Kazuki Matsuda1,2, Chihiro Hasegawa1,2, Yasuhiro Tsuji2, Takahiko Aoyama2
1MSD K.K., 2Nihon University
Background & Objective: The drug concentration time-course of monoclonal antibody (mAb) is known to exhibit nonlinear elimination due to target-mediated drug disposition (TMDD). The TMDD model facilitates the prediction of appropriate dosages for future studies based on mechanisms of action from key drivers, such as target occupancy and drug-target complex concentrations, which are informative for assessing efficacy and/or safety. However, establishing validated assays for both the drug and the target requires considerable time and funding. This study aims to investigate the identifiability of TMDD parameters through a simulation study under various scenarios using only mAb concentration data for modeling. Unless otherwise stated, identifiability refers to practical identifiability based on clinical trial design. Methods: As case examples, TMDD models with quasi-steady state approximations for denosumab (anti-RANKL mAb) and alirocumab (anti-PCSK9 mAb) were used. Compared to denosumab, the TMDD parameters for alirocumab showed that the turnover rate (kdeg) was approximately 40 times faster and target antigen concentration (Baseline) was approximately 7 times lower, while the internalization rate constant (kint) and the steady-state constant (Kss) were similar [1, 2]. In a simulation study, a standard first-in-human single-ascending dose study was assumed as a base scenario in terms of PK sampling points, number of dose levels and participants, and an analyte measured. In the trial design, the number of participants per dose level was six and the number of PK samples was 12 points per participant (0.5, 1, 2, 4, 8, 24, 96, 168, 336, 672, 1008, 1344 hours), which covers 3 times half-lifeof general mAb (approximately 20 days). Six dose levels were set so that the middle dose results in the mean drug concentration of ~Kss. The concentration was measured as free drug. Additional trial designs varying number of dose levels (3 or 9 doses) and another analyte measured (total concentration) was also investigated to evaluate the identifiability of TMDD parameters. Furthermore, the impact of initial parameter values on identifiability was assessed by altering the initial values of one of the TMDD parameters by a factor of 10 (or 1/10). Two hundred virtual studies were replicated for each trial design using NONMEM® 7.4, evaluating relative etimation error, relative bias (R-Bias), and relative mean square error (R-RMSE). Results: In all scenarios (except some cases where dependency of initial TMDD parameter values were investigated), all runs were converged successfully. The R-Bias for the population mean values of each TMDD parameter was less than 10%, and the R-RMSEs were all below 30%. The number of dose levels primarily influenced the R-RMSE, which decreased with an increase in dose levels, whereas the R-Bias remained relatively stable without regard to the number of dose levels. When total drug concentration data were used, all runs were again converged successfully. However, the R-Bias for kdeg was >50%, while there was no apparent change for kint and Kss compared to the case using free concentrations. No apparent trend in identifiability was observed between denosumab and alirocumab case examples. Assessment of impact of initial parameter values revealed that altering kdeg’s initial value affected convergence, R-Bias, or R-RMSE, but other initial parameter values did not influence identifiability. Discussion & Conclusion: This simulation study suggests that even with only free drug concentration data, parameter estimation for the TMDD models results in no critical issues, biases, or identifiability problems regardless of TMDD parameter values assessed, with two case examples of denosumab and alirocumab. This may be because both cases resulted in clearly different shapes of concentration-time profiles between lower and higher dose levels, which enables to precisely estimate TMDD parameters. Meanwhile, using total concentration data resulted in significant bias and imprecision, particularly for estimating kdeg, likely due to different PK profiles compared to the scenario using free drug concentration data, especially in the a phase. Estimates for kint and Kss were not biased. Given that kint and Kss correspond to Vmax and Km, respectively, using the Michaelis-Menten approximation could be viable when only total concentration data is available.
[1] Sutjandra, et al. Clin Pharmacokinet, 50(12):793-807, 2011. [2] Djebli, et al. Clin Pharmacokinet, 56(10):1155-1171, 2017.
Reference: PAGE 33 (2025) Abstr 11568 [www.page-meeting.org/?abstract=11568]
Poster: Methodology - Other topics