A healthy childhood environment helps to combat inherited susceptibility to obesity

Objectives To investigate the degree by which the inherited susceptibility to obesity is modified by environmental factors during childhood and adolescence. Design Cohort study with repeated measurements of diet, lifestyle factors and anthropometry. Setting The pan-European IDEFICS/I.Family cohort Participants 8,609 repeated observations from 3,098 children aged 2 to 16 years, examined between 2007 and 2014. Main outcome measures Body mass index (BMI) and waist circumference. Genome-wide polygenic risk scores (PRS) to capture the inherited susceptibility of obesity were calculated using summary statistics from independent genome-wide association studies of BMI. Gene-environment interactions of the PRS with sociodemographic (European region, socioeconomic status) and lifestyle factors (diet, screen time, physical activity) were estimated. Results The PRS was strongly associated with BMI (r2 = 0.11, p-value = 7.9 × 10−81) and waist circumference (r2 = 0.09, p-value = 1.8 × 10−71) in our cohort. The associations with BMI increased from r2=0.03 in 3-year olds to r2=0.18 in 14-year olds and associations with waist circumference from r2=0.03 to r2=0.14. Being in the top decile of the PRS distribution was associated with 3.63 times higher odds for obesity (95% confidence interval (CI): [2.57, 5.14]). We observed significant interactions with demographic and lifestyle factors for BMI as well as waist circumference. The risk of becoming obese among those with higher genetic susceptibility was ~38% higher in children from Southern Europe (BMI: p-interaction = 0.0066, Central vs. Southern Europe) and ~61% higher in children with a low parental education (BMI: p-interaction = 0.0012, low vs. high). Furthermore, the risk was attenuated by a higher intake of dietary fiber (BMI: p-interaction=0.0082) and shorter screen times (BMI: p-interaction=0.018). Conclusions Our results highlight that a healthy childhood environment might partly offset a genetic predisposition to obesity during childhood and adolescence.


76
Obesity is a complex multifaceted condition and its prevalence has been increasing 77 continuously over previous decades and has reached a high plateau in Western countries [1].

78
In 2015, a total of 107.7 million children and 603.7 million adults were obese. Although the 79 prevalence of obesity among children has been lower than that among adults, the rate of 80 increase in childhood obesity has been greater than the rate of increase in adult obesity, which 81 is most likely due to adverse changes of environmental and demographic factors with a direct 82 impact on children's health [2].

83
With the advent of genome-wide association studies (GWAS), it was shown that multiple 84 genetic loci increase the susceptibility to obesity [3,4]. However, genome-wide significant 85 variants identified in the first large-scale GWAS on body mass index (BMI) only account for a 86 small portion of BMI variation (~2.7%) [3]. A more recent genome-wide meta-analysis extended 87 the number of individuals from ~300,000 [3] to ∼700,000 [4], which consequently increased 88 the number of genome-wide significant SNPs from 97 to 751. Even these 751 genome-wide 89 significant SNPs account for only ∼6.0% of the variance of BMI [4]. However, genome-wide 90 estimates suggest that common variation accounts for >20% of BMI variation [3], which 91 highlights the polygenic architecture of BMI. More recently, whole genome data even increased 92 the fraction of variance of BMI accounted for by genetic variants, both common and rare, to 93 40% [5]. From twin studies we know that the heritability of BMI also depends on socioeconomic 94 status [6] and physical activity [7], suggesting that when socioeconomic status or physical 95 activity is high, genetic factors become less influential. Using candidate SNPs -either single 96 genotypes or <100 SNPs combined in a polygenic risk score (PRS), which is defined as a 97 weighted sum of BMI-related risk alleles -it was further shown that the genetic predisposition 98 to obesity is attenuated by a healthy lifestyle including physical activity [8,9] and adherence to 99 healthy dietary patterns [9][10][11][12][13][14][15]. However, most previous gene-environment (GxE) interaction during childhood and adolescence. Another limitation of previous gene-environment 103 interaction analyses is that they were based on <100 SNPs that reached genome-wide 104 significance in previous GWAS on BMI [3], which do not capture the whole polygenic risk profile 105 of obesity due to their low heritability. Khera et al. suggested that the power to predict BMI by 106 PRS can be improved by using lower p-value thresholds or even genome-wide approaches 107 [17]. Using a genome-wide polygenic risk score based on effect estimates from [3], Khera et 108 al. reported that the PRS-effect on weight and BMI z-scores emerges early in life and increases 109 until adulthood and that a high PRS is a strong risk factor for severe obesity and associated 110 diseases [17]. The authors suggested that given that the weight trajectories of individuals in 111 different PRS deciles start to diverge early in childhood, targeted strategies for obesity 112 prevention may have maximal effect when employed early in life. However, because lifestyle 113 factors were not considered in their study, it is not known to which degree the genetic 114 predisposition to obesity is modifiable by a healthy lifestyle early in life. Another limitation of

115
[17] is the use of weight and BMI as only proxies for obesity. Since several studies have shown 116 that classifying obesity using BMI alone misses an increasing proportion of individuals 117 categorized as obese [18,19], it is important to test the performance of BMI-PRS for the 118 prediction of waist circumference, which is proposed to be a better proxy for obesity-associated 119 metabolic abnormalities [20].

