Thanakorn Vongjarudech 1,2, Iván Noreña 3, Norbert Heinrich 3, Fred Njeleka 4, Daniel A. Mapamba 4, Bariki Mtafya 4, Thomas P.C. Dorlo 1, Mats O. Karlsson 1, Lilian T. Minja 4,5,6, Christina Manyama 4, Elin M. Svensson 1,2
1 Department of Pharmacy, Uppsala University (Uppsala, Sweden), 2 Department of Pharmacy, Pharmacology and Toxicology, Radboud University Medical Center (Nijmegen, The Netherlands), 3 Institue of Infectious Diseases and Tropical Medicine, LMU University Hospital, Munich (Munich, Germany), 4 National Institute for Medical Research-Mbeya Medical Research Centre (NIMR-MMRC) (, Tanzania), 5 Swiss Tropical and Public Health Institute (, Switzerland), 6 University of Basel (, Switzerland)
Objectives:
Tuberculosis (TB) remains a major global health crisis, with an estimated 10.7 million new cases and 1.23 million deaths annually [1]. Current response monitoring relies heavily on sputum culture. However, the Time to Positivity (TTP) in liquid culture (MGIT), a standard biomarker for bacterial burden, is limited by slow turnaround times (up to 42 days) [2], contamination, and a focus on replicating bacteria, potentially obscuring non-replicating population dynamics.
To bridge this gap, novel biomarkers that capture distinct bacterial subpopulations in sputum are being investigated. The Mycobacterium Tuberculosis – Molecular Bacterial Load Assay (TB-MBLA), a reverse transcriptase-qPCR targeting 16S rRNA to quantify viable bacteria [3], and PATHFAST TB Lipoarabinomannan (LAM) Ag test [4], a cell-wall glycolipid measured via immunoassay, offer rapid readouts compared to culture [4]. Despite their potential, the quantitative longitudinal relationship between these novel markers and the MGIT-TTP, specifically regarding baseline correlation, longitudinal dynamics, and censoring due to limits of quantification, remains poorly defined [5]. Furthermore, understanding the utility of biomarkers for identifying drug effects and differentiating between regimens is crucial for optimizing future TB treatment strategies.
This analysis leverages biomarker data from Phase II clinical trials of novel regimens, including the drug candidates sutezolid and delpazolid, with the aim of developing a multivariate joint modeling framework. The specific objectives were to:
1. Construct a joint pharmacometric model linking the longitudinal dynamics of MGIT-TTP, TB-MBLA, and the PATHFAST TB LAM Ag test, explicitly accounting for multi-type censoring.
2. Quantify the inter-biomarker correlation structure to determine the complementary information TB-MBLA and the PATHFAST TB LAM Ag test, particularly when MGIT-TTP, are censored or unavailable.
3. Characterize the exposure-response (ER) relationship of sutezolid and delpazolid simultaneously across all three biomarkers, thereby identifying the most informative metric for treatment efficacy.
Methods:
Data were pooled from 149 participants with drug-sensitive pulmonary TB enrolled in the SUDOCU [6] and DECODE [7] trials. Participants received various regimens of sutezolid (600–1200 mg), delpazolid (400–1200 mg), or control, alongside standard background therapy. Sputum was collected weekly from Week 0 to Week 12, yielding 3,483 TTP (MGIT), 1,605 TB-MBLA (LifeArc), and 1,638 PATHFAST TB LAM Ag test samples.
Biomarker data were log10-transformed. Individual PK exposures (AUC0-24h and Cmax at Week 2) were derived from published models [6,7]. Dynamics were characterized using linear or biphasic models, with a sigmoidal switching function. A multivariate joint model quantified inter-biomarker relationships via a full-block inter-individual variability (IIV) matrix on baseline burdens and elimination rates. IIV was modeled exponentially, with Box-Cox transformations applied where necessary to correct skewness [8]. Residual unexplained variability was described using additive or inverse-proportional error models on the log10 scale. Censored data (MGIT-TTP > 2.78 log10h [5], TB-MBLA < 2.11 or >7 log10copies/reaction, and PATHFAST TB LAM Ag test < 10 log10pg/mL) were handled using the M3 method [9,10]. Baseline demographics were evaluated as covariates. Drug effects were tested on slopes using linear or Emax functions, with the maximal effect of delpazolid, based on the observed exposure (AUC0-24h 50 mg·h/L; Cmax 11 mg/L). Parameters were estimated using SAEM with the interaction in NONMEM 7.5 supported by the PsN toolkit [11], followed by IMP for final objective function value (OFV), and the Fisher Information Matrix calculations of Relative Standard Errors (RSE). Results: Participants had a median age of 35 years (range:20-59), a weight of 53 kg (range:40.3-85), 23.5% female, and a chest X-ray severity score (Ralph score) of 65 (range:2-117) at baseline. The joint