Transcript No Slide Title

Development and Validation of the Physiologically Based iDEATM Predictive Model
G. Timony, D. A. Norris, G. D. Leesman, R. Retajczyk, S. Tran, Y. Chen, Y. Lee, N. Johnson, J. Castelo, K.-J. Lee, R. J. Christopher, P. Sinko, and G. Grass.
Trega Biosciences, 9880 Campus Point Drive, San Diego, CA 92121
Solubility
ABSTRACT
The equilibrium solubility of each compound was determined in simulated gastric fluid (SGF, pH 1.5) and simulated intestinal fluid (SIF, pH 5.0, 6.5, 7.0, and 7.5). One-half gram of each drug
compound was added to 5 ml of SGF or SIF and incubated at 37°C for at least 4 hours to attain saturation equilibrium. The pH was adjusted as necessary to maintain the correct pH. A 1 ml
sample was removed, filtered using a 0.45 Fm filter, and analyzed for drug concentration. Solubility was determined for each drug at each pH value in triplicate.
Purpose. To develop, optimize, and validate the physiologically-based iDEA predictive model for drug absorption. Methods. The simulation model used permeability (rabbit diffusion
chamber or Caco-2 cell), solubility and dose as inputs, and was developed using a database of 56 non-metabolized compounds and a total of 85 drug-dose combinations. Transit through the GI
tract was based on dispersed plug flow mixing kinetics. A set of parameters was designed to build a correlation between the in vitro data inputs and the in vivo clinical outcomes so that the
model could predict the fraction of dose absorbed into the portal vein (FDp) versus time. Genentech, Parke-Davis, Schering-Plough, SmithKline Beecham, Trega Biosciences, and others
provided the in vivo training data. The simulated FDp vs. time profiles, Cmax, and AUC values were compared to observed values that were calculated from oral and intravenous plasma
concentration versus time data. Eight compounds from the Biopharmaceutical Classification System (BCS) group were used in the optimization and evaluated as an internal validation set.
Three BCS compounds, which were not used in the optimization, were included as an external validation set. The observed FDp’s of the BCS compounds ranged from 47-100%. Results: The
compounds in the training set were diverse in terms of structure, solubility (0.05 ng/mL to >100 mg/mL), permeability (rabbit diffusion chamber: 0.059 to 118.010-6 cm/s; Caco-2 Cell 0.150
to 42.5 10-6 cm/s ), and absorption properties. When utilizing rabbit diffusion chamber permeability, the model predicted FDp for the 11 BCS compounds with a mean error of 5.3  8.9%.
The observed versus predicted r2 for FDp, Cmax and AUC were 0.92, 0.97, and 0.996, respectively. The model performed equally well when a single Caco-2 cell permeability value was used
instead of the rabbit diffusion chamber permeability data. Conclusions. The physiologically-based iDEA predictive model, trained on a diverse set of compound data and using either rabbit
diffusion chamber or Caco-2 cell permeability input, was successful in predicting the rate and extent of absorption for a partially external validation set consisting of 11 BCS compounds.
Figure 3: General Structure of iDEATM Predictive Model
Trns MF
The iDEATM predictive model is based on a multi-compartment representation of the human gastro-intestinal tract. The model is physiologic, using human intestinal flow rates, surface areas
and luminal pH values gathered from various literature sources. The flow model for the transit of soluble and insoluble drug is based on an approximation of dispersed plug flow (Figure 2).
The model accounts for the forward and retrograde movement of the solid and dissolved dosage form, accounts for dissolution of solid dosage form, adjusts solubility according the regional
