*Jatinder.sangha2@agr.gc.ca

Cost-effective high throughput screening for deoxynivalenol (DON) mycotoxin in grains, especially during early generations, in the breeding process is essential for selecting Fusarium head blight (FHB) resistance in wheat. We demonstrated the application of near infrared (NIR) spectroscopy with a wavelength ranging from 350-2500 nm to test DON concentrations in wholemeal flour from different durum wheat populations.

Chemometrics analysis involved using the GRAMS/AI software (Thermo Fisher) and R software to build models between DON concentration obtained from an enzyme linked immunosorbent assay (ELISA) and reflectance spectrum data obtained from NIR spectrometer (Labspec4). A small amount (~6 g) of wholemeal flour with DON concentration ranging from 0 to 70 ppm was prepared using nine different genetic populations (136 to 400 lines) of durum grown in FHB nurseries and scanned with NIR spectrometer.

The sample population was partitioned into a training population for model building and a test population for external validation (EV). Second derivative pre-processing of the spectral data using the Savitzky–Golay smoothing with 17 data points using GRAMS/AI provided the highest coefficient of determination (R2= 0.90) that predicted DON concentrations (R2=0.62) in test samples during EV. With R software based prediction models, both partial least squares (PLS) and orthogonal partial least squares (OPLS) methods generated reliable models than principal components regression (PCR) to predict DON concentration in wheat, with a predictive relevance (Q2) ranging from 0.78 to 0.89. The loading plots show wavelength peaks for DON specific signatures in the near infrared zone around 1926 nm and 1439 nm wavelengths.

These prediction results are potentially useful in screening FHB resistance by detecting and discarding wheat lines with high DON concentrations during early generation selections in the breeding process, which is higher throughput and more economical than ELISA or mass spectrometry. Further work aims to use these prediction models for evaluating unknown wheat populations for DON prediction.