Pbpk models for the prediction of in vivo performance of oral dosage forms

1
2 Review
4 PBPK models for the prediction of in vivo performance of oral dosage
5 forms
6
7
8 Edmund S. Kostewicz a,⇑
Q1 , Leon Aarons b
, Martin Bergstrand c
, Michael B. Bolger d
, Aleksandra Galetin b
,
9 Oliver Hatley b
, Masoud Jamei e
, Richard Lloyd f
, Xavier Pepin g
, Amin Rostami b,e
, Erik Sjögren h
,
10 Christer Tannergren i
, David B. Turner e
, Christian Wagner a
, Werner Weitschies j
, Jennifer Dressman a
11 a
Institute of Pharmaceutical Technology, Goethe University, Frankfurt/Main, Germany
12 b
Centre for Applied Pharmaceutical Research, School of Pharmacy and Pharmaceutical Sciences, The University of Manchester, United Kingdom
13 c
Pharmacometrics Research Group, Department of Pharmaceutical Biosciences, Uppsala University, Sweden
14 d
Simulations Plus, Inc., Lancaster, CA, United States
15 e
Simcyp Limited (a Certara Company), Blades Enterprise Centre, Sheffield, United Kingdom
16 f
Department of Drug Metabolism and Pharmacokinetics, GlaxoSmithKline, Ware, Hertfordshire, United Kingdom
17 g
Department of Biopharmaceutics, Pharmaceutical Sciences R&D, Sanofi Aventis, Vitry sur Seine Cedex, France
18 h
Department of Pharmacy, Uppsala University, Uppsala, Sweden
19 i
Medicines Evaluation CVGI, Pharmaceutical Development, AstraZeneca R&D Mölndal, Sweden
20 j
Department of Biopharmaceutics, University of Greifswald, Greifswald, Germany
21
22
2 4
a r t i c l e i n f o
25 Article history:
26 Received 28 May 2013
27 Received in revised form 27 August 2013
28 Accepted 11 September 2013
29 Available online xxxx
30 Keywords:
31 Physiologically-based pharmacokinetic
32 (PBPK) models
33 Predicting drug absorption
34 Innovative Medicines Initiative (IMI)
35 Oral Biopharmaceutical Tools (OrBiTo)
36 In vitro biopharmaceutical tools
37 Gastrointestinal physiology
38
3 9
a b s t r a c t
40Drug absorption from the gastrointestinal (GI) tract is a highly complex process dependent upon numer-
41ous factors including the physicochemical properties of the drug, characteristics of the formulation and
42interplay with the underlying physiological properties of the GI tract. The ability to accurately predict
43oral drug absorption during drug product development is becoming more relevant given the current chal-
44lenges facing the pharmaceutical industry.
45Physiologically-based pharmacokinetic (PBPK) modeling provides an approach that enables the plasma
46concentration–time profiles to be predicted from preclinical in vitro and in vivo data and can thus provide
47a valuable resource to support decisions at various stages of the drug development process. Whilst there
48have been quite a few successes with PBPK models identifying key issues in the development of new
49drugs in vivo, there are still many aspects that need to be addressed in order to maximize the utility of
50the PBPK models to predict drug absorption, including improving our understanding of conditions in
51the lower small intestine and colon, taking the influence of disease on GI physiology into account and fur-
52ther exploring the reasons behind population variability. Importantly, there is also a need to create more
53appropriate in vitro models for testing dosage form performance and to streamline data input from these
54into the PBPK models.
55As part of the Oral Biopharmaceutical Tools (OrBiTo) project, this review provides a summary of the
56current status of PBPK models available. The current challenges in PBPK set-ups for oral drug absorption
57including the composition of GI luminal contents, transit and hydrodynamics, permeability and intestinal
58wall metabolism are discussed in detail. Further, the challenges regarding the appropriate integration of
59results from in vitro models, such as consideration of appropriate integration/estimation of solubility and
60the complexity of the in vitro release and precipitation data, are also highlighted as important steps to
61advancing the application of PBPK models in drug development.
0928-0987/$ - see front matter Ó 2013 Published by Elsevier B.V.
http://dx.doi.org/10.1016/j.ejps.2013.09.008
Abbreviations: ABL, Aqueous Boundary Layer; BCS, Biopharmaceutics Classification System; BDDCS, Biopharmaceutical Drug Disposition Classification System; BE,
Bioequivalence; CYP, Cytochrome P450; Peff, effective permeability; EP, European Pharmacopoeia; ER, Extended Release; FE, Fold Error; GI, Gastrointestinal; IR, Immediate
Release; IVIVC, in vitro–in vivo Correlation; MMC, Myoelectric Motor Complex; MR, Modified Release; OrBiTo, Oral Biopharmaceutical Tools; BA, Oral Bioavailability; PBPK,
Physiologically Based Pharmacokinetic Modeling; QbD, Quality by Design; QC, Quality Control; UGT, UDP Glucuronosyltransferase; FDA, US Food and Drug Administration;
USP, United States Pharmacopeia.
⇑ Corresponding author. Address: Institute of Pharmaceutical Technology, Goethe University,Q2 Max-von-Laue Str. 9, 60438 Frankfurt/Main, Germany. Tel.: +49 69 798 296
92; fax: +49 69 798 296 94.
E-mail address: kostewicz@em.uni-frankfurt.de (E.S. Kostewicz).
European Journal of Pharmaceutical Sciences xxx (2013) xxx–xxx
Contents lists available at ScienceDirect
European Journal of Pharmaceutical Sciences
journal homepage: www.elsevier.com/locate/ejps
PHASCI 2875 No. of Pages 22, Model 5G
24 September 2013
Please cite this article in press as: Kostewicz, E.S., et al. PBPK models for the prediction of in vivo performance of oral dosage forms. Eur. J. Pharm. Sci.
(2013), http://dx.doi.org/10.1016/j.ejps.2013.09.008

It is expected that the ‘‘innovative’’ integration of in vitro data from more appropriate in vitro models
and the enhancement of the GI physiology component of PBPK models, arising from the OrBiTo project,
62 will lead to a significant enhancement in the ability of PBPK models to successfully predict oral drug
63 absorption and advance their role in preclinical and clinical development, as well as for regulatory
64 applications.
65 Ó 2013 Published by Elsevier B.V.
68
69
70 Contents
71 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
72 2. Current PBPK models and how they handle GI absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
73 2.1. The Simcyp Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
74 2.2. GastroPlus™. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
75 2.3. PK-SimÒ
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
76 2.4. Other in silico tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
77 2.4.1. MatLabÒ
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
78 2.4.2. STELLAÒ
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
79 2.4.3. GI-Sim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
80 3. Current challenges in PBPK setups for GI absorption. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
81 3.1. Composition of luminal contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
82 3.2. Transit and hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
83 3.3. Permeability, including transporters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
84 3.4. Intestinal wall metabolism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
85 3.5. Integration of results from in vitro models with PBPK modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
86 3.5.1. How to best handle gut solubility in Computational Oral Absorption Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
87 3.5.2. Accounting for the solubility/permeability interplay: the PBPK approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
88 3.5.3. Challenges for integration of results from in vitro release and precipitation models with PBPK modeling . . . . . . . . . . . . . . . . . . 00
89 4. PBPK models for predicting oral drug absorption: the future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
90 4.1. How to best combine PBPK with in vitro test input: striking the right balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
91 4.2. Predicting variability of response in the target patient population . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
92 4.3. Evaluating PBPK success: which yardsticks should we use? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
93 4.4. Applications of PBPK modeling in the context of the OrBiTo project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
94 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
95 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
96
97
98 1. Introduction
99 Physiologically-based pharmacokinetic (PBPK) models tradi-
100 tionally employ what is commonly known as a ‘‘bottom-up’’ ap-
101 proach. The concept is to describe the concentration profile of a
102 drug in various tissues as well as in the blood over time, based
103 on the drug characteristics, site and means of administration and
104 the physiological processes to which the drug is subjected. There-
105 by, PBPK modeling takes into account the factors influencing the
106 absorption, distribution and elimination processes (Rowland
107 et al., 2011). In PBPK modeling, parameters are determined a priori
108 from in vitro experiments and the physiology, utilizing in silico pre-
109 dictions to predict in vivo data. In one of the earliest invocations of
110 the PBPK approach, a Swedish physiologist and biophysicist, Teo-
111 rell, developed a five compartment scheme to reflect the circula-
112 tory system, a drug depot, fluid volume, kidney elimination and
113 tissue inactivation (Teorell, 1937a,b). The next advances came over
114 20 years later, when Edelman and Liebmann recognized that the
115 total body water was not equally accessible, but rather should be
116 divided into plasma, interstitial-lymph, dense connective tissue
117 and cartilage, inaccessible bone water, transcellular and intracellu-
118 lar components (Edelmann and Liebmann, 1959). A few years later,
119 physiological models were introduced to describe the handling of
120 drugs by the artificial kidney as well as to describe the pharmaco-
121 kinetics of thiopental and methotrexate (Bischoff and Dedrick,
122 1968; Bischoff et al., 1971; Dedrick and Bischoff, 1968). In the early
123 years however, in silico predictions were hampered by lack of
124 capacity in computing as well as large gaps in physiological
125knowledge. Further, the design of in vitro experiments, particularly
126in the area of predicting drug release, transport and metabolism,
127was still at a rather rudimentary stage. As knowledge in these areas
128grew, and more powerful computers became commonplace, it was
129possible to create better PBPK models and they became more
130widely used, such that for example, in 1979, a review of PBPK mod-
131els for anticancer drugs was published (Chen and Gross, 1979). As
132early as 1981, the brilliant pharmacokineticist, John Wagner, fore-
133saw applications of pharmacokinetics to patient care such as indi-
134vidualization of patient dose and dosage regimen, determination of
135the mechanism of drug–drug interactions, prediction of pharmaco-
136kinetics of drugs in man from results obtained in animals using
137physiologically based models, development of sophisticated com-
138puter programs to obtain population estimates of pharmacokinetic
139parameters and their variabilities, therapeutic monitoring and pre-
140diction of the time course of the intensity of pharmacological ef-
141fects. In the meantime, many of these ideas have been turned
142into reality, or are being turned into reality, through the applica-
143tion of PBPK models (Wagner, 1981).
144The first commercial software to attempt a comprehensive
145description of the gastrointestinal (GI) tract in the context of a
146PBPK model was GastroPlus™. The first version, introduced in
1471998, used a mixing-tanks-in-series approach to describe the
148movement of drug from one region in the GI tract to the next, with
149simple estimations of dissolution based on aqueous solubility and
150absorption rate constants based on existing pharmacokinetic data.
151Even at this stage, it was possible to obtain a reading on whether
152absorption (uptake across the GI mucosa) or solubility/dissolution
2 E.S. Kostewicz et al. / European Journal of Pharmaceutical Sciences xxx (2013) xxx–xxx
24 September 2013

