SEARCH

SEARCH BY CITATION

Keywords:

  • genetic risk score;
  • HbA1c;
  • nondiabetic;
  • replication;
  • SNP

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

Objectives

Glycated haemoglobin (HbA1c) is associated with cardiovascular disease risk in individuals without diabetes, and its use has been recommended for diagnosing diabetes. Therefore, it is important to gain further understanding of the determinants of HbA1c. The aim of this study was to investigate the effects of genetic loci and clinical and lifestyle parameters, and their interactions, on HbA1c in nondiabetic adults.

Design

Population-based cohort study.

Setting

Three northern provinces of the Netherlands.

Subjects

A total of 2921 nondiabetic adults participating in the population-based LifeLines Cohort Study.

Measurements

Body mass index (BMI), waist circumference, HbA1c, fasting plasma glucose (FPG) and erythrocyte indices were measured. Data on current smoking and alcohol consumption were collected through questionnaires. Genome-wide genotyping was performed, and 12 previously identified single-nucleotide polymorphisms (SNPs) were selected for replication and categorized as ‘glycaemic’ and ‘nonglycaemic’ SNPs according to their presumed mechanism(s) of action on HbA1c. Genetic risk scores (GRSs) were calculated as the sum of the weighted effect of HbA1c-increasing alleles.

Results

Age, gender, BMI, FPG, mean corpuscular haemoglobin, mean corpuscular haemoglobin concentration, current smoking and alcohol consumption were independent predictors of HbA1c, together explaining 26.2% of the variance in HbA1c, with FPG contributing 10.9%. We replicated three of the previously identified SNPs and the GRSs were also found to be independently associated with HbA1c. We found a smaller effect of the ‘nonglycaemic GRS' in females compared with males and an attenuation of the effect of the GRS of all 12 SNPs with increasing BMI.

Conclusions

Our results suggest that a substantial portion of HbA1c is determined by nonglycaemic factors. This should be taken into account when considering the use of HbA1c as a diagnostic test for diabetes.


Abbreviations
BMI

body mass index

FPG

fasting plasma glucose

GRS

genetic risk score

GWAS

genome-wide association study

Hb

haemoglobin

Ht

haematocrit

MCH

mean corpuscular haemoglobin

MCHC

mean corpuscular haemoglobin concentration

MCV

mean corpuscular volume

RBC

red blood cell

SD

standard deviation

SNP

single-nucleotide polymorphism

Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

It has been firmly established that glycated haemoglobin (HbA1c) is an index of long-term glucose control and a predictor of complications in patients with diabetes mellitus [1]. In addition, an association between HbA1c and cardiovascular disease risk has been shown in several studies in individuals without diabetes [2, 3]. An international expert committee recommended that HbA1c could be used as an indicator for the diagnosis of diabetes [4], but this is still strongly debated [5, 6]. Therefore, it is important to gain better insight into the factors that determine HbA1c in nondiabetic adults and their mechanisms of action.

Despite the fact that HbA1c is the most widely used and well-validated measure for determining average glycaemic control over time, discordance between HbA1c and other measures of glycaemic control is commonly observed. These differences between blood glucose data and HbA1c levels have generated controversy regarding the role of the sources of variation in HbA1c [7]. Evidence suggests that HbA1c and glucose level partly reflect different processes, particularly in the nondiabetic range of glucose tolerance [8, 9].

Variation in HbA1c is thought to be subject to both genetic and nongenetic determinants [10]. Nongenetic factors may explain about 40–60% of the HbA1c level in nondiabetic subjects [11, 12]. The two clinical variables that most clearly appear to influence HbA1c levels are red blood cell survival and blood glucose concentration. However, there are several other (glycaemia-independent) variables that can influence HbA1c, including race, smoking and age [13-15].

It has been estimated that the heritability of HbA1c is 40–60% [11, 12]. A total of 15 independent single-nucleotide polymorphisms (SNPs) that were significantly associated with HbA1c in nondiabetic adults were identified in two genome-wide association studies (GWASs) [16, 17]. The physiological mechanisms of regulation of HbA1c levels by these genetic loci remain unclear. It is considered that some SNPs modulate glycaemic physiology [18], whereas others may regulate nonglycaemic factors such as red blood cell (RBC) function [19].

