Morphometric Similarity Loss and Gray Matter Atrophy Align with Neurotransmitter and Mitochondrial Maps in Drug-Resistant Epilepsy

Jiaxi Zhao,1,2 Mingjun Gao,2,3 Lanling Zhou,2,3 Shiming Wan,1,2 Nianshun Liao,2 Sixun Yu,1– 3,* Xin Chen,2,3,* Haifeng Shu1– 4,* 1Department of Neurosurgery, The Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan, People’s Republic of China; 2Department of Neurosurgery, The General Hospital of Western Theater Command PLA, Chengdu, Sichuan, People’s Republic of China; 3College of Medicine, Southwest Jiaotong University, Chengdu, Sichuan, People’s Republic of China; 4T18-Issue Stress Injury and Functional Repair Key Laboratory of Sichuan Province, Chengdu, Sichuan, People’s Republic of China*These authors contributed equally to this workCorrespondence: Haifeng Shu, Department of Neurosurgery, The Affiliated Hospital of Southwest Medical University, No. 319, Section 3, Zhongshan Road, Lianxi Road Community, Beicheng Street, Jiangyang District, Luzhou, Sichuan, People’s Republic of China, Tel +0086-28-13908052010, Email [email protected]: Epilepsy imposes a substantial global burden, and drug-resistant epilepsy (DRE) accounts for a disproportionate share of morbidity due to persistent seizures and limited treatment options. Although MRI studies have reported gray-matter (GM) atrophy in DRE, its spatial specificity, accompanying changes in morphometric similarity network (MSN) connectivity strength, and the performance of an MRI-only supportive screening model for identifying DRE remain incompletely understood. We therefore tested whether GM atrophy clusters non-randomly in territories with high neurotransmitter-receptor, cellular, and mitochondrial distributions, whether MSNs are altered, and evaluated the feasibility of an MRI-only model to support DRE identification.Methods: This study first used voxel-based morphometry (VBM) to map group GM atrophy, then constructed MSN from structural MRI features to analyze global, regional, and graph-theoretic metrics; we next tested the spatial correlation of the GM atrophy map with neurotransmitter receptor and cellular/mitochondrial distributions under family-wise FDR control. Finally, we trained classifiers using LASSO-selected MRI features to develop an MRI-based screening/support tool. All analyses were performed separately in two independent cohorts.Results: Both cohorts showed temporo-limbic–anchored GM atrophy, with discovery-cohort stratification indicating broader thalamo–ventral temporal involvement in TLE and more focal cerebellar effects in non-TLE. MSNs showed preserved global indices with focal regional meanMS reductions (isthmus cingulate/medial orbitofrontal/pars triangularis), reproduced in TLE but not other subtype contrasts. The atrophy map co-localized with 5-HT1B and mGluR5 and with mitochondrial Complex I/IV (plus respiratory capacity and a neuronal subtype map) in discovery, while validation showed no FDR-significant correspondences and opposite directions for mGluR5 and respiratory capacity. The MRI-only panel achieved moderate external AUC (~0.75), consistent with a supportive screening application rather than diagnostic replacement.Conclusion: GM atrophy in DRE aligns with neurotransmitter and mitochondrial distributions and coincides with regional meanMS reductions; an MRI-only model aids DRE identification, though causality and clinical utility await validation in larger longitudinal/interventional studies.Keywords: voxel-based morphometry, morphometric similarity networks, JuSpace, clinical prediction model