Predicting Residues that Increase Antibiotic Resistance of CTX-M9 Enzymes using Molecular Simulation and Statistical Learning

Abstract

Over 2 million people annually contract a serious antibiotic resistant bacterial infection. Beta-lactamases contribute to antibiotic resistance by hydrolyzing beta-lactam antibiotics, the most commonly used class of antibiotics, and enzymes such as CTX-M9 confer resistance to nearly all commonly used outpatient beta-lactams. We seek to understand how point mutations of these enzymes, both within and away from the drug-binding pocket, alter drug resistance. We have performed molecular dynamics simulations of CTX-M9 and a number of point mutants that increase drug resistance. We then analyzed these with information theoretic methods to (1) prospectively identify residues that alter drug specificity and (2) explain the effect of previously identified gain of function mutations distant from the pocket. In order to globally identify residues contributing to drug activity, we employed positional mutual information, a non-linear measure of atom association based on movement, to detect residues most associated with the drug. We tested these predictions experimentally via alanine mutagenesis. High-ranking residues had a significantly greater loss of activity than low-ranking ones, suggesting that this metric can be used to identify residues important for drug resistance. We also utilized machine learning methods in the form of mutual-information-guided feature selection and decision trees to understand how the effect of distant mutations is transmitted to the drug-binding pocket. Using this technique, we were able to identify a small set of atoms in the binding pocket where particular positional changes identify enhanced drug resistance with >90% accuracy.