2.1 Search Strategy

A systematical search was conducted using PubMed, Google scholar, Scopus and Web of Science until January 2024 to retrieve all relevant for the FTO and variants and any significant keywords. The keywords and search strategy included: “obesity,” “overweight,” “body mass index,” “FTO,” “fat mass and obesity associated,” “children” and “adolescent.” The search was limited to only the English reports. Moreover, additional studies were revealed by a manual search including studies in systematic review and meta-analysis studies on the subject. The search strategy for each source was provided as Supporting Information. All the phase of this study was in line with recent version PRISMA guideline [29].

2.2 Inclusion and Exclusion Criteria

The studies that are included in this systematic and meta-analysis comprised the following criteria: case–control, cross-sectional and cohort studies that compare overweight/obese children with the healthy control group. Studies that contained genotype and allele frequency and determining the relation of FTO variants and overweight/obese in children and adolescents. Moreover, our exclusion criteria were comprised of (1) studies with no control group, (2) studies on adults, (3) studies on case group with secondary obesity or metabolic disorder, (4) case reports, animal studies, letters, abstract (without sufficient data), reviews and (5) studies with overlap genotype data of previous publications by same authors or groups. It should be noted that the studies in which Hardy–Weinberg equilibrium (HWE) was not observed was not excluded. We removed their effect from the final pooled estimate in sensitivity analysis.

2.3 Data Extraction

Eligible studies independently were extracted by four investigators (SC, ME and ST). The articles information, sample size, sex, ethnicity, age, BMI, SNPs reference number, genotypes distribution, genotyping methods and HWE states were extracted. The investigators discussed and resolved any disagreements. Sample size, HWE, control group, detection methods and ethnicity have been examined because they can influence the FTO gene polymorphism and overweight/obese risk relationship. Moreover, as different risk alleles were introduced for some SNPs of FTO gene, in order to be more uniform from methodologic perspective, we decided to use the risk alleles introduced in the dbSNP database (www.ncbi.nlm.nih.gov/snp/).

2.4 Outcomes

Upon reviewing the included studies, it was observed that the association was reported in three different ways. The majority of studies showed this association only in populations of obese children and/or adolescents. Some studies, on the other hand, demonstrated this association in a combined population of obese and overweight children. Finally, a small number of studies evaluated the association of FTO gene polymorphisms in the overweight children population. Therefore, the outcomes were categorised based on the composition of the samples in the included studies.

2.5 Statistical Analysis

For the meta- analysis R (version 4.2.1; R Core Team, 2022) was employed for analysis. In this study, we developed five different forms of analysis based on inheritance models. However, it should be noted that, as there are different risk alleles for different SNPs of FTO genes, the format demonstrating each model is symbolic and does not refer to a particular SNP. Also, we considered allele ‘A’ as the risk allele in this symbolic demonstration of the genetic models. Accordingly, the dominant model was defined based on the following format ‘AA + Aa versus aa’. The recessive model was defined as ‘AA versus Aa +aa’. Likewise, we consider two different models which we defined as follows: ‘AA versus aa’ (known as homozygote model) and ‘Aa versus aa’. Finally, we considered the allelic model based on frequency of each allele in the case and control group. To perform the meta-analysis, we divided our main data set into three groups. In one data set, we only included the studies which reported the effect of each of SNPs in a population of children and/or adolescents with obesity. The second data set only contained data of studies that evaluated the effect of SNPs on mixture of obese and overweight children and/or adolescents. Eventually, the third one comprised of the reports that only included the effect of SNPs on overweight children and/or adolescents. Then we used dichotomous data meta-analysis using ‘metabin’ command of ‘meta’ package of R [62]. To select which model to report, fixed versus random, we used the value of I². The I² values served as a measure of heterogeneity [63]. The fixed effect model was applied, if I² < 50% and p > 0.1 between studies. To measure publication bias in the collected studies funnel plots was created alongside checking Egger's test. It should be noted that due to restrictions exist for conducting such an analysis, we only evaluated the presence of possible publication bias for rs9939609. p values < 0.05 were considered statistically significant.

3.1 Description of Studies

In total, 4913 studies were identified by searching the electronic databases. After excluding duplicates and ineligible studies, 32 studies with 14,210 as case and 23,922 as control participants were included in the meta-analysis (Figure 1). As, in this study there was no specification over particular SNP, we included all the reported SNPs which their effect on risk of obesity and overweight in children and adolescents were evaluated. The SNPs which were reported at least in two different studies were considered for the final meta-analysis.

3.2 Search Results

We identified 20 titles during the initial systematic search (Figure 1). Then 12 additional studies were added after checking reference lists of relevant studies alongside manual search of Google Scholar. A total 32 articles met the inclusion criteria. Out of the 32 studies, nine studies were from China [43, 46, 49, 50, 52, 53, 55, 56, 60], three studies were from Spain [39, 42, 44], two studies were from India [35, 45]. Other studies were once seen in the following countries: Iran [61], Iraq [64], Poland [58], Brazil [54], Italy [51], Mexico [41], Pakistan [48], Portugal [40], Russia [32], Sweden [31], Turkey [37], United Kingdom [36], Czech Republic [59], Germany [30], Austria [34], Chile [47], Egypt [38] and Indonesia [33].

