Journal of Controlled Release 65 (2000) 5562 www.elsevier.

com / locate / jconrel

Development of predictive pharmacokinetic simulation models for drug discovery

a, a a,b a D.A. Norris *, G.D. Leesman , P.J. Sinko , G.M. Grass
b

Trega Biosciences, 9880 Campus Point Dr., San Diego, CA 92121, USA College of Pharmacy, Rutgers University, 160 Frelinghuysen Road, Piscataway, NJ 08854, USA Received 26 June 1999; accepted 10 July 1999

Abstract As discovery chemistry produces increased numbers of potential drug compounds, the use of ADME (absorption, distribution, metabolism, and excretion) properties is becoming increasingly important in the drug selection and promotion process. A computer simulation model has been developed and validated to predict ADME outcomes, such as rate of absorption, extent of absorption, etc. using a limited number of in vitro data inputs. The oral bioavailability of ganciclovir in dogs and humans was simulated using a physiologically based model that utilized many biopharmaceutically relevant parameters, such as the concentration of ganciclovir in the duodenum, jejunum, ileum, and colon at various dose levels and solubility values. The simulations were run and compared to dog and human in vivo data. The simulation results demonstrated that the low bioavailability of ganciclovir is limited by compound solubility rather than permeability due to partitioning as previously speculated. This technology provides a breakthrough in in silico prediction of absorption and with its continued development and improvement, will aid drug discovery and development scientists to produce better pharmaceutical products. 2000 Elsevier Science B.V. All rights reserved.
Keywords: Ganciclovir; Simulation models; Bioavailability; Humans; Dogs

1. Introduction Within the drug discovery environment, drug compounds must be selected based on their potential for success as a new drug product. The selection process, or screening, is classically accomplished using animal and / or in vitro model systems to determine the receptor-based activity prole for each candidate compound. Traditionally, promising drug candidates with acceptable receptor-based activity
*Corresponding author. Tel.: 11-858-410-6685; fax: 11-858410-6665. E-mail address: dnorris@trega.com (D.A. Norris)

are promoted to development and screened for ADME properties (oral absorption or permeability, distribution, metabolic stability, and excretion) or toxicity using numerous animal or in vitro models. Compounds which fail to demonstrate satisfactory ADME properties are removed from consideration as drug candidates. Prentis [1] observed, for a 20 year period (19681988), the primary reasons that drug candidates were dropped from development were pharmacokinetic difculties in man and a lack of efcacy. In 1997, this conclusion was conrmed by Kennedy who reported that the termination of the development of drug candidates other than for efcacy was poor pharmacokinetic (or ADME)

0168-3659 / 00 / $ see front matter 2000 Elsevier Science B.V. All rights reserved. PII: S0168-3659( 99 )00232-1

56

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62

characteristics [2]. Unfortunately, screening large numbers of compounds for pharmacokinetic properties, especially using animals, is a time and labor intensive process which limits its utility in drug discovery. In addition, although permeability data can be used to predict fraction dose absorbed in humans [35], these methods have broad limitations so that in vitro permeability data alone does not allow prediction of human absorption or activity before clinical data is collected. Numerous similar correlations using in vitro data have also been attempted and reported [610]. Although dissolution tests can be used to create models that will predict oral absorption, Dressman et al. [11] have concluded there is a real need to develop dissolution tests that better predict in vivo performance of drug products, indicating current dissolution based models may not be applicable to all situations. Furthermore, bioavailability studies using whole animals do not allow accurate prediction of human bioavailability [12]. A computer simulation model capable of screening compounds for ADME properties prior to promotion for development could drastically reduce the time and expense of total drug discovery and development. This is especially true for potential drugs which have absorption problems, are susceptible to drugdrug interactions or metabolic variability, or have large differences in bioavailability between animals and humans. This paper describes a physiologically based computer simulation model of the gastrointestinal tract capable of predicting oral absorption of drug compounds in mammals using a limited in vitro data set. The oral absorption properties of ganciclovir, a species and dose dependent low bioavailability drug [13], are used as an example to demonstrate the ability of the simulation model to accurately predict bioavailability in two mammalian species, humans and dogs.

