Life-Span Dynamics of Schizophrenia: Resting-State fMRI Analysis of Network Alterations
Poster No:
424
Submission Type:
Abstract Submission
Authors:
Sebastian Volkmer1, Stefan Fritze1, Geva Brandt1, Dusan Hirjak1, Emanuel Schwarz1
Institutions:
1Central Institute of Mental Health Mannheim, Mannheim, Baden-Württemberg
First Author:
Co-Author(s):
Introduction:
Schizophrenia's heterogeneity complicates identifying consistent biomarkers or targets. Normative modeling applied to resting-state fMRI (rsfMRI) may yield refined individual-level characterizations of network deviations. Building on computational psychiatry advances, we use a feature selection approach with lasso regression to derive age-predictive rsfMRI networks and compute individual-level normative deviations. We then relate these deviations to diagnosis, symptom severity, and polygenic risk scores, probing complex interactions between neural network alterations and genetic factors. By quantifying multivariate differences across the lifespan, this integrative approach offers new insights into schizophrenia's developmental and potentially neurodegenerative processes. Leveraging normative modeling to delineate personalized functional network deviations holds promise for advancing precision diagnostics and targeted interventions in schizophrenia.
Methods:
We processed large resting-state fMRI datasets from the Central Institute of Mental Health in Mannheim and publicly available cohorts. The data was processed with fmriprep version 23.2.3, and nilearn to calculate functional connectivity matrices based on the multi-subject dictionary learning atlas. After vectorizing functional connectivity matrices and applying ComBat to remove site effects, we retained approximately 3,000 healthy resting-state measurements as a normative reference. Using lasso regression, we identified age-predictive rs-fMRI features from this cohort. We then tested four independent schizophrenia spectrum disorder (SSD) cohorts (n=216) and healthy controls (n=444) by measuring their Mahalanobis distances to the normative reference subset of similarly aged individuals (±2 years). To enhance robustness against outliers, we employed the Minimum Covariance Determinant from scikit-learn and a geometric median in the Mahalanobis distance calculation. We correlated these distances with PANSS scores and polygenic risk scores. Finally, we examined each feature's individual contribution to the Mahalanobis distance.
Results:
Fitting an ordinary least squares model, we observed that patients with SSD demonstrated a significantly higher Mahalanobis distance compared to HC when matched within a ±2-year age range (p=6×10^(-4)). All analyses were adjusted for site, age, mean FD, and sex. A permutation test with 5,000 iterations confirmed that this mean difference between HC and SSD groups remained highly significant. Moreover, the Mahalanobis distance was negatively correlated with PANSS negative symptom scores (p=3×10^(-3)).
By examining the individual contributions to the Mahalanobis distance, we identified a key connection between a motor region and the posterior cingulate cortex. In our preliminary analysis, this connection displayed an age-related pattern: it increased from ages 18 to 35, then began to decrease, with the greatest group differences occurring within the 25–30 age range. Within the healthy and relatives PRS cohort, this same connection correlated strongly with PRS for trauma exposure (p=1×10^(-5)) and major depressive disorder (p=3×10^(-26)), with all PRS-related p-values corrected for false discovery rate.
By examining the individual contributions to the Mahalanobis distance, we identified a key connection between a motor region and the posterior cingulate cortex. In our preliminary analysis, this connection displayed an age-related pattern: it increased from ages 18 to 35, then began to decrease, with the greatest group differences occurring within the 25–30 age range. Within the healthy and relatives PRS cohort, this same connection correlated strongly with PRS for trauma exposure (p=1×10^(-5)) and major depressive disorder (p=3×10^(-26)), with all PRS-related p-values corrected for false discovery rate.
·Mahalanobis distance seperated by diagnosis across age bins.
Conclusions:
These findings suggest that normative modeling can characterize individualized network deviations in schizophrenia spectrum disorders. Changes in connectivity, especially between a motor region and the posterior cingulate cortex, follow a developmental pattern peaking in early adulthood. These deviations correlate with PANSS negative symptoms and polygenic risk for trauma and depression, indicating both environmental and genetic influences. This aligns with catatonic schizophrenia's links to trauma, fear responses, and motor symptoms. Further work may refine assessments, improve diagnostic strategies, and guide targeted interventions.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Genetics:
Genetics Other
Lifespan Development:
Aging
Modeling and Analysis Methods:
Classification and Predictive Modeling
fMRI Connectivity and Network Modeling 2
Keywords:
Aging
Computational Neuroscience
FUNCTIONAL MRI
Machine Learning
Schizophrenia
1|2Indicates the priority used for review