Reward and automatic processes in exercise behavior: A new approach and a systematic review

Background In a time of physical inactivity pandemic, attempts to better understand the factors underlying the regulation of exercise behavior is important. The dominant neuropsychological approach to exercise behavior explains physical activity as a reward. However, the opposite of physical exertion — behaviors minimizing energy cost — may also be a reward, which activates automatic reactions favoring the engagement in behaviors associated with lower energetic costs. Objective Our objective was to systematically review studies testing the automatic reactions triggered by stimuli associated with different types of exercise behavior (e.g., physical activity, sedentary behaviors) and energetic cost variations (e.g., behaviors minimizing energetic cost). Methods Two authors systematically searched, screened, extracted, and analyzed data from articles in the MEDLINE database. Results We included 26 studies. Three types of automatic processes were tested: Affective reactions, attentional capture, and approach tendencies. Results show that automatic reactions toward stimuli depicting exercise behaviors explained individuals’ level of physical activity. Brain imaging results show that stimuli associated with exercise behavior activate regions associated with reward, but these studies were scarce. Conclusion Reward is an important factor of exercise behavior. There is strong evidence showing that physical activity is a reward. While brain imaging results suggest that sedentary behaviors are also a reward, behaviors minimizing energetic cost have not been investigated so far. Additional studies are required to establish a strong and complete framework of reward in exercise behavior. Key points – Behavioral and brain imaging studies using different types of automatic behavior show that physical activity is a reward – Behaviors minimizing energetic cost have been essential to evolutionary survival and are likely to be a reward. However, experimental evidence remains scarce – The dominant neuropsychological approaches to exercise behavior may be incomplete, which may partly explain our current inability to counteract the pandemic of physical inactivity


Introduction
Twenty years ago, the World Health Organization (WHO) issued comprehensive guidelines for promoting physical activity among older adults [1]. Since then, the importance of physical activity for health has been increasingly emphasized and guidelines have been extended to all populations [2]. Today, however, one third of the adult population remains physically inactive and 80% of the adolescent population does not reach the recommended amount of physical activity [3]. Why do most people fail to exercise regularly [4]? What if a fundamental principle that leads us to minimize energetic cost has been neglected in exercise neuropsychology?
The dominant theoretical paradigm explains physical activity as a reward [5][6][7][8][9][10]. Here, we propose that the opposite of physical exertion, that is, behavior that minimizes energetic cost, is also an evolutionarily adaptive reward for the species. This new model has the potential to enhance our understanding of the basic neurophysiological processes governing automatic processes in exercise behavior. In addition to providing new fundamental knowledge, this model could help addressing a global health problem. Each year, physical inactivity is responsible for 13 million lost years of healthy life [11] and 5 million deaths worldwide [12]. To counteract the pandemic of physical inactivity [4], reconsidering the fundamental basis of the current approach to exercise behavior is needed.

Controlled and automatic processes in exercise behavior
In neuroscience and psychology, two types of processes are thought to govern the regulation of behaviors: Controlled and automatic processes [13][14][15][16]. The controlled processes are initiated intentionally, require cognitive resources, and operate within conscious awareness. Conversely, automatic processes are initiated unintentionally, tax cognitive resources to a much lesser extent, and occur outside conscious awareness [17,18]. These automatic processes can be problematic when they come into conflict with the controlled processes [16,19]. For example, an opportunity for sedentary behavior can automatically activate a behavioral response that competes with the conscious intention to adopt a physically active behavior, thereby preventing its implementation. Models testing the capacity of controlled processes for explaining exercise behavior have shown high levels of unexplained variance [20], whereas automatic processes have been shown to be critical [21][22][23]. This pervasive effect of automatic over controlled processes in explaining exercise behavior may suggest that the pandemic of physical inactivity [4] originates in automatic processes. Specifically, people may fail to exercise regularly despite conscious intentions to be active because behaviors minimizing energetic cost (BMEC) activate competing automatic processes. Here, BMEC are defined as any behavior resulting in energetic cost decrease, irrespective of the initial level of energy expenditure.

