Animal models have long served as the preclinical gatekeeper for tuberculosis drug development, yet they systematically fail to predict human pharmacokinetics, efficacy, and toxicity—a translation gap that has left TB drug discovery inefficient, costly, and ethically burdensome.
Quantitative systems pharmacology (QSP) modeling frameworks enable researchers to evaluate host-directed therapies (HDTs) against Mycobacterium tuberculosis by mechanistically linking cellular-level processes—such as autophagy induction via metformin—to disease progression outcomes [Mehta et al., 2022, Mehta et al., 2023; Model Code]. Rather than relying on animal models to screen HDT candidates, researchers can now simulate host-pathogen interactions across biological scales—from drug pharmacokinetics in specific lung compartments to immune cell responses and bacterial survival—using only in vitro data and mechanistic knowledge. This multiscale approach allows TB researchers to screen HDT targets, evaluate combination strategies with conventional antibiotics, and predict treatment outcomes before any animal work begins.
Physiologically-based pharmacokinetic (PBPK) models have similarly proven capable of predicting drug exposure in hard-to-access tissues like the central nervous system, eliminating the need for invasive animal sampling procedures that were once the only way to quantify drug concentrations in cerebrospinal fluid during tuberculosis meningitis treatment [Saleh et al., 2023; Mehta et al., 2022, Mehta et al., 2024]. The advantage is particularly stark for TB: the heterogeneous pulmonary microenvironment created by infection fundamentally alters drug disposition in ways animal models fail to capture, yet PBPK frameworks now mechanistically capture these pathophysiological changes.
The emerging standard can be “a priori in silico” workflow: in vitro data and mechanistic knowledge feed PBPK and QSP models before any animal studies, generating human-relevant predictions that then guide the design of smaller, targeted animal experiments—or replace them entirely [Mehta et al., 2024]. This is not speculation. For a disease where animal models have historically failed to predict human efficacy and toxicity, and where resistance threatens to undermine treatment gains, shifting preclinical decision-making to mechanistic, human-relevant in silico platforms offers both scientific rigor and ethical advantage. The infrastructure exists. The evidence is building. What remains is adoption.