Machine learning-based integration develops a hypoxia-derived signature for improving outcomes in glioma

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Glioma growth is often linked to a hypoxic microenvironment, but its clinical impact remains understudied.
• Hypoxia influences cellular communication within the tumor microenvironment, leading to distinct glioma subtypes.
• Two identified hypoxia-related glioma subtypes (C1 and C2) exhibit significant prognostic and molecular variations.

Purpose of the Study:

• To investigate the clinical implications of hypoxia in glioma.
• To identify and characterize hypoxia-related glioma subtypes.
• To develop a robust prognostic tool for glioma based on hypoxia-associated gene expression.

Main Methods:

• Single-cell RNA sequencing was employed to analyze cellular communication in hypoxic glioma.
• Machine learning algorithms were utilized to develop an 11-gene signature for outcome prediction.
• The 11-gene signature was validated across multiple cohorts using quantitative real-time PCR (RT-qPCR).

Main Results:

• Two distinct hypoxia-related glioma subtypes (C1 and C2) were identified, with C2 showing worse prognosis, increased immune/stromal cells, and higher immune checkpoint gene expression.
• An 11-gene signature was developed, accurately distinguishing high-risk from low-risk glioma patients and predicting overall and relapse-free survival.
• The developed risk score demonstrated superior accuracy compared to conventional clinical variables, molecular features, and 100 existing signatures.
• High-risk gliomas were associated with elevated expression of CD163, PD1, HIF1A, and PD-L1.

Conclusions:

• A hypoxia-related classification system for glioma can aid in treatment decision-making.
• The novel 11-gene signature serves as a reliable prognostic tool for glioma patients.
• Understanding hypoxia-driven tumor heterogeneity is crucial for improving glioma patient outcomes.