FTO genotype impacts food intake and corticolimbic activation

ABSTRACT

Background

Variants in the first intron of the fat mass and obesity-associated (FTO) gene increase obesity risk. People with “high-risk” FTO genotypes exhibit preference for high-fat foods, reduced satiety responsiveness, and greater food intake consistent with impaired satiety.

Objective

We sought central nervous system mechanisms that might underlie impaired satiety perception in people with a higher risk of obesity based on their FTO genotype.

Design

We performed a cross-sectional study in a sample that was enriched for obesity and included 20 higher-risk participants with the AA (risk) genotype at the rs9939609 locus of FTO and 94 lower-risk participants with either the AT or TT genotype. We compared subjective appetite, appetite-regulating hormones, caloric intake at a buffet meal, and brain response to visual food cues in an extended satiety network using functional MRI scans acquired before and after a standardized meal.

Results

Higher-risk participants reported less subjective fullness (χ² = 7.48, P < 0.01), rated calorie-dense food as more appealing (χ² = 3.92, P < 0.05), and consumed ∼350 more kilocalories than lower-risk participants (β = 348 kcal, P = 0.03), even after adjusting for fat or lean mass. Premeal, the higher-risk group had greater activation by “fattening” food images (compared with objects) in the medial orbital frontal cortex (β = 11.6; 95% CI: 1.5, 21.7; P < 0.05). Postmeal, the higher-risk subjects had greater activation by fattening (compared with nonfattening) food cues in the ventral tegmental area/substantia nigra (β = 12.8; 95% CI: 2.7, 23.0; P < 0.05), amygdala (β = 10.6; 95% CI: 0.7, 20.5; P < 0.05), and ventral striatum (β = 6.9; 95% CI: 0.2, 13.7; P < 0.05). Moreover, postmeal activation by fattening food cues within the preselected extended satiety network was positively associated with energy intake at the buffet meal (R² = 0.29, P = 0.04) and this relation was particularly strong in the dorsal striatum (R² = 0.28, P = 0.01), amygdala (R² = 0.28, P = 0.03), and ventral tegmental area/substantia nigra (R² = 0.27, P = 0.01).

Conclusion

The findings are consistent with a model in which allelic variants in FTO raise obesity risk through impaired central nervous system satiety processing, thereby increasing food intake. This study is registered at clinicaltrials.gov as NCT02483663.

INTRODUCTION

Genome-wide effects strongly influence appetitive behaviors (1–5), and single-gene hypomorphic mutations encoding elements of canonical pathways regulating energy intake and expenditure (e.g., LEP, LEPR, MC4R) can result in severe hyperphagia and early-onset obesity (6, 7). The more important single-gene influences at a population level, however, are specific single-nucleotide obesity-risk alleles within the first intron of the fat mass and obesity-associated (FTO) gene. Allelic variation in the first intron of FTO constitutes the most prevalent genetic association with obesity in many populations, with dose-related higher BMIs (in kg/m²) of ∼0.4 per allele (8). Behaviorally, people with “at-risk” FTO genotypes exhibit preference for high-fat foods (9), reduced satiety responsiveness (10), and greater food intake (11, 12), but not changes in energy expenditure or physical activity (11, 12).

Divergent central nervous system (CNS) mechanisms driving overconsumption in people with at-risk FTO genotypes have been suggested. fMRI studies have shown greater responsivity to food cues in the ventral striatum (nucleus accumbens) (13) and putamen (14), which could indicate heightened reward sensitivity or motivation for food intake, whereas findings of greater activation by food cues in the posterior fusiform gyrus suggest enhanced visual attention for food (15). In contrast, hyporesponsivity in the prefrontal cortex after a glucose load (16) and in the cingulate cortex, frontal gyri, and lenticular nucleus in higher BMI at-risk genotype carriers (14) imply reduced inhibitory or impulse control. Other studies have focused on dopamine signaling [which, in rodent studies, is regulated by FTO expression in midbrain dopaminergic neurons (17)] and suggest that reward learning may be altered, particularly when genetic polymorphisms that reduce dopamine signaling capacity co-occur (18). Central insulin sensitivity (19) and reduced postprandial suppression of acyl ghrelin (20) are potential hormonal links between high-risk FTO genotypes and impaired satiety. Although individually informative, no clear CNS signature of FTO-mediated obesity risk has emerged from this literature. Moreover, none of the aforementioned findings have been linked to food intake.

Using fMRI and measures of subjective appetite, appetite-regulating hormones, and food intake, we sought CNS mechanisms that might underlie impaired satiety in those with “high-risk” FTO genotypes. We proposed the following hypotheses: 1) that allelic variation in FTO would be associated with persistent postprandial brain activation by visual cues of calorically dense foods in an extended satiety network previously shown to mediate appetitive behaviors (1, 21); and 2) that these responses would be associated with greater ad libitum caloric intake.

