Estelle Yau (1,2), Andrés Olivares-Morales (2), Michael Gertz (2), Adam Darwich (1), Leon Aarons (1) and Kayode Ogungbenro (1)
(1) Centre for Applied Pharmacokinetic Research, University of Manchester, UK, (2) Roche Pharma and Early Development, Pharmaceutical Sciences, Roche Innovation Center Basel, Switzerland
Objectives: Physiologically-based pharmacokinetic (PBPK) models allow prediction of a drug’s pharmacokinetics (PK) by separating drug and system’s properties allowing their use for the investigation of “what if” type of questions. The multidimensional nature of PBPK models in terms of parameters and outputs generally hinders estimation of uncertain or unknown model parameters, where heterogeneous and subjective approaches for parameter estimation using PBPK models exist in the literature [1]. One possible approach to aid parameter estimation with PBPK models can be the use of Bayesian methods [2].However, before initiating parameter estimation procedures the application of parameter sensitivity analysis (SA) is generally recommended. Herein, a comprehensive sensitivity analysis was conducted to identify key parameters explaining variability/uncertainty in models for predicting drug distribution [3,4] and potentially reduce the dimensionality by excluding uninfluential parameters/tissue.
Methods: A global SA (GSA) was performed on Rodgers et al. mechanistic equations for tissue-to-plasma unbound partition coefficient (Kpu) predictions [3,4]. Sets of hypothetical strong basic, weak basic, acidic and neutral drugs (n=1000 for each drug class) were generated using realistic value ranges of physicochemical drug properties (lipophilicity [logP], plasma protein binding [fup], blood:plasma partition [BP] and acid/basic nature [pKa]). Partial rank correlation coefficients (PRCC) [5,6] were used to identify the most influential drug parameters within explored ranges on Kpu predictions. The significance of a non-zero PRCC value was tested using a two-sided Student’s t-test. Given the relationship between lipophilicity and plasma protein binding, several degrees of correlation between logP and fup were considered [7]. In addition, the sensitivity to physiological parameters was explored by incorporating 30% variability/uncertainty on different biological parameters (fractional tissue lipid volumes, fractional tissue water volumes, and acid phospholipid and proteins levels) [8].
Results: All tissues showed comparable drug parameter sensitivities to Kpu predictions for neutral and acidic compounds, while for weak and strong bases, some tissues showed distinct sensitivities that could be clustered into 4 representative tissues (adipose, heart, muscle and lung). Given the explored parameter space of hypothetical drugs, logP and fraction of drug unbound (fup) were generally the most influential parameters (and p<0.001) on Kpu predictions for different drug classes. For strong bases, BP was the most influential parameter on several tissue Kpu predictions. pKa for weak bases and pKa and fup for strong bases were the least influential parameters. For acids, fup, logP and pKa had similar sensitivity ranking to Kpu predictions. The PRCC analysis from different degrees of correlation between logP and fup showed similar results with logP remaining the most influential parameters for neutrals and weak bases, while for acids and weak bases logP was generally not the most sensitive parameter. Based on the assessment of prototypical single drugs per drug class, variability/uncertainty in tissue composition data (30% CV of input) had a limited impact on Kpu predictions (<30% CV on output) for all classes except for strong bases with low fup which could reach a CV of 45% in tissue Kpu predictions. Variability/uncertainty in acid phospholipid and in proteins levels were identified as having the most impact on Kpu predictions. This impact on Kpu predictions also propagated into volume at steady (Vss) predictions (>30% CV on Vss output).
Conclusions: Based on the GSA and within certain value ranges, if a dimensionality reduction is needed less influential parameters for Kpu predictions in each drug class might be assigned fixed values in the context of Bayesian parameter estimation. Variability in tissue composition, particularly related to acid phospholipid and protein concentrations can be influential when predicting Kpu for strong bases with low fup. This study represents the first step towards the development of a systematic Bayesian framework for PBPK parameter estimation in this project.
References:
[1] Margolskee et al. Eur J Pharm Sci (2017), 96, 610-625
[2] Tsamandouras et al. Br J Clin Pharmacol (2015), 79(1), 48-55.
[3] Rodgers & Rowland. J Pharm Sci (2005), 94(6), 1259-76.
[4] Rodgers & Rowland. J Pharm Sci (2006), 95(6), 1238-57.
[5] Marino et al. J Theor Biol (2008), 254(1), 178-96.
[6] Fenneteau et al. J Pharmacokinet Pharmacodyn (2009), 36(6):495-522.
[7] Yamazaki & Kanaoka. J Pharm Sci (2004), 93(6), 1480-94.
[8] Tsamandouras et al. Br J Clin Pharmacol (2015), 79(1), 48-55.
Reference: PAGE 28 (2019) Abstr 8979 [www.page-meeting.org/?abstract=8979]
Poster: Drug/Disease Modelling - Absorption & PBPK