153 would be limiting to the drug’s bioavailability. This was already a
154 very significant advance for scientists working in formulation
155 development, as it enabled goals for the formulation to be set on
156 a more realistic basis, recognizing that solubility and dissolution
157 problems are much more amenable to formulation solutions than
158 permeability limitations.
159 In addition to GastroPlus™, several other commercial PBPK
160 models such as Simcyp and PK-SimÒ
now have evolved descrip-
161 tions of the GI tract. Additionally, there are some software pro-
162 grams available which can be tailored to predict in vivo
163 performance of oral formulations, including MatlabÒ
and STELLAÒ
164 – these tend to be favored by academic groups. In the industry, as
165 well as utilizing the commercially available PBPK models, some
166 companies have ‘‘home-grown’’ models tailored to the specific
167 needs of their development programmes. All of these programmes
168 now strive to account for all relevant processes to the GI absorp-
169 tion of drugs, including release from the dosage form, decomposi-
170 tion/complexation in the GI tract, the various mechanisms of drug
171 uptake and efflux and first pass metabolism, whether this be in the
172 gut wall or liver, and to describe the interplay of these factors in
173 determining the rate and extent of drug absorption from the GI
174 tract.
175 As part of the Oral Biopharmaceutical Tools (OrBiTo) project
176 within the Innovative Medicines Initiative (IMI) framework, this
177 review provides a summary of the current status of PBPK models
178 available for predicting the in vitro performance of orally adminis-
179 tered drugs and their formulations. The current challenges in PBPK
180 set-ups for oral drug absorption including the composition of GI
181 luminal contents, transit and hydrodynamics, permeability and
182 intestinal wall metabolism are then discussed in detail. Further,
183 the challenges regarding the appropriate integration of results
184 from in vitro models, such as consideration of appropriate integra-
185 tion/estimation of solubility and the complexity of the in vitro re-
186 lease and precipitation data, are also highlighted as important
187 steps to advancing the application of PBPK models in drug
188 development.
189 2. Current PBPK models and how they handle GI absorption
190 Many drugQ3 companies are now building PBPK models for all
191 new candidate drugs early in the discovery and development cy-
192 cles. These models can be parameterized using in silico (distribu-
193 tion) and in vitro (intrinsic clearance) methods and can provide a
194 ‘‘ballpark’’ estimate of the human and/or animal plasma concentra-
195 tion vs. time profile prior to in vivo testing in animals. Furthermore,
196 they provide one of the most successful methods for scale-up from
197 animals to human. PBPK models also allow for an early estimate of
198 local organ tissue concentrations which can be tied to pharmaco-
199 dynamic models to get an estimate of the response in any particu-
200 lar tissue.
201 Whilst the use of PBPK models by the pharmaceutical industry
202 as both a prediction and mechanistic analysis tool to gain a holistic
203 analysis of drug absorption is fairly widespread, as described in the
204 following section the structure of different commercially available
205 PBPK models can vary considerably.
206 2.1. The Simcyp Simulator
207 The Advanced Dissolution Absorption and Metabolism (ADAM)
208 model (Jamei et al., 2009b) is a multi-compartmental GI transit
209 model fully integrated into the Simcyp human population-based
210 Simulator (Jamei et al., 2009a,b) as well as the rat, mouse and
211 dog simulators. The Simulator provides both pharmacokinetic
212 and pharmacodynamics models and a separate pediatric module.
213The ADAM model treats the GI tract as one stomach, seven
214small intestine and one colon compartment(s) (Fig. 1), within each
215of which, drug can exist in several states simultaneously viz.
216unreleased, undissolved (solid particles), dissolved or degraded.
217Dissolved drug is permitted to attain supersaturated levels con-
218trolled by the maximal allowable extent of supersaturation, its
219duration and first order precipitation rate constants. Drug can be
220dosed in a supersaturated state or supersaturation may be attained
221as a consequence of solubility change when moving from one re-
222gion of the GI tract to another. The most well-known example of
223this phenomenon is when the pH-dependent solubility of a weak
224base with pKa < 7 is reduced significantly upon transit from the
225fasted stomach to the duodenum. ADAM includes the population
226mean and inter-individual variability of regional luminal pH and
227bile salt concentrations in the fasted and fed states. It models the
228interplay of these factors upon solubility and dissolution rate
229(Persson et al., 2005a) via a bile micelle solubilization (Glomme
230et al., 2007) and a diffusion layer (Jamei et al., 2009b; Patel et al.,
231submitted for publication) model.
232Separate age-dependent functions define the (a) return of
233elevated fed stomach pH to basal (fasted) levels, and (b)
234population-specific fractions of achlorhydric individuals. A general
235North European Caucasian population has a fixed proportion (8%)
236of achlorhydric subjects while the Simcyp Japanese population
237has proportions that increase with age (for more detail see Jamei
238et al., 2009b). Differences such as these are captured in the various
239ethnic and disease populations available within the Simulator. The
240liver cirrhosis libraries, for example, define three levels of severity
241of disease and capture extended gastric emptying times decreases
242in gut Cytochrome P450 (CYP) 3A4, decreases in portal blood flow,
243increases in villous blood flow and lower levels of plasma binding
244proteins (Johnson et al., 2010). Also available are obesity, renal
245impairment disease libraries, a pediatric library (with built-in
246parameter age-dependencies as far as they have been quantified)
247and Japanese and Chinese ethnic libraries. When running a simula-
248tion the appropriate population should be selected and then the
249trial size (numbers and groups) together with age-range, gender
250proportions, fasted/fed status, fluid taken with dose and dosing
251regimen including staggering for up to four co-dosed drugs.
252A time-dependent fluid volume dynamics model handles basal
253luminal fluid, additional fluid taken with dose, biological fluid
254secretion rates, and fluid absorption rate in the fasted or fed state.
255Gastric emptying, intestinal transit times and their inter-individual
256variability are incorporated for the fasted and fed states. The
Fig. 1. Structure and features of the Simcyp ADAM Model: the relative lengths
(diameters) of the cylinders reflect the relative lengths (or diameters) of the GI
segments for a representative individual in vivo while the purple shading represents
the CYP 3A enzyme abundance distribution in the intestinal tract.
E.S. Kostewicz et al. / European Journal of Pharmaceutical Sciences xxx (2013) xxx–xxx 3
24 September 2013

