Genomics and proteomics to determine novel molecular subtypes and predict the response to immunotherapy and the effect of bevacizumab in glioblastoma.

ABSTRACT

Glioblastoma (GBM) is a highly aggressive, infiltrative malignancy that cannot be completely cured by current treatment modalities, and therefore requires more precise molecular subtype signatures to predict treatment response for personalized precision therapy. Expression subtypes of GBM samples from the Cancer Genome Atlas (TCGA) were identified using BayesNM and compared with existing molecular subtypes of GBM. Biological features of the subtypes were determined by single-sample gene set enrichment analysis. Genomic and proteomic data from GBM samples were combined and Genomic Identification of Significant Targets in Cancer analysis was used to screen genes with recurrent somatic copy-number alterations phenomenon. The immune environment among subtypes was compared by assessing the expression of immune molecules and the infiltration of immune cells. Molecular subtypes adapted to immunotherapy were identified based on Tumor Immune Dysfunction and Exclusion (TIDE) score. Finally, least absolute shrinkage and selection operator (LASSO) logistic regression was performed on the expression profiles of S2, S3 and S4 in TCGA-GBM and RPPA to determine the respective corresponding best predictive model. Four novel molecular subtypes were classified. Specifically, S1 exhibited a low proliferative profile; S2 exhibited the profile of high proliferation, IDH1 mutation, TP53 mutation and deletion; S3 was characterized by high immune scores, innate immunity and adaptive immune infiltration scores, with the lowest TIDE score and was most likely to benefit from immunotherapy; S4 was characterized by high proliferation, EGFR amplification, and high protein abundance, and was the most suitable subtype for bevacizumab. LASSO analysis constructed the best prediction model composed of 13 genes in S2 with an accuracy of 96.7%, and the prediction model consisting of 17 genes in S3 with an accuracy of 86.7%, and screened 14 genes as components of the best prediction model in S4 with an accuracy of 93%. To conclude, our study classified reproducible and robust molecular subtypes of GBM, and these findings might contribute to the identification of patients responding to immunotherapy, thereby improving GBM prognosis.