The aim of this study was to investigate a wide range of determinants of HbA1c in nondiabetic Dutch adults participating in the population-based LifeLines Cohort Study [20]. We selected genetic loci that have been established to be associated with HbA1c in previous GWASs [16, 17] and studied the individual and combined effects of these SNPs as well as the effects on HbA1c of clinical and lifestyle parameters and their interactions with genetic factors.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

LifeLines cohort study participants

The study population was derived from the LifeLines Cohort Study, an observational follow-up study in a large representative sample of the population of the three northern provinces of the Netherlands including three generations [20]. All participants completed a number of questionnaires and underwent a medical examination at baseline, and are being followed longitudinally with extensive standardized measurements. The study was approved by the Ethics Committee of the University Medical Centre Groningen.

Recruitment has been ongoing since the end of 2006, and since April 2009 the oldest unrelated subject(s) from each family have been selected for inclusion in the GWAS. For this study, we selected 3367 unrelated participants of western European descent who underwent the baseline medical examination, completed the questionnaires and from whom genome-wide data of good quality were available. We excluded participants with diabetes mellitus (= 156), either self-reported or diagnosed by a fasting plasma glucose (FPG) ≥7.0 mmol L−1. In addition, we excluded participants with significant anaemia (= 86), defined by a haemoglobin level <8.2 mmol L−1 for men and <7.0 mmol L−1 for women, and those without an HbA1c value (= 13), erythrocyte measurements (= 10) or FPG (= 181). The final study population comprised 2921 nondiabetic adults.

Anthropometry

Weight was measured to the nearest 0.1 kg and height to 0.1 cm by trained research staff using calibrated measuring equipment, with participants wearing light clothing. Body mass index (BMI) was calculated as weight/height squared (kg m−2). Subjects were considered overweight and obese with BMI values >25 kg m−2 and >30 kg m−2 respectively. Waist circumference, to the nearest 0.1 cm, was measured twice midway between the lowest rib and the top of the iliac crest at the end of gentle expiration. The mean of the two measurements was used in the analysis.

Laboratory analyses

For HbA1c analysis, a fasting whole blood sample (EDTA-anticoagulated) was collected and analysed using a turbidimetric inhibition immunoassay on a Cobas Integra 800 CTS analyser (Roche Diagnostics Nederland BV, Almere, the Netherlands). This method has been standardized against the reference method of the International Federation of Clinical Chemistry and Laboratory Medicine. Between-batch imprecision (coefficient of variation) was 2.1% for a mean HbA1c of 5.5%, and 1.9% for a mean HbA1c of 10.6%.

FPG was measured from fasting fresh venous plasma using the Roche glucose assay (hexokinase/glucose-6-phosphate dehydrogenase enzymatic reactions) on the Modular P analyser (Roche Diagnostics, Burgdorf, Switzerland).

Hb, haematocrit (Ht), RBC count, mean corpuscular volume (MCV), mean corpuscular haemoglobin (MCH) and mean corpuscular haemoglobin concentration (MCHC) were assessed using a Sysmex XE-2100 automated haematology analyser (Sysmex Corporation, Kobe, Japan). Gender-specific z-scores were calculated for Hb, Ht, RBC count, MCV, MCH and MCHC, to take into account differences between men and women. These gender-specific z-scores were used in all models.

Current smoking and alcohol consumption

Data were obtained from the questionnaires on current smoking and alcohol consumption, defined as any smoking and any alcohol consumption in the previous month (yes/no). Data on alcohol consumption were missing for 89 participants.

