Deep Learning-Guided Discovery of Celestolide as a Natural Allosteric Inhibitor Targeting Botrytis cinerea CYP51 and Its Application in Strawberry Preservation

Most CYP51 inhibitors act competitively and are prone to resistance, whereas allosteric inhibitors hold promise but are difficult to develop. In this study, we employed the neural relational inference framework alongside the Gaussian network-based deep learning model CorrSite2.0 to proactively identify allosteric sites on B. cinerea CYP51. We confirmed celestolide as a natural allosteric inhibitor, targeting CYP51. Celestolide forms strong interactions with CYP51 at the identified allosteric site, inducing conformational changes in CYP51 that suppress its enzymatic activity. These findings were consistent with in vitro CYP51 inhibition assays, where celestolide exhibited an IC₅₀ value of 133.28 μg/mL. Moreover, celestolide demonstrated potent antifungal activity against B. cinerea, increasing cell membrane permeability, causing membrane disruption, and ultimately leading to fungal cell death. In a gray mold infection model, celestolide effectively prevented fungal invasion of strawberries and significantly extended their shelf life by more than 1 week. These results highlight celestolide's potential as a natural preservative for postharvest fruit storage.