Simulation-guided engineering of antibiotics for improved bacterial uptake

The Gram-negative bacterial outer membrane poses a major obstacle to the development of much-needed antibiotics against drug-resistant infections. Its chemical composition and porin proteins differ from Gram-positive bacteria and mammalian cells, and heuristics developed for mammalian cell uptake apply poorly. Recently, machinelearning methods have predicted small-molecule uptake into Gram-negative bacteria, offering the possibility to rationally optimize this aspect of antibiotic lead development. Here, we report physics-based methods to prospectively predict Gram-negative bacterial uptake, select, and synthesize promising chemical derivatives targeting E. coli DNA gyrase B. Our methods do not require empirical parameterization and are readily adaptable to new chemical scaffolds. These physics-based predictions well capture experimentally measured uptake (r > 0.95) and are indeed predictive of antimicrobial activity (r > 0.92). These methods can be used prospectively in combination with target-binding simulations to optimize both bacterial uptake and target binding, overcoming important barriers to antibiotic lead generation before small-molecule synthesis.