Poster

Hye-Seon Kim¹, Stephen F. Harding^1,2, Olivia C. Haley², Laura Tibbs-Cortes², Todd A. Naumann¹, Susan P. McCormick¹, William T. Hay¹, Robert H. Proctor¹, Carson M. Andorf²
1. USDA-ARS, NCAUR, Mycotoxin Prevention and Applied Microbiology Research Unit, Peoria, IL 61604, USA.
2. USDA-ARS, Corn Insects and Crop Genetics Research Unit, Ames, IA 50011, USA.
Corresponding Author: Hye-Seon Kim, hyeseon.kim@usda.gov

Kim, Hye-Seon

hyeseon.kim@usda.gov

The ability of Fusarium graminearum to cause Fusarium head blight (FHB), and thereby threaten global cereal production, is mediated by the fungus’ capacity to secrete immunosuppressive proteins (effectors) and trichothecene mycotoxins (e.g., DON). Blocking production, secretion, or activity of critical effectors and/or the mycotoxins will prevent FHB. To address this, we combine the deep learning platform BindCraft for custom protein design with the molecular modeling capabilities of the Boltz framework. This hybrid pipeline bridges a key limitation of BindCraft—which cannot natively read chemical structures—by leveraging Boltz to evaluate how designed proteins interact with small target molecules. Here, we propose to use this integrated platform to generate custom protein binders that target three critical components of the F. graminearum–wheat interaction: 1) neutralize F. graminearum effector proteins, 2) alter the shape of the trichothecene detoxification enzyme Fhb7 to enhance its performance, and 3) block DON production by inhibiting the activity of a critical trichothecene biosynthetic enzyme (Tri5) in the fungus. For example, to design Fhb7 binders, we first used Boltz to predict the enzyme's activity against 13 structurally diverse trichothecene analogs. This analysis predicted that Fhb7 had greater detoxification activity against some analogs (15ADON and DON) than others (3ADON and NX-2) due to an unfavorable physical mismatch at the enzyme's active site. Our workflow focuses on engineering custom binders to alter the shape of Fhb7 and enhance its activity against the diversity of trichothecene analogs produced by F. graminearum. To translate information on variation in Fhb7 activity to regional management strategies, Fhb7 activity scores were integrated into "SNPVersity," an online catalog that maps genetic variations across 500 diverse F. graminearum isolates with their respective trichothecene chemotypes. By combining these tools, we predicted the efficacy of Fhb7 given the frequencies of different chemotypes of F. graminearum. The generative machine learning tools described here provide a novel approach to generate durable FHB resistance by altering functions of proteins that are critical in the F. graminearum-wheat interaction.

© Copyright 2026 by individual authors. All rights reserved. No part of this abstract or paper publication may be reproduced without prior permission from the applicable author(s).