Application of fluorescence spectroscopy and chemometrics to classify the spore level of nonfat dry milk and predict the process cheese emulsion properties

Abstract

Spore contamination is one of the major quality concerns in nonfat dry milk (NDM). Some thermophilic and mesophilic spores can survive severe heat treatment, leading to problems in food products that uses NDM as an ingredient. Powder manufacturers need to monitor the spore level throughout production and cease processing in the case of intolerable contamination caused by biofilm attachment and development on the equipment. Current spore enumeration methods are either time-consuming, labor intensive, or expensive, and thus are often not practical in common production facilities. Fluorescence spectroscopy, due to its high specificity and accuracy, can target dipicolinic acid (DPA), which is an intrinsic fluorescent compound in spores, to predict the spore level in contaminated powders. In this study, a total of 40 NDM samples were procured from commercial sources. Traditional plating methods were performed to obtain the reference values of spore count. The results ranged from 1.7 to 5.0 log CFU/g NDM. To enhance the fluorescence signal, pre-treatment steps included autoclaving, acidification, and centrifugation to extract the available DPA from spores present in the reconstituted NDM at 10% concentration. Classification models were constructed using partial least square discriminant analysis (PLSDA), random forest (RF), and forward selection quadratic discriminant analysis (FS-QDA), and models were validated by bootstrapping techniques. The highest classification accuracy was observed for random forest at 87% success rate. The fluorescence-based classification models can provide a rapid tool for industry to quickly determine the quality of processed milk powders. In another study, fluorescence spectroscopy was used to predict the emulsion characteristics of process cheese made by bench-top Thermomix™. Emulsion characteristics are the functional properties of process cheese in specific applications, and can be assessed through texture, fat droplet size, viscoelasticity, and other attributes related to its performance in final products. These properties vary from product to product, and are crucial to maintain in order to ensure consistent quality. However, this is difficult for process cheese manufacturers as they are dealt with varying natural cheese depending on the availability. Fluorescence spectroscopy and near infrared, as rapid and noninvasive techniques, are suitable for on-line measurement of the key emulsion attributes. Essential properties, such as texture profile, rheological properties, and particle size were analyzed by textural profile analyzer (TPA), dynamic stress rheometer (DSR), and dynamic light scattering analyzer, respectively, on 40 process cheese samples produced at the K-State Dairy Products Lab. These samples consisted of 5 batches of 8 combinations of two cheese ages, two mixing speeds, and two holding times. ANOVA results showed significant differences in functional properties existed between process cheese samples. Principle component analysis (PCA) projected the samples into two-dimensions to identify the important variation among samples and variables. Quantitative models based on partial least square regression (PLSR) were developed using tryptophan fluorescence emission spectra, vitamin A excitation spectra, and NIR short-wave region. The calibration models were validated by the leave-one-out method, and yielded the highest correlation coefficient of 0.73 between pre-processed NIR spectra and hardness, followed by particle size. Fluorescence showed limited success in predicting other emulsion attributes. The study showed that correlation existed between fluorescence and NIR spectra, and key emulsion characteristics in process cheese, yet the model was not accurate enough for implementation in the industry. Future study may utilize pilot-scale production unit and include more samples to improve the model performance.