Identifying putative key metabolites from fingerprinting metabolomics for the authentication of rice origin: A case study of Sengcu rice

• Published in: Journal of chemometrics 2022-12, Vol.36 (12), p.n/a
• Main Authors: Nguyen, Hoa Quynh, Tran‐Lam, Thanh‐Thien, Nguyen, Tung Ngoc, Quan, Thuy Cam, Dao, Yen Hai, Le, Giang Truong
• Format: Article
• Language: English
• Subjects: Authentication; Classification; Cluster analysis; Clustering; Discriminant analysis; Fingerprinting; Food; food authentication; Fraud; metabolite fingerprinting; Metabolites; partial least squares‐discriminant analysis; principal component analysis; Principal components analysis; Rice; UHPLC‐Q‐Orbitrap‐HRMS
• ISSN: 0886-9383; 1099-128X
• DOI: 10.1002/cem.3456

The expanding scale and nature of rice fraud in the global food system has caused major economic and human health concerns. Herein, an untargeted metabolomics approach was utilized for the discrimination between authentic and commercial Sengcu rice, a local specialty cultivated by terraced farming in northern Vietnam. A total of 8398 positive and 5250 negative mode compounds were introduced to multivariate analyses for the construction of classification models. Both principal component analysis and partial least squares‐discriminant analysis (PLS‐DA) clearly distinguished between authentic and commercial Sengcu rice. The optimized PLS‐DA models indicated that five positive (DMG, RSA, RCA, PAL, and BOSe) and six negative mode variables (PXP, RXP, TDHP, ISS, MXP, and RGB) was sufficient for validated model discrimination with a classification error rate less than 1.13 × 10−4 determined from repeated k‐fold cross validation. These putative signature metabolites clearly separated authentic and commercial Sengcu rice in the hierarchical clustering models. In addition, the isolated metabolite TDHP also reflected the difference in cultivation practices between authentic and commercial Sengcu rice. Overall, we have proposed an effective method for the identification of key metabolites from fingerprinting metabolomics, and it could serve as a fundamental approach for other in‐depth food authentication studies. There were 8398 ESI+ and 5250 ESI− metabolites employed for authenticating Sengcu rice. Sparse PLS‐DA models containing 5 ESI+ and 6 ESI− metabolites were optimized. Classification error rates of the sparse PLS‐DA models were less than 1.13 × 10−4. Hierarchical clustering clearly separated authentic from non‐authentic Sengcu rice.