Prediction of Oral Drug Bioavailability: from Animal-Based Extrapolation Towards the Application of Physiologically-Based Pharmacokinetic Modelling and Simulation

Prediction of oral drug bioavailability: from animal-based extrapolation towards the application of physiologically-based pharmacokinetic modelling and simulation

A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the Faculty of Medical and Human Sciences

2016

Andrés Miguel Olivares Morales

Manchester Pharmacy School

Table of contents

Table of contents ...... 2

List of Figures ...... 7

List of Tables ...... 11

List of Abbreviations ...... 13

Abstract ...... 18

Declaration ...... 19

Acknowledgements ...... 22

Chapter 1: General Introduction ...... 23

1.1 Introduction ...... 24

1.2 Oral bioavailability...... 26

1.3. Factors influencing oral bioavailability ...... 27

1.3.1 Oral drug absorption ...... 27

1.3.1.1 Intestinal permeability and absorption ...... 28

1.3.1.1.1 Anatomy of the human GI tract ...... 28

1.3.1.1.1.1 Structure of the small intestine ...... 29

1.3.1.1.1.2 Structure of the large intestine ...... 30

1.3.1.1.1.3 The unstirred water layer (UWL) ...... 31

1.3.1.1.2 Mechanisms for permeation of the intestinal membrane ...... 32

1.3.1.1.2.1 Transcellular passive diffusion ...... 33

1.3.1.1.2.2 Paracellular absorption ...... 33

1.3.1.1.2.3 Transporter-mediated absorption ...... 34

1.3.1.1.2.3.1 Intestinal efflux transporters ...... 34

1.3.1.1.2.3.2 Intestinal uptake transporters ...... 36

1.3.1.1.3 In vitro and in vivo investigation of permeability ...... 37

1.3.1.1.3.1 In vivo determination of permeability...... 37

1.3.1.1.3.2 in vitro and in silico methods for the determination of permeability...... 38

1.3.1.1.4 Permeability and the fraction absorbed ...... 41

1.3.1.1.5 BCS and the rate limiting steps for drug absorption ...... 41

1.3.1.2 Solubility and dissolution ...... 42

1.3.1.2.1 Solubility ...... 42

1.3.1.2.2 Drug release ...... 43

1.3.1.2.2 Dissolution ...... 44

1.3.1.2.3 Physiological factors that alter drug dissolution and solubility .. 45

1.3.2 Hepatic and intestinal first pass metabolism ...... 46

1.3.2.1 Hepatic first-pass metabolism and biliary excretions ...... 46

1.3.2.2 Intestinal first-pass metabolism...... 48

1.4 Prediction of human oral bioavailability ...... 49

1.4.1 Direct extrapolation from animal models ...... 49

1.4.2 Physiologically-based pharmacokinetic modelling and simulation ...... 54

1.4.2.1 PBPK models in drug development ...... 55

1.4.2.2 PBPK models for absorption and bioavailability ...... 56

1.5 Project aims and objectives ...... 60

1.6 List of manuscripts and author contribution statement ...... 62

1.8 References ...... 65

Chapter 2: Animal versus human oral bioavailability: do they correlate? ...... 82

Chapter 3: The Use of ROC Analysis for the Qualitative Prediction of Human Oral Bioavailability from Animal Data ...... 83

Chapter 5: Translating Human Effective Jejunal Intestinal Permeability to Surface-Dependent Intrinsic Permeability: a Pragmatic Method for a More Mechanistic Prediction of Regional Oral Drug Absorption ...... 85

Chapter 6: Development of a novel simplified PBPK absorption model to explain the higher relative bioavailability of the OROS formulation of oxybutynin ...... 86

6.1 Abstract ...... 87

6.2 Introduction ...... 88

6.3 Materials and Methods ...... 90

6.3.1 PBPK model development ...... 90

6.3.1.1 Development of OXY’s disposition model ...... 91

6.3.1.1.1 Disposition model parameters and parameter estimation ...... 93

6.3.1.2 Expansion of the mSAT model for mechanistic bioavailability predictions ...... 96

6.3.1.2.1 Drug Transit ...... 98

6.3.1.2.2 Dissolution and solubility ...... 99

6.3.1.2.3 Intestinal absorption ...... 100

6.3.1.2.4 Enterocyte compartments and intestinal metabolism ...... 101

6.3.2 OXY’s oral PK simulations and relative bioavailability predictions ...... 105

6.4 Results ...... 108

6.4.1 OXY’s disposition parameter estimation ...... 108

6.4.2 Mechanistic prediction of OXY’s oral pharmacokinetics ...... 110

6.4.3 Relative bioavailability between IR and OROS formulation ...... 111

6.5 Discussion ...... 116

6.6 Conclusion ...... 122

6.7 Acknowledgements ...... 122

6.8 References ...... 123

Chapter 7: Concluding remarks and future perspectives ...... 130

7.1 References ...... 137

Appendix A1: Supplementary Material for Chapter 2 ...... 138

A1.1 Supplementary Tables ...... 139

Appendix A2: Supplementary Material for Chapter 3 ...... 140

A2.1 Supplementary Tables ...... 141

A2.2 References ...... 148

Appendix A3: Supplementary material for Chapter 4 ...... 149

A3.1 Method employed for the calculation of the relative bioavailability and its 90% confidence interval (CI)...... 150

A3.2 Calculation method for the intrinsic clearance from in vivo clearance data 151

A3.3 Supplementary Tables ...... 153

A3.4 Supplementary Figures ...... 157

A3.5 Further discussion ...... 164

A3.5.1 Discussion on Impact of the intestinal P-gp distribution on the bioavailability of CR formulations...... 164

A3.5.2 Discussion on the analysis of the possible CYP3A4/P-gp interplay .... 164

A3.6 References ...... 167

Appendix A4: Supplementary material for Chapter 5 ...... 172

A4.1 Non-linear regression for the data reported by Wilson (1967) ...... 173

A4.2 Estimated segmental SA for the reference human intestine ...... 174

A4.3 Recalculation of the regional Peff values from their original references ..... 175

A4.3.1 Triamcinolone acetonide and hydrocortisone (Schedl 1965) ...... 175

A4.3.2 Hydrochlorothiazide, atenolol, furosemide, cimetidine and salicylic acid (Sutcliffe et al, 1988)...... 176

A4.3.3 Griseofulvin, ranitidine, paracetamol and talinolol (Gramatté, et al. 1994-1996) ...... 176

A4.4 mSAT model development...... 177

A4.4.1 Structure of the mSAT model ...... 177 5

A4.4.2 Optimization of the small intestinal transit time for the mSAT model 178

A4.4.3 Comparison of the mSAT model with alternative transit models ...... 179

A4.5 Regional fabs predictions from the mSAT model for solution and MR formulation...... 183

A4.6 Method for the application of the Peff,int approach to current mechanistic absorption models ...... 184

A4.7 References ...... 188

Appendix A5: Supplementary Material for Chapter 6 ...... 190

A5.1 STEP function implementation ...... 191

A5.2 mSAT luminal fluid dynamic model ...... 193

A5.3 System-related parameters for the extended mSAT model ...... 198

A5.4 Derivation of some of the drug-related parameters employed for the simulations ...... 200

A5.4.1 Calculation of the pH dependent solubility and ionization ...... 200

A5.4.2 Calculations of fmCYPj from in vitro data ...... 200

A5.4.3 In vitro release profile of the OROS formulation ...... 201

A5.4.4 Calculation of the isomer-specific intrinsic clearance ratio ...... 202

A5.4.5 Summary of the drug-dependent parameters employed for the simulations and their sources ...... 203

A5.5 Supplementary figures ...... 206

A5.6 NONMEM code for IV infusion fit ...... 208

A5.7 References ...... 212

Word count: 54,280

List of Figures

Page numbers are given as: thesis page number (publication page number) Figure 1.1 Schematic representation of the processes defining 26 oral bioavailability (F) Figure 1.2 Structure of the small intestinal membrane intestinal 30 membrane. Figure 1.3 Schematic representation of how compounds can 32 cross the intestinal membrane Figure 1.4 Schematic representation of principal transporters 35 expressed in membranes (apical and basolateral) of the enterocytes Figure 1.5 Schematic representation of the luminal processes 43 occurring after the oral administration of a solid dosage form. Figure 1.6 Generic structure of a whole body PBPK disposition 54 model. Each tissue is represented by a compartment. Figure 1.7 Schematic representation of the ADAM model 58 implemented within SimCYP Figure 1.8 Chemical structure of oxybutynin and its main 60 metabolite Figure 2.1 Plot of oral bioavailability (F) in animal species vs. 82 (282) oral bioavailability in humans (in percentage). Figure 2.2 Plots for the linear regression analysis by separated 82 (287) by species (in percentages) Figure 2.3 Box plots of median animal/human bioavailability 82 (288) ratios (RA/H) and interval between animal and human oral bioavailability. Figure 2.4 Plots for the linear regression analysis by separated 82 (288) by ion class (in percentages) Figure 2.5 Plot of the linear regression analysis for the general 82 (289) dataset, animal vs. human oral bioavailability. Figure 2.6 Plots for the linear regression analysis by classified 82 (289) by species (in percentages) Figure 3.1 Threshold based predictions of human oral 83 (722) bioavailability from animal data. Figure 3.2 (a) Pie chart of the distribution of the oral 83 (724) bioavailability data points employed for the analysis by species. (b) Venn diagram of the relationship between oral bioavailability data points for rat, dog and NHP. Figure 3.3 Averaged ROC curve for the human versus animal 83 (724) bioavailability dataset for all the preclinical species combined

Figure 3.4 Averaged ROC curves for the human versus animal 83 (725) bioavailability dataset by preclinical species Figure 3.5 Impact of the FP/FN cost ratio on the determination 83 (726) of the optimal thresholds by Eq. 3.4

Figure 3.6 Sensitivity and specificity as a function of Fanimal 83 (727) thresholds for rat (a), dog (b) and NHP (c). Figure 3.7 ROC curve for rat predictions of very low human 83 (727) bioavailability (Fhuman≤ 20%) dataset Figure 3.8 Number of compounds and BDDCS class 83 (728) distribution for rat (a), dog (b) and NHP (c) as a function of the outcome of the threshold based model. Figure 4.1 Schematic representation of the study design tree. 84 (35) Figure 4.2 Analysis of the relative bioavailability between CR 84 (37) and IR formulations of CYP3A4 substrates.

Figure 4.3 Impact of release rate (formulation) and CLint,CYP3A4 84 (38) on AUC (A), fa and FG (B) for non-P-gp substrates. Vmax,CYP3A4 was fixed at 2500 pmol/min/mg whereas the Km,CYP3A4 was varied (scenario IIa in Table 4.1).

Figure 4.4 Impact of release rate (formulation) and CLint,P-gp on 84 (39) AUC (A), fa and FG (B) for non CYP3A4 substrates. Jmax,P-gp was fixed at 300 pmol/min whereas the Km,P- gp was varied (scenario IIIa in Table 4.1). Figure 4.5 Impact of release rate (formulation), CLint,CYP3A4, and 84 (39) CLint,P-gp on AUC (A), fa and FG (B). Vmax,CYP3A4 was fixed at 2500 pmol/min/mg whereas the Km,CYP3A4 was varied (scenario Va in Table 4.1). Figure 4.6 The impact of release rate and CLint,CYP3A4 on the 84 (40) relative bioavailability (%) for BCS class 1 compounds. Figure 5.1 Illustrations of the changes on intestinal mucosal 85 (1179) surface area estimated by the different methods.

Figure 5.2 Upper panel, comparison between jejunal Peff,int 85 (1184) calculated by: a M1 (cylindrical SA), b M2 (mSA according to Wilson’s method) and c M3 (mSA according to Helander and Fändriks’ method). Lower panel, prediction of ileal absorption clearance (permeability clearance) employing jejunal Peff and ileal surface area using: d M1, e M2 and f M3. Figure 5.3 Comparison between different small-intestinal transit 85 (1185) models to the describe SITT data.

Figure 5.4 fabs (%) predictions using Peff values from Table 5.2 85 (1186) and the mSAT model.

Figure 5.5 Simulated Peff to fabs relationship for the mSAT 85 (1187) model.

Figure 5.6 Bar chart of the predicted fabs (overall and regional) 85 (1187) using the mSAT model and Peff,int for a subset of representatives drugs from Table 5.2 (colonic absorption was allowed). Figure 6.1 Schematic representation of the PBPK model 94 employed for OXY Predictions. Figure 6.2 OXY’s disposition model fit to the 5 mg IV infusion 110 data Figure 6.3 Model predictions of OXY oral pharmacokinetics 111 (racemic) after a multiple dose administration. Figure 6.4 mSAT model prediction of the pharmacokinetic of R- 113 OXY after the administration of three 5 mg IR formulations (A) and one 10 mg OROS formulation (B). Figure 6.5 mSAT predicted segmental and oral bioavailability 114 fractions for R-OXY IR and OROS formulations Figure 6.6 mSAT model prediction of the pharmacokinetic of R- 116 OXY after the administration of a 10 mg OROS formulation using two different permeability approaches. Figure A3.1 Impact of release rate (formulation) and CLint, CYP3A4 157 on AUC (A), fa and FG (B) for non-P-gp substrates. Vmax, CYP3A4 was fixed at 500 pmol/min/mg whereas the Km, CYP3A4 was varied (scenario Ia in Table 4.1). Figure A3.2 Impact of release rate (formulation) and CLint, CYP3A4 158 on AUC (A), fa and FG (B) for non-P-gp substrates. Km, CYP3A4 was fixed at 50 µM whereas the Vmax, CYP3A4 was varied (scenario Ib in Table 4.1). Figure A3.3 Impact of release rate (formulation) and CLint, CYP3A4 159 on AUC (A), fa and FG (B) for non-P-gp substrates. Km, CYP3A4 was fixed at 1 µM whereas the Vmax, CYP3A4 was varied (scenario IIb in Table 4.1).

Figure A3.4 Impact of release rate (formulation) and CLint, P-gp 160 (efflux) on AUC (A), fa and FG (B) for non-CYP3A4 substrates. Km, P-gpwas fixed at 150 µM whereas the Jmax, P-gp was varied (scenario IIIb in Table 4.1).

Figure A3.5 Impact of release rate (formulation), CLint, CYP3A4, and 161 CLint, P-gp on AUC (A), fa and FG (B). Km, CYP3A4 was fixed at 1 µM whereas the Vmax, CYP3A4 was varied, CLint, P-gp was fixed to 2 µL/min (scenario IVb in Table 4.1).

Figure A3.6 Impact of release rate (formulation), CLint, CYP3A4, and 162 CLint, P-gp on AUC (A), fa and FG (B). Jmax, P-gp was fixed at 300 pmol/min whereas the Km, P-gp was varied (scenario VIa in Table 4.1).

Figure A3.7 Impact of release rate (formulation), CLint, CYP3A4, and 163 CLint, P-gp on AUC (A), fa and FG (B). Km, P-gp was fixed at 150 µM whereas the Jmax, P-gp was varied, CLint, CYP3A4 was fixed to 2500 µL/min/mg (scenario Vb in Table 4.1). Figure A4.1 Nonlinear fit of an exponential model to Wilson’s 173 (1967) data (black solid circles). Figure A4.2 Schematic representation of the minimal Segment 178 Absorption and Transit (mSAT) model Figure A4.3 Comparison between different small-intestinal transit 180 models to describe SITT data. Figure A4.4 Simulated mass transfer along the intestinal segments 181 of the mSAT model.

Figure A4.5 Bar chart of the simulated overall and regional fabs 183 using the mSAT model and the permeability values from Table II (in the manuscript), when colonic absorption was allowed. Figure A4.6 Bar chart of the simulated overall and regional fabs 184 using the mSAT model and the permeability values from Table II (in manuscript) for a hypothetical CR formulation, when colonic absorption was allowed.

Figure A5.1 Illustration of the implementation of the discrete 193 tablet movement in the mSAT model using the STEPn function Figure A5.2 Schematic representation of the luminal fluid 195 dynamic model implemented within the mSAT model. Figure A5.3 mSAT fluid dynamic model predictions vs. observed 196 data Figure A5.4 Simulated luminal volumes in the different segments 197 of the mSAT model after the administration of 240 mL of water. Figure A5.5 In vitro release profile of R/S-OXY 10mg OROS 202 formulation. Figure A5.6 mSAT model prediction of the pharmacokinetic of S- 206 OXY after the administration of three 5 mg IR formulations (A) and one 10 mg OROS formulation (B). Figure A5.7 mSAT predicted segmental and oral bioavailability 207 fractions for S-OXY IR and OROS formulations.

List of Tables

Page numbers are given as: thesis page number (publication page number). Table 1.1 Previously reported interspecies oral 52 bioavailability comparisons. Table 2.1 Inclusion criteria for studies 82 (281) Table 2.2 Points extracted from Grass and Sinko plot 82 (282) Table 2.3 Interspecies oral bioavailability (F) for the 82 (283) selected compounds. Table 2.4 Linear regression analysis, afe and 82 (287) animal/human oral bioavailability ratio. Table 3.1 Definitions and Formulae for the Evaluation of 83 (723) the Binary Classification System Table 3.2 Area Under the ROC Curve for Animal Models 83 (725) Table 3.3 Cost Analysis Derived Optimal Thresholds for 83 (725) Fanimal and its Corresponding Evaluation Metrics Table 3.4 Cost Analysis Derived Optimal Thresholds for 83 (726) Fanimal and its Corresponding PPV and NPV

Table 3.5 Alternative Thresholds for Fanimal and its 83 (727) Corresponding Evaluation Metrics

Table 4.1 Different scenarios evaluated for CLint,CYP3A4 84 (36) CLint,P-gp for all BCS classes as a function of release rate. Table 4.2 Evaluated parameters and values. 84 (37) Table 5.1 Segmental SA Employed for the Calculation of 85 (1181) Peff,int from Peff Values (Open Perfusion System) Table 5.2 Intestinal Permeability Values (Double-Balloon 85 (1183) Technique) Employed for the Prediction of fabs Using the mSAT Model

Table 5.3 Regional Peff,int Values (Open Perfusion System) 85 (1184) and Prediction of the Ileal Absorption Clearance Estimated by Three Different Methods

Table 5.4 Predicted fabs Using the mSAT Model and the 85 (1186) Peff,int (Double-Balloon) Values from Table 5.2 Table 6.1 System-related parameters used in the mSAT 103 model Table 6.2 OXY’ drug-related parameters employed for the 108 simulations (racemic mixture and isomers) Table 6.3 Estimated OXY disposition parameters from the 109 IV infusion fit.

Table 6.4 Summary of the mSAT predicted vs. observed 115 pharmacokinetic parameters for OXY formulations (IR and OROS) Table A1.1 Supplementary information for the animal vs 139 human oral bioavailability comparison Table A2.1 Provisional BDDCS class for the drugs 141 employed in the study. Table A2.2 BDDCS class for the drugs employed in the 142 study. Table A2.3 Drug categories for high bioavailability drugs 146 (Fhuman ≥ 50%) according to the threshold based classification. Table A2.4 Drug categories for low bioavailability drugs 147 (Fhuman < 50%) according to the threshold based classification. Table A3.1 Input parameters for the Simcyp simulations 153 Table A3.2 Relative bioavailability studies between CR and 154 IR formulation reported in the literature. Table A3.3 Clearance and relative bioavailability values 156 employed for the comparison of the simulations with the observed data. Table A4.1 Segmental mSA estimated by the three different 174 methods for a reference man. Table A4.2 Physiological input parameters for the mSAT 182 model. Table A4.3 Mean (± standard deviation (SD)*) regional 186 surface area expansion factors for Method 3a

Table A4.4 Estimated mean regional Peff values (double 187 balloon technique) for the drugs listed in Table II using the SAEF method. Table A5.1 Mean residence time for the different segments 191 of the MSAT model Table A5.2 Estimated parameters for the luminal fluid 195 dynamics within the mSAT model Table A5.3 System-related parameters employed for the in 198 the mSAT model Table A5.4 CYP isoform-specific intrinsic clearance in 201 human liver microsomes (HLM) derived from recombinant microsomal data. Table A5.5 OXY’s drug-related parameters employed for the 204 simulations (racemic mixture and isomers)

List of Abbreviations aafe: absolute average fold error ACAT: Advanced Compartmental Absorption and Transit ADAM: Advanced Dissolution Absorption and Metabolism ADME: absorption, distribution, metabolism and elimination afe: average fold error

AGP: α1-acidglycoprotein AUC: area under the curve BBB: blood-brain barrier BCRP: breast cancer resistance protein BCS: Biopharmaceutical Classification System BDDCS: Biopharmaceutical Drug Disposition Classification System BE: bioequivalence BSA: body surface area BW: body weight Caco-2: Colorectal adenocarcinoma-2 cells CAT: Compartmental Absorption and Transit CCC: concordance correlation coefficient CES2: Carboxylesterase 2 CI: confidence interval CL: clearance

CLabs: Absorption clearance

CLint: intrinsic clearance

Cmax: maximum plasma/blood concentration CO: cardiac output COL: colon CPGA: 2-cyclohexyl-2-phenylglycolic acid CR: controlled release CSAEF: constant surface are expansion factor CV: coefficient of variation CYP: cytochrome P450 DDI: drug-drug interaction

DEOB: N-desethyloxybutynin DJF: duodeno-jejunal flexure DMSO: dimethyl sulfoxide Do/Dn: dose number DR: dissolution rate DUO: duodenum EMA: European Medicines Agency ER: extraction ratio F: Oral bioavailability fa /fabs: fraction of the dose absorbed in the gastrointestinal tract

Fanimal: oral bioavailability in animals species FDA: Food and Drug Administration

FG: fraction of the dose absorbed that escapes gut wall first pass metabolism

FH: fraction of the dose absorbed that escapes hepatic first pass metabolism

Fhuman: oral bioavailability in humans FN: false negative FOCE-I: first-order conditional with interaction maximum likelihood estimation method FP: False positive

Frel: relative bioavailability GI: Gastrointestinal HBA: hydrogen bond acceptor HDB: hydrogen bond donor HIM: human intestinal microsomes HIV: human immunodeficiency virus HLC: human liver cytosol HLM: human liver microsomes HMG-CoA: 3-hydroxy-3-methylglutaryl-coenzyme A ICV: Ileocecal valve IIV: inter-individual variability ILE: ileum IND: Investigational New Drug IR: immediate release

ISEF: inter-system extrapolation factor IV or iv: intravenous IVIVC: in vitro-in vivo correlations IVIVE: in vitro-in vivo extrapolation J: Youden’s index JEJ: jejunum LC-MS/MS: liquid chromatography with tandem mass spectrometry detection LIG: Loc-I-Gut LogD: logarithm of the distribution coefficient LogP: logarithm of the partition coefficient (octanol:water) LSI: length of the small intestine M&S: modelling and simulation M1: method 1 M2: method 2 M3: method 3 MCT: monocarboxylate transporters MDCK: Madin-Darby canine kidney cells MPPGL: milligrams of microsomal protein per gram of liver MR: modified-release MRP: multidrug resistance-associated protein MRT: mean residence time mSA: Mucosal surface area mSAT: minimal segmented absorption and transit MV: Small intestinal microvilli

MVcolon: colonic microvilli MW: molecular weight NDA: New Drug Applications NHP: Non-human primates NPV: Negative predictive value OATP: organic anion polypeptide transporter OCT: organic cation transporters ODE: ordinary differential equation OrBiTo: Oral Biopharmaceutical Tools

OROS: osmotic controlled-release oral delivery system OXY: oxybutynin PAMPA: parallel artificial membrane permeability assay PBPK: Physiologically-based pharmacokinetic(s) PC: plicae circulares or circular folds PD: pharmacodynamic(s) PDE: partial differential equation

Peff,int: intrinsic intestinal effective permeability

Peff: intestinal effective permeability (jejunal) PEPT1: peptide transporter 1 P-gp: P-glycoprotein PK: Pharmacokinetics PMAT: plasma membrane monoamine transporter PPV: positive predictive value PSA: polar surface area QSAR: quantitative structure-activity relationship R2: coefficient of determination RO5: Lipinski’s rule of 5 RoB: rest of the body ROC: receiver operating characteristics RSE: relative standard error RUV: residual unexplained variability SA: surface area SAEF: surface area expansion factors SD: standard deviation SFM: Segregated Flow Model SGF: simulated gastric fluid SI: small intestine(al) SIF: simulated intestinal fluid SITT: small intestinal transit time ST: stomach t: time tA: animal, high/low oral bioavailability threshold

16 tH: human, high/low oral bioavailability threshold TN: true negative TNR: true negative rate TP: true positive TPR: true positive rate UGT: UDP-Glucoronosyltransferases US: United States USP: United Stated Pharmacopeia UWL: unstirred water layer V: volume VL: Intestinal villi

Vss: Volume of distribution at steady state

Abstract

THE UNIVERSITY OF MANCHESTER Abstract of thesis submitted by Andrés Miguel Olivares Morales for the degree of Doctor of Philosophy, entitled: “Prediction of oral drug bioavailability: from animal-based extrapolation towards the application of physiologically-based pharmacokinetic modelling and simulation” Month and year of submission: February 2016

The majority of drugs available on the market are intended to be administered through the oral route. To achieve the desired therapeutic effect, an orally administered drug must first reach the systemic circulation and then its site of action. The fraction of the administered drug that reaches the systemic circulation is known as oral bioavailability and it is the product of the absorption and first-pass metabolism processes occurring in both the GI tract and the liver. The factors controlling bioavailability are manifold –both drug and physiologically related - and their complex interplay is key to defining a drug’s oral bioavailability. In drug discovery and development it is therefore pivotal to anticipate and understand the bioavailability of a drug candidate; a far from simple task, considering the multifactorial nature of the process. For that reason, the overall aim of this thesis was to provide different modelling and simulation (M&S) strategies that can be used for the prediction of oral bioavailability that can be of use in drug discovery and development. The first part of this thesis was focused on the evaluation of the use of bioavailability data obtained from pre-clinical species as a predictor of the human value, in a more traditional approach. In particular, the aim was to evaluate models that can quantitatively and qualitatively provide a relationship between animal and human bioavailability, by analysing trends in a large bioavailability dataset. This section demonstrated that although pre-clinical species cannot quantitatively predict bioavailability, the data obtained from them can be used for qualitative prediction of the human value. Nevertheless, such a modelling approach does not provide a mechanistic rationale of the factors affecting the bioavailability differences. Consequently, the second part of this thesis was focused on such mechanistic predictions. Particularly, we investigated the impact that drug release patterns can have on drug absorption and intestinal first pass metabolism, taking into account the physiological differences observed across the length of the human gastrointestinal (GI) tract. These release patterns are suspected to lead to bioavailability differences due to changes in the first-pass metabolism, especially for CYP3A substrates. Therefore we investigated this phenomenon applying a physiologically-based pharmacokinetic (PBPK) M&S approach: firstly, from the discovery point of view, using PBPK models in a prospective fashion to investigating the drug-related factors that might lead to such differences and secondly, from the development point of view, to predict the mechanistic differences in absorption and metabolism of oxybutynin, a drug known for its higher bioavailability when formulated as a modified release (MR) product. The latter was done by developing and applying a novel simplified PBPK model to predict such differences. The results of this work showed that the intestinal metabolism can be significantly reduced when having MR formulations of CYP3A substrates which, in some cases, can lead to higher relative bioavailability. Additionally, this thesis provided novel methods and models that have the potential to improve bioavailability predictions when using PBPK models, in particular for drugs formulated as MR

Declaration

No portion of the work referred to in this thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning.

Copyright i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the “Copyright”) and he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. iii. The ownership of certain Copyright, patents, designs, trade marks and other intellectual property (the “Intellectual Property”) and any reproductions of copyright works in the thesis, for example graphs and tables (“Reproductions”), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and commercialization of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy (see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=487 ), in any relevant Thesis restriction declarations deposited in the University Library, The University Library’s regulations (see http://www.manchester.ac.uk/library/aboutus/regulations ) and in The University’s policy on Presentation of Theses.

To Natalia

Acknowledgements

I would like to acknowledge and declare my gratitude to:

First and foremost my PhD supervisors, Professors Amin Rostami and Leon Aarons, for giving me the opportunity to work with them (the brightest minds I ever met) and for sharing their wisdom and experience with me. Also for giving me the chance to work to become an independent researcher and fulfil one of my dreams, especially for their encouragement and freedom during all this period that allowed me to give my best to build this thesis. In particular, I would like to thank Leon for inviting me to Ducie on Fridays where we shared good moments with friends and colleagues that made this process more manageable, especially in difficult times.

The Centre for Applied Pharmacokinetic Research (CAPKR) of the University of Manchester, which was my second home and my second family during this PhD, for integrating me and Natalia and for providing excellent discussions and good moments along this process. Especially I would like to thank to Drs Alex Galetin and Kayode Ogungbenro for their advice during all this period.

Lesley Wright, for her kind support and help with all the administrative matters related to the PhD.

My friends and colleagues in CAPKR, for all the good moments we enjoyed together. I hope we keep in touch and continue our friendship wherever we are.

My Chilean friends: Erik, Paulina, Christian, Natalia, Patricio, Anita, Hugo, Fiorella and of course, my friend and colleague Pablo. For your friendship, kindness and good moments we enjoyed during this period far from home.

My wife Natalia, because without your unconditional love and support this would not have been possible, thank you and I love you.

Finally, God and my parents and family, for supporting me in all the decisions I made during this period and for taking care of us during this time we have been abroad.

Chapter 1: General Introduction

1.1 Introduction

Several processes define the fate of a drug within the body after its administration. The study of such processes, i.e., absorption, distribution, metabolism and elimination (ADME), and the relationship between the drug administration and its concentration over time is what we define as pharmacokinetics (PK). On the other hand, the study of the relationship between drug in the body, its concentration and its therapeutic effect is what is known as pharmacodynamics (PD)[1]. To attain the desired therapeutic effect a drug must first reach its site of action. Given the practical issues associated with measuring drug concentrations at the site of action, however, plasma or blood concentrations are generally used as a valuable biomarker or even as a surrogate to establish the relationship between the drug administration, its concentration and its therapeutic effect, i.e. the PK/PD relationship [1, 2].

The development of a new drug involves a lengthy and resource intensive process whereby thousands of drug candidates are screened and tested in order to find a suitable lead that demonstrates the required balance between safety and efficacy for the treatment of a particular disease or condition. This process involves several stages of data-gathering, each one with its own objective, that are intended to support the drug’s safety and efficacy claims that might allow the submission of a drug for marketing-approval. These stages can be divided into: drug discovery, pre-clinical research, clinical research, regulatory submission and post-approval monitoring [3- 5].

The drug development process is however associated with large failure rates, where the success rate of a compound entering phase I is decreased significantly as we move forward to the next stage of development [6, 7]. Lack of adequate pharmacokinetics properties has been regarded as one of the factors responsible for such attrition rates, particularly lack of ability to deliver adequate drug concentration into the drug’s target [8]. Over the last decades, however, the increase understanding of the factors controlling a drug’s pharmacokinetic properties at early stages of the drug development has led to a decrease in the attrition rates attributed to pharmacokinetic issues; currently less than 10% of the causes of attrition are attributed to pharmacokinetics [6, 7]. This reduction comes through the use of better 24

in vitro profiling tools that allow early characterization of the drug ADME properties and the use of modelling and simulation (M&S) for the early detection of candidates with low chances of success [9, 10]. Nevertheless, there are safety and efficacy reasons for attrition that can be linked to underlying pharmacokinetic issues, for instance: a poor intestinal absorption of an oral candidate that might lead to variable and erratic pharmacological response, or the presence of a severe metabolic drug- drug interaction (DDI) that might be associated with an increased risk of developing toxicity [6, 7, 11, 12].

One of the key pharmacokinetic properties of a drug is its absolute oral bioavailability (F). This parameter is required in many stages of the drug development arena such as for the determination of the first in man (FIM) dose in early clinical stages or for the safety and toxicity assessment and it is an important regulatory requirement [13]. Moreover, low absolute bioavailability is associated with high interindividual variability (IIV) in the drug’s pharmacokinetic response that can possible lead to a variable or poor therapeutic outcome [14]. On the other hand the relative oral bioavailability (Frel) between formulations, i.e., clinical vs. commercial formulation, is of great interest during early drug development to evaluate the risk that a change in the formulation can entail in the latter development phases. Although it is not a strictly regulatory requirement, its determination is pivotal for such risk assessment, however this can lead to the conduction of lengthy and sometimes unnecessary clinical trials that can slow down the drug development process[15]. Therefore, it is pivotal to be able to make an accurate prediction of such parameters as early as possible in the development setting in order to optimize the resources in drug development and avoid the conduction of unnecessary clinical trials. This thesis is intended to provide different tools and examples that can be used for the predictions of oral drug bioavailability, by applying a variety of pharmacokinetic M&S approaches: from the direct correlation of bioavailability measurements in pre-clinical species to that in human, to the use and development of physiologically-based pharmacokinetic models combined with in vitro-in vivo extrapolation (IVIVE) for the mechanistic understanding of the oral bioavailability. For the latter approach, the focus will be put on the understanding of the impact that formulations can have on oral bioavailability and the interplay between intestinal

absorption and intestinal metabolism. Unless otherwise stated, absolute oral bioavailability will be simply referred as oral bioavailability throughout this thesis.

1.2 Oral bioavailability

Due to the ease of administration and its cost-effectiveness the oral route is, by and large, the preferred route of administration for the majority of drugs available in the market. Yet, the oral absorption process is far from simple and depends upon numerous factors [16, 17]. After oral administration the amount of drug that reaches the systemic circulation and becomes available to reach the site of action, relative to the total administered dose, is defined as oral bioavailability (F). Experimentally oral bioavailability can be measured by comparing the dose-normalized exposure (area under the concentration-time curve (AUC)) of an oral (PO) administration vs. an intravenous reference (IV) assuming that the systemic clearance (CLsys) remains similar in both administrations (Equation 1.1) [1, 18].

AUC Dose Equation 1.1 F po iv AUC Dose iv po

Figure 1.1. Schematic representation of the processes defining oral bioavailability (F). fa or fabs is the fraction of the dose that becomes absorbed in the GI tract; EG is the fraction of the absorbed drug that is eliminated in the GI wall, whereas FG is the fraction surviving such elimination; EH is the fraction of the dose that reaches the

portal circulation that is eliminated in the liver and FH is the fraction of the dose surviving such elimination.

Mechanistically, however, oral bioavailability encompasses several steps that occur after the drug administration (Figure 1.1). The first step is the gastrointestinal (GI) absorption, where the fraction of the administered dose that eventually becomes absorbed into the intestinal wall is known as fa or fabs. The second step involves the transfer of the drug from the absorption site to the systemic circulation. During this process the absorbed drug can be victim of pre-systemic extraction in both the intestinal membrane (EG) and the liver (EH). The fractions of the absorbed drug that survives extraction in the gut wall and the liver are known as FG and FH, respectively. Given that the aforementioned processes occur as a sequence, oral bioavailability can be mechanistically defined by Equation 1.2 [19].

F f  F  F Equation 1.2 a G H

1.3. Factors influencing oral bioavailability

The factors defining drug oral bioavailability are manifold. Broadly speaking they can be classified into two major categories: drug-related and physiologically-related factors. Drug related factors can be the drug’s physicochemical, biochemical and biopharmaceutical properties, whereas the physiological factors are inherent to the subjects/species where bioavailability is investigated [17, 20]. The following sections describe such factors and their impact on drug bioavailability from the perspective of the terms defined in Equation 1.2.

1.3.1 Oral drug absorption

The extent of oral drug absorption into the GI tract depends upon numerous factors [17]. In 1995 Amidon and co-workers published a seminal paper that stablished the basis for the mechanistic prediction of oral drug absorption [21]. This work was the culmination of several investigations related to the correlation between drug-related properties and their extent of absorption, such as the development of the concept of absorption potential (AP) that correlated solubility, dose and lipophilicity (LogP)

with fa [22]. The application of mass balance approaches to describe the dynamics and kinetics of drug absorption in the GI tract, by rationalizing Fick’s First Law of diffusion in the GI membrane, led to an equation that can describe the extent of drug absorption over time (Equation 1.3) [21-26]

t M ()t Equation 1.3 fa() t P w C w dAdt   A Dose 0 where the triple integral is accounting for the changes in drug concentration along the surface area (A) and the time (t) in the intestinal tube, M(t) is the drug mass absorbed over time, Pw is the permeability of the drug in the intestinal wall and Cw is the drug concentration next to the intestinal wall. Pw and Cw are parameters that can vary according to the position and time within the GI tract [21]. This equation highlighted the three key factors that define the extent of drug absorption: the drug’s ability to permeate the intestinal membrane, defined by the Pw coefficient, the drug’s concentration at the absorption site in the GI membrane, largely dependent upon its solubility and the GI luminal conditions, and the residence time of the drug in the GI lumen [21]. Each one of these factors is discussed below.

1.3.1.1 Intestinal permeability and absorption

1.3.1.1.1 Anatomy of the human GI tract In order to understand the permeation of a drug through the GI tract membrane it is pivotal to have a general understanding of the anatomical and physiological factors affecting drug absorption in the human GI tract. The human GI can be divided into different segments: mouth cavity, oesophagus, stomach, small intestine, and large intestine. Drug absorption can occur in almost all the segments of the GI tract, however stomach, small intestine and colon are considered as the most relevant for drug absorption [27, 28]. Even though drug absorption can occur in both stomach and colon, it is believed that the contribution of these segments to the overall absorption (fa) is minimal compared to that of the small intestine[29]. This proposition is mainly applicable to drugs formulated as immediate release (IR) whose absorption is believed to occur shortly after the drug administration (i.e., solutions, suspensions, tablets, etc.); the mean residence time of 28

a drug in the small intestine is around 3.32 hours and can vary between 2 to 5 hours in the fasted state [30-32]. In the cases where the absorption is not complete in the small intestine given the aforementioned timeframe, for example, when the drug is to be released for a prolonged period of time, the colon might play a key role in drug absorption [33, 34]. The colonic residence time varies between 12 to 36 hours and is highly variable. This prolonged time might be sufficient enough for a drug to eventually become absorbed [27, 30, 32, 35-37]. Albeit absorption from the stomach is considered to be minimal compared to that in the intestine, the gastric emptying time of a dosage form can significantly affect the drug absorption rate. Gastric emptying of a drug varies depending upon the drug formulation and prandial state. It is assumed that on average the gastric residence time is around 15 minutes for solutions administered in the fasted state. In contrast, in the fed state the gastric emptying of a non-disintegrating solid dosage form could take up to nine hours [30, 38-42]

1.3.1.1.1.1 Structure of the small intestine The small intestine is the major portion of the GI tract accounting for around 81% of the total intestinal length and is the major site for digestion and absorption of nutrients and xenobiotics. It consists of three major regions: duodenum, jejunum and ileum whose lengths are approximately 8%, 37% and 55% of the total length of the small intestine, respectively [27, 28, 43]. The mucosa is the membrane that covers the luminal portion of the intestine and consists of a single layer of epithelial cells over a vascularized lamina propria. This mucosa exhibits a series of folds, plicae circulares or circular folds, depressions, intestinal crypts, and projections, villi, that considerable increases the surface area compared to a cylinder, as shown in Figure 1.2 [27, 28, 43].

Figure 1.2 Structure of the small intestinal membrane intestinal membrane. Notice all the surface modifications, villi, intestinal crypts and circular folds (plicae circulares) which increase the small intestine surface area. In the villi, the absorptive cells (enterocytes) rest over the vessel rich mucosa where the absorbed nutrients, drugs and xenobiotics are transported to the portal circulation. The enterocytes display a brush border structure in the apical side formed by microvilli. Reproduced from [44], with permission.

The most abundant and relevant cellular type for drug absorption in the intestinal mucosa is the enterocyte [27, 28, 43]. The luminal portion of the enterocyte is denominated the apical portion, whereas the portion facing the lamina propria corresponds to the basolateral part. Each cell is bound to its neighbour through tight junctions that vary in number along the small intestinal tract [28, 45]. The apical portion of the enterocyte has thousands of microvilli (3000-7000 per cell) which define the denominated brush border and that are believed to further increase the surface area available for absorption [27, 28, 43].

1.3.1.1.1.2 Structure of the large intestine The large intestine, on the other hand, represents around 19% of the human intestine. It is divided into three mayor sections: caecum, colon and rectum. The caecum represents about 5% of the large intestine and is accountable for the majority of the microbial digestion, whereas the colon is responsible for most of the water and electrolyte absorption. The colon can be further divided into four regions: ascending colon, transverse colon, descending colon and recto-sigmoid colon [27, 28, 43]. Colonic absorption is believed to occur mainly in the cecum, ascending colon and part of the transverse colon. However, the colonic absorption might be limited 30

compared to that in the upper small intestine due the its lack of villi, tighter epithelial junctions, shorter microvilli in the colonic enterocytes, higher presence of secretive cells, and other factors that might make the colon an unfavourable environment for drug absorption in comparison to the small intestine [28, 33, 34, 46].

1.3.1.1.1.3 The unstirred water layer (UWL) Intestinal permeability can be limited in the small intestine by the presence of a carbohydrate rich glycocalix on top of the microvilli and a mucus layer. Additionally, some membrane components such as cholesterol could alter the permeability of small lipophilic molecules, whereas small polar molecules permeability may remain unaffected by these components [17, 47]. The unstirred water layer (UWL) encompasses the aforementioned glycocalix and mucus layer, but includes a relatively stagnant layer of water along the intestinal wall. The UWL is formed because it is practically impossible to stir the luminal content adjacent to the intestinal wall [48]. It is suggested that the UWL’s impact on drug absorption is more important for drugs displaying high permeability properties, given that the rate limiting step for absorption is considered the diffusion through the UWL. Whereas for a lowly permeable drug, the rate limiting step for absorption is the permeation of the drug through the intestinal membrane, where the impact of the UWL on the drug’s overall permeability is thought to be less important [48]. The UWL thickness (δ) is an operational value given by the difference from the apical side of the enterocyte to the point where the linear extrapolation of the concentration gradients reaches equilibrium with the well mixed luminal content [45, 48]. This concentration difference defines the thickness of the UWL. The reported values for the UWL thickness varies from 25 to 188 µm in humans for highly permeable solutes [48-51]. The actual impact of the UWL in drug absorption is disputed. Based on conscious human experimental determinations, it is suggested that the UWL impact on the rate and extent of absorption is negligible compared to the impact of the intestinal wall [48, 52, 53]. On the other hand a theoretical approach suggests that the UWL would have an impact on the effective permeability for highly permeable drugs [54].

1.3.1.1.2 Mechanisms for permeation of the intestinal membrane The intestinal permeability coefficient referred in Equation 1.3 is a representation of the membrane’s resistance to the passage of a drug from the luminal side of GI tract to the basolateral side and subsequently to the mesenteric circulation. A drug can permeate the GI membrane by different mechanisms that might co-exist simultaneously [55]. The predominance of either mechanism will depend upon the substrate and the membrane characteristics (Figure 1.3). There are two generally accepted mechanisms to describe the transport of a drug across the GI membrane, passive diffusion and active transport [17, 56, 57]. Passive diffusion is considered the most important mechanism for drug absorption [21, 58] and can be described by the means of two sub-processes: passive transcellular diffusion, involving the drug passing through the enterocyte membrane, and paracellular absorption, in which the drug passes through the pores of the tight junctions between the enterocytes. Active transport on the other hand can occur as either influx or efflux to the enterocyte cellular space and from the enterocyte to the luminal space, respectively.

Figure 1.3 Schematic representation of how compounds can cross the intestinal membrane. Passive diffusion through unstirred water layer (UWL), transcellular permeation, transporter mediated absorption, transporter mediated efflux, and passive 32

paracellular permeation through the pores in the junctional complex between the enterocytes. Adapted from [55].

1.3.1.1.2.1 Transcellular passive diffusion Transcellular passive diffusion will depend mainly upon the drug’s physicochemical properties and the membrane composition. The ability of a drug to penetrate the cell membrane will be favoured by high lipophilicity, low molecular weight, low polar surface area, unionized status, molecular flexibility, and the membrane composition [17, 56, 59-61]. Many of the aforementioned characteristics will however impact negatively on drug solubility. Therefore, in order to have a drug candidate with favourable absorption, a complex balance is required to satisfy both criteria. Lipinski’s rule of five (RO5) is widely used as a guide to find that balance during drug design stages [62]. RO5 suggests that a compound with more than five hydrogen bond donors (HBD), molecular weight of 500 Da or more, LogP of five or more and 10 or more hydrogen bond acceptors (HBA) will display unfavourable drug absorption characteristics. RO5 proposes that a violation of two or more of the above mentioned conditions will lead to poor absorption for passively diffused molecules [62, 63], this rule has proved to be useful as a predictor for acceptable oral absorption through passive diffusion in both human and animals [64, 65].

1.3.1.1.2.2 Paracellular absorption Paracellular diffusion refers to the drug passage across the intercellular space formed between the enterocytes. This mechanism requires the drug passage through the negatively charged and water filled pores into the junctional complex between the enterocytes, displaying a diameter 3 to 10 Å in man [57, 66]. The paracellular route was initially intended to explain the absorption of hydrophilic compounds, however in vivo human perfusion experiments suggest that absorption through this route is limited compared to transcellular passive diffusion and therefore restricted to small hydrophilic molecules (molecular weight (MW) < 200 Da and LogP < 0) [45, 56, 66]. This is consistent with the relatively small surface area available for paracellular absorption (about 0.01%), compared with the total surface of the small intestine

membrane, and the increasing density of the tight junctions when moving distally in the GI tract [45, 56].

1.3.1.1.2.3 Transporter-mediated absorption Even though passive diffusion remains as the predominant mechanism for drug absorption, many studies suggest that transporters play a key role for the absorption of drugs and xenobiotics. Over 400 different membrane transporters have been characterized in the human genome and their expressions vary between different cellular types and tissues. Transporters are mainly expressed in the membranes of the epithelial cells of the kidney (tubular cells), liver (hepatocytes) and intestine (enterocytes) in addition to endothelial cells in the blood-brain barrier (BBB) and are therefore of importance for the understanding of the ADME process [67]. The following section is an overview of the most important transporters involved in drug absorption, whereas comprehensive information about transporters, substrates and their localization within the different organs can be found elsewhere [67-71].

1.3.1.1.2.3.1 Intestinal efflux transporters Membrane transporters can be allocated within two large superfamilies: the ATP- binding cassette (ABC) and the solute carrier (SLC). The ABC superfamily encompasses a series of efflux proteins whose main function is to limit the intracellular amount of xenobiotics by extracting them against the concentration gradient by an ATP consuming mechanism. SLC proteins on the other hand are mainly influx transporters where the mechanism is generally driven by ion gradients generated by primary carriers [68, 69]. The most relevant members of the ABC superfamiliy include P-glycoprotein (P-gp, ABCB1), multidrug resistance proteins (MDR1-6, ABCC1-6), multidrug resistance-associated protein (MRP, ABCC) and breast cancer resistance protein (BCRP, ABCG2). The location of the transporters within the enterocytes is illustrated in Figure 1.4. P-gp, BCRP and MRP2 are mainly located in the apical side of the enterocytes, whereas some MRPs are located in the basolateral portion [67-69]. A variety of clinically relevant drugs are substrates for

these transporters (e.g. anticancer drugs, anti HIV drugs, immunosuppressants, etc.) [68].

Figure 1.4 Schematic representation of principal transporters expressed in membranes (apical and basolateral) of the enterocytes. Transporters in light blue and light blue arrows are uptake transporters whereas the light blue with black arrows represents the efflux transporters. Efflux transporters include P-gp, MRP, BRCP whereas uptake transporters include OATP, PEPT, OCTN, OCT PMAT, MCT. Adapted from [68].

P-gp is the most studied transporter in the intestinal membrane. Based on protein quantification, the location of P-gp in the human intestine varies along the intestinal tract increasing from the proximal duodenum to the distal ileum and the colon [72, 73]. P-gp substrates have different molecular characteristics, however hydrophobic, amphipathic or cationic molecules tend to show preference for P-gp [68]. P-gp substrates also include endogenous substances such as bile salts and steroid hormones, whereas known P-gp inhibitors are verapamil, immunosuppressant agents and grapefruit juice [68]. The impact of P-gp on the overall drug absorption will depend upon other properties such as the drugs permeability, membrane concentration and metabolism [69, 74]. Contrary to P-gp, protein quantification showed that BCRP expression along the human intestinal membrane tends to remain almost invariant, yet the expression levels show high interindividual variability [72]. According to mRNA quantification, BCRP and P-gp are the most expressed efflux transporters within the intestinal membrane [75]; this is consistent with the immunoquantification experiments performed by Tucker and co-workers in human duodenal samples [76]. On the other hand, the apparent MRP2 expression along the

human intestinal membrane tends to decrease from the duodenum towards the colon [75, 77]. BCRP and MRP2 show an overlapping affinity for P-gp substrates and for some inhibitors such as grapefruit juice [68]. Certain compounds can either induce and/or reduce their expression within the enterocytes and therefore affecting the activity (i.e. by up regulation or down regulation). These characteristics suggest that efflux transporters form part of a cellular defence mechanism against toxicity. Therefore, their functionality can impact on the drug absorption process [68].

1.3.1.1.2.3.2 Intestinal uptake transporters Solute carrier transporter superfamily members are located in both basolateral and apical sides of the enterocytes, and include peptide transporter 1 (PEPT1, SLC15A), organic cation transporters (OCT, OCTN, SLC22A), organic anion polypeptide transporter (OATP, SLCO), monocarboxylate transporters (MCT1, SLC16A), and plasma membrane monoamine transporter (PMAT, SLC29) (Figure 1.4). These uptake transporters play an important role for the absorption of nutrients such as peptides, amino acids, vitamins, sugars, endogenous substances, drugs and xenobiotics [68]. PEPT1 and the OATP transporters have been shown to be of particular importance for the drug absorption process [67, 68]. PEPT1 substrates are normally able to form peptide bond with the transporter. Peptide-like drugs such as ACE inhibitors, renin inhibitors, betalactamic antibiotics and some antivirals are PEPT1 substrates [68] .Therefore capacity of the transporter to uptake certain drugs can be limited by the presence of proteins or protein like substances, such as milk [78, 79]. On the other hand, OATP substrates include a wide range of amphiphilic organic compounds such as bile acids, thyroid hormones, prostaglandins, anionic peptides drug and xenobiotics [67] .Generally OATP are localized in the basolateral portion of the cells, however OATP2B1 and OATP1A2 have been detected in the apical membrane of the human enterocytes suggesting that they can play an important role in drug absorption [68]. The OATP2B1 substrate pool is narrow and includes HMG-CoA reductase inhibitors and glyburide. Animal and clinical studies indicated that OATP2B1 could play an important role in pravastatin absorption [68]. The OATP1A2 substrate pool on the other hand might be wider than that for OATP2B1[68]. Several fruit juices, such as apple, orange and grapefruit juice, have

been shown to inhibit OATP1A2 leading to clinically relevant food-drug interactions [80-82].

1.3.1.1.3 In vitro and in vivo investigation of permeability

1.3.1.1.3.1 In vivo determination of permeability

The Pw coefficient described Equation 1.3 can be determined by means of in vivo perfusion studies. Human permeability is usually determined in single pass perfusion experiments using open, semi-open or the double balloon (closed) perfusion systems [45]. The double balloon system or Loc-I-Gut is considered today to be the most accurate in vivo method for permeability determinations in humans [45, 83]. Given that the studies are performed in equilibrium conditions, the measured wall permeability is considered as an effective permeability (Peff) for the particular substrate [26, 84].In the double balloon technique the drug is administered in solution to conscious human volunteers employing a multilumen tube (Loc-I-Gut) that has two inflatable balloons attached to it. During the experiment the proximal balloon is positioned within the small intestine, right after the ligament of Treitz which represent the limit between duodenum and the beginning of the jejunum or duodenojejunal flexure (DJF). Once the test tube is positioned, the balloons are inflated in the upper jejunum and the 10 cm gap between them serves as the test segment for the perfusion system where disappearance of a drug is measured in order to obtain an estimate of its intestinal permeability [84, 85]. The reader is referred to the following references for more details about the procedure and the Loc-I-Gut technique, further discussion about Peff is given in Chapter 5 [84-87].

Jejunal Peff is calculated as per Equation 1.4 assuming that the absorption of the drug follows the hydrodynamics of a well-mixed tank [21, 24, 26]

QCCin out in Equation 1.4 Peff () SAC in

where, Qin is the perfusion flow rate, Cin and Cout are the water transport-corrected drug concentrations at the inlet and outlet portion of the test tube, respectively; SA is the surface area of the jejunal segment, defined as the surface area of a cylinder of a length (L) of 10 cm and a radius(r) of 1.75 cm (2πrL) [84, 86]. This calculation 37

assumes that any binding and/or drug loss during the perfusion experiment has been accounted for [84-87].

The derived Peff values are a reflexion of the intestinal membrane’s resistance to the passage of drug from the luminal side to the basolateral side, regardless of the transport mechanisms involved in such processes, i.e., transcellular, paracellular and/or carried mediated absorption [21, 26, 55]. In addition, jejunal Peff can be affected by the experimental conditions and the physiological factors that the drug might encounter during its passage through the intestinal membrane in the upper jejunum [45, 83, 88]. Although the double balloon technique provides accurate description of the human Peff values its application might be limited to a few drugs given its difficulty and the high cost associated with the technique [83]. Therefore in vitro and in silico methods are frequently employed as a surrogate of human Peff measurements.

1.3.1.1.3.2 in vitro and in silico methods for the determination of permeability One of the most common in vitro assays for measuring permeability is the use of cultured cell monolayers. The most common cells lines employed for the determination the in vitro apparent permeability (Papp) in drug development are Caco-2 and Madin-Darby canine kidney (MDCK) cells [89-91]. When cultured on semipermeable membranes both cell lines differentiate into columnar epitheliums showing similarities to that observed in humans intestine, making them useful for the investigation of permeability [90, 91]. In the in vitro system Papp is measured by adding drug in the apical side of the pre-incubated cell monolayers (donor chamber) and subsequently calculating the flux per unit of surface area by measuring drug appearance on the basolateral side of the cell monolayer (donor chamber), using Equation 1.5 [90, 91]

dQ 1 Equation 1.5 Papp () dt C0  SA

where dQ/dt is the flux, C0 is the initial concentration in the donor chamber, and SA is the area of the monolayer [90, 91]. 38

Each cell line provides it owns advantages and disadvantages for measuring Papp. For instance, Caco-2 cells are derived from human colorectal carcinoma, therefore, once differentiated into columnar epithelium in vitro, Caco-2 cells can be used to investigate different transport mechanisms such as transcellular, paracellular and transporter mediated permeability. However, one of the main disadvantages is their culture time. Caco-2 cells usually take around three to four weeks to differentiate into a columnar epithelium, in addition to the maintenance and feeding requirements. Caco-2 cells also tend to overexpress tight junctions, therefore underestimating the paracellular component of drug permeation. More details about the Caco-2 cells are given in the following references [89-93]. MDCK cells on the other hand require a shorter incubation time that Caco-2 cells, 2- 6 days [90, 94]. However, MDCK wild type cells tend to be heterogeneous giving variable results for permeability. MDCKI and MDCKII are the most commonly used strains for permeability assays, the latter being “leakier” than the former [95]. Given that these cells are derived from canine kidney epithelia, the expression of efflux transporter might be limited in the wild type MDCK cell compared to that in Caco-2 cells. For the investigation of efflux transport in MDCK cells, the cell lines are usually transfected with the MDR1 gene to overexpress P-gp. The MDCK-MDR1 cell type is widely used in drug discovery and development for the investigation of substrate affinity for P-gp and possible transporter-mediated drug interactions affecting intestinal absorption [89, 94-97].

Apparent permeability from both Caco-2 and MDCK-MDR1 cells have been shown to be correlated with human Peff and fa [91, 93, 97, 98]. The in vitro-in vivo extrapolation (IVIVE) from Caco-2 Papp to Peff has been reported for 20 drugs at pH 6.5 with an R2 value of 0.73 (Equation 1.6). The correlation was slightly improved to 0.85 when only the passively absorbed drugs where considered, as shown in Equation 1.7 [99]

logPPeff 0.6252  log app,2 C aco   0.3036 Equation 1.6

logPPeff 0.7524  log app,2 C aco   0.5441 Equation 1.7

A similar relationship has been stablished for Papp values derived from MDCK- MDR1 cells for 24 drugs measured in isotonic conditions at pH 7.4 (Equation 1.8). The reported R2 value was 0.71, whereas for the passively absorbed drugs (n=20) the correlation was 0.7 (Equation 1.9) [97].

logPPeff 0.712log  M D C K M D R1  1.05 Equation 1.8

logPPeff 0.829log  M D C K M D R1  1.30 Equation 1.9

Other in vitro methods for the determination of Papp include the parallel artificial membrane permeability assay (PAMPA) and the use of excised intestinal tissue from pre-clinical species or humans to measure Papp and the fa within the Ussing chamber [100, 101]. PAMPA is an artificial lipid membrane varying in composition and complexity that is employed for the measurement of drug’s passive permeation, particularly in drug discovery, and has been shown to provide good correlation with

Papp obtained in Caco-2 cell lines [102, 103]. On the other hand, the Papp values obtained from excised rat and human intestinal samples in the Ussing chamber have been shown to correlate directly with those of the human fa [101, 104, 105]. The reader is referred to the following references for more details about the aforementioned methods [100-102, 106-109].

Alternatively, Peff can be also predicted from the use of molecular descriptors such as polar surface area (PSA) and number of hydrogen bond donors. Equation 1.10 shows the correlation between Peff and the aforementioned molecular descriptors that was derived by means of a multivariate analysis (R2 =0.88) [110]. It must be highlighted, however, that this relationship was developed based on a limited number of compounds (n=13) and only applies for passively absorbed compounds with similar characteristics as the training set used in the multivariate analysis [110].

logPPSAHBDeff   2.546  0.11   0.278  Equation 1.10

1.3.1.1.4 Permeability and the fraction absorbed

Peff has been shown to be highly correlated with the human fa. The relationship between the fa and Peff was suggested by Amidon and co-workers for a variety of compounds whose Peff values were measured in human volunteers [21, 84].Currently, this relationship has been expanded to more than 30 different drugs and Peff is now used as indicator of drug absorption [21, 83, 84]. Assuming that a drug is completely dissolved upon administration and that does not experience any luminal degradation, the fa can be approximated from Peff as per Equation 1.11,

2 Peff Tres Equation 1.11 r fea 1

where Tres is the small intestinal residence time (circa 3.3 hours), and r is the intestinal radius (1.75 cm) [21, 66].

1.3.1.1.5 BCS and the rate limiting steps for drug absorption The work by Amidon’s group established the basis for understanding the rate limiting steps for drug absorption: dissolution, solubility and permeability [21] and served as the foundations for the Biopharmaceutics Classification System (BCS). The BCS classifies the drugs based on their solubility and permeably properties into four categories [21, 111]: Class 1, highly permeable and highly soluble; Class 2 highly permeable and lowly soluble; Class 3, lowly permeable and highly soluble; and Class 4, lowly permeable and lowly soluble [21]. A drug is classified as highly permeable in the BCS if its fa is greater or equal than 0.85 or 0.9 [21, 111, 112], whereas the solubility criteria is based on the calculation of an unitless dose number

(Do), as per Equation 1.12,

D ose Equation 1.12 D o  250m L C s where Dose (mg) is the highest dose of the drug product available on the market and

Cs is the drug’s minimal solubility(mg/mL) over the pH range of the GI tract. A drug is classified as highly soluble if the Do is less than 1 [21, 111, 112]. The BCS has been used widely in drug development as an indicator of the potential of drug to 41

become absorbed orally. However, it has also served for regulatory purposes. Some regulatory agencies, such the Food and Drug Administration (FDA) and the European Medicines Agency (EMA), grant waivers on the requirement to conduct clinical bioequivalence (BE) studies for IR drugs products containing drugs that belong to the Class 1 (FDA, EMA) and/or Class 3 (EMA) within the BCS, provided that the product meets the dissolution and quality requirements specified on their respective guidelines [111, 112].

1.3.1.2 Solubility and dissolution

As mentioned above, the second most relevant factor defining intestinal drug absorption is the drug’s concentration next to the intestinal wall (Cw in Equation 1.3). This concentration will depend mainly upon the drug solubility, its dissolution rate and the residence time of a drug in the given intestinal segment, considering the luminal conditions at the site of absorption [21].

1.3.1.2.1 Solubility One of the main assumptions in absorption modelling is that only drug in solution can become absorbed [21, 29]. When a drug is administered as an oral solution it will be almost immediately available for GI absorption. Depending upon its solubility in a given segment of the GI tract, the drug can precipitate or even display supersaturation. A solution is said to be supersaturated when the drug concentration in the solution is higher than its equilibrium solubility (Ceq) [113-115]. Drug solubility is influenced by physicochemical characteristics such as: molecular weight, ionization (pKa), lipophilicity (logP/LogD), number of H-bond donors, crystal form, among others [62, 63, 113, 116]. Drug solubility can be defined in many forms, however the most relevant solubility types are equilibrium solubility and kinetic solubility [113]. The former is regarded as the concentration of a drug in solution which is in contact with an excess amount of solid drug that is reached after a prolonged period of incubation employing, for instance, the shake-flash method [113]. Kinetic solubility on the other hand is usually measured in drug discovery and is measured as the concentration obtained immediately after the addition of a solution of the drug prepared in a rich solvent (usually DMSO) into an aqueous media [113]. The reader is referred to the reviews 42

by Lipinski and co-workers, Sugano, and Avdeef for further definitions and details about solubility measurements in drug discovery and development [62, 113, 117].

Several in silico methods can be used for the prediction of drug equilibrium solubility based on a compound structure and molecular descriptors, amongst other physicochemical properties. These models usually predict solubility within 2 to 10 fold error for neutral species, while for ionizable species the error is multiplied by the prediction error of the pKa [113]. The reader is referred to the following articles for more details about the methods and models [116, 118-123].

1.3.1.2.2 Drug release When a drug is administered in a solid dosage form (i.e., capsule or tablet) the drug concentration in the lumen is defined by the release rate of the drug from the formulation and the subsequent dissolution process (Figure 1.5). In general these processes depend largely upon the drug solubility and the biopharmaceutical properties of the drug product [17, 113, 124]

Figure 1.5. Schematic representation of the luminal processes occurring after the oral administration of a solid dosage form. For solid IR dosage forms the first step for dissolution is the disintegration of the dosage forms to release the drug content into the GI lumen (Figure 1.5). There are, however, formulations that are intended to control the rate of absorption by retaining the drug in the formulation for a specific period of time, thus releasing the drug at a specific rate or after a specific event [125-127]. Examples of such formulations include: hydrophilic matrix systems, osmotic pumps, enteric coated tablets, capsules and granules, gastro retentive systems, etc. For a comprehensive review on such

formulations and the modelling of the drug release the reader is referred to the following references [33, 125-132].

1.3.1.2.2 Dissolution The factors governing the dissolution rate were first described mathematically by Noyes and Whitney (1897) where they recognized that the dissolution rate was proportional to the difference between the drug solubility and the concentration at a given time. Further refinement to this relationship was done by Nernst and Brunner (1904) by applying the diffusion layer concept to the model developed by Noyes and Whitney [133], as shown in Equation 1.13,

SAD Equation 1.13 aq Adiss DRCs hV

where DR is the dissolution rate, SA is the surface area of the drug in contact with the dissolution media, Daq is the diffusion coefficient, h is the diffusion layer thickness,

Adiss amount of drug dissolved, and V is the volume of the dissolution media. This equation recognized the main biopharmaceutical factors that can be used in drug development to control the dissolution rate, for example: a particle size reduction will lead to an increased surface area, hence to an increased dissolution rate. Similarly, wetting agents such as surfactants can be used to reduce the interfacial tension between the solid and the aqueous media, thus increasing the contact surface area between the aqueous media and the solid particle. The use of emulsifying agents in the formulation can increase the solubility of lipophilic drugs with poor solubility issues, whereas changes in pH and buffer capacity can lead to changes in solubility for ionizable drugs, either increasing or decreasing the dissolution rate depending upon the GI conditions [113, 134-138]. Several variations have been proposed to Equation 1.13 in order to take into account the dissolution process from a more mechanistic perspective, and these variations have been implemented in several mechanistic absorption models (discussed below) and the details can be found elsewhere [139-143]. One such model is further discussed and implemented in Chapter 6.

Given the importance of dissolution and solubility for drug absorption, in vitro measurement of the dissolution rate has been widely used in drug development, serving both as a quality control measurement and as a prognostic tool for the in vivo performance of the orally administered drug [113, 136]. Prediction of the in vivo drug performance using dissolution test comes particularly at the hand of the development novel methodologies for the in vitro dissolution testing and the use of biorelevant dissolution media to stablish the so-called in vitro-in vivo correlations (IVIVC) [113, 143-147]

1.3.1.2.3 Physiological factors that alter drug dissolution and solubility The GI physiology could impact the dissolution process, thus altering drug absorption. As mentioned before the buffer strength and the pH most likely will affect the solubility and dissolution of ionizable compounds [115]. The pH in the normal fasted GI tract varies from 1 to 3.5 in the stomach to 8.0 in the descending colon, with a high degree of interindividual variation [32, 46, 148-150]. In the fed state these values can change from 1.7 to 6.4 in the stomach and 5.4 to 6.1 in the duodenum and 5.2 to 6.2 in the jejunum [149, 150]. A weak basic drug could therefore precipitate due to the pH change from one region to another of the GI tract [115, 117]. Additionally, bile salts can enhance the solubility of poorly soluble lipophilic drugs, either by decreasing the surface tension or by forming bile micelles [113]. This solubilisation has shown to be dependent on the bile salt concentration and will therefore increase during the fed state, given the postprandial increase in bile salt secretion [134]. In addition, the bile-salt solubilisation might also have impact on the drug permeation as the drug trapped in bile micelles might permeate the intestinal membrane at a slower rate [151]. The fluid volume available in the GI tract will also affect the dissolution rate (Equation 1.13). The volume of free water in the stomach has been estimated to be around 45 mL (range 13 - 72) in the fasted state, whereas in the fed state the volume of the stomach contents could be increased to 686 mL(range 534-859)[152]. In the small intestine and large intestine the free water volume has been measured to be 105 mL and 13 mL in the fasted state, respectively, whereas in the fasted state, the volume has been reported to be 54 mL (range 20 -156mL) and 11 mL (range 2- 97 mL), respectively [152]. Both gastric empting and GI transit time will affect drug dissolution, the longer the residence time

in one segment, the longer the time that a drug will spend in a particular GI region and vice versa. For instance, in the fasted state gastric emptying could be considerably delayed. Therefore a weak basic drug could benefit from a prolonged residence time in the stomach having more time to be dissolved in a more favourable acidic environment, although this might affect the absorption rate, as mentioned before given that de drug is less likely to be absorbed from the stomach [149, 153- 155]. The reader is referred to the review by Mudie and co-workers for more details about the physiological factors affecting drug dissolution [149].

1.3.2 Hepatic and intestinal first pass metabolism

The second limitation to oral drug bioavailability is the presystemic extraction by both the liver and the gut wall. First pass extraction is associated with poor bioavailability of a variety of drugs [156]. Metabolism is the principal mechanism for the elimination of most of the drugs on the market and is a major component of presystemic extraction [157, 158]. Metabolism can be divided into two phases; the first one (phase I) consists of an enzymatic oxidation of the parent drug and the second (phase II) consists of the enzymatic conjugation of the phase I product and/or the parent with glutathione, sulphate or glucuronic acid in order to enhance the compound’s hydrophilicity and therefore facilitate renal elimination [56, 159]. Among the metabolic enzymes, cytochrome P450s (CYP) (phase I) and the UDP- Glucoronosyltransferases (UGT) (phase II) are responsible for the majority of metabolic elimination. More than three quarters of the metabolic reactions of the 200 most prescribed drugs in 2002 are mediated by the CYP enzymes, whilst CYP3A4 is accountable for almost 46% of these reactions [158].

1.3.2.1 Hepatic first-pass metabolism and biliary excretions

The contribution of the liver to presystemic extraction and therefore to FH can occur by means of two mechanisms: first-pass metabolism and biliary excretion. The factors affecting the liver first pass metabolism are the same factors governing the hepatic clearance: blood flow, unbound fraction of drug in the blood/plasma, enzyme abundance and the affinity and the capacity of the metabolic enzymes to biotransform these drugs, expressed as hepatic intrinsic clearance (CLint) [17]. 46

Several models can be used for the description of the hepatic drug clearance and presystemic extraction ratio (ER), including: the well-stirred model, the parallel tube and dispersion model [160-162]. However, given its simplicity, the well-stirred model is the most commonly used model in drug development for the prediction of the in vivo hepatic clearance, as shown in Equation 1.14 [163, 164]

 Equation 1.14 fub CLint, liver CLQERQliver liver   liver   f CL Q ubint , liver liver

where CLliver, is the hepatic clearance, Qliver is the liver blood flow, fub is the drug’s fraction unbound in blood, and CLint,liver is the liver intrinsic clearance. The abundance of the metabolic enzymes in the liver is higher than in any other organ in the human body [156]. In Caucasians the total abundance of CYP enzymes in the liver has been reported to be 0.43 nmol/mg of microsomal protein, where the most abundant phase I enzymes is CYP3A, followed by CYP2C, CYP1A2, CYP2E1, CYP2A6 and CYP2D6 [165-167]

Biliary excretion, on the other hand, is affected by the drug efflux to the canalicular space between hepatocytes and it is believed to occur mainly through active transport [168, 169]. Transporters located in the canalicular membrane include P-gp, MRP2 and BCRP, therefore the substrate affinity for these transporters is an important factor for biliary excretion [67]. There are a number of drug specific physicochemical factors that can influence biliary excretion, such as molecular weight, polarity and lipophilicity [170, 171]. The impact of biliary excretion in drug bioavailability has been demonstrated for several drugs such as cardiovascular, anticancer, anti-infective and biologically active peptides. However, determination of the extent of biliary excretion in humans is very difficult, and therefore is mostly estimated by employing animal models, in vitro cell cultures or in silico models [169, 170].

Several in vitro systems have been proposed for the estimation of both the hepatic clearance and liver presystemic extraction within the IVIVE approach, these include primary hepatocytes and hepatocyte sub-cellular fraction such as human liver

microsomes (HLM) and human liver cytosol (HLC), as well as recombinant microsomes expressing a particular CYP isoform, amongst others [172-174]. Prediction of the human hepatic clearance using the aforementioned systems requires the use of physiologically-based scaling factors that take into account the abundance of the respective CYP enzymes as well as the abundance of the cellular systems and subcellular fractions in human [167, 173, 175]. The focus of this thesis will be put on the IVIVE approach using HLM microsomes and recombinant microsomes and is further described in Chapter 4 and Chapter 6.

1.3.2.2 Intestinal first-pass metabolism

The intestinal contribution to first-pass metabolism has been subject to increasing interest during the last decades and it is believed to play a major role for determining oral bioavailability [19, 159, 176]. As in the liver, CYP3A enzymes are the most abundant enzymes in the human intestinal wall, accounting for almost 80% of the small intestine CYP content [165]. CYP enzymes are strategically located at the villous tip of enterocytes and their presence varies along the small intestine [19, 165, 177, 178]. Quantification studies showed that CYP3A4 abundance decreases from the proximal duodenum towards the distal ileum, being highly expressed in the jejunal segment [72, 177-179]. This distribution pattern can have an impact on the bioavailability of CYP3A4 substrates administered as controlled release formulations [180, 181]. Even though the CYP3A4 abundance is about 1% of that in the human liver [182], it plays an important role in the intestinal pre-systemic extraction for several drugs [183-186]. In addition, it has been hypothesised that both CYP3A4 and P-gp might act together as barriers for the absorption of drugs and xenobiotics. This hypothesis is supported by their location within the enterocytes, overlapping affinity for several substrates and suggestions of coordinated regulation between them [187, 188]. However, there is conflicting evidence indicating that there is no shared regulation between CYP3A4 and P-gp and that the overlap in substrate affinities is just fortuitous. The latter is supported by differences in expression along the small intestinal membrane (i.e. P-gp abundance increases from proximal to distal small intestine, whilst CYP3A4 abundance decreases) and differences in affinity for certain substrates and/or inhibitors, such as digoxin and nifedipine [159]. Therefore, there is

no consensus about their possible coordinated role limiting the exposure to drugs and others xenobiotics [57, 159]. This hypothesis is further discussed in Chapter 4.

The factors influencing first-pass gut wall metabolism and therefore affecting the FG are similar to those governing the hepatic clearance: intestinal blood flow, intestinal fraction unbound and intestinal intrinsic clearance [97, 184, 189]. However, there is a complex interplay between drug absorption and intestinal first-pass metabolism, where the factors affecting drug absorption might have an impact on the rate and extent of intestinal first-pass extraction (i.e., the solubility, permeability, residence time, and interindividual variability) [74, 190, 191]. For a comprehensive review on intestinal first pass metabolism the reviews by Thelen and Dressman and Lin and co- workers are recommended [159, 176].

In vitro systems for the investigation of intestinal first pass in the IVIVE context are similar to that in the liver. These include: isolated enterocyte preparations, precision cut tissue slices, and human intestinal microsomes (HIM) [177, 178, 183, 192, 193]. For a comprehensive summary of the in vitro methods for the investigation of intestinal metabolism the reader is referred to the review by van de Kerkhof and co- workers [193]. Given that the intrinsic activities of the CYP3A enzymes in both liver and intestine have been reported to be similar [177, 182], the in vitro clearance data obtained in HLM can be used for the prediction of intestinal metabolism, as long as the differences in enzyme abundance between liver and intestine are taken into account [97, 177, 182]. Further details about this method can be found in Chapter 6.

