Factors related to differential occurrence rates of aflatoxins in maize grains: investigations with near infrared spectrometry and metabolite analysis

Aflatoxins are toxic secondary metabolites produced by aflatoxigenic moulds belonging to the Aspergillus section Flavi group. Aflatoxins contaminate food and feed with serious and fatal consequences. Therefore, toxin detection is critical for aflatoxin management. Current detection methods of aflatoxins involve sub-sampling to retrieve a representative sub-set for invasive analysis. However, as aflatoxins are heterogeneously distributed, sub-sampling contributes to errors. Also, invasive analysis is time consuming. Near infrared spectrometry (NIRS) is a non-invasive detection method that was explored for modeling contamination. Using a cross validation modeling, with ultra-high liquid chromatography (UPLC) as reference method, NIRS was explored for maize grain contamination by aflatoxigenic Aspergillus strains. Maize variety (33v62) and A. parasiticus (F75) strain were used. Maize grains at different stages of maturity had fungal colonisation in the following order: milk>dough>dent stages. NIRS pretreated data separated samples into clusters in principal component (PC) scores plot using the first two principal components. Data pretreatments involved Savitzy Golay first derivative and multiplicative scatter correction. The functional groups CH (1st overtone, 1st overtone combinations and 2nd overtone), OH (1st overtone, combinations), NH (1st overtone, combinations), C=O (2nd overtone), NH+OH (combinations), CH+CH (combinations) and CH+CC (combinations) were identified in plot loadings. These are mostly associated with sugar, amino acid and oils. The wavelengths that accounted for most of the variation were at 1220 nm, 1410 nm, 1690 nm and 1900 nm in PC1; and 1145 nm, 1400 nm, 1430 nm, 1890 nm and the combination band region in PC2. A single wavelength linear regression model was significant at 2198 nm with p < 0.0001 for the overall model and the multiple correlation coefficients of 0.94, and R2 = 0.88. Since fungal and toxin accumulation patterns are dynamic, a study was also conducted to understand how Aspergillus populations and aflatoxin concentrations can be influenced competing fungi in different environmental simulations (soil, tube and cotton wool). In the presence of Fusarium, Penicillium spp or Sarocladium zeae, even though starting populations were initially equal, A. flavus populations outnumbered those of the competing fungi at the end of the incubation period. However, aflatoxin levels were undetectable in competition with Fusarium spp. in the soil environment. This was the case, although the populations of the aflatoxigenic fungi were uninhibited. However, in the tube environment, fumonisin production was excluded in favour of aflatoxin accumulation in the grains. The tube environment also promoted the highest levels of aflatoxins. This demonstrated the influence of competing organisms and environmental conditions in aflatoxin production. Development of a model using NIR spectra and aflatoxin levels from field-inoculated samples was conducted. Development of a validation model with low aflatoxin distribution was challenging. Utilising individual kernels with more aflatoxin distribution was more beneficial. A partial least squares – discriminant analysis model using 300 ng/grain as model thresholds were optimal. Validation statistics were slope 0.48, offset 0.08, RMSECV 0.30 and R2 0.41. The difficulty of modeling aflatoxins in the field samples could be attributed to sample complexity and co-occurring fungi in grains. A better understanding of the changes within the grain matrix was explored using non-targeted metabolite analysis. Metabolite analysis showed that fungal dose classification by discriminant analyses was due to sugars, polar and non-polar acids, sugar alcohols and intermediates of the tricarboxylic acid cycle. Sugars were the most abundant molecule for dose classification at each of the maturity stages, with a total of 17, 23 and 16 sugars utilised for dose classification at the milk, dough and dent stages of maturity respectively. Fructose, sucrose and trehalose were among the important sugars for dose classification. Sugar alcohols included xylitol, and mannitol, while key amino acids were alanine, proline, serine, gamma-aminobutyric acid (GABA), valine, threonine, iso-leucine and glycine. Metabolites involved in the tricarboxylic acid pathway such as succinate, citric acid and malate were also used important for dose classification. Several of the sugars and the amino acids were significantly correlated (p0.05) with aflatoxin levels. NIRS modeling challenges for aflatoxins may be attributable to variation in aflatoxin levels due to competition by co-occurring fungi in the grain. Controlled single fungus contamination produced better NIRS models than models with field contaminated samples. The influence of sugars, amino acids, sugar alcohols and other metabolites may also mask aflatoxin concentrations used for modelling A. flavus and aflatoxin contaminated maize grains. Further understanding the grain chemistry changes that occur with aflatoxin accumulation, may be advantageous to explore NIRS calibrations in future studies