The influence of physical activity on neural responses to visual food cues in humans: A systematic review of functional magnetic resonance imaging studies

This systematic review examined whether neural responses to visual food-cues measured by functional magnetic resonance imaging (fMRI) are influenced by physical activity. Seven databases were searched up to February 2023 for human studies evaluating visual food-cue reactivity using fMRI alongside an assessment of habitual physical activity or structured exercise exposure. Eight studies (1 exercise training, 4 acute crossover, 3 cross-sectional) were included in a qualitative synthesis. Structured acute and chronic exercise appear to lower food-cue reactivity in several brain regions, including the insula, hippocampus, orbitofrontal cortex (OFC), postcentral gyrus and putamen, particularly when viewing high-energy-density food cues. Exercise, at least acutely, may enhance appeal of low-energy-density food-cues. Cross-sectional studies show higher self-reported physical activity is associated with lower reactivity to food-cues particularly of high-energy-density in the insula, OFC, postcentral gyrus and precuneus. This review shows that physical activity may influence brain food-cue reactivity in motivational, emotional, and reward-related processing regions, possibly indicative of a hedonic appetite-suppressing effect. Conclusions should be drawn cautiously given considerable methodological variability exists across limited evidence.


Introduction
Appetite and energy intake are influenced by an intricate network of homeostatic and hedonic mechanisms that integrate centrally in the brain (Berthoud et al., 2017). Hedonic feeding occurs in the absence of hunger and involves brain regions with established roles in reward, motivation and cognitive processing, which can be triggered by external cues (Dagher, 2012;Kenny, 2011). The biological control of appetite comprises multiple peripheral organ and tissue sites that send and receive hormonal, metabolic and chemical signals to and from the brain to influence hunger and satiety signalling (MacLean et al., 2017). Adipose and gastrointestinal derived hormones are prominent within the appetite control system which also encompasses a myriad of other orexigenic and anorexigenic signals including several neuropeptides and neurotransmitters (Lean and Malkova, 2016), lactate (McCarthy et al., 2020), various myokines (Grannell et al., 2022), and liver-derived bile acids (Perino et al., 2021). Notably, hormonal signals of acute and/or chronic nutritional status derived from adipose tissue and the digestive tract are integrated in the hypothalamus and brain stem (Hussain and Bloom, 2013), but also influence hunger and satiety through interaction with brain reward networks (Batterham et al., 2007;Zanchi et al., 2017). Specifically, the orexigenic signal ghrelin has been positively associated with activation in reward-related brain areas including the amygdala and insula, whereas opposing associations, in these same regions, are observed with anorexigenic hormones such as leptin, peptide YY (PYY) and glucagon-like peptide-1 (GLP-1) (Zanchi et al., 2017).
The development of blood-oxygen-level-dependent functional magnetic resonance imaging (fMRI) has advanced understanding of the neural underpinnings of eating behaviour, which are commonly interrogated during brief exposure to visual food-related cues (Dagher, 2012). A diverse network of brain regions appear sensitive to food cue exposure including areas implicated in reward (e.g., striatum, hypothalamus, orbitofrontal cortex [OFC], insula), processing emotions and memory (e.g., amygdala, hippocampus) and attention (e.g., posterior fusiform gyrus, inferior occipital gyrus, lingual gyrus) (Huerta et al., 2014;Tang et al., 2012;van der Laan et al., 2011;van Meer et al., 2015). Food cue reactivity in several of these regions appears greater when fasted vs. satiated (Führer et al., 2008), in response to stimuli with a higher perceived reward value (e.g., high-energy density vs. low-energy density (Goldstone et al., 2009;Killgore et al., 2003) and in individuals with overweight or obesity vs. lean individuals (Brooks et al., 2013;Meng et al., 2020;Pursey et al., 2014).
The relationship between physical activity and appetite control has been extensively studied due to its potential to influence weight management (Blundell et al., 2015;Donnelly et al., 2009). It is widely regarded that single bouts of moderate-to-vigorous intensity exercise transiently suppress appetite and the orexigenic peptide acylated ghrelin, and increase satiety-related peptides such as PYY and GLP-1 Schubert et al., 2014). These fluctuations are usually observed without compensatory changes in energy intake on the day of exercise, promoting the maintenance of a short-term energy deficit Schubert et al., 2013). The impact of exercise training on appetite is less well defined partly due to the challenges of measuring free living energy intake accurately (Dhurandhar et al., 2015). A recent meta-analysis of mainly self-reported energy intake concluded that compensatory appetite responses to exercise training appear small in magnitude (Beaulieu et al., 2021), which may be offset to some extent by augmented meal-induced satiety signalling (Beaulieu et al., 2016;King et al., 2009). In contrast, evidence with robust doubly labelled water assessments has demonstrated that compensatory increases in energy intake and appetite appear to counteract the increased energy expenditure induced by supervised exercise training (Martin et al., 2019).
In recent years, there has been increased interest in the link between physical activity and hedonic drivers of eating. Studies exploiting psychometric assessments of both 'liking' (pleasure) and 'wanting' (motivation to eat) of specific foods have shown that implicit wanting and explicit liking of high-fat foods is inversely associated with habitual physical activity levels (Oustric et al., 2018), and exercise training appears to reduce the motivation to consume high-fat foods (Beaulieu et al., 2020a;Riou et al., 2019). Although these findings support the notion that habitual physical activity and purposeful exercise beneficially influence food reward, findings are inconsistent (Beaulieu et al., 2020b). As an alternative to behavioural tools, some studies have used fMRI to discern the brain circuitries that influence hedonic responses to food in the context of habitual physical activity and structured exercise; however, this evidence has yet to be reviewed systematically.
The primary aim of this systematic review was to synthesise evidence from studies that have used fMRI to explore brain responses to visual food stimuli in relation to habitual physical activity or structured exercise (acute and chronic) manipulations. As a secondary aim, relationships between food cue reactivity in response to physical activity or exercise and other appetite-related parameters (appetite ratings and hormones) were explored.