1.4 Prediction of human oral bioavailability

1.4.1 Direct extrapolation from animal models

During pre-clinical development, research teams employ animal models to elucidate the pharmacokinetic characteristics of new drug candidates. The most commonly used pre-clinical species in such studies include mouse, rats, dogs and non-human primates (NHP) [194].

Human pharmacokinetic parameters for intravenous administration such as clearance, half-life and volume of distribution have been predicted empirically from pre-clinical species data by means of allometric scaling techniques [195, 196]. Allometric scaling requires the investigation of the desired pharmacokinetic parameters in one or more pre-clinical species, where the parameters of interest are regressed against a species body size descriptor (usually body weight) and the resulting regression equation can be used to predict the value of the parameters in humans [197-199]. However, one of the major drawbacks of allometric scaling is its empirical nature, which does not necessarily allows the understanding of the underlying mechanisms that can explain the species differences in the pharmacokinetic parameters, such as metabolic clearance [195]. Moreover, the usefulness of allometric scaling for the prediction of human pharmacokinetic parameters has been questioned due to large prediction errors that can arise from the allometric relationship and the use of arbitrary constants with little or no physiological meaning [200, 201].

As a part of the drug development process oral bioavailability is usually investigated in pre-clinical species and the resulting data is subsequently employed as a part of the lead selection/optimization process. In some research teams thresholds have been set up for oral bioavailability, for instance 20-30% in rats, as an indicator of acceptable human oral bioavailability for the new drug candidate, whereas a value below this limit would lead to further considerations and extensive scrutiny of the candidate in order to make the decision whether to continue or not with its development [56, 202]. Preclinical species seem to provide a good indication of the human oral absorption for certain drug candidates. Several groups have demonstrated acceptable correlations levels between the fa and intestinal permeability between humans and pre-clinical species, particularly for rat and monkeys, though this correlation was found to be poor in dogs [203-209]. When it comes to oral bioavailability, however, a similar relationship remains unclear. Several direct comparisons and correlations between human and animal bioavailability have been reported in the literature (Table 1.1).

One of the first attempts to summarize the relationship between human and animal bioavailability was done by Sietsema in 1989 where he collected bioavailability data for several drugs in humans, rodents, dogs and primates, showing a poor interspecies correlation [210], while in 2000 Mahmood also highlighted that allometric scaling was not an adequate tool for the direct prediction of human bioavailability [211]. From the analysis of the previous datasets it can be noticed that the bioavailability data obtained from preclinical species seems to underestimate the value in humans [64, 203, 207, 210, 212, 213]. While the reason for this underprediction might not be necessarily clear, as mentioned in the previous sections, the factors governing oral bioavailability are numerous and depend both on the drug-related properties and physiological, environmental and population characteristics. Therefore one can speculate that interspecies differences in physiological aspects, such as body size, blood flows, plasma proteins, GI tract morphology and physiology, enzymatic abundance in the GI tract and liver are probably the main contributors for the differences observed in bioavailability [46, 203, 213-219]. Most of the aforementioned comparisons and datasets (Table 1.1) were based on a limited number of compounds (< 50 compounds) [64, 203, 207, 211, 212], whereas the largest dataset is from 1989 and does not differentiate between preclinical species such as rats, mouse and/or rabbits nor include enough data points, especially for the analysis of the correlations between monkeys and humans [210]. Interestingly, one of the most cited articles when it comes to the comparison between animal and human oral bioavailability is based on Sietsema’s review on interspecies bioavailability data [220]. Therefore there is the need for an updated, wider and specific dataset that can be employed for the comparison between human and animal oral bioavailability, in order to definitely elucidate whether animal data can be used for the prediction of human oral bioavailability per se. This dataset can be also used for the development of quantitative and qualitative models for the extrapolation of bioavailability values from pre-clinical species to human, which can be of value in drug development. The latter points are amongst the aims of this thesis and this will be further discussed in Chapters 2 and 3.

Table 1.1 Previously reported interspecies oral bioavailability comparisons. Year Number of Species Correlation (R2) Comments Ref. Compounds 1989 409 rodents(rats, mice and rodents (0.4), dog (0.3) Bibliographic review. No information regarding [210] (14 prodrugs) rabbits), dogs and and primates (0.2) formulations and/or dose, not all the compounds primates(rhesus monkeys, listed have both animal and human F data cynomolgus monkeys, African green monkey and baboons)

2000 15 mice, rat, rabbit, guinea pig No correlation value Bibliographic review. No information regarding [211] or dog reported. MAE 33% formulation and/or dose, different allometric scaling method employed, correlation expressed as mean absolute error of the midpoints of the five methods employed for all the species. 2002 35 Monkeys Monkeys (0.5) Bibliographic review. Information about [207] formulations, no references for F available in the article

2005 56 ( 30 with Rats, dogs ( mostly beagles) No correlation between F Bibliographic review. Dose normalized AUC [64] absolute and monkey ( rhesus and in animal and humans was calculation for oral administration, no information bioavailability cynomolgus monkeys) reported, only a qualitative about dose and formulation. F data not published in information) prediction for the dose the article normalized AUC was reported.

2006 48 Rat Rat (0.29) Bibliographic review. No details about compounds [203] and F values on the article, no information about dose and formulation

2007 5 Cynomolgus monkey No correlation between F Animal data was experimentally determined, human [213] in animal and humans was data extracted from literature. reported 2009 8 African green monkey No correlation between F Animal data was experimentally determined, human [221] in animal and humans was data extracted from literature. reported

2010 13 Cynomolgus monkey No correlation between F Animal data was experimentally determined, human [212] in animal and humans was data extracted from literature. reported MAE, Mean average error. F, oral bioavailability.

1.4.2 Physiologically-based pharmacokinetic modelling and simulation

An alternative for the prediction of human oral bioavailability in drug discovery and development is to use in silico models that can provide a more mechanistic description of the bioavailability process. Physiologically-based pharmacokinetic (PBPK) models serve this purpose well, as they incorporate the known mechanisms describing the drug ADME process into the model structure. In contrast to more traditional pharmacokinetic models that summarize the complex ADME process using one or two compartments, PBPK models are multi-compartmental models where each compartment within the model is generally intended to represent a particular tissue, organ and/or process. The parameters of a PBPK model can be therefore naturally informed from known physiological and experimental values, such as the tissue volumes, blood flows, binding to plasma protein, etc.[222, 223]. This property allows the distinction between physiologically-related and drug-related parameters within the model, enabling the use of a generic PBPK model structure for the prediction of the ADME processes for one or several drugs, by just varying the drug-related input parameters in the model. A general structure of a PBPK model is shown in Figure 1.6 [222].

Figure 1.6 Generic structure of a whole body PBPK disposition model. Each tissue is represented by a compartment. For the ith tissue the drug is distributed to and from the tissue by means of a blood flow (Qi) limited process .Drug can be cleared (CL) in the liver(H) and/or the kidneys(R). RoB, rest of the body, HA and HV, hepatic artery and vein, respectively. Inspired from [222].

1.4.2.1 PBPK models in drug development

The origins of PBPK models can be traced back to the work by Teorell in 1937 [224]. Nevertheless, due to their mathematical complexity and the requirement of vast amounts of experimental data required for their parametrization, the use of PBPK models was relatively limited to a few drug applications, mainly by academia, and the environmental toxicology field, for safety and risk assessment [223, 225]. In the last decades, however, the development and application of PBPK models to solve drug-related problems have increased dramatically, judging by the number of scientific literature associated to PBPK models [222, 223]. The reasons for this increase are partly due to: the increase in computing power, the development of in silico methods for the prediction of the tissue to plasma partition coefficients, thus reducing the labour intensive animal experimentation to obtain the necessary parameters [226-230], the use of in vitro methods for the extrapolation of hepatic metabolism [172, 231], and the combination of PBPK models with the IVIVE approach [232], and the availability of user-friendly “off the shelf” PBPK software packages such as GastroPlus™ (Simulations Plus, Lancaster, CA, USA), PK-Sim® (Bayer Technology Services, Leverkusen, Germany) and SimCYP® (Certara, Sheffield, UK), amongst others [10, 233-237].

Recent reviews and case studies have highlighted the usefulness of PBPK models in drug discovery and development stages, particularly as a tool to support the decision making process, lead optimization and to reduce the conduction of unnecessary experiments and clinical trials, such as DDIs studies [10, 233-240]. There is also an increased confidence in the use of PBPK models for the prediction of pharmacokinetic parameters in discovery and development stages. For instance, clearance and volume of distribution at steady state (Vss) are well predicted, particularly when combining PBPK models and IVIVE approaches with in vivo data obtained in preclinical species. This approach has been shown to perform better than allometric scaling alone while at the same time it provides better insight into the mechanisms driving the drug’s ADME process [10, 233, 235, 236]. Nevertheless, there is still room for improvement, especially to overcome the translational issues that are found in the IVIVE approach, for instance, the prediction of human clearance derived only from in vitro systems [233, 241]. Furthermore, 55

given the high dimensionality of the inputs required by PBPK models, several safeguards and good practices have been invoked for the use of PBPK models for prospective ADME predictions, namely: verification of the quality of the input parameters, the verification of the model predictions in preclinical species prior to the human simulation, parameter sensitivity analysis and, most importantly, knowledge of the compound and verification that the model produces reasonable outputs [223, 233, 235, 239]

The use of PBPK models to support labelling information has also gained acceptance amongst regulatory agencies, particularly in the area of the prediction of DDIs, where PBPK have been widely employed [11, 233, 242-245]. Indeed, 33 of the total submissions for Investigational New Drug (IND) and New Drug Applications (NDA) received by the FDA between 2008 and 2012 contained PBPK modelling approaches to support drug labelling, the majority applied to DDIs predictions [246]. Moreover, given the predictive nature of the PBPK models, other applications are also possible, particularly to answer clinical pharmacological questions that might affect the drug exposure in special populations such as paediatrics, organ impairment, obesity, amongst others [246-250].

1.4.2.2 PBPK models for absorption and bioavailability

Another important application of PBPK models in drug development is the prediction of oral absorption and bioavailability issues. In fact, 9% of the aforementioned IND/NDA submissions applied PBPK modelling and simulation to answer absorption/bioavailability-related questions [246].

Mechanistic absorption models were firstly proposed by Amidon’s group after translating the dissolution, permeation and transit process in the GI tract into a differential equation form (Equation 1.3). The first successful absorption PBPK model was the Compartmental Absorption and Transit (CAT) model which was employed for the prediction of the fa for several structurally diverse drugs with known Peff [29, 31, 153]. In the CAT model, the drug transit along the small intestine

was modelled with the help of seven transit compartment, that were also intended to represent the different anatomical regions of the small intestine [29, 31, 153]. Subsequently, absorption PBPK models were improved to account for the different processes affecting oral absorption and bioavailability described in the previous sections, namely: dissolution, transit, precipitation, supersaturation, permeation, transport, intestinal metabolism, hepatic metabolism, etc. The Advanced CAT (ACAT) model was the first commercially available absorption PBPK model and lies within the core of the GastroPlus™ software package [251]. The ACAT model describes the human GI tract with nine segments, from stomach to colon, and each process within the GI lumen is described with compartmental differential equations. The ACAT model has been successfully applied for the prediction of different aspects of drug absorption and bioavailability [251]. Later on, Willman and co- workers (2004) developed an absorption PBPK model using partial differential equations (PDE) to describe the position-dependent changes in the absorption process across the length GI tract [252]. Recently, however, this model was updated into a compartmental structure similar to the ACAT model and is the absorption model of the PK-Sim® software package [253, 254]. Another popular model is the Advanced, Dissolution, Absorption and Metabolism (ADAM) model built within the SimCYP® Population-based ADME simulator [20, 255]. The ADAM model has a similar structure to that of the CAT and ACAT models, however its main difference (at least initially) was the incorporation of known inter-individual variability in the parameters affecting the ADME processes, thereby allowing the capacity to simulate not only the “reference individual” but populations and their variability [20, 255]. Other absorption models include the Segregated Flow Model (SFM) and the recently published GI-Sim models [256, 257]. In this thesis the ADAM model within SimCYP® will be employed and is shown Figure 1.7. This model will be further described in Chapter 4 and the relevant details of the model structure and parametrization can be found elsewhere [190, 255].

Figure 1.7 Schematic representation of the ADAM model implemetted within SimCYP®The stomach is represented by one compartment, the small intestine is represented by seven compartments and the colon is represented by one compartment. In each compartment the drug can be either in a solid state in the formulation (form), solid outside of the formulation (undiss) or dissolved (diss). The radii of the segments are represented by the radii of the cylinders, whereas the colour coding indicates bile enhanced solubility (green) and CYP3A4 abundance (purple). From [247], with permission ( originally adapted from [20]).

PBPK models have been used for the prospective predictions of oral bioavailability [236, 258]. Using the ACAT model within GastroPlus™ combined with a step-wise modelling and simulation strategy together with in silico, in vitro and in vivo data Parrott and co-workers (2005) predicted the oral bioavailability for 64 different compounds in the rat [236]. In their simulations, they could observe a systematic underprediction of the oral bioavailability; mainly attributed to the solubility issues that were not adequately captured by the model, particularly for poorly soluble compounds [236]. In another example, Paixao and co-workers employed a variation of the ACAT model combined with different approaches using, in vitro and in silico data for the prediction of oral bioavailability [258]. Their results showed good

bioavailability prediction when using in vitro data for Papp and CLint in the model, whilst solubility was predicted in silico. However, the predictive power of the model was considerably decreased when only in silico derived parameters were employed; highlighting the need for good quality inputs when using PBPK models [258]. It must be highlighted, that in both approaches the intestinal first pass metabolism was assumed to be negligible [236, 258]. The latter might not necessarily hold true especially for several drugs in the employed in the dataset used by Paixao and co- workers (2012) such as midazolam and nifedipine, which are known to undergo extensive intestinal first pass metabolism [97, 184, 259]. Therefore, although correctly predicted, the underlying mechanism driving bioavailability was not accurate.

The latter highlights one of the key issues affecting absorption PBPK models: the high dimensionality of the inputs. This could lead to the model providing the right answer but with the wrong mechanisms, consequently affecting its power to extrapolate the model predictions to a different scenario, such as DDIs in the case of intestinal metabolism. This high dimensionality was investigated systematically by Darwich and co-workers (2010). The authors investigated the theoretical interplay between the model parameters affecting dissolution, absorption, transport, and metabolism in the overall bioavailability [191]. This work underlined the complex dependency between absorption-related parameters and the intestinal first pass metabolism, especially for CYP3A substrates [191]. One such absorption related-parameter that has been overlooked in the previous works is how the GI release pattern of a drug might affect the intestinal-first pass. This release pattern can be controlled by the use special formulations such as modified release (MR) dosage forms, which have shown to have an impact of the bioavailability for several drugs [180, 181, 260]. In addition, this type of formulation might suffer from an even more complex interplay when it comes to the prospective predictions using PBPK models, as several absorption and metabolism-related factors that affect the overall bioavailability have been shown to change across the length of the GI tract, as highlighted in the previous sections. Therefore, this thesis will look into that particular interplay and provide methods to make better bioavailability predictions using PBPK models for MR dosages forms, which will be

discussed in Chapters 4 to 6. Chapter 6 of this thesis will be focused on a particular case when oral bioavailability is affected by the use of a MR formulation. Specifically, Chapter 6 will investigate the case of oxybutynin (OXY) in more detail by applying a PBPK modelling and simulation approach. Briefly, OXY is a highly permeable and soluble compound (BCS Class 1) that is rapidly absorbed after oral administration, yet its bioavailability is around 6 % due to an extensive first-pass metabolism in both the intestinal wall and the liver, which is mainly mediated by CYP3A4 [180, 261-263]. Interestingly, when OXY is formulated as a once a day (MR) OROS formulation, the relative bioavailability of OXY is around 153% compared to its IR tablet, while the exposure of its main metabolite, N- desethyloxybutynin (DEOB) is reduced by almost 30% [180]. The structure of OXY and DEOB are shown in Figure 1.8 and more details about the compound are given in Chapter 6.

Figure 1.8 Chemical structure of oxybutynin and its main metabolite

1.5 Project aims and objectives

The overall aim of this thesis is to provide models for the prediction of human oral bioavailability and its underlying processes, GI absorption and first pass metabolism. This will be done by applying different modelling and simulation techniques. However, it will be particularly focused on incorporating the complex interplay between the factors that can influence oral drug absorption and bioavailability, using an integrative modelling and simulation approach. This thesis will be presented in two main sections.

The first section of this thesis will be focused on the evaluation of the use of oral bioavailability data obtained in pre-clinical species as a predictor of the human value. In particular, the aim is to evaluate the development models that can quantitatively and qualitatively provide a relationship between animal and human bioavailability. For this purpose, the first step will be to construct a dataset of animal and human bioavailability by performing an extensive search in the published scientific literature. The second step will be to use the constructed dataset to evaluate the relationship between animal and human bioavailability, evaluating possible correlations and building models that can use the pre-clinical data for bioavailability predictions (Chapters 2 and 3).

The second section of this thesis is focused on the mechanistic prediction of drug oral absorption and first pass metabolism applying a PBPK modelling and simulation approach (Chapters 4 to 6). Specifically, the aim of this section is to investigate the impact that the drug release rate might have on drug absorption and intestinal first pass metabolism, considering the physiological differences observed across the length of the human GI tract. This section is divided in three interconnected sub- sections, each one with its own aims and objectives:

1. The aim of the fist sub-section will be to investigate the relative bioavailability between IR and MR formulation of CYP3A substrates, and to investigate the parameter space that can lead to a higher relative bioavailability for the MR formulation using an integrative PBPK modelling and simulation approach (Chapter4). 2. The aim of the second sub-section is to provide a novel, clear and simple methodology that can be used to parameterize regional intestinal permeability within the PBPK modelling and simulation approach, and to apply it for the prediction of the regional absorption in the GI tract using a PBPK model (Chapter 5). 3. Finally, the aim of the third sub-section will is to develop a simplified mechanistic absorption PBPK model that can be used for the prospective prediction of oral bioavailability, integrating all the knowledge gained in the previous sub-sections, and apply it to study the higher relative bioavailability

observed for the drug oxybutynin when is formulated as a once-a-day modified release formulation, compared to its IR counterpart (Chapter 6).

1.6 List of manuscripts and author contribution statement

In accordance with the University of Manchester guidance on the alternative thesis format for the Doctor of Philosophy degree, the following section describes the contribution of the candidate and co-authors to the manuscripts presented in this thesis that are either published or in preparation.

Chapter 2: Animal versus human oral bioavailability: do they correlate? Eur. J. Pharm. Sci. 57 (2014), 280-291

H. Musther: Set up the inclusion/exclusion criteria for the literature search, perform the analysis on the Grass and Sinko plot, write the introduction, part of the material/methods, part of the results and discussion, and conclusion sections of the manuscript. A. Olivares-Morales: Perform a literature search, collate literature data about human and animal oral bioavailability for 194 of a total of around 580 different compounds, statistical analysis of the entire dataset including regression analysis in Matlab, organization and management of the dataset, calculation of weighted means for the compounds with more than one reference for oral bioavailability, plotting the figures, data interpretation and analysis of the results, writing part of the materials/methods and results sections of the manuscript and contribute to the writing up of the manuscript and discussion. O.J.D. Hatley: Collate literature data about human and animal oral bioavailability for 1/3 of the total 580 compounds, and to contribute to the data analysis and discussion and the writing to the manuscript B. Liu: Collate literature data about human and animal oral bioavailability for 1/3 of the total 580 compounds, and to contribute to the data analysis and discussion and the writing to the manuscript A. Rostami-Hodjegan: Supervision and design of the research and to contribute to the writing process of the manuscript.

Chapter 3: The Use of ROC Analysis for the Qualitative Prediction of Human Oral Bioavailability from Animal Data Pharm Res. 31(2014), 720-730 A. Olivares-Morales: Led the research, literature search, design and conduct the research, perform the analysis of the data, interpretation of the results and writing of the manuscript and figures. O.J.D. Hatley: Provide expert advice on the preclinical data obtained from dogs and contribute to the discussion of the results and manuscript. D. Turner: Provide advice with the design of the analysis and contribute to the discussion of the manuscript. A. Galetin: Provide advice with the design of the analysis and contribute to the discussion of the manuscript. L. Aarons: Supervision and design of the research, expert opinion on the application of ROC analysis, and contribute to the writing and discussion of the manuscript. A. Rostami-Hodjegan: Supervision and design of the research, input in data analysis, and contribute to the writing and discussion of the manuscript.

Chapter 4: Analysis of the impact of controlled release formulation on oral drug absorption, gut wall metabolism and relative bioavailability of CYP3A substrates using a physiologically-based pharmacokinetic model Eur. J. Pharm. Sci. 67 (2015): 32-44 A. Olivares-Morales: Led the research, literature search, design and conduct the research, perform the analysis of the data, interpretation of the results and writing of the manuscript and figures. Y. Kamiyama: Perform the pilot study necessary to narrow dawn the multi-scale analysis A.S. Darwich: Help to design the research and to the interpretation of the outcome, contribute to writing the manuscript L. Aarons: Supervision and design of the research, input in data analysis, and contribute to the writing and discussion of the manuscript. A. Rostami-Hodjegan: Supervision and design of the research, input in data analysis, and contribute to the writing and discussion of the manuscript.

A. Olivares-Morales: Led the research, literature search, data collection, model development and implementation, analysis of the data, interpretation of the results and writing of the manuscript and figures. H. Lennernäs: Expert opinion on intestinal permeability, contribute to the discussion of the approach, contribution to the writing of the manuscript L. Aarons: Supervision and design of the research, and contribute to the writing and discussion of the manuscript. A. Rostami-Hodjegan: Supervision and design of the research, and contribute to the writing and discussion of the manuscript.

Chapter 6: Development of a novel simplified PBPK absorption model to explain the higher relative bioavailability of from the OROS formulation of oxybutynin [In preparation]

A. Olivares-Morales: Led the research, literature search, data collection, model development and implementation, analysis of the data, interpretation of the results and writing of the manuscript and figures. A. Ghosh: Provide the dataset for the analysis of the model outcome and contribute to the writing and discussion of the manuscript. L. Aarons: Supervision and design of the research, expert opinion on PBPK model building and implementation, contribute to the writing and discussion of the manuscript. A. Rostami-Hodjegan: Supervision and design of the research, and contribute to the writing and discussion of the manuscript.

1.8 References

1. M. Rowland and T. Tozer. Clinical pharmacokinetics: Concepts and Applications, Lippincott Williams and Wilkins1995. 2. L. Aarons. Physiologically based pharmacokinetic modelling: a sound mechanistic basis is needed. Br J Clin Pharmacol. 60:581-583 (2005). 3. C.P. Adams and V.V. Brantner. Estimating the cost of new drug development: is it really 802 million dollars? Health Aff (Millwood). 25:420- 428 (2006). 4. J.A. DiMasi, R.W. Hansen, and H.G. Grabowski. The price of innovation: new estimates of drug development costs. J Health Econ. 22:151-185 (2003). 5. L.B. Sheiner and J.-L. Steimer. Pharmacokinetic/Pharmacodynamic Modeling in Drug Development. Annu Rev Pharmacol Toxicol. 40:67-95 (2000). 6. H. Kubinyi. Drug research: myths, hype and reality. Nat Rev Drug Discov. 2:665-668 (2003). 7. I. Kola and J. Landis. Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov. 3:711-716 (2004). 8. R.A. Prentis, Y. Lis, and S.R. Walker. Pharmaceutical innovation by the seven UK-owned pharmaceutical companies (1964-1985). Br J Clin Pharmacol. 25:387-396 (1988). 9. E.H. Kerns and L. Di. Pharmaceutical profiling in drug discovery. Drug Discov Today. 8:316-323 (2003). 10. H.M. Jones, I.B. Gardner, and K.J. Watson. Modelling and PBPK simulation in drug discovery. AAPS J. 11:155-166 (2009). 11. A. Rostami-Hodjegan and G. Tucker. 'In silico' simulations to assess the 'in vivo' consequences of 'in vitro' metabolic drug-drug interactions. Drug Discov Today Technol. 1:441-448 (2004). 12. J. Lin, D.C. Sahakian, S.M. de Morais, J.J. Xu, R.J. Polzer, and S.M. Winter. The role of absorption, distribution, metabolism, excretion and toxicity in drug discovery. Curr Top Med Chem. 3:1125-1154 (2003). 13. P. Zou, Y. Yu, N. Zheng, Y. Yang, H.J. Paholak, L.X. Yu, and D. Sun. Applications of human pharmacokinetic prediction in first-in-human dose estimation. AAPS J. 14:262-281 (2012). 14. E.T. Hellriegel, T.D. Bjornsson, and W.W. Hauck. Interpatient variability in bioavailability is related to the extent of absorption: Implications for bioavailability and bioequivalence studies. Clin Pharmacol Ther. 60:601-607 (1996). 15. J.P. Rose. Effective Absorption Modeling in Relative Bioavailability Study Risk Assessment. The AAPS Journal. 14:895-903 (2012). 16. L.Z. Benet, F. Broccatelli, and T.I. Oprea. BDDCS applied to over 900 drugs. AAPS J. 13:519-547 (2011). 17. M.N. Martinez and G.L. Amidon. A mechanistic approach to understanding the factors affecting drug absorption: a review of fundamentals. J Clin Pharmacol. 42:620-643 (2002). 18. M. Gibaldi and D. Perrier. Absorption Kinetics and Bioavailability. Pharmacokinetics, Vol. 2, Informa Healthcare, New York, 1982, pp. 145-198.

19. L.Z. Benet, T. Izumi, Y. Zhang, J.A. Silverman, and V.J. Wacher. Intestinal MDR transport proteins and P-450 enzymes as barriers to oral drug delivery. J Control Release. 62:25-31 (1999). 20. M. Jamei, D. Turner, J. Yang, S. Neuhoff, S. Polak, A. Rostami-Hodjegan, and G. Tucker. Population-based mechanistic prediction of oral drug absorption. AAPS J. 11:225-237 (2009). 21. G.L. Amidon, H. Lennernas, V.P. Shah, and J.R. Crison. A theoretical basis for a biopharmaceutic drug classification: the correlation of in vitro drug product dissolution and in vivo bioavailability. Pharm Res. 12:413-420 (1995). 22. J.B. Dressman, G.L. Amidon, and D. Fleisher. Absorption potential: Estimating the fraction absorbed for orally administered compounds. J Pharm Sci. 74:588-589 (1985). 23. D.-M. Oh, R.L. Curl, and G.L. Amidon. Estimating the Fraction Dose Absorbed from Suspensions of Poorly Soluble Compounds in Humans: A Mathematical Model. Pharm Res. 10:264-270 (1993). 24. P.J. Sinko, G.D. Leesman, and G.L. Amidon. Predicting Fraction Dose Absorbed in Humans Using a Macroscopic Mass Balance Approach. Pharm Res. 8:979-988 (1991). 25. G.L. Amidon, P.J. Sinko, and D. Fleisher. Estimating Human Oral Fraction Dose Absorbed: A Correlation Using Rat Intestinal Membrane Permeability for Passive and Carrier-Mediated Compounds. Pharm Res. 5:651-654 (1988). 26. G.L. Amidon, J. Kou, R.L. Elliott, and E.N. Lightfoot. Analysis of models for determining intestinal wall permeabilities. J Pharm Sci. 69:1369-1373 (1980). 27. J. Valentin. Basic anatomical and physiological data for use in radiological protection: reference values. Ann ICRP. 32:1-277 (2002). 28. J.M. DeSesso and C.F. Jacobson. Anatomical and physiological parameters affecting gastrointestinal absorption in humans and rats. Food Chem Toxicol. 39:209-228 (2001). 29. L.X. Yu and G.L. Amidon. A compartmental absorption and transit model for estimating oral drug absorption. Int J Pharm. 186:119-125 (1999). 30. S.S. Davis, J.G. Hardy, and J.W. Fara. Transit of Pharmaceutical Dosage Forms through the Small-Intestine. Gut. 27:886-892 (1986). 31. L.X. Yu, J.R. Crison, and G.L. Amidon. Compartmental transit and dispersion model analysis of small intestinal transit flow in humans. Int J Pharm. 140:111-118 (1996). 32. J. Fallingborg, L.A. Christensen, M. Ingeman-Nielsen, B.A. Jacobsen, K. Abildgaard, and H.H. Rasmussen. pH-profile and regional transit times of the normal gut measured by a radiotelemetry device. Aliment Pharmacol Ther. 3:605-613 (1989). 33. P.J. Watts and L. Illum. Colonic drug delivery. Drug Dev Ind Pharm. 23:893- 913 (1997). 34. C. Tannergren, A. Bergendal, H. Lennernas, and B. Abrahamsson. Toward an increased understanding of the barriers to colonic drug absorption in humans: implications for early controlled release candidate assessment. Mol Pharm. 6:60-73 (2009). 35. A.J. Coupe, S.S. Davis, D.F. Evans, and I.R. Wilding. Nocturnal Scintigraphic Imaging to Investigate the Gastrointestinal Transit of Dosage Forms. J Controlled Release. 20:155-162 (1992).

36. M. Bouchoucha, G. Devroede, P. Arhan, B. Strom, J. Weber, P.-H. Cugnenc, P. Denis, and J.-P. Barbier. What is the meaning of colorectal transit time measurement? Dis Colon Rectum. 35:773-782 (1992). 37. M. Bouchoucha, G. Devroede, E. Dorval, A. Faye, P. Arhan, and M. Arsac. Different segmental transit times in patients with irritable bowel syndrome and "normal" colonic transit time: is there a correlation with symptoms? Tech Coloproctol. 10:287-296 (2006). 38. W. Weitschies, H. Blume, and H. Monnikes. Magnetic marker monitoring: high resolution real-time tracking of oral solid dosage forms in the gastrointestinal tract. Eur J Pharm Biopharm. 74:93-101 (2010). 39. F. Podczeck, N.C. Course, J.M. Newton, and M.B. Short. The influence of non-disintegrating tablet dimensions and density on their gastric emptying in fasted volunteers. J Pharm Pharmacol. 59:23-27 (2007). 40. S. Hellmig, F. Von Schoning, C. Gadow, S. Katsoulis, J. Hedderich, U.R. Folsch, and E. Stuber. Gastric emptying time of fluids and solids in healthy subjects determined by 13C breath tests: influence of age, sex and body mass index. J Gastroenterol Hepatol. 21:1832-1838 (2006). 41. A. Steingoetter, M. Fox, R. Treier, D. Weishaupt, B. Marincek, P. Boesiger, M. Fried, and W. Schwizer. Effects of posture on the physiology of gastric emptying: a magnetic resonance imaging study. Scand J Gastroenterol. 41:1155-1164 (2006). 42. L.X. Yu and G.L. Amidon. Saturable small intestinal drug absorption in humans: modeling and interpretation of cefatrizine data. Eur J Pharm Biopharm. 45:199-203 (1998). 43. International Commission on Radiological Protection. Report of the Task Group on Reference Man, Pergamon Press1975. 44. OpenStax College. The Small and Large Intestines. In O. CNX (ed.), http://cnx.org/contents/430d8e5a-f699-4949-9849-5b4ca1125fa2@4, 2013. 45. H. Lennernas. Human intestinal permeability. J Pharm Sci. 87:403-410 (1998). 46. T.T. Kararli. Comparison of the gastrointestinal anatomy, physiology, and biochemistry of humans and commonly used laboratory animals. Biopharm Drug Dispos. 16:351-380 (1995). 47. A.W. Larhed, P. Artursson, and E. Bjork. The influence of intestinal mucus components on the diffusion of drugs. Pharm Res. 15:66-71 (1998). 48. U. Fagerholm and H. Lennernäs. Experimental estimation of the effective unstirred water layer thickness in the human jejunum, and its importance in oral drug absorption. Eur J Pharm Sci. 3:247-253 (1995). 49. M.D. Levitt, J.K. Furne, A. Strocchi, B.W. Anderson, and D.G. Levitt. Physiological measurements of luminal stirring in the dog and human small bowel. J Clin Invest. 86:1540-1547 (1990). 50. A. Strocchi, G. Corazza, J. Furne, C. Fine, A. Di Sario, G. Gasbarrini, and M.D. Levitt. Measurements of the jejunal unstirred layer in normal subjects and patients with celiac disease. American Journal of Physiology - Gastrointestinal and Liver Physiology. 270:G487-G491 (1996). 51. M.D. Levitt, A. Strocchi, and D.G. Levitt. Human jejunal unstirred layer: evidence for extremely efficient luminal stirring. American Journal of Physiology - Gastrointestinal and Liver Physiology. 262:G593-G596 (1992).

52. W.L. Chiou. Effect of ‘unstirred’ water layer in the intestine on the rate and extent of absorption after oral administration. Biopharm Drug Dispos. 15:709-717 (1994). 53. A. El-Kattan and M. Varma. Oral Absorption, Intestinal Metabolism and Human Oral Bioavailability. In J. Paxton (ed.), Topics on Drug Metabolism. 10.5772/31087, InTech2012. 54. K. Sugano. Theoretical Framework III: Biological Membrane Permeation. Biopharmaceutics Modeling and Simulations. 10.1002/9781118354339.ch4, John Wiley & Sons, Inc.2012, pp. 64-121. 55. K. Sugano, M. Kansy, P. Artursson, A. Avdeef, S. Bendels, L. Di, G.F. Ecker, B. Faller, H. Fischer, G. Gerebtzoff, H. Lennernaes, and F. Senner. Coexistence of passive and carrier-mediated processes in drug transport. Nat Rev Drug Discov. 9:597-614 (2010). 56. V.H. Thomas, S. Bhattachar, L. Hitchingham, P. Zocharski, M. Naath, N. Surendran, C.L. Stoner, and A. El-Kattan. The road map to oral bioavailability: an industrial perspective. Expert Opin Drug Metab Toxicol. 2:591-608 (2006). 57. H. van de Waterbeemd, D.A. Smith, K. Beaumont, and D.K. Walker. Property-Based Design: Optimization of Drug Absorption and Pharmacokinetics. J Med Chem. 44:1313-1333 (2001). 58. P. Stenberg, K. Luthman, and P. Artursson. Virtual screening of intestinal drug permeability. J Control Release. 65:231-243 (2000). 59. D.F. Veber, S.R. Johnson, H.Y. Cheng, B.R. Smith, K.W. Ward, and K.D. Kopple. Molecular properties that influence the oral bioavailability of drug candidates. J Med Chem. 45:2615-2623 (2002). 60. K. Palm, P. Stenberg, K. Luthman, and P. Artursson. Polar Molecular Surface Properties Predict the Intestinal Absorption of Drugs in Humans. Pharm Res. 14:568-571 (1997). 61. K. Palm, K. Luthman, A.-L. Ungell, G. Strandlund, and P. Artursson. Correlation of drug absorption with molecular surface properties. J Pharm Sci. 85:32-39 (1996). 62. C.A. Lipinski, F. Lombardo, B.W. Dominy, and P.J. Feeney. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv Drug Deliv Rev. 46:3-26 (2001). 63. C.A. Lipinski. Drug-like properties and the causes of poor solubility and poor permeability. J Pharmacol Toxicol Methods. 44:235-249 (2000). 64. K.W. Ward, R. Nagilla, and L.J. Jolivette. Comparative evaluation of oral systemic exposure of 56 xenobiotics in rat, dog, monkey and human. Xenobiotica. 35:191-210 (2005). 65. M.V. Varma, R.S. Obach, C. Rotter, H.R. Miller, G. Chang, S.J. Steyn, A. El- Kattan, and M.D. Troutman. Physicochemical space for optimum oral bioavailability: contribution of human intestinal absorption and first-pass elimination. J Med Chem. 53:1098-1108 (2010). 66. H. Lennernäs. Does Fluid Flow Across the Intestinal Mucosa Affect Quantitative Oral Drug Absorption? Is It Time for a Reevaluation? Pharm Res. 12:1573-1582 (1995). 67. K.M. Giacomini, S.M. Huang, D.J. Tweedie, L.Z. Benet, K.L. Brouwer, X. Chu, A. Dahlin, R. Evers, V. Fischer, K.M. Hillgren, K.A. Hoffmaster, T. Ishikawa, D. Keppler, R.B. Kim, C.A. Lee, M. Niemi, J.W. Polli, Y. Sugiyama, P.W. Swaan, J.A. Ware, S.H. Wright, S.W. Yee, M.J. Zamek- 68

Gliszczynski, L. Zhang, and International Transporter Consortium. Membrane transporters in drug development. Nat Rev Drug Discov. 9:215- 236 (2010). 68. M. Estudante, J.G. Morais, G. Soveral, and L.Z. Benet. Intestinal drug transporters: An overview, Advanced Drug Delivery Reviews, Vol. 652013, pp. 1340-1356. 69. S. Shugarts and L.Z. Benet. The role of transporters in the pharmacokinetics of orally administered drugs. Pharm Res. 26:2039-2054 (2009). 70. L.M. Chan, S. Lowes, and B.H. Hirst. The ABCs of drug transport in intestine and liver: efflux proteins limiting drug absorption and bioavailability. Eur J Pharm Sci. 21:25-51 (2004). 71. J.R. Kunta and P.J. Sinko. Intestinal drug transporters: in vivo function and clinical importance. Curr Drug Metab. 5:109-124 (2004). 72. A. Bruyere, X. Decleves, F. Bouzom, K. Ball, C. Marques, X. Treton, M. Pocard, P. Valleur, Y. Bouhnik, Y. Panis, J.M. Scherrmann, and S. Mouly. Effect of variations in the amounts of P-glycoprotein (ABCB1), BCRP (ABCG2) and CYP3A4 along the human small intestine on PBPK models for predicting intestinal first pass. Mol Pharm. 7:1596-1607 (2010). 73. S. Mouly and M.F. Paine. P-glycoprotein increases from proximal to distal regions of human small intestine. Pharm Res. 20:1595-1599 (2003). 74. C.Y. Wu and L.Z. Benet. Predicting drug disposition via application of BCS: Transport/absorption/elimination interplay and development of a biopharmaceutics drug disposition classification system. Pharm Res. 22:11- 23 (2005). 75. G. Englund, F. Rorsman, A. Ronnblom, U. Karlbom, L. Lazorova, J. Grasjo, A. Kindmark, and P. Artursson. Regional levels of drug transporters along the human intestinal tract: co-expression of ABC and SLC transporters and comparison with Caco-2 cells. Eur J Pharm Sci. 29:269-277 (2006). 76. T.G. Tucker, A.M. Milne, S. Fournel-Gigleux, K.S. Fenner, and M.W. Coughtrie. Absolute immunoquantification of the expression of ABC transporters P-glycoprotein, breast cancer resistance protein and multidrug resistance-associated protein 2 in human liver and duodenum. Biochem Pharmacol. 83:279-285 (2012). 77. S. Berggren, C. Gall, N. Wollnitz, M. Ekelund, U. Karlbom, J. Hoogstraate, D. Schrenk, and H. Lennernas. Gene and protein expression of P- glycoprotein, MRP1, MRP2, and CYP3A4 in the small and large human intestine. Mol Pharm. 4:252-257 (2007). 78. T. Ogihara, T. Kano, T. Wagatsuma, S. Wada, H. Yabuuchi, S. Enomoto, K. Morimoto, Y. Shirasaka, S. Kobayashi, and I. Tamai. Oseltamivir (tamiflu) is a substrate of peptide transporter 1. Drug Metab Dispos. 37:1676-1681 (2009). 79. K. Morimoto, K. Kishimura, T. Nagami, N. Kodama, Y. Ogama, M. Yokoyama, S. Toda, T. Chiyoda, R. Shimada, A. Inano, T. Kano, I. Tamai, and T. Ogihara. Effect of milk on the pharmacokinetics of oseltamivir in healthy volunteers. J Pharm Sci. 100:3854-3861 (2011). 80. J.J. Lilja, K. Raaska, and P.J. Neuvonen. Effects of orange juice on the pharmacokinetics of atenolol. Eur J Clin Pharmacol. 61:337-340 (2005). 81. J. Imanaga, T. Kotegawa, H. Imai, K. Tsutsumi, T. Yoshizato, T. Ohyama, Y. Shirasaka, I. Tamai, T. Tateishi, and K. Ohashi. The effects of the SLCO2B1 c.1457C > T polymorphism and apple juice on the pharmacokinetics of 69

fexofenadine and midazolam in humans. Pharmacogenet Genomics. 21:84-93 (2011). 82. U.I. Schwarz, D. Seemann, R. Oertel, S. Miehlke, E. Kuhlisch, M.F. Fromm, R.B. Kim, D.G. Bailey, and W. Kirch. Grapefruit juice ingestion significantly reduces talinolol bioavailability. Clin Pharmacol Ther. 77:291-301 (2005). 83. H. Lennernas. Intestinal permeability and its relevance for absorption and elimination. Xenobiotica. 37:1015-1051 (2007). 84. H. Lennernas, O. Ahrenstedt, R. Hallgren, L. Knutson, M. Ryde, and L.K. Paalzow. Regional jejunal perfusion, a new in vivo approach to study oral drug absorption in man. Pharm Res. 9:1243-1251 (1992). 85. L. Knutson, B. Odlind, and R. Hallgren. A new technique for segmental jejunal perfusion in man. Am J Gastroenterol. 84:1278-1284 (1989). 86. T. Knutson, P. Fridblom, H. Ahlstrom, A. Magnusson, C. Tannergren, and H. Lennernas. Increased understanding of intestinal drug permeability determined by the LOC-I-GUT approach using multislice computed tomography. Mol Pharm. 6:2-10 (2009). 87. H. Lennernas, I.D. Lee, U. Fagerholm, and G.L. Amidon. A residence-time distribution analysis of the hydrodynamics within the intestine in man during a regional single-pass perfusion with Loc-I-Gut: In-vivo permeability estimation. J Pharm Pharmacol. 49:682-686 (1997). 88. A. Dahan, J.M. Miller, J.M. Hilfinger, S. Yamashita, L.X. Yu, H. Lennernas, and G.L. Amidon. High-permeability criterion for BCS classification: segmental/pH dependent permeability considerations. Mol Pharm. 7:1827- 1834 (2010). 89. A. Avdeef and K.Y. Tam. How well can the Caco-2/Madin-Darby canine kidney models predict effective human jejunal permeability? J Med Chem. 53:3566-3584 (2010). 90. J.D. Irvine, L. Takahashi, K. Lockhart, J. Cheong, J.W. Tolan, H.E. Selick, and J.R. Grove. MDCK (Madin–Darby canine kidney) cells: A tool for membrane permeability screening. J Pharm Sci. 88:28-33 (1999). 91. P. Artursson. Epithelial transport of drugs in cell culture. I: A model for studying the passive diffusion of drugs over intestinal absorbtive (Caco-2) cells. J Pharm Sci. 79:476-482 (1990). 92. P. Shah, V. Jogani, T. Bagchi, and A. Misra. Role of Caco-2 cell monolayers in prediction of intestinal drug absorption. Biotechnol Prog. 22:186-198 (2006). 93. P. Artursson, K. Palm, and K. Luthman. Caco-2 monolayers in experimental and theoretical predictions of drug transport. Advanced Drug Delivery Reviews. 46:27-43 (2001). 94. F. Tang, K. Horie, and R.T. Borchardt. Are MDCK cells transfected with the human MDR1 gene a good model of the human intestinal mucosa? Pharm Res. 19:765-772 (2002). 95. D.A. Volpe. Variability in Caco-2 and MDCK cell-based intestinal permeability assays. J Pharm Sci. 97:712-725 (2008). 96. D.A. Volpe. Drug-permeability and transporter assays in Caco-2 and MDCK cell lines. Future Med Chem. 3:2063-2077 (2011). 97. M. Gertz, A. Harrison, J.B. Houston, and A. Galetin. Prediction of human intestinal first-pass metabolism of 25 CYP3A substrates from in vitro clearance and permeability data. Drug Metab Dispos. 38:1147-1158 (2010).

98. P. Artursson and J. Karlsson. Correlation between oral drug absorption in humans and apparent drug permeability coefficients in human intestinal epithelial (Caco-2) cells. Biochem Biophys Res Commun. 175:880-885 (1991). 99. D. Sun, H. Lennernas, L.S. Welage, J.L. Barnett, C.P. Landowski, D. Foster, D. Fleisher, K.D. Lee, and G.L. Amidon. Comparison of human duodenum and Caco-2 gene expression profiles for 12,000 gene sequences tags and correlation with permeability of 26 drugs. Pharm Res. 19:1400-1416 (2002). 100. M. Kansy, F. Senner, and K. Gubernator. Physicochemical High Throughput Screening: Parallel Artificial Membrane Permeation Assay in the Description of Passive Absorption Processes. J Med Chem. 41:1007-1010 (1998). 101. H. Lennernäs, S. Nylander, and A.L. Ungell. Jejunal Permeability: A Comparison Between the Ussing Chamber Technique and the Single-Pass Perfusion in Humans. Pharm Res. 14:667-671 (1997). 102. A. Avdeef, P. Artursson, S. Neuhoff, L. Lazorova, J. Grasjo, and S. Tavelin. Caco-2 permeability of weakly basic drugs predicted with the double-sink PAMPA pKa(flux) method. Eur J Pharm Sci. 24:333-349 (2005). 103. M. Fujikawa, R. Ano, K. Nakao, R. Shimizu, and M. Akamatsu. Relationships between structure and high-throughput screening permeability of diverse drugs with artificial membranes: application to prediction of Caco- 2 cell permeability. Bioorg Med Chem. 13:4721-4732 (2005). 104. A. Sjoberg, M. Lutz, C. Tannergren, C. Wingolf, A. Borde, and A.L. Ungell. Comprehensive study on regional human intestinal permeability and prediction of fraction absorbed of drugs using the Ussing chamber technique. Eur J Pharm Sci. 48:166-180 (2013). 105. H. Lennernas. Animal data: the contributions of the Ussing Chamber and perfusion systems to predicting human oral drug delivery in vivo. Adv Drug Deliv Rev. 59:1103-1120 (2007). 106. D. Galinis-Luciani, L. Nguyen, and M. Yazdanian. Is PAMPA a useful tool for discovery? J Pharm Sci. 96:2886-2892 (2007). 107. A. Avdeef, S. Bendels, L. Di, B. Faller, M. Kansy, K. Sugano, and Y. Yamauchi. PAMPA--critical factors for better predictions of absorption. J Pharm Sci. 96:2893-2909 (2007). 108. A. Avdeef. The rise of PAMPA. Expert Opin Drug Metab Toxicol. 1:325-342 (2005). 109. A.L. Ungell, S. Nylander, S. Bergstrand, A. Sjoberg, and H. Lennernas. Membrane transport of drugs in different regions of the intestinal tract of the rat. J Pharm Sci. 87:360-366 (1998). 110. S. Winiwarter, N.M. Bonham, F. Ax, A. Hallberg, H. Lennernas, and A. Karlen. Correlation of human jejunal permeability (in vivo) of drugs with experimentally and theoretically derived parameters. A multivariate data analysis approach. J Med Chem. 41:4939-4949 (1998). 111. U.S. Food and Drug Administration. Guidance for Industry: Waiver of In Vivo Bioavailability and Bioequivalence Studies for Immediate-Release Solid Oral Dosage Forms Based on a Biopharmaceutics Classification System, http://www.fda.gov/downloads/Drugs/.../Guidances/ucm070246.pdf, 2000. 112. European Medicines Agency. Guideline on the investigation of bioequivalence,

http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guidelin e/2010/01/WC500070039.pdf, 2010. 113. K. Sugano, A. Okazaki, S. Sugimoto, S. Tavornvipas, A. Omura, and T. Mano. Solubility and dissolution profile assessment in drug discovery. Drug Metab Pharmacokinet. 22:225-254 (2007). 114. J. Brouwers, M.E. Brewster, and P. Augustijns. Supersaturating drug delivery systems: the answer to solubility-limited oral bioavailability? J Pharm Sci. 98:2549-2572 (2009). 115. K.J. Box, G. Volgyi, E. Baka, M. Stuart, K. Takacs-Novak, and J.E. Comer. Equilibrium versus kinetic measurements of aqueous solubility, and the ability of compounds to supersaturate in solution--a validation study. J Pharm Sci. 95:1298-1307 (2006). 116. S.H. Yalkowsky and S.C. Valvani. Solubility and partitioning I: Solubility of nonelectrolytes in water. J Pharm Sci. 69:912-922 (1980). 117. A. Avdeef. Solubility of sparingly-soluble ionizable drugs. Adv Drug Deliv Rev. 59:568-590 (2007). 118. Y.Q. Ran, Y. He, G. Yang, J.L.H. Johnson, and S.H. Yalkowsky. Estimation of aqueous solubility of organic compounds by using the general solubility equation. Chemosphere. 48:487-509 (2002). 119. N. Jain and S.H. Yalkowsky. Estimation of the aqueous solubility I: Application to organic nonelectrolytes. J Pharm Sci. 90:234-252 (2001). 120. G. Klopman, S. Wang, and D.M. Balthasar. Estimation of aqueous solubility of organic molecules by the group contribution approach. Application to the study of biodegradation. J Chem Inf Comput Sci. 32:474-482 (1992). 121. R. Kühne, R.U. Ebert, F. Kleint, G. Schmidt, and G. Schüürmann. Group contribution methods to estimate water solubility of organic chemicals. Chemosphere. 30:2061-2077 (1995). 122. M.H. Abraham and J. Le. The correlation and prediction of the solubility of compounds in water using an amended solvation energy relationship. J Pharm Sci. 88:868-880 (1999). 123. W.L. Jorgensen and E.M. Duffy. Prediction of drug solubility from structure. Adv Drug Deliv Rev. 54:355-366 (2002). 124. K. Sugano. Introduction to computational oral absorption simulation. Expert Opin Drug Metab Toxicol. 5:259-293 (2009). 125. H.N.E. Stevens, C.G. Wilson, P.G. Welling, M. Bakhshaee, J.S. Binns, A.C. Perkins, M. Frier, E.P. Blackshaw, M.W. Frame, D.J. Nichols, M.J. Humphrey, and S.R. Wicks. Evaluation of Pulsincap (TM) to provide regional delivery of dofetilide to the human GI tract. Int J Pharm. 236:27-34 (2002). 126. A. Rubinstein. Colonic drug delivery. Drug Discov Today Technol. 2:33-37 (2005). 127. R.A. Siegel and M.J. Rathbone. Overview of controlled release mechanisms. Fundamentals and Applications of Controlled Release Drug Delivery, Springer2012, pp. 19-43. 128. J. Siepmann and F. Siepmann. Mathematical modeling of drug delivery. Int J Pharm. 364:328-343 (2008). 129. G. Van den Mooter. Colon drug delivery. Expert Opin Drug Deliv. 3:111-125 (2006). 130. A. Streubel, J. Siepmann, and R. Bodmeier. Gastroretentive drug delivery systems. Expert Opin Drug Deliv. 3:217-233 (2006). 72

131. R. Conley, S.K. Gupta, and G. Sathyan. Clinical spectrum of the osmotic- controlled release oral delivery system (OROS), an advanced oral delivery form. Curr Med Res Opin. 22:1879-1892 (2006). 132. R. Langer. New methods of drug delivery. Science. 249:1527-1533 (1990). 133. A. Dokoumetzidis and P. Macheras. A century of dissolution research: from Noyes and Whitney to the biopharmaceutics classification system. Int J Pharm. 321:1-11 (2006). 134. S.D. Mithani, V. Bakatselou, C.N. TenHoor, and J.B. Dressman. Estimation of the increase in solubility of drugs as a function of bile salt concentration. Pharm Res. 13:163-167 (1996). 135. R.N. Gursoy and S. Benita. Self-emulsifying drug delivery systems (SEDDS) for improved oral delivery of lipophilic drugs. Biomed Pharmacother. 58:173-182 (2004). 136. J. Dressman, G. Amidon, C. Reppas, and V. Shah. Dissolution Testing as a Prognostic Tool for Oral Drug Absorption: Immediate Release Dosage Forms. Pharm Res. 15:11-22 (1998). 137. K. Johnson and A. Swindell. Guidance in the Setting of Drug Particle Size Specifications to Minimize Variability in Absorption. Pharm Res. 13:1795- 1798 (1996). 138. P. Harriott. Mass transfer to particles: Part I. Suspended in agitated tanks. AIChE Journal. 8:93-101 (1962). 139. R.J. Hintz and K.C. Johnson. The effect of particle size distribution on dissolution rate and oral absorption. Int J Pharm. 51:9-17 (1989). 140. J. Wang and D.R. Flanagan. General solution for diffusion-controlled dissolution of spherical particles. 2. Evaluation of experimental data. J Pharm Sci. 91:534-542 (2002). 141. J. Wang and D.R. Flanagan. General solution for diffusion-controlled dissolution of spherical particles. 1. Theory. J Pharm Sci. 88:731-738 (1999). 142. K. Sugano. Theoretical comparison of hydrodynamic diffusion layer models used for dissolution simulation in drug discovery and development. Int J Pharm. 363:73-77 (2008). 143. J.B. Dressman, K. Thelen, and E. Jantratid. Towards quantitative prediction of oral drug absorption. Clin Pharmacokinet. 47:655-667 (2008). 144. V.H. Sunesen, B.L. Pedersen, H.G. Kristensen, and A. Mullertz. In vivo in vitro correlations for a poorly soluble drug, danazol, using the flow-through dissolution method with biorelevant dissolution media. Eur J Pharm Sci. 24:305-313 (2005). 145. V.R.S. Uppoor. Regulatory perspectives on in vitro (dissolution)/in vivo (bioavailability) correlations. J Controlled Release. 72:127-132 (2001). 146. S. Klein. The use of biorelevant dissolution media to forecast the in vivo performance of a drug. AAPS J. 12:397-406 (2010). 147. E. Nicolaides, M. Symillides, J. Dressman, and C. Reppas. Biorelevant Dissolution Testing to Predict the Plasma Profile of Lipophilic Drugs After Oral Administration. Pharm Res. 18:380-388 (2001). 148. J. Dressman, R. Berardi, L. Dermentzoglou, T. Russell, S. Schmaltz, J. Barnett, and K. Jarvenpaa. Upper Gastrointestinal (GI) pH in Young, Healthy Men and Women. Pharm Res. 7:756-761 (1990). 149. D.M. Mudie, G.L. Amidon, and G.E. Amidon. Physiological parameters for oral delivery and in vitro testing. Mol Pharm. 7:1388-1405 (2010).

150. L. Kalantzi, K. Goumas, V. Kalioras, B. Abrahamsson, J.B. Dressman, and C. Reppas. Characterization of the human upper gastrointestinal contents under conditions simulating bioavailability/bioequivalence studies. Pharm Res. 23:165-176 (2006). 151. K. Sugano. Estimation of effective intestinal membrane permeability considering bile micelle solubilisation. Int J Pharm. 368:116-122 (2009). 152. C. Schiller, C.P. Frohlich, T. Giessmann, W. Siegmund, H. Monnikes, N. Hosten, and W. Weitschies. Intestinal fluid volumes and transit of dosage forms as assessed by magnetic resonance imaging. Aliment Pharmacol Ther. 22:971-979 (2005). 153. L. Yu. An Integrated Model for Determining Causes of Poor Oral Drug Absorption. Pharm Res. 16:1883-1887 (1999). 154. L.X. Yu and G.L. Amidon. Characterization of small intestinal transit time distribution in humans. Int J Pharm. 171:157-163 (1998). 155. B. Davies and T. Morris. Physiological Parameters in Laboratory Animals and Humans. Pharm Res. 10:1093-1095 (1993). 156. G.R. Wilkinson. Drug metabolism and variability among patients in drug response. N Engl J Med. 352:2211-2221 (2005). 157. D. Greenblatt. Presystemic extraction: mechanisms and consequences. The Journal of Clinical Pharmacology. 33:650-656 (1993). 158. L.C. Wienkers and T.G. Heath. Predicting in vivo drug interactions from in vitro drug discovery data. Nat Rev Drug Discov. 4:825-833 (2005). 159. K. Thelen and J.B. Dressman. Cytochrome P450-mediated metabolism in the human gut wall. J Pharm Pharmacol. 61:541-558 (2009). 160. M. Roberts and M. Rowland. A dispersion model of hepatic elimination: 1. Formulation of the model and bolus considerations. J Pharmacokinet Biopharm. 14:227-260 (1986). 161. K.S. Pang and M. Rowland. Hepatic clearance of drugs. I. Theoretical considerations of a “well-stirred” model and a “parallel tube” model. Influence of hepatic blood flow, plasma and blood cell binding, and the hepatocellular enzymatic activity on hepatic drug clearance. J Pharmacokinet Biopharm. 5:625-653 (1977). 162. M. Rowland, L. Benet, and G. Graham. Clearance concepts in pharmacokinetics. J Pharmacokinet Biopharm. 1:123-136 (1973). 163. R.S. Obach. Prediction of Human Clearance of Twenty-Nine Drugs from Hepatic Microsomal Intrinsic Clearance Data: An Examination of In Vitro Half-Life Approach and Nonspecific Binding to Microsomes. Drug Metab Dispos. 27:1350-1359 (1999). 164. K. Ito and J.B. Houston. Comparison of the use of liver models for predicting drug clearance using in vitro kinetic data from hepatic microsomes and isolated hepatocytes. Pharm Res. 21:785-792 (2004). 165. M.F. Paine, H.L. Hart, S.S. Ludington, R.L. Haining, A.E. Rettie, and D.C. Zeldin. The human intestinal cytochrome P450 "pie". Drug Metab Dispos. 34:880-886 (2006). 166. T. Shimada, H. Yamazaki, M. Mimura, Y. Inui, and F.P. Guengerich. Interindividual variations in human liver cytochrome P-450 enzymes involved in the oxidation of drugs, carcinogens and toxic chemicals: studies with liver microsomes of 30 Japanese and 30 Caucasians. J Pharmacol Exp Ther. 270:414-423 (1994).

167. K. Rowland Yeo, A. Rostami-Hodjegan, and G.T. Tucker. Abundance of Cytochromes P450 in Human Liver: A Meta-Analysis. Br J Clin Pharmacol. 57:687-688 (2004). 168. W.G. Levine. Biliary excretion of drugs and other xenobiotics. Annu Rev Pharmacol Toxicol. 18:81-96 (1978). 169. G. Ghibellini, E.M. Leslie, and K.L.R. Brouwer. Methods To Evaluate Biliary Excretion of Drugs in Humans: An Updated Review. Mol Pharm. 3:198-211 (2006). 170. X. Yang, Y.A. Gandhi, D.B. Duignan, and M.E. Morris. Prediction of biliary excretion in rats and humans using molecular weight and quantitative structure-pharmacokinetic relationships. AAPS J. 11:511-525 (2009). 171. W.E. Wright and V.D. Line. Biliary excretion of cephalosporins in rats: influence of molecular weight. Antimicrob Agents Chemother. 17:842-846 (1980). 172. J.B. Houston. Utility of in vitro drug metabolism data in predicting in vivo metabolic clearance. Biochem Pharmacol. 47:1469-1479 (1994). 173. N.J. Proctor, G.T. Tucker, and A. Rostami-Hodjegan. Predicting drug clearance from recombinantly expressed CYPs: intersystem extrapolation factors. Xenobiotica. 34:151-178 (2004). 174. T. Iwatsubo, N. Hirota, T. Ooie, H. Suzuki, N. Shimada, K. Chiba, T. Ishizaki, C.E. Green, C.A. Tyson, and Y. Sugiyama. Prediction of in vivo drug metabolism in the human liver from in vitro metabolism data. Pharmacol Ther. 73:147-171 (1997). 175. E.M. Howgate, K. Rowland Yeo, N.J. Proctor, G.T. Tucker, and A. Rostami- Hodjegan. Prediction of in vivo drug clearance from in vitro data. I: impact of inter-individual variability. Xenobiotica. 36:473-497 (2006). 176. J.H. Lin, M. Chiba, and T.A. Baillie. Is the Role of the Small Intestine in First-Pass Metabolism Overemphasized? Pharmacol Rev. 51:135-158 (1999). 177. O. von Richter, O. Burk, M.F. Fromm, K.P. Thon, M. Eichelbaum, and K.T. Kivisto. Cytochrome P450 3A4 and P-glycoprotein Expression in Human Small Intestinal Enterocytes and Hepatocytes: A Comparative Analysis in Paired Tissue Specimens[ast]. Clin Pharmacol Ther. 75:172-183 (2004). 178. M.F. Paine, M. Khalighi, J.M. Fisher, D.D. Shen, K.L. Kunze, C.L. Marsh, J.D. Perkins, and K.E. Thummel. Characterization of Interintestinal and Intraintestinal Variations in Human CYP3A-Dependent Metabolism. J Pharmacol Exp Ther. 283:1552-1562 (1997). 179. J.C. Kolars, K.S. Lown, P. Schmiedlin-Ren, M. Ghosh, C. Fang, S.A. Wrighton, R.M. Merion, and P.B. Watkins. CYP3A gene expression in human gut epithelium. Pharmacogenetics. 4:247-259 (1994). 180. S.K. Gupta and G. Sathyan. Pharmacokinetics of an oral once-a-day controlled-release oxybutynin formulation compared with immediate-release oxybutynin. J Clin Pharmacol. 39:289-296 (1999). 181. A. Sakr and M. Andheria. Pharmacokinetics of buspirone extended-release tablets: a single-dose study. J Clin Pharmacol. 41:783-789 (2001). 182. J. Yang, G.T. Tucker, and A. Rostami-Hodjegan. Cytochrome P450 3A expression and activity in the human small intestine. Clin Pharmacol Ther. 76:391 (2004). 183. M.F. Paine, D.D. Shen, K.L. Kunze, J.D. Perkins, C.L. Marsh, J.P. McVicar, D.M. Barr, B.S. Gillies, and K.E. Thummel. First-pass metabolism of

midazolam by the human intestine[ast]. Clin Pharmacol Ther. 60:14-24 (1996). 184. A. Galetin and J.B. Houston. Intestinal and hepatic metabolic activity of five cytochrome P450 enzymes: impact on prediction of first-pass metabolism. J Pharmacol Exp Ther. 318:1220-1229 (2006). 185. M. Kato, K. Chiba, A. Hisaka, M. Ishigami, M. Kayama, N. Mizuno, Y. Nagata, S. Takakuwa, Y. Tsukamoto, K. Ueda, H. Kusuhara, K. Ito, and Y. Sugiyama. The intestinal first-pass metabolism of substrates of CYP3A4 and P-glycoprotein-quantitative analysis based on information from the literature. Drug Metab Pharmacokinet. 18:365-372 (2003). 186. M. Kato. Intestinal first-pass metabolism of CYP3A4 substrates. Drug Metab Pharmacokinet. 23:87-94 (2008). 187. L.Z. Benet. The drug transporter-metabolism alliance: uncovering and defining the interplay. Mol Pharm. 6:1631-1643 (2009). 188. P.B. Watkins. The barrier function of CYP3A4 and P-glycoprotein in the small bowel. Advanced Drug Delivery Reviews. 27:161-170 (1997). 189. A. Galetin, H. Burt, L. Gibbons, and J.B. Houston. Prediction of time- dependent CYP3A4 drug-drug interactions: impact of enzyme degradation, parallel elimination pathways, and intestinal inhibition. Drug Metab Dispos. 34:166-175 (2006). 190. M. Jamei, G.L. Dickinson, and A. Rostami-Hodjegan. A framework for assessing inter-individual variability in pharmacokinetics using virtual human populations and integrating general knowledge of physical chemistry, biology, anatomy, physiology and genetics: A tale of 'bottom-up' vs 'top- down' recognition of covariates. Drug Metab Pharmacokinet. 24:53-75 (2009). 191. A.S. Darwich, S. Neuhoff, M. Jamei, and A. Rostami-Hodjegan. Interplay of metabolism and transport in determining oral drug absorption and gut wall metabolism: a simulation assessment using the "Advanced Dissolution, Absorption, Metabolism (ADAM)" model. Curr Drug Metab. 11:716-729 (2010). 192. I.A. de Graaf, R. de Kanter, M.H. de Jager, R. Camacho, E. Langenkamp, E.G. van de Kerkhof, and G.M. Groothuis. Empirical validation of a rat in vitro organ slice model as a tool for in vivo clearance prediction. Drug Metab Dispos. 34:591-599 (2006). 193. E.G. van de Kerkhof, I.A. de Graaf, and G.M. Groothuis. In vitro methods to study intestinal drug metabolism. Curr Drug Metab. 8:658-675 (2007). 194. C. Logan. Use of Animals for the Determination of Absorption and Bioavailability. Drug Bioavailability. 10.1002/9783527623860.ch8, Wiley- VCH Verlag GmbH & Co. KGaA2009, pp. 161-184. 195. R.S. Obach, J.G. Baxter, T.E. Liston, B.M. Silber, B.C. Jones, F. Macintyre, D.J. Rance, and P. Wastall. The Prediction of Human Pharmacokinetic Parameters from Preclinical and In Vitro Metabolism Data. J Pharmacol Exp Ther. 283:46-58 (1997). 196. T. Lave, P. Coassolo, and B. Reigner. Prediction of hepatic metabolic clearance based on interspecies allometric scaling techniques and in vitro-in vivo correlations. Clin Pharmacokinet. 36:211-231 (1999). 197. H. Boxenbaum. Interspecies scaling, allometry, physiological time, and the ground plan of pharmacokinetics. J Pharmacokinet Biopharm. 10:201-227 (1982). 76

198. I. Mahmood and J.D. Balian. Interspecies scaling: predicting pharmacokinetic parameters of antiepileptic drugs in humans from animals with special emphasis on clearance. J Pharm Sci. 85:411-414 (1996). 199. I. Mahmood and J.D. Balian. Interspecies scaling: predicting clearance of drugs in humans. Three different approaches. Xenobiotica. 26:887-895 (1996). 200. P.L. Bonate and D. Howard. Prospective allometric scaling: does the emperor have clothes? J Clin Pharmacol. 40:335-340 (2000). 201. I. Mahmood. Critique of prospective allometric scaling: does the emperor have clothes? J Clin Pharmacol. 40:341-344; discussion 345-346 (2000). 202. G.W. Caldwell, D.M. Ritchie, J.A. Masucci, W. Hageman, and Z. Yan. The new pre-preclinical paradigm: compound optimization in early and late phase drug discovery. Curr Top Med Chem. 1:353-366 (2001). 203. X. Cao, S.T. Gibbs, L. Fang, H.A. Miller, C.P. Landowski, H.C. Shin, H. Lennernas, Y. Zhong, G.L. Amidon, L.X. Yu, and D. Sun. Why is it challenging to predict intestinal drug absorption and oral bioavailability in human using rat model. Pharm Res. 23:1675-1686 (2006). 204. W.L. Chiou and A. Barve. Linear Correlation of the Fraction of Oral Dose Absorbed of 64 Drugs Between Humans and Rats. Pharm Res. 15:1792-1795 (1998). 205. Y.H. Zhao, M.H. Abraham, J. Le, A. Hersey, C.N. Luscombe, G. Beck, B. Sherborne, and I. Cooper. Evaluation of rat intestinal absorption data and correlation with human intestinal absorption. Eur J Med Chem. 38:233-243 (2003). 206. U. Fagerholm, M. Johansson, and H. Lennernas. Comparison between permeability coefficients in rat and human jejunum. Pharm Res. 13:1336- 1342 (1996). 207. W.L. Chiou and P.W. Buehler. Comparison of oral absorption and bioavailability of drugs between monkey and human. Pharm Res. 19:868-874 (2002). 208. M. Takahashi, T. Washio, N. Suzuki, K. Igeta, and S. Yamashita. Investigation of the Intestinal Permeability and First-Pass Metabolism of Drugs in Cynomolgus Monkeys Using Single-Pass Intestinal Perfusion. Biol Pharm Bull. 33:111-116 (2010). 209. W.L. Chiou, H.Y. Jeong, S.M. Chung, and T.C. Wu. Evaluation of Using Dog as an Animal Model to Study the Fraction of Oral Dose Absorbed of 43 Drugs in Humans. Pharm Res. 17:135-140 (2000). 210. W.K. Sietsema. The absolute oral bioavailability of selected drugs. Int J Clin Pharmacol Ther Toxicol. 27:179-211 (1989). 211. I. Mahmood. Can absolute oral bioavailability in humans be predicted from animals? A comparison of allometry and different indirect methods. Drug Metabol Drug Interact. 16:143-155 (2000). 212. T. Akabane, K. Tabata, K. Kadono, S. Sakuda, S. Terashita, and T. Teramura. A comparison of pharmacokinetics between humans and monkeys. Drug Metab Dispos. 38:308-316 (2010). 213. M. Takahashi, T. Washio, N. Suzuki, K. Igeta, Y. Fujii, M. Hayashi, Y. Shirasaka, and S. Yamashita. Characterization of gastrointestinal drug absorption in cynomolgus monkeys. Mol Pharm. 5:340-348 (2008). 214. H. Nishimuta, T. Nakagawa, N. Nomura, and M. Yabuki. Species differences in hepatic and intestinal metabolic activities for 43 human cytochrome P450 77

substrates between humans and rats or dogs, Xenobiotica. 10.3109/00498254.2013.787155:2013. 215. H. Nishimuta, K. Sato, Y. Mizuki, M. Yabuki, and S. Komuro. Species differences in intestinal metabolic activities of cytochrome P450 isoforms between cynomolgus monkeys and humans. Drug Metab Pharmacokinet. 26:300-306 (2011). 216. H. Nishimuta, K. Sato, Y. Mizuki, M. Yabuki, and S. Komuro. Prediction of the intestinal first-pass metabolism of CYP3A substrates in humans using cynomolgus monkeys. Drug Metab Dispos. 38:1967-1975 (2010). 217. K. Sugano. Physiology of Gastrointestinal Tract and Other Administration Sites in Humans and Animals. Biopharmaceutics Modeling and Simulations. 10.1002/9781118354339.ch6, John Wiley & Sons, Inc.2012, pp. 160-205. 218. M. Takahashi, T. Washio, N. Suzuki, K. Igeta, and S. Yamashita. The species differences of intestinal drug absorption and first-pass metabolism between cynomolgus monkeys and humans. J Pharm Sci. 98:4343-4353 (2009). 219. M. Martignoni, G.M. Groothuis, and R. de Kanter. Species differences between mouse, rat, dog, monkey and human CYP-mediated drug metabolism, inhibition and induction. Expert Opin Drug Metab Toxicol. 2:875-894 (2006). 220. G.M. Grass and P.J. Sinko. Physiologically-based pharmacokinetic simulation modelling. Adv Drug Deliv Rev. 54:433-451 (2002). 221. K.W. Ward, D.J. Coon, D. Magiera, S. Bhadresa, M. Struharik, and M.S. Lawrence. Exploration of the African green monkey as a preclinical pharmacokinetic model: oral pharmacokinetic parameters and drug-drug interactions. Xenobiotica. 39:266-272 (2009). 222. I. Nestorov. Whole body pharmacokinetic models. Clin Pharmacokinet. 42:883-908 (2003). 223. M. Rowland, C. Peck, and G. Tucker. Physiologically-based pharmacokinetics in drug development and regulatory science. Annu Rev Pharmacol Toxicol. 51:45-73 (2011). 224. T. Teorell. Kinetics of distribution of substances administered to the body I The extravascular modes of administration. Arch Int Pharmacodyn Ther. 57:205-225 (1937). 225. H.S.G. Chen and J.F. Gross. Physiologically based pharmacokinetic models for anticancer drugs. Cancer Chemother Pharmacol. 2:85-94 (1979). 226. P. Poulin, K. Schoenlein, and F.P. Theil. Prediction of adipose tissue: plasma partition coefficients for structurally unrelated drugs. J Pharm Sci. 90:436- 447 (2001). 227. P. Poulin and F.-P. Theil. A priori prediction of tissue:plasma partition coefficients of drugs to facilitate the use of physiologically-based pharmacokinetic models in drug discovery. J Pharm Sci. 89:16-35 (2000). 228. T. Rodgers and M. Rowland. Physiologically based pharmacokinetic modelling 2: Predicting the tissue distribution of acids, very weak bases, neutrals and zwitterions. J Pharm Sci. 95:1238-1257 (2006). 229. T. Rodgers, D. Leahy, and M. Rowland. Physiologically based pharmacokinetic modeling 1: Predicting the tissue distribution of moderate- to-strong bases. J Pharm Sci. 94:1259-1276 (2005). 230. L.M. Berezhkovskiy. Volume of distribution at steady state for a linear pharmacokinetic system with peripheral elimination. J Pharm Sci. 93:1628- 1640 (2004). 78

231. J.B. Houston and D.J. Carlile. Prediction of hepatic clearance from microsomes, hepatocytes, and liver slices. Drug Metab Rev. 29:891-922 (1997). 232. A. Rostami-Hodjegan. Physiologically based pharmacokinetics joined with in vitro-in vivo extrapolation of ADME: a marriage under the arch of systems pharmacology. Clin Pharmacol Ther. 92:50-61 (2012). 233. H.M. Jones, M. Dickins, K. Youdim, J.R. Gosset, N.J. Attkins, T.L. Hay, I.K. Gurrell, Y.R. Logan, P.J. Bungay, B.C. Jones, and I.B. Gardner. Application of PBPK modelling in drug discovery and development at Pfizer. Xenobiotica. 42:94-106 (2012). 234. N. Parrott and T. Lave. Applications of physiologically based absorption models in drug discovery and development. Mol Pharm. 5:760-775 (2008). 235. H. Jones, N. Parrott, K. Jorga, and T. Lavé. A Novel Strategy for Physiologically Based Predictions of Human Pharmacokinetics. Clin Pharmacokinet. 45:511-542 (2006). 236. N. Parrott, N. Paquereau, P. Coassolo, and T. Lave. An evaluation of the utility of physiologically based models of pharmacokinetics in early drug discovery. J Pharm Sci. 94:2327-2343 (2005). 237. N. Parrott, H. Jones, N. Paquereau, and T. Lavé. Application of Full Physiological Models for Pharmaceutical Drug Candidate Selection and Extrapolation of Pharmacokinetics to Man. Basic Clin Pharmacol Toxicol. 96:193-199 (2005). 238. S.S. De Buck, V.K. Sinha, L.A. Fenu, R.A. Gilissen, C.E. Mackie, and M.J. Nijsen. The Prediction of Drug Metabolism, Tissue Distribution, and Bioavailability of 50 Structurally Diverse Compounds in Rat Using Mechanism-Based Absorption, Distribution, and Metabolism Prediction Tools. Drug Metab Dispos. 35:649-659 (2007). 239. P. Zhao, M. Rowland, and S.M. Huang. Best Practice in the Use of Physiologically Based Pharmacokinetic Modeling and Simulation to Address Clinical Pharmacology Regulatory Questions. Clin Pharmacol Ther. 92:17-20 (2012). 240. P. Zhao, L. Zhang, J.A. Grillo, Q. Liu, J.M. Bullock, Y.J. Moon, P. Song, S.S. Brar, R. Madabushi, T.C. Wu, B.P. Booth, N.A. Rahman, K.S. Reynolds, E. Gil Berglund, L.J. Lesko, and S.M. Huang. Applications of Physiologically Based Pharmacokinetic (PBPK) Modeling and Simulation During Regulatory Review. Clin Pharmacol Ther. 89:259-267 (2011). 241. D. Hallifax and J.B. Houston. Evaluation of hepatic clearance prediction using in vitro data: Emphasis on fraction unbound in plasma and drug ionisation using a database of 107 drugs. J Pharm Sci. 101:2645-2652 (2012). 242. A. Galetin, M. Gertz, and J.B. Houston. Potential role of intestinal first-pass metabolism in the prediction of drug-drug interactions. Expert Opin Drug Metab Toxicol. 4:909-922 (2008). 243. A. Rostami-Hodjegan and G.T. Tucker. Simulation and prediction of in vivo drug metabolism in human populations from in vitro data. Nat Rev Drug Discov. 6:140-148 (2007). 244. European Medicines Agency. Guideline on the investigation of drug interactions, Vol. CPMP/EWP/560/95/Rev. 1 Corr. 2**, http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guidelin e/2012/07/WC500129606.pdf, 2012.

