The effect of GWAS identified BMI loci on changes in body weight among middle-aged danes during a five-year period

Authors

  • C. H. Sandholt,

    Corresponding author
    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
    • Funding agencies: This project was funded by the Lundbeck Foundation and produced by The Lundbeck Foundation Centre for Applied Medical Genomics in Personalised Disease Prediction, Prevention and Care (LuCamp, www.lucamp.org). The Novo Nordisk Foundation Center for Basic Metabolic Research is an independent Research Center at the University of Copenhagen partially funded by an unrestricted donation from the Novo Nordisk Foundation (www.metabol.ku.dk). Further funding came from the Danish Council for Independent Research (Medical Sciences). The Inter99 was initiated by Torben Jorgensen (PI), Knut Borch-Johnsen (co-PI), Hans Ibsen and Troels F. Thomsen. The steering committee comprises the former two and Charlotta Pisinger. The study was financially supported by research grants from the Danish Research Council, the Danish Centre for Health Technology Assessment, Novo Nordisk Inc., Research Foundation of Copenhagen County, Ministry of Internal Affairs and Health, the Danish Heart Foundation, the Danish Pharmaceutical Association, the Augustinus Foundation, the Ib Henriksen Foundation, the Becket Foundation, and the Danish Diabetes Association.

  • K. H. Allin,

    Corresponding author
    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
    • Funding agencies: This project was funded by the Lundbeck Foundation and produced by The Lundbeck Foundation Centre for Applied Medical Genomics in Personalised Disease Prediction, Prevention and Care (LuCamp, www.lucamp.org). The Novo Nordisk Foundation Center for Basic Metabolic Research is an independent Research Center at the University of Copenhagen partially funded by an unrestricted donation from the Novo Nordisk Foundation (www.metabol.ku.dk). Further funding came from the Danish Council for Independent Research (Medical Sciences). The Inter99 was initiated by Torben Jorgensen (PI), Knut Borch-Johnsen (co-PI), Hans Ibsen and Troels F. Thomsen. The steering committee comprises the former two and Charlotta Pisinger. The study was financially supported by research grants from the Danish Research Council, the Danish Centre for Health Technology Assessment, Novo Nordisk Inc., Research Foundation of Copenhagen County, Ministry of Internal Affairs and Health, the Danish Heart Foundation, the Danish Pharmaceutical Association, the Augustinus Foundation, the Ib Henriksen Foundation, the Becket Foundation, and the Danish Diabetes Association.

  • U. Toft,

    1. Research Centre for Prevention and Health, Glostrup University Hospital, Glostrup, Denmark
    Search for more papers by this author
  • A. Borglykke,

    1. Research Centre for Prevention and Health, Glostrup University Hospital, Glostrup, Denmark
    Search for more papers by this author
  • R. Ribel-Madsen,

    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
  • T. Sparso,

    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
  • J. M. Justesen,

    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
  • M. N. Harder,

    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    Search for more papers by this author
  • T. Jørgensen,

    1. Research Centre for Prevention and Health, Glostrup University Hospital, Glostrup, Denmark
    2. Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    3. Faculty of Medicine, University of Aalborg, Aalborg, Denmark
    Search for more papers by this author
  • T. Hansen,

    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    2. Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    3. Faculty of Health Sciences, University of Southern Denmark, Odense, Denmark
    Search for more papers by this author
  • O. Pedersen

    1. The Novo Nordisk Foundation Center for Basic Metabolic Research, Section of Metabolic Genetics, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    2. Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark
    3. Faculty of Health Sciences, University of Aarhus, Aarhus, Denmark
    Search for more papers by this author

  • Disclosure: The authors declare that there are no conflicts of interest associated with this manuscript.

  • Author contributions: CS and KA were responsible for study design and data analysis. RR, TS, JJ, and MH were responsible for genotype data generation. CS, KA, UT, AB, and TJ were responsible for data interpretation. TJ was PI initiating the study cohort. TH and OP were PIs initiating the genotyping project. CS and KA were responsible for writing the manuscript. All authors were involved in reviewing the manuscript and had final approval of the submitted and published versions.

  • C. H. Sandholt and K. H. Allin contributed equally to this article.

Abstract

Objective

Genome-wide association studies have identified genetic variants associating with BMI, however, it is un-clarified whether the same variants also influence body weight fluctuations.

