Reassessing relationships between appetite and adiposity in people at risk of obesity: a twin study using fMRI

Abstract

Background:

Neuroimaging studies suggest that appetitive drive is enhanced in obesity.

Objective:

To test if appetitive drive varies in direct proportion to the level of body adiposity after accounting for genetic factors that contribute to both brain response and obesity risk.

Subjects/Methods:

Participants were adult monozygotic (n=54) and dizygotic (n=30) twins with at least one member of the pair with obesity. Body composition was assessed by dual-energy X-ray absorptiometry. Hormonal and appetite measures were obtained in response to a standardized meal that provided 20% of estimated daily caloric needs and to an ad libitum buffet meal. Pre- and post-meal functional magnetic resonance imaging (fMRI) assessed brain response to visual food cues in a set of a priori appetite-regulating regions. Exploratory voxelwise analyses outside a priori regions were performed with correction for multiple comparisons.

Results:

In a group of 84 adults, the majority with obesity (75%), body fat mass was not associated with hormonal responses to a meal (glucose, insulin, glucagon-like peptide-1 and ghrelin, all P>0.40), subjective feelings of hunger (β=−0.01mm [95% CI −0.35, 0.34] P=0.97) and fullness (β=0.15mm [−0.15, 0.44] P=0.33), or buffet meal intake in relation to estimated daily caloric needs (β=0.28% [−0.05, 0.60] P=0.10). Body fat mass was also not associated with brain response to high-calorie food cues in appetite-regulating regions (Pre-meal β=−0.12 [−0.32, 0.09] P=0.26; Post-meal β=0.18 [−0.02, 0.37] P=0.09; Change by a meal β=0.29 [−0.02, 0.61] P=0.07). Conversely, lower fat mass was associated with being weight reduced (β=−0.05% [−0.07, −0.03] P<0.001) and greater pre-meal activation to high-calorie food cues in the dorsolateral prefrontal cortex (Z=3.63 P=0.017).

Conclusions:

In a large study of adult twins, the majority with overweight or obesity, the level of adiposity was not associated with excess appetitive drive as assessed by behavioral, hormonal, or fMRI measures.

1. Introduction

While energy homeostasis is controlled primarily by hypothalamic neurons, the downstream brain regions that enact appetitive behaviors are not functionally related to homeostasis, but rather to motivated behavior, reward perception, and attention [1,2]. Neural responses to high-calorie food cues, as measured by functional magnetic resonance imaging (fMRI), within several of these corticolimbic and insular areas positively predict choice of higher fat foods [3] and caloric intake [4]. Thus, activation by high-calorie food cues in key appetite-regulating regions is an objective marker of appetitive drive that can be utilized to understand the relationship of adiposity to central regulation of appetite and food intake. The brain is simultaneously receiving input from the gut and periphery on the state of energy stores via leptin from adipose tissue, circulating nutrient availability in the form of glucose, and meal ingestion via vagal afferents as well as falls in ghrelin or rises in glucagon-like peptide 1 (GLP-1) or insulin [5]. These physiologic signals are integrated with individual genetic, psychological, and behavioral factors such as voluntary efforts at weight control, to ultimately determine food choice, caloric intake, and weight status over time.

It has been theorized that, in obesity, the central nervous system component of this complex neurophysiologic system is characterized by altered appetitive drive and reward perception that promotes overeating of highly palatable foods [6,7]. fMRI findings suggest that, in the fasted-state, individuals with obesity present greater activation to pictures of high-calorie food in areas associated with reward, memory, attention, executive function and decision-making [8–17]; and when satiated, participants with obesity fail to reduce activation in areas related to reward, emotional processing and decision-making [14,16,18]. Some have argued that this represents hyperresponsivity [19,20] whereas others suggest that satiety signaling may be impaired [18]. However, findings of characteristic brain response patterns in obesity could instead reflect underlying genetic predispositions [4,21] (instead of a phenotype related to adiposity itself) a form of bias known as genetic confounding [22]. Heritability has been established for adiposity [23–26], eating behaviors [27–30] and neural responses to food cues [4,31]. Identical twins have more similar brain activation to high-calorie food images after a meal in brain areas related to satiety and reward than unrelated participants [4], even when twins are BMI discordant [31]. When two traits share the same genetic risk factor(s) – as seen for adiposity and neural responses to food cues [4,22] – a twin study approach can address issues related to genetic confounding.

