Journal of Molecular Medicine

, Volume 91, Issue 9, pp 1109–1115

Common variants near BDNF and SH2B1 show nominal evidence of association with snacking behavior in European populations

Authors

  • Sébastien Robiou-du-Pont
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
  • Loïc Yengo
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
  • Emmanuel Vaillant
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
  • Stéphane Lobbens
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
  • Emmanuelle Durand
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
  • Fritz Horber
    • Department of Internal MedicineLiechtensteinisches Landesspital
    • Faculty of MedicineUniversity of Bern
  • Olivier Lantieri
    • Institut inter Régional pour la Santé
  • Michel Marre
    • Endocrinology–Diabetology–NutritionBichat-Claude Bernard Hospital
    • University Denis Diderot Paris
  • Beverley Balkau
    • INSERM Centre de recherche en Epidémiologie et Santé des Populations U1018
    • University Paris
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
    • Genomic Medicine, Hammersmith HospitalImperial College London
    • Department of Genomics of Common Diseases, School of Public Health, Hammersmith HospitalImperial College London
  • David Meyre
    • CNRS-UMR8199Lille Pasteur Institute
    • Lille Nord de France University
    • Department of Clinical Epidemiology and BiostatisticsMcMaster University
Original Article

DOI: 10.1007/s00109-013-1027-z

Cite this article as:
Robiou-du-Pont, S., Yengo, L., Vaillant, E. et al. J Mol Med (2013) 91: 1109. doi:10.1007/s00109-013-1027-z

Abstract

We investigated the effect of 24 obesity-predisposing single nucleotide polymorphisms (SNPs), separately and in combination, on snacking behavior in three European populations. The 24 SNPs were genotyped in 7,502 subjects (1,868 snackers and 5,634 non-snackers). We tested the hypothesis that obesity risk variants or a genetic risk score increases snacking using a logistic regression adjusted for sex, age, and body mass index. The obesity genetic risk score was not associated with snacking (odds ratio (OR) = 1.00 [0.98–1.02], P value = 0.48). The obesity risk variants of two SNPs (rs925946 and rs7498665) close to the BDNF and SH2B1 genes showed nominal evidence of association with increased snacking (OR = 1.09 [1.01–1.17], P value = 0.0348 and OR = 1.11 [1.04–1.19], P value = 0.00703, respectively) but did not survive Bonferroni corrections for multiple testing. The associations of rs925946 and rs7498665 obesity risk variants with increased BMI (β = 0.180 [0.022–0.339], P value = 0.0258 and β = 0.166 [0.019–0.313], P value = 0.0271, respectively) were slightly attenuated after adjusting for snacking (β = 0.151 [−0.006 to 0.309], P value = 0.0591 and β = 0.152 [0.006–0.297], P value = 0.0413). Our data suggest that genetic predisposition to obesity does not significantly contribute to snacking behavior. The nominal associations of rs925946 and rs7498665 obesity risk variants near the BDNF and SH2B1 genes with increased snacking deserve further investigation.

Keywords

ObesityGenome-wide associationSingle nucleotide polymorphismSnacking behaviorBDNFSH2B1

Introduction

The prevalence of obesity has reached epidemic proportions throughout the world [1]. This recent epidemic is mainly attributed to environmental changes that promote overnutrition and sedentary lifestyle [1]. Obesity is a risk factor for type 2 diabetes, cardiovascular disease, and cancer and is associated with early mortality and shortened life expectancy [2]. Currently, obesity treatments such as lifestyle interventions, hypocaloric diets, or exercise training have shown a limited efficiency, while drug therapies provide modest weight loss and have potentially harmful side effects. Bariatric surgery is the only efficient procedure to reach significant long-term weight reduction, improvement of comorbidities, and reduced mortality rates [3].

