Physical Activity and Insulin Sensitivity Independently Attenuate the Effect of FTO rs9939609 on Obesity

The prevalence of obesity is high and continuously rising at worrisome rates worldwide. This is primarily due to environmental factors; however, genetic predisposition also plays an important role for an individual’s risk of obesity. To date, more than 900 genome-wide significant variants have been robustly associated with BMI (1).

FTO (fat mass and obesity associated) was the first locus where variants were firmly established to be associated with obesity (2), and FTO variants have the largest effect size for obesity reported for single common variants (1). These obesity-associated variants are located within intron 1 of FTO, and therefore likely affect transcriptional regulation rather than protein structure. It is, however, unclear which variant is causal in the FTO region, and through which gene the causal variant affects obesity. The region harbors several interesting candidate genes, mainly FTO, IRX3, IRX5, and RPGRIP1L, which could mediate the obesity effect. The expression of these genes has been shown to be affected by FTO variants primarily in either brain or adipose tissue (3–6). Moreover, mouse models have shown that RPGRIP1L knockout is associated with obesity (7), whereas FTO, IRX3, or IRX5 knockout is associated with lower body weight (4,8,9).

Interestingly, the effect of FTO variants on obesity is modified by several factors, including physical activity (PA) and insulin sensitivity (IS) (10,11). In particular, PA has been established to attenuate the effect of FTO variants on obesity (10–12); hence, the attenuation of the effect in physically active individuals was reported to be up to 30% (12). Alongside the interaction with PA, an equally strong interaction was reported for IS (11). These interactions could represent the same underlying interaction, as it is expected that active individuals are more sensitive to insulin. However, these interactions may also be independent, but this has, to date, not been tested. If independent, it would be relevant to include both factors in analyses assessing the association between FTO rs9939609 and relevant cardiometabolic phenotypes to fully understand the genomic mechanisms by which rs9939609 is linked to body weight.

Obesity is a major risk factor for increased mortality and for a range of comorbidities (13). In line with this, obesity-associated variants in FTO, either alone or combined with other variants in genetic risk scores, have been shown to be associated with obesity-related comorbidities, including type 2 diabetes, hypertension, ischemic heart disease, ischemic stroke, and peripheral artery disease (14), which is supported by large-scale genetic correlation studies (15). The FTO rs9939609 obesity-associated A allele has also been suggested to be associated with increased all-cause mortality independently of BMI (16), but this finding was not supported by a large meta-analysis (17).

To further elucidate the mechanisms underlying the link between variation in FTO and obesity, we aim to confirm the suggested interaction between FTO rs9939609 and IS on BMI and to assess whether this interaction is independent of the established PA interaction. Moreover, we aim to assess whether PA and/or IS modify the relationship between FTO rs9939609 and type 2 diabetes, cardiovascular disease outcomes, or all-cause mortality. Finally, we aim to elucidate the mechanisms underlying the observed associations by assessing the effect of the FTO intron 1 variant rs9939609 on RNA expression of nearby candidate genes in skeletal muscle tissue, and by assessing the possible physical interaction between the FTO intron 1 enhancer encompassing rs9939609 and the promoters of nearby genes in skeletal muscle cells.

For the association and interaction analyses, we assessed a cohort comprising up to 19,585 individuals. These individuals were collected as part of the population-based cohorts Danish study of Functional Disorders (DanFunD) (n = 6,657; collected 2012–2015), Intervention 1999 (Inter99) (n = 5,240; 1999–2001), Health2006 (n = 2,705; 2006–2008), monitoring trends and determinants in cardiovascular disease (Monica 10) (n = 2,391; 1993–1994), Health2010 (n = 1,311; 2010–2011), and Health2008 (n = 639; 2008–2009), as well as from the 1936 birth cohort (1936 BC) (n = 642; 1996–1997). For the functional analyses, we included a subset of men with muscle biopsy samples collected as part of the Adiposity Genetics (ADIGEN) study, comprising a total of 557 men originally selected either with BMI ≥ 31 kg/m² or at random from the population constituted by the draft board examination. All participants gave informed consent. The studies were approved by the Capital Regional Ethics Committee, Denmark (H-3-2011-081, H-3-2012-0015, KA 98155, KA 93054, KA 04130, KA-20060011, H-KA-20060011, H-4-2009-124, and KA 96008) and conducted according to the Declaration of Helsinki.