245. U.S. Food and Drug Administration. Guidance for Industry: Drug Interaction Studies —Study Design, Data Analysis, Implicationsfor Dosing, and Labeling Recommendations, U.S. Food and Drug Administration,, http://www.fda.gov/downloads/drugs/guidancecomplianceregulatoryinformat ion/guidances/ucm292362.pdf, 2012. 246. S.-M. Huang, D.R. Abernethy, Y. Wang, P. Zhao, and I. Zineh. The utility of modeling and simulation in drug development and regulatory review. J Pharm Sci. 102:2912-2923 (2013). 247. A.S. Darwich, D. Pade, B.J. Ammori, M. Jamei, D.M. Ashcroft, and A. Rostami-Hodjegan. A mechanistic pharmacokinetic model to assess modified oral drug bioavailability post bariatric surgery in morbidly obese patients: interplay between CYP3A gut wall metabolism, permeability and dissolution. J Pharm Pharmacol. 64:1008-1024 (2012). 248. A.N. Edginton and S. Willmann. Physiology-Based Simulations of a Pathological Condition. Clin Pharmacokinet. 47:743-752 (2012). 249. T.N. Johnson, A. Rostami-Hodjegan, and G.T. Tucker. Prediction of the Clearance of Eleven Drugs and Associated Variability in Neonates, Infants and Children. Clin Pharmacokinet. 45:931-956 (2012). 250. T.N. Johnson and A. Rostami-Hodjegan. Resurgence in the use of physiologically based pharmacokinetic models in pediatric clinical pharmacology: parallel shift in incorporating the knowledge of biological elements and increased applicability to drug development and clinical practice. Pediatric Anesthesia. 21:291-301 (2011). 251. B. Agoram, W.S. Woltosz, and M.B. Bolger. Predicting the impact of physiological and biochemical processes on oral drug bioavailability. Adv Drug Deliv Rev. 50 Suppl 1:S41-67 (2001). 252. S. Willmann, W. Schmitt, J. Keldenich, J. Lippert, and J.B. Dressman. A physiological model for the estimation of the fraction dose absorbed in humans. J Med Chem. 47:4022-4031 (2004). 253. K. Thelen, K. Coboeken, S. Willmann, J.B. Dressman, and J. Lippert. Evolution of a detailed physiological model to simulate the gastrointestinal transit and absorption process in humans, part II: extension to describe performance of solid dosage forms. J Pharm Sci. 101:1267-1280 (2012). 254. K. Thelen, K. Coboeken, S. Willmann, R. Burghaus, J.B. Dressman, and J. Lippert. Evolution of a detailed physiological model to simulate the gastrointestinal transit and absorption process in humans, part 1: oral solutions. J Pharm Sci. 100:5324-5345 (2011). 255. M. Jamei, S. Marciniak, K. Feng, A. Barnett, G. Tucker, and A. Rostami- Hodjegan. The Simcyp population-based ADME simulator. Expert Opin Drug Metab Toxicol. 5:211-223 (2009). 256. D. Cong, M. Doherty, and K.S. Pang. A New Physiologically Based, Segregated-Flow Model to Explain Route-Dependent Intestinal Metabolism. Drug Metab Dispos. 28:224-235 (2000). 257. E. Sjogren, J. Westergren, I. Grant, G. Hanisch, L. Lindfors, H. Lennernas, B. Abrahamsson, and C. Tannergren. In silico predictions of gastrointestinal drug absorption in pharmaceutical product development: application of the mechanistic absorption model GI-Sim. Eur J Pharm Sci. 49:679-698 (2013). 258. P. Paixao, L.F. Gouveia, and J.A. Morais. Prediction of the human oral bioavailability by using in vitro and in silico drug related parameters in a physiologically based absorption model. Int J Pharm. 429:84-98 (2012). 80

259. M. Gertz, J.B. Houston, and A. Galetin. Physiologically based pharmacokinetic modeling of intestinal first-pass metabolism of CYP3A substrates with high intestinal extraction. Drug Metab Dispos. 39:1633-1642 (2011). 260. M. Tubic-Grozdanis, J.M. Hilfinger, G.L. Amidon, J.S. Kim, P. Kijek, P. Staubach, and P. Langguth. Pharmacokinetics of the CYP 3A substrate simvastatin following administration of delayed versus immediate release oral dosage forms. Pharm Res. 25:1591-1600 (2008). 261. A. Olivares-Morales, Y. Kamiyama, A.S. Darwich, L. Aarons, and A. Rostami-Hodjegan. Analysis of the impact of controlled release formulations on oral drug absorption, gut wall metabolism and relative bioavailability of CYP3A substrates using a physiologically-based pharmacokinetic model. Eur J Pharm Sci. 67:32-44 (2015). 262. M. Yaich, M. Popon, Y. Medard, and E.J. Aigrain. In-vitro cytochrome P450 dependent metabolism of oxybutynin to N-deethyloxybutynin in humans. Pharmacogenetics. 8:449-451 (1998). 263. J. Douchamps, F. Derenne, A. Stockis, D. Gangji, M. Juvent, and A. Herchuelz. The pharmacokinetics of oxybutynin in man. Eur J Clin Pharmacol. 35:515-520 (1988).

Chapter 2: Animal versus human oral bioavailability: do they correlate?

Eur. J. Pharm. Sci. 57 (2014), 280-291

Helen Musther, Andrés Olivares Morales, Oliver J.D. Hatley, Bo Liu and Amin Rostami Hodjegan

European Journal of Pharmaceutical Sciences 57 (2014) 280–291

Contents lists available at ScienceDirect

European Journal of Pharmaceutical Sciences

journal homepage: www.elsevier.com/locate/ejps

Animal versus human oral drug bioavailability: Do they correlate? ⇑ Helen Musther a, Andrés Olivares-Morales b, Oliver J.D. Hatley b, Bo Liu a, Amin Rostami Hodjegan a,b, a Simcyp Limited (a Certara Company), Blades Enterprise Centre, John Street, Shefﬁeld S2 4SU, UK b Centre for Applied Pharmacokinetic Research, School of Pharmacy and Pharmaceutical Sciences, University of Manchester, Manchester, UK article info abstract

Article history: Oral bioavailability is a key consideration in development of drug products, and the use of preclinical spe- Received 15 April 2013 cies in predicting bioavailability in human has long been debated. In order to clarify whether any corre- Received in revised form 9 August 2013 lation between human and animal bioavailability exist, an extensive analysis of the published literature Accepted 13 August 2013 data was conducted. Due to the complex nature of bioavailability calculations inclusion criteria were Available online 26 August 2013 applied to ensure integrity of the data. A database of 184 compounds was assembled. Linear regression for the reported compounds indicated no strong or predictive correlations to human data for all species, Keywords: individually and combined. First in man pharmacokinetics The lack of correlation in this extended dataset highlights that animal bioavailability is not quantita- Oral drug absorption Drug development tively predictive of bioavailability in human. Although qualitative (high/low bioavailability) indications might be possible, models taking into account species-speciﬁc factors that may affect bioavailability are recommended for developing quantitative prediction. Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction estimation of the oral drug bioavailability in humans and their validity. The understanding of the absorption of oral dosage forms is a Understanding oral bioavailability is not just a drug develop- key consideration in drug development. Oral routes are preferred ment issue but it has regulatory implications as defined by the for being less invasive and more physiological and due to ease of many agencies such as FDA in their guidance for industry administration and patient compliance. However, compared to (FDA, 2003). These usually distinguish between the rate and the direct entry of the drug to systemic circulation that is achieved extent which the active ingredient or active moiety is absorbed through intravenous dosing, additional elements affecting the from a drug product and becomes available at the site of action. availability of the drug following oral administration must be con- Since measurement at the site of action is not practical, bioavail- sidered. These may include potential for degradation in stomach or ability calculation for extravascular administration acts as a gut lumen, metabolism in the gut wall and liver, permeability surrogate to determine the amount of drug reaching site of through the gut wall and incomplete release of the drug from the action relative to those from intravascular administration formulation. The molecular structure of the drug and constituents (Sietsema, 1989). of the dosage form can determine many of these processes and Subtle differences in the methods of calculating bioavailability they define how much of a drug reaches the systemic circulation. exist which may give rise to variable results for a given drug or With all of these factors in mind, the OrBiTo project is aiming to drug formulation. Without an understanding of these assump- deliver rational methods and a framework for predicting how oral- tions, comparison of various bioavailability measures would not ly-administered drugs will perform (OrBiTo, 2012). In doing so, it is be prudent. In the current drug development paradigm, adminis- important to recognise some of the current practices related to tration of drugs in various preclinical species prior to human clinical studies is common for variety of reasons. It is often assumed that data on drug absorption from animals could provide reasonable estimates of bioavailability in humans. However, ⇑ Corresponding author. Address: School of Pharmacy and Pharmaceutical whilst similarity of permeability and fraction absorbed to gut Sciences, Faculty of Medical and Human Sciences, University of Manchester, wall between animals and human is established (Chiou and Bar- Stopford Building, Oxford Road, Manchester M13 9PT, UK. Tel.: +44 161 3060634. ve, 1998; Chiou et al., 2000; Chiou and Buehler, 2002; Cao et al., E-mail addresses: [email protected] (H. Musther), andres.olivaresmorales 2006) there are considerable interspecies differences in first-pass @manchester.ac.uk (A. Olivares-Morales), [email protected] (O.J.D. Hatley), [email protected] (B. Liu), [email protected] (A. gut and liver metabolism. These differences can prevent conclud- Rostami Hodjegan). ing a level of overall bioavailability in humans based on the

0928-0987/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.ejps.2013.08.018 H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291 281 animal data. Comparisons and correlations of human and animal the commonly known comprehensive reports carried out by bioavailability have previously been reported in the literature, Grass and Sinko (2002), utilising the dataset published by Siet- and although these seem to indicate that animal values are not sema (1989). There has been no attempt to expand the data predictive of human bioavailability, they have mainly been within the 2002 report with any additional data published since limited to small sets of measurements, or comparisons within then or refine some ambiguities in the original report. Anecdotal one species. In some instances queries have been raised regard- evidence indicated that the number of data points in the pub- ing the treatment, analysis and sources of the data forming lished comparisons (within a scatter graph) were not consistent the reports. Furthermore, it is not clear whether formulations with the number of compounds that appeared in the original have been matched when comparing human with animal dataset. The reasons for this were not immediately clear from bioavailability; e.g., oral doses may have been given via solution the description given in the report. To assess the number of data or suspension to animals while human studies make use of solid points and their consistency with original source, the scatter plot formulations which might result in formulation-linked bioavail- of human vs. animal bioavailability in Grass and Sinko (2002) ability differences rather than solely a species difference. There was digitised using GetData Graph Digitizer v2.22 (Get Data may also be differences in study design such as use of the same Graph Digitizer, 2012), and the extracted data compared to that or different study groups for oral and iv administration which published in the original study by Sietsema (1989). In addition, may cloud the comparisons. These issues should not be over- the relationships between human and animal bioavailability, re- looked when making comparisons. ported in this original database (Sietsema, 1989) were reviewed. We report an extensive analysis of the published data References sources were obtained where available and checked conducted as part of OrBiTo project to clarify the relationships against criteria developed for ‘‘inclusion’’ which ensured the val- between human and animal bioavailability, paying specific atten- ues and the species were relevant to current study. Some studies tion to those issues described above. It is expected that this report were marked as ‘Rodent’ which were considered too broad in contributes to providing an answer to the question that whether a light of currently utilised preclinical species. Hence, all data correlation exists between the bioavailability in animals and relating to species other than mouse, rat, dog and non-human humans and whether such animal data can be used for predicting primates were discarded. human bioavailability; quantitatively or qualitatively. Additional compounds were identified using the human bioavailability database published by Varma et al. (2010). Some information on were obtained from other human vs. animal literature 2. Methods reports (Chiou and Buehler, 2002; Cao et al., 2006; Akabane et al., 2010). Where original data and references were not provided 2.1. Calculation of bioavailability in the publication, the authors were contacted and invited to clarify the sources of information. The overall bioavailability is often considered as a composite Finally, systematic literature searches being carried out using function of fraction released and absorbed into gut wall (F ), frac- abs PubMed and Google Scholar for the bioavailability values in human tion escaping first-pass gut wall metabolism (F ) and fraction G and their corresponding animal data. Original references were escaping first-pass hepatic metabolism (F ): H obtained and inspected in all cases.

F ¼ Fabs FG FH ð1Þ Calculation of oral bioavailability (F), and the definition of the 2.3. Inclusion and exclusion criteria fraction absorbed Fabs (which is one of its three components), is not unified. Pang and Rostami have recently commented on these Inclusion criteria (Table 1) were applied to ensure integrity (2011). Whilst one may consider the total oral drug bioavailability of the data and consistency between various researchers con- based on deducting the fraction ‘‘unabsorbed’’ (1 F) via analysis ducting the reviews. Mean bioavailability values were extracted of feces, in many cases the dose normalised relative area under directly from the publications. If iv and oral data had not been the curve (AUC) after oral and iv administration is used as a mea- obtained from the same individuals and they were from differ- sure of oral bioavailability. ent studies, bioavailability measures were considered unreliable There are implications in certain situations for using each of the due to potential effects of inter-subject variability. Where more above methods however in general they should produce the same than one dose was reported, the bioavailability for the lowest results. Disparities might occur when there are significant ele- dose was selected in order to minimise the potential impact ments of entero-hepatic recirculation or high first-pass metabo- of saturation effects. Information on formulations were lism in lung. When the F is defined as the fraction of given abs recorded. The details of strain and sex of animals utilised were dose that passes through the gut wall, the integration of all the noted for each reference, along with parameters relating to the mass transfer (alongside the GI tract) over the time period that compound type and use. Additional information were noted if absorption is happening may include the drug that originates from considered beneficial to the aims and objectives of the current entero-hepatic circulation. This leads to an apparent F can be- abs investigation (e.g. number of subjects where more than one come higher than 1 when traditional comparison of AUC after iv reference was found) and recorded in a ‘comments section’ of and oral administration is used to assess bioavailability (hence F database. Studies relating to controlled release formulations could be greater than 1). were discarded. Considering the differences between definitions used to determine F, it was essential to pay attention to methodologies used for calculating oral bioavailability before making comparisons between various species. Table 1 Inclusion criteria for studies.

2.2. Sources for human and animal bioavailability values 1. Oral and intravenous data should be established in the same group 2. Species should fall under category of Mouse, Rat, Dog or Non-Human Primate A number of reports have previously compared human and 3. AUC should be calculated to inﬁnity or absorption phase should be complete 4. Original study data (no review articles) must be included when possible animal bioavailability values for series of compounds. One of 282 H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291

2.4. Statistical analysis

For compounds with more than one bioavailability study, the weighted mean for the oral bioavailability was estimated by P n w x Pi¼1 i i x ¼ n ð2Þ i¼1wi

where xi was the mean oral bioavailability for the ith study, n in the number of studies, and the weights (wi) were the number of subjects in the ith study, respectively. For studies with an unknown number of subjects, the assigned weights corresponded to the median number of subjects employed in the rest of the studies, with values of 6 and 5 for the human and animal studies, respectively. Linear regression was performed for oral bioavailability in animal species (FAnimal,species) and human (FHuman) and the coefﬁ- cient of determination (R2) and the linear regression equation were recorded for each species and the whole dataset. A similar analysis was performed by grouping the compounds by ion class

(FAnimal,ionclass and FHuman,ionclass), and formulations (FAnimal,formulation and FHuman,formulation). In addition, accuracy of the FHuman prediction from FAnimal,species data was assessed by the ratio between animal and human oral bioavailability (RA/H) and average fold error (afe), Eqs. (3) and (4), respectively, whereas, for evaluation of Fig. 1. Plot of oral bioavailability (F) in animal species vs. oral bioavailability in humans (in percentage). Diamonds are for mouse, circles for rat, and triangles for the precision of the prediction, the concordance correlation dog and squares for non-human primates (NHP). coefficient (ccc) was calculated (Graham et al., 2012). All the statistical calculations where performed with the Statistical Toolbox within Matlab R2012a (The MathWorks Inc., Natick, further detail on the analysis of these data were not available for MA, USA) and Microsoft Excel 2010 (Microsoft Corporation, inspection. Redmond, WA, USA). 3.2. Literature search and description of the dataset FAnimal;species;i RA;i ¼ ð3Þ H FHuman;i The literature search resulted in a total of over 1000 studies, published between 1969 and 2012, representing around 450 ð1R logðR ÞÞ n A;i different compounds. The compounds with no FHuman and/or the Afe ¼ 10 H ð4Þ corresponding FAnimal,species data were removed from the dataset. From the original dataset, only 184 different compounds with both human and animal oral bioavailability were identified and 54 of 3. Results those compounds had more than one study for animal and/or human oral bioavailability. For the latter compounds, the weighted 3.1. Data extracted from previous reports of human vs. animal mean was calculated as described above. Finally, the number of bioavailability paired datasets with both animal and human bioavailability by species was 30, 122, 125 and 41 for the mouse, rat, dog and non- Digitisation of the scatter plot of human vs. animal bioavailabil- human primates (NHP), respectively. In addition,FAnimal,species was ity in the report by Grass and Sinko (2002) and comparisons to the plotted against FHuman (Fig. 1) and the final dataset is shown in Ta- information within the tables provided by Sietsema (1989) con- ble 3. Within species, the most frequent strains employed for the firmed that there were more data points than compounds in the oral bioavailability studies in animals were Sprague-Dawley and graph for all species (Table 2). It is worth noting that although Siet- Wistar for the rat (49% and 35%, respectively), Beagle and Mongrel sema collated bioavailability data from the literature for over 400 for the dog (66% and 19% respectively) and Rhesus and Cynomol- drugs, the human versus animal correlations were limited to gus monkey for the NHP (42% and 40%, respectively), whereas for approximately 70 compounds in total where the experimental bio- the mouse, no clear tendency was shown for the use of any availability values were available for human and at least one of the particular strain. In relation to ionic class, the majority of the other species. compounds in the dataset were basic followed by neutral, acidic Visual inspection of the tables, extracted data and the plot sug- and zwitterionic, representing 50.0%, 24.5%, 15.8%, and 9.8% of gests that multiple points may have been plotted for each com- the total compounds, respectively. The predominant formulation pound where a large range for bioavailability was reported, was solid (tablet and capsule) for human studies and liquid (solution and suspensions) in the case of animal studies; however for a large number of studies the formulations employed were not Table 2 informed. Points extracted from Grass and Sinko plot.

Species Number of points expected Number of points extracted 3.3. Correlation between animal and human bioavailability Rodent 40 61 Dog 43 76 As shown in Table 4, linear regression analysis revealed a poor Primates 11 15 correlation for the overall animal and human oral bioavailability Table 3 Interspecies oral bioavailability (F) for the selected compounds.

Compound Name Ionic class Mouse Rat Dog NHP Human Mean F (%) Range F n Mean F (%) Mean F (%) Range F Mean F (%) Range F n Mean F (%) Range F n Mean F (%) Range F n 5-Fluorouracil Neutral 28.0 28.6 26–33 3 Acarbose Base 4.0 1.4 Acebutolol Base 61.0 34.0 Acetylsalicylate (Aspirin) Acid 35.0 45.0 48.5 Acyclovir Base 2.18 90.7 54.2 Adefovir Base 8.0 12.0 Alprazolam Acid 7.5 92.0 Amitriptyline Base 96.5 46.0 Amlodipine Base 100.0 100.0 88.0 64.0 Amosulalol Base 22.1 58.5 57.4 100.0

Amoxicillin Zwitterion 76.8 77.4 280–291 (2014) 57 Sciences Pharmaceutical of Journal European / al. et Musther H. Antipyrine Neutral 73.0 100.0 Azathioprine Base 51.0 55.0 Azithromycin Base 46.0 97.0 37.0 Cefixime Acid 52.2 52.3 Cefuroxime Axetil Acid 23.1 33.5 Chlorpheniramine Base 9.4 47.0 Clonazepam Base 33.0 90.0 Cyclosporine Neutral 40.0 20.8 Dapsone Base 100.0 93.0 Diazepam Base 86.0 95.0 Doxazosin Base 50.0 60.0 63.0 Erythomycin Base 14.5 14–15 2 52.0 1.9 32.6 32–35 2 Estradiol Valerate Neutral 4.3 7.8 0.3–9 2 8.7 Estramustine phosphate Neutral 54.3 43.7 Ethambutol Base 76.0 87.4 77.3–97.4 2 Ethimizol Neutral 32.4 12.8 Ethinylestradiol Neutral 3.0 8.1 7.5–9 2 0.6 41.9 36–59 4 Ethosuximide Neutral 74.0 83.0 91.0 93.0 100.0 Ethylmorphine Base 5.7 26.0 28.0 Etoposide Neutral 10.4 8.9–13.9 3 13.4 12.9 54.0 52–57.3 2 Felodipine Neutral 20.0 17.0 16.0 Fenfluramine Base 15.0 89.3 Fenoterol Base 0.6 0.4–0.8 2 1.5 Fexofenadine Zwitterion 2.4 3.8 2.6–4.6 3 6.6 41.0 Finasteride Neutral 92.0 80.0 Fleroxacin Zwitterion 78.8 88.5 100.0 99.9 99–100 3 Fluconazole Neutral 75.0 80.0 100.0 90.0 Flumazenil Neutral 28.0 22.3 15–27.8 2 Flunisolide Neutral 55.7 99.2 55.0 4.9 20.0 Flunitrazepam Neutral 69.0 85.0 Fluvastatin Acid 35.0 86.4 41.0 16.0 29.0 Foscarnet Acid 10.0 13.1 9.1–17.1 2 Fosfomycin Acid 29.0 32.2 28–37 2 Furosemide Acid 28.9 46.5 58.3 43.4–71 8 Gabapentin Zwitterion 79.0 79.0 80.0 40.1 60.0 Ganciclovir Neutral 10.0 100.0 8.8 Gatifloxacin Zwitterion 60.5 98.5 96.0 Gitoxin Neutral 91.1 95.0 Glaziovine Base 98.5 84.3 Glyburide (Glibenclamide) Acid 14.7 85.3 82–89 3 Griseofulvin Neutral 3.9 51.2 283 Guanfacine Base 35.0 81.1 (continued on next page) 284 Table 3 (continued)

Compound Name Ionic class Mouse Rat Dog NHP Human Mean F (%) Range F n Mean F (%) Mean F (%) Range F Mean F (%) Range F n Mean F (%) Range F n Mean F (%) Range F n Hydralazine Neutral 37.7 33.6 31.3–35.4 2 Hydrochlorothiazide Neutral 30.7 63.1 60.2–67.5 2 Ibuprofen Acid 90.0 100.0 100.0 Idazoxan Base 12.6 66.0 34.0 Ifosphamide/ Ifosfamide Neutral 40.0 88.3 Indapamide Neutral 98.0 99.0 Indomethacin Acid 63.7 91.9 89–100 2 Isosorbide dinitrate Neutral 40.0 27.6 19–48 6 Isosorbide-2-mononitrate Neutral 100.0 100.0 Isosorbide-5-mononitrate Neutral 71.5 98.4 93–100 5 Isoxicam Neutral 99.0 100.0 .Mshre l uoenJunlo hraetclSine 7(04 280–291 (2014) 57 Sciences Pharmaceutical of Journal European / al. et Musther H. Itraconazole Neutral 27.1 16.6–34.9 2 55.0 Ketanserin Base 100.0 100.0 47.4 45–51 3 Ketorolac Acid 90.4 99.1 91–100 2 100.0 100.0 95.9 80.5–100 2 Lansoprazole Neutral 30.5 26.5 81.0 Levodopa Zwitterion 36.3 33.3 Levofloxacin Zwitterion 87.6 99.0 Levonorgestrel Neutral 67.0 9.0 22.0 9.0 89.0 Lidocaine (Lignocaine) Base 7.1 31.0 25.0 25–25 2 27.8 12.6–37 4 Linezolid Neutral 95.0 95.0 100.0 Lisuride Base 28.0 5.0 14.0 Lithium carbonate Neutral 97.9 94.5 Losartan Acid 55.7 35.8 Melagatran Zwitterion 13.0 72.0 7.7 Meloxicam Neutral 94.0 95.0 100.0 91.0 89–97 2 Menogaril Acid 33.2 11.8 33.0 33.6 Mepindolol Base 1.5 40.0 82.1 Mercaptopurine Neutral 8.0 4–12 2 12.0 Metformin Base 34.1 60.6 Methadone Base 0.0 79.0 79–79 2 Methylprednisolone Neutral 35.1 49.4 Metoclopramide Base 71.9 49–91 2 75.3 58.1–84 4 Metolazone Neutral 80.0 49.9 Metoprolol Base 25.0 44.5 41–50 5 Midazolam Neutral 2.3 14.9 2.3–45 8 15.1 1.9 1.6–2.1 3 33.8 24–48 8 Morphine Base 12.1 6.5–14.9 3 19.6 17.9–23 2 36.5 26–47.1 5 Moxifloxacin Zwitterion 78.0 78.0 91.0 48.5 45–52 2 82.0 Moxonidine Base 5.1 87.6 Nalbuphine Base 2.5 0.9–4.7 3 4.8 3.5–5.6 2 16.4 Naloxone Base 0.3 0.9 Naltrexone Base 0.8 91.5 100.0 Naproxen Acid 100.0 100.0 Naratriptan Base 71.0 95.0 63.0 Nefazodone Base 14.0 15.0 Nevirapine Neutral 91.0 93.0 Nicardipine Base 21.5 8.9 10.4 10.4 10.2–10.5 2 Nifedipine Neutral 9.3 50.0 43–63 7 Nimodipine Neutral 22.3 5.4 Nisoldipine Neutral 2.7 11.7 3.9 Nitrendipine Neutral 12.0 29.0 21.2 Nizatidine Base 72.4 94.4 96.7 94–100 2 Nomifensine Base 48.3 26.5 Norfenfluramine Base 19.0 85.3 Nufenoxole (SC–27166) Base 96.0 26.0 85.0 100.0 Ofloxacin Zwitterion 77.8 100.0 Omeprazole Neutral 19.3 6.4–40.8 2 15.0 55.6 40–70 3 Ondansetron Base 7.2 6–8.6 2 57.0 Oseltamivir acid Acid 30.0 35.0 73.0 79.0 Oxazepam Zwitterion 72.3 56–88.5 2 40.0 22.1 92.8 Phenobarbital Acid 96.7 91.0 93.0 11–15.5 3 Phenoxymethylpenic–illin (Penicillin V) Acid 29.0 11.6 11.0 5.5 48.1 Phenytoin Acid 36.0 78.5 69.9–90 3 Physostigmin Base 2.0 4.9 3.2–8.2 2 Pindolol Base 80.0 73.0 85.0 85.0 86.0 Piroxicam Zwitterion 100.0 89.0 100.0 Pravastatin Acid 12.3 32.3 19.1 Prazosin Base 30.5 23–38 2 68.0

Prednisolone Neutral 55.0 82.0 280–291 (2014) 57 Sciences Pharmaceutical of Journal European / al. et Musther H. Prednisone Neutral 38.0 77.3 69–80 3 Primaquine Base 25.0 96.0 Procainamide Base 85.0 80.9 75.3–83 2 Propoxyphene Base 25.3 18.0 Propranolol (±) Base 19.0 6.8 <1 22.0 ) Base 7.9 5.7–10.5 2 28.0 Propranolol ( Propranolol (+) Base 17.8 16.1–19.9 2 19.4 Propylthiouracil Acid 88.0 77.0 Pyridostigmine Base 33.6 7.6 14.3–7.6 2 Quinidine Base 57.0 73.3 70.0 Rabeprazole Base 24.3 51.8 Ranitidine Base 73.0 54.4 52–60 2 Reboxetine Base 21.0 5.0 90.0 46.0 97.0 Recainam Base 51.0 89.0 97.0 70.3 67–73 2 Remoxipride Base 8.0 <1 94.0 94.9 93–96 2 Rifabutin Base 44.0 20.0 Rifampin Zwitterion 89.0 94.8 Risedronate Acid 2.9 0.6 Risperidone Base 22.0 65.9 Rosiglitazone Zwitterion 100.0 99.0 Rosuvastatin Acid 19.0 20.1 Salbutamol Base 85.0 51.2 50–53 2 Salicylate Acid 92.0 100.0 95.2 Saquinavir Base 42.2 6.5 0.7 Selegiline Base 8.1 4.3 Sildenafil Base 17.0 33.5 54.0 40.4 38–41 2 Sitafloxacin Zwitterion 30.9 51.2 92.7 89.0 Sitagliptin Base 76.0 100.0 68.0 87.0 Sotalol Base 84.1 100.0 Sparfloxacin Zwitterion 61.6 58.3–63.3 2 84.4 77–91.9 2 79.7 92.0 Sulfisoxazole Zwitterion 77.0 100.0 Sulpiride Base 13.4 82.5 35.5 Sumatriptan Base 37.0 56.2 54–58 2 14.0 Tacrolimus Acid 22.0 26.2 6.5 15.5 15–17.8 2 Talinolol Base 29.0 17.1–52.1 3 65.0 55–68.9 2 Tamsulosin Base 14.4 29.7 100.0 Terazosin Base 67.8 82.0 Terodiline Base 23.0 90.8 90–92 2 Tetrabenazine Base 84.0 4.9 (continued on next page) 285 286 Table 3 (continued)

Compound Name Ionic class Mouse Rat Dog NHP Human Mean F (%) Range F n Mean F (%) Mean F (%) Range F Mean F (%) Range F n Mean F (%) Range F n Mean F (%) Range F n Theophylline Base 77.2 63.8–97.2 3 94.0 91–100 2 94.9 98.0 94–100 5 Tiagabine Acid 25.0 54.0 89.9 Timolol Base 29.0 61.0 Tinidazole Base 100.0 99.0 Tolterodine Base 14.2 11.0 60.5 43.0 Torsemide Acid 92.7 87.5–95.6 2 77.0 93.1 89–96 2 Tramadol Base 32.4 65.0 67.9 Trazodone Base 27.2 73.6 63–77 2 TRH Tartrate Base 1.5 12.6 2.0 Triazolam Neutral 19.0 16.0 53.0 Trovaﬂoxacin Zwitterion 68.0 58.0 85.0 89.1 87.6–91 2 .Mshre l uoenJunlo hraetclSine 7(04 280–291 (2014) 57 Sciences Pharmaceutical of Journal European / al. et Musther H. Valproic Acid Acid 94.4 93.0 94.4 92.8–96 2 Vardenaﬁl Base 10.0 27.3 14.5 Venlafaxine Base 12.6 59.8 6.5 44.0 Verapamil (±) Base 14.5 13.8–15.2 2 2.5 23.9 18–38 3 Verapamil () Base 7.4 1.7 1.5–1.9 2 20.0 Verapamil (+) Base 4.1 21.1 20.6–21.6 2 50.0 Warfarin Acid 85.8 93.0 Xamoterol Base 23.2 8–34 2 4.6 4.5–4.8 3 Zalcitabine Base 30.0 93.4 86–100 3 Zanamivir Base 3.0 2.0 Zolmitriptan Base 41.0 79.0 42.8 39–49 2 Zolpidem Base 27.0 66.2 65.8–66.6 2 Zopiclone Base 44.0 100.0 76.9 76.7–77 2

F, oral bioavailability; Range, F, range for the mean bioavailability values for the studies; n, number of studies for the calculation of the weighted mean. References for the bioavailability studies can be found in the supplementary material for the online version of this article. H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291 287

Table 4 Linear regression analysis, afe and animal/human oral bioavailability ratio.

2 Species Number of points Slope (b) Intercept (a) R ccc afe Median RA/H 5% lower percentile for RA/H 95% upper Percentile for RA/H All 318 0.553* 33.114 0.342 0.548 0.647 0.866 0.082 2.771 Mouse 30 0.507** 39.478 0.253 0.444 0.593 0.784 0.058 2.784 Rat 122 0.544* 35.759 0.287 0.470 0.583 0.723 0.075 2.777 Dog 125 0.580* 26.433 0.374 0.605 0.845 0.990 0.236 3.254 NHP 41 0.691* 32.942 0.694 0.698 0.417 0.592 0.051 1.042

2 Regression equation, FHuman = b FAnimal,species + a; R , coefﬁcient of determination; ccc, concordance correlation coefﬁcient; afe, average fold error; RA/H, animal and human oral bioavailability ratio. * p value vs. constant model < 0.001. ** p value vs. constant model < 0.005.

Fig. 2. Plots for the linear regression analysis by separated by species (in percentages), the coefﬁcient of determination (R2) for the linear regression are shown in each plot.

(a) Mouse F vs. human F; (b) Rat F vs. human F; (c) Dog F vs. human F and (d) Non-human primates (NHP) F vs. human F.

relationship (R2 = 0.342). Dog data, showed a minor improvement for every species in particular (Fig. 2). However, a prediction of 2 in the R value compared to the value from obtained from the FHuman from FAnimal,species, employing the linear regression model, whole dataset, whereas for the mouse and rat data, R2 values resulted in wide prediction intervals (PI), as shown in Figs. 5 were lower than for the dog. In contrast, R2 value obtained for and 6. Afe calculations showed values below the unity for the the NHP was higher than the value for the overall dataset and general dataset and for every species in particular. Calculated 288 H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291

Fig. 3. Box plots of median animal/human bioavailability ratios (RA/H) and interval between animal and human oral bioavailability. Triangles indicate 95% conﬁdence interval (CI) for the median values; Dashed line (---), indicate the upper limit for outliers representation.

Fig. 4. Plots for the linear regression analysis by separated by ion class (in percentages), the coefﬁcient of determination (R2) for the linear regression are shown in each plot.

(a) Mouse F vs. human F; (b) Rat F vs. human F; (c) Dog F vs. human F and (d) Non-human primates (NHP) F vs. human F. H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291 289

values of the concordance correlation coefﬁcient highlighted the lack of agreement between human and animal bioavailability for all species, suggesting a lack of precision in any quantitative prediction. In addition the median ratio between animal and hu-

man oral bioavailability (RA/H) showed similar results for the general dataset and almost all the species. The dog, however, showed

a median value close to the unity (median RA/H = 0.990), but with the highest interval of all the species (0.236 to 3.254 for the 5th and 95th percentile, respectively) (Fig. 3). A similar scenario occurs for the correlation analysis by ion class summarised in Table 2 and Fig. 4. Acidic drugs showed the highest R2 value (R2 = 0.549); followed by neutral and zwitterionic drugs, while the lowest R2 value was for basic compounds (R2 = 0.212) (see Tables 5 and 6). Grouping compounds by formulation type (Solution or Solid) shows no advantage over the weighted combination of data with Solution and Solid showing similar R2 values (Solution R2 = 0.339, Solid R2 = 0.328) to the overall relationship, all of which indicate a poor correlation.

Fig. 5. Plot of the linear regression analysis for the general dataset, animal vs. 4. Discussion human oral bioavailability. Diamonds are for mouse, circles for rat, and triangles for dog and squares for non-human primates (NHP). Solid line (–), linear regression line; Pointed line (), 95% conﬁdence interval (CI) for mean response; Dashed line The digitisation and careful re-analysis of the Grass and Sinko (---), 95% prediction interval (PI) for a future value. scatter plot raised a number of questions about the treatment of

Fig. 6. Plots for the linear regression analysis by classiﬁed by species (in percentages), the coefﬁcient of determination (R2) for the linear regression are shown in each plot. (a)

Mouse F vs. human F; (b) Rat F vs. human F; (c) Dog F vs. human F and (d) Non-human primates (NHP) F vs. human F. Solid line (–), linear regression line; Pointed line (), 95% conﬁdence interval (CI) for mean response; Dashed line (---), 95% prediction interval (PI) for a future value. 290 H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291

Table 5 the full details of equations employed, and they are transparent Linear regression analysis by ionic class. in describing methods of determination, formulations utilised Species Number of points Slope (b) Intercept (a) R2 and full details of subjects. However, some other reports provided Acid 53 0.686* 26.295 0.549 minimal or in occasions no information in some aspects related to Base 152 0.440* 35.891 0.212 the study and data analysis. Tracing used reference in other litera- Neutral 73 0.596* 30.897 0.409 ture back to the original study proved problematic in some cases, * Zwitterion 39 0.524 44.592 0.343 particularly where older papers were concerned. This led to exclu- 2 Regression equation, FHuman = b FAnimal,ionclass + a; R , coefficient of determination. sion of reports from current dataset when the original report was * p value vs. constant model < 0.001. not available and not analysed by authors of the current report. The regression analysis indicated that non-human primates are the most predictive amongst other species for human bioavailabil- 2 Table 6 ity (R of 0.7). Although the median relative bioavailability be- Linear regression analysis by formulation type. tween animal species and human was unbiased for dog (median value of the ratio being close to the unity), the wide range indi- Species Number of Slope Intercept R2 points (b) (a) cated imprecision of the values as a predictive measure with confidence. The regression plot for dog (Fig. 2) highlights the above Solution 57 0.157 32.128 0.339 Solid (Capsule/Tablet/ 30 0.524 34.153 0.328 conclusion showing the large degree of scatter and the poor corre- Solid) lation coefficient. For rat and mouse there was no indication of any

2 good correlation with the human data from any of the observed Regression equation, FHuman = b FAnimal,formulation + a; R , coefficient of determination. results. The results questioned the default assumption that bioavailability in rat or mouse can be a quantitative indication for human bioavailability. The dataset for non-human primates was far more limited than reported bioavailability data. Although it is clear that there are for rat or dog, with only 41 data points available. This is not more data points (pair of human-animal bioavailability data) than surprising considering the higher cost and more restricted ethical compounds due to multiple comparisons, not all points could be aspects associated with these studies. Further data could provide readily identified in the associated database of Sitesema based on more confidence in relatively high correlations observed. However, our re-analysis. No further detail on how the data were treated range of predicted bioavailability from non-human primates com- was given in the original review by Grass and Sinko and we could pared to observed human values (Fig. 6) was wide and indicated not resolve the disparities. It is plausible that multiple points are the qualitative rather than quantitative value. plotted to signify not just the mean values but also maximum When the formulations were matched, there was still an appar- and minimum reported values, where a large range of bioavailabil- ent lack of correlation. In addition, when exploring the full dataset, ity had been observed. The combination of these factors (lack of due to the use of solutions as the main route of administration in clear description of the methodology and the apparent mismatch animal studies (c.f. tablets in human), it might have been expected between the cited data and visualisation in the scatter plot) high- that a bias towards higher bioavailability values in animal should light anecdotal and occasional questions posed by those who be- be seen. However, this was not the case, suggesting that the formu- lieve animal data could be predictive of human bioavailability. lations did not have a significant impact on the correlations and bio- These issues indicate that any comparisons between species availability (note that all extended release formulations were should make an effort to clarify data extraction methodology and excluded from the database). This confirms that the notion that assumptions if the conclusions are to be used for defining drug other factors, such as metabolic differences between species, could development strategies with confidence. play a more important role in defining disparities human vs. animal Our methodology involved combining multiple studies by drug bioavailability. Accounting for such differences may improve calculating a weighted mean which was less ambiguous when understanding the differences and avoid over reliance on quantita- constructing scatter plots and correlations. However, it had the tive value of animal to human extrapolation of bioavailability. disadvantage that, where a large range of bioavailability values is reported, this is not captured in the correlations. Alternative strat- 5. Conclusion egies may involve separation of formulations, and a cursory analysis was undertaken utilising the new dataset and a reported An extended dataset to previously published reports was gener- formulation type (Solution or Solid). However, this is not an ideal ated for animal vs. human bioavailability data with clear inclusion scenario where the data in humans and animals are generated by criteria. This highlighted that there are not strong and predictive different research groups using different materials but it is imprac- linear correlations between overall and single species animal drug tical to attempt to apply this formulation matching criteria due to bioavailability and human values. Classification of high or low bio- the limitations it imposes on the dataset. Another consideration for availability could be achieved by setting certain cut-off points how- correlations could be related to weighting each of the data points ever quantitative models of oral drug bioavailability should be built based on the numbers subjects and animals used for each for each species based on understanding the physiologic, metabolic combination. and transporter related information affecting bioavailability. The current dataset is, to the best of our knowledge, the largest dataset published for investigating the correlation between animal and human oral bioavailability. However, in addition to the com- Acknowledgements plexity of the data analysis, the process of the new literature analysis illuminated further issues with performing a correlation of this A. O-M is recipient of a PhD grant awarded by CONICYT Chile, magnitude on bioavailability data. For instance, we had to discard Chilean Ministry of Education and a President’s Doctoral Scholar some of datapoints from previously published correlation studies Award from The University of Manchester. due to more stringent inclusion criteria. O.J.D.H. was funded by a PhD grant awarded through the CASE Clarity of the bioavailability studies in the literature varies award scheme, receiving support from both the MRC and widely. The information included in some publications provide AstraZeneca. H. Musther et al. / European Journal of Pharmaceutical Sciences 57 (2014) 280–291 291

The authors would like to thank Dr David Turner for his assis- Chiou, W.L., Jeong, H.Y., Chung, S.M., Wu, T.C., 2000. Evaluation of using dog as an tance in developing the concepts of this study; and James Kay animal model to study the fraction of oral dose absorbed of 43 drugs in humans. Pharm. Res. 17, 135–140. and Eleanor Savill for their help in preparing the manuscript, col- FDA, 2003. Guidance for Industry: Bioavailability and Bioequivalence Studies for lating and formatting the references. Orally Administered Drug Products – General Considerations. Center for Drug Evaluation and Research, Food and Drug Admininstration, US Department of Health and Human Services. (accessed 11.03.13). Get Data Graph Digitizer. 2012. Supplementary data associated with this article can be found, in (accessed 11.03.13). Graham, H., Walker, M., Jones, O., Yates, J., Galetin, A., Aarons, L., 2012. Comparison the online version, at http://dx.doi.org/10.1016/j.ejps.2013.08.018. of in-vivo and in-silico methods used for prediction of tissue: plasma partition coefﬁcients in rat. J. Pharm. Pharmacol. 64, 383–396. References Grass, G.M., Sinko, P.J., 2002. Physiologically-based pharmacokinetic simulation modelling. Adv. Drug Deliv. Rev. 54, 433–451. ORBITO, 2012. ORBITO | IMI – Innovative Medicines Initiative. (accessed 11.03.13). comparison of pharmacokinetics between humans and monkeys. Drug Metab. Pang, K.S., Rostami-Hodjegan, A., 2011. Fraction absorbed (Fabs): different connotations Dispos. 38, 308–316. and confusion for the literature? Biopharm. Drug Dispos. 32, 301–302. Cao, X., Gibbs, S.T., Fang, L., Miller, H.A., Landowski, C.P., Shin, H.C., Lennernas, H., Sietsema, W.K., 1989. The absolute oral bioavailability of selected drugs. Int. J. Clin. Zhong, Y., Amidon, G.L., Yu, L.X., Sun, D., 2006. Why is it challenging to predict Pharmacol. Ther. Toxicol. 27, 179–211. intestinal drug absorption and oral bioavailability in human using rat model. Varma, M.V., Obach, R.S., Rotter, C., Miller, H.R., Chang, G., Steyn, S.J., El-Kattan, A., Pharm. Res. 23, 1675–1686. Troutman, M.D., 2010. Physicochemical space for optimum oral bioavailability: Chiou, W.L., Barve, A., 1998. Linear correlation of the fraction of oral dose absorbed contribution of human intestinal absorption and ﬁrst-pass elimination. J. Med. of 64 drugs between humans and rats. Pharm. Res. 15, 1792–1795. Chem. 53, 1098–1108. Chiou, W.L., Buehler, P.W., 2002. Comparison of oral absorption and bioavailability of drugs between monkey and human. Pharm. Res. 19, 868–874.

Chapter 3: The Use of ROC Analysis for the Qualitative Prediction of Human Oral Bioavailability from Animal Data

Pharm Res. 31(2014), 720-730

Andrés Olivares Morales, Oliver J.D. Hatley, David Turner, Aleksandra Galetin, Leon Aarons and Amin Rostami Hodjegan

Pharm Res (2014) 31:720–730 DOI 10.1007/s11095-013-1193-2

RESEARCH PAPER

The Use of ROC Analysis for the Qualitative Prediction of Human Oral Bioavailability from Animal Data

Andrés Olivares-Morales & Oliver J. D. Hatley & David Turner & Aleksandra Galetin & Leon Aarons & Amin Rostami-Hodjegan

Received: 23 June 2013 /Accepted: 9 August 2013 /Published online: 27 September 2013 # Springer Science+Business Media New York 2013

ABSTRACT KEY WORDS BDDCS . interspecies . oral bioavailability . Purpose To develop and evaluate a tool for the qualitative pre- qualitative prediction . ROC analysis diction of human oral bioavailability (Fhuman) from animal oral bioavailability (Fanimal) data employing ROC analysis and to identify the optimal thresholds for such predictions. ABBREVIATIONS Methods A dataset of 184 compounds with known Fhuman and AUC Area under the ROC curve Fanimal in at least one species (mouse, rat, dog and non-human BCS Biopharmaceutical classification system primates (NHP)) was employed. A binary classification model for BDDCS Biopharmaceutical drug disposition classification Fhuman was built by setting a threshold for high/low Fhuman at 50%. system The thresholds for high/low Fanimal were varied from 0 to 100 to F Oral bioavailability generate the ROC curves. Optimal thresholds were derived from fa Fraction of the dose absorbed in the gastrointes- ‘cost analysis’ and the outcomes with respect to false negative and tinal tract false positive predictions were analyzed against the BDDCS class Fanimal Oral bioavailability in animals/species distributions. FG Fraction of the dose absorbed that escapes gut Results We successfully built ROC curves for the combined wall first pass metabolism dataset and per individual species. Optimal Fanimal thresholds were FH Fraction of the dose absorbed that escapes hepatic found to be 67% (mouse), 22% (rat), 58% (dog), 35% (NHP) first pass metabolism and 47% (combined dataset). No significant trends were ob- F human Oral bioavailability in humans served when sub-categorizing the outcomes by the BDDCS. FN False negative Conclusions Fanimal can predict high/low Fhuman with adequate sen- FP False positive sitivity and specificity. This methodology and associated thresholds can GI Gastrointestinal tract be employed as part of decisions related to planning necessary studies iv Intravenous during development of new drug candidates and lead selection. J Youden’s index NHP Non-human primates Electronic supplementary material The online version of this article NPV Negative predictive value (doi:10.1007/s11095-013-1193-2) contains supplementary material, which is PPV Positive predictive value available to authorized users. : : : : QSAR Quantitative structure-activity relationship A. Olivares-Morales O. J. D. Hatley A. Galetin L. Aarons ROC Receiver operating characteristics * A. Rostami-Hodjegan ( ) t Animal, high/low oral bioavailability threshold Centre for Applied Pharmacokinetic Research A School of Pharmacy and Pharmaceutical Sciences tH Human, high/low oral bioavailability threshold The University of Manchester, Stopford Building, Oxford Road TN True negative Manchester M13 9PT, UK TNR True negative rate e-mail: [email protected] TP True positive : D. Turner A. Rostami-Hodjegan TPR True positive rate Simcyp Limited, Blades Enterprise Centre, Sheffield, UK US-FDA United States Food and Drug Administration Qualitative Oral Bioavailability Prediction 721

INTRODUCTION trends were observed between animal and human bioavailability. However, issues were found with regards to the num- Oral bioavailability (F) is considered a key parameter during ber of data points in the plot as compared to the original drug development. It can be defined as the fraction of the dose publication from Sietsema (1989), in addition species defini- administered orally that reaches systemic circulation, as its tions and the fact of the publication was based on data from unchanged form, which becomes available at its intended site 1989 were suggestive that a new dataset was needed in order of action to produce the desired therapeutic effect. Even to perform a comprehensive analysis of the relationship be- though the latter is difficult to measure, it is usually assumed tween animal and human oral bioavailability. A recent study that the amount of drug at the site of action is proportional to by Musther and co-workers (16), addressed those needs by the amount of drug in plasma/blood. Oral bioavailability is introducing a more comprehensive and updated dataset and dependent upon the fraction of dose that is absorbed in the the correlations between animal and human oral bioavailabil- gastrointestinal (GI) tract (f a), as well as the fraction that ity were investigated. Their results were in agreement with the escapes first pass metabolism in both the GI tract and the analysis performed in the past by several groups (4,9, liver, F G and F H, respectively (Eq. 1)(1,2). 13,17,18). Amongst the species studied, NHP showed the stron- gest correlation with human, followed by poor correlations for ¼ Â Â ð Þ F f a F G F H 1 dog, rat and mouse. However the large prediction intervals suggestthatapointwisecorrelationbetweenhumanandpre- clinical species is not plausible. In terms of qualitative predic- Despite its importance, the information regarding oral tions, the study showed that mouse, rat and NHP underpredict drug bioavailability is not always available during the devel- human bioavailability. More interestingly, the median of the opment stages, as it usually requires the administration of an ratio between animal and human bioavailability was close to intravenous (iv) dose as a reference. However, due to safety unity, however the large intervals for the median ratio suggest and solubility reasons the iv dose is not always available; that these results should be treated carefully. Despite the sug- therefore, drug oral bioavailability is generally unknown until gested lack of predictability of human oral bioavailability from later stages in the development process. In addition, low oral preclinical species, the models are still employed during drug bioavailability is generally associated with higher inter indi- development. The information gathered from oral bioavailabil- vidual variability (3). ity studies in animal models is employed as part of the decision It is a common practice during drug development to em- making process of whether to continue or not with the develop- ploy animal models for the in vivo determination of safety, ment of any particular drug (19,20). efficacy and pharmacokinetic properties of a drug candidate Marketed drugs and drug candidates can be classified (1). The main goal of such studies is to predict drug’s according to extent of metabolism by means of the behaviour in humans/man based on animal data. The rela- Biopharmaceutics Drug Disposition System (BDDCS) (21). tionship between the animal models and human oral drug Similarly to the Biopharmaceutics Classification System bioavailability has been studied on several occasions. One of (BCS) (22), the BDDCS divides the compounds into four the first attempts to investigate this relationship was performed classes based on their aqueous solubility and permeability. by Sietsema in 1989, where a poor correlation was found However, the main difference between the two systems is the between animal (rodents, dogs and non-human primates permeability component. In the BCS, permeability relates to (NHP)) and human oral bioavailability (4). The lack of corre- intestinal permeability rate and the extent of absorption, lation could be explained by the interspecies differences in the whereas with the BDDCS permeability relates to the perme- factors governing oral bioavailability such as morpho- ability rate in the intestine and/or liver, which was found to be physiological differences in GI tract, abundance of trans- correlated with the extent of metabolism (21,23,24). The porters and metabolic enzymes and their regional distribu- BDDCS defines a highly soluble compound as a compound tion, given that physicochemical properties are inherent to whose highest regulatory approved strength is able to dissolve the drug and/or formulation (1,4–11). Nonetheless, cor- in 250 mL (or less) of water throughout a pH range of 1 to relations have been established for intestinal permeability 7.5 at 37°C. Likewise, the BDDCS defines highly permeable and/or the fraction of dose absorbed between humans compounds as compounds where 70% or more of the admin- and preclinical species, in particular for rat and NHP istered oral dose is metabolized. A compound is considered (10,12–14), suggesting that for those particular species, poorly metabolized if 50% or more of the administered dose is bioavailability differences may rely on the metabolic excreted in the urine or bile in its unchanged form (25–28). component. Thus, BDDCS Class 1 compounds are highly soluble and Based on Sietsema (1989) data, Grass and Sinko (15)plot- highly metabolized, Class 2 compounds are poorly soluble ted the relationship between animal and human oral bioavail- and highly metabolized, Class 3 compounds are highly soluble ability. The plot was similar to a scatter plot and therefore no and poorly metabolized and Class 4 compounds are poorly 722 Olivares-Morales et al. soluble and poorly metabolized. In addition, it has been In a similar fashion, Eq. 2 was modified to classify high and suggested that BDDCS class can be useful to estimate the low F animal, by setting up a threshold for animal oral bioavail- impact of intestinal transporters in drug absorption and meability (t A) as per Eq. 3. tabolism as well as propose possible food effect and clinically ; ≥ relevant drug-drug interactions (21,25–29). ¼ high if F animal; species tA ð Þ F animal; species ; < 3 A model can be evaluated based upon its ability to correctly low if F animal; species tA predict any particular outcome, where the prediction performance of a binary classification model can be evaluated by means of Receiver Operating Characteristic (ROC) analysis. Thus for F animal based predictions, false positives (FP) were The ROC space consists of a plot of the False Positive Rate compounds with high Fanimal and low Fhuman, and false nega- (FPR) as a function of the true positive rate (TPR). A binary tives (FN) were compounds with low Fanimal and high Fhuman classification is represented by a single point in the ROC space, (Fig. 1). If both Fanimal and Fhuman were high, the compound where a perfect classification will have a FPR of 0 and a TPR of was classified as a true positive (TP), similarly if both F animal 1. Likewise, a continuous system can be represented by curve in and F human were low, the compound was considered as a true theROCspaceandtheareaundertheROCcurve(AUC)can negative (TN). The predictions based on the animal data be employed as a measure of the performance of the predictions analysis for bioavailability were evaluated by calculating its made from the classification system or model. An AUC of 1 sensitivity, specificity, positive predictive value (PPV) and neg- corresponds to a perfect classification/prediction and an AUC of ative predictive value (NPV) for the determined animal 0.5 corresponds to a random classification/prediction (30–32). thresholds (tA), as shown in Table I. In the present study, we introduce a new model for the All the statistical analysis and the ROC curve generation ® categorical prediction of human oral bioavailability, stated as were implemented with Matlab 2012a and its statistical low or high, from animal data by employing a threshold deci- toolbox (The MathWorks Inc., Natick, MA, USA). The con- sion tool based upon ROC analysis. Oral bioavailability for struction of the ROC curves, for the overall dataset and by more than 180 compounds was analysed in different preclinical species, was achieved by varying t A form 0 to 100 and record- species to generate animal oral bioavailability thresholds that ing the error rates (Table I) for each threshold. The overall (all can be employed for the qualitative prediction of human oral the species combined) ROC curve was constructed by consid- bioavailability. Furthermore, the relationships between the ering all the datapoints within the dataset, including the resulting classifications were compared with the BDDCS classi- compounds with Fanimal values in more than one preclinical fication of the compounds employed for the analysis.

MATERIALS AND METHODS

Dataset Employed

A total of 184 different compounds with reported oral bioavailability in both human and preclinical species, namely mouse, rat, dog and NHP, were employed in this study. The oral bioavailability values for the compounds were collated from the literature by Musther and co-workers, as described elsewhere (16).

Binary Classification and ROC Analysis

A binary classification model was implemented by establishing a threshold (t H) for high and low human bioavailability at 50% (Eq. 2). A positive outcome (high F human) was defined when F human was greater or equal than 50%, while a negative outcome (low F human)occurred when F human waslessthan50%. Fig. 1 Threshold based predictions of human oral bioavailability from animal ; ≥ data. FN, False negatives; TP, True positives; TN, True negatives; FP, False ¼ high if F human tH ð Þ t t F human ; < 2 positive; A, Animal high/low bioavailability threshold; H, human high/low low if F human tH bioavailability threshold. Qualitative Oral Bioavailability Prediction 723

Ta b l e I Definitions and Formulae for the Evaluation of the Binary Classification low oral bioavailability) in the overall dataset, and for these System compounds the human threshold for high/low oral bioavail- Parameter Formula Probability ability was set up at 10%. The analysis was conducted using rat, dog and NHP data and the significance of the AUC and

Sensitivity or TPR TP P[ high Fanimal| high Fhuman] determination of optimal thresholds was performed, as TPþFN described above (Eq. 4). Specificity or TNR TN P[lowFanimal|lowFhuman] TN þFP

PPV TP P[ high Fhuman |highFanimal] BDDCS Classification TPþFP

NPV TN P[low Fhuman |lowFanimal] TN þFN The BDDCS classification for 155 of the compounds in the FPR 1−TNR P[ high Fanimal|lowFhuman] current study was obtained from the dataset published by Benet, and co-workers (34), while for the remaining compounds, a provisional BDDCS classification was given. All species. In addition, a bootstrap random resampling (n = individual drugs used for the analysis and their allocated 10000) ,with replacement, was performed for the determina- BDDCS classification are listed in Tables SI and SII of the tion of the average ROC curves and the bias corrected con- Supplementary Material. For the classification, the extent of fidence intervals for the resulting AUC, sensitivity and speci- metabolism and/or percentage of the dose excreted ficity. The curves were averaged by threshold averaging func- unchanged in the urine and/or bile was collated from the tion within Matlab®. The significance of the differences be- literature. With regards to solubility determination, for US- tween the resulting AUCs (for each species and the overall FDA approved drugs, the maximum dose strength was taken dataset) and the random classification (AUC=0.5) were cal- from the label, when possible; otherwise the maximum dose culated by the non-parametric Mann-Whitney U test. Deter- strength was extracted from published data. Aqueous solubil- minations of the optimal thresholds for the averaged ROC ity values were taken from published data when possible, curves were calculated as the interception of a line of slope, S otherwise the “Mass Solubility” value within SciFinder® (Eq. 4), with the averaged ROC curve. chemical properties was employed (35). High/low solubility costðÞ FP −costðÞ TN N was assigned using Eq. 6,whereDn is the dose number, Dmax S ¼ Â ð4Þ costðÞ FN −costðÞ TP P is the maximum dose strength (mg) and Cs is the aqueous solubility (mg/mL). High solubility was assigned to compounds with Dn equal or greater than 1, while low solubility Where, cost(FP) and cost(FN) are the costs of FP and FN, was assigned for compounds with a Dn less than 1. respectively; cost(TN) and cost(TP) are the costs of TN and TP, Dmax =250 mL respectively; N and P, are the number of positives and negatives Dn ¼ ð6Þ C s values based on Fhuman data (32). The net costs for TN and TP were assumed to be 0, initially the net costs for FP and FN were assumed equal. Subsequently, the cost ratio between FP and FN Class distribution was then compared within each of the was varied in order to evaluate the impact of the cost assumptions threshold-based outcome groups (i.e. true positives (TP), true on the determination of the optimal threshold points. negatives (TN), false positives (FP) and false negatives (FN)) for For rat, dog and NHP, alternative thresholds were deter- the rat, dog and NHP. Significance of the difference between mined by visually comparing the relationship between animal the BDDCS class distribution within each of the out- threshold (tA) versus its resulting sensitivity and specificity; the come groups (i.e. TP, TN, FP, FN) were evaluated by closest points to the intersection lines between the latter two visual inspection of the plots and by employing Fisher’s were considered as alternative thresholds. New thresholds exacttestimplementedinthe R statistics Package were compared against the optimal thresholds derived from (http://www.r-project.org/). cost analysis employing Youden’s index ( J) (Eq. 5), where the maximum value for J is 1 for a perfect classification, whereas the minimum value is 0 for a threshold with no predictive RESULTS power (33).

J ¼ sensitivity þ specificity − 1 ð5Þ The analysed dataset consisted in total of 318 data points for animal and human oral bioavailability divided in 30, 122, 125 and 41 pairwise correlations between mouse, rat, dog and An additional ROC analysis was performed for the com- NHP with human oral bioavailability, respectively (Fig. 2a). pounds in the dataset whose Fhuman values were ≤20% (very For some of the compounds in the dataset oral bioavailability 724 Olivares-Morales et al. values were available for more than one species as shown in Fig. 2b for rat, dog and NHP. The summary of resulting ROC curves for the total dataset of 318 Fanimal values and per individual species are shown in Table II. Combination of all F animal values resulted in a smooth ROC curve with an AUC around 0.8 (Fig. 3). A similar outcome was observed employing individual dog and rat data (Fig. 4b, c) with ROC curve AUC for dog data slightly higher than for the rat (0.8 and 0.7, respectively). In contrast, ROC curves generated for mouse and NHP resulted in a step- like curve rather than a smooth curve (Fig. 4a, d), primarily attributed to the limited number of data points employed in their construction. However, the AUC values of the latter ROC curves were higher than for the overall dataset and the rest of the species. Interestingly, NHP showed an AUC value close to the unity (0.96). With regards to the significance Fig. 3 Averaged ROC curve for the human versus animal bioavailability tests for the above AUCs - all the species combined and by dataset for all the preclinical species (mouse, rat, dog and NHP) combined. species- showed high significance levels compared to the ran- The dashed line corresponds to the line for random classification, AUC=0.79 dom classification (AUC=0.5) yet for mouse data, p-value was for the overall dataset. slightly higher than for the rest of the species. The ‘cost analysis’ determinations of the optimal Fanimal thresholds are summarized in Tables III and IV For the FN, FP, TN and TP according to the aforementioned thresh- overall dataset, threshold was 47%, in agreement with the olds are shown in Tables SIII and SIV of the Supplementary 50% value employed for the human bioavailability threshold Material. with resulting specificity and sensitivity of 0.82 and 0.66, The NPV and PPV values showed an opposite relationship respectively. A similar situation occurred for dog data, though to their corresponding sensitivities and specificities; this trend the threshold was slightly higher than for the overall dataset, was evident for all of the species and the overall dataset. All of whereas specificity and sensitivity were balanced with values the corresponding PPV values were above 0.8, with the ex- closeto0.75.Thehighestthresholdwasfoundwhen ception of the rat (0.72). Interestingly, NPV values were employing mouse data (tA =67%), which gave rise to perfect similar for almost all of the species and the overall dataset; specificity but this should be viewed with caution given the however, corresponding values were higher for NHP. The limited data available (n =30). Rat and NHP showed thresh- analysis of the impact of the ratio between the cost of FP and olds <50%, whilst the corresponding specificity for the rat was FN on the resulting thresholds showed a sigmoid like increase lower than for any species investigated. In agreement with the on the threshold value when varying the ratio from 0.0025 to AUC values, NHP showed the highest specificity and sensitiv- 20 (Fig. 5). However, in some regions the thresholds remained ity for its optimal threshold. Details of the drugs classified as invariant to changes in the FP/FN ratios.

Fig. 2 (a) Pie chart of the distribution of the oral bioavailability data points employed for the analysis by species, mouse (n =30), rat (n =122),dog(n =125)and non-human primates (NHP) (n =41). (b) Venn diagram of the relationship between oral bioavailability data points for rat, dog and NHP. The area of the circles represents the number of compounds with oral bioavailability data for both animal species and humans, the areas of the interception represents the number of compounds with bioavailability data for more than one species. Qualitative Oral Bioavailability Prediction 725

Fig. 4 Averaged ROC curves for the human versus animal bioavailability dataset by preclinical species. (a) Mouse ROC curve, AUC=0.82; (b) Rat ROC curve, AUC=0.73; (c) Dog ROC curve, AUC=0.80; (d) NHP ROC curve, AUC=0.96; Dashed line corresponds to the line for random classification.

Visual inspection of the plots in Fig. 6 led to alternative were lower than the values achieved by the aforementioned animal thresholds at 28%, 54%, and 31% for rat, dog and thresholds (Tables III and V). NHP, respectively (Table III). Even though specificity and ROC Analysis of the reduced dataset of compounds with sensitivity were balanced for all the species, NPV and PPV very low F human was not possible for NHP due to the lack of showed similar relationships to the previous analysis. Yet, the compounds. Analysis of dog data (n =11) showed that this alternative thresholds showed no improvement of the overall preclinical animal model cannot be applied to the categorical predictability of animal data for the particular species as prediction of very low bioavailability compounds since AUC compared with the former thresholds derived from the cost was not significantly different from the random classification analysis. Youden’s index ( J) values for alternative thresholds (data not shown). Nevertheless, for rat data (n =34) ROC

Table II Area Under the ROC Curve for Animal Models Table III Cost Analysis Derived Optimal Thresholds for FAnimal and its Corresponding Evaluation Metrics Species na AUC Lower 95% confi- Upper 95% confi- p-value b b a a dence interval dence interval Species Opt. tA (%) Specificity (95% CI) Sensitivity (95% CI) J

All 318 0.786 0.734 0.835 <0.0001 All 47 0.82 (0.75, 0.88) 0.66 (0.59, 0.73) 0.48 Mouse 30 0.819 0.613 0.936 <0.005 Mouse 67 1.00 (1.00, 1.00) 0.67 (0.41, 0.86) 0.67 Rat 122 0.731 0.634 0.818 <0.0001 Rat 22 0.60 (0.46, 0.72) 0.77 (0.67, 0.86) 0.37 Dog 125 0.796 0.708 0.870 <0.0001 Dog 58 0.80 (0.67, 0.89) 0.70 (0.59, 0.80) 0.50 c NHP 41 0.963 0.871 0.992 <0.0001 NHP 35 1.00 (1.00, 1.00) 0.84 (0.65, 0.96) 0.84 a n , number of data points employed for the determination of the ROC curve Opt. tA,optimalthresholdforFanimal; 95% CI, 95% confidence interval; J, b determined by bootstrap (n =10000) Yo u d e n’sIndexasperEq.5 c NHP, non-human primates a determined by bootstrap (n =10000) 726 Olivares-Morales et al.

F Table IV Cost Analysis Derived Optimal Thresholds for Animal and its Cor- distribution within each threshold model classes (FN, TP, responding PPV and NPV TN and FP) for any of the preclinical species.

Species Opt. tA (%) NPV PPV

All 47 0.64 0.84 DISCUSSION Mouse 67 0.67 1.00 Rat 22 0.66 0.72 Several groups have suggested that pointwise correlation of oral Dog 58 0.67 0.82 bioavailability between preclinical species and human is almost NHP 35 0.80 1.00 non-existent and, therefore, data from such studies in preclinical species should be treated with caution if the intention is to predict Opt. tA, optimal threshold for Fanimal; NPV, negative predictive value; PPV, positive predictive value human oral bioavailability quantitatively (4,15,16,36). The current study performed a systematic analysis of F animal data from four different species, including 184 compounds across all analysis was possible (Fig. 7) and the results showed that the rat BDDCS classes. Rather than attempting to predict a particular model provided a significantly improvement on the predictions value for Fhuman employing Fanimal data, ROC analysis performed as compared to the random classification (AUC=0.77; 0.52– here showed that the animal data can be employed for a cate- 0.91, 95% confidence interval (CI)). In this case, the threshold gorical prediction of high or low human bioavailability. The area for high/low Fanimal was 7.4% with a corresponding sensitivity under the ROC curve is representative of the probability of – – of 0.95(0.75 1.00, 95% CI) and specificity of 0.67(0.40 0.88, correctly classifying Fhuman as high and low when employing 95% CI), indicating that rat can be used for predictions of very Fanimal data: in our analysis the probability was around 80% low values of human bioavailability predictions. which can be considered as a high value (30). Even though mouse Provisional BDDCS classification was assigned to 29 of the data showed similar performance, the reduced number of data compounds in the dataset (Table SI in the Supplementary points employed in the analysis limit us from making any con- Material). The majority of the compounds analysed were clusions about the utility of this preclinical model for making this classified as Class 1 (47%), followed by Classes 3 (24%), 2 type of prediction. On the contrary, the higher abundance of rat (22%) and 4 (7%). The proportion of compounds classified as and dog data during preclinical development stages was reflected BDDCS Class 1 to 4 for every category after the application of in the smother ROC curves built for those particular species and the classification model (i.e., TP, FN, TN and FP) is repre- the resulting high probabilities for a correct classification sented in Fig. 8. The results from Fisher’stestshowedno employing those species. This finding is in contrast to previous significant difference (p value >0.05) between the overall attempts to establish correlations between animal and human BDDCS class distribution (“Ini.” In Fig. 8)andthe oral bioavailability (10,13,15,16,37). Albeit with a much smaller sample size than rat and dog, NHP showed the best performance of all the species with a probability for a correct classification around 100%. The results are consistent with previous findings for this particular species from pointwise correlation analyses (13,16). However, the cost and ethical implications on the use of NHP during preclinical development will limit their use to later stages of the drug development process. The resulting thresholds for the cost analysis in rat, dog and NHP are consistent with the correlation analysis performed by Musther and co-workers (16). Rat and NHP proved to be underpredictive of human oral bioavailability and in the same fashion the thresholds for high/low F Animal were lower than the threshold for high/low F human. In contrast, threshold for dog (58%) was relatively close to the human value, which is consistent with the observed trend in the dog to neither underpredict nor overpredict human oral bioavailability. The model achieved high TPR (>0.70) and acceptable FPR Fig. 5 Impact of the FP/FN cost ratio on the determination of the optimal (≤0.40) employing the aforementioned thresholds for all the thresholds by Eq. 4. Sky-blue line and circles, thresholds for the combined preclinical species. The former suggests that it is unlikely to dataset; Blue line and upper triangles, mouse thresholds; Red line and squares, rat thresholds; Green line and lower triangles, dog thresholds; Yellow line and have low oral bioavailability values in human when having diamonds, NHP thresholds. high oral bioavailability in any of the preclinical species. Qualitative Oral Bioavailability Prediction 727

Fig. 6 Sensitivity and specificity as a function of Fanimal thresholds for rat (a), dog (b)andNHP(c). An increase on the thresholds will increase the specificity but at the same time will decrease its sensitivity, the thresholds closer to the intercept between the two lines were chosen as alternative thresholds. Red line and cross (-+-), specificity; Blue line and asterisk (-*-), sensitivity.

When we determined the optimal thresholds by an alter- (Fig. 6) and thus the thresholds represent the best estimator native method, the results were similar to the ones based on based on a balanced situation. cost analysis for rat, dog and NHP. In addition, the new Even though rat predictions showed higher FPR than the thresholds did not provide any significant improvement of rest of the preclinical species, its convenience and availability the ability to predict human high/low bioavailability from during preclinical development stages make it one the animal data. The optimal thresholds may also be determined commonest species employed to generate bioavailability data by other alternative methods, such as the Euclidean distance for new drug candidates. Our threshold for rat predictions of between the ROC curve and the hypothetical perfect classifi- high/low F human was 22%, consistent with similar thresholds cation (e.g., point 0, 1 in the ROC space) (38). However, this (F ≥20–30%) for the evaluation of rat F animal during drug particular method does not take into account the nature of the development published by different groups in the industrial dataset unlike the ‘cost analysis’ determination based on Eq. 4. setting (19,20). In contrast to the current analysis, previously Determination of the optimal thresholds according to Eq. 4 proposed thresholds fail to provide any indication of the will depend upon the assigned net cost of FP and FN. In a expected range for human bioavailability and also the corre- conservative fashion, a higher cost can be assigned to the FP sponding TPR and FPR for the rat based predictions. From and the thresholds will move to a higher value and the oppo- the analysis of the reduced dataset for very low human oral site will occur when assigning a higher net cost to FN. How- bioavailability (<20%), rat data showed the best performance ever, as shown in Fig. 5, the increase of the ratio follows a step across species for prediction of F human in this range, with the like increase in the thresholds (showing insensitive regions to probability for a correct prediction of high/low F human of 77%. the assigned net cost ratios). For example a two-fold increase on the net cost of a FP compared to the net cost of FN will increase the threshold for both rat and dog, whereas a three- fold increase will have the same impact on the resulting thresholds as a tenfold increase. However, moving the thresholds will affect the corresponding sensitivity and specificity

Ta b l e V Alternative Thresholds for Fanimal and its Corresponding Evaluation Metrics

Species Alt. tA Specificity Sensitivity NPV PPV Youden’s (%) (95% CI)a (95% CI)a Index ( J)

Rat 28 0.65 (0.51, 0.77) 0.67 (0.55, 0.77) 0.6 0.72 0.33 Dog 54 0.72 (0.58, 0.83) 0.73 (0.62, 0.83) 0.67 0.78 0.46 NHP 31 0.94 (0.67, 1.00) 0.88 (0.69, 0.96) 0.83 0.96 0.82

Alt. tA, alternative threshold for Fanimal;95% CI, 95% confidence interval; J ’ NPV, negative predictive value; PPV, positive predictive value; , Youdens Fig. 7 ROC curve for rat predictions of very low human bioavailability Index as per Eq. 5 (Fhuman ≤20%) dataset. The dashed line corresponds to the line of random a determined by bootstrap (n =10000) classification. 728 Olivares-Morales et al.