257 lengths and radii of intestinal regions are individualized based
258 upon covariation with body surface area and jejunal–ileal ratios,
259 enabling individualized regional absorptive surface areas to be
260 determined (Jamei et al., 2009b).
261 Gut wall passive and active permeability (Peff,man) can be pre-
262 dicted from in vitro permeability measurements (Caco-2, etc.) or
263 using QSAR-type models; regional Peff differences can be specified.
264 The dog and mouse models also include a mechanistic model that
265 accounts for transcellular, paracellular and mucus/unstirred layer
266 permeability, calculates regional absorptive surface area (SA) from
267 villous dimensions (and in humans the plicae circulares), and fur-
268 ther scales to effective SA by the method of (Oliver et al., 1998).
269 Bile micelle partition of drug affects free fraction in luminal fluids
270 (Buckley et al., in press) and thus can provide an additional food ef-
271 fect where bile salt concentrations are elevated particularly in the
272 fed state.
273 Drug entering the enterocytes may be metabolised by gut wall
274 enzymes and/or subject to efflux by gut wall transporters in a com-
275 plex interplay (Darwich et al., 2010). Regional abundances of gut
276 wall enzymes (CYPS: 2C9, 2C19, 2D6, 2J2, 3A4, 3A5) and transport-
277 ers (P-glycoprotein (P-gp), MPRP2, BCRP) are incorporated where
278 information is available (Harwood et al., 2012) a generic apical
279 influx transporter functionality is provided. Segregated villous
280 blood flows to each compartment of the intestinal tract are scaled
281 up 1.3-fold in the fed state to account for increased postprandial
282 blood perfusion. Enterohepatic recirculation of drug is handled
283 with different patterns of gallbladder accumulation or by-pass
284 according to fasted or fed status. In the fasted state gallbladder
285 drug release is linked to late Phase II/early Phase III of individual-
286 ized Interdigestive Migrating Motor Complex (IMMC) cycle times.
287 Metabolic and transporter-mediated drug–drug interactions in
288 the intestinal wall can be modeled separately in each compartment
289 (Neuhoff et al., submitted for publication).
290 ADAM has been adapted to model the impact of bariatric sur-
291 gery in morbidly obese patients on drug BA including Roux-en-Y
292 gastric bypass, biliopancreatic diversion with duodenal switch,
293 sleeve gastrectomy and jejunoileal bypass (Darwich et al., 2012).
294 In the absence of clinical studies in such population groups these
295 models provide useful guidance for adjusting dose regimens. In an-
296 other application, Patel et al. (submitted for publication) described
297 the prediction of formulation-specific food effects for three formu-
298 lations of the Biopharmaceutics Classification System (BCS)/
299 Bipharmaceutical Drug Disposition Classification System (BDDCS)
300 Class 2 drug nifedipine using the ADAM model (Fig. 2). BCS/BDDCS
301 classification and QSAR-based approaches failed in this regard be-
302 cause they consider drug properties alone rather than those of the
303 formulation. The nifedipine ADAM model was based solely upon
304 in vitro data, aside from prior knowledge of negligible renal clear-
305 ance (pre-clinical species) and that nifedipine is readily absorbed
306 from the colon.
307 The ADAM model can also be used to establish physiologically-
308 based (PB) in vitro–in vivo correlations (IVIVC) (PB-IVIVCs). Tradi-
309 tional deconvolution methods are generally empirical and do not
310 deconvolute in vivo dissolution separately from GI transit, perme-
311 ation or first-pass metabolism. Where there is significant gut wall
312 or hepatic first-pass metabolism of drug establishment of robust
313 relationships between in vitro and deconvoluted in vivo dissolution
314 profiles can become difficult, perhaps requiring complex non-lin-
315 ear functions. PBPK models separately handle the different mecha-
316 nisms influencing drug disposition. Thus PB-IVIVC models should
317 permit the development of improved/simplified IVIVCs for BCS I
318 or II drugs with significant first pass extraction and inter-individual
319 variability. It also opens up the possibility to develop IVIVCs for: (i)
320 CR formulations of BCS III drugs (e.g. gastro-retentive) where
321 absorption is governed by the complex interplay of release,
322 transit/gastro-retention and permeability rather than being solely
323release-limited, and (ii) BCS IV compounds whose absorption
324may be controlled by a similar interplay of multiple processes. This
325approach has thus far been successfully applied to the develop-
326ment and validation of IVIVC for CR formulations of the BCS Class
327I, high first pass extraction (CYP 2D6 substrate) drug metoprolol
328and also employed in the design of new formulations and in under-
329standing population variability (Patel et al., 2012a,b) (Fig. 3).
3302.2. GastroPlus™
331GastroPlus™ is a whole-body PBPK model distributed by Simu-
332lations Plus (Simulations Plus, Inc., Lancaster, CA, USA). Currently
333version 8 is available. GastroPlus™ uses the Advanced Compart-
334mental Absorption Transit (ACAT™) mechanistic absorption model
335(Fig. 4) which has been described in the literature (Agoram et al.,
3362001) to model the absorption of oral formulations from the GI
337tract.
338The GastroPlus™ implementation of PBPK provides an internal
339module called PEAR Physiology™ (Population Estimates for
340Age-Related Physiology) for the calculation of organ physiologies
341for American (Western) and Japanese (Asian) human models for
342any age from 0 to 85 years old and also will generate unique
343new populations using user-defined coefficients of variation for
344population simulations. The GastroPlus™ PBPKPlus™ module will
345automatically calculate tissue:plasma partition coefficients (Kp)
346or convenient import of user-specified Kp and fut values from
347tab-delimited ASCII text files. Beagle dog, rat, monkey, and mouse
348physiologies are also available for modeling preclinical in vivo data.
349Building models of the preclinical species can improve the
350confidence in estimation of input parameters for the human model.
351When in vitro or preclinical data are not available, model parame-
352ters can be fitted for individual tissues to best match observed
353data.
354The ACAT model for oral absorption describes each of the
355following including: drug release from the formulation, solubility,
Fig. 2. Prediction of food effects on the PK profiles of nifedipine immediate and
controlled release formulations (see Patel et al., submitted for publication in this
issue of EJPS).
24 September 2013

356 dissolution/precipitation rate, chemical stability, permeability,
357 carrier mediated influx and efflux and gut wall metabolism using
358 differential equations, most of which are defined in the user
359 manual. Most of the equations involve linear kinetics, whilst
360 Michaelis–Menten nonlinear kinetics is used to describe saturable
361 metabolism and carrier-mediated transport.
362 Input data are in vitro, in vivo or in silico estimates of the
363 drug- or formulation-related parameters, e.g., drug aqueous
364 solubility–pH relationship, permeability, particle size distribution
365 and formulation type. Physiological parameters such as GI transit
366 time, pH, absorptive surface area, bile salt concentrations in each
367 compartment, pore size and density, compartment dimensions
368 and fluid content, etc. are built into the model but can be modified
369 by the user. Many of the input parameters required for the Gastro-
370 Plus model can alternatively be predicted from structure using the
371 optional ADMET Predictor™ module or the full ADMET Predictor
372 program (Simulations Plus, Inc., Lancaster, CA, USA).
373 In GastroPlus™ (version 8) the formulation types that can be se-
374 lected include both Immediate Release (IR) formulations (solution,
375 suspension, tablet, and capsule) and Controlled Release (CR) for-
376 mulations (enteric-coated or other form of Delayed Release (DR)).
377 For CR, release of either dissolved material (drug in solution) or
378 undissolved material (solid particles, which then dissolve
379according to the selected dissolution model) can be evoked. For
380CR formulations, percent released vs. time can be specified as tab-
381ular release-time entries or alternatively using a single or double
382Weibull function.
383The drug must be in solution before it can be absorbed. The pKa
384value(s) and the Solubility Factor (SF – the ratio of the solubility of
385the ionized to unionized forms) are used to predict a pH vs.
386solubility curve under aqueous conditions. This solubility is then
387corrected for the influence of bile salts based on a theoretical rela-
388tionship or alternatively a directly measured in vitro solubility
389determined in biorelevant media can be used instead. The pKa va-
390lue(s) and SF can be fitted to measured solubility data in aqueous
391buffers. If the drug concentration in a compartment exceeds the
392solubility of drug in that compartment, the drug may precipitate.
393In this case, a drug may precipitate rapidly or may remain in a
394supersaturated state. However, the actual precipitation time is
395not easy to predict and may depend on the drug as well as the for-
396mulation used. GastroPlus™ uses single or multiple first order
397exponential precipitation or a complete mechanistic nucleation
398and growth model that can account for formulation effects due
399to nucleation inhibitors and solubilizers (Lindfors et al., 2008).
400When using the mechanistic nucleation and growth model, the
401simulation outputs include the size, time, and degree of
Fig. 3. Predicted profiles (lines) vs. observed (diamonds) for fast, medium and slow release formulations of metoprolol.
Fig. 4. Structure and features of the Advanced Compartmental Absorption Transit (ACAT™) Model.
24 September 2013

