|Home | About | Journals | Submit | Contact Us | Français|
In 2007, an association of single nucleotide polymorphisms (SNPs) in the fat mass and obesity-associated (FTO) gene region with body mass index (BMI) and risk of obesity was identified in multiple populations, making FTO the first locus unequivocally associated with adiposity. At the time, FTO was a gene of unknown function and it was not known whether these SNPs exerted their effect on adiposity by affecting FTO or neighboring genes. Therefore, this breakthrough association inspired a wealth of in silico, in vitro, and in vivo analyses in model organisms and humans to improve knowledge of FTO function. These studies suggested that FTO plays a role in controlling feeding behavior and energy expenditure. Here, we review the approaches taken that provide a blueprint for the study of other obesity-associated genes in the hope that this strategy will result in increased understanding of the biological mechanisms underlying body weight regulation.
Overweight and obesity, defined as body mass indices (BMI) >25 and >30, respectively, are associated with premature death through increased risk of many chronic diseases, including type 2 diabetes, cardiovascular disease and cancer . Over the last three decades, the prevalence of overweight and obesity have increased rapidly and the latest World Health Organization (WHO) estimates suggest that 1.6 billion adults (aged 15 years and over) were overweight and 400 million were obese in 2005. These figures are predicted to rise to 2.3 billion overweight and over 700 million obese adults by 2015 (http://www.who.int/mediacentre/factsheets/fs311/en/index.html). Obesity is therefore a major international public health threat and economic burden. Although environmental factors, such as little physical activity and over-eating, have driven the recent rise in the numbers of people who are overweight or obese, genetic factors are estimated to account for 40–90% of the population variation in BMI [2–4]. It is hoped that identifying the genetic factors underlying the heritable risk of obesity will contribute to our basic knowledge of the biology of energy balance, and might even highlight molecules and pathways that can be targeted for therapeutic intervention. In 2007, single nucleotide polymorphisms (SNPs) within the fat mass and obesity-associated gene (FTO) became the first to be associated reproducibly with human body mass. We review the implications of genetic association between SNPs in the FTO gene region and BMI in humans, the various studies undertaken and the challenges in progressing from a genetic association to new biological insight. We also provide an overview of possible future directions for research in this field.
Before 2007, despite huge efforts using genome-wide linkage studies and candidate gene association studies, no genetic variation had been unequivocally associated with BMI and risk of obesity in population studies . In recent years, however, genome-wide association studies (GWAS), which test the correlation between SNPs across the entire genome and trait variation in a sample of individuals, have succeeded in identifying variants associated reproducibly with complex traits. GWAS for type 2 diabetes (T2D) detected strong association between common SNPs in the FTO region and risk of T2D [6–8]. However, subsequent analyses showed that the association between FTO SNPs and T2D was mediated by an association with BMI . The association between FTO SNPs and BMI and the risk of being overweight or obese has been confirmed in multiple populations [6,7,9–27]. The effect of FTO SNPs on BMI is modest, with those individuals homozygous for the risk allele weighing, on average, 3 kg more than those homozygous for the protective allele .
Association does not necessarily mean causation. BMI-associated SNPs lie within a 47 kilobase (kb) linkage disequilibrium (LD) block encompassing parts of the first two introns as well as exon 2 of FTO. Thus, the association signal could be due to correlation between FTO intronic SNPs and variation elsewhere in the gene or control elements of other genes. Indeed, the transcription start site of the retinitis pigmentosa GTPase regulator-interacting protein 1-like (RPGRIP1L) gene (homolog of murine Rpgrip1l, also known as Ftm) is in close proximity (~400 bp) to the 5’ end of FTO . It is therefore possible that SNPs in FTO are associated with obesity through an effect on RPGRIP1L. Efforts to fine-map the association signal to causative variant(s) are underway (Box 1).
