# Cerebrospinal Fluid Genetics Enhance Risk Stratification in Bipolar Disorder

## TL;DR

This study uses cerebrospinal fluid genetics to improve risk prediction in bipolar disorder, offering a scalable alternative to direct CSF collection.

## Contribution

A novel genetic proxy strategy for CSF biomarkers in bipolar disorder, enabling scalable risk stratification.

## Key findings

- Four distinct genetic clusters were identified, each with unique clinical and biological features.
- A multimodal polygenic risk score model outperformed clinical-only prediction (C-index = 0.77).
- Genetically informed CSF biomarkers reduced false-negative rates by 84% in low-risk subgroups.

## Abstract

Bipolar disorder (BD) research confronts challenges: blood‐based biomarkers offer limited insights into neurobiology, while cerebrospinal fluid (CSF) collection is clinically unusual. Linking genetic susceptibility to pathophysiology remains crucial for biologically informed risk stratification. We integrated cohort data and genome‐wide association study (GWAS) summary statistics: the largest BD meta‐analysis, CSF multi‐omics profiles including 3107 proteomic and 2602 metabolomic participants, and a validation cohort of 247,834 UK Biobank participants. Unsupervised clustering revealed four single‐nucleotide variant (SNV) clusters: metabolic‐imbalance, metabolic‐active, human leukocyte antigen (HLA)+immune, and HLA‐immune. These clusters exhibited distinct clinical features, with the metabolic‐imbalance cluster showing multi‐directional associations with 21 psychiatric traits, while the HLA‐immune cluster was associated with emotional instability in BD patients (odds ratio [OR] = 1.14, p = 0.027). The optimized multimodal cluster‐specific polygenic risk scores (PRS) model significantly outperformed clinical‐only prediction factors (C‐index = 0.77), with the metabolic‐imbalance PRS contributing a 22.6% incremental predictive value (hazard ratio [HR] = 1.23, 95% CI: 1.04–1.45, p = 0.016). Risk reclassification showed an 84% reduction in false‐negative rates in the low‐risk subgroup, identifying a high‐risk layer with a 17.6‐fold increased BD incidence. Altogether, genetically informed substitutes for CSF biomarkers emerged as a scalable tool for risk prediction, overcoming the barriers of CSF collection while capturing neurobiological heterogeneity.

This study introduced a novel genetic proxy strategy that connects the CSF biomarkers with the clinical applications in BD research. First, it delineated a genetically anchored CSF biomarker network. Second, it demonstrated the clinical utility of multimodal PRS. Finally, it provided a scalable framework for risk stratification. These insights pave the way for precision monitoring and therapeutic strategies tailored to the heterogeneous pathophysiology of BD.