Exercise intervention modulates food reward in obesity: efficacy, mechanisms, and prospects
1School of Physical Education, Yunnan Normal University, Kunming, China
2School of Physical Education, Shihezi University, Shihezi, China
3Department of Scientific Research, Henan Sport University, Zhengzhou, China
4School of Physical Education, Yunnan Minzu University, Kunming, China
*Correspondence: Yucheng Jiang, 2427187828@qq.comAbstract
Aberrant food reward processing is a key contributor to the development and maintenance of obesity, and exercise intervention has been proposed as a feasible way to correct this maladaptive eating pattern. Nevertheless, inconsistent research findings mean its real efficacy and internal regulatory mechanisms remain controversial. This work adopted a systematic review design following standardized literature screening and bias evaluation procedures to sort out all qualified controlled trials focusing on exercise-induced changes in food reward among obese individuals, then synthesized existing evidence and conducted subgroup analyses to explore multiple potential moderators. Our synthesis demonstrated that exercise is effective at modulating aberrant food reward, with its outcomes jointly shaped by exercise protocols, personal traits and study design. Three interrelated regulatory axes were identified: (i) the cognitive-neural pathway that tunes mesolimbic dopamine, opioid and endocannabinoid signaling as well as prefrontal and hippocampal function; (ii) the physiological-endocrine pathway that balances gut hormones and microbiota while mitigating central insulin and leptin resistance and chronic inflammation; (iii) the psychological-behavioral pathway that strengthens inhibitory control, reduces food attentional bias and improves emotional regulation. Noticeable limitations including high outcome heterogeneity, insufficient causal evidence and the absence of personalized exercise schemes still restrict current research progress. We recommend future research to combine multi-dimensional reward measurements, multimodal omics and causal inference to unpack core mechanisms, and build phenotype-matched exercise regimens to advance obesity intervention from population-based strategies to individualized precision treatment.
1Introduction
Obesity has become a major global public health challenge. According to the World Health Organization, in 2022, the number of children and adolescents aged 5–19 years with obesity exceeded 160 million worldwide, and the prevalence of obesity in this age group increased from 2% in 1990 to 8% in 2022. During the same period, the number of adults with obesity surpassed 890 million, with approximately 16% of adults aged 18 years and older living with obesity, and the global prevalence of adult obesity more than doubled compared with 1990 (1). Among the contributing factors, aberrant eating behavior, as a core pathogenic factor in the development and maintenance of obesity, can substantially exacerbate obesity risk through multiple pathways, including disruption of physiological homeostasis and heightened neural activation (2). Therefore, shifting the focus of obesity prevention and treatment from mere weight control to the correction of aberrant eating behaviors is key to moving the prevention and control gateway upstream and transforming management strategies toward a prevention-oriented and precision intervention paradigm.
The core feature of aberrant eating behavior in individuals with obesity lies in maladaptive food reward processing, manifested as heightened reward sensitivity to food cues, with its underlying mechanisms involving multisystem dysregulation across cognitive-neural, physiological-endocrine, and psychological-behavioral domains. Exercise intervention, as a non-pharmacological approach that integrates both psychological modulation and physiological enhancement, holds promise for correcting this aberrant behavior through the synergistic modulation of multisystem functions. However, current research findings on the effects of exercise on food reward remain markedly controversial, with different exercise regimens producing varying and even contradictory effects on food reward behaviors (3–6). This complex relationship not only engenders uncertainty regarding the efficacy of exercise interventions but also impedes the elucidation of the underlying mechanisms.
In light of this, through a systematic review and critical appraisal of the empirical evidence on the effects of exercise on food reward processing, this review aims to elucidate the modulatory effects of exercise interventions on food reward, to delineate the potential neural, physiological, and psychological mechanisms, and to clarify current research limitations and future directions. We seek to deepen the theoretical understanding of the mechanisms underlying food reward processing in individuals with obesity and to provide an empirical basis and theoretical reference for developing exercise-based weight management strategies in practice.
2Methods
2.1Literature search strategy
This study strictly followed the PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. A combination of subject headings and free-text terms was used to systematically search eight Chinese and English databases. The Chinese databases included CNKI, VIP, Wanfang, and SinoMed; the English databases included Web of Science, PubMed, Embase, and the Cochrane Library. In addition, a supplementary search was conducted by manually screening the reference lists of the included studies to minimize the risk of overlooking relevant studies. The search covered the period from the inception of each database to May 15, 2026.
The search terms were structured around three core dimensions: “exercise intervention,” “food reward,” and “obesity.” The search terms for exercise intervention were sport OR athletics OR exercise OR training OR fitness OR physical exercise OR physical activity OR physical education OR leisure activity OR recreation. The search terms for food reward were reward OR food reward OR hedonic eating OR food craving OR food preference OR food choice OR appetite OR eating behavior OR feeding behavior. The search terms for obesity were overweight OR obesity OR obese. Terms within each dimension were combined with the Boolean operator OR, and the three dimensions were combined with AND. The search strategy was adapted to the syntactic conventions of each database. Taking PubMed as an example, the complete search string is as follows:
2.2Inclusion and exclusion criteria
Inclusion criteria: (1) Population: Individuals with a clear diagnosis of overweight or obesity based on internationally recognized BMI criteria. (2) Intervention: Any form of single or structured exercise intervention program. (3) Comparator: No exercise intervention (e.g., usual care, health education, waitlist) or an alternative exercise intervention differing in intensity or type from the intervention group. (4) Outcomes: Reporting of at least one quantitative assessment measure related to food reward. (5) Study design: Randomized controlled trials or non-randomized controlled trials.
