Assessing Opportunities for Statistical Analysis of Quantitative 13 C-NMR Data; Application to Classification and Component Analysis of Honey

HOIJEMBERG, PABLO ARIEL; HA, CHRISTINE; LELINHO, JOSEPH G.; PELCZER, ISTVÁN

Charlotte, NC

Conferencia; Second Annual Practical Applications of NMR in Industry Conference (PANIC); 2014

PANIC Organizing Committee

Analysis of complex materials, such as sugar mixtures of honey, can be greatly helped by using quantitative 13C-NMR spectroscopy. 13C-NMR spectra have a superb dispersion, are closely correlated with the actual chemical structure, show minimal overlap due to the singlet nature of the resonances. The sensitivity limitations can be highly overcome by using a 13C-detecion optimized cryoprobe. Honey is a complex natural product, which consists of multitude of closely related sugar components. The actual composition of these will provide a characteristic fingerprint of the honey related to its origin, treatment, conditioning, possible contamination and adulteration. Therefore credible and trustworthy characterization of honey samples is essential. Mixture analysis most often relies on statistical comparison of the input data leading to diagnostic clustering and information about individual components. Next to PCA and supervised discriminant analysis methods (PLS-DA and O-PLS-DA) component analysis can also be assisted by calculating STOCSY [1] traces starting from well-selected driver peaks. While such analysis is commonplace for 1H-NMR spectra, 13C-NMR has not been used for this purpose almost at all so far. Sensitivity and quantitative nature are two major obvious concerns; both of them can be largely avoided by using cryoprobes. One additional significant issue is consistency of chemical shifts or peak alignment, which has to be very accurate given the small linewidth relative to the possible chemical shift fluctuations. The alternative is to use either well-designed strategic binning or virtual binning through artificially reducing spectral resolution in general. A relatively novel approach is to deconstruct the original spectrum to individual lines (in our case using careful curve fit deconvolution) and pass the digital set to off-line analysis software (Excel, Matlab, etc.), although the issue of alignment yet remains to be resolved. Each of these approaches has their own benefits and drawbacks. We have taken quantitative 13C-NMR of about a dozen different honey samples from a local apiary and subjected this set, as a test case, to a variety of multivariate statistical analyses and STOCSY component extraction. In the poster we present the comparative assessment of the results.