Genetic Variant in HK1 Is Associated With a Proanemic State and A1C but Not Other Glycemic Control–Related Traits

Clinical characteristics and data available on the studied populations are reported in Table 1. The study protocol was approved by the local ethics committee, and participants from all of the studies described (and the parents of children) signed an informed consent form.

The Data from the Epidemiological Study on the Insulin Resistance Syndrome (D.E.S.I.R.) cohort is a longitudinal French general population described elsewhere (10,26). We analyzed 4,590 nondiabetic D.E.S.I.R. participants successfully genotyped for rs7072268, of whom 3,795 were examined throughout the 9-year study.

The Swiss cohort study of obese adults has previously been described (27). All of the subjects were recruited for obesity surgery. We analyzed 2,363 nondiabetic participants successfully genotyped for rs7072268.

The Northern Finland 1986 Birth Cohort (NFBC1986) is a prospective 1-year birth cohort including all Finnish Caucasian mothers with children whose expected date of birth fell between 1 July 1985 and 30 June 1986 in the two northernmost provinces of Finland (28). Clinical examination at 15–16 years of follow-up was conducted between August 2001 and June 2002. We analyzed 5,287 nondiabetic participants successfully genotyped for rs7072268 in the NFBC1986 cohort.

The Haguenau community-based cohort of young adults investigates long-term consequences of being born small for gestational age and has previously been described (29). Briefly, subjects born between 1971 and 1985 were identified from a population-based registry of Haguenau (France). Non–European ancestry subjects are estimated to be <0.1% of the general population (29). At a mean age of 22 years, participants under overnight fasting conditions underwent a medical examination for assessment of anthropometric and clinical parameters. We analyzed 1,455 nondiabetic participants successfully genotyped for rs7072268.

French children and adults with European ancestry from families with a history of obesity were recruited at the Centre National de la Recherche Scientifique (CNRS)-UMR8090 unit (Lille, France) through an ongoing national media campaign (30). We analyzed 5,261 nondiabetic participants successfully genotyped for rs7072268.

We analyzed 7,447 unrelated French individuals with type 2 diabetes ascertained from the French type 2 diabetes family and Obesity family studies, collected by the CNRS-UMR8090 unit, from the Endocrinology-Diabetology Department of the Corbeil-Essonnes Hospital (7), and from the Diabhycar/Diab2-Néphrogène/Surdiagène study (31). We used 5,380 unrelated normoglycemic participants (age at exam ≥40 years) as control subjects (ascertained by the D.E.S.I.R. cohort; the SU.VI.MAX study, which has previously been described [32], and the French type 2 diabetes family and obesity family studies).

For each population, glycemic status was defined according to 1997 American Diabetes Association criteria (2): normal glucose was defined as FPG <6.1 mmol/l without hypoglycemic treatment, and type 2 diabetes was defined as FPG ≥7.0 mmol/l or treatment with antidiabetic agents. For the Corbeil study, overt nephropathy was defined as microalbuminuria levels ≥30 mg/24 h or ≥20 mg/l in two of three urinary takings.

Genotyping of SNP rs7072268 was performed using a TaqMan assay according to the manufacturer's instructions (no. C-30005592-10; Applied Biosystems, Foster City, CA). Allelic discrimination was performed by capillary electrophoresis analysis using an Applied Biosystems 3730xl DNA Analyser and GeneMapper 3.7 software. The genotype success rate was at least 98%, and no deviation (P > 0.05) from Hardy-Weinberg equilibrium was observed in any of the examined populations. Genotyping of MTNR1B-rs10830963, GCK-rs1799884, G6PC2-rs560887, and SLC30A8-rs13266634 in the D.E.S.I.R. study had previously been reported (10,19,33,34).