Fig. 8 Number of compounds and BDDCS class distribution for rat (a), dog (b)andNHP(c) as a function of the outcome of the threshold based model. (d, e and f) BDDCS class distribution for the outcome (in percentage of the outcome groups) for rat, dog and NHP, respectively. Ini., initial number of compounds; TP, compounds classified as true positives, FN, compounds classified as false negatives; TN, compounds classified as true negatives; FP,compounds classified as false positives.

The threshold (7.4%), in this case was close the human thresh- classes. Nevertheless, a different distribution can be expected old (10%), with a high TPR of 0.95 and with a small decrease for new drug candidates, with a higher tendency for Class 2 in the FPR to 0.33, confirming further the suitability of the rat and 3 drugs (23,25). However, our analysis on the outcome of data for qualitative predictions of F human during drug devel- the classification model showed no clear tendency for any opment. Nevertheless, the shape of the ROC curve (Fig. 7) particular BDDCS Class. One reason for this observation suggests that it is necessary to obtain more data to validate this can be attributed to the fact that BDDCS is a categorical tool conclusion. with no clear distinction between overall metabolism and first The dataset employed for the generation of this predictive pass metabolism. In addition, the binary nature of both BCS model was based upon successful candidates and hence a and BDDCS does not account for the continuity of the prop- relatively high prevalence (58%) of compounds with high erties employed within them- e.g., solubility, intestinal perme- Fhuman values was observed. It would be of interest to test this ability and/or extent of metabolism- which might be an issue model with a dataset reflective of the “true” drug development for the middle range of those properties and therefore affect- process. However, we do not expect a high variation on the rat ing the actual class distribution. Finally, interspecies differ- threshold nor the resulting TNR and TPR , as our threshold ences in metabolic activity are not accounted for by the was in line with the ones employed in the development setting existing BDDCS, and therefore the class distribution for a and the latter are insensitive to the prevalence of high Fhuman particular drug could be different in a particular preclinical values (33). species. ROC analysis for the evaluation of a predictive model of Even though our methodology was applied on animal human bioavailability has been applied before. Langdon and oral bioavailability data, a similar approach could be

Barret (39) developed a model for the prediction of oral bio- employed to its constituents-fa,FG and FH-which might availability in both rat and human based on QSAR and genetic be of more interest for some groups involved in preclinical programing (GP). The evaluation of such a model was based on development. However, we suggest that a more mechanis- the ability of the model to predict high (F ≥33%) and low tic approach should be attempted if the intention is to (F <33%) bioavailability in both human and rat. Interestingly quantitatively predict drug’s behaviour in man, taking in TNR and FNR were in agreement with results of our study. consideration the physiological differences between the BDDCS distributions of the drugs investigated in our anal- preclinical species and humans in addition to drug’sin- ysis were in agreement with the observed distribution for trinsic characteristic that can impact on drug’soralbio- marketed drugs, where Classes 1 and 3 are the predominant availability and its individual components. Qualitative Oral Bioavailability Prediction 729

CONCLUSION 7. Nishimuta H, Sato K, Mizuki Y, Yabuki M, Komuro S. Species differences in intestinal metabolic activities of cytochrome P450 isoforms between cynomolgus monkeys and humans. Drug Metab A new method for categorical prediction of high and low Pharmacokinet. 2011;26:300–6. human bioavailability from animal bioavailability data was 8. Zamek-Gliszczynski MJ, Lee CA, Poirier A, Bentz J, Chu X, Ellens developed by employing ROC analysis. Oral bioavailability of H, et al. ITC recommendations for transporter kinetic parameter more than 50% in animals can successfully predict high oral estimation and translational modeling of transport-mediated PK and DDIs in humans. Clin Pharmacol Ther. 2013;94:64–79. bioavailability in human, with high TPR and low FPR. A 9. Akabane T, Tabata K, Kadono K, Sakuda S, Terashita S, similar scenario can be expected for bioavailability values Teramura T. A comparison of pharmacokinetics between humans equal or greater than 22%, 58% and 35% in rat, dog and and monkeys. Drug Metab Dispos. 2010;38:308–16. NHP. Even though NHP was the best predictor of F ,rat 10. Cao X, Gibbs S, Fang L, Miller H, Landowski C, Shin H-C, et al.Why Human is it challenging to predict intestinal drug absorption and oral bioavail- was shown to be the best predictor for low human oral ability in human using rat model. Pharm Res. 2006;23:1675–86. bioavailability, supporting the use of this animal model for 11. Bueters T, Juric S, Sohlenius-Sternbeck AK, Hu Y, Bylund J. Rat the F predictions in mid to early stages of drug develop- poorly predicts the combined non-absorbed and presystemically human – ment. Thresholds proposed in the current study can be metabolized fractions in the human. Xenobiotica. 2013;43:607 16. 12. Chiou WL, Barve A. Linear correlation of the fraction of oral dose employed in the pharmaceutical industry as part of the tool absorbed of 64 drugs between humans and rats. Pharm Res. box of methods for making decisions related to planning 1998;15:1792–5. necessary studies during the development of new drug candi- 13. Chiou WL, Buehler PW. Comparison of oral absorption and bio- dates and lead selection. availability of drugs between monkey and human. Pharm Res. 2002;19:868–74. 14. Zhao YH, Abraham MH, Le J, Hersey A, Luscombe CN, Beck G, et al. Evaluation of rat intestinal absorption data and correlation with ACKNOWLEDGMENTS AND DISCLOSURES human intestinal absorption. Eur J Med Chem. 2003;38:233–43. 15. Grass GM, Sinko PJ. Physiologically-based pharmacokinetic simula- A.O-M. is recipient of a PhD grant awarded by CONICYT tion modelling. Adv Drug Deliv Rev. 2002;54:433–51. Chile, Chilean Ministry of Education and a President’sDoc- 16. Musther H, Olivares-Morales A, Hatley OJD, Liu B, and Rostami- Hodjegan A. Animal versus human oral drug bioavailability: do they toral Scholar Award from The University of Manchester. correlate? Eur J Pharm Sci. 2013. doi:10.1016/j.ejps.2013.08.018. O.J.D.H. is funded by a PhD grant awarded by Medical 17. Takahashi M, Washio T, Suzuki N, Igeta K, Yamashita S. The Research Council UK and AstraZeneca. This work was con- species differences of intestinal drug absorption and first-pass metab- tributed to Orbito IMI project (http://www.imi.europa.eu/ olism between cynomolgus monkeys and humans. J Pharm Sci. 2009;98:4343–53. content/orbito) as sideground. The authors would like 18. Ward KW, Nagilla R, Jolivette LJ. Comparative evaluation of oral acknowledge the fruitful comments and discussion made by systemic exposure of 56 xenobiotics in rat, dog, monkey and human. the members of the Centre for Applied Pharmacokinetic Xenobiotica. 2005;35:191–210. Research of The University of Manchester, in particular to 19. Caldwell GW, Ritchie DM, Masucci JA, Hageman W, Yan Z. The new pre-preclinical paradigm: compound optimization in early and Adam Darwich, Kayode Ogungbenro and Nikolaos late phase drug discovery. Curr Top Med Chem. 2001;1:353–66. Tsamandouras. 20. Thomas VH, Bhattachar S, Hitchingham L, Zocharski P, Naath M, Surendran N, et al.Theroadmaptooralbio- availability: an industrial perspective. Expert Opin Drug Metab Toxicol. 2006;2:591–608. REFERENCES 21. Wu C-Y, Benet LZ. Predicting drug disposition via application of BCS: transport/absorption/ elimination interplay and development of a biopharmaceutics drug disposition classification system. Pharm 1. Lin JH. Species similarities and differences in pharmacokinetics. Res. 2005;22:11–23. Drug Metab Dispos. 1995;23:1008–21. 22. Amidon GL, Lennernas H, Shah VP, Crison JR. A theoretical basis 2. Jamei M, Marciniak S, Feng K, Barnett A, Tucker G, Rostami- for a biopharmaceutic drug classification: the correlation of in vitro Hodjegan A. The Simcyp® population-based ADME simulator. drug product dissolution and in vivo bioavailability. Pharm Res. Expert Opin Drug Metab Toxicol. 2009;5:211–23. 1995;12:413–20. 3. Hellriegel ET, Bjornsson TD, Hauck WW. Interpatient variability in 23. Broccatelli F, Cruciani G, Benet LZ, Oprea TI. BDDCS Class bioavailability is related to the extent of absorption: implications for prediction for new molecular entities. Mol Pharm. 2012;9:570–80. bioavailability and bioequivalence studies. Clin Pharmacol Ther. 24. Chen ML, Amidon GL, Benet LZ, Lennernas H, Yu LX, The BCS. 1996;60:601–7. BDDCS, and regulatory guidances. Pharm Res. 2011;28:1774–8. 4. Sietsema WK. The absolute oral bioavailability of selected drugs. Int 25. Benet LZ. The role of BCS (biopharmaceutics classification system) J Clin Pharmacol Ther Toxicol. 1989;27:179–211. and BDDCS (biopharmaceutics drug disposition classification sys- 5. Chhabra RS, Pohl RJ, Fouts JR. A comparative study of xenobiotic- tem) in drug development. J Pharm Sci. 2013;102:34–42. metabolizing enzymes in liver and intestine of various animal species. 26. Benet LZ. Predicting drug disposition via application of a Drug Metab Dispos. 1974;2:443–7. biopharmaceutics drug disposition classification system. Basic Clin 6. Nishimuta H, Nakagawa T, Nomura N, Yabuki M. Species differ- Pharmacol Toxicol. 2010;106:162–7. ences in hepatic and intestinal metabolic activities for 43 human 27. Custodio JM, Wu CY, Benet LZ. Predicting drug disposition, cytochrome P450 substrates between humans and rats or dogs. absorption/elimination/transporter interplay and the role of food Xenobiotica. 2013. doi:10.3109/00498254.2013.787155. on drug absorption. Adv Drug Deliv Rev. 2008;60:717–33. 730 Olivares-Morales et al.

28. Benet L, Amidon G, Barends D, Lennernäs H, Polli J, Shah V, et al. 35. Chemical Abstracts Service. SciFinder®, version 2013, American The use of BDDCS in classifying the permeability of marketed drugs. chemical society. https://scifinder.cas.org/scifinder/view/scifinder/ Pharm Res. 2008;25:483–8. scifinderExplore.jsf (accessed March 30, 2013). Calculated using 29. Shugarts S, Benet LZ. The role of transporters in the pharmaco- Advanced Chemistry Development (ACD/Labs) Software V11.02 kinetics of orally administered drugs. Pharm Res. 2009;26:2039– (© 1994–2013 ACD/Labs). 54. 36. Shanks N, Greek R, Greek J. Are animal models predictive for 30. Hanley JA, McNeil BJ. The meaning and use of the area under a humans? Philosophy, Ethics, and Humanities in Medicine. 2009;4:2. receiver operating characteristic (ROC) curve. Radiology. 37. Chiou WL, Jeong HY, Chung SM, Wu TC. Evaluation of using dog 1982;143:29–36. as an animal model to study the fraction of oral dose absorbed of 43 31. Fawcett T. An introduction to ROC analysis. Patttern Recogn Lett. drugs in humans. Pharm Res. 2000;17:135–40. 2006;27:861–74. 38. Pérez-Garrido A, Helguera AM, Borges F, Cordeiro MNDS, Rivero 32. Metz CE. Basic principles of ROC analysis. Semin Nucl Med. V, Escudero AG. Two new parameters based on distances in a 1978;8:283–98. receiver operating characteristic chart for the selection of classifica- 33. Bewick V, Cheek L, Ball J. Statistics review 13: receiver operating tion models. J Chem Inf Model. 2011;51:2746–59. characteristic curves. Crit Care. 2004;8:508–12. 39. Langdon WB, Barrett SJ. Genetic programming in data mining for drug 34. Benet LZ, Broccatelli F, Oprea TI. BDDCS applied to over 900 discovery. In Ghoshand A, Jain L (eds), Evolutionary computation in drugs. AAPS J. 2011;13:519–47. data mining, Vol. 163, Springer Berlin Heidelberg; 2005. p. 211–235.

Eur. J. Pharm. Sci. 67 (2015): 32-44

Andrés Olivares Morales, Yoshiteru Kamiyama, Adam S Darwich, Leon Aarons and Amin Rostami Hodjegan

European Journal of Pharmaceutical Sciences 67 (2015) 32–44

Contents lists available at ScienceDirect

European Journal of Pharmaceutical Sciences

journal homepage: www.elsevier.com/locate/ejps

Analysis of the impact of controlled release formulations on oral drug absorption, gut wall metabolism and relative bioavailability of CYP3A substrates using a physiologically-based pharmacokinetic model

Andrés Olivares-Morales a, Yoshiteru Kamiyama a,b, Adam S. Darwich a, Leon Aarons a, ⇑ Amin Rostami-Hodjegan a,c, a Centre for Applied Pharmacokinetic Research, Manchester Pharmacy School, The University of Manchester, Manchester, UK b Discovery Drug Metabolism & Pharmacokinetics Management, Analysis & Pharmacokinetics Research Labs., Astellas Pharma Inc., Ibaraki, Japan c Simcyp Limited, Blades Enterprise Centre, Shefﬁeld, UK article info abstract

Article history: Controlled release (CR) formulations are usually designed to achieve similar exposure (AUC) levels as the Received 4 July 2014 marketed immediate release (IR) formulation. However, the AUC is often lower following CR compared to Received in revised form 20 October 2014 IR formulations. There are a few exceptions when the CR formulations have shown higher AUC. This study Accepted 24 October 2014 investigated the impact of CR formulations on oral drug absorption and CYP3A4-mediated gut wall metab- Available online 5 November 2014 olism. A review of the current literature on relative bioavailability (Frel) between CR and IR formulations of CYP3A substrates was conducted. This was followed by a systematic analysis to assess the impact of the Keywords: release characteristics and the drug-speciﬁc factors (including metabolism and permeability) on oral bio- Controlled release formulations availability employing a physiologically-based pharmacokinetic (PBPK) modelling and simulation PBPK models Gut wall metabolism approach. From the literature review, only three CYP3A4 substrates showed higher Frel when formulated Oral bioavailability as CR. Several scenarios were investigated using the PBPK approach; in most of them, the oral absorption of CYP3A CR formulations was lower as compared to the IR formulations. However, for highly permeable com- P-glycoprotein pounds that were CYP3A4 substrates the reduction in absorption was compensated by an increase in

the fraction that escapes from ﬁrst pass metabolism in the gut wall (FG), where the magnitude was dependent on CYP3A4 afﬁnity. The systematic simulations of various interplays between different parameters demonstrated that BCS class 1 highly-cleared CYP3A4 substrates can display up to 220% higher relative bioavailability when formulated as CR compared to IR, in agreement with the observed data collected from the literature. The results and methodology of this study can be employed during the formulation development process in order to optimize drug absorption, especially for CYP3A4 substrates. Ó 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http:// creativecommons.org/licenses/by/3.0/).

1. Introduction from pre-systemic metabolism in both the gut wall (FG) and the liver (FH)(Lin et al., 1999). The magnitude of oral drug absorption and systemic availability Formulation characteristics can play a critical role in the drug are consequences of the interplay between parameters related to absorption process. This applies in particular for drugs for which the drug itself, drug product (formulation), study condition and dissolution, solubility and/or permeability characteristics repre- the system, i.e., the human body. Hence, drug-specific physico- sent the limiting steps for oral absorption, namely, drugs that do chemical and biopharmaceutical characteristics, together with not belong to class 1 in the Biopharmaceutics Classification System anatomical and physiological factors, will determine a drug’s oral (BCS) (Amidon et al., 1995; Wilding, 1999). The BCS defines four bioavailability (F) in a given scenario. F is the product of the frac- classes based on a compound’s aqueous solubility and intestinal tion of the drug that is absorbed (fa) and the fractions that escape permeability (high solubility and high permeability (class 1), low solubility and high permeability (class 2), high solubility and low permeability (class 3), low solubility and low permeability (class ⇑ Corresponding author at: Centre for Applied Pharmacokinetic Research, Man- chester Pharmacy School, The University of Manchester, Stopford Building, Oxford 4)) (Amidon et al., 1995). In general, the selection of a specific for- Road, Manchester M13 9PT, UK. Tel.: +44 161 3060634. mulation is based on its minimal negative impact on the drug E-mail addresses: [email protected] (A. Olivares- absorption rate, i.e., immediate release (IR) formulations. However, Morales), [email protected] (A. Rostami-Hodjegan). http://dx.doi.org/10.1016/j.ejps.2014.10.018 0928-0987/Ó 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/3.0/). A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 33 there are circumstances for which controlling the release rate of Despite almost complete absorption, both buspirone and oxy- the drug from the formulation into the gastrointestinal (GI) lumen butynin display an oral bioavailability of around 4% and 6%, respec- is desirable (Langer, 1990). Hence, understanding the potential tively, due to extensive first-pass metabolism in the gut wall and kinetics of oral absorption for the so-called controlled release liver (Douchamps et al., 1988; Gammans et al., 1985; Lukkari (CR) dosage forms and the prediction of their behaviour based on et al., 1998; Mizushima et al., 2007; Yaich et al., 1998; Zhu et al., in vitro information is valuable. 2005). Cytochrome P450 (CYP) 3A4 is believed to be the main CR formulations provide certain advantages when compared to enzyme responsible for the metabolism of oxybutynin and buspi- their IR counterparts. CR formulations can reduce peak to trough rone (Douchamps et al., 1988; Gammans et al., 1985; Lukkari fluctuations in the plasma concentration–time profile (compared et al., 1998; Mizushima et al., 2007; Yaich et al., 1998; Zhu et al., to multiple-dose administration of an IR product), hence reducing 2005). Therefore it has been hypothesized that the observed differ- fluctuation-related side effects and/or sub-therapeutic concentra- ences between CR and IR formulations are a consequence of the tions. CR formulations can increase the exposure over time of drugs distribution pattern of CYP3A along the small intestine (Gupta with a short elimination half-life, and can be used to target deliv- and Sathyan, 1999; Sakr and Andheria, 2001a,2001b; Tubic- ery into distal regions of the intestine (e.g. colon), or where there is Grozdanis et al., 2008). The abundance of CYP3A varies along the a need for targeted delivery for the treatment of a specific disease, membrane of the small intestine, being higher in the upper region such has Crohn’s disease (Langer, 1990; Rubinstein, 2005; and decreasing towards the distal region and colon (Berggren et al., Thombre, 2005). This can lead to an increased patient compliance. 2007; Paine et al., 1997; Zhang et al., 1999). Therefore, the CR for- Furthermore, CR formulations can be of use in drug development mulation of such drugs would release most of its drug content into when the standard IR formulation is not an alternative due to unfa- intestinal regions with a lower abundance of CYP3A, thus poten- vourable pharmacokinetic properties of the drug candidate tially bypassing the CYP3A-mediated first pass metabolism. This (Langer, 1990; Rubinstein, 2005; Thombre, 2005). hypothesis is supported by an observed reduction in the exposure One of the main goals when developing a CR formulation of a of the metabolites of both buspirone and oxybutynin when admin- marketed drug is to achieve, at least, the same exposure as the istered as a CR formulation vs. their IR formulations (Gupta and equivalent dose of their IR counterpart. In general however the rel- Sathyan, 1999; Sakr and Andheria, 2001a,2001b). The reduction ative bioavailability of a CR formulation compared to its IR coun- in exposure of oxybutynin’s metabolite, N-desethyloxybutynin, terpart is expected to be less than 100% (European Medicines could also explain the reported improvements in the safety profile Agency, 2013). Several physiological factors can influence the of oxybutynin when formulated as a CR (Gupta et al., 1999; observed differences in systemic exposure between IR and CR. A Sathyan et al., 2001). CR formulation is intended to release its drug content within 12– Despite the fact that clinical evidence might support the afore- 24 h, in contrast the small intestinal transit time is around 2–5 h mentioned hypothesis, there are no clear indications whether this (Davis et al., 1986; Fallingborg et al., 1989; Yu et al., 1996). There- higher relative bioavailability would be observable for all CYP3A fore a majority of the dose should be released into distal regions of substrates when formulated as CR. Due to the complex relationship the small intestine and the colon, where the residence time in the between absorption and first pass metabolism in the GI tract colon is about 12–24 h (Coupe et al., 1992; Davis et al., 1986; (Darwich et al., 2010) it might prove difficult to differentiate the Fallingborg et al., 1989). The extended release may limit the main driving forces behind this observed phenomenon, i.e., colonic absorption potential for a drug formulated as CR as, in general, absorption window vs. a decreased gut wall metabolism in the the distal regions of the intestine provide a less favourable envi- colon, or both (Tannergren et al., 2009). To our knowledge however ronment for drug absorption. For instance, the reduced surface there is a paucity of studies investigating these bioavailability dif- area available for absorption in the distal region of the GI tract ferences in a prospective manner. In addition, no attempts have may limit the absorption for poorly permeable compounds been made to either elucidate the drug and formulation properties (Tannergren et al., 2009; Watts and Lllum, 1997), the intestinal associated with the occurrence of such phenomenon or to correlate pH increases towards the distal portion of the intestine conse- its magnitude to the aforementioned drug’s physicochemical, bio- quently limiting the aqueous solubility of basic compounds pharmaceutical and biochemical properties. (Fallingborg et al., 1989). Finally, the lack of bile salts, less fluid vol- Due to the multifactorial nature of the problem, modelling and ume in the colon, differences in the regional permeability and pos- simulation (M&S), in particular physiologically-based pharmacoki- sible degradation by colonic microflora can also have a negative netic (PBPK) M&S, can be useful for the prospective analysis of the impact on the drug absorption of CR formulations (Lennernas, impact of such properties on the absorption and first past metabo- 2014a; Schiller et al., 2005; Sutton, 2009; Tannergren et al., 2009). lism of CR formulations of CYP3A substrates. In silico PBPK models Regardless of the unfavourable conditions for the absorption in integrate current knowledge of both the system, i.e., morphophys- the distal regions of the GI tract, there are a few examples in the iological factors (and their population characteristics) and drug literature were CR formulations of a marketed drug can display properties that may influence oral drug absorption (Jamei et al., higher relative bioavailability compared to their IR formulations. 2009c). This approach has the advantage to allow the theoretical For instance, a single-dose study of a CR formulation of buspirone exploration of the interplay between the system and the drug (5-hydroxytryptamine 1A (5-HT) partial agonist) showed a relative properties and therefore hypothesize on the main driving forces bioavailability of 170–190% as compared to a similar dose of an IR that control drug absorption, transport and metabolism (Darwich formulation (Sakr and Andheria, 2001b) producing an almost 3.3- et al., 2010). fold higher exposure at steady-state (Sakr and Andheria, 2001a). Herein the relative bioavailability between CR and IR formula- For oxybutynin (anticholinergic), the CR formulation displayed a tions of CYP3A substrates was investigated in order to understand relative bioavailability of 153% as compared to the IR formulation how the physicochemical, biochemical and pharmaceutical prop- (Gupta and Sathyan, 1999). Additional studies have showed that erties of a drug (or drug product) can affect its oral bioavailability. the CR formulation of oxybutynin significantly reduced the anti- Firstly, a literature survey was performed to collate clinical studies cholinergic side-effects of oxybutynin as compared to the IR for- in which the pharmacokinetics of CYP3A4 substrates were simulta- mulation, without reducing the efficacy of oxybutynin for the neously investigated in both IR and CR formulations. Secondly, a treatment of urinary incontinency (Comer and Goa, 2000; Gupta systematic analysis was performed to investigate the impact that et al., 1999; Sathyan et al., 2001). drug release characteristics and the drug-related physicochemical 34 A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 and biochemical properties defining oral bioavailability have on carrier mediated intestinal uptake and/or efflux. The intestinal oral drug absorption and CYP3A4-mediated intestinal first pass regional distribution pattern of a given transporter is incorporated metabolism. This was performed using in silico PBPK M&S. The and is expressed relative to the abundance in the jejunum (Mouly aims of this study were to investigate possible mechanisms and Paine, 2003; Jamei et al., 2009c). It is also assumed that the involved in the observed differences in oral bioavailability between absorption from the stomach is negligible compared to the absorp-

IR and CR formulations by analysing the trends in fa, FG, and the tion in the small intestine and the colon. The drug is absorbed into systemic exposure (AUC). In addition, an attempt was made to the enterocyte compartment, where enzymatic first pass metabo- identify the parameter space associated with the higher relative lism can occur by either CYPs and/or UDP-glucuronosyltransferas- bioavailability of drugs formulated as CR compared to their IR es (UGTs), following Michaelis–Menten kinetics; with only the counterparts and to correlate simulations with the observed clini- drug’s free fraction (fraction unbound (fu)) being susceptible to cal data gathered from the literature search. metabolism. Alternatively, the Qgut model (Yang et al., 2007) can be employed for the estimation of the first pass gut wall metabolism. The distribution of CYPs and UGTs enzymes along the GI tract 2. Materials and methods is also incorporated in the ADAM model. The non-metabolized fraction enters the portal vein by means of blood flow limited pro- 2.1. Literature survey cesses and subsequently enters the liver, where additional first pass metabolism can occur prior to reaching the systemic circula- A literature survey was conducted using PubMed and Google tion. A detailed description of the ADAM model within the SimcypÒ Scholar in order to identify studies in which the pharmacokinetics population-based simulator can be found elsewhere (Jamei et al., of CYP3A4 substrates formulated as IR and CR was investigated. 2009b,2009c). The selection of the ADAM model was based on its The search was restricted only to those studies in which the phar- capability to simulate drug absorption and first pass metabolism, macokinetic parameters of both formulations were investigated in taking into account the factors that have an impact on these the same set of subjects, ideally healthy adult volunteers. In order processes. to avoid any possible food effects on the absorption parameters, only studies for which the formulations were administrated in fasted conditions were considered. The main pharmacokinetic 2.3. Study design and parameter selection parameter of interest was the AUC. Whenever reported, the relative bioavailability between the IR and CR formulation, in terms To investigate the impact of different formulations and the rel- of the AUC ratio (CR/IR) and its 90% confidence interval was evant drug properties on fa, FG, and AUC a factorial study was employed. Otherwise it was calculated employing an approxima- designed (Fig. 1). A set of five release profiles, representative of five tion of the Fieller’s Theorem (Fieller, 1954; Motulsky, 2010) using different formulations, were defined by varying the release rate 1 1 the reported AUCs, only when both CR and IR formulations were constant (krel) from 0.096 h to 4.6 h in Eq. (1) investigated in the same set of subjects. The detailed calculation krelt method is described in the Supplementary Material. FrelðtÞ¼1 e ð1Þ

where Frel(t) is the fraction of the dose released from the formula- 2.2. Simulations and PBPK model tion as a function of time (h). The five release profiles were representative of two immediate release (IR) tablets and three For the analysis of the impact of the controlled release formula- controlled release (CR) tablets. The profiles were designed to release tions on fa, FG and systemic exposure, a series of simulations were 90% of the drug content within 0.5, 1, 6, 12 and 24 h, resulting in a 1 conducted employing the Advanced Dissolution Absorption and krel of 4.6, 2.3, 0.38, 0.19, and 0.096 h , respectively (t90). Six drug- Metabolism (ADAM) model within the SimcypÒ population-based specific parameters were selected based on their importance in simulator (Jamei et al., 2009b) Version 12 Release 2 (Simcyp Lim- defining oral bioavailability and were systematically modified to ited, Sheffield, UK). The ADAM model is a PBPK absorption model generate a set of virtual compounds. The modified parameters that integrates the drug physicochemical and biopharmaceutical included: solubility (mg/mL); human jejunal effective permeability, 4 properties (e.g. release profile, solubility, permeability, particle Peff (10 cm/s); maximal CYP3A4-mediated metabolic rate, size, affinity for metabolic enzymes, etc.) and the human physiol- Vmax,CYP3A4 (pmol/min/mg microsomal protein); CYP3A4 affinity, ogy (e.g. gastric empting, intestinal transit times, GI fluid volumes, Km,CYP3A4 (lM); maximal P-gp-mediated efflux rate, Jmax,P-gp metabolic enzyme abundances, blood flows, bile secretion, etc.) (pmol/min); and P-gp affinity, Km,P-gp (lM). In addition, each and their variability (Jamei et al., 2009b, 2009c). Within the ADAM parameter was assigned five different values. Hence, the number model the anatomy of the human GI tract is represented by nine of virtual compounds amounted to 15,625. For each virtual com- consecutive segments (stomach, duodenum, jejunum 1 and 2, pound five simulations were carried out, one for each of the release ileum 1–4, and colon). Each segment is described as a smooth cyl- profiles described above, resulting in a total of 78,125 simulations inder with the anatomical and physiological characteristics of each (57). The specific ranges for each parameter were derived from segment accounted for, i.e., fluid dynamics, pH, bile salt concentra- the literature and were representative of the values obtained exper- tion, surface area, blood flows, gut wall mass and volume, etc. Drug imentally. Unless otherwise stated, the values were representative transit throughout the segments is modelled as first order unidi- of approximately the 1st, 25th, 50th, 75th and 99th percentile of rectional process, from the stomach to the colon. In each segment the range reported in the selected references. the amount of drug is distributed between four different states: Aqueous solubility values were derived by rearranging the dose drug in formulation, drug released (undissolved), drug dissolved, number (Dn) equation (Amidon et al., 1995) into Eq. (2), and and drug degraded in the lumen. The dissolution rate can either employing the Dn values as reported by Benet et al. (2011), only be inputted from an in vitro dissolution profile and/or estimated for the compounds for which the authors reported the experimen- from a built-in diffusion layer model (DLM), it is assumed that only tal aqueous solubility. The dose employed for the estimation of the dissolved drug can be absorbed. Drug absorption into the gut wall solubility as function of the Dn was 30 mg. The reason for selecting is modelled as a first order process depending on the drug’s intes- this dose was based on an exploratory study initially performed for tinal permeability and the segment’s physiological characteristics. buspirone, where administered the dose for the CR formulation When required, Michaelis–Menten kinetics can be used to model was 30 mg (Sakr and Andheria, 2001a,2001b). The aforementioned A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 35

verted back to Peff in the ADAM model, using the same equation. This was done in order to estimate the absorption rate constant

(ka,i) in each of segments of the ADAM model (Jamei et al., 2009c). Enzyme kinetic parameters, i.e., intrinsic metabolic clearance

(CLint), Vmax and Km, for CYP3A4-mediated metabolism in human liver microsomes (HLM) were obtained from the review by Bu for

113 compounds (Bu, 2006). Reported Vmax and Km values were employed directly as no correlation was observed between them. The CYP3A4-mediated intrinsic metabolic clearance was calculated

from the ratio between the Vmax and Km, assuming linear conditions (Vmax/Km). Vmax and Km values were limited, when possible, to those that in combination generated CLint,CYP3A4 values within the CLint,CYP3A4 range reported by Bu (2006). Transporter kinetic parameters, i.e., Jmax and Km, for the P-gp- mediated efﬂux in Caco-2 cell monolayers were obtained from the work of Troutman and Thakker (2003) for 8 different P-gp substrates. In the same way as for the CYP3A4 kinetic parameters, P-gp-related parameters were treated as independent, and the

intrinsic clearance (CLint,P-gp(efﬂux)) was calculated from the ratio between Jmax and Km assuming linear conditions. Limitations were applied as described above to match the reported CLint,P-gp(efﬂux) values (Troutman and Thakker, 2003).

2.4. Model assumptions and simulations

A Simcyp ‘‘compound file’’ was created based on the reported physicochemical characteristics, protein binding and blood-to- plasma ratio for the compound buspirone (Gammans et al., 1986; Gertz et al., 2011; Shibata et al., 2002). The ‘‘compound file’’ was then modified and used as a template to generate a set of virtual compounds from the combinations of the aforementioned parameters. The ionic class of the virtual compounds was set to be neutral in order to simplify the analysis and to reduce the number of combinations that could be derived from accounting for the different ionic classes. The drug’s dissolution rate was estimated using the diffusion layer model built-into the SimcypÒ ADAM model, Fig. 1. Schematic representation of the study design tree. Values 1–5 in each circle where the drug was assumed to be a monodispersed powder with represent the different values that each parameter could take. ‘‘Sim’’ represent the result of a simulation. Each result is a product of the combinations of the different an initial particle radius of 30 lm. Peff values were estimated from values of each parameter. The parameters are as follows: release rate constant (krel), the calculated Papp,Caco-2 values using the default method in the Ò solubility (sol), apparent permeability in Caco-2 cells monolayers (Papp), maximum Simcyp simulator for passively absorbed drugs (Sun et al., metabolic rate by CYP3A4 (V ), substrate affinity for CYP3A4 (K ), maximum max m 2002), Peff was kept constant throughout all the intestinal seg- efflux rate by P-gp (Jmax) and substrate affinity for P-gp (Km). ments. Elimination was assumed to occur only by means of procedure allowed us to evaluate the impact of solubility, regard- CYP3A4-mediated metabolism, both in the liver and the GI tract, less of the selected dose. which was estimated from the aforementioned enzyme kinetics parameters of CYP3A4. The fraction of drug unbound in the entero-

Dose=250 ml cytes (fu,gut) was assumed to be 1 as per Yang et al. (2007). The rest Solubility ¼ ð2Þ Ò Dn of the parameters were kept as Simcyp default values. The input parameters are summarized in Table S1 of the Supplementary Human jejunal effective permeability was obtained from the Material. report by Lennernas (2007). Peff values were converted to apparent The virtual trials were simulated assuming a representative passive permeability in Caco-2 cell monolayers (Papp,Caco-2 population. The values employed were those from the ‘‘healthy (106 cm/s)) employing the relationship reported by Sun and co- volunteers’’ population library within SimcypÒ, assuming no vari- workers (Eq. (3))(Darwich et al., 2010; Sun et al., 2002). This con- ability for the system parameters. A ‘‘minimal’’ PBPK model was version was performed to account for the passive component of the used to describe the disposition and systemic elimination of the intestinal permeability described within Peff, whereas the active simulated compounds (Rowland Yeo et al., 2010). The oral dose component was explicitly accounted by the simulations of the P- was set to 30 mg, administered under fasted conditions together gp-mediated efflux (described below). with 250 mL of water; with sampling up to 36 h post dose (Sakr Log Peff þ0:5441 and Andheria, 2001a,2001b). Simulations were carried out using P ¼ 10 0:7224 ð3Þ app;Caco-2 the SimcypÒ Batch processor on a Dell OptiPlex 7010 PC (Intel Core The use of the aforementioned correlation entails some limita- i7-3770, 16 GB Ram) running Microsoft Windows 7 Enterprise tions mainly due to the limited number of compounds on which it (Dell Corp. Ltd., Berkshire, UK). is based (n = 13), the observed mild correlation (r2 = 0.85), and the associated wide prediction intervals. Thus, a note of caution is rec- 2.5. Data analysis ommended before its application. Nevertheless, for the work performed herein, once the Papp,Caco-2 range was obtained using In order to analyse the simulated data the study tree was sub- the aforementioned correlation, the Papp,Caco-2 values were con- categorized into the four classes described in the BCS, thus leading 36 A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 to a reduction in the number of combinations analysed (from resulting from the use of the mean in vitro metabolic data 78,125 to 12,500) by limiting the values for solubility and perme- (Hallifax et al., 2010; Hallifax and Houston, 2012) the intrinsic met- ability from five to two values each. Selection of the solubility and abolic clearance in HLM was back calculated from the in vivo sys- permeability values was based on the BCS cut-off criteria for high/ temic clearance employing either the well-stirred model low soluble and permeable compounds. Solubility was considered (Rowland et al., 1973) or the dispersion model (Roberts and high if its corresponding Dn was less than 1, whereas for a Dn equal Rowland, 1986). The details of the calculations are described in or greater than 1, solubility was considered as low (Amidon et al., the Supplementary Material. 1995). Permeability was considered high if the calculated fraction absorbed was equal or greater than 0.9, and a value below 0.9 ln 10 krel ¼ ð6Þ was considered as low permeability (U.S. Food and Drug t90 Administration, 2000). The fraction absorbed was calculated employing Eq. (4) (Amidon et al., 1995; Sinko et al., 1991) 3. Results 2Peff R TSI f a ¼ 1 e ð4Þ The literature survey was successful in retrieving and identify- where R is the mean radius of the small intestine (1.75 cm) and TSI is ing 17 studies of 11 different compounds that met the inclusion the mean transit time in the small intestine (3.32 h) (Lennernäs criteria (Fig. 2). The compounds were identified to belong to classes et al., 1992; Yu et al., 1996). 1–3 of the BCS. Based on the 17 studies uniquely identified in this Data analysis was carried out using Matlab 2013a (The Math- investigation, 23 data points were derived for the analysis of the works Inc., Natick, MA, USA). The analysis was focused on the relative bioavailability between CR and IR formulations, 8 of which impact of the release rate constant (krel), and the drug specific were directly given in the reports whilst the rest were calculated parameters on the simulation outcome (fa, Fg and AUC). Several from the information given in the reports. The detailed information scenarios were evaluated for the impact of both CYP3A4 and P- in terms of AUC ratios, 90% confidence intervals and their refer- gp clearance employing a ‘‘one-at-a-time’’ method, i.e., fixing most ences are shown in Table S2 of the Supplementary Material. of the parameters and varying the parameters of interest. These were accomplished by either fixing Vmax,CYP3A4/Jmax,P-gp, and vary- 3.1. Parameter range and values ing Km (CYP3A4/P-gp) or vice versa. The scenarios evaluated are described in Table 1. Amongst the scenarios described in Table 1, the cases in which a The simulated parameters and their ranges are summarized in Table 2. Solubility varied from 105 to 104 mg/mL as derived from CR formulation showed higher relative bioavailability (Frel) than the corresponding IR formulation were investigated in further Eq. (2). The range of solubility values was truncated to a minimum of 0.001 mg/mL and a maximum of 100 mg/mL in order to improve detail. Frel was calculated using Eq. (5) the computational performance of the simulations. Human Peff ran- AUC 4 F ¼ MR 100 ð5Þ ged from 0.04 to 10 10 cm/s. Calculated Papp,Caco-2 values (Eq. rel 6 AUCIR (3)) varied from 0.01 to 80 10 cm/s, covering the range from low to highly permeable compounds (Lennernas, 2007). The where AUC was the AUC of the IR formulation with a k of 4.6 h1 IR rel V and K range varied from 1 to 10,000 pmol/min/ and AUC was the AUC of any of the other formulations evaluated. max,CYP3A4 m,CYP3A4 MR mg microsomal protein and 1–10,000 lM, respectively. J The simulations were compared, in terms of release characteristics, max,P-gp and K ranges were 1–1500 pmol/min and 1–2,000 lM, relative bioavailability and metabolic clearance, with the observed m,P-gp respectively. data derived from the literature search. The latter was performed only for compounds with similar physicochemical properties as the simulated compounds and for those for which the main meta- 3.2. Cut-off values for BCS classification bolic enzyme was CYP3A4, i.e., the CYP3A4 is responsible for 50% or more of the compound’s metabolic clearance (fmCYP3A4 P 0.5). The values that defined the limits for high and low solubility Whenever possible the release characteristics of the literature com- were 10 mg/mL (Dn = 1.2) and 1.0 mg/mL (Dn = 0.12), respectively. pounds were derived from the in vitro release profiles where the Likewise, the value for high permeability was 5 106 cm/s corresponding krel was estimated according to its t90 (Eq. (6)) other- (fa 0.89) whereas for low permeability, the value was 6 wise these were approximated based on the information described 0.5 10 cm/s (fa 0.34). For both solubility and permeability, in the product label and/or clinical studies. With regards to the met- the selected cut-off values coincided with the 25th and 50th per- abolic clearance, in order to avoid any possible underpredictions centile of their selected range (values 2 and 3 in Fig. 1).

Table 1

Different scenarios evaluated for CLint,CYP3A4 CLint,P-gp for all BCS classes as a function of release rate.

Scenario Description CLint,CYP3A4 (lL/min/mg) CLint,P-gp (lL/min)

Vmax (pmol/min/mg) Km (lM) Jmax (pmol/min) Km (lM) Ia CYP3A4 (m) Fixed (500) Variable – – Ib CYP3A4 (m) Variable Fixed (50) – – IIa CYP3A4 (h) Fixed (2500) Variable – – IIb CYP3A4 (h) Variable Fixed (1) – – IIIa P-gp (m) – – Fixed (300) Variable IIIb P-gp (m) – – Variable Fixed (150) IVa CYP3A4 (h) & P-gp (m) Fixed (2500) Variable Fixed (300) Fixed (150) IVb CYP3A4 (h) & P-gp (m) Variable Fixed (1) Fixed (300) Fixed (150) Va CYP3A4 (h) & P-gp (m) Fixed (2500) Fixed (1) Fixed (300) Variable Vb CYP3A4 (h) & P-gp (m) Fixed (2500) Fixed (1) Variable Fixed (150)

(h), high; (m), medium. A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 37

(CLint,CYP3A4 P 2500 lL/min/mg) BCS classes 2 and 3 drugs (Fig. S3A).

For scenarios Ia-IIb the BCS classiﬁcation had an effect on fa, where fa decreased when moving from BCS class 1 to class 4. CLint,CYP3A4 had no impact on fa. In addition, CR formulations systematically displayed lower fa compared to the IR formulations, where the fa was reduced as a consequence of a reduced krel. A similar trend was observed for almost all of the scenarios evaluated in

Table 1. The magnitude of the differences in fa, as a result of changing krel, was higher for highly permeable compounds (BCS classes 1 and 2). On the contrary, FG showed an opposite trend as compared to that of fa. The CR formulations showed higher FG than their IR counterparts, the increase was inversely related to the decrease

in drug release rate. The magnitude of the increase in FG was dependent on the CLint,CYP3A4 and was typically observed for virtual compounds with CLint,CYP3A4 equal to or greater than 200 lL/min/ mg. For compounds displaying a low afﬁnity to CYP3A4, the differ-

ences in FG were almost imperceptible (Figs. 3B and S1B–S2B). On the contrary, for compounds with high afﬁnity for CYP3A4, the dif-

ference in FG as a function of both release rate and CLint,CYP3A4 was highly marked (scenario IIb; Fig. S3B).

3.4. Simulations: P-gp substrates

For the simulated P-gp substrates (scenarios IIIa and IIIb in Table 1) the relationship between AUC and drug release was similar to that observed for the CYP3A4 substrates. Nevertheless, irre- Fig. 2. Analysis of the relative bioavailability between CR and IR formulations of spectively of the values for CLint,P-gp, the AUC decreased as the CYP3A4 substrates. The markers represent the mean relative bioavailability and the release rate was reduced, this was more pronounced for low solu- lines represent the 90% conﬁdence intervals. The marker shape and colour represents the BCS class for each compound: class 1 (blue circle), class 2 (green ble compounds (BCS classes 2 and 4; Figs. 4A and S4A). For BCS square) and class 3(red triangle). (For interpretation of the references to colour in class 1 compounds, CLint,P-gp values between 0.007 and 30 lL/min this ﬁgure legend, the reader is referred to the web version of this article.) had almost no impact on the AUC. However, a decrease in the

AUC was observed when CLint,P-gp was set to 300 lL/min (Figs. 4A and S4A). No differences were noticeable when ﬁxing either Table 2 J or K . As for the CYP3A4 substrates, the f was lower Evaluated parameters and values. max,P-gp m,P-gp a for CR formulations than for their IR counterparts, and decreased Parameter Value as the release rate decreased. On the contrary to what was seen

12345 for CYP3A4 substrates, altering CLint,P-gp had an impact on the fa, 1 where the impact on f was dependent upon the CL values krel (h ) 4.6 2.3 0.38 0.19 0.096 a int,P-gp Solubility (mg/mL) (Dn) 0.001 0.1 1 10 100 and BCS classiﬁcation. The fa of BCS class 2 compounds was the P , ( 106 cm/s) (120) (12) (1.2) (0.12) (0.012) app Caco-2 most sensitive to changes in CLint,P-gp (Figs. 4B and S4B). Since 0.01 0.5 5 10 80 the aforementioned compounds were not subject to metabolism, Vmax,CYP3A4 (pmol/min/mg) 1 100 500 2500 10,000 neither the release rate nor the CLint,P-gp had an impact on FG. Km,CYP3A4 (lM) 1 10 50 100 10,000

Jmax,P-gp(efﬂux) (pmol/min) 1 30 300 500 1500 K (lM) 1 50 150 300 2000 m,P-gp(efﬂux) 3.5. Simulations: CYP3A4 and P-gp substrates

Scenarios IVa–Vb in Table 1 describe the simulations carried out 3.3. Simulations: CYP3A4 substrates for virtual compounds with overlapped afﬁnity for both CYP3A4

and P-gp. When CLint,CYP3A4 was varied, and using a fixed CLint,P-gp In general, a reduction in release rate, i.e., changing from an IR (2 lL/min), no significant differences were observed between the formulation to a CR formulation, was associated with a decrease in new AUC trend compared to the trend observed for CYP3A4 sub- AUC for a majority of the CYP3A4 substrates (Figs. 3A and S1A– strates only (Figs. 5A and S5A). A similar outcome was obtained S3A). However, in certain cases, the AUC either remained constant when the analysis was performed from the P-gp point of view, as compared to the IR formulation or increased when the CR for- i.e., varying CLint,P-gp and using a fixed CLint,CYP3A4 (2500 lL/min/ mulations were employed; dependent on both BCS class and mg); the observed trends were similar to that for P-gp substrates

CLint,CYP3A4. When Vmax,CYP3A4 was kept fixed (scenarios Ia and IIa alone (Figs. S6–7B). Likewise, both fa and FG followed almost a sim- in Table 1), the increase in exposure was only observed for BCS ilar pattern as the observed for CYP3A4 or P-gp substrates only class 1 CYP3A4 substrates with CLint,CYP3A4 values equal to or (Figs. 5B and S5–7B). Although the overall trend remained the greater than 250 lL/min/mg (Figs. 3A and S1A). A similar situation same, subtle changes were observed in the trends of fa and FG in was observed when Km,CYP3A4 was fixed to the ‘medium’ value (sce- response to changes in the CLint,CYP3A4 or CLint,P-gp, respectively; nario Ib in Table 1) though the CLint,CYP3A4 necessary to observe a an increase of CLint,CYP3A4 led to an increase in fa (Fig. 5B), likewise, similar change in exposure was reduced to 50 lL/min/mg an increase in CLint,P-gp resulted in a small increase on the FG (Fig. S2). The use of a low Km,CYP3A4 in scenario IIb, i.e., high affinity (Figs. S6–7B). These changes were dependent of both release rate for CYP3A4, resulted in a similar outcome. However, the AUC also and BCS classification, as the increase in fa was more prominent remained constant for CR formulations of highly cleared for IR formulations of BCS class 2 compounds (Figs. 5B and S5B), 38 A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44

Fig. 3. Impact of release rate (formulation) and CLint,CYP3A4 on AUC (A), fa and FG (B) for non-P-gp substrates. Vmax,CYP3A4 was fixed at 2500 pmol/min/mg whereas the Km,CYP3A4 was varied (scenario IIa in Table 1). For plots A and B, the subplots represent the different BCS classes (1–4), whereas the symbols in each plot represent different CLint,CYP3A4 values: upper triangle (2500), circle (250), square (50), diamond (25), and lower triangle (0.25). For the plots in the right hand side (B), the green lines and open symbols represent FG, whereas the black lines and filled symbols represent the fa. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) whereas the impact of CLint,P-gp on FG was perceptible only for IR formulations (simvastatin, buspirone and oxybutynin). In contrast, formulations of BCS class 1 compounds (Fig. S6A). a majority of the drugs showed either similar or lower relative bioavailability when formulated as CR. Judging from the BCS point of

3.6. Relative bioavailability of CYP3A4 substrates view an a priori trend for either higher of lower Frel was not clear. For instance CR formulations of ﬂuvastatin (BCS class 1) and simva-

Analysis of the relative bioavailability (Frel) of CR formulations statin (BCS class 2), both highly permeable compounds, showed showed that highly (CYP3A4) cleared BCS class 1 simulated com- opposite results in terms of Frel (Fig. 2). Whereas CR formulations pounds could display up to a 220% higher Frel compared to the IR of low permeable compounds, such as propiverine and gepirone formulations. When the trends for the simulations were compared (both BCS class 3), showed similar Frel to their IR formulations. with similar compounds derived from the literature survey, i.e., Therefore this justified the use of more mechanistic and multivar- BCS class 1 and mainly CYP3A4 cleared, there was a very good iate models such as PBPK for M&S purposes in order to accommo- agreement between the simulated Frel and the observed data date several factors influencing the observed differences. (Fig. 6). The back-calculated CYP3A4 clearance values (HLM) from the in vivo systemic clearance are reported in Table S3 of the Sup- 4.2. Impact of release rate on oral drug absorption and bioavailability plementary Material. A general trend towards a reduction in drug exposure (AUC) 4. Discussion was observed in simulations when varying the release rate, i.e., moving from an IR formulation to a CR formulation. These results 4.1. Analysis of the relative bioavailability between CR and IR were anticipated as, in general the CR formulations are intended formulations of CYP3A4 substrates to release the majority the drug content further distally in the intestine (e.g., distal ileum and colon), where the distal regions of Due to the selected inclusion criteria for the search, the analysis the GI tract provides unfavourable conditions for drug absorption was limited only to 11 different compounds (Fig. 2). A larger set of compared to the upper regions of the small intestine (Lennernas, drugs could have been included for this analysis if, for instance, the 2014a; Schiller et al., 2005; Sutton, 2009; Tannergren et al., calculations of relative bioavailability were performed between 2009). This assumption is supported by the observed decrease in different subjects and groups, i.e., the IR data was taken from one fa when switching from IR to CR formulations (Figs. 3–5B). Interest- study whereas the CR data was taken from a separate study. How- ingly the decrease in fa was observed for all the scenarios evaluated ever, this would have confounded the impact of the formulation irrespectively of BCS class, CYP3A4 clearance, and/or P-gp efflux. with the inter-individual variability of the kinetics, leading to var- These results are in line with the work by Tannergren et al. iable Frel. Therefore these studies were not considered. Of the total (2009), where they investigated the colonic absorption and bio- drugs investigated, only three drugs formulated as CR showed sta- availability of several compounds, compared to that in upper tistically significant higher relative bioavailability than their IR regions of the GI tract. For BCS class 1 compounds, the relative A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 39

Fig. 4. Impact of release rate (formulation) and CLint,P-gp on AUC (A), fa and FG (B) for non CYP3A4 substrates. Jmax,P-gp was fixed at 300 pmol/min whereas the Km,P-gp was varied (scenario IIIa in Table 1). For plots A and B, the subplots represent the different BCS classes (1–4), whereas the symbols in each plot represent different CLint,P-gp values: upper triangle (300), circle (6), square (2), diamond (1), and lower triangle (0.15). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5. Impact of release rate (formulation), CLint,CYP3A4, and CLint,P-gp on AUC (A), fa and FG (B). Vmax,CYP3A4 was ﬁxed at 2500 pmol/min/mg whereas the Km,CYP3A4 was varied

(scenario Va in Table 1). CLint,P-gp was fixed to 2 lL/min. For plots A and B, the subplots represent the different BCS classes (1–4), whereas the symbols in each plot represent different CLint,CYP3A4 values: upper triangle (2500), circle (250), square (50), diamond (25), and lower triangle (0.25). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) 40 A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44

Fig. 6. The impact of release rate and CLint,CYP3A4 on the relative bioavailability (%) for BCS class 1 compounds (see methods section). Different release rates (krel) represent 1 different formulations. Km,CYP3A4 was ﬁxed to 1 lM whereas Vmax,CYP3A4 was variable. For the IR control (krel = 4.6 h ) Frel was 100%. The colour bar indicates the Frel values (%) represented in the 3D surface. The circles and line represent the mean Frel and its 90% conﬁdence interval for BCS class 1 compounds mainly cleared by CYP3A4

(fmCYP3A4 P 0.5): buspirone sustained release (BUSP F1), buspirone extended release (BUSP F2), oxybutynin extended release (OXYB), quetiapine extended release (QETP) and cyclobenzaprine extended release (CBZP).

colonic bioavailability was considered good compared to that in of this study as the values Papp values were subsequently the upper regions of the intestine. In this study the Frel between back-transformed into Peff using the same equation by the ADAM the IR and CR formulations for low CYP3A4 afﬁnity BCS class 1 model. compounds, varied between 49% and 80% (mean: 66%) in agree- A similar overestimation could arise for colonic solubility. In the ment with the value reported by Tannergren et al. (2009) present study all compounds were treated as neutral and therefore

(Frel P 70%). On the other hand, the simulated relative absorption, regional differences in the intestinal pH, which are accounted for in fa,rel, for the same compounds varied between 66% and 88% (mean: the ADAM model, did not affect intestinal solubility of the com- 72%). Where Tannergren, and co-workers, reported values between pounds. This may in particular lead to an overestimation of colonic 39% and 127% with a mean of 82% (Tannergren et al., 2009). For solubility of basic compounds, whereas an opposite situation can

BCS classes 3 and 4, however, Tannergren found a low Frel in the occur for acidic compounds, for which the solubility is higher in colon (Frel < 50%). In the current simulation study, Frel varied the upper regions of the GI tract. There are also many in vivo factors between 42% and 68% for BCS class 3 compounds, and 23% and that might contribute to the possible under/overestimation of drug

53% for BCS class 4 compounds, whereas fa,rel varied between 58– dissolution and solubility within the GI tract. For instance the over- 76% and 34–61% for BCS classes 3 and 4 compounds, respectively. simplified composition of the small intestinal and colonic fluids in The latter might indicate an overestimation of the absorption for available PBPK absorption models, as well as the actual fluid vol- BCS classes 3 and 4 compounds in our simulations. This could be umes available to dissolve the drug might affect such estimations due to an overestimation of colonic permeability, in our study (Sjogren et al., 2014). we employed a constant Peff value throughout all intestinal seg- Furthermore, several biopharmaceutical and physicochemical ments within the ADAM model, however this might not be neces- properties, known to influence drug absorption, were not taken sarily the case. It has been suggested that the reduced surface area into account in this study, i.e. particle size and its distribution; and increased number of tight junction in the colon could limit the excipients; and in particular the drug release mechanism, which permeability of passively absorbed compounds (Lennernas, was oversimplified in this study; just to name a few (Martinez 2014a), thus permeability could vary along the GI tract, in particu- and Amidon, 2002). Consideration of such factors would have sig- lar for the colon. This was not taken into account in the simula- nificantly increased the number of simulations to be performed, tions, and could lead to this possible overestimation of fa,rel. thus complicating any subsequent analysis. Those simulations Nevertheless, more data has been sort in order to support the exis- were out of the scope of this work. tence of a differential permeability along the GI tract (Lennernas, 2014b). Another possible source of error that might explain those 4.3. Understanding of the higher relative bioavailability observed for differences was the use of Eq. (3) to correlate Papp,Caco-2 with Peff CR formulations of CYP3A4 substrates (and vice versa). This equation is associated with large prediction intervals and therefore this can affect the Peff predictions (Sun One of the main goals of this work was to identify the parameter et al., 2002). However this is unlikely to affect the overall outcome space in which a drug, formulated as CR, would display higher A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 41 relative bioavailability than the corresponding IR formulation. aforementioned study contained a mixture of gelatine and leci-

The above results clearly indicated absorption – fa – to be reduced thin intended to improve the wettability of simvastatin in the for all the CR formulation as compared to the IR formulations. formulation and promote the formation of microemulsions or Still, in the case of the simulated CYP3A4 substrates, the reduction even micelles, thus improving simvastatin’s dissolution. In fact, in fa seemed to be compensated by an increase in FG (Figs. 3B and the solubility of simvastatin was increased more than 5000 times S1B–S3B), that is, a reduction in the CYP3A4-mediated first pass when it was embedded in the mixture of gelatine and lecithin intestinal metabolism. For some of the simulated compounds, this (Tubic-Grozdanis et al., 2008). It is not clear whether the CR compensation was translated into similar exposure levels of CR for- formulation employed in the study by Jang et al. (2010) used mulations as compared to IR. The proposed explanation is based on the same approach to increase the solubility of simvastatin. Yet, the distribution of the CYP3A abundance along the GI tract. As dis- the exposure of the CR formulation was similar to that of cussed previously in this manuscript, the CYP3A enzymes decrease Tubic-Grozdanis et al. (2008). towards the distal regions of the human GI tract (Berggren et al., Another factor that might have influenced the observed differ- 2007; Paine et al., 1997; Zhang et al., 1999), this pattern is taken into ences in simvastatin’s exposure between IR and CR formulations account in the ADAM model. As a result, when a CR formulation can be the fact that simvastatin is a prodrug that is converted to releases its drug content into the distal regions of the intestine, the simvastatin acid (the active form) in vivo (Prueksaritanont et al., drug would encounter less CYP3A enzymes on its way towards the 2005). This process can occur by means of chemical and enzymatic portal circulation, thus reducing the CYP3A-mediated intestinal first hydrolysis in both the gut wall and lumen, therefore differences pass metabolism. In this study the impact on the AUC was however the enzyme levels along the gut wall membrane could explain only noticeable for highly permeable (BCS classes 1 or 2) and highly some of the observed differences in simvastatin’s exposure cleared drugs (CLint,CYP3A4 P 250 lL/min/mg). This seems reason- (Alvarez-Lueje et al., 2005; Prueksaritanont et al., 2005; Satoh able as the differences in absorption between CR and IR formulation, et al., 2002). However, due to the similar exposure observed for for BCS classes 2 and 4 compounds, would be too high to be simvastatin acid between the IR and CR formulations, we believe compensated by a reduction in the intestinal first pass metabolism. that these differences are predominately due to differences in the Nevertheless, a similar exposure level as the IR formulation was CYP3A-mediated metabolism of simvastatin (Jang et al., 2010; observed for the CR formulations for some of the BCS class 3 Tubic-Grozdanis et al., 2008) compounds (high CLint,CYP3A4 P 2500 lL/min/mg). This could be a Another aspect of this simulation study that may result in dis- product of the aforementioned overestimation in absorption. BCS crepancies between simulated and observed data is the attempt to class 1 compounds, on the other hand, are more likely to be absorbed describe a hypothetical BCS class 1 drug. However, the physiochem- in distal regions of the GI tract (Tannergren et al., 2009). Thus, for this ical, biopharmaceutical, and affinity properties employed herein type of compounds, the reduction in intestinal metabolism could were not necessarily intended to represent those for the drugs used lead to AUC levels higher than that observed for IR formulations for the comparison (i.e., oxybutynin, buspirone, etc.). Finally, in our (Figs. 3A and S3A). study, the fraction of drug unbound in the enterocytes was assumed

A relative bioavailability of up to 220% was observed for the to be 1. This assumption can affect FG estimations, as only the free simulated CR formulations of highly CYP3A4-cleared compounds drug concentration in the enterocyte would be available for metab-

(CLint,CYP3A4 P 2500 lL/min/mg) (Fig. 6). These results were in olism (Darwich et al., 2010; Heikkinen et al., 2012; Sinha et al., good agreement with the clinical observations for CR release for- 2012). This parameter is highly sensitive and this might affect the mulations, for buspirone, oxybutynin, quetiapine and cyclobenza- results of the simulations when there is binding to the enterocytes prine, where the increase in relative bioavailability in the CR (Gertz et al., 2010; Yang et al., 2007). Nevertheless, this was not formulations was dependent upon an apparent reduction in met- the case, as the simulations performed herein were not meant to abolic clearance of the aforementioned compounds. The use of represent any particular compound, rather they were representa- in vivo data for the determination of the in vitro intrinsic clear- tive of hypothetical cases, and thus the CLint,CYP3A4 range should be ance for the analysis in Fig. 6 seemed justiﬁed as the in vitro val- considered as an unbound intrinsic clearance. ues would have underpredicted the in vivo clearance for oxybutynin and buspirone. The in vitro clearance, varied between 268 and 442 lL/min/mg (Gertz et al., 2011; Zhu et al., 2005) for 4.4. Impact of the intestinal P-gp distribution and possible CYP3A4/P- buspirone, and 78–278 lL/min/mg for oxybutynin (Mizushima gp interplay on the bioavailability of CR formulations et al., 2007; Yaich et al., 1998), whereas the value determined from the in vivo clearances (Table S3) were 5454 lL/min/mg The results for the simulated P-gp substrates were consistent and 2932 lL/min/mg for buspirone and oxybutynin, respectively. with the previous work by Darwich et al. (2010). In general both

This underprediction was also observed, to a lesser extent, for absorption and exposure were decreased when CLint,P-gp was cyclobenzaprine, whereas for quetiapine an in vitro value similar increased. No impact on FG was observed as function of the to the in vivo value was observed (Table S3). The mechanisms CLint,P-gp, in this scenario no intestinal metabolism was considered. behind said underpredictions when using human liver micro- In addition, no significant differences in terms of absorption and somes are still unknown; however it has been attributed to fac- exposure were observed between the IR and CR formulations as tors such as the ionization, binding to plasma proteins, and product of variable P-gp clearance (Fig. 4). clearance model inaccuracies (Berezhkovskiy, 2011; Hallifax When the analysis was performed on compounds with over- et al., 2010; Hallifax and Houston, 2012; Poulin, 2013; Poulin lapped affinity for both CYP3A4 and P-g, no significant differences et al., 2012). Simvastatin (BCS class 2) represent an interesting were observed in the trend for AUC compared to that of simulated case that was not in agreement with the simulated Frel across CYP3A4 or P-gp substrates alone (Fig. 5). In the same line, only the defined parameter space. Even though simvastatin is classi- minor differences in the trends for fa and FG were observed. These fied as BCS class 2 the CR formulation showed 2–3-fold higher subtle differences might be an indication of a possible competition relative bioavailability that the IR formulation. One of the reasons between CYP3A4 and P-gp for the substrate in the enterocyte com- for such disagreement with the simulated data was the use of an partments within the ADAM model. However, the reasons for such enabling CR formulation in one of the simvastatin studies differences are not clear yet. Further discussion about these results (Tubic-Grozdanis et al., 2008). The formulation employed in the is included in Sections 5 and 6 of the Supplementary Material. 42 A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44

4.5. Similar studies and the use of PBPK model for formulation absorption differences. For P-gp substrates that were not subject development to first-pass metabolism, no clear differences between the CR and IR formulation were observed. Finally, an interplay between Previous multi-scale studies have investigated the complex CYP3A4 and P-gp was observed for IR formulations, however, more interplay between the factors governing drug absorption and intes- data is needed to investigate the mechanism of such phenomena. tinal first pass metabolism and absorption such as the study by Darwich et al. (2010), using the same ADAM model, or the study Conflict of interest by Heikkinen et al. (2012) using the Advanced Compartmental Absorption and Transit (ACAT) model in Gastroplus™. Neverthe- The authors declare no conflict of interest. A.R-H. is currently on less, to our understanding, this is the first study that has investi- a part-time secondment to Simcyp Ltd. (a Certara company) and gated the impact of the release characteristics from the holds shares in Certara. The SimcypÒ simulator is freely available, formulation on oral bioavailability, specially focused on the inter- following completion of the training workshop, to approved mem- play between the physicochemical, biopharmaceutical and bio- bers of academic institutions and other non-for-profit organiza- chemical properties. tions for research and teaching purposes. From a biopharmaceutics point of view, there are an increasing number of examples of the use of PBPK models for the optimiza- Author contributions tion of new dosage forms, in particular for CR formulations. Some of these examples have recently been reviewed by Brown et al. A.O-M, A.S.D, L.A and A.R-H wrote the manuscript; A.O-M, A.S.D, (2012). The use of PBPK models for the evaluation of the impact L.A and A.R-H designed the study; Y.K and A.O.M performed liter- of biopharmaceutical properties on absorption has recently been ature search, A.O.M performed the simulations; Y.K, performed encouraged by the regulatory agencies such as by the United States pilot study; A.O-M analysed the data. Food and Drug Administration (Zhang and Lionberger, 2014). In addition, our study provides a systematic analysis of the available data on the relative bioavailability of CYP3A4 substrates as well as Acknowledgements the impact of drug- and formulation-specific factors on the oral bioavailability. The outcome of this study can be considered as a A.O-M. is recipient of a PhD grant awarded by CONICYT Chile, first step in the line of providing examples of possible applications Chilean Ministry of Education and a President’s Doctoral Scholar of PBPK M&S in the formulation development process, in particular Award from The University of Manchester. The authors would like for the evaluation of the possible impact of controlled release dos- acknowledge the fruitful comments and discussion made by the age forms on the drug candidate’s absorption and bioavailability. members of the Centre for Applied Pharmacokinetic Research This applies in particular for drugs candidates that are considered (CAPKR) of The University of Manchester, in particular to Aleksan- as CYP3A4 substrates; however more work is needed in order to dra Galetin, Nikolaos Tsamandouras and Alison Margolskee. This fully validate this approach. Due to the complexity of the analysis, project is an associated (‘‘sideground’’) contribution to the IMI Oral we simplified several aspects that would have a clear impact on Biopharmaceutical Tools (OrBiTo) project (http://www.imi.euro- predicted Frel. One of them was to assume a virtual reference pa.eu/content/orbito). human, thus eliminating the inter-individual variability on the physiological factors that influence drug absorption (Jamei et al., Appendix A. Supplementary material 2009a). The parameters employed for the simulations (Tables S1), the results from the literature search (Tables S2 and S3), the methodol- 5. Conclusion ogy employed for the calculations and the results for the scenarios not shown in this manuscript (Figs. S2–S6) and further discussion A factorial sensitivity analysis was performed for the investiga- about the outcome of the simulations involving P-gp and the possi- tion of the differences between immediate release and controlled ble interplay with CYP3A4 can be found in the electronic Supple- release formulations on drug absorption, first pass metabolism mentary Material. and systemic exposure. This was complemented with a literature Supplementary data associated with this article can be found, in survey of the observed differences in oral bioavailability of CR for- the online version, at http://dx.doi.org/10.1016/j.ejps.2014.10.018. mulations of CYP3A4 substrates. The use of a PBPK absorption model allowed the simultaneous consideration of both formulation and drug-specific properties. In general, a reduced absorption was References observed when employing a controlled release formulation. The results matched previous observations made for colonic absorption Alvarez-Lueje, A., Valenzuela, C., Squella, J.A., Nunez-Vergara, L.J., 2005. Stability (Tannergren et al., 2009). However, in some cases the reduction in study of simvastatin under hydrolytic conditions assessed by liquid chromatography. J. AOAC Int. 88, 1631–1636. fa was compensated by a reduction in intestinal metabolism, thus Amidon, G.L., Lennernas, H., Shah, V.P., Crison, J.R., 1995. A theoretical basis for a leading to a net increase in systemic exposure. This increase was biopharmaceutic drug classification: the correlation of in vitro drug product both permeability and CYP3A4-affinity dependent. In addition, CR dissolution and in vivo bioavailability. Pharm. Res. 12, 413–420. formulations of highly CYP3A4-cleared compounds were more Benet, L.Z., Broccatelli, F., Oprea, T.I., 2011. BDDCS applied to over 900 drugs. AAPS J. likely to display higher relative bioavailability than the IR formula- 13, 519–547. Berezhkovskiy, L.M., 2011. The corrected traditional equations for calculation of tions. The simulations were in agreement with the observed clini- hepatic clearance that account for the difference in drug ionization in cal data for a number of CYP3A4 substrates. This study provided extracellular and intracellular tissue water and the corresponding corrected further support to the hypothesis that the observed higher relative PBPK equation. J. Pharm. Sci. 100, 1167–1183. Berggren, S., Gall, C., Wollnitz, N., Ekelund, M., Karlbom, U., Hoogstraate, J., Schrenk, bioavailability of CR formulations of highly cleared CYP3A4 could D., Lennernäs, H., 2007. Gene and protein expression of P-glycoprotein, MRP1, be due to differences in the intestinal first pass metabolism. The MRP2, and CYP3A4 in the small and large human intestine. Mol. Pharm. 4, 252– outcome of this simulation study can be taken as a first step, as 257. Brown, J., Crison, J., Timmins, P., 2012. Predicting feasibility and characterizing drug-specific simulations are required in order to fully support performance of extended-release formulations using physiologically based the PBPK approach for investigation of these metabolic and pharmacokinetic modeling. Ther. Delivery 3, 1047–1059. A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44 43

Bu, H.Z., 2006. A literature review of enzyme kinetic parameters for CYP3A4- Mizushima, H., Takanaka, K., Abe, K., Fukazawa, I., Ishizuka, H., 2007. Stereoselective mediated metabolic reactions of 113 drugs in human liver microsomes: pharmacokinetics of oxybutynin and N-desethyloxybutynin in vitro and in vivo. structure–kinetics relationship assessment. Curr. Drug Metab. 7, 231–249. Xenobiotica 37, 59–73. Comer, A.M., Goa, K.L., 2000. Extended-release oxybutynin. Drugs Aging 16, 149– Motulsky, H., 2010. Intuitive Biostatistics: A Nonmathematical Guide to Statistical 155, Discussion 156-147. Thinking. Oxford University Press. Coupe, A.J., Davis, S.S., Evans, D.F., Wilding, I.R., 1992. Nocturnal scintigraphic Mouly, S., Paine, M., 2003. P-glycoprotein increases from proximal to distal regions imaging to investigate the gastrointestinal transit of dosage forms. J. Controlled of human small intestine. Pharm. Res. 20, 1595–1599. Release 20, 155–162. Paine, M.F., Khalighi, M., Fisher, J.M., Shen, D.D., Kunze, K.L., Marsh, C.L., Perkins, J.D., Darwich, A.S., Neuhoff, S., Jamei, M., Rostami-Hodjegan, A., 2010. Interplay of Thummel, K.E., 1997. Characterization of interintestinal and intraintestinal metabolism and transport in determining oral drug absorption and gut wall variations in human CYP3A-dependent metabolism. J. Pharmacol. Exp. Ther. metabolism: a simulation assessment using the ‘‘advanced dissolution, 283, 1552–1562. absorption, metabolism (ADAM)’’ model. Curr. Drug Metab. 11, 716–729. Poulin, P., 2013. Prediction of total hepatic clearance by combining metabolism, Davis, S.S., Hardy, J.G., Fara, J.W., 1986. Transit of pharmaceutical dosage forms transport, and permeability data in the in vitro–in vivo extrapolation methods: through the small intestine. Gut 27, 886–892. emphasis on an apparent fraction unbound in liver for drugs. J. Pharm. Sci. 102, Douchamps, J., Derenne, F., Stockis, A., Gangji, D., Juvent, M., Herchuelz, A., 1988. 2085–2095. The pharmacokinetics of oxybutynin in man. Eur. J. Clin. Pharmacol. 35, 515– Poulin, P., Kenny, J.R., Hop, C.E.C.A., Haddad, S., 2012. In vitro–in vivo extrapolation 520. of clearance: modeling hepatic metabolic clearance of highly bound drugs and European Medicines Agency, 2013. Guideline on the pharmacokinetic and clinical comparative assessment with existing calculation methods. J. Pharm. Sci. 101, evaluation of modified release dosage forms . Prueksaritanont, T., Qiu, Y., Mu, L., Michel, K., Brunner, J., Richards, K., Lin, J., 2005. Fallingborg, J., Christensen, L.A., Ingeman-Nielsen, M., Jacobsen, B.A., Abildgaard, K., Interconversion pharmacokinetics of simvastatin and its hydroxy acid in dogs: Rasmussen, H.H., 1989. PH-profile and regional transit times of the normal gut effects of gemfibrozil. Pharm. Res. 22, 1101–1109. measured by a radiotelemetry device. Aliment. Pharmacol. Ther. 3, 605–614. Roberts, M., Rowland, M., 1986. A dispersion model of hepatic elimination: 1. Fieller, E.C., 1954. Some problems in interval estimation. J. R. Stat. Soc. Ser, B 16, formulation of the model and bolus considerations. J. Pharmacokinet. Biopharm. 175–185. 14, 227–260. Gammans, R.E., Mayol, R.F., Labudde, J.A., 1986. Metabolism and disposition of Rowland, M., Benet, L., Graham, G., 1973. Clearance concepts in pharmacokinetics. J. buspirone. Am. J. Med. 80, 41–51. Pharmacokinet. Biopharm. 1, 123–136. Gammans, R.E., Mayol, R.F., MacKenthun, A.V., Sokya, L.F., 1985. The relationship Rowland Yeo, K., Jamei, M., Yang, J., Tucker, G.T., Rostami-Hodjegan, A., 2010. between buspirone bioavailability and dose in healthy subjects. Biopharm. Drug Physiologically based mechanistic modelling to predict complex drug–drug Dispos. 6, 139–145. interactions involving simultaneous competitive and time-dependent enzyme Gertz, M., Harrison, A., Houston, J.B., Galetin, A., 2010. Prediction of human inhibition by parent compound and its metabolite in both liver and gut—the intestinal first-pass metabolism of 25 CYP3A substrates from in vitro clearance effect of diltiazem on the time-course of exposure to triazolam. Eur. J. Pharm. and permeability data. Drug Metab. Dispos. 38, 1147–1158. Sci. 39, 298–309. Gertz, M., Houston, J.B., Galetin, A., 2011. Physiologically based pharmacokinetic Rubinstein, A., 2005. Colonic drug delivery. Drug Discovery Today: Technol. 2, 33– modeling of intestinal first-pass metabolism of CYP3A substrates with high 37. intestinal extraction. Drug Metab. Dispos. 39, 1633–1642. Sakr, A., Andheria, M., 2001a. A comparative multidose pharmacokinetic study of Gupta, S.K., Sathyan, G., 1999. Pharmacokinetics of an oral once-a-day controlled- buspirone extended-release tablets with a reference immediate-release release oxybutynin formulation compared with immediate-release oxybutynin. product. J. Clin. Pharmacol. 41, 886–894. J. Clin. Pharmacol. 39, 289–296. Sakr, A., Andheria, M., 2001b. Pharmacokinetics of buspirone extended-release Gupta, S.K., Sathyan, G., Lindemulder, E.A., Ho, P.-L., Sheiner, L.B., Aarons, L., 1999. tablets: a single-dose study. J. Clin. Pharmacol. 41, 783–789. Quantitative characterization of therapeutic index: application of mixed-effects Sathyan, G., Chancellor, M.B., Gupta, S.K., 2001. Effect of OROSÒ controlled-release modeling to evaluate oxybutynin dose-efficacy and dose-side effect delivery on the pharmacokinetics and pharmacodynamics of oxybutynin relationships. Clin. Pharmacol. Ther. 65, 672–684. chloride. Br. J. Clin. Pharmacol. 52, 409–417. Hallifax, D., Foster, J.A., Houston, J.B., 2010. Prediction of human metabolic clearance Satoh, T., Taylor, P., Bosron, W.F., Sanghani, S.P., Hosokawa, M., Du, B.N.L., 2002. from in vitro systems: retrospective analysis and prospective view. Pharm. Res. Current progress on esterases: from molecular structure to function. Drug 27, 2150–2161. Metab. Dispos. 30, 488–493. Hallifax, D., Houston, J.B., 2012. Evaluation of hepatic clearance prediction using Schiller, C., Frohlich, C.P., Giessmann, T., Siegmund, W., Monnikes, H., Hosten, N., in vitro data: emphasis on fraction unbound in plasma and drug ionisation Weitschies, W., 2005. Intestinal fluid volumes and transit of dosage forms as using a database of 107 drugs. J. Pharm. Sci. 101, 2645–2652. assessed by magnetic resonance imaging. Aliment. Pharmacol. Ther. 22, 971– Heikkinen, A.T., Baneyx, G., Caruso, A., Parrott, N., 2012. Application of PBPK 979. modeling to predict human intestinal metabolism of CYP3A substrates – an Shibata, Y., Takahashi, H., Chiba, M., Ishii, Y., 2002. Prediction of hepatic clearance evaluation and case study using GastroPlus. Eur. J. Pharm. Sci. 47, 375–386. and availability by cryopreserved human hepatocytes: an application of serum Jamei, M., Dickinson, G.L., Rostami-Hodjegan, A., 2009a. A framework for assessing incubation method. Drug Metab. Dispos. 30, 892–896. inter-individual variability in pharmacokinetics using virtual human Sinha, V.K., Snoeys, J., Osselaer, N.V., Peer, A.V., Mackie, C., Heald, D., 2012. From populations and integrating general knowledge of physical chemistry, biology, preclinical to human – prediction of oral absorption and drug–drug interaction anatomy, physiology and genetics: a tale of ‘bottom-up’ vs ‘top-down’ potential using physiologically based pharmacokinetic (PBPK) modeling recognition of covariates. Drug Metab. Pharmacokinet. 24, 53–75. approach in an industrial setting: a workflow by using case example. Jamei, M., Marciniak, S., Feng, K., Barnett, A., Tucker, G., Rostami-Hodjegan, A., Biopharm. Drug Dispos. 33, 111–121. 2009b. The SimcypÒ population-based ADME simulator. Expert Opin. Drug Sinko, P.J., Leesman, G.D., Amidon, G.L., 1991. Predicting fraction dose absorbed in Metab. Toxicol. 5, 211–223. humans using a macroscopic mass balance approach. Pharm. Res. 8, 979–988. Jamei, M., Turner, D., Yang, J., Neuhoff, S., Polak, S., Rostami-Hodjegan, A., Tucker, G., Sjogren, E., Abrahamsson, B., Augustijns, P., Becker, D., Bolger, M.B., Brewster, M., 2009c. Population-based mechanistic prediction of oral drug absorption. AAPS J. Brouwers, J., Flanagan, T., Harwood, M., Heinen, C., Holm, R., Juretschke, H.P., 11, 225–237. Kubbinga, M., Lindahl, A., Lukacova, V., Munster, U., Neuhoff, S., Nguyen, M.A., Jang, S.B., Lee, Y.J., Lim, L.A., Park, K.M., Kwon, B.J., Woo, J.S., Kim, Y.I., Park, M.S., Kim, Peer, A., Reppas, C., Hodjegan, A.R., Tannergren, C., Weitschies, W., Wilson, C., K.H., Park, K., 2010. Pharmacokinetic comparison of controlled-release and Zane, P., Lennernas, H., Langguth, P., 2014. In vivo methods for drug absorption immediate-release oral formulations of simvastatin in healthy Korean subjects: – comparative physiologies, model selection, correlations with in vitro methods a randomized, open-label, parallel-group, single- and multiple-dose study. Clin. (IVIVC), and applications for formulation/API/excipient characterization Ther. 32, 206–216. including food effects. Eur. J. Pharm. Sci. 57, 99–151. Langer, R., 1990. New methods of drug delivery. Science 249, 1527–1533. Sun, D., Lennernas, H., Welage, L., Barnett, J., Landowski, C., Foster, D., Fleisher, D., Lennernas, H., 2007. Intestinal permeability and its relevance for absorption and Lee, K.-D., Amidon, G., 2002. Comparison of human duodenum and caco-2 gene elimination. Xenobiotica 37, 1015–1051. expression profiles for 12,000 gene sequences tags and correlation with Lennernas, H., 2014a. Human in vivo regional intestinal permeability: importance permeability of 26 drugs. Pharm. Res. 19, 1400–1416. for pharmaceutical drug development. Mol. Pharm. 11, 12–23. Sutton, S., 2009. Role of physiological intestinal water in oral absorption. AAPS J. 11, Lennernas, H., 2014b. Regional intestinal drug permeation: biopharmaceutics and 277–285. drug development. Eur. J. Pharm. Sci. 57, 333–341. Tannergren, C., Bergendal, A., Lennernas, H., Abrahamsson, B., 2009. Toward an Lennernäs, H., Ahrenstedt, Ö., Hällgren, R., Knutson, L., Ryde, M., Paalzow, L., 1992. increased understanding of the barriers to colonic drug absorption in humans: Regional jejunal perfusion, a new in vivo approach to study oral drug absorption implications for early controlled release candidate assessment. Mol. Pharm. 6, in man. Pharm. Res. 9, 1243–1251. 60–73. Lin, J.H., Chiba, M., Baillie, T.A., 1999. Is the role of the small intestine in first-pass Thombre, A.G., 2005. Assessment of the feasibility of oral controlled release in an metabolism overemphasized? Pharmacol. Rev. 51, 135–158. exploratory development setting. Drug Discovery Today 10, 1159–1166. Lukkari, E., Taavitsainen, P., Juhakoskr, A., Pelkonen, O., 1998. Cytochrome P450 Troutman, M., Thakker, D., 2003. Efflux ratio cannot assess P-glycoprotein-mediated specificity of metabolism and interactions of oxybutynin in human liver attenuation of absorptive transport: asymmetric effect of P-glycoprotein on microsomes. Pharmacol. Toxicol. 82, 161–166. absorptive and secretory transport across caco-2 cell monolayers. Pharm. Res. Martinez, M., Amidon, G., 2002. A mechanistic approach to understanding the 20, 1200–1209. factors affecting drug absorption: a review of fundamentals. J. Clin. Pharmacol. Tubic-Grozdanis, M., Hilfinger, J.M., Amidon, G.L., Kim, J.S., Kijek, P., Staubach, P., 42, 620–643. Langguth, P., 2008. Pharmacokinetics of the CYP 3A substrate simvastatin 44 A. Olivares-Morales et al. / European Journal of Pharmaceutical Sciences 67 (2015) 32–44

following administration of delayed versus immediate release oral dosage Yang, J., Jamei, M., Yeo, K.R., Tucker, G.T., Rostami-Hodjegan, A., 2007. Prediction of forms. Pharm. Res. 25, 1591–1600. intestinal first-pass drug metabolism. Curr. Drug Metab. 8, 676–684. U.S. Food and Drug Administration, 2000. Guidance for industry: waiver of in vivo Yu, L.X., Crison, J.R., Amidon, G.L., 1996. Compartmental transit and dispersion bioavailability and bioequivalence studies for immediate-release solid oral model analysis of small intestinal transit flow in humans. Int. J. Pharm. 140, dosage forms based on a biopharmaceutics classification system. . Zhang, Q.-Y., Dunbar, D., Ostrowska, A., Zeisloft, S., Yang, J., Kaminsky, L.S., 1999. Watts, P.J., Lllum, L., 1997. Colonic drug delivery. Drug Dev. Ind. Pharm. 23, 893– Characterization of human small intestinal cytochromes P-450. Drug Metab. 913. Dispos. 27, 804–809. Wilding, I.R., 1999. Evolution of the biopharmaceutics classification system (BCS) to Zhang, X., Lionberger, R.A., 2014. Modeling and simulation of biopharmaceutical oral modified release (MR) formulations; what do we need to consider? Eur. J. performance. Clin. Pharmacol. Ther. 95, 480–482. Pharm. Sci. 8, 157–159. Zhu, M., Zhao, W., Jimenez, H., Zhang, D., Yeola, S., Dai, R., Vachharajani, N., Mitroka, Yaich, M., Popon, M., Medard, Y., Aigrain, E.J., 1998. In-vitro cytochrome P450 J., 2005. Cytochrome P450 3A-mediated metabolism of buspirone in human dependent metabolism of oxybutynin to N-deethyloxybutynin in humans. liver microsomes. Drug Metab. Dispos. 33, 500–507. Pharmacogenetics 8, 449–451.