Obesity is a highly heritable disease (heritability estimates of 40–80 %), and the recent rise of an “obesogenic” environment has enhanced the influence of adiposity-related genetic factors [4]. Loss-of-function mutations in eight genes involved in the neuronal differentiation of the paraventricular nucleus and in the leptin/melanocortin pathway have been linked with Mendelian forms of human obesity, with hyperphagia as a common feature [5]. More recently, studies on polygenic obesity have identified 54 common variants associated with body mass index (BMI) or obesity, mainly by genome-wide association studies [5]. Several of the likely causal obesity genes identified through genetic association studies (FTO, MC4R, POMC, SH2B1, BDNF, PCSK1, NPC1, NRXN3, and NEGR1) are highly expressed or known to act in the central nervous system, indicating a potential role for central regulation of food intake in polygenic obesity susceptibility, in line with monogenic forms of human obesity [6].

Consistent evidence of association of the two major contributors to polygenic obesity (SNP rs1421085/rs9939609 in FTO and SNP rs17782313 near MC4R) with food intake/food behavior-related endophenotypes has been reported in the literature [6]. Bauer et al. have reported significant associations between obesity gene variants recently identified by genome-wide association study (GWAS) (SH2B1, KCTD15, MTCH2, NEGR1, and BDNF), dietary intake, and nutrient-specific food preference [7]. More recently, McCaffery et al. studied 16 gene variants at 11 loci identified by GWAS in 2,075 participants from the Look AHEAD clinical trial and reported significant associations between obesity-predisposing variants at the FTO, BDNF, SH2B1, and TNNI3K loci and parameters related to food pattern or food consumption [8]. These observations are consistent with the fact that food intake-related parameters are heritable and are strongly correlated with adiposity [9]. Eating rate in children is a highly heritable trait (heritability, h2 = 62 %) [9] and snacking behavior has been positively associated with the risk of obesity in some but not all studies [10, 11]. However, with the notable exception of the Bauer and McCaffery studies, the possible impact of the obesity loci recently identified by GWAS on food behavior-related traits has been poorly investigated. In addition, investigating the effect of a genetic obesity predisposition score on food behavior has never been investigated. This prompted us to assess the single or combined effect of 24 obesity-predisposing variants on snacking behavior in 7,502 European individuals from three independent studies. Our prior hypothesis was that obesity risk alleles may be associated with an increased risk of snacking. For the two SNPs nominally associated with snacking behavior (rs925946 near BDNF and rs7498665 near SH2B1), we assessed whether their association with BMI was mediated by snacking behavior.

Materials and methods

Subjects

The study protocols were all approved by local ethics committees and informed consent was obtained from each subject before participating in the studies, in accordance with the Declaration of Helsinki. For children younger than 18 years, verbal consent was obtained and parents provided a written informed consent. All participants were European Caucasians. The three populations involved in this study are described in Table 1. We selected 1,080 children (818 obese and 262 nonobese) from 624 French pedigrees recruited by the “Centre National de la Recherche Scientifique” with at least one obese child. The second population included 2,145 obese Swiss subjects recruited for obesity surgery in the Lindberg Clinic from Winterthur. The third population included 4,277 French adults from the Data from Epidemiological Study on Insulin Resistance syndrome (D.E.S.I.R.) general population.
Table 1

Phenotypic characteristics of the studied populations

 

French obese children

Swiss obese

D.E.S.I.R.

n

1,080

2,145

4,277

% male

48.33 %

26.06 %

49.89 %

Mean BMI (kg/m2)

26.11

43.53

24.74

Range of BMI

11.94–54.36

30.00–90.70

15.40–53.60

Mean age (year)

11.24

41.80

46.54

Range of age

6–17

14–81

30–65

% snacking (N)

46.30 % (497)

38.23 % (815)

13.21 % (556)

Phenotyping

The 97th BMI percentile was used as the threshold to define childhood obesity, according to the recommendations of the European Childhood Obesity Group study. To calculate the BMI Z score and percentile thresholds for childhood obesity, we used the French growth charts provided by Rolland-Cachera et al. [12]. Adult obesity was defined by a BMI ≥30 kg/m2 as recommended by the International Obesity Task Force. All the carriers of the loss of function rare mutations in MC4R, PCSK1, and SIM1 were excluded from the obese populations.