ously published model [13], was designed to account for physiological parameters effecting oral absorption from the mammalian GI tract, such as pH, transit time, and intestinal surface area. The model was also designed to account for drug compound parameters, such as permeability, solubility and dissolution rate. The model was developed by incorporating these parameters into a series of computerimplemented differential equations capable of predicting the rate and extent of absorption into the portal vein, and drug concentrations and rates of absorption from individual regions of the GI tract. In addition to novel advances in the basic design and structure of the physiological model, it was recognized that a signicant quantity of data would be required for model development. While a single corporate database may not provide a sufcient quantity of compounds to produce a robust model, a consortium of databases containing a larger number of compounds with diverse and varied characteristics was needed to supply enough data to permit development of a predictive model. Therefore, clinical pharmacokinetic (PK) results, including oral and intravenous data, and in vitro data was obtained from a consortium of pharmaceutical companies. The in vitro data included rabbit intestinal tissue permeability data, and physiochemical and pharmaceutical properties of the drugs (solubility, dissolution rate, and dose) (Table 2). The raw oral and intravenous data was analyzed by a consistent PK model in order to determine the actual percent dose absorbed into the portal vein (FDp%) as a function of time. Compounds with ,10% total metabolic clearance were selected and included in the development of the model. KineticaE (InnaPhase Corporation, Philadelphia, PA) was used to t the PK data to obtain best t plasma proles for each drug. The plasma proles were compared to the simulated FDp% and t 50 (time for 50% drug absorption) to determine the predictive capability of the model.

2. Methods

2.2. Physiological model testing 2.1. Physiological model development

A physiologically based computer simulation model of the gastrointestinal (GI) tract (IDEAE, NaviCyte, Inc., San Diego, CA), based on a previThe physiological model was tested for its ability to predict the FDp% of three drug compounds (atenolol, naproxen, and ganciclovir) in both dogs and humans. In vitro data (permeability and solu-

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62 Table 1 Permeability and solubility values for selected drugs Atenolol Pe (duodeneum) Pe (jejunum) Pe (ileum) Pe (colon) Solubility (mg / ml)
a b a

57

Ganciclovir [24]
26

Naproxen [24] 3.06310 26 9.86310 26 13.31310 26 39.37310 26 8.00 b

1.29310 3.50310 26 5.36310 26 0.62310 26 25.42 b

1.10310 2.19310 26 4.06310 26 3.80310 26 4.30 c

26

Permeability determined in rabbit tissue (cm / s). 378C, pH 7.0. c 258C, pH 7.0 [25].

bility) used in the predictions are shown in Table 1. The PK data was analyzed to determine the FDp% of all three drugs. The FDp% values from the PK analysis were compared to the values predicted using the model. For ganciclovir, a dose response relationship was simulated using the model. Oral doses from 0.06 mg / kg to 570 mg / kg in dogs and 0.014 to 143 mg / kg in humans were simulated while holding all other parameter values constant. The FDp% for each dose level was compared to in vivo and clinical bioavailability data to determine the ability of the model to predict dose response relationships. The concentration of ganciclovir in each region of the intestine was determined at three dosage levels (50, 500, 5000 mg in dogs; 10, 100, 1000 mg in humans) to determine if the concentration of ganciclovir within the intestinal regions was directly inuencing the bioavailability. Independent simulations were determined for each dosage level. The effect of solubility on the maximum dose absorbed was determined by varying the maximum solubility of ganciclovir (input data) from the actual value to innity. Using this analysis the dose dependent bioavailability was determined at each solubility level to conrm whether or not the bioavailability at higher doses was limited by the solubility in both humans and dogs.

FDp% and t 50 was tested using a representative collection of drug compounds exhibiting a wide range of permeabilities, solubilities, and doses (Table 2). The FDp% calculated from PK drug concentrations were compared to the simulated FDp% from the physiological model for humans (Fig. 1). A reasonably good correlation coefcient is reported (r 2 .0.92). Even though some variation existed the model was found to accurately distinguish between well and poorly absorbed drugs. The physiological model was also used to predict FDp% in both humans and dogs for atenolol, ganciclovir, and naproxen and the values compared to literature values for each of these compounds (Table 3). Good agreement was observed for all drugs in both species. The physiological model was also evaluated for its ability to predict a dose response curve for ganciclovir in dogs and humans, since this drug exhibits
Table 2 Relative in vitro characteristics of drug compounds Compound a1 a2 a3 a4 a5 a6 a10 b1 b2 b3 b5 b6 Permeability 1111 11 1 1 1 1111 1111 11111 1111 1 1111 1 Solubility 1111 111 1 1111 111 11 11111 11111 11 1 11 11111 Dose 11111 11111 1111 11 1111 1111 1 1 11 111 111 111

3. Results

3.1. Testing the physiological model s predictive capability

The ability of the physiological model to predict

11111 indicates greatest relative value, 1 indicates least relative value.

