Non-Targeted Nuclear Magnetic Resonance Analysis for Food Authenticity: A Comparative Study on Tomato Samples

Abstract

Non-targeted NMR is widely accepted as a powerful and robust analytical tool for food control. Nevertheless, standardized procedures based on validated methods are still needed when a non-targeted approach is adopted. Interlaboratory comparisons carried out in recent years have demonstrated the statistical equivalence of spectra generated by different instruments when the sample was prepared by the same operator. The present study focused on assessing the reproducibility of NMR spectra of the same matrix when different operators performed individually both the sample preparation and the measurements using their spectrometer. For this purpose, two independent laboratories prepared 63 tomato samples according to a previously optimized procedure and recorded the corresponding 1D ¹H NMR spectra. A classification model was built using the spectroscopic fingerprint data delivered by the two laboratories to assess the geographical origin of the tomato samples. The performance of the optimized statistical model was satisfactory, with a 97.62% correct sample classification rate. The results of this work support the suitability of NMR techniques in food control routines even when samples are prepared by different operators by using their equipment in independent laboratories.

Keywords: NMR; fingerprint; food control; geographical origin; inter-laboratory comparison; metabolomic analysis; method validation; traceability.

Figure 1

A typical 1D ¹H NOESY spectrum of an aqueous extract of a tomato sample (400 MHz). The main classes of metabolites identified via comparison with reference compounds are indicated by increasing numbering. The full chemical shift assignment is reported in Table 1. “W” refers to the residual water signal. The chemical shift scale is referenced to the TSP-d₄ singlet at 0 ppm.

Figure 2

(a) PC1/PC2 scores plot related to PCA and (b) DModX plots using a dataset composed of all the spectra registered by Lab1 and Lab2. The observations are indicated as yellow diamonds and red triangles for spectra produced by Lab1 and Lab2, respectively.

Figure 3

ROC curves for models (a) M1 and (b) M4.