Analysis of RNA polymerase II trigger loop mutations using simulations and machine learning

RNA polymerase catalyzes RNA synthesis by the help of the coordinated actions of a near active site domain called the trigger loop (TL). TL is documented to have functions in nucleotide selection, positioning, fidelity and backtracking, while the mechanisms of these events and the exact roles of TL are unknown. In this study, we performed a comprehensive analysis of the TL mutations using molecular dynamics (MD) simulations and deep learning techniques. We generated a numerical continuum of the complete phenotypical map of the TL mutants using a second generation of genetic fitness data by a deep learning approach. We showed that amino acid sequences of the TL mutants could predict the continuous phenotypes at 0.68 R2 correlation. Incorporation of the structural data from the MD simulations into these predictive models has a minor effect in the predictions. On the other hand, MD data brought important insights on the structural outcomes of the distinct phenotypes that are clustered based on a variational autoencoder (VAE) model. This study showed that most of the lethal and some of the loss-of-function (LOF) mutants have large distances between the near active site residues and the incoming nucleoside triphosphate (NTP) suggesting a direct mechanism for affecting the catalysis, while a subset of LOF mutants manifested closer distances for NTP that suggests a more indirect mechanism. We showed that mutants with the gain-of-function (GOF) phenotype directly or indirectly interrupted a hydrophobic pocket previously known as stabilizing the open TL. Overall, this study provides a systematic description of structural outcomes of the distinct TL mutations.