Dopaminergic activity and exercise behavior in anorexia nervosa.

Driven exercise (i.e., the tendency to exercise in excess to influence weight/shape or regulate emotion) is difficult to manage in the context of anorexia nervosa, and is associated with poorer treatment outcomes, and psychological and medical severity. Driven exercise is observed in a considerable number of those diagnosed with anorexia nervosa; however, to date, this hallmark symptom remains poorly understood. Dopamine signaling is implicated in motivating and maintaining appetitive behavior among patients with eating disorders; but, much less is known about the role of dopamine signaling specific to the symptom of driven exercise. An improved understanding of this biobehavioral mechanism may inform the etiology of driven exercise in anorexia nervosa, with the potential to impact future research and treatment efforts. This review describes the role that dopamine serves in maintaining symptoms in the context of anorexia nervosa, and synthesizes current relevant evidence on exercise in AN and related dopaminergic activity. Throughout, theoretical implications are discussed, along with critical directions for future research.


Introduction
Anorexia nervosa (AN) is a deadly psychiatric disorder associated with significant medical morbidity and mortality (1), marked distress and impairment (2), and considerable treatment costs (3). Given these concerns, as well as the significant global health burden associated with AN (4), efforts have increasingly been made to better understand key biobehavioral mechanisms that influence the pathogenesis and maintenance of AN, including dopamine (DA) signaling (5). DA has been studied extensively for its role in behavioral motivation driven exercise as exercise dependence likens it to other forms of behavioral addiction (e.g., substance misuse, gambling), which are robustly supported by literature confirming the role of DA in the maintenance of these behavior patterns (40). Another leading model of driven exercise in EDs refers to this symptom as compulsive exercise, whereby repetitive behaviors may serve to reduce anxiety and are therefore reinforced by reductions in negative affect (11)(12)(13)33). Exercise dependence and compulsive exercise represent competing theoretical models for the maintenance of exercise behaviors in EDs, whereby these behaviors are either positively reinforced by rewarding experiences or negatively reinforced by way of reduction in negative affect, respectively. In both of these models, exercise behavior may become entrenched and habitual, despite adverse consequences (e.g., extreme weight loss; cardiac risk). While habit-related behavior is typically considered less DA-mediated, in both models, learning occurs via either positive or negative reinforcement, in behavioral reward processes that intricately involve DA signaling (41).

Dopamine: mechanisms of action in anorexia nervosa
Basic animal and human research supports the notion that physical activity is influenced by neurochemical pathways associated with reward (42,43). Specifically, ventral tegmental area (VTA) DA neurons serve a central role in motivated behavior and reward processing (44)(45)(46)(47). DA is thought to operate primarily through projections from the midbrain VTA through the mesolimbic pathway into the ventral striatum of the basal ganglia in the forebrain (i.e., nucleus accumbens [NAc]), as well as the mesocortical pathway to the prefrontal cortex. These dual pathways provide the initial groundwork for generating conditioned responses (48,49), both in reinforcing tendencies to approach stimuli that are associated with primary rewards (e.g., food), and for maintaining habit strength once behaviors have been learned (46).
The role of DA in behavioral stimulus-reward associations includes two hypotheses relating to its mechanism of action: DA mediates (1) the pursuit of a reward based on attribution of incentive salience (i.e., a cognitive process that confers a wanting attribute), and (2) the 'stamping in' of anticipated future rewards (i.e., learning/habit formation) (44,47,(50)(51)(52). Of note, DA was historically thought to also mediate the hedonic impact of a reward (i.e., liking) (53), but the hypothesis that DA directly impacts the experience of pleasure is currently less well-supported (54). In contrast, hypotheses related to incentive salience (i.e., wanting) and learning/habit formation are well-supported and differentiated in experimental testing (47); however combined research findings suggest that the actions of DA are neither necessary, nor sufficient to independently and fully explain learning/habit formation (5). Instead, evidence suggests that habit formation may be less DA-dependent whereas DA is necessary for wanting, and sufficient to support incentive salience (55,56). Of note, and particularly when discussing exercise behavior, it is possible to want something in the absence of liking it. Further, for an individual who experiences reward value from driven exercise, an interaction likely occurs between these related processes (i.e., wanting, learning/ habit formation); however, most evidence suggests that DA is exerting its influence through incentive salience attribution that leads associated rewards to become wanted (50).
