Thi Ngoc Anh Vu 1,2, Yun Min Song 3, Hung Manh Pham 1, Soyoung Lee 1,2, Sang Kyum Kim 1,2, Jae Kyoung Kim 3,4,5, Jung-woo CHAE 1,2,6, Hwi-yeol Yun 1,6
1 College of Pharmacy, Chungnam National University (Daejeon, Republic of Korea), 2 Institute of Drug Research and Development, Chungnam National University (Daejeon, Republic of Korea), 3 Biomedical Mathematics Group, Institute for Basic Science (Daejeon, Republic of Korea), 4 Department of Mathematical Sciences, Korea Advanced Institute of Science and Technology (Daejeon, Republic of Korea), 5 Department of Medicine, College of Medicine, Korea University (Seoul, Republic of Korea), 6 Department of Bio-AI Convergence, Chungnam National University (Daejeon, Republic of Korea)
Introduction
In conventional IVIVE–PBPK (In Vitro-In Vivo Extrapolation and Physiologically Based Pharmacokinetic) modeling, hepatic intrinsic clearance (CLint) is typically derived using the standard quasi–steady state approximation (sQSSA), which assumes that total enzyme concentration (ET) is negligible relative to the Michaelis constant (Km) [1-7]. This assumption is generally valid under physiological steady-state conditions [1-4]; however, it may become unreliable when enzyme abundance changes dynamically, such as during CYP3A4 induction by rifampicin [1, 5, 6]. Under induction, the ET/Km ratio increases and the classical Michaelis–Menten-based scaling can introduce systematic bias in exposure predictions. To address this limitation, we incorporated a refined total quasi–steady state approximation (tQSSA)-based intrinsic clearance framework into IVIVE–PBPK modeling [1, 5-7].
Objective
This study aimed to integrate a refined tQSSA-based intrinsic clearance equation into a Simcyp PBPK platform and quantitatively evaluate its predictive performance compared with the conventional sQSSA approach under both baseline and rifampicin-induced CYP3A4 conditions.
Methods
The refined intrinsic clearance framework was incorporated into a Simcyp-based IVIVE–PBPK by embedding C/C++ functions inside Lua scripting to create a custom Ordinary Differential Equation step overriding the default sQSSA clearance. Under baseline physiological conditions, intrinsic clearance was adjusted using a tQSSA-based correction accounting for the finite relationship between ET and Km. Under induction conditions, enzyme-mediated clearance was further adjusted using the fraction metabolized (fm) and fold-induction factor (Eind), allowing dynamic modulation of metabolic capacity. Two CYP3A4 substrates were evaluated: midazolam (IV 1 mg and 2.5 mg) and nifedipine (PO 20 mg). Simulations were performed in Simcyp under baseline conditions and rifampicin 600 mg once-daily induction conditions. Model performance was assessed using concentration-based metrics, including root mean square error (RMSE), average absolute fold error (AAFE), 2-fold prediction accuracy.
Results
Under baseline conditions, both formulations demonstrated comparable predictive performance. For midazolam IV (1 mg) across the four published datasets (Phimmasone et al., 2001; Kharasch et al., 2011; Shin et al., 2016; and Kharasch et al., 2004), AUC values were slightly closer to observations with tQSSA than with sQSSA, with RMSE ranging from 0.111–0.283 for sQSSA and 0.108–0.245 for tQSSA, and overlapping AAFE values with identical 2-fold prediction accuracy of 88.2–100% for both formulations. A similar pattern was observed for nifedipine 20 mg PO under baseline conditions (Holtbecker et al., 1996), where overall performance metrics were comparable (RMSE 0.433 vs 0.440; AAFE 2.07 vs 2.08). AUC showed a modest improvement toward observed values with tQSSA (179.775 vs 165.074; observed 221.187), whereas t1/2 remained overpredicted by both models across midazolam and nifedipine datasets.