Behaviors minimizing energetic cost as a reward
To counteract the lack of physical activity, reconsidering our current view of the psychological and neural mechanisms regulating exercise behavior is urgently needed. Here, we argue that a fundamental principle pushing individuals to minimize energetic cost has been insufficiently considered in the dominant approaches to exercise behavior. While the rewarding value of exercise has been widely studied [5][6][7][8][9][10], the potential reward associated with BMEC has been disregarded so far. Previous literature has mostly viewed physical activity as a natural reward, whereas the potential rewarding value of BMEC has been insufficiently considered.
Reward is the positive value ascribed to an object, a behavioral act, or an internal physical state [24], through multiple neuropsychological components [25][26][27]. The "wanting" (or desire) component is the positive value resulting from the relevance of the behavior for the needs of the individual [26,28,29]. The "liking" component is the positive value resulting from the hedonic pleasure associated with the performance of the behavior [25]. Reward triggers automatic processes that can initiate, sustain, and change behavior adaptively between different available options and plays a key role in optimizing the allocation of resources necessary for evolutionary survival [30]. As BMEC determine behavioral adaptation on short (e.g., walking vs. running [31]) and evolutionary timescales (e.g., quadrupedalism vs. bipedalism [32]), it is difficult to conceive of these behaviors differently from a reward.
1.3. From an automatic to a controlled trigger of physical activity Besides BMEC, physical activity has also been necessary for development and evolutionary survival. For example, physical activity can be viewed as a necessary means to achieve motor learning and development, explaining why children are naturally inclined to exert physical effort in play periods [33]. Furthermore, physical activity is triggered when individuals need to search for food or shelter, interact with competitors, and avoid predators [34]. Particularly, food and physical activity are thought to be part of the same cycle, where alternating periods of food scarcity and abundance are associated with higher and lower physical activity, respectively [35]. This model is supported by findings in rodents and monkeys showing that physical activity increases during restricted feeding [36,37]. This increase of physical activity during food restriction is interpreted as foraging behavior that conferred a decisive advantage for survival in periods of food scarcity [36,38]. Since the goal of this increased energetic expenditure is energy replenishment, the optimization principle is at work. This principle is based on cost minimization [39]. For example, individuals automatically adapt their step frequency and walking speed in real-time to optimize energy costs [40], and learn to minimize the physical efforts required to obtain a specific reward [41]. In addition, individuals who sustained physical activity for longer periods were more likely to find food. Therefore, species that developed processes alleviating pain and fatigue during physical activity, such as analgesia, sedation, and anxiolysis [5][6][7][8][9][10], were more likely to survive. Physical activity likely became a reward due to these convergent processes. In modern occidental societies, the food-physical activity circle is broken. During the past century, food became abundantly available without the requirement of physical activity, owing to advances in farming, agriculture, industrialization, transportation, and storage. As a result, modern society is perpetually in the phase of abundant food associated with lower physical activity. Food scarcity, the automatic trigger of physical activity, no longer exists and needs to be replaced by a controlled trigger, such as conscious intentions, to regulate energy balance. For example, individuals may decide to start running to reduce their risk of disease incidence or to improve their physical health. Then, following an extensive repetition of physical activity, it may transition from a controlled to an automatic behavior [42]. Following up on the previous example, once engaged in running, the release of neurotransmitters associated with this behavior will have a hedonic effect. The repetition of this hedonic effect may be learned [43] and result in the association of the hedonic effect and the behavior (i.e., running). Once this automatic association is consolidated in memory, an environmental stimulus (e.g., seeing another individual running) can automatically trigger a positive evaluation of the stimulus, which will in turn evoke preparatory responses favoring the engagement in running, such as approach tendencies toward running. The positive reinforcement taking place during volitional physical activity, such as sport and fitness-related activities, can also benefit nonvolitional physical activity, such as occupational work, activities of daily living, fidgeting, spontaneous muscle contraction, or actively maintaining posture [44,45]. In sum, engaging in exercise in modern society is triggered by controlled processes. However, the automatic processes promoting BMEC, which are particularly salient in the current environment, threaten controlled processes that promote exercise. Therefore, sustaining regular and frequent exercise requires important self-control resources [46].

