PROJECT SUMMARY Reward-guided decision-making and impulse control are disrupted after chronic cocaine use. These changes have been attributed to altered functions in brain circuits critical for computation of reward predictions, action policies, prediction errors and attention. ‘Reward prediction’ signals reflect the reward the animal expects to receive as a result of behavior or presentation of a stimulus. ‘Action policies’ are rules that govern behavior that are triggered by external stimuli or context, and are thought to underlie habits. Both reward predictions and action policies are modified when there are violations in predictions known as ‘reward prediction errors’. ‘Signed’ reward prediction errors reflect the valence associated with an error, strengthening or weakening the associability between cues, responses and outcomes. ‘Unsigned’ prediction errors reflect the surprise induced by errors which lead to increases in ‘attention’ so that learning can occur. We have uncovered neural correlates of these constructs and the relationship between them by recording from multiple brain areas as rats perform a reward- guided decision-making task in which we unexpectedly varied the delay to and size of reward across several trial blocks. We have shown that nucleus accumbens core (NAc) and anterior insula (AI) encode reward predictions, firing strongly for cues that predict more valued reward, whereas firing in dorsal lateral striatum (DLS) is highly associative, encoding action policies such as stimulus-response associations and contextual bias signals. We have also shown that midbrain dopamine (DA) neurons increase firing to unexpected reward and decrease firing to unexpected reward omission. During learning these signed prediction errors transfer to cues, with cues predicting more valued reward eliciting stronger firing. Unlike firing of DA neurons in ventral tegmental area (VTA), our work has shown that firing in basolateral amygdala (ABL) and anterior cingulate cortex (ACC) better reflects integrated unsigned prediction error signals and attention, increasing during unexpected up- and down-shifts in value at the time of the error and during cue sampling on subsequent trials. Here, we propose that activation of ACC increases attention and engages model-based mechanisms that govern goal-directed behavior and associated mechanisms in AI, NAc and VTA, and that chronic cocaine reduces the fidelity of reward prediction and prediction error signals, in part, by disrupting ACC function. Further, we propose that epigenetic changes in ACC, NAc and DLS after chronic drug use alter cue selectivity in a way that promotes fast, automatic behavior by altering the start point and rate of signals that drive actions. In this resubmission of my renewal application we propose to further model the normal circuit and behavior, and its disruption after cocaine-exposure by bi-directionally manipulating neural signals via optogenetics and epigenetic (histone deacetyl...