We therefore enrolled adult twins in which at least one member of the pair had obesity to test whether appetitive drive and neural response to food cues in appetite-regulating brain regions varies in proportion to body adiposity, with and without accounting for genetic influences. Assessments of appetitive drive included subjective appetite and objective food intake as well as hormonal excursions evoked by a meal and brain response to food cues in a set of selected corticolimbic and insular areas that were previously shown to positively correlate with caloric intake [4]. To parse the contribution of genetic factors, we first mimicked a standard study design by analyzing the overall sample as if it was composed of unrelated individuals. To minimize genetic confounding, we then compared within-pair differences in body fat mass (FM) to within-pair differences in measures of appetitive drive and neural response to food cues. In monozygotic (MZ) twins, within-pair comparisons are inherently controlled for age, sex, genetic influences, family environment, and unmeasured potential confounders. We hypothesized that FM would be related to 1) pre-meal brain response to food cues in appetite-regulating regions and 2) the degree of reduction in brain response to food cues induced by a standardized meal, but relationships would be attenuated when genetic confounding was controlled for in within-pair analyses.

2. Subjects and Methods

2.1. Participants

Eighty-four MZ (n=54) and same-sex dizygotic (DZ, n=30) twins were recruited from the community-based Washington State Twin Registry [32]. Inclusion criteria were: at least one twin from the pair had BMI ≥ 30 kg/m², age 18–50 y, BMI 18.5–50 kg/m², weight stable for at least 6 mo., and twins raised together until ≥ age of 15 y. Exclusion criteria were: current participation in a formal weight-loss program, use of weight-loss medication or medications that impact appetite, history of weight-loss surgery or eating disorders; major medical problems including diabetes; food allergies or vegan or vegetarian diet; smoking or heavy alcohol consumption; and contraindication to MRI. Seven MZ complete pairs were originally recruited for a prior randomly selected twin sample [21], but also met criteria for the current study in which at least one member of the pair has obesity and were included. Procedures were approved by the University of Washington Human Subjects Committee. All participants provided written informed consent.

2.2. Procedures.

Twin pairs underwent study procedures (Figure 1) on the same day staggered by 30 min. Twin order was counterbalanced by BMI and assigned by a statistician who did not interact with the participants [21]. Standardized meals were titrated to estimated daily caloric needs as calculated by the Mifflin-St. Jeor equation and an activity factor of 1.3 [33] and based on self-reported height and weight obtained at screening. Participants were offered a standardized breakfast (10% of the participant’s daily caloric need), then, three hours after breakfast, the pre-meal fMRI session, followed by the consumption, within 15 min, of a standardized meal consisting of macaroni and cheese (50% fat, 40% carbohydrate, 10% protein), representing 20% of the daily caloric requirement, and, 30 min after the end of the first fMRI, the post-meal fMRI session, which was immediately followed by an ad libitum buffet (~5000 kcal provided) served in a private room where participants had 30 min to choose and consume food. Total caloric intake and macronutrient percentages (%) were surreptitiously measured by weighing all uneaten food (ProNutra, Viocare Inc. Princeton, NJ, USA). Additional details are in [21].

Figure 1. Study overview.