Implications
In the fields of psychology and neuroscience, the neutral nature of sitting and lying positions has always been taken for granted. If BMEC are a reward, this assumption suddenly becomes questionable. Reward perception has been shown to be dependent on individuals' physiological state [47]. For example, thirsty participants show higher perceptual readiness to drinking-related stimuli [48] and hungry participants show stronger automatic approach reactions toward food-related stimuli [49]. Therefore, if BMEC are a reward, the reward associated with sitting and lying positions depends on the participant's maximal exercise capacity and recent exercise history (e.g., did the participant come by bike or bus). Accordingly, it is urgent to investigate the rewarding nature of BMEC. Until this point has been clarified, a precautionary principle should be applied and scientists in the field of neuroscience, psychology, and exercise should prospectively adjust their experimental designs to discard this potential bias. Physical activity should be monitored during the days/hours preceding the experiment and this information should be included in the models as a covariate. Additionally, if BMEC are a reward, the pandemic of physical inactivity is driven by an automatic resistance to the intended engagement in exercise. Therefore, public health policies take the wrong approach. Part of the massive investment aiming at increasing conscious intentions to be active should be redirected toward the development of research projects aiming at understanding the mechanisms underlying this automatic resistance and interventions aiming at reducing it.

Objective
Rewarding stimuli trigger different types of automatic processes such as attention capture (i.e., reward captures attention), affective reactions (i.e., reward produces hedonic pleasure), and approach tendencies (i.e., reward predisposes to physical approach). Here, our objective was to systematically review studies investigating automatic reactions to stimuli associated with different types of exercise behavior (e.g., physical activity and sedentary behaviors) and energetic cost variations (e.g., behaviors minimizing energetic cost). Figure 1 presents the flow diagram of our search strategy. The potential studies were identified by searching the electronic MEDLINE database via PubMed. We searched for all available records starting from January 2000 until June 2017 using the following combination of keywords in the title or abstract of the article: (exercise OR "physical activity" OR "sedentary behavior") AND (reward OR automatic OR impulsive OR implicit OR non-conscious).

Eligibility criteria and study selection 2.2.1. Inclusion criteria
To be included in this systematic review, the article had to 1) be published in a peer-reviewed journal and written in English, 2) report original data collected from humans, 3) test the automatic behaviors or brain activation triggered by stimuli associated with different types of exercise behavior (e.g., physical activity, sedentary behaviors, and BMEC), and 4) the reward should be assessed during the presentation of stimuli related to exercise behavior or during the performance of this behavior.

Exclusion criterion
Self-reported measures were excluded as they appear to be less appropriate to measure the automatic processes associated with reward [68]. For a review of the relationship between self-reported habit index and exercise behavior, please see [69].

Exclusion selection
Five steps were used to select the articles meeting the inclusion criteria. If there was a doubt at any step, the article was kept for further inspection. At step 1, articles not written in English were excluded. At Step 2, articles not reporting original experimental data were excluded (e.g., reviews, meta-analyses, commentary, technical reports, case studies). At Step 3, articles were excluded if they did not involve a human population. At step 4, articles were excluded if they did not test the automatic behaviors or brain activation triggered by rewarding stimuli. At step 5, articles exclusively using self-reported measures of these processes were excluded. We performed reference screening and forward citation tracking on the articles remaining after step 5.