Exclusion criteria: (1) Qualitative studies, reviews, meta-analyses, conference abstracts, editorials, commentaries, and case reports. (2) Animal experiments, in vitro cell experiments, and other non-human experimental studies (such studies are cited only as supporting evidence in the mechanistic discussion and are not included in the formal effectiveness analysis). (3) Duplicate publications of the same study data (only the version with the most complete data or the earliest publication date is included). (4) Studies for which the full text is unavailable or data reporting is incomplete, and the information remains unobtainable after contacting the authors. (5) Non-Chinese or non-English language publications.
2.3Study selection and data extraction
Two uniformly trained reviewers independently performed the study selection according to the inclusion and exclusion criteria. The specific process was as follows: First, bibliographic records retrieved from each database were imported into EndNote X9 reference management software for automatic deduplication, supplemented by manual verification. Second, titles and abstracts were screened to exclude obviously irrelevant studies, and the number of excluded records was recorded. Finally, full texts were obtained for further screening, reasons for exclusion were documented, and eligible studies were ultimately included. Disagreements during the screening process were resolved by discussion; if consensus could not be reached, a third reviewer arbitrated.
Prior to formal data extraction, a standardized data extraction form was developed based on the PICOS framework of this study, with reference to the data extraction specifications in the Cochrane Handbook for Systematic Reviews of Interventions and the core outcome indicators of this study. After the initial draft of the form was completed, two reviewers independently conducted a pilot extraction on three sample studies. Consistency assessment was performed for the categorical items in the form, and the inter-rater agreement yielded a Cohen’s Kappa > 0.80, indicating good consistency. Based on the pilot extraction results, variable definitions, field formats, and value assignment rules were optimized and refined, and the form was finalized after review by the entire research team. All data were independently extracted by two reviewers and then cross-checked, with final verification by a third reviewer. When key information was missing or unclearly reported, the corresponding author of the original study was contacted whenever possible to obtain supplementary data.
The extracted information included: (1) Population: total sample size, sample sizes of the intervention and control groups, participant age (mean ± standard deviation or age range), BMI (mean ± standard deviation or z-score); (2) Intervention: type of exercise intervention, total intervention duration, frequency, session duration, exercise intensity; (3) Comparator: specific content of the control condition; (4) Outcomes: quantitative measures related to food reward (e.g., explicit liking, explicit wanting, implicit wanting, and relative preference as assessed by the Leeds Food Preference Questionnaire); (5) Study design: study type (e.g., randomized controlled trial, non-randomized controlled trial), time points of food reward assessment (e.g., pre-meal/post-meal), risk of bias assessment results; (6) Other information: first author, year of publication, country where the study was conducted, etc.
2.4Risk of Bias assessment
Based on the study design of the included literature, corresponding internationally standardized tools were used to assess the risk of bias. For randomized controlled trials, the Cochrane Risk of Bias tool (RoB 1.0) was employed, evaluating seven domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias. Each domain was rated as “low risk,” “unclear risk,” or “high risk.” For non-randomized intervention studies, the Risk Of Bias In Non-randomized Studies – of Interventions (ROBINS-I) tool was applied, assessing seven domains: confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selective reporting. Each domain was rated as “low risk,” “moderate risk,” “serious risk,” or “critical risk.” The overall risk of bias was determined according to the ROBINS-I criteria: an overall “low risk” rating was assigned if all domains were low risk; an overall “moderate risk” rating was assigned if all domains were “low risk” or “moderate risk”; an overall “serious risk” rating was assigned if at least one domain was “serious risk” and no domain was “critical risk”; and an overall “critical risk” rating was assigned if any domain was “critical risk.” All risk of bias assessments were performed independently by two reviewers. Prior to the assessment, reviewers received uniform training to ensure consistent understanding of the criteria. During the assessment, each reviewer independently reviewed the full texts, evaluated each domain, and recorded the judgments. After completion, results were cross-checked. Any disagreements were resolved through discussion or, when necessary, by arbitration with a third reviewer to reach consensus.
3Results
3.1Study selection process
3.2Basic characteristics of the included literature
After screening, 15 publications were included (5–19), comprising 25 intervention groups that reported the independent effects of exercise interventions on food reward-related outcomes in individuals with obesity. The included studies were published between 2012 and 2024 and were conducted across multiple countries, including the United States, the United Kingdom, France, Norway, and Australia. Study designs included both randomized controlled trials and non-randomized controlled trials. The sample sizes of the intervention groups ranged from 12 to 120 participants. Exercise intervention protocols encompassed both acute and chronic exercise. The intervention modalities included aerobic exercise and interval training, with intervention durations ranging from 4 to 24 weeks, session durations of 10–60 min, and frequencies ranging from once per week to more than five times per week. Exercise intensity varied from moderate to high. The time points for food reward assessment ranged from pre-meal to several hours post-meal, and also encompassed multiple time windows from immediately post-exercise to several hours post-exercise. The primary tools for assessing food reward included the LFPQ, the PFS, and fMRI to measure participants’ neural responses to food cues. The basic characteristics of the included studies are shown in Table 1.
