Cerebral μ-opioid and CB1-receptor systems have distinct roles in human feeding behavior

Eating behavior varies greatly between healthy individuals, but the neurobiological basis of these trait-like differences in feeding remains unknown. Central μ-opioid receptors (MOR) and cannabinoid CB1-receptors (CB1R) regulate energy balance via multiple neural pathways, promoting food intake and reward. Because obesity and eating disorders have been associated with alterations in brain’s opioid and endocannabinoid signaling, the variation in MOR and CB1R systems could potentially underlie distinct eating behavior phenotypes, also in non-obese population. In this retrospective positron emission tomography (PET) study, we analyzed [11C]carfentanil PET scans of MORs from 92 healthy subjects (70 males and 22 females), and [18F]FMPEP-d2 scans of CB1Rs from 35 subjects (all males, all also included in the [11C]carfentanil sample). Eating styles were measured with the Dutch Eating Behavior Questionnaire (DEBQ). We found that lower cerebral MOR availability was associated with increase in external eating – individuals with low MORs reported being more likely to eat in response to environment’s palatable food cues. CB1R availability was negatively associated with multiple eating behavior traits. We conclude that although MORs and CB1Rs overlap anatomically and functionally in the brain, they have distinct roles in mediating individual feeding patterns.


Introduction
Obesity is one of the leading public health issues, resulting from individuals' long-term excessive energy intake in relation to energy expenditure (1). Yet, humans vary greatly in their choices and habits related to food intake quantity and quality i.e. eating behavior (2,3). Trait-like eating behaviors have been associated with multiple clinical eating disorders in addition to obesity (4)(5)(6)(7), but also nonobese individuals vary in how they control their feeding (8). Interacting with peripheral hormones, central nervous system (CNS) integrates hunger and satiety signals with environmental stimuli for regulating food intake (1). Large-scale genome-wide association studies have identified limbic system, hippocampus and hypothalamus to be key regions in the CNS contributing to individual's body mass index (BMI) and eating behavior (9,10). Central regulation of feeding is however constantly challenged by the modern environment characterized by abundance of palatable and energy-dense food products, promoting feeding independently of metabolic needs (11,12).
Palatability and hedonic properties of food are centrally mediated by µ-opioid receptor (MOR) system (13,14). Both endogenous and exogenous opioids stimulate feeding, especially via hedonic hotspots of nucleus accumbens, insula and frontal cortex (15)(16)(17)(18). Conversely, opioid antagonists reduce food intake and related hedonic responses in rodents (18) and humans (19,20). Human positron emission tomography (PET) studies have revealed that obesity associates with decrease of MORs in appetite regulating brain areas (21,22), and insular MORs are lowered in patients with bulimia nervosa proportionally to fasting behavior (23). Central MOR system function varies considerably also in healthy humans (24), and traits linked with feeding control such as impulsivity are associated with MOR availability (25). Nevertheless, the association between the MOR system and specific patterns of eating behavior remains elusive.
Feeding is also regulated by brain's endocannabinoid system, which overlaps anatomically and functionally with the central MORs (26). The most abundant central cannabinoid receptors are the CB 1 -receptors (CB 1 Rs), which regulate food intake through circuits of ventral striatum, limbic system and hypothalamus (27,28). Functional interplay between MOR and CB 1 R systems is highlighted in animal studies, where CB 1 R-antagonists and MOR-antagonists have synergistic effect on reducing food intake (29), and CB 1 R-antagonist can be used to block MOR-agonist induced food intake and vice versa (30). MOR-agonists also directly increase endocannabinoid concentration and CB 1 R-agonists increase opioid concentration in the brain, including nucleus accumbens (31,32). In humans, CB 1 R-antagonist rimonabant showed promise as an anti-obesity drug, but was withdrawn due to psychiatric side effects (33). More nuanced understanding of CB 1 R system and feeding is clearly required to enable further pharmacological advancement.