After an overnight fast (≥ 9 h), study visits started at 0800 AM with intravenous line placement and a blood draw. Blood samples and appetite ratings by visual analog scale (VAS) were collected throughout the visit. Participants consumed a standardized breakfast (Bkfst), titrated to represent 10% of daily caloric needs, underwent dual-energy X-ray absorptiometry (DXA) scan and completed behavioral questionnaires. Participants underwent magnetic resonance imaging (MRI) #1, consumed, within 15 min, a standardized meal of macaroni and cheese (50% fat, 40% carbohydrate, 10% protein), representing 20% of daily caloric needs, then underwent MRI #2, followed by an ad libitum buffet meal served in a private room where she/he had 30 min to choose and consume food from with presented food varying in its macronutrient composition and energy density (~5000 kcal).

2.3. Anthropometrics

Weight and height were measured in light clothing after voiding. Participants self-reported their adult maximum weight (excluding pregnancy for women), which was then used to calculate the % weight-reduced from lifetime maximum weight according to the equation: adult maximum weight minus current weight, divided by adult maximum weight, multiplied by 100.

2.4. Body composition

Total body fat and lean mass were assessed by dual-energy X-ray absorptiometry (DXA; GE Lunar Prodigy or iDXA with correction factor, GE Healthcare, Milwaukee, WI, USA) [4].

2.5. Laboratory assessment

Blood samples were obtained for measurements of fasting leptin (Human RIA assays, Millipore), glucose (hexokinase method), and insulin (Tosoh immunoenzymometric assay); HOMA-IR was calculated [34]. Glucose, insulin, ghrelin (Human RIA assays, Millipore) and glucagon-like peptide-1 (GLP-1) (ELISA Human total GLP-1 assay, ALPCO) levels were analyzed pre- and post-meal.

Pre-meal concentrations were obtained before the standardized meal. Thirty and 60 min after the meal, samples were taken from which the post-meal peak or nadir values were determined to be at 30 min post-meal for insulin and GLP-1 (peaks), and at 60 min post-meal for glucose (peak) and ghrelin (nadir). The % change with a meal was calculated subtracting the post-meal from the pre-meal level, dividing by the pre-meal level, and multiplying by 100. All assays were performed in duplicate. Samples from both twins in the pair were batched together. Intra- and inter-assay coefficients of variation were 9.2% and 5.3% for leptin, 3.0% and 6.5% for GLP-1 and 3.9% and 11.5% for ghrelin [4].

2.6. Behavioral assessments

Visual analog scale (VAS) appetite ratings of hunger and fullness were completed before and after eating [35] on a zero (I am not hungry/full at all) to 100 mm (I have never been more hungry/full) scale.

The Three-Factor Eating Questionnaire Revised 18-item version (TFEQ-R18) [36] was applied as a validated measure of patterns of eating behavior (cognitive restraint, emotional eating and uncontrolled eating) to augment appetite and hormone assessments.

2.7. Functional MRI

2.7.1. Images and imaging paradigm

Study images were easily recognizable food items rated as compatible (“low calorie”) or not compatible (“high-calorie”) with dieting in order to lose weight and non-food objects [37]. Image selection and validation are detailed in the Suppl. Methods. The fMRI paradigm consisted of 13 blocks of 10 photographs each shown for 2.4 s each. Object blocks alternated with high- and low-calorie food blocks (Suppl. Figure 1A). The order of blocks was counterbalanced between twin pairs but was the same for each member of a pair.

2.7.2. Acquisition and processing

Acquisition parameters and image processing procedures are previously published [21] and are detailed in Suppl. Methods.

2.7.3. Region of interest (ROI) approach

A ROI approach was employed and included a priori regions that are markers of appetitive drive because activation by high-calorie food cues within these regions is known to be related to satiety and/or to predict food choice and total caloric intake [3,4]. ROIs included were left and right dorsal and ventral striatum, left and right amygdala, left and right insula, medial orbitofrontal cortex, substantia nigra/ventral tegmental area (Suppl. Figure 1B). The latter was anatomically defined as previously described [21], all other ROI masks were established with a functional-anatomic approach [3]. Briefly, functional maps were generated at the group level (whole-brain mixed-effects model, P<0.05, uncorrected, accounting for twins and including all participants), for high- and low-calorie (both vs. objects) contrasts during both pre- and post-meal sessions. Maps were summed to create a group level food vs. object activation map representing a minimal level of responsivity to food cues. Masks were restricted to activated voxels within a priori identified anatomic regions (25% probability based on Harvard-Oxford atlas) [38] and applied to individual-level data. Extracted mean ROI activation (in parameter estimates) was then averaged into a single value for brain activation within all a priori ROIs [21].