BMI was calculated based on measures of weight and height. Fasting serum insulin and fasting plasma glucose concentrations were measured by standard automated methods. In all study samples, PA was dichotomized into high and low based on five categories of self-reported work and leisure time activity. The low-PA group comprised individuals categorized as physically passive, and the high-PA group comprised individuals categorized as light, medium, hard, or very hard physically active. IS was defined based on the HOMA insulin resistance index ([fasting serum insulin × fasting plasma glucose]/22.5). To facilitate comparison of effect sizes, we converted the HOMA insulin resistance index to an index of IS (HOMA-IS) defined as 100 – HOMA insulin resistance index. HOMA-IS was dichotomized based on tertiles; the lowest tertile was categorized as low IS, and the two highest tertiles as high IS in each study sample, separately. All-cause mortality was defined based on data from the cause-of-death register and the central person register, and cardiovascular disease events and type 2 diabetes were defined based on ICD-8 and ICD-10 codes retrieved from patient registers (https://www.finngen.fi/en/researchers/clinical-endpoints) (Supplementary Table 1).

For DanFunD, Inter99, Health2006, Health2010, Health2008, and 1936 BC individuals, FTO rs9939609 genotype information was extracted from Infinium OmniExpress-24 v1.3 Chip (Illumina Inc., San Diego, CA) data, while Monica 10 genotypes were from Infinium Global Screening Array 24 v2.0 (Illumina), and ADIGEN genotypes were from the CoreExome Chip v1.0 (Illumina). All data sets were subjected to standard quality control.

Biopsies were taken at a follow-up visit from 71 obese (BMI ≥31 kg/m² at the time of draft board examination) and 74 age-matched nonobese individuals (BMI <31 kg/m² at the time of draft board examination) from the ADIGEN cohort; of these, we had rs9939609 genotype data available for 140 (obese, n = 67; nonobese, n = 73). The biopsies were taken from the vastus lateralis muscle in the right thigh using a thin Bergström needle and were snap frozen in liquid nitrogen. RNA was extracted, and expression of ∼47,000 transcripts was measured using the HumanExpression HT-12 Chip (Illumina) as previously described (18). The array preprocessing was performed in R with the lumi package (19) and included quantile normalization, log₂ transformation, and removal of undetected probes (Th < 0.01). We were interested in assessing the expression of the candidate genes, IRX3, IRX5, FTO, and RPGRIP1L; however, RPGRIP1L was eliminated in the quality control.

Cis-regulatory activity was measured using H3K27ac and H3K4me1 histone modifications for active regulatory regions in chromatin immunoprecipitation sequencing data from human skeletal muscle–derived myotubes (GSE126099). Physical interactions between promoters and putative enhancer elements were measured using Promoter Capture Hi-C from myotubes (18), using interactions with an enrichment score of >2. Data were plotted in relation to University of California Santa Cruz Genome Browser meta genes on hg38 using tidyGenomeBrowser (https://github.com/MalteThodberg/tidyGenomeBrowser).

We applied linear regression for association analyses between FTO rs9939609 and quantitative traits, and logistic regression for dichotomous traits. We calculated Pearson correlation coefficients across the five groups of PA and continuous levels of HOMA-IS to assess the correlation between these traits and between each of them and BMI. We used basic two-way interaction models for PA and IS separately to estimate their respective effect modification performance on the FTO genotype association with BMI, a joint two-way interaction model, to test whether the effect modifications were independent, and a three-way interaction model to further test the relationship between the interactions. Analyses of all-cause mortality and cardiometabolic outcomes were run as Cox regressions using age as the underlying timescale and FTO rs9939609 as exposure. The final censoring date was 31 December 2017, leading to these follow-up intervals for each cohort: DanFunD, 2.5–5.2 years; Inter99, 16.9–18.8 years; Health2006, 9.6–11.6 years; Monica 10, 23.1–24.5 years; Health2010, 6.2–7.9 years; Health2008, 8.0–9.3 years; and 1936 BC, 20.3–21.7 years. For the significantly associated outcomes, we tested whether PA or IS modified the effect of rs9939609 on these outcomes. For all analyses, effect estimates and P values were obtained with an additive genetic model adjusted for sex, age at examination as a restricted cubic spline with four knots, and cohort, as well as with and without adjustment for BMI as a restricted cubic spline with four knots.

