Mid-infrared spectroscopy can be applied to authenticate A2 milk.

Due to a genetic variation in β-casein, A2 milk is more easily digestible than regular milk (A1); presence of the amino acid proline instead of histidine in position 67 of the peptide chain prevents the release of β-casomorphin-7 during digestion. This study evaluated the application of mid-infrared (MIR) spectroscopy as a rapid, noninvasive, and routinely large-scale method to authenticate the A2 variant in Holstein cow milk. Spectral, genetic, and milk quality (fat, protein, lactose, and SCC) data from 2,270 milk samples from 2 consecutive routine milk controls were retrieved from 1,356 animals from 6 farms located in the same area that raised both A1 and A2 cows. Genetic information included β-casein, κ-casein, and β-lactoglobulin variants. Milk compositional differences were statistically assessed before the spectral modeling. Then, a preliminary principal component analysis (PCA) on spectra information was conducted, followed by a partial least squares discriminant analysis (PLS-DA) with 30% of the samples as the test set. Results indicated that milk quality was similar across all protein fractions but differed slightly among farms (P < 0.05). The preliminary spectral evaluation revealed that the first 2 components of the PCA explained 73.2% of the variance. Still, it could not segregate A1 and A2 milk samples based on β-casein genetic information. The PLS-DA model revealed the lowest balanced accuracy in the training and testing set for the genotype A1A1 (50%). For genotypes A1A2 and A2A2, a better balanced accuracy was recorded in the training than in the testing set and slightly greater for A2A2 than for A1A2. For A1A2, balanced accuracy was 80% for the training set and 81% for the testing set. For A2A2, the balanced accuracy was 81% for the training set and 82% for the testing set. Moreover, balanced accuracy improved when only considering 2 levels, A1 milk (comprising genotypes A1A1 and A1A2) and A2 milk (genotype A2A2), reaching 94% for the training set and 88% for the testing set. In conclusion, MIR spectral information is a promising method to authenticate A2 milk based on a PLS-DA model.