58

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62

Fig. 1. Comparison of FDp% calculated from plasma level time curves and FDp% simulated using the IDEAE model.

dose dependent absorption [13]. At increasing dosing levels, the model was able to simulate the decrease in FDp% as the dose approached |50 mg / kg in dogs and |0.5 mg / kg in humans (Fig. 2). To further investigate dose dependent absorption, the concentration of ganciclovir within the duodenum, jejunum, ileum and colon was simulated at three dose levels in dogs and humans (Figs. 3 and 4). At the lowest doses (Figs. 3A, 4A) (where maximum bioavailability is observed in each species), the ganciclovir concentrations remained below the solubility limit for all time points. At the intermediate doses (Figs. 3B, 4B), (approximately 50% bioavailability), the ganciclovir concentrations were at the solubility limit for a short amount of time in some regions of the intestine, and at the highest dose levels (Figs. 3C, 4C), (where bioavailability is very low), the concentrations were at the solubility limit for a considerable amount of time in all regions. The simulated concentration of ganciclovir in the colon was at the solubility limit in both species
Table 3 FDp% for atenolol, ganciclovir, and naproxen in humans and dogs Humans Simulation Atenolol Gancliclovir Naproxen 48 3 100

Fig. 2. Dose response curve FDp% for ganciclovir in dogs and humans. m, FDp% in dogs simulated using the computer model; ., FDp% in dogs determined using pharmacokinetic data, d FDp% in humans simulated using the computer model, j FDp% in humans determined using pharmacokinetic data. Mean dog weight for simulation data517.5 kg, mean human weight570 kg.

at the higher two doses for all time points and, therefore, not shown in the gures. The solubility of ganciclovir was entered as a parameter into the simulation model and was varied between the actual solubility values and innity. A plot of mass of ganciclovir absorbed (mg) versus dose (mg / kg) was constructed based on varying solubility (Figs. 5 and 6). For humans and dogs, a maximum mass absorbed was observed when the actual solubility was entered in the simulation. An increase in the maximum mass absorbed was observed as the entered solubility values were increased. At innite solubility, the dose dependent absorption was no longer observed. For humans, an approximately .300-fold increase in solubility was necessary to eliminate dose dependent absorption. For dogs, only a .30-fold increase was necessary.

Dogs Literature 50 [26] 4 [21,26,28] 95 [26] Simulation 87 35 75 Literature 83 [27] 35 70 [29]

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62

59

Fig. 3. The concentration of ganciclovir in the duodenum, jejunum, and ileum of dogs after simulation of three dosage levels. (A) 50 mg dose, (B) 500 mg dose, (C) 5000 mg dose. () duodenum, (?????) jejunum, (- - -) ileum.

Fig. 4. The concentration of ganciclovir in the duodenum, jejunum, and ileum of humans after simulation of three dosage levels. (A) 10 mg dose, (B) 100 mg dose, (C) 1000 mg dose. () duodenum, (?????) jejunum, (- - -) ileum.

60

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62

Fig. 5. Simulations demonstrating the effect of increased solubility on the mass of ganciclovir absorbed in dogs. The amount absorbed for increasing dosing is shown for the actual solubility (m), solubility 33 (.), solubility 330 (d), and innite solubility (j).

4. Discussion Previous efforts to predict oral absorption in humans have used uptake and transport studies to calculate a permeability through tissue or cell monolayers [3,5] alone or in conjunction with other physiological or system parameters to predict extent of absorption. Historically, the methods have utilized a two tank perfect mixing model [3] and / or macroscopic mass balance approach [5]. More recently physiologically based models have been described in the literature [13,14]. The predictive capabilities of

Fig. 6. Simulations demonstrating the effect of increased solubility on the mass of ganciclovir absorbed in humans. The amount absorbed for increasing dosing is shown for the actual solubility (m), solubility 330 (.), solubility 3300 (d), and innite solubility (j).

