AI-based neuroanatomical prediction of cytokine profiles in schizophrenia: linking peripheral inflammation to central receptor expression
N. Sirbu1, S.M. Ortega1, T. Pollak1, N. Koutsouleris1,2, P.A. Lalousis1
1Institute of Psychiatry, Psychology & Neuroscience (IoPPN), Department of Psychosis Studies, London, United Kingdom, 2Ludwig-Maximilian-University, Department of Psychiatry, Munich, Germany
Background: Growing evidence highlights immune system dysregulation in schizophrenia, with consistent reports of altered cytokine levels in peripheral blood and cerebrospinal fluid (CSF), suggesting that inflammatory processes may contribute to the observed heterogeneity in disease mechanisms. Structural neuroimaging studies have identified cortical thinning, ventricular enlargement, and subcortical abnormalities associated with immune activation and elevated cytokines.
Recent clustering approaches have revealed biologically distinct immune subgroups within schizophrenia, each showing unique brain-structure and cognitive profiles. However, the relationship between individual cytokines and specific neuroanatomical patterns remains poorly understood. Because cytokines exert diverse and context-dependent actions through their regional receptor distribution, mapping brain regions whose grey matter variability is associated with peripheral cytokine levels, and examining whether these regions are particularly responsive to cytokine signalling via local receptor expression, may clarify how immune processes modulate neural structure and function in schizophrenia.
Recent clustering approaches have revealed biologically distinct immune subgroups within schizophrenia, each showing unique brain-structure and cognitive profiles. However, the relationship between individual cytokines and specific neuroanatomical patterns remains poorly understood. Because cytokines exert diverse and context-dependent actions through their regional receptor distribution, mapping brain regions whose grey matter variability is associated with peripheral cytokine levels, and examining whether these regions are particularly responsive to cytokine signalling via local receptor expression, may clarify how immune processes modulate neural structure and function in schizophrenia.
Aims & Objectives: To determine whether structural MRI features can predict individual cytokine concentrations in schizophrenia and to evaluate whether the predictive neuroanatomical regions are spatially enriched for expression of their respective cytokine receptors in the human brain.
Method: Data were drawn from the Australian Schizophrenia Research Bank (185 patients, 101 controls; mean age = 39 ± 10 years). Grey matter (GMV), white matter (WMV), and CSF volumes were derived using CAT12/SPM12 segmentation of 1.5 T T1-weighted MRI (Neuromorphometrics atlas; 408 ROIs).
Support Vector Machine (SVM) classifiers were trained to discriminate high vs. low cytokine expression (CRP, IFN-γ, IL-1β, IL-2, IL-6, IL-8, IL-10, IL-12p70, TNF-α) using pooled cross-validation, adjusting for age, sex, site, and total intracranial volume. Feature importance was quantified using sign-based consistency (FDR p < 0.05).
Predictive brain regions were mapped to the Allen Human Brain Atlas to test for regional co-expression of the corresponding cytokine receptor genes.
Support Vector Machine (SVM) classifiers were trained to discriminate high vs. low cytokine expression (CRP, IFN-γ, IL-1β, IL-2, IL-6, IL-8, IL-10, IL-12p70, TNF-α) using pooled cross-validation, adjusting for age, sex, site, and total intracranial volume. Feature importance was quantified using sign-based consistency (FDR p < 0.05).
Predictive brain regions were mapped to the Allen Human Brain Atlas to test for regional co-expression of the corresponding cytokine receptor genes.
Results: SVM models achieved significant cytokine-specific prediction:
IL-1β (BAC = 61.1%, AUC = 0.60, p = 0.001), IFN-γ (58.4%, 0.60, p = 0.009), IL-6 (58.9%, 0.62, p = 0.010), and IL-12p70 (60.0%, 0.61, p = 0.005).
Predictive features spanned frontal, temporal, subcortical, and cerebellar regions.
Voxel-based morphometry revealed that IFN-γ, IL-6, and IL-12p70 models were dominated by thalamic, basal-ganglia, and cingulo-cerebellar structures, regions integral to immune-neural regulation.
Gene-expression mapping demonstrated significant enrichment of receptor-positive regions among predictive brain areas for IFN-γ (overlap = 20, p < 0.0001), IL-6 (21, p < 0.0001), and IL-12p70 (19, p < 0.0001), but not IL-1β (p = 0.60). Receptor transcripts showed the highest cortical expression for IL12RB2 (frontal pole > 5.0 log₂), IFNGR1/2 (parahippocampal and temporal > 7.8), and IL6ST (striatum > 10.5).
IL-1β (BAC = 61.1%, AUC = 0.60, p = 0.001), IFN-γ (58.4%, 0.60, p = 0.009), IL-6 (58.9%, 0.62, p = 0.010), and IL-12p70 (60.0%, 0.61, p = 0.005).
Predictive features spanned frontal, temporal, subcortical, and cerebellar regions.
Voxel-based morphometry revealed that IFN-γ, IL-6, and IL-12p70 models were dominated by thalamic, basal-ganglia, and cingulo-cerebellar structures, regions integral to immune-neural regulation.
Gene-expression mapping demonstrated significant enrichment of receptor-positive regions among predictive brain areas for IFN-γ (overlap = 20, p < 0.0001), IL-6 (21, p < 0.0001), and IL-12p70 (19, p < 0.0001), but not IL-1β (p = 0.60). Receptor transcripts showed the highest cortical expression for IL12RB2 (frontal pole > 5.0 log₂), IFNGR1/2 (parahippocampal and temporal > 7.8), and IL6ST (striatum > 10.5).
Discussion & Conclusions: This study provides the first integrative evidence that variability in grey matter, white matter, and cerebrospinal fluid volumes can predict peripheral cytokine levels in schizophrenia, and that these brain volumetric patterns colocalise with regional expression of their corresponding cytokine receptors. The results suggest a biologically coherent link between peripheral immune signalling and central brain architecture, implicating thalamo-cortico-cerebellar circuits in immune-related neuropathology.
Although limited by its cross-sectional design and cytokine-specific variability, this proof-of-concept framework establishes a foundation for future multimodal studies aimed at advancing biologically informed stratification of schizophrenia.
Although limited by its cross-sectional design and cytokine-specific variability, this proof-of-concept framework establishes a foundation for future multimodal studies aimed at advancing biologically informed stratification of schizophrenia.
In sessions