2.7.4. Exploratory analysis

A voxelwise approach assessed correlations between FM and brain activation in all voxels outside of the a priori ROIs. These voxels were mapped with FEAT v6.0 (part of FSL) using a FLAME statistical model. The obtained Z statistic images (Gaussianized T/F) were corrected for our non-independent sample of twins and for multiple comparisons with a cluster-threshold correction: individual voxel threshold at Z>2.3 and corrected cluster significance threshold of P=0.05. A cluster mask was applied to all participants and the mean parameter estimates were extracted from this region for correlational analyses.

2.8. Statistics

Data are reported as mean ± standard deviation (SD), unless otherwise noted. Normality was confirmed for continuous variables, transformations applied when indicated for statistical testing and back-transformed for graphical presentation. Potential confounding by age and sex was evaluated separately for each analysis, and identified confounders were included as covariates in models. Proportions of categorical variables are compared by chi-squared tests. Generalized estimating equation regression models or linear mixed models with restricted maximum likelihood estimation were used to account for the non-independence of twins in overall analyses among all twins [39]. In within-pair analyses limited to MZ twin pairs, simple and multiple linear regressions tested unadjusted and adjusted associations. For within-pair analyses, MZ twins within each pair were ordered based on their total FM (twin 1 and twin 2). Twin analyses were further adjusted for between pair effects of age and sex. When multiple comparisons were performed beyond our a priori hypotheses, Bonferroni corrected P-values were used to judge significance. Statistical analyses and graphing were completed with STATA (version 15.1; StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX, USA: StataCorp LLC.) and GraphPad Prism (version 8 for Windows, GraphPad Software, La Jolla, CA, USA).

3. Results

3.1. General characteristics

Table 1 shows participant characteristics for the overall sample. Age ranged from 18.6 – 48.6 y and BMI from 20.6 – 43.6 kg/m². Due to recruitment of pairs in which the index twin had obesity, the majority of the sample fell in the overweight or with obesity category (94%). Participants, while required to be weight stable for the past 6 mo., were on average 6% weight-reduced from their lifetime maximum weight (range 0 – 25.4%). Greater FM was strongly associated with smaller % of weight-reduced from lifetime maximum weight, independent of sex (Figure 2A). Greater FM was also associated with higher fasting levels of leptin (β=0.99 ng/mL P<0.001) and HOMA-IR (β=0.08 P<0.001), independent of sex. MZ pairs’ characteristics are in Suppl. Table 1.

Table 1.

Participants characteristics

General Characteristics (n=84 participants)
Age, years		31.4 ± 9.1
Female, %		54.8
Zygozity, % MZ		64.3
BMI, kg/m2		32.8 ± 4.7
BMI category, %	with obesity	75
	overweight	19
	normal weight	6
Fat mass, kg		38.8 ± 10.6
Weight reduced from lifetime maximum weight, %		6.3 ± 5.7
Hormone Profile
Fasting Leptin, ng/mL		34.7 ± 22.8
HOMA-IR		2.8 ± 1.7
Insulin	Pre-meal, μU/mL	8.0 ± 5.3
	Post-meal peak, μU/ml	344* ± 22.5
	Change with a meal, %	402.9 ± 380.3
Glucose #	Pre-meal, mg/dL	90.4 ± 7.6
	Post-meal peak, mg/dL	103.5* ± 11.0
	Change with a meal, %	16.7 ± 12.1
GLP-1	Pre-meal#, pmol/L	8.3 ± 10.2
	Post-meal peak, pmol/L	9.8* ± 10.1
	Change with a meal, %	36.6 ± 40.3
Ghrelin #	Pre-meal, pg/mL	900.8 ± 337.4
	Post-meal nadir, pg/mL	812.2* ± 284.2
	Change with a meal, %	−9.7 ± 11.2
Behavioral and Appetite Ratings
VAS Fullness	Pre-meal	17.9 ± 14.7
	Post-meal	48.7* ± 21.0
VAS Hunger	Pre-meal #	67.3 ± 17.6
	Post-meal	32.9* ± 20.0
TFEQ-R18	Cognitive restraint	13.9 ± 3.0
	Uncontrolled eating	19.3 ± 4.5
	Emotional eating	6.7 ± 2.4
Consumption at ad libitum buffet
Total calories, kcal		1172.0 ± 592.0
Daily caloric needs consumed, %		48.7 ± 20.2

