Although antibiotics have been used for more than 70 years, we are not yet able to predict how antibiotic concentration affects activity (i.e. antibiotic pharmacodynamics) even in the simplest settings, e.g. E. coli growth in vitro3-7. Our inability to design rational treatment strategies is illustrated by the substantial improvements in treatment that have been made solely based on expert opinion even after decades of clinical practice8-12. Currently, most dosing recommendations are based on those regimens that perform best during a lengthy and expensive series of trial-and-error experiments. Many drug candidates fail during this testing process, and for those candidates that do make it through, the best regimen may well be missed. This trial-and-error approach also limits opportunities for the improvement of dosing for existing drugs and may slow down the development of new promising antibiotics13. Rational dosing of new combination regimens using multiple drugs is even more complex. Antibiotic synergy and antagonism cannot usually be predicted and the nature of the drug-drug interaction may change depending on drug concentration14. Furthermore, differences in the pharmacokinetic and pharmacodynamic profiles of drugs used in combination can facilitate the selection of resistance during multi-drug treatment15,16.
There are many reasons that the development of new drugs is slow and expensive. While many of these are unavoidable and associated with ensuring safety and efficacy, delays and additional costs are partially attributable to bottlenecks that occur as candidate compounds are weeded out during library screening, pre-clinical research, and clinical trials. Late failures of drug candidates are especially problematic given the time and financial investments made during the long development process. These late failures are often due to relapse, either because of persisting bacteria or resistance evolution, an outcome that early trials of drug efficacy are not designed to assess. For example, in the recent trial assessing moxifloxacin for TB, despite excellent early success17, patients were more likely to relapse with the new regimen compared to standard therapy.
We require new tools to rationally design dosing regimens to maximize the benefits of existing antibiotics and to shorten the development process for new antibiotics18. The development of models that can inform optimal dosing strategies from data collected in early phases of antibiotic development (e.g. drug-target binding and transmembrane permeability) could accelerate the drug development process and help to identify promising compounds that should be prioritized. In particular, mathematical models that predict relapse from pre-clinical and early clinical data would be tremendously helpful19.
We use a novel mechanistic modeling framework that makes explicit links between chemical reaction kinetics (i.e. drug-target association and dissociation), effects on bacterial growth and death, and the population dynamics of bacteria within infected hosts. This modeling framework has been able to reproduce (and mechanistically explain) observed differences in antibiotic pharmacodynamics (Abel zur Wiesch et al., 2015; Abel zur Wiesch et al., 2017). Our mechanistic models differs from previous models20 which have explicitly built in these pharmocodynamic effects.