Dense sampling of choices links high learning rates to obesity and low reward sensitivity to binge eating

Mounting evidence shows that obesity is associated with alterations in dopamine transmission. However, in humans, corresponding changes in dopamine-dependent behavior, such as reward learning with increasing BMI, have not been conclusively established and dissociated from pathological eating behavior. Here, we provide a principled assessment of differences in reinforcement learning (RL) related to obesity or pathological eating, such as binge eating (BE), using a large sample of 370 participants enriched for high BMI and BE. To dissociate trait and state components of RL, participants completed up to 31 runs of a novel RL game across weeks (1,486,950 observed choices). We used hierarchical Bayesian models to estimate the mean and variability of parameters for each individual. Across runs, higher BMI was predicted by higher and more variable learning rates using elastic net models ({Delta}R2=.19; pperm<.0001) although performance did not differ (r=-0.04, pboot=.46). In contrast, BE and the dimension loss of control were characterized by lower and more variable reward sensitivity ({Delta}R2=.19, controlled for BMI; pperm<.0001) leading to fewer points (r=.10, pboot=.030). Crucially, patients with BE disorder also showed attenuated reward sensitivity in the nucleus accumbens during anticipation in an effort task that required no learning (p=.006). We conclude that obesity-related differences in reward learning are best explained by changes in learning, whereas BE and loss of control were mostly driven by reduced reward sensitivity. Notably, repeated assessments revealed that increased variability across states contribute to both obesity and BE. Our findings highlight the necessity to complement neurobiological research with behavioral precision mapping to derive mechanistic insight into multidimensional disorders such as obesity and BE disorder.