| Study | Country | Age, BMI (kg/m2) | n (women) | Control | Exercise intervention program | Assessment | Intervention outcomes | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CG | IG | Intervention | Type | Period (wks) | Session (min) | Freq (wk−1) | Intensity | Timing | Measures | |||||
| Ross et al. (5) | USA | 50.76 ± 10.38 31.19 ± 4.41 | 75(52) | NR | AE | 2 | 12 | NR | 1 | moderate intensity | NR | PFS | Food reward sensitivity ↓ | |
| Beaulieu et al. (6) | UK | CG:41.4 ± 10.7, 31.4 ± 3.7 IG: 43.2 ± 7.5, 30.6 ± 3.8 | 15 (9) | 46 (30) | No exercise control | AE | 2 | 12 | NR | 5 | 70% HRmax | NR | LFPQ | Explicit liking for high-fat foods – Explicit wanting for high-fat foods NR Implicit wanting for high-fat foods ↓ Relative preference for high-fat foods NR |
| Miguet et al. (7) | France | 13.3 ± 0.9 2.3 ± 0.2 (BMIz) | 33 (21) | 33 (21) | No exercise control | HIIT | 1 | NR | 15 | NR | 70–90% HRmax | PreM (Ex30)/ PostM0 (Ex30+) | LFPQ | PreM (Ex30): all outcomes – PostM0 (Ex30+): Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods ↓ Relative preference for high-fat and sweet foods ↓ |
| Fillon et al. (8) | France | 12.8 ± 1.4 2.2 ± 0.4 (BMIz) | 17 (8) | 17 (8) | No exercise control | AE (Exercise pre-meal group) | 1 | NR | 30 | NR | 65% VO2peak | PreM (Ex0)/ PostM0 (Ex60) | LFPQ | PreM (Ex0): all outcomes – PostM0 (Ex60): Explicit liking for high-fat foods ↓; Explicit liking for sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods NR |
| Fillon et al. (8) | France | 12.8 ± 1.4 2.2 ± 0.4 (BMIz) | 17 (8) | 17 (8) | No exercise control | AE (Exercise post-meal group) | 1 | NR | 30 | NR | 65% VO2peak | PostM0 (Ex60) | LFPQ | Explicit liking for high-fat foods ↓; Explicit liking for sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods NR |
| Pelissier et al. (9) | France | 13 ± 1 35.7 ± 4.1 98.8 ± 0.7 (BMI percentile) | 17 (11) | 17 (11) | No exercise control | AE | 1 | NR | 38 | NR | 65% VO2peak | PreM | LFPQ | Explicit liking for high-fat foods –; Explicit liking for sweet foods ↓ Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – |
| Siroux et al. (10) | France | 13.1 ± 1.3 4.48 ± 0.25 (BMIz) | 13 (8) | 13 (8) | No exercise control | AE | 1 | NR | 30 | NR | 65% VO₂peak | PreM (Ex45)/ PostM0 (Ex45+)/ PostM6h (Ex7.5 h) | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – |
| Martins et al. (11) | Norway | 33.4 ± 10.0 32.3 ± 2.7 | 12 (7) | 12 (7) | No exercise control | AE | 1 | NR | 27 | NR | 70% HRmax | PreM (Ex0) | LFPQ | Explicit liking for high-fat foods – Explicit wanting for high-fat foods – Implicit wanting for high-fat foods – Relative preference for high-fat foods – |
| Martins et al. (11) | Norway | 33.4 ± 10.0 32.3 ± 2.7 | 12 (7) | 12 (7) | No exercise control | HIIT | 1 | NR | 18 | NR | 85–90% HRmax | PreM (Ex0) | LFPQ | Explicit liking for high-fat foods – Explicit wanting for high-fat foods – Implicit wanting for high-fat foods – Relative preference for high-fat foods – |
| Martins et al. (11) | Norway | 33.4 ± 10.0 32.3 ± 2.7 | 12 (7) | 12 (7) | No exercise control | HIIT | 1 | NR | 9 | NR | 85–90% HRmax | PreM (Ex0) | LFPQ | Explicit liking for high-fat foods – Explicit wanting for high-fat foods NR Implicit wanting for high-fat foods – Relative preference for high-fat foods – |
| Fillon et al. (12) | France | 13.1 ± 1.4 2.3 ± 0.3 (BMIz) | 15 (9) | 15 (9) | No exercise control | AE (Ex60 pre-meal group) | 1 | NR | 30 | NR | 65% VO2peak | PreM (Ex45)/ PostM0 (Ex60+) | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – |
| Fillon et al. (12) | France | 13.1 ± 1.4 2.3 ± 0.3 (BMIz) | 15 (9) | 15 (9) | No exercise control | AE (EX180 pre-meal group) | 1 | NR | 30 | NR | 65% VO2peak | PreM (Ex165)/ PostM0 (Ex180+) | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – |
| Fillon et al. (13) | France | 12.7 ± 1.3 2.2 ± 0.4 (BMIz) | 18 (6) | 18 (6) | No exercise control | AE (Ex90 pre-meal group) | 1 | NR | 30 | NR | 65% VO2peak | PreM (Ex15)/ PostM0 (Ex45)/ PostM6h15 (Ex6h45) | LFPQ | Explicit liking for high-fat foods – Explicit wanting for high-fat foods NR Implicit wanting for high-fat foods– Relative preference for high-fat foods NR |
| Fillon et al. (13) | France | 12.7 ± 1.3 2.2 ± 0.4 (BMIz) | 18 (6) | 18 (6) | No exercise control | AE (Ex90 pre-meal group) | 1 | NR | 30 | NR | 65% VO2peak | PreM (Ex75)/ PostM15 (Ex105)/ PostM5h15 (Ex6h45) | LFPQ | PreM (Ex75)/PostM15 (Ex105): all outcomes – PostM5h15 (Ex6h45): Explicit liking for high-fat foods↓ Explicit wanting for high-fat foods NR Implicit wanting for high-fat foods – Relative preference for high-fat foods NR |
| Cornier et al. (14) | USA | 38.2 ± 9.5 33.3 ± 4.3 | 12(5) | NR | AE | 2 | 24 | 15–60 | 5 | 60–75% VO₂max | NR | fMRI | Neural responses to food cues in bilateral parietal lobes, left insula, etc. ↓ | |
| Hopkins et al. (15) | UK | 18–55 30.8 ± 3.5(women) 30.5 ± 4.7(men) | 46(30) | NR | AE | 2 | 12 | NR | 5 | 70% HRmax | NR | LFPQ | Explicit liking for high-fat foods – Explicit wanting for high-fat foods NR Implicit wanting for high-fat foods – Relative preference for high-fat foods NR | |
| Martins et al. (16) | Norway | 34.4 ± 8.8 33.3 ± 2.9 | 14 | NR | AE | 2 | 12 | 32 | 3 | 70% HRmax | NR | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – | |
