“A newly developed NMR-based screening method can help the wine industry to differentiate grape variety, a significant factor for grape producers, winemakers and labelling.”
1H-Nuclear Magnetic Resonance (NMR) provides inherently quantitative and highly reproducible information and is therefore a well-adapted technique for both targeted and non-targeted screening.
From one single data set it is possible to both quantify targeted compounds and to classify samples into groups.
An NMR-based screening method has been developed for wine and its potential for the differentiation of grape variety has been demonstrated.
Materials and Methods
600 authentic varietal german wines from 10 grape varieties have been investigated.
Sample preparation consisted of the addition of 10% of buffer and subsequent pH adjustment to pH 3.1. Analyses were performed under full automation, on an AVANCE III 400 NMR Spectrometer equipped with a 5
mm 1H/D BB probehead with z gradients.
Two 1H NMR experiments (a standard single pulse experiment and a multiple suppression experiment) and a fast 2D J-resolved experiment have been acquired for each sample.
Multivariate statistical data analysis (based on Linear Discriminant Analysis) has been applied in order to build a classification model for grape variety prediction.
MonteCarlo embedded cross-validation techniques have been used to assess the predictive power of the model.
Results and Discussion
In wine the high concentration of ethanol and water saturates the detector and makes the detection of other components difficult in a standard 1H experiment (Fig. 1). However, this fast experiment can be used for direct quantification of the ethanol content of wine.
Selective suppression of water and ethanol signals by means of an 8-bands suppression experiment increases the dynamic range of compounds detected. An example of such a spectrum is given in Figure 2.
The concentration range of compounds that can be detected and quantified is of 4–5 orders of magnitude (Fig. 3).
As the 8-bands suppressed spectra contain many more signals (Fig. 2), they were thus used for the building of a classification model for grape variety differentiation.
The results of the MonteCarlo cross validation analysis shows the potential of the model for the prediction of grape variety (Fig. 4).
The grape varieties sufficiently represented could be differentiated, except for Grau and Weissburgunder (Pinot gris/Pinot blanc). By discarding the lowest represented variety, Silvaner (only 13 samples), correct prediction probabilities of 96% are achieved.
The combination of fast 1H NMR fingerprinting with multivariate analysis has shown promising results for grape variety prediction of wine.
This method has already been applied successfully to the differentiation of wine vintage as well as the differentiation between wine aged in barrels or with oak chips.