Large-scale testing of antimicrobial lethality at single-cell resolution predicts mycobacterial infection outcomes

In vitro antibiotic testing is important for guiding therapy and drug development. Current methods are focused on growth inhibition in bulk bacterial populations but often fail to accurately predict treatment responses. Here we introduce Antimicrobial Single-Cell Testing (ASCT), a large-scale live-cell imaging approach that quantifies bacterial killing in real time at single-cell resolution. By tracking over 140 million mycobacteria and analysing ~20,000 time-kill curves, we identify key determinants of antibiotic killing and its clinical relevance. For Mycobacterium tuberculosis, we found that drug-specific killing dynamics in starved bacteria, rather than growth inhibition or killing of growing cells, predict regimen efficacy in mice and humans. Extending this approach to Mycobacterium abscessus and comparing 405 bacterial strains, we show that antibiotic killing is also a genetically encoded bacterial trait (drug tolerance). We demonstrate that tolerance patterns cluster by antibiotic targets, identify a phage protein that modulates antibiotic killing, and show that strain-specific killing dynamics are associated with individual patient outcomes independent of drug resistance. Together, these findings establish a framework that reveals how drug properties and bacterial diversity shape treatment responses, offering a path to more effective and personalized therapies.