402 supersaturation for initial particle formation as well as the maxi-
403 mal degree of supersaturation achieved and the dynamics of parti-
404 cle size changes and absolute numbers in all of the GI
405 compartments.
406 Since the influence of bile salts on solubility, diffusion coeffi-
407 cient, dissolution rate, and precipitation is well documented, Gas-
408 troPlus™ automatically adjusts the concentration of bile salts in
409 each compartment for fasted and fed states, with the bile being al-
410 most completely reabsorbed by the distal end of the small intes-
411 tine. In the GastroPlus model, a solubilization ratio based on logP
412 (Mithani et al., 1996) can be used for the drug if measured
413 solubilities in biorelevant media (Fasted State Simulated Gastric
414 Fluid (FaSSGF) and Fasted and Fed State Simulated Intestinal Fluid
415 (FaSSIF and FeSSIF)) are not available. The solubilization ratio can
416 also be predicted from measured solubility in FaSSGF, FaSSIF and
417 FeSSIF. There are a number of choices for describing the dissolution
418 kinetics, including the Nernst–Brunner modification of the
419 Noyes–Whitney equation, Wang–Flanagang, Tekano’s Z-factor
420 and instantaneous dissolution model. The initial particle size can
421 be described by a distribution (Normal and Log-Normal) or tabu-
422 lated data to describe more complex particle size distributions.
423 The solubility of very small particles (nanoparticles) may be in-
424 creased via the Kelvin effect. The NanoFactor Effect is applied to
425 the dissolution equation which effectively increases solubility (as
426 well as concentration gradient to create faster absorption).
427 Loss of the drug via chemical/metabolic pH-dependent degrada-
428 tion in the lumen, e.g. valacyclovir hydrolyzed back to acyclovir at
429 neutral pHs and above (Sinko and Balimane, 1998) is determined
430 by interpolation from an input table of degradation rate (or half-
431 life) vs. pH, and the pH in each compartment.
432 Absorption rate in the ACAT model depends on the effective
433 permeability of the drug (transcellular and/or paracellular) and
434 the physiological Absorption Scale Factor (ASF) for each compart-
435 ment as well as the time-dependent concentration gradient be-
436 tween lumen and enterocyte (transcellular) or portal vein
437 (paracellular). ASF is used to account for changes in surface-area-
438 to-volume ratio along the GI tract, changes in distribution coeffi-
439 cient and regional permeability due to changes in pH, and changes
440 in paracellular pore size and porosity. These factors also form the
441 basis of a mechanistic model for scaling gut physiology between
442 preclinical and human studies.
443 In addition, in vitro Km values for influx transporters can be used
444 to accurately simulate nonlinear dose dependence for substrates
445 including valacyclovir, valganciclovir, gabapentin, and amoxicillin
446 (Bolger, Lukacova et al., 2009).
447 Local variations in absorption/exsorption e.g. due to different
448 expression of active influx or efflux transporter(s) are calculated
449 using included distributions of various transporters from a number
450 of publications. These are the relative expression levels of the
451 transporter within the intestinal compartments and unlike en-
452 zymes, these values are not related to the liver.
453 Gut wall metabolism is based on local expression levels of en-
454 zymes in each enterocyte compartment included in the program.
455 Scaling of gut expression levels with respect to whole liver enzyme
456 content is used for the default expression levels of all of the com-
457 mon highly expressed CYP and UDP Glucuronosyltransferase (UGT)
458 enzymes in the gut. The accuracy of GastroPlus™ to estimate drug
459 concentrations in the intestinal enterocytes and the ability to accu-
460 rately reproduce nonlinear dose dependence for substrates of CYP
461 enzymes using only in vitro metabolic data has been demonstrated
462 in the literature for UK-343,664, a P-gp and CYP3A4 substrate. This
463 example represented an example of conducting such a simulation
464 using mechanistic oral absorption compartmental absorption and
465 transit models (Abuasal et al., 2012). GastroPlus™ has also been
466 used to accurately model the nonlinear dose dependence for sub-
467 strates of ATP-binding cassette efflux transporters and substrates
468for both CYP enzymes and efflux transport (Abuasal et al., 2012;
469Tubic et al., 2006).
470GastroPlus™ has capabilities which allows for input into pro-
471jects from discovery pharmacokinetics through clinical develop-
472ment. For example, it has been implemented in in vitro–in vivo
473extrapolations in animals and humans (e.g. De Buck et al., 2007),
474clinical formulation development and implementation of quality
475by design (e.g. Crison et al., 2012) and generation of IVIVC (e.g.
476Mirza et al., 2013).
477Recently, GastroPlus™ was used by GlaxSmithKline (GSK) in the
478evaluation of different formulations of a weak acid compound
479developed as a salt. Whilst the current clinical formulation con-
480tained the sodium salt (Formulation A), the Project Team wanted
481to investigate the behavior of a formulation containing the less sol-
482uble free acid in either a nanomilled (Formulation B) or micronised
483(Formulation C) form. Following investigation in clinical study,
484whilst exposure was only slightly reduced for Formulation B, expo-
485sure was significantly lower for Formulation C compared to the
486current formulation.
487GastroPlus™ was therefore used to investigate whether it was
488possible to obtain similar exposure levels to Formulation A using
489formulation C by either increasing the dose slightly or using a
490nanomilled formulation. The input parameters to the GastroPlus™
491model were the measured solubility in Britton–Robinson buffers at
492pH 2, 4, 6, 8, 10 and 12, solubility in SGFsp, FaSSIF and FeSSIF buf-
493fers, permeability in MDCK cells, and plasma protein binding. The
494human liver clearance was predicted from in vitro–in vivo Extrap-
495olation. The tissue:plasma partition coefficients (Kps) were pre-
496dicted using the default Rodgers-Single (Lukacova) Kp method
497(Lukacova et al., 2008). Formulation-specific input parameters
498were the particle size distribution, the enhancement in solubility
499for the salt (Formulation A), and the enhancement in solubility
500due to the nanoparticle effect (Formulation B).
501The GastroPlus™ model predicted that it would not be possible
502to get equivalent exposure to Formulation A with Formulation C at
503a modest increase in dose. The model also showed that the phar-
504macokinetics of Formulation B would be insensitive to changes in
505particle size in the nm range. This study, in which a mechanistic
506model of formulations is constructed to understand the impact of
507change is in alignment with the U.S. Food and Drug Administration
508(FDA) Quality-by-design (QbD) initiative (Yu, 2008).
5092.3. PK-SimÒ
510PK-SimÒ
is a whole-body PBPK model distributed by Bayer
511Technology Services GmbH (Leverkusen, Germany). Currently, ver-
512sion 5 is available.
513PK-SimÒ
5 allows quantitative pharmacokinetic predictions not
514only in humans, but also in preclinical mammals like Beagles, dogs
515(mongrels), mice, minipigs, monkeys, and rats. Whilst mean popu-
516lation values of the anatomy and physiology for each species are
517available in the software, these values can be altered by the user
518(PK-SimÒ
, 2012).
519The software contains detailed information of the age, gender,
520and race-related physiological parameters of various human study
521populations including black, white and Mexican Americans
522(NHANES, 1997), Europeans (Valentin, 2002); and Asians (Tanaka
523and Kawamura, 1996). Additionally, the different populations can
524be further modified, thereby enabling simulations to be conducted
525in special patient populations including obese and the renally im-
526paired, in addition to children and the elderly. Using the Population
527Module, simulations can be performed in a virtual population with
528up to 1000 individuals (each having different anatomical and phys-
529iological properties).
530As shown in Fig. 5, the structure of PK-SimÒ
5 is based on com-
531partmental models. In the current version, the model structure of
24 September 2013

532 the GI tract consists of 12 segments (representing the stomach,
533 duodenum, upper and lower jejunum and ileum, respectively, cae-
534 cum, ascending, transverse, descending, and sigmoid colon, and
535 rectum). Each of the different intestinal segments contains an addi-
536 tional mucosal segment (Thelen et al., 2011, 2012). The detailed
537 description of the gut wall (including the mucosa) enables the
538 implementation of various processes that take place in the gut
539 wall, such as first pass metabolism, active transport mechanisms,
540 and additionally encompasses drug–drug and drug–protein inter-
541 actions. However, these reactions are not limited to the GI com-
542 partments and may be implemented in any relevant organ or
543 tissue (PK-SimÒ
, 2012).
544 In PK-SimÒ
5, all organs are presented as compartments and
545 each compartment is subdivided into a vascular and an extravascu-
546 lar space. The vascular space is then further subdivided into plasma
547 and red blood cells, whilst the extravascular space is further subdi-
548 vided into the interstitium and the cellular space. The fractions of
549 lipid, protein, and water in each organ and sub-compartment are
550 embedded in the software, enabling the calculation of tissue–plas-
551 ma-partition coefficients using five different approaches (PK-SimÒ
552 standard (PK-SimÒ
, 2012), Rodgers and Rowland (Rodgers et al.,
553 2005a,b; Rodgers and Rowland, 2006, 2007), Schmitt (Schmitt,
554 2008), Poulin and Theil (Poulin et al., 2001; Poulin and Theil,
555 2000, 2002a,b), and Berezhkovskiy (Berezhkovskiy, 2004a,b)). Dif-
556 ferential mass-balance-equations not only describe the transport
557 of the drug from one sub-compartment to the other sub-compart-
558 ment but also describe the transport from one organ to another.
559 For drugs applied via the oral route, the underlying assumption
560 is that the drug needs to dissolve during its passage through the GI
561 tract before being absorbed. PK-SimÒ
5 allows the calculation of
562 the pH-dependent solubility of ionizable compounds using the
563 Henderson–Hasselbalch equation. To describe the solubility of
564 the drug, the software allows the input of one solubility value,
565 either in a blank buffer or in biorelevant media (e.g. FaSSIF or FeSS-
566 IF). However, the current version of the software does not support
567 an a priori calculation of the drug’s bile dependent solubility.
568To describe the GI dissolution characteristics of an oral formu-
569lation, seven different input functions can be selected. These in-
570clude the situation for an oral solution, user-defined dissolution
571which enables a direct upload of dissolution data from an Excel file,
572Weibull kinetics, a ‘‘lint 80’’ dissolution function (which describes
573linear dissolution until 80% of the drug is dissolved, then hyper-
574bolic extrapolation to 100% dissolution), particle-size dependent
575dissolution based on Noyes–Whitney kinetics, zero order, and first
576order dissolution kinetics.
577Neither a supersaturation ratio or precipitation kinetics can be
578used to describe GI supersaturation or precipitation in PK-SimÒ
5795. Instead, precipitation is accommodated in the simulation by
580selection of either the ‘‘dissolved’’ or the ‘‘particle dissolution’’
581function. In this case, the drug is allowed to precipitate according
582to its pKa, its pH-dependent solubility profile, and the pH gradient
583along the GI tract in a kind of ‘‘reverse dissolution’’ step. Further, it
584is possible to take into account the possibility of the precipitate
585going back into solution (PK-SimÒ
, 2012).
586To describe the permeability of the drug across the gut wall,
587experimentally determined values from Caco 2 or PAMPA assays
588can be used directly in the simulations. However, if these data
589are not available, the transcellular intestinal permeability can be
590calculated using the basic physicochemical properties of the drug,
591including lipophilicity, the effective molecular weight, and the pKa
592(PK-SimÒ
, 2012). The software is able to accommodate situations
593in which the drug is a substrate for active transport mechanisms,
594active efflux, influx, and Pgp-like transport mechanisms.
595PK-SimÒ
5 describes first pass metabolism using various kinetic
596models, such as Michaelis–Menten or first order kinetics. For this
597purpose, the corresponding enzyme abundance for each organ
598(such as the liver or the various GI compartments) can be incorpo-
599rated into the software. Age-dependent ontogeny for a broad vari-
600ety of CYP and UGT enzymes are taken into consideration in PK-
601SimÒ
5. Additionally, specific drug–protein- and drug–drug-inter-
602actions can be defined at the enzyme level. Scaling from in vitro
603to in vivo clearance is also possible (PK-SimÒ
, 2012).
Fig. 5. Structure of the absorption model used in PK-SimÒ
(Version 5 and upwards). The model also includes the large intestine (not shown in this figure) which has a similar
structure to the small intestine compartment (reproduced with permission from Thielen et al. (2012)).
24 September 2013