pH, and calculates the compound flux in each region of the intestine (Figure 3). The correlation of in-vitro solubility and permeability to human absorption was achieved by optimizing the
value of a series or proprietary adjustment parameters against absorption parameters obtained through the pharmacokinetic analysis of the in-vivo training set data.
Diss
Trns MB
Trns MS01
Trns SM01
Solubility at pH 6.5
Fraction Absorbed (%)
Trns SB02
Trns SB
Flux
0.01
5
10
15
20
25
30
35
40
45
50
70.0
60.0
50.0
40.0
30.0
20.0
1.0e-5
1.0e-6
Figure 4: iDEATM Predictions for Training Set Compounds
1.0e-5
80
0.0
Compound rank order
70
1.0e-7
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
All Compound Doses rank order
5
10
15
20
25
30
35
40
Compound rank order
45
50
55
5
10
15
20
25
30
35
40
45
50
55
Compound rank order
60
50
40
30
85% w ithin FDp Criterion
Mean Error = 11%
10
Pharmacokinetic Analysis
10000
1000
100
Where FDp(t) = fraction dose absorbed to the portal vein at time t, D = dose, FDp =
fraction dose absorbed to the portal vein at infinity, t50 is the time for 50% of the
dose to be absorbed, and Pec is a parameter related to the slope of the error function.
Rabbit Intestinal Permeability Assay
Compound permeabilities through duodenum, jejunum, ileum and distal colon were determined in the apical -> basolateral direction using vertical Ussing-type low volume diffusion chambers
at 37°C. Donor and receiver chambers were each filled with 1.5 mL of pre-warmed Ringer’s buffer containing 25 mM glucose, pH 7.4. Samples(0.5mL) were collected from the receiver
chamber at 30. 45, 60, 75 and 90 minutes after experiment initiation. Removed sample volume was immediately replaced with 0.5 mL Ringer’s buffer containing 25 mM glucose. Compound
permeability was calculated using the equation below:
V dC
A C0 dt
Where Pe is the effective permeability in cm/s, V is the receiver chamber volume in milliliters,
A is the surface area available for transport (cm2), Co is the donor drug concentration, and dC/dt
is the slope of the best fit line through the concentration versus time profile in the receiver
chamber.
Caco-2 Effective Permeability Assay
0.31cm2
Effective permeability (Peff) was measured in the apical -> basolateral direction in 20-23 day old caco-2 cell cultures (Passage number 30-40) grown on a
filter. The donor side of the
chamber was dosed at a concentration of 100 uM in 300 ul of Ringers buffer at pH 7.4 containing a final concentration of 1.0% DMSO. The receiver side of the chamber contained 1200 ul of an
identical buffer. Samples were incubated at 37° C, in a 95% humidity chamber containing 5% CO2. 100 ul samples from the receiver side of the chamber were taken at 30, 50, 70, and 90
minutes post experiment initiation. Four replicates of each sample were performed on each day. Bioanalysis was performed using LC-MS, HLPC-UV or LSC. Peff was calculated using the
dX / dt
following formula:
A * C0 * 60
where X = mass transported, A = surface area and Co = initial donor drug concentration
Mean (SD)
Solubility
(mg/ml)
45.5 (42.1)
Rabbit Intestinal
Permeability X 106
(cm/sec)
9.3 (7.1)
Caco-2
Permeability X 106
(cm/sec)
19 (14)
FDp (%)
86.6 ( 19.7)
Low
0.01
30
40
60
70
80
1
90
1
100
10
100
1000
10000
1000000
AUC: Known vs Predicted
100000
10000
1000
100
r2 = 0.96
10
1
100000
1
iDEATM Predicted Cmax (ng/mL)
10
100
1000
10000
100000
1000000
iDEATM Predicted AUC (ng/mL*hr)
Figure 5: iDEATM Predictions for BCS Compounds
High
FDp: Known vs Predicted
>100
100
90
80
70
60
50
40
30
20
10
0
Cmax: Known vs Predicted
100000
Rabbit Permeability Input:
Mean Error =5.3%
Caco-2 Permeability Input:
Mean Error = 5.9%
2
r > 0.89
1000000
10000
1000
Rabbit Permeability Input
Caco-2 Permeability Input
100
r2 > 0.97
10
0
20
40
IDEA
0.38
50
Know n Cm ax (ng/m L)









