Quantitative prediction of formulation-specific food effects and their population variability from in vitro data with the physiologically-based ADAM model: A case study using the BCS/BDDCS Class II drug nifedipine

Abstract

Quantitative prediction of food effects (FE) upon drug pharmacokinetics, including population variability, in advance of human trials may help with trial design by optimising the number of subjects and sampling times when a clinical study is warranted or by negating the need for conduct of clinical studies. Classification and rule-based systems such as the BCS and BDDCS and statistical QSARs are widely used to anticipate the nature of FE in early drug development. However, their qualitative rather than quantitative nature makes them less appropriate for assessing the magnitude of FE. Moreover, these approaches are based upon drug properties alone and are not appropriate for estimating potential formulation-specific FE on modified or controlled release products. In contrast, physiologically-based mechanistic models can consider the scope and interplay of a range of physiological changes after food intake and, in combination with appropriate in vitro drug- and formulation-specific data, can make quantitative predictions of formulation-specific FE including the inter-individual variability of such effects. Herein the Advanced Dissolution, Absorption and Metabolism (ADAM) model is applied to the prediction of formulation-specific FE for BCS/BDDCS Class II drug and CYP3A4 substrate nifedipine using as far as possible only in vitro data. Predicted plasma concentration profiles of all three studied formulations under fasted and fed states are within 2-fold of clinically observed profiles. The % prediction error (%PE) in fed-to-fasted ratio of C_max and AUC were less than 5% for all formulations except for the C_max of Nifedicron (%PE = −29.6%). This successful case study should help to improve confidence in the use of mechanistic physiologically-based models coupled with in vitro data for the anticipation of FE in advance of in vivo studies. However, it is acknowledged that further studies with drugs/formulations exhibiting a wide range of properties are required to further validate this methodology.

Introduction

Assessment of potential food effects (FE) on the rate and extent of absorption of orally dosed drugs is an important part of drug development particularly for poorly soluble drugs (Yasuji et al., 2012). As such the US FDA provides guidance for the consideration of FE in relation to bioavailability and bioequivalence studies (FDA, 2002). Understanding the impact of food on exposure to, and the efficacy of, a new drug as early as possible may avoid relabeling and safety issues at later stages of development. Classification systems such as the Biopharmaceutical Classification System (BCS) (Amidon et al., 1995, Custodio et al., 2008) and the Biopharmaceutical Drug Disposition Classification System (BDDCS) (Wu and Benet, 2005, Custodio et al., 2008), FeSSIF/FaSSIF solubility ratio- and Quantitative Structure–Activity Relationship (QSAR)-based approaches (Singh, 2005, Gu et al., 2007) are used to anticipate FEs during early development stages mainly because of simplicity of calculation and minimal requirements for input data. Such approaches are neither intended nor able to:

• consider the full scope and interplay of postprandial physiological changes;

• predict plasma concentration profiles in fasted and fed states;

• provide quantitative information about population variability;

• quantitatively predict the changes in AUC, C_max, and T_max (with the exception of QSARs for predicting AUC change developed by Singh (2005)).

Furthermore these approaches are based upon characteristics of the drug alone. Hence they are applicable only to Immediate Release (IR) formulations where formulation-related properties are expected to have minimal impact on the kinetics of drug absorption. Thus, in general, these approaches cannot predict the nature and extent of FE for modified/controlled release (MR or CR) formulations which may be formulation-specific (Welling, 1996). In contrast, with appropriate in vitro data, population-based mechanistic models are able to integrate all available physiological and anatomical (or system) data, and drug- and formulation-specific information to predict pharmacokinetics (PK) and pharmacodynamics (PD) not only for an ‘average’ or ‘typical’ human’ but also for a population of individuals (Jamei et al., 2009a). A wide range of food-related system changes can be incorporated as model parameters to attempt to account for the clinically observed phenomena, viz. blood flow, gastric residence time, luminal pH, bile salt concentrations, fluid volumes, etc. (Jamei et al., 2009b, Jamei et al., 2009c).

Mechanistic PBPK models have at least two major advantages over empirical models. First, they can be used in advance of information from the clinic to anticipate drug pharmacokinetics. The models can be modified iteratively as further in vitro or clinical information becomes available and can be used to inform what and when additional data are collected or measured. Second, mechanistic models provide a rational framework within which to propagate the impact of the known inter-individual variability of parameters such as gut lumen pH and bile salt concentrations, fasted–fed status, transit times, enzyme abundances, and tissue volumes; covariate models for physiological parameters can be incorporated where such information is available to build as realistic virtual individuals as possible (Jamei et al., 2009a). Many applications related to age, sex, ethnicity and disease impact have been recently summarised by Rostami-Hodjegan (2012) while a particular example of the application of such models in the area of oral absorption was provided by Darwich et al., 2012, Darwich et al., 2013 who used the models to understand and demonstrate changes to drug bioavailability following bariatric surgery in morbidly obese patients.

Mechanistic PBPK models have also been used for the quantitative prediction of FE on the PK of IR formulations based upon measured biorelevant solubility/dissolution and other in vitro and in vivo disposition parameters (Jones et al., 2006, Turner et al., 2012, Heimbach et al., 2013). However, to our knowledge, there are no reports either on the use of mechanistic models to predict FE for CR formulations or on the prediction of formulation-specific differences in FE for BCS Class II drugs based exclusively upon in vitro data.

Herein the Advanced Dissolution, Absorption and Metabolism (ADAM) (Jamei et al., 2009c) model in conjunction with the 14 organ full Physiologically-Based Pharmacokinetic (PBPK) model of the Simcyp Simulator (Fig. 1) is used to predict FE on one IR (Procardia®, an immediate release soft gelatin capsule) (Reitberg et al., 1987), and two CR formulations ([i] ADALAT OROS®, an osmotically-driven Gastrointestinal Therapeutic System (GITS) and [ii] Nifedicron, encapsulated mini tablets) (Schug et al., 2002c) with different mechanisms of drug release; results are compared with those obtained using other methods. Nifedipine (NIF), a BCS Class II (poor solubility, high permeability) and BDDCS Class II (poor solubility, high metabolism) compound, was selected as a model drug for this case study because it is one of the most extensively studied drugs in the clinic for FE due to significant differences in the nature and extent of FE with different formulations (Challenor et al., 1987, Reitberg et al., 1987, Rimoy et al., 1989, Ueno et al., 1989, Armstrong et al., 1997, Toal et al., 1997, Toal et al., 2012, Schug et al., 2002a, Schug et al., 2002b, Schug et al., 2002c, Toal, 2004, Wonnemann et al., 2006, Wonnemann et al., 2008, Anschutz et al., 2010).

An IR formulation of NIF (Procardia) is reported to have significant reduction in C_max, increase in T_max, and negligible increase in AUC when given with food (Reitberg et al., 1987) while CR formulations tend to exhibit the opposite effect (increased C_max, and increased AUC) (Schug et al., 2002c, Toal et al., 2012). Moreover, there are wide variations between different CR formulations of the rate and extent of absorption in the fasted state and even more so in the fed state (Toal et al., 2012). As a result concerns have been raised about switching between the innovator reference product Adalat OROS and other generic products without patient monitoring (Meredith, 2007, Toal et al., 2012).

The aim of this study is to provide an example of the use of mechanistic PBPK models in combination with in vitro data alone to predict variations in the PK of both the same formulation in the fasted and fed states as well as between different formulations; as far as possible clinical data were not used to parameterise the models rather only to assess model performance.