| Martins et al. (16) | Norway | 34.4 ± 8.8 33.3 ± 2.9 | 16 | NR | HIIT | 2 | 12 | 20 | 3 | 85–90% HRmax | NR | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – | |
| Martins et al. (16) | Norway | 34.4 ± 8.8 33.3 ± 2.9 | 16 | NR | HIIT | 2 | 12 | 10 | 3 | 85–90% HRmax | NR | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – | |
| Thivel et al. (17) | France | 12–16 31.8 ± 3.8 | 12(6) | NR | AE (Concentric exercise group) | 2 | 12 | NR | 3 | NR | NR | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods – | |
| Thivel et al. (17) | France | 12–16 34.8 ± 5.5 | 12(6) | NR | AE (Eccentric exercise group) | 2 | 12 | NR | 3 | NR | NR | LFPQ | Explicit liking for high-fat and sweet foods – Explicit wanting for high-fat and sweet foods – Implicit wanting for high-fat foods –; Implicit wanting for sweet foods ↓ Relative preference for high-fat foods ↑; Relative preference for sweet foods ↓ | |
| Martin et al. (18) | USA | 48.9 ± 11.4 31.5 ± 4.7 | 61 (45) | 59 (43) | NR | AE (Low-dose group) | 2 | 24 | NR | ≥5 | 65–85%VO₂peak | NR | FPQ | Explicit liking for high-fat and sweet foods NR Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods NR Relative preference for high-fat and sweet foods – |
| Martin et al. (18) | USA | 48.9 ± 11.4 31.5 ± 4.7 | 61 (45) | 59 (43) | NR | AE (High-dose group) | 2 | 24 | NR | ≥5 | 65–85%VO₂peak | NR | FPQ | Explicit liking for high-fat and sweet foods NR Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods NR Relative preference for high-fat and sweet foods ↓ |
| Alkahtani et al. (19) | Australia | 29 ± 3.7 30.7 ± 3.4 | 10 (0) | 10 (0) | NR | MIIT | 2 | 4 | 37.5 | 3 | 45 ± 20% VO₂peak | NR | LFPQ | Explicit liking for high-fat foods ↑ Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods NR |
| Alkahtani et al. (19) | Australia | 29 ± 3.7 30.7 ± 3.4 | 10 (0) | 10 (0) | NR | HIIT | 2 | 4 | 18.7 | 3 | 90% VO₂peak | NR | LFPQ | Explicit liking for high-fat foods ↓ Explicit wanting for high-fat and sweet foods NR Implicit wanting for high-fat and sweet foods – Relative preference for high-fat and sweet foods NR |
3.3Risk of Bias assessment results
Of the 15 included studies, 11 were randomized controlled trials and were assessed for risk of bias using the Cochrane Risk of Bias tool (RoB 1.0). The remaining four studies were non-randomized controlled trials and were evaluated using the ROBINS-I tool. Overall, the risk of bias in the 15 included studies stemmed primarily from two sources: in the randomized controlled trials, the key limitations were insufficient reporting of randomization details and the inherent challenges of blinding; in the non-randomized intervention studies, the key limitation was confounding bias arising from single-arm designs or baseline imbalance between groups. The detailed risk-of-bias assessments for each domain are presented in Tables 2, 3.
| Study | Random sequence generation | Allocation concealment | Blinding of participants and personnel | Blinding of outcome assessment | Incomplete outcome data | Selective reporting | Other sources of bias |
|---|---|---|---|---|---|---|---|
| Miguet et al. (7) | Unclear | Unclear | High | Low | Low | Low | Low |
| Fillon et al. (8) | Unclear | Unclear | High | Low | Low | Low | Low |
| Pelissier et al. (9) | Unclear | Unclear | High | Low | Low | Low | Low |
| Siroux et al. (10) | Unclear | Unclear | High | Low | Unclear | Low | Low |
| Martins et al. (11) | Unclear | Unclear | High | Low | Low | Low | Low |
| Fillon et al. (12) | Unclear | Unclear | High | Low | Low | Low | Low |
| Fillon et al. (13) | Unclear | Unclear | High | Low | Low | Low | Low |
| Martins et al. (16) | Unclear | Unclear | High | Low | Unclear | Low | Unclear |
| Thivel et al. (17) | Unclear | Unclear | High | Low | Unclear | Low | Unclear |
| Martin et al. (18) | Low | Low | High | Low | Unclear | Low | Low |
| Alkahtani et al. (19) | Unclear | Unclear | High | Low | Low | Low | Low |
| Study | Confounding | Selection of participants | Classification of interventions | Deviations from intended interventions | Missing data | Measurement of outcomes | Selective reporting | Overall risk of bias |
|---|---|---|---|---|---|---|---|---|
| Ross et al. (5) | Moderate | Low | Low | Moderate | Low | Low | Low | Moderate |
| Beaulieu et al. (6) | Moderate | Low | Low | Moderate | Low | Low | Low | Moderate |
| Cornier et al. (14) | Moderate | Low | Low | Moderate | Low | Low | Moderate | Moderate |
| Hopkins et al. (15) | Moderate | Low | Low | Low | Moderate | Low | Moderate | Moderate |
3.4Effects of exercise intervention on food reward processing in individuals with obesity
From the 15 included studies, a total of 25 intervention arms reporting the independent effects of exercise interventions on food reward behavior in individuals with obesity were extracted. Among these, 10 intervention arms demonstrated an inhibitory effect on at least one food reward dimension, with no dimension showing enhancement; 13 intervention arms observed no significant changes in any food reward dimension; 1 intervention arm reported a mixed effect (a significant increase in one dimension and a significant decrease in another); and 1 intervention arm reported an enhancing effect on food reward.