120
In this study, 1) we show the prediction capacity of the PRS proposed in [17] for BMI as well 121 as for waist circumference of European children and adolescents and 2) analyze its interaction 122 with parental education, region of residence, selected dietary variables and physical activity to 123 investigate to which degree the inherited susceptibility to obesity in children is modified by

130
The pan-European IDEFICS/I.Family cohort [21,22] is a multi-center, prospective study on the 131 association of social, environmental and behavioral factors with children's health status.

132
Children were recruited through kindergarten or school settings in Belgium, Cyprus, Estonia,

133
Germany, Hungary, Italy, Spain andSweden. In 2007/2008, 16,229 children aged between 2 134 and 9.9 years participated in the baseline survey. Follow-up surveys were conducted after two 135 (FU1, N = 11,043 plus 2,543 newcomers) and six years (FU2, N = 7,117 plus 2,512 newly 136 recruited siblings). Physical examinations covered a broad spectrum of parameters according 137 to a detailed and standardized study protocol. Questionnaires were completed by parents for 138 children younger than 12 years. In the second follow-up (FU2), adolescents of 12 years of age 139 or older reported for themselves. All questionnaires were developed in English and translated 140 into local languages. The quality of translations was checked by back translation into English.

141
The study was conducted in agreement with the Declaration of Helsinki; all procedures were 142 approved by the local ethics committees and written and oral informed consents were obtained 143 from the parents, their children and adolescents, respectively, as applicable. Children were 144 selected for a whole-genome scan based on their participation in the individual study modules.

259
We found that the PRS-Khera provided the best prediction of BMI (see Table S1 for details on 260 the characteristics of the other PRS). PRS-Khera was strongly associated with BMI (r 2 = 0.11, 261 p-value = 7.9 x 10 -81 ) and waist circumference (r 2 = 0.09, 1.8 x 10 -71 ) in our study population 262 (

265
The correlation between PRS-Khera and BMI increased along the age range, from a squared  Table S2). Similar trends were found for waist circumference, for 268 which the squared correlation with PRS-Khera was r 2 = 0.03 [0.01, 0.07] in 3-year olds and r 2 269 = 0.14 [0.08, 0.22] in 14-year olds ( Figure 1 and Table S2). This increase of correlation by age 270 group was confirmed in our sensitivity analyses using other genome-wide PRS ( Figure S4 and 271 p-value interaction = 0.0246, Figure 2 and Table S4). Interactions were confirmed in our 285 sensitivity analyses using other genome-wide PRS ( Figure S5). We did not find significant  Table S4).

289
The genetic susceptibility to a high BMI was further modified by intake of dietary fiber and

295
Interactions between PRS-Khera and the fruit and vegetable score or MVPA were not 296 significant.

299
In our pan-European cohort of children aged 2 to 16 years, we found a strong association of a 300 polygenic risk score of obesity with BMI as well as with waist circumference and this 301 association increased by age. We observed a prediction r 2 of 18% in 14-year olds, which is 302 even higher than in the original study containing mainly adults [4]. We further found significant we observed gene-environment interactions with (1)

380
Our study also has several limitations. First, measurement errors of self-reported lifestyle 381 behaviors are inevitable. However, measurement error in environmental exposure typically positive findings but reduces the statistical power to detect subtle interactions. Second, the 384 use of PRS derived from associations with BMI in the analyses of waist circumference led to 385 slightly lower prediction accuracy for waist circumference than for BMI. However, since PRS-

386
Khera is known to be a strong risk factor for severe obesity and associated health outcomes

387
[17], we decided to use this PRS for both obesity measurements.

390
Our study showed significant interactions between the polygenic risk for an increased BMI and 391 sociodemographic and behavioral factors that affect BMI as well as waist circumference.

392
Among children with a high genetic risk, we identified children from Southern Europe, children       [24,25]. Boys with a BMI zscore > 2.29 and girls with a BMI z-score > 2.19 were defined as obese [24,25]. Associations adjusted for region of residence, sex, age, parental education, vegetable score. Z-scores for BMI and waist circumference were calculated according to [24,25]. Boys with a BMI z-score > 2.29 and girls with a BMI z-score > 2.19 were defined as obese [24,25].

Figure Legends
Figure 1. Squared correlation (r 2 with 95% confidence intervals) of PRS-Khera with BMI and waist circumference in dependence of age. Squared correlations could not be calculated for ≥15-year old children due to the small sample size in these age groups (see Tables S1 & S2). Waist circumference was not measured in 2-year old children.

Figure 2. Interactions between PRS-Khera and sociodemographic factors on BMI and waist circumference.
Associations between PRS and BMI / waist circumference are shown in different strata (beta estimates and 95% CIs) as well as in the whole study population (red line). Raw pvalues (pr) and FDR-adjusted p-values (pa) are given for the test of deviations of the association between PRS and obesity in one subgroup in comparison to the reference category (interaction). The category without p-values is the reference category.