18. DEVELOPMENTAL BOTTLENECK GENES IN DORSOLATERAL PREFRONTAL CORTEX RISK NETWORKS REVEAL SHARED AND DISORDER-SPECIFIC ARCHITECTURE IN SCHIZOPHRENIA AND BIPOLAR DISORDER

Background Schizophrenia (SCZ) and bipolar disorder (BD) genetic risks are highly polygenic and largely non-coding, suggesting that their effects are mediated through gene regulatory networks. However, most studies rely on static models that fail to capture developmental dynamics. We aimed to identify bottleneck genes -genes mediating connectivity between risk genes (SCZ, BD)- in the dorsolateral prefrontal cortex (DLPFC), disentangle shared from disease-specific mechanisms. Methods We analyzed DLPFC RNA-seq data from 264 neurotypical individuals using a sliding-window approach to identify 205 age-resolved co-expression networks. Shortest paths between SCZ/BD risk genes (PGC3) were computed, and path-restricted betweenness centrality (BC-risk) was estimated across time. Two permutation frameworks (10,000 iterations each) identified: (i) shared bottlenecks through edge-weight permutation, and (ii) disease-specific mediator genes through risk-gene label permutation. Temporal dynamics were characterized using Leiden clustering based on Tanimoto similarity. Cellular specificity of bottleneck-risk coupling was evaluated using single-nucleus co-expression networks from two independent datasets. Enrichment across cortical cell types was assessed using permutation testing. Results We identified 437 SCZ- and 338 BD-significant mediator genes (p < 0.0001). Among these, 76 SCZ and 51 BD genes were prioritized as bottlenecks, with 23 shared between disorders, indicating conservation of core network architecture. However, 47.8% of shared bottlenecks belonged to different temporal clusters, suggesting disorder-specific developmental deployment. Temporal clustering revealed three communities: Perinatal–Juvenile (peak ≤18y), Adult (18–40y), and Lifespan (steady increase). Two developmental transitions emerged at 16.6–18.5 and 37.7–48.6 years, reflecting coordinated shifts in bottleneck activity. Risk-label permutation identified 30 SCZ-specific mediator genes, 19 overlapping with bottlenecks, whereas BD showed only one disease-specific intermediate gene. Single-nucleus analyses revealed disorder- and epoch-specific enrichment patterns (p < 0.05). In BD, enrichment involved L2/L6 excitatory neurons and neurogliaform interneurons during the Perinatal–Juvenile stage, L4 excitatory neurons and VIP interneurons during adulthood, and L2 excitatory neurons across the Lifespan. In SCZ, enrichment involved L2/L6 excitatory neurons during the Perinatal–Juvenile stage, L2/L3 excitatory neurons during adulthood, and L2/L3/L6 excitatory neurons with VIP interneurons across the Lifespan, with the strongest signal in Lifespan L2 neurons (p < 0.001). Discussion SCZ genetic risk propagates through dynamic bottleneck genes organized into a conserved topological backbone along with disease-specific mediators, indicating that risk transmission depends on both stable architecture and selective disease-specific routing. Developmental transitions, particularly in late adolescence, align with SCZ onset and reflect network reorganization. Comparison with BD revealed conserved bottlenecks but distinct temporal deployment. The absence of overlap between SCZ- and BD-specific intermediates indicates that shared architecture is balanced by disorder-specific mechanisms of genetic risk propagation. Single-nucleus analyses further supported disorder-specific temporal and cell-type deployment of bottleneck-mediated risk propagation.