METHODS

Participants

Although considered to be individuals for analyses of FTO genotype, participants were enrolled in a twin study of genetic influences on obesity and were recruited from the community-based Washington State Twin Registry (22). Briefly, participants were between the ages of 18 and 50 y, had a BMI of 18.5–50, and had no major medical problems, dietary restrictions (e.g., vegan or vegitarian diet), heavy alcohol consumption, daily smoking, use of medications that impact appetite (e.g., atypical antipsychotics), history of bariatric surgery, eating disorders, excessive exercise, or current participation in a weight-loss program.

The sample included 42 monozygotic (MZ) pairs and 15 same-sex dizygotic (DZ) pairs (Supplemental Figure 1). Twenty randomly selected MZ pairs were recruited across the range of eligible BMIs (1); the remaining 22 MZ and 15 DZ pairs were recruited such that one member of the twin pair had a BMI of ≥30 in order to enrich the sample for individuals with obesity. The University of Washington Human Subjects Division approved all study procedures. Participants provided written informed consent.

Study procedures

Study procedures have been described in detail elsewhere (1) and are shown in Supplemental Figure 2. Participants arrived fasting and each twin pair completed all study procedures on the same day (staggered by 30 min). Following height and weight measurements, antecubital IV placement, and a fasting blood draw at 0800, individuals were provided a standardized breakfast (t0: 0800 for twin 1, 0830 for twin 2) representing 10% of their estimated daily caloric requirements (calculated with the Mifflin-St Jeor equation and standard activity factor of 1.3) (23). Following the first MRI session (t210), participants consumed a standardized meal of macaroni and cheese (representing 20% of estimated daily caloric requirements) and then underwent the second MRI session (t270). One hour following the standardized meal (t300), participants were presented with an ad libitum buffet and had 30 min to select and consume food of their choice.

Body composition

Body composition was measured by dual X-ray absorptiometry (81% of subjects on a General Electric Lunar Prodigy and 19% on a GE Lunar iDXA using a correction factor). EnCore software (platform version 16.2, General Electric Medical Systems) was used to estimate subcutaneous and abdominal visceral adipose tissue mass (24, 25).

Behavioral measures

Eating behavior was assessed by the Three-Factor Eating Questionnaire 18-item version (26) and the Revised Restraint Scale (27). The International Physical Activity Questionnaire Short Form assessed self-reported physical activity (www.ipaq.ki.se). Visual analog scale ratings of hunger and fullness were completed serially to assess appetite (28). Food appeal ratings of a subset of photographs viewed during the fMRI were completed on a 10-point Likert scale from “not at all” to “extremely” appealing (21). Ad libitum total caloric and macronutrient intake was surreptitiously monitored (ProNutra; Viocare Technologies) (1) by a buffet meal comprised of foods exceeding estimated energy needs (∼5000 kcal total) and varying in hedonic appeal and caloric content (Supplemental Figure 3).

Plasma glucose and hormone analyses

Blood samples were collected in tubes containing EDTA plus protease inhibitor cocktail (Sigma-Aldrich), aprotinin (Sigma-Aldrich), and dipeptidyl peptidase-4 inhibitor (Millipore). Samples were cold centrifuged (2000 rpm for 10 min at 4°C), and plasma divided into aliquots and stored at –80°C until analyzed. Automated assays were used to measure plasma glucose concentrations (hexokinase method) and plasma insulin concentrations (Tosoh immunoenzymometric assay). Total glucagon-like peptide-1 (GLP-1) concentration was measured by ELISA (Human Total GLP-1 assay; ALPCO), and total ghrelin and leptin concentrations were determined by Human RIA assays (Millipore). All assays were performed in duplicate, and all samples from each twin pair were run in the same batch. Intra- and interassay coefficients of variation were 9.2% and 5.3% for leptin, 3.9% and 11.5% for ghrelin, and 3.0% and 6.5% for GLP-1.

fMRI paradigm, acquisition, and processing

The visual food cue images and imaging paradigm have been described in detail elsewhere (1). Images of foods were validated in an independent sample and rated as to whether they should (“nonfattening”; e.g., fruit, vegetables) or should not (“fattening”; e.g., desserts, French fries) be eaten when trying to lose weight (29). The terminology “fattening” and “nonfattening” encompasses the foods’ energetic properties and social connotations. Nonfood images depicted common objects (e.g., pencils, electronics). The paradigm included 13 blocks of 10 images each (7 nonfood objects, alternating with 3 nonfattening and 3 fattening food blocks).