Data are reported mean ± SD, unless otherwise noted. % change with a meal was calculated subtracting the post-meal from the pre-meal level, dividing by the pre-meal level and multiplying by 100. Peak values for Insulin and GLP-1 were obtained 30 min after a meal; Peak value for Glucose and nadir value for Ghrelin were from 60 min after a meal intake.

^#n=83.

^*P<0.001 compared to pre-meal values by generalized estimating equation regression models. MZ, monozygotic; GLP-1, glucagon-like peptide 1; VAS, visual analog scale; TFEQ-R18, three-factor eating questionnaire revised 18-item version.

Figure 2. Body fat mass and objective measures of appetite.

Association of total body fat mass with (A) weight reduction from lifetime maximum weight (square-root transformed for testing and back-transformed for graphing purposes, β=−0.05 %, independent of sex), (B) total caloric intake at the ad libitum buffet (β=17.0 kcal), and (C) the percentage of daily caloric needs consumed (β=0.28 %, both adjusted for age and sex). n=84. P-values were determined by generalized estimating equation regression models. Pearson’s correlation coefficients are presented for descriptive purposes.

3.2. Relationships of body fat mass to hormonal and behavioral measures of appetitive drive

In response to the meal, insulin, glucose, and GLP-1 concentrations significantly increased, whereas ghrelin declined (all P<0.001, adjusted for sex). Greater FM was associated with higher pre-meal levels of insulin (β=0.21 μU/mL P<0.001) and lower pre-meal levels of ghrelin (β=−9.09 pg/mL P=0.001), independent of sex. Conversely, FM was not associated with pre-meal levels of glucose (β=0.02 mg/dL P=0.71) or GLP-1 (β=−0.002 pmol/L P=0.98), or the % change by a meal for insulin (β=−3.98 % P=0.40), glucose (β=−0.05 % P=0.68), GLP-1 (β=−0.12 % P=0.68), or ghrelin (β=0.04 % P=0.66, all adjusted for sex). Subjective hunger and fullness, as assessed by VAS scores, also significantly changed with the consumption of the standardized meal (hunger β=−34.4 mm P<0.001; fullness β=30.8 mm P<0.001). However, FM was not associated with pre-meal (hunger β=−0.01 mm P=0.97; fullness β=0.15 mm P=0.33) or with the meal-induced change in hunger (β=0.26 mm P=0.28) or fullness score (β=−0.20 mm P=0.43), independent of age and sex. Greater FM, independent of sex, was associated with lower TFEQ-R18 cognitive restraint scores (β=−0.06 P=0.01), but the finding did not meet a Bonferroni-adjusted significance level of P<0.004, nor did associations of uncontrolled (β=0.05 P=0.18) or emotional eating subscale scores (β=0.03 mm P=0.14).

Ad libitum buffet meal consumption after the standardized meal and both fMRI exams averaged close to 1 200 kcal (range 14.6 – 3 255.6 kcals), which represented almost 50% of participant’s estimated daily caloric needs (Table 1). Greater FM was related to an increased total caloric intake, independent of age and sex (Figure 2B), but was not significantly associated with the % of estimated daily caloric needs consumed (Figure 2C). Fat mass was unrelated to macronutrient consumption (Protein β=0.01 % P=0.83; Fat β=0.02 % P=0.86; Carbohydrate β=−0.02 % P=0.83).