Data extraction
Data were extracted from the included articles and summarized in Table 1. In this table, we report 1) the type of population (e.g., age, healthy individuals, individuals with pathologies such as anorexia nervosa, respiratory disease, or obesity), 2) the technique used to investigate brain substrates (fMRI), 3) the type of measure used to assess behavioral performance (e.g., reaction time), 4) the type of task used (i.e., Implicit Association Test, Manikin task, Visual Dot Probe Task, imagined scenarios), 5) the type of reward used and the format in which the reward was presented (pictures, words, imagined stimuli), 6) the content of the reward, i.e., whether the reward was related to specific sports and fitness (e.g., scheduled physical activity such as running, swimming) and sedentary-related activities (e.g., specific activity associated with a low energy expenditure such as watching TV, playing video games) or more to the general concept of action/effort (e.g., active, energetic, vigorous) and inaction/rest (e.g., weak, frail, inactive), and 7) whether recent physical activity history was controlled.

Literature search
The primary search retrieved 1664 potentially relevant articles. Of the 1664 screened articles, disagreement occurred in 31 cases (2%), which were all resolved by discussion. This selection yielded 110 potentially relevant full-text articles, which were then reviewed. All articles remained after step 1. At Step 2, 19 articles were excluded because they did not report original experimental data. At Step 3, 26 articles were excluded because they did not involve humans. At step 4, 26 articles were excluded because automatic behaviors were not triggered by rewarding stimuli. At step 5, 18 articles were removed because they relied exclusively on selfreports. Five were added after reference screening and forward citation tracking. Finally, 26 articles were included (Table 1). Three articles of this final reported fMRI analyses.

Recent exercise history
Recent exercise history was not controlled in 80.0% of the studies. The studies controlling for this state (20.0%) used a self-reported questionnaire (7-day physical activity recall or an adapted version of the International Physical Activity Questionnaire) to assess the amount of overall physical activity performed during the past week. One study used an accelerometer to objectively measure this information.

Automatic affective processes
Sixteen studies were designed to investigate automatic affective processes. Individuals with higher levels of physical activity showed more positive automatic affective reactions toward active behaviors compared to individuals with lower levels of physical activity [80][81][82][83][84][85][86]. For example, Bluemke and collaborators [82] used a priming task in which exercise (e.g., to swim, to jog) or control (e.g., to read, to eat) verbs were presented before participants had to quickly categorize positive (e.g., athletic, strong) and negative (e.g., exhausted, tense) target words. Results showed that physically active individuals were faster at categorizing positive target words after exercise primers, whereas inactive students were faster with negative words. Using an IAT contrasting words associated with physical activity (e.g., workout, cross-train, run) and sedentary behaviors (sit, rest, snooze), a study revealed that individuals who were explicitly identified as exercisers had more positive automatic affective reactions toward exercise, as compared to non-exercisers [81]. Additionally, participants who reported greater habitual levels of physical activity also had more positive automatic affective reactions toward exercise compared to participants who reported less habitual physical activity levels. However, one study did not find significant associations between automatic affective reactions and physical activity [87]. Specifically, using an Extrinsic Affective Simon Task with low (reading, resting, and watching television), moderate (walking, cycling, swimming), and high intensity activities words (running, training, and exercising), results showed no differences in the automatic attitudes toward physical activity between a group of children with obesity and a matched control group [87]. Studies also showed that positive automatic affective reactions toward physical activity can prospectively predict physical activity [23,[89][90]. Using the Single Category Implicit Association Test (SC-IAT) [91], a variant of the IAT enabling the measurement of attitudes toward a specific target concept (e.g., physical activityonly) rather than relative attitudes between two targets (physical activity vs. sedentary behaviors), Conroy and collaborators [23] showed that automatic affective reactions toward physical activity positively predicted the number of daily steps over one week, above and beyond controlled processes (e.g., behavioral intentions, outcome expectations). Using the same SC-IAT, Rebar and collaborators [90] revealed that automatic affective reactions toward physical activity prospectively predicted the objectively measured level of physical activity over the next two weeks, above and beyond physical activity intentions. Furthermore, using an IAT contrasting stimuli associated with physical activity and sedentary behaviors, automatic affective reactions toward physical activity predicted adherence to a 14-week heath and exercise course [88] and selfreported recreational physical activity six months after the end of a pulmonary rehabilitation program [89]. However, still using an IAT, a longitudinal study in endometrial cancer survivors did not demonstrate evidence supporting the fact that automatic affective reactions prospectively predicted daily minutes of exercise [92]. Finally, two studies examined how changes in automatic affective reactions were linked to physical activity [93,94]. The first study used the same SC-IAT and revealed that positive changes in affective reactions toward physical activity (i.e., from unfavorable to more favorable automatic evaluations) were associated with an increased self-reported physical activity over one-week [94]. The second study was designed to experimentally manipulate automatic affective reactions using an evaluative conditioning procedure [93]. Participants learned to associated pictures related to exercise (individuals engaging in individual and team sports, such as swimming or basketball) and non-exercise activities (individuals engaging non-physical activity such as watching TV or playing on a gaming console) with pictures associated with positive (individual relaxing in the sun) and negative (individual experiencing neck pain) affective feeling or experiences. Results revealed that participants who learned to associate exerciserelated pictures with positive affective pictures and nonexercise-related pictures with negative affective pictures (i.e., acquisition of positive associations) reduced their negative automatic affective reactions toward physical activity and selected higher intensities on a self-paced cycling task compared to participants in a control condition. A study showed that imagining a positive experience associated with physical activity led to more positive automatic affective reactions toward physical activity [95]. This study also showed more positive affective reactions toward physical activity in frequent exercisers [95].