For analyses of RNA expression according to FTO rs9939609 genotype, as well as PA and IS group, we applied Matrix eQTL (20) (R version 3.5.0) using linear regression adjusted for age, BMI at examination categorized as normal weight, overweight, or obesity, and 15 probabilistic estimations of expression residual factors to account for complex nongenetic factors (21).

All summary data generated or analyzed during this study are included in the published article (and its online supplementary files). Restrictions apply to the individual-level data that support the findings of this study, which are therefore not publicly available. Data are available upon reasonable request, and with relevant permissions, from A.L. and T.H.

The FTO rs9939609 A allele was strongly associated with higher BMI (effect per A allele [95% CI], β = 0.45 [0.36, 0.54] kg/m², P = 5.6 × 10⁻²³), more modestly with lower IS (β = −0.06 [−0.10, −0.01], P = 0.0094) but not with PA (odds ratio [OR] = 1.01 [0.96, 1.05], P = 0.770). When adjusting for BMI, the association with IS was no longer significant (Table 1).

We tested for correlation between PA, IS, and BMI. Between PA and IS, correlation was low (Pearson correlation coefficient, 0.09), hence allowing largely independent interactions of these variables. Moreover, the correlation between BMI and PA was also low (−0.07), whereas the correlation between BMI and IS was modest (−0.43).

We observed an interaction between self-reported PA and FTO rs9939609 on levels of BMI. In individuals belonging to the low-PA group, the estimated per-allele effect (SE) was β = 0.68 (0.08) kg/m², and this effect was attenuated in individuals belonging to the high-PA group (β = −0.32 [0.10] kg/m², P = 0.0013) (Fig. 1 and Supplementary Table 2). This corresponded to a 47% smaller effect of the rs9939609 obesity-associated A allele on BMI in the high-PA group, compared with individuals in the low-PA group. Similarly, we observed an interaction between rs9939609 and IS, relative to BMI. In individuals with low IS, the estimated per-allele effect of rs9939609 was β = 0.61 (0.07) kg/m², and this effect was attenuated in individuals with high IS (β = −0.31 [0.09] kg/m², P = 0.00028) (Fig. 1 and Supplementary Table 2). This corresponded to a 51% smaller effect of the rs9939609 obesity-associated A allele on BMI in individuals with high IS, compared with individuals with low IS.

Next, we used a joint two-way interaction model, to assess whether the observed interactions between FTO rs9939609 and PA and IS, respectively, on BMI were independent or represented the same underlying interaction. This model allowed for independent effects of each of the interactions. For the group with low PA and low IS, the estimated per-allele effect of the FTO rs9939609 A allele was β = 0.73 (0.09) kg/m² in the joint two-way interaction model. This effect was attenuated by high PA with estimated β = −0.20 (0.09) kg/m² (P = 0.023), and by high IS with estimated β = −0.28 (0.09) kg/m² (P = 0.0011) (Fig. 1 and Supplementary Table 3). Hence, the interactions appeared largely independent. Moreover, no clear interaction pattern was observed when applying the three-way interaction model (Supplementary Table 4).

