Comparing a mechanistic with an empirical approach to assess resistance development of antibacterials in vitro.
K.Bergmann (1), O. Ackaert (1), N. Haddish-Berhane (2), A. Betts (2)
(1) LAP&P Consultants BV, Leiden, The Netherlands; (2) Translational Research Group, Department of Pharmacokinetics, Dynamics & Metabolism, Pfizer, Groton (CT), USA
Objectives: Already early in the preclinical phase, resistance development of bacteria against antibacterials can be observed. The use of dynamic models, already in an early stage, for rapid screening and/or more mechanistic insight in the resistance development is therefore important. The aim of the current study was to compare a series of models to describe resistance development and consequently describe the time course of the bacterial count.
Methods: The pharmacodynamic properties of 3 antibacterial compounds (two Novel siderophore conjugated Beta-Lactams and one LpxC inhibitor) were investigated in in vitro static concentration time kill experiments against Klebsiella pneumoniae (KP-1487) and Pseudomonas aeroginosa (PA-UC12120). The resulting bacterial count-time profiles were analysed with NONMEM using two approaches. The first approach used an empirical one-population adaptation model. The bacteria kill was described by a (sigmoidal) Emax relationship and the EC50 value was allowed to change with time and/or compound concentration to describe the resistance development of the bacteria. The second approach was a mechanistic two-population mutation model. This model takes into consideration that the susceptible bacteria can mutate to become resistant to the antibiotic. For this matter a new model is proposed, taking into account the probability of a mutation occurring. For both approaches the M3-method was applied to take into account the samples below the limit of quantification (LOQ).
Results: Both the empirical adaptation model and the mechanistic mutation model adequately captured the development of resistance during drug exposure to the 3 compounds. Moreover the models described the time course of the bacterial count well. In addition, for the mechanistic model, implementing the new estimation method was necessary to obtain more realistic mutation rates. Both approaches resulted in similar human dose predictions to obtain -1 and -2 log kill after 24h.
Conclusions: A tool box, including an empirical model with an adaptive EC50 (including various adaptation functions, depending on concentration and/or time) and a mechanistic mutation model was developed to analyze in vitro bacterial count-time profiles. Depending on the aim of the analysis and on the available data, the empirical model (e.g. for rapid screening) or the mechanistic mutation model (e.g. for more mechanistic insight) might be preferred.
