Abstract Resistance to cytotoxic therapy in glioblastoma (GBM) is largely driven by DNA repair mechanisms. Proteasome inhibition may sensitize tumors through suppression of NF-κB survival signaling and MGMT transcription. However, biological mechanisms that modulate tumor-intrinsic sensitivity remain poorly understood. 58 patients with recurrent MGMT-unmethylated GBM, undergoing sequential bortezomib-temozolomide in an ongoing Phase II trial were analyzed. Prospective machine learning (ML) response prediction employed multimodal features, integrating whole-exome sequencing, longitudinal deep learning tumor segmentation (n=116 mpMRIs), clinical variables, and quality-of-life measures. Patients were split chronologically into training (n=43) and prospective validation (n=15) cohorts. Post hoc pathway burden analysis stratified non-objective responders by unsupervised clustering, identifying distinct genomic resistance mechanisms. 14/58 patients (24%) achieved objective RANO response. ML modelling achieved validation AUC 0. 91 (permutation p=0. 0260), where HSPBP1 9 bp insertion emerged as the dominant predictor, consistent with impaired chaperone-mediated protein folding and accumulation of nascent proteins for degradation. This finding aligns with prior preclinical evidence that disruption of HSP70-dependent protein quality control amplifies proteasome-inhibitor-induced proteotoxic stress. The mutation was present in 100% of RANO-responders versus 59% in the remaining patients (p=0. 0027, OR100) suggesting this mutation confers sensitivity to proteasome inhibition. However, objective response was only observed when accompanied by preserved downstream apoptosis and NF-κB signaling. Responders exhibited significantly lower loss-of-function (LoF) mutation burden in apoptosis, NF-κB, cell cycle regulation, and RTK signaling (all p0. 03). Pathway burden clustering identified three distinct resistance mechanisms: I) deficient apoptosis machinery; II) proteasome-independent survival pathways; and III) insufficient proteotoxic stress. Survival differed across groups (overall log rank p=0. 013), where cluster I yielded the poorest OS (median 15. 5 months vs 20. 9 in RANO-responders, p0. 01). A radiogenomic association also emerged: baseline contrast-enhancing (CE) /non-enhancing (NE) volume ratio correlated with apoptosis LoF burden (Spearman’s ρ=0. 454, p0. 001), where lower CE/NE ratios denoted preserved apoptotic capacity and greater treatment sensitivity. This manifested clinically by 4. 7-fold higher response rate in patients with CE/NE ratio ≤0. 324 and low LoF burden (OR=13. 42, p=0. 0026). Although clinical trials are inherently not powered for ML endpoints, our unique use of primary data in a prospective clinical treatment setting provides actionable insights. HSPBP1 mutation, concurrent with preserved apoptosis and NF-κB pathway integrity, emerged as predictive biomarker of clinical benefit from proteasome inhibition in recurrent MGMT-unmethylated GBM. Citation Format: Marianne H. Hannisdal, Mohummad A. Rahman, Nello Blaser, Leif Oltedal, Judit Haaz, Arvid Lundervold, Petter Brandal, Tora S. Solheim, Dorota Goplen, Martha Chekenya. HSPBP1 9 bp insertion mutation concurrent with pro-apoptosis and NF-κB genomic integrity predicts response to bortezomib-mediated proteasome inhibition in recurrent glioblastoma treated in NCT03643549 trial abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts) ; 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86 (8Suppl): Abstract nr LB014.
Hannisdal et al. (Fri,) studied this question.