With respect to individual food reward dimensions, 21 intervention arms reported the effects of exercise on explicit liking. Of these, 5 showed a significant decrease in explicit liking after exercise, 1 showed a significant increase in explicit liking for high-fat foods, and 15 observed no significant effect. Eleven intervention arms reported on explicit wanting, all of which observed no significant effect. Twenty-one intervention arms reported on implicit wanting; among them, 3 showed a significant decrease in implicit wanting after exercise, and 18 observed no significant effect. Fifteen intervention arms reported on relative preference; of these, 2 showed a significant decrease in relative preference after exercise, 1 reported a mixed effect (a significant decrease in relative preference for sweet foods and a significant increase in relative preference for high-fat foods), and 12 observed no significant effect.
With respect to exercise type, 12 intervention arms reported the effects of acute exercise on food reward. Of these, 5 showed an inhibitory effect on at least one food reward dimension, and 7 observed no significant effect. Thirteen intervention arms reported the effects of chronic exercise on food reward. Of these, 5 showed an inhibitory effect on at least one food reward dimension, 1 reported a mixed effect (a significant increase in one dimension and a significant decrease in another), 1 reported an enhancing effect on food reward, and 6 observed no significant effect.
4Discussion
4.1Mechanisms underlying the effects of exercise intervention on food reward processing in obesity
The accumulating evidence reviewed above supports the efficacy of exercise interventions in modulating food reward processing in individuals with obesity. Beyond this phenomenological level, however, a critical question concerns the mechanisms through which these effects are achieved; the following sections synthesize current evidence across multiple levels of analysis to outline a tentative mechanistic framework. Before examining the specific mechanisms, it is important to clarify the nature and hierarchy of the evidence discussed below. The mechanistic pathways proposed in this review are derived from evidence spanning multiple levels of inquiry, including rodent models of diet-induced obesity, human neuroimaging studies, and observational data from exercise interventions. While animal studies provide valuable insights into molecular and circuit-level mechanisms that are not feasible in humans, their translation to clinical practice requires caution. Throughout this section, findings from animal experiments are presented as candidate mechanisms rather than established causal pathways in humans, and we explicitly indicate where evidence is preliminary or requires further validation. Where human data are available, they are given greater interpretive weight, and we note the distinction between correlational neuroimaging findings and causal evidence.
4.1.1Cognitive-neural mechanisms
4.1.1.1Exercise modulates dopaminergic and glutamatergic pathways in the nucleus accumbens to ameliorate food reward function
The dopaminergic pathway projecting from the VTA to the NAc and the glutamatergic pathways projecting from the PFC and amygdala to the NAc have been identified in animal models as the core neural circuitry regulating food reward (20). In rodent models, chronic high-fat diet consumption induces compensatory suppression of the reward system, characterized by deficient DA function and excessive glutamatergic plasticity, thereby driving overeating to compensate for the reward deficit. Regular exercise, however, can normalize reward dysfunction by modulating these pathways, evidence that provides animal-level support for understanding the regulation of human food reward (21). Specifically, in the DAergic pathway, exercise has been shown to not only directly elevate DA levels in the NAc of obese mice, but also promote the expression of tyrosine hydroxylase (TH) in the substantia nigra pars compacta–dorsal striatum pathway and the VTA–NAc pathway, indirectly increasing striatal DA availability and effectively facilitating neurotransmission in these circuits (20). Concurrently, exercise upregulates D2R levels in the NAc of obese mice, ameliorating the downregulation of D2R expression and the aberrant VTA–NAc pathway activity linked to enhanced DA–D1R binding induced by long-term high-fat diet (21).
In the glutamatergic pathway, studies in obese rodents have shown that the number of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) lacking the GluA2 subunit on the cell membrane of NAc neurons is significantly increased. The high Ca2+ permeability of these receptors directly enhances glutamatergic synaptic transmission efficacy, thereby lowering the reward threshold (20). Exercise intervention not only directly reduces the number of AMPAR- and calcium-permeable AMPA receptor (CP-AMPAR)-positive cells on the NAc cell membrane surface, but also increases the phosphorylation level of the GluA2 subunit within AMPARs and its occupancy at synapses. The incorporation of the GluA2 subunit renders AMPARs impermeable to Ca2+, thereby attenuating the excessive encoding of highly rewarding foods by the NAc circuitry and reducing pathological eating motivation. Crucially, animal studies further reveal that the DAergic function restored by exercise can suppress the synaptic trafficking of CP-AMPARs, demonstrating an interactive mechanism through which DA signaling negatively regulates glutamatergic synaptic plasticity (22, 23).
These findings from animal models provide mechanistically plausible explanations for the exercise-induced changes in reward-related brain activity observed in human neuroimaging studies. However, it is important to recognize that the molecular and synaptic mechanisms identified in rodents represent candidate pathways rather than being definitively established in humans. Direct evidence linking these specific molecular changes to alterations in human reward behavior is currently lacking, and the translational gap between animal models and human obesity remains to be systematically addressed.
4.1.1.2Exercise modulates the opioid and endocannabinoid systems to ameliorate hedonic eating behavior
The EOS plays a central role in regulating hedonic eating and motivational behavior. Pathological obesity is frequently accompanied by a global reduction in brain MOR density, which results in an enhanced anticipatory reward response upon stimulation by food cues (24). In human intervention studies, aerobic exercise has been shown to induce upregulation of MOR expression, with this effect widely distributed across reward-related brain regions including the ventral striatum, medial prefrontal cortex, anterior cingulate cortex, and insula (25).
