Mapping symptom dimensions in early psychosis through neurotransmitter-informed co-localization of structural and functional brain alterations

Traditional diagnostic tools like the Positive and Negative Syndrome Scale for Schizophrenia (PANSS) and the Scale for the Assessment of Negative Symptoms (SANS) assess broad symptom domains but often miss the nuanced patterns seen in early psychosis. This limited dimensional resolution poses a challenge for developing targeted interventions and advancing precision psychiatry. In this study, we applied a dimensional approach to patients with recent-onset psychosis (ROP) to investigate the neurochemical basis of symptom dimensions. Using machine learning, we examined how the co-localization of structural and functional brain alterations with neurotransmitter maps from healthy individuals distinguishes patients with dominant symptom profiles from healthy controls, aiming to improve diagnostic precision and understanding of underlying disease mechanisms.

The sample comprised 454 ROP patients from the Munich (N = 139) and PRONIA (N = 315) cohorts (mean age = 27.1 ± 7.5 years; 168 females), as well as 522 healthy controls (HC) from PRONIA (mean age = 25.6 ± 6.0 years; 306 females). To identify symptom dimensions, Orthogonal Projective Non-Negative Matrix Factorization (OPNMF) was applied to all PANSS and SANS items, generating three-, four-, and five-factor solutions.

A linear support vector machine was used to classify ROP patients with high loadings on each symptom factor versus HC, employing a nested cross-validation framework (inner CV: 10×10; outer CV: 10×10). Classification relied on cortical and subcortical correlations of grey matter volume (GMV), from T1-weighted magnetic resonance imaging, and fractional amplitude of low-frequency fluctuations (fALFF) in the slow-3, slow-4, and slow-5 bands, from resting-state functional MRI, with the distribution of 25 normative neurotransmitter maps using JuSpace [1] methodology. Imaging data were corrected for age, sex, and site: GMV was adjusted via dynamic standardization to construct a normative reference sample and fALFF via partial correlations and offset correction.

The factor solution with the highest explained variance (89.3%) included four symptom dimensions: avolition-asociality, expressive deficits, cognitive disorganization, and positive symptoms. Correlations with the fALFF slow-3 frequency band yielded the best performance for avolition-asociality (balanced accuracy [BAC]: 58.1%, AUC: 0.55, sensitivity: 63.8%, specificity: 52.3%) and cognitive disorganization (BAC: 73.7%, AUC: 0.76, sensitivity: 71.4%, specificity: 75.9%). For expressive deficits and positive symptoms, the fALFF slow-4 band provided the highest accuracy (expressive deficits: BAC: 65.0%, AUC: 0.70, sensitivity: 63.0%, specificity: 67.1%; positive symptoms: BAC: 63.3%, AUC: 0.66, sensitivity: 67.8%, specificity: 58.9%).

These findings suggest that symptom dimensions in recent-onset psychosis are differentially associated with specific functional patterns co-localizing with normative neurotransmitter distributions. The integration of dimensional symptom modelling with neurochemical-informed neuroimaging offers a promising path toward more biologically grounded classification approaches.

[Table presented]