Metabolic Insight into Glioma Heterogeneity: Mapping Whole Exome Sequencing to In Vivo Imaging with Stereotactic...

Area of Science:

• Neuro-oncology
• Medical imaging
• Genomics

Background:

• Intratumoral heterogeneity (ITH) in glioma complicates diagnosis and treatment due to diverse metabolic profiles linked to genomic alterations.
• Multiparametric MRI characterizes ITH spatially and functionally but cannot directly assess underlying metabolic activities.
• A gap exists in integrating pathology, genomic data, and metabolic insights for glioma characterization.

Purpose of the Study:

• To demonstrate machine learning's potential in predicting cellular and molecular tumor characteristics in glioma.
• To correlate noninvasive multiparametric MRI data with genomic alterations in glioma.
• To explore the link between glioma genomics, metabolism, and imaging phenotypes.

Main Methods:

• Combined stereotactic biopsy and open-craniotomy for sample collection in ten treatment-naïve glioma patients.
• Performed voxel-wise analysis of multiparametric MRI (T1w, T1w-CE, T2w, T2w-FLAIR, DWI) and whole-exome sequencing.
• Utilized regression-based Generalized Additive Models (GAM) and receiver operating characteristic (ROC) analyses to predict genomic alterations (IDH1, TP53, EGFR, PIK3CA, NF1).

Main Results:

• Achieved a mean Area Under the Curve (AUC) of 0.75 ± 0.11 across five gene targets and 31 MR contrast combinations.
• Individual AUCs reached 0.96 for IDH1 and TP53 with T2w-FLAIR/ADC, and 0.99 for EGFR with T2w/ADC.
• Demonstrated the potential to predict exome-wide mutation events from noninvasive in vivo imaging.

Conclusions:

• Noninvasive imaging combined with deep learning can predict glioma's genomic mutation events.
• The identified genomic alterations (IDH1, TP53, EGFR, PIK3CA, NF1) are crucial in glioma metabolic pathways and heterogeneity.
• This approach offers a pathway to indirectly assess glioma's metabolic landscape and refine targeted therapies by addressing genomic heterogeneity.