Attentional bias
Four studies were designed to investigate attentional bias [96][97][98][99]. Overall, results showed higher tendency to approach rather than to avoid physical activity, irrespective of the individuals' level of exercise. Results also showed that physically active individuals showed an attentional bias toward stimuli associated with physical activity, whereas physically inactive individuals showed an attentional bias toward stimuli associated with sedentary behaviors [96][97][98][99].
One study used a Stroop color-naming task in which participants were instructed to quickly indicate the font color of words related to physical activity (e.g., energetic, vigorous, muscle), sedentary (e.g., unmotivated, lethargic, unfit), or control words (e.g., synthetic, suburban, varied) [96]. In this task, the difference in reaction time between exercise and control words and between sedentary and control words was used to infer the degree of attentional bias toward exercise and sedentary behaviors, respectively. Results revealed that regular exercisers (i.e., participants with an athletic identity) showed an attentional bias for exerciserelated stimuli, whereas non-exercisers showed an attentional bias for sedentary-related words.
Another study using a VDP based on pairs of words, with one word related to physical activity associated with a neutral word (e.g., throw-cloth, football-sentence, tennis-devote), revealed a positive correlation between physical activity during the previous week and attentional bias toward words related to physical activity [98]. Another study used a VDP based on pairs of pictures, with one picture of an object related to exercise (e.g., football, stretching bands, field hockey stick, Frisbee) associated with a control picture where the exercise-related object was replaced by a non-exerciserelated object (e.g., remote control, vacuum cleaner, beer bottle) [97]. Results showed an attentional bias toward physical activity in men, irrespective of their habitual level of physical activity, whereas only active women demonstrated such a bias toward physical activity [97].
Another study tested attentional bias in adult patients with anorexia nervosa using eye-tracking [99]. Specifically, this study used a viewing task in which anorexia nervosa patients, physically active participants (i.e., at least 5 h per week of endurance sports), and physically inactive participants (i.e., only performing recreational physical exercise) were presented pairs of pictures, one related to an active situation (i.e., a young female athlete engaging in various physical activity) and one related to an inactive situation (i.e., a young female athlete engaging in various passive situations). They were instructed to freely explore picture pairs presented for 3 s on a computer screen. Results revealed that anorexia nervosa patients and physically active participants had a greater attentional bias toward stimuli associated with physical activity than physically inactive participants. Additionally, in anorexia nervosa patients, attentional bias toward physical activity-related stimuli strongly correlated with self-reported amount of physical activity.