Individuals with AN typically report wanting an alteration in their body weight or shape, or wanting to remain at their current weight, for fear of weight gain (57). Other hallmark symptoms of AN include wanting to engage in dietary restriction, or behaviors that are compensatory for caloric intake (e.g., self-induced vomiting; exercise) (57). Of note, wanted behaviors reported among those with AN reflect similarities with the compulsive nature of behavior in obsessive-compulsive disorder, as well as the dependent nature of behavior in substance use disorder (58). In the context of AN, it can be difficult to determine whether the conferral of wanting is either a state-or trait-dysregulation, as changes in DA signaling among those who are acutely ill with AN may be normative responses to starvation (59). Specifically, mesolimbic DA neurons have been implicated in the motivational aspects of feeding behavior (60), and physiologic factors resulting from sustained weight loss stimulate food intake (61); however, this foraging impetus is in direct conflict with a high drive for thinness and body dissatisfaction in individuals with AN (8,62). This conflict is made more complex by midbrain DA activity which may also serve to initially reward dieting (59).
To further explore the question of state-versus trait-level dysregulation in the DA system, some work examining patients with AN during and following treatment suggests that aberrations in DA signaling may contribute to the etiology of the disorder (9,63). For example, one study of adolescent females found that when acutely ill, reward response was elevated for unexpected reward receipt and omission, and prediction error in those with AN, relative to healthy controls (64). Of note, prediction error and unexpected reward omission normalized with weight gain, but unexpected reward receipt remained significantly elevated, suggesting this may be a premorbid trait level difference specific to those who develop AN. There is consistent evidence that homovanilic acid, a major metabolite of DA, is decreased in the central nervous system among those with AN who are underweight and in the acute phase of illness (9,65). Further, a negative correlation has been demonstrated between striatal D2 receptor availability and body mass (66), and there is evidence that polymorphism of DA D2 receptor genes may be a susceptibility factor in the development of AN (67). Another study found that individuals with AN who were weight restored showed increased D2/D3 receptor binding in the ventral striatum; these findings suggest that either upregulated D2/D3 receptor activity or decreased intrasynaptic DA concentration -or both, might exist even post-recovery (63). While distinguishing changes that are due to trait from those that are related to state among those with AN has been a considerable confound in the research of this disorder (68), aberrations in the DA system following weight restoration suggest a trait level contribution to characteristic symptoms of AN.

DA signaling and neural circuity.
Accounting for differences in reward response, the ventral regions of the basal ganglia, including the VTA, play a pivotal role in reward and reinforcement, as well as in the development of addictive behaviors and habit formation (69). Despite what is generally known about DA, the mesolimbic reward system and projections to the NAc, a considerable gap in knowledge remains, specifically in understanding the precise circuitry that supports differential responding to appetitive versus aversive stimuli (70). Recent work with a rodent model suggests that the encoding of either reward or aversion in the NAc is closely associated with a cell's anatomical location, supporting the notion that a 'hedonic hotspot' (i.e., area most responsible for pleasure associated with intrinsic rewards) is site-specific within the limbic area (54,71,72). Specifically, DA in the ventromedial NAc was found to encode salience of a cue, whereas DA in the dorsal medial NAc and dorsal lateral NAc was found to encode the value signal. Further, DA signaling in the ventrolateral and central ventromedial NAc seemed to encode both value and salience (71). These preliminary findings support a prior study where DA terminals in the ventromedial NAc were activated by aversive cues, whereas DA signaling to reward-predictive cues were associated with lateral subregions of the NAc (70). To date, no studies have specifically examined these processes on a neurocircuit level, among patients with AN, arguably an important line of future inquiry.

