Sensitivity analysis of a two-state bacterial model for prediction of clinical short-term bactericidal efficacy in tuberculosis: GSK3036656 case study
Silvia Maria Lavezzi (1), Laura Iavarone (1), Pablo Gamallo (2), Ana Novakovic (1), Chao Chen (3), Georgios Vlasakakis (3)
(1) Parexel Clinical Pharmacology, Modeling and Simulation, (2) GSK Global Health Medicines R&D, (3) GSK Clinical Pharmacology Modeling and Simulation
Objectives: GSK3036656 is a potent anti-tuberculosis (TB) drug with a novel mechanism of action, currently being investigated in Phase 2 combination trials. Day 14 early bactericidal activity (EBA) from a recently completed study in patients (NCT03557281) were shown to be in close agreement with simulated EBA values from a translational two-state bacterial (2SB) model [1,2]. The current work expands with a sensitivity analysis of the 2SB model to explore the impact of varying parameters and of the slow-growing bacteria on EBA prediction.
Methods: A two-compartment population pharmacokinetic (PK) model for GSK3036656 was developed based on a first-in-human study in healthy subjects (NCT03075410) and on study NCT03557281 [1,3]. The 2SB model includes fast- (F) and slow-growing bacterial populations (S), with growth rates rF and rS, respectively, their proportion at steady state being F:S. Drug effect is assumed to have a maximum, EmaxF and EmaxS, on F and S, and potency IC50F and IC50S [1,2].
PK-2SB simulations were performed: (i) with typical PK and 2SB parameters without inter-subject variability and residual variability (RV), (ii) using individual PK parameters and baseline bacterial counts from patients in study NCT03557281 (without RV). Dosing regimens were as in study NCT03557281: 1, 5, 15, and 30 mg once a day (QD) for 14 days, with loading doses of 3, 15, 30 and 75 mg on Day 1, respectively.
The impact of varying 2SB model parameters was explored via increasing/decreasing rF, rS, EmaxF, EmaxS, IC50F and IC50S one at a time by 10-fold. The values tested for F:S were 91:9 (almost all bacteria at steady state are F) and 50:50 (bacteria at steady state are equally distributed into F and S).
The importance of S was assessed via testing the following assumptions: (i) drug has no effect on S, (ii) S is not observable in sputum samples, and (iii) S is not included in equations.
Simulated bacterial load was then compared with reference values and with observed values [1]. The analysis was performed using NONMEM 7.4 and R 4.1.
Results: When exploring the impact of varying 2SB model parameters, typical EBA for all doses (1-30 mg QD) was >20% higher than the reference value when rF decreased and when EmaxS increased by 10-fold; it was instead >20% lower when rF or rS increased by 10-fold and when EmaxF decreased by 10-fold. Decreasing IC50F or increasing EmaxF by 10-fold impacted only on 1 mg QD EBA (>20% higher), and this dose was the less impacted by F:S equal to 91:9 (doses>1 mg: EBA >20% higher). Decreasing EmaxS by 10-fold impacted slightly only on 30 mg QD EBA (approx. -20%), and this dose was the less impacted by IC50F increased by 10-fold (doses<30mg: EBA >20% lower). The EBA values simulated for NCT03557281 patients were in part or all (depending on dose level) higher than the observed EBA inter-quartile range (IQR) when rF decreased or when EmaxS increased by 10-fold for doses 1-30 mg QD, when IC50F or IC50S decreased by 10-fold or when EmaxF was increased by 10-fold for 1 mg QD only, and when F:S was 91:9 for 1-15 mg QD only. The simulated patient EBA values were all lower than the observed EBA IQR when IC50F increased by 10-fold for 5 mg QD only, and when rF increased or when EmaxF decreased by 10-fold for 5-15 mg QD.
When the importance of the S population was assessed, typical EBA was at most ~20% lower than the reference value for each dose when drug effect on S was null. When S was considered as not observable or was removed altogether from the model, typical EBA was >20% higher. For NCT03557281 patient simulations, EBA values were within the observed EBA IQR when drug effect on S was null, while they were in part or all higher than the observed EBA IQR when S was considered as not observable or not included in equations.
Conclusions: All 2SB model parameters except IC50S were shown to have relevant (>20%) impact on typical EBA, however the magnitude of impact was dependent on the direction of change, indicating that sensitivity is local. Furthermore, the impact of 2SB model parameters on EBA values might be confounded in practice by inter-subject and/or residual variability.
Both fast- and slow-growing population dynamics appear fundamental to characterize EBA, while drug effect on the slow-growing population has less of an impact on this outcome.
To further evaluate the capability of the 2SB model to predict TB treatment effect, more work is warranted, possibly using preclinical and clinical data from various compounds.
References:
[1] Lavezzi, SM et al. Population modeling of GSK3036656 pharmacokinetics and early bactericidal activity. ASCPT (2023)
[2] Muliaditan, M. & Della Pasqua, O. Evaluation of pharmacokinetic-pharmacodynamic relationships and selection of drug combinations for tuberculosis. Br. J. Clin. Pharmacol. 87, 140–151 (2021).
[3] Tenero, D. et al. First-time-in-human study and prediction of early bactericidal activity for GSK3036656, a potent leucyl-tRNA synthetase inhibitor for tuberculosis treatment. Antimicrob. Agents Chemother. 63, e00240-19 (2019).