Daniele Boaretti 1, Roberto Visintainer 1, Micha Levi 2, Veronique Dartois 3, Federico Reali 1
1 Fondazione The Microsoft Research - University Of Trento Centre For Computational And Systems Biology (Rovereto, Italy), 2 Gates Medical Research Institute (Cambridge, USA), 3 Hackensack Meridian Health (Nutley, USA)
Objectives: Tuberculosis (TB), one of the deadliest infectious diseases worldwide [1], still requires therapy optimization to eradicate Mycobacterium tuberculosis residing in hard-to-treat granuloma lesions. Bacterial clearance is impacted by drug diffusion through the cellular lesion surrounding the non-vascularized caseum; therefore, drug penetration is a key determinant of treatment efficacy and pharmacodynamic target attainment (TA). Combining experimental data from anti-TB treatments in rabbit models with simulation tools can support the identification of the most effective treatments for eradicating M. tuberculosis infection [2]. Here, we present a study of 16 anti-TB drugs, focusing on their penetration kinetics into the cellular lesion and caseum by extending and adapting a published minimal physiologically based pharmacokinetic (mPBPK) mouse model [2] to TB-infected rabbits.
Methods: Concentrations of 16 anti-TB drugs [3, 4], namely, pyrazinamide (PZA), moxifloxacin (MXF), rifampicin (RIF), pretomanid (PRE), rifapentine (RPT), isoniazid (INH), delamanid (DEL), ethambutol (EMB), clofazimine (CFZ), linezolid (LZD), mCLB703, bedaquiline (BDQ, B), sutezolid (SUT), sorfequiline (TBAJ-876), TBAJ-587, and quabodepistat (QBS), were measured in plasma, lung, cellular lesions, and caseum from TB-infected rabbits following single-drug treatment. Based on the previous lumping strategy [2], the model was updated for rabbits by adjusting the physiological parameters, and including granulomas through two concentric compartments, namely cellular lesion and caseum. Drug penetration into cellular lesions and caseum was assumed to occur via passive diffusion driven by pulmonary capillary blood flow. To ensure consistency between simulations and experiments, each treatment was simulated for either a single day or up to 14 days to reproduce experimental steady-state, using daily oral dosing at human-equivalent doses when available. Estimated parameters included the drug absorption rate, total body clearance, inflow rates from lung to cellular lesion and from cellular lesion to caseum, and the drug consumption rate at the site of action. A global scaling factor for tissue partition coefficients was introduced, as proposed in previous modeling frameworks [5], to modulate tissue distribution across compartments. Model fitting was performed using the nlmixr2 R package [6] (R version 4.4). Parameter uncertainty was evaluated by visual inspection of an ensemble of 200 simulations generated from drug-specific parameter uncertainty. Non-compartmental analysis was conducted in plasma, lung, cellular lesion, and caseum by computing the maximum concentration (Cmax) and area under the concentration–time curve (AUC). TA was assessed in plasma, lung, cellular lesion, and caseum at steady-state for human-equivalent doses, including the minimum bactericidal concentration to eradicate 90% viable bacteria under aerobic, nutrient-rich conditions (MBC90) and the concentration that kills 90% of bacteria in ex vivo caseum (casMBC90). We varied granuloma diameter (2–6 mm; 50% caseum) and caseum fraction (10–90%; 3.5 mm) independently to assess the sensitivity of TA to lesion geometry.
Results: The lesion-extended mPBPK model reproduced pharmacokinetic profiles for all 16 drugs across plasma, lung, cellular lesions, and caseum and non-compartmental metrics generally agreed with observed data (Cmax and AUC typically within two-fold). Estimated inflow rates from lung to cellular lesion were low for most compounds, supporting the hypothesis of slow drug penetration into granulomatous tissue. Among single agents, MXF, RPT, and diarylquinolines (BDQ, TBAJ-876, TBAJ-587) showed comparatively higher lesion exposure and TA, including in caseum for selected targets. In cellular lesion, most drugs stay near 100% efficacious time over diameter and caseum changes. Sensitivity is seen for PZA and modestly INH, EMB, and CFZ, where efficacious time drops as lesions enlarge or as caseum fraction changes. In caseum, for many drugs TA is close to zero regardless of diameter or composition changes, while TBAJ876, TBAJ587, and RPT show the broadest TA, with a few active only in narrow ranges. In regimen simulations, diarylquinoline-containing combinations, as well as DEL- and QBS-based therapies, provided greater coverage relative to MBC90 and casMBC90, consistent with their higher potency, i.e., lower MBC90 and casMBC90 values. Across regimens, compounds with low unbound fraction in caseum (including BDQ, TBAJ-587, TBAJ-876, DEL, QBS, and PRE) contributed most TA in cellular lesions and caseum.
Conclusion: Extending the mPBPK framework to explicitly represent cellular lesions and caseum provides mechanistic insight into drug penetration within TB granulomas and enables quantitative prediction of tissue-specific TA. This approach supports comparative evaluation and optimization of anti-TB drugs and multidrug regimens and represents a step toward mechanism-based prioritization of therapies aimed at sterilizing caseous lesions.
References:
References:
[1] W. H. Organization, Global Tuberculosis Report 2025. 2025.
[2] F. Reali et al., “A minimal PBPK model to accelerate preclinical development of drugs against tuberculosis,” Frontiers in Pharmacology, vol. 14, p. 1272091, 2024.
[3] A. Bustion et al. “The Kinetics of Bedaquiline Diffusion in Tuberculous Cavities Open a Window for the Emergence of Resistance”, The Journal of Infectious Diseases”, vol . 232, 3, Pages e431–e441, 2025
[4] R Savic et al., “Lesion penetration predicts tuberculosis regimen success: a validated translational tool for de-risking late-stage drug development”, PREPRINT (Version 1) available at Research Square [https://doi.org/10.21203/rs.3.rs-6890010/v1], 2025
[5] S. A. Peters, “Evaluation of a generic physiologically based pharmacokinetic model for lineshape analysis,” Clinical pharmacokinetics, vol. 47, pp. 261–275, 2008.
[6] M. Fidler et al., “Nonlinear mixed-effects model development and simulation using nlmixr and related R open-source packages,” CPT: pharmacometrics & systems pharmacology, vol. 8, no. 9, pp. 621–633, 2019.
Reference: PAGE 34 (2026) Abstr 12062 [www.page-meeting.org/?abstract=12062]
Poster: Drug/Disease Modelling - Infection