MicroNIR spectroscopy: A potential analytical tool for the classification of origin and prediction of δ13C and δ15N values of lamb meat

The hand-held JDSU MicroNIR 1700 was used to explore the classification of origin and prediction of δ13C and δ15N values of lamb meat samples (n=30) from three regionally unique farms in South Africa. Lambs were raised extensively and consumed a diet mainly consisting of lucerne/alfalfa (Medicago sativa) in the Rûens (RU) or grass in the Free-State (FS) or Karoo bushes/shrubs in the Hantam Karoo/Calvinia (HK/CAL) region. Preceding research of descriptive sensory analysis and stable isotope ratio analysis revealed distinct sensory and isotopic differences of the lamb meat obtained from the different farms, which were related to diet linked to origin. Since the above-mentioned analyses can be time-consuming and costly, the aim of the current work was to investigate NIRS as a potential tool for rapid and effective classification of lamb meat. The prediction of isotope values may also serve as an indirect method for the indication of diet related to origin.
Meat samples (n=30) were homogenised, defatted, freeze-dried and δ13C and δ15N determined by isotopic mass spectrometry. Samples were scanned (triplicate) in diffuse reflectance mode between 950-1650 nm using the JDSU MicroNIR 1700 spectrometer. Mean spectra (per sample) were used for data analysis. The full and selected (1100-1600 nm) NIR wavelength ranges were compared. Unscrambler X 10.3 was used for pre-treatment and chemometric analysis. Two pre-treatments: multiplicative scattering correction (MSC) and Savitzky-Golay second derivative (15 point) (SG) were compared. Principal component analysis (PCA) and partial least square discriminant analysis (PLS-DA) (±0.5 cutoff criteria) methods were applied for classification of origin, and partial least squares regression (PLSR) method for the prediction of isotope ratios. For PLS-DA a calibration (n=21) and validation (n=9) sample set were used. For PLSR the whole sample set (n=30) was used for calibration and the validation set (n=9) used in PLS-DA, for prediction.
PCA revealed grouping of samples based on origin. SG transformation at both wavelength ranges showed optimum separation. RU grouped separately from FS and HK/CAL, while FS and HK/CAL were only partially separated. All PLS-DA models achieved 100% correct classification, while SG pre-treatment produced models with the best R2 (0.88 and 0.89) values. The best PLSR calibration models for predicting δ13C and δ15N values were obtained with the R2 of 0.81 and 0.89, respectively. The δ13C was less well predicted than δ15N. Overall, the highest absorption were observed at 1190 nm (C-H second overtones), 1380 nm (Ar-OH first overtones), 1430-1470 nm (R-OH first overtones) and 1510 nm (N-H first overtones). Differences in chemical components of lamb meat (caused by different diets) are likely the reason for discrimination between samples. HK/CAL associated with strong absorption in the 1510 nm range, likely due to its higher δ15N value. Similarly FS had a higher δ13C value and associated with 1190 nm absorption. The results preliminary confirmed the potential of NIRS for the discrimination of origin and estimation of isotope ratios of lamb meat. The results also serve as baseline data for future work, where robust classification model development for authentication will be explored using a larger sample size.