A structure-guided computational screening approach for predicting plant enzyme-metabolite interactions

Plants are molecular factories that have spent millions of years evolving the enzymes needed to synthesize diverse primary and specialized metabolites. Despite the wealth of metabolites that plants produce, many of the enzymes responsible for generating these molecules have yet to be identified. For enzymes with known substrates, the extent of substrate promiscuity and small-molecule regulation remains unexplored. Many computational methods for identifying metabolic enzymes focus on gene-based approaches that rely on transcriptomics, metabolomics, and comparative genomics. With new AI-based tools for accurate protein structure prediction, protein-based strategies that screen a library of small molecules against a high-quality protein model can facilitate the identification of substrates, products, or inhibitors. Virtual screening has been used for structure-based drug design in the pharmaceutical industry for decades and easily translates to investigating plant metabolic enzymes. Here, we present a method for rapid, user-friendly, and open-source virtual screening using the Arabidopsis thaliana UGT74F2 with a curated library of specialized metabolites and herbicides and AutoDock Vina as an example. This method may be applied broadly to metabolic enzymes, and compound libraries can be easily adapted. Compounds are ranked based on their relative binding affinities and the resulting binding modes are evaluated using a molecular visualization program, like PyMOL. Because this is a computational approach, results from the virtual screen will need to be validated using in vitro or in vivo activity, binding, or inhibition assays. Virtual screening may aid in identifying substrates for enzymes of unknown function, revisiting substrate selectivity, or identifying natural or synthetic inhibitors.