Detection of Edible Plant Oil Adulteration by Triacylglycerol Profiles Using an Atmospheric Pressure Chemical Ionization Source and MS³ Ion Trap Mass Spectrometry

Abstract

An adulteration of high‐price oil has been an important concern in agri‐food fraud problems. Triacylglycerols are the main components of edible plant oils, which play an important role as target for adulteration detections. The objective is to evaluate the utility of triacylglycerol profiles in the detection of adulteration in high priced oils. An effective method is established to detect adulteration of high priced oils based on triacylglycerol profiles and chemometrics. Triacylglycerol profiles of edible oils are analyzed by an APCI‐MS3‐IT‐MS. All triacylglycerol compounds are quantified by normalization of the chromatographic peak area in selected reaction monitoring (SRM) mode based on MS3 fragment ion pairs, which are used to eliminate interference from the isobaric species of triacylglycerol. The results of the hierarchical cluster analysis (HCA) and random forest (RF) classification algorithm indicated that peanut, soybean, sesame, sunflower seed, and linseed oils are completely classified into five groups based on their triacylglycerol profiles. Finally, a recursive support vector machine (R‐SVM) discriminant model is established, which can successfully identify adulteration of high priced oil with cheaper edible oils at a concentration of as low as 4% with an accuracy of 93.7%. Practical Applications: The triacylglycerol profiles of edible plant oils are analyzed by RPLC‐APCI‐IT‐MS. The SRM scan mode based on MS3 fragment ion pairs can eliminate interference of triacylglycerol isomers. The discriminant model for adulterated high‐price oils is established by R‐SVM. HCA, and RF can classify the five kinds of edible plant oil. The triacylglycerol of five edible plant oils is analyzed by RPLC‐APCI‐MS3‐IT‐MS. The triacylglycerol data matrix is submitted using unsupervised (HCA) and supervised (RF) chemometric methods to classify five kinds of edible plant oil. A discriminant model established with the R‐SVM can be used to identify adulterated linseed oil samples (≥4%) with an accuracy of 93.7%.