Ambient mass spectrometry for rapid authentication of milk from Alpine or lowland forage

Abstract

Metabolomics approaches, such as direct analysis in real time-high resolution mass spectrometry (DART-HRMS), allow characterising many polar and non-polar compounds useful as authentication biomarkers of dairy chains. By using both a partial least squares discriminant analysis (PLS-DA) and a linear discriminant analysis (LDA), this study aimed to assess the capability of DART-HRMS, coupled with a low-level data fusion, discriminate among milk samples from lowland (silages vs. hay) and Alpine (grazing; APS) systems and identify the most informative biomarkers associated with the main dietary forage. As confirmed also by the LDA performed against the test set, DART-HRMS analysis provided an accurate discrimination of Alpine samples; meanwhile, there was a limited capacity to correctly recognise silage- vs. hay-milks. Supervised multivariate statistics followed by metabolomics hierarchical cluster analysis allowed extrapolating the most significant metabolites. Lowland milk was characterised by a pool of energetic compounds, ketoacid derivates, amines and organic acids. Seven informative DART-HRMS molecular features, mainly monoacylglycerols, could strongly explain the metabolomic variation of Alpine grazing milk and contributed to its classification. The misclassification between the two lowland groups confirmed that the intensive dairy systems would be characterised by a small variation in milk composition.

Figure 1

Flow chart of the experimental design and statistical analysis of the (+ / −) DART-HRMS metabolites. After DART-HRMS data pre-processing (TIC normalisation, signal alignment and signal filtering), the four datasets (two dilutions per two ion modes) were submitted to low-level data fusion and the merged dataset was randomly separated into a training (n = 63) and a test (n = 25) set. A partial least squares discriminant analysis (PLS-DA) was performed on the training set and the outcomes were plotted in a scatter gram (see Fig. 2). A hierarchical cluster analysis (HCA) was performed on the 25 selected ions (coefficient > 30) by Pearson distance criterion and generating a heat map (see Fig. 3). The 25 selected ions were used to build a linear discriminant analysis (LDA) model that was validated on the test set (see Table 4).

Figure 2

PLS-DA scores plot based on (+ / −) DART-HRMS metabolites on the train set (n = 63). Ninety-five percent ellipses confidence intervals (0.95-CI) are drawn around each centroid of groupings. MMS, mix crop/maize silages (blue circles); HAY, permanent meadow and lucerne hays (green circles); APS, Alpine pasture (red circles).

Figure 3

Heatmap obtained by hierarchical clustering analysis (HCA) of the selected milk (+ /−) DART-HRMS ions. The red (positive) and blue (negative) colour scales indicate the degree of correlation between metabolic ions and feeding system; the two shorter Pearson’s distance-tree clusters among the feeding systems (columns) and metabolites (rows) are represented by the branch height (the lower a node is vertical, the more similar its subtree is).