The ECS also plays an important role in hedonic eating behavior, and the obese state is often associated with dysregulated ECS signaling in key brain regions. Currently, human evidence on exercise-induced ECS modulation remains limited; however, findings from rodent studies provide valuable mechanistic insights into how exercise may regulate food reward processing. In the PFC, chronic exercise significantly upregulates CB1 receptor expression in rodents, enhancing local endocannabinoid signaling tone. This molecular change is hypothesized to strengthen top-down cognitive control, which may explain the improved food-related decision-making observed in human exercise trials (26). In the hippocampus, enhanced CB1 receptor signaling is proposed as a key molecular substrate for neuroplasticity related to dietary hedonic memory. In rodent models, chronic exercise specifically enhances hippocampal ECS function, evidenced by elevated expression of the CB1 receptor gene (Cnr1) alongside widespread upregulation of genes encoding synthetic and degradative enzymes. These changes collectively promote the turnover and signaling efficacy of hippocampal endocannabinoids (27). In the hypothalamus, exercise downregulates CB1 receptor expression in rodents, which attenuates CB1-mediated inhibition of GABAergic inputs to proopiomelanocortin (POMC) neurons in the arcuate nucleus. This in turn enhances anorexigenic signaling via α-melanocyte-stimulating hormone and reduces excitatory drive to orexigenic neurons in the lateral hypothalamic area (27, 28).
Notably, an interesting observation from animal studies is that long-term exercise and palatable diet induce similar ECS regulatory patterns in key brain regions—upregulation of CB1 receptor expression in the prefrontal cortex and hippocampus and downregulation of CB1 expression in the hypothalamus—yet, against this background, exercise attenuates weight gain. This phenomenon tentatively suggests that exercise may not simply substitute for food reward signals; rather, through neural adaptive mechanisms similar to those engaged by hedonic eating, it may form functional compensation in circuits involved in the integration of metabolism and reward, potentially contributing to a positive intervention against unhealthy eating behavior. However, this interpretation remains speculative, as it is derived from rodent data, and the extent to which these ECS-mediated mechanisms operate in humans—particularly in the context of exercise interventions for obesity—has yet to be determined.
4.1.2Physiological-endocrine mechanisms
4.1.2.1Exercise modulates gastrointestinal hormones and the gut microbiome to regulate food reward
Dysregulation of the energy intake control pathway in individuals with obesity constitutes an important physiological basis for subsequent hedonic eating. Regular exercise can exert beneficial effects on energy balance by modulating the function of the gut–brain axis, with appetite-regulating hormones and the gut microbiome playing key roles in this process. At the gastrointestinal hormone level, research has shown that acute high-intensity exercise can significantly alter neural responses to images of high- and low-calorie foods in reward-related brain regions of individuals with obesity, a change associated with decreased concentrations of acylated ghrelin and increased release of peptide YY (PYY) (32). Furthermore, multiple studies have confirmed that exercise can reduce acylated ghrelin levels and promote increased concentrations of various anorexigenic gut peptides, such as PYY, glucagon-like peptide-1 (GLP-1), and pancreatic polypeptide, in overweight/obese individuals (7, 33, 34). Evidence indicates that these hormonal changes are associated with attenuated neurocognitive responses of reward-related brain regions to visual food cues (32).
Beyond hormonal regulation, the gut microbiome serves as another key mediator of gut-brain axis effects on food reward processing. Observational studies in humans have shown that gut microbiome dysbiosis frequently coexists with food reward system dysfunction in individuals with obesity, though a direct causal link between the two has not been established in human populations. Preclinical rodent studies have demonstrated that both high-intensity interval training and cold-water swimming can effectively ameliorate high-fat diet-induced gut dysbiosis, as evidenced by increased α-diversity, enrichment of beneficial bacterial taxa (e.g., Lachnospiraceae and Prevotellaceae), and enhanced microbial metabolic functions (e.g., the tricarboxylic acid cycle activity) (35, 36). This finding from rodent models provides indirect preclinical evidence for the potential pathway whereby exercise modulates food reward processing via gut microbiota remodeling. The exercise-reshaped gut microbiota may, in turn, reduce the transmission of gut-derived inflammatory factors to the central nervous system, alleviate neuroinflammation in reward-related brain regions, and consequently modulate the function of key molecules involved in DA metabolism (e.g., TH and dopamine receptors), as well as indirectly influence central food reward perception and motivation through metabolites such as short-chain fatty acids (37).
It should be noted that, although existing evidence preliminarily suggests the potential for exercise to regulate food reward through the gut microbiome, direct causal evidence linking “microbiota alterations” to “reward behavior modulation” is still lacking. Furthermore, most of the mechanistic evidence linking the gut microbiome to food reward regulation is derived from animal models; whether analogous gut–brain–reward interactions occur in humans following exercise, and whether such effects are of sufficient magnitude to influence eating behavior, remain open questions. Future studies should integrate fecal microbiota transplantation, multi-omics approaches, and brain imaging techniques to systematically validate this pathway in both animal models and human populations.
4.1.2.2Exercise alleviates chronic inflammation to restore reward circuitry function
The systemic chronic inflammation triggered by excessive adipose tissue accumulation constitutes an important pathological mechanism underlying aberrant food reward processing in individuals with obesity. Studies have demonstrated that inflammatory cytokines can disrupt the function of reward-related neural circuits through multiple pathways. Specifically, interleukin-6 (IL-6) can inhibit DA synthesis by reducing BH₄ bioactivity, while tumor necrosis factor-α (TNF-α) decreases DA release, accelerates its reuptake, and induces MOR desensitization and epigenetic suppression (38). Moreover, TNF-α also impairs AMPAR/NMDAR function at glutamatergic synapses (39). At the metabolic level, inflammatory cytokines such as TNF-α, IL-6, and interleukin-1β (IL-1β) can induce central insulin resistance and leptin resistance, thereby disrupting the integration of energy and reward signals (39, 40). Collectively, these alterations lead to attenuated reward signaling, diminished pleasure, and aberrant eating motivation. Multiple studies have shown that exercise intervention can significantly improve the inflammatory profile of overweight and obese populations, as evidenced by reductions in pro-inflammatory factors such as high-sensitivity C-reactive protein (hs-CRP), IL-6, TNF-α, and IL-1β, accompanied by enhanced secretion of anti-inflammatory cytokines such as interleukin-10 (IL-10) (41, 42). This suggests that exercise may indirectly modulate the food reward system by regulating the inflammatory state. However, direct evidence supporting the exercise–inflammation–reward modulation pathway remains lacking, and the specific molecular and neural circuit mechanisms await further investigation.