We assessed the effect of FTO rs9939609 on mortality and obesity-related comorbidities, where cardiometabolic disease outcomes were defined based on health register data. The FTO rs9939609 obesity-associated A allele was significantly associated with higher risk of all-cause mortality (hazard ratio [HR] [95% CI]; HR = 1.08 [1.01, 1.15], P = 0.019), hypertension (HR = 1.07 [1.01, 1.13], P = 0.019), ischemic stroke (HR = 1.11 [1.01, 1.23], P = 0.037), and type 2 diabetes (HR = 1.18 [1.07, 1.31], P = 0.0015) (Fig. 2 and Supplementary Table 5). When adjusting the analyses for BMI, the effect estimates for all-cause mortality and ischemic stroke remained similar and the associations significant, and thus appeared partly independent of BMI (Fig. 2 and Supplementary Table 5). We tested whether PA or IS modified the effect of rs9939609 on significantly associated outcomes. High PA was associated with a lower effect of rs9939609 on risk for all-cause mortality (HR = 0.95 [0.83, 1.10]), hypertension (HR = 0.91 [0.81, 1.02]), and type 2 diabetes (HR = 0.87 [0.70, 1.08]), albeit nonsignificant. Similarly, high IS was associated with a nonsignificant lower effect of rs9939609 on the risk of hypertension (HR = 0.95 [0.85, 1.06]) and type 2 diabetes (HR = 0.87 [0.68, 1.12]). Contrary to these observations, both high PA and IS were associated with a higher effect of rs9939609 on risk of ischemic stroke (PA, HR = 1.32 [1.05, 1.66], P = 0.019; IS, HR = 1.30 [1.05, 1.60], P = 0.015) (Supplementary Table 6).

To explore the potential mechanism underlying the association between variation in FTO and obesity, and how this might be modified by PA and IS, we assessed the effect of FTO rs9939609 genotype on RNA expression in skeletal muscle biopsies and on mapped enhancer-promoter interactions in the FTO region in muscle cells.

We assessed the expression of relevant genes in skeletal muscle biopsies from 140 men. Interestingly, we found that FTO expression was significantly affected by the FTO rs9939609 genotype (β = 0.03 [0.01], P = 0.011), with the highest expression in homozygous carriers of the obesity-associated A allele (Fig. 3). This association was significant even when adjusted for BMI, and thus appeared independent of BMI. In contrast, the FTO rs9939609 genotype had no significant impact on the expression of IRX3 (β = −0.07 [0.06], P = 0.254) or IRX5 (β = −0.005 [0.02], P = 0.819) in skeletal muscle (Fig. 3A). We were unable to assess the impact of FTO rs9939609 on the expression of RPGRIP1L in our data, because of low expression levels. These findings were supported by data from the Genotype-Tissue Expression (GTEx) portal, where we queried expression data across 49 tissues, and found that rs9939609 only affected the expression of FTO, with the same direction of effect, namely, with the highest FTO expression in homozygous A allele carriers, and only in skeletal muscle (normalized effect size [NES]; NES = 0.12, P = 4.2 × 10⁻⁶). Moreover, in GTEx data of muscle tissue, rs9939609 was not associated with altered expression of either IRX3 (NES = −0.05, P = 0.100), IRX5 (NES = 0.04, P = 0.380), or RPGRIP1L (NES = −0.01, P = 0.760). Also, in GTEx data, RPGRIP1L showed low expression in muscle. We tested whether FTO expression differed across groups of high and low PA and IS, respectively, to elucidate whether any of these factors could possibly counteract the effect of the rs9939609 genotype, but we were unable to detect any differences in FTO expression across the PA and IS groups (Supplementary Table 7). Chromatin conformation data from muscle cells showed that rs9939609 was located within a regulatory active region, which was in physical contact with the FTO promoter, suggesting that the region encompassing rs9939609 could act as an enhancer for the FTO gene in skeletal muscle (Fig. 3B). Of note, the enhancer also showed interaction with IRX3 and IRX5 (Supplementary Fig. 1).

In a study sample comprising up to 19,585 individuals, we confirmed previously reported interactions of PA and IS with the association of FTO rs9939609 with BMI. Interestingly, we found that the interactions were largely independent. Moreover, the obesity-associated FTO rs9939609 A allele was associated with increased risk of all-cause mortality, hypertension, ischemic stroke, and type 2 diabetes, but only the associations with all-cause mortality and ischemic stroke appeared partly independent of BMI. Effect estimates indicated that high PA might reduce the impact of rs9939609 on all-cause mortality, hypertension, and type 2 diabetes and that high IS might reduce the impact on hypertension and type 2 diabetes. Contrary to our hypothesis, high PA and high IS both seemed to increase the impact of rs9939609 on ischemic stroke. Gene expression data from human skeletal muscle biopsies, as well as data from the GTEx portal, indicated that the FTO rs9939609 variant might mediate its effect through altered expression of FTO in skeletal muscle. The functional impact of rs9939609 on FTO expression was supported by a physical interaction between an enhancer region encompassing rs9939609 and the FTO promoter in muscle cells.