3.3. Fat mass and brain activation by visual food cues

Body FM was not associated with the average brain activation to high-calorie food cues (vs. objects) within a priori ROIs measured pre-meal, post-meal, or as change by a meal, adjusted for age and sex; (Figure 3). To explore whether being weight-reduced was influencing the negative results (being weight-reduced was related to both lower FM (Figure 2A) and reduced brain activation by a meal (β=−2.47, P=0.045, adj for age and sex)), we performed a fully adjusted analysis by including the % weight-reduced from lifetime maximum weight as a covariate in the model and again found no associations (Pre-meal: β=−0.08 P=0.52; Post-meal: β=0.14 P=0.21; Change by a meal: β=0.21 P=0.25). There were no individual a priori ROIs in which activation was associated with FM (Suppl. Table 4). Analyses of secondary contrasts low-calorie food (vs. object) or high vs. low-calorie food cues, no significant associations were found (Suppl. Table 2).

Figure 3. Body fat mass and brain activation by visual food cues in a priori regions of interest (ROIs).

Association of total body fat mass with brain activation in our a priori ROIs to high-calorie visual food cues (vs. objects) (A) pre-meal (β=−0.12, n=83), (B) post-meal (β=0.18, n=81) and (C) change by a meal (β=0.29, n=81, all adjusted for age and sex). P-values were determined by generalized estimating equation regression models. Pearson’s correlation coefficients are presented for descriptive purposes.

Finally, using a within-pair twin study approach among MZ twin pairs only (n=27 pairs; Suppl. Table 1), we found that within-pair differences in FM were unrelated to within-pair differences in either activation to high-calorie foods (vs. objects) within a priori ROIs pre-meal (β=0.26 P=0.91), post-meal (β=1.01 P=0.54) or change in activation by a meal (β=0.76 P=0.78).

3.4. Exploratory voxelwise analysis of brain activation outside a priori ROIs

Pre-meal, no significant clusters demonstrated positive associations between FM and activation by high-calorie food cues (vs. objects), but a negative correlation was observed in the right frontal pole, representing the dorsolateral pre-frontal cortex (DLPFC) (Figure 4A and Suppl. Table 3). Furthermore, the association between greater FM and reduced activation in the DLPFC to high-calorie food cues was also present for the contrast of high vs. low-calorie food images (Suppl. Table 3). A cluster in the bilateral occipital pole was found in which the change in activation by a meal was negatively associated with FM (Suppl. Table 3). Due to previous demonstration of DLPFC interaction with appetitive drive and its role in inhibiting motivational drive [40,41], extracted mean parameter estimates from the identified DLPFC cluster were used to describe the behavioral, hormonal, and appetitive correlates of DLPFC activation.

Figure 4. Dorsolateral prefrontal cortex activation and eating behavior and appetitive response to a meal.

Voxelwise analysis outside a priori ROIs identified that total body fat mass was associated with (A) pre-meal activation by high-calorie food cues (vs. objects) in a cluster representing the dorsolateral prefrontal cortex (DLPFC), corrected for our non-independent sample of twins and for multiple comparisons with a cluster-threshold correction: individual voxel threshold at Z>2.3 and corrected cluster significance threshold of P=0.05. MNI coordinates of peak location of the cluster are X=61, Y=184, Z=92. Associations of pre-meal DLPFC activation by high-calorie food cues (vs. objects) with (B) TFEQ-R18 uncontrolled eating subscale scores (β=−0.05, n=83, adjusted for fat mass), (C) fat intake (%) in an ad libitum buffet (β=−0.06 %, n=83, adjusted for age and sex). ROI, region of interest; MNI, Montreal Neurological Institute. TFEQ-R18, Three-Factor Eating Questionnaire Revised 18-item version. P-values were determined by generalized estimating equation regression models. Pearson’s correlation coefficients are presented for descriptive purposes.

