Abstract
Convergent evolution has resulted in nonhomologous enzymes that contain similar active sites that catalyze the same primary and secondary reactions. Comparing how these enzymes achieve their reaction promiscuity can yield valuable insights to develop functions from the optimization of latent activities. In this work, we have focused on the promiscuous amidase activity in the esterase from Bacillus subtilis (Bs2) and compared with the same activity in the promiscuous lipase B from Candida antarctica (CALB). The study, combining multiscale quantum mechanics/molecular mechanics (QM/MM) simulations, deep machine learning approaches, and experimental characterization of Bs2 kinetics, confirms the amidase activity of Bs2 and CALB. The computational results indicate that both enzymes offer a slightly different reaction environment reflected by electrostatic effects within the active site, thus resulting in a different reaction mechanism during the acylation step. A convolutional neural network (CNN) has been used to understand the conserved amino acids among the evolved protein family and suggest that Bs2 provides a more robust protein scaffold to perform future mutagenesis studies. Results derived from this work will help reveal the origin of enzyme promiscuity, which will find applications in enzyme (re)design, particularly in creating a highly active amidase.