604 An additional feature of PK-SimÒ
is that it can be combined with
605 the MoBiÒ
software (Bayer Technology Services GmbH, Leverku-
606 sen, Germany) which not only enables a completely new PBPK
607 model to be built from scratch but also allows modifications of
608 existing PK-SimÒ
models.
609 With PK-SimÒ
, the description of the anatomy and physiology
610 for different species, enables, together with the MoBiÒ
software,
611 a detailed description of the events taking place during the absorp-
612 tion process and also immediately post absorption. Further, the
613 application of the software is not just limited to pharmacokinetic
614 simulations, but also enables the link between pharmacokinetics
615 and pharmacodynamics to be examined. A search of the literature
616 reveals that PK-SimÒ
is mostly used for describing highly complex
617 pre- and post-absorptive processes, such as metabolism, mass-bal-
618 ance, and pharmacogenomics in different CYP phenotypes or pop-
619 ulation subgroups (Dickschen et al., 2012; Eissing et al., 2011b;
620 Willmann et al., 2009a), the influence of drugs on the pharmacoki-
621 netics and pharmacodynamics of hormonal systems (Claassen
622 et al., 2013; Eissing et al., 2011a) and pharmacokinetic scaling ap-
623 proaches (Strougo et al., 2012; Weber et al., 2012). In contrast,
624 there are fewer publications that deal with describing drug absorp-
625 tion in humans and preclinical mammals (Thelen et al., 2010; Will-
626 mann et al., 2007, 2009b, 2003, 2010).
627 Recently, PK-SimÒ
was used to investigate the lack of dose-lin-
628 earity in nifedipine pharmacokinetics (Thelen et al., 2010). In vivo
629 data following oral administration of a soft gelatine capsule con-
630 taining 5 mg, 10 mg, or 20 mg nifedipine in an IR format showed
631 a dose-linear increase of the AUC. In contrast, linearity was not ob-
632 served for Cmax following administration, with sub-proportional in-
633 creases at the 20 mg dose (Raemsch and Sommer, 1983). To
634 estimate the factors that led to the marked reduction of Cmax values
635 in the absorption of nifedipine from the 20 mg dose, biorelevant
636 dissolution tests in FaSSGF were performed with both 10 mg and
637 20 mg nifedipine. In the dissolution tests, precipitation of nifedi-
638 pine was observed for the 20 mg formulation, but not for the
639 10 mg formulation. To describe the intralumenal dissolution of
640 the drug, a Weibull function was fitted to the in vitro dissolution
641 profiles of both formulations (10 mg and 20 mg) and implemented
642 in the software. As nifedipine is known to exhibit significant first
643 pass metabolism via CYP 3A enzymes, Michaelis–Menten kinetics
644 were used to accommodate both small intestinal and hepatic
645 CYP-mediated metabolism.
646 The simulation results showed that nifedipine plasma profiles
647 could be simulated well by combining the in vitro dissolution test
648 results with the PBPK software, and it could be demonstrated that
649 the onset in nifedipine absorption following administration of
650 20 mg nifedipine was limited by gastric drug precipitation. The
651 software, combined with in vitro dissolution tests, thus served as
652 an important tool for investigating the intralumenal dissolution
653 behavior of this poorly soluble drug, Additionally, this could only
654 be achieved using a quantitative description of the gut wall metab-
655 olism of nifedipine (Thelen et al., 2010).
656 2.4. Other in silico tools
657 2.4.1. MatLabÒ
658 Increased availability of commercial software platforms facili-
659 tates wider use of PBPK modeling as a ‘learn and confirm’ tool at
660 different stages of drug development (Jones et al., 2011b; Parrott
661 and Lave, 2008; Rostami-Hodjegan, 2012; Rowland et al., 2011;
662 Sinha et al., 2012). In addition to commercial packages, a number
663 of studies have reported the application of in-house, user custom-
664 ized models built in either MatlabÒ
, Berkeley Madonna, MoBiÒ
or
665 STELLAÒ
(Bouzom et al., 2012; Gertz et al., 2013, 2011; Jones
666 et al., 2012; Kambayashi and Dressman, 2012). These customized
667 models provide a certain flexibility and can be extended/used for
668multiple purposes beyond generic PBPK, e.g., PBPK-PD (Claassen
669et al., 2013), to account for dissolution and precipitation kinetics
670(Kambayashi and Dressman, 2012) or modeling of in vitro trans-
671porter kinetics and other cellular process (Korzekwa et al., 2012;
672Menochet et al., 2012).
673An example of the implementation of the mechanistic intestinal
674absorption model in MatlabÒ
has been reported recently for CYP3A
675substrates with high intestinal extraction (Gertz et al., 2011). The
676intestinal model within the whole body PBPK framework followed
677the principles of compartmental absorption and transit model (Yu
678and Amidon, 1999). For all the drugs investigated, absorption was
679considered to occur from the small intestinal compartments, with
680the exception of saquinavir, where colonic absorption was also
681incorporated, in agreement with previous reports (Agoram et al.,
6822001). The intestinal model allowed description of the changes of
683drug amount in the intestinal lumen corresponding to different
684intestinal segments of duodenum, jejunum and ileum and ac-
685counted for both undissolved and dissolved drug. Inclusion of the
686heterogeneous expression levels of CYP3A and efflux transporters
687along the small intestine allowed the prediction of enterocytic
688drug concentration in different intestinal segments and assess-
689ment of any potential nonlinearity/saturation of the intestinal
690first-pass. The latter has been implicated in the under-prediction
691of intestinal availability (FG) observed with the use of minimal
692models as the QGut model (Gertz et al., 2010, 2011; Yang et al.,
6932007). In addition to FG predictions, this custom built PBPK model
694in MatlabÒ
allowed the assessment of apparent i.v. and oral clear-
695ance, with the majority of the drugs predicted within 3-fold of the
696observed data.
697In another study, a cyclosporine PBPK model constructed in
698MatlabÒ
was applied to predict the effect of different formulations
699(NeoralÒ
and SandimmuneÒ
) on the magnitude of the inhibition of
700CYP3A4 metabolism and efflux via P-gp in different intestinal seg-
701ments, together with the impact on hepatic counterparts. Due to
702slower absorption of cyclosporine from SandimmuneÒ
formula-
703tion, the predicted reduction of transporter and enzyme activity
704was lower and the effect was more protracted in comparison to
705NeoralÒ
(microemulsion) (Gertz et al., 2013).
706Depending on the specific questions that need to be addressed,
707application of a reduced or semi-PBPK model can in some instances
708be more advantageous than a whole body model. This is in partic-
709ular the case when the focus of the analysis is on a specific organ,
710either due to drug–drug interaction or absorption concerns (e.g., li-
711ver and intestine) or safety issues (e.g., muscle in the case of stat-
712ins). By doing so, the number of tissue compartments is minimized
713and tissues of less relevance/interest can be lumped together; this
714is often performed by grouping together tissues which show com-
715parable perfusion and distribution equilibrium characteristics (ra-
716pid or slow) (Nestorov et al., 1998). In some instances, the model
717can be reduced to either just a central compartment or with a con-
718sideration of an additional peripheral compartment, while keeping
719a mechanistic description of metabolism/transporter processes in
720the organs of interest, e.g., intestine (Ito et al., 2003; Quinney
721et al., 2010; Zhang et al., 2009). These reduced models are more
722amenable to parameter optimization whilst still retaining
723physiological relevance and extrapolative power. The reduced
724PBPK models have been used to describe nonlinear drug
725disposition (e.g., clarithromycin) (Quinney et al., 2010) and to
726predict the magnitude of intestinal interaction in addition to liver,
727as illustrated in the midazolam and diltiazem example (Zhang
728et al., 2009).
7292.4.2. STELLAÒ
730A simple integrated PBPK model into which a user can build
731functionality can be achieved through the use of the STELLAÒ
732model platform (Cognitus Ltd., North Yorkshire, UK). STELLAÒ
24 September 2013

733 (Structural Thinking Experimental Learning Laboratory with Ani-
734 mation) is an icon-based model building and simulation tool that
735 was first introduced in the 1980s. In contrast to commercially
736 available PBPK models available, through the use of a graphical
737 interface, the user constructs and runs the simulation by building
738 a graphical representation of the model.
739 The utilization of STELLAÒ
in PBPK modeling was first realized
740 in the late 1980s (Grass and Morehead, 1989; Washington et al.,
741 1990). In one early publication, STELLAÒ
was used to examine
742 the potential influence of CR formulations of moexipril on the po-
743 tential effects on plasma concentrations (Grass and Morehead,
744 1989). This was achieved by the input of in vitro dissolution data
745 into the model, Since then, the functionality of the models have be-
746 come more elaborate and the STELLAÒ
model has been used to not
747 only simulate but also accurately predict human plasma profiles
748 through the addition of using biorelevant in vitro dissolution data
749 and, more recently, also in vitro precipitation kinetics (e.g. (Juene-
750 mann et al., 2011; Nicolaides et al., 2001; Shono et al., 2011; Wag-
751 ner et al., 2012).
752 As an example, STELLAÒ
software has been successfully used to
753 accurately predict plasma profiles for different fenofibrate lipid-
754 based formulations under fasting conditions in humans (Fei et al.,
755 2013). For the simulations, in vitro data from dispersion/dissolu-
756 tion (of the dosage forms), precipitation as well as re-dissolution
757 of the precipitate were taken into account in the model. Using
758 the STELLAÒ
model map shown in Fig. 6, the plasma profiles for
759 the different formulations were in very good agreement with the
760 in vivo plasma profiles for these formulations.
761 Whilst the STELLAÒ
platform is very useful if the basis of data
762 available is limited, it is essential that in vivo pharmacokinetic data,
763 in particular the elimination rate constant, are available if the goal
764 is to simulate/predict plasma profiles. Further, when several fac-
765 tors can directly influence the absorption characteristics including
766 first-pass metabolism, interaction with P-gp and regional differ-
767 ences in absorption, all of these parameters would need to be in-
768 cluded in the model. In such cases, the model map may become
769 quite complicated to set up and it may be more relevant to perform
770 these simulations in one of the commercially available software
771 packages that have the various functionalities already incorporated
772 in the model.
773 2.4.3. GI-Sim
774 The internal AstraZeneca absorption model GI-Sim deploys a
775 compartmental physiological model for the GI tract in combination
776 with up to three compartments to describe the plasma concentra-
777 tion–time profile. The human physiological model adopted in
778GI-Sim constitutes of nine GI compartments coupled in series;
779the stomach (1), the small intestine (2–7) and the colon (8–9)
780(Fig. 7) (Sjögren et al., 2013). The description of the underlying
781physiology has been reported previously (Yu and Amidon, 1998,
7821999; Yu et al., 1996a). The physiological parameters for the differ-
783ent GI compartments were adopted from Heikkinen et al. (2012a).
784In GI-Sim, undissolved particles and dissolved molecules flow
785from one intestinal compartment to the next. In contrast, the bile
786salt micelles present in the small intestine are modeled with a con-
787stant concentration and a calculated micellar volume fraction of
7880.0002 (Persson et al., 2005b) in each small intestinal compart-
789ment. Each compartment is ideal, i.e. concentrations of dissolved
790and undissolved drug, pH, etc. are the same throughout the
791compartment, apart from a thin Aqueous Boundary Layer (ABL)
792at the intestinal wall. Each compartment has a specific biorelevant
Fig. 6. Model map used in STELLAÒ
to simulate plasma fenofibrate concentrations
in plasma following oral lipid formulations (reproduced with permission from Fei
et al. (2013)).
Fig. 7. A schematic view of the absorption model used in GI-Sim.
24 September 2013