Genotyping, quality control and imputation

For this study, genome-wide genotyping was performed in all participants using the Illumina HumanCytoSNP12 v2 beadchip assay (Illumina, Inc; San Diego, CA, USA). Genotypes were called with the Illumina GenomeStudio software package (Illumina, Inc). SNPs were excluded if the genotype call rate was <95%, the minor allele frequency was <1% or the Hardy-Weinberg P-value was <10−4. Subjects were excluded if the sample call rate was <95%, if the subject was not Caucasian (assessed with EigenStrat [21]), or if the subject was highly related to another individual [22] and had a lower sample call rate. The resulting data set contained 257,581 SNPs from 3367 individuals. Untyped SNPs were imputed with IMPUTE using the HapMap CEU Phase II reference set (release 22, build 36) [23].

Selected SNPs

For this study we selected genome-wide significant SNPs from two large GWASs [16, 17]. Soranzo et al. [16] identified 11 independent SNPs and Paré et al. [16, 17] found four independent SNPs associated with HbA1c. Of these 15 SNPs, two had an imputation quality score <0.5 (rs1387153 and rs16926246) in our sample, and rs730497 was in perfect linkage disequilibrium (LD) with rs1799884. These three SNPs were excluded from the analyses and the remaining 12 were used for replication in the current study (Table A1). Because these SNPs were selected for hypothesis-based replication on the basis of existing GWAS evidence, there was no need to apply strict adjustment for multiple testing.

We categorized the 12 SNPs according to the action they are considered to have on HbA1c. The loci of SNPs rs552976, rs1402837, rs1799884 and rs13266634 have previously been associated with fasting glucose or reduced glucose-induced insulin secretion [24-27]. Consequently, these ‘glycaemic’ loci were expected to affect HbA1c through a glycaemic pathway. By contrast, the other SNPs are thought to influence HbA1c via ‘nonglycaemic’ erythrocyte and iron biology pathways [16, 19].

A genetic risk score (GRS) was calculated for each individual as the sum of the HbA1c-increasing alleles of all 12 SNPs weighted by their respective effect sizes as observed by Soranzo et al. and Paré et al. [16, 17, 28]. In addition, we calculated a GRS for the four ‘glycaemic’ and eight ‘nonglycaemic’ SNPs separately. We combined the effects of the SNPs to calculate the GRSs irrespective of the significance of the individual SNPs in the current study because they were identified as true positive signals in previous GWASs. For construction of the GRSs by simply adding the SNP effect sizes, independence of the SNP effects is assumed. Therefore, we checked for LD between the associated SNPs reported in earlier GWASs and hence excluded rs730497 (see above) from the analyses. Wray et al. in a previous review of the use of GRSs noted that, in the long term, prediction of this additive genetic risk can be combined with predictions from environmental risk factors and nonadditive genetic factors, and the interactions between these variables [29]. Except for the inclusion of nonadditive factors (i.e. gene–gene interactions), this is the method we have used in the current study. Of note, we multiplied the reported SNP effect sizes by 10.93, which is the factor used to convert HbA1c in percentages to HbA1c in mmol mol−1 [30]. Consequently, the effect size of the GRSs is also given in mmol mol−1.

Statistical methods

We used multiple linear regression to investigate the relations between all potential determinants and HbA1c. First a model with only age and gender was analysed. In the next steps, each determinant was added to the age and gender model separately. Then, a forward-stepwise approach with all clinical and lifestyle factors was used to derive a final predictive model of HbA1c including only significant terms. Thus, age and gender were forced into the model and the other clinical and lifestyle factors were added subsequently by stepwise modelling. Because BMI and waist circumference were highly correlated (r = 0.80, < 0.001), we initially only included BMI in the modelling. The GRSs were each separately added to this so-called final predictive model to determine their effect on HbA1c as well as the proportion of variance explained.

Gene–'environment' interactions were calculated by multiplying the GRS of all 12 SNPs by the other (‘nongenetic’) determinants. Initially, all gene–environment interactions were tested separately for their association with HbA1c in the first model (only age and gender included). Thereafter, significantly associated gene–environment interactions were added to the final predictive model to test their independent association with HbA1c.

All analyses were performed with SPSS version 16.0 (SPSS Inc., Chicago, IL, USA). A value of < 0.05 was considered statistically significant.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

The characteristics of the study population are shown in Table 1.

