGRAM59 (19.9)131 (44.1)107 (36.0)
HMGA2 rs1531343C/GC = 0.101.10DIAGRAM+2 (0.7)53 (17.9)242 (81.5)
ADAMTS9 rs4607103C/TT = 0.261.09DIAGRAM161 (54.2)118 (39.7)18 (6.1)
TSPAN8 rs7961581C/TC = 0.291.09DIAGRAM26 (8.8)119 (40.1)152 (51.2)
MTNR1B rs10830963G/CG = 0.321.09MAGIC36 (12.1)119 (40.1)142 (47.8)
BCL11A rs243021A/GA = 0.441.08DIAGRAM+58 (19.5)146 (49.2)93 (31.3)
SLC22A18AS rs231362G/AA = 0.431.08DIAGRAM+93 (31.3)152 (51.2)52 (17.5)
ZBED3-AS1 rs4457053G/AG = 0.351.08DIAGRAM+32 (10.8)142 (47.8)123 (41.4)
PROX1 rs340874C/TT = 0.421.07MAGIC104 (35.0)134 (45.1)59 (19.9)
GCK rs917793T/AT = 0.231.07MAGIC18 (6.1)103 (34.7)176 (59.3)
TSGA13 rs972283G/AA = 0.441.07DIAGRAM+100 (33.8)132 (44.6)64 (21.6)
VPS33B rs8042680A/CA = 0.391.07DIAGRAM+52 (17.5)129 (43.4)116 (39.1)
HNF1A rs7957197T/AA = 0.151.07DIAGRAM+213 (71.7)80 (26.9)4 (1.4)
DGKB rs2191349T/GG = 0.441.06MAGIC94 (31.7)146 (49.2)57 (19.2)
GCKR rs780094C/TT = 0.381.06MAGIC114 (38.4)143 (48.2)40 (13.5)
PLEKHF2 rs896854T/CC = 0.491.06DIAGRAM+75 (25.3)154 (51.9)68 (22.9)
BCL2A1 rs11634397G/AA = 0.331.06DIAGRAM+138 (46.5)121 (40.7)38 (12.8)
KCNQ1 rs2237895C/AC = 0.421.24Danish population48 (16.2)151 (50.8)98 (33.0)
TCF7L2 rs12255372T/GT = 0.361.43Wang et al. (2013)33 (11.1)149 (50.2)115 (38.7)

GRS for predictive modelling of GDM risk

Using 36 SNPs previously associated with T2D,[11, 16] we constructed a simple-count GRS ranging from 0 to 72, where each individual was scored on the basis of their genotype. Carriers of the risk allele could get a score of ‘1’ if they were heterozygotes (carriers of only one risk allele for a specific SNP) or a score of ‘2’ if they were homozygotes (carriers of two risk alleles for a specific SNP) for the risk allele. We then created a weighted GRS per participant by multiplying the number of risk alleles present per SNP by the β-estimate reported for this SNP in the meta-analysis of glucose and insulin-related traits consortium (MAGIC) and diabetes genetics replication and meta-analysis (DIAGRAM+) studies,[30-32] and from two other meta-analyses.[33, 34] Then, to make the GRS more specific to our population, we used the new explained-variance GRS model proposed by Che et al.[35] to account for the minor allele frequency of each SNP: explained-variance GRS = log10(Odds ratio)√[2Minor Allele Frequency(1−Minor Allele Frequency)]. In the case where the minor allele was not the allele associated with T2D risk, we considered the frequency of the risk allele as the substitute in the explained-variance GRS formula. Then, we summed the results over the 36 SNPs. The β-estimates are the natural log of the odds ratios listed in Table 1, and the minor allele frequencies were derived from the ALLELE procedure in sas genetics 9.3 (SAS Institute Inc., Cary, NC, USA).

