Neurobiological substrates of the positive formal thought disorder in schizophrenia revealed by seed connectome-based predictive modeling

Formal thought disorder (FTD) is a core symptom cluster of schizophrenia, but its neurobiological substrates remain poorly understood. Here we collected resting-state fMRI data from 276 subjects at seven sites and employed machine-learning to investigate the neurobiological correlates of FTD along positive and negative symptom dimensions in schizophrenia. Three a priori, meta-analytically defined FTD-related brain regions were used as seeds to generate whole-brain resting-state functional connectivity (rsFC) maps, which were then compared between schizophrenia patients and controls. A repeated cross-validation procedure was realized within the patient group to identify clusters whose rsFC patterns to the seeds were repeatedly observed as significantly associated with specific FTD dimensions. These repeatedly identified clusters (i.e., robust clusters) were functionally characterized and the rsFC patterns were used for predictive modeling to investigate predictive capacities for individual FTD dimensional-scores. Compared with controls, differential rsFC was found in patients in fronto-temporo-thalamic regions. Our cross-validation procedure revealed significant clusters only when assessing the seed-to-whole-brain rsFC patterns associated with positive-FTD. RsFC patterns of three fronto-temporal clusters, associated with higher-order cognitive processes (e.g., executive functions), specifically predicted individual positive-FTD scores (p = 0.005), but not other positive symptoms, and the PANSS general psychopathology subscale (p > 0.05). The prediction of positive-FTD was moreover generalized to an independent dataset (p = 0.013). Our study has identified neurobiological correlates of positive FTD in schizophrenia in a network associated with higher-order cognitive functions, suggesting a dysexecutive contribution to FTD in schizophrenia. We regard our findings as robust, as they allow a prediction of individual-level symptom severity.