Identification and quantification of bovine meat adulteration by using NIR spectroscopy

Food authentication is a major concern in food quality. The increasing processing of meat into frozen and/or cooked products has made it more common to find commercial meat frauds.
Many analytical methods are available for minced meat authentication. However, they all are invasive, expensive and require sophisticated laboratory procedures. On the contrary, recent scandals have highlighted the necessity of rapid methods for the meat supply chain inspection to support fair trade. Therefore, the aim of this research was to evaluate the suitability of NIR spectroscopy for the quantification of minced beef adulteration with turkey meat.
Eleven batches of beef bottom round and eleven turkey breast meat samples were used. Each batch was separately minced and used to prepare in duplicate seven mixtures of bovine meat added with different percentages of turkey meat: 5-10-15-20-30-40-50% (w/w). For each batch, two samples of pure bovine meat (0%-adulteration) and two samples of pure turkey meat (100%-adulteration) were also prepared, providing a total of 198 samples.
All fresh meat samples were vacuum packed, frozen, and stored at -18°C for 6 months. Then the samples were thawed (4°C, 16 h) and, immediately after spectroscopic analyses, cooked in a microwave oven (4 min, 450 W).
NIR spectra of all samples were recorded in duplicate using a FT-NIR spectrometer (MPA, Bruker Optics) fitted with an integrating sphere (12500-3900 cm-1). The spectral data were standardized by different spectral pre-processing methods (SNV, MSC, first or second derivative). Quantitative calibration models were developed using Partial Least Squares (PLS) regression and classification models were developed by PLS Discriminant Analysis (PLS-DA). The models were validated using both “leave-one batch-out” cross-validation and an external test set created with samples belonging to three different batches. The chemometric calculations were performed by using the PLS_Toolbox working under MatLab 8.3.0.
PLS was applied to spectral data in order to evaluate the possibility to predict the amount of turkey meat in minced beef. The results indicate that it is more accurate and robust to detect turkey meat in beef when the meat is freshly prepared. However, the results for cooked and frozen meat were also acceptable. Variable selection/reduction helped to achieve more robust and simple PLS models that allow quantifying turkey meat in beef with an error of around 10%. This error appears to be a little bit high for commercial screening purposes. Therefore, it was decided to establish classification models (PLS-DA) for the identification of adulterations with 20% turkey meat or more (threshold value selected because of economic reasons). Also the results from PLS-DA classification models showed that it is more feasible to detect meat adulteration in fresh meat, with sensitivities and specificities values of around 90% for external prediction.
The obtained PLS models demonstrate the ability of NIR spectroscopy to predict the percentage of adulteration of bovine meat with the lower grade turkey meat. The NIR spectroscopic method thus represent an inexpensive alternative to protein- and DNA-based methods commonly used for species identification, particularly in a preliminary screening step where high accuracy is not required.