Abstract
Fraud is endemic throughout the food industry and a major concern to producers and consumers. Adulteration or substitution with an inferior product may cause economic or health problems. Milk is a foodstuff that is widely consumed around the world; however, it is also one of the top fraudulent foods. Consumers are placing much greater value on authenticity and origin of their purchase; thus, it is important to establish methods to authenticate food. New Zealand milk is regarded to be of prime quality and hence particularly susceptible to attempts to pass of milk from other sources as being from New Zealand. Therefore, developing effective tools for verifying the geographic origin of milk is particularly important in the New Zealand context.
The aim of this work is to explore the use of metabolomic profiling as a tool for the discrimination of raw milk samples from New Zealand, Sri Lanka and China (Inner Mongolia), according to their geographic origins. Multiple advanced analytical techniques were combined, including 1H-NMR (nuclear magnetic resonance) and Q-Tof-MS (quadrupole time-of-flight mass spectrometry) coupled with UHPLC (ultra-high-performance liquid chromatography). Over two hundred low molecular weight compounds were detected by untargeted omics-based techniques. Following proper data pre-processing, multivariate analyses such as PCA (principal component analysis) and PLS-DA (partial least squares-discriminative analysis) were applied to explore groupings in the data.
In Part I, the milk from the different countries could be completely separated while samples from different regions of New Zealand behaved similarly. The candidate biomarkers for distinguishing samples from New Zealand and Sri Lanka, or New Zealand and China/Inner Mongolia were selected, such as amino acids, organic acids or their derivatives, according to the VIP values of PLS-DA and p-values of Wilcoxon's Rank-Sum test.
In order to select the robust markers that were geographic origin-based, some factors that influence the milk metabolite profile were investigated. These factors were divided into two types, raw milk-based (Part II), e.g. raw milk from different lactation stages, and milk process-based (Part III), including pasteurization, drying processes and storage. Multivariate data analysis, PCA, PLS-DA or PLSR (partial least squares regression), were used to select the metabolites that were strongly influenced by these processes. These metabolites were not suitable for milk geographic origin discriminant analysis as they were most likely to be influenced by some other interference factors.
In Part IV, the candidate markers, selected in Part I, were further screened according to the results obtained from Part II and Part III. Some of the candidate markers were removed considering their instability and were not used for discriminant modelling. The metabolites or ions acetyl carbohydrate, citrate, lactate detected by 1H-NMR and creatinine, valporic acid, 2-methylbutyroylcarnitine, thiocyanate, 397.0020 ([M-H]-), 255.0887 ([M-H]-), 356.9826 ([M-H]-), N-acetyl-D-glucosamine, quinic acid and L-glutamic acid 5-phosphate detected by UHPLC-QToF/MS were used to distinguish the raw milk samples from New Zealand and Sri Lanka. In addition, the metabolites or ions N-acetyl-D-glucosamine, riboflavin, N(α)-benzyloxycarbonyl-L-leucine, L-glutamic acid, thiabendazole, thiocyanate, 227.9973 ([M-H]-) and 356.9826 ([M-H]-) monitored by UHPLC-QToF/MS were selected and used to differentiate the raw milk samples from New Zealand and China/Inner Mongolia. The model was also optimized further and used for commercial milk powder authentication based on their raw milk origins.
In conclusion, untargeted metabolomic analysis by NMR and MS-based techniques, combined with multivariate data analysis, proved to be effective tools for milk traceability analysis. By considering the possible interfering factors, the discriminant PLS-DA model built on the basis of the markers was successfully applied to distinguish the origins of commercial milk powder samples. However, some metabolites could not be identified or accurately quantified due to the limitations of the untargeted analysis. Future research could be carried out in improving the identity of the selected biomarkers with the aid of some other techniques or pure standards.