Table 1. Characteristics of the study population
CharacteristicsValues
  1. BMI, body mass index; MCH, mean corpuscular haemoglobin; MCHC, mean corpuscular haemoglobin concentration; MCV, mean corpuscular volume; RBC, red blood cell; Values are means ± SD or number (%)

Age (years)55.5 ± 9.9
Male (%)1176 (40.3)
BMI (kg m−2)26.5 ± 4.1
Normal (%)1112 (38.1)
Overweight (%)1322 (45.3)
Obese (%)487 (16.7)
Waist circumference (cm)92.9 ± 11.7
Current smoking [n (%)]566 (19.4)
Alcohol consumption [n (%)]2344 (80.3)
HbA1c (%)5.6 ± 0.3
HbA1c (mmol mol−1)38 ± 3
Fasting plasma glucose (mmol L−1)5.1 ± 0.5
Haemoglobin (mmol L−1)8.7 ± 0.7
Haematocrit (L L−1)0.42 ± 0.03
RBC count (×10−12 L−1)4.7 ± 0.4
MCV (fL)90.4 ± 4.2
MCH (fmol)1.9 ± 0.1
MCHC (mmol L−1)20.6 ± 0.6

Clinical and lifestyle parameters and HbA1c

There were positive associations between HbA1c and age, BMI, waist circumference, FPG, haematocrit, RBC count and current smoking, and negative associations with haemoglobin level, MCV, MCH, MCHC and alcohol consumption (Table 2). Gender was not associated with HbA1c in the baseline model (including only age and gender).

Table 2. Associations between HbA1c and demographic, clinical and lifestyle parameters and genetic risk factors
 HbA1c (mmol mol−1) 
  1. BMI, body mass index; FPG, fasting plasma glucose; GRS, genetic risk score; MCH, mean corpuscular haemoglobin; MCHC, mean corpuscular haemoglobin concentration; MCV, mean corpuscular volume; RBC, red blood cell; SNP, single-nucleotide polymorphism; R2: explained variance in HbA1c; ∆R2: additional explained variance compared with the explained variance of the age and gender model; *Linear regression model with risk factors separately added to the baseline model including age and gender, †Erythrocyte indices are expressed as z-scores, ‡For alcohol consumption, the model with age and gender consisted of = 2832 participants with a value of R2 (%) of 9.6, §Replicated SNPs.

 BetaCIP-valueR2 (%)
Baseline model    10.8
Age (years)0.110.10–0.12<0.001 
Gender−0.13−0.37–0.110.276 
Clinical and lifestyle parameters Beta*CIP-value∆R2 (%)
BMI (kg m−2)0.130.10–0.16<0.0012.5
Waist circumference (cm)0.050.04–0.06<0.0012.4
FPG (mmol L−1)2.292.07–2.51<0.00111.2
Haemoglobin−0.13−0.25 to −0.020.0270.1
Haematocrit0.170.05–0.280.0060.2
RBC count0.230.12–0.35<0.0010.5
MCV−0.14−0.25 to −0.020.0250.1
MCH−0.45−0.57 to −0.34<0.0011.8
MCHC−0.62−0.73 to −0.50<0.0013.2
Current smoking0.700.40–1.00<0.0010.6
Alcohol consumption−0.94−1.26 to −0.63<0.0011.1
Genetic factors SNP Beta*CIP-value∆ R2 (%)
rs5529760.10−0.07–0.270.2350.0
rs1402837 0.280.08–0.480.007§0.2
rs1799884 0.19−0.04–0.410.0990.1
rs13266634 −0.02−0.20–0.160.8620.0
rs2779116 0.03−0.16–0.220.7450.0
rs1800562 0.14−0.21–0.500.4360.0
rs6474359 0.35−0.05–0.750.0880.1
rs4737009 −0.26−0.46 to −0.050.014§0.2
rs7072268 −0.34−0.54 to −0.140.0010.3
rs7998202 0.17−0.11–0.440.2360.0
rs1046896 0.200.00–0.400.046§0.1
rs855791 −0.11−0.28–0.060.2060.0
GRS of all 12 SNPs0.320.15–0.48<0.0010.4
Glycaemic GRS0.280.05–0.510.0180.2
Nonglycaemic GRS0.320.09–0.540.0050.2