Automatic approach tendencies
Three studies were designed to investigate automatic approach tendencies toward physical activity and sedentary behaviors [21,22,100]. Overall, results showed that automatic approach tendencies toward physical activity positively predicted physical activity, whereas automatic approach tendencies toward sedentary behaviors negatively predicted physical activity [21,22]. All these studies used a manikin task based on pictograms representing physical activity and an active lifestyle (e.g., a pictogram of running, swimming, cycling) or rest and sedentary lifestyle (e.g., a pictogram of watching TV, lying on the sofa, resting). Two studies showed that automatic approach tendencies toward physical activity predicted higher involvement in nonvolitional physical activity in a laboratory context over and above intentions to be physically active [21] and free time spent in physical activity over one week as measured with an accelerometer [22]. Moreover, automatic approach tendencies toward sedentary behaviors negatively predicted involvement in physical activity.
A study was designed to test whether automatic approach tendencies toward physical activity and sedentary behaviors can be manipulated using approach bias modification training, and subsequently impact exercise behaviors [100]. Results showed that participants trained to systematically approach physical activity and avoid sedentary behaviors spent longer periods of time exercising in the laboratory after the training compared to participants systematically trained to approach sedentary behaviors and avoid physical activity [100].

Brain substrates
Three studies reported potential brain substrates of reactions triggered by stimuli associated with physical activity and sedentary behaviors using fMRI. Results showed that some brain areas activated in response to these stimuli were consistent with areas highlighted in the reward literature [101][102][103]. However, an area shown to be involved in both the wanting and liking components of reward, the nucleus accumbens, was not reported in any of the 3 studies, thereby calling for further investigation. One study was conducted to identify the neural correlates involved in the control of brisk walking [101]. Young healthy individuals were asked to imagine themselves in three situations: Brisk walking in a long corridor, standing, and lying while their brain activity was measured using fMRI. Results revealed a stronger activation during mental imagery of brisk walking compared to mental imagery of standing or lying in areas associated with reward: insula, pallidum and caudate. Another study examined inhibition response to active-and inactive-related stimuli [103] in anorexia nervosa patients, physically active participants (at least 5 h per week of endurance sports), and physically inactive participants (casual physical exercise) using a go/no-go task including stimuli associated with physical activity (e.g., a physically active person) and physical inactivity-(e.g., a physically inactive person) related pictures. The brain areas activated in this study were not related to reward. The last study tested the neural responses to pictures of physical activity and sedentary behaviors in a sample of overweight versus lean women [102]. Participants were asked to watch physical activities, sedentary activities, and landscape pictures presented during fMRI scanning.
Results revealed an increased activation in brain areas associated with reward (amygdala, putamen, limbic lobe) when viewing pictures of physical activities compared to sedentary activities and control stimuli. Sedentary stimuli also activated the amygdala. Additionally, as body mass index increased, the activation of the right putamen decreased. Finally, overweight women showed a decreased activation when watching sedentary compared with control stimuli in the insular cortex. These imaging results should be considered with caution as the experiments were not designed to investigate the brain regions associated with reward.

Discussion
The objective of this work was to 1) propose a new model of reward and automatic exercise behavior, and 2) to systematically review studies investigating automatic reactions to stimuli associated with different types of exercise behavior (e.g., physical activity and sedentary behaviors) and energetic cost variations (e.g., behaviors minimizing energetic cost). We included 26 studies. Three types of automatic processes were tested: Affective reactions, attentional capture, and approach tendencies.

Main findings
Overall, results showed that automatic processes favored physical activity rather than sedentary behaviors. Results also showed that individuals' level of exercise affected automatic processes toward exercise-related stimuli. Specifically, physically active and inactive individuals showed an attentional bias toward stimuli associated with physical activity and sedentary behaviors, respectively. Individuals reporting higher levels of physical activity showed higher positive automatic affective associations toward stimuli associated with exercise compared to individuals with lower levels of exercise. These affective associations prospectively predicted physical activity above and beyond controlled precursors of exercise behavior. Finally, results showed that automatic processes could be experimentally manipulated to increase the level of physical activity. The fMRI results were scarce and suggested that physical activity and sedentary behaviors are related to reward processes. Specifically, activations of the basal ganglia (pallidum and caudate), amygdala, putamen, and prefrontal cortex are in line with the reward studies in the field of eating or addictive behaviors. However, an area shown to be involved in both the wanting and liking components of reward, the nucleus accumbens, was not reported in any of the three studies, thereby calling for further investigation. Studies specifically testing the brain substrates of reward in exercise behavior are still scarce and additional research investigating the brain substrates triggered by rewarding stimuli associated with physical activity, sedentary behaviors, and BMEC is required.

