Determination of Loline Alkaloids and Mycelial Biomass in Endophyte-Infected Schedonorus Pratensis by Near-Infrared Spectroscopy and Chemometrics.

Near infrared spectroscopy (NIRS) is an accurate, fast and nondestructive technique whose use in predicting forage quality has become increasingly relevant in recent decades. Epichloë-infected grass varieties are commonly used in areas with high pest pressure due to their better performances compared to endophyte-free varieties. The insect resistance of Epichloë-infected grasses has been associated with four main groups of endophyte secondary metabolites: ergot alkaloids, indole-diterpenes, lolines and peramine. Concentrations of these alkaloids are usually measured with high performance liquid chromatography or gas chromatography analysis, which are accurate methods but relatively expensive and laborious. In this paper, we developed a rapid method based on NIRS to detect and quantify loline alkaloids in wild accessions of Schedonorus pratensis infected with the fungal endophyte Epichloë uncinata. The quantitative NIR equations obtained by modified partial least squares algorithm had coefficients of correlation of 0.90, 0.78, 0.85, 0.90 for N-acetylloline, N-acetylnorloline and N-formylloline and the sum of the three, respectively. The acquired NIR spectra were also used for developing an equation to predict in planta fungal biomass with a coefficient of correlation of 0.75. These results showed that the use of NIRS and chemometrics allows the quantification of loline alkaloids and mycelial biomass in a heterogeneous set of endophyte-infected meadow fescue samples.

Prediction of the type of milk and degree of ripening in cheeses by means of artificial neural networks with data concerning fatty acids and near infrared spectroscopy.

The present study addresses the prediction of the time of ripening and type of mixtures of milk (cow's, ewe's and goat's) in cheeses of varying composition using artificial neural networks (ANN). To accomplish this aim, neural networks were designed using as input data the content of 19 fatty acids obtained with GC-FID of the cheese fat and scores obtained from principal component analysis (PCA) of NIR spectra. The best model of neuronal networks for the identification of the type of mixtures of milk was obtained using the information concerning the fatty acid concentration (80% of correct results in the training phase and 75% in the validation phase). Regarding the information of the near-infrared (NIR) spectra a neural network was designed. The aforesaid neural network predicted the ripening of cheeses with 100% accuracy in both training and in validation.