In search of relief: understanding the underlying learning mechanisms of stress-induced overeating

Emotional eating, defined as the consumption of high-caloric, ultra-processed foods in response to negative affective states including stress, is a widespread behavioral pattern. While such behavior may have a short-term adaptive function (i.e., temporarily alleviating distress), it becomes maladaptive when chronic. Excessive emotional or stress overeating is indeed a well-established risk factor for the development of obesity, binge eating, and related metabolic conditions (1). Two dominant theoretical frameworks attempt to explain this phenomenon. The Emotional Eating Theory (EET)(2) is primarily applied to subclinical populations and proposes that stress-induced eating serves as an affect regulation strategy. Specifically, eating is thought to alleviate negative emotional states, thereby reinforcing future eating episodes through negative reinforcement. In contrast, the Incentive Sensitization Theory (IST)(3), frequently applied to clinical overeating, attributes excessive consumption to dopaminergic dysregulation, whereby repeated exposure to palatable food enhances anticipatory 'wanting' responses, dissociated from the actual pleasure (‘liking’) experienced during consumption. While both frameworks have generated valuable insights, each suffers from specific limitations that constrain their explanatory power. EET, for example, assumes that eating alleviates stress, but the emotional mechanism underlying this presumed relief remains theoretically underdefined and empirically unmeasured. Typically, studies infer the presence of relief from a reduction in negative affect following food consumption. It conflates the decline in stress with emotional recovery, missing the momentary, rewarding nature of relief that may arise during eating under stress, rather than after stress has passed. Thereby, this operationalization fails to isolate relief as a distinct, temporally bound, and positively valenced emotional state. Importantly, recent research in fear extinction and threat avoidance has begun to redefine relief as a measurable, intrinsic form of positive emotion with its own neurobiological profile. Relief activates key reward-related brain regions (e.g., ventral striatum, orbitofrontal cortex) (4,5,6) and may act as a powerful reinforcer (7). Within the emotional eating context, we propose that relief, when experienced during food consumption under stress, may act as a rewarding event. If sufficiently pleasant, this relief may reinforce the behavior through mechanisms parallel to food-reward, thereby contributing to future overeating. Conversely, its absence may explain why some individuals suppress eating under stress: if no rewarding relief occurs, the behavior is not reinforced. In parallel, IST has generated a robust body of neuroimaging research linking compulsive eating to increased activity in dopaminergic ‘wanting’ systems, primarily through passive paradigms involving cue exposure. However, these designs often fail to capture the active, real-time processes of food selection and consumption. Moreover, IST-based studies have repeatedly demonstrated high inter-individual variability in how the brain encodes ‘wanting’ and ‘liking’, but the origins of this variability remain elusive (8). To address these gaps, we propose a conceptual bridge between EET and IST via the construct of relief, a transitory, positively valenced emotional state that emerges during stress-alleviating behaviors such as eating (9). Based on Papalini's (2024) integrative model (9), we hypothesize that this consumption-evoked relief constitutes a rewarding internal state that strengthens future eating responses. By directly measuring relief during eating in a stressful context, we aim to identify whether it functions as a reinforcing mechanism that links emotional regulation (as described by EET) to the neural reward sensitization processes (as proposed by IST). Furthermore, we posit that inter-individual differences in the pleasantness of this relief experience may account for the variability observed in both craving strength and compulsive eating behaviors. In this way, relief may act as a latent mediator, potentially explaining why some individuals engage in stress-induced overeating while others do not. To formalize this model, the present experiment include two sessions (within subject design), a stress plus relief session: S+R, and a stress-only session: S. At the start of the S+R session, participants select one snack from six available options (three sweet, three salty food items) to be consumed later. Next, participants complete the Partial Reinforcement Learning (PRL) phase. During this task, they learn to associate one of two button presses with the delivery of a chosen favorite snack reward under a 50% reinforcement rate across 70 trials. Each successful trial results in the earning of one unit of the selected favorite snack, which they can later consume following the instruction of the researcher. A two-minute resting condition follows. Participants then undergo the Montreal Imaging Stress Task (MIST), which is used to induce psychosocial stress (10). They solve timed arithmetic problems under the believe that they are competing against the average of a group of previous participants matched on age and education. The experimenter remains in the room during the task, and an evaluator is also present to enhance social-evaluative threat (both keep the interaction with the participant as limited as possible). The task lasts 25 minutes and written negative feedback is delivered on the screen regardless of actual performance. Following the MIST, participants enter a first feedback condition, during which they are asked to engage with their emotional response to the feedback (‘Your performance did fall short. You’ll need to perform the mathematical task again later’) for two minutes. This is followed by a second feedback condition, where participants are again asked to re-focus on their emotional experience. Once fully engaged, they are instructed to bring a single piece of their chosen snack to their mouth and begin eating it while continuing to engage with their emotional state. These conditions (first and second feedback) last two minutes each. Next, participants enter the devaluation phase, during which they consume the snack ad libitum until reaching satiation. This phase lasts between five and ten minutes. Following devaluation, the extinction phase begins. Here, participants perform the same button-pressing task used during the PRL phase, but no real-time feedback is provided. While participants are informed that their responses are still being recorded and will be rewarded at the end of the session, no snack is delivered during this phase. The session concludes with a termination condition. Participants are informed that the experiment has ended but are asked to remain in the room for approximately five minutes while the experimenters save some data. During this period, they are free to consume any remaining snack. The S session mirrors the structure of the S+R session but omits two key conditions. After the first feedback condition, participants proceed directly to the extinction task without receiving food or undergoing devaluation. No snack is available after termination of the task. Throughout both sessions, physiological data including pupil dilation, skin conductance level, and heart rate are recorded continuously. Salivary cortisol samples and blood pressure (systolic and diastolic) are collected at three time points: following the PRL phase, after the MIST, and after the first feedback condition. The snack is weighed at three time points: beginning of the task, after the termination of the task, and after the participant leaves the lab. Visual Analog Scale (VAS) ratings assessing relief, hunger levels, fullness, stress, anxiety, unpleasantness of the negative feedback, food wanting as well as food liking, are obtained at multiple time points over the sessions.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.