Neural markers of abstinence in alcohol dependence: Insights from Reward Learning

Abstract

Maladaptive reward learning and decision-making circuity are key factors in the onset and progression of alcohol use disorder and have therefore emerged as key targets for neuropsychological and pharmacological interventions. Probabilistic reversal learning studies have consistently reported impaired learning in recently detoxified alcohol dependent (AD) participants. However, the neural and behavioural changes associated with reward learning which occur throughout abstinence remain unexplored. Here, we show that AD participants, with mean abstinence of 20 months, exhibit intact behavioural performance within an electroencephalography (EEG) probabilistic reversal learning task. Reinforcement learning modelling reveals reward and punishment related learning rates and exploration rates are comparable between AD and healthy control (HC) participants, suggesting recovery of even the nuanced aspects of learning in longer term abstinence. However, EEG analysis indicates that AD, compared to HC participants, show globally elevated event-related potential (ERP) feedback related negativity (FRN) following reward valuation. Furthermore, Feedback-P3 valence prediction error signal is negatively associated with abstinence duration indicating a potential state marker of AD recovery. We then employ unsupervised machine learning (canonical polyadic tensor decomposition) to identify spatiotemporal EEG patterns of reward valuation in a purely data-driven manner. Classification analysis shows these tensor components can predict group membership with 80.4% accuracy. By probing group differences in tensor components, we discover early hyperfunctioning in centro-frontal regions linked to alcohol dependence and associated with early abstinence. The clinically meaningful EEG biomarkers presented here could guide the development of more targeted treatments and support big data approaches to objective patient monitoring.