We analyzed the effect of SNP rs7072268 on quantitative traits using linear regression models under an additive model adjusted for age, sex, and BMI. To take into account familial relationships within the French obesity pedigrees, we tested the association between rs7072268 and glucose homeostasis-related traits using Gaussian models of generalized estimated equations (GEEs) performed with STATA software. The estimates of the effect of rs7072268 on quantitative traits and their standard errors for each separate population were combined in the meta-analyses using the weighted inverse normal method. The overall effect and its CI were estimated using the inverse variance method implemented in the “meta.summaries” function of the R RMETA package. The effect of rs7072268 on diabetic status was assessed using a logistic regression model adjusted for age, sex, and BMI. In the D.E.S.I.R. participants, the effect of the rs7072268 genotype on quantitative traits was assessed in nondiabetic individuals at baseline and using repeated measures at 3-, 6-, and 9-year follow-up visits. We used mixed models for analyses of repeated measures adjusted for age, sex, and BMI. Using the QUANTO software, we estimated what significant effects of rs7072268 on glucose homeostasis–related parameters we could expect in the related meta-analyses, with a detection power of 80%. Given the analyzed sample sizes, small effects of HK1 rs7072268 (estimated at β ≤ 0.1 ) on glucose homeostasis–related parameters can be detected with a power of 80%. All statistical analyses were performed with R (version 2.6.1), SPSS (version 14.0 for Windows), QUANTO (version 1.2), and STATA software (version 5.0).

Homeostasis model assessment of pancreatic β-cell function (HOMA-B) was calculated as follows: HOMA-B = (20 × fasting serum insulin)/(FPG − 3.5), where fasting serum insulin is in milliunits per liter and FPG is in millimoles per liter (35). Homeostasis model assessment of insulin resistance (HOMA-IR) was calculated as follows: HOMA-IR = (FPG × fasting serum insulin)/22.5, where fasting serum insulin is in picomoles per liter and FPG is in millimoles per liter (35).

The insulinogenic index, the insulin sensitivity index (ISI), and the disposition index (DI) were calculated from an oral glucose tolerance test (OGTT) according to the following formulas:

ISI and DI were only calculated in the French obese pedigrees as measurements of serum insulin and plasma glucose were available at 0, 30, 60, 90, and 120 min after glucose load. In the Haguenau study, measurements of serum insulin and plasma glucose were only available at 0, 30, and 120 min after glucose load.

We first genotyped SNP rs7072268 in 4,590 middle-aged nondiabetic individuals from the French D.E.S.I.R. population (mean age 47 years) and in 2,363 Swiss nondiabetic obese adults (mean age 41 years) (Table 1). After an additive genetic model adjusted for age, sex, and BMI was applied, the rs7072268–T allele showed a consistent association with increased A1C in the D.E.S.I.R. study at baseline and over the 9-year follow-up (β = 0.023%_A1C [95% CI 0.016–0.031], P = 1.76 × 10⁻³, and β = 0.022%_A1C [0.016–0.029], P = 3.93 × 10⁻⁴, respectively; Table 2) and in the Swiss obese adults sample set (β = 0.046%_A1C [0.032–0.060], P = 9.46 × 10⁻⁴; Table 2). These results were unchanged when the additive genetic model was adjusted for age and sex only (data not shown). When we also included FPG level in the linear regression model, the significance of the effect on A1C was stronger in both studies and in a meta-analysis of the D.E.S.I.R. baseline data and the Swiss obese samples (n = 6,953; β = 0.029%_A1C [0.018–0.040], combined P = 2.22 × 10⁻⁷; Table 2).

We then assessed the impact of the rs7072268–T allele on glucose homeostasis–related traits in the D.E.S.I.R. and Swiss samples. After applying an additive genetic model adjusted for age, sex, and BMI, we did not find significant associations between rs7072268 and any glucose-related traits including fasting glucose, fasting insulin, HOMA-B, and HOMA-IR (Table 3).

To further support these paradoxical findings, we tested the effect of rs7072268 on the same fasting traits in 12,003 additional nondiabetic individuals ascertained from the NFBC1986 study (age at examination 16 years), the French Haguenau cohort (mean age 22 years), and French obesity pedigrees including both children and adults (mean age 11 and 46 years, respectively) (Table 1). A1C levels were not measured in these sample sets. After applying an identically adjusted additive genetic model, we did not find significant associations with any of these traits as analyzed in each cohort or in the overall combined meta-analysis (Table 3). Furthermore, analyses of glucose and insulin levels after an oral glucose load in 1,440 individuals from Haguenau and in 1,055 children and 2,294 adults from the French obesity pedigrees did not show any significant associations (Table 4).