model successfully characterized the simultaneous dynamics. The estimated median baseline bacterial burden was 2.09 (0.8%RSE) log10h for MGIT-TTP, 7.07 (1.7%RSE) log10copies/reaction and 4.20 (2.4%RSE) log10pg/mL for PATHFAST TB LAM Ag test. Baseline random effects showed a moderate-to-strong correlation: MGIT-TTP:TB-MBLA -0.52 (21.5%RSE), TTP:PATHFAST TB LAM Ag test -0.59 (15.8%RSE), and TB-MBLA:PATHFAST TB LAM Ag test 0.75 (31.9%RSE). MGIT-TTP and TB-MBLA exhibited biphasic trajectories, whereas the PATHFAST TB LAM Ag test followed a linear trajectory. MGIT-TTP transitioned from an initial fast rate (0.036 (5.8%RSE) log10h/day) at Day 8.6 (5%RSE) to a slower rate (0.010 (3.8%RSE) log10h/day). TB-MBLA transitioned at Day 35 (1.5%RSE) from a slower initial rate of 0.070 (2.1%RSE) to a faster rate of 0.086 (2.5%RSE) log10copies/reaction/day, potentially reflecting assay saturation at high baseline burdens or a compressed dynamic range due to log transformation. The PATHFAST TB LAM Ag test demonstrated a constant decline rate of 0.033 (5.1%RSE) log10pg/mL/day throughout the treatment period. Significant inter-biomarker correlations were identified on the rates of decline. The initial killing rates of MGIT-TTP and TB-MBLA were strongly correlated, 0.71 (19.7%RSE), while the late-phase (sterilization) killing rates showed strong correlation across all three biomarkers [MGIT-TTP:TB-MBLA 0.67 (11.7%RSE), MGIT-TTP:PATHFAST TB LAM Ag test (0.74 (10.3%RSE), and TB-MBLA:PATHFAST TB LAM Ag test 0.76 (9.6%RSE)]. Age was the only significant covariate (-0.04% decline rate/year). E-R analysis identified significant relationships for both drugs (ΔOFV-10.95). Sutezolid Cmax enhanced killing in both early and late phases (10.7%, 17.5%RSE at a median of 0.881mg/L). Delpazolid exposure AUC0-24h significantly influenced the late-phase slope (16.4% increase, 13.1% RSE at a median of 36mg·h/L). Model performance was comparable across PK metrics, with minimal separation between the final model and alternatives using sutezolid Cmax on early phase only and delpazolid Cmax (ΔOFV-10.79) or AUC0-24h (ΔOFV-10.11) on late phase only. Notably, when evaluated at these same median exposures, drug effect estimates in the joint model were lower than separate MGIT-TTP-only analyses (sutezolid: 14.3%, 61%RSE; delpazolid: 27.3%, 53%RSE), likely due to the additional arms in the pooled dataset. However, the joint framework substantially improved precision, reducing RSEs from >50% to <20%. Biomarker-specific scaling factors were not significant, suggesting a shared pharmacological effect across assays.
Conclusion:
This first joint framework for MGIT-TTP, TB-MBLA, and PATHFAST TB LAM Ag test demonstrates that TB-MBLA and PATHFAST TB LAM Ag test are robust biomarkers with comparable E-R sensitivity to MGIT-TTP, despite having less than half the number of clinical samples. By quantifying strong inter-marker correlations and improving parameter precision, this approach supports integrating TB-MBLA and PATHFAST TB LAM Ag test into Phase IIb/III trials to complement culture-based endpoints and accelerate regimen evaluation.
References:
Funding:
The SUDOCU and DECODE clinical trials were conducted by the PanACEA consortium, which is part of the EDCTP2 programme supported by the European Union (grant number TRIA2015-1102). Additional support was provided by the German Federal Ministry of Education and Research (BMBF; 01KA1701) and the Netherlands Organisation for Health Research and Development (ZonMw). This work was funded, in part, by a Veni project (E.M.S., Project No. 09150161910052) financed by the Dutch Research Council (NWO), by the Swedish Heart-Lung Foundation (Hjärt-Lungfonden; E.M.S., project no. 20230736) and by LigaChem Biosciences. T.D. was supported by the Swedish Research Council (VR 2022-01251). LifeArc provided the TB-MBLA kits, and PHC Corporation provided the PATHFAST TB LAM Ag test assay kits used in this study.
References:
1. World Health Organization. Geneva: World Health Organization; 2025.
2. Diacon AH et al. Eur J Clin Microbiol Infect Dis. 2010 Dec:29:1561–5.
3. LifeArc. 2018 Aug.
4. Akinaga A et al. J Clin Microbiol. 62:e00629-24.
5. Dufault SM et al. Int J Antimicrob Agents. 2025 Feb:65:107404.
6. Heinrich N et al. Lancet Infect Dis. 2025 Nov:25:1208–18.
7. Minja LT et al. Lancet Infect Dis. 2025 Nov:25:1219–29.
8. Petersson KJF et al. Pharm Res. 2009 Sep:26:2174–85.
9. Beal SL. J Pharmacokinet Pharmacodyn. 2001 Oct:28:481–504.
10. Wijk M et al. CPT Pharmacometrics Syst Pharmacol. 2025 Jun:14:1042–9.
11. Lindbom L et al. Comput Methods Programs Biomed. 2005 Sep:79:241–57.
Reference: PAGE 34 (2026) Abstr 11902 [www.page-meeting.org/?abstract=11902]
Poster: Oral: Drug/Disease Modelling - Other Topics