Altered intrinsic thalamic and subcortical functional connectivity in temporal lobe epilepsy with focal-to-bilateral tonic-clonic seizures
Background. Temporal lobe epilepsy (TLE) is the most common drug-resistant focal epilepsy, and a substantial subgroup develops focal-to-bilateral tonic-clonic seizures (FBTCS) carrying heightened risk for SUDEP, cognitive decline, and poor surgical outcomes. The subcortical network mechanisms that index seizure generalization, their dependence on onset laterality, and their utility as biomarkers are poorly defined. Methods. Resting-state fMRI was acquired from 166 TLE patients (120 FBTCS+, 46 FBTCS-; 71 right TLE, 95 left TLE) and 119 healthy controls. Functional network connectivity and graph-theoretic analyses quantified clustering coefficient, degree centrality, local efficiency, and intrinsic connectivity contrast across 16 bilateral thalamic regions (Aim 1) and 54 ROIs spanning thalamus, hippocampus, amygdala, caudate, putamen, globus pallidus, and nucleus accumbens (Aim 2). A prespecified rmMANCOVA cascade tested covariate robustness of FBTCS effects. Aim 3 used nested 10x5 cross-validated linear SVMs with bootstrap and DeLong inference, prespecified sensitivity analyses, and permutation testing to compare connectivity versus standard clinical features for FBTCS classification. Results. Aim 1 showed reduced cross-hemispheric ventral anterior thalamic connectivity in TLE, with more pronounced ipsilateral dysfunction in right TLE; only right-TLE FBTCS+ patients showed ipsilateral VA disruption that worsened with illness duration. Aim 2 identified six bilateral subcortical subnetworks with reduced clustering coefficient in left ventroposterior lateral thalamus, left ventral anterior caudate, and right ventroposterior putamen in FBTCS+ patients. Right TLE FBTCS+ patients showed elevated bilateral nucleus accumbens cross-hemispheric connectivity that tracked with recent seizure burden, dissociating a state-dependent accumbens marker from trait-like thalamo-striatal desegregation. In Aim 3, connectivity outperformed clinical features in pooled TLE (AUC 0.618 vs 0.418, delta-AUC = +0.200, DeLong p = .005) and most strongly in right TLE (AUC 0.657 vs 0.311, delta-AUC = +0.346, DeLong p = .002). Left TLE was null across all panels, serving as a within-study negative control. Ventroposterior thalamus and nucleus accumbens emerged as top discriminators. Conclusions. Subcortical connectivity abnormalities in TLE organize into limbic-thalamo-striatal subnetworks with state- versus trait-like dissociation. Connectivity features discriminate FBTCS status beyond clinical features in pooled and right TLE, while left TLE remains null. These signatures motivate prospective external validation as candidate FBTCS risk-stratification biomarkers and targets for personalized treatment.