UC San Diego

Rationale: Eating disorders are highly debilitating, and sometimes fatal, psychiatric conditions linked to alterations in reinforcement learning that may maintain illness and interfere with treatment. Computational psychiatry frameworks offer a novel approach for eliciting latent learning and decision-making mechanisms implicated in psychopathology and show promise for optimizing treatments. This dissertation applied computational methodology (e.g., theory-driven mathematical modeling) and neuroimaging to investigate reinforcement learning across eating disorder diagnoses, contexts, and valences to characterize their role in eating disorder phenomenology. Study 1 showed that adolescents with anorexia nervosa restricting type (AN-R; n = 36) displayed augmented habit-based reward learning compared to age-matched adolescent controls (n = 28), with higher goal-directed reward learning associated with greater orbitofrontal cortex-to-nucleus accumbens functional connectivity. This supports theories that suggest dietary restriction in AN-R may be habit-based. Study 2 investigated valence-dependent learning asymmetry, or learning bias, in young adults with AN-R, AN binge-eating/purging type (AN-BP) and bulimia nervosa (BN; n = 54, mean age = 22.4) compared to undergraduates (n = 40, mean age = 20.1), using a probabilistic gambling card game. Participants with a binge-eating/purging phenotype (AN-BP, BN) showed slower learning rates and reduced learning bias compared to both participants with a restricting phenotype (AN-R) and to undergraduates. Learning bias was negatively correlated with binge eating severity across phenotypes, further suggesting that attenuated positive learning bias is associated with binge eating. Study 3 examined aversive interoceptive reversal learning during breathing resistance in adults with AN-R, AN-BP and BN (n = 67) compared to age-matched controls (n = 25). Participants with AN-R showed faster learning rates to seen and unseen cues, compared to participants with a binge-eating/purging phenotype (AN-BP, BN). Higher learning rates in this group were associated with lower insula-to-anterior prefrontal cortex functional connectivity, with the control group showing the opposite association. Results support theories suggesting faster aversive interoceptive learning contributes to restrictive eating. Relevance: Collectively, these transdiagnostic studies demonstrate alterations in latent learning mechanisms across eating disorder phenotypes, developmental stages, valences, and contexts that may critically inform not only conceptualizations of eating disorder pathology, but also efforts to develop idiographic treatments and precision psychiatry practices.