In a forward-stepwise multiple linear regression model, age, gender, BMI, FPG, MCH, MCHC, current smoking and alcohol consumption were independent predictors of HbA1c. This final predictive model with eight ‘nongenetic’ factors explained 26.2% of the variance in HbA1c (Table 3), of which FPG alone explained 10.9%. We also used forward-stepwise modelling to derive a final predictive model including waist circumference instead of BMI; however, the results remained largely unchanged. Defining alcohol use (g day−1) and smoking (number of cigarettes per day) as continuous instead of dichotomous variables did not improve (for alcohol use) or only very slightly affected (for smoking: 0.3%) the explained HbA1c variance in the model.

Table 3. Multivariate associations between HbA1c and clinical and lifestyle parameters and genetic risk factors
 HbA1c (mmol mol−1)  
Beta*CIP-valueR2 (%)
  1. BMI, body mass index; FPG, fasting plasma glucose; GRS, genetic risk score; MCH, mean corpuscular haemoglobin; MCHC, mean corpuscular haemoglobin concentration; SNP, single-nucleotide polymorphism; R2: explained variance in HbA1c (%); ∆R2: added explained variance of the GRSs to the model (%); Model IV: Model III + GRS of all 12 SNPs; Model IVa: Model III + glycaemic GRS; Model IVb: Model III + nonglycaemic GRS; *Multiple linear regression model.

Model I     9.6
Age (years)0.110.10–0.12<0.001 
Gender−0.14−0.38–0.100.263 
Model II    20.5
Age (years)0.080.07–0.09<0.001 
Gender0.350.12–0.580.003 
FPG (mmol L−1)2.252.03–2.48<0.001 
Model III    26.2
Age (years)0.070.06–0.09<0.001 
Gender0.250.03–0.480.029 
BMI (kg m−2)0.050.02–0.080.001 
FPG (mmol L−1)2.131.91–2.36<0.001 
MCH−0.21−0.33 to −0.090.001 
MCHC−0.56−0.68 to −0.44<0.001 
Current smoking0.810.53–1.09<0.001 
Alcohol consumption−0.73−1.02 to −0.44<0.001 
 Beta*CIP-value∆ R2 (%)
Model IV
GRS of all 12 SNPs0.200.05–0.350.0100.2
Model IVa
Glycaemic GRS0.230.02–0.440.0320.2
Model IVb
Nonglycaemic GRS0.15−0.06–0.350.1510.1

Genetic factors and HbA1c

In the current study we replicated three of the twelve SNPs previously identified from GWASs for HbA1c: rs1402837, rs4737009 and rs1046896 (Table 2). SNP rs7072268 was also significantly associated with HbA1c but, in contrast to other studies [17, 19], we found a negative effect of the T-allele. Hence, this finding cannot be considered a replication. The three GRSs were all positively and significantly associated with HbA1c, indicating a cumulative effect of the different loci (Table 2). Adding the GRS of all 12 SNPs to the final model with independent nongenetic risk factors increased the percentage of explained variance in HbA1c by 0.2% (Table 3). The effect of the ‘nonglycaemic GRS’ on HbA1c markedly attenuated when it was added to the final predictive model with the eight independent nongenetic risk factors, whereas the effect of the ‘glycaemic GRS’ on HbA1c remained largely the same.

Gene–environment interactions

Two interactions between the GRS of all 12 SNPs and ‘nongenetic’ factors remained significant after inclusion in the final multivariate model. The effect of the GRS of all 12 SNPs on HbA1c was significantly smaller in women compared with men. In addition, the effect of the GRS of all 12 SNPs attenuated with increasing BMI. Both interactions together added 0.3% to the explained variance in HbA1c (Table A2).

