Reinforcement learning deficits in at-risk mental state for psychosis: the role of primary negative symptoms

Abstract

Negative symptoms such as amotivation begin to emerge in the at-risk mental state for psychosis (ARMS), and in some cases, these symptoms are idiopathic (i.e., primary negative symptoms, PNS). While deficits in reinforcement learning (RL) are evident in patients with first-episode and chronic schizophrenia and are associated with the severity of negative symptoms, whether these deficits are present during ARMS warrants further investigation. Additionally, it is unclear whether these deficits, especially those driven by rewards, are particularly characteristic of ARMS individuals with PNS. This study compared 97 ARMS individuals and 37 healthy controls (HC) on probabilistic RL. We then compared ARMS individuals with PNS (n = 15), those without PNS (n = 82), and HC on probabilistic RL. Reward- and punishment-driven RL (both trial-by-trial and procedural) and value-guided choice were assessed using two probabilistic RL tasks: Go/NoGo (GNG) and Gain vs. Loss-Avoidance (GLA) tasks. Compared with HC, ARMS individuals showed impairments in overall procedural RL on both RL tasks, as well as subtle, task-specific alterations in value-guided choice. Furthermore, ARMS-PNS individuals, relative to HC, showed particular deficits in reward-driven RL (trial-to-trial RL on GLA task and procedural RL on both tasks), as well as impaired positive expected value representations to guide decision-making on GLA task. Extending the results in patients with established psychosis, our findings indicate that deficits in RL exist before the onset of a psychotic disorder. With a modest sample of ARMS-PNS individuals, preliminary evidence suggests that deficits in reward-driven RL may be more specific to PNS.