2025 American Academy of Neurology Abstract Website

Objective:

To assess the reproducibility and relevance of published glioblastoma (GBM) prognostic signatures in a real-world dataset.

Background:

GBM is the most common malignant brain tumor in adults. Overall, the prognosis for GBM is poor; the five-year survival rate is 7.2% and almost all GBMs recur after treatment. An increasing number of prognostic gene signatures have been published using different techniques and data. However, these largely have not been validated in independent datasets.

Design/Methods:

Ten known prognostic signatures in GBM, from studies published between 2019 to 2024, were analyzed both independently and as a 76-gene meta-signature. Clinical and mRNA expression data from four Cleveland Clinic Foundation (CCF) GBM clinical trials were utilized. Prognostic performance was also evaluated in open-source datasets from The Cancer Genome Atlas (TCGA) and the Chinese Glioma Genome Atlas (CGGA). Gene signature scores were computed using single-sample gene set enrichment analysis (ssGSEA). Survival analysis using Cox proportional hazards regression and log-rank tests was performed to compare overall survival, using median ssGSEA to stratify risk groups. Using sigQC, we identified and assessed the prognostic performance of 5 genes from the meta-signature that is both highly expressed and highly variable.

Results:

The meta-signature was not prognostic in the CCF data (p=0.3), but was prognostic in both TCGA (p=0.004) and CGGA (p<0.0001). None of the individual signatures were prognostic in CCF data. All gene signatures except one were prognostic in one or both TCGA and CGGA, highlighting the variability in performance between gene signatures across different datasets. The 5-gene set and none of the combinations were prognostic in CCF data.

Conclusions:

The meta-signature and individual gene signatures demonstrated inconsistent prognostic performances. This suggests the lack of applicability of published prognostic gene signatures and highlights the need to construct a more robust, novel signature to inform clinical care.