Snacking behavior can be defined as being “all foods and drinks taken outside the context of the three main meals.” We previously described the questionnaires used to assess snacking behavior in the populations included in this study [13, 14]. In brief, snacking behavior was measured in the population of French obese children populations by an in-house questionnaire completed by trained physicians or dieticians. The question asked was “snacking between meals” with only two answers possible: “yes” or “no.” The Swiss subjects completed a validated auto-questionnaire about eating behavior disorder, using the “Diagnostic and Statistical Manual of Mental Disorders, 4th edition.” In this population, snacking behavior was defined using four criteria: (1) high frequency of eating occasions: five to seven per day, (2) regular eating between meals, (3) various types of food eaten (e.g., crisps, cheese, biscuits, chocolate, yogurt, fruit), and (4) eating in response to emotional cues or hunger. Only people who answered positively to the four aforementioned questions were classified as snackers. The eating behavior of individuals from the D.E.S.I.R. was measured by a questionnaire completed by a physician. The questionnaire includes the field “actual eating behavior disorders” and three questions, (1) snacking, (2) anorexia, and (3) bulimia, are asked with only two answers possible, yes or no. The total daily food intake measurement was only available in the D.E.S.I.R. population and was calculated by the NAQA equation that estimates the daily intake of calories, from a 23-item food frequency questionnaire.

Genotyping

Genomic DNA was extracted from blood samples. We selected the 24 obesity-predisposing SNPs according to the following criteria: (1) genome-wide significance level of association (P < 5 × 10−8) for BMI and/or obesity traits reported in at least one study and (2) genome-wide level of significance (P < 5 × 10−8) reported in a population of European ancestry. Original publications that refer to the selected SNPs have been cited in a previous publication [15]. The 24 SNPs are: rs2844479 (AIF1), rs6265 and rs925946 (BDNF), rs6013029 (CTNNBL1), rs7647305 (ETV5), rs7138803 (FAIM2), rs1421085 and rs6499640 (FTO), rs10938397 (GNPDA2), rs11084753 (KCTD15), rs1424233 (MAF), rs17782313 (MC4R), rs10838738 (MTCH2), rs2815752 (NEGR1), rs1805081 (NPC1), rs6232 and rs6234 (PCSK1), rs4712652 (PRL), rs10508503 (PTER), rs12145833 (SDCCAG8), rs10913469 (SEC16B), rs7498665 (SH2B1), rs6548238 (TMEM18), and rs17150703 (TNKS).

SNP genotyping was performed using SNPlex or Assay-on-Demand kits (Applied Biosystem, Foster City, CA, USA) with a mean call rate of 99.2 % (range, 93.7 to 100 %) (Supplementary Table 1). Probe details and experimental conditions are available on request. A subset of 593 subjects were genotyped twice and the mean concordance rate was 99.9 % (98.6 to 100 %) (Supplementary Table 2). All the SNPs were in the Hardy–Weinberg equilibrium in the three populations (P > 0.001, Supplementary Table 1).

Statistical analysis

All statistical analyses used the free software R 2.13. We performed a logistic regression model to study the association of SNPs/genotype score with snacking behavior. The regression models were adjusted for sex, age, and BMI. We added the pedigree number as a random effect variable in a mixed logistic regression to take into account family relatedness of French obese children. An additive model of inheritance was used in all analyses. Starting from the assumption that obesity risk alleles increase the risk of snacking behavior, we used one-sided P values. A genetic predisposition score was calculated by summing the alleles predisposing to obesity for the 24 SNPs. We used an unweighted score as Janssen et al. previously showed that weighting had no major impact on the score [16]. When we calculated the genotype score, we discarded all individuals with more than one missing genotype and imputations were performed for the remaining missing genotype. This imputation was performed for each SNP individually using the arithmetic average of the coded genotypes observed for all the individuals successfully genotyped. The Hardy–Weinberg equilibrium was tested using a chi-square test in combination with permutations and bootstrapping (“genetics” R 2.13 package). The summary statistics obtained in the three populations for the binary trait snacking behavior were meta-analyzed using the Mantel–Haenszel method with a fixed effect model. We performed 109 statistical tests in total. We first tested associations between snacking, BMI, sex, and age in the whole sample (three tests). We then assessed the associations of 24 obesity-predisposing variants and of the genetic risk score with snacking behavior in the French obese children, in the Swiss obese adults, in the D.E.S.I.R. general population, and in the pooled population (100 tests). We investigated the association of rs925946 SNP in BDNF and of rs7498665 SNP in SH2B1with BMI before and after adjustment for snacking in addition to sex, age, and study (four tests). Finally, we tested the association of rs925946 SNP in BDNF and of rs7498665 SNP in SH2B1 with snacking behavior after adjusting for total daily calorie intake in addition to sex, age, and BMI in the D.E.S.I.R. population (two tests). P values <0.05 before Bonferroni correction were considered as nominally significant. After applying the Bonferroni correction for multiple testing, P values < 4.6 × 10−4 (=0.05/109) were considered as significant.