Since our results thus far suggested that the effect of rs7072268 on A1C was not due to differences in glycemic status, we assessed the impact of rs7072268 on RBC-related parameters available in D.E.S.I.R. and the Swiss obese adults sample set and also in 5,229 participants from the NFBC1986 study (where RBC-related traits but not A1C were measured). After an additive genetic model adjusted for age, sex, and BMI was applied, our combined analysis demonstrated an association between the rs7072268–T allele and decreased hematocrit (n = 11,492; β = −0.13%_hematocrit [95% CI −0.20 to −0.06], combined P = 2.26 × 10⁻⁴; Table 5) and decreased hemoglobin levels (β = −0.044 g/dl [−0.071 to −0.017], combined P = 1.43 × 10⁻³; Table 5). Combined case-control studies for anemia (stringently defined by hemoglobin ≤12 g/dl for women and ≤13 g/dl for men; 669 cases) from the same cohorts further supported the anemic effect of the rs7072268–T allele (odds ratio [OR] 1.13 [95% CI 1.01–1.27]; combined P = 0.032). We next studied the effects of variation at rs7072268 on mean corpuscular hemoglobin (MCH) and mean corpuscular volume (MCV) indexes: because the P values for heterogenity in effects on both traits were <0.05, our analysis was performed in each cohort in isolation. In Swiss obese adults, the rs7072268–T allele associates with both decreased MCH and MCV parameters (β = −0.21 pg/cell [95% CI −0.28 to −0.14], P = 2.16 × 10⁻³, and β = −0.56 × 10⁻¹⁵ l/cell [−0.72 to −0.38], P = 1.29 × 10⁻³, respectively; Table 5), suggesting a microspherocytocytic anemic state. In the D.E.S.I.R. participants, the RBC count also showed a negative association with the rs7072268–T allele both at baseline (β = −0.018 × 10¹²/l [95% CI −0.025 to −0.011], P = 8.01 × 10⁻³; Table 5) and over the 9-year follow-up (β = −0.020 × 10¹²/l [−0.027 to −0.014], P = 9.63 × 10⁻⁴, respectively; Table 5).

The rs7072268–T allele was also associated with decreased hemoglobin level in 1,924 French type 2 diabetic subjects from the Corbeil Hospital cohort, in whom this parameter was measured (β = −0.13 g/dl [95% CI −0.16 to −0.09], P = 7.66 × 10⁻⁴; Table 5). When the presence of overt nephropathy, the microalbuminuria level, or the albumin-to-creatinine ratio were introduced in the linear regression model, this association remained significant (P < 1.5 × 10⁻³), suggesting that the effect of HK1 on RBC is independent of diabetes-linked kidney disease. We also identified in type 2 diabetic subjects a trend for association between the rs7072268–T allele and decreased MCV (Table 5).

In a combined meta-analysis including nondiabetic and type 2 diabetic participants, the rs7072268–T allele strongly associated with decreased hemoglobin levels (n = 13,416; β = −0.054 g/dl [95% CI −0.076 to −0.031], combined P = 3.74 × 10⁻⁶; Table 5). In addition, the trend for an increased risk for clinical anemia was further supported (836 cases; OR 1.13 [95% CI 1.02–1.25]; combined P = 0.018).

We then assessed the contribution of rs7072268 to type 2 diabetes risk in 7,447 French type 2 diabetic individuals and 5,380 unrelated normoglycemic French control subjects (age at exam ≥40 years). The type 2 diabetes case-control analysis only displayed a nominal association between the rs7072268–T allele and increased risk of type 2 diabetes (OR 1.07 [95% CI 1.00–1.14], P = 0.045; Table 6). These findings were not supported by GWA studies meta-analyses carried out by the DIAGRAM+ consortium, including 8,130 type 2 diabetic and 38,987 control European participants (OR 0.98 [0.94–1.02]; P = 0.40) (M. McCarthy, unpublished data). Therefore, the weak HK1 rs7072268 effect on increased type 2 diabetes risk, found in our samples, is not supported by other European populations.

We then analyzed the contribution of four previously reported genetic determinants of A1C (MTNR1B-rs10830963 [9,34], GCK-rs1799884 [20], G6PC2-rs560887 [20], and SLC30A8-rs13266634 [20]) on A1C levels in the D.E.S.I.R. cohort. We confirmed the contribution of these SNPs to A1C levels in ∼4,500 nondiabetic individuals from the D.E.S.I.R. study at baseline—except for SLC30A8-rs13266634, which displayed only a trend for association with A1C levels (P_MTNR1B = 2.25 × 10⁻⁴, P_GCK = 1.32 × 10⁻⁴, P_G6PC2 = 2.31 × 10⁻⁶, and P_SLC30A8 = 0.063; Table 7). Analysis of HK1-rs7072263 combined with the four other SNPs demonstrated a significant additive effect on A1C levels (β_{per allele} = 0.032%, P = 1.49 × 10⁻¹⁵; Fig. 1). Individuals carrying seven or more “high-A1C” alleles (n = 415; ∼11% of the European population) showed a mean 0.17% increase in A1C compared with individuals carrying fewer than two high-A1C alleles (n = 219; Fig. 1).