Whereas linkage disequilibrium (LD) aids the discovery of genetic associations, because the investigator does not need to test all variant(s) in a given genomic region to detect association signals, it is a hindrance to fine-mapping efforts. This is because high correlation between SNPs makes it difficult to distinguish between the effects of true causal variants and correlated non-causal variants. Consequently, it is not clear which causal variant(s) underlie the association between BMI and SNPs in FTO (which reside within a 47 kb LD block). So far, fine-mapping attempts have focused on selecting tagging SNPs within the associated genomic region and genotyping or imputing these in additional study populations of different ancestry [6,9,15,19,25,26]. The advantage of using study cohorts of different ancestry lies in their varying patterns of LD. For example, LD is generally weaker in populations of African descent. Two recent studies attempted to refine the association signal from FTO in individuals of African ancestry, but none of the SNPs tested was associated with obesity in both cohorts [6,15]. Fine-mapping studies have been done with Japanese  and Old Amish Order participants . Some of the SNPs associated with BMI in these populations have overlapped with associated SNPs in Europeans but the SNPs most strongly associated with BMI vary between cohorts. For example, the strongest SNP (rs1861868) in the Old Amish Order study had not been previously associated with obesity. This might reflect interstudy differences in the causal variant(s) or differences in levels of LD between tested SNPs and the same causal variant(s). It is worth mentioning that in a European fine-mapping study cohort, three of the four most significantly associated SNPs (rs17817449, rs3751812 and rs1421085) were in putative functional regions (based on conservation and predicted impact on regulatory domains) but these SNPs were highly correlated with each other and their effects could not be distinguished . Interestingly, despite an earlier demonstration that associated SNPs in the FTO region were likely to be tagging the same signal rather than multiple, independent signals , Tonjes et al. discovered an independent association with BMI from SNPs in FTO introns 2 and 3, downstream of the original signal, in Sorbian individuals . This suggests that this genomic region might harbor multiple variants that influence susceptibility to obesity, which might complicate future fine-mapping attempts.
Since the discovery of the FTO signal, additional GWAS have succeeded in identifying many additional novel obesity loci [12,28–33]. Before these findings can be converted into clinical benefits, however, it is necessary to determine the biological mechanisms by which confirmed obesity susceptibility variants impact BMI.
When associations between SNPs at FTO and BMI were first discovered [6,7,9], little was known about the function of the FTO gene product. Fto was first identified in the mouse by positional cloning  as one of the genes within a 1.6 megabase deletion on chromosome 8 responsible for the Fused toes (Ft) phenotype . Mice homozygous for the deletion died mid-gestation and exhibited severe malformations of the head and face, central nervous system (CNS) developmental defects [36,37], randomized left–right asymmetry , polydactyly and growth retardation. Heterozygotes displayed fused toes and enlargement of the thymus [34,35]. However, no obesity or thinness was reported in these mice. Nevertheless, as Fto was only one of six genes deleted in the Ft mouse (the others were the Iroquois B cluster of genes, Irx3, Irx5 and Irx6, and two other genes, Fts and Rpgrip1l), it was not known which, if any, of these phenotypic manifestations were due to Fto deficiency. Interestingly, a phenotype similar to that of Ft mice was seen in a human patient harboring a small chromosomal duplication on 16q12.2, a region that includes the FTO gene .