To further explore the interactions between the GRSs and gender, we stratified the population according to gender and calculated the effect of the three GRSs on HbA1c in men and women separately (Table A3). In men we found a positive and highly significant association between the ‘nonglycaemic GRS’ and HbA1c, whereas in women there was no significant association. Thus, the observed gender difference in the effect of the GRS of all SNPs on HbA1c is largely explained by a difference in the effect of the ‘nonglycaemic GRS’. In addition, we stratified the population into normal, overweight and obese individuals and calculated the effect of the three GRSs on HbA1c in these three groups separately (Table A4). There was a positive association between all three GRSs and HbA1c in normal and overweight individuals, but no association between all three GRSs and HbA1c in obese individuals.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

The aim of this study was to investigate the genetic and nongenetic determinants of HbA1c in nondiabetic adults to evaluate the relative contribution of different factors (glycaemic vs. nonglycaemic and genetic vs. nongenetic) to HbA1c levels in individuals without diabetes. We found that age, gender, BMI, FPG, MCH, MCHC, current smoking and alcohol consumption were independently associated with HbA1c. A final predictive model with these eight nongenetic determinants explained 26.2% of the variance in HbA1c, of which FPG contributed less than half (i.e. 10.9%). In addition, we were able to confirm in our study population the previously identified association between three of 12 SNPs and HbA1c. The GRSs were also independently associated with HbA1c and explained 0.2% of the variance in HbA1c. There was a lower effect of the nonglycaemic GRS in women than men and an attenuation of the effect of the GRS of all 12 SNPs with increasing BMI. To our knowledge, this is the first study to identify these gene–environment interactions, therefore, they need to be viewed with some caution until independently confirmed in other population-based cohorts.

Because erythrocytes are freely permeable to glucose, the rate of formation of HbA1c is directly proportional to the ambient glucose concentration. However, a lack of agreement between HbA1c and other measures of glycaemic control is common. The reason for this remains unclear and the role of other sources of variation in HbA1c remains controversial. From the findings of a recent study investigating the relationships amongst HbA1c, FPG and 2-h postload plasma glucose, it was concluded that there is only a moderate correlation between glucose and HbA1c in the general population [8]. About 20 years ago, Yudkin et al. suggested that the degree of glucose intolerance may explain only one third of the variance in HbA1c in a nondiabetic population [9]. This suggestion is in line with our findings that the final predictive model with eight demographic, clinical and lifestyle parameters explained 26.2% of the variance in HbA1c, with only almost half due to FPG.

Glycation of haemoglobin A (HbA) begins during erythropoiesis and continues throughout the lifespan of the RBC in the circulation. Consequently, RBC lifespan determines the duration of exposure of haemoglobin to glucose and thereby HbA1c levels. Increased erythrocyte turnover, as observed in patients with haemolytic anaemia, results in lower HbA1c levels [31]. By contrast, higher HbA1c levels were observed in patients with iron deficiency anaemia, probably because of a relatively high proportion of old erythrocytes [32, 33]. In line with our findings, Koga et al. demonstrated that MCH was negatively associated with HbA1c in premenopausal women; they suggested that this was explained by menstrual blood loss that causes iron deficiency and thereby relatively low levels of MCV and MCH [34]. As our study population consisted of men and women (pre as well as postmenopausal), menstrual blood loss cannot be the only explanation of the observed relation between MCH and HbA1c. It is possible that a relative lack of iron caused by factors other than blood loss also leads to a relatively long lifespan of erythrocytes in individuals with lower MCH. Cohen et al. concluded that erythrocyte survival varies sufficiently amongst haematologically normal individuals to cause differences in HbA1c [35]. If MCH is related to this variance in erythrocyte survival, this could explain the observed differences in HbA1c. More studies are needed to investigate the mechanisms by which erythrocyte indices influence HbA1c levels. However, the relatively high contribution of MCHC and MCH to the explained variance in HbA1c highlights that factors other than glycaemia determine HbA1c levels.

