Recent investigations have shed light on the genetic mechanism of adiposity and related traits. Several large studies have demonstrated an association between FTO gene polymorphisms and weight, BMI and related traits [3, 6, 7]. Moreover, the FTO rs9939609 SNP was also associated with PCOS, and its effect was mediated, as in the general population, by adiposity . Interestingly, it was postulated in the within-patient analysis that the impact of this particular genetic variant could be larger in individuals with PCOS than in the general population . A greater effect of rs9939609 SNP variants on weight-related variables in women with PCOS was also observed in several other studies [24–27]. However, none of these studies was adequately powered to confirm that the greater effect in women with PCOS was significantly different from that in the general population.
In this study, a meta-analysis demonstrated that each additional copy of the effect allele (A/C) was associated with an average increase in BMI of 0.19 z score units and a mean weight gain of 0.20 z score units. This corresponded to around a 3.3 kg/m2 and a 9.6 kg difference in BMI and body weight units, respectively, between TT and AA/CC homozygotes. Importantly, this effect was highly consistent between the PCOS studies included in this meta-analysis, with no evidence of heterogeneity between the analysed populations. The impact of FTO variants on BMI and weight is significantly stronger than those observed for the general female population as reported in the previous studies of FTO conducted on adult participants not selected for PCOS [3, 6, 15, 33–35].
Unexpectedly, the analysis revealed a variation between the studies with regard to the inheritance model of the FTO polymorphisms, which seemed to be related to the BMI in the investigated populations. Phenotypic manifestation of the heterozygous FTO variant existed only in studies with the lowest mean BMI in the TT group. In trials with a higher mean BMI in the TT group, a single variant of an effect allele was not associated with an apparent increase in obesity indices. For those studies, the recessive model fits best. To our knowledge, this phenomenon has not been observed in the general population.
The current study has several possible limitations. First, our PCOS cohorts with a mean age ranging from 25 to 32 years old were younger than those reported in the general female population studies. Effect sizes adjusted for age were not available for either reference group, either as unpublished data or from earlier publications [3, 6]. Nevertheless, the age adjustment in women with PCOS had a very limited impact on both the observed association between FTO and BMI and the results of comparisons between the patients with PCOS and the women in the general population, which remained almost unchanged. Moreover, the FTO-associated increase in BMI reported by Frayling et al  was comparable in the young and older cohorts, with no evidence of between-group heterogeneity. The per allele increase in logBMI z scores in the young female ALSPAC Mothers cohort (N = 6,376; mean age 28.4 years) and the NFBC1996 cohort (N = 2,306; each patient sampled at 31 years old) was not significantly different from the effect size observed in elderly women from the BWHS cohort (N = 3,244; mean age 68.8 years) as shown by the heterogeneity test (p = 0.79; I
2 = 0%, for all three cohorts) . It thus seems unlikely that the effect of FTO variants on BMI in PCOS is due to the young age of the women examined.
Another possible limitation includes a lack of direct comparison with control cohorts consisting of women without PCOS matched with the analysed study populations. Such comparisons were either not made in the original studies [23, 25, 27] or were based on a very limited sample size, precluding quantitative analysis . Nevertheless, we compared our study data with results, published and unpublished, from two large cohorts representing the general female population (GIANT Consortium, unpublished data) . This analysis showed a statistically larger effect of the FTO variants in PCOS than in the reference groups [3, 6]. Second, although the possibility of a publication bias in reports examining the impact of the FTO gene in PCOS cannot be entirely excluded, it seems unlikely in light of the results of the statistical tests (Egger's and Begg's tests). Moreover, all studies regarding PCOS and the FTO gene were included even if they were originally designed not to test for the difference in impact of the FTO genotypes on weight or BMI.
The sample size of 2,548 individuals is smaller than those of many genetic studies comparing effect sizes. In part, this number reflects the limited number of studies in which PCOS women were genotyped for the FTO SNP as we included all available studies referring to FTO in PCOS. The problem of the limited number of PCOS patients who were genotyped may be, at least in part, associated with the need to follow a diagnostic algorithm for this disease that is costly and time-consuming. We acknowledge that, while our study tested the hypothesis that FTO genetic variation has a greater impact on BMI and body weight in the context of PCOS, a biological explanation for this phenomenon remains to be elucidated. The unique metabolic setting of PCOS, which includes significant alterations in sex steroid hormone levels (hyperandrogenaemia and significant extraglandular aromatisation of androgens), reduced sex hormone-binding globulin levels, and insulin resistance, may modify the influence of the FTO genotypes . Alternatively, gene–gene interactions may contribute to the greater influence of FTO. Further studies are needed to elucidate the roles of biochemical and genetic factors affecting the action of FTO in PCOS.
The major studies that have demonstrated an impact of FTO on body weight have been conducted primarily on white populations [3, 37–39]. This initial finding in white cohorts was confirmed in a Mexican population as well as in multiple studies including a recent large-scale meta-analysis of the East and South Asian population [12, 14–16]. In spite of initial data suggesting that FTO had no clear impact on phenotype in African cohorts , the effect of the gene variants on obesity was eventually established in a new genome-wide association study in a population with a predominantly African ancestry in the USA . The lack of effect in the initial African study may have been due to the impact of environmental circumstances, such as seasonal food shortages and high physical activity in the African population [40, 41].
Results from one study of the Chinese population, which could not be included in our meta-analysis due to a lack of numerical data, suggested no clear association between FTO allelic variation and obesity-related traits in both healthy individuals and women with PCOS . Subjects from studies included in our meta-analysis were mainly white. Consequently, our findings cannot be generalised to other ethnic groups. Our observations might not be specific to PCOS, as there might exist other phenotypes and diseases in which the effect of the FTO genotypes is larger than in the general population. Finally, we should point out that a slightly greater effect of FTO genotypes was observed in heavier diabetic cohorts than in the general population . Nevertheless, the magnitude of the current finding and its presence in obese and non-obese populations with PCOS suggest that it has a different mechanism in PCOS.
In conclusion, our meta-analysis, based on eight distinct cohorts, shows that the per allele effect of FTO polymorphisms on BMI and weight seems to be more than two times greater than the effect found in large population-based studies. Our results suggest that the metabolic context or specific polygenic background of PCOS modifies the influence of FTO on weight and BMI.