Clues from the Ft mouse and human chromosomal duplication were not sufficient to shed light on the specific effects of FTO. Moreover, FTO was not annotated in the public databases as having homology to any other known gene. The discovery that this mysterious gene was associated with human obesity [6,7,9] inspired further research aimed at elucidating its functional properties. In silico analyses of the human FTO sequence revealed homologs in other vertebrates (from fish to mammals) and marine algae (from unicellular photosynthetic picoplankton to multicellular seaweed) [40–43]. Sequence analysis showed that FTO shares features with Fe (II) and 2-oxoglutarate (2OG) oxygenases [41,43]. These enzymes catalyze oxidative reactions on multiple substrates using non-heme iron as a co-factor and 2OG as a co-substrate [41,43]. Within this superfamily, FTO is most similar to the Escherichia coli enzyme AlkB and its eukaryotic homologs, which can repair DNA methylation damage by hydroxylating methyl groups on the DNA leading to their removal . These data suggest that FTO might act as a demethylase. Sequence analysis also predicted that human FTO and its vertebrate homologs are globular proteins that carry a nuclear localization signal and are unlikely to be targeted to membranes or organelles [42,43]. This prediction was confirmed by in vitro studies, which showed that murine Fto is indeed a 2OG oxygenase that can catalyse nucleic acid demethylation . It is conceivable, therefore, that the nucleic acid demethylation activity of FTO might regulate the expression of genes involved in metabolism and that dysregulation of this process might lead to obesity.
Caution is always recommended when interpreting gene function from in vitro studies, as the activity of molecules might differ in whole organisms and under different conditions: therefore, animal models are used to further elucidate in vivo function.
Early studies of mouse Fto and human FTO mRNA expression showed that both are ubiquitous, with particularly high levels of expression in the brain and hypothalamus [7,9,34]. These were intriguing results because the hypothalamus is a key site for regulation of energy balance, and genes responsible for monogenic obesity function in the hypothalamus to regulate appetite . Further studies of Fto mRNA expression in wild-type rodent tissues confirmed its ubiquitous expression, with high levels of expression in hypothalamic regions known to play important roles in the regulation of energy intake and expenditure, and suggested that expression of Fto might be regulated by nutritional status [40,41]. In the fasted state (that is, when there is a strong stimulus for eating), mice exhibit a significant reduction in hypothalamic Fto mRNA expression compared to fed controls. This effect is not rescued by supplementation with the anti-starvation hormone leptin [41,46], which suggests that the reduced hypothalamic Fto expression observed during fasting is independent of leptin levels. In support of this conclusion, hypothalamic Fto expression is reduced in fasted Lepob mice (which lack leptin) compared to fed Lepob and control mice . These studies suggest that Fto is downregulated during fasting and upregulated during feeding, and that variation in Fto resulting in decreased expression or activity might provide a signal that promotes feeding and obesity.
In contrast to the mouse data, Fto expression was increased significantly in the hypothalamus of food-deprived and food-restricted rats . Two possible reasons have been suggested to explain this discrepancy: different sensitivity to starvation in mice and rats, or different times at which samples were taken in the two studies. In rats, in addition to hypothalamic expression, Fto was widely and consistently expressed in brain regions related to circadian rhythmicity  and therefore its expression might vary at different times of day. Upregulation of hypothalamic Fto in fasted rats suggests that high levels of Fto protein might stimulate food intake, which is the opposite effect predicted from its expression in wild type mice . More recent data in rats disagree with this study, finding that over-expression of Fto mRNA in the hypothalamus decreased food intake, whereas a 40% decrease in Fto protein led to increased food consumption . These data might contrast with previous data in rats because the measurements and manipulation of Fto were limited to a specific region of the hypothalamus (the arcuate nucleus), where the expression of Fto is very high. To date, all studies of rodents have been done over prolonged periods of fasting, which might better resemble starvation rather than fasting and hence some of the observations might reflect that. There is a need for studies with shorter fasting times in rodents to better understand the regulation of Fto expression during this time.
Given that variants in FTO are associated with human obesity and that Fto mRNA levels appear to be regulated in response to feeding and fasting, its expression was determined in relation to changes in obesity. In six mouse models of obesity (Ay, Lepob, Leprdb, Cepfat, tub and mice with diet-induced obesity (DIO); Glossary), hypothalamic Fto expression did not differ significantly from that in wild type mice . This study assessed Fto gene expression in other metabolically relevant tissues, such as fat (adipose) tissue. Adipose tissue acts as a fat store and synthesizes and secretes a variety of proteins (such as leptin) that influence appetite and metabolism at distal sites. Expression of Fto in mesenteric fat was reduced significantly compared to wild type in all mouse models except DIO. This is interesting because murine Fto gene expression was shown to be downregulated under fasting conditions, suggesting that obese mouse models mimic the fasted state, possibly contributing to their over-eating.