Results

We first tested associations between snacking, BMI, sex, and age in the pooled population. BMI was positively correlated with snacking behavior (β = 0.031 95 % confidence interval (C.I.) [0.021–0.041], P value = 1.13 × 10−9, adjusted for age, sex, and study). Women nibbled significantly more than men (odds ratio (OR) = 2.04 95 % C.I. [1.78–2.34], P value = 3.64 × 10−25, adjusted for BMI, age, and study), and older people snacked less than the young (using age as a quantitative variable β = −0.020, 95 % C.I. [−0.026 to 0.013], P value = 3.73 × 10−9, adjusted for sex, BMI, and study).

We then investigated the association of the 24 obesity-predisposing SNPs with snacking behavior in the entire population. No SNP association survived the multiple testing correction, but two SNPs, rs925946 near BDNF (OR = 1.09 95 % C.I. [1.01–1.17], P value = 3.48 × 10−2) and rs7498665 near SH2B1 (OR = 1.11 95 % C.I. [1.04–1.19], P value = 7.03 × 10−3), showed nominal evidence of associations with snacking, the obesity risk allele being associated with a higher risk of snacking (Table 2). Associations performed in each cohort separately are shown in Supplementary Table 3. We did not find any association between the obesity genotype score and snacking (OR = 1.00 95 % C.I. [0.98–1.02], P value = 0.48) in the entire population (Table 2).
Table 2

Meta-analysis of the association of gene variants with snacking after adjustment for sex, age, and BMI

Gene

SNP

Risk allele

R.A.F. (%)

OR (95 % C.I.)

P value

PHeterogeneity

AIF1

rs2844479

A

60.64

0.96 (0.89–1.02)

1.42 × 10−01

0.399

BDNF

rs6265

C

78.70

1.04 (0.96–1.14)

2.09 × 10−01

0.886

BDNF

rs925946

T

27.57

1.09 (1.011.17)

3.48 × 1002

0.178

CTNNBL1

rs6013029

T

5.65

1.01 (0.87–1.17)

4.55 × 10−01

0.463

ETV5

rs7647305

C

78.76

0.93 (0.85–1.01)

8.11 × 10−02

0.660

FAIM2

rs7138803

A

35.11

0.98 (0.91–1.04)

2.83 × 10−01

0.368

FTO

rs1421085

C

41.33

0.98 (0.91–1.05)

3.00 × 10−01

0.558

FTO

rs6499640

A

62.23

1.04 (0.97–1.11)

1.90 × 10−01

0.675

GNPDA2

rs10938397

G

43.91

0.95 (0.89–1.02)

1.33 × 10−01

0.746

KCTD15

rs11084753

G

65.15

0.99 (0.92–1.06)

3.68 × 10−01

0.705

MAF

rs1424233

T

48.13

1.07 (0.99–1.15)

5.36 × 10−02

0.456

MC4R

rs17782313

C

24.01

1.05 (0.97–1.13)

1.68 × 10−01

0.783

MTCH2

rs10838738

G

33.70

1.01 (0.94–1.09)

3.80 × 10−01

0.948

NEGR1

rs2815752

A

63.63

0.97 (0.90–1.05)

2.53 × 10−01

0.577

NPC1

rs1805081

T

59.91

0.97 (0.90–1.04)

2.28 × 10−01

0.613

PCSK1

rs6232

C

4.04

1.04 (0.88–1.22)

3.61 × 10−01

0.152

PCSK1

rs6234

C

25.28

0.98 (0.90–1.06)

3.40 × 10−01

0.456

PRL

rs4712652

A

54.34

1.00 (0.93–1.07)

4.79 × 10−01

0.039

PTER

rs10508503

C

90.32

0.93 (0.83–1.05)

