Chemometrics in the analysis of complex volatile chemical mixtures

CARMEN ROSSINI; MARÍA SOL BALBUENA; WALTER M FARINA; ANDRÉS GONZÁLEZ

Congreso; III Congress of the Latin American Association of Chemical Ecology (ALAEQ).; 2014

ALAEQ

Introduction: GC-MS analysis of complex mixtures with underlying biological questions involves the identification and peak-integration of individual compounds for subsequent multivariant statistical analysis. This strategy is time consuming, and often implies the subjective exclusion of minor components. Raw GC-MS data can also be analyzed by data-driven information extraction, an approach known as chemometrics. Here we present a comparative analysis of both methods, using cuticular hydrocarbons (CHCs) from honeybees. Material and methods: CHCs were extracted from hive and forager bees after visiting artificial feeders with different sucrose-solution treatments. GC-MS data (19 samples/treatment) were first analyzed by manual integration of 48 peaks, with subsequent Principal Component Analysis (PCA) of net CHC amounts. For the chemometric approach, GC-MS files were converted to Network Common Data Form (NetCDF) and imported into MZmine2.10, defining parameters for mass detection, chromatogram building and peak deconvolution. Chromatograms normalized for retention times and peak areas were then aligned (RANSAC tool), generating a matrix of 1235x76 that was subjected to PCA. Results: Both methods produced identical results, with a clear separation of hive bees from forager bees with respect to their CHCs, and no separation of forager bees according to feeding treatment. Furthermore, eight out of ten most relevant CHCs singled out by their PCA loadings were the same in both methods. While the conventional method took approximately one week for processing the experimental data set, the chemometric approach took approximately 30 min for file conversion and additional 30 min to produce the first PCA results. Conclusions: While expertise is the key factor for identifying biologically relevant compounds, we here show that data-driven methods are efficient and reliable tools when complex volatile profiles are under investigation. A quantitative overview of such data sets can be achieved within one day, and the main compounds to focus on for identification can be efficiently selected.