Determination of Salmonella sp in whole milk by NIR and Chemometrics

It is well known that milk and milk products provide a wealth of nutrition benefits. Nonetheless, during milking, bulk tank milk can be contaminated by Salmonella sp bacteria in the udder or through fecal matter. A significant problem related to this harmful microorganism is its determination, which is very time consuming and requires specific conditions in order to provide confident results. For example, nowadays, the classical culture method employed of detecting Salmonella sp in milk includes four stages of the detection process and depending on the need to obtain confirmations, it lasts from 5 to 7 days: pre-enrichment in non-selective liquid medium; selective enrichment in liquid media; plating on selective media; serological and biochemical identification of suspected colonies. Classical culture methods are in use by many laboratories and they are laborious to perform, demand large volumes usage of liquid and solid media and reagents, and encompass time-consuming procedures both in operation and data collection. The ideal method to determine Salmonella sp should bring up advantages as relative low cost, fast, not being sample destructive, reagents free and, consequently, not generating residues. It would drive the method to the possibility of online monitoring, and would provide a good solution to this issue. Thus, this research aimed to offer a PLS-DA model based on Near Infrared Spectroscopy (NIRS) to determine Salmonella sp in whole milk. To this, 26 different whole milk samples were acquired. Each sample was UHT sterilized and consisted in a 1.0L milk box, from which it was collected 1.0mL portion to be set to NIR spectra acquisition in a JDSU MicroNIR® equipment, capable to capture information from 900 to 1600nm. After that, a 5.0mL aliquot of this pure milk (from the box) was transferred to an erlenmeyer in order to be spiked with 1.0mL of Salmonella sp solution. This 5.0mL was set to rest during 10 minutes and then it was collected 1.0 mL to be set to NIR measurement. In order to avoid contamination, all the procedures regarding safe handling were observed. After spectra acquisition, the data were organized as a matrix, were each sample was placed in a row and variables in the columns. All data handling was performed using Matlab® and PLS toolbox®. PLS-DA models were mean-centered and cross-validation was done through leave one out method. The results were enough acceptable to suggest that this method can be used in milk industry, mainly considering that there was no wrongly discriminated sample, besides providing appropriated models. The main model parameters to the whole milk with 6 latent variables were calculated and provided 1.0 to sensitivity and specificity of calibration and prediction to both classes, RMSEC was 0.189743, RMSECV was 0.269708, RMSEP was 0.200702. The R2 for Calibration was 0.85599 to class 1.0 and 0.85599 to class 2, R2 of Cross Validation was 0.714588 to class 1.0 and 0.714588 to class 2 and R2 for Prediction was 0.862665 to class 1.0 and 0.862665 to class 2.