Homeostatic hormones and DA signaling.
In the context of appetitive behavior, hormones that regulate our homeostatic system can interact with, and override, DA signaling (60). Specifically, leptin inhibits DA neurons whereas ghrelin activates them; the interaction of these hormones with DA signaling may contribute to disturbances in food-related cognitive processing evidenced among those with AN (73). There is also some evidence that α-melanocyte-stimulating hormone, a hormone associated with appetite control, may modulate an exaggerated DA release in the lateral hypothalamus (LHA) during food anticipation and consumption (74). Among rodents, elevation in LHA DA levels was inversely related to sucrose intake, suggesting that this DA release surrounding feeding behavior may be associated with hedonic satiety, and decrease motivation to obtain food (74).
As a general consequence of caloric restriction in AN, reduced leptin signaling in the VTA is posited to increase DA signaling, and drive increased physical activity (75). This phenomenon has been demonstrated in an animal model, whereby leptin injections in the lateral ventricle and VTA of rodents under caloric restriction led to suppressed running wheel activity (75). Ghrelin has also been shown to mediate the link between reward and physical activity in a rodent model. Specifically, local administration of ghrelin into the VTA increased both intracellular DA levels in the NAc, as well as greater locomotor activity; a similar effect was found with administration of ghrelin into the lateral dorsal tegmental area (76).
Among human patients with AN who report driven exercise, a negative correlation has been shown between increased activity and leptin levels (77,78). Notably, low leptin levels are characteristic of acute AN (79), which may be further reduced in the context of high activity (78). Further investigation of associations between physical activity (i.e., daily step counts) with specific biological markers determined that ghrelin (80) and urinary cortisol (81) are positively associated with physical activity in inpatients with AN. Recent systematic review of differences in cortisol concentrations in adolescents with EDs, and specifically AN, determined that cortisol levels were higher among those with AN compared to healthy controls, and a reduction in cortisol levels occurred after nutritional recovery (82). Given that hypercortisolemia seems to be correlated with both self-reported compulsivity in exercise behavior and actual activity level among inpatients with AN (81), cortisol may be directly implicated in problematic exercise behavior. Taken together, via human and animal models, both food-related cognitive processing and also driven exercise behavior in AN may be affected by homeostatic biomarkers like ghrelin, leptin and cortisol, via their dynamic effects on the DA system (73).

Activity Based Anorexia (ABA).
Efforts to better understand the etiology of AN and the role of DA within it have included a widely used rodent model mimicking features of human AN: the activity-based anorexia (ABA) model (83). The ABA model demonstrates that when rats are fed one meal a day and allowed to run on an activity wheel, they stop eating, run excessively, and expire due to starvation (84). While the behavior of these animals is functionally similar to humans who self-starve, the ABA model cannot account for psychosocial components of ED (e.g., anxiety), that motivate driven exercise among humans. However, this rodent model includes aberrant reward response (85), anhedonia (86), and hippocampal processes that modulate anxiety (87), all of which align with psychosocial human models of compulsive exercise (12). Both in the ABA model, as well as in research with patients diagnosed with AN, considerable attention has been devoted to the role that DA may play in the etiology and maintenance of this disorder (6).
