Abstracts

Rationale: Treatment of focal epilepsy often involves multiple failed anti-seizure medication (ASM) trials before seizures are controlled or a patient is considered for surgery. To better prognosticate which patients will develop severe focal epilepsy, we sought to identify inflammation-associated protein biomarkers in plasma early in the disease course. Protein concentrations were combined to build a predictive algorithm that provides the probability of ASM response. Here we demonstrate the predictive ability of this test and demonstrate it is robust despite confounding variables.

Methods: Plasma was collected from 57 participants upon enrollment in the Human Epilepsy Project 1 (HEP1), a multicenter prospective observational study of people with newly treated focal epilepsy, aged 12-60 years, without progressive neurological disease, large MRI lesions, or major comorbidities. Participants were studied for approximately 3 years and were stratified into a group with no seizures (NS, n=20) and a frequent focal impaired awareness seizures (FIAS) group despite ASMs (FS, n=37). Seizure frequency was assessed via seizure diaries and medical records. We measured concentrations of 51 different inflammatory proteins using six simultaneous multiplex ELISA panels, and quantitated using Meso Scale Discovery Platform. A subset of proteins was used to construct a predictive algorithm to distinguish the two cohorts. Proteins were selected by considering the mean concentration differences between cohorts, as measured by the effect size, the relative classification importance, measured using a bootstrapped random forest selection technique and the uniqueness of information content measured using the Akaike information criterion.

Results: FS group (n=37) participants met at least one of the following criteria: 0.5 FIAS per month; or ≥ 1 FIAS in ≥ 23% of the months enrolled (3^rd quartile); or ≥ 1 FIAS in each of the last three months enrolled. There were no significant demographic differences between the FS and NS (n=20) groups. Plasma drawn at enrollment—before ASM response was known—was tested for the concentration of 51 proteins. Three proteins—P-Cadherin, MMP-9 and MIF—effectively predicted seizure control with an AUC of 0.88, sensitivity of 87%, specificity of 85% and a cross-validated accuracy of 83% when combined into an algorithm. Potential confounding variables including choice of ASM, time elapsed from last seizure, and seizure semiology did not influence the results.

Conclusions: An algorithm consisting of concentrations of three peripherally circulating proteins—P-Cadherin, MMP-9 and MIF—predicts ASM response in newly treated focal epilepsy participants, with a cross-validated accuracy of 83%. This tool may allow early identification of focal epilepsy patients who may prove refractory to ASMs and require more aggressive treatment. Furthermore, study of P-Cadherin, MMP-9 and MIF may inform the development of novel treatments for focal epilepsy.

Funding: Please list any funding that was received in support of this abstract.: Funding for the Human Epilepsy Project was provided by the Epilepsy Study Consortium via grants from UCB, Pfizer, Lundbeck, Eisai, Vogelstein, Sunovion. Andrews Foundation, Henry & Katherine Chesbrourgh, SK Life Science, and Engage.