Model-based rationale for drug combinations in tuberculosis
Morris Muliaditan (1,2), Oscar Della Pasqua (1,2)
(1) Clinical Pharmacology and Therapeutics Group, UCL, London, UK, (2) Clinical Pharmacology Modelling & Simulation, GlaxoSmithKline, Uxbridge, UK
Objectives: Pulmonary tuberculosis (TB) is an infectious disease caused by M. Tuberculosis (Mtb). The first-line regimen consists of rifampicin (R), isoniazid (H), ethambutol (E) and pyrazinamide (Z) daily for two months followed by R and H daily for additional four months, with doses depending on the patient’s body weight. Despite the evidence of efficacy in clinical trials and clinical practice, the rationale underpinning the selection of present and novel combination therapies for TB remains empirical. Such empirical dose selection may explain the recent failure of three major phase 3 trials aimed to shorten current first-line therapy. Novel methods are hence required to better integrate pharmacokinetic-pharmacodynamic (PKPD) data arising from different preclinical experimental protocols for rational selection of drug combinations for clinical development. The availability of parametric approaches that enable quantification of the contribution of each drug in the combination therapy separately from the growth and death rate of the Mtb bacteria has remained limited. In fact, most of those methods are not readily applicable for the evaluation and ranking of combination regimens under current experimental protocols in TB. In addition, available methods describing drug interactions are not suitable to guide dose selection in clinical development. To overcome some of the current challenges in TB drug development, this analysis aimed to demonstrate how non-linear mixed effect modelling in conjunction with simulation methods can be used to integrate PKPD data from various preclinical experiments. Second, different scaling methods for selection of the dose and drug combination were evaluated.
Methods: Standard of care drugs (R, H, Z and E) were used as paradigm compounds. PKPD data of untreated BALB/c mice or treated with R, RH, RZ, RE, RZH, RZE and RZEH (5 days per week, once daily) for 8 weeks were extracted from literature (1-4). All drugs were administered via oral gavage. Mice were infected either by aerosol or intravenous route. In most experiments, the following standard doses were given: 10 mg/kg (R), 150 mg/kg (Z) and/or 100 mg/kg (E). Doses of H varied between experiments and ranged from 1.56 to 50 mg/kg. One experiment (1) studied the microbial clearance following 10-50 mg/kg R.
Disease progression, as measured by colony forming unit (CFU) counts, was described using a two-state model which assumed the existence of fast (F) and slow-growing (S) Mtb populations. Net growth rates of each population was assumed to be the same in mice as in human, whereas the inoculum and the host carrying capacity was estimated for each preclinical experiment or clinical study.
R was subsequently selected as backbone therapy and a PKPD model of R monotherapy was developed. The additional contribution of subsequent candidates was based on change in potency of the backbone regimen, as described by fast-growing (IC50-F), slow-growing (IC50-S) or both populations. Following this strategy, a stepwise covariate model building was performed. This was followed by simulations to assess the predictive performance of the PKPD models for early bactericidal activity (EBA0-14) in patients treated with R mono-therapy for which clinical data is available (5). The inoculum in TB patients was assumed to be 100 CFU/ml sputum while treatment was assumed to start approximately two months after infection. The PKPD relationship were initially assumed to be the same as in mice while differences in drug disposition and bacterial load at onset of treatment were taken into account. Published population PK of R was used to simulate exposure in TB patients (6).
Alternative scaling strategies were then explored to improve the predictive performance of the in vivo PKPD model for EBA0-14 in patients. Strategies included scaling for, but not limited to, differences in infection routes in the mouse models (intravenous vs. high-dose aerosol vs. low-dose aerosol), protein binding, F:S Mtb ratio or combination of the aforementioned factors. External validation of the selected scaling factors was subsequently performed using datasets from patients treated with R, RH, RE and HRZE (7, 8). Data analysis was performed using a non-linear mixed effect approach as implemented in NONMEM 7.3.
Results: Z was found to be the best companion for R (1.9x increase in IC50-F and IC50-S). In contrast to current clinical practice, it was found that the addition of H worsened the antibacterial effects of RZ in a dose dependent manner. Addition of E had no effect to the antibacterial activity of RZ. External validation demonstrated adequate performance for prediction of microbial clearance in mice infected via high-dose aerosol intravenous route who were treated with various combination regimens.
EBA0-14 in patients following R mono-therapy could not be predicted by scaling for difference in disposition and initial bacterial load alone. On the other hand, scaling for difference in protein binding, F:S Mtb ratio or combination of both factors did yield adequate predictions. This latter approach was chosen for subsequent simulations. EBA0-14 in patients who received either (up to 2 fold) higher doses of R or WHO recommended RHZE dosing regimens were also predicted. RH and RE data showed higher variability than predicted by the model.
Conclusion: We have shown that it is possible to systematically integrate different experimental protocols and describe the effect of combination treatments on the parameters of interest. The evaluation of drug combinations by non-linear mixed effect modelling using preclinical PKPD data provides a more robust rationale for selection of drug combinations than empirical choices. It is anticipated that the proposed parametric approach may allow for the assessment of the contribution of companion drugs in novel combination therapies. This analysis furthermore demonstrates that EBA0-14 in TB patients can be predicted from in vivo experiments. Finally, the data shows that RZ or eventually RZE are sufficient as a backbone therapy in prospective novel combination regimens against TB.
 Almeida et al. (2009). Antimicrob. Agents Chemother 53: 4178-4184
 Hu et al. (2015). Front Microbol 6:641
 Almeida et al. (2014). J Mycobac Dis 4:145
 Grosset et al. (2011). Am J Respir Crit Care Med 183: A1832
 Rustomjee et al. (2008). Antimicrob Agents Chemother. 52: 2831-4
 Smythe et al. (2012). Antimicrob. Agents Chemother 56: 2091-2098
 Jindani et al. (2003). Am J Respir Crit Care Med. 167: 1348-1354
 Dawson et al. (2015) Lancet. 385: 1738-1747
The research leading to these results has received funding from the Innovative Medicines Initiative Joint Undertaking under grant agreement n°115337, resources of which are composed of financial contribution from the European Union’s Seventh Framework Programme (FP7/2007-2013) and EFPIA companies’ in kind contribution.