Predictive methods for computational metalloenzyme redesign – a test case with carboxypeptidase A
Computational metalloenzyme design is a multi-scale problem. It requires treating the metal coordination quantum mechanically, extensive sampling of the protein backbone, and additionally accounting for the polarization of the active site by both the metal cation and the surrounding protein (a phenomenon called electrostatic preorganization). We bring together a combination of theoretical methods that jointly offer these desired qualities: QM/DMD for mixed quantum-classical dynamic sampling, quantum theory of atoms in molecules (QTAIM) for the assessment of electrostatic preorganization, and Density Functional Theory (DFT) for mechanistic studies. Within this suite of principally different methods, there are both complementarity of capabilities and cross-validation. Using these methods, predictions can be made regarding the relative activities of related enzymes, as we show on the native Zn2+-dependent carboxypeptidase A (CPA), and its mutant proteins, which are hypothesized to hydrolyze modified substrates. For the native CPA, we replicated the catalytic mechanism and the rate in close agreement with the experiment, giving validity to the QM/DMD predicted structure, the DFT mechanism, and the QTAIM assessment of catalytic activity. For most sequences of the modified substrate and tried CPA mutants, substantially worsened activity is predicted. However, for the substrate mutant that contains Asp instead of Phe at the C-terminus, one CPA mutant exhibits a reasonable activity, as predicted across the theoretical methods. CPA is a well-studied system, and here it serves as a testing ground for the offered methods.