The current study
Accumulating evidence suggests that variation in central MOR and CB 1 R function could be linked to feeding and pathological eating behavior traits in humans, but it remains unresolved what facets of feeding they govern in humans. Individual differences in eating patterns can be conceptualized based on the psychological mechanisms that contribute to or attenuate development of overweight. In such conceptualization, emotional eating refers to reactive overeating to distress or negative emotions, while external eating refers to tendency to overeat in response to attractive foodcues. Finally, restrained eating refers to the tendency to eat less than desired (34)(35)(36). The emotional and external overeating are based on psychosomatic and externality theories of eating behavior, while restrained eating dimension centers around food intake self-inhibition (37). Such consistent patterns contribute to differences in weight gain and maintenance (34,38), and they can be measured using

Subjects
The study subjects were historical controls retrieved from the AIVO neuroinformatics database

Eating behavior assessment with the DEBQ
The Dutch Eating Behavior Questionnaire (DEBQ) (37) was used to quantify eating behavior. The DEBQ is a 33-item questionnaire with Likert-type scoring in each item (response options ranging from 1−5, from "Never" to "Very often"). It is divided in three dimensions measuring different behavioral traits: Emotional eating, External eating and Restrained eating (34)(35)(36). The DEBQ subscales have good internal consistency, dimensional validity and test-retest reliability (4,7,37,39).

Image processing and modeling
PET images were pre-processed similarly using automated processing pipeline, Magia (40).
[ 11 C]carfentanil data preprocessing has been described previously (24). MOR availability was expressed as [ 11 C]carfentanil binding potential (BP ND ), which is the ratio of specifically bound radioligand to that of nondisplaceable radioligand in tissue (41). Occipital cortex served as the reference region (42). CB 1 R availability was expressed as [ 18 F]FMPEP-d 2 volume of distribution (V T ), which was quantified using graphical analysis by Logan (43). The frames starting at 36 minutes and later since injection were used in the model fitting, since Logan plots became linear after 36 minutes (43). Plasma activities were corrected for plasma metabolites as described previously (44).

Statistical analysis
The primary study question was whether the DEBQ subscales (Emotional eating, External eating,   Figure 1.

Associations between µ-opioid receptor availability and eating behavior
Higher External eating score was associated with lower [ 11 C]carfentanil BP ND bilaterally in multiple brain areas (Figure 2). Cortical associations were found in cingulate and frontotemporal areas, while subcortical effects were prominent in nucleus accumbens, caudate, putamen, insula, hippocampus and amygdala. Associations with Total DEBQ or other subscale scores were not statistically significant. Results were essentially similar when controlling for sex, smoking and BMI. We also checked the association of External eating and BP ND in subsamples of 70 males and 22 females. In males, the association was similar to that of the full sample (Supplementary Figure 2), while in the female subsample there were no significant associations, likely due to limited statistical power.

Associations between CB 1 -receptor availability and eating behavior
Total DEBQ score was associated with lower [ 18 F]FMPEP-d 2 V T in multiple brain regions including bilateral anterior cingulate cortex, nucleus accumbens, caudate, putamen, insula, hippocampus, amygdala and thalamus (Figure 3). Full-volume associations with distinct DEBQ subscales and V T were not statistically significant. Controlling for BMI did not alter the results. associated with higher Total DEBQ score, age as a nuisance covariate. Shown are clusters where p < 0.05, FWE corrected at cluster level.

Regional analysis of central receptor availability and eating behavior
Results of the ROI-analysis were consistent with the full volume results for both tracers (Figure 4).
For [ 18 F]FMPEP-d 2 , the association directions of all DEBQ subscales were negative, but the 95% confidence intervals overlapped with zero. Visualization of the regional relationships in representative ROIs is presented in Figure 5.

Discussion
Our main finding was that higher DEBQ scores were associated with lower central availability of µopioid and CB 1 -receptors in healthy, non-obese humans. MOR and CB 1 R systems however showed distinct patterns of associations with specific dimensions of self-reported eating: While CB 1 Rs were associated in general negatively with different DEBQ subscale scores (and most saliently with the Total DEBQ score), MORs were specifically and negatively associated with externally driven eating only. Our results add support to the view that endogenous opioid and endocannabinoid systems underlie interindividual variation in feeding, with distinct effects on eating behavior measured with DEBQ.

