Multi-dimensional omics characterization in glioblastoma identifies the purity-associated pattern and prognostic gene signatures

Abstract

Background: The presence of tumor-associated stroma and tumor-infiltrated immune cells have been largely reported across glioblastomas. Tumor purity, defined as the proportion of tumor cells in the tumor, was associated with the genomic and clinicopathologic features of the tumor and may alter the interpretation of glioblastoma biology.

Methods: We use an integrative approach to infer tumor purity based on multi-omic data and comprehensively evaluate the impact of tumor purity on glioblastoma (GBM) prognosis, genomic profiling, and the immune microenvironment in the Cancer Genome Atlas Consortium (TCGA) cohort.

Results: We found that low tumor purity was significantly associated with reduced survival time. Additionally, we established a purity-relevant 5-gene signature that was an independent prognostic biomarker and validated it in the TCGA, CGGA and GSE4412 cohort. Moreover, we correlated tumor purity with genomic characteristics and tumor microenvironment. We identified that gamma delta T cells in glioblastoma microenvironment were positively correlated with purity and served as a marker for favorable prognosis, which was validated in both TCGA and CGGA dataset.

Conclusions: We observe the potential confounding effects of tumor purity on GBM clinical and molecular information interpretation. GBM microenvironment could be purity-dependent, which provides new insights into the clinical implications of glioblastoma.

Keywords: Glioblastoma; Tumor heterogeneity; Tumor immunity; Tumor microenvironment; Tumor purity.

Fig. 1

a The workflow of this study. b Heatmap of clinical and molecular characteristics of glioblastoma patients in TCGA-GBM cohort (n = 583). c The data distribution of tumor purity estimates. d Correlations (Spearman’s Rho) between tumor purity estimates inferred by different methods

Fig. 2

a Boxplots showing comparisons between tumor purity (CPE scores) between transcriptome molecular subtypes. For each comparison, data were analyzed using student’s t-test or one-way ANOVA. Box plot center, box, and whiskers correspond to the median, IQR and 1.5xIQR (interquartile range), respectively. b Kaplan–Meier curves for overall survival according to tumor purity. c Workflow of construction of 5-gene purity-associated signature. d Kaplan–Meier curves for overall survival devided by risk score in TCGA-GBM dataset. e Risk score is an independent prognostic factor in TCGA-GBM dataset. CL classical, ME mesenchymal, NE neural, PN proneural

Fig. 3

a GO enrichment analysis revealed enrichment of immune-related pathways in low purity samples. b GSEA enrichment analysis revealed enrichment of specific KEGG pathways in low purity samples. c Differentially enriched REACTOME pathways in samples with low tumor purity (left) or high tumor purity (right). d Differences in pathway activity were analyzed using GSVA and t values were shown from a linear model

Fig. 4

a Oncoprint summarizing recurrently altered genes and their distribution in TCGA-GBM high- purity samples (upper panel) or low-purity samples (lower panel). (b, c) Correlation plot showing Spearman’s Rho between purity and mutation count or subclone numbers

Fig. 5

a The landscape of immune cell infiltrates sorted by increasing purity in TCGA-GBM RNA-seq dataset. Immune cell infiltrates were estimated by ssGSEA algorithm. b The correlation between the proportion of immune cell infiltrates and survival (upper panel) or purity estimates (lower panel) in TCGA-GBM or CGGA RNA-seq cohort. Purity values in CGGA cohort were inferred by ESTIMATE method. c Correlation plot showing Spearman’s Rho between cell types in TCGA-GBM. d Scatter plot of correlation of tumor purity and CYT (a geometric mean of GZMA and PRF1; y-axis in log2 scale). e Correlation between immune checkpoints gene expression (TCGA RNA-seq dataset) and tumor purity. Pearson’s correlation coefficients (r) are stated