We then assessed the effect of MTNR1B-rs10830963, GCK-rs1799884, G6PC2-rs560887, and SLC30A8-rs13266634 on FPG levels and RBC-related parameters including RBC count, hemoglobin, and hematocrit levels. As previously reported (9,10,19,33), the four SNPs are strongly associated with FPG levels (Table 7). SNPs GCK-rs1799884, G6PC2-rs560887, and SLC30A8-rs13266634 are not associated with RBC-related parameters (Table 7). In contrast, the MTNR1B-rs10830963–T allele associates with decreased RBC count and hemoglobin and hematocrit levels (β = −0.017 × 10¹²/l [95% CI −0.025 to −0.001], P = 0.022; β = −0.055 g/dl [−0.076 to −0.033], P = 0.011; and β = −0.19% hematocrit [−0.25 to −0.12], P = 4.13 × 10⁻³, respectively; Table 7).

Our data unambiguously demonstrate that HK1 rs7072268 strongly associates with increased A1C levels in European general populations, as reported by Pare et al. (20). In contrast, we failed to find any further association with other quantitative metabolic traits commonly used to monitor glucose control. In addition, it is unlikely that HK1 rs7072268 significantly increases risk for type 2 diabetes. Our data suggest that the effect of HK1 variation on A1C levels may be due to a molecular mechanism involving RBC function rather than related to impaired blood glucose homeostasis. In this regard, we found that the HK1 rs7072268–T allele increasing A1C is strongly associated with reduced hemoglobin and hematocrit levels (Spearman correlation between hematocrit and hemoglobin levels in nondiabetic subjects from D.E.S.I.R.: r² = 0.94; P < 0.0001). In addition, the rs7072268–T allele contributes to an increase in the risk of clinical anemia. However, this result has to be confirmed in large-scale and more powered case-control studies. In support of our findings, dea mice with an HK1 deficiency also display lower RBC count and hemoglobin and hematocrit levels (22). Indeed, these mice show severe anemia, with extensive tissue iron deposition and marked reticulocytosis, which results from significant intravascular hemolysis (22). Approximately 20 patients with nonspherocytic hemolytic anemia due to HK1 deficiency have been described thus far (21), but there is no information available about their A1C levels. SNP rs7072268 is located in the first intron of the HK1 isoform, HK1-R, specifically expressed in RBC and is in intermediate linkage disequilibrium with a common nonsynonymous coding SNP, rs1133189 (according to the HapMap CEU population: r² = 0.58). Although we have no obvious information about the truly causative common SNPs in the HK1 locus associated with anemia (that might be obtained from fine-mapping studies), we speculate they may impair HK1 expression or the maturation of this hexokinase enzymatic isoform in reticulocytes and in mature RBCs, as known in monogenic HK1 deficiency (21,23).

In RBCs, the oxygen affinity of hemoglobin is strongly regulated by 2,3-biphosphoglycerate (2,3-DPG) produced by a bypass in glycolysis (21). Increasing 2,3-DPG levels cause a decreased oxygen affinity and thus improve the transfer of oxygen to tissues and ameliorate the anemic state. HK1 deficiency contributes to decrease 2,3-DPG levels and thus annuls its beneficial effect (21). HK1 is also known to bind in mitochondria to the voltage-dependent anion channels, known as mitochondrial porins (39). Mitochondrial-associated hexokinase activity has been shown to protect cells from entering apoptosis via the blockade of the interaction of the proapoptotic BAX with the voltage-dependent anion channels (40,–42). We speculate that HK1 variation may impair the HK1 antiapoptotic effect in reticulocytes (i.e., the precursors of RBCs), as well as in kidney and brain where HK1 is expressed (21,43). It may have deleterious effects on maturation of RBCs and on erythropoiesis via decreased synthesis of kidney and brain erythropoietin (Epo).