Chapter 5: Translating Human Effective Jejunal Intestinal Permeability to Surface- Dependent Intrinsic Permeability: a Pragmatic Method for a More Mechanistic Prediction of Regional Oral Drug Absorption

AAPS J. 17(5) (2015):1177-1192

Andrés Olivares Morales, Hans Lennernäs, Leon Aarons and Amin Rostami Hodjegan

The AAPS Journal, Vol. 17, No. 5, September 2015 ( # 2015) DOI: 10.1208/s12248-015-9758-0

Research Article

Translating Human Effective Jejunal Intestinal Permeability to Surface-Dependent Intrinsic Permeability: a Pragmatic Method for a More Mechanistic Prediction of Regional Oral Drug Absorption

Andrés Olivares-Morales,1 Hans Lennernäs,2 Leon Aarons,1 and Amin Rostami-Hodjegan1,3,4

Received 27 January 2015; accepted 20 March 2015; published online 19 May 2015

Abstract. Regional intestinal effective permeability (Peff) values are key for the understanding of drug absorption along the whole length of the human gastrointestinal (GI) tract. The distal regions of the GI tract (i.e. ileum, ascending-transverse colon) represent the main sites for GI absorption when there is incomplete absorption in the upper GI tract, e.g. for modiﬁed release formulations. In this work, a new and pragmatic method for the estimation of (passive) intestinal permeability in the different intestinal regions is being proposed, by translating the observed differences in the available mucosal surface area along the human GI

tract into corrections of the historical determined jejunal Peff values. These new intestinal Peff values or Bintrinsic^ Peff(Peff,int) were subsequently employed for the prediction of the ileal absorption clearance (CLabs,ileum) for a set of structurally diverse compounds. Additionally, the method was combined with a semi- mechanistic absorption PBPK model for the prediction of the fraction absorbed (fabs). The results showed that Peff,int can successfully be employed for the prediction of the ileal CLabs and the fabs. Peff,int also showed to be a robust predictor of the fabs when the colonic absorption was allowed in the PBPK model, reducing the overprediction of fabs observed for lowly permeable compounds when using the historical Peff values. Due to its simplicity, this approach provides a useful alternative for the bottom-up prediction of GI drug absorption, especially when the distal GI tract plays a crucial role for a drug’s GI absorption. KEY WORDS: intestinal permeability; oral drug absorption; physiologically based pharmacokinetic modelling.

INTRODUCTION absorption process. Due to the established correlation

between human effective jejunal permeability (Peff) and the Human intestinal permeability is a key drug-related fraction of the administered oral dose that becomes absorbed parameter governing the gastrointestinal (GI) drug in the GI tract (fabs or fa), Peff has been widely used for the a priori estimation of drug absorption in different stages of the

Electronic supplementary material The online version of this article drug development process (1–3). Peff is one of the fundamen- (doi:10.1208/s12248-015-9758-0) contains supplementary material, tal biopharmaceutical parameters in the Biopharmaceutics which is available to authorized users. Classification System (BCS) and has been extensively used in 1 Centre for Applied Pharmacokinetic Research, Manchester Phar- drug development (2,4,5). As a consequence, it has often macy School, The University of Manchester, Stopford Building, been applied in mechanistic GI absorption models, i.e. Oxford Road, Manchester, M13 9PT, UK. ™ 2 GastroPlus (6), PK-Sim® (7,8), SimCYP® (9) and some in- Department of Pharmacy, Uppsala University, BOX 580, Uppsala, house absorption models (10,11), for the Bbottom-up^ predic- Sweden. 3 Certara, Blades Enterprise Centre, Sheffield, UK. tion of the rate and extent of oral drug absorption. 4 To whom correspondence should be addressed. (e-mail: Human Peff is usually determined in single-pass perfu- [email protected]) sion experiments using open, semi-open or double-balloon ABBREVIATIONS: BCS, Biopharmaceutical Classification System; (closed) perfusion systems (12), where the latter is considered fl CLabs, Absorption clearance; DJF, Duodeno-jejunal exure; fabs, to be the most accurate in vivo method for Peff determina- Fraction of the dose absorbed in the gastrointestinal tract; GI, tions in humans (3,12). In the double-balloon technique, the Gastrointestinal; ICV, Ileocecal valve; LIG, Loc-I-Gut; mSA, Muco- drug is single-pass perfused in solution to conscious human sal surface area; mSAT, minimal segmented absorption and transit volunteers employing a multilumen tube. The reader is model; MV, Small-intestinal microvilli; MVcolon, Colonic microvilli; referred to the following references for more details about PBPK, Physiologically based pharmacokinetic; PC, Plicae circularis or the clinical procedure, the data analysis and the double- circular folds; Peff, Jejunal intestinal effective permeability; Peff,int, – Intrinsic intestinal effective permeability; P , Intestinal membrane balloon technique (1,2,13 16). m fl permeability; SA, Surface area; SAEF, Surface area expansion factor; The derived Peff values are a re ection of the intestinal SI, Small intestine; SITT, Small-intestinal transit time; UWL, membrane’s resistance to the passage of the drug across the Unstirred water layer; VL, Intestinal villi apical membrane of the GI tract, regardless of the

1177 1550-7416/15/0500-1177/0 # 2015 American Association of Pharmaceutical Scientists 1178 Olivares-Morales et al.

mechanisms involved, i.e. unstirred water layer (UWL) human intestine (36,37) into the human Peff values and to permeation, transcellular permeation, paracellular perme- evaluate the ability of the new permeability values to predict the ation and/or carried-mediated absorption/efflux (2,17–19). In GI absorption using a semi-mechanistic PBPK absorption model. addition, jejunal Peff can be affected by the experimental conditions and the physiological factors that the drug might MATERIALS AND METHODS encounter during its passage through the intestinal membrane in the upper jejunum (3,12,20). Determination of the Mucosal Surface Area in the Small Intestinal Peff is a time- and position-dependent parameter, Intestine and the Colon and its magnitude is a function of the nature of the small-intestinal membrane and the luminal conditions (1–3,18,21).Thelatterisof The human intestine has several structures known to special interest in the cases where jejunal Peff alone might not increase the mucosal surface area (mSA) compared to that of necessarily explain the drug absorption in certain regions of the a cylinder. In the small intestine (SI), the circular folds or plicae human GI tract. This is the case of the distal ileum and/or the colon, circularis (PC), villi (VL) and enterocyte microvilli (MV) are where the luminal conditions and the structure of the intestinal commonly believed to increase the surface area by factors of 3, membrane differ significantly with those of the upper jejunum 10, and 20, respectively (23,37,38). In the colon, the colonic (22,23). These distal intestinal sites play a key role in the absorption haustra, the crypts and the colonic microvilli (MVcolon)perform of drugs contained in modified release (MR) dosage forms, where a similar function (22). It has been shown, however, that the due to the nature of the formulation and the prolonged colonic amplifications in mSA generated by the circular folds and villi residence time, a great proportion of the given dose is designed to decrease from the proximal SI to the distal SI. In addition, a be absorbed in such regions (22,24). Regional permeability decrease in mSA has been observed from the colonic microvilli measurements are considered as an important parameter for the compared to that of the small intestine (36,39). understanding and the mechanistic modelling of GI drug absorp- For this work, three different methods for the estimation tion. Nevertheless, the available regional Peff estimates are sparse of the intestinal surface area were employed as follows: (a) and limited to a small set of compounds. Given that such values method 1 (M1) or control assumes that the mSA of the were determined mostly using open perfusion systems, their intestine is that of a cylinder; (b) method 2 (M2) assumes that comparison with the ones obtained from the double-balloon the mSA of the SI decreases exponentially, as per the work of technique is limited (4,5). Recent works have highlighted the Wilson (36); and (c) method 3 (M3) assumes that the mSA of necessity for regional Peff estimates in humans, which are of the SI decreases gradually, based on recently published particular interest during early stages of drug and formulation region-specific surface area expansion factors (SAEFs) (39). development (4,5,21,25). Yet, due to the vast economic cost A detailed description of the three methods is given below, associated with clinical perfusion studies, further investigations on and they are illustrated in Fig. 1a. regional Peff determination might be limited in the near future. Mechanistic absorption models can be employed as a M1: Cylindrical Surface Area (Control) complement for the understanding of drug permeation along the different segments of the GI tract. For instance, the majority This method assumed that the surface area (SA) in all of the aforementioned models can incorporate some of the the intestinal segments was that of a cylinder (Fig. 1a), π physiological factors known to affect the drug’s regional absorp- SAi=2× ×Ri×Li, where Ri and Li are the radii and lengths of tion. Remarkable progress has been made in the field of solubility the given intestinal segments (i), respectively. The anatomical and dissolution, where factors such as the pH-dependent parameters, i.e. lengths and radii, of the different intestinal solubility for ionisable compounds, variable GI fluid volumes, segments were derived from the literature and were meant to B ^ supersaturation and precipitation, presence of bile micelles and represent the values of a reference man of 70 kg of weight bile salt-mediated solubility enhancement, to name a few, have and 170 cm of height (23,40). The anatomical length of the small intestine (L )was already been incorporated in these models (6,7,26–30). Never- si assumed as 670.7 cm (41), and the length of each small intestinal theless, in terms of regional intestinal membrane permeability segment was calculated as a percentage of L ; 8, 37 and 55% for (once the intraluminal and intracellular processes have been si the duodenum, jejunum and ileum, respectively (23). The radius accounted for), there is a need for improvements (6–8,26,28). of the duodenum (r )wasassumedtobe2.37cm(23), whereas While in most of these mechanistic models, regional differ- duo for the jejunum and the ileum, the radii were assumed to be 1.75 ences in the expression/abundance of intestinal transporters are and 1.5 cm, respectively (5,15). With regard to the colon, the already accounted to some extent (6–8,26,28), the approach with total length (L ) was assumed to be 104.34 cm, and it was regard to the passive permeation along the GI tract is still not well col calculated from Eq. 1 for the same reference man (40). defined. For example, in the physiologically based pharmacokinetic (PBPK) modelling of orally administered drugs, it is a common practice to assume that colonic absorption is insignificant compared Lcol ¼ 0:52 Â height þ 18:5cm ð1Þ to that in the small intestine. Thus, colonic absorption, by default, is not allowed in such models (10,11,31–34).However,basedonthe For this work, colonic absorption was assumed only to large number of MR dosage forms labelled for once-daily occur in the ascending colon (acol) and the length of the administration, absorption from the colon appears to be a very segment (Lacol) was considered to be a 16% of Lcol (22,23). crucial and common process (22,35). The radius of the ascending colon was assumed to be 2.42 cm The aim of this study was to derive regional intestinal (42). The aforementioned anatomical parameters were also permeability estimates by translating the observed differences in employed for the calculation of the mucosal surface area in intestinal mucosal surface area along the different segments of the M2 and M3. Surface Area, Permeability and Oral Drug Absorption 1179

Fig. 1. Illustrations of the changes on intestinal mucosal surface area estimated by the different methods. a Changes in the surface area expansion factors (SAEFs) along the length of the small intestine (as a percentage), the SAEF is deﬁned as the ratio between the surface area estimated by a given method and the surface area of a cylinder. b Total mucosal surface area estimated from the different methods (M1 to M3), whereas CSAEF is the total mucosal surface area estimated by applying a constant SAEF of 600-fold along the small intestine. c An illustration of the total surface area estimated for each of the regions of the small intestine and the ascending colon, for the different methods (M1 to M3) and when using the CSAEF. The values were calculated assuming a reference human intestine (see main text for details on the anatomical values employed for the calculations)

M2: Exponential Decrease (Wilson) respectively. More details about the fitting process can be found in Section 1 of the Supplementary Material. This method was based on the work by Wilson (1967), Equation 2 was integrated with respect to the longitudi- who reported a correlation between the mSA and the nal position (xi+j) between the DJF and the ICV in order to longitudinal position between the beginning of the jejunum derive the cumulative mSA, as shown in Eq. 3 fl fi (duodenojejunal exure, DJF) and the nal portion of the þ Zx j i λ 2 ÀÁ ÀÁ λ x jþi ileum (ileocecal valve, ICV) (36). This method accounts for 1 L þ mSA jþi x jþi ¼ yxjþi dx jþi ¼À 1 À e j i ð3Þ the amplification in SA product of the presence of circular λ2 0 folds and villi along the SI (Fig. 1a). Herein, Wilson’s data was digitized using GetData Graph Digitizer v2.26 (http:// where mSAj+i is the cumulative mucosal surface area at getdata-graph-digitizer.com/) and fitted by an exponential any position of the aforementioned segment. To obtain model (Eq. 2) using non-linear least squares regression with thetotalmucosalsurfaceareainthewholeSI,asimilar the Blsqnonlin^ function of the Optimization Toolbox within relationship for the duodenum was needed; however, no ’ Matlab 2014a (The Mathworks Inc., Natick, MA, USA) data for such a correlation was provided in Wilson s work. This relationship was assumed to be proportional to the ÀÁ λ longitudinal position in the duodenal segment (xduo)and 2 þ L þ x j i yxjþi ¼ λ1e j i ð2Þ similar to that of the initial portion of the jejunum, i.e. the intercept (λ ) of the exponential relationship generat- 2 1 where, y(xj+i) is the ratio of mSA per unit of serosal length (cm / ed for the jejunum and ileum (Eq. 4). cm), xj+i is the longitudinal position (cm) along the segment comprised between the DJF and the ICV and Lj+i is the length mSAduoðÞ¼xduo λ1xduo ð4Þ (cm) of the jejunum plus the ileum (j+i) (0.92×Lsi). The 2 regression coefficients, λ1and λ2, were found to be 164.16 cm / cm and −3.33 respectively. The coefficients of variation (CV) of Combining Eqs. 3 and 4 and accounting for the mSA the parameter estimates were found to be 0.65 and 35.03%, expansion due to the intestinal microvilli (MVsi) of 20-fold 1180 Olivares-Morales et al.

(23,43), the cumulative mucosal surface area at any longitu- (44–49). This was done in order to allow the incorporation of the dinal position (x) of the SI is given by Eq. 5. different mSA calculation methods for the estimation of the 8 regional Peff values; the details of the re-calculation procedure λ ≤ : > 1MVsixif x 0 08Lsi > − : can be found in Section 3 the Supplementary Material. > : x 0 08Lsi < 0 92 λ2 : λ MV L 0:08− 1−e 0 92Lsi if L > x > 0:08L ðÞ¼ 1 si si λ si si For the re-calculation, the value of the SA term in Eq. 6 mSAsi x > 2 > 0:92 ÀÁ depended upon which of the aforementioned mSA estimation > λ2 : λ1MVsiLsi 0:08− 1−e if x > Lsi λ2 methods were used. For M1, the mSA was calculated as ð5Þ described above for a cylinder of radius (ri) and length (Li), where ri was assumed either 1.75 cm (upper jejunum) or For the ascending colon, the only structure considered to 1.5 cm (distal ileum) and Li was the length of the test portion effectively increase the mSA available for absorption was the of the perfusion tube (cm). For M2, the mSA available colonic microvilli (MVcolon); its increase in colonic mSA was (Eq. 5) across the length of the test portion of the multiple assumed to be 6.4-fold (39). lumen tube was employed (44–49). This mSA was defined as the difference in cumulative mucosal surface area between M3: Gradual Decrease (Helander and Fändriks, H&F) the distal (xfinal,i) and proximal (xinitial,i) collection orifices of the perfusion tube. The positions required to calculate the This method was based on a recent work by Helander and mSA according to M2, xfinal,i and xinitial,i, were assumed to be Fändriks (2014) where they reported mean SA expansion the same for all the experiments and were representative of factors generated by the different structures present in the SI the positions informed for the jejunal and ileal perfusion and the colon (Fig. 1a)(39). The segmental cylindrical SA was experiments (44–49). This was done due to the high calculated as per M1 and for each intestinal segment, the variability reported in the positioning of the test segments of following SAEF were applied: (a) circular folds (for every the multiple lumen tube or the lack of information about it in segment in the SI), 1.57-fold; (b) duodenal villi and microvilli, some of the references. For the jejunal experiments, xinitial, 6.5- and 9.2-fold, respectively; (c) jejunal villi and microvilli, 8.6- jejunum was assumed to be 10 cm distally from the DJF, and 14.1-fold, respectively; (d) ileal villi and microvilli, 4.5- and whereas for the ileal experiments, xinitial, ileum, it was assumed 15.7-fold, respectively, and (e) colonic microvilli, 6.4-fold (39). to be 110 cm from the same intestinal landmark. The The total intestinal mSA calculated for each of the positions of the distal portions, xfinal, were estimated by methods described above was contrasted with the mSA adding the lengths of the test segments (Li) to both jejunal fi resulting from the use of the classical ampli cation factors and ileal initial positions (xfinal, jejunum=xinitial, jejunum+Li and described in the literature; 3-fold for the circular folds, 10-fold xfinal, ileum=xinitial, ileum+Li). Finally, for M3, the SA term was for the villi and 20-fold for the intestinal microvilli (22,23,37) calculated using the cylindrical SA described above for M1 (Fig. 1b). and the SAEF described in M3 for the jejunum and the ileum. The detailed regional estimates of the mSA derived from the Analysis of the Differences in Regional Intestinal aforementioned calculations methods are shown Table I. Permeability For the analysis, Peff,i values derived using the SA according to M1 were considered as the control values, whereas A comparison between the reported human regional Peff for the other two methods regional Peff values were considered values was performed in order to elucidate the possible as Bintrinsic^ values, Peff,int(i,k), where the subscript i stands for impact that the mSA assumptions might have in the regional the region (jejunum or distal ileum) and the subscript k refers to Peff determinations. This was done for a recently published the mSA calculation method. To support the hypothesis of the dataset of 11 drugs whose in vivo absorption was measured in existence of a relatively similar Peff,int along the membrane of different segments of the human SI during single-pass open the GI tract, regional Peff values were contrasted in order to perfusion experiments (4,5). Regional Peff values were identify possible similarities between Peff along the SI, i.e. originally derived by Lennernäs by re-arranging the reported jejunal vs ileal. It is worth noting that Peff,int is not intended to absorption data to a parallel tube model (4,5,12,18). Selection represent pure membrane permeability (Pm)(50); instead, as for of this model was based on its ability to describe the Peff, Peff,int is a lumped parameter involving several permeation hydrodynamics of the absorption process during open and/ mechanisms (2,17–19). or semi-open single-pass perfusion experiments (12,18), as The evaluation of the agreement was done by visual shown in Eq. 6 inspection, i.e. comparing the plots between the regional Peff values for each method, and by calculation of error metrics Qin Cout;i such as: the average fold error (afe), for measuring bias Peff;i ¼ ln ð6Þ SAi Cin;i (over/underprediction); the absolute average fold error (aafe), for measuring absolute spread between the values, and the concordance correlation coefficient (ccc), as a where the subscript i in Peff,i stands for either the upper jejunum measure of their agreement. The metrics were calculated as or distal ileum, Qin is the perfusate flow rate (volume/time), Cin described in a previous report (51), where a good agreement and Cout are the respective water transport-corrected concen- between Peff values was considered when both the afe and trations of drug entering and leaving the test segment and SAi is aafe were within the twofold error, i.e. afe between 0.5 and 2, the segment’s surface area. and aafe less than 2. For the ccc, a moderate equivalence (ccc Herein, the reported regional Peff,i values were re-calculated greater than 0.90 (52)) was considered as good agreement from the absorption data originally reported in the literature between regional Peff values. Surface Area, Permeability and Oral Drug Absorption 1181

Table I. Segmental SA Employed for the Calculation of Peff,int from Peff Values (Open Perfusion System)

mSA jejunum mSA ileum Ratio 4 2 c 4 2 e Segment (×10 cm ) (×10 cm ) (mSAjej/mSAile) a b d Length xi, jejunum xi, ileum Compounda (cm) (cm) M1 M2 M3(cm) M1 M2 M3 M1 M2 M3

Hydrocortisone 15 10 0.0165 4.52 3.14 110 0.0141 2.89 1.57 1.26 1.56 2.00 Triamcinolone 15 10 0.0165 4.52 3.14 110 0.0141 2.89 1.57 1.26 1.56 2.00 acetonide Paracetamol 30 10 0.0330 8.71 6.28 110 0.0283 5.57 3.14 1.26 1.56 2.00 Salicylic acid 80 10 0.0880 20.6 16.7 110 0.0754 13.2 8.36 1.26 1.56 2.00 Hydrochlorothiazide 80 10 0.0880 20.6 16.7 110 0.0754 13.2 8.36 1.26 1.56 2.00 Atenonol 80 10 0.0880 20.6 16.7 110 0.0754 13.2 8.36 1.26 1.56 2.00 Furosemide 80 10 0.0880 20.6 16.7 110 0.0754 13.2 8.36 1.26 1.56 2.00 Cimetidine 80 10 0.0880 20.6 16.7 110 0.0754 13.2 8.36 1.26 1.56 2.00 Talinonol 30 10 0.0330 8.71 6.28 110 0.0283 5.57 3.14 1.26 1.56 2.00 Griseofulvine 20 10 0.0220 5.95 4.19 110 0.0188 3.81 2.09 1.26 1.56 2.00 Ranitidine 30 10 0.0330 8.71 6.28 110 0.0283 5.57 3.14 1.26 1.56 2.00 a Lengths of the test segments during open perfusion experiments derived from references (4,5,44–49) b,e xi, jejunum and xi, ileum, are the assumed positions (with respect to the DJF) of the proximal portion of the test segment of the tube employed during the single-pass open perfusion studies; this position was needed to calculate the mSA according to M2 (Eq. 5) c,e For the estimation of the mSA, the jejunal and ileal radii were assumed as 1.75 and 1.5 cm, respectively, as per references (4,5)

The Use of Jejunal Peff,int(k) for the Prediction of Distal Development of a Semi Mechanistic-PBPK Model Absorption for the Prediction of the Regional Fraction Absorbed

In the same line as described above, the ability to predict To further explore the possibility of using Peff,int(k) for the observed ileal absorption clearance (CLabs, ileal(observed)) the predictions of the regional drug absorption, a semi- using the aforementioned jejunal Peff, jejunum(k) values (de- mechanistic PBPK model was developed. The new model, rived from single-pass perfusion experiments) combined with or minimal segmented absorption and transit (mSAT) model, ileal mSA data was also evaluated. The observed absorption is based on the compartmental absorption and transit (CAT) clearance (CLabs, ileal(observed)) was derived from the absorp- model developed by Yu and co-workers (31–33). However, tion data reported in the original references of the single-pass the mSAT model describes the human GI tract by means of perfusion experiments (44–49). This was calculated as per five physiologically defined GI compartments: stomach, Eq. 7 duodenum, jejunum, ileum and ascending colon, as shown in Figure S2 of the Supplementary material. These five com- ¼ − Cout;ileum ð Þ partments are the minimal number of compartments required CLabs;ileumðÞ observed Qin;ileumln 7 Cin;ileum to represent the GI tract, considering all the small-intestinal segments involved in drug absorption following oral administration: hence the terms Bminimal^ and Bsegmented^. fl where Qin is the perfusate ow rate (volume/time) and Several assumptions were made for the estimation of GI Cin,ileum and Cout,ileum are the water transport-corrected drug absorption following oral dosing when using the mSAT concentrations of drug entering and leaving the ileal model in this study, including: (a) no significant absorption segment. The detailed calculation method is shown in can occur from the stomach (33); (b) the drug is emptied from Section 3 of the Supplementary Material. The predicted the stomach by means of a first-order process (53–55); (c) ileal absorption clearance (CLabs, ileum(predicted)) was estimated drug release and dissolution are instantaneous (unless stated according to Eq. 8 differently), and—as in the first development of the CAT model—no precipitation and/or changes in solubility due to CL ; ðÞ¼ P ; ðÞ; Â mSA ðÞ ð8Þ abs ileum predicted eff int jejunum k ileum k differences in luminal pH or bile salts were included in this version of the mSAT model (31–33); (d) all the GI compartments are well mixed, and the drug concentration is where Peff,int (jejunum, k) is the jejunal Peff,int, estimated by any of the aforementioned mSA calculation methods (k), homogenous within each GI compartment (31–33); (e) the and mSAileum(k) is the corresponding ileal mSA. The lengths of each of the intestinal compartment are represen- surface area values employed for such calculation are tative of their anatomy (23,40); (f) drug transit between the showninTableI. small-intestinal compartments is related to the segment’s The predictions were evaluated, in terms or accuracy and anatomical length (23,40) and the mean small-intestinal precision, by visual inspection and by using the same error transit time (SITT) (31,32,56); (g) drug absorption occurs metrics, and criteria for success, as the one described above only by means of non-saturable processes, where only the for the comparison between Peff values. dissolved drug can be absorbed (2); (h) there is no drug 1182 Olivares-Morales et al. degradation in the GI lumen and (i) the colonic transit is model to describe the SITT data, a comparison was per- described as a first-order process (57,58). It is clear that these formed by using different transit models, e.g. three compart- assumptions are valid to make when the regional differences ment first order, CAT etc. More details about the fitting and in permeation is the main biopharmaceutical parameter the comparison with the alternative models can be found in affecting both rate and extend of drug absorption. the Section 4 of the Supplementary Material. The mSAT model was implemented in Matlab 2014a as a Finally, the segment-specific absorption rate constant system of 14 ordinary differential equations (9 in the case of (ka,n), necessary to describe the passage of the dissolved drug assuming complete and instantaneous release/dissolution) to from the intestinal lumen across the intestinal wall, is shown describe the drug transit and absorption in each of the GI in Eq. 11 segments. Mass balance equations were implemented in order mSAnðÞk ka;nkðÞ¼ Peff;intðÞk Â DFÂ ð11Þ to estimate the regional and overall fabs. The system was Vn evaluated using Matlab’s ode15s solver for stiff differential equations. The most relevant details of the mSAT model are where, Peff,int(k) is the intrinsic effective permeability, deter- given below, and the detailed differential equations can be mined with the double-balloon technique and calculated as ’ found in the Appendix 1. described below, mSAn(k) is the segment s mSA, determined The drugs used in the modelling and simulation (M&S) with any of the methods described previously (k), Vn is the ’ fl were assumed to be administered as solution. Within the luminal segment s cylindrical volume and DF is degree of atness fi compartments of the mSAT model, the drug mass can be coef cient that accounts for the rather elliptical shape of the transferred along the segments with no possibility for precipita- intestine compared to that a full cylinder (1.7) (59). All the tion. If required for simulations purposes, the drug can be aforementioned system-related input parameters for the modelled in the solid form (undissolved /unreleased), where the mSAT model are summarized in Table S2 of the Supplemen- combined drug release and dissolution can be controlled by an tary Material. empirical first-order release constant, krel.Thefirst order gastric − 1 Prediction of the fabs Using the mSAT Model and Peff,int(k) emptying rate constant (kge) was assumed to be 4 h ,whichis the inverse of the reported mean gastric residence time for liquids (53), whereas for the ascending colon, the first-order rate The mSAT model was employed to evaluate the use of − 1 Peff,int (k) for the prediction of fabs, in particular when colonic constant (kcol)wasassumedtobe0.0667h (57). Contrary to the CAT model, the mSAT model describes absorption was considered. This approach also served as the SI by only three anatomically defined segments. Thus, in validation of the mSAT model for such predictions. The analysis order to optimally describe the drug’s transit along the SI, a was performed on 10 drugs whose Peff values were determined non-linear transfer between each of the small-intestinal using the double-balloon technique (Table II)(3,33). Peff,int (k) segments was assumed. The mass (A(t)) transfer from an was derived from Peff data, by multiplying it by the cylindrical intestinal segment to its adjacent compartment was described SA of the jejunal segment where the double-balloon segment by a Weibull function (Eq. 9) was positioned (SALoc-I-Gut), and then dividing it by the available jejunal mSA estimated by any of the methods (k)for β −ðÞt=αn the same segment. The length of the test segment (LLoc-I-Gut) AtðÞ¼A0 Â 1−e ð9Þ was assumed 10 cm and the jejunal radius (rjej,Loc-I-Gut)was assumed as 1.75 cm. Thus, SALoc-I-Gut was estimated as 109.96 cm2 (2×π×1.75 cm×10 cm), as shown in Eq. 12 where A0 is the initial amount of drug in said segment, β is the Weibull shape parameter and αn is a segment-dependent scale 2 Peff Â SALocÀIÀGut Peff Â 109:96cm parameter. The scale parameter was defined for the nth small- Peff;intðÞk ¼ ¼ ð12Þ mSALocÀIÀGutðÞk mSALocÀIÀGutðÞk intestinal segment as, an=fn×SITT×γ,wherefn is the fractional length of the intestinal segment (with respect to Lsi), SITT is the mean small-intestinal transit time (3.32 h) (32)andγ is a dimensionless coefficient. Selection of the Weibull function for Given the calculation method of mSA for the Loc-I-Gut this process was based on its flexibility and similarity to the segment according to M3, where SAEF are directly applied to exponential model when β=1. A representative differential the cylindrical SA, Eq. 12 can be reduced to the following equation describing the mass (A ) transfer between the small- n expression (Eq. 13): intestinal segments is shown in Eq. 10 Peff Â SALoc‐I‐Gut Peff;intðÞ M3 ¼ dAn mSA ‐ ‐ ðÞ ¼ k ; − ðÞt Â A − −k ; ðÞt Â A ; n ¼ 1; 2; 3 ð10Þ Loc I Gut M3 dt t n 1 n 1 t n n Â Â π Â Â ¼ Peff 2 r je j LLoc‐I‐Gut Â π Â Â Â β−1 2 r je j LLoc‐I‐Gut SAEFjejunum β t where kt;nðÞ¼t Â . αn αn ¼ Peff ð Þ fi β γ 13 The coef cients, and ,werefoundtobe2.01 SAEFjejunumðÞ M3 (CV=5.6%) and 1.57 (CV=2.90%), respectively. These were obtained by fitting the system described in Eq. 10 to the data on SITT’s distribution reported by Yu and co-workers (32). where rjej is the jejunal radius, LLoc-I-Gut is the length of the The fitting was done using the lsqnonlin function in Matlab Loc-I-Gut segment and SAEFjejunum is the combined SAEF 2014a. In addition, to evaluate the performance of the mSAT for the jejunal segment according to M3. Further details Surface Area, Permeability and Oral Drug Absorption 1183 about this relationship can be found in Section 6 of the The regional variation in intestinal mSA according to all Supplementary Material. the methods is illustrated in Fig. 1c. Despite being shorter The calculated Peff,int(k) values (Table II) were combined than the ileum, the jejunum displayed the largest mSA of all with the physiological parameters described previously in the small-intestinal segments. Depending on the method (M2 order to predict the overall and regional fabs using the mSAT or M3), its mSA was around 35 to 280% higher than that of model, where the main focus was to investigate the impact of the ileum, the second biggest contributor in mSA for the SI. allowing (ascending) colonic absorption for the fabs predic- The detailed segmental mSA for all the methods can be tions. Due to the importance of the colon for the absorption found in Table S1 of the Supplementary Material. of MR formulations, the investigation was performed assuming both the administration of a solution and a MR Similarities in Regional Intestinal Permeability formulation. The MR profile was simulated by assigning a −1 krel of 0.19 h to the first-order release/dissolution model, i.e. The estimated jejunal and ileal Peff,int values are 90% of the drug content is expected to be released and summarized in Table III. Due to differences in the mSA dissolved within 12 h of the administration of the dosage values, the magnitude of the Peff,int estimates varied between form. the estimation methods. However, for all the methods, only The fabs predictions were evaluated by visual inspection small differences between the jejunal and ileal Peff,int values and by using the same error metrics and criteria for success as were observed (Fig. 2a–c). This agreement was confirmed by the one described above for the prediction of the ileal CLabs. the correlation concordance coefficient between the regional In addition, the theoretical relationship between fabs and Peff Peff, int values as follows: 0.93, 0.92 and 0.88 for M1, M2 and was evaluated for a wide range of simulated Peff values in M3, respectively. The absolute average fold errors, on the order to validate the utility of the mSAT model for the other hand, varied between 1.60, 1.62 and 1.83, for M1, M2 prediction of the fabs from Peff data. and M3, respectively, whereas the average fold errors, varied between 0.95, 1.27 and 1.63, for M1, M2 and M3, respectively. RESULTS The latter suggest that, for all the methods, the overall differences between jejunal and ileal Peff or Peff,int seems to Intestinal Surface Area be consistent and within the twofold error. Yet, when using M2 and M3, the ileal Peff,int tended to be slightly higher than The results of the total mSA involved in GI absorption the jejunal counterpart, where the differences were more are illustrated in Fig. 1b.Forarepresentativehuman marked for highly permeable compounds (BCS classes 1 and intestine, the surface area available for oral absorption was 2), as shown in Fig. 2(b, c). approximately 0.73, 79 and 98 m2, for M1, M2 and M3, respectively. However, when the classical SA expansion Prediction of Ileal Absorption factors were applied to the cylindrical SA (600-fold), the available mSA was considerably greater (435 m2) than the The predictions of CL using jejunal P data and ones estimated by M2 and M3 (Fig. 1b). abs,ileal eff,int ileal mSA according to Eq. 8 are shown in Table III and Fig. 2d–f. There was good agreement between the observed and predicted CLabs,ileal values, especially for M1 and M2 Table II. Intestinal Permeability Values (Double-Balloon Technique) (Fig. 2e, f). The absolute average fold error between Employed for the Prediction of fabs Using the mSAT Model observed and predicted CLabs,ileal was less than twofold for all the methods (1.60, 1.62 and 1.83 for M1, M2 and M3, b c d BCS Peff,int (M1) Peff,int (M2) Peff,int (M3) − − − respectively). However, a general trend towards the Compound classa (×10 4 cm/s) (×10 4 cm/s) (×10 4 cm/s) underprediction of CLabs,ileal was observed when using M2 Enalaprilat 3 0.22 0.0008 0.0012 and M3 (Fig. 2e, f). This was reflectedinanafeof0.78and Furosemide 4 0.31 0.0010 0.0016 0.61, for M2 and M3, respectively. In addition, for all the Terbutaline 3 0.50 0.0017 0.0026 methods, the ccc was slightly less than when the Peff,int Atenolol 3 0.53 0.0018 0.0028 values were compared directly; the values varied between Metoprolol 1 1.50 0.0052 0.0079 0.88, 0.86 and 0.80, for M1, M2 and M3, respectively. In Propranolol 1 2.69 0.0093 0.014 thesamelineasfortheP comparisons, the differ- Fluvastatin 1 2.81 0.0096 0.015 eff,int ences between the observed and predicted values were Antipyrine 1 5.61 0.019 0.029 Naproxen 1 8.00 0.027 0.042 more marked for highly permeable compounds Ketoprofen 1 8.50 0.029 0.045 (Fig. 2e, f). This was particularly pronounced when using M3 for the predictions (Fig. 2f). Nevertheless, the overall a BCS classification extracted from reference (4) results suggest that jejunal Peff,int can be used for the b Peff values extracted from reference (33) ’ c prediction the CLabs,ileal, as long as the segment smSAis Peff,int values re-calculated from the data in the third column (Peff,int accounted for. (M1)) using Eq. 12 and the mSA estimated for the test segment used in double-balloon single-pass perfusion experiments; this value was 4 2 estimated as 3.20×10 cm , as per M2 (Eq. 5). The initial position of mSAT Model and fabs Predictions the test segment was assumed to be immediately after the DJF (14) d Peff,int values re-calculated from the data in the third column (Peff,int The performance of different transit models to describe the (M1)) using Eq. 12 and the mSA estimated for the test segment used in double-balloon single-pass perfusion experiments; this value was mean SITT data is shown Fig. 3. Both the CAT model (seven estimated as 2.09×104 cm2 , as per M3 transit compartment) and the mSAT model adequately 1184 Olivares-Morales et al.

Table III. Regional Peff,int Values (Open Perfusion System) and Prediction of the Ileal Absorption Clearance Estimated by Three Different Methods

b b d Peff,int (jejunum) Peff,int (ileum) Predicted ileal −4 −4 −4 3 (×10 cm/s) (×10 cm/s) CLabs (×10 cm /s) c BCS Observed ileal CLabs − Compound classa M1 M2 M3 M1 M2 M3(×10 4 cm3/s) M1 M2 M3

Hydrocortisone 2 8.79 0.032 0.046 5.57 0.027 0.050 787 1242 928 724 Triamcinolone acetonide 2 4.16 0.015 0.022 9.94 0.049 0.090 1405 588 439 343 Paracetamol 1 4.79 0.018 0.025 7.17 0.036 0.065 2028 1356 1012 790 Salicylic acid 1 2.67 0.011 0.014 4.12 0.024 0.037 3108 2010 1500 1171 Hydrochlorothiazide 4 0.18 0.00079 0.0010 0.16 0.00089 0.0014 117 139 104 81 Atenonol 3 0.38 0.0016 0.0020 0.27 0.0016 0.0025 206 283 211 165 Furosemide 4 0.48 0.0021 0.0025 0.22 0.0013 0.0020 166 365 272 213 Cimetidine 3 0.77 0.0033 0.0040 0.30 0.0017 0.0027 228 580 433 338 Talinonol 4 0.31 0.0012 0.0016 0.37 0.0019 0.0034 106 88 66 52 Griseofulvine 2 7.47 0.028 0.039 11.57 0.057 0.10 2181 1407 1051 820 Ranitidine 3 1.74 0.0066 0.0091 1.41 0.0072 0.013 400 492 367 287 a BCS classiﬁcation extracted from references (4,5) b Open perfusion regional permeability values were re-calculated from references (4,5,44–49) and the mSA values shown in Table I c Observed regional absorption clearances were re-calculated from references (4,5,44–49) employing a parallel tube model (Eq. 7); the details for the calculations can be found in the Section 2 of the Supplementary Material d Predicted values from Eq. 8, using jejunal Peff data derived by the open perfusion system and ileal mSA data from Table I described the mean SITT data. In contrast, the alternative three non-linear transfer function, in this case Weibull, in order to compartment transit model models showed to be inadequate to adequately describe the mean SITT data when using a different describe such data. This conﬁrmed the necessity of introducing a number of transit compartments (other than 7).

Fig. 2. Upper panel, comparison between jejunal Peff,int calculated by: a M1 (cylindrical SA), b M2 (mSA according to Wilson’s method) and c M3 (mSA according to Helander and Fändriks’ method). Lower panel, prediction of ileal absorption clearance (permeability clearance) employing jejunal Peff and ileal surface area using: d M1, e M2 and f M3. Solid light circles, BCS class 1; solid light squares, BCS class 2; solid dark upper triangles, BCS class 3; and solid dark lower triangles, BCS class 4. Black solid line represents the line of unity, whereas the dashed grey lines represent the twofold error Surface Area, Permeability and Oral Drug Absorption 1185

Fig. 3. Comparison between different small-intestinal transit models to the describe SITT data. The lines represent the cumulative percentage of the dose reaching the colon for the different SITT models. Red solid line, mSAT model (Weibull transfer between segments); dot-dashed cyan line, full CAT model (seven transit compartments); dashed blue line,CAT model with only three compartments (same ﬁrst-order transit rate constant for all the segments); dotted green line, CAT model with only three compartments, where the transit was fractionally divided for each segment (based on the segment’s length). The solid dots are the observed cumulative percentage of the dose reaching the colon, as per reference (32)

The fabs predictions using the mSAT model and Peff,int for the confirmed M3 as the most robust approach for fabs predictions, 10 drugs listed in Table II aresummarizedinFig.4 and Table IV. with and without allowing colonic absorption (Fig. 5). When colonic absorption was not allowed in the mSAT model, the predictions using Peff,int either calculated by M1 or M3 were in Simulated Colonic Contribution to the fabs good agreement with the observed values (Fig. 4a, c). This was supported by the absolute average fold error, correlation The simulated regional intestinal contribution to the concordance coefficient and average fold error as follows: 1.24, overall fabs is shown in Fig. 6a for a set of representative drugs 0.90 and 1.21, for M1, and 1.19, 0.95 and 1.11, for M3, respectively. (increasing in permeability): enlaprilat, metoprolol and When using Peff,int calculated by M2, however, a general trend ketoprofen. All the drugs listed in Table II are shown towards the underprediction of the fabs was observed (Fig. 4b); the Figure S5 of the Supplementary Material. From Fig. 6a and corresponding aafe, ccc and afe were found to be 1.32, 0.88 and Figure S5 it can be observed that for M1, the colonic 0.83. Despite the aforementioned differences, for all the methods, absorption constituted a significant proportion of the total the overall error (aafe) between the observed and predicted fabs fabs, especially for lowly permeable drugs (enalaprilat, furo- remained less than 1.5-fold, thus confirming the utility of the semide, terbutaline and atenolol). This was not the case of the mSAT model combined with Peff,int for fabs predictions. predictions using either M2 or M3, where the colonic The outcome changed dramatically when colonic absorption contribution to the total fabs remained relatively small was allowed in the mSAT model (Fig. 4d, e). When employing compared to that of the overall SI. This was consistent with Peff,int calculated by M1 (cylindrical surface area), the fabs was what is expected for the in vivo colonic fabs given the nature systematically overpredicted (Fig. 4d). This was reflectedinthe of the colon (35). For the simulated MR profiles, Fig. 6b and changes to the ccc (0.33), aafe (1.49) and afe (1.49). Using M2, the Figure S6, the colonic contribution to the total fabs increased predictions were slightly improved compared to that in the gradually with the increase in permeability (for all the previous scenario (Fig. 4b vs e); the ccc changed to 0.91, the aafe methods). For all the drugs listed in Table II, when using to 1.26 and the afe to 0.89. Using M3, however, the predictions M1, the relative colonic contribution to the fabs was higher remained almost invariant in terms of ccc (0.94), aafe (1.22) and than for the whole SI combined (Fig. 6b and Figure S6), afe (1.15). The simulated relationship between Peff,int and fabs whereas for M2 and M3, this proportion was dependent upon using the mSAT model, for a wider range of Peff,int values, the permeability value. 1186 Olivares-Morales et al.

Fig. 4. fabs (%) predictions using Peff values from Table II and the mSAT model. Upper panel, no colonic absorption allowed in the mSAT model: a M1 (cylindrical SA), b M2 (mSA according to Wilson’s method), c M3 (mSA according to Helander and Fändriks’ method). Lower panel, colonic absorption allowed in the mSAT model: d M1, e M2 and f M3. Black solid lines represent the line of unity, whereas the dashed grey lines represent the twofold error

DISCUSSION implicitly encompasses several mechanisms involved in membrane permeation, such as: diffusion through the unstirred The work performed herein provides a mechanistically water layer, passive transcellular permeation, carrier mediat- sound alternative to the bottom-up estimation of the regional ed absorption/secretion and passive paracellular permeation intestinal absorption of drugs administered orally. This ap- (12,17,60). When Peff is experimentally determined from proach is based on the rationalization of Peff by the available absorption data (e.g. luminal disappearance rate of the drug), mSA along the different regions of the human intestine. its value is scaled by the SA available in the test segment, As discussed earlier in this manuscript, Peff is an effective usually assuming a cylindrical SA (13). This SA does not property related to the drug passage from the lumen and reﬂect all available structures in the SI that might increase the across the apical intestinal epithelium. Once the intraluminal surface area compared to a cylinder, i.e. circular folds, and intracellular processes are accounted for, Peff itself intestinal villi and intestinal microvilli (Fig. 1).

Table IV. Predicted fabs Using the mSAT Model and the Peff,int (Double-Balloon) Values from Table II

Predicted fabs (%) mSAT model Predicted fabs (%) mSAT model [no colonic absorption] [colonic absorption allowed]

a b Compound BCS class fabs (%) observed M1 M2 M3 M1 M2 M3 Enalaprilat 3 10 37.4 15.9 30.0 75.5 18.2 32.8 Furosemide 4 55 47.4 21.3 38.8 83.8 24.2 42.2 Terbutaline 3 44 63.7 32.1 54.2 92.4 36.2 58.3 Atenolol 3 56 65.5 33.5 56.0 93.1 37.7 60.2 Metoprolol 1 88 92.4 66.4 87.3 99.4 71.9 90.3 Propranolol 1 92 98.0 83.8 96.0 99.9 88.2 97.5 Fluvastatin 1 95 98.2 84.8 96.4 99.9 89.0 97.8 Antipyrine 1 97 99.7 96.1 99.4 100 97.9 99.7 Naproxen 1 99 99.9 98.4 99.8 100 99.3 99.9 Ketoprofen 1 100 99.9 98.7 99.8 100 99.4 99.9 a BCS classiﬁcation from reference (4) b Observed fabs data from reference (33) Surface Area, Permeability and Oral Drug Absorption 1187

Fig. 5. Simulated Peff* to fabs relationship for the mSAT model. a Colonic absorption not allowed. b Colonic absorption allowed. The lines represent different methods for the estimation of Peff,int. Black solid line, M1; grey long-dashed line, M2; and grey short-dashed line, M3. *For M2 and M3, Peff is re-calculated internally by the mSAT model according to the mSA available in the double-balloon segment. The insert shows the same plots in the semi logarithmic scale

Despite this simpliﬁcation, Peff, as it is, has been an effective where Asi,lumen is the amount of drug in the SI’s lumen, SAsi is predictor of both the rate and extent the GI absorption the SA of the SI and SALoc-i-Gut is the SA of the double- (2,16,61). This discrepancy can be explained by the following balloon segment. The total mSA of this segment can be differential equation that describes the drug absorption from the represented as: 2×π×rLIG×LLIG×SAEF. The term SAEF is the SI (assuming the SI as single compartment) (1,2): combined SA expansion factor due all the aforementioned intestinal structures. For the single compartment case, the main assumption is that the structure of the small-intestinal dAsi;lumen CLabs SAsi membrane is relatively the same across the whole length of ¼ − Â Â Asi;lumen ð14Þ dt SALoc‐i‐Gut Vsi the SI; therefore, its SA is given by: 2×π×rsi×Lsi×SAEF.

Fig. 6. Bar chart of the predicted fabs (overall and regional) using the mSAT model and Peff,int for a subset of representatives drugs from Table II (colonic absorption was allowed). Upper panel (a), drugs assumed to be administered in solution; lower panel (b), drugs assumed to be administered as a MR formulation. Each bar represents a different method for the estimation of the absorption

(M1, M2 and M3), whereas the shades of grey indicate the proportion of the contribution to the fabs from each intestinal segment 1188 Olivares-Morales et al.

Under this assumption, the SAEF terms cancels out in Eq. 14, Since both M1 and M2 provided relatively similar results, so that the intestinal absorption can be predicted from Peff, to this point, the use of Peff,int instead of Peff did not provide regardless of SA employed for its estimation (cylindrical vs any advantage (or disadvantage) for the estimation of mucosal). However, when the structure of the membrane is regional absorption. The reason for the lack of difference significantly different of that where Peff was measured, i.e. the can be explained from the ratio between jejunal and ileal colon, the latter assumption does not necessarily hold true, mSA employed for the estimation of CLabs(ileum,k) (Eq. 8): and the segment-dependent SAEF has to be considered for 1.26, for M1, and 1.56, for M2 (Table I). These similarities can the estimation of the intestinal SA and absorption. be also explained by the similar nature of the intestinal Based on this theoretical framework, we hypothesized that membrane along the human SI; while there are SA differ- when using the mSA for the estimation of Peff,anBintrinsic Peff^ ences between the ileum and the upper jejunum, they are not (Peff,int) can be obtained which would remain relatively constant as dramatic as when the SA is compared with, for instance, along the whole human intestine. Under the aforementioned the colonic membrane. Yet, no direct colonic Peff data is assumption, the regional differences in absorption should be a available to perform such comparison (4,5). Further investi- product of the variable conditions along the GI tract and the gations were performed by evaluating the use of Peff,int variable nature of the intestinal membrane, where the differ- compared to Peff for the predictions of fabs. This analysis ences in mSA are assumed to play a major role in defining such was performed using the newly developed mSAT model. differences in absorption. This assumes however that the Contrary to the traditional CAT model, where the mass differences in the membrane composition are relatively small transfer along the SI is described by a series of (seven) transit across the human intestine, which might not be necessary true as compartments, all with the same mean residence time (31– the small-intestinal and colon membrane have different physi- 33), the mSAT model performs a similar task using only three cochemical properties (62–64) as well as differences in the compartments. These three compartments, however, are mucuslayerandUWLcomposition(65). physiologically sound, so that the lengths and radii are To support the approach of employing the mSA differ- representative of each one of the small intestinal segments ences along the intestine for the estimation of regional GI (Tables S1 and S2). In addition, the mean drug (mass) absorption, the first step was to provide good estimates of the residence time is related to the segment’s length, where the changes in the available mSA along the GI tract. The data longer the segment, the longer its residence time. This was from Wilson [36] provided a good starting point for such an implemented by assuming a non-linear mass transfer between approach, whereas the data recently published by Helander the segments using a Weibull function (Eq. 10). and Fändriks (2014) served as an appropriate complement for Alternative models were also considered in order to such estimations (36,39). Both mSA estimation methods describe small-intestinal transit. For example, the data were assume a decrease in the mSA from the proximal to the also fitted by using an equally divided first-order transit rate distal regions of the GI tract (Fig. 1a). This is consistent with constant (kt,si=3/SITT) in all the segments or by using a the observed reduction in abundance and length of the proportionally divided transit rate constant (kt,n=1/fn×SITT), circular folds and villi when moving from the proximal where fn is the fractional length of the small-intestinal towards the distal SI (23,36,39). For instance, when using segment. However, as shown in Fig. 3, none of these M2 and M3, the ileal mSA was estimated as 36 and 74% of alternative models adequately described the mean SITT data that of the jejunum, respectively, whereas for M1, this (31). Therefore, the non-linear Weibull transfer implemented relationship was estimated as 127% (Table S1). In the colonic in the mSAT model was considered as the best alternative to case, due to the lack of circular folds and villi, the ratio the full CAT model (Fig. 3). between jejunal and ascending colonic surface area was Under the same assumption as for the CAT model, i.e. no reduced to 0.4 and 0.3%, for M2 and M3, respectively colonic absorption, the fabs predictions using the mSAT model (Table S1). The total surface area for the SI was around 79 were similar to that of the CAT model using the same validation and 98 m2 when estimated using either M2 or M3, dataset (33). The agreement between the observed and respectively. These values were in relatively good agreement predicted fabs when using the mSAT model (Fig. 4a–c) suggests 2 with the value of 71 m reported by Willmann and co-workers that both the mSAT model and Peff,int can be used for the (8). The reported value by Helander and Fändriks [39] was prediction of GI absorption, especially when using the classical 2 around 30 m , due to the use of the physiological length of the approach (Peff)andPeff,int estimated by M3. Yet, when using small intestine in their calculations (291 cm), 43% of the small Peff,int calculated by M2, a systematic underprediction of fabs was intestinal length employed in our study (670.7 cm) (39). In observed for almost all the drugs and BCS classes (compared to contrast, the value of 435 m2 obtained using the classical M1 and M3, as shown in Fig. 4b and Table IV). This amplification factors (600-fold) seems to over-estimate the SA underprediction was probably due to the combination of two of the SI compared to the current estimates (36,39). factors: (a) higher jejunal mSA estimated for the double-balloon Further support for the possible use of Peff,int for the segment in M2, and hence lower Peff,int values (Table II), and (b) prediction of regional oral absorption was provided by the a decreased mSA in the distal regions of the intestine in M2 comparison between regional Peff,int values in Table III. Overall, when compared to M3 (Table S1). there was good agreement between jejunal and ileal Peff,int The main differences between the two methods, classical (Fig. 2a–c), especially when using either M1 or M2 methods. A Peffvs Peff,int, were observed when colonic absorption was similar scenario was observed when the ileal absorption clearance included in the simulations (Table IVand Fig. 4d–f). It has been was predicted using jejunal Peff, int and the ileal mSA (Fig. 2d, f), suggested that due to a reduced SA and higher abundance of in line with our hypothesis of the existence of a relatively similar tight junctions (i.e. assumed smaller paracellular area) in the Peff,int along the membrane of the human intestine. colonic membrane, intestinal permeability and possibly Surface Area, Permeability and Oral Drug Absorption 1189

absorption should be reduced compared to that in the SI of Peff to derive Peff,int. The latter might limit the application (4,5,22,25,66,67). This was appropriately captured when the of this approach when such mechanisms are required to be predictions were made using Peff,int, estimated by both M2 and explicitly accounted for the estimation of the overall rate and M3, without affecting the overall fabs predictions (Figs. 4d–f and extent of drug absorption. However, the development of a 6a and Figure S5). In fact, based on the aafe, ccc and afe, the fully mechanistic model to describe regional intestinal per- predictions for M3 remained almost unaffected compared to meability was beyond the scope of this work. that when colonic absorption was excluded (Fig. 4). Several simplifications were made during this analysis, When employing the classical approach for permeability one of the most relevant being the assumption of complete (M1), however, the overall fabs was systematically overestimated; dissolution and no precipitation. This assumption might affect the ccc was reduced by almost two thirds compared to that when the predictions made for drugs whose solubility might limit colonic absorption was excluded. This overestimation might their oral absorption, i.e. BCS class 2 and 4 drugs. For those explain why it is a common practice to assume negligible colonic drugs, accounting for the GI physiological factors affecting absorption during PBPK modelling of orally administrated drugs the solubility, dissolution and precipitation along the GI tract in solution, suspensions or IR formulations (31,33,34). When is absolutely necessary, especially when it comes to the using Peff,int (M3), however, there is no need to make such prediction and understanding of their regional intestinal assumptions, as the model seems to be able to capture such absorption (4,25,83–85). However, for our simulations, that differences in absorption (Fig. 6a and Figure S5). assumption seemed reasonable as the majority of the drugs The aforementioned assumption has especial implica- listed in Table II can be classified as highly soluble (BCS tions when it comes to the modelling of oral MR formula- classes 1 and 3), with the exception of furosemide, which is tions. Current approaches for PBPK modelling of drugs poorly soluble weak acid (pKa=3.9), yet its solubility is formulated as oral MR involve either the need to perform a expected to be high at the intestinal pH range (6–7.4) (85,86). sensitivity analysis on the colonic permeability or to optimize One of the main advantages of the approach developed it based on observed clinical data (34,68–71). Whereas some in this work (Peff,int) is that, given its simplicity, it can be drugs have been successfully modelled using jejunal Peff (i.e. readily implemented in the current intestinal mechanistic M1) as a fixed value along all the GI segments within the PBPK models. This can be done by means of applying the PBPK model (72), others might not necessarily benefit from SAEF to the Peff values determined by the double-balloon such an approach, due to the inherent risk of overestimations technique; this will derive Peff (or Peff,int) values for each in the fabs, especially for lowly permeable drugs (73). Thus, intestinal segment. The SAEF coefficients and an example of the use of Preff int seems to be an appropriate alternative how to apply them can be found in Section 5 of the during the bottom-up prediction of oral absorption of drugs Supplementary Material. administered as MR or at least a good starting point (Fig. 6b). It is clear that reliable regional human intestinal perme- CONCLUSION ability estimates are key for the successful prediction of drug absorption from the distal GI tract and to validate the A new approach for the prediction of regional intestinal approach presented herein. Due to the elevated cost of the absorption was proposed based on the scaling of in vivo- clinical investigation of regional human Peff, the use of animal determined jejunal Peff by the available mSA along the models and in vitro systems such as the Ussing chamber human GI tract to derive an intrinsic Peff (Peff, int). This combined with excised human intestinal fragments can approach was successfully employed for the prediction of the provide a reliable alternative for such estimates ileal absorption of several structurally diverse compounds. (3,60,65,66,74,75). In addition, one of the main goals of the Peff,int was combined with a newly developed semi- Pan-European project Oral Biopharmaceutical Tools mechanistic absorption PBPK model for the prediction of (OrBiTo), funded by the Innovative Medicines Initiative fabs, where the predictions showed a good agreement with the (IMI), is to provide the necessary information to improve observed data. In addition, the new approach showed to be our understanding of how orally administered drugs become robust when the colonic absorption was allowed in the PBPK absorbed from the GI tract and to generate better in vitro and model, by reducing the observed overprediction of fabs when in silico tools that allow a better prediction of their in vivo using the classical Peff. Therefore, due to its simplicity, the performance (http://www.orbitoproject.eu/objectives). There- new approach provides a useful alternative for the bottom-up fore, the aforementioned permeability investigations are part prediction of oral drug absorption, especially when the distal of the main goals of the project (76). GI tract plays a crucial role for a drug’s oral absorption. One of the major drawbacks of the newly proposed approach is its dependence on human jejunal in vivo Peff ACKNOWLEDGMENTS values for the prediction of regional intestinal absorption. Those jejunal Peff values are currently limited to only 30 A.O-M.isrecipientofaPh.D.grantawardedby drugs (3–5). Therefore, its application to drugs not listed in CONICYT Chile, Chilean Ministry of Education, and a the current datasets depends on the availability of in vivo, President’s Doctoral Scholar Award from The University of in vitro or in silico methods for the prediction of jejunal Peff Manchester. The authors would like to acknowledge the fruitful (59,60,77–82). Due to the use of Peff for this study, regional comments and discussions made with members of the Centre for differences in the transport mechanisms of the drugs involved Applied Pharmacokinetic Research (CAPKR) of The Univer- herein were not explicitly considered (e.g. UWL permeation, sity of Manchester. This project is an associated (Bsideground^) carrier-mediated transport, paracellular permeability). In- contribution to the IMI Oral Biopharmaceutical Tools (OrBiTo) stead, their impact was implicitly accounted for by the use project (http://www.imi.europa.eu/content/orbito). 1190 Olivares-Morales et al.

