Stable isotope resolved metabolomics classification of prostate cancer cells using hyperpolarized NMR data

Abstract

Metabolic fingerprinting is a strong tool for characterization of biological phenotypes. Classification with machine learning is a critical component in the discrimination of molecular determinants. Cellular activity can be traced using stable isotope labelling of metabolites from which information on cellular pathways may be obtained. Nuclear magnetic resonance (NMR) spectroscopy is, due to its ability to trace labelling in specific atom positions, a method of choice for such metabolic activity measurements. In this study, we used hyperpolarization in the form of dissolution Dynamic Nuclear Polarization (dDNP) NMR to measure signal enhanced isotope labelled metabolites reporting on pathway activity from four different prostate cancer cell lines. The spectra have a high signal-to-noise, with less than 30 signals reporting on 10 metabolic reactions. This allows easy extraction and straightforward interpretation of spectral data. Four metabolite signals selected using a Random Forest algorithm allowed a classification with Support Vector Machines between aggressive and indolent cancer cells with 96.9% accuracy, -corresponding to 31 out of 32 samples. This demonstrates that the information contained in the few features measured with dDNP NMR, is sufficient and robust for performing binary classification based on the metabolic activity of cultured prostate cancer cells.