Computational glycosyltransferases masked deoxynivalenol toxicity and halted FHB spread in wheat grains

Contamination of cereal grains by deoxynivalenol (DON) poses a significant threat to food safety. This toxic secondary metabolite, produced by F. graminearum, acts as a virulent factor to promote mycelium spread in plant tissue. Enzymatic degradation has emerged as a promising strategy for managing DON contamination, through effective detoxification that not only reduces the toxin's harmful effects but also impairs fungal invasion by disrupting its virulence mechanism. However, the availability of efficient DON detoxification enzymes remains limited.

Glycosyltransferases (GTs) are a class of important enzymes responsible for toxic compounds detoxification and natural products biosynthesis, but only a small fraction of GTs has been experimentally characterized. This study aimed to develop a deep learning model for predicting GTs' promiscuity and computationally identify new DON detoxification enzymes.

The transformer framework was applied to develop a GT-specific enzyme promiscuity prediction model. After computational prediction and preliminary verification, a series of biological assays including mycotoxin tolerance, seedling inoculation, and single floret injection were conducted to assess the resistance of transgenic wheat lines expressing newly identified enzymes against DON and F. graminearum infection.

Using the model, two UGTs capable of detoxifying DON were successfully identified. Transgenic wheat lines demonstrate increased resistance to Fusarium infection, evidenced by reduced mycotoxin accumulation and milder disease symptoms.

This work highlights a promising data-driven approach for functional enzyme discovery in agricultural biotechnology. By enabling the identification of enzymes that mitigate mycotoxin contamination, this strategy contributes to improving food safety and strengthening the resilience of food systems against future biotic stressors.