Exercise Training Reduces Reward for High-Fat Food in Adults With Overweight/Obesity

Kristine Beaulieu; Mark Hopkins; Catherine Gibbons; Pauline Oustric; Phillipa Caudwell; John Blundell; Grahamfinlayson


Med Sci Sports Exerc. 2020;52(4):900-908. 

In This Article



Men and women with overweight and obesity age 18–55 yr were recruited via poster advertisements and e-mail lists at the University of Leeds, United Kingdom, and surrounding areas. Participants were screened on the following inclusion criteria: body mass index (BMI) between 26.0 and 38.0 kg·m−2, nonsmoker, inactive (≤2 h·wk−1 of exercise over the previous 6 months), weight stable (±2 kg for previous 3 months), not currently dieting or participating in a weight loss regime, no history of eating disorders, not taking any medication known to affect metabolism or appetite, and acceptance of the study foods. Participants were asked to keep lifestyle habits and activities constant throughout the study. The study was approved by the Leeds West NHS Research Ethics Committee (09/H1307/7). Participants provided written informed consent before taking part. The study was registered under international standard trials approval (ISRCTN47291569).

Study Design

Forty-six participants (exercisers; 16 men/30 women, mean [SD] age, 43.2 [7.5] yr) completed a 12-wk exercise intervention in which they exercised 5 d·wk−1 under supervision of research staff in the Human Appetite Research Unit, University of Leeds, United Kingdom, between November 2011 and July 2013. Aspects of these data have been previously reported.[9,15–17]

Fifteen control participants (controls; 6 men/9 women, mean [SD] age, 41.4 [10.7] yr) completed 12 wk of maintaining current low activity levels between July 2012 and July 2013. The nonexercising controls were not made aware of the exercise arm of the study; participants were requested not to change their dietary or exercise patterns for the duration of the study.

At baseline and postintervention, food reward and eating behavior variables were assessed (described hereinafter). Food reward was measured during HFAT and HCHO probe days before and after a fixed-energy meal. These days were separated by at least 1 d and in a randomized crossover order. Before each laboratory session, participants were instructed to maintain their usual diet, not to engage in physical activity for 24 h, refrain from consuming alcohol for 24 h, and fast overnight (10–12 h).


Anthropometrics and Body Composition. At baseline and postintervention, participants completed a measurement day. Anthropometrics and body composition measures were taken while participants were wearing tight-fitting clothing and a swim cap. Standing height without shoes was measured to the nearest 0.1 cm using a stadiometer (Leicester height measure, SECA, United Kingdom). Body weight was measured using an electronic balance and recorded to the nearest 0.1 kg (BodPod; Life Measurement, Inc., Concord, CA). Fat mass, fat-free mass, and percentage body fat were estimated via air displacement plethysmography (BodPod) following the manufacturer's instructions.

Eating Behavior Questionnaires. Psychometric questionnaires were completed after the body composition measures. The Three-Factor Eating Questionnaire is a validated 51-item instrument that measures three dimensions of eating behavior: cognitive control of restraint (i.e., concern over weight gain and the strategies adopted to prevent this), disinhibition of eating (i.e., tendency of an individual to overeat and to eat opportunistically in the obesogenic environment), and susceptibility to hunger (i.e., extent to which feelings of hunger are perceived and how these sensations result in food intake).[18] The Binge Eating Scale is a validated 16-item questionnaire that assesses the severity of binge eating.[19] The questions are based on both behavioral characteristics (e.g., amount of food consumed) and the emotional, cognitive response (e.g., guilt or shame).

Test Meals. At baseline and postintervention, participants completed two separate probe meal days in which they consumed foods that were either HFAT or HCHO. The ingredients of the foods provided during the meal days were covertly manipulated to be as follows: 10.4 (SD, 1.1) kJ·g−1, 37.7% carbohydrate, 54.4% fat, and 7.9% protein for HFAT, or 6.6 (SD, 0.8) kJ·g−1, 72.4% carbohydrate, 19.3% fat, and 8.3% protein for HCHO. Four hours after a standardized breakfast (ad libitum on baseline probe day and quantities consumed replicated on postintervention probe day; see meal details in Table, Supplemental Digital Content 1, Food items and macronutrient composition of the ad libitum HFAT and HCHO breakfast at baseline, http://links.lww.com/MSS/B826), the participants consumed a fixed-energy lunch composed of food items providing 3347 kJ (matched for weight across HFAT and HCHO; see meal details in Table, Supplemental Digital Content 2, Food items and macronutrient composition of the fixed energy HFAT and HCHO lunches, http://links.lww.com/MSS/B827). Foods were designed to be similar in appearance and palatability between conditions.

Food Reward. The Leeds Food Preference Questionnaire (LFPQ;[20]) was administered during the HFAT and HCHO meal days before and after fixed-lunch consumption (3347 kJ) to assess food reward behaviors (liking and wanting) in the hungry and fed states. The LFPQ computes scores of implicit wanting and explicit liking for high-fat (>50% energy) and low-fat (<20% energy) foods images matched for familiarity, sweetness, protein, and acceptability.