Methods

Among 3,982 adult individuals that attended both a baseline and a five-year follow-up examination in the Danish Inter99 intervention study, a genetic risk score (GRS) was constructed based on 30 BMI variants to address whether it is associated with body weight changes. Moreover, it was examined whether the effect of lifestyle changes was modulated by the GRS.

Results

The GRS associated strongly with baseline body weight, with a per risk allele increase of 0.45 (0.33-0.58) kg (P = 2.7 × 10−12), corresponding to a body weight difference of 3.41 (2.21-4.60) kg comparing the highest (≥ 30 risk alleles) and lowest (≤ 26 risk alleles) risk allele tertile. No association was observed with changes in body weight during the five years. Changes in lifestyle, including physical activity, diet and smoking habits associated strongly with body weight changes, however, no interactions with the GRS was observed.

Conclusion

The GRS associated with body weight cross-sectionally, but not with changes over a five-year period. Body weight changes were influenced by lifestyle changes, however, independently of the GRS.

Introduction

During the last decade a dramatic increase in the prevalence of obesity has occurred worldwide. The number of obese individuals was 500 million globally in 2008 [1] and this number is projected to rise substantially towards 2030 [2]. A major part of this pandemic can be attributed to environmental changes, where high intake of energy dense food and sedentary lifestyles are becoming more widespread; however, some individuals seem more prone to gain weight indicating a substantial genetic component as well. Body mass index (BMI) is a widely used surrogate measure of obesity, and its heritability estimated from twin and family studies ranges from 0.47 to 0.90 and 0.24 to 0.81, respectively. Heritability estimates are generally highest during childhood and decrease with age [3]. Genome-wide association studies (GWAS) have within the last years been successful in identifying loci harboring single nucleotide polymorphisms (SNPs) associated with BMI in adult populations [4-8]. Large meta-analyses performed by the GIANT (Genetic Investigation of ANthropometric Traits) consortium, comprising up to 249,796 individuals in total, have confirmed SNPs in 32 loci associating with BMI at genome-wide significance level. Although, the effect sizes of individual risk alleles are modest, ranging from 0.06 to 0.39 kg/m2, carrying multiple risk alleles would expectedly have a substantial influence on BMI. However, it is not clarified whether cumulating the same variants that predispose to increased levels of BMI cross-sectionally also influence body weight fluctuations over time.

Hence, in the prospective population-based Inter99 intervention study, we examined whether changes in body weight over five years can be attributed to a genetic risk score (GRS) based on BMI loci identified through the GWAS approach. Moreover, we examined whether the effect of lifestyle changes on body weight fluctuations could be modulated by the GRS.

Methods

Study population

The present study is based on data from the Danish population-based Inter99 study (ClinicalTrials.gov ID-no: NCT00289237), which is a non-pharmacological intervention study for ischemic heart disease. The primary aim of the study was to evaluate the effect of a multifactorial lifestyle intervention approach on the prevention of ischemic heart disease [9]. Secondary objectives were to evaluate the effect on other lifestyle-related diseases such as diabetes [10]. A random sample of 13,016 individuals living in Copenhagen County from seven different age groups (30–60 years, grouped with five-year intervals) was drawn from the Civil Registration System and further prerandomized into high-intensity (90%) and low-intensity (10%) intervention groups. A total of 6,784 (52%) attended the baseline health examination, and all were included in the intervention program (6,091 in the high-intensity and 693 in the low-intensity intervention group) and followed for five years. All participants received individual lifestyle counselling at the baseline examination, focused on habits of smoking, physical activity, dietary intake, and use of alcohol. The risk of ischemic heart disease within a ten year period was estimated applying the Copenhagen Risk Score [11]. All high-risk participants (both in the high-intensity and low-intensity intervention group) were re-invited for repeated individualized lifestyle counselling and health examination after one and three years. High-risk individuals in the high-intensity intervention group were offered additional group-based lifestyle counselling focussing on either diet and exercise or smoking cessation. The groups consisted of 15–20 participants who met up to six times for two hours during the six months following health examinations at baseline and after one and three years. All baseline participants were re-invited to a health examination after five years and 66% (N = 4,511) returned. The group-based lifestyle counselling intervention and its effect has been described in detail previously, but conclusively no significant effect was demonstrated on five-year changes in BMI, smoking, physical activity, dietary habits, or alcohol consumption [12]. Written informed consent was obtained from all participants and the study was approved by the Scientific Ethics Committee of the Capital Region of Denmark (KA 98 155).