these earlier models, however, have been limited by their basic design model structure approaches, as well as by the amount and diversity of compound data employed to develop, validate and test the models. Another approach to predicting oral absorption involves the development of in vitro to in vivo correlations (IVIVC) based on dissolution rates. Numerous methods have been described recently to build these correlations [79,15,16]. Unfortunately, a dissolution based approach considers only the release of drug from the dosage form and disregards the transport of drugs across the intestinal mucosa. For drugs in Class III (high solubility, low permeabilty) of the Biopharmaceutical Classication Scheme (BCS) [17], the permeability may be the rate-limiting factor in absorption and therefore must be considered. Since this model strives to simulate absorption for all drugs, a dissolution based model would have been insufcient to simulate permeability limited absorption. In building IDEAE, drug compounds with a wide range of permeabilities, solubilities, and doses were used (Table 2) so that condent determinations could be made when testing unknown compounds. Access to the most diverse drug database was made more feasible by enrolling a consortium of companies having existing databases of different compounds. This permitted a more robust model to be constructed more quickly since each member contributed a set of compounds unique from those submitted by other members. Also, previous work to predict fraction dose absorbed have correlated a single parameter, such as absorption potential [3], intestinal effective permeability [5] or Caco-2 cell monolayer permeabilities [4] to the fraction dose absorbed. In contrast, the physiological model we developed has been exploited to determine the rate of absorption, and then calculate FDp%. Therefore, a direct comparison between the fraction dose absorbed in humans, as calculated from PK data, to the FDp% from the simulation model is performed. Although this relationship should be linear with a slope of |1.0, a reasonably high regression coefcient was achieved using a representative collection of compounds, and the physiological model was able to correctly classify compounds as well or poorly

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62

61

absorbed drugs. The absence of complete linearity in predicting FDp% was likely due to the low number of compounds used for the development of the initial model. Table 3, which shows the extent of absorption for three representative compounds, reinforces the ability of the model to predict FDp% for both well and poorly absorbed compounds. Ganciclovir, an acyclic nucleoside analogue, is a selective antiviral agent which has shown activity against herpes simplex viruses I an II, cytomegalovirus, varicella zoster, and Epstein Barr virus [18,19]. It is approved by the FDA for use in the treatment of cytomegalovirus in immunocompromised patients. Due to the poor bioavailability of ganciclovir (,10%) [20,21], high doses (1000 mg three times a day, or 500 mg six times a day) are required to achieve acceptable plasma levels in humans. It has been speculated that the poor bioavailability is due to its lipophilicity and membrane partitioning, as evidenced by an octanol-to-water partition coefcient of 0.022 [22,23]. The poor bioavailability and dose dependent absorption in both humans and dogs [13] made ganciclovir a good model drug to test the physiological models capabilities. As shown in Fig. 2, a decrease in FDp% was observed for dogs at approximately 50 mg / kg, and the model accurately predicted the decrease in FDp% as the dose was increased. Additionally, at |10 mg / kg the observed FDp% in humans is ,10% [20,21], and the model was able to predict the FDp% at this level accurately. Given the predictive capability of the physiological model, the model was also used to determine possible factors responsible for ganciclovirs low FDp%. These simulations support the concept of solubility limited ux. As the dose is increased, the concentration reaches saturation for longer periods of time and decreases the amount of drug that can be absorbed based on the permeability (Figs. 24). Figs. 5 and 6 show how solubility limits the mass absorbed. At the actual solubility, the maximum mass of drug that can be absorbed is achieved at the highest doses shown. By articially increasing the solubility the total mass absorbed can be increased to an amount directly proportional and linear to the dose. Therefore, it is speculated that the low bioavailability of ganciclovir is attributed not to the poor partitioning of the drug into membranes, but

rather the high quantities of drug required to achieve therapeutic effectiveness and the low solubility of ganciclovir. In summary, a physiologically based computer model to simulate oral absorption in mammals was developed and tested. The model was found to accurately respond to changes in permeability, surface area, transit time, solubility, and dissolution rate. When tested with a set3 of drugs, the model was able to generate FDp% proles similar to that observed in vivo in humans, and was able to simulate oral bioavailability for both well and poorly absorbed compounds in both humans and dogs. This technology provides a breakthrough in in silico prediction of absorption and with its continued development and improvement, will aid drug discovery and drug development scientists to produce better pharmaceutical products.