We observed a large effect-modifying impact of PA and IS on the relationship between FTO rs9939609 and BMI. Hence, the effect of the obesity-associated rs9939609 A allele was 47% smaller in high-PA compared with low-PA individuals, and 51% smaller in individuals with high IS compared with individuals with low IS. For PA, this effect was larger than the up to 30% effect reduction previously reported in Europeans (10–12). While our results strongly supported the PA interaction, the observed effect size could be affected by the relatively smaller sample size and potential differences in how information on PA was obtained. With respect to IS, the observed effect modification for HOMA-IS was similar to that previously reported for BIGTT-Si (an oral glucose tolerance test–derived index of insulin sensitivity) in up to 5,554 individuals from the Inter99 cohort (11), also included in this study. Importantly, our findings indicated that IS and PA are equally strong modifying factors of the relationship between FTO rs9939609 and obesity. Hence, the two interactions were not just a reflection of inactive individuals being less sensitive to insulin, and therefore both PA and IS are relevant to include in FTO rs9939609 association analyses.

The independence of the interactions was also supported by only a weak correlation between PA and IS. Even though rs9939609, besides BMI, was also associated with HOMA-IS, this association should not be able to drive the observed interaction of IS with the rs9939609-BMI association. A strong correlation between BMI and either PA or IS could have caused some uncertainty of the effect estimates in the interaction analyses, but these correlations were only modest in our data set. Therefore, both interactions appeared robust.

The FTO rs9939609 obesity-associated A allele was associated with cardiometabolic outcomes, which is in line with previous observations for this variant alone or in combination with other BMI-associated variants (14,22,23). Moreover, except for all-cause mortality and ischemic stroke, these associations were not significant when adjusting for BMI, which could support the finding that BMI is causally linked to most of these cardiometabolic outcomes (14). However, it has also been reported that FTO-related cardiovascular disease outcomes and type 2 diabetes risk are at least partly independent of BMI (22–24). With respect to all-cause mortality, for which BMI has also been shown to be causally linked (25), we observed an association with FTO rs9939609 independent of BMI, with a larger effect size than previously reported (17). Possible BMI-independent effects of FTO could be mediated by its fundamental role during postnatal development and be accentuated by cumulative effects across the many tissues where FTO is expressed. We hypothesized that the observed associations with cardiometabolic traits and all-cause mortality, similarly to the BMI association, could be effect modified by PA and IS. For hypertension and type 2 diabetes, we observed nonsignificant attenuation of the association with rs9939609 for both PA and IS, and for all-cause mortality only for PA. Also, contrary to our hypothesis, both high PA and high IS seemed to accentuate the impact of rs9939609 on ischemic stroke. However, these effect modifications should be substantiated by additional studies with larger sample sizes and greater statistical power.

Our results also suggested that the obesity-associated rs9939609 A allele was associated with increased expression of FTO in muscle, apparently independent of BMI, which was supported by similar findings from GTEx data. In a previous study, differences in FTO expression in skeletal muscle according to FTO rs9939609 genotype were reported to be nonsignificant, but the highest expression was observed in homozygous carriers of the A allele (26). In other studies, FTO variants have been reported to be associated with altered expression of the nearby genes IRX3, IRX5, and RPGRIP1L (3–6). For IRX3 and IRX5, the altered expression mediated by FTO variants was reported in human brain cells, where it was shown to affect feeding behavior and food preference (27,28), and in preadipocytes, where it was shown to affect browning and storage of fat (6). For RPGRIP1L, the altered expression mediated by FTO variants was reported in brain, where it resulted in disrupted leptin signaling in primary cilia and, thus, altered satiety signaling (29,30). We observed no effect of rs9939609 on the expression of IRX3, IRX5, or RPGRIP1L in skeletal muscle tissue in our data or in data from the GTEx portal. Moreover, the RPGRIP1L expression in muscle, particularly in our data but also in GTEx data, was low, indicating that RPGRIP1L is unlikely to play a major functional role in skeletal muscles. Taken together, the FTO locus likely harbors several causal genes, mediating the association between variants in intron 1 of FTO and obesity, possibly including IRX3, IRX5, RPGRIP1L, and FTO. The effects of the causal genes might be pleiotropic across multiple tissues, likely with effects in the brain, preadipocytes, and muscle, and vary according to developmental stage, which is supported by both functional studies and genetic association studies (3–6,31). Our findings support the proposed hypothesis of genetic association signals being explained by several genetic enhancer variants, which, in combination, contributes to the regulation of gene expression across tissues (5,32). Additional studies, with larger sample sizes and across multiple tissues, are needed to further elucidate the genetic and mechanistic architecture of the FTO locus.