Notably, while certain aspects of rodent models are difficult to reconcile in human experience, a recent comprehensive review of research on the ABA model determined that DA signaling and dysregulation in related reward circuitry was similar in rodents as in humans with AN (88). Rodents with typical signs of ABA show elevations in D2 receptor expression in the caudate putamen (89); while no research to date has specifically examined D2 receptors in patients with AN who report driven exercise, as noted above, dysregulated D2 receptor function has been noted in patients with AN (67). Treating rodents with olanzapine (i.e., an atypical antipsychotic that blocks DA and serotonin receptors) in an ABA study paradigm reduces wheel running activity (90); another study reported that olanzapine, but not fluoxetine (i.e., selective serotonin reuptake inhibitor) was effective in reducing ABA among mice (91). Together, these studies suggest that reducing activity via olanzapine treatment is likely due to its mechanism of action in altering DA, rather than serotonin signaling. Several ABA studies demonstrate that DA alteration, and specifically D2/D3 receptor antagonism, seem to have beneficial effects in rodents via increasing food intake and body weight, and reducing wheel running activity (90,92,93). In adult human AN, the use of olanzapine in a clinical trial demonstrated modest weight gain, but no change in psychological symptoms; in particular, with regards to driven exercise behavior, in this trial there was no mitigation of obsessional symptoms or overconcern with gaining weight (94). Retrospective chart review of adolescents with AN who were treated in a specialized ED program also found modest increases in weight among those who received aripiprazole, a partial D2 agonist, compared to those who had not; this study did not specifically include assessment of compulsivity, other psychological symptoms, or activity (95). In summary, given preliminary findings that suggest olanzapine may not mitigate psychological aspects of AN but may assist in weight gain, studies that antagonize D2/D3 receptors warrant further substantiation in humans with AN (92), and particularly among those who report driven exercise.

Neurobiological models of symptom maintenance in anorexia nervosa
To more broadly explain the mechanisms of action related to DA and the maintenance of AN symptoms, two neurobiologically informed theories involving striatal function have been proposed: a reward deficit model (96,97), and a habit-related model (98,99). Described below, each of these models implicates DA activity, and together reflect a continuum in connectivity from reward-to habit-related brain regions that may help to explain how AN symptoms arise, and become maintained.

Reward-deficit model.
The reward-deficit model (figure 1) posits that there are broad deficits in the ventral frontostriatal network in AN (96,97), which may be evidenced in under-response to stimuli that typically yield a primary reward response (e.g., highly palatable food) (100), and increased response to disorder-salient cues (e.g., thinness) (101). In the reward-deficit model of AN, altered DA functioning is thought to affect the reward valuation of stimuli, resulting in reduced response to disorder-nonspecific reward stimuli and increased response to disorder-specific stimuli.
When considering driven exercise behavior as a symptom of AN, relative to the rewarddeficit model, an individual might endorse increased experience of positive affect in response to any form of exercise. In particular, they might report elation in anticipation of, or in response to, intense cardiovascular exercise. In turn, they may also report decreased enjoyment, irritability and discontent when asked to participate in a sedentary activity that others without driven exercise symptomology find rewarding (e.g., watching a movie in a preferred genre).

Habit-centered model.
The habit-centered model (figure 1) view initially aligns with the reward-deficit model in positing that by way of dysfunction in the ventral striatum, a symptom may be at first goaldirected, and motivated by incentive salience, particularly to disorder-specific cues (e.g., by wanting weight loss). However, this theory diverges from the reward-deficit model in postulating that over time, an AN symptom may shift to a stimulus-control response (i.e., habitual pattern), and be maintained, despite negative consequences (102). Over-training of reward responses to disorder-relevant stimuli is thought to eventually foster a compulsive reliance on rigid and automatic ED behaviors, independent of momentary DA responding (103). On the neural level, this shift from behavior that is flexible to that which is relatively automatic might represent a transition from prefrontal to striatal control of behavior (104). The habit-centered model thus posits that once an individual has fully developed AN (i.e., after initial stages of illness, when symptoms first develop), dysfunctions in the dorsal frontostriatal network may supercede those in more ventral, reward-oriented brain regions (99,105).