In contrast, under rifampicin induction (600 mg q.d.), clear divergence emerged. For midazolam 1 mg IV (Phimmasone et al., 2001; Kharasch et al., 2004), sQSSA showed increased prediction error (AAFE up to 2.396; 2-fold prediction accuracy 28.6%), whereas tQSSA reduced AAFE to 1.352 and improved 2-fold prediction accuracy to 92.9%, with RMSE decreasing from 0.408 to 0.204. AUC was markedly lower with sQSSA (7.448 vs 14.560), while tQSSA was closer to observations (13.234). For midazolam 2.5 mg IV (Shin et al., 2016), sQSSA yielded lower AUC (22.118 vs 39.345), whereas tQSSA closely matched observed exposure (41.148), with AAFE decreasing from 2.080 to 1.403 and 2-fold prediction accuracy improving from 37.5% to 100%. A similar pattern was observed for nifedipine 20 mg PO under induction (Holtbecker et al., 1996). AUC improved from 7.538 to 20.804 vs 18.099 observed, with AAFE decreasing from 2.52 to 1.32 and 2-fold prediction accuracy increasing from 11.1% to 88.9%. Improvements under induction were primarily reflected in AUC and Cmax rather than t1/2.
Discussion
Under baseline physiological conditions, sQSSA and tQSSA yield similar predictions, consistent with the ET ≪ Km assumption. However, under CYP3A4 induction where enzyme abundance increases substantially, the classical sQSSA formulation introduces systematic underprediction of systemic exposure. Incorporation of the refined tQSSA framework consistently reduced RMSE and AAFE, shifted AFE toward unity, and substantially improved 2-fold prediction accuracy. Notably, AUC improved from approximately 0.4–0.6-fold with sQSSA to near unity (~0.9–1.15-fold) with tQSSA across both midazolam and nifedipine datasets. These findings extend our previous “Beyond Michaelis–Menten” study conducted in PK-Sim, established the mechanistic advantage of tQSSA over sQSSA [6], by demonstrating consistent external performance gains within the Simcyp platform under clinically relevant induction scenarios, supporting its broader application in dynamic PBPK modeling.
References:
References
[1] Kim, J.K. & Tyson, J.J. Misuse of the Michaelis–Menten rate lawfor protein interaction networks and its remedy. PLoS Comput.Biol (2020) 16, 1–21.
[2] Back, H.m., Yun, H.y., Kim, S.K. & Kim, J.K. Beyond the Michaelis–Menten: accurate prediction of In vivo hepatic clearance for drugs with low KM. Clin Transl Sci (2020) 13(6), 1199–1207.
[3] Choi, B., Rempala, G.A. & Kim, J.K. Beyond the Michaelis- Mentenequation: accurate and efficient estimation of enzyme kinetic parameters. Sci Rep. (2017) 7(1), 17018.
[4] Segel, L.A. & Slemrod, M. The quasi- steady- state assumption: acase study in perturbation. SIAM Rev (1989) 31, 446–477.
[5] N. A. T. Vu, Y. M. Song, Q. T. Tran, et al. Beyond the Michaelis–Menten: Accurate Prediction of Drug Interactions Through Cytochrome P450 3A4 Induction. Clin Pharmacol Ther. (2023) 113(5), 1048–105.
[6] J. Lee, Y. M. Song, H.-y. Yun et al. Beyond Michaelis-Menten: A modified enzyme kinetics equation improves drug metabolism prediction in bottom-Up PBPK modeling. Eur J Pharm Sci. (2025) 1:214:107286.
[7] N. A. T. Vu, Y. M. Song, S. K. Kim, et al. Beyond the Michaelis–Menten: Evaluation of a tQSSA-Based IVIVE Approach for Predicting In Vivo Intrinsic Clearance From Hepatocyte Assays, CPT Pharmacometrics Syst Pharmacol. (2026) 15(2):e70169.
Reference: PAGE 34 (2026) Abstr 12090 [www.page-meeting.org/?abstract=12090]
Poster: Drug/Disease Modelling - Other Topics