Figure 2: Dispersed Plug Flow Model Applied to
Human Intestinal Transit of Indocyanine Green
20
iDEATM FDp Predictions
Know n FDp (%)
t

1

t50

D FDp 
2
FDp (t ) 
1  erf 1
2
t

Pec t50



Table 1: Diversity of the Validation Set: Model
Drugs (n=11) from the Biopharmaceutics
Classification System
10
r2 = 0.89
10
0
0
Cmax: Known vs Predicted
100000
90
10.0
55
FDp: Known vs Predicted
100
1.0e-6
Observed AUC (ng/m L*hr)
0.1
80.0
Observed Cm ax (ng/m L)
1
90.0
Permeability (cm/s)
10
1.0e-4
Permeability (cm/s)
100
100.0
Trns SA02
Ints A2
Ints A1
1000
20
Best-fit curves were determined for the intravenous and oral plasma concentration data of each drug compound at each dosage level in order to estimate the fraction of dose absorbed to the
portal vein (FDp). A two-compartment disposition model with elimination from the central compartment was used to fit the IV and PO curves simultaneously using weighted non-linear
regression. The oral input function was fit to the equation below:
Flux
Trns SA01
Jejunum Permeability
Caco-2 Permeability
Know n FDp
A diverse collection of 56 marketed drugs and drug failures was assembled from
consortium members and commercially available substances. All drugs were not metabolized in man
or had low hepatic clearance and thus were not subject to significant first pass metabolism. The following pharmacokinetic data was available each of the compounds: Plasma concentration vs
time curves following both oral and intravenous administration to healthy human subjects, data from human mass balance studies, and in-vitro metabolic stability data (human hepatocytes).
The solubility, permeability and pharmacokinetic properties of the training set were very diverse (Figure 1).
Prec Ints S2
RESULTS
Fraction of Dose Absorbed (%)
iDEATM
Peff (cm / sec) 
Trns SF01
Prec
Solubility, pH 6.5 (mg/ml)
Selection of Training Set Compounds
Trns MS02
Trns SM02
Trns SF
Figure 1: In-Vitro and In-Vivo Diversity of Training Set Compounds
METHODS
Diss
Trns MB02
Ints S1
The discovery and development process required to bring new drugs to market is both time consuming and expensive. Although recent practices such as high throughput screening and
combinatorial chemistry have increased the number of compounds secured at early phases of drug discovery, this has not translated directly to improvements in the rate or volume of
compounds moving ahead into development. A contributing factor is the bottleneck imposed by the need to conduct in-vivo pharmacokinetic evaluations on large numbers of compounds in
order to select the subset of compounds with desirable ADME properties in humans. The physiologically-based iDEA predictive model was designed to provide discovery and development
scientists with a tool to estimate the rate and extent of absorption of new chemical entities in humans, using only simple in-vitro measurements as inputs, and thereby eliminate the ADME
bottleneck. An overview of the development, optimization, and validation of the physiologically-based iDEA predictive model is presented here..
Ints M2
Physiologic Model
INTRODUCTION
.
Pe 
Trns MF01
Ints M1
Know n AUC (ng/m L*hr)
trega
TM
60
Predicted FDp(%)
25.0
80
100
10
100
1000
10000
100000
AUC: Known vs Predicted
100000
10000
Rabbit Permeability Input
Caco-2 Permeability Input
1000
r 2 > 0.99
100
100
iDEATM Predicted Cm ax (ng/m L)
1000
10000
100000
1000000
iDEATM Predicted AUC (ng/m L*hr)
CONCLUSIONS
0.02
 A diverse and unique database of solubility, permeability and human pharmacokinetic data for 56 non-metabolized compounds was assembled to train the iDEATM predictive model.
43.0
 The iDEATM predictive model requires only solubility and permeability (rabbit diffusion chamber or caco-2 cell) and dose as inputs.
 The iDEATM predictive model was constructed using a compartmental framework which incorporates human intestinal physiologic parameters.
47.2
100
From: N.F.H. Ho, J.Y. Park and W.I. Higuchi. Advancing Quantitative
and Mechanistic Approaches in Interfacing Gastrointestinal Drug
Absorption Studies in Animals and Humans. In “Animal Models for Oral
Drug Delivery in Man” W. Crouthamel and A.L. Sarapu Eds. APHA,
1973.
 The flow model utilized, based on an approximation of dispersed plug flow, provides an accurate representation of human gastro-intestinal transit.
 A correlation of in-vitro solubility and permeability to human absorption was achieved by optimizing the value of a series of proprietary adjustment parameters against absorption parameters
obtained through the pharmacokinetic analysis of the in-vivo training set data.
 The iDEATM predictive model accurately predicted the relevant biopharmaceutical outcomes (FDp, Cmax and AUC) for the training set, and for a partially external training set consisting of 11
BCS compounds (See Presentation Number 3512, Leesman et al, for the result of a blinded external validation of the iDEATM predictive model model).