Application of reflectance spectroscopy to identify maize genotypes and aflatoxin levels in single kernels

Authors: AOUN, MERIEM - Cornell University; SIEGEL, CHLOE - Cornell University; WINDHAM, GARY - Retired ARS Employee; Williams, William; NELSON, REBECCA - Cornell University

Submitted to: Food Control
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/6/2022
Publication Date: 8/8/2022
Citation: Aoun, M., Siegel, C., Windham, G.L., Williams, W.P., Nelson, R.J. 2022. Application of reflectance spectroscopy to identify maize genotypes and aflatoxin levels in single kernels. Food Control. 15(4):324-342. https://doi.org/10.3920/WMJ2021.2750.
DOI: https://doi.org/10.3920/WMJ2021.2750

Interpretive Summary: Aflatoxin contamination of corn grain represents a danger to humans or animals that consume the grain. Aflatoxin contamination, a chronic problem in the Southeast, reduces the value and marketability of the grain. Identifying contaminated grain before it reaches the food chain reduces losses and assures a safer food supply. Analytical methods to detect and quantify aflatoxin require an investment and training and are not generally available to farmers. Spectroscopy is a rapid, non-destructive, and low-cost analytical technique that has the potential to complement more resource-intensive analytical methods. We compared the performance of two instruments: a research-grade ultraviolet-visible-near infrared (UV-Vis-NIR) spectrometer that measures reflectance from 304-1085'nm, and a miniaturized NIR spectrometer that measures reflectance from 740-1070 nm. Both systems were used to predict genotype identity and aflatoxin levels in maize kernels from a single genotype and across 10 genotypes. Compared to the UV-Vis-NIR instrument, the NIR device had similar classification accuracies for AF thresholds of 100 and 1,000 ppb, and higher accuracy for the aflatoxin threshold of 20 ppb. Spectral classification accuracies outperformed visual sorting and the bright greenish yellow fluorescence test. Spectral features in the UV, visible, and NIR regions were found to be associated with aflatoxin levels. Grain testing and sorting based on the identified aflatoxin-associated wavelengths will be useful in efforts to reduce aflatoxin exposure.

Technical Abstract: Spectroscopy is a rapid, non-destructive, and low-cost analytical technique that has the potential to complement more resource-intensive analytical methods. We explored the use of spectral methods to differentiate maize genotypes and assess aflatoxin contamination in maize kernels. We compared the performance of two instruments: a research-grade ultraviolet-visible-near infrared (UV-Vis-NIR) spectrometer that measures reflectance from 304-1085'nm, and a miniaturized NIR spectrometer that measures reflectance from 740-1070 nm. Both systems were used to predict genotype identity and AF level in maize kernels from a single genotype and across 10 genotypes. The classification accuracy of 10 maize genotypes (n = 1,170 kernels) using the UV-Vis-NIR instrument was 72% in calibration sets and 71% in validation sets. The accuracy dropped to 65-66% when the NIR device was applied on 740 kernels. The classification accuracy of aflatoxin-contaminated kernels (n = 247) from a single genotype using the UV-Vis-NIR instrument was 71%, 82%, and 92% in validation sets for AF thresholds of 20, 100, and 1,000 parts per billion (ppb), respectively. Using the same spectrometer, aflatoxin classification accuracy in kernels (n= 872) selected from 10 genotypes was 67%, 90%, and 95% in validation sets for aflatoxin thresholds of 20, 100, and 1,000 ppb, respectively. Compared to the UV-Vis-NIR instrument, the NIR device had similar classification accuracies for AF thresholds of 100 and 1,000 ppb, and higher accuracy for the AF threshold of 20 ppb. Spectral classification accuracies outperformed visual sorting and the bright greenish yellow fluorescence test. Spectral features in the UV, visible, and NIR regions were found to be associated with aflatoxin levels.