A Machine Learning Analysis of TCGA Expression Data to Finding Signatures for “Normal-Like” IDH-WT Diffuse Gliomas with a Longer Survival

Abstract

Classification of primary CNS tumours is currently achieved by complementing histologic analysis with molecular information, in accordance with the WHO guidelines, and aims at providing accurate prognosis and optimal patient management. cIMPACT-NOW update 3 now recommends grading diffuse IDH-wild type astrocytomas as grade IV glioblastomas if they bear one or more of the following molecular alterations: EGFR amplification, TERT promoter mutation, and whole-chromosome 7 gain combined with chromosome 10 loss. In this reanalysis of the Cancer Genome Atlas (TCGA) glioma expression datasets, we identified 14 IDH-wt infiltrating astrocytic gliomas displaying a “normal-like (NL)” transcriptomic profile associated with a longer survival rate. Some of these tumours would be considered as GBM-equivalents with the current diagnostic algorithm. A k-nearest neighbors model was used to identify 3-gene signatures able to identify NL IDH-WT gliomas. Genes such as C5AR1 (complement receptor) SLC32A1 (vesicular gamma-aminobutyric acid transporter), and SMIM10L2A (long non-coding RNA) were overrepresented in these signatures which were validated further using the Chinese Glioma Genome and Ivy Glioblastoma Atlases. They showed high discriminative power and correlation with survival. This finding could lead to the validation of an immunohistochemical or PCR test which would facilitate classification of IDH-WT astrocytomas with unclear histological grading. Furthermore, associated signaling pathways might represent novel treatment targets for aggressive tumours.

LEARNING OBJECTIVES

This presentation will enable the learner to:

1. 1. Reconsider recent updates in the WHO classification of infiltrating gliomas.

2. 2. Discuss advanced bioinformatics profiling of the brain cancer transcriptome.