Statistical analyses

Analyses were conducted with sas 9.2 (SAS Institute Inc.). Comparisons were performed through the General Linear Model procedure and using the type-III sum of squares (for unbalanced study design). Means and standard deviations (SDs) were computed for participants' characteristics, anthropometric, and metabolic variables (Table 2). Continuous variables were tested for normality of distribution, and log transformations of skewed variables were used in subsequent analyses, where necessary. All genotype distributions were tested for any deviation from Hardy–Weinberg equilibrium using the ALLELE procedure in sas genetics 9.3 (SAS Institute Inc.). Significance testing for linkage disequilibrium coefficient D was obtained using a chi-square test, likelihood ratio, and Fisher exact test (P ≤ 0.01). Receiver operating characteristics (ROC) curves were developed to evaluate the ability of each index to predict pre-diabetes and T2D states. Statistical significance was established at P ≤ 0.05.

Table 2. Participants' characteristics, and anthropometric and metabolic variables (n = 296)
 GDM n = 214Controls n = 82 P
Mean ± SD or n (%)95% CIMean ± SD or n (%)95% CI
  1. P values derived from a general linear model adjusted for age, parity, oral hypoglycaemic agents, ethnicity, and BMI.

  2. a

    P values derived from a general linear model.

  3. b

    P values derived from a general linear model adjusted for age, parity, oral hypoglycaemic agents, and ethnicity.

  4. c

    Data were log10 transformed.

Characteristicsa
Age, years36.4 ± 4.935.7–37.035.6 ± 5.234.5–36.80.26
Parity2.1 ± 0.82.0–2.22.2 ± 1.02.0–2.40.34
Time between index pregnancy and metabolic testing, years3.52 ± 1.973.25–3.793.82 ± 2.273.33–4.310.26
White189 (93.6%)76 (97.4%)
Impaired fasting glucose42 (19.63)1 (1.22)
Impaired glucose tolerance79 (36.92)10 (12.2)
Pre-diabetes, including type 2 diabetes135 (63.08)12 (14.63)
Type 2 diabetes40 (18.69)1 (1.22)
Normoglucotolerant79 (36.92)70 (85.37)
Oral hypoglycaemic agents7 (3.27)0 (0.0)
Insulin0 (0.0)0 (0.0)
Anthropometric indicatorsb
Weight, kg73.4 ± 17.571.1–75.868.7 ± 13.965.7–71.70.02
BMI, kg/m2c27.7 ± 6.526.8–28.525.6 ± 4.924.5–26.60.007
Waist circumference, cm91.1 ± 14.689.1–93.085.6 ± 11.983.1–88.20.003
Metabolic outcomes b
Fasting plasma glucose, mmol/lc5.92 ± 1.095.78–6.075.32 ± 0.385.24–5.40<0.0001
2-hour post-OGTT glucose, mmol/lc8.33 ± 2.837.95–8.716.00 ± 1.615.65–6.35<0.0001
Fasting insulin, ρmol/l83.54 ± 46.8277.24–89.8595.95 ± 39.4087.42–104.480.0003
2-hour post-OGTT insulin, ρmol/l613.07 ± 424.54555.65–670.49488.22 ± 346.88413.14–563.300.07
Area under the curve for glucose, mmol/l per minutec1063.61 ± 269.661027.40–1099.82795.02 ± 162.55759.62–830.42<0.0001
Area under the curve for insulin, ρmol/l per minute60 726.07 ± 32 538.662245.38–56 325.1358 667.04 ± 30 843.4751 950.02–65 384.060.97
HOMA-IS indexc0.61 ± 2.170.32–0.900.54 ± 1.540.21–0.870.03
Insulin secretion58.85 ± 31.0054.66–63.0473.36 ± 33.2066.13–80.600.0001
Insulinogenic indexc135.83 ± 121.62119.46–152.21319.01 ± 793.52146.20–491.82<0.0001
A1C (%)5.59 ± 3.875.54–5.655.34 ± 1.985.29–5.38<0.0001

