Corlett and the group from Cambridge seek to gain insights into mechanisms related to the formation of delusions, one of the core positive symptoms of schizophrenia, using fMRI and an associative learning task, and as a pharmacological model of psychosis, the ketamine challenge in normal volunteers. They report a number of significant and interesting correlations that they interpret within the framework of the reinforcement learning model of associative learning. Right prefrontal activity associated with unexpected negative feedback during the placebo phase (a prediction error signal) correlates with two measures of “early” delusional thinking during higher-dose ketamine challenge. Measures of sensory disturbances during high-dose ketamine correlate with measures of early delusional thinking. And low-dose ketamine-related changes in the response to feedback are correlated with early delusional thinking during high-dose ketamine. Ketamine also appears to have an effect on the acquisition of associations within this paradigm. The authors conclude that their result implicates altered frontal function and disruptions in error-dependent learning in associative processes that lead to referential thinking and delusion formation. They also implicate a role for NMDA neurotransmission in this deficit.

The paper is very important since it tests a new theoretical model of delusions based upon an explicit model of error-dependent associative learning. Having such an explicit model provides the authors not only with increased insight into mechanisms, but also the ability to design cognitive activation experiments that can test for the presence of predicted deficits as well as their underlying neural circuitry. For this reason, the work is likely to have a very significant impact on our thinking about a previously elusive topic: how delusions form in the brains of people with schizophrenia.

A limitation of the paper is that the findings are correlational and the authors do not precisely specify, in computational terms, the nature of the abnormality in feedback-related learning that leads to delusions. However, they do localize the deficit to the prefrontal cortex and indicate their belief that the primary deficit is associated with the processing of prediction errors that, in turn, depend upon glutamatergic neurotransmission. There are other alternative interpretations of their results which they acknowledge in their discussion, such as the possibility that greater activation in the prefrontal cortex to error feedback is a marker of poorer prefrontal efficiency and that this is associated with psychosis proneness. A second alternative account which they discuss is the possibility that the quality of sensory processing in individual subjects drives the correlations. A third, non-NMDA account of their data involves altered dopaminergic neurotransmission, given that the work of Shultz and Montague have strongly implicated a role for dopaminergic neural activity in prediction error signaling during reinforcement-based associative learning.

One way to test the general PFC inefficiency theory would be to see if activity during other non-associative learning tasks predicts psychosis proneness under ketamine challenge, or that the error-dependent PFC signal predicts non-NMDA-related psychosis-like states such as those induced by cannabis or amphetamine. Given the explicit nature of the reinforcement learning model, there will likely be a number of task manipulations that would manipulate specific parameters in computational models of the task that could be tested using fMRI. One expects that this paper will generate many additional studies that will test the authors’ model and various alternative views regarding the significance of their results for understanding the cognitive and neural mechanisms underlying delusion formation in schizophrenia.

View all comments by Cameron Carter