The mechanism by which HK1-related anemia increases A1C levels is unknown. Using a conditional regression model, we failed to clearly show that the HK1 effect on A1C was affected by adjustment for the hemoglobin or hematocrit levels (supplemental Table A1, available in the online appendix, available at http://diabetes.diabetesjournals.org/cgi/content/full/db09-0652/DC1). This may suggest that the hemoglobin or hematocrit levels would explain a small variance of A1C. However, larger studies are needed for confirmation of these findings. A higher turnover of the RBC pool should diminish protein glycation as a result of the reduced hemoglobin half-life (5). Alternatively, we speculate that the enhanced accumulation of unprocessed glucose resulting from the HK1 deficiency may favor hemoglobin glycation within RBCs, which in turn may increase the RBC death rate via their impaired deformability (44). Importantly, anemia due to iron deficiency often seen in late pregnancy also causes increased A1C levels (45), and A1C levels significantly decrease after iron or vitamin B12 treatment in patients with iron or vitamin B12 deficiency anemia, respectively (46,47). Therefore, different anemia-inducing mechanisms increase A1C levels.

Other genes associated with RBC-related parameters may also interfere with the glycation of hemoglobin. In this regard, our present data suggest that genetic variation in MTNR1B (encoding melatonin receptor 2), which strongly influences both A1C and fasting glucose (9), also associates with decreased RBC count and hemoglobin and hematocrit levels. Melatonin is a neurohormone mainly involved in the regulation of circadian rhythms. Recently, Bozek et al. (48) provided evidence of a circadian oscillation of Epo gene expression in the kidney, a tissue that strongly expresses MTNR1B in rats (49 ). In contrast, three other genetic determinants of A1C (GCK, G6PC2, and SLC30A8) modulate fasting glucose but do not influence hematologic parameters measured in our cohorts. Altogether, A1C levels seem to be largely genetically determined (Fig. 1), possibly via the modulation of blood glucose or hematologic parameters.

As both the American Diabetes Association and the European Association for the Study of Diabetes have proposed to use A1C as a criterion for type 2 diabetes diagnosis (an individual with A1C <6% is considered as nondiabetic), both genetic and environmental factors (including iron and vitamin B12) interacting with RBC function and survival have to be taken into consideration to better interpret A1C levels in the general population. Furthermore, diabetes by itself is a known cause for anemia through a range of deleterious mechanisms (44), and it would be important to better determine the impact of anemia on A1C assays.

In conclusion, our study presents mechanisms that may underlie the consistent association between HK1 genetic variation and A1C but also identifies for the first time a gene contributing to a common proanemic state. At a time when the utility of GWA studies is debated for disease prediction (50), our study highlights the power of GWA to identify physiological determinants of complex conditions such as anemia having serious implications for health.

A.B. is funded by a research fellowship from the Conseil Régional du Nord Pas de Calais (France) and the CNRS. This study was supported by the French Agence Nationale de la Recherche (ANR-08-GENO-001-01) and the European Union (Integrated Project EuroDia LSHM-CT-2006-518153 in the Sixth Framework Programme [FP6] of the European Commission [to P.F.]). The Diab-2-Néphrogène/Surdiagène study received a 2003 AFD grant. The NFBC1986 is supported by the European Commission (contract no. QLG1-CT-2000-01643), Biocenter, the University of Oulu, and the Academy of Finland.

The D.E.S.I.R. study was supported by INSERM, CNAMTS, Lilly, Novartis Pharma, sanofi-aventis, the Association Diabète Risque Vasculaire, the Fédération Française de Cardiologie, La Fondation de France, ALFEDIAM, ONIVINS, Ardix Medical, Bayer Diagnostics, Becton Dickinson, Cardionics, Merck Santé, Novo Nordisk, Pierre Fabre, Roche, and Topcon. No other potential conflicts of interest relevant to this article were reported.

We are grateful to all patients and their families for participation in the genetic study. We thank Marianne Deweirder, Frédéric Allegaert, and Emmanuel Vaillant for their technical assistance and their precious management of DNA samples; Stefan Gaget and Sophie Gallina for bioinformatics support; and Sylvie Poulain, Philippe Delfosse, and Philippe Gallina for the recruitment of families with a history of obesity and type 2 diabetes. We are grateful to the DIAGRAM+ consortium for providing GWA studies data on rs7072268, and we particularly thank Mark McCarthy and Benjamin Voight. The Diab-2-Néphrogène/Surdiagène study acknowledges the participating patients and physicians and the staff of the CIC Poitiers, PHRC (Projet Hospitalier de Recherche Clinique). We thank Leena Peltonen for providing NFBC1986 DNA samples.