Results

The clinical, anthropometric, and metabolic characteristics of the study participants are shown in Table 2. There were no statistical differences between women with prior GDM and the control group when comparing age, parity, and time to follow-up after delivery. Weight, BMI, and waist circumference were higher in women with prior GDM than in women of the control group (≤ 0.02 for all). Fasting plasma glucose, 2-hour plasma glucose, fasting insulin, area under the curve for glucose, HOMA-IS, and A1C levels were significantly higher in women with prior GDM compared with the control group (≤ 0.03 for all). Insulin secretion and insulinogenic indices were lower in women with prior GDM (≤ 0.0001 for all), compared with women from the control group. The area under the curve for insulin and 2-hour post-OGTT insulin level did not statistically differ among groups. A total of 19.6% of women with prior GDM were diagnosed with impaired fasting glucose, 36.9% with impaired glucose tolerance, and 18.7% were diagnosed with T2D, compared with 1.2, 12.2, and 1.2%, respectively, in the control group. Thus, among GDM women, 63.1% were identified as pre-diabetic 3 years after delivery.

All SNPs were in Hardy–Weinberg equilibrium. As shown in Table 3, the GRSs were higher in women with prior GDM compared with women from the control group (< 0.0001 for all). The associations remained significant after further adjustments for age, BMI, parity, oral hypoglycaemic agents and ethnicity (data not shown). As for the simple-count GRS, women with prior GDM had an average of 38.6 ± 3.9 risk alleles, whereas the control groups were carriers of 37.3 ± 3.2 risk alleles (< 0.0001).

Table 3. Differences in GRS between women with prior GDM and controls (n = 296)
 Prior GDM (mean ± SD) n = 21495% CIControls (mean ± SD) n = 8295% CI P
Simple-count GRS38.6 ± 3.938.1–39.137.4 ± 3.236.7–38.1<0.0001
Weighted GRS2.01 ± 0.271.98–2.051.96 ± 0.221.91–2.00<0.0001
Explained-variance GRS1.21 ± 0.181.18–1.231.17 ± 0.151.13–1.20<0.0001

Given that women with prior GDM had a higher T2D-associated GRS, we tested whether this GRS was also predictive of pre-diabetes or T2D in women with prior GDM. In a general linear model, the explained-variance GRS significantly differed between pre-diabetic women and women with prior GDM who remained normoglucotolerant at testing (1.20 ± 0.18 versus 1.17 ± 0.17, respectively, < 0.0001). The ROC curves (Figure 1) evaluated the discriminative power of the explained-variance GRS alone, and of age and BMI, or the additive effects of the three parameters together, between pre-diabetic women and normoglucotolerant women. The explained-variance GRS ROC curve had an area under the curve of 0.6122, whereas the age and BMI ROC curve had an area under the curve of 0.6269. When adding the explained-variance GRS into the logistic regression, we obtained an area under the curve of 0.6672, thus improving the predictive value.

Figure 1.

ROC curves: pre-diabetic (n = 135) versus normoglucotolerant (n = 79) women.

Using the same statistical model as previously described, we tested the difference in the GRS between women with prior GDM who remained normoglucotolerant at testing and women with T2D. All GRSs significantly differed between groups (< 0.0001 for all), as shown in Table 4. ROC curves were plotted using specificity and sensitivity data from a logistic regression, including the effects of age and BMI (area under the curve = 0.6761), explained-variance GRS (area under the curve = 0.5478), and the additive effects of age, BMI, and explained-variance GRS (area under the curve = 0.6772), as shown in Figure 2.

Table 4. Differences in GRS between normoglucotolerant and T2D status among women with prior GDM (n = 119)
 Normoglucotolerant (mean ± SD) n = 7995% CIT2D (mean ± SD) n = 4095% CI P
Simple-count GRS38.1 ± 4.037.2–39.038.7 ± 4.537.3–40.1<0.0001
Weighted GRS1.95 ± 0.251.90–2.012.00 ± 0.281.91–2.09<0.0001
Explained-variance GRS1.17 ± 0.171.13–1.201.20 ± 0.181.14–1.25<0.0001
Figure 2.

ROC curves: normoglucotolerant (n = 79) versus type 2 diabetic (n = 40) women.