When considering driven exercise behavior as a symptom of AN, relative to the habitcentered model, an individual might report that early on in their disorder, they participated in occasional runs, a few per week, which initially had felt pleasurable and in line with their desire to lose weight. Over time, the amount of running they report feeling obligated to perform on a daily basis may become regimented and obligatory, such that even when the weather is inclement and running outside is objectively miserable, completing a certain number of miles is considered non-negotiable.
In summary, the reward-deficit model suggests a lack of reward response to non-disorder specific cues, with no deficit noted when registering reward response to disorder-relevant cues. The habit-centered model essentially extends the reward-deficit model, reflecting what might transpire when intact reward responses (to disorder-relevant cues) become overrehearsed and entrenched. In the context of AN, behaviors that result in instantaneous or expedient weight loss may be particularly rewarding when initially implemented. For example, when considered on a behavioral level, the reward-deficit model would support an explanation for how exercise may be initially rewarding for those with AN. At first, weight loss is tangibly witnessed soon after initiation of increased cardiovascular activity. Over time, negative consequences will likely arise from regular excessive exercise (e.g., injury; exercising at the sacrifice of other necessary or meaningful activities), and engagement in this behavior will be less rewarding (i.e., necessitating larger doses for the same effect; feeling compelled to exercise even in inclement weather). The habit-centered model would support why an individual might continue exercise over time, perhaps fearing that cessation of this learned 'habit' may lead to even worse perceived negative consequences, i.e., weight gain.
In efforts to better understand the synchrony of these two neural networks (i.e., reward and habit), recent work compared resting-state functional connectivity patterns between patients with AN and healthy controls (106). Findings suggested that compared with healthy controls, those with AN demonstrated widespread dysconnectivity in areas implicated in habit learning, as well as poor functional coordination between key areas associated with reward processing; together, these findings lend credence to both the habit-centered and reward-deficit models of AN (106).

Biological study of exercise and reward
Acutely ill with anorexia nervosa.-The limited research that has investigated neurobiological correlates of exercise in AN has identified differences in brain structure, including larger hippocampal volumes among females with AN who reported pretreatment excessive exercise (107).
Turning from structure to connectivity and function, study of reward and related learning in human AN has focused on disorder-specific cues (e.g., food) and presumed rewards (e.g., money), with very little attention specifically paid to exercise behavior. In neuroimaging studies that have employed food-related image cues (i.e., provoking core AN symptoms), findings support an increase in medial pre-frontal cortex (PFC) activity (108,109). However, when completing tasks that challenge cognitive control, restrained eaters such as those with AN demonstrated reduced PFC activity, as compared to healthy controls (110). In one study using functional magnetic resonance imaging (fMRI) and two affective go/no go tasks to probe inhibitory control, individuals with AN showed reduced response inhibition for food and non-food images in the putamen, as compared with healthy athlete and non-athlete controls (111). It is notable that this study found hypoactivation in the putamen of those with AN, independent of stimulus category, particularly given prior work that has identified the implication of increased activity of the putamen in stimulus-response habit learning (112). In contrast, exercise-specific cues (e.g., an image of an individual doing physical exercise) resulted in an exaggerated response in the PFC and cerebellum in AN patients, which the authors proposed reflects increased response inhibition and reward attributed to exercisespecific cues (111). These results align with the reward-deficit model described above, which highlights altered DA functioning resulting in increased reward response to disordercompatible stimuli (96,97).
Other work probing reward response used eye-tracking to compare visual processing of active and non-active cues (i.e., pictures of individuals exercising or sedentary) among patients with AN, athlete controls, and non-athlete controls (113). Those with AN and athlete controls demonstrated greater attentional engagement and increased pleasure ratings toward exercise-related stimuli, compared to non-athlete controls. Notably, self-report trait reward sensitivity and attentional orientation toward activity images were only evidenced among those with AN, suggesting that this patient population attributes more reward value to cues associated with exercise than non-clinical controls. A study using computational tasks also demonstrated evidence that relative to controls, those currently ill or recovered from AN were more motivated by rewards associated with exercise (114). In this study, using a progressive ratio breakpoint task, those with AN were willing to expend more effort for access to a small amount of exercise, as compared to healthy controls. Taken together, both of these studies showed implicit and explicit evidence suggesting that those with AN (including those who are recovered) find exercise-related cues more rewarding. However, it is important to note that task-related activity may not generalize to real-world clinical situations. Further, it is quite possible that neural correlates of both goal-directed (i.e., reward-based) and compulsive (i.e., habit-based) behavior may simultaneously or alternately activate; thus, parsing these processes in the context of AN and exercise behavior is a line of inquiry that warrants future research.