3.5. Behavioral, hormonal, and appetitive correlates of DLPFC activation

Self-reported % weight-reduced from lifetime maximum weight was not associated with DLPFC activation (β=2.06 P=0.36). In addition, DLPFC activation was unrelated to hormonal parameters (fasting leptin, HOMA-IR, % change with a meal of insulin, glucose, GLP-1 and ghrelin, adjusted for sex, Suppl. Table 5).

However, DLPFC activation was negatively associated with TFEQ-R18 uncontrolled eating scores (β=−0.05 P=0.002), even when analyses were adjusted for FM (Figure 4B). No significant association was seen to TFEQ-R18 cognitive restraint (β=−0.003 P=0.83, unadjusted) or emotional eating (β=−0.002 P=0.86, unadjusted) subscale scores.

At the ad libitum buffet, DLPFC activation to high-calorie food cues was negatively associated with the % of energy intake from fat (Figure 4C). On the contrary, no association was seen between DLPFC activation and total caloric intake (β=−2.34 kcal P=0.16), protein (β=0.01 % P=0.52) or carbohydrate consumption (β=0.05 % P=0.09), or % of estimated daily caloric needs consumed (β=−0.07 % P=0.26, all adjusted for age and sex).

4. Discussion

In a sample of adults predominantly with overweight and obesity, we found no evidence that FM was associated with average brain activation to visual food cues in a selected set of appetite-regulating regions, either pre-meal or in response to a meal. Relationships were also notably absent in twin models that account for shared genetic influences on appetite and adiposity. Although contrary to our hypotheses, the results regarding appetitive drive were consistent across a variety of measures including subjective appetite, monitored food intake, hormonal response to a meal, and brain activation to visual food cues by fMRI analyses that assessed appetitive drive by targeting brain regions in which activation by high-calorie food cues is predictive of total caloric intake [4]. Lower FM was instead correlated with a greater self-reported weight-reduction from lifetime maximum weight and pre-meal activation to high-calorie visual food cues in a brain area fundamental for inhibitory control, the DLPFC, identified in an analysis exploring regions outside our a priori ROIs.

Meta-analyses [15–17,42] have previously reported that, pre-meal, the brain activation to visual food cues in participants with obesity is increased in reward areas, and post-meal, there is sustained brain activation by high-calorie food cues after a meal in areas of reward and appetitive processing when compared to participants with normal weight [18]. But, others have suggested no clear relationships to obesity exist [43]. Using BMI as an indirect measure of adiposity, two studies (n=26 [11] and n=44 [8]) reported positive correlations between brain activation to high-calorie food images and BMI in brain areas related to reward and cognitive control, such as the putamen, caudate, insula [11] and ventral striatum [8]. Other publications (n=40 [13] and n=13 [10]) reported negative associations between brain activation to visual food cues and BMI in the prefrontal cortex [13], orbitofrontal cortex [10] and anterior cingulate cortex [10,13]. In contrast, Luo et al [44] (n=13) and Liebmann et al [12] (n=24) reported no correlation between BMI and brain activation in the striatum [44] and hippocampus [12] but a positive association existed with the measurement of waist circumference [44]. These contradictory findings could be explained by methodological factors such as samples consisting of only female participants [10,11,44], participants from a single ethnic group [44], restricted to participants of normal weight [10], or lack of control for age, satiety state, and being weight-reduced [43,45]. In the current study, the lack of a positive relationship between fMRI-assessed brain response to food cues in the selected regions thus contradicts some of the accumulated neuroimaging literature as well as theories that hyperresponsivity to reward is a driving factor, beyond inherited influences, in obesity [6,7].