Discussion

Main findings

According to the present study, a new genetic score including 36 SNPs previously associated with T2D was found to be associated with GDM. This study also provides evidence that this genetic score was predictive of pre-diabetes and T2D among women with prior GDM years after delivery. To our knowledge, this is the first study to show that combining genetic information within a score generates a predictor of both GDM and the progression to pre-diabetes and T2D among women with prior GDM.

Strengths and limitations

The strengths of this study include the use of the new explained-variance GRS model, because this new weighted method incorporates odds ratios from meta-analyses and the risk allele frequencies. Within the same odds ratio, disease risk will vary depending on the risk allele frequencies. It has been shown that when the sample size is small, the explained-variance GRS consistently yielded better performance than the weighted GRS.[35] In the present study, we used both approaches (explained-variance GRS and weighted GRS) and observed similar results. In addition, the SNPs that were studied were identified from large meta-analyses and recent GWAS.[12, 14] Moreover, the use of an OGTT to assess glucose and insulin responses is more sensitive for the detection of impaired glucose tolerance, pre-diabetes, and T2D.[24] Confounding factors such as age, time since the last pregnancy, and BMI were also considered in the present study. Additional adjustments for oral hypoglycaemic agents and ethnicity did not alter the results. Some limitations for this study should be acknowledged. Metabolic testing before pregnancy would have been interesting, as pre-pregnancy obesity and abnormal glucose levels are believed to be strong predictors of T2D. Also, adding more SNPs to the model, and not only SNPs that predict T2D, but others that prevent individuals developing T2D, may show better results as the new explained-variance GRS takes into account the minor allele frequency and the β-estimate from odds ratios.

Interpretation

Several studies have already reported associations between several SNPs and GDM. Huopio et al.[36] showed that rs10830963 of MTNR1B was significantly associated with GDM, fasting glucose levels, and reduced insulin secretion. Stuebe et al.[10] showed that GDM was associated with SNPs within TCF7L2 (rs7901695), MTNR1B (rs10830963), and GCK (rs780094), which are associated with T2D and fasting glucose levels in non-pregnant women. In addition, a higher number of risk alleles in women with a history of GDM was also reported by Lauenborg et al.,[37] where they observed associations in 11 SPNs, recently associated with T2D, with GDM, and demonstrated the combined additive effect of all T2D susceptibility alleles on the predictive risk of GDM, and performed ROC curves to assess the discriminative accuracy of these genetic variants in GDM. They showed that T2D risk alleles additively increase the risk of GDM.

The discriminative accuracy of genetic variants associated with T2D and GDM can be assessed looking at the area under the curves evaluated by ROC curves. The ROC curves were elaborated for the prediction of T2D risk based on 36 SNPs (included in a GRS), clinical characteristics (age and BMI), and both. Interestingly, the area under the curve of age and BMI was a stronger predictor of T2D risk when the explained-variance GRS was considered, suggesting that genetic variants included in the explained-variance GRS added to known predictive factors of T2D risk (Figures 1 and 2). In comparison, a recent GWAS has shown ROC curves with an area under the curve of 0.66 (95% CI 0.63–0.68) for the predictive modelling of T2D based on SNPs and clinical characteristics combined (age, sex, and BMI).[38] Previous studies have reported the area under the curves for genetic information, age, and BMI, and obtained a large range of results (0.73, 0.86, and 0.58), depending on the number of gene variants tested.[37, 39, 40] Differences between results from the present study and prior studies looking at genetic discrimination for T2D can be explained by the fact that in our study, ROC curves were evaluated only in women with prior GDM (GDM versus controls in the other studies), and therefore women were already more likely to develop T2D because of their history of GDM. Nevertheless, we observed that the GRS predicts progression to pre-diabetes and T2D, even in this high-risk population.