793 pH and as a consequence the solubility of an ionisable compound
794 changes along the GI tract. Within a compartment, particles may
795 either dissolve or grow and dissolved molecules may partition into
796 the bile salt micelles or be transported across the intestinal mem-
797 brane. The area available for absorption in each compartment is
798 calculated, assuming an ideal tube, from the compartment vol-
799 umes and a mean radius (weighted by the segments length) of
800 1.15 cm. An area amplification factor, which decreases in the distal
801 compartments, is included to account for how folds, villi and
802 microvilli structures affect the area available for absorption (Mudie
803 et al., 2010; Willmann et al., 2004).
804 In GI-Sim, the pH-dependent solubility of a compound is de-
805 scribed by the Henderson–Hasselbalch equation and dissolved un-
806 charged molecules may also partition into the micelles in the small
807 intestinal compartments (Sjögren et al., 2013). The rate of dissolu-
808 tion and crystalline nucleation is described by Fick’s law together
809 with the Nielsen stirring model while the rate of crystalline nucle-
810 ation is modeled by a modified version of the crystalline nucleation
811 theory (Lindfors et al., 2008; Nielsen, 1961).
812 The membrane transport process in GI-Sim is modeled as a se-
813 rial diffusion through the ABL of thickness L, with permeability
814 PABL, and a membrane, with the permeability Pm. Together they
815 constitute a barrier to membrane transport and absorption with
816 the total effective permeability P, described by:
817
1
P
¼
1
PABL
þ
1
f0 Á Pm
ð1Þ
819819
820 where f0 is the uncharged fraction. The human effective permeabil-
821 ity input in GI-Sim is estimated from an established correlation be-
822 tween measured human effective permeability and apparent
823 permeability from the Caco-2 model for a number of reference
824 drugs. To account for differences in ABL thickness due to inadequate
825 stirring in vitro, the correlation is based on Pm rather than P. Undis-
826 solved drug particles diffuse slower than free drug monomers
827 across the ABL but contain many molecules and may therefore con-
828 tribute substantially to the PABL. In GI-Sim, the diffusion coefficient
829 of particles across the ABL is inversely proportional to the particle
830 radius. The general effect of particles is that PABL increases with
831 increasing concentration of particles. This effect will be especially
832 important for small nanosized particles and contribute to the
833 improved absorption provided by such formulations. GI-Sim also
834 includes other functionalities such as luminal degradation, drug re-
835 lease profiles e.g. for CR formulations as well as physiology models
836 for pre-clinical species. The contribution of gut wall and first-pass
837 liver metabolism to BA is estimated as a combined first pass effect.
838 The accuracy of the absorption model in GI-Sim regarding pre-
839 diction of fabs and plasma exposure in humans has been evaluated
840 by simulating the fabs of twelve compounds reported to be
841 incompletely absorbed in humans due to permeability, solubility
842and/or dissolution rate limited absorption (Sjögren et al., 2013).
843The overall predictive performance of GI-Sim was good as >95%
844of the predicted Cmax and AUC were within a 2-fold deviation from
845the clinical observations and the predicted plasma AUC was within
846one standard deviation of the observed mean plasma AUC in 74% of
847the simulations (Fig. 8). GI-Sim also captured the trends in dose
848and particle size dependent absorption of the study drugs as exem-
849plified by AZ2 and digoxin (Fig. 9). In addition, GI-Sim was also
850shown to be able to predict the increase in absorption and plasma
851exposure achieved with nano formulations. Based on the results,
852the performance of GI-Sim was shown to be suitable for early risk
853assessment as well as to guide decision making in pharmaceutical
854formulation development.
8553. Current challenges in PBPK setups for GI absorption
8563.1. Composition of luminal contents
857From an oral drug release and absorption perspective, the most
858important features of the GI tract in humans are the stomach, small
859intestine and proximal large intestine. In addition, secretions from
860the various accessory organs (including the gall bladder and
861pancreas) that supply the small intestine play a significant role.
862It is also important to consider that the GI tract is not a static
863environment. Rather, not only does the physiological state alter
864between fasting and fed conditions, but also as the drug/dosage
865formulation transits through the GI tract, the environment to
866which it is exposed is also continuously changing. To complicate
867things even further, inter-individual variability in GI physiology
868is an additional aspect that needs to be considered in the PBPK
869model.
870The most important parameter of the fasted state stomach is
871the acidic pH, which is typically below pH 2, but can range
872between 1 and 7.5 (Dressman et al., 1990; Lindahl et al., 1997).
873Following food intake, gastric pH almost instantaneously increases
874to between 4 and 7, depending on the nature of the meal
875(Apostolopoulos et al., 2006; Carver et al., 1999; Dressman et al.,
8761990; Kalantzi et al., 2006a). Gastric pH thereafter reduces to fast-
877ing levels within approximately 2–5 h after a solid meal (Dressman
878et al., 1990; Kalantzi et al., 2006a; Russell et al., 1993).
879In the fasting stomach, the contents are typically hypo-osmotic
880and influenced by the volume of water administered. Another
881important parameter is surface tension, which even under fasting
882conditions is much lower than that of water (72 mN/m) and with
883food can be even lower, depending on the nature of the meal
884(Efentakis and Dressman, 1998). In the stomach there are also
885two main digestive enzymes that may be important in terms of
886drug delivery. Pepsin, a proteolytic enzyme, is able to catalyze
887the degradation of peptides and protein drugs. Also lipases, which
Fig. 8. OvervieQ9 w of the overall predictive performance of GI-Sim. The graphs depict the accuracy in predicted AUC (A) and Cmax (B). Error bars in observed AUC and Cmax
represent standard deviation. Different drugs are indicated by color and shape, the solid line and the dotted lines represent the line of unity and a 2-fold difference,
respectively.
24 September 2013