Methods
This review was conducted in accordance with the PRISMA 2020 guidelines (Page et al., 2021), and was registered prospectively in the PROSPERO database (CRD42020193938).

Literature search
Seven electronic databases (PubMed, Scopus, SPORT Discus, Psy-cINFO, PsycArticles, The Cochrane Library, and ClinicalTrials.gov) were searched by two researchers independently (AMD and TS) for eligible studies from inception to 09 December 2020. Searches in SPORT Discus, PsychINFO and PsycArticles were conducted simultaneously through EBSCOhost. An updated search was performed by two researchers independently (AMD and TS) in February 2023 to identify any additional eligible studies. The reference lists of all eligible studies and relevant review articles were searched to identify further eligible studies that were not captured during the searches. The search strategy was developed with an academic librarian (NR) and consisted of keywords related to 'functional magnetic resonance imaging', 'visual food cues', 'physical activity' and 'appetite'. Details of the full search strategy are presented in the supplementary content.

Inclusion and exclusion criteria
Studies were eligible for inclusion if human participants underwent at least one fMRI scan to examine the neural response to visual food cues in conjunction with an assessment of physical activity or a structured exercise exposure. For clarity, we define physical activity as any bodily movement through skeletal muscle activity whereas exercise, as a subset of physical activity, describes bodily movements that are planned, structured and purposeful. This distinction between physical activity and structured exercise is maintained throughout. All types of study were eligible for inclusion and no restrictions were placed on participants' age, sex, adiposity status, alcohol intake, smoking or publication language. Studies involving participants with drug dependence, a condition affecting eating behaviour or metabolism, or taking medication known to influence the outcomes were excluded. Studies were also excluded if brain imaging data were not obtained with fMRI during a visual food cue reactivity task. Both self-reported and objectively quantified habitual physical activity, and acute (single bout) or chronic (training) structured exercise interventions were eligible for inclusion. Experimental (acute and chronic exercise) studies were required to have either a control arm involving the same participants as the exercise arm, independent control and exercise groups, or pre-and post-intervention fMRI measurements. Only studies with identical procedures in the exercise and control arms, with the exception of the exercise manipulation, were eligible for inclusion.

Study selection
Search results were transferred to an online systematic review software platform (Covidence, Veritas Health Innovation, Melbourne, Australia) for the removal of duplicates and study screening. The fulltext of potential studies deemed relevant based on the title and abstract were retrieved and reviewed against the inclusion and exclusion criteria. All screening was performed independently by two researchers (AMD and TS) and a third researcher (AET) resolved any conflicts if a consensus could not be reached.

