Smarter mouse studies could speed tuberculosis drug testing

Tuberculosis, caused by Mycobacterium tuberculosis (Mtb), remains a major global health challenge, and developing new treatment regimens is essential to improve outcomes. Early, non-clinical testing helps decide which candidate regimens should move forward, and the relapsing mouse model (RMM) is a common tool for that step. But RMM studies are animal-, labor-, and time-intensive, demanding significant resources during non-clinical development. With many regimens in the pipeline, researchers led by Alexander Berg sought ways to make RMM testing leaner without sacrificing the quality of the information they provide. Rather than running every possible experimental design in the lab, the team used computational simulations to create many “virtual” studies that mimicked groups of mice treated for selected durations with control and hypothetical anti-TB regimens. By analyzing those simulated outcomes with model-based methods, the investigators could compare what was put into the simulations with what their analysis would estimate, focusing on a key metric used to compare regimens: time to 95% cure probability (T 95 ). This approach allowed them to test different study designs rapidly and see which attributes could be changed to reduce animal use while preserving useful results.

The team simulated relapse outcomes from virtual studies that represented groups of mice receiving various treatment durations with control and hypothetical anti-TB regimens, then subjected the simulated data to model-based analysis. They compared the known input values used to generate the simulations with the model estimates of T 95 and assessed both bias and precision across competing designs. Using this stochastic, simulation-based approach, the researchers identified alternative RMM study designs that required 28% fewer mice yet still produced low bias in estimates and maintained precision for T 95 estimation within +/− 1-2 weeks for most regimens. The method tested how changing study attributes affected the ability to recover the true time to 95% cure probability, highlighting which design changes had minimal impact on decision-critical estimates. By evaluating many virtual scenarios before running animal studies, the approach provided evidence-based guidance on how to allocate resources more efficiently in non-clinical development.

The findings suggest that RMM studies can be redesigned to use substantially fewer animals while still giving informative, decision-ready data. By keeping bias low and T 95 precision within a narrow window (+/− 1-2 weeks), the alternative designs evaluated could help researchers prioritize the most promising regimens without unnecessary animal use or wasted effort. This matters for programs managing large pipelines of candidate regimens, because reducing the number of mice per study cuts costs, labor, and time while preserving the statistical quality needed to compare treatments. The simulation-based workflow also offers a repeatable way to test study options before committing to resource-intensive experiments, promoting improved animal stewardship and more thoughtful resource utilization during non-clinical development. Ultimately, using virtual studies and model-based analysis can help ensure RMM results remain reliable for decision making while aligning studies with ethical and practical constraints.