888 are found in small quantities in the gastric fluid, are important for
889 initiating fat digestion in the stomach. As such they may also influ-
890 ence the behavior of lipid based formulations. (Porter et al., 2008).
891 The next main site is the small intestine, the pH of which is
892 influenced by the pH of the gastric contents entering the small
893 intestine and the buffering nature of the pancreatic secretions.
894 Additionally, direct bicarbonate secretion contributes to the rise
895 in pH along the small intestine towards the ileum. Under fasting
896 conditions, the pH in the upper small intestine is highly variable
897 with mean values of pH 6.5, but can range anywhere between as
898 low as 2 just below the pylorus up to around 7 in the mid to distal
899 duodenum. pH values in the upper small intestine can be influ-
900 enced by the phase of the Myoelectric Motor Complex (MMC)
901 (Woodtli and Owyang, 1995). Following food, the pH gradually de-
902 creases due to the emptying of acidic gastric contents and then re-
903 turns to fasting state values following complete gastric emptying.
904 By the mid to distal ileum, pH has increased to around 7.5–8.0
905 (Dressman et al., 1990), predominantly due to a combination of
906 absorption of nutrients and direct, local bicarbonate secretion,
907 leading to a lack of food effect on the pH value in the distal small
908 intestine.
909 Through the action of bile acids, cholesterol and other lipids, the
910 proximal small intestine contains a mixture of mixed micelles, lip-
911 osomes and emulsion droplets. These aggregates can not only en-
912 hance the solubility of lipophilic drugs but can also interact with
913 the dosage form or even with specific excipients contained within
914 the formulation. Under fasting conditions, intestinal chyme
915 contains bile acids in the 2–6 mM concentration range and low
916 concentrations of phospholipids (0.19–0.26 mM). Under fed condi-
917 tions, the concentrations of bile salts can increase to a mean value
918 of around 15 mM, and approximately 3 mM for phospholipids
919 (Kalantzi et al., 2006b; Persson et al., 2005a, 2006) (Bergström
920 et al., 2013 in this issue of EJPS). Under both fasting and fedQ4 condi-
921 tions, due to the constant presence of surface-active components,
922 surface tension is low, approximately 30 mN/m (Persson et al.,
923 2005a) and can thus exert a direct impact on drug wettability
924 and hence drug dissolution.
925 The presence of food triggers the secretion of digestive enzymes
926 from the pancreas into the small intestine. There are three major
927 types of enzymes secreted for the digestion of carbohydrates
928 (amylases), lipids (lipases) and protein (proteases). From a drug
929 absorption perspective, lipases are the most relevant in terms of
930 their ability to digest lipid formulations and the proteases and their
931 influence on the stability of peptide drugs.
932 In contrast to the proximal GI tract, the colonic fluids are less
933 well characterized and this is reflected by the less detailed descrip-
934 tion of the distal regions of the GI environment in the PBPK models
935 currently available. The main relevant activities important to drug
936 dissolution and absorption are fermentation by the large bacterial
937 population present and reabsorption of water and electrolytes. Due
938 to the fermentation of food residues into short chain fatty acids
939(SCFA), the pH drops from the terminal ileum to ascending colon
940to approximately pH 5.5–6.0 (Diakidou et al., 2009; Nugent et al.,
9412001). Thereafter, following absorption of the SCFA’s, pH increases
942in the distal colon to approximately 7 in the descending colon/rec-
943tum. Whilst the concentration of bile salts in the large intestine are
944relatively low compared to the small intestine, the surface tension
945of the colonic fluids remains quite low (approximately 40 mN/m)
946(Diakidou et al., 2009).
947In addition, PBPK models need to consider the large number of
948metabolically active bacteria present in the terminal GI tract and
949their mechanism degradation, which could also influence drug sta-
950bility (McConnell et al., 2008; Sousa et al., 2008) and ultimately the
951fraction of dose absorbed.
9523.2. Transit and hydrodynamics
953The GI transit of drug substance in both solid and liquid forms
954(i.e. in solution or suspension) can be of great importance for the
955absorption of drugs due to the distinctly diverse physiochemical
956conditions in different regions of the GI tract (Varum et al.,
9572010). A complicating factor in describing these processes is that
958the GI distribution for solid particles and liquids are inherently dif-
959ferent (Davis et al., 1986; Varum et al., 2010). Therefore, it is nec-
960essary to consider whether the drug remains within the
961formulation and hence is moving in the form of a solid particle,
962or if it is dissolved or suspended in the intestinal fluids. Whilst
963the majority of the PBPK models currently available (Agoram
964et al., 2001; Jamei et al., 2009b; Willmann et al., 2009b) recognize
965this general difference, most effort has been put into characterizing
966the GI distribution of dissolved/suspended drug probably as a con-
967sequence of focusing on predicting the absorption of poorly soluble
968drugs and for these cases, transit has often been assumed similar
969for suspended and dissolved drug.
970Under fasting conditions, the transit of solid formulations
971through the upper GI tract (stomach and small intestine) is primar-
972ily governed by the MMC (Coupe et al., 1991; Husebye, 1999). This
973results in a distinct cyclic pattern of electromechanical activity that
974triggers peristaltic waves that originate from the stomach and
975propagates through the small intestine. An MMC cycle consists of
9764 distinct phases, reoccurring every 1.5–2 h in the fasting state
977(Dooley et al., 1992; Sarna, 1985). Postprandially, MMCs disappear,
978to be replaced by a digestive motor activity characterized by regu-
979lar mixing and propelling movements that optimize nutrient
980absorption (De Wever et al., 1978).
981Larger solid objects such as capsules or single unit tablets have
982been demonstrated to have significantly different GI transit pat-
983terns (with respect to gastric emptying and colon transit times)
984compared to solutions or small solid units (e.g. pellets) (Abrahams-
985son et al., 1996; Davis et al., 1986; Varum et al., 2010). For gastric
986emptying, there is a dependency on the size of the formulation and
987this is most obvious when the dose is administered together with
Fig. 9. Observed (symbols) and predicted (dotted lines) plasma concentration time profiles of AZ2 (dose dependent/solubility limited absorption) (A) and digoxin (particle
size/dissolution limited absorption) (B). Administration of solutions are indicated by a star (Ã
).
24 September 2013

988 food (Davis et al., 1986; Khosla and Davis, 1990). Smaller units and
989 dissolved drug are typically emptied significantly faster than larger
990 units. This is in line with the stomach’s functionality of grinding
991 the solids down to a manageable size before emptying into the
992 duodenum. For large non-disintegrating capsules given in combi-
993 nation with heavy meals, gastric emptying has been reported as
994 late as 10 h post dosing (Davis et al., 1986). When administered
995 in the fasting state, gastric emptying is generally fast and for most
996 pharmaceutical granules and pellets with diameter less than 2 mm
997 are typically emptied within a less than 1 h (Bergstrand et al., 2011,
998 2009; Davis et al., 1986).
999 Also movement within the stomach can play a significant role
1000 especially on the absorption characteristics from MR formulations.
1001 Following dosing in the fed state, where MR formulations may re-
1002 main longer in the stomach, transit between the proximal stomach
1003 (fundus) and the distal stomach (antrum) can have an impact on
1004 the release and subsequent absorption characteristics (Weitschies
1005 et al., 2005). Apart from the fact that drug substance released in the
1006 proximal stomach is emptied from the stomach more slowly than
1007 drug release in the distal stomach, there can also be differences in
1008 the dissolution rate due to lower mechanical stresses and higher
1009 postprandial pH in the proximal compared to the distal stomach
1010 (Bergstrand et al., 2011, 2012b, 2009). Therefore, the inclusion of
1011 within-stomach movement in PBPK models might be an important
1012 consideration for the prediction, especially for MR formulations.
1013 Small intestinal transit is less dependent on the size of the for-
1014 mulation and concomitant food intake. If anything, solid particles
1015 appear to transit slightly faster than the liquid content (Hardy
1016 et al. 1986; Yuen, 2010). Fluid volumes along the GI tract have been
1017 measured by Schiller and co-workers using water-sensitive Mag-
1018 netic Resonance Imaging (MRI) in 12 healthy volunteers (6 female)
1019 (Schiller et al., 2005). The result from this study provided evidence
1020 that GI fluid is not continuously available throughout the GI tract
1021 but is found in cluster. Therefore, assuming fixed volumes for the
1022 individual small intestinal compartments, as in the case in the gen-
1023 eric PBPK models, may lead to incorrect predictions.
1024 It would also be important for PBPK models to consider the ex-
1025 tent to which the volume of fluid within the gut lumen (Vlumen)
1026 changes with time as a result of fluid intake, secretion and reab-
1027 sorption, as this could have a significant effect on the dissolution
1028 of a drug and hence the concentration presented to enzymes and
1029 transporters within the enterocyte.
1030 GI transit is a heterogeneous process for a solid dosage form,
1031 which sometimes moves quickly, sometimes slowly, experiencing
1032 fluids of varying composition, and being subject to varying peri-
1033 staltic pressures (Weitschies et al., 2010). A typical small intestinal
1034 transit time for solid dosage forms in healthy subjects is reported
1035 to range between 2 and 4 h. However, there is a considerable be-
1036 tween and within subject variability, with values ranging between
1037 0.5 and 9.5 h. It should also be recognized that a considerable part
1038 of the small intestinal residence time is spent at rest rather than
1039 the object moving continuously through this region (Weitschies
1040 et al., 2010).
1041 Assuming that fluid movement along the GI tract is consistent
1042 with the rate of gastric emptying and small intestine transit,
1043 ordinary differential equations can be used to simulate the change
1044 in fluid volume in the stomach and each intestinal segment. The
1045 gastric emptying and small intestine transit time, regional fluid
1046 secretion and reabsorption rates and the baseline fluid volumes
1047 are all affecting the regional fluid volume. As expected, the
1048 between and within subject variability of any of these processes
1049 can contribute to the observed systemic variability.
1050 Before entering the colon, solid dosage formsQ5 typically stagnate
1051 in the cecum for a variable period of time. Food intake stimulates
1052 emptying of cecum into the colon, a mechanism that is known as
1053 the gastro-ileocecal reflex (Schiller et al., 2005; Shafik et al.,
10542002). Further, transit of dosage forms through the colon is highly
1055variable and appears to occur during periods of relatively fast
1056movement followed by long periods of rest (Adkin et al., 1993).
1057The movement periods may be stimulated by food intake but this
1058is not always the case (Weitschies et al., 2010). The size of the par-
1059ticles has been shown to influence the movement through colon
1060with small solid pellet particles moving slower than larger single
1061unit formulations (Abrahamsson et al., 1996; Adkin et al., 1993;
1062Davis et al., 1984). At present, these aspects of GI transit have
1063not been characterized in enough detail to be represented well in
1064the PBPK models.
1065The between and within subject variability is large with respect
1066to almost all aspects of GI transit, even for homogeneous healthy
1067populations studied under highly controlled conditions. This vari-
1068ability can for many substances and formulations also propagate
1069into large variability in systemic exposure and treatment success.
1070Taking into account the fact that disease and pharmacological
1071treatments can also alter GI transit (Varum et al., 2010) and the fact
1072that feeding habits generally vary more in daily life than in a con-
1073trolled clinical study setting, the true expected population variabil-
1074ity in clinical practice is likely to be substantially larger.
1075Another important consideration is the influence of circadian
1076rhythm on GI transit. Whilst few studies have been undertaken
1077to examine this more closely, it is known that a relatively longer
1078gastric residence time has been described for tablets administered
1079at night time compared to day time (Coupe et al., 1992a,b) and that
1080the total residence time in the colon is likely prolonged due to pat-
1081terns of bowel movement (Sathyan et al., 2000). It is hoped that
1082further studies evaluating circadian rhythm on GI transit will be
1083undertaken so that this can be appropriately incorporated into
1084PBPK models and thereby enable simulations to be achieved for
1085the purpose of predicting the influence of dosing regimens other
1086than the typical morning dosing schedule.
1087PBPK models are also serving an important function to carry out
1088predictions for pediatric populations (Barrett et al., 2012; Björk-
1089man, 2005). Similar to the lack of knowledge as to how GI transit
1090differs in elderly and diseased subjects, little is known about po-
1091tential differences in the pediatric population. The current knowl-
1092edge base has been summarized in two recent review articles
1093(Bowles et al., 2010; Kaye, 2011). Incorporation of this information
1094in PBPK models could aid in development of formulations better
1095tailored to special needs in the pediatric population.
10963.3. Permeability, including transporters
1097Physiologically-based intestinal models contain enterocyte
1098compartments corresponding to a particular intestinal segment;
1099the entry of the drug into enterocyte is defined by its effective per-
1100meability (Lennernas, 2007) and a radius of that intestinal segment
1101(Yu et al., 1996b). Therefore, drug permeability, together with
1102in vitro metabolic clearance data, represents one of the key input
1103parameters in physiologically-based intestinal models (Gertz
1104et al., 2011; Heikkinen et al., 2012a; Jamei et al., 2009b; Pang
1105and Chow, 2012; Yang et al., 2007). With these methods, perme-
1106ability in the lower GI tract can be estimated based on surface area
1107differences for transcellular and passive absorption, but for para-
1108cellular and active uptake mechanisms, estimations of permeabil-
1109ity in these regions remains a challenge.
1110In vitro apparent permeability (Papp) data may be generated in
1111either non-cell based (PAMPA) or cell based systems (MDCK,
1112Caco-2, LLC-PK1). Transfected cell lines used for the assessment
1113of passive permeability should have low expression levels of
1114endogenous transporters (Hilgendorf et al., 2007) or should be
1115used in the presence of a an inhibitor of active processes e.g., P-
1116gp inhibitor GF120918 (Thiel-Demby et al., 2009), to allow unbi-
1117ased estimation of passive permeability. In vitro permeability assay
24 September 2013