fMRI analyses

Image acquisition and processing methods are described in the Supplemental Methods. A priori regions of interest (ROIs) within an extended satiety network were selected because activation by food cues in these regions was previously shown to be a marker of satiety or predictor of food choice (21) that is influenced by inherited factors (1). ROIs included the medial orbital frontal cortex (mOFC) and bilateral ventral striatum, amygdala, dorsal striatum, and insula. We used a combined anatomic and functional criteria approach to create masks for each region. ROI masks (Supplemental Figure 4) were first anatomically defined based on the Harvard-Oxford probabilistic atlas (30) with a minimum probability criterion of 25%, and then further restricted to voxels exhibiting a minimum level of responsivity to fattening food compare with objects (whole-brain; P < 0.05, uncorrected using a mixed-effects model that accounted for twins). Pre- and postmeal scans of all subjects were used to derive functional criteria to avoid introducing bias related to FTO genotype or satiety state. The resultant masks were then applied to subject-level data and mean parameter estimates were extracted for contrasts of interest. Initial analyses combined all 5 regions to test group differences (higher- compared with lower-risk groups) in mean brain activation in our extended satiety network. To investigate potential group differences within the individual regions comprising this network, region-specific analyses were then performed; the ventral tegmental area/substantia nigra (VTA/SN) was also included because of its key role in dopaminergic signaling [the VTA/SN mask was anatomically defined previously (31, 32)]. For descriptive purposes, the anatomic locations of activation across the extended satiety network and VTA/SN were mapped using the Local Analysis of Mixed Effects (Oxford Centre for Functional MRI of the Brain). Z statistic images (Gaussianized T/F) were corrected for multiple comparisons using a cluster-threshold correction with an individual voxel threshold at Z > 1.65 (voxel height) and a family-wise error (FWE)-corrected cluster significance threshold of P = 0.05 (voxel extent, FWE-corrected across all ROIs examined). Using a voxel-wise approach, exploratory analyses of all voxels outside of the pre-hypothesized ROIs of the extended satiety network and VTA/SN were completed in the same manner except that a Z threshold of 2.3 was used.

Genotyping and FTO obesity risk group definition

Genomic DNA was extracted from whole blood. Zygosity was determined by genotyping the twins for a set of 10 short tandem repeat markers. Polymerase chain reaction was used to amplify DNA fragments in 20-µL reactions using primers flanking the FTO polymorphism (primers available upon request). Genotyping of 4 polymorphisms at FTO (rs9939609, rs1421085, rs8050136, and rs9930506) was performed by Sanger sequencing. The primary single nucleotide polymorphism (SNP) of interest was the rs9939609; however, rs9939609 and rs8050136 were in complete linkage disequilibrium in this sample. Distributions of rs1421085 and rs9930506 genotypes (obtained to improve coverage across racial groups) are shown in Table 1.

TABLE 1.

Participant characteristics for higher- and lower-risk FTO genotypes¹

	Lower risk,	Higher risk,
	N = 94	N = 20	P value
Genotype	TT,AT	AA
Female, %	60	30	0.02
MZ, %	77	60	0.13
Age, y	30.5 ± 9.4	27.2 ± 7.1	0.07
Race			0.70
Caucasian, %	79	80
Asian, %	2	0
Black, %	8	15
Mixed or other, %	12	5
BMI, kg/m2	30.0 ± 5.8	32.7 ± 6.1	—2
Fat mass,3 kg	33.1 ± 13.0	37.0 ± 13.2	0.23
Percent fat,3 %	37.2 ± 7.5	38.4 ± 7.6	0.56
Lean mass,3 kg	53.6 ± 9.1	57.5 ± 9.3	0.08
Percent lean,3 %	60.6 ± 6.9	59.6 ± 7.0	0.56
Visceral fat mass,3 g	1017 ± 768	1077 ± 781	0.80
rs1421085			<0.0001
TT,CT, %	98	20
CC, %	2	80
rs9930506			<0.0001
AA,GA, %	94	25
GG, %	6	75

¹Values are means ± SDs unless otherwise indicated. The recessive model of inheritance was used to determine lower-risk and higher-risk groups for SNP rs9939609. FTO, fat mass and obesity-associated gene; MZ, monozygotic.

²BMI not reported because the recessive model was chosen based on BMI differences.

³Body composition data are adjusted for sex. Group comparisons are unadjusted and were made by chi-square test (categorical) or by linear regression with a robust estimator of variance.

To define our “lower” and “higher” risk groups, we selected the inheritance model that provided the best fit for the association of FTO rs9939609 genotype with BMI. The recessive model (lower risk = TT or AT compared with higher risk = AA) provided the best fit (F_1,112 = 3.27) compared with the additive (TT compared with AT compared with AA; F_2,111 = 1.62), dominant (TT compared with AA or AT; F_1,112 = 0.26), and homozygous (AA compared with TT; F_1,55 = 2.59) models. Using the recessive model, 20 individuals were identified as having a higher risk for obesity based on FTO rs9939609 genotype (AA). The lower-risk group was composed of 94 individuals with either a TT or AT genotype (Supplemental Table 1) except for analyses of appetite, hormone, and fMRI data in which 3 subjects were excluded due to unrelated illness during study procedures (n = 91). The sample size for fMRI analyses premeal was 110 individuals (n = 90 lower risk, 20 higher risk) and 108 postmeal (n = 88 lower risk, 20 higher risk) due to unsuccessful MRI scans.

