Authors :
Presenting Author: Tareq Alzaher, MD – Jordan University of Science and Technology
Ayah Al-Bzour, MD – Jordan University of Science and Technology
Sara Abualinin, MD – Jordan University of Science and Technology
Noor Al-Bzour, MD – Jordan University of Science and Technology
Yousef Badran, MD – Jordan University of Science and Technology
Mohammad Samarah, MD – Jordan University of Science and Technology
Bayan Alsarayreh, MD – Jordan University of Science and Technology
Nour Othman, MD – Jordan University of Science and Technology
Ahmed Yassin, MD – Jordan University of Science and Technology
Rationale:
Approximately one-third of individuals with epilepsy experience resistance to anti-seizure medications (ASMs). Emerging evidence suggests that neuroinflammation plays a critical role in both seizure generation and drug resistance. This study aims to identify immune biomarkers predictive of ASM response and seizure localization by integrating transcriptomic and immune profiling with machine learning (ML).
Methods:
We analyzed transcriptomic datasets from peripheral blood (GSE143272; n=91) and resected brain tissue (GSE256068; n=135) using differential gene expression and pathway enrichment techniques. Differentially expressed genes (DEGs) in epilepsy were intersected with inflammation-related gene sets from the Molecular Signatures Database (MSigDB). Immune cell infiltration and microenvironment features were estimated using xCELL. These immune metrics, including ImmuneScore, StromaScore, and neutrophil to lymphocyte ratio (NLR), were used to train a Random Forest Classifier (RFC) to predict ASM response and seizure focus. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and classification accuracy.
Results:
We identified 335 DEGs, with 11 linked to inflammation, including HLA-C, PTGS2, and CD69. Valproate responders (n=16) showed significantly reduced infiltration of adaptive immune cells compared to non-responders (n=9), such as CD4⁺ T-cells (p=0.004), CD8⁺ T-cells (p=0.017), Th1 cells (p=0.008), and NK cells (p=0.017). These immune cell infiltration patterns were not observed in carbamazepine or phenytoin subgroups. RFC models accurately predicted valproate response (AUC=0.92, accuracy=0.90) and seizure localization (accuracy=0.98), clearly distinguishing temporal from frontal lobe epilepsy. The top performing features were TMEM140, HLA-C, and CD69 in the inflammation-genes model, while the StromaScore, ImmuneScore, and NLR were the top features in the immune model.
Conclusions:
Our findings reveal that neuroinflammation influences both treatment response and seizure localization in epilepsy. Valproate responders showed distinct immune suppression suggesting inflammation status could guide therapy selection. These results highlight that neuroinflammation can be used as a potential predictive biomarker and a potential target for tailoring valproate use or immunomodulatory treatment.
Funding: None.