FTO effect on energy demand versus food intake

Abstract

Arising from: J. Fischer et al. Nature 458, 894–898 (2009)10.1038/nature07848; Fischer et al. reply

An intronic single nucleotide polymorphism (SNP) (rs9939609) close to the fat mass and obesity associated gene (FTO) was the first SNP to be discovered with common variants linked to body mass index¹; at least seven studies in humans have implicated this SNP with variations in food intake and satiety^{2,3,4,5,6,7,8}, and four studies have rejected an effect on energy expenditure normalized for body weight^2,5,6,8. Fischer et al.⁹ recently constructed a mouse in which the homologous Fto gene was inactivated (Fto^-/-) and showed that these mice were protected from obesity. This observation strongly implicates the effects of the intronic SNP rs9939609 as arising due to an effect on the closest gene (FTO). However, the suggested mechanism underlying this effect in mice was opposite to that in humans. The Fto^-/- mice showed no significant differences in food intake relative to wild-types litter-mates⁹ but had an elevated metabolic rate. The apparent contrasting effects of the gene in humans and mice is worthy of closer investigation.

Main

The difference in body fatness between Fto^-/- and Fto^+/? mice was about 3.5 g in males and 1 g in females at 20 weeks old (Fig. 3 in ref. 9). One gram of fat is equal to about 39 kJ of energy. Assuming the mice weaned at 3 weeks of age, then over 17 weeks (120 days) the energy imbalance necessary to generate this difference is about 0.33 kJ day^-1 for females and about 1.16 kJ day^-1 for males. The energy density of standard rodent chow is about 17 kJ g^-1. Hence the difference in food intake between genotypes required to generate the observed differences in body fatness was about 0.02 g day^-1 in the females and 0.07 g day^-1 in males. Food intake in rodents is potentially much easier to study than in humans because food intake in rodents can be measured accurately, they can be fed from a single food source removing the influence of differences in macronutrient composition and they can be monitored for much longer periods—exceptionally over their entire lives. Nevertheless, given that individual variability in food intake in mice has a standard deviation of about 0.3 g day^-1 then the required sample size to detect these effect sizes, using a 3 level one-way ANOVA (as analysed in ref. 9) with a power of 80% and α = 0.05 is 355 per group for males, and 4,337 for females. Fischer et al.⁹ only presented data for food intake in female mice. They used a total sample of 31 female animals. The post hoc estimated power to detect the effect size of 0.02 g day^-1 in food intake was only 5.1%. Many studies of the effects of genes on food intake are underpowered and this critique could have been levelled at any number of recent studies. An example of a correctly powered study is that of ref. 10 on the effects of insulin receptor substrate-1 (Irs1) null mice. In that study using a sample of 12 null and 12 wild-type mice it was shown that the effect of disruption of Irs1 on longevity did not come about because of an effect of the inactivation of Irs1 on food intake. Although the sample size in this latter study may not seem very different from that used by Fischer et al.⁹, the key difference between the studies is the magnitude of the effect size that is being detected. In ref. 10 the effect size is 50× greater than that being detected by Fischer et al.⁹, consequently the power to detect or reject this effect was 98%. Sample sizes in experiments like those conducted in ref. 9 need to be sufficient to detect the effect size that generates the observed difference in fatness. Clearly this experiment was underpowered and the rejection of an effect of the Fto genotype on food intake has a strong likelihood of being a type II error.

The second potential issue with the findings of Fischer et al.⁹ relates to the estimated energy expenditure. First, there is the same power issue highlighted above with respect to food intake, except that there is even less power in the expenditure measures as the total sample is only 23 individuals. More importantly, to evaluate the role of expenditure differences Fischer et al. (Fig. 4 in ref. 9) made a simple division of the metabolism by the lean body mass (LBM). This is a common practice to attempt to normalize for body size effects. However, simple division by LBM can generate a spurious elevation of metabolic rate if the intercept of the relationship between metabolism and LBM is not zero¹¹, as is often the case. The suggested increased metabolism in the Fto^-/- animals is potentially an artefact of using this analysis method. The metabolic rates of these animals may only seem higher because the expenditure has been divided by a smaller lean body mass. If these data had been analysed using ANCOVA¹¹ the effect of genotype would very probably disappear (but then the absence of an effect could also be a type II error because of the power issue).

Overall the construction of the Fto^-/- mouse is a great achievement that identifies the FTO gene as the prime candidate being affected by the intronic rs9939609 SNP. Given the roles of many genes on both intake and expenditure it is entirely possible that the mechanisms by which Fto influences energy balance in the mouse, as claimed by Fischer et al.⁹, really do contrast the effects in humans^{2,3,4,5,6,7,8}. Unfortunately the issue of lack of power and the complexity of normalization of energy expenditure measurements mean that at present it is impossible to judge.