Determination of a robust metabolic barcoding model for chemotaxonomy in Aizoaceae species : expanding morphological and genetic understanding

Doctoral Thesis

2016

Permanent link to this Item
Authors
Journal Title
Link to Journal
Journal ISSN
Volume Title
Publisher
Publisher

University of Cape Town

License
Series
Abstract
The use of metabolic fingerprints as taxonomic markers is becoming more common. Many studies have found that by comparing the vast metabolic fingerprints of closely related species to each other, secondary metabolites tend to be unique to the samples of individual species and are identified in clustering algorithms as the variables responsible for species-specific clustering. A holistic approach to metabolic fingerprinting was thus employed to assess the stability of various metabolomic markers and finally to distinguish taxonomically difficult Aizoaceae species. Many secondary metabolites are not constitutively produced. Because at least some Aizoaceae species facultatively use crassulacean acid metabolism (CAM), there was a potentially interesting molecular switch that could be monitored for transitions in metabolic fingerprints. In order to contextualise the changes in carbon uptake, 20 different climate, nutrient, physiological, and other variables were monitored over the course of 12 months to build up a store of species-specific information to use in model optimisation across 5 Aizoaceae species (Galenia africana, Aridaria noctiora, Carpobrotus edulis, Ruschia robusta, and Tetragonia fruticosa) using two Crassulaceae species as CAM controls (Cotyledon orbiculata and Tylecodon wallichii ). Metabolic fingerprints of the leaves of various Aizoaceae species were generated using LC/TOFMS, following which Principal Components Analysis (PCA) was used to identify the LC-MS ions which distinguished the species from each other, or in statistical terms, were informative. Once isolated, this subset of informative data was established as metabolic barcodes for the identification of the study species. A machine learning algorithm, Random Forest, was used to build a classification model based on the metabolic barcodes which was then trained on various trends from the factors monitored over the year. The use of these trends in the development of a classification model based on metabolic barcodes resulted in a highly robust classification model for species identification. Clustering analysis of a subset of ions which corresponded to compounds previously isolated from Aizoaceae species did not show species-specific clustering and was inevitably biased by compounds from species with a greater number of studies focusing on compound isolation. Ideally, this model should be expanded to include other species from the Aizoaceae family to further check robustness of the model. Application of this model to these and other species could facilitate not only species identification and distribution, but also the identification of novel chemical constructs associated with particular species.
Description
Keywords

Reference:

Collections