Geographical identification of Chinese wine based on chemometrics combined with mineral elements, volatile components and untargeted metabonomics

Abstract

Identifying the geographic origin of a wine is of great importance, as origin fakery is commonplace in the wine industry. This study analyzed the mineral elements, volatile components, and metabolites in wine using inductively coupled plasma-mass spectrometry, headspace solid phase microextraction gas chromatography-mass spectrometry, and ultra-high-performance liquid chromatography-quadrupole-exactive orbitrap mass spectrometry. The most critical variables (5 mineral elements, 13 volatile components, and 51 metabolites) for wine origin classification were selected via principal component analysis and orthogonal partial least squares discriminant analysis. Subsequently, three algorithms-K-nearest neighbors, support vector machine, and random forest -were used to model single and fused datasets for origin identification. These results indicated that fused datasets, based on feature variables (mineral elements, volatile components, and metabolites), achieved the best performance, with predictive rates of 100% for all three algorithms. This study demonstrates the effectiveness of a multi-source data fusion strategy for authenticity identification of Chinese wine.

Keywords: Geographical origin; Machine learning algorithms; Mineral element; Untargeted metabolomics; Volatile component; Wine.

Fig. 1

Heat map (A) of the differential distribution of 12 mineral elements in wine samples (geographic sources: BHW, Bohai Bay; HLS, Ningxia Helanshan Dongluo; and HZ, Huai Zhuo Basin); score plots of the principal coordinate analysis (PCA) model (B) and the orthogonal partial least squares discriminant analysis (OPLS-DA) model (C) for the mineral elements of wines; results of cross-validation of the 200 calculations using the permutation test (D); variable importance in projection (VIP) plots (E).

Fig. 2

Radar plots of volatile constituent species (A) and mass concentration (B) in wine samples from three production regions in China; plots of scores from the principal coordinate analysis (PCA) model (C) and the orthogonal partial least squares discriminant analysis (OPLS-DA) model (D) for volatile constituents; results of cross-validation of 200 calculations using the permutation test (E); and plots of the variable importance in projection (VIP) (F).

Fig. 3

Ultra-perofmance liquid chromatography-quadrupole time-of-flight-mass spectrometry (UHPLC-Q-Exactive Orbitrap MS) total ion chromatograms of Cabernet Sauvignon wines from different production areas in electrospray ionization-positive (ESI+) and ESI−negative (ESI−) modes (A and B), principal coordinate analysis (PCA) model score plots (C and D), orthogonal partial least squares discriminant analysis (OPLS-DA) model score plots (E and F), and cross-validation results of 200 calculations using the substitution test (G and H).

Fig. 4

Type composition of significantly different metabolites in wines from three regions.

Fig. 5

Heatmap visualization of electrospray ionization-positive (ESI+) mode (A) and ESI-negative (ESI−) mode. (B) Differential metabolites in wine samples and boxplots of the relative amounts of each group of compounds in ESI+ mode (C—F) and ESI− mode (G–J) in wine samples from each region.