Recovered from anorexia nervosa.-In efforts to address the stability of reward attribution outside the context of acute illness, some work has used an acute phenylalanine/ tyrosine depletion (APTD) method to transiently decrease DA signaling among those recovered from AN (i.e., mimicking DA levels during acute illness). For example, one study found that those who were recovered from AN implicitly appraised underweight and exercise cues (as measured via eye-blink startle response to photos of thin and physically active female bodies) as more rewarding than did healthy controls (115). Notably, these effects were not demonstrated when recovered individuals were DA depleted, suggesting that this reward attribution process may, in part, be DA-dependent. These findings support the reward-deficit model, whereby positive incentive salience (i.e., wanting) appears to be mediated by DA processes, which contribute to AN symptom maintenance.
In contrast, in another study that used APTD methodology, among those who were recovered from AN, no differences were evidenced in willingness to work for exercise as a reward during a behavioral task paradigm (116). Notably, this study found that decreasing DA does not reduce motivation to exercise in those recovered from AN, but does so among healthy controls; these findings suggest that in AN, driven exercise develops largely independent of DA-mediated reward processes and instead, becomes dependent on neurocircuitry that regulates compulsive, habit-informed behavior. These results support the aforementioned habit-focused model, and reflect potential vulnerability to disorder-specific behavioral patterns, even after weight restoration and recovery. While there is no current circuit model that comprehensively explains habit formation across disorders (117), a shift from the reward-deficit model to a habit centered model in AN may reflect the status of DA receptor density and affinity throughout the progression of the disorder. Specifically, intracellular DA deficits, and/or upregulated DA receptor activity resulting from sustained weight loss in AN may facilitate a shift from goal-directed behavior (i.e., reward) to stimulus control (i.e., habit) (98). Evidence that upregulated D2/D3 receptor binding continues after weight restoration (63) may provide support for why some AN habits remain entrenched, with suspended time required for full recovery from this disorder (98).
Taken together, results from preliminary studies highlighted here that use DA-depletion methods appear to support aspects of both reward-deficit and habit-focused models of reward processing.

Discussion
In humans with AN, limited research has specifically attended to the mechanism of action that DA may have in motivating and maintaining the symptom of driven exercise. However, DA is broadly implicated in reward response, and intricately related to both the rewardrelated and habit-focused models of AN (96,98). In addition, study of hyperactivity in rodents has robustly supported the role of DA in impacting ABA, with consistent evidence that disrupting DA signaling can mitigate wheel running activity (90,92,93). Of the studies highlighted in this review that focus specifically on exercise in the context of human AN, some work has compared those with AN with healthy controls, while other studies have downregulated DA function among those who were recovered from AN. In the former, evidence suggests increased neural response to exercise stimuli in the PFC and cerebellum (112), greater visual attention to exercise stimuli (113), and greater willingness to work for an exercise reward among those with AN (114), compared to healthy controls. While these findings broadly suggest that those with AN find exercise more rewarding, it remains uncertain why those with AN were more attended to, or willing to work for exercise. For example, rather than attributing more reward value to exercise, willingness to work may instead reflect a desire to avoid negative affect (e.g., low mood associated with current weight; guilt for eating; compulsivity in habitual exercise behavior). If exercise is motivated by avoidance of negative affect, the symptom is negatively reinforced, and akin to escape rather than reward. In either case, the use of an in-laboratory paradigm (i.e., pictures of female bodies exercising) to elicit reward response may not replicate the lived experience and salience of exercise itself, including physiological responses to activity (e.g., increased heart rate, endocannabinoid release). Future work will benefit from studies of in vivo exercise in the context of AN, that can capture both the reward as well as negative reinforcement value of physical activity.