Anthropometric traits and lifestyle measures

At the health examinations at baseline and at the five-year follow-up visit weight (kg) was measured in light indoor clothes and without shoes. Waist circumference (to the nearest cm) was measured in standing position midway between the iliac crest and the lower costal margin. Height was measured at the baseline examination but not at the follow-up visit. It was measured without shoes and to the nearest 0.5 cm. BMI was calculated as kg/m2, and BMI standard deviation score (BMI-SDS) as (individual BMI – mean population BMI)/SD of BMI (4.6 kg/m2). All lifestyle factors were estimated from self-reported questionnaire data. A three-point dietary score was developed based on a food frequency questionnaire [13] and the method was validated using 28-day diet history and biomarker analysis (N = 264) [14]. In short, the dietary score was based on questions regarding the intake of fruits, raw and boiled vegetables, vegetarian dishes, fish, and fat (both spread on bread and for preparation) to get a rough index of the overall quality of dietary habits, which were divided into three categories: 1) unhealthy, 2) moderate, and 3) healthy. Alcohol intake was estimated as the total units of alcohol per week. The level of physical activity was estimated summing the time spent actively commuting (min/week) and time spent on leisure time physical activity (min/week). Subsequently, four categories were created: 1) 0–113 min/week, 2) 143–225 min/week, 3) 255–340 min/week, and 4) 450–720 min/week [15]. Smoking habits were divided into four classes: 1) never smoked, 2) former smoker, 3) occasional smoker, and 4) daily smoker [16]. Education was categorized into four categories: 1) basic education (up to high school), 2) low education (<2 years of vocational training), 3) medium education (2–4 years of vocational training or equivalent), and high education (>4 years or an academic degree). Questionnaires were filled out both at the baseline and the follow-up examination allowing for changes in lifestyle factors to be estimated. However, intake of sugar-sweetened beverages was only available from the follow-up questionnaire.

Genotyping and SNP selection

6,377 participants from the Inter99 cohort were genotyped by the Metabochip [17] on an Illumina HiScan (Illumina, San Diego, CA). Genotypes were called using the Genotyping module (version 1.9.4) of GenomeStudio software (version 2011.1, Illumina) and custom cluster data generated from 5,865 Danish DNA samples analyzed on the same HiScan. 250 individuals were excluded during quality control removing closely related individuals (N = 119), individuals with an extreme inbreeding coefficient (N = 25), individuals with a call rate <90% (N = 30), individuals with mislabelled sex (N = 11), and individuals with a high discordance rate to previously genotyped SNPs (N = 65), leaving 6,347 individuals who passed all quality control criteria. The average call rate for all SNPs on the Metabochip was 99.0%.

The SNPs for the present study were selected based on meta-analyses of 46 GWAS (N = 123,865) and subsequent replication in 34 studies (N = 125,931) that have confirmed 32 SNPs associating with BMI at genome-wide significance level (P < 5 × 10−8) [8]. All genotypes were retrieved from Metabochip data, with 26 of the SNPs being present, four (NRXN3 rs10150332, RBJ rs713586, PTBP2 rs1555543, ZNF608 rs4836133) being captured by perfect proxies (r2 = 1), and two (LRP1B rs2890652 and MTIF3 rs4771122) not being represented on the Metabochip, and hence, a total of 30 SNPs are included in the study (Supporting Information Table 1). Proxy search was performed based on 1000 Genome Pilot 1 data for linkage disequilibrium estimation using SNP annotation proxy search (SNAP) (http://www.broadinstitute.org/mpg/snap/).

Table 1. Study population characteristics at baseline and at five-years follow-up
 BaselineFollow-upChange from baseline to follow-upP for change from baseline to follow-up
  1. Data in Table 1 are balanced meaning that only individuals who had the given measurement at the baseline examination and the follow-up examination are included.

  2. Quantitative traits are given as mean (SD) except for alcohol consumption which is given as median (IQR) because of a skewed distribution. Mean changes from baseline to follow-up are reported as mean (95% CI). Differences are tested applying a paired t-test. Normal weight was defined as BMI < 25 kg/m2, overweight as 25 ≤ BMI < 30 kg/m2 and obesity as BMI ≥ 30 kg/m2.