Statistics

Nonnormally distributed variables were transformed. Adjusted medians were compared by the Wilcoxon rank-sum test. For testing hypotheses between groups, simple and multiple linear regression with a robust estimate of variance due to inclusion of twin data or linear mixed models with the restricted maximum likelihood estimation were used and were adjusted for sex unless otherwise specified. Adjusted means and 95% CIs are presented. Intraclass correlations (ICCs) were calculated using linear mixed models with the restricted maximum likelihood estimation. When correlational analyses included all subjects and were not testing the effect of genetics, we used generalized estimating equations with the identity link, a robust standard error, and an independent covariance structure to adjust variance for the inclusion of twin pairs within the models. Statistics and graphing were completed using STATA (v. 13.1) and GraphPad Prism (v. 6.00).

RESULTS

Participant characteristics

Sixty-two participants were women and 52 were men. There was a significantly lower proportion of women in the higher-risk group (Table 1), and therefore all comparisons by risk group were adjusted for sex. Enrichment of the sample for obese persons resulted in higher- and lower-risk groups that were similar in body composition (Table 2).

TABLE 2.

Behavioral characteristics for higher- and lower-risk FTO genotypes¹

	Lower risk	Higher risk	P value
Eating behavior
TFEQ subscales2
Cognitive restraint	14.0 (13.3, 14.6)	12.5 (11.0, 13.9)	0.08
Uncontrolled eating	19.1 (18.2, 20.1)	19.9 (17.8, 21.9)	0.51
Emotional eating	6.4 (6.0, 6.9)	6.8 (5.8, 7.9)	0.54
Restraint scale2	15.0 (14.0, 16.0)	16.0 (13.9, 18.2)	0.35
Physical activity
IPAQ3
Total activity, MET-min/wk	2989 (1023–5415)	2978 (1365–5581)	0.76
Vigorous activity, MET-min/wk	469 (409–1369)	619 (169–1939)	0.49
Subjective satiety
VAS score2
Premeal hunger, mm	65.5 (61.8, 69.2)	67.2 (59.0, 75.5)	0.74
Premeal fullness, mm	20.7 (17.8, 23.6)	11.3 (4.98, 17.7)	0.005
Postmeal hunger, mm	39.5 (35.4, 43.6)	43.6 (34.5, 52.7)	0.36
Postmeal fullness, mm	44.8 (40.8, 48.8)	36.1 (27.2, 45.0)	0.04

¹All data are adjusted for sex. Group comparisons for eating behavior and subjective satiety measures were made by multiple linear regression with a robust estimator of variance. For physical activity, adjusted medians were compared by Wilcoxon rank-sum test. VAS scores represent the mean satiety rating during the MRI examinations. FTO, fat mass and obesity-associated gene; IPAQ, International Physical Activity Questionnaire; MET, metabolic task equivalent; TFEQ, Three-Factor Eating Questionnaire; VAS, visual analog score.

²Means (95% CIs).

³Medians (IQRs).

Appetite and food appeal

Overall hunger ratings were not significantly different (χ² = 0.81, P = 0.37; Figure 1A). However, the higher-risk group consistently reported themselves as less full (χ² = 7.48, P < 0.01), including during both fMRI sessions (Figure 1B and Table 2). Sensitivity analyses showed that lower fullness ratings were independent of the effects of age, fat mass, lean mass, or BMI (data not shown). Compared with the lower-risk group, higher-risk individuals rated fattening foods as more appealing overall (χ² = 3.92, P < 0.05; Figure 1C), and further adjustment for fat mass did not alter the finding (χ² = 3.96, P < 0.05). There were no genotypic differences in appeal ratings for nonfattening food (Figure 1D).

FIGURE 1.

Satiety, food appeal, and caloric intake in higher- compared with lower-risk FTO genotypes. All participants rated themselves as equally hungry across the study visit day (A). Individuals at higher risk for obesity (genotype AA) consistently rated themselves as less full across the visit day (P < 0.01, main effect of group) (B). Overall, appeal ratings for both “fattening” (C) and “nonfattening” (D) foods were lower postmeal [P < 0.001, P < 0.01, respectively; main effect of meal (a compared with b)]; however, individuals at higher risk for obesity rated fattening foods as more appealing overall [P < 0.05; main effect of group (c compared with d)]. Participants in the higher-risk group consumed more calories based on calculated daily caloric requirement at the ad libitum buffet (E). In panels A and B, arrows indicate consumption of a meal: up-arrow, breakfast (10% of estimated daily caloric need); down-arrow, standardized meal (20% of estimated daily caloric need); double-arrow, ad libitum buffet. Gray bars represent fMRI scans. Data are mean ± SEM, P values were determined by linear mixed models (A–D) and linear regression (E), and all models are adjusted for sex (lower risk, n = 91; higher risk, n = 20). **P < 0.05 compared with lower-risk group. FTO, fat mass and obesity-associated gene; VAS, visual analog scale.

