Construction and validation of risk prediction model for glioblastoma associated with cancer stem cells and disulfidptosis.

Background: Glioblastoma (GBM) is the most common malignant brain tumor, and effective therapeutic strategies remain scarce. Therefore, the study aims to screen biomarkers to reveal the molecular mechanisms of cancer stem cells (CSCs) and disulfidptosis in GBM therapy. Methods: The transcriptome data and clinical data for GBM were retrieved from The Cancer Genome Atlas (TCGA) database. The GSE74187 was obtained from the Gene Expression Omnibus (GEO) database. Firstly, the messenger RNA (mRNA)-based stemness index (mRNAsi) and disulfidptosis scores were computed, and weighted gene co-expression network analysis (WGCNA) was performed in TCGA-GBM using mRNAsi and disulfidptosis scores as traits to obtain the key module genes. Secondly, differentially expressed genes (DEGs) of the disease and normal groups in TCGA-GBM were screened, and DEGs were used to cross the key module gene to obtain intersection genes. Thirdly, univariate Cox, least absolute shrinkage and selection operator (LASSO), and multivariate Cox regression analyses were performed on intersection genes in TCGA-GBM to screen for biomarkers, and the biomarkers were used to construct survival risk score model. Meanwhile, clinical characteristics, immune infiltration, and drug sensitivity prediction of biomarkers were studied. To clinically validate the identified biomarkers, reverse transcription quantitative polymerase chain reaction (RT-qPCR) analysis was conducted on 20 samples. Results: LOXL1, LOXL4, and SP6 were obtained by three regression analyses to build survival risk score model, and they were verified in GSE74187. Clinical pathological features analysis found that risk score and isocitrate dehydrogenase (IDH) status were independent prognostic factors for GBM. Immune-related analysis showed that the risk score had positive correlation with PDCD1, NRP1, TGFB1, and PVRL2. In silico drug sensitivity prediction suggested differential responses to 88 compounds between risk groups. Furthermore, preliminary molecular docking analysis indicated potential binding affinity of A_443654 and A_770041 with the biomarker proteins, highlighting these compounds as candidates for further experimental investigation. The RT-qPCR results revealed significant overexpression of LOXL1 and SP6 in the tumor group compared to the control group, while LOXL4 expression remained with no significant difference between the groups. Conclusions: A risk prediction model based on LOXL1, LOXL4, and SP6 was constructed and validated. The model, rather than individual genes, provides valuable insights, providing valuable insights for the diagnosis of GBM in the field of CSC and disulfidptosis.