Ensemble-Based Computational Enzyme Design: Methods and Applications

Abstract

Designing artificial enzymes to meet the increasing demand for industrial applications holds immense societal significance. However, despite successes in de novo enzyme design methodologies, artificial enzymes exhibit low catalytic efficiency compared to natural ones. In this thesis, we studied structural changes occurring during the evolution of a low-activity de novo Kemp eliminase HG3 (k_cat/K_M 146 M⁻¹s⁻¹). We observed a rigidification of catalytic residues, a better active site pre-organization and a wider active site entrance. Using backbone ensembles from ensemble refinement (ER) of crystallographic data of HG3, our ensemble-based enzyme design method allowed a high prediction accuracy of HG4, a highly efficient enzyme we engineered (k_cat/K_M 103,000 M⁻¹s⁻¹) containing key first and second-shell mutations found during evolution. Building upon these findings, we established a benchmark of our ensemble-based method using backbone ensembles from various methods obtained from a weakly active Kemp Eliminase, HG3. Our results showed that ensembles derived from PertMin, a method we previously developed, and ER accurately predicted the conformation of its optimized variant, HG4. This benchmark served as a guide for improving the catalytic efficiency of HG3 and KE70 (k_cat/K_M 57 M⁻¹ s⁻¹). Designed sequences displayed catalytic efficiency that improved by 100-250-fold, comparable to screening thousands of variants from rounds of directed evolution. Finally, we benchmarked our ensemble-based design method for predicting to stabilize the multiple transition states of the reaction catalyzed by BH32.12, an evolved variant of BH32 catalyzing the Morita-Baylis-Hillman reaction. Our research advances the field of enzyme design by combining evolutionary insights with computational approaches, leading to the creation of highly efficient artificial enzymes for various applications.