Background: Phenotypic tolerance describes the situation in which bacteria are resistant to antibiotics but are genotypically susceptible. Such tolerance may be important for optimal antibiotic therapy. Objectives: 1) To develop a mechanism-based PD model that describes the potential phenotypic tolerance of P. aeruginosa at high initial inocula (CFUo; CFU: colony forming units) to ceftazidime. 2) To propose strategies of model qualification forin vitro bacterial count data.
Methods: Time-kill experiments were performed in duplicate with ceftazidime concentrations of 0, 0.5, 1, 2, 8, and 32 times the minimal inhibitory concentration (MIC) against P. aeruginosa PAO1 (MIC: 2 mg/L for ceftazidime) with log10 CFUo of 6.2, 7.3, and 8.0. Bacterial counts were determined at 7 (or more) time points from 0 to 24 h (0 to 48 h for log CFUo of 7.3). Models were developed in Berkeley Madonna (version 8.3.11). Bacterial counts on log-scale were fitted by candidate models in NONMEM VI (method: FOCE). Internal cross-validation (ICV) was performed by leaving out all data at one inoculum, re-estimating model parameters, and predicting bacterial counts for the inoculum not used during re-estimation.
Results: For log CFUo of 6.2, concentrations of 2 x MIC and above achieved a net bacterial killing of 3 to 4 log10at 8 and 24 h. At log CFUo of 7.3, there was no net killing up to 4 h and bacterial counts decreased linearly on log-scale between 4 and 48 h by 0.9 log for the 8 x MIC curve and by 1.6 log for the 32 x MIC curve. For log CFUo of 8.0, bacterial counts showed neither a decline nor growth at ceftazidime concentrations from 0.5 to 32 x MIC. The proposed PD model comprised a susceptible and a genotypically resistant population. Bacterial death was described by a stimulation of autolysin activity by ceftazidime where autolysis prevented successful bacterial replication. It was hypothesized that bacteria released signal molecules that cause a loss of autolysin activity and a prolonged generation time for bacterial replication. Results of linear regression of observed vs. population predicted log10 (CFU/mL) were: slope 1.02, intercept -0.12, r=0.97. Median [25-75% percentile] of the ratio of predicted and observed bacterial counts not used during re-estimating in the ICV was 1.54 [0.75-4.17] for log CFUo 6.2, 1.01 [0.71-2.53] for log CFUo 7.3 and 0.90 [0.63-1.09] for log CFUo 8.0.
Conclusions: In vitro time-kill experiments showed a pronounced tolerance of P. aeruginosa at high CFUo whose potential clinical significance needs to be evaluated. 2) The proposed model yielded unbiased and precise curve fits and good predictive performance and may be useful to optimize dosage regimens. 3) Internal cross-validation was very helpful to assess the predictive performance across various initial inocula.
J. Bulitta is supported by Johnson & Johnson with a post-doctoral fellowship.