Current issues and perspectives
Besides the major findings reported above, this systematic review highlighted several theoretical and methodological issues that should be addressed in the future.

BMEC as a reward
The studies included in this review focused on automatic behaviors triggered by stimuli associated with physical activity. When stimuli associated with sedentary behavior were used, they were considered as a control condition most of the time. None of the studies were specifically designed to investigate the automatic behaviors or brain activation triggered by sedentary behaviors or BMEC. These findings reveal a knowledge gap in the literature of exercise behavior and highlight the necessity to address the potential rewarding value of BMEC and sedentary behaviors in future studies.

Neurophysiological studies are needed
This review showed higher tendency to approach rather than to avoid physical activity, irrespective of the individuals' level of exercise. While this result supports the fact that physical activity is a reward, it does not discard the possibility that cost minimization is also one. What we observe may result from facilitation processes but may also result from the competition of facilitation and inhibition processes supporting two different rewards (i.e., physical activity and BMEC). Purely behavioral work may never resolve this uncertainty and a neural approach is therefore needed.

Recent exercise history
Very few studies controlled for exercise history over the days or hours preceding the experiment. This lack of control is an issue worth considering because in other contexts, the perception of reward has been found to be dependent on the physiological state [47,104]. For instance, thirsty participants showed higher perceptual readiness to drinking-related stimuli [48] and hungry participants showed stronger automatic approach reactions toward food-related stimuli [50,105]. As mentioned in the Implications section, if BMEC are a reward, the reward associated with sitting and lying positions depends on the participant's maximal exercise capacity and recent exercise history (e.g., did the participant come by bike or bus). These factors should therefore be considered during the conception of the study design.

Specific behaviors
This systematic review revealed that all the included studies focused on specific exercise behaviors (e.g., running, dancing, swimming) and/or sedentary behaviors (e.g., watching television, reading, video gaming). However, specific exercise behaviors can be rewarding due to the energetic cost they are associated with, but they can also be rewarding due to other factors. For example, the pleasure associated with a picture of an individual playing soccer may reflect the pleasure felt when watching this sport on TV, not the actual experience of playing soccer. The pleasure associated with a video gaming stimulus unlikely solely stems from the fact that the individual plays in a seated position. The pleasure associated with a skiing-related stimulus, the positive value ascribed to the stimulus (i.e., reward), may result from the liking component of speed perception, not only to value ascribed to the energetic cost associated with skiing. Therefore, it is difficult to infer strong conclusions from studies using an approach based on specific exercise behaviors.

Conclusion
Overall, results showed that automatic reactions toward stimuli depicting exercise behavior explained levels of physical activity. Imaging results showed that some brain regions associated with reward were activated by stimuli associated with physical activity and sedentary behaviors, but these studies remain scarce. These results highlight the importance of reward in exercise behavior. This systematic review also reveals a knowledge gap that further highlights the necessity to reassess the veracity of the dominant neuropsychological approaches to exercise behavior. Neurophysiological techniques may afford the establishment of a strong and complete framework of reward in exercise behavior. Finally, this review draws an emerging line of research that has the potential to initiate the development of individualized and efficient interventions to counteract the pandemic of physical inactivity.

Conflict of Interest
Authors declare that they have no conflict of interest. Author contributions MPB and BC conceived the new model of exercise behavior; MPB and BC performed the systematic review; MPB and BC wrote the first draft; All authors contributed to improve the manuscript.