In studies that investigated DA and exercise among those recovered from AN, startle potentiation was elevated in response to exercise stimuli, but not under circumstances where DA was depleted (i.e., DA-state dependent) (115). In contrast, those recovered from AN were still more motivated to work for exercise than healthy controls, even when DA was depleted (i.e., DA-independent) (116). Taken together, these studies support the rewarddeficit and habit-focused models of AN, respectively, but notably cannot fully explain whether vulnerability to hyperactivity is present before illness onset. Symptoms that are DAindependent may be important indicators of potential susceptibility towards illness-related routines even after recovery.
Given the medical acuity of AN, and the high prevalence of driven exercise in this patient population, it is critical that future work investigate biobehavioral mechanisms like DA function that serve to initiate and maintain this symptom. There are several critical lines of inquiry that may guide future research, including whether reward response specific to exercise stimuli is state (relative to acute illness) or trait (premorbid to the illness) dependent. While behavioral and neurobiological response to exercise cues is very likely a combination of both DA-mediated and other interactive mechanisms, efforts to accurately distinguish whether vulnerability for driven exercise pre-dates AN onset warrants further targeted study of DA receptor antagonism.
Secondly, future research design would ideally distinguish between reward response and negative reinforcement (i.e., escape from or avoidance of negative affect). In the psychosocial models of exercise detailed previously, exercise in the context of AN is posited to reflect either its rewarding properties (i.e., exercise dependence) (39) or alternatively, its anxiolytic properties (i.e., compulsive exercise) (12). Study of neural response and DA signaling may help to reconcile which one of these models is more accurately reflective of either reward-or negatively-reinforced behavior, at different time points over the course of illness and recovery. Even within one model, i.e., the compulsive exercise model, positive and negative reinforcement may both be implicated. For example, compulsivity may create distress when one is not able to engage in intended exercise; subsequently, exercise may become negatively reinforcing in alleviating this distress. By the same token, compulsive exercise might also be positively reinforced, by factors such as mood improvement, or praise from others in regards to improved fitness. Considered together, the degree to which positive and/or negatively-reinforced behavior is implicated in driven exercise behavior, and the degree to which this learning is DA-dependent, warrants further inquiry. Finally, the current (albeit minimal) research on reward and exercise in AN, appears to support proposed neurobiological models of human AN (i.e., reward deficit and habitfocused models) (30,31). However, while some appetitive AN symptoms are posited to be at first be reward-related and then shift to being more habit-controlled, it remains unclear if this neurocircuitry shift also happens for exercise behavior, and if so, when it may occur during the course of illness. While some parallels between ABA and human AN have drawn criticism, DA signaling and the reward systems are similar across species, with compelling evidence that D2/D3 receptor antagonism may decrease physical activity (92,93). These findings suggest that further investigation of DA signaling, and receptor antagonism specifically related to exercise behavior among humans with AN is warranted.

Summary and Conclusions
Limited research has specifically attended to the mechanism of action that DA may have in motivating and maintaining the symptom of driven exercise in the context of AN. Preliminary evidence suggests increased reward response to exercise stimuli in laboratory paradigms among those acutely ill with AN, compared to healthy controls. Among those recovered from AN, mixed findings indicate that response to exercise cues are both DAmediated, potentially DA-independent, as well as premorbid to illness onset. Future research is critical to improving understanding of the biobehavioral mechanisms that maintain exercise behavior in AN, with specific lines of inquiry in DA-related reward processing.