Conﬂict of Interest The authors declare no conﬂict of interest. Jejunum (jej) and ileum (ile) (n=2, 3): ÀÁ dAsolid;n ¼ wtðÞ Â A ; − − k þ wtðÞ Â A ; ðA7Þ dt n−1 solid n 1 rel n solid n APPENDIX 1: MSAT MODEL EQUATIONS

Small-Intestinal Transit: dAdiss;n For the small-intestinal segments, the time-dependent ¼ wtðÞ− Â Adiss;n−1 þ krel Â Asolid;n ðA8Þ dt n 1 transit rate for the duodenum (duo), jejunum (jej) and ileum − ðÞ þ Â SAn;k Â (ile) was defined as per Eq. A1, a detailed explanation of the wtn Peff;intðÞk DF Vn parameters and subscripts is given in the Materials and Â Methods section of the manuscript. Adiss;n β β−1 ðÞ¼ ðÞ ¼ Â t ð Þ kt;n t wtn Â Â γ Â Â γ A1 f n SITT f n SITT dAwall;n SAn;k ¼ Peff;intðÞk Â Â DFÂ Adiss;n ðA9Þ dt Vn where kt,n is the time-dependent transit rate constant for the nth small-intestinal segment, defined by a Weibull transfer β ’ rate function, w(t)n, is the Weibull s shape parameter, fn is Ascending colon (acol): the fractional length (with respect to Lsi) of the nth small- intestinal segment, SITT is the mean small-intestinal transit ÀÁ dAsolid;acol ðÞ time and γ is a dimensionless coefficient. ¼ wt Â Asolid;ile− krel þ kt;acol Â Asolid;acol ðA10Þ dt ile Mass transfer equations: The general differential equations describing the amount dA j;segment of drug in each compartment ( dt ) are described below dA ; (Eqs. A2-A12), where j stands for the drug’s state within the diss acol ¼ ðÞ Â þ Â ð Þ wtile Adiss;ile krel Asolid;acol A11 dA ; dt intestinal segment, i.e. solid in the lumen ( solid segment), dt − þ Â SAacol;k Â dAdiss;segment kt;acol Peff ;intðÞk DF dissolved/released in the lumen ( dt ) and/or absorbed Vacol dAwall;segment in the intestinal wall ( dt ). However, for most of the Â Adiss;acol analysis performed in this work, the drug was assumed to be administered as a solution with no possibility for precipitation. dAwall;acol SAacol;k Stomach (st): ¼ Peff;intðÞk Â Â DFÂ Adiss;acol ðA12Þ dt Vacol

ÀÁ dAsolid;st ¼ − k þ k Â A ; ðA2Þ dt rel ge solid st REFERENCES

1. Sinko PJ, Leesman GD, Amidon GL. Predicting fraction dose dAdiss;st ¼ krel Â Asolid;st−kge Â Adiss;st ðA3Þ absorbed in humans using a macroscopic mass balance approach. dt Pharm Res. 1991;8(8):979–88. doi:10.1023/A:1015892621261. Duodenum (duo), n=1 2. Amidon GL, Lennernas H, Shah VP, Crison JR. A theoretical basis for a biopharmaceutic drug classiﬁcation: the correlation of in vitro drug product dissolution and in vivo bioavailability. Pharm Res. 1995;12(3):413–20. ÀÁ dAsolid;duo 3. Lennernas H. Intestinal permeability and its relevance for ¼ k Â A ; − k þ wtðÞ Â A ; ðA4Þ – – dt ge solid st rel duo solid duo absorption and elimination. Xenobiotica. 2007;37(10 11):1015 51. doi:10.1080/00498250701704819. 4. Lennernas H. Human in vivo regional intestinal permeability: importance for pharmaceutical drug development. Mol Pharm. 2014;11(1):12–23. doi:10.1021/mp4003392. dAdiss;duo 5. Lennernas H. Regional intestinal drug permeation: ¼ kge Â Adiss;st þ krel Â Asolid;duo ðA5Þ dt biopharmaceutics and drug development. Eur J Pharm Sci. – SAduo;k 2014;57:333 41. doi:10.1016/j.ejps.2013.08.025. − wtðÞ þ P ; ðÞÂ Â DF duo eff int k V 6. Agoram B, Woltosz WS, Bolger MB. Predicting the impact of duo physiological and biochemical processes on oral drug bioavail- – Â Adiss;duo ability. Adv Drug Deliv Rev. 2001;50(Suppl 1(0)):S41 67. doi:10.1016/S0169-409X(01)00179-X. 7. Thelen K, Coboeken K, Willmann S, Burghaus R, Dressman JB, Lippert J. Evolution of a detailed physiological model to simulate the gastrointestinal transit and absorption process in dAwall;duo SAduo;k ¼ Peff;intðÞk Â Â DFÂ Adiss;duo ðA6Þ humans, part 1: oral solutions. J Pharm Sci. 2011;100(12):5324– dt Vduo 45. doi:10.1002/jps.22726. Surface Area, Permeability and Oral Drug Absorption 1191

8. Willmann S, Schmitt W, Keldenich J, Lippert J, Dressman JB. A administration: inter-individual variability of cimetidine absorp- physiological model for the estimation of the fraction dose tion. J Pharm Pharmacol. 2009;61(7):891–9. doi:10.1211/jpp/ absorbed in humans. J Med Chem. 2004;47(16):4022–31. 61.07.0008. doi:10.1021/jm030999b. 28. Jamei M, Turner D, Yang J, Neuhoff S, Polak S, Rostami- 9. Jamei M, Marciniak S, Feng K, Barnett A, Tucker G, Rostami- Hodjegan A, et al. Population-based mechanistic prediction of Hodjegan A. The Simcyp population-based ADME simulator. oral drug absorption. AAPS J. 2009;11(2):225–37. doi:10.1208/ Expert Opin Drug Metab Toxicol. 2009;5(2):211–23. doi:10.1517/ s12248-009-9099-y. 17425250802691074. 29. Woltosz WS, Bolger MB, Lukacova V. Modeling of absorption. 10. Sjogren E, Westergren J, Grant I, Hanisch G, Lindfors L, Methods Mol Biol. 2012;929:313–36. doi:10.1007/978-1-62703- Lennernas H, et al. In silico predictions of gastrointestinal drug 050-2_13. absorption in pharmaceutical product development: application 30. Carlert S, Lennernas H, Abrahamsson B. Evaluation of the use of the mechanistic absorption model GI-Sim. Eur J Pharm Sci. of Classical Nucleation Theory for predicting intestinal crystal- 2013;49(4):679–98. doi:10.1016/j.ejps.2013.05.019. line precipitation of two weakly basic BSC class II drugs. Eur J 11. Gertz M, Houston JB, Galetin A. Physiologically based pharma- Pharm Sci. 2014;53:17–27. doi:10.1016/j.ejps.2013.12.004. cokinetic modeling of intestinal first-pass metabolism of CYP3A 31. Yu LX, Lipka E, Crison JR, Amidon GL. Transport approaches substrates with high intestinal extraction. Drug Metab Dispos. to the biopharmaceutical design of oral drug delivery systems: 2011;39(9):1633–42. doi:10.1124/dmd.111.039248. prediction of intestinal absorption. Adv Drug Deliv Rev. 12. Lennernas H. Human intestinal permeability. J Pharm Sci. 1996;19(3):359–76. doi:10.1016/0169-409X(96)00009-9. 1998;87(4):403–10. doi:10.1021/js970332a. 32. Yu LX, Crison JR, Amidon GL. Compartmental transit and 13. Lennernas H, Ahrenstedt O, Hallgren R, Knutson L, Ryde M, dispersion model analysis of small intestinal transit flow in humans. Paalzow LK. Regional jejunal perfusion, a new in vivo approach Int J Pharm. 1996;140(1):111–8. doi:10.1016/0378-5173(96)04592-9. to study oral drug absorption in man. Pharm Res. 33. Yu LX, Amidon GL. A compartmental absorption and transit 1992;9(10):1243–51. doi:10.1023/A:1015888813741. model for estimating oral drug absorption. Int J Pharm. 14. Knutson L, Odlind B, Hallgren R. A new technique for 1999;186(2):119–25. doi:10.1016/S0378-5173(99)00147-7. segmental jejunal perfusion in man. Am J Gastroenterol. 34. Kesisoglou F, Wu Y. Understanding the effect of API properties 1989;84(10):1278–84. on bioavailability through absorption modeling. AAPS J. 15. Knutson T, Fridblom P, Ahlstrom H, Magnusson A, Tannergren 2008;10(4):516–25. doi:10.1208/s12248-008-9061-4. C, Lennernas H. Increased understanding of intestinal drug 35. Rubinstein A. Colonic drug delivery. Drug Discov Today permeability determined by the LOC-I-GUT approach using Technol. 2005;2(1):33–7. doi:10.1016/j.ddtec.2005.05.021. multislice computed tomography. Mol Pharm. 2009;6(1):2–10. 36. Wilson JP. Surface area of the small intestine in man. Gut. doi:10.1021/mp800145r. 1967;8(6):618–21. 16. Lennernas H, Lee ID, Fagerholm U, Amidon GL. A residence- 37. DeSesso JM, Jacobson CF. Anatomical and physiological parameters time distribution analysis of the hydrodynamics within the affecting gastrointestinal absorption in humans and rats. Food Chem intestine in man during a regional single-pass perfusion with Toxicol. 2001;39(3):209–28. doi:10.1016/S0278-6915(00)00136-8. Loc-I-Gut: in-vivo permeability estimation. J Pharm Pharmacol. 38. Wilson TH. Intestinal absorption. Philadelphia: W.B. Saunders 1997;49(7):682–6. doi:10.1111/j.2042-7158.1997.tb06092.x. Company; 1962. 17. Sugano K, Kansy M, Artursson P, Avdeef A, Bendels S, Di L, 39. Helander HF, Fändriks L. Surface area of the digestive et al. Coexistence of passive and carrier-mediated processes in tract—revisited. Scand J Gastroenterol. 2014;49(6):681–9. drug transport. Nat Rev Drug Discov. 2010;9(8):597–614. doi:10.3109/00365521.2014.898326. doi:10.1038/nrd3187. 40. Valentin J. Basic anatomical and physiological data for use in 18. Amidon GL, Kou J, Elliott RL, Lightfoot EN. Analysis of radiological protection: reference values. Ann ICRP. 2002;32(3– models for determining intestinal wall permeabilities. J Pharm 4):1–277. doi:10.1016/s0146-6453(03)00002-2. Sci. 1980;69(12):1369–73. 41. Hounnou G, Destrieux C, Desme J, Bertrand P, Velut S. 19. Fagerholm U, Lennernäs H. Experimental estimation of the Anatomical study of the length of the human intestine. Surg effective unstirred water layer thickness in the human jejunum, Radiol Anat. 2002;24(5):290–4. doi:10.1007/s00276-002-0057-y. and its importance in oral drug absorption. Eur J Pharm Sci. 42. Sadahiro S, Ohmura T, Yamada Y, Saito T, Taki Y. Analysis of 1995;3(5):247–53. doi:10.1016/0928-0987(95)00027-B. length and surface-area of each segment of the large-intestine 20. Dahan A, Miller JM, Hilfinger JM, Yamashita S, Yu LX, according to age, sex and physique. Surg Radiol Anat. Lennernas H, et al. High-permeability criterion for BCS classi- 1992;14(3):251–7. doi:10.1007/Bf01794949. fication: segmental/pH dependent permeability considerations. 43. Sinclair TS, Jones DA, Kumar PJ, Phillips AD. The microvillus Mol Pharm. 2010;7(5):1827–34. doi:10.1021/mp100175a. in adult jejunal mucosa—an electron microscopic study. Histo- 21. Dahan A, Lennernas H, Amidon GL. The fraction dose absorbed, pathology. 1984;8(5):739–46. in humans, and high jejunal human permeability relationship. Mol 44. Gramatte T, Oertel R, Terhaag B, Kirch W. Direct demonstra- Pharm. 2012;9(6):1847–51. doi:10.1021/mp300140h. tion of small intestinal secretion and site-dependent absorption 22. Watts PJ, Illum L. Colonic drug delivery. Drug Dev Ind Pharm. of the [beta]-blocker talinolol in humans [ast]. Clin Pharmacol 1997;23(9):893–913. doi:10.3109/03639049709148695. Ther. 1996;59(5):541–9. 23. International Commission on Radiological Protection. Report of 45. Gramatte T, Richter K. Paracetamol absorption from different the Task Group on Reference Man: Pergamon Press; 1975. sites in the human small intestine. Br J Clin Pharmacol. 24. Coupe AJ, Davis SS, Evans DF, Wilding IR. Nocturnal 1994;37(6):608–11. scintigraphic imaging to investigate the gastrointestinal transit 46. Gramatte T, el Desoky E, Klotz U. Site-dependent small of dosage forms. J Control Release. 1992;20(2):155–62. intestinal absorption of ranitidine. Eur J Clin Pharmacol. doi:10.1016/0168-3659(92)90161-J. 1994;46(3):253–9. doi:10.1007/BF00192558. 25. Tannergren C, Bergendal A, Lennernas H, Abrahamsson B. 47. Gramatte T. Griseofulvin absorption from different sites in the Toward an increased understanding of the barriers to colonic human small intestine. Biopharm Drug Dispos. 1994;15(9):747– drug absorption in humans: implications for early controlled 59. doi:10.1002/bdd.2510150903. release candidate assessment. Mol Pharm. 2009;6(1):60–73. 48. Sutcliffe FA, Riley SA, Kaserliard B, Turnberg LA, Rowland M. doi:10.1021/mp800261a. Absorption of drugs from the human jejunum and ileum. Br J 26. Thelen K, Coboeken K, Willmann S, Dressman JB, Lippert J. Clin Pharmacol. 1988;26(2):P206–7. Evolution of a detailed physiological model to simulate the 49. Schedl HP. Absorption of steroid hormones from the human gastrointestinal transit and absorption process in humans, part II: small intestine. J Clin Endocrinol Metab. 1965;25(10):1309–16. extension to describe performance of solid dosage forms. J doi:10.1210/jcem-25-10-1309. Pharm Sci. 2012;101(3):1267–80. doi:10.1002/jps.22825. 50. Amidon GL, Sinko PJ, Fleisher D. Estimating human oral fraction 27. Willmann S, Edginton AN, Kleine-Besten M, Jantratid E, dose absorbed: a correlation using rat intestinal membrane Thelen K, Dressman JB. Whole-body physiologically based permeability for passive and carrier-mediated compounds. Pharm pharmacokinetic population modelling of oral drug Res. 1988;5(10):651–4. doi:10.1023/a:1015927004752. 1192 Olivares-Morales et al.

51. Musther H, Olivares-Morales A, Hatley OJ, Liu B, Rostami Hodjegan application to food effects. AAPS J. 2009;11(1):45–53. A. Animal versus human oral drug bioavailability: do they correlate? doi:10.1208/s12248-008-9079-7. Eur J Pharm Sci. 2014;57:280–91. doi:10.1016/j.ejps.2013.08.018. 71. Lukacova V, Woltosz WS, Bolger MB. Prediction of modified 52. McBride G. A proposal for strength-of-agreement criteria for release pharmacokinetics and pharmacodynamics from in vitro, Lin’s concordance correlation coefficient. NIWA Client Report: immediate release, and intravenous data. AAPS J. HAM2005-062. 2005. 2009;11(2):323–34. doi:10.1208/s12248-009-9107-2. 53. Yu LX, Amidon GL. Saturable small intestinal drug absorption 72. Patel N, Polak S, Jamei M, Rostami-Hodjegan A, Turner DB. in humans: modeling and interpretation of cefatrizine data. Eur J Quantitative prediction of formulation-specific food effects and Pharm Biopharm. 1998;45(2):199–203. doi:10.1016/S0939- their population variability from in vitro data with the 6411(97)00088-X. physiologically-based ADAM model: a case study using the 54. Read NW, Al-Janabi MN, Holgate AM, Barber DC, Edwards BCS/BDDCS Class II drug nifedipine. Eur J Pharm Sci. CA. Simultaneous measurement of gastric emptying, small bowel 2014;57:240–9. doi:10.1016/j.ejps.2013.09.006. residence and colonic filling of a solid meal by the use of the 73. Olivares-Morales A, Kamiyama Y, Darwich AS, Aarons L, gamma camera. Gut. 1986;27(3):300–8. Rostami-Hodjegan A. Analysis of the impact of controlled 55. Davis SS, Hardy JG, Fara JW. Transit of pharmaceutical dosage release formulations on oral drug absorption, gut wall metabo- forms through the small-intestine. Gut. 1986;27(8):886–92. lism and relative bioavailability of CYP3A substrates using a doi:10.1136/Gut.27.8.886. physiologically-based pharmacokinetic model. Eur J Pharm Sci. 56. Yu LX, Amidon GL. Characterization of small intestinal transit 2015;67:32–44. doi:10.1016/j.ejps.2014.10.018. time distribution in humans. Int J Pharm. 1998;171(2):157–63. 74. Lennernäs H, Nylander S, Ungell AL. Jejunal permeability: a doi:10.1016/S0378-5173(98)00174-4. comparison between the Ussing chamber technique and the 57. Bouchoucha M, Devroede G, Dorval E, Faye A, Arhan P, Arsac single-pass perfusion in humans. Pharm Res. 1997;14(5):667–71. M. Different segmental transit times in patients with irritable doi:10.1023/a:1012121632357. bowel syndrome and Bnormal^ colonic transit time: is there a 75. Lennernas H. Animal data: the contributions of the Ussing correlation with symptoms? Tech Coloproctol. 2006;10(4):287– chamber and perfusion systems to predicting human oral drug 96. doi:10.1007/s10151-006-0295-9. delivery in vivo. Adv Drug Deliv Rev. 2007;59(11):1103–20. 58. Bouchoucha M, Thomas SR. Error analysis of classic colonic doi:10.1016/j.addr.2007.06.016. transit time estimates. Am J Physiol Gastrointest Liver Physiol. 76. Lennernas H, Aarons L, Augustijns P, Beato S, Bolger M, Box 2000;279(3):G520–7. K, et al. Oral biopharmaceutics tools—time for a new 59. Sugano K. Estimation of effective intestinal membrane perme- initiative—an introduction to the IMI project OrBiTo. Eur J ability considering bile micelle solubilisation. Int J Pharm. Pharm Sci. 2014;57:292–9. doi:10.1016/j.ejps.2013.10.012. 2009;368(1–2):116–22. doi:10.1016/j.ijpharm.2008.10.001. 77. Stenberg P, Luthman K, Artursson P. Virtual screening of 60. Lennernas H. Human jejunal effective permeability and its correla- intestinal drug permeability. J Control Release. 2000;65(1– tion with preclinical drug absorption models. J Pharm Pharmacol. 2):231–43. doi:10.1016/S0168-3659(99)00239-4. 1997;49(7):627–38. doi:10.1111/j.2042-7158.1997.tb06084.x. 78. Gertz M, Harrison A, Houston JB, Galetin A. Prediction of 61. Fagerholm U, Johansson M, Lennernas H. Comparison between human intestinal first-pass metabolism of 25 CYP3A substrates permeability coefficients in rat and human jejunum. Pharm Res. from in vitro clearance and permeability data. Drug Metab 1996;13(9):1336–42. doi:10.1023/A:1016065715308. Dispos. 2010;38(7):1147–58. doi:10.1124/dmd.110.032649. 62. Brasitus TA, Dudeja PK. Regional differences in the lipid composi- 79. Akamatsu M, Fujikawa M, Nakao K, Shimizu R. In silico tion and fluidity of rat colonic brush-border membranes. Biochim prediction of human oral absorption based on QSAR analyses Biophys Acta. 1985;819(1):10–7. doi:10.1016/0005-2736(85)90189-0. of PAMPA permeability. Chem Biodivers. 2009;6(11):1845–66. 63. Brasitus TA, Dudeja PK. Alterations in the physical state and doi:10.1002/cbdv.200900112. composition of brush border membrane lipids of rat enterocytes 80. Tavelin S, Taipalensuu J, Soderberg L, Morrison R, Chong SH, during differentiation. Arch Biochem Biophys. 1985;240(1):483– Artursson P. Prediction of the oral absorption of low- 8. doi:10.1016/0003-9861(85)90054-2. permeability drugs using small intestine-like 2/4/A1 cell mono- 64. Brasitus TA, Dudeja PK, Bolt MJ, Sitrin MD, Baum C. Dietary layers. Pharm Res. 2003;20(3):397–405. doi:10.1023/ triacylglycerol modulates sodium-dependent D-glucose trans- A:1022699920043. port, fluidity and fatty acid composition of rat small intestinal 81. Winiwarter S, Bonham NM, Ax F, Hallberg A, Lennernas H, brush-border membrane. Biochim Biophys Acta. Karlen A. Correlation of human jejunal permeability (in vivo) of 1989;979(2):177–86. doi:10.1016/0005-2736(89)90433-1. drugs with experimentally and theoretically derived parameters. 65. Dahlgren D, Roos C, Sjogren E, Lennernas H. Direct in vivo A multivariate data analysis approach. J Med Chem. human intestinal permeability (Peff) determined with different 1998;41(25):4939–49. doi:10.1021/jm9810102. clinical perfusion and intubation methods. J Pharm Sci. 2014. 82. Sun D, Lennernas H, Welage LS, Barnett JL, Landowski CP, doi:10.1002/jps.24258. Foster D, et al. Comparison of human duodenum and Caco-2 66. Sjoberg A, Lutz M, Tannergren C, Wingolf C, Borde A, Ungell gene expression profiles for 12,000 gene sequences tags and AL. Comprehensive study on regional human intestinal perme- correlation with permeability of 26 drugs. Pharm Res. ability and prediction of fraction absorbed of drugs using the 2002;19(10):1400–16. doi:10.1023/A:1020483911355. Ussing chamber technique. Eur J Pharm Sci. 2013;48(1–2):166– 83. Schiller C, Frohlich CP, Giessmann T, Siegmund W, Monnikes H, 80. doi:10.1016/j.ejps.2012.10.007. Hosten N, et al. Intestinal fluid volumes and transit of dosage 67. Ungell AL, Nylander S, Bergstrand S, Sjoberg A, Lennernas H. forms as assessed by magnetic resonance imaging. Aliment Membrane transport of drugs in different regions of the intestinal Pharmacol Ther. 2005;22(10):971–9. doi:10.1111/j.1365- tract of the rat. J Pharm Sci. 1998;87(3):360–6. doi:10.1021/js970218s. 2036.2005.02683.x. 68. Johnson TN, Zhou D, Bui KH. Development of physiologically 84. Sutton SC. Role of physiological intestinal water in oral absorption. based pharmacokinetic model to evaluate the relative systemic AAPS J. 2009;11(2):277–85. doi:10.1208/s12248-009-9087-2. exposure to quetiapine after administration of IR and XR 85. Fallingborg J, Christensen LA, Ingeman-Nielsen M, Jacobsen formulations to adults, children and adolescents. Biopharm Drug BA, Abildgaard K, Rasmussen HH. pH-profile and regional Dispos. 2014;35(6):341–52. doi:10.1002/bdd.1899. transit times of the normal gut measured by a radiotelemetry 69. Watson KJ, Davis J, Jones HM. Application of physiologically device. Aliment Pharmacol Ther. 1989;3(6):605–13. doi:10.1111/ based pharmacokinetic modeling to understanding the clinical j.1365-2036.1989.tb00254.x. pharmacokinetics of UK-369,003. Drug Metab Dispos. 86. Granero GE, Longhi MR, Mora MJ, Junginger HE, Midha KK, 2011;39(7):1203–13. doi:10.1124/dmd.111.038224. Shah VP, et al. Biowaiver monographs for immediate release 70. Parrott N, Lukacova V, Fraczkiewicz G, Bolger MB. Predicting solid oral dosage forms: furosemide. J Pharm Sci. pharmacokinetics of drugs using physiologically based modeling– 2010;99(6):2544–56. doi:10.1002/jps.22030.

Chapter 6: Development of a novel simplified PBPK absorption model to explain the higher relative bioavailability of the OROS formulation of oxybutynin

In preparation

Andrés Olivares-Morales, Avijit Ghosh, Leon Aarons and Amin Rostami-Hodjegan

6.1 Abstract

A new model that describes the small intestinal transit time using a reduced number of small intestinal compartments has been recently proposed. This model, or minimal Segmented Transit and Absorption model (mSAT), was combined with a new method to parametrize human intestinal effective permeability to predict regional gastrointestinal (GI) absorption (fabs). Herein this model was extended and applied for the prediction of oral bioavailability and pharmacokinetics of oxybutynin and its enantiomers (R and S oxybutynin). In addition the model was employed to provide a mechanistic explanation of the higher relative bioavailability observed for oxybutynin’s modified-release OROS formulation compared to its immediate-release (IR) counterpart. The expansion of the mSAT model involved the incorporation of mechanistic equations for the prediction of release, transit, dissolution, permeation and first pass metabolism, in both the gut wall and the liver. The predicted pharmacokinetics of OXY after oral administration for both the IR and OROS formulations were in close agreement with the observed data. The predicted absolute bioavailability for the IR formulation was within 5% of the observed value and the model adequately predicted the higher relative bioavailability observed for the OROS formulation, 170% vs. 139%, respectively. From the model predictions it can be noticed that the higher bioavailability observed for the OROS formulation was mainly attributable to differences in intestinal first-pass metabolism (FG) rather than due to a higher colonic fabs, thus confirming previous hypotheses. The predicted absorption was almost 70% lower for the OROS formulation compared to the IR formulation, whereas the FG was almost eight fold higher than in the IR formulation.

These results provide further support to the hypothesis of an increased FG as the main factor responsible for the higher bioavailability of OXY’s OROS formulation vs the IR.

6.2 Introduction

In recent years there has been an increase in the use of physiologically-based pharmacokinetics (PBPK) models in drug development, particularly in the preclinical and early clinical stages. Numerous articles coming from academia, industry and regulatory agencies have highlighted the benefits of applying such models in the drug discovery and development arena [1-6]. By incorporating some of the mechanisms driving the pharmacokinetics of a drug, PBPK models allows one to distinguish between drug-related and the physiological factors controlling drug absorption, disposition, metabolism and elimination (ADME). This characteristic enables one to integrate measured in vitro or in silico drug properties into the model in the so-called in vitro-in vivo extrapolation (IVIVE) approach, enabling the use of such models in a truly prospective fashion[6]. PBPK models have been increasingly applied in oral absorption and biopharmaceutics. This has been partly due to the development and availability of mechanistic absorption models such as the ones included in commercial software packages like SimCYP® (ADAM) [7], GastroPlus™ (ACAT) [8] and PK-Sim® [9, 10], as well as some in-house developments coming from academia and industry [11-14].

Oral absorption and bioavailability are dependent upon numerous physiological and drug-related factors interacting simultaneously. These factors are the key to defining the different steps of the oral absorption process, i.e., transit, dissolution, release, permeation, transport, and metabolism [15, 16]. Given the complex nature of these drug-physiology interactions, prospective predictions of oral bioavailability within the PBPK framework are still a challenging task [17-19]. Nevertheless, significant efforts have recently been made in order to improve our understanding of such a complex interplay. For instance, the Innovative Medicines Initiative (IMI) OrBiTo project has amongst its goals to enrich our knowledge of the physiological and biopharmaceutical properties that define in vivo drug absorption and to provide new in vitro and in silico tools that can help to make better predictions of the in vivo drug product performance [20].

One of the key biopharmaceutical factors defining oral drug absorption are formulation characteristics. In a recent study we investigated the impact that 88

modified/controlled release (MR/CR) formulations might have on the oral bioavailability of substrates of the cytochrome P450 (CYP) 3A4 enzymes using a prospective PBPK modelling and simulation (M&S) approach [21]. By employing this approach we were able to evaluate the interplay between drug and physiological- related factors governing intestinal absorption and first-pass metabolism, and in particular to identify the possible scenarios where MR formulations of a CYP3A substrate might display higher relative bioavailability (Frel) compared to its immediate released (IR) counterparts [21]. In our previous study, it was shown that highly cleared CYP3A-substrates belonging to Class 1 within the biopharmaceutics classification system (BCS) are more likely to display higher Frel when formulated as MR [21].The mechanism proposed for this phenomenon is an increased intestinal availability (FG) due to a decreased intestinal first-pass metabolism as a result of the lower abundance of CYP3A enzymes in the distal gastrointestinal (GI) tract, where most of the drug contained in the MR formulation is likely to be released and absorbed [21-24]. Another interesting outcome of the study was the observed trend to over predict fabs of MR formulations belonging to BCS Classes 2 and 3 [21, 25]. This over prediction was attributed to an overestimation of the colonic absorption due to the PBPK approach employed for the study. That is, intestinal effective permeability

(Peff) was assumed to be same for all intestinal segments implemented in the absorption model, including the colon [21]. In an attempt to address this issue, and given the lack of an implementation of regional intestinal (passive) permeability within in the PBPK framework, two novel approaches to implement regional Peff for prospective PBPK simulations were recently proposed [26]. The approaches were based on the translation of regional intestinal variations in the available mucosal surface area (mSA) into segment-dependent permeability and absorption[26]. When the approaches were combined with a novel simplified absorption PKPK model, or minimal Segmented Absorption and Transit (mSAT) model, they showed a potential to decrease the observed overestimation of the colonic fabs [26], especially when applying the so-called Method 3 (M3) which was based on the intestinal mSA values derived recently by Helander and Fändriks [26, 27].

One of the limitations of the prospective PBPK analysis of the bioavailability differences between IR and MR formulations of CYP3A substrates was the lack of

drug-specific simulations necessary to provide stronger support for the prospective modelling and simulation study outcome [21]. Therefore the present work was designed as a continuation of the work performed in the previous studies, with the main aim to provide a drug-specific example to support the hypothesis of a reduced first pass metabolism as the main driving mechanism of the higher Frel of CYP3A substrates formulated as MR [21]. In particular, this study investigated the bioavailability differences observed for the MR formulation of oxybutynin (OXY). OXY is a highly-cleared antimuscarinic drug with anticholinergic, spasmolytic and local anaesthetic properties employed for the treatment of urge urinary incontinency due to over activity of the detrusor muscle[28]. Base on mass balance and solubility studies, OXY is a highly permeable and soluble compound (BCS Class 1) [21, 29, 30]. After oral administration, OXY is rapidly absorbed and undergoes extensive first-pass metabolism in both the intestinal wall and the liver mainly mediated by CYP3A4 and its elimination is mainly by means of metabolism; only 0.02% of the dose is eliminated unchanged in the urine [28, 29, 31]. When formulated as a once a day (MR) OROS formulation, the relative bioavailability of OXY was around 153% compared to its IR tablet, while the exposure of its main metabolite, N-desethyloxybutynin (DEOB) was reduced by almost 30% [28]. This reduction in exposure was translated into an improved safety profile, due to a reduction in the incidence of dry mouth episodes, OXY’s main side- effect, which is mainly attributed to its active metabolite, yet keeping the same efficacy [32, 33]. In this study OXY’s bioavailability differences will be investigated by means of a PBPK modelling and simulation approach, combining in vivo, in vitro and in silico data together with an extended version of the previously developed mSAT model. The model will be expanded to prospectively predict OXY’s oral bioavailability for both IR and MR formulations [26].

6.3 Materials and Methods

6.3.1 PBPK model development

The PBPK model employed for the mechanistic pharmacokinetic prediction of OXY was developed using a two-step approach. Firstly, a compound-specific semi-

physiological PBPK model was developed and optimized with intravenous (IV) data to describe OXY’s disposition. Secondly, the optimized OXY’s disposition model was coupled with an extended version of the generic mSAT model to predict oral absorption and bioavailability using and in vitro-in vivo extrapolation (IVIVE) approach. The two-stage approach was chosen as it reduces the confounding issues that can arise when drug’s disposition is not properly considered in the prediction of oral pharmacokinetics, thus providing better information about the predictive performance of the oral absorption model (mSAT). The details of the different steps of the model development are described in the following sections:

6.3.1.1 Development of OXY’s disposition model

The semi-physiological PBPK model employed to describe OXY’s disposition was developed based on the approach proposed by Cao and Jusko (2012), whereby the model structure represents a hybrid between a whole-body PBPK model and a compartmental or mammillary pharmacokinetic model [34]. The advantage of employing such an approach, compared to the use of traditional compartmental models, is that the former allows the use of prior physiological and anatomical knowledge to define the model structure. Therefore under this approach, the model parameters are constrained between physiologically plausible limits [34]. This approach appropriately distinguishes between drug-related and system specific parameters, as the model can be informed from known drug-related properties, such as plasma protein binding and/or metabolic routes [34]. It should be noted however that the model structure was not the product of the reduction of a whole-body PBPK model using formal reduction techniques such as “proper lumping” [35-37]. In contrast, the model development was data driven, whereby the number of tissue compartments was defined based on the best description of the observed clinical data. The selected structure of the disposition model is shown in Figure 6.1A. The model is comprised of two anatomically defined compartments, systemic blood and liver, and three additional empirical tissue compartments, necessary to describe OXY’s disposition after an intravenous (IV) infusion [29]. All of the aforementioned compartments were assumed to be well-mixed, where drug transport into and from the tissues was assumed to occur by means of perfusion-limited processes. 91

The drug’s concentration in the ith non-eliminating empirical tissue was described by Equation 6.1,

dCtissuei,, C tissuei Equation 6.1 VQCtissuei,,  tissuei () blood  dt K b,, tissue i

where Vtissue,i represents the tissue volume (L), Ctissue,i is the drug’s concentration in the tissue (ng/mL), Qtissue,i the tissue blood flow (L/h) and Kb,tissue is an empirical tissue to blood partition coefficient necessary to describe the extent of OXY’s distribution into the tissue compartment. The dynamics in the blood compartment were described by Equation 6.2,

dC1 3 C Equation 6.2 VRQCQblood   ()    liver blood in tissuei,, tissuei HV dt Ki 1 K btissue,, bliver 3 ()QQC    tissuei, liver blood i 1

where Vblood is the systemic blood volume (L), Rin, is the drug’s infusion rate after IV administration (mg/h), QHV is the hepatic vein’s blood flow, Cliver is concentration in the liver tissue, Kb,liver is the drug’s liver tissue to blood partition coefficient and Qliver is the combined hepatic and non-villous splanchnic blood flow [38-40].The latter parameter was defined the sum of the arterial blood supply to the liver and the non- villous portal blood flow (i.e., oesophagus, stomach, gut tissue, pancreas, upper large intestine, lower large intestine, and spleen) [38-40]. OXY’s systemic elimination was assumed to occur exclusively through hepatic metabolism, given that only less than 0.02% of the administered dose was found unchanged in the urine after IV administration [29]. The dynamics of the liver compartment were described by Equation 6.3,

dC 4 Equation 6.3 VQCQCliver ()()    liver entn,, entn liver blood dt n 1 C liver ()QH V  f up  C L liver  K b, liver

where Qent,n and Cent,n are the villous blood flow and concentration entering the liver from the nth enterocyte compartment, respectively (these values were assumed zero for IV administrations), fup is the drug’s fraction unbound in plasma and CLliver is

OXY’s hepatic clearance (L/h). CLliver was scaled from the drug’s unbound intrinsic microsomal clearance (CLint(u), L/h/mg of microsomal protein) using the following equation [41] Equation 6.4 CLCLLWMPPGLliverint( u )  

where, LW is the liver weight (g), and MPPGL the amount of microsomal protein per gram of liver (mg/g).

6.3.1.1.1 Disposition model parameters and parameter estimation The majority of the system-related parameters employed in OXY’s semi- physiological disposition model were derived from the literature and are summarized in Table 6.1. The parameters were intended to represent those of a “reference individual”, i.e., Caucasian male, 70 kg and 1.70 m. The remaining model parameters were obtained by fitting the model shown in Figure 6.1A to OXY’s plasma concentration-time profile obtained after a 5mg IV infusion published by Douchamps and co-workers [29]. The data was digitized using GetData Graph Digitizer v2.26 (http://getdata-graph-digitizer.com/). The estimated parameters were: CLint(u), Qtissue,i, Vtissue,I (for tissues 1 to 3) and an empirical unbound tissue to plasma partition coefficient (Kpu,tissue). Kpu was assumed to be the same for all the empirical tissues. The Kpu was converted into a Kb value, needed for the model equations, using the following equation [42],

Kfpu, i up Equation 6.5 K bi,  BP where BP is OXY’s blood to plasma ratio.

Figure 6.1 Schematic representation of the PBPK model employed for OXY Predictions. A, Semi-physiological PBPK model employed to describe OXY’s distribution. B, extended minimal Segmented Absorption and Transit (mSAT) model th for oral bioavailability predictions. Vtissue,i, volume of the i tissue; Qtissue,i, blood th flow entering and leaving the i tissue; QHV, hepatic vein blood flow; Qliver, blood flow entering the liver; CLliver, total liver clearance. kGE, gastric emptying rate th constant. For the n intestinal segment: kdiss,n, dissolution rate constant; ka,n, absorption rate constant; CLn, intestinal clearance ; kt,n; transit rate (constant or time varying). More details with respect to the parameters can be found in the Methods Section.

One of the main requisites for the disposition model was that the sum of the blood flows and tissue weights should be equal to the cardiac output (CO) and body weight

(BW), respectively [34]. To implement these constraints, Qtissue,i and Vtissue,i were defined by Equation 6.6 and Equation 6.7, respectively,

Qtissuei,,, f COi (1  f COliver )  C O Equation 6.6

fBWi,,,(1  ( f BWliver  f BWblood ))  BW Equation 6.7 Vtissue, i   tissues

th where fco,i and fBW,i are the fractions of CO and BW for the i empirical tissue, respectively, fCO,liver is the fraction of the cardiac output corresponding to the blood flow entering the liver (Table 6.1), fBW,liver and fBW,blood are the fractions of the BW corresponding to the weight of the liver and blood, respectively and ρtissues is the tissue density (kg/L).The aforementioned fractions were parameterized using a logistic-normal transformation as suggested by Tsamandouras and co-workers [43]. An example of this parametrization is shown in Equation 6.8,

 e 1 Equation 6.8 f  1  ee121  e 2 f  2  ee121 1 f  3  ee121

where θ1 and θ2 are the parameters needed to be estimated (on the logistic scale). This approach not only constrains the individual fractions to be between 0 and 1 but also imposes the constraint that the sum of the fractions should be equal to 1 [43]. The selected parameterization also contributed to the structural identifiability of the model as the parameters needed to be estimated for the empirical tissues (fractions of volumes and blood flows) were also reduced from six to four. The assumption of a single unique Kpu to describe the extent of tissue distribution in all the empirical tissues was also based on identifiability grounds, given that as an independent Kpu for each tissue could not be uniquely identified given the model structure [44, 45]. The latter was corroborated by performing a model structural identifiability analysis prior to the parameter estimation procedure [44, 45].The analysis was done using the “IdentifiabilityAnalysis” package for Mathematica (Wolfram Research, Inc., Champaign, IL, USA) [46, 47], where the model was shown to be at least locally identifiable, i.e., there is a finite set of solutions leading to the same input/output relationship [44, 46, 47]. For details on the Identifiability Analysis procedure and the Mathematica Package, readers are referred to [46, 47]. Finally, the model parameters were estimated in NONMEM version 7.3 (ICON Development Solutions, Hanover, Maryland, USA). The model was implemented using the LSODA 95

differential equation solver (ADVAN13) subroutine in NONMEM [48]. The estimation was performed using the first order conditional estimation with interaction method (FOCE-I), assuming no random effects on the model structural parameters given that only mean data was available for the IV plasma concentration profile [29]. For the estimation, both the data and model output were transformed into natural logarithms (transform both sides) and an additive error model (on the log scale) was assumed for the residual unexplained variability (RUV). CLint(u) and Kpu,tissues were also logged for the purposes of parameter estimation. The latter was done to stabilize the estimation and to prevent the model parameters taking negative values in the differential equations [49, 50].

6.3.1.2 Expansion of the mSAT model for mechanistic bioavailability predictions

The mSAT model is a multi-compartmental absorption model that has been recently proposed and used for the prediction of the fraction absorbed using different intestinal permeability approaches [26]. The model structure was based on the original Compartmental Absorption and Transit (CAT) model developed in the late 90’s by Yu and co-workers [51-53]. The main difference with respect to the CAT model however, was that the new model describes the small intestine with only three anatomically defined compartments, duodenum, jejunum and ileum, instead of seven in the CAT model. In order to adequately describe the mean small intestinal transit time (SITT) with a reduced number of compartments, the mSAT model was implemented with a Weibull transit function that was optimized based upon the same SITT data used for the development of the CAT model [26, 53]. As in the first version of the CAT model, the mSAT model structure was kept relatively simple as the initial goal was only to predict the fraction absorbed based on permeability data [26]. For this study the mSAT model was expanded for the mechanistic prediction of oral absorption and bioavailability following a similar structure to that of the ACAT and ADAM models implemented within GastroPlus™ and the SimCYP® simulator, respectively [8, 54]. The expanded mSAT model structure is shown in Figure 6.1B. The main segments of the mSAT model are stomach (ST), duodenum (DUO), jejunum (JEJ), ileum 96

(ILE), and ascending colon (COL) [26]. For each nth GI segment, the drug amount can be modelled either in the solid (Asolid,n) state or the dissolved state (Adiss,n), where no explicit compartments were implemented for the drug contained in the formulation. The model assumes that all the GI compartments are well mixed [51- 53]; only dissolved drug can be absorbed [55]; absorption can only occur by means of non-saturable process; no significant absorption can occur from the stomach compartment [51]; no drug degradation can occur in the luminal portions of the GI compartments; and the lengths of the GI segment are representative of their anatomical lengths [38, 56]. The model was implemented in Matlab 2014a (The Mathworks Inc., Natick, MA, USA) and the ordinary differential equations (ODE) were numerically evaluated using the ode15s solver for stiff ODEs. The equations describing the drug’s dynamics within the compartments of the mSAT model are summarized below (Equation 6.9 to Equation 6.18) and the model details are given in the following sections. Stomach

dAsolid, st Equation 6.9 STEPst  INPUTt()  k GE  A solidst,  DR st dt

dAdiss, st Equation 6.10 D Rst  k G E  A diss, st dt

Duodenum

dAsolid, duo Equation 6.11 STEPduo  INPUTt()  k GE  A solidst,  DR duo dt

w() tduo A solidduo,

dAdiss, duo Equation 6.12 kGE  A dissst,,,  D R duo (()) w t duo  k aduo  A dissduo dt

Jejunum and ileum (n=2, 4)

dAsolid, n Equation 6.13 STEPn  INPUTt()()  wt n1  A solid , n 1  DR n dt

w() tn A solid, n

dAdiss, n Equation 6.14 w()(()) tn1  A dissn , 1  D R n  w t n  k an ,  A dissn , dt

 1 Equation 6.15  t wt()n  f SITT  f  SITT  LSI,, n LSI n

Ascending colon

dAsolid, col Equation 6.16 STEPcol  INPUTt()()  wt ile  A solidile,  DR col dt

kAtcol,, solidcol

dAdiss, col Equation 6.17 w()() tile  A dissile,,,,  D R col  k tcol  k acol  A disscol dt

All the intestinal segments

dCent, n Equation 6.18 Ventn,,,,,,  k an  A dissn () Q entn  C L entn  C entn dt

6.3.1.2.1 Drug Transit Drug mass in the stomach compartment (solid and dissolved) was transferred to the adjacent segment by means of a first-order process controlled by the gastric emptying rate constant, kGE [57-59]. In the case of disintegrating solid immediate- release (IR) dosage forms (or suspensions) the initial conditions (at t=0) in the stomach compartment (Equation 6.9) were set to the administered dose (Asolid,st(0) = dose). However, in the case of non-disintegrating solid dosage forms, the solid drug was transferred to the adjacent segment by means of a discrete process depending upon the mean residence time (MRT) in the stomach (1/kGE); this also applied to the rest of the GI compartments. For non-disintegrating solid dosage forms, transfer of the solid mass was implemented with a step function (STEP) following the method described by Hénin and co-workers [60].When the STEP function is used the initial conditions of the stomach compartment are set to zero and an input function (INPUT(t)) for the solid mass needs to be used. This INPUT(t) function can take any form, for example a zero 98

order input rate, simulated release profile, an in vitro release profile, etc. More details about the implementation of the STEPn function in the mSAT model can be found in Section A5.1 of the Supplementary Material. For the small intestinal compartments, the transit of the drug particles to the adjacent segment (solid and dissolved) was implemented by a time-varying Weibull function, w(t)n (Equation 6.15), where β and γ are dimensionless coefficients with a value of 2.01 and 1.57, th respectively [26], fLSI,n is the fractional length of the n small intestinal segment

(with respect to the total length of the small intestine, LSI), fLSI,n was assumed as 0.08, 0.37 and 0.55 for duodenum, jejunum and ileum, respectively [56]. For the ascending colon, the drug transit was assumed as a first order process depending upon the ascending colon transit rate constant (kt,col).

6.3.1.2.2 Dissolution and solubility

The segment-dependent dissolution rate (DRn) can be either inputted from in vitro dissolution studies or predicted using derivations of the Noyes-Withney /Nernst- Brunner equation for drug dissolution [61]. Particularly for OXY, the dissolution rate was predicted by using a modification of the model proposed by Wang and Flanagan for spherical particles dissolving over time [54, 62, 63] as shown in Equation 6.19,

dAsolid, n Equation 6.19 DRn   dt 12 33 3 DAAaq  solid( 0 ), n  solidn , Adiss, n 11  S n     r V h()() t r t p0, lum n eff

where Asolid(0),n is the initial amount of solid drug in the given GI segment, this mass was calculated in Matlab by integrating the cumulative amount of solid drug entering each intestinal segment at each iteration, whereas for the stomach compartment this mass was assumed equals to the dose, Daq is the aqueous diffusion coefficient 2 (cm /h), r0 is the initial particle radius (cm), ρp is the particle density (mg/mL), Sn is the segment-dependent aqueous solubility (mg/mL), Vlum,n is the segment-dependent luminal fluid volume (mL), heff(t) is the effective diffusion layer thickness (cm) [64, 65] and r(t) is the particle radius (cm) at a given time. The main assumptions of Equation 6.19 are that the spherical particles are in a well-stirred media, the particles dissolve isotopically, that the total number of particles remains constant across the 99

system, there is an immediate precipitation when the dissolution rate takes positive value (i.e., no super-saturation was allowed in the model) and all the particles have the same initial radius (r0) [62]. The time-varying radius for the dissolving spherical particles was calculated using Equation 6.20 [62, 63, 66]. When the particle radius reached a critical value of 10-9cm, the dissolution was assumed complete and the dissolution rate was assumed to be zero. heff(t), on the other hand, was assumed to be equal to r(t) in the case of particles with radius smaller than 30 µm, otherwise it was assumed equal to an empirical maximum value of 30 µm [54, 65-68]. 1 Equation 6.20 3 Asolid, n r() t r0  A solid( 0 ), n

OXY’s segmental solubility (Sn) was calculated according to its intrinsic solubility

(S0), pKa, and segment-depended pH with the use of the Henderson-Hasselbalch equation for monoprotic bases [69]. Details of such calculations can be found in Section A5.4 of the Supplementary Material.

Luminal fluid volumes (Vlum,n), were determined using an empirical fluid dynamics model with similar characteristics to the model proposed by Jamei and co-workers [54]. The model assumes that fluid movements along the GI tract are driven by gastric emptying and intestinal transit time, taking into account volume fluctuations due to fluid intake, intestinal fluid secretion and reabsorption [54]. The empirical nature of the model is due to the fact that the GI fluid secretion and reabsorption parameters were obtained by fitting the model to free intestinal water data [70]. This data was obtained by magnetic resonance imaging (MRI) after the intake of 240 mL of water in 12 healthy volunteers under fasting conditions [70]. The details of the model and the fitting can be found in the Section A5.2 of the Supplementary Material.

6.3.1.2.3 Intestinal absorption Absorption of the dissolved drug was modelled as a first order process depending upon a segment-specific first order absorption rate constant (ka,n) as shown by Equation 6.21

100

Equation 6.21 2 PDFSAEFIeff   n  ratio, n k an,  rn

where Peff is the drug’s effective intestinal permeability, rn is the radius of the intestinal segment, DF is the degree of flatness coefficient, that accounts for changes in surface area to volume ratio due to the elliptical shape of the human intestine (compared to a cylinder), assumed to be 1.7 as suggested by Sugano (2009) [71], th SAEFn is the ratio between surface area amplification factors of the n intestinal segment with respect to that of the jejunum, where Peff is measured [26, 72]. This ratio takes into account regional variations in drug absorption due to changes in the available mucosal surface area (mSA) and it is necessary for the implementation or segment-dependent intestinal permeation [26]. SAEFn differences were implemented using the Method 3 (M3) proposed by Olivares-Morales and co-workers (2015), where SAEFn takes values of 0.49, 1.00, 0.58 and 0.033 for the duodenum, jejunum, ileum and ascending colon, respectively, Iratio,n is the ratio between the fraction of unionized drug at the segment’s pH with respect to that in the jejunum (Iratio,n = funionized,n/funionized,jejunum). This ratio takes into account regional differences in intestinal permeability due to changes in ionization compared to that of the upper jejunum, where Peff is measured [72, 73]. Iratio,n was calculated using the Henderson- Hasselbalch equation based on the segment’s luminal pH and OXY’s pKa (Section A5.4.1 of the Supplementary Material).

6.3.1.2.4 Enterocyte compartments and intestinal metabolism Mechanistic enterocyte compartments were implemented in the mSAT model to predict OXY’s regional intestinal metabolism. Equation 6.18 shows the general th structure of such compartments where Cent,n is the OXY’s concentration in the n enterocyte compartment (mg/L), Vent,n is the volume of the enterocyte compartment

(L). These volumes were calculated by multiplying the mSAn of the given intestinal segment by the respective enterocyte height (without accounting for the surface area th expansion due to microvilli) [9, 27], Qent,n is the villous blood flow of the n enterocyte compartment (L/h) and CLent,n is the segment-dependent enterocyte clearance. This clearance was calculated according to Equation 6.22

101

m fm CL A Equation 6.22 CL  CYPjint( u ) CYPjentn ( ), ent, n   A j 1 CYPj() liver

where fmCYPj is the fraction of the total intrinsic clearance (CLint(u)) associated to the jth CYP isoform. This value was derived from the in vitro data reported by Mizushima and co-workers [74] and the details of such calculations can be found in

Section A5.4.2 of the Supplementary Material. ACYPj(liver) is the mean liver th abundance of the j CYP isoform (pmol/mg microsomal protein) [75] and ACYPj(ent),n is the absolute abundance of the jth CYP isoform in the nth enterocyte compartment (pmol) [22, 76, 77]. Equation 6.22 assumes that the intrinsic clearances (per pmol of enzyme) are equal in both liver and the intestinal wall [41, 78] and that there is no binding of the drug in the enterocyte compartments (fugut=1) [11, 79]. All the system- related parameters for the mSAT model were derived from the literature and they are summarized in Table 6.1. Finally, the model was implemented with mass balance equations that allowed the estimation of the extent of drug dissolution, absorption (fabs) and first pass metabolism in the intestine and the liver (FG and FH). The extended model was combined with the disposition model developed in the previous section and implemented together in Matlab.

102

Table 6.1. System-related parameters used in the mSAT model Parameter [units] Value Ref. * Reference body weight (BW) [kg] 70 A5.3 Reference height (HT) [m] 1.70 A5.3 Reference body surface area (BSA) 1.81 A5.3 [m2] Tissue density (ρtissue) [kg/L] 1.040 A5.3 Cardiac output (CO) [L/h] 350.37 A5.3 Blood weight [kg] (fraction of BW) 5.53 (0.079) A5.3 Blood volume (Vblood) [L] 5.32 A5.3 Liver-specific parameters Liver weight[kg] (fraction of BW) 1.82 (0.026) A5.3 Liver density (ρliver) [kg/L] 1.080 A5.3 Liver volume(Vliver) [L] 1.69 A5.3 Liver blood flow (Qliver) [L/h] 68.32 (0.195) A5.3 (fraction of CO) Hepatic vein blood flow (QHV) 89.34 (0.255) A5.3 [L/h] (fraction of CO) MPPGL [mg/g] 40 A5.3 CYP abundances (ACYPj(liver),n) A5.3 [pmol/mg] ACYP3A4liver 137 A5.3 ACYP2C9liver 73 A5.3 ACYP2C19liver 14 A5.3 ACYP2D6liver 8 A5.3 GI tract parameters Gastric emptying rate constant (kGE) 4 A5.3 [h-1] Mean intestinal transit time (SITT) 3.32 A5.3 Ascending colon transit rate 0.098 A5.3 -1 constant (kt,col) [h ] Degrees of flatness coefficient (DF) 1.7 A5.3 Small intestinal length (LSI) [cm] 670.7 A5.3 Segment Regional intestinal parameters DUO JEJ ILE COL Radius (rn) [cm] 2.37 1.75 1.5 2.42 A5.3 Length (Ln) [cm] 53.7 248.2 368.9 16.7 A5.3 Fractional length (fLSI,n) 0.08 0.37 0.55 - A5.3 Cylindrical volume [mL] 9.47×102 2.39×103 2.61×103 3.07×102 A5.3 2 4 5 5 3 Mucosal surface area (mSAn) [cm ] 7.50×10 5.19×10 3.86×10 1.62×10 A5.3 Surface area scaling factor 0.49 1.00 0.58 0.033 A5.3 ratio(SAEFn) Enterocyte height [µm] 32.2 32.2 32.2 35.1 A5.3 Enterocyte compartment volume 0.0262 0.119 0.079 8.9×104 A5.3 (Vent,n)[L] Enterocyte compartment blood flow 1.33 6.24 9.25 4.20 A5.3 (Qent,n) [L/h] (0.0038) (0.0178) (0.0264) (0.012) (fraction of CO) Intestinal CYP abundances (ACYPj(ent),n) [pmol] ACYP3A4ent,n 9,110 36,060 21,030 0 A5.3 ACYP2C9ent,n 1770 7,030 4,100 0 A5.3 ACYP2C19ent,n 210 820 480 0 A5.3

103

ACYP2D6ent,n 110 440 240 0 A5.3 Segment Regional mSAT luminal ST DUO JEJ ILE COL parameters pH 1.5 6.4 6.6 7.1 6.5 A5.3 Baseline water volume (Vlum,n) [mL] 35 6 13 24 5 A5.3 *,More details and references for the system-related parameters can be found in Section A5.3 of the Supplementary Material (Table A5.3).

104

6.3.2 OXY’s oral PK simulations and relative bioavailability predictions

OXY oral pharmacokinetic profile was mechanistically predicted using the combined mSAT absorption and disposition model (Figure 6.1). For the predictions, OXY’s drug-specific input parameters were either derived from literature or calculated using in silico equations. A summary of the parameters and their values are provided in Table 6.2. More details about the parameter selection and/or calculations can be found in Section A5.4 of the Supplementary Material.

Two different oral pharmacokinetic studies were predicted using the mSAT model. In the first study, the pharmacokinetics of OXY was investigated after the administration of three 5 mg IR tablets (every 5 hours) to eight healthy volunteers [29]. The tablets were administered under fasting conditions together with 100 mL of water and blood samples were collected up to 15 h post dose. OXY plasma concentrations were measured using a validated assay [29]. This study was simulated with the intention to evaluate the model capacity to mechanistically predict OXY’s oral pharmacokinetic, particularly given the fact that some of the study participants were also part of the IV study used for the development of OXY’s disposition model [29]. The model prediction were evaluated graphically and contrasted with observed mean clinical data in terms of accuracy of the prediction of mean pharmacokinetic parameters such as, absolute bioavailability (F), area under de curve (AUC0-t) and maximum plasma concentration (Cmax).

The second study was a relative bioavailability study between OXY’s IR tablet and its controlled-release (OROS) formulation. The clinical data of this study was kindly provided by Janssen Pharmaceutica. The study was conducted in 41 healthy volunteers, males and females, where each subject received three 5 mg IR release tablets (every 8 hours) and one 10 mg controlled-release (OROS) formulation in a cross-over fashion. The formulations were administered in the morning under fasted conditions together with a 240 mL glass of water. Blood samples were collected up to 48 hours post dose and were analysed for OXY using a validated LC-MS/MS assay [80]. In contrast to the previous study, a stereo-selective assay was employed for the determination of OXY plasma concentrations and the concentrations of OXY enantiomers, R and S-oxybutynin, were reported [80]. It has been shown that OXY 105

enantiomers display stereo-selective pharmacokinetics, mainly attributable to differences in volume of distribution and clearance between the enantiomers [74]. To overcome this issue the aforementioned differences were accommodated into the mSAT model predictions, adapting OXY’s parameters with the observed in vitro parameters for each enantiomer [74, 81]. For instance, changes in volume of distribution between enantiomers were mainly attributed to plasma protein binding differences [74, 81]. Hence, stereo-selective plasma protein binding was implemented in the model by employing the reported fup for each enantiomer (Table

6.2). This led to changes in the volume of distribution as the Kb,tissue is related to fup by Equation 6.5. The clearance differences, on the other hand, were accommodated assuming that the CLint,(u) of the racemic mixture represented an average between the intrinsic clearances of the enantiomers. Thus the enantiomer-specific CLint(u) was calculated by multiplying the CLint(u) value of the racemic mixture by the CLint(u) ratio between the enantiomers (R/SCLint,ratio). The latter parameter was defined for each enantiomer as the ratio between the observed in vitro CLint(u) of the given enantiomer and the average in vitro CLint(u) between them. More details on the derivation of the aforementioned parameters can be found in Section A5.4.4 of the Supplementary Material. For the relative bioavailability predictions, both IR and OROS formulations were simulated as per the study protocol, whereas an additional IV profile was simulated to estimate the absolute bioavailability of each formulation. The drug was assumed to be administered as a racemic mixture for prediction of the luminal processes (i.e., dissolution, release). However, once the drug entered the intestinal wall compartments of the mSAT model, the dose was divided into enantiomer fractions. The compartments within the mSAT model where used to prospectively estimate bioavailability fractions for each formulation (fa, fG and fH). In the case of the OROS formulation, the observed mean in vitro release profile was used as the INPUT function (Equation 6.9). The in vitro release profile was digitized from the literature [80, 82]. This profile was measured in an USP apparatus VII in different dissolution media such as water, simulated gastric fluid (SGF) and simulated intestinal fluid (SIF) [82]. Given the results of the release study, it was assumed that OXY’s release rate from the OROS formulation was not affected by luminal changes in pH and/or fluid volumes [82]. The INPUTn profile can be found in the Section A5.4.3 of the

106

Supplementary Material. Since OXY content in the OROS is released as a suspension [83, 84], the dissolution rate of the suspended drug was calculated using Equation 6.19. Lastly, an additional simulation was conducted for the OROS formulation using the classical approach for intestinal permeability, i.e., Method 1 (M1) in [26], SAEFn ratio was assumed 1 in all segments. This was done in order to evaluate the benefits of implementing the segment-dependent permeability approach for the prediction of OXY’s pharmacokinetics when administered as an OROS formulation [26].

107

Table 6.2. OXY’ drug-related parameters employed for the simulations (racemic mixture and isomers)

Parameter [units] OXY R-OXY S-OXY Ref.** racemic Molecular weight (MW) [g/mol] 357.5 - - A5.5 pKa (base) 8.04 - - A5.5 LogD7.4 2.98 - - A5.5 LogP 3.71 - - A5.5 Intrinsic solubility(S0) [mg/mL] 0.012 - - A5.5 3 Particle density (ρp) [mg/mL] 1.20×10 - - A5.5 -3 Initial particle radius (r0) [cm] 1.00×10 - - A5.5 Aqueous diffusion coefficient 0.025 - - A5.5 2 (Daq)[cm /h] Blood to plasma ratio (BP) 0.686 0.71 0.682 A5.5 -3 -3 -3 Fraction unbound in plasma (fup) 3.40×10 4.70×10 2.75×10 A5.5 -6 Apparent permeability (Papp) [×10 21.9 - - A5.5 cm/s] Jejunal effective permeability (Peff) 4.31 - - A5.5 [×10-4 cm/s] 3 3 3 Kpu(liver) 1.18×10 1.00×10 1.53×10 A5.5 Kpu,tissues Estimated - - See methods CLint(u) Estimated - - See methods R/SCLint,ratio 1 0.89 1.1 A5.5 fmCYP3A4 0.80 0.79 0.81 A5.5 fmCYP2C9 0.12 0.13 0.11 A5.5 fmCYP2C19 0.07 0.08 0.05 A5.5 fmCYP2D6 0.01 0.00 0.03 A5.5 Segment Regional luminal parameters ST DUO JEJ ILE COL Segmental solubility (Sn) [mg/mL] 12 0.54 0.34 0.12 0.43 A5.5 Luminal ionization ratio (Iratio,n) n/a 0.64 1.00 2.94 0.80 A5.5 -, the parameter was the same as for racemic oxybutynin. **, Details and the sources of each parameter value can be found in Section A5.4.5 of the Supplementary Material (Table A5.5)

6.4 Results

6.4.1 OXY’s disposition parameter estimation

The estimation of the parameters for the semi-physiological disposition model was successfully completed and the parameters estimates are summarized in Table 6.3. The structural model parameters were obtained with relative good precision; all the relative standard errors (RSE) were below 25%.The model provided a good fit to the infusion data as in shown in Figure 6.2 [29]. It is worth mentioning that given the 108

model structure and its local identifiability properties [44, 46, 47] an exchange between the parameters, Qtissue,1to Qtissue,2, and Vtissue,1to Vtissue,2, can be made without affecting either fit or the input-output relationship of the model. The NONMEM code for the estimation can be found in Section A5.6 of the Supplementary Material.

Table 6.3. Estimated OXY disposition parameters from the IV infusion fit.

Parameter Estimate RSE (%)

CLint(u) [µL/min/mg] 3,944 5.0% a Qtissue,1(fCO,1) [L/h] 9.76 (0.0346) 23% a Qtissue,2(fCO,2) [L/h] 253 (0.897) 1.9% a Qtissue,3(fCO,3) [L/h] 19.4 (0.0688) 16% a Vtissue,1(fBW,1) [L] 50.8 (0.775) 3.1% a Vtissue,1(fBW,1) [L] 6.42 (0.098) 19% a Vtissue,1(fBW,1) [L] 8.33 (0.127) 8.9%

Kpu,tissues 469 10% RUV [%CV]b 6.54 35% a values in parenthesis represent the estimated fractions of CO and BW. The relative standard errors (RSE) were calculated as: 100× (standard error/estimate).

For CLint(u) and Kpu,tissue the RSE are reported in the normal scale, these were calculated using normal/log-normal reverse algebra. The estimates of the fractions of the CO and BW (hence blood flows and volumes) were obtained on the logistic scale. These values were transformed back to fractions, in the logistic-normal scale, using Equation 6.8. The RSE for such fractions were obtained by simulations:1×107 random samples were drawn from a multivariate-normal distribution and then transformed back to the logistic-normal scale, where summary statistics (mean and standard deviations) of the resulting vectors were calculated using Matlab 2014a. More details of the approach are given in [43].

2 b RUV is expressed as a coefficient of variation (%CV) calculated as: 100e  1

109

Figure 6.2 OXY’s disposition model fit to the 5 mg IV infusion data. The solid black line represents the mSAT model prediction whereas the solid grey circles and error bars are the respective mean and standard error of the mean (SEM) of the observed clinical data ( digitized from [29]). The insert shows the same plot on the semi- logarithmic scale.

6.4.2 Mechanistic prediction of OXY’s oral pharmacokinetics

The predicted oral pharmacokinetic profile of OXY’s IR tablet (3 doses, every 5 h) is shown in Figure 6.3. There was a good agreement between the observed clinical data and the mSAT model prediction [29]. The predicted AUC0-15h was 32.5 ng×h/mL, comparable with the reported value of 33.7 ± 7.9ng×h/mL [29]. Cmax, on the other hand, was predicted to be 6.24 ng/mL whereas the reported Cmax was 7.55 ± 2.22 ng/mL [29]. The estimated absolute bioavailability for the racemic OXY was 5.94%, whereas the reported absolute bioavailability was 6.2 ± 1.2 % [29].

110

Figure 6.3 Model predictions of OXY oral pharmacokinetics (racemic) after a multiple dose administration. The solid black line represents the mSAT model prediction whereas the solid grey circles and error bars are the respective mean and standard error of the mean (SEM) of the observed clinical data (digitized from [29]).

6.4.3 Relative bioavailability between IR and OROS formulation

The mSAT model was able to predict the pharmacokinetics of OXY enantiomers when formulated as IR and OROS. The pharmacokinetic predictions of the R isomer (R-OXY) are shown Figure 6.4 whereas the predictions of S-OXY can be found in the Supplementary Material (Figure A5.6). Table 6.4 summarizes the observed and predicted relevant pharmacokinetic parameters, stratified by formulation and enantiomers. There was a good agreement between the observed clinical data and the mSAT predictions (Figure 6.4A and Figure A5.6). Nevertheless, a general trend towards the underestimation of the oral pharmacokinetics of the IR formulation could be observed for both enantiomers (Figure 6.4A and Figure A5.6A). This underestimation tended to be more prominent for the S enantiomer than for the R enantiomer, both in terms of Cmax and AUC0-48 (Table 6.4). The predicted Cmax and

AUC0-48 were within ±20% of the observed values for R-OXY, whereas the bias for the S enantiomer was -51% and -37% for Cmax and AUC0-48, respectively. For the OROS formulations, underestimation of the pharmacokinetic parameters was also observed, though to a lesser extent than for the IR formulation. The latter was 111

particularly observed for the R enantiomer (Figure 6.4B) where both AUC0-48 and

Cmax were within 6% of the observed values The parameters for the S enantiomer, on the other hand, showed a bias between 20 to

28% for AUC0-48 and Cmax, respectively. The relative bioavailability (Frel) predictions of the OROS formulation were consistent with that of the observed values, both enantiomers displayed higher Frel when formulated as OROS compared to the IR. The Frel predictions tended to be overestimated by a factor of 20%. Mechanistic predictions of the intestinal processes driving oral absorption and bioavailability are shown in Figure 6.5 and the parameters are summarized in (Table 6.4). Both enantiomers were predicted to be well absorbed from the IR formulations, the fabs values were close to 1. For the OROS formulation, this fraction was predicted to be reduced to by almost a 70% (Table 6.4)The predicted intestinal first-pass metabolism for the IR formulation was high (FG≤ 0.13), whereas for the OROS formulations the FG was predicted to be almost 8 fold higher. For the IR formulation, the majority of the absorption and intestinal elimination was predicted to occur in the jejunum segment (Figure 6.5), while for the OROS formulation the absorption was predicted to occur mainly in the distal ileum and ascending colon, with a limited intestinal first-pass only in the ileal segment. The predicted hepatic availability (FH) for both formulations was practically the same.

112

Figure 6.4 mSAT model prediction of the pharmacokinetic of R-OXY after the administration of three 5 mg IR formulations (A) and one 10 mg OROS formulation (B). The solid black line represents the mSAT model prediction, whereas the solid grey circles and lines represent the individual observed clinical data (kindly provided by Janssen Pharmaceutica). The insert shows the same plot in the semi-logarithmic scale.

113

Figure 6.5 mSAT predicted segmental and oral bioavailability fractions for R-OXY IR and OROS formulations. A, predicted fabs; dark blue bars represent the IR formulation and the light blue bars represent the OROS formulation. A, predicted fraction of the administered dose metabolized in the intestinal segments (EG); dark green bars represent the IR formulation and the light green bars represent the OROS formulation.

114

Table 6.4. Summary of the mSAT predicted vs. observed pharmacokinetic parameters for OXY formulations (IR and OROS) Observed values mSAT model predictions (mean ±SD) AUC C AUC C Isomer Formulation 0-48 max F 0-48 max F F f F F [ng×h/mL] [ng/mL] rel [ng×h/mL] [ng/mL] rel abs G H IR 21.7 ±13.0 3.28 ± 2.16 n/a 17.3 2.71 n/a 0.078 1.00 0.13 0.61 R-OXY OROS 18.6 ± 10.5 0.99 ± 0.59 1.39 ± 0.44 19.5 0.93 1.70 0.13 0.28 0.80 0.58 IR 31.6 ± 15.8 6.58 ± 3.71 n/a 20.2 3.24 n/a 0.075 1.00 0.10 0.71 S-OXY OROS 34.4 ±17.3 1.84 ± 0.97 1.72 ± 0.49 27.4 1.32 2.04 0.15 0.28 0.79 0.69 SD, standard deviation.

115

Finally, Figure 6.6 shows the simulated pharmacokinetic profile obtained employing the two different permeability approaches (R-OXY). It can be noticed that when using the classical approach for permeability, i.e., employing the same Peff values for all the intestinal segments, a significant overestimation of the distal absorption would have been predicted for OXY. Therefore the use of the newly proposed regional permeability approach (M3) gives rise to a much more accurate prospective prediction [26], particularly important given the type of nature of the drug (BCS class 1) and the prolonged release from the OROS formulation (24 h).

Figure 6.6 mSAT model prediction of the pharmacokinetic of R-OXY after the administration of a 10 mg OROS formulation using two different permeability approaches. The solid black line represents the mSAT model prediction using segment-dependent permeability (M3 method), whereas the solid blue line is the model prediction using the same Peff value in all the intestinal segments (M1 method)[26]. Circles and lines represent the individual observed clinical data (kindly provided by Janssen Pharmaceutica).

6.5 Discussion

The recently proposed mSAT model was successfully expanded and applied to the mechanistic prediction of the oral bioavailability differences between OXY’s IR and OROS formulation in a purely prospective fashion. The approach employed herein involved several steps that were necessary to appropriately distinguish between OXY’s disposition and absorption processes in the model predictions. 116

The first step consisted in the development of a disposition model capable of describing OXY’s distribution and elimination processes as well as to mechanistically account for its pre-systemic extraction when combined with the absorption model. Several structures were evaluated for the disposition model, from a three-compartment model with a mechanistic liver compartment to a whole body PBPK model. The chosen model was that of Figure 6.1A. This model represents a hybrid between the full PBPK model and the compartmental model, providing the flexibility of the compartmental model for the fitting purposes, yet retaining the mechanistic nature of the PBPK model where it was needed, such as in the liver compartment. As previously described, the model was constrained in terms of the volumes of the empirical tissues and their blood flows following the approach suggested by Cao and Jusko (2012). These constraints were implemented during the estimation process in NONMEM with the help of the logistic-normal transformations proposed by Tsamandouras and co-workers (2015). As shown in Figure 6.2, the fitted model provided an accurate description of the observed IV infusion data [29] and the model parameters were precisely estimated (Table 6.3).

The selected modelling approach required the fitting of both the intrinsic microsomal clearance and an empirical Kpu value. The former could have been informed purely from the reported in vitro data, where the unbound in vitro formation CLint in HLM has been reported to be 225 and 278 µL/min/mg for R-OXY and S-OXY, respectively [74]. However, there were indications that suggested that the use of such in vitro value would have led to under-predictions in both the systemic clearance and first pass metabolism of OXY [21]. This underestimation has been previously reported in the literature, particularly for highly cleared and highly protein bound substrates similar to OXY [85, 86]. For these reasons fitting the microsomal clearance to the IV infusion data seemed to be a reasonable choice. This was supported by the estimated CLint of 3,944 µL/min/mg for racemic OXY (Table 6.3). However, this estimate assumes that only the CYP-mediated metabolism was responsible for OXY’s elimination, which might not necessarily true as there is evidence of the formation of an inactive metabolite, 2-cyclohexyl-2-phenylglycolic acid (CPGA), which is mediated by Carboxylesterases (CES2) that can be found in both the liver and the intestinal wall [87, 88]. Nevertheless, the CPGA formation was

117

assumed to contribute minimally to OXY’s overall elimination, given that the reported in vitro CPGA formation CLint in HLM was 5.9 µL/min/mg; a value considerably smaller that the reported formation CLint of OXY’s main metabolite, NDEO [74, 88].

The fitted Kpu value was assumed to be the same for all the empirical tissues. This assumption was mainly based on model structural identifiability grounds that otherwise would have deemed the model as unidentifiable [44, 45]. The decision to estimate a Kpu value, instead of directly estimating the Kb needed in the model equations, was to allow for the prediction of the pharmacokinetics of OXY enantiomers. While there is a difference in plasma binding between the two enantiomers, we have assumed that the Kpu value to be the same for both the racemic OXY and its enantiomers (R and S-OXY).This assumption was based on the fact that

OXY, a strong base (pKa>7), is predominantly bound to α1-acidglycoprotein (AGP), which is mainly located in the circulating plasma. Thus the impact of OXY’s binding to AGP within tissues was assumed to be minimal [81, 89]. Consequently, differences in OXY’s volume of distribution can be mainly attributed to plasma protein binding differences between the enantiomers (and racemic mixture), rather than to tissue binding [74, 81, 89]. These differences were accommodated in the disposition model by using the enantiomer specific fup value in the calculation of the

Kb, which in turn affects the volume of distribution (Equation 6.5). The latter also highlights the usefulness of the chosen hybrid-PBPK model structure employed in this work, as this extrapolation could have not been possible when using a traditional three-compartment model to describe OXY’s disposition.

The second step in the modelling approach involved the expansion of the mSAT model for the mechanistic prediction of OXY oral bioavailability in a purely prospective manner. As previously described in the methods section, the model was expanded by incorporating mechanistic equations for the prediction of the most relevant absorption processes using Matlab: transit, release, dissolution, permeation and intestinal metabolism. Using the expanded mSAT model, the predictions of the oral pharmacokinetic profile of the racemic OXY after the administration of three 5 mg IR tables in the “reference individual” were in good agreement with the mean observed data (Figure 6.3). Similarly, the predicted absolute bioavailability of the

118

racemic OXY was within 5% of the reported mean value [29]. Since OXY’s disposition was already accounted by the model in the previous step, the predictions of oral bioavailability were mainly attributed to the mechanistic nature of the mSAT model and the use of the appropriate system-related and drug-specific parameters for such predictions (Table 6.1 and Table 6.2); the latter provided further support for the use of this model for mechanistic oral absorption predictions [26]. Since OXY belongs to Class 1 within the BCS (highly permeable and soluble), there were no major difficulties in terms of the prediction of the dissolution and permeation processes for the IR formulation within the mSAT model [21]. These processes were not considered to be the rate limiting step for OXY’s absorption [29]. In contrast, prediction of OXY’s first pass metabolism was considered the key step for the bioavailability predictions of OXY’s IR formulation. Judging by the outcome of the bioavailability predictions, the expanded mSAT model did a reasonable job for such predictions. The latter also highlights the importance of having good disposition data available for the purpose of judging mechanistic bioavailability predictions, particularly due to the possible confounding issues that can arise from the use of the oral data to calibrate some of the absorption parameters in the model, particularly when the predictions are not in agreement with the observed clinical data. The latter might lead to biased estimates of the absorption parameters due to the fact that the disposition parameters were biased in the first instance [43, 90].

Once the ability of the mSAT model to prospectively predict the oral pharmacokinetic profile was established, the third and final step of the modelling approach was the mechanistic investigation of the relative bioavailability between OXY’s IR and OROS formulations. The pharmacokinetic profiles of both formulations were successfully predicted by the model for the “reference individual” as shown in Figure 6.4 and summarized in Table 6.4. However, as it can be seen in Figure 6.4, there is a large inter-individual variability (IIV) in the observed pharmacokinetic profiles. This IIV was not captured by the mSAT model as it can only provide predictions for the “reference individual”. This is in contrast to available software packages that can include such variability in the model predictions such as SimCYP® and PK-Sim®.

119

In terms of relative bioavailability, the model predictions were consistent with the observed trend of higher exposure when using the OROS formulation compared to the IR tablets (Table 6.4)[28, 32]. However the model slightly overestimated the magnitude of the relative bioavailability of the OROS formulation, mainly due to a 0.6 fold underestimation of the exposure of the IR formulations (particularly for S- OXY). This underestimation could be due to the assumptions made with regards to the intrinsic clearances of each isomer in the model predictions, such as the use of the in vitro R/SCLint,ratio to derive the isomer-specific CLint(u) employed in the mSAT model. However, considering that these predictions were based purely on available literature data and no further optimization were made in any of absorption model parameters, this bias was considered acceptable.

From a mechanistic perspective, the model predictions with regards to the bioavailability fractions were consistent with the previous multi-factorial simulation study [21]. The predicted fabs of R-OXY and S-OXY were close to the unity, in agreement with the mass balance studies suggesting the complete absorption of the racemic mixture [29]. Due to the use of the differential permeability approach employed in the model, the segmental contribution to the overall absorption can be estimated for the different formulations [26]. Most of the drug content from the IR formulation was predicted to be absorbed in the upper GI segments (Figure 6.5A). Similarly, predictions can be made with regards to the fate of the drug in the enterocyte compartments (Figure 6.5B) where most of the drug content was expected to be eliminated by the CYP enzymes located in the upper regions of the GI segments [21-24]. As shown in Table 6.4, the predicted FG of the IR formulation was approximately 0.13 for both isomers. This value was in close agreement with the observed FG of 0.11 to 0.14; calculated from the reported absolute bioavailability and the in vivo clearance using the standard formulae [29, 91]. The model predicted an FH close to 0.61 and 0.71 for IR formulation of R-OXY and S-OXY, respectively. These values were slightly higher than the estimated FH between 0.44 and 0.58 for the racemic mixture [29, 91]. The segmented structure of the mSAT model was necessary for the bioavailability predictions of the OXY’s OROS formulation, particularly when the model structure was combined with the regional permeability approach (M3) [26]. Otherwise the absorption of the OROS formulation would have

120

been considerably overestimated when using the same Peff value for all the segments of the mSAT model, as shown in Figure 6.6. This highlights the importance of accounting for the segmental-permeability difference in the PBPK framework [26, 72, 92]. Even though there is no available clinical data with regards to the regional absorption and intestinal first pass metabolism of the OXY’s OROS formulations. The model outcome can be used to explain the mechanism proposed for the higher relative bioavailability of OXY’s OROS formulation. It was clear from Figure 6.5 and Table

6.4 that the fabs for the OROS formulation was considerably reduced compared to the

IR formulation. However, the predicted FG was almost eight times higher for the

OROS formulation compared to the IR. Since the predicted FH remained almost the same for both formulations (IR and OROS), the model provided strong support for the hypothesis that the main factor responsible to the higher relative bioavailability observed for OXY’s OROS formulation was an increased intestinal availability [21, 28]. This increased availability was mainly due to release of the majority of the drug content from the OROS formulation in the distal regions of the GI tract, where the abundance of CYP is reduced compared to the upper GI tract [22-24]. This outcome not only corroborates the findings from previous work, but also can be seen as a further support to the fact that there is an interesting interplay between absorption and metabolism along the GI that can be explored in formulation development. This phenomenon could be of importance for substrates similar to OXY, where a MR release formulation can be developed to either increase or maintain exposure levels observed with the IR counterparts.

One of the main limitations of this study was the lack of predictions including variability and uncertainty. This variability might be incorporated in the model through the model parameters either from literature sources or estimated from data and it can be of importance to understand the possible differences in OXY’s FG to its full extent. Another important aspect that was not accounted for in this work was the fate of OXY’s main metabolite, DEOB. This metabolite has clinical implications as it has been associated with the occurrence of OXY’s side effects [28, 32, 33]. However, there was a paucity of data available in the literature regarding the metabolite and did not allow its incorporation in the current model using a “bottom-

121

up” approach. Finally, the results highlighted the usefulness of the newly proposed mSAT model for oral pharmacokinetic predictions. However, these result should be considered carefully as more compounds need to be tested with this particular model.

6.6 Conclusion

A newly proposed mechanistic absorption model was expanded and employed for the “bottom-up” predictions of oral bioavailability of oxybutynin IR and OROS formulations. The new model was able to capture the bioavailability differences observed between OXY formulations, where the OROS formulation displayed higher relative bioavailability than its IR counterpart. The model predictions suggest that this higher bioavailability was mainly due to an increased intestinal availability (FG) product of the decreased intestinal first-pass metabolism in the distal regions of the GI tract, where the abundance of the CYP enzymes are decreased.

