Transcriptomic signature of temozolomide treatment-effect heterogeneity in IDH-wildtype glioblastoma.

2084 Background: Adjuvant temozolomide (TMZ) is standard of care for IDH-wildtype glioblastoma (GBM), yet clinical benefit is heterogeneous and not fully explained by MGMT promoter methylation. We developed a transcriptomic signature specifically predicting TMZ benefit rather than overall prognosis using a causal inference framework. Methods: Transcriptomic and clinical data from 448 patients with histologic grade 4 IDH-wildtype GBM treated with radiotherapy alone or radiotherapy plus TMZ across six cohorts were analyzed. A training set (N=322; TCGA, CGGA-325/693, GLASS) and an independent validation set (N=126; GSE7696, CGGA-301) were used. Data were normalized, batch-corrected, and screened for outliers. To address non-random treatment assignment, inverse probability of treatment weighting (IPTW) based on propensity scores from age, sex, and MGMT status was applied. A 15-gene Temozolomide Sensitivity Score (TSS) was derived using univariate prognostic filtering followed by LASSO modeling of gene-by-treatment interactions. Final gene weights were obtained from a doubly adjusted multivariable Cox model incorporating IPTW and clinical covariates. Performance was assessed by comparing adjusted hazard ratios (HRs) for TMZ benefit across TSS-defined subgroups. Results: Five genes emerged as significant independent modulators of TMZ response: C2CD4A, SLC35E3, and DNASE1L3 predicted sensitivity, whereas APOBEC3B and PMAIP1 predicted resistance. In the training cohort, TSS-sensitive patients derived significant benefit from TMZ (median overall survival, 21.0 vs 7.7 months; adjusted HR, 0.20; 95% CI, 0.10–0.39; P<0.001), whereas TSS-resistant patients did not (14.4 vs 13.1 months; adjusted HR, 0.89; 95% CI, 0.53–1.51; P=0.674), with a significant difference in treatment effect (P<0.001). Validation confirmed these findings: TSS-sensitive patients benefited (21.9 vs 12.8 months; adjusted HR, 0.32; 95% CI, 0.16–0.62; P<0.001), while TSS-resistant patients did not (13.7 vs 15.1 months; adjusted HR, 1.25; 95% CI, 0.70–2.23; P=0.453), showing differential efficacy (P=0.003). Age, sex, and MGMT status were balanced between groups. Conclusions: A 15-gene transcriptomic signature developed through causal inference and interaction modeling is associated with differential temozolomide benefit in IDH-wildtype glioblastoma independent of MGMT status. This signature may identify patients less likely to benefit from temozolomide, offering a promising tool for de-escalation strategies and enrollment in precision clinical trials.