Novel applications and methods for the computer-aided understanding and design of enzyme activity

Abstract

Despite great progress over the past several decades in the development and application of computer-aided tools for engineering enzymes for a vast array of industrial applications. rational enzyme design remains an ongoing challenge in biotechnology. This thesis presents a set of novel applications and methods for the computer-aided understanding and design of enzyme activity. In the first part. we apply biophysical modeling approaches in order to design non-native substrate specificity in a key enzymatic step (the thiolase-catalyzed condensation of two acyl-CoA substrates) of an industrially useful de novo metabolic pathway. We present a model-guided. rational design study of ordered substrate binding applied to two biosynthetic thiolases. with the goal of increasing the ratio of C6/C4 products formed by the 31HIA pathway, 3-hydroxyhexanoic acid and 3-hydroxybutyric acid. We identify thiolase mutants that result in nearly ten-fold increases in C6/C4 selectivity. Our findings can extend to other pathways that employ the thiolase for chain elonglation, as well as expand our knowledge of sequence-structure-function relationship for this important class of enzymes. In the second part, we apply methods from machine learning to an ensemble of reactive and non-reactive, but "almost reactive" molecular dynamics trajectories in order to elucidate catalytic drivers in another industrially important model enzyme system, ketol-acid reductoisomerase. Using a small number of molecular features, we show that we can identify conformational states that are highly predictive of reactivity at specific time points relative to the progress of the prospective catalytic event and also that provide mechanistic insight into the reaction catalyzed by this enzyme. We then present a novel theoretical framework for evaluating the contribution to the overall catalytic rate of the conformational states found in the previous part to be predictive of reactivity. Leveraging a computational enhanced sampling technique called transition interface sampling, we show that trajectories sampled in such a manner as to selectively visit the conformations predicted to be characteristic of reactivity exhibit rate constants many orders of magnitude greater than trajectories not required to visit these reactive conformations. The results of this analysis illustrate the importance of incorporating dynamical information into existing frameworks for biocatalyst design.