3.3 Demographics

An overview of general characteristics of the included studies is displayed in Table S1. The sample size varied from 24 to 1423 in case group and from 45 to 4022 in healthy participants. Overall, there was 14,930 participants in case group and 24,765 in control group. Most of the studies based their inclusion criteria for guidelines, using different cut-offs such as BMI or percentiles for definition of obesity and overweight among the population of children and/or adolescents.

3.4 Effect of Different FTO Gene SNPs on Risk of Obesity

Among the evaluated SNP of FTO gene in children and adolescents with obesity, only five SNPs (rs9939609, rs1421085, rs1861868, rs1477196 and rs17817449) were assessed in several studies. Among these five SNPs, two of them, rs9939609 (OR: 1.32, 95% CI: 1.14; 1.54) and rs1421085 (OR: 1.73, 95% CI: 1.26; 2.380), increased the risk of obesity in the dominant model (Figure 2 and Figure S1). The same SNPs showed similar effects in the recessive and allelic models (Figure 2 and Figures S2 and S5). Besides, in homozygote model, rs9939609 and rs1421085 showed strong association with obesity (OR: 1.72, 95% CI: 1.28; 2.31 and OR: 2.32, 95% CI: 1.38; 3.89, respectively). In the other model (Aa vs. aa), rs1421085 showed significant association (Figure 2). Forest plot for all five SNPs are available in the Figures S1–S5.

3.5 Effect of Different FTO Gene SNPs on Risk of Obesity and Overweight

The effect of only two SNPs, rs9939609 and rs9935401, were evaluated on the combined population of obese and overweight children and adolescents in different studies. Both of these SNPs have risen the risk of obesity and overweight in our target population. According to our analysis, the OR of the rs9939609 in dominant and recessive models were 1.28 (95% CI: 1.09; 1.51) and 1.59 (95% CI: 1.29; 1.97), respectively. Regarding the rs9935401, the highest level of the associations was observed in homozygote 1.77 (95% CI: 1.39; 2.27) and recessive 1.56 (95% CI: 1.38; 1.77) models (Figure 3).

3.6 Effect of Different FTO Gene SNPs on Risk of Overweight

The number of studies which assessed the effect of SNPs of FTO genes on risk of overweight were very few. Aside from rs9939609, only one other SNP, rs6499640, was evaluated in different studies. Interestingly none of these two SNPs showed any association with an increased risk of being overweight in children, but in the allelic model of rs9939609 which showed a mere harmful effect (A vs. T) (OR: 1.12, 95% CI: 1.01–1.25) (Figure 4).

3.7 Qualitative Assessment

Aside from the SNPs which were included in the meta-analysis, there were some other SNPs evaluated only in one report. Table 1 summarises the effect of these SNPs in population of children and adolescents. In this regard some reports indicated that rs9921255, rs6499640 and rs72066790 had association with increased risk of obesity. There were two more SNPs (rs9928094 and rs9930333) which showed correlation with high risk of obesity and over weight in a population of Spanish children and adolescents. As these associations are only observed in a single study, further validations are needed to provide a solid picture over the correlation of these SNPs with obesity and overweight.

3.8 Publication Bias

A funnel plot was utilised to examine the publication bias of the included articles. These curves demonstrate the correlation between polymorphisms and the risk of obesity in allelic, dominant, recessive and co-dominant genetic analysis models. The symmetrical shape of the funnel plots indicated that there is no publication bias in our study of FTO polymorphisms. Also, and based on our analysis (only for rs9939609 in obesity subgroup), the Egger test was insignificant for all models. The output of our analysis in this respect is presented in Figure S6.

3.9 Sensitivity Analysis

In most included studies, the genotypes are compatible with HWE which was interesting to be statistically significant at p ≤ 0.05. To exclude the effect of incompatible HWE studies, we conducted a sensitivity analysis. Accordingly, after removing the studies with incompatible HWE, the pooled estimates of all models were increased for rs9939609 in all genetic models (Figure S7–S11). Moreover, after careful inspection of the data set, it was realised that, if we applied the same strategy to other SNPs, the possibility of conducting meta-analysis will be violated. The rs1421085 was an exception, since in all the studies, the HWE was confirmed for this particular SNP.

4.1 Limitation and Strengths

To the best of our knowledge, this meta-analysis is the first to comprehensively evaluate the association of FTO gene polymorphisms with childhood obesity. Previous studies mainly focused on the effect of FTO rs9939609 on childhood obesity. However, in this study, we aimed to provide a comprehensive overview of the other FTO gene polymorphisms alongside FTO rs9939609 itself. Decentralising the focus from one polymorphism to the others is important as it allows for consideration of their association and opens up new research avenues. This is critical because our study revealed that many other polymorphisms showed a stronger association with obesity and overweight than FTO rs9939609 in the same studies.

Despite these important findings, there are still some limitations in this study that should be acknowledged. Firstly, only limited ethnicities were included which influenced the level of association observed. Therefore, conducting GWAS studies in different ethnicities and considering multiple polymorphisms can provide a more precise and comprehensive understanding of the association between FTO gene polymorphisms and childhood obesity.

Furthermore, obesity is a complex disorder and focusing only on genetic factors may provide a biased view of the disease's pathophysiology. Therefore, it is crucial to include other factors such as behavioural and environmental parameters in the analysis. Unfortunately, due to lack of data, our study was unable to measure the effects of other parameters such as environment and diet factors alongside their interactions with genetic parameters.