Raman-based machine-learning platform reveals unique metabolic differences between IDHmut and IDHwt glioma.

Area of Science:

• Biomedical Optics
• Molecular Pathology
• Computational Biology

Background:

• Formalin-fixed, paraffin-embedded (FFPE) tissues are crucial for cancer diagnosis and biobanking.
• Background noise from embedding media has limited FFPE tissue use in Raman spectroscopy.
• Developing methods to analyze FFPE tissues with Raman spectroscopy is essential for unlocking their full potential.

Purpose of the Study:

• To develop a computational workflow (APOLLO) for classifying glioma subtypes using Raman spectroscopy on FFPE tissue slides.
• To identify novel Raman spectral markers for distinguishing glioma subtypes and molecular characteristics.
• To demonstrate the feasibility of label-free molecular analysis of archived FFPE tissues.

Main Methods:

• Utilized spontaneous Raman spectroscopy for molecular fingerprinting of 46 FFPE glioma samples.
• Employed machine learning (SVM, random forest) to build classifiers for tumor/non-tumor, IDH1 status, and methylation subtypes.
• Validated key Raman frequencies using stimulated Raman spectroscopy and mass spectrometry.

Main Results:

• Developed APOLLO, a workflow distinguishing tumor from non-tumor FFPE tissue based on Raman spectra.
• Identified novel Raman peaks associated with DNA and proteins in tumors.
• Successfully differentiated IDH1 mutant from wild-type gliomas, noting higher cholesterol esters in IDH1 mutant tumors.
• Achieved high accuracy in classifying clinically relevant glioma methylation subtypes (G-CIMP-high vs. G-CIMP-low).

Conclusions:

• Label-free Raman spectroscopy can effectively classify glioma subtypes from FFPE slides.
• APOLLO provides a method to extract meaningful biological information from archived FFPE tissues.
• This approach holds promise for future applications in analyzing other cancer types using FFPE samples.