Participants, no. (%)3,982 (100)3,982 (100)  
Women, no. (%)1,992 (50)1,992 (50)  
Age, years46.7 (7.7)52.1 (7.7)5.38 (5.37-5.39) 
Weight, kg77.9 (15.7)79.1 (16.1)1.18 (1.01-1.34)3.2 × 10−42
Waist circumference, cm86.2 (13.0)89.3 (13.2)3.02 (2.83-3.22)6.8 × 10−179
BMI, kg/m226.1 (4.4)26.5 (4.5)0.39 (0.34-0.45)2.8 × 10−41
BMI categories, no. (%)    
Normal weight1,795 (45)1,633 (41)−162 (−9) 
Overweight1,584 (40)1,631 (41)47 (3)1.6 × 10−21
Obese603 (15)718 (18)115 (20) 
Alcohol consumption, units per week8.0 (3.0-15.0)7.5 (3.5-14.0)−0.61 (−0.90- −0.31)4.8 × 10−5
Physical activity, no. (%)     
0-113 min/week404 (11)391 (11)Less active919 (25) 
143-225 min/week833 (23)833 (23)No change1,859 (51)0.66
255-340 min/week1,942 (53)2,005 (55)More active900 (24) 
450-720 min/week499 (14)449 (12)   
Dietary score, no. (%)     
Unhealthy diet538 (14)341 (9)Unhealthier diet404 (11) 
Moderate healthy diet2,692 (71)2,480 (65)No change2,426 (64)1.2 × 10−57
Healthy diet588 (15)997 (26)Healthier diet988 (26) 
Smoking, no. (%)     
Daily1,123 (28)840 (21)More intensive smoking127 (3) 
Occasionally154 (4)140 (4)No change3,404 (86)2.5 × 10−44
Previous1,095 (28)1,409 (36)Less intensive smoking or cessation422 (11) 
Never1,581 (40)1,564 (40)   

The SNP quality of the 30 selected SNPs was estimated based on genotype call rate (> 95%), Hardy-Weinberg equilibrium (P > 0.002) or cross-hybridization with the X-chromosome, and all 30 SNPs passed these filters (HWE; min. P = 0.04, max. P = 1.00 and call rate; min. 99.1%, max. 99.9%).

Genetic risk score

Genotypes were coded according to the number of BMI increasing alleles, for 18 SNPs the minor allele and for 12 SNPs the major allele were considered risk increasing, with risk allele frequencies ranging from 4% to 87%. An un-weighted and a weighted GRS were calculated. The un-weighted GRS summarizes the number of BMI increasing alleles for each individual. For the weighted GRS the number of BMI increasing alleles was multiplied with the reported per allele effect size for each SNP [8]. To be included in the study the participant should be Dane by self report, genotype information should be retrieved for all 30 SNPs, and body weight should be available both at baseline and follow-up.

Statistical analyses

Data were analyzed using the STATA statistical software (version 12.1; StataCorp, College Station, TX, USA) and RGui version 2.13.1 (http://www.r-project.org/). A two-sided P-value less than 0.05 was considered statistically significant. T-tests were used to test for differences in anthropometric traits at baseline and at follow-up. Differences between attendees and nonattendees were tested using t-tests for normally distributed quantitative traits, Kruskal-Wallis rank tests for non-normally distributed quantitative traits and chi square tests for categorical traits. Linear regression models were used to test for associations between the GRS and anthropometric traits at baseline as well as changes in these traits during the five-year follow-up period. All analyses were made on a balanced dataset including only individuals who attended both the baseline and follow-up examinations. Analyses were performed assuming an additive genetic model (Figure 1A and B) and the GRS was entered into the models both as a continuous variable and as tertiles. In addition to the trait of interest and the GRS, models included age and sex, and when changes over time were analyzed we also included baseline levels of the trait of interest. The GRS was also tested in a fully adjusted model including the level of physical activity, diet intake, alcohol consumption, smoking habits, and educational level. When changes in body weight during the five-year period were analyzed, changes in lifestyle factors and baseline educational level were included. As the intensity of the intervention has previously been reported to be without effect on five-year changes in BMI [12], this was not included in the fully adjusted model. Interactions between the GRS and changes in lifestyle factors were tested by introducing a two-factor interaction term into the regression models. The explained variance (R2) was estimated using ordinary least squares regression.

