Rational evolution of a recombinant DNA polymerase for efficient incorporation of unnatural nucleotides by dual-site boosting

Abstract

Machine learning modelling assisting function-oriented enzyme engineering is normally built on predefined protein sequence space. However, efficient defining the determinant amino acid positions upon which the combinatorial mutation library is constructed is still a challenge in protein science. Herein, we present a comprehensive investigation of modifying a recombinant DNA polymerase for efficient incorporating one unnatural nucleotide, including the identification of key sites/regions, machine learning-assisted mutants screening, and the underlying mechanism of kinetics boosting. By using hundreds of training points and only dozens of testing samples, we found that one highly engineered enzyme’s catalytic efficiency can be further improved by one order of magnitude by specific mutation on two sites, 485I and 451L. Compared to the position 485 which is known to dominate local conformation of B-family DNA polymerases, 451 is a split-new active site discovered by our approach. A novel allosteric regulation mechanism is underlying the apparent synergy of 485I and 451L on the kinetics boosting. As a result, a “half-closed” conformation of the binding pocket and a cooperative binding of both primer and template DNA strands on the protein accelerated the processes of substrate’s incorporation, molecular recognition, and releasing of incorrect nucleotides. These findings have implications in guiding the function-tuning of DNA polymerases for a broad range of biotechnological applications.