Interaction between genetic susceptibility to obesity and food intake on BMI in Finnish school-aged children

Diet modulates the genetic risk of obesity, but the modulation has been rarely studied using genetic risk scores (GRSs) in children. Our objectives were to identify single nucleotide polymorphisms (SNPs) that drive the interaction of specific foods with obesity and combine these into GRSs. Genetic and food frequency data from Finnish Health in Teens study was utilized. In total, 1142 11-year-old subjects were genotyped on the Metabochip array. BMI-GRS with 30 well-known SNPs was computed and the interaction of individual SNPs with food items and their summary dietary scores were examined in relation to age- and sex-specific BMI z-score (BMIz). The whole BMI-GRS interacted with several foods on BMIz. We identified 7–11 SNPs responsible for each interaction and these were combined into food-specific GRS. The most predominant interaction was witnessed for pizza (p < 0.001): the effect on BMIz was b − 0.130 (95% CI − 0.23; − 0.031) in those with low-risk, and 0.153 (95% CI 0.072; 0.234) in high-risk. Corresponding, but weaker interactions were verified for sweets and chocolate, sugary juice drink, and hamburger and hotdog. In total 5 SNPs close to genes NEGR1, SEC16B, TMEM18, GNPDA2, and FTO were shared between these interactions. Our results suggested that children genetically prone to obesity showed a stronger association of unhealthy foods with BMIz than those with lower genetic susceptibility. Shared SNPs of the interactions suggest common differences in metabolic gene-diet interactions, which warrants further investigation.

Food-specific GRS and their interactions.The associations of food-specific GRS, food intake, and their interaction on BMIz were tested in two models (Table 3): model 1 was adjusted only for sex, while model 2 was additionally adjusted for physical activity and sleep duration (fully adjusted).The fully adjusted interactions were further illustrated in Fig. 1.The interactions were validated for pizza, sweets and chocolate, sugary juice drink, and hamburger and hotdog when the adjusted mean effect sizes differed between the low-and high-risk groups.
The most predominant interaction was marked for pizza: it associated inversely with b − 0.130 (95% CI − 0.23; − 0.031) with BMIz in those with low GRS, while positively with b 0.153 (95% CI 0.072; 0.234) with BMIz in those having high GRS.Sugary juice drink followed the same pattern with the exception that the association among the low GRS group did not reach formal significance.Significant interactions were noted for hamburger and hotdog (p = 0.027) and sweets and chocolate (p = 0.011): the verification followed the same pattern but without formal significance.Shared SNPs. Figure 2 illustrates the shared SNPs between the four food items.In total, we identified 15 out of 30 SNPs presenting an interaction.Interestingly, 33% of the SNPs were shared between pizza, sweets and chocolate, sugary juice drink, and hamburger and hotdog.These SNPs are close to the following genes (expressed in high magnitudes in these tissues): NEGR1 (brain), SEC16B (liver/pancreas), TMEM18 (bone), GNPDA2 (non-specific), and FTO (non-specific).The description of the SNPs and genes is presented in Supplementary Table 3.

