TB drug rankings shift with scale, detection limits, and bacterial load

Tuberculosis (TB) begins when people inhale Mycobacterium tuberculosis (Mtb), and the infection often leads to lung granulomas — complex, spheroidal clumps of immune cells and bacteria. Inside many granulomas is caseum, a central necrotic region where bacteria can be quarantined and made harder for drugs to reach. That structural complexity helps explain why TB treatment requires multiple antibiotics given for long periods, and why different patients and even different granulomas inside the same lung respond differently. Clinical trials that test which drug combinations shorten treatment or reduce toxicity are essential but expensive and difficult, and they can’t easily reveal the biological reasons behind wide variation in outcomes. To tackle this problem, Denise E. Kirschner and colleagues used HostSim, a whole-host, mechanistic, multi-scale computational model of Mtb infection. HostSim follows infection and immune responses across molecular, cellular, tissue, organ, and whole-host scales. The researchers built a heterogeneous virtual cohort to run virtual clinical trials, letting them explore treatment responses in many distinct simulated hosts without the time and cost of standard trials. They also added drug behavior into the model to better mimic real treatments.

The team extended HostSim by newly integrating pharmacokinetics / pharmacodynamics so drug concentrations and effects could be simulated over time and space inside the virtual hosts. With that capability, they modeled commonly prescribed TB antibiotic regimens, including HRZE and BPaL, and tracked both experimental and clinical measurements across scales. This approach allowed the researchers to identify which simulated hosts and which individual granulomas showed the most improvement with a given regimen, and to probe which biological mechanisms drove differences in treatment outcome. By comparing simulated outcomes to published drug rankings, they were able to virtually recreate several rankings from the literature. Crucially, they found that many methods for ranking treatment efficacy are strongly influenced by how “improvement” is defined and, in some cases, by the detection threshold of CFU. Other differences in rankings depended on the initial bacterial burden of hosts or granulomas. Their results indicate that different metrics of regimen optimality can be orthogonal — measuring different aspects of success — and that this may explain seemingly contradictory findings across prior studies.

These findings matter for how researchers and clinicians interpret drug comparisons and design trials. If rankings of regimens change depending on spatial scale (whole host versus individual granulomas), the detection limit of CFU tests, or the starting bacterial load in a patient, then a single ranking or single outcome measure may miss important trade-offs. HostSim’s virtual trials suggest that investigators should think carefully about which definitions of improvement they use, and consider reporting multiple metrics so that a regimen’s strengths and weaknesses are clear. For public health campaigns aiming to shorten treatment or reduce toxicity, the work implies that optimal choices may differ across patients or even across lesions within a patient. Using multi-scale, mechanistic models like HostSim can help untangle these sources of heterogeneity, guide which laboratory and clinical measurements matter most, and offer a way to test hypotheses about why some trials give different answers than others. Overall, the study suggests a path toward more nuanced and mechanistically informed evaluations of TB regimens.