In line with other studies, we found a negative association between alcohol consumption and HbA1c [36, 37]. Facchini et al. found that low-to-moderate alcohol consumption in healthy men and women was associated with enhanced insulin-mediated glucose uptake and lower plasma glucose and insulin concentrations in response to oral glucose [38]. The resulting lower levels of glycaemia could explain the lower levels of HbA1c related to alcohol consumption. Our finding of higher HbA1c levels in current smokers is also in line with results of earlier studies [14, 39, 40]. Even after adjusting for potential confounders such as abdominal obesity, there is a positive association between current smoking and HbA1c. Glycotoxins in cigarette smoke may induce the higher rate of glycation of HbA [41] or the relatively high degree of tissue hypoxia [42] may explain increased HbA1c levels in smokers [43].

In the current study we were able to replicate three of the previously reported significant SNPs associated with HbA1c, but the contribution of these SNPs to the explained variance in HbA1c is low. In contrast to other studies, we found a negative association between the T-allele of SNP rs7072268 and HbA1c, although this may have been a false-positive finding [17, 19]. Fluctuation of effects of individual SNPs is no longer a problem when the effects of all SNPs are added to calculate a GRS. This GRS can be used to determine the extent of the effect of all SNPs together on HbA1c. The GRS of all 12 SNPs was independently associated with HbA1c, but contributed only 0.2% to the explained variance in HbA1c. This small contribution of the genetic factors to the explained variance in HbA1c cannot be explained by the fact that it is introduced into the multivariate model after the clinical and lifestyle parameters, as the genetic factors also show only modest association with HbA1c in the univariate analyses (Table 2).

The effect of the ‘nonglycaemic GRS’ on HbA1c markedly attenuated when it was added to the final predictive model with the eight independent nongenetic risk factors. This attenuation could be due to the fact that MCHC and/or MCH act as mediator variables. If the action of these SNPs on HbA1c is at least in part through their effect on MCHC and/or MCH, the effect of the GRS will be reduced when MCHC and/or MCH is also included in the model. The effect of the glycaemic GRS on the other hand was largely unchanged.

The nonglycaemic GRS was positively associated with HbA1c in men but not in women, demonstrating a gender difference in the effect of these nonglycaemic SNPs on HbA1c. In addition, the effect of the GRS of all 12 SNPs was attenuated with increasing BMI. It is possible that the effect of being obese overrides the potential genetic effect of the SNPs.

An important strength of our study is that we were able to investigate the association between a wide range of nongenetic risk factors and HbA1c in a large population-based study. Consequently, we could adjust for potential confounding and investigate the independent associations between the risk factors and HbA1c. A limitation of this study is the lack of measures of glycaemia other than FPG, which does not capture variation in glycaemia throughout the day. Therefore, the relatively modest contribution of FPG to the explained variance in HbA1c was not unexpected. However, repeated measurements of glucose levels are not feasible in a cohort study with nondiabetic participants. More importantly, in spite of this lack of other measures of glycaemia the contribution of nonglycaemic factors to the explained variance in HbA1c is evident. Another potential limitation is that the oldest unrelated subject(s) from each family participating in the LifeLines Cohort Study were selected for inclusion in the GWAS. As a consequence, the sample in the current study is older and therefore not entirely representative of all participants in the LifeLines Cohort Study.

In conclusion, age, gender, BMI, FPG, MCH, MCHC, current smoking and alcohol consumption are independently associated with HbA1c. In addition, we were able to replicate the association between three previously identified SNPs and HbA1c in our study population. The added effect of the previously identified nonglycaemic SNPs is different in men and women, and the additional effect of all 12 SNPs was attenuated with increasing BMI. Our results suggest that a substantial part of the individual differences in HbA1c in the general population are determined by nonglycaemic factors. This should be taken into account when considering using HbA1c as a diagnostic test for diabetes.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

We thank Behrooz Alizadeh, Annemieke Boesjes, Marcel Bruinenberg, Noortje Festen, Ilja Nolte, Lude Franke and Mitra Valimohammadi for their help in creating the GWAS database; and Rob Bieringa, Joost Keers, René Oostergo, Rosalie Visser and Judith Vonk for data collection and validation. We are grateful to the study participants, the staff from the LifeLines Cohort Study and Medical Biobank Northern Netherlands and the participating general practitioners and pharmacists.