6.7 Acknowledgements

A.O-M. is recipient of a Ph.D. grant awarded by CONICYT Chile, Chilean Ministry of Education and a President’s Doctoral Scholar Award from The University of Manchester. The authors would like acknowledge Janssen Pharmaceutica for providing the observed clinical data for the relative bioavailability between oxybutynin’s IR and OROS formulation, as well as Drs Luca Marciani (University of Nottingham) and Deana Mudie (University of Michigan) for providing insight on the luminal free water data employed in the development of the fluid dynamics model implemented in the mSAT model. Finally we would acknowledge the fruitful comments and discussions made with members of the Centre for Applied Pharmacokinetic Research (CAPKR) of The University of Manchester, particularly Thierry Wendling, Nikos Tsamandouras and Aleksandra Galetin. This project is an in kind contribution from the University of Manchester to the IMI Oral Biopharmaceutical Tools (OrBiTo) project (http://www.imi.europa.eu/content/orbito).

122

6.8 References

1. M. Rowland, C. Peck, and G. Tucker. Physiologically-based pharmacokinetics in drug development and regulatory science. Annu Rev Pharmacol Toxicol. 51:45-73 (2011). 2. P. Zhao, M. Rowland, and S.M. Huang. Best Practice in the Use of Physiologically Based Pharmacokinetic Modeling and Simulation to Address Clinical Pharmacology Regulatory Questions. Clin Pharmacol Ther. 92:17-20 (2012). 3. P. Zhao, L. Zhang, J.A. Grillo, Q. Liu, J.M. Bullock, Y.J. Moon, P. Song, S.S. Brar, R. Madabushi, T.C. Wu, B.P. Booth, N.A. Rahman, K.S. Reynolds, E. Gil Berglund, L.J. Lesko, and S.M. Huang. Applications of Physiologically Based Pharmacokinetic (PBPK) Modeling and Simulation During Regulatory Review. Clin Pharmacol Ther. 89:259-267 (2011). 4. H.M. Jones, Y. Chen, C. Gibson, T. Heimbach, N. Parrott, S.A. Peters, J. Snoeys, V.V. Upreti, M. Zheng, and S.D. Hall. Physiologically based pharmacokinetic modeling in drug discovery and development: A pharmaceutical industry perspective. Clin Pharmacol Ther. 10.1002/cpt.37:n/a-n/a (2014). 5. A. Rostami-Hodjegan, I. Tamai, and K.S. Pang. Physiologically based pharmacokinetic (PBPK) modeling: It is here to stay! Biopharm Drug Dispos. 33:47-50 (2012). 6. A. Rostami-Hodjegan. Physiologically based pharmacokinetics joined with in vitro-in vivo extrapolation of ADME: a marriage under the arch of systems pharmacology. Clin Pharmacol Ther. 92:50-61 (2012). 7. M. Jamei, S. Marciniak, K. Feng, A. Barnett, G. Tucker, and A. Rostami- Hodjegan. The Simcyp population-based ADME simulator. Expert Opin Drug Metab Toxicol. 5:211-223 (2009). 8. B. Agoram, W.S. Woltosz, and M.B. Bolger. Predicting the impact of physiological and biochemical processes on oral drug bioavailability. Adv Drug Deliv Rev. 50 Suppl 1:S41-67 (2001). 9. K. Thelen, K. Coboeken, S. Willmann, R. Burghaus, J.B. Dressman, and J. Lippert. Evolution of a detailed physiological model to simulate the gastrointestinal transit and absorption process in humans, part 1: oral solutions. J Pharm Sci. 100:5324-5345 (2011). 10. S. Willmann, W. Schmitt, J. Keldenich, J. Lippert, and J.B. Dressman. A physiological model for the estimation of the fraction dose absorbed in humans. J Med Chem. 47:4022-4031 (2004). 11. M. Gertz, J.B. Houston, and A. Galetin. Physiologically based pharmacokinetic modeling of intestinal first-pass metabolism of CYP3A substrates with high intestinal extraction. Drug Metab Dispos. 39:1633-1642 (2011). 12. E. Chen, G. Tai, and H. Ellens. The Importance of Villous Physiology and Morphology in Mechanistic Physiologically-Based Pharmacokinetic Models. Pharm Res. 31:305-321 (2014). 13. E. Sjogren, J. Westergren, I. Grant, G. Hanisch, L. Lindfors, H. Lennernas, B. Abrahamsson, and C. Tannergren. In silico predictions of gastrointestinal

123

drug absorption in pharmaceutical product development: application of the mechanistic absorption model GI-Sim. Eur J Pharm Sci. 49:679-698 (2013). 14. D. Cong, M. Doherty, and K.S. Pang. A New Physiologically Based, Segregated-Flow Model to Explain Route-Dependent Intestinal Metabolism. Drug Metab Dispos. 28:224-235 (2000). 15. A.S. Darwich, S. Neuhoff, M. Jamei, and A. Rostami-Hodjegan. Interplay of metabolism and transport in determining oral drug absorption and gut wall metabolism: a simulation assessment using the "Advanced Dissolution, Absorption, Metabolism (ADAM)" model. Curr Drug Metab. 11:716-729 (2010). 16. M.N. Martinez and G.L. Amidon. A mechanistic approach to understanding the factors affecting drug absorption: a review of fundamentals. J Clin Pharmacol. 42:620-643 (2002). 17. I. Poggesi, J. Snoeys, and A. Van Peer. The successes and failures of physiologically based pharmacokinetic modeling: there is room for improvement. Expert Opin Drug Metab Toxicol. 0:1-5 (2014). 18. P. Poulin, R.D.O. Jones, H.M. Jones, C.R. Gibson, M. Rowland, J.Y. Chien, B.J. Ring, K.K. Adkison, M.S. Ku, H. He, R. Vuppugalla, P. Marathe, V. Fischer, S. Dutta, V.K. Sinha, T. Björnsson, T. Lavé, and J.W.T. Yates. PHRMA CPCDC initiative on predictive models of human pharmacokinetics, part 5: Prediction of plasma concentration–time profiles in human by using the physiologically-based pharmacokinetic modeling approach. J Pharm Sci. 100:4127-4157 (2011). 19. N. Parrott, N. Paquereau, P. Coassolo, and T. Lave. An evaluation of the utility of physiologically based models of pharmacokinetics in early drug discovery. J Pharm Sci. 94:2327-2343 (2005). 20. H. Lennernas, L. Aarons, P. Augustijns, S. Beato, M. Bolger, K. Box, M. Brewster, J. Butler, J. Dressman, R. Holm, K. Julia Frank, R. Kendall, P. Langguth, J. Sydor, A. Lindahl, M. McAllister, U. Muenster, A. Mullertz, K. Ojala, X. Pepin, C. Reppas, A. Rostami-Hodjegan, M. Verwei, W. Weitschies, C. Wilson, C. Karlsson, and B. Abrahamsson. Oral biopharmaceutics tools - time for a new initiative - an introduction to the IMI project OrBiTo. Eur J Pharm Sci. 57:292-299 (2014). 21. A. Olivares-Morales, Y. Kamiyama, A.S. Darwich, L. Aarons, and A. Rostami-Hodjegan. Analysis of the impact of controlled release formulations on oral drug absorption, gut wall metabolism and relative bioavailability of CYP3A substrates using a physiologically-based pharmacokinetic model. Eur J Pharm Sci. 67:32-44 (2015). 22. M.F. Paine, M. Khalighi, J.M. Fisher, D.D. Shen, K.L. Kunze, C.L. Marsh, J.D. Perkins, and K.E. Thummel. Characterization of Interintestinal and Intraintestinal Variations in Human CYP3A-Dependent Metabolism. J Pharmacol Exp Ther. 283:1552-1562 (1997). 23. J.C. Kolars, K.S. Lown, P. Schmiedlin-Ren, M. Ghosh, C. Fang, S.A. Wrighton, R.M. Merion, and P.B. Watkins. CYP3A gene expression in human gut epithelium. Pharmacogenetics. 4:247-259 (1994). 24. A. Bruyere, X. Decleves, F. Bouzom, K. Ball, C. Marques, X. Treton, M. Pocard, P. Valleur, Y. Bouhnik, Y. Panis, J.M. Scherrmann, and S. Mouly. Effect of variations in the amounts of P-glycoprotein (ABCB1), BCRP (ABCG2) and CYP3A4 along the human small intestine on PBPK models for predicting intestinal first pass. Mol Pharm. 7:1596-1607 (2010).

124

25. C. Tannergren, A. Bergendal, H. Lennernas, and B. Abrahamsson. Toward an increased understanding of the barriers to colonic drug absorption in humans: implications for early controlled release candidate assessment. Mol Pharm. 6:60-73 (2009). 26. A. Olivares-Morales, H. Lennernas, L. Aarons, and A. Rostami-Hodjegan. Translating Human Effective Jejunal Intestinal Permeability to Surface- Dependent Intrinsic Permeability: a Pragmatic Method for a More Mechanistic Prediction of Regional Oral Drug Absorption. AAPS J. 17:1177- 1192 (2015). 27. H.F. Helander and L. Fandriks. Surface area of the digestive tract - revisited. Scand J Gastroenterol. 49:681-689 (2014). 28. S.K. Gupta and G. Sathyan. Pharmacokinetics of an oral once-a-day controlled-release oxybutynin formulation compared with immediate-release oxybutynin. J Clin Pharmacol. 39:289-296 (1999). 29. J. Douchamps, F. Derenne, A. Stockis, D. Gangji, M. Juvent, and A. Herchuelz. The pharmacokinetics of oxybutynin in man. Eur J Clin Pharmacol. 35:515-520 (1988). 30. E. Miyamoto, S. Kawashima, Y. Murata, Y. Yamada, Y. Demizu, H. Kontani, and T. Sakai. Physico-chemical properties of oxybutynin. Analyst. 119:1489-1492 (1994). 31. M. Yaich, M. Popon, Y. Medard, and E.J. Aigrain. In-vitro cytochrome P450 dependent metabolism of oxybutynin to N-deethyloxybutynin in humans. Pharmacogenetics. 8:449-451 (1998). 32. G. Sathyan, M.B. Chancellor, and S.K. Gupta. Effect of OROS® controlled- release delivery on the pharmacokinetics and pharmacodynamics of oxybutynin chloride. Br J Clin Pharmacol. 52:409-417 (2001). 33. S.K. Gupta, G. Sathyan, E.A. Lindemulder, P.-L. Ho, L.B. Sheiner, and L. Aarons. Quantitative characterization of therapeutic index: Application of mixed-effects modeling to evaluate oxybutynin dose-efficacy and dose-side effect relationships. Clin Pharmacol Ther. 65:672-684 (1999). 34. Y. Cao and W.J. Jusko. Applications of minimal physiologically-based pharmacokinetic models. J Pharmacokinet Pharmacodyn. 39:711-723 (2012). 35. I.A. Nestorov, L.J. Aarons, P.A. Arundel, and M. Rowland. Lumping of whole-body physiologically based pharmacokinetic models. J Pharmacokinet Biopharm. 26:21-46 (1998). 36. A. Dokoumetzidis and L. Aarons. Proper lumping in systems biology models. IET Syst Biol. 3:40-51 (2009). 37. A. Dokoumetzidis and L. Aarons. A method for robust model order reduction in pharmacokinetics. J Pharmacokinet Pharmacodyn. 36:613-628 (2009). 38. J. Valentin. Basic anatomical and physiological data for use in radiological protection: reference values. Ann ICRP. 32:1-277 (2002). 39. L.R. Williams and R.W. Leggett. Reference values for resting blood flow to organs of man. Clin Phys Physiol Meas. 10:187-217 (1989). 40. P.J. Matheson, M.A. Wilson, and R.N. Garrison. Regulation of Intestinal Blood Flow. J Surg Res. 93:182-196 (2000). 41. E.M. Howgate, K. Rowland Yeo, N.J. Proctor, G.T. Tucker, and A. Rostami- Hodjegan. Prediction of in vivo drug clearance from in vitro data. I: impact of inter-individual variability. Xenobiotica. 36:473-497 (2006).

125

42. T. Rodgers, D. Leahy, and M. Rowland. Physiologically based pharmacokinetic modeling 1: Predicting the tissue distribution of moderate- to-strong bases. J Pharm Sci. 94:1259-1276 (2005). 43. N. Tsamandouras, T. Wendling, A. Rostami-Hodjegan, A. Galetin, and L. Aarons. Incorporation of stochastic variability in mechanistic population pharmacokinetic models: handling the physiological constraints using normal transformations. J Pharmacokinet Pharmacodyn. 10.1007/s10928-015-9418- 0:1-25 (2015). 44. J.A. Jacquez and T. Perry. Parameter estimation: local identifiability of parameters. Am J Physiol. 258:E727-736 (1990). 45. K.R. Godfrey and W.R. Fitch. The deterministic identifiability of nonlinear pharmacokinetic models. J Pharmacokinet Biopharm. 12:177-191 (1984). 46. J. Karlsson, M. Anguelova, and M. Jirstrand. An Efficient Method for Structural Identiability Analysis of Large Dynamic Systems, 16th IFAC Symposium on System Identification, Vol. 162012, pp. 941-946. 47. M. Anguelova, J. Karlsson, and M. Jirstrand. Minimal output sets for identifiability. Math Biosci. 239:139-153 (2012). 48. R. Bauer. NONMEM Users Guide: Introduction to NONMEM 7.3.0, ICON Development Solutions, Hanover, MD, USA, 2013. 49. G. Langdon, I. Gueorguieva, L. Aarons, and M. Karlsson. Linking preclinical and clinical whole-body physiologically based pharmacokinetic models with prior distributions in NONMEM. Eur J Clin Pharmacol. 63:485-498 (2007). 50. R.J. Carroll and D. Ruppert. Transformation and weighting in regression, CRC Press1988. 51. L.X. Yu and G.L. Amidon. A compartmental absorption and transit model for estimating oral drug absorption. Int J Pharm. 186:119-125 (1999). 52. L.X. Yu, E. Lipka, J.R. Crison, and G.L. Amidon. Transport approaches to the biopharmaceutical design of oral drug delivery systems: prediction of intestinal absorption. Adv Drug Deliv Rev. 19:359-376 (1996). 53. L.X. Yu, J.R. Crison, and G.L. Amidon. Compartmental transit and dispersion model analysis of small intestinal transit flow in humans. Int J Pharm. 140:111-118 (1996). 54. M. Jamei, D. Turner, J. Yang, S. Neuhoff, S. Polak, A. Rostami-Hodjegan, and G. Tucker. Population-based mechanistic prediction of oral drug absorption. AAPS J. 11:225-237 (2009). 55. G.L. Amidon, H. Lennernas, V.P. Shah, and J.R. Crison. A theoretical basis for a biopharmaceutic drug classification: the correlation of in vitro drug product dissolution and in vivo bioavailability. Pharm Res. 12:413-420 (1995). 56. International Commission on Radiological Protection. Report of the Task Group on Reference Man, Pergamon Press1975. 57. L.X. Yu and G.L. Amidon. Saturable small intestinal drug absorption in humans: modeling and interpretation of cefatrizine data. Eur J Pharm Biopharm. 45:199-203 (1998). 58. N.W. Read, M.N. Al-Janabi, A.M. Holgate, D.C. Barber, and C.A. Edwards. Simultaneous measurement of gastric emptying, small bowel residence and colonic filling of a solid meal by the use of the gamma camera. Gut. 27:300- 308 (1986). 59. S.S. Davis, J.G. Hardy, and J.W. Fara. Transit of Pharmaceutical Dosage Forms through the Small-Intestine. Gut. 27:886-892 (1986).

126

60. E. Hénin, M. Bergstrand, J. Standing, and M. Karlsson. A mechanism-Based Approach for Absorption Modeling: The Gastro-Intestinal Transit Time (GITT) Model. The AAPS Journal. 14:155-163 (2012). 61. A. Dokoumetzidis and P. Macheras. A century of dissolution research: from Noyes and Whitney to the biopharmaceutics classification system. Int J Pharm. 321:1-11 (2006). 62. J. Wang and D.R. Flanagan. General solution for diffusion-controlled dissolution of spherical particles. 1. Theory. J Pharm Sci. 88:731-738 (1999). 63. K.C. Johnson. Comparison of methods for predicting dissolution and the theoretical implications of particle-size-dependent solubility. J Pharm Sci. 101:681-689 (2012). 64. K. Sugano. Theoretical comparison of hydrodynamic diffusion layer models used for dissolution simulation in drug discovery and development. Int J Pharm. 363:73-77 (2008). 65. R.J. Hintz and K.C. Johnson. The effect of particle size distribution on dissolution rate and oral absorption. Int J Pharm. 51:9-17 (1989). 66. A. Okazaki, T. Mano, and K. Sugano. Theoretical dissolution model of poly- disperse drug particles in biorelevant media. J Pharm Sci. 97:1843-1852 (2008). 67. K. Sugano. Computational oral absorption simulation for low-solubility compounds. Chem Biodivers. 6:2014-2029 (2009). 68. K. Johnson. Dissolution and Absorption Modeling: Model Expansion to Simulate the Effects of Precipitation, Water Absorption, Longitudinally Changing Intestinal Permeability, and Controlled Release on Drug Absorption. Drug Dev Ind Pharm. 29:833-842 (2003). 69. A. Avdeef. Solubility of sparingly-soluble ionizable drugs. Adv Drug Deliv Rev. 59:568-590 (2007). 70. D.M. Mudie, K. Murray, C.L. Hoad, S.E. Pritchard, M.C. Garnett, G.L. Amidon, P.A. Gowland, R.C. Spiller, G.E. Amidon, and L. Marciani. Quantification of gastrointestinal liquid volumes and distribution following a 240 mL dose of water in the fasted state. Mol Pharm. 11:3039-3047 (2014). 71. K. Sugano. Estimation of effective intestinal membrane permeability considering bile micelle solubilisation. Int J Pharm. 368:116-122 (2009). 72. H. Lennernas. Human in vivo regional intestinal permeability: importance for pharmaceutical drug development. Mol Pharm. 11:12-23 (2014). 73. M. Zur, M. Gasparini, O. Wolk, G.L. Amidon, and A. Dahan. The Low/High BCS Permeability Class Boundary: Physicochemical Comparison of Metoprolol and Labetalol. Mol Pharm. 11:1707-1714 (2014). 74. H. Mizushima, K. Takanaka, K. Abe, I. Fukazawa, and H. Ishizuka. Stereoselective pharmacokinetics of oxybutynin and N-desethyloxybutynin in vitro and in vivo. Xenobiotica. 37:59-73 (2007). 75. K. Rowland Yeo, A. Rostami-Hodjegan, and G.T. Tucker. Abundance of Cytochromes P450 in Human Liver: A Meta-Analysis. Br J Clin Pharmacol. 57:687-688 (2004). 76. E. Sjogren, B. Abrahamsson, P. Augustijns, D. Becker, M.B. Bolger, M. Brewster, J. Brouwers, T. Flanagan, M. Harwood, C. Heinen, R. Holm, H.P. Juretschke, M. Kubbinga, A. Lindahl, V. Lukacova, U. Munster, S. Neuhoff, M.A. Nguyen, A. Peer, C. Reppas, A.R. Hodjegan, C. Tannergren, W. Weitschies, C. Wilson, P. Zane, H. Lennernas, and P. Langguth. In vivo methods for drug absorption - Comparative physiologies, model selection,

127

correlations with in vitro methods (IVIVC), and applications for formulation/API/excipient characterization including food effects. Eur J Pharm Sci. 57:99-151 (2014). 77. M.F. Paine, H.L. Hart, S.S. Ludington, R.L. Haining, A.E. Rettie, and D.C. Zeldin. The human intestinal cytochrome P450 "pie". Drug Metab Dispos. 34:880-886 (2006). 78. J. Yang, G.T. Tucker, and A. Rostami-Hodjegan. Cytochrome P450 3A expression and activity in the human small intestine. Clin Pharmacol Ther. 76:391 (2004). 79. J. Yang, M. Jamei, K.R. Yeo, G.T. Tucker, and A. Rostami-Hodjegan. Prediction of intestinal first-pass drug metabolism. Curr Drug Metab. 8:676- 684 (2007). 80. M. Pitsiu, G. Sathyan, S. Gupta, and D. Verotta. A semiparametric deconvolution model to establish in vivo–in vitro correlation applied to OROS oxybutynin. J Pharm Sci. 90:702-712 (2001). 81. A. Shibukawa, N. Ishizawa, T. Kimura, Y. Sakamoto, K. Ogita, Y. Matsuo, Y. Kuroda, C. Matayatsu, T. Nakagawa, and I.W. Wainer. Plasma protein binding study of oxybutynin by high-performance frontal analysis. J Chromatogr B Analyt Technol Biomed Life Sci. 768:177-188 (2002). 82. G. Sathyan, R.R. Dmochowski, R.A. Appell, C. Guo, and S.K. Gupta. Effect of antacid on the pharmacokinetics of extended-release formulations of tolterodine and oxybutynin. Clin Pharmacokinet. 43:1059-1068 (2004). 83. U.S. Food and Drug Administration. Ditropan XL® Product Label. http://www.accessdata.fda.gov/drugsatfda_docs/label/2003/17577se8- 033,18211se8-016,20897slr010_ditropan_lbl.pdf (accessed 06/01/2015 2015). 84. R. Conley, S.K. Gupta, and G. Sathyan. Clinical spectrum of the osmotic- controlled release oral delivery system (OROS), an advanced oral delivery form. Curr Med Res Opin. 22:1879-1892 (2006). 85. D. Hallifax and J.B. Houston. Evaluation of hepatic clearance prediction using in vitro data: Emphasis on fraction unbound in plasma and drug ionisation using a database of 107 drugs. J Pharm Sci. 101:2645-2652 (2012). 86. J.A. Foster, J.B. Houston, and D. Hallifax. Comparison of intrinsic clearances in human liver microsomes and suspended hepatocytes from the same donor livers: clearance-dependent relationship and implications for prediction of in vivo clearance. Xenobiotica. 41:124-136 (2011). 87. M. Taketani, M. Shii, K. Ohura, S. Ninomiya, and T. Imai. Carboxylesterase in the liver and small intestine of experimental animals and human. Life Sci. 81:924-932 (2007). 88. Y. Sato, A. Miyashita, T. Iwatsubo, and T. Usui. Conclusive Identification of the Oxybutynin-Hydrolyzing Enzyme in Human Liver. Drug Metab Dispos. 40:902-906 (2012). 89. T. Rodgers, D. Leahy, and M. Rowland. Tissue distribution of basic drugs: Accounting for enantiomeric, compound and regional differences amongst β- blocking drugs in rat. J Pharm Sci. 94:1237-1248 (2005). 90. N. Tsamandouras, A. Rostami-Hodjegan, and L. Aarons. Combining the 'bottom up' and 'top down' approaches in pharmacokinetic modelling: fitting PBPK models to observed clinical data. Br J Clin Pharmacol. 79:48-55 (2015).

128

91. M.V. Varma, R.S. Obach, C. Rotter, H.R. Miller, G. Chang, S.J. Steyn, A. El- Kattan, and M.D. Troutman. Physicochemical space for optimum oral bioavailability: contribution of human intestinal absorption and first-pass elimination. J Med Chem. 53:1098-1108 (2010). 92. E. Sjögren, D. Dahlgren, C. Roos, and H. Lennernäs. Human in Vivo Regional Intestinal Permeability: Quantitation Using Site-Specific Drug Absorption Data. Mol Pharm. 10.1021/mp500834v:(2015).

129

Chapter 7: Concluding remarks and future perspectives

130

Prediction of human oral bioavailability in drug discovery and development is a challenging task. Herein different modelling and simulation approaches have been proposed and evaluated to fit such a purpose, from the holistic prediction of oral bioavailability based on pre-clinical species data to a more mechanistic and focused approach, applying state of the art PBPK models to evaluate the impact that MR formulations have on drug absorption and intestinal first pass metabolism. Each approach that was investigated in this thesis provides its own advantages and disadvantages.

The direct extrapolation of the pre-clinical bioavailability data to that in humans is a simple approach that can be readily implemented in drug development, especially considering the minimal amount of data required for the predictions and the availability of the pre-clinical bioavailability data obtained during the drug candidate’s ADME, safety and efficacy assessment. However, as shown in this thesis, one of the main disadvantages of such an approach is the lack of a quantitative and precise correlation between the bioavailability in the pre-clinical species to that in humans, where the majority of the pre-clinical species evaluated showed a weak correlation for bioavailability to that in human. This was demonstrated by building and analysing what is at the moment one of the most extensive pre-clinical and human bioavailability datasets available in the published literature. Nevertheless, there is still room to expand and improve such a dataset, either by adding more compounds that were not included in our search criteria and/or by populating the dataset with missing drug-related properties, such as permeability, solubility, affinity for metabolizing enzymes and/or transporters, etc. This expansion could allow the analysis of trends that were out of the scope of this work and that can lead to a better understanding of the reasons behind the apparent lack of correlation between the bioavailability observed in pre-clinical species to that in humans. Despite the lack of a quantitative correlation, this thesis also showed that pre-clinical bioavailability data could, at least, be employed for the qualitative prognosis of whether the human oral bioavailability could be expected to be high or low. This was demonstrated through the application of ROC analysis to the previously developed bioavailability dataset. The results of this analysis were in line with previous works suggesting that some preclinical species, particularly rats and NHPs, can at least provide a qualitative indication of the expected human oral bioavailability. The ROC-derived optimal pre-

131

clinical bioavailability thresholds, predicting high or low human oral bioavailability, were also similar to values employed by certain industrial discovery and development groups for candidate selection and lead optimization. However, in contrast to the ROC analysis, the latter thresholds are usually the result of practical experience that has been accumulated throughout the years in the discovery groups rather than a systematic statistical analysis. The ROC-derived thresholds, on the other hand, are associated with the probabilities of making an accurate prediction of the expected human bioavailability for a given value in pre-clinical species (sensitivity and specificity), which can be of value when using pre-clinical data for decision making or lead optimization. One of the main disadvantages of the work presented in Chapters 2 and 3 is that they do not provide a mechanistic rationale for the relationship between pre-clinical and human oral bioavailability. For instance, in Chapter 2 both rat and NHP tended to underpredict human oral bioavailability. This was consistent with their respective bioavailability thresholds derived from the ROC analysis of 22% and 35%. Given that the fa appears to be correlated between human and the aforementioned preclinical species, at least for passively absorbed drugs, one can speculate as to whether the underprediction is the product of the species differences in first-pass metabolism due to differences in activity and/or expression of metabolizing enzymes. Yet, due the nature of the analysis performed herein, i.e., evaluating oral bioavailability as a whole, it is hard to provide a more insightful explanation with respect to the mechanisms driving such bioavailability differences. It would be therefore of interest to perform a similar analysis from the point of view of the fractions that define oral bioavailability (fa, FG and FH). However, such an analysis could be time consuming and resource demanding, given the necessity of an expanded dataset. Finally, although the analysis performed in Chapters 2 and 3 could still be of value in drug development, particularly given their simplicity, the current trend in drug discovery and development is to try to anticipate the ADME properties of a drug candidate applying a more mechanistic approach, linking the drug-related properties to the expected ADME profile using PBPK models. Such an approach can provide insights about the mechanisms driving the candidate’s bioavailability, and therefore can be used for the anticipation of bioavailability issues such as the DDI liability or dose non-linearity due to solubility problems, to name a few examples.

132

Consequently, the second part of this thesis was aimed at providing examples of more mechanistic bioavailability predictions through the use of PBPK models. This thesis was particularly focused on the investigation of the interplay between absorption and CYP-mediated intestinal metabolism from the point of view of MR dosage forms, a variable that had been overlooked in previous works. As mentioned earlier, the use of PBPK models provides the advantage of a mechanistic insight into the bioavailability process; however it comes at the cost of an increased amount of data required for such predictions, translated into the number of input parameters required by the model. This high dimensionality on the input parameters and the complex interplay between drug-related and physiologically-related parameters, particularly in the absorption process, can also lead to difficulties on the interpretation of the predictions derived from such models, having some times the “right answer” but with the wrong mechanism. Herein we investigated the cases when MR formulations of CYP3A substrates displayed higher relative bioavailability than their respective IR formulation through PBPK modelling and simulation. The approach presented in Chapter 4 was contextualized in the drug discovery setting, where the goal was to identify the drug parameter space that would lead to such scenarios, either to take advantage of them or to prevent them. This is an example of the usefulness of PBPK models to answer “what if” type questions in drug discovery and development. However, the systematic analysis presented in Chapter 4 also highlighted the complex interplay between the input parameters and difficulty of the interpretation of the model outputs. As an example, we performed 78,125 simulations using the SimCYP® batch processor that were latter reduced with the help of the BCS classification and the analysis of specific scenarios, yet the interpretation of the overall results was still difficult. Moreover, one of the main issues with the use of this exploratory analysis is the assumption that the model produces a reasonable output and that the most relevant mechanism defining the drug’s bioavailability are incorporated into the model structure, hence defining the adequacy of the model to evaluate our hypothesis. The former could be difficult to evaluate without the knowledge of the particular compound in development, thus limiting the exploratory nature of the analysis, but it can be informed from similar compounds for which literature data exist, whereas the latter will depend on the current state of art in terms of translational absorption predictions and the availability of such knowledge in the

133

PBPK model structure. For these reasons, we restricted our exploratory analysis to the use of input parameter that had been previously observed in the literature, in an attempt to represent a reasonable parameter space and thus provide sensible outcomes. Also we used SimCYP®’s ADAM model which, to our knowledge, included the most relevant mechanisms driving drug absorption and bioavailability that were necessary for our research, such as dissolution, permeation, release, and intestinal first pass metabolism. Yet, the work could have been carried out with any other commercially available software or even with an in house development. The advantage of the use of commercially available software is the ease of use and readily availability of the model structure and parameter libraries, such as human physiological parameters. However, the main disadvantage is their closed structure that can limit the implementation of alternative models or mechanisms that were not included in the particular release, in our case we used SimCYP® version 13. This could be achieved by developing an in house model using scripting/programing languages such as Matlab, R, Phyton, etc. as done latter on in Chapter 6; however the process of parameter gathering and the subsequent model validation can be resource intensive and time consuming for the analysis performed in Chapter 4. One of the difficulties we encountered in our PBPK simulation analysis was the overprediction of the fa for some of the MR formulations, particularly the ones with substrates belonging to the BCS Class 3. This overprediction was attributed to the assumption that Peff was the same for all the intestinal segments within the ADAM model. This assumption was considered reasonable at the time given that the common approaches to evaluate distal permeability, such as sensitivity analysis or parameter estimation, were not applicable in a prospective scenario, where the clinical data is not yet available, as it is in the discovery setting. This highlighted the lack a simple methodology to implement regional permeability, which was not part of SimCYP® at the time of the analysis. Therefore, in Chapter 5 we developed a simple method for the implementation of regional permeability for the PBPK framework which showed the potential to reduce the overprediction observed when Peff was assumed to be the same in all intestinal segments of the GI tract, including the colon. This method might oversimplify the mechanisms driving permeability and intestinal absorption; however it can provide a useful starting point for the implementation of regional absorption, especially when there is no access to a proprietary models such as the ones implemented in GastroPlus™ (Opt LogD

134

model), PK-Sim® and the newer versions of SimCYP® that include the so-called

MechPeff model (version 14 onwards)[1-3] The aim of the final Chapter of this thesis was to provide a mechanistic explanation of the higher relative bioavailability observed for oxybutynin’s OROS formulations and, at the same time, provide the necessary drug-specific example needed as a validation of the outcomes obtained in the exploratory PBPK modelling and simulation approach, where we hypothesised about the possibility of a higher FG in MR formulations being responsible for the higher oral bioavailability of substrates like oxybutynin. The oxybutynin’s simulations were done using a specially developed simplified absorption model, the mSAT model, which also included the semi-mechanistic model for regional permeability parametrization, previously developed in Chapter 5. Therefore Chapter 6 can be seen as an integrative approach of concepts developed in Chapters 4 and 5. One of the main issues when it comes to prospective absorption PBPK prediction is the use of the clinical data to calibrate the model predictions in the so-called “middle out” approach [4]. When done properly, there is the possibility of gaining considerable information about the drug-related properties defining the drug absorption process. However, given that oral bioavailability is a complex interplay between absorption and pre-systemic extraction, the estimation of drug-related parameters or even physiologically-related parameters using only oral pharmacokinetic data to inform the model could lead to confounded results, where the parameter estimates can be a mixture of the information contained in the absorption and disposition processes. Therefore, the modelling approach employed for oxybutynin involved an estimation step using only IV data, whereas the absorption modelling was performed in a fully “bottom up” fashion using only drug- related parameters derived from the literature or calculated using known in silico equations. We believe that this approach provides a better insight on the absorption and bioavailability predictions; however given the scarcity of reliable IV data its applicability could be limited. The mSAT model predictions were able to capture the bioavailability differences observed for oxybutynin OROS formulation compared to its IR counterpart, where the mechanism for such higher bioavailability was an increased FG, as suggested in Chapter 4. This highlights the usefulness of the newly developed mSAT model for the absorption and first-pass metabolism prediction. Yet one has to bear in mind that oxybutynin is a highly soluble and permeable BCS Class

135

1 compound, for which the PBPK absorption predictions pose a less challenging task, as compared to poorly soluble and permeable compounds. Consequently, before validating its potential for drug absorption predictions, the usefulness of the mSAT model for absorption prediction needs to be evaluated in a large pool of compounds, particularly belonging to BCS Classes 2 and 3. Given that the predictions in Chapter 6 were only based on a “reference individual”, the high degree of interindividual variability observed in oxybutynin’s pharmacokinetic profiles (IR and OROS) was not adequately captured by the mSAT model predictions. This variability can have a significant impact on both fa and FG predictions and therefore could limit the subsequent use of the model to evaluate scenarios such as metabolic DDIs. As mentioned before in Chapter 6, this variability could be incorporated in the model from a bottom up perspective, as in SimCYP®, or by applying the middle out approach. The latter can be achieved either with a population pharmacokinetic approach using non-linear mixed effect modelling or with the use of hierarchical modelling within the Bayesian framework [4]. Given the relatively simple structure of the mSAT model and the availability of the necessary parameters for its implementation in any modelling platform, the exploration of such modelling approaches pose an interesting challenge as a continuation of the research presented herein. In conclusion, this thesis provided several examples and methodologies that can be used for the prediction of oral bioavailability in drug development, from simple correlation models based on pre-clinical data to the use of more complex PBPK models to answer “what if?” questions. The use of either method will depend upon the questions asked in drug discovery and development. However, given the increased trend for rational drug discovery development, the use of an integrative approach, like the one provided by PBPK models, seems to be the natural path to follow.

136

7.1 References

1. K. Thelen, K. Coboeken, S. Willmann, R. Burghaus, J.B. Dressman, and J. Lippert. Evolution of a detailed physiological model to simulate the gastrointestinal transit and absorption process in humans, part 1: oral solutions. J Pharm Sci. 100:5324-5345 (2011). 2. B. Agoram, W.S. Woltosz, and M.B. Bolger. Predicting the impact of physiological and biochemical processes on oral drug bioavailability. Adv Drug Deliv Rev. 50 Suppl 1:S41-67 (2001). 3. D. Pade, D. Turner, M. Jamei, and A. Rostami Hodjegan. Prediction of Passive Regional Effective Intestinal Permeability in Mouse Using ‘MechPeff’: A Mchanistic Model Utilizing only Drug Physicochemical Parameters as Inputs Presented at the 2013 AAPS Annual Meeting and Exposition poster R6070, San Antonio, Texas, 2013. 4. N. Tsamandouras, A. Rostami-Hodjegan, and L. Aarons. Combining the 'bottom up' and 'top down' approaches in pharmacokinetic modelling: fitting PBPK models to observed clinical data. Br J Clin Pharmacol. 79:48-55 (2015).

137

Appendix A1: Supplementary Material for Chapter 2

138

A1.1 Supplementary Tables

Table A1.1 Supplementary information for the animal vs human oral bioavailability comparison Data presented in a digital format (Microsoft Excel file). The file can be found in the CD (“Appendix_A1.xlsx”) or alternatively in the following hyperlink: http://goo.gl/tha5mz

139

Appendix A2: Supplementary Material for Chapter 3

140

A2.1 Supplementary Tables

Table A2.1 Provisional BDDCS class for the drugs employed in the study**. Provisional BDDCS Name Molecular Weight Ion Class Class Amusalol 380.46 base 1 Estramustine 440.41 neutral 2 phosphate Ethimizol 210.24 neutral 1 Ethylmorphine 313.391 base 1 Fenfluramine 231.26 base 3 Fenoterol 303.35 base 1 Flunisolide 434.498 neutral 1 Gatifloxacin 375.4 zwitterion 4 Gitoxin 920.989 neutral 1 Glaziovine 297.35 base 3 Idazoxan 204.23 base 1 Isoxicam 335.34 neutral 2 Lisuride 338.45 base 1 Melagatran 429.52 zwitterion 4 Menogaril 541.55 acid 2 Mepindolol 262.35 base 3 Metolazone 365.84 neutral 4 Moxonidine 241.68 base 3 Nisoldipine 388.42 neutral 2 Nomifensine 238.33 base 1 Norfenfluramine 203.21 base 1 Nufenoxole 387.52 base 2 Physostigmine 275.35 base 1 Recainam 263.38 base 4 Remoxipride 371.27 base 1 Terodiline 281.44 base 2 TRH Tartrate 362.38 base 3 Trovafloxacin 416.35 zwitterion 4 Xamoterol 339.39 base 3 ** Additional information about the references and values employed for the classification can be found in a digital format (Microsoft Excel file). The file can be found in the CD (“Appendix_A2_TA21_TA22.xlsx”) or alternatively in the following hyperlink: http://goo.gl/6yxwVs

141

Table A2.2 BDDCS class for the drugs employed in the study**. Molecular Name Ion Class BDDCS Classa Weight 5-Fluorouracil 130.08 neutral 1 Acarbose 645.605 base 1 Acebutolol 336.426 base 1 Acetylsalicylate 180.15 acid 1 Acyclovir 225.21 base 4 Adefovir 273.186 base 3 Alprazolam 351.8 acid 1 Amitriptyline 277.403 base 1 Amlodipine 408.879 base 1 Amoxicillin 365.4 zwitterion 3 Antipyrine 188.226 neutral 1 Azathioprine 277.263 base 1 Azithromycin 748.984 base 3 Cefixime 453.46 acid 4 Cefuroxime Axetil 510.47448 acid 3 Chlorpheniramine 274.8 base 1 Clonazepam 315.72 base 1 Cyclosporine 1202.61 neutral 2 Dapsone 248.31 base 2 Diazepam 284.74022 base 1 Doxazosin 451.4751 base 1 Erythomycin 733.93 base 3 Estradiol Valerate 272.39 neutral 1 Ethambutol 204.31 base 3 Ethinylestradiol 296.41 neutral 1 Ethosuximide 141.17 neutral 1 Etoposide 588.56 neutral 3 Felodipine 384.26 neutral 2 Fexofenadine 501.656 zwitterion 3 Finasteride 372.55 neutral 1 Fleroxacin 369.34 zwitterion 4 Fluconazole 306.28 neutral 3 Flumazenil 303.29 neutral 1 Flunitrazepam 313.3 neutral 1 Fluvastatin 411.47 acid 1 Foscarnet 126 acid 3 Fosfomycin 138.06 acid 3 Furosemide 330.75 acid 4 Gabapentin 171.24 zwitterion 3 Ganciclovir 255.23 neutral 3 Glyburide 494.01 acid 2 Griseofulvin 352.77 neutral 2 Guanfacine 246.1 base 3 Hydralazine 160.18 neutral 1 Hydrochlorothiazide 297.74 neutral 3 Ibuprofen 206.28 acid 2 Ifosphamide 261.09 neutral 1 Indapamide 365.84 neutral 1 Indomethacin 357.79 acid 2 Isosorbide dinitrate 236.14 neutral 1 Isosorbide-2-mononitrate 191.14 neutral 1 142

Isosorbide-5-mononitrate 191.14 neutral 1 Itraconazole 705.64 neutral 2 Ketanserin 395.43 base 2 Ketorolac 255.27 acid 3 Lansoprazole 369.37 neutral 2 Levodopa 197.19 zwitterion 1 Levofloxacin 361.37 zwitterion 3 Levonorgestrel 312.45 neutral 4 Lidocaine 234.34 base 1 Linezolid 337.35 neutral 1 Lithium carbonate 73.891 neutral 3 Losartan 422.92 acid 2 Meloxicam 351.41 neutral 2 Mercaptopurine 152.18 neutral 2 Metformin 129.17 base 3 Methadone 309.45 base 1 Methylprednisolone 374.47 neutral 1 Metoclopramide 299.8 base 3 Metoprolol 267.37 base 1 Midazolam 325.77 neutral 1 Morphine 285.34 base 1 Moxifloxacin 401.44 zwitterion 3 Nalbuphine 357.45 base 1 Naloxone 327.37 base 1 Naltrexone 341.41 base 1 Naproxen 230.26 acid 2 Naratriptan 335.47 base 3 Nefazodone 470.01 base 2 Nevirapine 266.3 neutral 2 Nicardipine 479.53 base 1 Nifedipine 346.34 neutral 2 Nimodipine 418.44 neutral 2 Nitrendipine 360.36 neutral 2 Nizatidine 331.46 base 3 Ofloxacin 361.37 zwitterion 3 Omeprazole 345.42 neutral 1 Ondansetron 293.37 base 1 Oseltamivir acid 284.35 acid 3 Oxazepam 286.71 zwitterion 2 Penicillin V 350.39 acid 4 Phenobarbital 232.23 acid 1 Phenytoin 252.27 acid 2 Pindolol 248.32 base 3 Piroxicam 331.35 zwitterion 2 Pravastatin 424.53 acid 3 Prazosin 383.4 base 1 Prednisolone 360.44 neutral 1 Prednisone 358.43 neutral 2 Primaquine 259.35 base 1 Procainamide 235.33 base 3 Propoxyphene 339.47 base 2 Propranolol 259.34 base 1 Propranolol (-) 259.34 base 1 Propranolol (+) 259.34 base 1 143

Propylthiouracil 170.23 acid 1 Pyridostigmine 181.21 base 3 Quinidine 324.42 base 1 Rabeprazole 359.44 base 1 Ranitidine 314.4 base 3 Reboxetine 313.39 base 1 Rifabutin 847 base 2 Rifampin 822.94 zwitterion 2 Risedronate 283.11 acid 3 Risperidone 410.48 base 1 Rosiglitazone 357.43 zwitterion 1 Rosuvastatin 481.54 acid 3 Salbutamol 239.31 base 3 Salicylate 160.11 acid 1 Saquinavir 670.84 base 2 Selegiline 187.28 base 1 Sildenafil 474.58 base 1 Sitafloxacin 409.81 zwitterion 3 Sitagliptin 407.31 base 3 Sotalol 272.36 base 3 Sparfloxacin 392.4 zwitterion 1 Sulfisoxazole 267.3 zwitterion 4 Sulpiride 341.43 base 3 Sumatriptan 295.4 base 1 Tacrolimus 804.02 acid 2 Talinolol 363.49 base 3 Tamsulosin 408.51 base 1 Terazosin 387.43 base 1 Tetrabenazine 317.42 base 2 Theophylline 180.16 base 1 Tiagabine 375.55 acid 2 Timolol 316.42 base 1 Tinidazole 247.27 base 1 Tolterodine 325.49 base 1 Torsemide 348.42 acid 2 Tramadol 263.38 base 1 Trazodone 371.86 base 2 Triazolam 343.21 neutral 1 Valproic acid 144.21 acid 1 Vardenafil 488.6 base 1 Venlafaxine 277.4 base 1 Verapamil (-) 454.6 base 1 Verapamil (+) 454.6 base 1 Verapamil 454.6 base 1 Warfarin 308.33 acid 2 Zalcitabine 211.22 base 3 Zanamivir 332.31 base 3 Zolmitriptan 287.36 base 1 Zolpidem 307.39 base 1 Zopiclone 388.81 base 1 ** Additional information about the references and values employed for the classification can be found in a digital format (Microsoft Excel file). The file can be found in the CD (“Appendix_A2_TA21_TA22.xlsx”) or alternatively in the following hyperlink: http://goo.gl/6yxwVs 144

a, BDDCS class extracted from [1]

145

Table A2.3 Drug categories for high bioavailability drugs (Fhuman ≥ 50%) according to the threshold based classification.

True positives (TP) False negatives (FN) Acyclovir dog Itraconazole rat Quinidine rat, dog Acyclovir rat Amlodipine rat, dog Ketorolac rat, dog, Ranitidine dog Alprazolam rat Amosulalol rat, dog, NHP NHP Reboxetine dog, NHP Cefixime dog Amoxicillin dog Lansoprazole rat Recainam rat, dog, NHP Clonazepam dog Antipyrine dog Levofloxacin rat Remoxipride dog Etoposide rat, dog, NHP Azathioprine NHP Linezolid rat, dog Rifampin rat Fenfluramine rat Dapsone dog Lithium carbonate Risperidone rat Furosemide dog Diazepam dog NHP Rosiglitazone rat Glyburide dog Doxazosin rat, dog Meloxicam rat, dog salbutamol dog Griseofulvin rat Ethambutol NHP Metformin rat Salicylate rat, dog Hydrochlorothiazide Ethosuximide rat, dog, NHP Metoclopramide rat Sitafloxacin rat, NHP NHP Finasteride dog Moxifloxacin rat, dog, Sitagliptin rat, dog, NHP Lansoprazole dog Fleroxacin rat, dog NHP Sotalol dog Levonorgestrel rat, dog, Fluconazole rat, dog Naltrexone rat Sparfloxacin rat, dog, NHP Flunitrazepam rat Naproxen dog NHP Mepindolol rat, dog Furosemide rat Naratriptan rat, dog Sulfisoxazole rat Methadone dog Gabapentin rat, dog, NHP Nevirapine rat Talinolol rat Moxonidine rat Gatifloxacin rat, dog Nizatidine rat, dog Terazosin rat Nifedipine NHP Gitoxin dog Nufenoxole rat, NHP Theophylline rat, dog, Norfenfluramine rat Glaziovine dog Ofloxacin rat NHP Nufenoxole dog Guanfacine NHP Oseltamivir acid rat, Tiagabine rat Omeprazole rat dog Ibuprofen rat, NHP dog Tinidazole dog Ondansetron, rat Ifosphamide rat Oxazepam rat Torsemide rat, dog Oxazepam dog Indapamide dog Phenobarbital dog Tramadol rat, dog Prazosin dog Indomethacin rat Pindolol rat, dog, NHP Trazodone rat Prednisolone dog Isosorbide-2- Piroxicam dog, NHP Trovafloxacin rat, dog, Rabeprazole dog mononitrate rat Prednisone NHP NHP Reboxetine rat Isosorbide-5- Primaquine rat Valproic acid rat, dog Remoxipride rat mononitrate dog Procainamide dog Warfarin dog Sitafloxacin dog Isoxicam dog Propylthiouracil dog Zolpidem rat Tamsulosin rat, dog Zopiclone rat, dog Terodiline dog Tiagabine dog Timolol dog Triazolam rat Verapamil (+) rat, dog

146

Table A2.4 Drug categories for low bioavailability drugs (Fhuman < 50%) according to the threshold based classification.

True negatives (TN) False positives(FP) Acarbose dog Idazoxan rat Propranolol (+) dog 5-Fluorouracil rat Acetylsalicylate dog Levodopa dog Pyridostigmine dog Acebutolol rat Adefovir rat Lidocaine dog Risedronate rat Acetylsalicylate rat Chlorpheniramine dog Lidocaine rat, NHP Rosuvastatin rat Amitriptyline dog Cyclosporine dog Lisuride NHP Saquinavir rat Azithromycin rat, dog Erythomycin rat, dog, Melagatran rat Selegiline dog Cefuroxime axetil rat NHP Menogaril dog, NHP Sildenafil dog Ethimizol rat Estradiol valerate rat, Mercaptopurine NHP Sulpiride rat Flumazenil rat dog Metoprolol NHP Sumatriptan dog Flunisolide rat Estramustine Midazolam rat, dog, Tacrolimus dog Fluvastatin rat phosphate dog NHP Tolterodine rat Ganciclovir dog Ethinylestradiol rat, Morphine rat, dog TRH Tartrate rat, Idazoxan dog dog, NHP Nalbuphine rat, dog dog Isosorbide dinitrate rat Ethylmorphine rat, dog Nefazodone dog Vardenafil rat, dog Ketanserin rat, dog Felodipine rat, dog Nicardipine rat, dog, Venlafaxine rat, NHP Lisuride rat Fenoterol rat NHP Verapamil (-) rat, Losartan rat Fexofenadine rat, NHP Nisoldipine rat, dog dog Melagatran dog Flunisolide NHP Nitrendipine rat, dog Verapamil dog, NHP Methylprednisolone rat Fluvastatin dog, NHP Nomifensine dog Xamoterol dog Metolazone dog Foscarnet dog Penicillin V rat, dog, Nimodipine rat Fosfomycin dog NHP Rifabutin rat Ganciclovir rat Phenytoin dog Sildenafil rat Hydralazine dog Physostigmine rat Sulpiride dog Pravastatin rat, dog Sumatriptan rat Propoxyphene dog Tacrolimus rat Propranolol rat, dog, Tetrabenazine rat NHP Tolterodine dog Propranolol (-) dog Venlafaxine dog Zolmitriptan rat, dog

147

A2.2 References

1. L.Z. Benet, F. Broccatelli, and T.I. Oprea. BDDCS applied to over 900 drugs. AAPS J. 13:519-547 (2011).

148

Appendix A3: Supplementary material for Chapter 4

149

A3.1 Method employed for the calculation of the relative bioavailability and its 90% confidence interval (CI).

The relative bioavailability between CR formulations and IR formulations was calculated from the mean AUC data reported for the different studies that were collated from the literature. As a rule, the AUC employed for such calculations was the AUC from zero to the last measured time point (AUC0-t). In the cases where such a parameter was not reported the AUC from zero to infinity was employed (AUC0- inf). In the case of multiple dose studies, where the aforementioned AUC0-inf was not reported, the AUC at steady state (AUCss) was employed.

a For the estimation of the ratio between the formulations (CR and IR), , and its b 90% confidence interval an approximation of the Fieller’s theorem [1, 2] was employed, Equation A3.1

a Equation A3.1 C Ia t , df SE a (1 ) bbbg(1 )

where a is the mean AUC for the CR release formulation and b is the mean AUC for

the IR formulation, t ,df is the inverse of the cumulative t distribution (two tailed) for a significance level of α (0.05) and degrees of freedom (df) is equal to na+nb-2, na and nb are the number of subjects in the study for the estimation of a and b,

a respectively, and SE is the combined standard error for the ratio and it is b calculated using Equation A3.2

a SESE22 Equation A3.2 SE(1  g ) ab  a 22 b b(1 g ) a b

where the quantity g is given by Equation A3.3, and SEa and SEb are the standard error of the means of a and b, respectively.

150

t22 SE Equation A3.3 g   ,df b b 2

A3.2 Calculation method for the intrinsic clearance from in vivo clearance data

The calculation of the intrinsic hepatic clearance (CLint,h) from the in vivo systemic plasma clearance (CLsys) was calculated according to the well stirred model as shown in Equation A3.4[3]. The hepatic plasma clearance (CLh) was estimated by subtracting any renal contribution to the systemic clearance when necessary.

CLQ h Equation A3.4 CL  h int, h CL fQ()h up h BP

All the calculations were based on an “reference individual”( 70 kg and 1.7 m), the liver blood flow (Qh) was assumed to be 20.7 ml/min/kg [4]. The intrinsic clearance in human liver microsomes (HLM) was calculated using Equation A3.5 assuming a liver weight (LW) of 1,718.4g (Simcyp’s mean value for the healthy volunteer population) and a microsomal protein per gram of liver (MPPGL) of 40 mg/g[5].

CLint,h Equation A3.5 CLint, HLM  LWMPPGL

When the hepatic blood clearance (CLh,blood, in Equation A3.6) exceeded the liver blood flow (i.e., buspirone), the hepatic intrinsic clearance was estimated by optimization of the CLint,h term in the dispersion model[6]( Equations A3.7 and A3.8). The optimization was conducted in Matlab 2013a (The Mathworks Inc., Natick, MA, USA) using the “fminunc” function of the Matlab’s Optimization

Toolbox. The results of the CLint, HLM calculations are summarized in Table A3.3

CLh Equation A3.6 CLh, blood  BP

151

where BP is the drug’s blood to plasma ratio.

Equation A3.7 4a CLQhh 1  aa11  2222D n D n (1a ) e  (1  a ) e where Dn is 0.17[7] and a is defined by Equation A3.8

f C L Equation A3.8 a14 upint, h  D n BPQ h

152

A3.3 Supplementary Tables

Table A3.1. Input parameters for the Simcyp simulations

Parameter Value Reference /comments Molecular weight (g/mol) 422 [8] LogP (octanol:water) 2.63 [8] Ionic class Neutral N/A pka N/A N/A Fraction unbound in plasma 0.05 [9] Blood plasma ratio (B:P) 0.81 [10]

Fraction unbound gut wall (fu,gut) 1 Simcyp default Intrinsic solubility (mg/mL) variable (Table 4.2) See methods -6 Papp, Caco-2 (10 cm/s) variable (Table 4.2) See methods Maximum supersaturation ratio 10 Simcyp default Precipitation Rate const. (1/h) 4 Simcyp default Particle radius (µm) 30 Simcyp default Particle density (g/mL) 1.2 Simcyp default heff (µm) 30 Simcyp default D (10-4 cm2/min) 4.021 Simcyp default Luminal degradation No Simcyp default Distribution model Minimal PBPK

Vss (L/kg) 0.937153 Estimated (Simcyp’s method 2) Elimination Enzyme kinetics (HLM)

Vmax, CYP3A4 variable ( Table 4.2) See methods

Km, CYP3A4 variable (Table 4.2) See methods fu, mic 1 Transport Transporter Kinetics Caco-2 cell monolayers

Jmax, P-gp (Efflux) variable (Table 4.2) See methods

Km, P-gp (Efflux) variable (Table 4.2) See methods

heff, Diffusion layer thickness; D, aqueous diffusion coefficient at 37°C; Vss, Steady state volume of distribution.

153

Table A3.2. Relative bioavailability studies between CR and IR formulation reported in the literature.

b Compound AUCCR SD NCR AUCIR SD NIR Ratio Lower Upper BCS/ Reference

(ng×h/mL) (ng×h/mL) (AUCCR/AUCIR) 90% CI 90% CI BDCSS Fluvastatina 0.37 0.29 0.47 1 [11, 12] Fluvastatina 0.39 0.31 0.49 1 [11, 12] Fluvastatin 564.00 360.96 12 1165.00 559.20 12 0.48 0.31 0.71 1 [12, 13] Nifedipine 686.00 370.44 25 973.00 379.47 25 0.71 0.56 0.88 2 [12, 14] Tacrolimus 1.02 0.91 1.13 2 [15] (+)-Tramadol 4.88 5.1E-4 29 5.48 5.2E-4 29 0.89 0.89 0.89 1 [16, 17] (-)-Tramadol 3.95 5.1E-4 29 4.41 5.2E-4 29 0.90 0.90 0.90 1 [16, 17] Tramadola 0.90 0.85 0.95 1 [17, 18] Tramadola 1.00 0.97 1.03 1 [17, 19] Cyclonezaprinea 0.91 0.85 0.98 1 [12, 20] Gepirone 51.80 27.30 12 54.90 25.60 12 0.94 0.65 1.34 3 [21-23] Gepirone 55.00 33.70 12 54.90 25.60 12 1.00 0.66 1.46 3 [21-23] Gepirone 55.30 28.20 12 54.90 25.60 12 1.01 0.70 1.42 3 [21-23] Propiverine 554.00 215.00 10 559.00 330.00 10 0.99 0.69 1.53 3 [24-26] Propiverine 1560.00 670.00 10 1520.00 737.00 10 1.03 0.72 1.49 3 [24-26] Propiverine 798.00 282.00 10 733.00 437.00 10 1.09 0.76 1.68 3 [24-26] Quetiapinea 1.04 0.92 1.19 1 [12, 27] Oxybutynina 1.53 1.17 1.75 1 [28-30] Buspirone 19.14 21.28 18 11.97 14.64 16 1.60 0.78 3.59 1 [10, 31, 32]

154

Buspirone 20.00 22.58 17 14.64 12.90 17 1.67 0.78 3.69 1 [10, 32, 33] Buspirone 15.56 14.12 33 9.10 14.25 33 1.71 1.05 3.26 1 [10, 32, 33] Simvastatin 44.17 27.83 12 20.47 19.04 12 2.16 1.27 4.21 2 [12, 34] Simvastatin 44.92 14.88 7 14.81 7.28 7 3.03 2.07 4.74 2 [12, 35]

AUCCR, mean AUC for the controlled release formulation; AUCIR, mean AUC for the immediate release formulation; NCR, number of subjects employed for the calculation of AUCCR; NIR, number of subjects employed for the calculation of AUCIR ; BCS/BDCSS, biopharmaceutics classification system/biopharmaceutics drug disposition system classification system. a the relative bioavailability (AUC ratio) and its 90% confidence interval was reported in the study. b duplicated compounds are different studies and/or diffrenet CR formulations.

155

Table A3.3. Clearance and relative bioavailability values employed for the comparison of the simulations with the observed data.

Compound fup BP CLh, in vivo CLint,u (HLM) in vitro CLint,u (HLM) in vivo krel Reference (L/h) (µL/min/mg) (µL/min/mg) (h-1) Buspirone (F1) 0.05 0.81 119 268 to 244 5454a 0.288 [8-10, 33, 36] Buspirone (F2) 0.05 0.81 119 268 to 244 5454a 0.105 [8-10, 33, 36] Cyclobenzaprine 0.07 0.6* 41.34 153b 368c 0.132 [37-40] Oxybutynin 0.0067 0.69* 34.1 78 to 278 2932 0.144 [29, 41-43] Quetiapine 0.17 1.26* 138d 253e 206f 0.144g [44-47]

fup, fraction unbound in plasma; BP, blood to plasma ratio; CLh, in vivo, in vivo hepatic plasma clearance; CLint,u (HLM) in vivo, unbound intrinsic clearance in human liver microsomes calculated from the in vivo clearance data; CLint,u (HLM) in vitro, unbound intrinsic clearance in human liver microsomes determined from in vitro data. * estimated using Simcyp’s QSAR model. a The values were calculated from the dispertion model as described previously. b the fraction unbund in the incubation ( fu,inc) was estimated from the equation derived by Hallifax and Houston [48], asuming a protein concentration of 1 mg/mL in the incubation. c based on in vitro inhibition studies the proportion of cyclobenzaprine metabolized by CYP3A4 was assumed to be 50%. d Based on apparent clearance from oral data (CLpo) e The unbound intrisic clearance for CYP3A4 was reported in the recombinant system (CLint,u rhCYP3A4), this was converted into CLint,u (HLM) by multypling it by the CYP3A4 abundance in the liver (137 pmol/mg protein) and by an inter system extrapolation factor (ISEF) of 0.21[49]. f The intrinsic clearance reported by Johnson and co-workers (2014) was directly used, given that their calculation method was similar the one employed for the other compounds. g t90 assumed 16 hours as per clinical study and oxybutynin OROS release profile

156

A3.4 Supplementary Figures

Figure A3.1. Impact of release rate (formulation) and CLint, CYP3A4 on AUC (A), fa and FG (B) for non-P-gp substrates. Vmax, CYP3A4 was fixed at 500 pmol/min/mg whereas the Km, CYP3A4 was varied (scenario Ia in Table 4.1). For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, CYP3A4 values: upper triangle (500), circle (50), square (10), diamond (5), and lower triangle (0.05). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

157

Figure A3.2. Impact of release rate (formulation) and CLint, CYP3A4 on AUC (A), fa and FG (B) for non-P-gp substrates. Km, CYP3A4 was fixed at 50 µM whereas the Vmax, CYP3A4 was varied (scenario Ib in Table 4.1).For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, CYP3A4 values: upper triangle (0.02), circle (2), square (10), diamond (50), and lower triangle (200). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

158

Figure A3.3. Impact of release rate (formulation) and CLint, CYP3A4 on AUC (A), fa and FG (B) for non-P-gp substrates. Km, CYP3A4 was fixed at 1 µM whereas the Vmax, CYP3A4 was varied (scenario IIb in Table 4.1). For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, CYP3A4 values: upper triangle (1), circle (100), square (500), diamond (2500), and lower triangle (10000). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

159

Figure A3.4. Impact of release rate (formulation) and CLint, P-gp (efflux) on AUC (A), fa and FG (B) for non-CYP3A4 substrates. Km, P-gpwas fixed at 150 µM whereas the Jmax, P-gp was varied (scenario IIIb in Table 4.1). For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, P-gp values:. upper triangle (0.007), circle (0.2), square (2), diamond (3), and lower triangle (10). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

160

Figure A3.5. Impact of release rate (formulation), CLint, CYP3A4, and CLint, P-gp on AUC (A), fa and FG (B). Km, CYP3A4 was fixed at 1 µM whereas the Vmax, CYP3A4 was varied, CLint, P-gp was fixed to 2 µL/min (scenario IVb in Table 4.1). For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, P-gp values: upper triangle (300), circle (6), square (2), diamond (1), and lower triangle (0.15). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

161

Figure A3.6. Impact of release rate (formulation), CLint, CYP3A4, and CLint, P-gp on AUC (A), fa and FG (B). Jmax, P-gp was fixed at 300 pmol/min whereas the Km, P-gp was varied (scenario VIa in Table 4.1). CLint, CYP3A4 was fixed to 2500 µL/min/mg. For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, P-gp values: upper triangle (300), circle (6), square (2), diamond (1), and lower triangle (0.15). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

162

Figure A3.7. Impact of release rate (formulation), CLint, CYP3A4, and CLint, P-gp on AUC (A), fa and FG (B). Km, P-gp was fixed at 150 µM whereas the Jmax, P-gp was varied, CLint, CYP3A4 was fixed to 2500 µL/min/mg (scenario Vb in Table 4.1). For plots A and B, the subplots represent the different BCS classes (1-4), whereas the symbols in each plot represent different CLint, P-gp values: upper triangle (0.007), circle (0.2), square (2), diamond (3), and lower triangle (10). For the plots in the right hand side (B), the green lines and open symbols represent the FG, whereas the black lines and filled symbols represent the fa.

163

A3.5 Further discussion

A3.5.1 Discussion on Impact of the intestinal P-gp distribution on the bioavailability of CR formulations

The results for the simulated P-gp substrates were consistent with the previous work by Darwich and co-workers (2010). In general both absorption and exposure were decreased when CLint, P-gp was increased [50]. However, the impact of CLint, P-gp on the exposure of BCS class 1 compounds was minimal. BCS class 2 compounds seemed to be the most affected by the P-gp-mediated efflux (Figure 4.4 and Figure A3.4). It has been suggested that transporters would have little clinical impact for BCS/BDDCS (Biopharmaceutical Drug Disposition System) class 1 compounds, whereas for class 2 compounds the impact of efflux transporters would be clinically noticeable. A similar statement has been made for Classes 3 and 4 compounds [51, 52], which was in agreement with our results for simulated P-gp substrates. Nevertheless, in our study a clear reduction in the AUC was observed for BCS class

1 compounds for which CLint, P-gp was equal to 300 µL/min (Figure 4.4 A and Figure A3.4A). This clearance may however be highly exaggerated as the reported apparent in vitro CLint, P-gp (efflux) in Caco-2 cell monolayers varies from 0.8 to 5.5 µL/min for several drugs [53]; as a reference, the intrinsic clearance of the well- known P-gp-probe substrate digoxin was reported to be 2.5 µL/min [53, 54]. It has been suggested that the expression of P-gp increases towards the distal regions of the human intestine, a pattern that is accounted for in the ADAM model [55, 56]. However, no significant absorption differences were observed between the IR and CR formulations as function of varying P-gp clearance. These results were in line with the findings by Tannergren and co-workers (2009), where no difference in colonic absorption were found for P-gp substrates compared to non-substrates [57].

A3.5.2 Discussion on the analysis of the possible CYP3A4/P-gp interplay

It has been hypothesized that the combined role of CYP3A4 and P-gp acts as a barrier for the absorption of xenobiotics [58, 59]. This is based on several factors such as: the overlapped substrate affinity between CYP3A4 and P-gp, similar location in the villous tips of the enterocytes and a possible coordinated regulation of

164

both transporter and metabolic enzyme [60]. One of the proposed mechanisms is the possible recirculation of substrates by P-gp, leading to an increased residence time in the intracellular space of the enterocytes and therefore an increase of the probability for CYP3A4-mediated metabolism [58-63]. Several in silico modelling approaches have backed up this hypothesis[50, 64-66], as well as in vitro studies employing CYP3A4-transfected Caco-2 cells to simultaneously analyse the metabolism and P- gp-mediated efflux [67, 68]. There is still however an ongoing debate whether or not there is a synergistic (or additive) mechanism between the enzyme and transporter leading to a reduction in oral drug bioavailability, and thus more experimental data has been encouraged[69-71]. Herein, the differences between IR and CR formulation for the combined CYP3A4 and P-gp substrates were investigated. The simulations were performed on high affinity CYP3A4 substrates with a fixed efflux CLint, P-gp, similar to the one for digoxin (CLint, P-gp (efflux) ≈2 µL/min) with the intention to represent the worst case scenario. Likewise, in another scenario, the CLint, CYP3A4 was fixed to the value where we observed an increased AUC for CR formulations (2500 µL/min/mg). Yet, no significant differences on the AUC trend were observed for the combined CYP3A4 and P-gp substrates compared to the trends observed when they were analysed separately (e.g., Figure 4.5A and Figure A3.6A). In contrast, although not noticeable on the simulated drug exposure, there were signs of a minor interaction between CYP3A4 and P-gp affecting both fa and FG. In Figure 4.5B a minor increase in fa as a function of the increase in CLint, CYP3A4 for IR formulations can be noticed. This was more noticeable for BCS class 2 drugs. The mechanism for this minor increase is yet unknown, although it could be a result of competition between

CYP3A4 and P-gp for substrate elimination; the utilised Km, CYP3A4 value for this simulation was 1 µM whereas the Km, P-gp was 150 µM. On the other hand, in Figure

A3.6B, a minor increase in FG as a function of the increased CLint, P-gp, solely for IR formulations of BCS class 1 compounds was observed. Similarly as for fa, the mechanism for this increase is yet unknown. However, there is the possibility that the P-gp efflux could shift the absorption of the drug towards the distal regions of the GI tract where there is less abundance of CYP3A4, thus reducing the CYP3A4-mediated metabolism. This would only be noticeable for BCS class 1 drugs which exhibit unimpaired absorption in the distal regions of the intestine [57]. Nevertheless, it

165

would be difficult to observe these differences in vivo, as the changes in AUC are minimal (Figure A3.6A and Figure A3.7A). Therefore more data is needed in order to confirm the above mentioned suggestions.

166

A3.6 References

1. E.C. Fieller. Some Problems in Interval Estimation. Journal of the Royal Statistical Society Series B-Statistical Methodology. 16:175-185 (1954). 2. H. Motulsky. Intuitive biostatistics: a nonmathematical guide to statistical thinking, Oxford University Press2010. 3. K.S. Pang and M. Rowland. Hepatic clearance of drugs. I. Theoretical considerations of a “well-stirred” model and a “parallel tube” model. Influence of hepatic blood flow, plasma and blood cell binding, and the hepatocellular enzymatic activity on hepatic drug clearance. J Pharmacokinet Biopharm. 5:625-653 (1977). 4. M. Kato, K. Chiba, A. Hisaka, M. Ishigami, M. Kayama, N. Mizuno, Y. Nagata, S. Takakuwa, Y. Tsukamoto, K. Ueda, H. Kusuhara, K. Ito, and Y. Sugiyama. The intestinal first-pass metabolism of substrates of CYP3A4 and P-glycoprotein-quantitative analysis based on information from the literature. Drug Metab Pharmacokinet. 18:365-372 (2003). 5. Z.E. Barter, M.K. Bayliss, P.H. Beaune, A.R. Boobis, D.J. Carlile, R.J. Edwards, J.B. Houston, B.G. Lake, J.C. Lipscomb, O.R. Pelkonen, G.T. Tucker, and A. Rostami-Hodjegan. Scaling factors for the extrapolation of in vivo metabolic drug clearance from in vitro data: reaching a consensus on values of human microsomal protein and hepatocellularity per gram of liver. Curr Drug Metab. 8:33-45 (2007). 6. M. Roberts and M. Rowland. A dispersion model of hepatic elimination: 1. Formulation of the model and bolus considerations. J Pharmacokinet Biopharm. 14:227-260 (1986). 7. M.S. Roberts and M. Rowland. Correlation between in-vitro microsomal enzyme activity and whole organ hepatic elimination kinetics: analysis with a dispersion model. J Pharm Pharmacol. 38:177-181 (1986). 8. M. Gertz, J.B. Houston, and A. Galetin. Physiologically based pharmacokinetic modeling of intestinal first-pass metabolism of CYP3A substrates with high intestinal extraction. Drug Metab Dispos. 39:1633-1642 (2011). 9. Y. Shibata, H. Takahashi, M. Chiba, and Y. Ishii. Prediction of hepatic clearance and availability by cryopreserved human hepatocytes: an application of serum incubation method. Drug Metab Dispos. 30:892-896 (2002). 10. R.E. Gammans, R.F. Mayol, and J.A. Labudde. Metabolism and disposition of buspirone. The American Journal of Medicine. 80:41-51 (1986). 11. H.R. Xu, W.L. Chen, N.N. Chu, X.N. Li, and J.R. Zhu. The difference in pharmacokinetics and pharmacodynamics between extended-release fluvastatin and immediate-release fluvastatin in healthy Chinese subjects. J Biomed Biotechnol. 2012:386230 (2012). 12. L.Z. Benet, F. Broccatelli, and T.I. Oprea. BDDCS applied to over 900 drugs. AAPS J. 13:519-547 (2011). 13. D. Barilla, P. Prasad, M. Hubert, and K. Gumbhir-Shah. Steady-state pharmacokinetics of fluvastatin in healthy subjects following a new extended release fluvastatin tablet, Lescol® XL. Biopharm Drug Dispos. 25:51-59 (2004). 167

14. C.B. Toal. Formulation dependent pharmacokinetics--does the dosage form matter for nifedipine? J Cardiovasc Pharmacol. 44:82-86 (2004). 15. U.S. Food and Drug Administration. ASTAGRAF XL™ Product Label. http://www.accessdata.fda.gov/drugsatfda_docs/label/2013/204096s000lbl.pd f (accessed 11/03/2014 2014). 16. H. Malonne, B. Sonet, B. Streel, S. Lebrun, S. De Niet, A. Sereno, and F. Vanderbist. Pharmacokinetic evaluation of a new oral sustained release dosage form of tramadol. Br J Clin Pharmacol. 57:270-278 (2004). 17. M.A. Lassoued, F. Khemiss, and S. Sfar. Comparative study of two in vitro methods for assessing drug absorption: Sartorius SM 16750 apparatus versus Everted Gut Sac. J Pharm Pharm Sci. 14:117-127 (2011). 18. M. Raber, H.U. Schulz, M. Schurer, U. Bias-Imhoff, and H. Momberger. Pharmacokinetic properties of tramadol sustained release capsules. 2nd communication: investigation of relative bioavailability and food interaction. Arzneimittelforschung. 49:588-593 (1999). 19. M. Raber, H.U. Schulz, M. Schurer, S. Krupp, and H. Momberger. Pharmacokinetic properties of tramadol sustained release capsules. 3rd communication: investigation of relative bioavailability under steady state conditions. Arzneimittelforschung. 49:594-598 (1999). 20. M. Darwish, E.T. Hellriegel, and F. Xie. Single-dose pharmacokinetics of once-daily cyclobenzaprine extended release 30 mg versus cyclobenzaprine immediate release 10 mg three times daily in healthy young adults : a randomized, open-label, two-period crossover, single-centre study. Clin Drug Investig. 28:793-801 (2008). 21. C.J. Timmer and J.M. Sitsen. Pharmacokinetic evaluation of gepirone immediate-release capsules and gepirone extended-release tablets in healthy volunteers. J Pharm Sci. 92:1773-1778 (2003). 22. R.J. Behme, D. Brooke, R.F. Farney, and T.T. Kensler. Characterization of polymorphism of gepirone hydrochloride. J Pharm Sci. 74:1041-1046 (1985). 23. P. Dogterom, J.M. Huisman, R. Gellert, and A. Verhagen. Pharmacokinetics of Gepirone in Subjects with Normal Renal Function and in Patients with Chronic Renal Dysfunction. Clin Drug Investig. 22:513-522 (2002). 24. K. May, T. Giessmann, D. Wegner, R. Oertel, C. Modess, S. Oswald, M. Braeter, and W. Siegmund. Oral absorption of propiverine solution and of the immediate and extended release dosage forms: influence of regioselective intestinal elimination. Eur J Clin Pharmacol. 64:1085-1092 (2008). 25. K.O. Haustein and G. Huller. On the pharmacokinetics and metabolism of propiverine in man. Eur J Drug Metab Pharmacokinet. 13:81-90 (1988). 26. W. Siegmund, J. Siegert, K. Richter, F. Schnabel, C. Feustel, and W. Kirch. Influence of a fat-rich meal on bioavailability of extended-release and immediate-release propiverine. J Clin Pharmacol. 52:681-690 (2012). 27. C. Figueroa, M. Brecher, J.E. Hamer-Maansson, and H. Winter. Pharmacokinetic profiles of extended release quetiapine fumarate compared with quetiapine immediate release. Prog Neuropsychopharmacol Biol Psychiatry. 33:199-204 (2009). 28. S.K. Gupta and G. Sathyan. Pharmacokinetics of an oral once-a-day controlled-release oxybutynin formulation compared with immediate-release oxybutynin. J Clin Pharmacol. 39:289-296 (1999).

