Multi-modal deep phenotyping reveals biologically coherent subgroups in treatment-resistant schizophrenia

Background:

Schizophrenia (SCZ) is a highly heterogeneous psychiatric disorder characterized by substantial variability in clinical presentation, cognitive impairment, treatment response, and underlying biological mechanisms. Treatment-resistant schizophrenia (TRS), typically defined by insufficient response to standard antipsychotic therapies, represents a major clinical and therapeutic challenge, affecting a substantial proportion of patients and contributing disproportionately to disease burden. However, the biological definition of TRS remains poorly characterized due to inconsistent diagnostic frameworks and divergent classification strategies across studies (Rakitzi & Georgila, 2024). Recent advances in multi-modal data integration and genetically informed stratification approaches offer new opportunities to identify biologically meaningful subgroups that transcend conventional diagnostic categories. This study aimed to evaluate the biological coherence and clinical validity of multiple TRS classification strategies through deep phenotyping integrating genetic, neurocognitive, and clinical data.

Methods:

We applied the CASTom-iGEx pipeline (Trastulla et al., 2024) to genotype data from 188 schizophrenia spectrum disorder (SSD) patients and 174 controls to derive genetically regulated gene- and pathway-level expression profiles across three brain-relevant tissues: dorsolateral prefrontal cortex (DLPC), hippocampus, and cerebellum. All patients were projected onto four predefined biologically informed subgroups previously identified in 24,764 patients from 36 European Psychiatric Genomics Consortium cohorts with DLPC model. We systematically compared three TRS classification strategies: (1) cross-sectional TRS definitions based on current treatment resistance status, (2) longitudinal TRS using the CDP-STAR model to identify core persistent TRS, and (3) a combined strategy prioritizing STAR-defined classifications while incorporating broader cohort definitions. Associations between subgroup membership and key clinical variables, including demographics, cognitive performance, symptom severity, and clozapine exposure, were assessed using regression models adjusted for age, sex, and ancestry principal components. Polygenic risk scores (PRS), gene concordance analyses, and pathway-level profiling were used to evaluate biological differentiation and cross-model consistency. Finally all the analyses were performed in external validation cohort to replicate key findings.

Results:

Projection analyses identified four clinically distinct subgroups. Cluster 3 showed significant enrichment of confidently classified STAR core TRS patients (p < 0.05), whereas cross-sectional and combined TRS definitions demonstrated weaker subgroup specificity. Verbal fluency performance and clozapine treatment history significantly differentiated subgroup membership after covariate adjustment (p < 0.05), supporting clinical validity. Gene concordance analyses across all three brain models revealed distinct biological architectures between TRS and broader SSD classifications, suggesting treatment resistance may involve mechanisms partially independent from general schizophrenia severity. Preliminary pathway-level analyses suggested potential biological differentiation between TRS and non-TRS profiles, although specific pathway enrichment patterns remain under ongoing investigation. Notably, biologically relevant signals were detectable even using cross-sectional TRS definitions, although longitudinal classifications demonstrated greater subgroup specificity.

Conclusions:

Multi-modal deep phenotyping identified biologically coherent schizophrenia subgroups with clinically relevant distinctions in treatment resistance. TRS appears to involve partially distinct molecular mechanisms beyond general schizophrenia severity, although precise pathway-level characterization requires further validation. While longitudinal TRS definitions may offer stronger biological specificity, cross-sectional classifications may still provide utility in large-scale genetic studies when integrated within comprehensive deep phenotyping frameworks. These findings support data-driven refinement of TRS definitions and may contribute to future biomarker development and personalized therapeutic strategies.

References

Rakitzi, S., & Georgila, P. (2024). Treatment-Resistant Schizophrenia: Evidence-based Pharmacotherapy, Cognitive Behavioral Psychotherapy and Rehabilitation. Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-64623-2 Trastulla, L., Dolgalev, G., Moser, S., Jiménez-Barrón, L. T., Andlauer, T. F. M., von Scheidt, M., Budde, M., Heilbronner, U., Papiol, S., Teumer, A., Homuth, G., Völzke, H., Dörr, M., Falkai, P., Schulze, T. G., Gagneur, J., Iorio, F., Müller-Myhsok, B., Schunkert, H., & Ziller, M. J. (2024). Distinct genetic liability profiles define clinically relevant patient strata across common diseases. Nature Communications, 15(1), Article 1. https://doi.org/10.1038/s41467-024-49338-2