Joint analysis of single-cell RNA sequencing and bulk transcriptome reveals the heterogeneity of the urea cycle of astrocytes in glioblastoma

Background: Glioblastoma (GB) is incurable with a dismal prognosis. Single-cell RNA sequencing (scRNA-seq) is a pivotal tool for studying tumor heterogeneity. The dysregulation of the urea cycle (UC) frequently occurs in tumors, but its characteristics in GB have not been illuminated. This study integrated scRNA-seq UC scores and bulk RNA-seq data to build a GB prognostic model. Methods: Samples from 3 pairs of GB patients were collected for scRNA-seq analysis. GB-mRNA expression data, clinical data, and SNV mutation data were sourced from the Cancer Genome Atlas (TCGA). GB-mRNA expression data and clinical data were downloaded from the Chinese Glioma Genome Atlas (CGGA). GB RNA-seq data and clinical data were obtained from Gene Expression Omnibus (GEO) database. The R package Seurat was applied for scRNA-seq data processing. UMAP and TSNE were used for dimensionality reduction. UCell enrichment method was employed to score each astrocyte. Monocle algorithm was applied for pseudotime trajectory analysis. CellChat R package was applied for cell communication analysis. Cell labeling was performed on the results of the nine subclusters of astrocytes. The GSE138794 dataset was used to validate the results of single-cell classification. For bulk RNA-seq, univariate Cox and LASSO analyses were undertaken to screen prognostic genes, while multivariate Cox regression analysis was applied to set up a prognostic model. The differences between high-risk (HR) and low-risk (LR) groups were studied in terms of immune infiltration, sensitivity to antitumor drugs, etc. We verified the effect of the marker gene on the function of GB cells at the cellular level. Results: The analysis of scRNA-seq data yielded 7 core cell types. Further clustering of the largest proportion of astrocytes resulted in 9 subclusters. UC score and pseudotime analysis revealed the heterogeneity and differentiation process among subclusters. Subcluster 8 was annotated as an immature astrocyte (marker: CXCL8), and this cell cluster had a higher UC score. The results were validated in the GSE138794 dataset. Combining UC scores, we performed univariate Cox and LASSO to select prognostic genes on bulk RNA-seq data. A prognostic model based on 5 feature genes (RGS4, CTSB, SERPINE2, ID1, and CALD1) was established through multivariate Cox analysis. In addition, patients in the HR group had higher immune infiltration and immune function. Final cell experiments demonstrated that 5 feature genes were highly expressed in GB cells. CALD1 promoted the malignant phenotype of GB cells. Conclusion: We set up a novel prognostic model for predicting the survival of GB patients by integrating bulk RNA-seq and scRNA-seq data. The risk score was closely correlated with immune infiltration and drug sensitivity, pinpointing it as a promising independent prognostic factor.