Application of FT-IR spectroscopy to analyze casein in cow milk.

Infrared spectrometry is, at present, the most frequently applied methodology for the composition analysis of milk and dairy products. The determination of fat, protein and lactose has been described during recent decades. The introduction of Fourier Transform Infra-red (FT-IR) technology in combination with the application of multi-dimensional procedures (i.e. principal component regression, partial least squares) has improved this methodology, opening new perspectives for the simultaneous and routinely determination of many new parameters such as casein, urea, specific sugars, etc.. The aim of our study was to develop on MilkoScanTM FT 120 (Foss Electric, Hillerød, Denmark) a calibration curve for the analysis of casein in cow milk and to execute a preliminary validation. The calibration curve was developed on 89 individual milk samples collected from 4 dairy herds in the Grana-Padano cheese district. In order to obtain a higher variability of milk protein content and composition, in each herd milk samples from cows in early and late lactation were collected. On milk samples, casein separated at pH 4.6 and protein content were measured using Kjeldahl method (N x 6.38) and the results, expressed as casein percentage of whole milk, were used as reference values. On each milk sample the full spectra with MilkoScanTM FT 120 was collected twice. The samples were sorted according to casein content and homogenously subdivided in two data subset: the 1st (2/3 of the samples) was used to develop the calibration curve; the 2nd (1/3 of the samples) was used, as external dataset, to validate the calibration curve by cross validation. These elaborations were performed using Regression and Partial Least Squares procedures of SAS. The casein content in milk samples used to develop the calibration curve ranged between 2 and 3.5% (w/v) with a mean value of 2.64%. The calibration curve that better fits the casein reference values considers 13 different wavelength areas included in 7 factors. The standard error of the calibration curve was 0.054 as absolute value and 2% as relative value. The intercept of this curve was 0.0001 with a slope of 0.999. Also the standard error of prediction obtained with the cross validation using the external dataset was very low (0.057). The intercept and the slope of the validation curve were close to the optimal values and respectively 0.1 and 0.97. The repeatability parameters, evaluated on all samples, were very low (sr=0.011 as absolute value and RSDr=0.42% as relative value). The results obtained in this study, considering the scarcity of literature information in this area, seem very promising. The calibration curve, after a better validation in different conditions, could be included in the milk routine analysis with different purposes.