Food intake

A higher-risk FTO genotype was associated with consuming ∼350 more total calories at the ad libitum buffet compared with individuals with a lower-risk genotype (β = 348 kcal; 95% CI: 45, 650 kcal; P = 0.03; d = 0.73). Adjusting for fat (P = 0.03) or lean mass (P = 0.04) did not alter the finding, and caloric intake determined by the percentage of daily caloric needs consumed at the buffet was also greater (Figure 1E). In contrast, macronutrient percentages consumed did not differ between the genotype groups (percentage of fat: 34.9% compared with 34.1%, P = 1.0; percentage of carbohydrate: 50.8% compared with 50.9%, P = 1.0; percentage of protein: 14.2% compared with 14.9%, P = 0.40). In order to test the extent to which variation in FTO explained overall genetic similarity in food intake (1, 12), we used a twin analytic approach. The overall sex-adjusted ICC for caloric intake at the buffet meal among MZ pairs was 66% (95% CI: 48%, 84%; P < 0.0001). The DZ pair ICC was 36% (95% CI: 0%, 82%; P = 0.08), but CIs were wide. Models adjusted for FTO genotype did not markedly change these estimates, reflecting the strength of genome-wide influences on appetitive behavior relative to variation at a single gene.

Appetite-regulating hormones and glucose

Fasting leptin concentrations and HOMA-IR estimates did not differ between FTO obesity risk groups (Figure 2A, B), even when models were also adjusted for fat mass (data not shown). Figure 2C–F depicts the study profiles of untransformed glucose, insulin, ghrelin, and GLP-1 values, stratified by FTO genotype group. Higher- and lower-risk groups did not differ in visit day profiles for glucose (χ² = 0.05, P = 0.83), log insulin (χ² = 0.48, P = 0.49), log total ghrelin (χ² = 1.87, P = 0.17), or log GLP-1 (χ² = 1.53, P = 0.22). However, there was a significant interaction (time × group) for log insulin (χ² = 19.87, P < 0.01); insulin concentrations after eating the standardized meal were elevated in the higher-risk compared with the lower-risk genotype group (Figure 2D; z = 2.65, P < 0.01). This tendency toward higher postmeal insulin concentrations persisted in a fully adjusted model that also included blood glucose concentrations as a covariate (t = 1.96, P = 0.052). We therefore examined excursions in meal-related satiety signals from before to after the standardized meal. The higher-risk genotype group had a greater percentage change in blood glucose concentration (13.6% compared with 8.6%, P = 0.03) and insulin concentrations (627% compared with 301%, P = 0.02) compared with the lower risk group. GLP-1 did not differ between the groups (33% compared with 31%, P = 0.83), but ghrelin decreased more from before to after the test meal among the higher-risk group (–3.6% compared with 2.8%, P = 0.01). Findings were minimally altered by adjusting for fat mass (data not shown).

FIGURE 2.

Appetite-regulating hormones and glucose in higher- compared with lower-risk FTO genotypes. Participants at higher risk for obesity had similar fasting plasma leptin concentrations (A) and calculated HOMA-IR scores (B) compared with those at lower risk for obesity. The groups also had similar glucose (C) and insulin (D) profiles across the study visit day; however, after the standardized meal (t270), higher-risk participants had higher plasma insulin values. Plasma concentrations of ghrelin (E) and GLP-1 (F) did not differ between groups. In panels C–F, arrows indicate consumption of a meal [i.e., up-arrow, breakfast (10% of daily caloric need); down-arrow, standardized meal (20% of estimated daily caloric need)], and gray bars represent fMRI scans. Data are means ± SEMs, P values were determined by linear mixed models and are adjusted for sex (lower risk, n = 91; higher risk, n = 20). *P < 0.001 compared with lower-risk group. FTO, fat mass and obesity-associated gene; GLP-1, glucagon-like peptide 1.

