评估用于胶质母细胞瘤患者反应分层的生物标志物可重复性

Purpose: Glioblastoma (GBM) is the most common and aggressive primary malignant brain tumor, with a median survival of only 15 months. Effective therapeutic advances remain limited, with few clinical trial patients exhibiting robust responses to experimental therapies. This has led to efforts to identify molecular biomarker signatures capable of stratifying patient responses either prognostically (pre-treatment) or diagnostically (post-treatment). This study aims to assess the robustness and reproducibility of published transcriptional GBM biomarker signatures for predicting clinical outcomes in GBM trials. Methods: We analyzed published RNA-seq data from 4 clinical trials involving oncolytic virotherapy and immune checkpoint inhibitors, as well as multiple cohorts of patients treated via the Stupp protocol. The performance of reported immune gene signatures was evaluated across all datasets using both grouped and continuous survival modeling. Intra-patient signature variation between different biopsy sites and timepoints was assessed using longitudinal, multi-site sampling collected by the Break Through Cancer Accelerating GBM Therapies TeamLab. LASSO, multivariate Cox regression, and risk scoring with FDR adjustment were used to identify novel signatures with cross-cohort utility. Key Findings: No single published signature consistently predicted survival across all analyzed trials. A post-treatment ssGSEA antitumor cytokine signature that was associated with survival in the CAN-3110 OV trial (R = 0.74, p < 0.01) was also significantly associated with improved survival in an adjuvant anti-PD1 trial (R = 0.75, p = 0.03) and neoadjuvant anti-PD1 trial (R = 0.30, p = 0.05), but not in standard-of-care controls. PAM clustering of MCP immune signatures, as reported for the DNX-2401 OV+Anti-PD1 trial, also stratified post-treatment survival in the CAN-3110 OV trial (p < 0.01), with the coldest TME subtype consistently predicting worse survival. However, the prognostic value of the most immune enriched TME subtype was inconsistent and did not extend to the Anti-PD1 trials. Furthermore, multi-site, longitudinal biopsy samples revealed marked heterogeneity in biomarker signatures between biopsies from individual patients, even when assessed at the same timepoint. We've identified several candidate transcriptomic signatures that show promise for generalizable prognostic value in GBM patient cohorts and are characterizing their spatiotemporal variability. Conclusions: While no individual immune signature is universally predictive in GBM immunotherapy, certain post-treatment cytokine signatures and TME stratifications are significant in multiple immunotherapy contexts, though not in standard-of-care patient cohorts. Marked spatiotemporal heterogeneity in these signatures underscores the need for composite prognostic markers resilient to sampling variance.