Figure 1.

Baseline weight (A) and five-year weight changes (B) as a function of number of risk alleles. Black dots indicate mean body weight or body weight change and error bars indicate standard errors. The solid line indicates the regression line fitted to the data.

Results

A total of 5,924 individuals had information on GRS and body weight at baseline and of these 3,982 attended the re-examination. Generally, individuals that did not attend the follow-up examination had a higher body weight, a larger waist circumference, were less physically active, consumed an unhealthier diet and smoked more intensively at baseline compared to attendees; however, no difference in GRS was observed (Supporting Information Table 1). During the five-year period mean body weight increased with 1.18 (95% CI; 1.01-1.34) kg (P = 3.2 × 10−42), and the mean waist circumference increased accordingly with 3.02 (2.83-3.22) cm (P = 6.8 × 10−179) (Table 1). Collectively, the level of physical activity was unchanged (P = 0.66), whereas, the attendees became generally healthier with respect to dietary habits, smoking, and alcohol consumption, (P = 1.2 × 10−57, P = 2.5 × 10−44, and P = 4.8 × 10−5, respectively) (Table 1).

GRS and changes in anthropometrics

The GRS was constructed using SNPs in BMI loci that all previously have been investigated cross-sectionally in the Inter99 study, however, with 11 being represented by other SNPs (r2 > 0.7) than in the present study (Supporting Information Table 2) [8], [18-22]. Of the 30 loci in the present study, 11 associated significantly with body weight, with effect sizes ranging from 0.70 (0.09-1.30) to 1.71 (0.28-3.14) kg (Supporting Information Figure 1A). Among the 3,982 follow-up attendees the median number of BMI increasing alleles was 28 ranging from 16 to 41 (Figure 1A).

Table 2. Association between tertiles of genetic risk score and anthropometric traits and lifestyle factors
 Genetic risk score 
1st tertile2nd tertile3rd tertileP
  1. Data in Table 2 are balanced meaning that only individuals who had the given measurement at the baseline examination and the follow-up examination are included.

  2. Quantitative traits are given as mean (SD) except for risk alleles which is given as median (range) and alcohol consumption which is given as median (IQR) because of a skewed distribution.

Genetic risk score25 (16-26)28 (27-29)31 (30-41) 
Participants, no. (%)1,442 (36)1,380 (35)1,160 (29) 
Women, no. (%)717 (50)687 (50)588 (51)0.87
Age, years46.7 (7.7)46.8 (7.7)46.6 (7.7)0.76
Baseline    
Body weight, kg76.3 (14.6)78.1 (16.0)79.7 (16.6)3.6 × 10−11
BMI, kg/m225.6 (4.0)26.1 (4.4)26.7 (4.7)1.3 × 10−11
BMI-SDS    
Waist circumference, cm85.1 (12.2)86.2 (13.4)87.7 (13.4)3.6 × 10−10
Alcohol consumption, units per week8.0 (3.0-15.0)7.0 (3.0-15.0)8.0 (3.0-14.8)0.13
Physical activity, no. (%)   0.36
0-113 min/week141 (11)144 (11)119 (11) 
143-225 min/week320 (24)284 (22)229 (21) 
255-340 min/week699 (53)659 (52)584 (54) 
450-720 min/week161 (12)188 (15)150 (14) 
Dietary score, no. (%)   0.33
Unhealthy diet207 (15)188 (14)143 (13) 
Moderate healthy diet947 (69)935 (71)810 (73) 
Healthy diet224 (16)201 (15)163 (15) 
Smoking, no. (%)   0.90
Daily404 (28)387 (28)332 (29) 
Occasionally54 (4)49 (4)51 (4) 
Previous406 (29)377 (27)312 (27) 
Never560 (39)560 (41)461 (40) 
Change from baseline to follow-up    
Change in body weight, kg1.3 (5.1)1.1 (5.4)1.1 (5.7)0.81
Change in waist circumference, cm3.4 (6.3)3.0 (6.4)2.7 (6.3)0.14
Change in alcohol consumption, units per week−0.5 (8.5)−0.8 (9.6)−0.6 (8.0)0.15
Change in physical activity, no. (%)   0.89
Less active333 (25)314 (25)272 (25) 
No change659 (50)643 (50)557 (51) 
More active329 (25)318 (25)253 (23) 
Change in dietary score, no. (%)   0.99
Unhealthier diet147 (11)140 (11)117 (10) 
No change882 (64)839 (63)705 (63) 
Healthier diet349 (25)345 (26)294 (26) 
Change in smoking, no. (%)   0.06
Increased smoking40 (3)36 (3)51 (4) 
No change1,224 (86)1,201 (87)979 (85) 
Decreased smoking160 (11)136 (10)126 (11) 