168

29. J. Douchamps, F. Derenne, A. Stockis, D. Gangji, M. Juvent, and A. Herchuelz. The pharmacokinetics of oxybutynin in man. Eur J Clin Pharmacol. 35:515-520 (1988). 30. E. Miyamoto, S. Kawashima, Y. Murata, Y. Yamada, Y. Demizu, H. Kontani, and T. Sakai. Physico-chemical properties of oxybutynin. Analyst. 119:1489-1492 (1994). 31. A. Sakr and M. Andheria. A comparative multidose pharmacokinetic study of buspirone extended-release tablets with a reference immediate-release product. J Clin Pharmacol. 41:886-894 (2001). 32. M. Sheikhzadeh, S. Rohani, M. Taffish, and S. Murad. Solubility analysis of buspirone hydrochloride polymorphs: measurements and prediction. Int J Pharm. 338:55-63 (2007). 33. A. Sakr and M. Andheria. Pharmacokinetics of buspirone extended-release tablets: a single-dose study. J Clin Pharmacol. 41:783-789 (2001). 34. S.B. Jang, Y.J. Lee, L.A. Lim, K.M. Park, B.J. Kwon, J.S. Woo, Y.I. Kim, M.S. Park, K.H. Kim, and K. Park. Pharmacokinetic comparison of controlled-release and immediate-release oral formulations of simvastatin in healthy Korean subjects: a randomized, open-label, parallel-group, single- and multiple-dose study. Clin Ther. 32:206-216 (2010). 35. M. Tubic-Grozdanis, J.M. Hilfinger, G.L. Amidon, J.S. Kim, P. Kijek, P. Staubach, and P. Langguth. Pharmacokinetics of the CYP 3A substrate simvastatin following administration of delayed versus immediate release oral dosage forms. Pharm Res. 25:1591-1600 (2008). 36. M. Zhu, W. Zhao, H. Jimenez, D. Zhang, S. Yeola, R. Dai, N. Vachharajani, and J. Mitroka. Cytochrome P450 3A-mediated metabolism of buspirone in human liver microsomes. Drug Metab Dispos. 33:500-507 (2005). 37. U.S. Food and Drug Administration. AMRIX® Clinical Pharmacology and Biopharmaceutics Review. http://www.accessdata.fda.gov/drugsatfda_docs/label/2013/204096s000lbl.pd f (accessed 11/03/2014 2014). 38. H.B. Hucker, S.C. Stauffer, A.J. Balletto, S.D. White, A.G. Zacchei, and B.H. Arison. Physiological disposition and metabolism of cyclobenzaprine in the rat, dog, rhesus monkey, and man. Drug Metab Dispos. 6:659-672 (1978). 39. G.A. Winchell, J.D. King, C.M. Chavez-Eng, M.L. Constanzer, and S.H. Korn. Cyclobenzaprine pharmacokinetics, including the effects of age, gender, and hepatic insufficiency. J Clin Pharmacol. 42:61-69 (2002). 40. R.W. Wang, L. Liu, and H. Cheng. Identification of human liver cytochrome P450 isoforms involved in the in vitro metabolism of cyclobenzaprine. Drug Metab Dispos. 24:786-791 (1996). 41. G. Sathyan, R.R. Dmochowski, R.A. Appell, C. Guo, and S.K. Gupta. Effect of antacid on the pharmacokinetics of extended-release formulations of tolterodine and oxybutynin. Clin Pharmacokinet. 43:1059-1068 (2004). 42. H. Mizushima, K. Takanaka, K. Abe, I. Fukazawa, and H. Ishizuka. Stereoselective pharmacokinetics of oxybutynin and N-desethyloxybutynin in vitro and in vivo. Xenobiotica. 37:59-73 (2007). 43. A. Shibukawa, N. Ishizawa, T. Kimura, Y. Sakamoto, K. Ogita, Y. Matsuo, Y. Kuroda, C. Matayatsu, T. Nakagawa, and I.W. Wainer. Plasma protein binding study of oxybutynin by high-performance frontal analysis. J Chromatogr B Analyt Technol Biomed Life Sci. 768:177-188 (2002).

169

44. G.V. Bakken, I. Rudberg, H. Christensen, E. Molden, H. Refsum, and M. Hermann. Metabolism of Quetiapine by CYP3A4 and CYP3A5 in Presence or Absence of Cytochrome B5. Drug Metab Dispos. 37:254-258 (2009). 45. C.L. DeVane and C.B. Nemeroff. Clinical pharmacokinetics of quetiapine: an atypical antipsychotic. Clin Pharmacokinet. 40:509-522 (2001). 46. T.N. Johnson, D. Zhou, and K.H. Bui. Development of physiologically based pharmacokinetic model to evaluate the relative systemic exposure to quetiapine after administration of IR and XR formulations to adults, children and adolescents. Biopharm Drug Dispos. 35:341-352 (2014). 47. S.W. Grimm, N.M. Richtand, H.R. Winter, K.R. Stams, and S.B. Reele. Effects of cytochrome P450 3A modulators ketoconazole and carbamazepine on quetiapine pharmacokinetics. Br J Clin Pharmacol. 61:58-69 (2006). 48. D. Hallifax and J.B. Houston. Binding of drugs to hepatic microsomes: comment and assessment of current prediction methodology with recommendation for improvement. Drug Metab Dispos. 34:724-726; author reply 727 (2006). 49. Y. Chen, L. Liu, K. Nguyen, and A.J. Fretland. Utility of Intersystem Extrapolation Factors in Early Reaction Phenotyping and the Quantitative Extrapolation of Human Liver Microsomal Intrinsic Clearance Using Recombinant Cytochromes P450. Drug Metab Dispos. 39:373-382 (2011). 50. A.S. Darwich, S. Neuhoff, M. Jamei, and A. Rostami-Hodjegan. Interplay of metabolism and transport in determining oral drug absorption and gut wall metabolism: a simulation assessment using the "Advanced Dissolution, Absorption, Metabolism (ADAM)" model. Curr Drug Metab. 11:716-729 (2010). 51. J.M. Custodio, C.Y. Wu, and L.Z. Benet. Predicting drug disposition, absorption/elimination/transporter interplay and the role of food on drug absorption. Adv Drug Deliv Rev. 60:717-733 (2008). 52. C.Y. Wu and L.Z. Benet. Predicting drug disposition via application of BCS: Transport/absorption/elimination interplay and development of a biopharmaceutics drug disposition classification system. Pharm Res. 22:11- 23 (2005). 53. M. Troutman and D. Thakker. Efflux Ratio Cannot Assess P-Glycoprotein- Mediated Attenuation of Absorptive Transport: Asymmetric Effect of P- Glycoprotein on Absorptive and Secretory Transport Across Caco-2 Cell Monolayers. Pharm Res. 20:1200-1209 (2003). 54. M.E. Taub, L. Podila, D. Ely, and I. Almeida. Functional assessment of multiple P-glycoprotein (P-gp) probe substrates: influence of cell line and modulator concentration on P-gp activity. Drug Metab Dispos. 33:1679-1687 (2005). 55. S. Mouly and M. Paine. P-Glycoprotein Increases from Proximal to Distal Regions of Human Small Intestine. Pharm Res. 20:1595-1599 (2003). 56. S. Oswald, C. Groer, M. Drozdzik, and W. Siegmund. Mass Spectrometry- Based Targeted Proteomics as a Tool to Elucidate the Expression and Function of Intestinal Drug Transporters. AAPS J. 10.1208/s12248-013- 9521-3:(2013). 57. C. Tannergren, A. Bergendal, H. Lennernas, and B. Abrahamsson. Toward an increased understanding of the barriers to colonic drug absorption in humans: implications for early controlled release candidate assessment. Mol Pharm. 6:60-73 (2009). 170

58. P.B. Watkins. The barrier function of CYP3A4 and P-glycoprotein in the small bowel. Advanced Drug Delivery Reviews. 27:161-170 (1997). 59. L.Z. Benet, C.-Y. Wu, M.F. Hebert, and V.J. Wacher. Intestinal drug metabolism and antitransport processes: A potential paradigm shift in oral drug delivery. J Controlled Release. 39:139-143 (1996). 60. L.Z. Benet. The drug transporter-metabolism alliance: uncovering and defining the interplay. Mol Pharm. 6:1631-1643 (2009). 61. L.Z. Benet, T. Izumi, Y. Zhang, J.A. Silverman, and V.J. Wacher. Intestinal MDR transport proteins and P-450 enzymes as barriers to oral drug delivery. J Control Release. 62:25-31 (1999). 62. L.Z. Benet and C.L. Cummins. The drug efflux-metabolism alliance: biochemical aspects. Adv Drug Deliv Rev. 50 Suppl 1:S3-11 (2001). 63. L. Benet, G. Amidon, D. Barends, H. Lennernäs, J. Polli, V. Shah, S. Stavchansky, and L. Yu. The Use of BDDCS in Classifying the Permeability of Marketed Drugs. Pharm Res. 25:483-488 (2008). 64. R. Badhan, J. Penny, A. Galetin, and J.B. Houston. Methodology for development of a physiological model incorporating CYP3A and P- glycoprotein for the prediction of intestinal drug absorption. J Pharm Sci. 98:2180-2197 (2009). 65. K. Ito, H. Kusuhara, and Y. Sugiyama. Effects of intestinal CYP3A4 and P- glycoprotein on oral drug absorption--theoretical approach. Pharm Res. 16:225-231 (1999). 66. T. Watanabe, K. Maeda, C. Nakai, and Y. Sugiyama. Investigation of the effect of the uneven distribution of CYP3A4 and P-glycoprotein in the intestine on the barrier function against xenobiotics: a simulation study. J Pharm Sci. 102:3196-3204 (2013). 67. C.L. Cummins, W. Jacobsen, and L.Z. Benet. Unmasking the dynamic interplay between intestinal P-glycoprotein and CYP3A4. J Pharmacol Exp Ther. 300:1036-1045 (2002). 68. C.L. Cummins, W. Jacobsen, U. Christians, and L.Z. Benet. CYP3A4- transfected Caco-2 cells as a tool for understanding biochemical absorption barriers: studies with sirolimus and midazolam. J Pharmacol Exp Ther. 308:143-155 (2004). 69. U. Christians, V. Schmitz, and M. Haschke. Functional interactions between P-glycoprotein and CYP3A in drug metabolism. Expert Opin Drug Metab Toxicol. 1:641-654 (2005). 70. K. Thelen and J.B. Dressman. Cytochrome P450-mediated metabolism in the human gut wall. J Pharm Pharmacol. 61:541-558 (2009). 71. R.A.B. van Waterschoot and A.H. Schinkel. A Critical Analysis of the Interplay between Cytochrome P450 3A and P-Glycoprotein: Recent Insights from Knockout and Transgenic Mice. Pharmacol Rev. 63:390-410 (2011).

171

Appendix A4: Supplementary material for

Chapter 5

172

A4.1 Non-linear regression for the data reported by Wilson (1967)

The data form Wilson’s work was digitized using GetData Graph Digitizer v2.26 (http://getdata-graph-digitizer.com/) and fitted by an exponential model[1]. This fitting was done using the using the “lsqnonlin” function of the Optimization Toolbox within Matlab 2014a (The Mathworks Inc., Natick, MA, USA).The method performs non-linear least squares regression with the Levenberg-Marquardt algorithm, minimizing the ordinary least squares (OLS) objective function (objfun) described by Equation A4.1

n Equation A4.1 objfun(()(,)) obsx y xp 2  pred n 1

where, ypred(x,p) is the model prediction for a given x and a vector of parameters, p. The function lsqnonlin finds the set of parameters (p) that minimize Equation A4.1. The result of the fitting is shown in Figure A4.1

Figure A4.1. Nonlinear fit of an exponential model to Wilson’s (1967) data (black solid circles). The solid black line is the regression line; the dashed redline is the 95% confidence interval (CI); and the dashed blue line represents the 95% prediction intervals (PI). The precision (CV) of the coefficients, λ1 and λ2, was found to be 0.66% and 35%, respectively.

173

A4.2 Estimated segmental SA for the reference human intestine

Table A4.1 Segmental mSA estimated by the three different methods for a reference man.

Segment Length Radius mSA Volume (cm) (cm) (cm2) (cm3) M1 M2 M3 Duodenum 53.66 2.37 7.99×102 1.76×105 7.50×104 9.47×102 Jejunum 248.16 1.75 2.73×103 4.49×105 5.19×105 2.39×103 Ileum 368.89 1.50 3.48×103 1.60×105 3.86×105 2.61×103 Total SI (1) 670.7 - 7.00×103 7.85×105 9.80×105 Ascending 16.69 2.42 Colon 2.54×102 1.62×103 1.62×103 3.07×102 Total Colon (2) 104.34 2.42 1.59×103 1.02×104 1.02×104 1.92×103 Total 775.04 - Intestine(1+2) 8.59×103 7.95×105 9.90×105 -

174

A4.3 Recalculation of the regional Peff values from their original references

A4.3.1 Triamcinolone acetonide and hydrocortisone (Schedl 1965)

The absorption data informed by Schedl[2] was expressed in terms of the percentage absorbed or fraction absorbed (fabs) from the given test segment (Equation A4.2)

C out Equation A4.2 fabs1   PEG ratio C in

where, Cout and Cin are the concentrations of drug leaving and entering the test segment, respectively; PEGratio is the ratio of the non-absorbable marker, polyethylene glycol (PEG), entering and leaving the test segment. This ratio was employed to correct the concentration for any changes due to the net fluid transfer in the segment (absorption or secretion). Thus, the fluid-corrected concentration ratio (

C ' out ) can be defined as shown in Equation A4.3 C in

CCout' out Equation A4.3 PEGratio CC in in

This ratio was employed for the calculation of the regional absorption clearance

(CLabs,i ) and (Peff,i) according to Equation A4.4 and Equation A4.5, respectively.

C out ' Equation A4.4 C Labs, i Q in ln() C in

C out ' 1 Equation A4.5 Peff, i  Q in ln()  CSA in

Qin was informed as 0.25 mL/s; the surface area (SA) term in Equation A4.5 can be either the cylindrical SA or the mucosal SA (mSA) in the test segment, calculated as described in the methods section of the manuscript.

175

A4.3.2 Hydrochlorothiazide, atenolol, furosemide, cimetidine and salicylic acid (Sutcliffe et al, 1988)

The absorption data informed by Sutcliffe and co-workers [3] was expressed in terms of the drug lost (percentage) from the intestinal segment. This was assumed as the percentage of drug absorbed or fabs in the given segment. The values reported were corrected by fluid movements (net secretion or reabsorption).This was determined by the use of a non-absorbable marker (PEG 4000). The calculations of CLabs,i and Peff,i for each drug and each segment were performed as described above, using the reported Qin of 0.0833 mL/s.

A4.3.3 Griseofulvin, ranitidine, paracetamol and talinolol (Gramatté, et al. 1994-1996)

The absorption data informed in the works by Gramatté, and co-workers [4-7], differs from the data informed in the previous studies. In this case the reported -1 absorption is expressed in term of the net absorption rate (Δabs, drug) [µg (Li min) ], where Li is the length of the test segment employed during the perfusion experiment. The mass balance equation employed for such calculations is described by Equation A4.6[8]

 QCQC  Equation A4.6 absdrug, in in out out

where, and Cin and Cout represent the drug concentrations entering and leaving the test segment, respectively; Qin and Qout are the corresponding drug flow rates entering and leaving the test segment, determined according to Equation A4.7 and Equation A4.8

[]PEG perf Equation A4.7 Qin Inf S p []PEG p

Equation A4.8 []PEG p QQout in []PEG d

176

where, Inf is the perfusate infusion rate (mL/min); [PEG]perf, [PEG]p, and [PEG]d are the corresponding PEG concentrations in the perfusate, proximal, and distal collection ports of the test segment; Sp is the sampling rate (mL/min) from the proximal collection port of the test segment. By rearranging Equation A4.6 and combination with Equation A4.3 and Equation A4.8, the fluid transfer-corrected

C ' concentration ratio ( out ) can be derived (Equation A4.9), C in

 Equation A4.9 CCout' out abs, drug PEGratio 1  CCQC in in in in

where, the denominator in the right hand side of Equation A4.9 (QinCin) is the informed drug’s perfusion rate. This ratio is needed to estimate CLabs,i and Peff,i , as - per Equation A4.4 and Equation A4.5. For ranitidine, the perfusion rate [µg (Li min) 1] was not informed in the original reference [6], therefore this was assumed as the informed initial perfusion rate (ml/s) multiplied by the drug nominal concentration in the perfusate

A4.4 mSAT model development

A4.4.1 Structure of the mSAT model

The mSAT model describes the gastrointestinal (GI) tract by means of five compartments. These compartments are meant to represent the anatomical segments of the GI tract that are relevant for drug absorption: stomach, duodenum, jejunum, ileum and ascending colon. A schematic representation of the mSAT model is shown in Figure A4.2.

177

Figure A4.2. Schematic representation of the minimal Segment Absorption and Transit (mSAT) model representing the human GI tract by five consecutive compartments: stomach, the small intestine ( duodenum(duo), jejunum(jej) and ileum(ile)), and the large intestine ( ascending colon). A detailed explanation of the model and model parameters can be found in the material and methods section of the manuscript.

A4.4.2 Optimization of the small intestinal transit time for the mSAT model

The intestinal transit time data form Yu and co-workers [9] was digitized using GetData Graph Digitizer v2.26 (http://getdata-graph-digitizer.com/) and fitted simultaneously by the system of differential equations shown in Equation A4.10.

dA Equation A4.10 duo  k() t  A dt t, duo duo dA jej k()() t  A  k t  A dt tduo,, duo tjej jej dA ile k()() t  A  k t  A dt tjej,, jej tile ile

dAcol kt, ile() t A ile dt

178

This system describes the transit of particles along the small intestine where,

 1  t kttn, () is the segment-specific (and time dependent) transit rate nn between the intestinal segments; β is the Weibull shape parameter, and αn is the segment-dependent scale parameter. The scale parameter was defined for the nth

small intestinal segment as, ann f  SITT   ; where, fn is the fractional length of the intestinal segment (with respect to the total length of the small intestine, Lsi); SITT, is the mean intestinal transit time, assumed as 199 min [9] ; and γ is a proportionality coefficient (assumed the same for all the segments) .The fitting was performed following the same procedure as the one described for the fitting of Wilson’s data , this time the estimated parameters were β and γ. The coefficients, β and γ, were found to be 2.01 and 1.57, and the relative standard errors (RSE%) associated to the parameter estimates were relatively low, 5.6% and 2.90%, for β and γ, respectively.

A4.4.3 Comparison of the mSAT model with alternative transit models

The ability to describe the SITT data of the mSAT model was contrasted with that of different transit models. All the parameters employed for the mSAT transit model (Weibull) and the alternative models are described in Table A4.2. The evaluated alternative models are described below: 1. The traditional compartmental transit model described by Yu and co-workers (CAT)[9]. This model describes the mean intestinal transit by series of 7 transit compartments, each one with the same mean residence time (MRT). For this model a first order transfer rate is assumed. The transit rate constant,

kt, is assumed to be the same for all the transit compartments and is defined

as: kt = 7/SITT [9]. 2. A similar model to the one described above but, instead of seven intestinal transit compartments, the transit was described by only three compartments.

For each intestinal compartment the transit rate constant (kt) was given by 3/SITT. The third model evaluated was similar to model described previously. However, the main difference was the treatment of the transit rate constant between each intestinal -1 compartment, this was given by kt,n = (fn×SITT) , where fn is the fractional length of

179

the intestinal segment. The fractional lengths of the duodenum, jejunum and ileum were assumed as 0.08, 0.37, and 0.55, respectively (Table A4.2). Therefore the MRT for each segment was assumed as 16, 74, and 109 minutes, for the duodenum, jejunum and ileum, respectively. The description of the SITT data by the different models is shown in Figure A4.3. Both the Weibull transit model and the full CAT model adequately described the SITT data [9]. The alternative transit models, on the other hand, tended to overestimate the amount of drug reaching the colon prior to the first 210 minutes, while underestimating it after that time.

Figure A4.3. Comparison between different small-intestinal transit models to describe SITT data. The lines represent the cumulative fraction of dose reaching the colon for the different small intestinal transit models. Red solid line, mSAT model (Weibull transfer between segments); dot-dashed cyan line, full CAT model (seven transit compartments); dashed blue line, CAT model with only three compartments ( same first-order rate constant for all the segments); Dotted green line, CAT model only three compartments, where the transit was fractionally divided for each segment ( based on the segment’s length). The solid dots are the observed cumulative percentage of the dose reaching the colon, as per reference [9].

For a given dose, the simulated mass transfer along different intestinal segment of the mSAT model is shown in Figure A4.4.

180

Figure A4.4. Simulated mass transfer along the intestinal segments of the mSAT model. The lines indicate the percentage of the dose in each segment as a function of time. Dashed blue line, duodenum; dotted green line, jejunum; solid red line, ileum; and dot-dashed cyan line, colon. The solid dots are the observed cumulative percentage of the dose reaching the colon, as per reference [9].

181

Table A4.2. Physiological input parameters for the mSAT model.

Parameter Value Ref. Degrees of flatness (DF) 1.7 [10] Stomach Gastric emptying time 0.25 h [11] -1 (rate constant, kge) (4 h )

Small intestine Length (Lsi) 670.7 cm [12] Mean intestinal transit time (SITT) 3.32 h [9] Duodenum Fractional length (fduo) 0.08 [13] Radius (rduo) 2.37 cm [13] Jejunum Fractional length (fjej) 0.37 [13] Radius (rjej) 1.75 cm [14] Ileum Fractional length (file) 0.55 [13] Radius (rile) 1.50 cm [14]

Colon Total colonic length (Lcol) 104.34 See main text Ascending colon Fractional length (facol) 0.16 [13, 15] Radius(racol) 2.42 cm [16] -1 Transit rate constant (kcol) 0.0667 h [17]

182

A4.5 Regional fabs predictions from the mSAT model for solution and MR formulation.

Figure A4.5. Bar chart of the simulated overall and regional fabs using the mSAT model and the permeability values from Table II (in the manuscript), when colonic absorption was allowed. Each bar represent a different method for the estimation of the absorption (M1, M2, and M3), whereas the shades of grey indicate the proportional contribution to the fabs from each intestinal segment described in the mSAT model.

183

Figure A4.6. Bar chart of the simulated overall and regional fabs using the mSAT model and the permeability values from Table II (in manuscript) for a hypothetical CR formulation, when colonic absorption was allowed. Each bar represent a different method for the estimation of the absorption (M1, M2, and M3), whereas the shades of grey indicate the proportional contribution to the fabs from each intestinal segment described in the mSAT model.

A4.6 Method for the application of the Peff,int approach to current mechanistic absorption models

A simple method to apply the principles derived in this work to any multi- compartmental intestinal absorption model (similar to the CAT model or any of its derivations) is described below [18, 19].

To calculate Peff,int from Peff data the following steps are required:

Peff SA LocIGut  P eff 2  r jej  L LocIGut   P eff Peff ,int    mSA2  rL   SAEF SAEF LocIGut  jej LocIGut   jejunum jejunum 184

where SAEFjejunum, is the combined surface area expansion factor (SAEF) due to all the structures that increase surface area in the upper jejunum (circular folds, intestinal villi and microvilli).

Using Peff,int and the available mSA, the intestinal drug absorption from any segment (n) of the human intestine, can be estimated as follows:

dAPP mSA2  rLSAEF   n eff  n AA   eff  n n n   ... nn2 dt SAEFjejunum V n SAEF jejunum  rL n n 2  P eff SAEFn An rn SAEF jejunum

dA where is the drug absorption rate in the nth intestinal segment, mSA is the dt segment’s mucosal surface area, and Vn is the segment’s cylindrical volume. The

2  P expression eff represents the traditional segment-dependent absorption rate rn constant (ka,n), which is employed in the CAT model and its derivations [18, 19]. On SAEF the other hand, the ratio n can be derived from the data presented in this SAEF jejunum work, where M3 showed the to be the best for the prediction of intestinal absorption

[20]. Therefore, by multiplying Peff by the aforementioned segment-dependent SAEF ratios an estimate of regional intestinal permeability can be obtained. The ratios are summarized in Table A4.3 (derived from the data recently published by Helander and Fandriks [20]).

185

Table A4.3. Mean (± standard deviation (SD)*) regional surface area expansion factors for Method 3a

Segment Circular Intestinal Microvilli Combined SAEF ratiod folds villi SAEFd (Segment/ Jejunum) Duodenum 1.57 6.5 ± 0.87b 9.2 ± 4.11b 93.89 ± 27.92 0.49 ± 0.22 Jejunum 1.57 8.6 ± 1.77b 14.1 ± 6.75b 190.38 ± 63.19 1.00 ± 0.47 Ileum 1.57 4.5 ± 0.49c 15.7 ± 2.20c 110.92 ± 12.55 0.58 ± 0.20 Ascending - - 0.033 ± 6.4 ± 2.69c 6.4 ± 2.69 colon 0.018 *In the original reference the error data was informed as standard error of the mean SD SEM  (SEM), this was converted to SD using the standard formula: n aExpansion factors were extracted from reference [20]. bbased on 5 samples. cbased on 6 samples. dSD calculated using standard propagation of error formulas [21].

Consequently, by applying the segmental SAEF ratios (Table A4.3) to the jejunal

Peff values for the drugs listed in Table II of the manuscript, a segmental Peff can be obtained as shown in Table A4.4. In addition, by performing Monte Carlo simulations using the derived SD for the aforementioned SAEF ratio, an estimate of the uncertainty around the regional Peff estimates can be obtained (assuming a lognormal distribution), product of the uncertainty of the SAEF ratios applied. However, it should be kept in mind that these SAEF values were derived from a limited number of samples and that the variability of the Peff values themselves is not accounted by the SAEF values.

186

Table A4.4. Estimated mean regional Peff values (double balloon technique) for the drugs listed in Table II using the SAEF method.

Compound Mean regional intestinal effective permeability (Peff)a (cm/h) Duodenum Jejunum Ileum Colon Enalaprilat 0.039 0.079 0.046 0.0027 Furosemide 0.054 0.11 0.064 0.0037 Terbutaline 0.089 0.18 0.10 0.0061 Atenolol 0.094 0.19 0.11 0.0064 Metoprolol 0.27 0.54 0.31 0.018 Propranolol 0.48 0.97 0.57 0.033 Fluvastatin 0.50 1.01 0.59 0.034 Antipyrine 1.00 2.02 1.18 0.068 Naproxen 1.42 2.88 1.68 0.10 Ketoprofen 1.51 3.06 1.78 0.10 a Jejunal Peff values extracted from reference [18].

187

A4.7 References

1. J.P. Wilson. Surface area of the small intestine in man. Gut. 8:618-621 (1967). 2. H.P. Schedl. Absorption of steroid hormones from the human small intestine. J Clin Endocrinol Metab. 25:1309-1316 (1965). 3. F.A. Sutcliffe, S.A. Riley, B. Kaserliard, L.A. Turnberg, and M. Rowland. Absorption of Drugs from the Human Jejunum and Ileum. Br J Clin Pharmacol. 26:P206-P207 (1988). 4. T. Gramatte, R. Oertel, B. Terhaag, and W. Kirch. Direct demonstration of small intestinal secretion and site-dependent absorption of the [beta]-blocker talinolol in humans[ast]. Clin Pharmacol Ther. 59:541-549 (1996). 5. T. Gramatte and K. Richter. Paracetamol absorption from different sites in the human small intestine. Br J Clin Pharmacol. 37:608-611 (1994). 6. T. Gramatte, E. el Desoky, and U. Klotz. Site-dependent small intestinal absorption of ranitidine. Eur J Clin Pharmacol. 46:253-259 (1994). 7. T. Gramatte. Griseofulvin absorption from different sites in the human small intestine. Biopharm Drug Dispos. 15:747-759 (1994). 8. J.S. Fordtran, F.C. Rector, Jr., M.F. Ewton, N. Soter, and J. Kinney. Permeability characteristics of the human small intestine. J Clin Invest. 44:1935-1944 (1965). 9. L.X. Yu, J.R. Crison, and G.L. Amidon. Compartmental transit and dispersion model analysis of small intestinal transit flow in humans. Int J Pharm. 140:111-118 (1996). 10. K. Sugano. Estimation of effective intestinal membrane permeability considering bile micelle solubilisation. Int J Pharm. 368:116-122 (2009). 11. L.X. Yu and G.L. Amidon. Saturable small intestinal drug absorption in humans: modeling and interpretation of cefatrizine data. Eur J Pharm Biopharm. 45:199-203 (1998). 12. G. Hounnou, C. Destrieux, J. Desme, P. Bertrand, and S. Velut. Anatomical study of the length of the human intestine. Surg Radiol Anat. 24:290-294 (2002). 13. International Commission on Radiological Protection. Report of the Task Group on Reference Man, Pergamon Press1975. 14. H. Lennernas. Regional intestinal drug permeation: biopharmaceutics and drug development. Eur J Pharm Sci. 57:333-341 (2014). 15. P.J. Watts and L. Illum. Colonic drug delivery. Drug Dev Ind Pharm. 23:893- 913 (1997). 16. S. Sadahiro, T. Ohmura, Y. Yamada, T. Saito, and Y. Taki. Analysis of Length and Surface-Area of Each Segment of the Large-Intestine According to Age, Sex and Physique. Surg Radiol Anat. 14:251-257 (1992). 17. M. Bouchoucha, G. Devroede, E. Dorval, A. Faye, P. Arhan, and M. Arsac. Different segmental transit times in patients with irritable bowel syndrome and "normal" colonic transit time: is there a correlation with symptoms? Tech Coloproctol. 10:287-296 (2006). 18. L.X. Yu and G.L. Amidon. A compartmental absorption and transit model for estimating oral drug absorption. Int J Pharm. 186:119-125 (1999).

188

19. P.J. Sinko, G.D. Leesman, and G.L. Amidon. Predicting Fraction Dose Absorbed in Humans Using a Macroscopic Mass Balance Approach. Pharm Res. 8:979-988 (1991). 20. H.F. Helander and L. Fandriks. Surface area of the digestive tract - revisited. Scand J Gastroenterol. 49:681-689 (2014). 21. H. Ku. Notes on the use of propagation of error formulas. Journal of Research of the National Bureau of Standards. 70:(1966).

189

Appendix A5: Supplementary Material for Chapter 6

190

A5.1 STEP function implementation

The function for the discrete transport of drug mass along the different GI segments of the mSAT model was implemented according to the method proposed by Hénin and co-workers [1]. The STEP function was defined as the sum of individual logistic STEP functions (I_STEP) for each segment of the mSAT model. Each individual function can take values between 0 and 1 and is defined by Equation A5.1

e t Equation A5.1 ISTEP_  IP  eet  n where t is time (h); γ is a coefficient defining the sigmoidicity of the logistic function, where the higher the value the steeper the switch between 0 and 1. These values were fixed to a high number (100) for the simulation purposes; and IPn is the inflection point for the logistic function of the nth GI segment of the mSAT model.

IPn was defined as the cumulative mean residence time (MRT) for the given GI segment in the mSAT model (IPn = MRTn). So for the stomach (ST) IPST will be equal to its MRT (Table A5.1), whereas for the duodenum IPDUO will be equal to the sum of the MRTST and MRTDUO and so on for the rest of the GI segments of the mSAT model.

Table A5.1 Mean residence time for the different segments of the MSAT model

Segment Equation MRT [h] Stomach 1/kge 0.25 Duodenum 1/(0.08×SITT) 0.27 Jejunum 1/(0.37×SITT) 1.23 Ileum 1/(0.55×SITT) 1.83 Ascending colon 1/kt,col 10.2

One of the drawbacks of using the I_STEP function is that given the nature of the logistic function the inflection point flattens as time increases. Following the recommendations made by Hénin and co-workers (2012) a new function was defined (TOTAL_STEP) as follows: 5 Equation A5.2 TOTALSTEPISTEP_ 1 _  n n 1

191

Subsequently TOTAL_STEP was used as the input for the new function in order to re-scale the time and avoid flattening of the inflection points. The new function was defined for the nth segment of the mSAT model as per Equation A5.3

eTOTALSTEP_  Equation A5.3 TABPOS_ n  eeTOTALSTEPNEWIP__ where NEW_IP is the new inflection point for the re-scaled function. This parameter took values of 1.5, 2.5, 3.5, 4.5 and 5.5 for the stomach, duodenum, jejunum, ileum th and ascending colon, respectively. Finally, the STEPn function for the n GI compartment of the mSAT model was defined a shown below: Equation A5.4 STEPTABPOSTABPOSn__ n1 n

where for the stomach compartment TAB_POSST-1 was assumed to be 1. The implementation of the STEPn function for the OROS formulation is illustrated in Figure A5.1 over a period of 24 h.

192

Figure A5.1 Illustration of the implementation of the discrete tablet movement in the mSAT model using the STEPn function [1]. The y axis indicates the GI segment and the solid black line represent the tablet position at a given time.

A5.2 mSAT luminal fluid dynamic model

A simple water secretion and reabsorption model was implemented within the mSAT model to describe the luminal fluid dynamics in the compartments of the mSAT model after the ingestion of a glass of water. The model structure was based on the approach proposed by Jamei and co-workers[2], whereby movements of fluids within the luminal segments are given by the gastric emptying rate (kGE) and the small intestinal transit time (SITT). The steady state luminal fluid volumes are therefore controlled by the zero order secretion rates and first order reabsorption rates in the different luminal compartments, as shown in Equation A5.5

dVlumen, n Equation A5.5 ksec,n () k Tn ,  k reabn ,  V lumenn , dt

th where Vlumen,n is the luminal fluid volume in the n GI segment of the mSAT model

(mL); ksec,n is the zero order secretion rate (mL/min); kT,n is the first order transit rate 193

-1 -1 constant(min ); and kreab,n is the first order reabsorption constant (min ). The main difference between the model proposed by Jamei and co-workers (2009) and the one implemented within the mSAT model was the empirical nature of the model secretion and reabsorption parameters. Whilst in the model proposed by Jamei and co-workers these parameters were derived from the literature, in the present model the parameters were obtained by fitting the model to the water dynamic data recently published by Mudie and co-workers [3]. Therefore the model implemented within the mSAT model can be regarded as an operational model rather than a fully mechanistic luminal fluid dynamic model. The structure of the model is shown in Figure A5.2. The model was fitted to the data provided by Mudie and co-workers (2014) and the parameters that were optimized are highlighted in red in Figure A5.2. The fitting was done in Matlab 2014a using the “lsqnonlin” function included in Matlab’s optimization toolbox. The model assumed that the first-order reabsorption constant are the same for all the small intestinal segments ( kreab,duo= kreab,jej= kreab,ile=kreab,SI), whereas the zero order secretion rate for a given intestinal segment was assumed to be proportional to the length of that particular segment. This assumption eliminated the necessity of estimating an individual secretion rates for each small intestinal segment within the mSAT model, thus only a global secretion rate was estimated (ksec,SI). This rate was subsequently divided for each segment according to the respective fractional length (fLSI,n) of the small intestinal segment in the mSAT model. For the ascending colon, no secretion was assumed and the first- order reabsorption rate was adjusted a posteriori to a value of 0.074 min-1 in order to provide a steady state volume in the ascending colon close to 4.7 mL, the value reported by Sutton [4]. The small intestinal transit time was fixed and the transit rate constant used for this model was assumed to be first order, as opposed to the particle movement in mSAT model that is assumed to follow a time-dependent Weibull function. No differences were observed in the model parameters and the fit when either approach was chosen (first order vs. Weibull). The simplicity of the first order transit provides an advantage when it comes to repeated administration of water into the system. The model fit to the Mudie and-co-workers (2014) data is shown in Figure A5.3 and the parameter estimates are summarized in Table A5.2. The remaining parameters of the model are the same as for the mSAT model (Table 6.1

194

in the manuscript). The simulated segmental volumes using the fluid dynamic model are shown in Figure A5.4 after the administration of a 240 mL glass of water.

Figure A5.2. Schematic representation of the luminal fluid dynamic model implemented within the mSAT model. The parameters in red were obtained by fitting the model to the data reported by Mudie and co-workers[3], whereas the parameters in black were fixed.

Table A5.2. Estimated parameters for the luminal fluid dynamics within the mSAT model

Parameter Estimate RSE (%) Water gastric emptying rate constant (kge,water 0.0754 13.8% )[min-1] Stomach secretion rate (ksec,st)[mL/min] 1.70 20.9% Intestinal reabsorption rate constant 0.681 10.4% -1 (kreab,SI)[min ] Total intestinal secretion rate (ksec,SI) [mL/min] 48.6 14.4% RSE relative standard error of the parameter estimates.

195

Figure A5.3. mSAT fluid dynamic model predictions vs. observed data from [3]. The solid grey circles represent the mean luminal water volume in the stomach at a given time point, whereas the solid grey squared represent the same data for the small intestine. The error bars illustrated the standard error of the mean (SEM).

196

Figure A5.4. Simulated luminal volumes in the different segments of the mSAT model after the administration of 240 mL of water. A plot on the normal scale. Blue solid line - stomach (ST); green dashed line - duodenum (DUO); red dashed-dotted line - jejunum (JEJ); cyan dotted line - ileum (ILE); purple solid line with plus marker - ascending colon(col).B plot on the semi-logarithmic scale

197

A5.3 System-related parameters for the extended mSAT model

The system related parameters for the mSAT model are summarized in Table A5.3. Table A5.3 System-related parameters employed for the in the mSAT model Parameter [units] Value Ref. Reference body weight (BW) [kg] 70 [5] Reference height (HT) [m] 1.70 [5] Reference body surface area (BSA) 1.81 Calc. a [m2] Tissue density (ρtissue) [kg/L] 1.040 [6] Cardiac output (CO) [L/h] 350.37 [7]b Blood weight [kg] (fraction of BW) 5.53 (0.079) [6] Blood volume (Vblood) [L] 5.32 [6] Liver-specific parameters Liver weight [kg] (fraction of BW) 1.82 (0.026) [5, 6, 8] Liver density (ρliver) [kg/L] 1.080 [9] Liver volume(Vliver) [L] 1.69 [5, 6] Liver blood flow (Qliver) [L/h] (fraction 68.32 (0.195) [5, 6] of CO) Hepatic vein blood flow (QHV) [L/h] 89.34 (0.255) [5, 6] (fraction of CO) MPPGL [mg/g] 40 [10] CYP abundances (ACYPj(liver),n) [pmol/mg] ACYP3A4liver 137 [11] ACYP2C9liver 73 [11] ACYP2C19liver 14 [11] ACYP2D6liver 8 [11] GI tract parameters - Gastric emptying rate constant (kGE) [h 4 [2, 12] 1] Mean intestinal transit time (SITT) 3.32 [13] Ascending colon transit rate constant 0.098 [14] -1 (kt,col) [h ] Degrees of flatness coefficient (DF) 1.7 [15] Small intestinal length (LSI) [cm] 670.7 [16] Segment Regional intestinal parameters DUO JEJ ILE COL Radius (rn) [cm] 2.37 1.75 1.5 2.42 [17-21] Length (Ln) [cm] 53.7 248.2 368.9 16.7 [5, 16, 20] Fractional length (fLSI,n) 0.08 0.37 0.55 - [20] Cylindrical volume [mL] 9.47×10 2.39×10 2.61×1 3.07×1 Calc.c 2 3 03 02 2 d Mucosal surface area (mSAn) [cm ] 7.50×10 5.19×10 3.86×1 1.62×1 Calc. 4 5 05 03 Surface area scaling factor 0.49 1.00 0.58 0.033 [22, 23] ratio(SAEFn) Enterocyte height [µm] 32.2 32.2 32.2 35.1 [24-27] Enterocyte compartment volume 0.0262 0.119 0.079 8.9×104 Cal,e

198

(Vent,n) [L] Enterocyte compartment blood flow 1.33 6.24 9.25 4.20 [5, 8, (Qent,n) [L/h] (0.0038) (0.0178) (0.0264 (0.012) 28] (fraction of CO) ) Intestinal CYP abundances (ACYPj(ent),n) [pmol] ACYP3A4ent,n 9,110 36,060 21,030 0 [29-31] ACYP2C9ent,n 1770 7,030 4,100 0 [29-31] ACYP2C19ent,n 210 820 480 0 [29-31] ACYP2D6ent,n 110 440 240 0 [29-31] Segment Regional mSAT luminal parameters ST DUO JEJ ILE COL pH 1.5 6.4 6.6 7.1 6.5 [32-34] Baseline water volume (Vlum,n) [mL] 35 6 13 24 5 [3, 4] Cal., Calculated a calculated using the Du Bois and Du Bois equation for body surface area (BSA) based on body weight (BW, kg) and height (HT, cm) [35]: BSABWHT0.007184  0.425  0.725  . b calculated as function of BW using the formula [7]: COBW11.2 0.81 . c calculated based on the segment’s lengths and radii. d calculated using the cylindrical surface area and the scaling factors proposed by Helander and Fändriks [22, 23] e calculated by multiplying the mucosal surface area ( without accounting for villi expansion) by the enterocyte height [23, 36]

199

A5.4 Derivation of some of the drug-related parameters employed for the simulations

A5.4.1 Calculation of the pH dependent solubility and ionization pH dependent solubility and ionization at the given segment of the mSAT model were calculated using the Henderson-Hasselbalch equation. Given that oxybutynin (OXY) is a monoprotic basic drug Equation A5.6 was employed for the segmental solubility calculations (Sn)

()pK a pH n SSn 0 10  1 Equation A5.6

th where S0 is the drug’s intrinsic solubility and pHn is the luminal pH of the n intestinal segment. Due to the likely precipitation of salts when the calculated theoretical solubility exceeds intrinsic solubility by 3-4 orders of magnitude, the calculated solubility was limited by a solubility factor of 1000 in the case of basic drugs and 10000 in the case of acidic drugs [37].

The fraction of the drug unionized (funionized,n) at a given segment was calculated for OXY using Equation A5.7.

 pH 10 n Equation A5.7 f  unionized, n  pH 10n  10  pKa

A5.4.2 Calculations of fmCYPj from in vitro data

The contribution of a particular CYP isoform to the total intrinsic clearance (fmCYPj) of OXY isomers was calculated from measured in vitro elimination rates reported by Mizushima and co-workers[38]. These rates were measured by incubating the substrate, either R-OXY or S-OXY, in media containing recombinant human liver microsomes expressing a particular CYP 450 isoform (Gentest Corporation). The incubations were carried out at a final microsomal concentration of 5 pmol P450/mL and a final protein concentration of 0.2 mg/mL [38]. The unbound concentration for the incubation was 1 µM (Cu). This concentration was 10 times below the reported

Km values of either R or S oxybutynin [38] and thus the elimination rates (v, pmol/min/pmol of P450) were considered as intrinsic clearances (CLint = v/Cu) [38]. These rates were later corrected by the specific abundance of CYP P450 enzymes in the liver using the values reported by Rowland-Yeo and co-workers in order to 200

obtain the clearance per milligram of microsomal protein in the recombinant system [11] as shown in Equation A5.8

vAj CYPliver, j Equation A5.8 CLint, j  C uj,

th where CLint,j is the intrinsic clearance for a particular j CYP isoform in the recombinant system (µL/min/mg microsomal protein) and ACYPliver,j is the liver abundance of a particular CYP isoform (pmol/mg microsomal protein) (Table A5.3). To be able to extrapolate the intrinsic clearance from the recombinant system to human liver microsomal data the values were corrected by using the inter-system extrapolation factor for the particular jth CYP isoform (ISEFj).The ISEFj values were derived from the literature [39, 40]. Therefore, the total intrinsic clearance (CLint,total) was calculated as per Equation A5.9, whereas the fmCYPj was calculated by dividing the intrinsic clearance, in human liver microsomes, for the particular CYP isoform by the total intrinsic clearance [40]. The aforementioned values are summarized in Table A5.4.

m Equation A5.9 CLCLISEF int,total  int, j j  j 1

Table A5.4. CYP isoform-specific intrinsic clearance in human liver microsomes (HLM) derived from recombinant microsomal data.

a CYP ISEF Intrinsic clearance (HLM) fmCYP isoform (µL/min /mg microsomal protein) R-OXY S-OXY R-OXY S-OXY RACEMICb CYP2C9 0.54 64.4 84.5 0.13 0.11 0.12 CYP2C19 0.25 38.6 40.2 0.08 0.05 0.07 CYP2D6 0.15 1.65 19.3 0.00 0.03 0.01 CYP3A4 0.21 401 595 0.79 0.81 0.80 Total - 505 738 1.00 1.00 1.00 a ISEF values extracted from [40] b average between the fmCYP for each isomer.

A5.4.3 In vitro release profile of the OROS formulation

The in vitro release profile of a 10 mg OROS formulation was measured in the USP apparatus VII. The profile remained almost invariant to changes in the dissolution media [41] and thus the release rate of OXY (mg/h) as a function of time was used as

201

an input function (INPUT) for the OROS formulation for the solid drug in the mSAT model. This profile was digitized from the work of Pitsiu and co-workers [42] and is shown in Figure A5.5. Due to the variable step size of the differential equation solver (ode15s) in Matlab, the profile was linearly interpolated between the time points using the “interp1” function within Matlab, whereas after 24 hours the release rate was assumed to be zero.

Figure A5.5 In vitro release profile of R/S-OXY 10mg OROS formulation. The grey line and solid squares represent the cumulative percentage of the dose released, whereas the dashed blue line represents the R/S-OXY release rate.

A5.4.4 Calculation of the isomer-specific intrinsic clearance ratio

For the simulations involving OXY isomers some assumptions were made with respect to the isomer-specific intrinsic clearance. CLint(u) was obtained from the fitting of the disposition model to the IV infusion data. The value, however, represented the intrinsic clearance for the racemic mixture of OXY. Therefore the in vitro data from Mizushima and co-workers [38] was used to provide a distinction between the clearances for each isomer. The reported in vitro CLint(u) in human liver microsomes for the R-OXY was 225 µL/min/mg, whereas the same value for the S isomer was 278 µL/min/mg. The average CLint(u) between both isomers was 252 µL/min/mg. Subsequently, the intrinsic clearance for each isomer was calculated by 202

multiplying the CLint(u) of the racemic mixture by the ratio between the isomer- specific CLint(u) and the average in vitro CLint(u)(R/SCLint,ratio). This calculation assumes that both isomers contribute equally to the intrinsic clearance of the racemic mixture. This assumption might not necessarily be accurate as there are reports suggesting that the individual isomer contribution to the overall in vitro intrinsic clearance might differ between isomers. For instance, the total intrinsic clearance of the racemic mixture of omeprazole was 19 µL/min/mg, whereas the values for the isomers were 27.2 and 12.2µL/min/mg for the R and S isomer, respectively[43]. The proportional contribution of each isomer to the in vitro intrinsic clearance of the racemic mixture of omeprazole was therefore 45:55 for the R and S isomer, respectively [43]. Nevertheless, this ratio was close to our 50:50 assumption and, to our knowledge, there are no reports available in the literature where the intrinsic clearance of the racemic OXY and its isomer were simultaneously analysed and reported under the same experimental conditions. Thus, our assumption seemed fair for the current simulations.

A5.4.5 Summary of the drug-dependent parameters employed for the simulations and their sources

OXY’s parameters employed for the simulations and their sources are summarized in Table A5.5

203

Table A5.5 OXY’s drug-related parameters employed for the simulations (racemic mixture and isomers)

Parameter [units] OXY racemic R-OXY S-OXY Ref.

Molecular weight (MW) 357.5 - - [44] [g/mol] pKa (base) 8.04 - - [45] LogD7.4 2.98 - - [46] LogP 3.71 - - Calculated a b Intrinsic solubility (S0) [mg/mL] 0.012 - - [45] 3 Particle density (ρp) [mg/mL] 1.20×10 - - Assumed -3 Initial article radius (r0) [cm] 1.00×10 - - Assumed Aqueous diffusion coefficient 0.025 - - [47]c 2 (Daq) [cm /h] Blood to plasma ratio (BP) 0.686 0.71 0.682 [38] -3 -3 -3 Fraction unbound in plasma (fup) 3.40×10 4.70×10 2.75×10 [38, 48] d Apparent permeability (Papp) 21.9 - - [49] [×10-6 cm/s]c Jejunal effective permeability 4.31 - - [49, 50]e (Peff) [×10-4 cm/s] 3 3 3 f Kpu (liver) 1.18×10 1.00×10 1.53×10 [51] Kpu,tissues Estimated - - See methods CLint(u) Estimated - - See methods g R/SCLint,ratio 1 0.89 1.1 [38] h fmCYP3A4 0.80 0.79 0.81 [38] h fmCYP2C9 0.12 0.13 0.11 [38] h fmCYP2C19 0.07 0.08 0.05 [38] h fmCYP2D6 0.01 0.00 0.03 [38] Segment Regional luminal parameters ST DUO JEJ ILE COL Segmental solubility (Sn) 12 0.54 0.34 0.12 0.43 Calculated [mg/mL] Luminal ionization ratio (Iratio,n) n/a 0.64 1.00 2.94 0.80 Calculated - the parameters are the same as for the racemic oxybutynin. a back-calculated from the experimental logD value using the following equation: pKa pH logPD log  log(1  10 ) b intrinsic solubility in water measured at 37°C. c at 37°C.Calculated using OXY’s MW and the equation published by Avdeef [47]: DMW0.3564 0.453 aq d apical to basolateral (A→B) at isotonic pH 7.4 in MDCK-MDR1 cell monolayers e calculated using the correlation published by Gertz and co-workers for Papp and Peff in the same cellular system and similar experimental conditions [50]: log(PP ) 0.829  log( )  0.476 eff ap 204

f calculated using the mechanistic equations published by Rodgers and co- workers[51] g estimated as the ratio between the average in vitro CLint (CLint,av) and the isomer CLint using the data from Mizushima and co-workers [38]. Further details of the calculation are given in Section 4.4. h estimated from data in recombinant human liver microsomes published by Mizushima and co-workers [38].

205

A5.5 Supplementary figures

Figure A5.6 mSAT model prediction of the pharmacokinetic of S-OXY after the administration of three 5 mg IR formulations (A) and one 10 mg OROS formulation (B). The solid black line represents the mSAT model prediction, whereas the solid grey circles and lines represent the individual observed clinical data (kindly provided by Janssen Pharmaceutica). The insert shows the same plot on the semi-logarithmic scale.

206

Figure A5.7. mSAT predicted segmental and oral bioavailability fractions for S- OXY IR and OROS formulations. A predicted fabs: dark blue bars represent the IR formulation and the light blue bars represent the OROS formulation. B predicted fraction of the administered dose metabolized in the intestinal segments (EG): dark green bars represent the IR formulation and the light green bars represent the OROS formulation.

207

A5.6 NONMEM code for IV infusion fit

$PROBLEM MSAT_model IV CAO_and_Jusko 3 Tissues $INPUT ID TIME AMT RATE CONC=DROP MDV CMT EVID LNCONC=DV $DATA …/….csv IGNORE=# $SUBROUTINE ADVAN13 TOL9

$MODEL ; NUM/ DESCRIPTION COMP=(DUO) ;1 DUODENUM COMP=(JEJ) ;2 JEJUNUM COMP=(ILE) ;3 ILEUM COMP=(COL) ;4 COLON COMP=(LIVER) ;5 LIVER COMPARTMENT COMP=(BLOOD,DEFDOSE) ;6 BLOOD COMPARTMENT COMP=(TIS1) ;7 T1 COMP=(TIS2) ;8 T2 COMP=(TIS3) ;9 T3

$PK ; ------PARAMETERS TO BE ESTIMATED------

TVCLINT_3A4 = THETA(1) +ETA(1) ; INTRISIC CLEARANCE FOR CYP3A LGITCO_T1 = THETA(2); LOGIT OF THE FRACTION OF CO THE TISSUE 1 LGITCO_T2 = THETA(3); LOGIT OF THE FRACTION OF CO THE TISSUE 2 LGITWT_T1 = THETA(4) ; LOGIT OF THE FRACTON OF THE BW TISSUE 1 LGITWT_T2 = THETA(5) ; LOGIT OF THE FRACTON OF THE BW TISSUE 2 TVKPU_TIS = THETA(6) ; LN(FKPU_tis) CLINT_3A4= EXP(TVCLINT_3A4) KPU_TIS = EXP(TVKPU_TIS)

;------MODEL PARAMETERS PHYSIOLOGICAL------BW = 70 ; BODY WEIGHT CO = 350.37 ; CARDIAC OUTPUT L/H ;------SMALL INTESTINAL VOLUMES FOR THE ENTEROCYTES--- VENT_DUO =0.026255 ; VENT_JEJ =0.11863 ; VENT_ILE =0.079092 ; VENT_COL =0.00089094 ;

; FRACTION OF WEIGHT THE TISSUES WITH PHYSIOLOGICAL VALUES------FW_LIVER = 0.026; FW_BLOOD = 0.079; FW_ROB = 1- (FW_LIVER + FW_BLOOD); REST OF THE BODY 0.895

FFWT_T1 = EXP(LGITWT_T1)/(EXP(LGITWT_T1) + EXP(LGITWT_T2) + 1); FRACTION OF WT OF THE REST OF THE BODY (INVERSE LOGIT OF THE) FFWT_T2 = EXP(LGITWT_T2)/(EXP(LGITWT_T1) + EXP(LGITWT_T2) + 1); FRACTION OF WT OF THE REST OF THE BODY (INVERSE LOGIT OF THE) FFWT_T3 = 1/(EXP(LGITWT_T1) + EXP(LGITWT_T2) + 1); FRACTION OF WT OF THE REST OF THE BODY (INVERSE LOGIT OF THE)

FWT_T1 = FFWT_T1*FW_ROB FWT_T2 = FFWT_T2*FW_ROB FWT_T3 = FFWT_T3*FW_ROB

208

;------TISSUE WEIGHTS------WT_LIVER = FW_LIVER*BW; WT_BLOOD = FW_BLOOD*BW; WT_TIS1 = FWT_T1*BW; WT_TIS2 = FWT_T2*BW; WT_TIS3 = FWT_T3*BW;

;---FRACTION OF CARDIAC OUTPUT THAT GOES TO THE TISSUES------FCO_DUO = 0.0038; FCO_JEJ = 0.0178; FCO_ILE = 0.0264; FCO_COL = 0.012; FCO_VILLI = 0.06; FCO_LIVER = 0.195; FCO_HV = FCO_LIVER + FCO_VILLI; ; FRACTION OF THE CARDIAC OUTPUT FOR THE REST OF THE BODY FCO_ROB = 1-FCO_HV; 0.745

;---TISSUE BLOOD FLOOWS------Q_DUO = FCO_DUO*CO; Q_JEJ = FCO_JEJ*CO; Q_ILE = FCO_ILE*CO; Q_COL = FCO_COL*CO;

Q_LIVER = FCO_LIVER*CO ; PORTAL VEIN BLOOD FLOW L/H Q_HV = FCO_HV*CO ; HEPATIC VEIN BLOOD FLOW L/H

FFCO_T1 = EXP(LGITCO_T1)/(EXP(LGITCO_T1) + EXP(LGITCO_T2) + 1) ; FRACTION OF CO OF THE REST OF THE BODY (INVERSE LOGIT) FFCO_T2 = EXP(LGITCO_T2)/(EXP(LGITCO_T1) + EXP(LGITCO_T2) + 1) ; FRACTION OF CO OF THE REST OF THE BODY (INVERSE LOGIT) FFCO_T3 = 1/(EXP(LGITCO_T1) + EXP(LGITCO_T2) + 1) ; FRACTION OF CO OF THE REST OF THE BODY (INVERSE LOGIT OF THE)

FCO_T1 = FFCO_T1*FCO_ROB ; FCO_T2 = FFCO_T2*FCO_ROB ; FCO_T3 = FFCO_T3*FCO_ROB ;

Q_TIS1 = FCO_T1*CO; Q_TIS2 = FCO_T2*CO; Q_TIS3 = FCO_T3*CO;

; ------TISSUE VOLUMES------DEN_TIS = 1.040 ; TISSUE DENSITY DEN_LIV = 1.080 ; LIVER DENSITY

V_BLD = WT_BLOOD/DEN_TIS; V_LIV = WT_LIVER/DEN_LIV; V_TIS1 = WT_TIS1/DEN_TIS; V_TIS2 = WT_TIS2/DEN_TIS; V_TIS3 = WT_TIS3/DEN_TIS;

;------LIVER SPECIFIC PARAMETERS FOR CLEARANCE------L_RO = 1080 ; LIVER DENSITY g/L MPPGL = 40 ; MEAN MILIGRAMS OF PROTEIN PER GRAM OF LIVER ACYP3A_LIV = 137 ; [pmol/mg microsomal protein] CL_UNIT = 60/(1000000); unit conversion from [ul/min/mg] to [l/h/mg] 209

KGTOG = 1000; UNIT CONVERSION KG TO CL_LIVER = (CLINT_3A4*CL_UNIT)*MPPGL*WT_LIVER*KGTOG

;------INTESTINE SPECIFIC PARAMETERS FOR CLEARANCE---- ; ABUNDANCE OF THE CYP ENZYMES ACYP3A_DUO = 9110 ; [pmol], abundance of CYP3A4 in the duodenum ACYP3A_JEJ = 36060 ; [pmol], abundance of CYP3A4 in the jujunum ACYP3A_ILE = 21030 ; [pmol], abundance of CYP3A4 in the ileum ACYP3A_COL = 0 ; [pmol], abundance of CYP3A4 in the colon

CLI_DUO = ((CLINT_3A4*CL_UNIT)/ACYP3A_LIV)*ACYP3A_DUO; [l/h] CLI_JEJ = ((CLINT_3A4*CL_UNIT)/ACYP3A_LIV)*ACYP3A_JEJ; [l/h] CLI_ILE = ((CLINT_3A4*CL_UNIT)/ACYP3A_LIV)*ACYP3A_ILE; [l/h] CLI_COL = ((CLINT_3A4*CL_UNIT)/ACYP3A_LIV)*ACYP3A_COL; [l/h]

;------DRUG RELATED PARAMETERS ------

BP = 0.686 ; BLOOD TO PLASMA RATIO FUP = 0.0034 ; FRACTION UNBOUND IN PLASMA FUB = FUP/BP ; FRACTION UNBOUND IN BLOOD FUGUT =1 ; FRACTION UNBOUND IN THE GUT KB_LIV = 6.0003 ; OXYBUTYNI'S LIVER KB (R&R) fup = 0.0034 KB_ALL = KPU_TIS*FUB; KB_TIS1 = KB_ALL; KB_TIS2 = KB_ALL KB_TIS3 = KB_ALL;

;------SCALE CONCENTRATIONS ------S6=(V_BLD*BP)/1000; SCALE PLASMA TO BLOOD AND THEN DIVIDE BY 1000 $DES ;------INTESTINAL WALL DADT(1)=-(Q_DUO/VENT_DUO)*A(1) - (CLI_DUO/VENT_DUO)*A(1) DADT(2)=-(Q_JEJ/VENT_JEJ)*A(2) - (CLI_JEJ/VENT_JEJ)*A(2) DADT(3)=-(Q_ILE/VENT_ILE)*A(3) - (CLI_ILE/VENT_ILE)*A(3) DADT(4)=-(Q_COL/VENT_COL)*A(4) - (CLI_COL/VENT_COL)*A(4) ;------DISPOSITION PART------DADT(5)= (Q_DUO/VENT_DUO)*A(1) + (Q_JEJ/VENT_JEJ)*A(2) + (Q_ILE/VENT_ILE)*A(3) & + (Q_COL/VENT_COL)*A(4) + (Q_LIVER/V_BLD)*A(6) - (Q_HV/(V_LIV*KB_LIV))*A(5) & - ((FUB*CL_LIVER)/(KB_LIV*V_LIV))*A(5); LIVER DADT(6)= (Q_HV/(V_LIV*KB_LIV))*A(5) + (Q_TIS1/(KB_TIS1*V_TIS1))*A(7) + (Q_TIS2/(KB_TIS2*V_TIS2))*A(8)& +(Q_TIS3/(KB_TIS3*V_TIS3))*A(9) - (Q_LIVER/V_BLD)*A(6) -& (Q_TIS1/V_BLD)*A(6) - (Q_TIS2/V_BLD)*A(6)-(Q_TIS3/V_BLD)*A(6); BLOOD DADT(7)= (Q_TIS1/V_BLD)*A(6) - (Q_TIS1/(KB_TIS1*V_TIS1))*A(7); TISSUE 1 DADT(8)= (Q_TIS2/V_BLD)*A(6) - (Q_TIS2/(KB_TIS2*V_TIS2))*A(8); TISSUE 2 DADT(9)= (Q_TIS3/V_BLD)*A(6) - (Q_TIS3/(KB_TIS3*V_TIS3))*A(9); TISSUE 3

$ERROR

210

;-----Log Domain------IPRED = F DUMVAL = 0 ; Prevent Zero values IF(F.EQ.0) DUMVAL =0.0001 ; Prevent zero values IF (CMT.EQ.6) THEN IPRED = log(F+DUMVAL) Y = IPRED + EPS(1) IRES = DV - IPRED W = 1 IWRES = IRES/W ENDIF

$THETA (7.6) ; ln(CLINT_3A4) (-0.7) ; LGITCO_T1 (2.6) ; LGITCO_T2 (1.8) ; LGITBW_T1 (-0.3) ; LGITBW_T2 (6.15) ; ln(KPU_TIS)

$OMEGA 0 FIX ; PPVCL

$SIGMA 0.1

$ESTIMATION PRINT=1 METHOD=1 INTERACTION NOABORT MAXEVAL=9999 SIG=3 SIGL=9 $COVARIANCE UNCONDITIONAL PRINT=E SIGL=9 $TABLE

211

A5.7 References

1. E. Hénin, M. Bergstrand, J. Standing, and M. Karlsson. A mechanism-Based Approach for Absorption Modeling: The Gastro-Intestinal Transit Time (GITT) Model. The AAPS Journal. 14:155-163 (2012). 2. M. Jamei, D. Turner, J. Yang, S. Neuhoff, S. Polak, A. Rostami-Hodjegan, and G. Tucker. Population-based mechanistic prediction of oral drug absorption. AAPS J. 11:225-237 (2009). 3. D.M. Mudie, K. Murray, C.L. Hoad, S.E. Pritchard, M.C. Garnett, G.L. Amidon, P.A. Gowland, R.C. Spiller, G.E. Amidon, and L. Marciani. Quantification of gastrointestinal liquid volumes and distribution following a 240 mL dose of water in the fasted state. Mol Pharm. 11:3039-3047 (2014). 4. S.C. Sutton. Role of physiological intestinal water in oral absorption. AAPS J. 11:277-285 (2009). 5. J. Valentin. Basic anatomical and physiological data for use in radiological protection: reference values. Ann ICRP. 32:1-277 (2002). 6. R.P. Brown, M.D. Delp, S.L. Lindstedt, L.R. Rhomberg, and R.P. Beliles. Physiological Parameter Values for Physiologically Based Pharmacokinetic Models. Toxicol Ind Health. 13:407-484 (1997). 7. I. Nestorov. Modelling and simulation of variability and uncertainty in toxicokinetics and pharmacokinetics. Toxicol Lett. 120:411-420 (2001). 8. L.R. Williams and R.W. Leggett. Reference values for resting blood flow to organs of man. Clin Phys Physiol Meas. 10:187-217 (1989). 9. A. Heinemann, F. Wischhusen, K. Püschel, and X. Rogiers. Standard liver volume in the caucasian population. Liver Transpl Surg. 5:366-368 (1999). 10. Z.E. Barter, M.K. Bayliss, P.H. Beaune, A.R. Boobis, D.J. Carlile, R.J. Edwards, J.B. Houston, B.G. Lake, J.C. Lipscomb, O.R. Pelkonen, G.T. Tucker, and A. Rostami-Hodjegan. Scaling factors for the extrapolation of in vivo metabolic drug clearance from in vitro data: reaching a consensus on values of human microsomal protein and hepatocellularity per gram of liver. Curr Drug Metab. 8:33-45 (2007). 11. K. Rowland Yeo, A. Rostami-Hodjegan, and G.T. Tucker. Abundance of Cytochromes P450 in Human Liver: A Meta-Analysis. Br J Clin Pharmacol. 57:687-688 (2004). 12. L.X. Yu and G.L. Amidon. Saturable small intestinal drug absorption in humans: modeling and interpretation of cefatrizine data. Eur J Pharm Biopharm. 45:199-203 (1998). 13. L.X. Yu, J.R. Crison, and G.L. Amidon. Compartmental transit and dispersion model analysis of small intestinal transit flow in humans. Int J Pharm. 140:111-118 (1996). 14. M. Bouchoucha, G. Devroede, E. Dorval, A. Faye, P. Arhan, and M. Arsac. Different segmental transit times in patients with irritable bowel syndrome and "normal" colonic transit time: is there a correlation with symptoms? Tech Coloproctol. 10:287-296 (2006). 15. K. Sugano. Estimation of effective intestinal membrane permeability considering bile micelle solubilisation. Int J Pharm. 368:116-122 (2009).

212

16. G. Hounnou, C. Destrieux, J. Desme, P. Bertrand, and S. Velut. Anatomical study of the length of the human intestine. Surg Radiol Anat. 24:290-294 (2002). 17. T. Knutson, P. Fridblom, H. Ahlstrom, A. Magnusson, C. Tannergren, and H. Lennernas. Increased understanding of intestinal drug permeability determined by the LOC-I-GUT approach using multislice computed tomography. Mol Pharm. 6:2-10 (2009). 18. H. Lennernas. Regional intestinal drug permeation: biopharmaceutics and drug development. Eur J Pharm Sci. 57:333-341 (2014). 19. P.J. Watts and L. Illum. Colonic drug delivery. Drug Dev Ind Pharm. 23:893- 913 (1997). 20. International Commission on Radiological Protection. Report of the Task Group on Reference Man, Pergamon Press1975. 21. S. Sadahiro, T. Ohmura, Y. Yamada, T. Saito, and Y. Taki. Analysis of Length and Surface-Area of Each Segment of the Large-Intestine According to Age, Sex and Physique. Surg Radiol Anat. 14:251-257 (1992). 22. A. Olivares-Morales, H. Lennernas, L. Aarons, and A. Rostami-Hodjegan. Translating Human Effective Jejunal Intestinal Permeability to Surface- Dependent Intrinsic Permeability: a Pragmatic Method for a More Mechanistic Prediction of Regional Oral Drug Absorption. AAPS J. 17:1177- 1192 (2015). 23. H.F. Helander and L. Fandriks. Surface area of the digestive tract - revisited. Scand J Gastroenterol. 49:681-689 (2014). 24. T. Eide. A morphometrical analysis of dysplasia in small adenomas of the large instestine. Virchows Archiv A. 410:119-124 (1987). 25. M.G. Tabbaa, A.T.R. Axon, and M.F. Dixon. Enterocyte dimensions in patients with abnormal intestinal permeability. Eur Journal of Gastroenterology and Hepatology. 6:607-610 (1994). 26. P.S. Lipski, M.K. Bennett, P.J. Kelly, and O.F. James. Ageing and duodenal morphometry. J Clin Pathol. 45:450-452 (1992). 27. P. Crowe and M. Marsh. Morphometric analysis of small intestinal mucosa IV. Determining cell volumes. Virchows Archiv A. 422:459-466 (1993). 28. P.J. Matheson, M.A. Wilson, and R.N. Garrison. Regulation of Intestinal Blood Flow. J Surg Res. 93:182-196 (2000). 29. E. Sjogren, B. Abrahamsson, P. Augustijns, D. Becker, M.B. Bolger, M. Brewster, J. Brouwers, T. Flanagan, M. Harwood, C. Heinen, R. Holm, H.P. Juretschke, M. Kubbinga, A. Lindahl, V. Lukacova, U. Munster, S. Neuhoff, M.A. Nguyen, A. Peer, C. Reppas, A.R. Hodjegan, C. Tannergren, W. Weitschies, C. Wilson, P. Zane, H. Lennernas, and P. Langguth. In vivo methods for drug absorption - Comparative physiologies, model selection, correlations with in vitro methods (IVIVC), and applications for formulation/API/excipient characterization including food effects. Eur J Pharm Sci. 57:99-151 (2014). 30. M.F. Paine, H.L. Hart, S.S. Ludington, R.L. Haining, A.E. Rettie, and D.C. Zeldin. The human intestinal cytochrome P450 "pie". Drug Metab Dispos. 34:880-886 (2006). 31. M.F. Paine, M. Khalighi, J.M. Fisher, D.D. Shen, K.L. Kunze, C.L. Marsh, J.D. Perkins, and K.E. Thummel. Characterization of Interintestinal and Intraintestinal Variations in Human CYP3A-Dependent Metabolism. J Pharmacol Exp Ther. 283:1552-1562 (1997). 213

32. J. Fallingborg, L.A. Christensen, M. Ingeman-Nielsen, B.A. Jacobsen, K. Abildgaard, and H.H. Rasmussen. pH-profile and regional transit times of the normal gut measured by a radiotelemetry device. Aliment Pharmacol Ther. 3:605-613 (1989). 33. D.F. Evans, G. Pye, R. Bramley, A.G. Clark, T.J. Dyson, and J.D. Hardcastle. Measurement of gastrointestinal pH profiles in normal ambulant human subjects. Gut. 29:1035-1041 (1988). 34. M. Koziolek, M. Grimm, D. Becker, V. Iordanov, H. Zou, J. Shimizu, C. Wanke, G. Garbacz, and W. Weitschies. Investigation of pH and Temperature Profiles in the GI Tract of Fasted Human Subjects Using the Intellicap® System. J Pharm Sci. 10.1002/jps.24274:n/a-n/a (2014). 35. D. Du Bois and E. Du Bois. Clinical calorimetry: Tenth paper a formula to estimate the approximate surface area if height and weight be known. Arch Intern Med. XVII:863-871 (1916). 36. K. Thelen, K. Coboeken, S. Willmann, R. Burghaus, J.B. Dressman, and J. Lippert. Evolution of a detailed physiological model to simulate the gastrointestinal transit and absorption process in humans, part 1: oral solutions. J Pharm Sci. 100:5324-5345 (2011). 37. A. Avdeef. Solubility of sparingly-soluble ionizable drugs. Adv Drug Deliv Rev. 59:568-590 (2007). 38. H. Mizushima, K. Takanaka, K. Abe, I. Fukazawa, and H. Ishizuka. Stereoselective pharmacokinetics of oxybutynin and N-desethyloxybutynin in vitro and in vivo. Xenobiotica. 37:59-73 (2007). 39. N.J. Proctor, G.T. Tucker, and A. Rostami-Hodjegan. Predicting drug clearance from recombinantly expressed CYPs: intersystem extrapolation factors. Xenobiotica. 34:151-178 (2004). 40. Y. Chen, L. Liu, K. Nguyen, and A.J. Fretland. Utility of Intersystem Extrapolation Factors in Early Reaction Phenotyping and the Quantitative Extrapolation of Human Liver Microsomal Intrinsic Clearance Using Recombinant Cytochromes P450. Drug Metab Dispos. 39:373-382 (2011). 41. G. Sathyan, R.R. Dmochowski, R.A. Appell, C. Guo, and S.K. Gupta. Effect of antacid on the pharmacokinetics of extended-release formulations of tolterodine and oxybutynin. Clin Pharmacokinet. 43:1059-1068 (2004). 42. M. Pitsiu, G. Sathyan, S. Gupta, and D. Verotta. A semiparametric deconvolution model to establish in vivo–in vitro correlation applied to OROS oxybutynin. J Pharm Sci. 90:702-712 (2001). 43. X.-Q. Li, L. Weidolf, R. Simonsson, and T.B. Andersson. Enantiomer/Enantiomer Interactions between the S- and R- Isomers of Omeprazole in Human Cytochrome P450 Enzymes: Major Role of CYP2C19 and CYP3A4. J Pharmacol Exp Ther. 315:777-787 (2005). 44. Chemical Abstracts Service. SciFinder®, version 2013, American Chemical Society. https://scifinder.cas.org/scifinder/view/scifinder/scifinderExplore.jsf (accessed June 6, 2015). Calculated using Advanced Chemistry Development (ACD/Labs) Software V11.02 (© 1994-2013 ACD/Labs). 45. E. Miyamoto, S. Kawashima, Y. Murata, Y. Yamada, Y. Demizu, H. Kontani, and T. Sakai. Physico-chemical properties of oxybutynin. Analyst. 119:1489-1492 (1994). 46. R.S. Plumb, W.B. Potts, P.D. Rainville, P.G. Alden, D.H. Shave, G. Baynham, and J.R. Mazzeo. Addressing the analytical throughput challenges in ADME screening using rapid ultra-performance liquid 214

chromatography/tandem mass spectrometry methodologies. Rapid Commun Mass Spectrom. 22:2139-2152 (2008). 47. A. Avdeef. Leakiness and size exclusion of paracellular channels in cultured epithelial cell monolayers-interlaboratory comparison. Pharm Res. 27:480- 489 (2010). 48. A. Shibukawa, N. Ishizawa, T. Kimura, Y. Sakamoto, K. Ogita, Y. Matsuo, Y. Kuroda, C. Matayatsu, T. Nakagawa, and I.W. Wainer. Plasma protein binding study of oxybutynin by high-performance frontal analysis. J Chromatogr B Analyt Technol Biomed Life Sci. 768:177-188 (2002). 49. E. Callegari, B. Malhotra, P.J. Bungay, R. Webster, K.S. Fenner, S. Kempshall, J.L. LaPerle, M.C. Michel, and G.G. Kay. A comprehensive non- clinical evaluation of the CNS penetration potential of antimuscarinic agents for the treatment of overactive bladder. Br J Clin Pharmacol. 72:235-246 (2011). 50. M. Gertz, A. Harrison, J.B. Houston, and A. Galetin. Prediction of human intestinal first-pass metabolism of 25 CYP3A substrates from in vitro clearance and permeability data. Drug Metab Dispos. 38:1147-1158 (2010). 51. T. Rodgers, D. Leahy, and M. Rowland. Physiologically based pharmacokinetic modeling 1: Predicting the tissue distribution of moderate- to-strong bases. J Pharm Sci. 94:1259-1276 (2005).

215