The effect sizes of common variants influencing T2D risk are low, and genetic predispositions may explain approximately 5–10% of the variance;[41] however, the estimated effect-size distribution suggested the existence of increasingly large numbers of susceptibility SNPs with decreasingly small effects.[42] SNPs with intermediate minor allele frequency (5–20%) contained an unusually small number of susceptibility loci and explained a relatively small fraction of heritability compared with what would be expected from the distribution.[43] For these reasons, the explained-variance GRS was elaborated by weighting each risk allele by its respective effect size on T2D risk and its minor allele frequency. In the present study, the explained-variance GRS indicated a genetic difference between GDM women and the control group, and also between pre-diabetic (including T2D) women and women with prior GDM who remained normoglucotolerant after testing. This finding is consistent with previous studies reporting that a combination of all risk alleles into a weighted GRS was significantly associated with the risk of T2D.[16] A higher GRS is also associated with less probability of returning to a normoglucotolerant state, and with an increased risk of developing T2D.[16]

Consistent with the literature, results of anthropometric measurements and glucose homeostasis clearly demonstrated metabolic deteriorations in women with prior GDM, compared with women without GDM.[2, 5] In the present study, women with prior GDM were overweight, had higher glycaemic traits, and had decreased insulin sensitivity. Pre-diabetes and T2D were more prevalent in women with prior GDM compared with women from the control group. Similarly, in Polish women with GDM, the cumulative incidence of impaired glucose tolerance and T2D over a 3-year follow-up was 41.5%.[44] Hunger-Dathe et al.[45] showed that the prevalence of impaired glucose tolerance was higher for pre-diabetes (28–53%) and for T2D (11–43%) in a longer follow-up after GDM (up to 3–15 years). By taking into account the new 2013 guidelines from the Canadian Diabetes Association,[20] more women were actually at risk because of lower cut-off values. In 2009, a meta-analysis with 675 455 women and 10 859 T2D cases demonstrated that women with GDM had a nearly 7.5-fold increased risk of predicted T2D.[5] Kim et al.[6] reported in a systematic review that more than 70% of women with GDM will eventually progress to T2D. The identification of factors associated with subsequent T2D development in women with GDM may determine subjects at risk, in which case more aggressive preventive care could be provided to prevent the development of pre-diabetes and T2D. Indeed, lifestyle intervention has been shown to attenuate the genetic contribution to T2D risk.[16, 46]

Conclusion

In conclusion, findings from the present study suggest that a genetic score is associated with both GDM and progression to pre-diabetes and T2D in women with prior GDM. In addition, we observed that including a genetic score and traditional risk factors for T2D, such as age and BMI, resulted in a greater discrimination of the risk of pre-diabetes in women with prior GDM. This result suggests that using an explained-variance GRS combined with traditional risk factors for T2D is likely to be more accurate in predicting future risk of progression to T2D among women with prior GDM.

Disclosure of interests

The authors did not declare any conflicts of interest.

Contribution to authorship

HC and JV contributed equally to this work. HC and JV performed the statistical analyses, interpreted the data, and wrote the article; JV and VG participated in the recruitment of the women and collected data. JR was the principal investigator and designed the study, supervised the research, directed the data analysis and interpretation, and assisted with the manuscript preparation. SJW, MCV, and AT assisted with the development of the research study design and protocol, as well as data analysis and interpretation. All the authors revised the article.

Details of ethics approval

The data collection scheme and the study protocol were approved by the Laval University Hospital Research Center Ethics Committee (ref. no. 129.05.10) on 27 May 2009.

Funding

This work was supported by the Fonds de la recherche du Québec en santé (FRQS) and the Canadian Institutes of Health Research (CIHR) (grant number OOP-98026).

Acknowledgements

We thank all the participants of this study. We would also like to express our gratitude to Geneviève Faucher and Ann-Marie Paradis, from Laval University, for their help with the recruitment of participants and with technical support. This work is supported by the Canadian Institutes for Health Research (CIHR) and by the Fonds de la recherche du Québec en santé (FRQS). André Tchernof is a research chair in bariatric and metabolic. Marie-Claude Vohl holds a Tier 1 Canada Research Chair in Genomics Applied to Nutrition and Health. Julie Robitaille is the recipient of a Junior Investigator scholarship from the FRQS.

Ancillary