The glioblastoma virome across tumor and extracellular compartments.

Background: Glioblastoma (GBM) is the most aggressive primary brain tumor in adults, marked by rapid progression and poor prognosis. GBM may also harbor a complex virome, defined here as the total repertoire of viral elements within tumors, encompassing both exogenous viruses and reactivated endogenous retroviruses. While viral signatures have been observed in other cancers, their transcriptional activity in GBM and their distribution across tumor cells and extracellular compartments remain largely unexplored. Methods: We analyzed 45 long- and small-RNA libraries from glioma stem-like cells, their extracellular fractions (exosomes, microvesicles, ribonucleoprotein complexes), and normal brain controls (BioProject PRJNA360129). Taxonomic profiling and functional annotation were performed using established pipelines (MetaPhlAn4, BLASTn, UniProtKB, InterPro), with fresh media controls to exclude background contamination. To validate discriminative power, viral RNA profiles were classified using machine learning -Random Forest models trained on nine species identified as glioblastoma-enriched by LEfSe. Results: We detected ~3.41 million viral reads, with >99% from GBM cells and extracellular compartments and <1% from normal brain. Reads spanned 36 species across 11 phyla. Nine species were consistently enriched in glioblastoma: Saimiriine gammaherpesvirus 2, Human endogenous retrovirus K, Camelpox virus, Pestivirus A, Macacine gammaherpesvirus 5, Avian myelocytomatosis virus, Rabbit fibroma virus, Omegapapillomavirus 1, and Cyprinid herpesvirus 2 with genes mapping to oncogenic pathways. Random Forest models showed long-RNA profiles perfectly discriminated against tumor from normal samples (AUC = 1.0), whereas small-RNA profiles carried limited signal (AUC = 0.5). Tumor-extracellular comparisons revealed strong compartment-specific long-RNA signatures (AUC = 1.0), with partial overlap between tumor cells and exosomes in small-RNA data (AUC = 0.75). Conclusion: Glioblastoma contains a distinct viral transcriptome; long-RNA profiles show strong diagnostic power, and compartment-specific signatures suggest biomarker potential.