Discussion
Initially, we observed interactions between the whole BMI-GRS and certain foods on BMIz in school-aged children from Finland.Further investigations demonstrated that each interaction was driven by 7-11 SNPs.When combining these SNPs into food-specific GRS we verified an interaction for pizza, sweets and chocolate, sugary juice drink, and hamburger and hotdog.Thus, children bearing more risk alleles for obesity showed a Table 1.Background characteristics of the participants in low-and high-risk groups, using the median number (= 27) of risk alleles as the cut-off, reported as mean (SD), if not indicated otherwise.www.nature.com/scientificreports/stronger association of weight-promoting foods on BMI than those with fewer obesity-risk alleles.Importantly, there were no differences in the consumption frequency of these foods between the groups with varying genetic susceptibility, suggesting that the effect originates from a different kind of response to food than from a difference in consumption pattern.No interactions were observed for dietary summary scores describing overall eating habits or summary food scores for sugary foods or fruits and vegetables.
The health profile of the interacting foods was considered weight-promoting based on earlier studies due to their energy-dense-nutrient-poor characteristics 43 .The most predominant interaction was observed with pizza on BMIz, e.g., a positive association with BMIz in the high-risk group while opposite in the low-risk group.Pizza consumption is the top contributor for intakes of total energy, saturated fat and sodium in US children and teens, with a daily consumption frequency of 20% 44 .In our study, once a week/in two weeks were the most common consumption patterns of pizza (> 70%).In a systematic review 45 , TMEM18 and FTO were linked with total energy and fat intakes, thus partly supporting our findings.
In total, we identified 15 out of 30 SNPs as being responsible for the observed interactions for pizza, sweets and chocolate, sugary juice drink, and hamburger and hotdog.Interestingly, 33% of the interacting SNPs were shared between the foods.These included SNPs in or near genes NEGR1 (rs2815752), SEC16B (rs543874), TMEM18 (rs2867125), GNPDA2 (rs10938397) and FTO (rs1421085).Except for TMEM18, the other SNPs were previously shown to drive the interaction of fried foods, e.g., any deep-fried foods enjoyed at home or away from home on BMI in three US cohorts 40 , while an independent effect was noted only for FTO.The FTO (rs1421085) gene has been associated repeatedly with various obesity phenotypes in different study designs and populations 46 (Supplementary Table 3), and its expression is aggregated in primary adipocytes 47 .The variant rs1421085 in the first intron of the FTO gene regulates the adipocyte-thermogenesis pathway by interacting with other genes (ARID5B, IRX3, and IRX5) 47 .Previous reports have witnessed multiple interactions of FTO variant rs1421085 with the intake of fiber 48 , dietary variation, alcohol consumption, and sedentary behaviors on BMI among adults 49 .
Our unique finding concerned TMEM18 (rs2867125), which has been associated with pediatric BMI 50,51 , but here it presented an interaction with several foods.The contribution of any GRS or SNP may vary with age and in different stages of life 13 .The total BMI-GRS used here was significantly associated with BMIz and explained 3.7% of the variance in children 52 , which is somewhat higher than reported in adults 22 .Studies looking at genetic interaction with diet on BMI in pediatric cohorts are scarce 51 but informative, since food consumption is more naïve and less affected by social acceptance in children than in older age groups.Thus, our results on TMEM18 may imply that the BMI trajectory in childhood is modified by the food intake, e.g., most likely energy-rich foods provide more support for growth.
Although the weight and waist differed by the genetic susceptibility to obesity; other lifestyle factors including sleep duration, and leisure-time physical activity (LTPA) were similar between the groups.Furthermore, Table 2. P value for interaction between BMI-GRSs and food items regarding BMIz.The analyses were adjusted for sex, leisure-time physical activity, sleep duration and 1st and 2nd principal coordinates (PC) for population structure.a Sum of risk alleles.b Effect sizes from Speliotes et al. 21.c Effect sizes from Fin-HIT 52 .d Effect size is the ratio between Fin-HIT and Speliotes.e Dicotomized BMI-GRS.Significant values are in [bold].
Unweighted BMI-GRS a BMI-GRS  www.nature.com/scientificreports/sleep duration and LTPA only marginally affected the results, suggesting that the interaction was independent of these factors.Certain risk variants of FTO (rs9939609), TMEM18 (rs4854344), and NRXN3 (rs10146997) have been reported to increase the vulnerability to metabolic conditions in children under sleep deprivation 53 , thus interacting with lifestyle factors, but that was not observed here.Sugary juice drinks are widely consumed among young Finnish children and enjoyed with a snack, instead of milk or juice at meals 54,55 .The healthiness of sugary juice drinks is frequently discussed as the dilute berryderived squash contains mainly sugar and provides energy, but barely nutrients.Frequent sugary juice drinkers will likely evolve with time into consumers of carbonated sugary-sweated beverages (SSB), which are deemed as weight-promoting foods 56 .Additionally, two earlier studies have demonstrated SSB to interact with the obesityprone genotype 38,39 .Similar to our finding on sugary juice drink, the reported magnitude of association between SSB and BMI was greater among those genetically prone to obesity, implying that the downstream effects after consuming SSB differ between the individual, making obese-prone more vulnerable to weight gain.Furthermore, Brunkwall's study 39 highlighted that the SSB-BMI interaction was mainly driven by one SNP -rs1555543, close to gene PTBP2, among middle-aged Swedish individuals.The same SNP has demonstrated an interaction with smoking on BMI in the Pakistani population 57 .However, we did not observe any interaction of rs1555543 in our sample, possibly due to the young age of the participants.
The ultimate strength of the study is that we used a cohort of school-aged children whose food consumption is likely less affected by social acceptance.Although mis-and underreporting are common challenges in dietary assessment, it is shown that amongst 11-12-year-old children that the FFQ is a valid method and independent of BMI, implying that social acceptance and desirability are less common in children than in older age groups 58 .Thus, we observed no differences in food consumption frequencies.However, we did not address portion sizes, which may differ by BMI 59 .The study was facilitated by a previously reported association of BMI-GRS and BMIz 52 , relying on 30 well-characterized SNPs.Our results may be generalized to a comparable European population with a similar socioeconomic background.Based on our earlier work 52 , using a GRS with more SNPs would most likely result in similar outcomes, as the GRSs present with corresponding associations with BMIz.
Due to the limited sample size and using the tailor-made Metabochip array only obesity SNPs were considered.Future studies with larger sample size and genome-wide coverage of SNPs are warranted for broader A B Figure 1.Food-specific GRS and their confirmed interactions.The results are divided into two panels for clarity and based on the food-specific GRS presented with b-coefficients with 95% CI.The most predominant interaction was marked for pizza.Other notable interactions were sugary juice drink, hamburger and hotdog, and sweets and chocolate but without formal significance.The figure was made with PRISM version 9.5.0 (https:// www.graph pad.com/).investigations of the interactions between genes and diet.The food frequency questionnaire (FFQ) covered 16 food items and was considered suitable for 11-year-old children to comprehend 60,61 .However, it might have been too narrow to distinguish between foods with varying health profiles, e.g., all dairy products were considered together without considering differences in the nutrient content.Thus, we might have lost some of the information.Because power for detecting interactions is typically much lower than power for main effects, we raised the Type I error rate to 20% when assessing interactions as suggested 62,63 .On the other hand, this might increase the chance of false positive results.However, we illustrated the association in subgroups as well.
In conclusion, the interacting foods with the genetic risk of obesity were mainly weight-promoting in Finnish children.Our results point out that children genetically prone to obesity showed a stronger association of unhealthy foods with BMIz than those with lower genetic susceptibility.Since a part of the SNPs driving the interactions were shared between the weight-promoting foods, this implies metabolic differences among genetically prone individuals, which warrants further studies in this and other geographically diverse populations.