Fto expression differs under feeding and fasting conditions and displays tissue-specific differences in mouse models of obesity, but it is not known whether these differences are the cause or the consequence of obesity. To further investigate whether differences in Fto expression or function can cause increases or decreases in fat mass, mice harboring Fto mutations were generated.
Two mouse models of Fto deficiency have been reported: a null mutation (Fto−/−) resulting in the complete absence of Fto protein expression , and a partial loss-of-function mutation with reduced Fto protein levels (Table 1) . The partial loss-of-function Fto mouse model carries a point mutation resulting in a change in the amino acid sequence from isoleucine to phenylalanine at position 367 (I367F). Although this residue is outside the catalytic core, it is located within a ~20 amino acid block conserved throughout vertebrates that defines a new functional domain . In vitro experiments demonstrated that, although the full-length I367F protein was correctly localized to the nucleus, it displayed lower levels of expression in mammalian cells and resulted in reduced catalytic activity, possibly through altering the Fto dimerization state. However, the exact role of different Fto dimerization states in energy balance is poorly understood. The mutant Fto I367F protein retains partial function, which is likely to account for some of the observed differences between the phenotypes of these two animal models. Namely, although both models demonstrated reduced body weight and fat mass, this starts early in life in the case of Fto−/− mice, whereas weight reduction in FtoI367F mice has a maturity-onset. Also, there is no discernible phenotype in the Fto null heterozygous mice, whereas the phenotype of the heterozygous FtoI367F is very similar to that of their homozygous mutant littermates. The importance of I367 in dimerization might help explain why the FtoI367F heterozygous mice have a phenotype: the I367F substitution might have a dominant negative effect by disrupting the function of wild type Fto through formation of heterodimers. The weight reduction observed in the two models is also very different: Fto null mice have a 30–40% weight reduction compared to wild type littermates, whereas the FtoI367F mutants have only a 10% reduction in weight. Another significant difference between the two models is that only Fto−/− mice exhibited growth retardation and early perinatal death [48,49]. On a high-fat diet, both models exhibited reduced weight gain and reduced white adipose tissue compared to controls [48,49]. These findings indicate that disruption of Fto activity can protect against diet-induced obesity.
These studies then investigated whether reduced growth and adiposity in the Fto mouse models is due to decreased energy intake, increased energy expenditure or both. Compared to wild type littermate controls, there was no difference in absolute food intake in Fto-deficient mouse models [48,49]. However, Fto−/− mice ate relatively more given their reduced body weight and size . These results demonstrated that reduced fat mass in Fto-deficient mice was not due to reduced food intake. However, both models exhibited higher levels of energy expenditure (higher metabolic rate) such that overall the animals gained less weight and were protected from obesity [48,49]. The increase in energy expenditure was unrelated to levels of physical activity (indeed, paradoxically there was a reduction in locomotor activity in Fto null mice but this was not observed in the FtoI367F mutant) but was potentially mediated by increased sympathetic nervous system (SNS) activity, a system that originates in the spinal cord with projections to peripheral tissues and is used by the hypothalamus to regulate energy homeostasis [48,49]. This increased SNS activity might be promoting lipolysis and thermogenesis in adipose tissue and muscle.