Data extraction and synthesis
Data were extracted independently from eligible studies by two researchers (AMD and TS) into a standardised electronic spreadsheet. Information was extracted on the following parameters: (1) study details (authors, publication date, sample size); (2) participant characteristics (age, sex, body mass index (BMI), handedness); (3) details of the physical activity assessment or exercise exposure; (4) details of the food cue paradigm; (5) energy intake and meal provision; (6) fMRI data acquisition; (7) fMRI data analysis; (8) fMRI results (peak voxel coordinates, anatomical region, cluster size, peak z score of cluster activation); and (9) non-fMRI outcomes (appetite perceptions, appetite-related hormones). If the standard error of the mean (SEM) was reported, the SEM was multiplied by the square root of the sample size to derive the standard deviation (Higgins et al., 2022). Due to the small number of eligible studies and heterogeneity in experimental design, it was not possible to conduct a co-ordinate based meta-analysis and, therefore, a qualitative synthesis of the findings is presented.
Data for contrasts comparing food (high-and/or low-energy density) vs. non-food cues and direct comparisons of high-vs. low-energy density food cues derived from whole brain and regions of interest (ROI) analyses are presented. It should be noted that various terminology was used to distinguish the 'high' and 'low' food image categories in the eligible studies including high vs. low hedonic value, high vs. low energy, high-vs. low-calorie and palatable vs. non-palatable. For consistency, we refer to 'high-energy density' and 'low-energy density' in the results and discussion, but the terminology presented in the original articles is retained in Table 2. All activated clusters are reported in the results and associated tables and figures, but the discussion and interpretation focus on brain regions activated in at least two studies.

Quality assessment
The National Heart, Lung, and Blood Institute (NHLBI) quality assessment tools (National Institutes of Health, 2014) were used to assess the quality of eligible studies by two reviewers independently (AMD and TS) and a consensus was reached through discussion. The quality assessment tools for (i) 'observational cohort and cross-sectional studies'; and (ii) 'before-after studies with no control group' were utilised according to the study design of the eligible studies. The original assessment tools were used but question 10 in the before-after tool was modified to replace the requirement for studies to report P-values with anatomical coordinates of activated clusters of voxels as follows: 'Did the statistical methods examine changes in outcome measures from before to after the intervention? Were anatomical coordinates for significant voxel clusters reported for the pre-to-post changes?'. the removal of duplicates, 3453 studies underwent title and abstract screening, and the full text was screened for 17 articles which resulted in eight studies that met the inclusion criteria for this review. Studies were classified based on study design: (1) exercise training interventions [n = 1]; (2) acute crossover studies [n = 4]; and (3) cross-sectional studies [n = 3]. An overview of the participant characteristics and physical activity or exercise exposure for each study is presented in Table 1. A summary of the fMRI protocol and key findings are provided in Table 2, and detail pertaining to fMRI data acquisition and analysis is shown in Table 3. The brain regions activated for each contrast in the eligible studies is collated in Table 4.

Study characteristics
One pre-post uncontrolled exercise training intervention met the inclusion criteria which recruited 12 men and women with overweight or obesity (Cornier et al., 2012) (Table 1). Supervised treadmill walking was performed five days/week with the duration and intensity increasing across the six-month intervention (Table 1). A fasted fMRI scan lasting ~13 mins with high-energy density food and non-food objects was completed at baseline and on two occasions after the intervention, once within 30 mins of an acute exercise bout and once without exercise for 24 h (Table 2). Cornier et al. (2012) reported statistically significant reductions in fat mass and body fat percentage, but not body mass, after exercise training. Reduced reactivity to food cues (high-energy density food vs. non-food) in the bilateral parietal cortices, left insula and visual cortex was observed after chronic exercise (24 h after the last bout) compared to baseline (Fig. 2, Table 2, Table 4). Pre-to-post intervention food cue reactivity was not statistically different when the post-intervention scan was completed within 30 mins of acute exercise (Table 2).

Correlations
The BOLD signal change from pre-to post-intervention (24 h after exercise) in the insula was positively associated with the change in body mass (r = 0.76), fat mass (r = 0.78) and circulating leptin concentrations (r = 0.80). The BOLD signal change in the visual cortex (r = 0.91) and the hypothalamus (r = − 0.76) from pre-to post-intervention (24 h after exercise) was associated with the change in blood leptin concentrations.

Table 1
Characteristics of studies investigating the influence of physical activity with food cue reactivity using functional magnetic resonance imaging. BMI, body mass index; MET, metabolic equivalent of task; MRI, magnetic resonance imaging; MVPA, moderate-to-vigorous intensity physical activity; NM, not measured; NR, not reported; SD, standard deviation; V˙O 2 max, maximum oxygen uptake.