1.57 × 10−01

0.408

SDCCAG8

rs12145833

T

85.50

0.94 (0.86–1.03)

1.43 × 10−01

0.872

SEC16B

rs10913469

C

16.81

0.99 (0.90–1.08)

4.19 × 10−01

0.248

SH2B1

rs7498665

G

34.45

1.11 (1.041.19)

7.03 × 1003

0.440

TMEM18

rs6548238

C

82.39

0.99 (0.90–1.09)

4.26 × 10−01

0.026

TNKS

rs17150703

A

10.62

1.02 (0.90–1.13)

4.14 × 10−01

0.075

Genotype score

   

1.00 (0.98–1.02)

4.81 × 10−01

0.996

Nominally significant P values (P < 0.05) are presented in bold

R.A.F. risk allele frequency in the general D.E.S.I.R. population, OR odds ratio, C.I. confidence interval, P value one-sided P value after adjustment for sex, age, and BMI

To investigate if the BMI increasing effects of rs925946 SNP in BDNF and of rs7498665 SNP in SH2B1 were mediated by a pattern of snacking behavior, we tested their association with BMI before and after adjustment for snacking in addition to sex, age, and study. Before adjustment for snacking, rs925946 in BDNF and rs7498665 in SH2B1 showed a nominal evidence of association with BMI (β = 0.180 95 % C.I. [0.022–0.339], P value = 2.58 × 10−2 and β = 0.166 95 % C.I. [0.019–0.313], P value = 2.71 × 10−2, respectively). Further adjustment for snacking slightly attenuated the association of rs925946 in BDNF and rs7498665 in SH2B1 with BMI (β = 0.151 95 % C.I. [−0.006 to 0.309], P value = 5.91 × 10−2 and β = 0.152 95 % C.I. [0.006–0.297], P value = 4.13 × 10−2, respectively).

To assess whether the association of rs925946 (BDNF) and rs7498665 (SH2B1) SNPs with snacking behavior was independent of total daily energy intake, we tested their association with snacking behavior before and after adjusting for total daily calorie intake in addition to sex, age, and BMI in the D.E.S.I.R. population. Before adjustment for total daily energy intake, rs925946 in BDNF (OR = 1.21 95 % C.I. [1.07–1.36], P value = 5.03 × 10−3) and rs7498665 in SH2B1 (OR = 1.11 95 % C.I. [0.99–1.24], P value = 6.97 × 10−2) showed, respectively, nominal and borderline nominal evidence of association with snacking behavior in the D.E.S.I.R. study (Supplementary Table 3). Results were not modified by further adjustment for total daily energy intake (OR = 1.18 95 % C.I. [1.05–1.33], P value = 1.19 × 10−2 and OR = 1.11 95 % C.I. [0.99–1.25], P value = 6.27 × 10−2, respectively).

Discussion

In this study, we investigated in a population of 7,502 Europeans the association between snacking and 24 polymorphisms previously identified by GWAS to be associated with obesity-related traits. This is the first time, to our knowledge, that a study has investigated the impact of genetic predisposition to obesity on food behavior, using a multiple SNP predisposition score. The main result of this study is the lack of association between the obesity genetic risk score and snacking behavior. Snacking is positively associated with obesity at least in some studies [10, 11] and in line with this observation has been positively associated with BMI in our population. However, our candidate approach does not support the view that snacking behavior mediates genetic predisposition to obesity. We observed nominal associations for rs925946 (BDNF) and rs7498665 (SH2B1) SNPs with snacking. The fact that the associations of rs925946 (BDNF) and rs7498665 (SH2B1) SNPs with snacking are nominal and do not survive corrections for multiple testing may be related to the suboptimal power of our study at a Bonferroni-adjusted stringent level of significance (Supplementary Fig. 1). Although these associations require more investigations, some additional arguments make the association of these SNPs with nibbling plausible. In rodents, both the nutritional state and type of diet regulate BDNF and SH2B1 expression levels in the brain [1720]. Inactivation of BDNF or SH2B1 in mouse models leads to hyperphagia and obesity [21, 22]. Central administration of BDNF or neuron-specific SH2B1 overexpression suppresses appetite, induces weight loss, and protects against high-fat diet obesity in rodents [2325]. Deletions or de novo inversion at the BDNF locus and deletions in the 16p11.2 chromosomal area including the SH2B1 gene or rare mutations in the SH2B1 gene are associated with Mendelian forms of hyperphagic obesity in humans [2629]. Furthermore, common variants in/near SH2B1 and BDNF have been associated with polygenic obesity [30] and also with the modulation of food intake and feeding behavior. Common SNPs in or near BDNF have been associated with nutrient intake [7], food consumption (i.e., meat, eggs, dairy product) [8], anorexia, bulimia, and binge eating disorders [31, 32]. The SH2B1 rs7498665 variant has been associated with total, saturated, and monounsaturated fat and with carbohydrate intake [7] and with dairy product consumption [8].