The lack of fMRI findings was corroborated by subjective ratings of appetite and meal-induced excursions of peripheral appetite-regulating hormones, which also showed no relation to FM. Yet, adipose tissue mass was linearly related to leptin concentrations, suggesting that variability in FM was being appropriately signaled by peripheral tissue. Leptin has well-documented anorexigenic effects in rodent models [46] and in humans [47], but we documented no appetite suppression despite higher leptin concentrations. Central leptin resistance has been proposed as a potential mechanism for this phenomenon [48], but its presence cannot be discerned by the current study design. There was also no evidence that higher FM contributed differentially to overeating at the ad libitum buffet as the % of estimated daily caloric needs consumed was equivalent across all levels of adiposity. In sum, within a well-controlled experiment with a satiety stimulus that was matched to caloric needs, appetitive drive was not increased in relation to FM, hormonal response to the meal was intact, and food intake beyond caloric needs was not found among weight stable individuals at higher FM.

Appetitive drive is, of course, not the only influence on body adiposity, in part because it is modulated by inhibitory control [40]. Weight-reduced adults [41] and successful weight-loss maintainers [49,50] engage prefrontal cortical regions governing executive functioning and dietary control [51,52]. Weight-loss outcomes improve when inhibitory regions are activated by food cues, including the DLPFC and inferior frontal gyrus, among others [53,54], and disinhibition predicts weight regain [55]. Our exploratory analyses, accounting for the relatedness of twins within the sample, revealed that right DLPFC activation to high-calorie food cues negatively correlated with FM. Participants demonstrating greater pre-meal activation in the DLPFC also had lower scores for uncontrolled eating and consumed less calories from dietary fat at a buffet meal. One hypothesis based on these findings is that among individuals at elevated polygenic risk of obesity, since at least one member of the twin pair had obesity, some have engaged in self-regulation via inhibitory control mechanisms [43].

The current study is among the largest to date and used a gold standard measure of adiposity, in contrast to prior studies utilizing BMI [43,56]. However, our design differs in other important ways. First, we enrolled a sample who were either with obesity or were at risk for obesity based on relatedness to their twin with obesity. As a result of these selection criteria, the sample included few participants of normal weight. Second, we carefully controlled for the state of satiety by titrating the caloric load of the meal according to estimated daily energy needs. Prior studies of response to a meal have used eucaloric load [57,58] that may have accentuated appetite differences related to adiposity because participants with overweight and obesity received a lower percentage of their estimated energy needs. Third, through a twin study design, we were able to perform analyses that controlled for genetic influences on brain response to food cues [4,21,22]. These within-pair analyses strengthen the interpretation that there was no independent relationship between adiposity and brain response to food cues in appetite-regulating regions. This study has important limitations. The twin study design controls for polygenic inherited factors but cannot speak to specific genetic polymorphisms [4,59,60]. Moreover, our group of DZ twins was not large enough to confidently execute within-pair analyses to evaluate familial environment effects. We acknowledge that our measure of weight reduction was self-reported. In addition, our a priori hypothesis-testing was restricted to ROIs and their average activation and excluded other regions involved in the control of ingestion [61]. To ameliorate this limitation, voxelwise analyses were performed outside the a priori ROIs, however, these analyses were exploratory, thus, the DLPFC findings must be interpreted with caution. Further investigation of the relation of visual cortex activation to adiposity is warranted as well as larger studies designed for whole brain exploration under stringent cluster size thresholding [62].

In conclusion, the findings argue against conceptualizations that increasing adiposity is uniformly characterized by appetitive drive beyond that required to meet energetic needs and emphasize the heterogeneity of patients with overweight and obesity in regards to eating behavior. Future studies on brain functional connectivity, particularly between the DLPFC and key appetite-regulating regions, are needed to unravel network communication and continue to illuminate the complex reciprocal relationships of inhibitory control and brain regulation of appetite.

Highlights.

Neuroimaging studies suggest that appetitive drive is enhanced in obesity

Brain response to food cues in appetite-regulating regions was assessed in adult twins

Controlled for genetics, fat mass was unrelated to regional brain response to food cues

Lower fat mass correlated with greater activation in the dorsolateral prefrontal cortex

In adult twins, the level of adiposity was not associated with excess appetitive drive