Brain activation across an extended satiety network

Mean activation by fattening food (compared with objects) across all ROIs in the extended satiety network did not differ between the higher- and lower-risk genotype groups either before (10.7; 95% CI: 5.0, 16.4 compared with 7.6; 95% CI: 4.9, 10.2; P = 0.28, d = 0.24) or after (8.2; 95% CI: 2.6, 13.8 compared with 7.4; 95% CI: 4.8, 10.1; P = 0.83, d = 0.06) the standardized meal, nor did responses to nonfattening food images (data not shown). There were, however, regional differences between the FTO obesity risk groups. The higher-risk group had significantly greater activation prior to the standardized meal by fattening food (compared with objects) in the mOFC (Figure 3) and tended toward greater activation within the ventral striatum. After eating, the higher-risk group also had greater activation in the VTA/SN by fattening food. In comparison, the higher-risk group showed significantly less activation by nonfattening food (compared with objects) in the ventral striatum after eating and tended to have less activation in the dorsal striatum premeal and amygdala postmeal. Accordingly, the contrast of fattening compared with nonfattening food cues emphasized the higher risk group's significantly greater pre-meal activation in the ventral striatum and postmeal activation in the amygdala, ventral striatum, and VTA/SN. Figure 4 depicts the anatomic locations of greatest activation for these findings.

FIGURE 3.

Differences in brain activation between FTO genotype groups within regions of an extended satiety network. Gray panels depict the results for brain activation in response to fattening food cues (top: compared with objects; bottom: compared with nonfattening images). Premeal (left panel), the regions in which individuals at higher risk for obesity (based on FTO genotype) had significantly greater activation by fattening food cues than the lower risk group were the mOFC and ventral striatum, and a trend was present for the amygdala. Postmeal (right panel), the amygdala, ventral striatum, and VTA/SN also demonstrated significantly greater activation by fattening food cues in the higher-risk group. In contrast, participants at lower risk for obesity showed significantly greater activation by images of nonfattening foods (middle panel) in the ventral striatum (postmeal) and trends were present in the dorsal striatum (premeal) and the amygdala (postmeal). Data are β weights ± 95% CI, P values were determined by linear regression and are adjusted for sex (premeal: lower risk, n = 90; higher risk, n = 20; postmeal: lower risk, n = 88; higher risk, n = 20). *P < 0.05 compared with other group (Cohen's d = 0.44–0.62), ^#P < 0.085 compared with other group (Cohen's d = 0.39-0.46). FTO, fat mass and obesity-associated gene; mOFC, medial orbital frontal cortex; VTA/SN, ventral tegmental area/substantia nigra.

FIGURE 4.

Anatomic locations of differences in activation between higher- and lower-risk FTO genotype groups. To provide anatomic specificity to the results presented in Figure 3, the locations of activation differences between FTO-based obesity risk groups were mapped using a descriptive voxelwise approach limited to the extended satiety network and VTA/SN. The left panel (premeal) shows clusters of greater activation in individuals at higher compared with lower risk (A, fattening > objects; B, fattening > nonfattening). The right panel (postmeal) shows clusters with greater activation in the higher-risk compared with the lower-risk groups (C, fattening > objects; E, fattening > nonfattening) and clusters with greater activation in lower-risk compared with higher-risk groups (D, nonfattening > objects). Z statistic maps were corrected for multiple comparisons and were thresholded at Z > 1.65 and a cluster significance threshold of P = 0.05 (FWE corrected across all ROIs examined; premeal: lower risk, n = 90; higher risk, n = 20; postmeal: lower risk, n = 88; higher risk, n = 20). Color scales provide Z values of functional activation. Montreal Neurological Institute coordinates are indicated. FTO, fat mass and obesity-associated gene; FWE, family-wise error; mOFC, medial orbital frontal cortex; ROI, region of interest; VTA/SN, ventral tegmental area/substantia nigra.

Correlations of activation within an extended satiety network with measured food intake

In order to test whether postmeal brain activation in ROIs was related to subsequent ad libitum buffet meal intake, we performed correlational analyses among all participants. Mean activation across the extended satiety network by fattening foods (compared with objects) was positively associated with eating more kilocalories at the buffet meal (Figure 5; R² = 0.29, P = 0.04) and a greater percentage of estimated daily calorie needs (R² = 0.09, P = 0.04, unadjusted). Regionally, positive correlations between activation and ad libitum food intake were found for the VTA/SN (R² = 0.27, P = 0.01), dorsal striatum (R² = 0.28, P = 0.01), and amygdala (R² = 0.28, P = 0.03). A trend was present for the mOFC (R² = 0.25, P = 0.06).

FIGURE 5.

Greater ad libitum caloric intake is related to postmeal brain activation by fattening food cues within an extended satiety network. A correlational analysis performed among all participants showed that brain activation measured in the satiated state (after intake of a standardized meal) was positively associated with subsequent ad libitum caloric intake at a buffet meal. For each individual, parameter estimates were calculated as the mean activation by fattening food cues (compared with objects) across all regions within the extended satiety network (amygdala, insula, dorsal striatum, ventral striatum, and mOFC). Pearson's correlation coefficient and P value were derived from generalized estimating equation. Data are adjusted for sex (n = 108). mOFC, medial orbital frontal cortex.