Table 2
A summary of the food cue paradigm presented during the functional magnetic resonance imaging scan and key findings of the eligible studies.   (Crabtree et al., 2014;Evero et al., 2012;Saanijoki et al., 2018) or children (Masterson et al., 2018) (Table 1). Exercise involved a single session of 30-60 mins of cycling or treadmill walking/running which was completed at an average intensity of 70% of maximum oxygen uptake or 67-83% of maximal heart rate ( Table 1). The control trial involved a time-matched period of rest in all studies (Table 1).
Study measures were performed after fasting for 3 h (Saanijoki et al., 2018) or after an overnight fast (Crabtree et al., 2014;Evero et al., 2012;Masterson et al., 2018) (Table 2). In all studies, a single food cue reactivity fMRI scan (duration ~6-16 mins) was performed in each trial commencing between ~3 and 80-110 mins after completion of the exercise or rest period (Table 2). All studies presented two food image categories encompassing foods of 'high' and 'low' energy density determined according to energy content or palatability (Table 2). Non-food object images were included as a control in three studies (Crabtree et al., 2014;Evero et al., 2012;Saanijoki et al., 2018), whereas one study used a blurred version of the food images (Masterson et al., 2018) (Table 2).

Study findings
Three of the four acute crossover studies defined contrasts investigating the neural response to food cues (high-and/or low-energy density) compared to non-food cues (Crabtree et al., 2014;Evero et al., 2012;Saanijoki et al., 2018) (summarised in Table 2 and Table 4). In whole-brain analysis, Evero et al. (2012) reported lower reactivity to food (high-and/or low-energy density) vs. non-food cues after exercise than control in the left postcentral gyrus and several reward-related regions including the right insula, right putamen and right rolandic operculum (Fig. 3). Another study also observed lower food cue (highand low-energy density combined) reactivity in other reward-and attention-related brain regions (right posterior cingulate, left hippocampus, left OFC) after exercise (Crabtree et al., 2014) (Fig. 3). Conversely, higher food cue responsiveness has also been reported after exercise most notably in the right insula (Crabtree et al., 2014) (Fig. 3) and the left precuneus (Evero et al., 2012). Furthermore, Crabtree et al. (2014) identified divergent brain food cue responsiveness according to the energy value of the food images with reactivity in reward-and attention-related regions generally higher when viewing low-energy density food cues and lower when viewing high-energy density food cues (Crabtree et al., 2014) (see Table 2 and Table 4 for further detail). One study observed no statistically significant differences in the neural response to food (high-and low-energy density combined) vs. non-food images between exercise and control (Saanijoki et al., 2018).
Three of the four acute crossover studies defined contrasts to compare neural responses between high-and low-energy density food cues (Crabtree et al., 2014;Masterson et al., 2018;Saanijoki et al., 2018). Reduced reactivity to high-vs. low-energy density food cues was observed in the left pallidum and right superior occipital gyrus after exercise compared to control in the study by Crabtree et al. (2014). Masterson et al. (2018) reported greater reactivity to high-vs. low-energy density food cues in the right medial temporal lobe and left hippocampus in the exercise compared to the control trial. When analysing the main effect of trial (independent of food image category), this study also reported that BOLD activity was lower in the left postcentral gyrus and higher in the right anterior insula after exercise compared to control. Saanijoki et al. (2018) found no statistically significant differences in the contrast of high-vs. low-energy density food cues between exercise and rest.

Correlations
Three of the four studies explored correlations between exerciseinduced changes in food cue reactivity and appetite-related outcomes (Crabtree et al., 2014;Evero et al., 2012;Saanijoki et al., 2018). Two studies found no statistically significant correlations between the exercise-induced BOLD signal change for any contrast and either ratings of perceived appetite (hunger, desire to eat, fullness, prospective food consumption) (Crabtree et al., 2014;Evero et al., 2012) or concentrations of appetite-related hormones (total ghrelin, acylated ghrelin, total PYY) (Crabtree et al., 2014). Exercise-induced changes in thirst were inversely associated with lower BOLD activity in the pallidum after exercise in the high-vs. low-energy density food contrast (r = − 0.58) (Crabtree et al., 2014). Another study reported that exercise-induced changes in fullness were negatively associated with BOLD activity in the left caudate (r = − 0.62), whereas exercise-induced changes in prospective food consumption were positively associated with BOLD activity in the left precentral gyrus (r = 0.75) in response to high-vs. low-energy density food cues (Saanijoki et al., 2018).

