Opposing associations of default mode network and subcortical network efficiency with IQ in children with drug-resistant epilepsy.
Paediatric drug-resistant epilepsy (DRE) is associated with IQ deficits only partially explained by clinical seizure variables (for example, age of onset, seizure burden, seizure location). Given that intelligence depends on efficient large-scale brain networks, structural connectomics provides a complementary mechanistic framework for explaining residual IQ variance. We tested whether brain-network architecture explains additional IQ variance using global and network-averaged graph-theoretic metrics derived from streamline-count weighted diffusion MRI connectomes. Seventy-one children with DRE (50 focal epilepsy; 21 multifocal epilepsy) and 15 control participants underwent diffusion and T1-weighted MRI. For each participant, a 253 × 253 structural connectome was constructed, edge weights were defined as the number of streamlines, and graph metrics were computed using the Brain Connectivity Toolbox. Nodal metrics were averaged within Yeo's seven functional networks plus a Subcortical network. Associations with IQ were examined using correlations and general linear models (single-network and eight-predictor models). Finally, mediation analyses tested whether network metrics explained IQ differences between controls and children with DRE. Global metrics were not associated with IQ (all p > 0.05). Regionally, higher Salience-network betweenness centrality showed a nominal negative association with IQ in the multifocal subgroup (p = 0.040*, adjusted R2 = 0.142). In the eight-network GLM, higher nodal efficiency within the Default Mode Network (DMN) was positively associated with IQ (B = 122.406, p = .013), whereas higher nodal efficiency within the Subcortical Network was negatively associated with IQ (B = -51.942, p = .012). These regional associations did not survive Bonferroni correction. Exploratory mediation analyses suggested that opposing DMN and Subcortical Network effects partially accounted for the observed group difference in IQ. These findings are hypothesis-generating, as the regional associations with IQ were nominal and did not survive family-wise correction. Nevertheless, the mediation effects survived correction, highlighting opposing regional, rather than global, network alterations as candidate correlates of IQ variability in paediatric DRE.