Exploratory analyses outside the extended satiety network

Exploratory voxel-wise analyses identified greater activation by fattening food (compared with objects) premeal among the higher-risk genotype group in a large area of right intracalcarine cortex that included both primary visual cortex and visual association regions (Supplemental Table 2). In contrast, the lower-risk group demonstrated greater activation in the left angular gyrus, which plays a role in semantic processing (33). The lower-risk group also showed a number of regions in which nonfattening food (compared with object) images elicited greater activation, including the left inferior frontal gyrus, frontal pole, and cerebellum. Postmeal, the higher-risk group had greater activation by fattening food cues (compared with objects) in a region of the midbrain (red nucleus) adjacent to the VTA/SN. The lower-risk group showed greater activation by fattening food in the precuneus, bilateral angular gyrus, and middle frontal gyrus (Brodmann area 8). Additionally, lower-risk subjects had greater activation by nonfattening foods (compared with objects) in the lingual gyrus, subcallosal cortex, and cingulate gyrus.

DISCUSSION

It is well established that the AA genotype at the rs9939609 intronic variant of FTO confers a higher risk of obesity. The current study found that these higher-risk individuals show greater postprandial activation by visual cues of calorically dense foods in brain regions involved in satiety perception (21). Behaviorally, higher-risk individuals also reported less satiety; rated calorically dense, “fattening” foods as more appealing; and ate more total calories at a buffet meal than did similarly obese participants who were either homozygous for the T allele or heterozygous. Moreover, greater postprandial activation by high-calorie food cues across the extended satiety network was directly linked to greater ad libitum caloric intake. These findings provide evidence for a CNS mechanism driving poor satiety responsiveness (10) and overconsumption (12) in people with high-risk FTO genotypes. There was no evidence that peripheral hormonal responses to the meal explained the lack of CNS satiety response. Hyperresponsivity to food cues could also play a role, because higher-risk individuals showed greater activation by fattening food images premeal in the mOFC. In contrast, the lower-risk TT/AT genotype group had higher activation by low-calorie food cues, especially postmeal. Overall, both the behavioral and fMRI findings support a mechanism in which allelic variation in FTO may raise obesity risk through its impact on CNS satiety processing. The corticolimbic regions implicated—the VTA/SN, amygdala, and ventral striatum—specifically suggest that the meal was less effective in suppressing motivation (34) for highly energetic food in the higher-risk individuals.

Existing literature demonstrates that CNS processing of visual food cues has an inherited basis (1, 35). In accordance with these results, individuals with high-risk FTO genotypes have shown altered brain responses to food cues in a number of studies, but the affected regions and direction of effects are not consistent. These inconsistencies may derive partly from some studies’ emphasis on testing within certain ROIs [e.g., nucleus accumbens (13), posterior fusiform gyrus (15), and putamen (14)] or from recognized interactions of genotype with nutritional state (16, 20) and BMI (14). Consequently, findings from studies performed in normal-weight (15, 18, 20) or, on average, overweight (16) subjects might vary from those performed in obese subjects. In addition, individuals with high-risk genotypes who maintain a normal weight into adulthood despite genetic susceptibility may have neural responsiveness that is not typical of high-risk allele carriers. To address variability based on nutritional state and BMI, the current study evaluated participants before and after a meal and achieved similar adiposity levels in the higher- and lower-risk groups.

In a study by Karra et al. (20), 10 high-risk but normal-weight AA rs9939609 men exhibited less activation by food cues (all food) than did low-risk TT men (n = 10) in the VTA/SN, hypothalamus, posterior insula, globus pallidus, thalamus, and hippocampus when fasted. In contrast, we saw no regions in the extended satiety network premeal in which the higher-risk group had significantly less activation. Similar to Karra et al. (20), we found that lower-risk individuals had greater activation by low-calorie food cues, particularly when fed, but the regions identified in the 2 studies did not overlap. Wiemerslage et al. (14) studied 13 AA and 17 TT rs9939609 FTO genotype men across a range of BMIs after an overnight fast. They found greater responsivity to high-calorie food images among AA genotype men in the posterior cingulate cortex, cingulate gyrus, precuneus, and cuneus in whole-brain analyses and in the putamen in an ROI analysis. We also found greater activation by high-calorie food cues among higher-risk subjects premeal, but identified different affected regions: the mOFC, ventral striatum, and, in exploratory analyses, occipital and lateral occipital visual association regions. Our ventral striatal findings are consistent with a study showing that children with AA or AT rs9939609 genotypes had greater responsivity to food commercials in the nucleus accumbens (13). Our exploratory findings in the occipital lobe are consistent with those of a large study (n = 77) showing that AA genotype adults had enhanced response to food cues within posterior fusiform cortical areas that process visual information (15). Finally, when studied 30 min after a glucose load, Heni et al. (16) report that high-risk individuals (based on the rs8050136 genotype) had less prefrontal cortical activation by food cues, suggesting reduced inhibitory control. In sum, allelic variation in FTO affects CNS processing of food cues across regions directing visual identification of food, motivation for food, and cognitive control of behavior. Prior and current findings have not yet coalesced around specific regions. Some studies, but not all, point to greater responsivity to highly energetic foods in high-risk FTO genotypes, particularly in a satiated state and among higher BMI subjects.