4.1.2.3Exercise ameliorates insulin and leptin signaling pathways to optimize food reward regulation
Obesity-related insulin resistance induced by long-term high-fat diet consumption leads to compensatory elevations in peripheral and central insulin levels, which in turn impair insulin signal transduction in the NAc. This impairment is specifically manifested as disrupted phosphorylation of the insulin receptor (InsR) and its substrate insulin receptor substrate 2 (IRS2), as well as reduced activation of the downstream key effector protein kinase B (Akt), thereby weakening the regulatory capacity of insulin over DA release (43). Studies have demonstrated that regular aerobic exercise can effectively reverse this signaling deficit by upregulating InsR/IRS2 expression and enhancing Akt phosphorylation in the NAc, thus restoring insulin signaling sensitivity (43, 44). The restored insulin signaling pathway can further inhibit downstream glycogen synthase kinase-3β (GSK3β) activity (43) and promote the expression of the DA synthesis rate-limiting enzyme TH in the VTA via insulin signaling (44), thereby optimizing the function of the DAergic reward pathway, reducing excessive seeking of high-fat foods, and facilitating weight control and the restoration of metabolic homeostasis.
Leptin, as a key adipokine conveying energy storage information to the central nervous system, not only regulates energy balance by stimulating the hypothalamic arcuate nucleus to induce satiety, but also participates in inhibiting food-related reward by acting on the VTA (45). The precise mechanisms through which leptin resistance develops and impacts reward circuitry have been extensively studied in animal models. In obese rodents, dysregulated adipose tissue secretion triggers hyperleptinemia and central leptin resistance, rendering leptin unable to effectively suppress the hypothalamic feeding center or constrain the excessive responsiveness of the mesolimbic system to highly rewarding foods, thereby manifesting as failed satiety signaling and enhanced reward-driven eating (46). A meta-analysis has confirmed that aerobic, resistance, and combined exercise can all significantly reduce circulating leptin levels in overweight/obese individuals (47). This effect, combined with findings from animal models, suggests that exercise may reverse central leptin resistance—not merely through passive reductions in leptin secondary to fat loss, but potentially by restoring the sensitivity of reward-related brain regions to leptin signaling. However, this mechanistic pathway is largely derived from animal studies on how leptin resistance develops and affects reward circuitry. Whether these mechanisms operate similarly in humans directly determines whether exercise-induced changes in leptin sensitivity translate into meaningful improvements in food reward behavior in individuals with obesity—a question that requires further investigation.
4.1.3Psychological-behavioral mechanisms
4.1.3.1Exercise enhances inhibitory control to reduce food reward sensitivity
Individuals with obesity often exhibit deficits in PFC-mediated inhibitory control, leading to heightened impulsive responses to highly rewarding food cues. Exercise has been shown to enhance food-related inhibitory control, thereby helping individuals with obesity overcome their elevated food reward sensitivity and reduce impulsive eating behavior. Empirical studies have demonstrated that both a single bout of high-intensity interval exercise and continuous moderate-intensity aerobic exercise can improve the inhibitory control of overweight individuals in response to high-calorie food stimuli (48). This benefit may be associated with exercise-induced increases in dlPFC blood flow and changes in neurotransmitter release (49). Further research has revealed that, following a single bout of high-intensity interval exercise, individuals not only exhibit enhanced inhibitory control but also show significantly reduced delay discounting of food rewards, with concurrent decreases in hedonic evaluation of high-energy-density foods and the tendency to select ideal portion sizes (49). This suggests that exercise-enhanced inhibitory control can effectively intervene in and modify the rapid reward evaluation and decision-making processes triggered by food cues, thereby reducing impulsive eating behavior by improving the cognitive efficacy of dietary self-regulation.
4.1.3.2Exercise modulates attentional Bias to reduce the salience of high-calorie food cues
Individuals with obesity generally exhibit an attentional approach bias toward high-calorie foods, characterized by attention being easily captured by food cues and difficulty in disengaging. This automatic cognitive processing pattern significantly increases the risk of cue-induced craving and overeating in food-rich environments. Exercise can serve as an effective intervention to reshape the attentional processing of food cues. Studies have shown that acute exercise can enhance the attention of individuals with obesity toward the health attributes of food while reducing their attentional bias toward high-calorie food cues (50, 51). Neuroelectrophysiological evidence further reveals that, following acute resistance exercise, overweight/obese individuals exhibit shortened latencies of the N2 and P3 components and reduced N2 amplitude when processing food cues. This indicates that exercise can not only accelerate the speed of attentional orientation and allocation toward food cues in overweight/obese individuals, but also reduce the automatic attentional capture by food-related stimuli (51). Such alterations in attentional processing patterns may be related to the attenuated responses of brain regions involved in reward, memory, and visual processing to high-calorie food cues, with BDNF potentially playing a key mediating role (52, 53).
