Raman spectroscopy with multivariate analysis:characterization and classification of biomarkers of pancreatic cancer

Abstract

Aims: Pancreatic cancer remains an insidious condition for which early diagnosis would improve prognosis. Biospectroscopy has been proposed as a reagent-free, non-destructive approach towards screening in biofluids1. Raman spectroscopy generates bio-fingerprint spectra. Within a computational framework, such spectra may objectively classify presence or absence of underlying disease. Methods: A total of n = 13 plasma samples (10 from patients with pancreatic cancer and 3 controls) and n = 15 urine samples (12 pancreatic cancer and 3 controls) were obtained following ethical approval. Ten Raman spectra were acquired using 10% laser power from 200 \ensuremathμl of each sample deposited and dried on slides. Pre-processed spectra were analysed by principal component analysis using MATLAB software2. Vector loadings analysis determined principle variables (or peaks) responsible for discriminating cancer vs. controls. Results: Raman spectra were readily derived. Identifying differences between spectra from different categories is challenging3, so computational algorithms are required. Each spectrum is reduced to a single point in n-dimensional hyperspace, allowing comparison between different spectra. Following vector loading analysis, several wavenumbers appear important, namely, variables in the 1720?1755 cm?1 region and the 900?1000 cm?1 spectral regions. Conclusions: The variables between 1755?1720 cm?1 are associated with C = O stretching vibrations of aldehydes and lipids. Wavenumbers between 900?1000 cm?1 represent spectral regions of DNA/RNA vibrations. Spectral fingerprints combined with computational analysis offers exciting opportunities in biomarker development, screening and diagnosis. For conditions such as pancreatic cancer, the possibility of inexpensive, point-of-care testing has enormous possibilities for monitoring at-risk individuals.