Satiety and reward are integrated concepts, because the rewarding aspects of food must be devalued to inhibit further food intake. The current VTA/SN and striatal findings are intriguing in light of hypotheses that dopamine-dependent regulation of reward is a neurobiological factor in FTO’s influence on obesity risk (17–19). Furthermore, satiety-promoting signals, such as leptin and insulin, also suppress activation by high-calorie food cues within dopaminergic pathways (36), possibly through direct action at receptors on VTA/SN dopamine neurons (37, 38) or indirectly via leptin action in lateral hypothalamic neurons that innervate the VTA (39). Although not directly studied in the current report, the findings of impaired satiety and regions implicated are compatible with models proposing a role for dopamine signaling in FTO-mediated obesity risk.

The molecular basis linking allelic variation within the first intron of FTO to obesity risk remains an active area of investigation. Notably, the FTO genetic variants associated with adiposity are in noncoding DNA in the gene's first intron. Over 80 variants in this region are in linkage disequilibrium, so that the specific variant(s) responsible for the associated phenotypes are not yet fully resolved (8, 40). Functional gene candidates in the vicinity that are potentially linked—by remote effects on gene regulatory elements—to allelic variation in FTO include retinoblastoma-like 2 (41), iroquois homeobox 3 and 5 (40, 42), and retinitis pigmentosa GTPase regulator-interacting protein-1 like (RPGRIP1L) (43). Recent data suggest, for example, that the obesity-risk (A) allele at the rs8050136 site alters binding of the transcriptional regulator cut-like homeobox 1 (CUX1) (43). As a result, both FTO and RPGRIP1L expression are decreased, thereby diminishing neural responsiveness to leptin (43) and increasing food intake and fat mass in mice (44). The human data presented herein are informative about the translational relevance of the CUX1 mechanism because, in the current sample, the A allele at rs8050136 was in complete linkage disequilibrium with the A allele of rs9939609. Both loci are common in individuals of Caucasian or European descent, the race of >80% of the sample. Thus, the findings are consistent with a molecular mechanism whereby high-risk alleles in FTO impair CNS leptin signaling and increase meal size (45), perhaps via attenuation of leptin's actions potentiating satiety signaling in the hindbrain (46, 47) or modulating mesolimbic dopamine signaling, or a combination of these affects (39).

The current study is the largest to date to use fMRI to assess brain responses to visual food cues in relation to allelic variation in FTO and the first to link these CNS responses to caloric intake. The genotype groups were well matched for fat mass, fasting hormone concentrations, and HOMA-IR, providing a means to query differences based on rs9939609 genotype relatively independently of adiposity or insulin resistance. However, we could have underestimated the effects of FTO genotype if non-FTO susceptibility factors for obesity act via neuronal pathways that overlap with those affected by the FTO gene or if the effects depend on the presence of other high-risk polymorphisms (e.g., ANKK1) (19). In rodents, intermittent access to high-fat diet raises hypothalamic fto expression and motivation for sucrose (48), and we cannot rule out that higher-risk subjects’ dietary patterns preceding the study influenced the result. Finally, these cross-sectional data cannot link FTO to weight gain.

These findings support the apparent primacy of increased caloric intake as the proximate mechanism for the effects of FTO alleles on adiposity (49). The specific intronic SNP(s) and the molecular mechanisms by which they convey their effects are under investigation (20, 44). Studies of the interplay of homeostatic and dopamine-dependent regulatory mechanisms may be especially revealing.

Notes

Supported by funding provided by DK089036, DK098466, DK52431, and DK026687. Additional assistance was provided by the University of Washington's Nutrition Obesity Research Center (P30 DK035816), Diabetes Research Center (P30 DK017047), and the Institute of Translational Health Sciences (UL1 TR000423).

Supplemental Methods, Supplemental Tables 1 and 2, and Supplemental Figures 1–4 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

Abbreviations used:

CNS

central nervous system

DZ

dizygotic

FTO

fat mass and obesity-associated gene

FWE

family-wise error

GLP-1

glucagon-like peptide-1

ICC

intraclass correlation

mOFC

medial orbital frontal cortex

MZ

monozygotic

ROI

region of interest

SNP

single nucleotide polymorphism

VAS

visual analog scale

VTA/SN

ventral tegmental area/substantia nigra.