Conflict of interest statement

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

None of the authors has any conflicts of interest to declare.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix

Appendix

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. Conflict of interest statement
  9. References
  10. Appendix
Table A1. The 12 selected SNPs used for replication in this study
SNPChrPosition*Effect allele/other alleleNearest locus
  1. Chr, chromosome; Pos, position; SNP, single-nucleotide polymorphism; *NCBI Genome build 36, release 22.

Glycaemic
rs5529762169 499 684 G/A G6PC2/ABCB11
rs1402837 2169 465 600T/C G6PC2
rs1799884 744 195 593T/C GCK
rs13266634 8118 253 964T/C SLC30A80A
Nonglycaemic
rs2779116 1156 852 039T/C SPTA1
rs1800562 626 201 120G/A HFE
rs6474359 841 668 351T/C ANK1
rs4737009 841 749 562G/A ANK1
rs7072268 1070 769 919T/C HK1
rs7998202 13112 379 869G/A ATP11A/TUBGCP3
rs1046896 1778 278 822T/C FN3K
rs855791 2235 792 882G/A TMPRSS6
Table A2. Final multivariate model for HbA1c including gene–environment interactions
 HbA1c (mmol mol−1)P-valueR2 (%)
Beta*CI
  1. BMI, body mass index; FPG, fasting plasma glucose; GRS, genetic risk score; MCH, mean corpuscular haemoglobin; MCHC, mean corpuscular haemoglobin concentration; SNP, single-nucleotide polymorphism; R2: explained variance in HbA1c (%);*Multiple linear regression model.

Model    26.7
Age (years)0.070.06–0.09<0.001 
Gender1.520.52–2.520.003 
BMI (kg m−2)0.200.08–0.320.001 
FPG (mmol L−1)2.111.88–2.34<0.001 
MCH−0.21−0.33 to −0.090.001 
MCHC−0.55−0.67 to −0.43<0.001 
Current smoking0.830.55–1.11<0.001 
Alcohol consumption−0.73−1.02 to −0.44<0.001 
GRS of all 12 SNPs2.130.98–3.27<0.001 
Interaction GRS of all 12 SNPs × gender−0.40−0.71 to −0.090.011 
Interaction GRS of all 12 SNPs × BMI−0.05−0.09 to −0.010.012 
Table A3. Effect of GRSs on HbA1c in male and female subjects
 Male (= 1134)Female (= 1698)
HbA1c (mmol mol−1)HbA1c (mmol mol−1)
Beta*CIP-valueBeta*CIP-value
  1. GRS, genetic risk score; SNP, single-nucleotide polymorphism; *Linear regression model adjusted for the final predictive model including age, body mass index, fasting plasma glucose, mean corpuscular haemoglobin, mean corpuscular haemoglobin concentration, current smoking and alcohol consumption.

GRS of all 12 SNPs0.420.18–0.670.0010.05−0.14–0.240.634
GRS of glycaemic SNPs0.29−0.06–0.630.1020.19−0.08–0.450.169
GRS of nonglycaemic SNPs0.490.16–0.820.004−0.09−0.35–0.170.494
Table A4. Effect of GRSs on HbA1c in normal, overweight and obese subjects
 Normal (= 1091)Overweight (= 1278)Obese (= 463)
HbA1c (mmol mol−1)HbA1c (mmol mol−1)HbA1c (mmol mol−1)
Beta*CIP-valueBeta*CIP-valueBeta*CIP-value
  1. GRS, genetic risk score; SNP, single-nucleotide polymorphism; *Linear regression model adjusted for the final predictive model including age, gender, body mass index, fasting plasma glucose, mean corpuscular haemoglobin, mean corpuscular haemoglobin concentration, current smoking and alcohol consumption.

GRS of all 12 SNPs0.300.07–0.530.0120.280.05–0.510.017−0.25−0.66–0.150.218
GRS of glycaemic SNPs0.430.11–0.750.0090.21−0.11–0.540.197−0.19−0.74–0.370.512
GRS of nonglycaemic SNPs0.13−0.18–0.450.4100.330.01–0.640.041−0.27−0.79–0.260.326