Discrimination of geographical origin of lentils (Lens culinaris Medik.) using 1H NMR fingerprinting and multivariate statistical analysis

Lentil (Lens culinaris Medik.) is the fourth most important pulse crop in the world after bean (Phaseolus vulgaris L.), pea (Pisum sativum L.), and chickpea (Cicer arietinum L.). Canada is the world's largest exporter of lentils, while in Italy lentils are a minor legume and can be found in restricted areas. However, Italian lentils present unique and characteristic qualities giving them a higher value, so that many of them have obtained international and national marks linked to their geographical origins, such as "protected geographical indication" (PGI), "traditional food products" (PAT) and Slow Food Presidium. For these reasons, there is a growing demand for analytical methods able to certify the declared geographical origin of lentils, in order to protect consumers and producers from fraud and unfair competition. In the present work, non-targeted 1H-NMR fingerprinting, in combination with different multivariate statistical analysis techniques, was used to classify lentils according to their geographical origin. In particular, 85 lentil samples from two different countries, i.e. Italy and Canada, were collected from retail markets and analysed by using an optimized 1H-NMR protocol. Principal component analysis showed partial grouping of samples on the basis of origin with overlapping zones. Therefore, two class-modeling techniques such as Soft Independent Modelling of Class Analogy (SIMCA) and UNEQual dispersed classes (UNEQ) and three discriminant techniques, such as k - Nearest Neighbor (k-NN), Linear Discriminant Analysis (LDA), Partial Least Squares - Discriminant Analysis (PLS-DA), were used and the performances of the resulting models were compared. The best average recognition and cross-validation prediction abilities, 100% and 93.7% respectively, were obtained by the LDA model, performed on a set of 20 principal components previously selected by a stepwise decorrelation procedure. The other models, except the SIMCA one, also showed good performances (above 90%). All tested statistical models were validated by evaluating the prediction abilities on an external set of lentil samples. LDA model showed the best results with an external prediction ability of 100%, but also the other models showed remarkable performances (above or near 90%). These findings demonstrated the suitability of the methods developed to discriminate geographical origin of lentils and confirmed the applicability of the NMR data, in combination with chemometrics, to solve geographic origin issues of foodstuffs.