Central µ-opioid receptors and external eating behavior
External eating -the tendency to feed when encountering palatable food cues such as advertisements -was associated with lowered MOR availability in cortico-limbic and striatal regions, which are major brain areas processing environmental food cues and mediating reward (49). A bulk of studies have shown that these regions are activated by mere perception of food cues or anticipation of feeding (50)(51)(52), and our recent work shows that lowered MOR availability is associated with amplified hemodynamic responses to food images in the same regions (14). Higher score on external eating is associated with increased food craving (53) and cue-induced palatable food intake (35,54), and may also contribute to short-term weight gain (38). Altogether these results suggest that central MOR system has an important role in modulating particularly this kind of impulsive feeding that may lead to overweight.
Previous PET studies have established that feeding triggers endogenous opioid release in humans (21,55). Binge eating disorder (BED) is accompanied with downregulated central MORs and high External and Emotional eating scores (56). Morbid obesity is also associated with lowered central MOR availability (21,22), possibly reflecting receptor downregulation due to repeated overstimulation following feeding. In minipigs, already 12 days of high sucrose intake and following central endogenous neurotransmitter release downregulates MORs in cingulate and prefrontal cortices, nucleus accumbens and elsewhere in striatum (57). The present findings extend the role of MORs in obesity and eating disorders to different feeding patterns in healthy subjects.
Healthy humans vary considerably in central MOR availability (24), and it is also possible that lowered MOR availability constitutes a genetically determined (58) risk factor for externally driven eating behavior. In healthy humans, trait impulsivity is associated with central MOR availability (25). Increased cue-reactivity is prevalent feature of behavioral addictions (59), and patients with BED and pathological gambling have reduced availability of central MORs as measured with in vivo PET (60). It is thus possible that subjects with lower MOR availability are susceptible for increased external eating in modern environment where they are consistently bombarded with feeding cues in advertisements and food shelves in supermarkets (11). However, the present data are purely cross-sectional and longitudinal human studies are needed to further disentangle the causes and the effects between the decrease of MORs in relation to external eating.

Central CB 1 -receptors and eating behavior
Higher Total DEBQ score associated with lower availability of central CB 1 Rs, and ROI-level modeling suggested a consistent negative association with all DEBQ subscales. Compared with the [ 11 C]carfentanil model, wider posterior distributions reflect the uncertainty arising from smaller [ 18 F]FMPEP-d 2 sample size. Brain's endocannabinoid system is a major homeostatic signaling system, with CB 1 Rs abundant in all known food intake regulating central regions (61). In previous brain PET studies, lowered CB 1 R availability has been associated with increased BMI (62), while globally upregulated CB 1 Rs have been found in anorexia nervosa (63). These opposite phenotypes on body adiposity spectrum could potentially result from corresponding alterations from CB 1 Rregulated food intake behaviors. Indeed, stimulation of CB 1 Rs by pharmacological agonists or endocannabinoids promotes food intake and amplifies the rewarding properties of feeding (64). In contrast, antagonism of the CB 1 Rs by rimonabant (withdrawn anti-obesity drug, Acomplia) effectively reduces food intake and increases energy expenditure, but in many patients with the cost of psychiatric symptoms (e.g. depressive mood, suicidality, anxiety) (61). Accordingly, the endocannabinoid system function has been connected to several other essential behavioral processes in addition to feeding (e.g. stress-coping, emotion regulation, pain perception) (65,66). Being this diverse and complex regulatory system, it may not be possible to pinpoint single distinct aspect of food intake behavior mediated by CB 1 Rs. Rather, our results add support to central CB 1 Rs role in regulation of multiple eating behavior traits, with implications on both homeostatic and hedonic feeding (67).

Limitations and methodological considerations
The [ 11 C]carfentanil data were pooled from three PET scanners, which may produce minor variance in outcome measures (46). However, this was accounted for in the analyses by adding the PET scanner as a nuisance covariate to all full-volume and regional analyses. The sample studied with [ 11 C]carfentanil consisted predominantly of males, and our statistical power was compromised for detecting potential sex differences. Also all subjects of the [ 18 F]FMPEP-d 2 subsample were males, and thus conclusions regarding CB 1 Rs might not be generalizable to females. Eating behavior was assessed by self-reports, rather than by direct observations. DEBQ has however been found to successfully identify clinically relevant eating styles (4,5). Our study had a cross-sectional design, and although we found evidence of eating behavior's association with MOR and CB 1 R systems, whether these receptor systems' alterations directly promote future weight gain is to be examined in longitudinal studies. Finally, in a single PET scan it is not possible to determine the exact proportions for causal factors to the altered receptor availability, which could potentially be affected by changes in receptor density, affinity or endogenous ligand binding.

Conclusions
Low cerebral MOR availability is associated with increased externally triggered eating behavior.
Modern obesogenic environment may promote food consumption via engaging the opioidergic link of the reward circuit whose tone is linked with this kind of impulsive eating. Central CB 1 Rs are in turn associated with multiple eating behavioral traits measured with DEBQ, consistent with endocannabinoid system's role as a major homeostatic regulatory system at the intersection of feeding, emotion and behavior.
TaK: Corresponding and first author, study design, study coordination, data acquisition, data modeling, statistical analysis, interpretation of the results, tables and figures, main writer of the manuscript. ToK: Study design, data modeling, statistical analysis, interpretation of the results,

Supplementary Material
Supplementary Material accompanies this paper.