Untargeted Metabolomics of Plant Samples using HPLC-DAD and Gaussian Mixture Models

Abstract

Abstract/SummaryO_ST_ABSPremiseC_ST_ABSPlants produce millions of different chemical compounds, contributing greatly to their physiology and evolutionary trajectories. Most untargeted metabolomic methods are inaccessible, either due to upfront instrument costs or intensive technical training. More accessible methods using diode array detectors often only utilize a few wavelengths, preventing high-throughput observation of total metabolic diversity. MethodsLeaves from the genera Betula, Magnolia, Rosa, and Viburnum were collected, dried and ground, extracted, and analyzed by HPLC-DAD. Chromatographic data was then processed in a curated R pipeline, and resulting resolved peaks were clustered by absorbance spectra using Gaussian Finite Mixture Models (GMMs). To assess clustering, GMM was compared to a more traditional linear discriminant analysis (LDA) method, with clusters identified through literature searches. ResultsSignificant associations between the abundances of chemical classes and whole-metabolome alpha and beta diversity indices were recovered. In general, GMMs performed better than other classification methods like LDA, especially between classes that share common features like non-flavonoid phenolics and flavonoids. DiscussionWe show that our method can easily extract relevant class-level diversity of metabolite profiles among closely related species, genotypes, and ecotypes. Regardless of underlying research question, our method extends the usage of DAD beyond restricted targeted analyses and increases the accessibility of untargeted metabolomics.