The lack of association between snacking and rs1421085 and rs6499640 (FTO), rs17782313 (MC4R) and rs6232, and rs6234 (PCSK1) was unexpected, as experiments in rodents and humans have convincingly linked these three genes with the central control of energy intake [5]. One possible explanation is that these genes may not be associated with snacking but with other dietary phenotypes such as daily energy intake, satiety, or food preferences [7]. Another possibility is that the power of the present study was limited for the detection of subtle genetic effects. We have 80 % power to detect an OR of 1.15 in a large range of risk allele frequencies (0.1 < RAFs < 0.9) and an OR of 1.10 for RAFs between 0.3 and 0.7 at a P value level of 0.05 (Supplementary Fig. 1), suggesting that the weaker effects (OR < 1.10) may have been missed in our study. The lack of association may also be due to heterogeneity in the definition and in the collection of the snacking phenotype or to heterogeneity in the ascertainment of the studied populations, representing three important limitations in our study [33]. However, we were unable to find high levels of between-study genetic heterogeneity in our populations (Table 2). The last explanation is that the obesity-predisposing SNPs are mostly not associated with traits related to food intake or feeding behaviors, and that genetic variants contributing to snacking are in great part independent from those influencing BMI and obesity risk.

Our data do not favor the view that the association of genetic variants for obesity is driven by snacking behavior (gene–snacking–BMI triangular model). First, we did not find any association for 22 of the 24 gene variants conclusively associated with obesity-related traits. Second, the nominal associations of rs925946 near BDNF and rs7498665 near SH2B1 with snacking behavior did not survive the corrections for multiple testing. Third, we observed that the association between rs925946 (BDNF), rs7498665 (SH2B1), and BMI was only slightly attenuated after adjusting the linear regression model for snacking, as a binary variable. This suggests that the association of the two SNPs with adiposity may only be mediated in part by snacking behavior. If the nominal associations of rs925946 near BDNF and rs7498665 near SH2B1 with snacking behavior are confirmed in independent studies, our data argue at best for a selective association of obesity SNPs with snacking, a result in line with the post-GWAS by Bauer et al. and McCaffery et al. for additional traits related to food intake/food behavior patterns [7, 8]. These data support the view that obesity-predisposing variants may increase adiposity by physiological mechanisms other than food intake or food behavior patterns. This hypothesis is strengthened by two observations: (1) recent GWAS for obesity-related traits has identified candidate genes involved in adipocyte biology [34] or vulnerability to bacterial infection [35], suggesting a complex etiology for obesity, and (2) a large fraction of common variants identified by GWAS are located in or close to loci of unknown function that remains to be discovered.

In conclusion, our study showed nominal associations of rs925946 (BDNF) and rs7498665 (SH2B1) with snacking behavior that deserves further investigation, but there was no association between the genetic obesity predisposition score and snacking. Our data support the view that genetic predisposition to obesity does not play a major role in snacking behavior. Hypothesis-free approaches such as GWAS may help to identify common gene variants contributing to snacking behavior in the next future.

Acknowledgments

We thank all the participants in the studies. This work was supported by the “Agence National de la Recherche” and by the “EU-funded EUROCHIP FP7 consortium.” D.M. is funded by a Tier 2 Canada Research Chair.

Conflict of interest

The authors declare no conflict of interest.

Supplementary material

109_2013_1027_MOESM1_ESM.pdf (262 kb)
ESM 1(PDF 261 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2013