Integration of clinical characteristics and molecular signatures of the tumor microenvironment to predict the prognosis of neuroblastoma.

This study aimed to analyze the clinical characteristics, cell types, and molecular characteristics of the tumor microenvironment to better predict the prognosis of neuroblastoma (NB). The gene expression data and corresponding clinical information of 498 NB patients were obtained from the Gene Expression Omnibus (GEO: GSE62564) and ArrayExpress (accession: E-MTAB-8248). The relative cell abundances were estimated using single-sample gene set enrichment analysis (ssGSEA) with the R gene set variation analysis (GSVA) package. We performed Cox regression analyses to identify marker genes indicating cell subsets and combined these with prognostically relevant clinical factors to develop a new prognostic model. Data from the E-MTAB-8248 cohort verified the predictive accuracy of the prognostic model. Single-cell RNA-seq data were analyzed by using the R Seurat package. Multivariate survival analysis for each gene, using clinical characteristics as cofactors, identified 34 prognostic genes that showed a significant correlation with both event-free survival (EFS) and overall survival (OS) (log-rank test, P value < 0.05). The pathway enrichment analysis revealed that these prognostic genes were highly enriched in the marker genes of NB cells with mesenchymal features and protein translation. Ultimately, USP39, RPL8, IL1RAPL1, MAST4, CSRP2, ATP5E, International Neuroblastoma Staging System (INSS) stage, age, and MYCN status were selected to build an optimized Cox model for NB risk stratification. These samples were divided into two groups using the median of the risk score as a cutoff. The prognosis of samples in the poor prognosis group (PP) was significantly worse than that of samples in the good prognosis group (GP) (log-rank test, P value < 0.0001, median EFS: 640.5 vs. 2247 days, median OS: 1279.5 vs. 2519 days). The risk model was also regarded as a prognostic indicator independent of MYCN status, age, and stage. Finally, through scRNA-seq data, we found that as an important prognostic marker, USP39 might participate in the regulation of RNA splicing in NB. Our study established a multivariate Cox model based on gene signatures and clinical characteristics to better predict the prognosis of NB and revealed that mesenchymal signature genes of NB cells, especially USP39, were more abundant in patients with a poor prognosis than in those with a good prognosis. KEY MESSAGES: Our study established a multivariate Cox model based on gene signatures and clinical characteristics to better predict the prognosis of NB and revealed that mesenchymal signature genes of NB cells, especially USP39, were more abundant in patients with a poor prognosis than in those with a good prognosis. USP39, RPL8, IL1RAPL1, MAST4, CSRP2, ATP5E, International Neuroblastoma Staging System (INSS) stage, age, and MYCN status were selected to build an optimized Cox model for NB risk stratification. These samples were divided into two groups using the median of the risk score as a cutoff. The prognosis of samples in the poor prognosis group (PP) was significantly worse than that of samples in the good prognosis group (GP). Finally, through scRNA-seq data, we found that as an important prognostic marker, USP39 might participate in the regulation of RNA splicing in NB.