Model Description

We designed a novel parsimonious physiologically based pharmacokinetics model to describe anti-TB drug ADME in the mouse. Starting from the works of (Jones and Rowland-Yeo, 2013; Stader et al., 2019; Lee et al., 2020), we iteratively lumped compartments that are marginally involved in pulmonary TB infection to streamline the model diagram. As a result of an extensive validation step, adipose, brain, bone, heart, muscles, pancreas, and skin were grouped together in a compartment called “other”, whilst interstitial spaces were embedded into corresponding tissues and organs. The lumped compartment closely resembles the clustering identified by (Yau et al., 2023) to reduce PBPK models based on rat tissue composition. The final reduced model includes nine compartments. Eight of them are physiologically- derived, venous blood, arterial blood, lung, kidney, liver, spleen, gut, and lumped tissues, whilst one is treatment-specific to accommodate the oral route of administration. In contrast to (Ryu et al., 2022), we kept systemic bloodstreams and lungs separated to gain PK insights into the target tissue and fit the experimental data. The gut, spleen, liver, and kidney, which are pivotal or contributor tissues for absorption, distribution, and elimination phases, were explicitly modeled. First-order reactions for absorption and elimination processes and for drug exchange among compartments were found to best support the PK datasets. Mechanisms of gut-reabsorption did not lead to substantial fit improvements and thus were discarded for parsimony. Rodgers and Rowland’s method was chosen for tissue-to-plasma partition coefficient computation since comparative studies demonstrated its enhanced performances across both drug classes (>70% compounds within threefold of experimental values) and tissues (from 66.1% in the brain to 92.7% accuracy in the heart) (Jones et al., 2011; Graham et al., 2012). Under the experimental setting of plasma and lung sample collection, the proposed design satisfies the global structural identifiability criterium for the two parameters to estimate, i.e., absorption and total clearance rates, mitigating one of the major drawbacks ascribed to whole-body PBPK models (Chiş et al., 2011; Läer and Khalil, 2011; Peters and Dolgos, 2019). In addition, compared to a reference full-body PBPK model (Peters, 2008; Stader et al., 2019), our ODE system can be simulated more than three times faster, which results in a substantial speed up that enables larger what-if investigations.