Body weight increased with increasing number of risk alleles (Figure 1A) and the un-weighted GRS associated strongly with baseline body weight, with a per allele increase of 0.45 (0.33-0.58) kg (P = 2.7 × 10−12) (Table 3), corresponding to a body weight difference of 3.41 (2.21-4.60) kg comparing the highest (≥ 30 risk alleles) with the lowest (≤ 26 risk alleles) tertile of BMI increasing alleles (Table 2). A more extreme comparison of individuals with 19 or less risk alleles (N = 25) versus 36 or more risk alleles (N = 34) resulted in a body weight difference of 10.76 kg. An association corresponding to the association with body weight was observed between the un-weighted GRS and baseline BMI-SDS with a per allele effect of 0.03 (0.02-0.04) (P = 8.5 × 10−12) (data not shown). In contrast, the GRS did not associate with changes in body weight during the five-year follow-up period (per allele change 0.02 (−0.03-0.07) kg, P = 0.49) (Figure 1B, Table 3). This observation was reflected in only two of the 30 SNPs associating with changes in body weight during the five-year follow-up period and they both associated with a decrease in body weight (Supporting Information Figure 1B). In line with these results, the explained proportion of the variance (R2) was 3.11% for baseline body weight versus 1.40% for changes in body weight. Neither when stratifying the individuals into a weight loss nor weight gain group did the GRS have an effect on the body weight change (data not shown).

Table 3. Association between differentiated genetic risk scores and baseline body weight and body weight changes
 BaselineFollow-up
 Number of variants in the genetic risk scorePer allele effect size in body weight, kg (95% CI)PPer allele effect size in body weight change, kg (95% CI)P
  1. Data in Table 3 are balanced meaning that only individuals who had the given measurement at the baseline examination and the follow-up examination are included.

  2. a

    Baseline per allele effects were adjusted for the level of physical activity, diet, smoking habits, alcohol consumption and educational level. Per allele effects in body weight changes were adjusted for changes in the level of physical activity, diet, smoking habits, alcohol consumption and educational level at baseline.

Complete genetic risk score300.45 (0.33-0.58)2.7 × 10−120.02 (−0.03 to 0.07)0.49
Complete genetic risk score, fully adjusteda300.49 (0.35-0.62)2.3 × 10−120.01 (−0.04 to 0.07)0.61
Genetic risk score without FTO290.44 (0.31-0.57)4.3 × 10−110.02 (−0.03 to 0.07)0.52
Genetic risk score based on SNPs identified prior to GIANT meta-analysis140.62 (0.44-0.80)1.1 × 10−11−0.006 (−0.08 to 0.06)0.86
Genetic risk score based on SNPs identified in GIANT meta-analysis160.27 (0.10-0.45)0.0030.04 (−0.03 to 0.11)0.26
Weighted complete genetic risk score303.23 (2.42-4.04)6.2 × 10−150.06 (−0.25 to 0.38)0.70
Weighted complete genetic risk score, fully adjusteda303.45 (2.58- 4.31)6.8 × 10−150.11 (−0.23 to 0.45)0.53
Weighted genetic risk score without FTO293.69 (2.74-4.64)3.4 × 10−140.05 (−0.32 to 0.42)0.79
Weighted genetic risk score based on SNPs identified prior to GIANT meta-analysis143.40 (2.48-4.32)5.3 × 10−130.02 (−0.34 to 0.38)0.93
Weighted genetic risk score based on SNPs identified in GIANT meta-analysis162.50 (0.84-4.17)0.0030.21 (−0.43 to 0.85)0.52

