Functional clustering within glioblastoma identifies a network-integrated subtype with improved survival

Glioblastoma (GBM) is traditionally viewed as disrupting brain network connectivity. Recent evidence of tumor-neuron synaptic interactions suggests gliomas may actively engage with the brain. We hypothesized that GBM tumors contain functionally distinct intratumoral subregions that differentially couple with resting-state networks (RSNs) and carry prognostic significance.

We applied fuzzy c-means clustering to resting-state fMRI data from 190 GBM patients to identify and count the number of intratumoral functional subregions. We used the participation ratio, a PCA-derived measure of signal dimensionality, as a continuous variable complement to cluster number. Cluster-level BOLD time series were correlated with seven canonical RSNs and compared to tumor location-matched voxels in 347 healthy subjects. Associations with survival and preoperative seizures were evaluated using Kaplan-Meier analysis, restricted mean survival time (RMST), multivariable Cox proportional hazards modeling, and multivariable logistic regression.

Fifty-two of 190 GBMs exhibited multiple intratumoral functional clusters. Multi-cluster tumors were associated with longer survival and higher seizure prevalence. Elevated RSN coupling was more prevalent in multi-cluster tumors as compared to single-cluster tumors, with elevated coupling strongly predicting multi-cluster status. Distinct clusters often showed preferential coupling to different RSNs.

A subset of GBM contains functionally distinct subregions that selectively engage large-scale brain networks, a pattern associated with favorable survival and increased seizure risk. These findings support a model in which certain tumors function as active participants within neurovascular networks, indicating a novel network-integrated subtype of GBM with implications for prognosis and disease staging.