Neural Evidence for Disrupted Predictive Coding in First Episode Psychosis

1631 

Abstract Submission 

Angela Wang¹, Eric Rawls², Collin Teich², Angus MacDonald¹, Scott Sponheim³

¹Department of Psychology, University of Minnesota, Minneapolis, MN, ²Department of Psychiatry and Behavioral Sciences, University of Minnesota, Minneapolis, MN, ³Minneapolis Veteran Affairs HealthCare System, Minneapolis, MN

Angela Wang  
Department of Psychology, University of Minnesota
Minneapolis, MN

Eric Rawls, PhD  
Department of Psychiatry and Behavioral Sciences, University of Minnesota
Minneapolis, MN

Collin Teich  
Department of Psychiatry and Behavioral Sciences, University of Minnesota
Minneapolis, MN

Angus MacDonald, PhD  
Department of Psychology, University of Minnesota
Minneapolis, MN

Scott Sponheim, PhD  
Minneapolis Veteran Affairs HealthCare System
Minneapolis, MN

Predictive coding is a theoretical framework that suggests our brain constantly attempts to predict sensory inputs based on past experiences to guide our learning and behavior. Prediction error (PE), the mismatch between our priors (predictions) and our actual sensory input, allows the brain to update these sensory predictions for the future. In patients with psychosis, maladaptive inferences and behavior are theorized to result from atypical updating of priors following PEs. However, few studies have tested this proposition directly. Using a three armed-bandit task with EEG, we sought to understand the impact of neural PEs on subsequent sensory updating in patients with first episode psychosis (PwP) and healthy controls (HC). Our goal was to directly test for links between neural PEs (reward positivity [RewP], P300) and subsequent sensory updating (P1, N1). This link is hypothesized to be present in HC, but not in PwP, implying that predictive coding partially underlies reinforcement learning deficits in PwP.

We collected 32-channel EEG simultaneously with 3T fMRI from 36 HC and 34 PwP. We employed a 3-armed bandit task, which instructed participants to select one out of three options, resulting in a win of +1 points or an omission of +0 points. We separated feedback trials, and the immediately following stimulus trials, according to Outcome (win/omission) and PE (low/high). Thus, event-related potentials for the following feedback conditions were extracted: Win-HighPE, Win-LowPE, Lose-HighPE, and Lose-LowPE. A mixed model approach was used to calculate main effects of Group (HC/PwP), Outcome (win/omission), and PE (low/high), as well as interactions between Group and the within-subject factors of Outcome and PE. Pearson correlations were calculated between feedback-locked potentials (RewP, P300) and the immediately subsequent stimulus-locked potentials (P1, N1), separately for HC and PwP. Furthermore, correlations between stay/switch choices and ERP components were also analyzed.

Within our analysis of feedback-locked potentials, the RewP displayed significantly greater activity to wins versus losses (t = 3.14, p < 0.01) and to highPE versus lowPE (t = 4.53, p < 0.001), as well as an interaction indicating that the greatest RewP amplitudes were for Win-highPE trials. P300 showed significantly greater activity in highPE versus lowPE (t = 5.37, p < 0.001). We found no group differences between HC and PwP in the RewP or P300. We then directly tested for links between neural PEs and subsequent sensory updating. This analysis revealed significant correlations in HC between RewP/P300 amplitudes with the subsequent visual P1, in Lose-HighPE condition (rs > 0.51, ps < 0.001). This correlation was absent in PwP. Moreover, in HC, switch behavior was correlated with RewP and P300 in the Lose-LowPE condition (rs > 0.35, ps < 0.03). However, in PwP, switch behavior was correlated with P1 in the Lose-LowPE condition (r = -0.52, p < 0.01), and N1 in the Win-HighPE and Win-LowPE conditions (rs > 0.39, ps < 0.02).

Our findings provide evidence that both HC and PwP experience PE, indicating that PE signaling itself is not disrupted in PwP. However, in HC, reward processing (RewP, P300) modulates subsequent sensory inputs (P1). This modulation is absent in PwP, suggesting a dysfunctional link between PE signaling and sensory updating. In addition, within HC, neural reinforcement processing predicts reinforcement learning behaviors, while for PwP, these behaviors are instead predicted by sensory components. Overall, our results support the hypothesis that PEs should drive sensory updating, and that dysfunctions in this predictive coding impact maladaptive learning and behavior in psychosis. Future research will incorporate simultaneously recorded fMRI data to directly localize brain regions underlying predictive coding deficits, which could be later targeted for brain-based interventions such as neuromodulation.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ²

Decision Making

Learning and Memory Other

EEG/MEG Modeling and Analysis ¹

Attention: Visual

Cognition

Computational Neuroscience

Electroencephaolography (EEG)

Learning

Psychiatric

Schizophrenia

Vision

Other - Predictive Coding, Reward Processing; Psychosis; Sensory Updating