Distiller’s dried grains with solubles (DDGS) is the major by-product of biofuel plants that use corn as raw material. This cereal is commonly contaminated by toxic substances known as mycotoxins. Fusarium stands out among the various genera producing these toxins, as it includes the main producers of fumonisins (FBs). In poultry, ingestion of FBs triggers numerous negative effects upon animal performance. DDGS has been used as an alternative to protein sources in poultry diets due to its nutritional characteristics. As well as other mycotoxins, FBs present in corn are concentrated in DDGS during its processing, and mycotoxicological monitoring is essential before its use in feed formulations. Advanced methodological tools, such as the Near Infrared reflectance spectroscopy (NIRS), have been gaining ground in the Industry due to the speed in analysis of multiple parameters. Thus, this work aimed to develop a method for predicting FBs B1 and B2 in DDGS using NIRS in association with chemometric methods. One hundred and sixty-six DDGS samples were used to build the models. Two datasets were created: a calibration group, containing 132 samples, and an external validation group, including 34 samples. The material was received and promptly analyzed at the Laboratory of Mycotoxicological Analyses (LAMIC), Santa Maria (Brazil), by Liquid Chromatography Coupled to Tandem Mass Spectrometry (LC-MS/MS; reference method) throughout 2020; another fraction was used for optical data collection in order to build the spectra library. The spectra were obtained in a Foss XDS Rapid Content® Analyzer and the data were extracted and converted into a JCAMP file. The final spectral data were exported to conduct the chemometric analyses via the Unscrambler v.9.7 software (CAMO, Norway). FBs levels -1 -1 varied from 250 to 8,360 μg.kg ; mean value and standard deviation were 3,897 and 1,627 μg.kg , respectively. The models were evaluated separately for FB1 and FB2. Partial least squares was the regression method applied in the models, using cross-validation. The calibration results for FB1 and FB2 were, respectively: correlation coefficient, 0.90 and 0.89; coefficient of determination, 0.80 and 0.79; root mean square error of prediction, 527 and 237; and residual prediction deviation, 2.30 and 2.20. Values of the external validation dataset were compared with the levels obtained via LC-MS/MS and the Mann-Whitney test was applied.
No statistical difference was found between the groups (p: 0.085), thus indicating a satisfactory predictive ability and confirming the potential of NIRS to predict FBs in DDGS.