Alessandro Ravoni1, Enrico Mastrostefano1, Davide Moretti1, Elia Onofri1, Paolo Tieri1, Filippo Castiglione1,5, Aristides Dokoumetzidis2,3, Evangelos Karakitsios1, Salvatore D’Agate1,4, Umberto Villani1,4, Oscar Della Pasqua1,4
1Istituto per le Applicazioni del Calcolo “Mauro Picone” (IAC), CNR National Research Council of Italy, 2Department of Pharmacy, University of Athens, 3Pharma-Informatics Unit, Athena Research Center, 4Clinical Pharmacology and Therapeutics Group, University College London, 5Medicine Department, Pulmonary Systems Medicine, University of Florida
Introduction: The primary objective of this investigation was to integrate a well-established agent-based model (ABM) of the immune response (C-ImmSim, [1, 2]), a recent physiologically based pharmacokinetic (PBPK) model of antitubercular drug distribution in the human lung [3], and a pharmacokinetic-pharmacodynamic (PKPD) model of the pharmacodynamic effects [4]. This integration aims to characterise the combined effects of the immune system (IS) and drug treatment in patients with active tuberculosis (TB). Specifically, we model the interaction between Mycobacterium tuberculosis (Mtb), the host immune response, and pharmacological treatment to capture the dynamic effects of drug exposure on bacterial clearance. As a case study, we focus on rifampicin (RIF), a first-line antitubercular drug [5], evaluating its efficacy in monotherapy within the integrated framework. Methods: C-ImmSim is a programme that simulates the response of the IS using a stochastic ABM, where agents represent immune cells and key molecules interacting over time. Mtb is modelled as an agent that can transition between different states: extracellular, intracellular (i.e., within macrophages), fast-growing, slow-growing, and dormant. Bacteria replicate within macrophages and can exit upon cell lysis, while some may enter a dormant state, becoming undetectable to the IS. Dormant bacteria can later reactivate, potentially leading to disease relapse [6]. Using a system of differential equations, the PBPK model tracks the concentration of the drug after administration. Specifically, it describes the time course of unbound RIF concentration across different lung compartments, namely intracellular water, extracellular water, and the lesion. Time-integrated concentration values derived from the PBPK model provide the average drug exposure at each discrete time step of the ABM simulation. In the ABM model, extracellular water concentration affects Mtb in the tissue, intracellular concentration corresponds to the drug inside non-infected cells, while the concentration of the drug inside infected macrophages is evaluated through the lesion compartment. Both slow- and fast-growing Mtb are influenced by this concentration, as available data suggest a shift in their ratios over the course of the infection (i.e., varying metabolic phenotypes), which prevents separate characterization of each subpopulation. In vitro estimates of bactericidal activity (i.e., from experiments excluding the IS) were used to define a death probability in the ABM, dependent on both drug concentration and Mtb state. A calibration process was used to estimate the drug’s differential effects on each bacterial subpopulation, ensuring alignment with independent studies of RIF monotherapy in humans [7]. Results: Using the Approximate Bayesian Computation (ABC) method [8], we successfully calibrated our hybrid model to reproduce realistic epidemiological distributions of disease states, including active, latent, dead, and cleared infections. Virtual patients with active TB were then treated with RIF to match observed bacterial load, allowing us to estimate the drug’s impact on different bacterial phenotypes in a way consistent with real-world data. Additionally, we quantified the IS contribution to infection control under treatment. Our findings show that while IS and RIF play distinct roles in bacterial clearance, the drug enhances the IS’s killing capacity compared to untreated cases. Conclusion: The integration of these models enables us to estimate RIF’s effect on different Mtb phenotypes and quantify the IS’s role in controlling TB infection under treatment. Future work will focus on advanced calibration and multidrug therapy simulations to reproduce real patient data from standard treatment regimens.
[1] Rapin N, Lund O, Bernaschi M, Castiglione F. Computational immunology meets bioinformatics: the use of prediction tools for molecular binding in the simulation of the immune system. PLoS One. 2010; 5(4):e9862. [2] Stolfi P, Castiglione F, Mastrostefano E, Di Biase I, Di Biase S, Palmieri G, Prisco A. In-silico evaluation of adenoviral COVID-19 vaccination protocols: Assessment of immunological memory up to 6 months after the third dose. Front Immunol. 2022; 13:998262. [3] Karakitsios E, Dokoumetzidis A. Extrapolation of lung pharmacokinetics of antitubercular drugs from preclinical species to humans using PBPK modelling. J Antimicrob Chemother. 2024;79(6):1362-1371. Erratum in: J Antimicrob Chemother. 2024;79(9):2411. [4] Muliaditan M, Della Pasqua O. Evaluation of pharmacokinetic-pharmacodynamic relationships and selection of drug combinations for tuberculosis. Br J Clin Pharmacol. 2021; 87(1):140-151. [5] World Health Organization (WHO). Treatment of tuberculosis guidelines. World Health Organization, ISBN 9789241547833. [6] Mastrostefano E, Ravoni A, Onofri E, Tieri P, Castiglione F. Harnessing computational models to uncover the role of the immune system in tuberculosis treatment. 2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pages 3725–3732. [7] Gyselen A. Rifampicin in the retreatment and original treatment of advanced pulmonary tuberculosis. Bull Int Union Tuberc. 1970; 43:60-3 [8] Klinger E, Rickert D, Hasenauer J. pyABC: distributed, likelihood-free inference. Bioinformatics. 2018; 34(20):3591-3593.
Reference: PAGE 33 (2025) Abstr 11642 [www.page-meeting.org/?abstract=11642]
Poster: Methodology - New Modelling Approaches