Computational modelling and neural correlates of reinforcement learning following three-week escitalopram: a double-blind, placebo-controlled semi-randomised study

Reinforcement learning is a fundamental aspect of adaptive behaviour, since it involves the acquisition and updating of associations between actions and their outcomes based on the rewarding or punishing consequences. Acute experimental manipulations of serotonin have provided compelling evidence for its role in reinforcement learning. However, it remains unknown how more chronic manipulation of serotonin, which holds greater clinical relevance, affects reinforcement learning and the underlying neural mechanisms. Consequently, we aimed to investigate the effect of a three-week administration of the SSRI, escitalopram, on a reinforcement learning paradigm during functional magnetic resonance imaging. The study used a double-blind, placebo-controlled design with 64 healthy volunteers. Participants were semi-randomised, ensuring matched groups for age, sex and intelligence quotient (IQ), to receive either 20 mg of escitalopram (n = 32) or placebo (n = 32) for at least 21 days. We analysed group differences in reinforcement learning using both analysis of covariance as well as innovative hierarchical Bayesian modelling of the reinforcement learning task. Escitalopram reduced learning from punishment during punishment trials. A key novel finding was that there was decreased activation of the intraparietal sulcus in the escitalopram group when compared to the placebo group during reward trials. The involvement of the intraparietal sulcus suggests that escitalopram affects the encoding of value outcome, which may lead to reduced reinforcement sensitivity, and thereby impacting adaptive learning from feedback. Understanding these mechanisms may help to optimize SSRI treatment to mitigate clinical symptoms and improve quality of life for neuropsychiatric patients, by elucidating serotonin's effects on affect, cognition, and behaviour.