Quantifying Subpopulation Synergy for Antibiotic Combinations via Mechanism-Based Modeling and a Sequential Dosing Design

Abstract

Quantitative modeling of combination therapy can describe the effects of each antibiotic against multiple bacterial populations. Our aim was to develop an efficient experimental and modeling strategy that evaluates different synergy mechanisms using a rapidly killing peptide antibiotic (nisin) combined with amikacin or linezolid as probe drugs. Serial viable counts over 48 h were obtained in time-kill experiments with all three antibiotics in monotherapy against a methicillin-resistant Staphylococcus aureus USA300 strain (inoculum, 10 8 CFU/ml). A sequential design (initial dosing of 8 or 32 mg/liter nisin, switched to amikacin or linezolid at 1.5 h) assessed the rate of killing by amikacin and linezolid against nisin-intermediate and nisin-resistant populations. Simultaneous combinations were additionally studied and all viable count profiles comodeled in S-ADAPT and NONMEM. A mechanism-based model with six populations (three for nisin times two for amikacin) yielded unbiased and precise (r ‫؍‬ 0.99, slope ‫؍‬ 1.00; S-ADAPT) individual fits. The second-order killing rate constants for nisin against the three populations were 5.67, 0.0664, and 0.00691 liter/(mg · h). For amikacin, the maximum killing rate constants were 10.1 h ؊1 against its susceptible and 0.771 h ؊1 against its less-susceptible populations, with 14.7 mg/liter amikacin causing half-maximal killing. After incorporating the effects of nisin and amikacin against each population, no additional synergy function was needed. Linezolid inhibited successful bacterial replication but did not efficiently kill populations less susceptible to nisin. Nisin plus amikacin achieved subpopulation synergy. The proposed sequential and simultaneous dosing design offers an efficient approach to quantitatively characterize antibiotic synergy over time and prospectively evaluate antibiotic combination dosing strategies.

Substantial Impact of Altered Pharmacokinetics in Critically Ill Patients on the Antibacterial Effects of Meropenem Evaluated via the Dynamic Hollow-Fiber Infection Model

“…Models were evaluated based on the S-ADAPT objective function (Ϫ1ϫlog-likelihood), standard diagnostic plots, visual predictive checks, and plausibility of the parameter estimates (14,54,65). Models with one, two, and three bacterial subpopulations and which described the effect of meropenem via direct bacterial killing, the inhibition of successful bacterial replication, or both were evaluated (54,66,67).…”

Optimization of Polymyxin B in Combination with Doripenem To Combat Mutator Pseudomonas aeruginosa

“…In brief, it was assumed polymyxin B is competitively displaced by Ca 2ϩ and Mg 2ϩ at the outer membrane of P. aeruginosa, which is the initial site of polymyxin action (41). Two types of synergy, namely, "subpopulation synergy" [i.e., polymyxin B killing the doripenem-resistant subpopulation(s) and vice versa] and "mechanistic synergy" were considered to explain the interaction between polymyxin B and doripenem (40). Mechanistic synergy was evaluated based on the assumption that polymyxin B could enhance killing by doripenem, that doripenem could enhance killing by polymyxin B, or both (see Fig.…”

“…The polymyxin B concentration was assumed to be constant throughout the experiments since polymyxin B is known to be stable in vitro (42). The modeling methods used were described previously (20,(43)(44)(45)(46)(47)(48).…”

“…A life cycle growth model as previously described was used to model bacterial growth kinetics dividing each subpopulation into a vegetative and a replicating state (39,40). Polymyxin B activity was characterized by target site binding, as previously described by Bulitta et al (20).…”

“…Candidate models were estimated by simultaneously fitting the viable count profiles using the importance sampling Monte Carlo parametric expectation maximization method (pmethod ϭ 4) in the parallelized S-ADAPT software (version 1.57) facilitated by the SADAPT-TRAN pre-and postprocessing tool (46). An additive residual error variance model on log 10 scale was used for bacterial counts of Ն100 CFU/ml, and an error model containing a Poisson error was used for bacterial counts that were Ͻ100 CFU/ml (20,40).…”