Combining Chemometrics and Spectroscopy for Determination of Phenolic Molecules in Pinot Noir Wines

Abstract

Numerous compounds effect the overall flavor profile and stability of wines. For red wines, the polyphenolic family of compounds contribute to features such as astringency and color. A promising approach to phenolic determination involves using spectroscopy combined with chemometric analysis tools. A dataset consisting of 155 young Pinot noir wines from New Zealand were measured using ultraviolet-visible (UV-Vis) and mid-infrared (MIR) spectroscopy as well as highperformance liquid chromatography (HPLC). Partial least squares (PLS) regression models were developed to predict HPLC-derived concentrations of phenolic molecules using the UV-Vis and MIR datasets. The PLS models of the UV-Vis spectra resulted in robust prediction of total phenolic content as well as monomeric phenolic compounds such as malvidin-3-glucoside, catechin, and caftaric acid. Of the 22 phenolic compounds modeled, only 3 PLS models resulted in r 2 values less than 0.6, while 15 models had r 2 values greater than 0.8. The findings indicate that UV-Vis spectroscopy can predict a wide range of phenolic molecules in a fraction of the time and cost required for HPLC. MIR spectroscopy resulted in adequate models of total phenolic content and tannins but was significantly worse than UV-Vis for predicting flavonols, anthocyanins, and cinnamic acids. The prediction performance of MIR was likely harmed by noise resulting from variability in the laboratory temperature. MIR may be useful in on-line applications due to faster acquisition times, but otherwise, UV-Vis appears to be strictly superior to MIR for phenolic determination. Partial least squares discriminant analysis (PLS-DA) models of the UV-Vis, MIR, and HPLC data were used to predict which wines originated from the Marlborough, Martinborough, and Central Otago regions. The three datasets resulted in 67-80% classification accuracies, with HPLC having the best performance. Resveratrol was identified as the phenolic compound that was most indicative of regionality. The final study reported in this thesis investigated the potential of Raman spectroscopy for analyzing Pinot noir wines. Several practical issues with Raman were identified, including poor signal-to-noise ratio, long measurement times, and sample overheating. Overall, Raman was significantly worse than relevant alternative spectroscopic methods.