1118 development, impact of the cell line, surfactants and time points
1119 have been discussed in detail elsewhere (Avdeef et al., 2007; Chiu
1120 et al., 2003; Matsson et al., 2005; Polli et al., 2001; Thiel-Demby
1121 et al., 2009). Use of either isotonic or gradient pH (6.5 and 7.4)
1122 for the permeability assays will affect the fraction ionized and
1123 potentially bias Papp (A–B) estimates depending on the physico-
1124 chemical properties of the compound investigated (Neuhoff et al.,
1125 2003). Assessment is generally performed at a single concentration
1126 below the anticipated luminal drug concentration which may
1127 result in over-estimation of the contribution of active efflux
1128 processes.
1129 Alternatively, effective permeability (Peff) can be obtained from
1130 in vivo measurements by a perfusion system (Loc-I-Gut) in the
1131 upper jejunum at a pH 6.5 and at therapeutic dose of the drug. This
1132 method provides a net estimate of paracellular, passive transcellu-
1133 lar diffusion and transporter mediated uptake or efflux; however,
1134 availability of such data is limited to approximately 35 compounds
1135 reported by Lennernas and colleagues (Knutson et al., 2009;
1136 Lennernas, 2007; Lennernas et al., 1993). A number of studies have
1137 reported the correlation between Papp in either Caco-2 or MDCK
1138 and Peff values obtained in upper jejunum (Gertz et al., 2010; Sun
1139 et al., 2002); these literature or in-house generated regression
1140 analyses for the same dataset can subsequently be used to predict
1141 the Peff for any new developing drug. It is important to note that
1142 available Peff values are associated with large inter-individual
1143 variability (on average 70%) which in some cases exceeds 100%
1144 (e.g., amoxicillin) (Chiu et al., 2003; Winiwarter et al., 1998). The
1145 use of these empirical regression equations for drugs with
1146 Papp < 10 nm/s may be problematic considering the large scatter
1147 of the data in this range (Gertz et al., 2010; Sun et al., 2002).
1148 Drug permeability is often incorporated in the intestinal models
1149 as a hybrid parameter together with enterocytic blood flow, as in
1150 the QGut model (Gertz et al., 2010; Yang et al., 2007), enabling
1151 description of either permeability or perfusion rate limited scenar-
1152 ios. Predictive utility of permeability data obtained in different cell
1153 lines has recently been assessed for a diverse set of 25 CYP3A
1154 substrates. Differences in the FG prediction success using the QGut
1155 model were minor regardless of the cellular system used (MDCK
1156 and Caco-2), with high degree of prediction accuracy for drugs
1157 with in vivo FG > 0.5. In contrast, imprecision was increased for a
1158 subset of 11 drugs with high intestinal extraction (Gertz et al.,
1159 2010). In the same study, use of permeability data predicted from
1160 polar surface area and hydrogen bonding potential resulted in the
1161 most biased FG predictions and significant under-prediction trend.
1162 This in silico approach is not recommended for drugs with high po-
1163 lar surface area (e.g., saquinavir), as the validity of the existing
1164 regression equation (Winiwarter et al., 1998) was not established
1165 for drugs within that chemical space.
1166 The increase in the complexity of intestinal models over the
1167 years is associated with increased availability of some of the sys-
1168 tem parameters, e.g., heterogeneous expression levels of CYP3A
1169 and efflux transporters along the small intestine (Berggren et al.,
1170 2007; Harwood et al., 2012; Mouly and Paine, 2003; Paine et al.,
1171 2006, 1997; Tucker et al., 2012). However, in contrast to some met-
1172 abolic enzymes, absolute abundance data for many intestinal
1173 transporters, regional differences in these estimates and inter-indi-
1174 vidual variability are generally lacking. In addition, information on
1175 the correlation between expression of different transporters (e.g.,
1176 P-gp vs. BCRP) or transporter-enzyme expression (e.g., BCRP vs.
1177 CYP3A4) in the same individuals is limited for intestine.
1178 Kinetic characterization of intestinal transporters and metabolic
1179 enzymes over the range of drug concentrations is the preferable
1180 way of obtaining in vitro input parameters for the mechanistic
1181 intestinal models, allowing the model to account for any potential
1182 saturation of the protein of interest. Lack of such extensive in vitro
1183 kinetic data for the intestinal efflux transporters (together with
1184transporter abundance and inconsistency in scaling), can be a lim-
1185iting factor for the mechanistic prediction of oral drug absorption.
1186Recently, compartmental modeling approaches have been dis-
1187cussed for the in vitro determination of kinetic parameters for ef-
1188flux transporters, highlighting the limitation of commonly
1189applied enzyme kinetic principles directly to monolayer flux data,
1190as this leads to incorrect parameter estimates (Kalvass and Pollack,
11912007; Korzekwa et al., 2012; Zamek-Gliszczynski et al., 2013). The
1192analysis has also emphasized the need to consider intracellular
1193rather than the media drug concentration as relevant for the inter-
1194action with the efflux transporters; however, such mechanistic
1195transporter kinetic data are currently not available for a large num-
1196ber of drugs.
11973.4. Intestinal wall metabolism
1198A recent analysis of the relative contributions of the fraction ab-
1199sorbed (Fa), FG and the fraction escaping hepatic elimination (FH) on
1200BA on 309 drugs studied in human has indicated that for 30% of the
1201compounds FG was <0.8, highlighting the importance of incorporat-
1202ing intestinal metabolism in both BA and dose predictions in drug
1203discovery and development (Varma et al., 2010). However, this is
1204often not the case and the lack of consideration of extrahepatic
1205metabolism or incorrect assumption of its minimal relevance
1206may result in under-prediction of oral clearance (Poulin et al.,
12072011). The estimation of FG is confounded by the difficulties in
1208defining the exact contribution of the intestine from conventional
1209i.v/oral dosing strategies either in human (Galetin et al., 2010) or
1210in vivo animal models without employing more labor intensive
1211cannulation based studies (Matsuda et al., 2012; Quinney et al.,
12122008), and a comprehensive knowledge of species differences in
1213intestinal metabolism.
1214Enterocytes contain a range of Cytochrome P450 (CYPs) (de Wa-
1215ziers et al., 1990; Paine et al., 2006), and conjugation enzymes (e.g.,
1216UGTs, sulfotransferases) (Riches et al., 2009; Strassburg et al.,
12172000). Intestinal enzymes show a large intra- as well as inter-indi-
1218vidual variability (Paine et al., 1997; Thummel et al., 1996) and dif-
1219ferential expression along the length of the small intestine, with
1220the highest level in the proximal regions (Galetin et al., 2010; Paine
1221et al., 2006); comparable distribution patterns in both expression
1222and activity along the human intestine have been reported for
1223CYP and UGTs (Paine et al., 1997; Strassburg et al., 2000; Tukey
1224and Strassburg, 2001; Zhang et al., 1999). Zonal enzyme expression
1225in the intestine reflects the need to compartmentalize the intestine
1226within the PBPK models in order to reflect these regional differ-
1227ences in metabolism (Gertz et al., 2011; Jamei et al., 2009b; Pang,
12282003). This is of particular relevance for modeling of metabolism of
1229compounds administered as Extended Release (ER) formulation
1230designed to escape extensive first-pass metabolism by dissolution
1231in regions of low metabolic capacity (Paine et al., 1997), or drugs
1232for which solubility may be altered by changes in GI pH or bile salt
1233secretions (Dressman and Reppas, 2000).
1234In vitro metabolism data may be obtained using intestinal
1235microsomes and corresponding cofactors for CYP and UGT metab-
1236olism (Cubitt et al., 2009; Gertz et al., 2010) or cytosol (Cubitt et al.,
12372011) to account for pathways such as sulfation. Caution should be
1238applied when utilizing microsomal data from samples obtained by
1239microsomal scraping due to reported reduced enzyme activity and
1240protein yield in comparison to enterocyte elution (Galetin and
1241Houston, 2006). Alternatively, hepatic microsomes can be used
1242for the initial assessment of intestinal CLint following the normali-
1243zation for enzyme abundance, as illustrated in the case of CYP3A
1244substrates where CYP3A4 metabolic activities were comparable
1245between the liver and intestine once expressed per pmol CYP3A
1246(Galetin and Houston, 2006; Gertz et al., 2010; von Richter et al.,
12472004). This approach has limitations if there are uncertainties
24 September 2013

Pbpk models for the prediction of in vivo performance of oral dosage forms

Pbpk models for the prediction of in vivo performance of oral dosage forms

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Ähnlich wie Pbpk models for the prediction of in vivo performance of oral dosage forms

Ähnlich wie Pbpk models for the prediction of in vivo performance of oral dosage forms (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

Pbpk models for the prediction of in vivo performance of oral dosage forms