Study characteristics
Three cross-sectional studies met the eligibility criteria with all studies recruiting male and female adults (Killgore et al., 2013;Luo et al., 2018;Drummen et al., 2019) (Table 1). Mean BMI ranged from 24.5 to 32.3 kg/m 2 and mean BMI category ranged from lean to obesity (Killgore et al., 2013;Luo et al., 2018), or was restricted to individuals classified with overweight/obesity and at elevated risk of type 2 diabetes (Drummen et al., 2019) (Table 1). Habitual levels of physical activity (and sedentary time) were assessed using a self-report questionnaire in two studies (Killgore et al., 2013;Drummen et al., 2019) and repeated physical activity recall interviews in one study (Luo et al., 2018) (Table 1). Functional brain imaging was conducted after an overnight fast (Drummen et al., 2019), one hour after food intake (Killgore et al., 2013) or 20-30 mins after the ingestion of 300 mL water either with or without 75 g of glucose (Luo et al., 2018) (Table 2). Two studies included two food categories encompassing foods of 'high' and 'low' energy density determined based on energy content (Killgore et al., 2013;Drummen et al., 2019), whereas one study included high-energy density food cues only (Luo et al., 2018) (Table 2). All studies included a category of non-food object images. The duration of stimulus presentation ranged from ~4-9 mins (Table 2).

ROI analysis
No association between any physical activity index and BOLD activity in response to food (highand low-calorie) vs. non-food in the nucleus accumbens.

Study findings
Two of the three cross-sectional studies defined contrasts investigating the neural response to food cues (high-and/or low-energy density) compared to non-food cues (Luo et al., 2018;Drummen et al., 2019) (summarised in Table 2 and Table 4). After adjustment for age and BMI, Drummen et al. (2019) reported a negative correlation between the Baeke leisure index and BOLD activity in response to food vs. non-food cues in the right thalamus, left middle cingulate gyrus and right precuneus. In the study by Luo et al. (2019), higher levels of moderate-to-vigorous intensity physical activity were associated with lower reactivity to high-energy density vs. non-food cues after glucose ingestion in the composite ROI (average of 10 ROIs), right precuneus, middle insula, left postcentral gyrus and left amygdala. When stratified by adiposity status, the inverse association for the composite ROI was Table 3 An overview of the functional magnetic resonance imaging data acquisition and data analysis parameters of the eligible studies. only apparent in individuals with obesity and not in individuals who were lean. Sedentary behaviour was positively associated with the BOLD signal change in the middle insula and left amygdala in response to high-energy density vs. non-food cues after glucose ingestion. One of the three cross-sectional studies defined contrasts to compare neural responses between high-and low-energy density food cues (Killgore et al., 2013) (Table 2, Table 4). Greater self-reported weekly exercise was associated with lower reactivity to high-vs. low-energy density food cues in the medial OFC and left anterior insula.

Study quality assessment
A summary of the study quality assessment is displayed in Fig. 4 and a full breakdown for each study is presented in Supplementary Table 2. Studies scored favorably on items related to reporting of objectives, exposures and outcomes, with the exception of cross-sectional studies which did not use valid/reliable tools for assessing habitual physical activity. Cross-sectional studies also scored poorly on items related to causality and blinding of researchers assessing outcomes. Most studies did not justify the study sample size or present a formal power calculation, and it was often unclear whether recruited participants were representative of the wider population. Some studies also lacked information on recruited participants including characteristics that may influence the outcome measures such as smoking status, alcohol/drug dependency and current dieting practices. Overall, the studies in this review had various methodological designs, and few provided adequate information, suggesting that future experimental studies with more robust designs are required.

Discussion
This systematic review aimed to consolidate findings from observational and experimental (acute and chronic) studies on the influence of habitual physical activity and structured exercise on brain responses to food cues assessed using fMRI. Our review identified six brain regions each reported in at least two studies (insula, hippocampus, OFC, putamen, postcentral gyrus and precuneus) that appear sensitive to food cue exposure in relation to habitual physical activity or structured exercise manipulations. The cross-sectional evidence reviewed suggests that higher levels of habitual physical activity are associated with lower reactivity to high-energy density food cues in brain regions linked to reward including the insula and OFC. These findings appear in line with the consensus from experimental studies indicating that structured acute and chronic exercise may lower food-cue responsiveness in several motivation-and reward-related brain regions (e.g., insula, hippocampus, OFC, putamen) particularly in response to images of high-energy density foods. However, findings across the limited literature reviewed are not consistent which is likely due to the considerable heterogeneity in methodological approaches.