Microarray analysis of white adipose tissue, liver and skeletal muscle detected some differences in gene expression in FtoI367F mutants compared to wild type mice . As expected, given the known link between excess adipose tissue and induction of inflammatory responses, in mutant mice with reduced white adipose tissue mass, expression of multiple genes involved in inflammation was downregulated in adipose tissue. In contrast, upregulation in adipose tissue of some genes involved in fatty acid catabolism might explain, in part, the lower fat mass. Upregulation of genes involved in fatty acid synthesis in adipose tissue and upregulation of fatty acid synthase and genes involved in carbohydrate metabolism in muscle might reflect secondary adaptations to a lower supply of fatty acid (due to smaller fat reserves). Indeed, FtoI367F mutants demonstrated increased carbohydrate metabolism relative to fat metabolism. Hypothalamic neuropeptide expression did not differ between wild type and mutant in the fasted state, but Npy expression was lower in fed mutant mice. Given that increased Npy stimulates food intake, this finding suggests that Fto-deficient mice are more sensitive to satiety and are thereby protected from obesity through over-eating. A blunting of Npy induction was seen in Fto−/− mice in the fasted state, lending support to the suggestion that Fto promotes Npy expression . It is conceivable that the nucleic acid demethylation activity of Fto might provide a mechanism through which it affects expression of these other genes.
The phenotypic characterization of both the Fto−/− and FtoI367F mice support the idea that association of FTO SNPs with human obesity arises via regulatory or functional effects on FTO rather than other genes in the region. Although the Fto null mutation could impact regulation of other genes, the non-synonymous mutation in FtoI367F mice is most likely to exert its functional effects on Fto alone . Furthermore, both models suggest that alleles associated with increased risk of obesity will cause up- or dysregulation of FTO and that inhibition of FTO might protect against obesity [48,49].
Although animal models are useful tools to investigate gene function, they do not necessarily reflect the situation in humans. Recently, for example, a loss-of-function non-synonymous mutation resulting in an arginine to glutamine change at position 316 in FTO was reported to segregate with an autosomal recessive disease in a large Palestinian Arab consanguineous family with nine affected members . The syndrome includes post-natal growth retardation, head and face dysmorphism, severe psychomotor delay, functional brain deficits and, in some patients, brain malformations, cardiac defects, genital abnormalities and cleft palate. These developmental characteristics seem more similar to the Ft mouse than either of the Fto mouse models. This discrepancy between the effects of human and mouse mutations (Table 1) might reflect differences between the actions of FTO in humans and mice or, alternatively, multiple roles for FTO that are disparately affected by the mouse and human mutations. This is an intriguing area of investigation for the future and results from different animal models (e.g. transgenic Fto over-expression in mice) are eagerly anticipated. The lack of clinical obesity in patients carrying R316Q and family members is perhaps consistent with the relationship between loss of Fto function and leanness in mice. More recently, however, loss-of-function heterozygous mutations in FTO were found in both lean and obese humans, suggesting that haploinsufficiency does not protect against obesity completely, though the relationship might be complicated by other obesity-promoting factors . Other differences between the effects of mouse Fto mutations and human FTO SNPs (Table 1) are perhaps less surprising because FTO SNPs in the human population are likely not strong loss-of-function variants, particularly as the most strongly associated SNPs are intronic and therefore postulated to have an effect on gene regulation.
Despite recent progress, the mechanism by which SNPs in FTO influence human body mass remains elusive. Multiple processes could plausibly contribute to the risk of obesity, including neurological circuits governing appetite and whole-body energy expenditure, as well as peripheral pathways involved in energy expenditure. Loss-of-Fto function appears to reduce fat mass in mice, at least in part, through increased energy expenditure but not decreased energy intake [48,49]. However, the study of intermediate phenotypes in humans showed that FTO SNPs are associated with appetite and food intake but not energy expenditure (Table 1). Interestingly, data from rodents suggested that Fto might affect neuropeptide Y expression in the hypothalamus, which in turn is known to impact feeding behavior. An investigation of the association between FTO SNPs/expression and neuropeptide levels in human hypothalamus might therefore provide a mechanism for the modulatory effect of FTO SNPs on appetite, although it would be challenging to obtain human hypothalamic material. Another potential area of future investigation is the role of FTO in circadian rhythms, given its expression in relevant brain regions in rats . Aberrant circadian rhythms have been linked to metabolic disease and obesity .
