Virtual Brain Modelling Reveals Altered Network Dynamics and Predicts Symptom Severity in Schizophrenia: Relevance to Antipsychotic Treatment Response and Resistance.

Background: Schizophrenia (SZ) is a heterogeneous disorder characterized by positive, negative, cognitive and disorganisation symptoms, with pleomorphic phenotype and poorly understood aetiology. Dopamine dysregulation represents the most replicated in vivo pathophysiology finding. However, accumulating evidence suggests altered large-scale brain connectivity and excitation/inhibition balance disruption. While quantifying these alterations would help clarify SZ neurophysiology and guide therapy, single-patient identification remains impractical. This is particularly relevant for treatment-resistant SZ (TRS) whose pathophysiology remains unclear. Methods: Using virtual brain modelling, we simulated large-scale brain dynamics from functional and structural MRI data in 43 healthy controls and 33 SZ patients, stratified into TRS and non-TRS. We used the Wong-Wang model to derive four network parameters reflecting GABAergic and NMDA receptor-mediated glutamatergic transmission: global coupling (G), local inhibitory (Ji) and excitatory (JNMDA) coupling, and recurrent excitation (w+). We performed simulations in the whole brain and in resting-state networks, compared model parameters between groups and tested their association with Positive and Negative Syndrome Scale (PANSS) scores. Results: Compared to non-TRS, TRS patients showed increased JNMDA and decreased w+ at the whole-brain level, increased JNMDA and G in the dorsal attention network, lower G in the limbic network and increased w+ in the visual network. Moreover, Ji in the default mode network predicted negative symptom severity while JNMDA in the limbic and ventral attention networks predicted total PANSS scores and disorganisation. Conclusions: These findings suggest that TRS is characterized by network-specific alterations in excitation/inhibition profiles and support the use of personalised multiscale modelling to identify symptom biomarkers.