Treg-driven magnetic resonance radiomics in glioblastoma: a multicenter and cross-species model for prognostic biomarker discovery.

Abstract

Glioblastoma (GBM) is the most aggressive primary brain tumor in adults. This study aimed to develop and validate a regulatory T cell (Treg)-associated magnetic resonance imaging (MRI) radiomics model and assess its prognostic value, cross-species similarity, and interpretability across multicenter cohorts. Tregs retained prognostic significance for patients with GBM under World Health Organization CNS 5 classification. In murine models, Treg depletion suppressed tumor growth and reduced the infiltration of myeloid-derived suppressor cells, tumor-associated macrophages, and exhausted CD8T cells. The in vitro co-culture of Tregs with GL261 cells significantly reduced radiation-induced apoptosis. We constructed a Treg-associated radiomics model comprising six features and validated the prognostic value of the radiomics score (RS) across multicenter clinical imaging cohorts (First Affiliated Hospital of Fujian Medical University, n = 111, hazard ratio [HR] = 2.178, 95% confidence interval [CI]: 1.269-3.740; Second Affiliated Hospital of Zhejiang University School of Medicine, n = 126, HR = 1.664, 95% CI: 1.049-2.640; The Cancer Genome Atlas, n = 86, HR = 1.811, 95% CI: 1.089-3.010). Treg depletion experiments also confirmed causal associations between two radiomic features and Treg infiltration. RS stratification correlated with hypoxia, glycolysis, interleukin (IL)-2/signal transducer and activator of transcription (STAT) 5, and IL-6/Janus kinase/STAT3 signaling pathways and immune checkpoints and genomic alterations. The MRI radiomics model enables noninvasive prediction of Treg infiltration and serves as a robust, generalizable prognostic biomarker for GBM. These findings provide evidence for biological causal associations and cross-species similarities between radiomics features and Treg infiltration.