Model-informed drug development for antimicrobials: translational pharmacokinetic-pharmacodynamic modelling of apramycin to facilitate prediction of efficacious dose in complicated urinary tract infections

Abstract

Objectives: The use of mouse models of complicated urinary tract infection (cUTI) has usually been limited to a single timepoint assessment of bacterial burden. Based on longitudinal in vitro and in vivo data, we developed a pharmacokinetic-pharmacodynamic (PKPD) model to assess the efficacy of apramycin, a broad-spectrum aminoglycoside antibiotic, in mouse models of cUTI.

Methods: Two Escherichia coli strains were studied (EN591 and ATCC 700336). Apramycin exposure-effect relationships were established with in vitro time-kill data at pH 6 and pH 7.4 and in mice with cUTI. Immunocompetent mice were treated with apramycin (1.5-30 mg/kg) starting 24 h post-infection. Kidney and bladder tissue were collected 6-96 h post-infection for cfu determination. A PKPD model integrating all data was developed and simulations were performed to predict bacterial burden in humans.

Results: Treatment with apramycin reduced the bacterial load in kidneys and bladder tissue up to 4.3-log compared with vehicle control. In vitro and in vivo tissue time-course efficacy data were integrated into the PKPD model, showing 76%-98% reduction of bacterial net growth and 3- to 145-fold increase in apramycin potency in vivo compared with in vitro. Simulations suggested that an 11 mg/kg daily dose would be sufficient to achieve bacterial stasis in kidneys and bladder in humans.

Conclusions: PKPD modelling with in vitro and in vivo PK and PD data enabled simultaneous evaluation of the different components that influence drug effect, an approach that had not yet been evaluated for antibiotics in the cUTI model and that has potential to enhance model-informed drug development of antibiotics.

Figure 1.

Schematic representation of the pharmacokinetic-pharmacodynamic (PKPD) model. The model includes two bacterial subpopulations, one susceptible (1. Main) and one resistant (2. Resistant). In each subpopulation, the bacteria may exist in one of two discrete states: (S) antibiotic-susceptible proliferating bacteria and (D) dormant bacteria unsusceptible to the antibiotic. The bacteria enter the D state as a response to high population densities. k_a, absorption rate constant; k_g, bacterial growth rate constant; k_death, bacterial death rate constant; k_drug, rate constant describing the drug effect (modelled as a power model); k_sr, rate constant describing the transfer from the susceptible to the dormant population as a response to high population densities; k, kidney; b, bladder; SC, compartment for subcutaneous administration; C, central (plasma) compartment. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Figure 2.

Visual predictive checks (VPCs) of the final model based on pH-adjusted in vitro time–kill data. Dots represent observed data, lines represent the 95% CI of the median (shaded area). Each panel represents data for a specific bacterial strain (namely, EN591 and ATCC 700336) under different pH conditions (pH 6 or pH 7.4) at different apramycin concentrations. Data below the limit of quantification (10 log₁₀ cfu/mL) are plotted as −0.2 (MIC for EN591: 8 and 32 mg/L at pH 7.4 and pH 6, respectively; MIC for ATCC 700336: 4 and 16 mg/L at pH 7.4 and pH 6, respectively). This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Figure 3.

Visual predictive checks of the final in vivo PKPD model. The model was developed from combined plasma PK and kidney and bladder PD data in mice infected with either (a) EN591 or (b) ATCC 700336 bacterial strains. Dots represent observed PD data (log₁₀ cfu/organ), and lines represent the 95% CI of the median (shaded area). Each panel represents data for an organ (either kidney or bladder) at different apramycin doses. Apramycin was administered subcutaneously twice daily for 3 days starting at 24 h after bacterial inoculation. This figure appears in colour in the online version of JAC and in black and white in the print version of JAC.

Figure 4.

In vivo PD predictions in mice. Predictions in mice infected with either (a) EN591 or (b) ATCC 700336 bacterial strains that underwent different dosing regimens of apramycin and were not included in the original modelling. Dots represent observed PD data (log₁₀ cfu/organ), and lines represent the 95% CI of the median (shaded area) from 500 simulations considering RSEs in all model parameters. Each panel represents data for an organ (either kidney or bladder) at different apramycin doses. Apramycin was administered subcutaneously twice daily for 3 days starting at either (a) 24 h or (b) 96 h after bacterial inoculation.

Figure 5.

Predicted efficacy in humans for the studied strains. (a) EN591 and (b) ATCC 700336. Expected bacterial densities (log₁₀ cfu/organ) are depicted for 48 h after a 30 min IV infusion of apramycin at 0.3, 1.2, 3.6, 10.8 or 30 mg/kg. Human unbound plasma PK profiles were predicted for a 75 kg patient with a creatinine clearance of 120 mL/min. Each panel represents data for an organ (either kidney or bladder) at the different apramycin doses explored previously in human healthy volunteers. Lines represent median and shaded area represents 95% CI of simulations considering inter-individual variability (IIV), residual error and RSEs in all model parameters.