Methods
We have conducted a cross-sectional analysis of 1142 Finnish children.For this study, we utilized the background characteristics, genotype data, and anthropometric measurements from the Finnish Health in Teens cohort (Fin-HIT), launched in 2011 as a school-based cohort study, initially comprising 11,407 Finnish children aged between 9 and 12 years.The details of the Fin-HIT cohort are described elsewhere 64 .The Coordinating Ethics Committee of the Hospital District of Helsinki and Uusimaa has approved the study protocol (169/13/03/00/10) and written informed consent was obtained from all participants and their parents.All study procedures adhered to the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.
DNA extraction, genotyping, quality control, and generation of genetic risk score.The participants provided saliva samples by using the Oragene® DNA (OG-500) Self-Collection Kit (DNA Genotek Inc., Ottawa, Ontario, Canada).DNA was extracted using an automated protocol with the chemagic DNA Saliva Kit (PerkinElmer, Wellesley, Massachusetts).DNA samples (n = 1368) were randomly selected from the Fin-HIT cohort and subjected to genotyping with the Cardio-Metabochip (Illumina, Inc., San Diego, California) at the Finnish Institute for Molecular Medicine Technology Centre (Helsinki, Finland) as explained elsewhere 52 .The number of individuals and SNPs included in the final analysis after QC was 1142 and 125,187 with a total a Chi-Square, Missing values in low-and high-group, respectively: b n = 55 and n = 52, c n = 42 and n = 27, d n = 4 and n = 5, e n = 39 and n = 33, f n = 62 and n = 47, g n = 13 and n = 16, h n = 13 and n = 16, i n = 9 and n = 8, j n = 10 and n = 2, k n = 8 and n = 2, l n = 7 and n = 4, m n = 36 and n = 18, n n = 11 and n = 8, o n = 17 and n = 13, p n = 11 and n = 6, q n = 9 and n = 5, r n = 9 and n = 3, s n = 10 and n = 7, t n = 7 and n = 6, u n = 10 and n = 7, v n = 7 and n = 4, w n = 9 and n = 5, x n = 9 and n = 1.

Figure 2 .
Figure 2. Venn diagram of shared SNPs by food items.Five (33%) SNPs interactions (in red intersection) were shared between pizza, sweets and chocolate, sugary juice drink, and hamburger and hotdog.The figure was made through R version 4.2.2 (https:// posit.co/ produ cts/ open-source/ rstud io/).