To eliminate the possibility that the association between the GRS and baseline body weight was driven mainly by the strongest locus, we excluded FTO rs1558902 from the GRS, and results remained similar showing a per allele increase in body weight of 0.44 (0.31-0.57) kg (P = 4.3 × 10−11) (Table 3). When we included only the 14 SNPs identified prior to the GIANT meta-analysis [8], the per allele increase in body weight was 0.62 (0.44-0.80) kg (P = 1.1 × 10−11), whereas, a GRS based only on the 16 SNPs discovered in the GIANT meta-analysis showed a per allele effect 0.27 (0.10-0.45) kg (P = 0.003). Based on the estimated effect sizes and the number of SNPs included in the GRSs a theoretical difference in body weight between individuals carrying no risk alleles and individuals carrying the maximum number of risk alleles could be calculated. This would correspond to ∼27 kg for the complete GRS, ∼26 kg for the GRS without FTO, ∼17 kg for the GRS based on SNPs identified prior to the GIANT meta-analysis and ∼9 kg for the GRS based on SNPs identified in the GIANT meta-analysis.

None of the differentiated GRSs were associated with changes in body weight during the five-year follow-up period. Weighing each of the tested GRSs by incorporating the reported SNP effect sizes [8] showed similar association patterns as the un-weighted GRSs (Table 3).

The GRS also associated strongly with waist circumference at baseline (P = 1.4 × 10−10) but not with changes during the five-year follow-up period (P = 0.70). Sensitivity analyses, weighing as well as excluding and including SNPs, showed similar association pattern as body weight when analyzing waist circumference (data not shown).

GRS × lifestyle interactions

Increased level of physical activity (P = 5.7 × 10−11), a healthier diet (P = 2.9 × 10−4), and less alcohol consumption (P = 0.01) associated with decreased body weight gain (Figure 2). Smoking cessation was associated with substantial body weight gain, whereas individuals who became more intensive smokers decreased their body weight (P = 7.4 × 10−39). These associations remained similar after multifactor-adjustment for changes in all confounding lifestyle factors and baseline educational level (data not shown). We stratified individuals according to changes in physical activity, dietary habits, smoking, or alcohol consumption, as either healthier, unhealthier, or no change, during the five-year period, however, the GRS was not associated with body weight changes in either of these categories (Figure 2). Accordingly, no interactions were observed between the GRS and changes in physical activity (Pint = 0.86), dietary habits (Pint = 0.48), smoking (Pint = 0.04), or alcohol consumption (Pint = 0.56).

Figure 2.

Body weight changes by changes in lifestyle. Numbers in parentheses indicate the percentage of women, columns indicate mean body weight change, whereas error bars indicate standard errors. Black dots indicate mean body weight change per GRS risk allele and error bars indicate 95% confidence intervals. Pint indicates P-values for interaction between GRS and lifestyle factors.

Discussion

In the Danish Inter99 population 3,982 participants were followed for five years with both lifestyle intervention counselling and health examinations. Overall, the level of physical activity was unchanged among these participants during the five-year period, whereas dietary, smoking, and alcohol habits became generally healthier. Still, an average increase in body weight of 1.18 kg was observed in the study population. The observed weight change is, albeit smaller, comparable to the weight change of 1.52 kg observed during four years of follow-up in a study of 120,877 U.S. subjects [23]. A GRS based on 30 validated BMI variants identified through GWAS associated strongly with body weight both at baseline and at the follow-up examination, whereas no statistically significant influence was observed for changes in body weight during five years. Nevertheless, based on results from the present study, we cannot exclude a minor effect of the GRS on changes in body weight. As a per allele change in body weight of 0.02 (−0.03-0.07) kg was observed, we can, however, with 95% confidence exclude a per allele effect on body weight change lager than 0.07 kg. This suggests that among middle-aged adults the GRS associates with the overall or usual level of body weight but not with fluctuations in body weight over time. It also suggests that relevant lifestyle changes are of benefit for weight changes independently of currently known BMI variants.