Table 4
Summary of the brain regions activated for each contrast in the eligible studies.
Study group: 1, exercise training interventions; 2, acute crossover studies; 3, cross-sectional studies. Studies are listed under all contrasts that were explored in the analysis even if no activated clusters were reported. For study groups 1 and 2, arrows reflect direction of BOLD signal change during the food cue task in response to chronic or acute exercise (downwards blue arrow: decrease; upwards red arrow: increase). For study group 3, arrows reflect direction of association between the BOLD signal change during the food cue task and the index of physical activity or sedentary time (downwards blue arrow: negative; upwards red arrow: positive). Brain regions highlighted in bold represent regions activated in at least two studies. 1 Data reported from model adjusted for body mass index and age. 2 Results represent pre-to-post intervention change when the post exercise measures occurred 24 h after the last exercise bout. 3 Data reported from model adjusted for obesity status and sex. 4 Results displayed for the trial-by-stimulus interaction only. The main effect of trial also revealed lower activity after exercise vs. control in the left postcentral gyrus and right anterior insula. DLPFC, dorsolateral prefrontal cortex; MVPA, moderate-to-vigorous intensity physical activity; OFC, orbitofrontal cortex.
The present review identified the insula as the brain region most consistently activated by food cue exposure in relation to habitual physical activity or structured exercise (in five out of eight studies).
Specifically, observational studies suggest habitual physical activity is negatively, and sedentary time is positively associated with insula food cue reactivity particularly in response to high-energy density food Fig. 2. Clusters of activated voxels in response to high-energy density food vs. non-food object images after exercise training in the study by Cornier et al. (2012). Blue circles represent separate clusters of activated voxels and denote reduced food cue reactivity after exercise training compared to baseline: 1-3 = parietal cortex (right); 4-5 = parietal cortex (left); 6-8 = insula (left); 9-10 = visual cortex (right). Figures were generated by the authors using the FMRIB Software Library (FSL) viewer FSLeyes (version 0.34.2, FMRIB Centre, Oxford, UK) and the Multi-image Analysis GUI (version 4.1, Research Imaging Institute, University of Texas Health Science Centre, San Antonio, USA) with clusters of activated voxels overlayed on the MNI-152 template. Fig. 3. Clusters of activated voxels in response to food (high-and low-energy density combined) vs. non-food cues after acute exercise in the studies by Evero et al. (2012) and Crabtree et al. (2014). No activated clusters were reported in the study by Saanijoki et al. (2018). Each circle represents a separate cluster of activated voxels from a given study in the whole brain analysis. Green circles denote reduced food cue reactivity after exercise in Evero et al. (2012): 1 = postcentral gyrus (left); 2 = putamen (right); 3 = insula (right); 4 = rolandic operculum (right); 5 = supramarginal gyrus (right). Blue circles denote reduced food cue reactivity after exercise in Crabtree et al. (2014): 6 = posterior cingulate (right); 7 = precentral gyrus (left); 8 = precentral gyrus (right); 9 = hippocampus (left); 10 = orbitofrontal cortex (left). Red circles denote increased food cue reactivity after exercise in Crabtree et al. (2014): 11 = insula (right); 12 = middle temporal gyrus (right). images (Killgore et al., 2013;Luo et al., 2018). This finding is supported by experimental studies demonstrating lower reactivity to food cues in the insula after a single exercise bout (Evero et al., 2012) or prolonged exercise training (Cornier et al., 2012). The insula is recognised as the primary gustatory cortex but also plays important roles in food craving, integrating interoceptive information from the gut, memory for food cues and reward processing (Pelchat et al., 2004;Balleine and Dickinson, 2000;de Araujo et al., 2012;Small, 2010). Consequently, the lower insula BOLD response to food cues may suggest habitual physical activity and structured exercise (acute and chronic) reduce the saliency of food rewards and cravings.
Other key brain regions identified in this review include the hippocampus, OFC and putamen which, along with the insula, are thought to play prominent roles in eating behaviour (Dagher, 2012;Watts et al., 2022) and are often responsive to food cue exposure (Huerta et al., 2014;Tang et al., 2012;van der Laan et al., 2011;van Meer et al., 2015). The OFC encodes the incentive value of food stimuli according to motivational state (Gottfried et al., 2003;Arana et al., 2003), whereas the putamen (situated in the dorsal striatum) is implicated in food motivation through dopaminergic projections (Volkow et al., 2002). Findings synthesised from this review demonstrate that acute exercise lowers BOLD activity in response to food cues (high-and/or low-energy density) in the OFC (Crabtree et al., 2014) and putamen (Evero et al., 2012), with activity in the OFC also inversely associated with self-reported weekly exercise in response to high-vs. low-energy density food cues (Killgore et al., 2013). Whilst these observations appear consistent with Fig. 4. Summary of study quality assessment presented as a percentage across the eligible studies. Study quality assessed using The National Heart, Lung, and Blood Institute (NHLBI) quality assessment tools (National Institutes of Health, 2014). Panel A: study quality for exercise training interventions assessed using the 'before-after studies with no control group' tool. Panel B: study quality for acute crossover and cross-sectional studies assessed using the 'observational cohort and cross-sectional studies' tool. The full criteria for each tool and a full breakdown for each study is presented in Supplementary Table 2. lower expected food reward and motivation to eat, exercise may provoke divergent responses according to the energy density of foods presented with evidence that the appeal of low-energy density foods may be enhanced. This is supported by Crabtree et al. (2014) who reported greater responsiveness to low-energy density food cues after acute exercise in the insula and putamen. Although the implication of this finding from a single study should be interpreted cautiously, it may highlight a positive role of exercise given stimuli with higher perceived reward value typically evoke greater food cue reactivity in the brain (Goldstone et al., 2009;Killgore et al., 2003).
The hippocampus is thought to influence feeding behaviour by integrating episodic meal-related memories with information from interoceptive signals and the external environment (Parent et al., 2022;Kanoski and Grill, 2017). Based on evidence in the present review, this region of the temporal lobe appears sensitive to modulation with acute exercise. Specifically, a single exercise bout was shown to increase hippocampus BOLD activity (high-vs. low-energy density food cues) in children (Masterson et al., 2018), whereas lower food cue responsiveness was apparent in men after acute exercise with presentation of high-energy density food cues in particular appearing to drive this response (Crabtree et al., 2014). The reason for the conflicting direction of response in these investigations is unclear but it is notable that the precise anatomical location in the hippocampus was not reported in either study. Evidence from rodent models suggests metabolic signals of energy status primarily act on ventral hippocampal neurons whereas dorsal hippocampal neurons have been linked to the inhibitory effects of meal-related memory on eating (Parent et al., 2022;Kanoski and Grill, 2017). Further work is required to clarify the implications of altered hippocampus food cue reactivity with exercise in humans, particularly considering this region was not identified in any of the exercise training or cross-sectional studies.
Reactivity to high-energy density food (vs. non-food) cues in the precuneus, located in the caudal region of the cingulate cortex, was greater after exercise in one eligible acute crossover study (Evero et al., 2012). This is in accordance with a previous study reporting heightened precuneus BOLD activity in response to smoking-related images after a 10-min moderate-intensity exercise bout (Janse Van Rensburg et al., 2009). Conversely, the observational studies in this review demonstrated that self-reported measures of physical activity were inversely associated with food cue reactivity in the precuneus (Drummen et al., 2019;Luo et al., 2018). Although the proposed functions of the precuneus are diverse, this region has been linked to visuo-spatial processing, attention shifts between objects and episodic memory retrieval (Cavanna and Trimble, 2006). Consequently, it could be hypothesised that exercise may increase attention towards rewarding (food) stimuli at least transiently in the immediate post-exercise period, but the saliency of food stimuli may be diminished in attention-related brain regions in those with higher habitual physical activity levels.
Evidence collated in this review suggests that single exercise bouts may reduce reactivity to foods of low-energy density in the postcentral gyrus (Crabtree et al., 2014;Evero et al., 2012), with BOLD activity in this region when viewing high-energy density vs. non-food cues also inversely associated with self-reported levels of habitual moderate-to-vigorous intensity physical activity (Luo et al., 2018). The postcentral gyrus is the site of the primary somatosensory cortex with principal functions in proprioception and touch sensations. Although this region of the parietal lobe appears receptive to food stimuli (Tang et al., 2012;Huerta et al., 2014), the implications of altered postcentral gyrus BOLD activity to food cues in the context of physical activity and structured exercise are unclear and require further investigation.
Our systematic review identified few relationships of brain food cue reactivity with hormonal and subjective indicators of appetite in response to exercise. Such relationships within the acute crossover studies were restricted to inverse associations between exercise-related changes in food cue-related BOLD activity with thirst in the pallidum (Crabtree et al., 2014) and fullness in the caudate (Saanijoki et al., 2018), whereas food cue-related BOLD activity in the precentral gyrus was positively related to prospective food consumption (Saanijoki et al., 2018). In contrast, no correlations were apparent between exercise-related changes in brain food cue responsiveness and the appetite-related hormones ghrelin and PYY (Crabtree et al., 2014). These peptides secreted from the gastrointestinal tract provide signals to the brain about hunger and satiety on a meal-by-meal basis (Begg and Woods, 2013) and fluctuate in response to single exercise bouts in directions expected to suppress appetite (Schubert et al., 2014;Dorling et al., 2018). With exercise training, diminished insula food cue reactivity was associated with greater exercise-induced weight/fat loss and decreases in leptin concentrations (Cornier et al., 2012). Given the lack of consistent correlations apparent in the literature, these associations should be interpreted tentatively until the relevance for eating behaviour can be established.

