Plasma Proteins Predict Epilepsy Years Before Diagnosis

Epilepsy affects tens of millions globally, yet clinical diagnosis still relies heavily on patient history, medical records, and imaging during seizures—with no reliable blood-based biomarker available. This study represents a major step toward changing that, leveraging one of the world's largest plasma proteomic datasets to map the molecular landscape preceding epilepsy onset.

Using plasma proteome data from 52,372 UK Biobank participants free of epilepsy at baseline, researchers ran longitudinal Cox proportional hazard models across 2,920 proteins over a mean follow-up of nearly 14 years, during which 440 participants developed incident epilepsy. After rigorous Bonferroni correction, 103 proteins emerged as significantly associated with epilepsy risk. NEFL (HR 2.13, 95% CI 1.85–2.46) and GDF15 (HR 1.82, 95% CI 1.60–2.07) were the strongest predictors, with p-values as low as 3.36×10⁻²⁵. Notably, both proteins remained significant across short-term (≤5 years) and long-term (>5 years) follow-up windows, suggesting sustained biological relevance.

Protein trajectory analyses revealed abnormal changes in epilepsy-associated protein levels up to 15 years before diagnosis—a finding with significant implications for early detection. Pathway enrichment and protein network analyses consistently highlighted immune response mechanisms, identifying four central hub proteins. The 103 epilepsy-associated proteins were also correlated with neuroimaging data from brain regions implicated in epileptogenesis and showed stronger associations with environmental stress-related variables than with polygenic risk scores, suggesting that acquired rather than purely genetic factors may drive much of the proteomic signal.

Subgroup analyses stratified by age, sex, diabetes status, and kidney function largely replicated the primary findings. Subtype-specific proteins were identified for focal epilepsy (LRG1, HAVCR2, VSIG4, SPINK1) and generalized epilepsy (BCAT1, GDF15, SOD2, GAGE2A). A machine learning predictive model incorporating top-ranked proteins demonstrated meaningful discriminative ability for future epilepsy risk, and several identified proteins were flagged as candidate drug targets based on existing pharmacological databases.

These findings collectively offer a roadmap for developing blood-based screening tools and mechanism-informed therapies for epilepsy—a disease where cause-specific treatment remains largely unavailable.