Direct use of TXRF spectral signal for multivariate data analysis: a new strategy for food fingerprint

Total-reflection X-ray fluorescence (TXRF) spectroscopy is a well-established technique used for the elemental analysis of samples in several scientific fields. Compared to other atomic spectroscopies (ICP-OES, ICP-MS, AAS and XRF), TXRF allows fast analyses of different types of matrixes (water, biological tissues and fluids, minerals, composites, ores, etc.) using a very small amount of sample after a simple sample-preparation. These advantages make the technique very suitable for the analysis of food where reliable and fast analytical methods are crucial to quantify mineral nutrients and potentially toxic elements, to trace the food and to identify frauds. For these purposes, and specifically for food traceability, the detected elements in food are usually quantified (using different approaches which can influence the accuracy of the quantification) and then processed with multivariate statistical analysis. Following this approach, TXRF has more or less the same applicability of other atomic spectroscopies which, on the other hand, may show better limits of detection and offer a wider number of quantifiable elements, in particular trace elements. However, compared to other atomic spectroscopies like ICP-OES, ICP-MS, AAS, TXRF gives a continuum spectrum (generally in the range from 0 to 20kV) as an output, which can be used, in combination with chemometric approaches, as a fingerprint of the sample. While this approach was already applied on XRF spectra [1,2], it has never been applied on TXRF ones. Moreover, to our knowledge, XRF spectra, and in particular TXRF ones, have never been used for food fingerprinting. For this reason, in the present work we developed, tested and validated a new method for food fingerprinting and traceability using TXRF spectra. In order to fulfill this goal, a group of 24 different genotypes of beans (Phaseolus vulgaris) grown in two different places in the same region (Veneto, Italy) but at different altitudes (sea level vs 1000 m a.s.l.) were analyzed. The elemental quantitative data as well as the TXRF signals were processes by unsupervised and supervised multivariate techniques with the aim at developing a classification method for geographical origin of beans according to the growing site.