Limitations of current evidence and future directions
Several methodological considerations should be noted when interpreting the overview of evidence in this review. Evidence is restricted to a limited number of observational and experimental studies which precluded quantitative synthesis in a meta-analysis. Apart from the insula, most brain regions were only identified in one experimental study (see Table 4) and the five other regions (hippocampus, OFC, postcentral gyrus, precuneus, putamen) that met the threshold for further discussion were only identified in two or three out of the eight eligible studies. This lack of consistency will invariably be underscored by the considerable heterogeneity that exists in the design of current studies including differences in scanning protocols, food cue paradigms, measurement timing, data processing and analysis, participant characteristics and the protocol for exercise exposure or habitual physical activity assessment.
Further work is required to expand the limited evidence base with greater standardisation of fMRI paradigms (e.g., MRI scanner, food stimulus, task paradigm, analysis pipeline) across research centres needed to improve the reliability of conclusions and potential to amalgamate data statistically using meta-analytical techniques. As an alternative measure of food reward and attentional bias, reaction time can be captured concurrently during fMRI food cue paradigms but was not reported in the studies eligible for this review. Considering recent evidence demonstrating that acute exercise increases attentional bias towards food cues assessed using eye tracking during a computer food task (Flack et al., 2022), measuring food cue-related reaction time and BOLD responses simultaneously would enhance understanding of the influence of physical activity/exercise on hedonic appetite responses. Future observational studies should quantify physical activity and sedentary time using accelerometry and inclinometry as self-report tools that have dominated the extant literature are prone to overreporting (Dhurandhar et al., 2015). Out of the eight eligible studies, one acute crossover study reported no effect on brain food cue reactivity in healthy men (Saanijoki et al., 2018), but notably the fMRI scan was performed 80-110 mins after exercise which may have missed important brain effects in closer proximity to exercise. The optimum timing of fMRI acquisition in relation to acute exercise bouts has yet to be established and represents a research avenue to address.
It is vital that further chronic studies involving randomised controlled/crossover designs that integrate an appropriate control arm are conducted to explore potential causal links between exercise and food cue reactivity. Future experimental work should also target clinically relevant populations such as those with overweight or obesity who are likely to exhibit heightened food cue reactivity (Brooks et al., 2013;Meng et al., 2020;Pursey et al., 2014) in addition to individuals who are losing weight or are in the weight-reduced state. Individual variability in subjective appetite and hormonal (acylated ghrelin, total PYY) responses to acute exercise has been identified using a replicated crossover design showing some individuals experience a greater magnitude of exercise-induced appetite suppression than others (Goltz et al., 2018). Whether similar exercise-related response heterogeneity exists in neural responses to visual food cues has not been explored but represents an important future research direction.

