Correlating the 1H NMR fingerprinting of a collection of red lentil varieties with seed colour and morphology using advanced statistical analyses

This study aimed to investigate the biochemical basis of seed morphological traits in red lentils that are important for lentil producers in relation to quality, consumers' preferences and commercial value. To achieve this objective, proton Nuclear Magnetic Resonance (1H NMR) spectroscopy combined with multivariate statistical analyses was employed. A collection of 64 red lentil varieties exhibiting diversity in seed colour, size, weight, and cotyledon pigmentation was analysed. Aqueous extracts of the seeds were profiled using 1H NMR, and spectra were processed into bucketed variables. Partial Least Squares Regression and Multiple Linear Regression were applied to assess relationships between spectral data and continuous morphological traits: lightness (L*), chromatic indexes (a*, b*), Hundred Kernel Weight, and seed size. For categorical traits like cotyledon colour, Partial Least Squares Discriminant Analysis (PLS-DA) and binomial logistic regression were used. Variable Importance in Projection scores helped to identify key metabolite buckets significantly contributing to trait prediction. Metabolites such as leucine, fructose, and phenolic compounds were positively associated with seed size and weight, while NAD+ and short-chain fatty acids showed negative associations. Cotyledon colour classification achieved high accuracy (up to 100 %) using both PLS-DA and logistic models, with amino acids like leucine and alanine linked to yellow pigmentation and tryptophan and citrate linked to orange. Overall, the study demonstrates that 1H NMR fingerprinting, combined with rigorous statistical modelling, effectively elucidates the multivariate relationships between metabolomic profiles and key agronomic traits, providing a valuable tool for phenotypic prediction and lentil breeding.