An undeniable increase in consumption and sales of plant food supplements (PFS), especially in the western world, is observed. High consumer demands increased the vulnerability of these products to malpractices such as adulteration and fraud, which further led to opening up various illicit channels for their purchase. Therefore, some of these products may be ‘suspicious’ and a risk for consumers health.
This study aims at developing a screening approach for detection of four regulated plants commonly encountered in PFS claiming weight-loss as indication. However, PFS usually are composed of mixtures of different plants resulting into a not so straightforward approach for their detection. In order to tackle this problem, a feasibility study based on a combination of chemometrics and multidimensional fingerprints recorded using chromatography was carried out.
Chromatographic fingerprints were developed using ultra-high performance liquid chromatography with diode array detection. The obtained multidimensional fingerprint data for a herbal reference standard (absorbance × wavelengths × time points) were used for selection of orthogonal wavelengths by conducting a correlation analysis. After preparation of triturations, recording of the fingerprints and creation of datasets (corrected for shifts in retention times using correlation optimized warping), partial least squares – discriminant analysis was applied to construct classification models. Two approaches for modelling were explored: (a) binary modelling and (b) multiclass modelling. The correct classification rate % (ccr%) for cross validation, modelling and external test prediction were used to select the optimal model for each plant. For both binary modelling and multiclass modelling, a ccr% above 85% for external test set validation was observed for all the plants.
As a proof of concept, 12 weight-loss PFS, seized by the Belgian Federal Agency for Medicines and Health Products, were classified using the above developed models. 10 out of 12 samples were suggestively found to be positive by binary modelling whereas 11 were found positive with multiclass modelling. Despite both models having good ccr%, multiclass modelling was unable to detect multiple plants in samples. Therefore, a preference of binary models over multiclass models could be made. Furthermore, it can be concluded that the combination of chemometrics and multi-wavelength chromatographic fingerprinting is a promising approach for identifying plants using in herbal mixtures.