We hypothesized that PA and/or IS may counteract the obesity-increasing effect of the FTO rs9939609 variant, possibly by downregulation of FTO expression in muscle. However, we were unable to show any difference in FTO expression in groups stratified according to either PA or IS level. This could, however, be due to a limited sample size, or possible effects might be confounded because of the nature of the data set, originally collected as an obesity case-control sample rather than a population-based sample. Nonetheless, previous studies strongly support our hypothesis. High-intensity exercise significantly reduces the mRNA expression of FTO in skeletal muscle (33). Moreover, it was suggested that AMP-activated protein kinase (AMPK) was implicated in the downregulation of FTO expression largely independent of rs9939609 genotype (33). These findings are in line with the fact that exercise is a well-known mechanism for activation of AMPK in human muscles (34,35). In mouse muscle cell lines, AMPK activation leads to downregulation of FTO expression and activity, as well as lower lipid accumulation, whereas inhibition of AMPK leads to increased expression of FTO and increased lipid accumulation (36). Some studies have also suggested a link between lower IS and higher FTO expression in adipose tissue and muscle (37,38); however, other studies find no such link (39–41). The modifying effect of IS might also be mediated through AMPK, as it has been shown that AMPK activity is reduced in white adipose tissue of individuals with marked obesity, and insulin resistance, compared with weight-matched and insulin-sensitive individuals (42,43). Hence, these studies support the hypothesis that PA and maybe IS may affect FTO expression in skeletal muscle, and thereby counteract the obesity-inducing effects of the FTO rs9939609 variant.

Future studies could further elucidate the modification of the effect of FTO rs9939609, and address some of the limitations of the current study. These limitations included no explicit consideration of causality, and, therefore, interpretations were restricted to estimated cross-sectional associations being defined by the underlying regression models used. Another limitation was the lack of inclusion of other factors known to modify the effect of FTO rs9939609 on obesity, including alcohol consumption, sleep duration, and dietary composition. Moreover, PA was self-reported, and IS was estimated based on the HOMA insulin resistance index; therefore, studies with objectively measured PA and IS in a large sample size would be of great interest to confirm our findings.

In conclusion, the effect of FTO rs9939609 on obesity, and potentially also on some cardiometabolic traits, was modified by PA and IS independently. Our data support the notion that exercise and/or improved IS can counteract genetic predisposition not only to obesity but likely also to certain cardiometabolic outcomes.

Funding. This study was supported by The Novo Nordisk Foundation (grant number NNF18CC0034900), A.E.J. and M.K.A. were supported by The Danish Diabetes Academy (grant numbers PDMI002-18 and PDMI001-18), where funds were contributed by AstraZeneca, and S.E.S. was funded by The Novo Nordisk Foundation Copenhagen Bioscience PhD Programme (grant number NNF18CC0033668).

Duality of Interest. A.E.J. and M.K.A. received funding in the form of salary from AstraZeneca via The Danish Diabetes Academy. At the time of this study, E.L.O., J.G., and C.J.R. were employed by AstraZeneca and hold shares in the company. AstraZeneca had no involvement in either the collection, analysis, or interpretation of the data or the writing of the manuscript. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. M.K.A., L.Ä., C.H.S., and T.H. were involved in the conceptualization, design, and conduct of the study. E.L.O., N.G., A.A., O.P., K.W., R. B., T.I.A.S., A.L., J.G., C.J.R., and T.H. curated the data and supervised the study. M.K.A., L.Ä., J.B.-J., A.E.J., S.E.S., M.T., and L.M.P. did the formal analyses and interpreted the results. M.K.A. wrote the first draft of the manuscript, and all authors edited, reviewed, and approved the final version of the manuscript. M.K.A. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.