A novel gene signature based on five glioblastoma stem-like cell relevant genes predicts the survival of primary glioblastoma

Abstract

Purpose: Primary glioblastoma (pGBM) is the most common and lethal type of neoplasms in the central nervous system, while the existing biomarkers, lacking consideration on the stemness changes of GBM cells, are not specific enough to predict the complex prognosis respectively. We aimed to build a high-efficiency prediction gene signature related to GBM cell stemness and investigate its prognostic value in primary glioblastoma.

Methods: Differentially expressed genes were screened in GSE23806 database. The selected genes were then verified by univariate Cox regression in 591 patients from four enormous independent databases, including the Chinese Glioma Genome Atlas (CGGA), TCGA, REMBRANDT and GSE16011. Finally, the intersected genes were included to build the gene signature. GO analysis and GSEA were carried out to explore the bioinformatic implication.

Results: The novel five-gene signature was used to identify high- and low-risk groups in the four databases, and the high-risk group showed notably poorer prognosis (P < 0.05). Gene ontology (GO) terms including "immune response", "apoptotic process", and "angiogenesis" were picked out by GO analysis and GSEA, which revealed that the gene signature was highly possibly related to the stemness of GSCs and predicting the prognosis of GBM effectively.

Conclusion: We built a gene signature with five glioblastoma stem-like cell (GSC) relevant genes, and predicted the survival in four independent databases effectively, which is possibly related to the stemness of GSCs in pGBM. Several GO terms were investigated to be correlated to the signature. The signature can predict the prognosis of glioblastoma efficiently.

Keywords: GSC; Gene signature; Glioma; Primary GBM; Survival.

Fig. 1

Five GSC relevant genes were identified. a The differentially expressed status in GSE23806 between glioblastoma stem-like cell lines and conventional glioma cell lines. b The expression pattern of five included genes in CGGA database, arranged by the rising risk score of each patient. c The coefficients of five included genes

Fig. 2

Relationship between the signature risk score and the clinicopathologic characteristics. a The clinicopathologic information of patients in CGGA database, arranged by the increasing risk score. b The distribution of risk score in patients stratified by clinical information is not significantly different, except the age (P > 0.05, NS). c–e Distribution of the risk score in patients stratified by IDH status(c), MGMT promoter methylation (d), and TCGA subtypes (e). ***P < 0.001, ****P < 0.0001. f Multivariate Cox regression with clinical information and risk score for survival

Fig. 3

The performance of the risk score in databases. The overall survival of high- and low- risk score group in CGGA database (a), TCGA (b), REMBRANDT (c) and GSE16011 (d) is quite different

Fig. 4

Altered functional characteristics related to the gene signature. a, b GO terms where the positively and negatively correlated genes enriched in. c Hallmarks enriched in the high-risk score group. d GO terms and KEGG pathway enriched in the low-risk score group