A translational toolkit to predict clinical Phase IIb/III outcomes from preclinical mouse studies
Janice JN Goh
UCSF
Objectives
Despite known drug regimens, tuberculosis (TB) treatment remains long and complex, highlighting the urgent need to identify shorter but equally efficacious regimens. Thousands of possible 3-4 drug combinations exist but clinical trials are resource intensive and only the most promising regimens can be tested. Developing regimen ranking tools to select only the most efficacious regimens for clinical testing would thus help hasten drug development. We previously developed a pharmacokinetic-pharmacodynamic (PK-PD) translational platform with bacterial dynamics that used EC50 as a portable metric between mice and humans to reliably predict short-term monotherapy outcomes in Phase IIa trials from preclinical mouse studies1. Building on this platform, we have included predictions for Phase IIb (time to stable culture conversion, (TSCC)) and Phase III (proportion of TB patients with relapse 6 months post treatment). This allows us to rank regimens by their predicted clinical outcomes.
Methods
We compiled a database of preclinical mouse PK and PD data with 2, 3 and 4 way regimens of bedaquiline, pretomanid, moxifloxacin and pyrazinamide (BPaMZ) e.g. BPaMZ, PaMZ, PaZ, MZ. PD data included mouse bacterial burden (log10CFU/lung) and relapse 3 months post-treatment for the tested drug regimens. A bacterial dynamics model along with the PK-PD model accounting for drug-drug interactions in combination was used to ensure that we could estimate the true drug effect in mice separate from the immune system2.
To better reflect long-term treatment outcomes, we split our EC50 estimates from the drug combination PK-PD models in mice into a fast (0-28 days) and slow EC50 (>28 days)4. Mouse PK-PD model parameters were then translated to clinical parameters for prediction by correction for each drug’s fraction unbound between human and mice. Simulations of mouse bacterial burden (log10CFU/lung) were then carried out for all regimens to compare and rank their efficacy.
To model relapse, we estimated a new term for bacterial regrowth (Knet) in the bacterial dynamics model using mouse CFU in lung at the end of treatment and mouse lung CFU relapse data 3 months post-treatment. The clinical bacterial burden over time with treatment was simulated using our fast-slow EC50 PKPD model for 3-6 months. After, regrowth was simulated using Knet from the remaining CFU for 6 months post-treatment to predict relapse. The proportion of patients without relapse 6 months post-treatment was used as the primary outcome to rank regimen efficacy, and model prediction was evaluated against clinical data where available.
Results
Our integrated PK-PD model successfully predicted TSCC over 8 weeks of treatment for all regimens with available clinical data as validation. Simulations suggest both PaMZ and BPaMZ were able to achieve TSCC with 4 months of treatment.
We also successfully extend our platform to predict Phase III relapse outcomes. Our model was also able to predict 6-month relapse probability post treatment (PaMZ, NC-006) as verified with VPC, showing that Knet was also a clinically translatable parameter for prediction of relapse post treatment. Further exploration of the addition of B to the regimen showed adding B would greatly improve the sterilizing activity of the regimen and have a lower proportion of patient relapse at 4 months (BPaMZ median proportion relapse 0.98 +/- 0.0203, 95% prediction interval) compared to PaMZ alone (PaMZ proportion relapse 0.84+/- 0.0664, 95% prediction interval). This suggests that BPaMZ is an efficacious combination that can shorten treatment duration by 2 months, compared to standard of care HRZE.
Conclusion
Building off our previous platform for Phase IIa trials, we have created an extended translational platform that also predicts long term Phase IIb and Phase III outcomes. By accounting for drug-drug interactions in the PD model, and also modeling EC50 in both fast and slow bacterial growth states, we were able to predict long term outcomes and relapse 6 months post-treatment with good accuracy. This helps us to better rank the best performing regimens. BPaMZ and PaMZ as a case study, we have demonstrated our translational platform can predict Phase II and III outcomes prior to actual trials, allowing us to better prioritize the regimens most likely to succeed. Future validation with other drug combinations will be carried out as well.
References
- Ernest, J. P. et al. Translational predictions of phase 2a first-in-patient efficacy studies for antituberculosis drugs. bioRxiv 2023.01.18.524608 (2023) doi:10.1101/2023.01.18.524608.
- Zhang N. et al. Mechanistic Modeling of Mycobacterium tuberculosis Infection in Murine Models for Drug and Vaccine Efficacy Studies. Antimicrob. Agents Chemother. 64, e01727-19 (2020).
- Wicha, S. G., Chen, C., Clewe, O. & Simonsson, U. S. H. A general pharmacodynamic interaction model identifies perpetrators and victims in drug interactions. Nat. Commun. 8, 1–11 (2017).
- Muliaditan, M. & Della Pasqua, O. Evaluation of pharmacokinetic-pharmacodynamic relationships and selection of drug combinations for tuberculosis. Br. J. Clin. Pharmacol. 87, 140–151 (2021).