Temozolomide-associated inflammatory-repair remodeling in recurrent glioblastoma exposes a ROCK-linked therapeutic vulnerability

Abstract

Background Glioblastoma (GBM) adapts to therapy through coordinated malignant-cell and microenvironmental responses, but the mechanisms linking treatment-associated inflammation to tumor-cell phenotypic plasticity remain poorly defined. Here, we investigated whether preoperative temozolomide exposure is associated with inflammatory-repair remodeling and a ROCK-linked therapeutic vulnerability in recurrent GBM using a window-of-opportunity clinical trial cohort. Methods We integrated patient-resolved single-cell transcriptomics, spatial RNA profiling and multiplex protein imaging with experimental perturbations and orthotopic glioblastoma models. In the presurgical window-of-opportunity subgroup of NCT05236036, four patients with recurrent glioblastoma received additional preoperative temozolomide (TMZ; 150 mg/m2/day for 5 days) before resection and were compared with three recurrent comparators who did not receive additional preoperative TMZ. Results Across seven tumors, malignant-cell inflammatory activity covaried with integrin-binding and wound-healing programs (median partial ρ = 0.45 and 0.42, respectively). Tumors exposed to preoperative TMZ showed enrichment of inflammatory, chemokine, adhesion and wound-healing programs (GSEA q < 0.05), with the direction of enrichment preserved in all leave-one-patient-out analyses. In two tumors exposed to preoperative TMZ, CXCL12 localized to vascular territories, whereas chemokine and wound-repair programs increased near injury-reference regions. Across seven specimens analyzed by multiplex protein imaging, MYL9 abundance correlated with local CXCL12 (ρ = 0.61), phosphorylated MYPT1 (ρ = 0.48) and nuclear phosphorylated STAT3 (ρ = 0.45), with positive associations in every specimen. Experimentally, TMZ increased CXCL8 and reactive-state markers and enhanced subsequent scratch closure, while CXCL8 and CXCL12 increased MLC2 phosphorylation. Fasudil attenuated TNF-NF-κB, inflammatory-response and IFN-γ-response programs in TMZ-treated cells and reduced scratch closure and p-MLC2 in complementary assays. Across three orthotopic models, fasudil plus TMZ prolonged survival versus TMZ alone (model-stratified HR, 0.21; 95% CI, 0.07-0.61; P = 0.001). Conclusions These findings link treatment-associated inflammatory-repair programs with cytoskeletal remodeling and support further translational evaluation of fasudil plus TMZ in glioblastoma.