At present, the strongest associations between FTO SNPs and BMI belong to intronic SNPs, which might have a role in the regulation of FTO and/or nearby genes. It is important to note, however, that associated SNPs are not necessarily the causal SNPs underlying the association. It is possible that fine-mapping the causal variant(s) could shed light on the biological mechanism impacting body mass, for example by determining whether the causal variant(s) change the amino acid sequence of FTO or whether they lie within a control element affecting expression of other genes in the region, such as RPGRIP1L. However, fine-mapping the association signal might be difficult because the obesity-associated SNPs lie within a 47 kb LD block in which the effects of causal variant(s) could be indistinguishable from other, highly correlated proxies. Under these circumstances it might prove more important to understand the biological effect of the risk haplotype (rather than the causal variants themselves) on genes and pathways. FTO obesity-associated SNPs do not appear to influence FTO expression but could be investigated for their effect on expression of its neighboring gene RPGRIP1L. Nevertheless, there is now some evidence to suggest that the associations with BMI are indeed mediated through FTO (Box 2); however, joint effects via FTO and RPGRIP1L, or other genes, have not been dismissed completely. A recent report suggested that the obesity-associated FTO region contains highly conserved non-coding elements that might be controlling expression of IRX3 (encoding a transcription factor important in development). Though none of the obesity-associated SNPs fall within these elements, the possibility that the association signal results from correlation with causal variant(s) within these regions has not been excluded .
Given the close proximity between FTO and RPGRIP1L, it is possible that SNPs in FTO are associated with obesity through a direct regulatory effect on RPGRIP1L or an indirect effect (through linkage disequilibrium with RPGRIP1L or with RPGRIP1L regulatory site variation) on RPGRIP1L function. For example, the obesity-associated FTO SNP, rs8050136, lies within a Cut-like 1 (CUTL1) transcription factor binding site. CUTL1 preferentially binds the rs8050136 allele associated with protection against obesity . Short interfering (si) RNA-mediated knockdown of CUTL1 in fibroblasts decreased both FTO and RPGRIP1L expression . RPGRIP1L is located in the basal body of cilia, where it regulates sonic hedgehog (SHH) signaling . Ciliary defects have been associated with obesity in mice and humans [62,63]; indeed, the disruption of genes involved in cilia formation or maintenance leads to obesity [62,63]. It is therefore possible that human obesity is influenced by FTO SNPs via an effect on RPGRIP1L expression and ciliary function. However, several pieces of evidence suggest that this association is driven by a direct effect on FTO:
It is clear that the journey undertaken by FTO has generated many questions as well as answers. Perhaps the most important lesson from the extensive work on FTO over the last three years is the promise that other obesity loci identified by GWAS could lead to new understanding of the underlying biology of human adiposity. Despite the small effect sizes and predictive value of the common variants identified to date, these hits often represent previously unsuspected obesity genes. The story of FTO research provides valuable lessons for the study of other obesity-associated genes. Ultimately, it is hoped that knowledge of susceptibility gene function will highlight novel genes and molecules that could be targeted for therapeutic intervention. For example, the data presented in this review indicate that FTO deficiency protects against obesity. However, the devastating effects of a loss-of-function non-synonymous mutation in FTO in a human family emphasizes that care should be taken in developing drugs to alleviate obesity by reducing the expression or inactivating catalytic activity of FTO . Clinical and translational effects should not be expected in the near future but GWAS and subsequent investigation of obesity genes promises exciting new insights into common obesity in the next few years.
Inês Barroso acknowledges funding from the Wellcome Trust (grant 077016/Z/05/Z) and the Cambridge NIHR Comprehensive Biomedical Research Centre. We thank Dr Andrew Nelson for helpful comments on the draft manuscript.