Before the procedure, screening of the images used in the task was completed by each participant to improve internal validity. If a participant did not know or recognize, or would never/rarely eat a particular food item used in the study, replacement images were chosen from a database of images of similar composition.

The LFPQ is composed of two tasks: one based on subjective ratings (explicit liking) and the other based on a forced-choice task (implicit wanting). During the measure of explicit liking, participants were presented with one food image at a time, in a randomized order, and rated the extent to which they like each food (How pleasant would it be to taste this food now?). Participants made their ratings using a 100-mm visual analog scale. Implicit wanting was assessed by asking participants to select as fast as possible between successive pairs of foods from specific categories the food "they most want to eat now." Scores for implicit wanting were computed from mean response times adjusted for frequency.[21] To calculate liking and wanting fat appeal bias as a measure of hedonic preference for high-fat relative to low-fat foods, low-fat scores were subtracted from high-fat scores; thus, a positive score indicates greater explicit liking/implicit wanting toward high-fat compared with low-fat foods. The LFPQ has been validated in a wide range of research (e.g., Ref.[22]

Twelve-week Exercise Intervention

During the 12-wk supervised exercise intervention (5 d·wk−1), each exercise session was individually prescribed to expend 2092 kJ at an intensity of 70% of age-predicted heart rate maximum, and to ensure compliance to the exercise prescription, the duration and intensity of each exercise session were recorded (Polar RS400, Polar, Finland). A selection of aerobic exercise equipment was available (i.e., treadmill, rower, cycle ergometer, and elliptical) from which the participants were free to choose and change within each session as long as they met the energy expenditure requirements. The duration needed to expend 2092 kJ at 70% heart rate maximum at baseline was calculated based on the relationship between heart rate, V̇O2, and V̇CO2 for each individual during an incremental maximal aerobic capacity (V̇O2max) test. This test was also performed at week 6 of the intervention to account for changes in energy metabolism and after the intervention to assess overall changes in cardiorespiratory fitness with the intervention. Total exercise-induced energy expenditure during the intervention was 116.98 ± 15.56 MJ, which represented >98% of the prescribed exercise-induced energy expenditure. V̇O2max was not measured in the controls.

Statistical Analyses

Data are presented as mean (SD), unless specified otherwise. Data were analyzed using the statistical package SPSS version 21. Data were checked for outliers before statistical analyses, and one of the controls had a change score (baseline to postintervention) in both liking and wanting that was 5 SD below the mean; therefore, this participant was excluded from the analysis. Independent-sample t-tests were used to evaluate differences in participant characteristics at baseline. Repeated-measures ANOVA with group (exercisers, controls), week (baseline, postintervention), condition (HFAT, HCHO), and state (hungry, fed), where appropriate, were used to assess changes in outcome variables. Where appropriate, Greenhouse–Geisser probability levels were used to adjust for nonsphericity, and post hoc analyses were performed using the Bonferroni adjustment for multiple comparisons. Where missing data were present, completers and intent-to-treat analyses (ITT) were conducted with the last observation carried forward method. To assess the associations among changes in food reward (overall mean of the two conditions and two states), eating behavior traits and body composition, Pearson's correlations were conducted in the whole group and in exercisers and controls separately.

The magnitude of the mean weight change (exercisers–controls) was interpreted against a minimal clinically important difference of 2.5 kg,[23] where a small clinically important effect was defined between 2.5 and 7.5 kg (3 × 2.5 kg), a moderate effect between 7.5 and 15 kg (6 × 2.5 kg), and a large effect >15 kg.[24] The magnitude of the mean waist circumference change was interpreted against a minimal clinically important difference of 2 cm,[4,25] where a small clinically important effect was defined between 2 and 6 cm, a moderate effect between 6 and 9 cm, and a large effect >9 cm.[24]

Following the American Statistical Association's policy statement on P values,[26] all P values from specified statistical models were reported along with effect size and confidence intervals (CI) to help determine compatibility of the data with the interpretation of findings. We have avoided referring to any outcome as "statistically significant" on the basis of a particular P value. Estimated marginal mean differences (MΔ) are reported (MΔEx–C, exercisers–controls; MΔPre–Post, postintervention–baseline; MΔHF–HC, HFAT–HCHO; MΔH–Fed, fed–hungry), as well as effect sizes as partial η squared (η2p) and 95% CI of the mean difference (95% CI). Because there are no benchmark values for η2p in the context of repeated measures designs,[27] the effect sizes were interpreted cautiously alongside the mean differences within the field of human appetite.

Based on G*Power (v3.1), in order to detect an interaction in liking or wanting (η2 p = 0.03) between two groups and two repeated measurements (r ≈ 0.8, based on prior data from our research group[9]) with α = 0.05 and 1 − β = 0.8, a total sample size of 28 was required.