Assessment of Table Grape Quality Levels by using Headspace Solid Phase Microextraction Mass spectrometry-based Electronic Nose (HS-SPME MS-eNose) and 1H Nuclear Magnetic Resonance spectroscopy (1H NMR) fingerprinting techniques in combination with Multivariate Statistical Analyses

Introduction: The Quality Level (QL) of table grape is defined through sensory evaluation of the its overall appearance by a five-point rating scale (from 5: excellent to 1: extremely poor). Since this evaluation is dependent on subjective judgments, the aim of this work was to develop methods based on MS-eNose and 1H NMR data for an objective QL assessment of two table grape cultivars, i.e. Vittoria and Italia. Methods: Table grape bunches were stored at 5 and 10 °C and sampled after a number of days needed to reach each QL. Berries were homogenized, centrifuged and the supernatant was analyzed by HS-SPME MS-eNose and 1H NMR. LDA and PLS-DA were applied on data to discriminate samples based on the five QLs (five class discrimination) and on their marketability/non-marketability (two class discrimination). Results: The model performances were expressed in terms of the prediction ability calculated by V-fold cross-validation procedure (CV=20%). For both cultivar, NMR model average prediction abilities ranged from 92 to 93%, and from 76 to 79%, in the two and five class discrimination, respectively. Better results were obtained in case of the MS-eNose models with mean prediction abilities ranged from 98 to 100% (two class discrimination), and from 86 to 100% (five class discrimination). In particular, the best prediction abilities of 100% were obtained in case of cv. Vittoria for MS-eNose PCA-LDA in both discriminations (two and five class discriminations) and for MS-eNose PLS-DA in the two class discimination. Conclusions: Taking into account the results obtained, both analytical technologies adopted herein could be used as valid and rapid tools for the table grape quality evaluation. In details, although NMR data can provides additional information on the identification of quality marker metabolites MS-eNose predictive models showed better performances in both types of discriminations.