4.1.3.3Exercise improves emotion regulation to alleviate emotional eating
Emotional eating is a key psychological-behavioral factor in the development and maintenance of obesity, essentially representing the use of highly rewarding food intake as a maladaptive strategy for coping with negative emotions. Regular exercise, through its mood-enhancing benefits, can help individuals with obesity rebuild healthy emotion regulation strategies, thereby disrupting the aberrant association between negative emotions and food reward. A large body of evidence demonstrates that exercise can effectively alleviate symptoms of anxiety and depression, enhance stress coping capacity, and promote psychological resilience—benefits that have direct implications for the improvement of emotional eating behavior in individuals with obesity. A 12-week exercise intervention significantly reduced anxiety symptoms and the consequent binge eating behavior in individuals with obesity (54). A 10-month intervention study further found that exercise not only reduced emotional eating behavior, but also decreased post-consumption food liking and attenuated reward responsivity during the transition from hunger to satiety (55). This indicates that the effects of exercise extend beyond behavioral suppression of eating to fundamentally reshape the hedonic system, weakening the reward value of food as an emotional compensator and thereby reestablishing the balance between the homeostatic and hedonic systems (see Figure 2).
5Conclusions, limitations, and future directions
Existing evidence indicates that the modulation of food reward by exercise in individuals with obesity is characterized by both effect complexity and mechanistic systematicity. At the effect level, exercise intervention represents a potential means to regulate food reward behavior in individuals with obesity, with its effects moderated by multiple factors across three domains: (1) exercise intervention protocol-related factors, including exercise modality, intervention duration, frequency, session duration, exercise intensity, and energy expenditure (energy balance); (2) individual characteristics-related factors, including sex, age, body composition, metabolic health status, baseline physical activity level, habitual dietary patterns, exercise experience, and psychiatric comorbidities; and (3) study design-related factors, including the timing of food reward assessment, concomitant pharmacological treatment, and the food environment. At the mechanistic level, exercise intervention can achieve synergistic regulation through integrated neural-physiological-psychological multisystem pathways. In terms of cognitive-neural mechanisms, exercise can modulate the mesolimbic DA pathway, regulate EOS/ECS function, and restore the function of brain regions including the PFC and hippocampus, thereby achieving the integration of reward-related neural information. In terms of physiological-endocrine mechanisms, exercise can regulate gastrointestinal hormones and gut microbiota function, alleviate chronic low-grade inflammation, and ameliorate central insulin and leptin resistance, restoring the normal regulation of reward circuitry by metabolic signals. In terms of psychological-behavioral mechanisms, exercise can enhance inhibitory control, modulate attentional bias toward food cues, and improve emotion regulation, thereby attenuating the psychological drive for hedonic eating.
While systematically elucidating the modulatory effects of exercise on food reward in individuals with obesity, this review also reveals core limitations that urgently need to be addressed. First, the heterogeneity and inconsistency of exercise intervention effects are prominent, with similar exercise protocols even inducing opposite changes in food craving across different studies, suggesting that the individualized moderating variables influencing the exercise–reward relationship have yet to be effectively identified and integrated. Second, the key pathways through which exercise regulates the reward system have not been fully elucidated; although existing studies suggest that exercise may influence reward processing by improving inflammatory status and modulating the gut microbiota, direct causal evidence linking these pathways to improvements in reward behavior remains lacking. Third, the scientific basis for exercise prescription is weak; although current research has confirmed the moderating roles of factors such as intensity and type, the construction of personalized exercise prescriptions based on the core components of food reward (“liking” and “wanting”) still lacks a theoretical framework and empirical foundation. Fourth, the available evidence predominantly focuses on aerobic exercise and interval training, whereas intervention studies on resistance exercise remain extremely limited, constraining head-to-head comparisons of effects between different exercise modalities and the analysis of mechanistic differences. Fifth, the analysis of moderating variables is primarily based on semi-quantitative synthesis through subgroup stratification and qualitative discussion, lacking rigorous quantitative testing; moreover, the limited sample sizes within certain subgroups and the presence of confounding among variables to varying degrees mean that the observed moderating trends should be interpreted as exploratory findings. Sixth, the included primary studies commonly present inherent limitations such as inadequate reporting of randomization details and the intrinsic difficulty of blinding participants in exercise interventions, which introduced some degree of risk of bias. This reduces the overall strength of the evidence, weakening the robustness of the conclusions, and constraining the external validity of the findings.
Future research should strive to promote a scientific paradigm shift from population-level effects to individualized responsiveness in exercise interventions. Specifically, Future research should strive to promote a scientific paradigm shift from population-level effects to individualized responsiveness in exercise interventions. Specifically, First, systematically compare the differential effects of exercise parameters, individual baseline characteristics, and study design factors on various dimensions of food reward, evaluate the moderating effects of these variables, and identify the core sources of effect heterogeneity. Second, integrate multimodal behavioral, physiological, and neuroimaging data to elucidate the neural, physiological, and psychological mechanisms through which exercise modulates food reward. Third, develop a personalized exercise parameter matching system based on the “liking” and “wanting” reward phenotypes to inform precision exercise prescription. Fourth, conduct head-to-head comparisons of resistance exercise with aerobic and combined exercise in populations with obesity, incorporating measurements of myokines and skeletal muscle endocrine function to clarify the modality-specific modulatory effects of different exercise modalities. Fifth, strictly adhere to clinical trial reporting standards, fully disclose key methodological details such as randomization and allocation concealment, and implement outcome assessor blinding whenever feasible to reduce methodological bias and enhance the strength of evidence.
Data availability statement
Publicly available datasets were analyzed in this study. This data can be found here: No original raw data was generated in this narrative review. All analyses and conclusions are based on previously published peer-reviewed literature, which can be accessed via public academic databases including PubMed, Web of Science, Scopus and CNKI. Therefore, no data repository or accession number is available for this manuscript.
Ethics statement
Ethical approval was not required for the study involving humans in accordance with the local legislation and institutional requirements. Written informed consent to participate in this study was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and the institutional requirements.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that Generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.