Combining biomarkers sharpens TB drug trial signals

Tuberculosis drug trials rely on measuring how much bacteria patients cough up over time, but the common culture tests are slow and often produce missing or contaminated results. That slows research and makes early clinical trials expensive and uncertain. M. McClean and colleagues set out to test whether a newer molecular test, the Tuberculosis Molecular Load Bacterial Assay (TB-MBLA), could serve as a faster, more reliable marker in early bactericidal activity (EBA) studies, and whether combining different measures could give a clearer picture of how a drug works. They used TB-MBLA (LifeArc (R) ) on sputum samples collected from all 78 patients in the PanACEA BTZ-043 Phase Ib/IIa trial. Rather than relying on any single readout, the team also created a joint biomarker by applying probabilistic principal component analysis (pPCA) to integrate TB-MBLA with colony forming units (CFU) and time-to-positivity (TTP) culture data. The goal was to see if this combined approach could reduce data loss, reveal biological signals that single tests miss, and provide a practical framework for modelling treatment response in short EBA trials.

To evaluate markers for dose-exposure-response and exposure–response relationships, the researchers reexamined the original stage IIa dose-response and stages Ib/IIa pharmacokinetics–pharmacodynamics (PK-PD) exposure–response analyses. They applied linear and non-linear mixed models to TB-MBLA alone and to the first principal component (PC1) derived from their pPCA integration of TB-MBLA, CFU, and TTP. TB-MBLA alone did not show a detectable exposure–response effect in the PK-PD analysis across the 14-day treatment window, a contrast with the culture-based measures CFU and TTP, which did demonstrate exposure–response signals. When the team combined biomarkers into the joint PC1 marker, they observed a significant Emax exposure–response between days 0–3, though the effect was less pronounced than when using CFU and TTP individually. The study shows that pPCA can be applied as a modelling framework and that a latent joint component combining CFU and TTP can improve detection of treatment effects compared with either biomarker on its own.

These results carry practical implications for how early TB drug studies are designed and interpreted. First, the TB-MBLA RT-qPCR assay did not replace culture measures for detecting BTZ-043 exposure–response over the short 14-day window in this trial, so relying on TB-MBLA alone could miss signals that CFU and TTP catch. Second, combining biomarkers via latent variable techniques like pPCA can help overcome the well-known problem of missing or contaminated culture data and can make detection of treatment effects more robust. The authors suggest that EBA trials should be designed with attention to the drug–bacterial subpopulation axis relevant to the investigational compound, to maximise the chance of observing real effects. More broadly, latent variable and joint modelling approaches offer an effective and efficient framework for modelling Mycobacterial tuberculosis load in EBA trials and could help streamline early-phase drug development by making analyses less vulnerable to single-test limitations.