Various reasons could explain the missing association between the GRS and body weight changes. Firstly, as the present study was restricted to individuals aged 30–61 years it cannot be excluded that the studied GRS influences weight gain earlier in life. This scenario is supported by a study of a GRS based on 11 BMI variants associating with a faster tempo of weight gain during the first 11 years of life, but not later [24]. Generally, the hypothesis of a stronger genetic influence on BMI early in life is substantiated by stronger heritability estimates among children which decrease towards adulthood [3]. A larger effect of environmental factors on body weight variance in adulthood could dilute and make the genetic impact more heterogenic, e.g. through gene × environment interactions. Changes in lifestyle, including physical activity and dietary and smoking habits were strong predictors of changes in body weight during the five-year period; however, neither differentiated effect of the GRS according to changes in lifestyle, nor interactions between the GRS and lifestyle changes, could be established in the present study. In accordance with our findings, a study including ∼12,000 white Europeans observed no association between a GRS based on 12 BMI variants and changes in BMI during a three to four year follow-up period [25]. However, this study found that the level of baseline physical activity modulated the effect of the GRS on BMI changes. We studied the interaction between the GRS and changes in physical activity on body weight changes and found no interaction. Neither did we observe an interaction between the GRS and baseline physical activity (data not shown). Recently, a study of ∼33,000 individuals with European ancestry, reported that the association between a GRS based on 32 BMI variants and adiposity appeared to be more pronounced with greater intake of sugar-sweetened beverages [26], however, we detected no such interaction (data not shown).

Secondly, an explanation of the missing association between the GRS and body weight changes could be that the genetic background of usual levels of body weight and body weight fluctuations differs substantially. The GRS applied in the present and previous studies [24, 25], is based on genetic variants which all were identified in cross-sectional GWAS aimed at uncovering genetic variation associating with the levels of BMI at random points of time in life, which could explain their missing impact on fluctuations in body weight. Accordingly, the GRS explained a larger proportion of the baseline body weight variance (3.11%) compared to the variance of body weight changes during the five-year follow-up period (1.40%). The explained variance of BMI at baseline was slightly higher (3.45%) than for body weight at baseline, which could reflect the fact that the genetic variants included are identified using BMI as body composition measure. Although our study suggests that known obesity genetics does not influence body weight changes in adults, prospective genetic discovery studies, and well-powered prospective studies of gene × environment interactions may enlighten whether genetics besides from determining the usual level of BMI also influences weight changes during adult life, either alone or in interaction with environmental factors.

Whereas exclusion of the FTO variant did not substantially influence the relationship between the GRS and baseline body weight, the theoretical difference in body weight between individuals carrying no risk alleles and maximal number of risk alleles was notably larger for the GRS based on all 30 SNPs compared to the scores including only SNPs identified prior to the GIANT meta-analysis or SNPs identified in the meta-analysis. Together, our results suggest that inclusion of SNPs with relative low effect sizes do not infer uncertainty in the statistical models but strengthen the association between a GRS and body weight.

To obtain the most reliable, unbiased and transferable weighing of the GRS, we used effect sizes reported in the largest meta-analysis of BMI GWAS [8], acknowledging that the effect sizes consist of independent estimates for the variants known prior to the meta-analysis and discovery estimates for the newly identified SNPs. However, this weighted GRS showed similar results as the un-weighted GRS, suggesting that for the variants identified at present a simple sum of risk alleles may adequately describe the cumulative effect of carrying multiple risk alleles. The lacking effect of weighting the GRS could be attributable to the type of variants included; all being common and exerting modest effects. Nevertheless, future identification of rare variants with large effect sizes may substantiate the use of a weighed GRS. Moreover, characterization of the biological targets of the SNPs would enable construction of pathway-centerd GRS improving the specificity thereby acknowledging the heterogeneity of obesity.

In conclusion, a GRS based on 30 BMI SNPs identified in GWAS associated with baseline body weight but not with changes in body weight during a five-year follow-up period. Changes in known body weight modulating lifestyle factors, including physical activity, diet, and smoking habits, associated strongly with body weight changes, however, no modulating effect of the GRS was demonstrated. This suggests that accumulation of known BMI variants may influence adiposity in an early stage of life determining the usual body weight level, whereas fluctuations in body weight during adult life are strongly influenced by changes in lifestyle irrespectively of currently known BMI variants.

Acknowledgments

We are indebted to the staff and participants of the Inter99 Study for their important contributions. The authors wish to thank A. Forman, T. Lorentzen, B. Andreasen and G.J. Klavsen for technical assistance and A.L. Nielsen, G. Lademann and M.M.H. Kristensen for management assistance.

Ancillary