Conclusion
This systematic review provides evidence that habitual physical activity and structured exercise may influence brain food cue reactivity in regions linked to pleasure, motivation to eat, episodic memory and attention. Structured acute and chronic exercise exposures appear to reduce brain food-cue reactivity in motivation-and reward-related regions particularly when viewing high-energy density food cues and may enhance the appeal of low-energy density food cues. This is supported by cross-sectional evidence showing self-reported habitual physical activity may be inversely associated with lower brain activation in food responsive reward regions. Consequently, there is currently little evidence that habitual physical activity and structured exercise perturbs brain appetite responses in a direction associated with enhanced pleasure of food and motivation to eat. These findings should be interpreted cautiously until further research with consistent methodological approaches is conducted to expand the limited evidence base.

Declaration of Competing Interest
LJJ has current funding from Entrinsic Beverage Company LLC, Entrinsic Bioscience, LLC, Volac International Ltd, Herbalife Europe Ltd, Bridge Farm Nurseries and Decathlon SA, and has previously received funding from PepsiCo Inc. and British Summer Fruits; has performed consultancy for PepsiCo Inc. and Lucozade, Ribena Suntory, and has received conference fees from PepsiCo Inc. and Danone Nutricia. In all cases, monies have been paid to LJJs institution and not to LJJ.