References
[1] R.A. Prentis, Y. Lis, S.R. Walker, Pharmaceutical innovation by the seven UK-owned pharmaceutical companies (1964 1985), Br. J. Clin. Pharmacol. 25 (1988) 387396. [2] T. Kennedy, Managing the drug discovery / development interface, DDT 2 (1997) 436444. [3] J.B. Dressman, G.L. Amidon, D. Fleisher, Absorption potential: estimating the fraction absorbed for orally administered compounds, J. Pham. Sci. 74 (1985) 588589. [4] P. Artursson, J. Karlsson, Correlation between oral drug absorption in humans and apparent drug permeability coefcients in human intestinal epithelial (Caco-2) cells, Biochem. Biophys. Res Commun. 175 (1991) 880885. [5] G.L. Amidon, P. Sinko, D. Fleisher, Estimating human oral fraction dose absorbed: a correlation using rat intestinal membrane permeability for passive and carrier-mediated compounds, Pharm. Res. 5 (1988) 651654. [6] J.E. Polli, J.R. Crison, G.L. Amidon, Novel approach to the analysis of in vitroin vivo relationships [published erratum appears in J Pharm Sci 1997 Feb;86(2):268], J. Pharm. Sci. 85 (1996) 753760. [7] K. Ishii, Y. Katayama, S. Itai, Y. Ito, H. Hayashi, In vitro dissolution tests corresponding to the in vivo dissolution of clarithromycin tablets in the stomach and intestine, Chem. Pharm. Bull. (Tokyo) 43 (1995) 19431948. [8] J.B. Dressman, D. Fleisher, Mixing-tank model for predicting dissolution rate control of oral absorption, J. 75 (2) (1986) 109109. [9] A. Dunne, T. OHara, J. Devane, A new approach to modelling the relationship between in vitro and in vivo drug dissolution / absorption [In Process Citation], Stat. Med. 18 (1999) 18651876.

62

D. A. Norris et al. / Journal of Controlled Release 65 (2000) 55 62 mg / d dosing regimens in human immunodeciency virusand cytomegalovirus-seropositive patients, Clin. Ther. 17 (1995) 425432. S.A. Spector, D.F. Busch, S. Follansbee, K. Squires, J.P. Lalezari, M.A. Jacobson, J.D. Connor, D. Jung, A. Shadman, B. Mastre, Pharmacokinetic, safety, and antiviral proles of oral ganciclovir in persons infected with human immunodeciency virus: a phase I / II study. AIDS Clinical Trials Group, and Cytomegalovirus Cooperative Study Group, J. Infect. Dis. 171 (1995) 14311437. J. Lavelle, S. Follansbee, C.B. Trapnell, W.C. Buhles, K.G. Griffy, D. Jung, A. Dorr, J. Connor, Effect of food on the relative bioavailability of oral ganciclovir, J. Clin. Pharmacol. 36 (1996) 238241. E.J. Benjamin, B.A. Firestone, R. Bergstrom, M. Fass, I. Massey, I. Tsina, Y.Y. Lin, Selection of a derivative of the antiviral agent 9-[(1,3-dihydroxy-2- propoxy)-methyl]guanine (DHPG) with improved oral absorption, Pharm. Res. 4 (1987) 120125. N. Jezyk, W. Rubas, G.M. Grass, Permeability characteristics of various intestinal regions of rabbit, dog, and monkey, Pharm. Res. 9 (1992) 15801586. S. Budavari, M.J. ONeil, A. Smith, P.E. Heckelman, J.F. Kinneary (Eds.), 12th Edition, The Merck Index, Merck & Co., Inc, Whitehouse Station, NJ, 1996. Physicians Desk Reference, 52nd Edition, Medical Economics Company, Inc, 1998. J. McAinsh, B.F. Holmes, Pharmacokinetic studies with atenolol in the dog, Biopharm. Drug Dispos. 4 (1983) 249 261. R.D. Anderson, K.G. Griffy, D. Jung, A. Dorr, J.D. Hulse, R.B. Smith, Ganciclovir absolute bioavailability and steadystate pharmacokinetics after oral administration of two 3000mg / d dosing regimens in human immunodeciency virusand cytomegalovirus-seropositive patients, Clin. Ther. 17 (1995) 425432. H.H. Frey, B. Rieh, Pharmacokinetics of naproxen in the dog, Am. J. Vet. Res. 42 (1981) 16151617.

[10] M.J. Ginski, J.E. Polli, Prediction of dissolutionabsorption relationships from a dissolution / Caco-2 system, Int. J. Pharm. 177 (1999) 117125. [11] J.B. Dressman, G.L. Amidon, C. Reppas, V.P. Shah, Dissolution testing as a prognostic tool for oral drug absorption: immediate release dosage forms, Pharm. Res. 15 (1998) 1122. [12] W.K. Sietsema, The absolute oral bioavailability of selected drugs, Int. J. Clin. Pharmacol. Ther. Toxicol. 27 (1989) 179211. [13] G.M. Grass, Simulation models to predict oral drug absorption from in vitro data, Adv. Drug Delivery Rev. 23 (1997) 199219. [14] L.X. Yu, G.L. Amidon, Saturable small intestinal drug absorption in humans: modeling and interpretation of cefatrizine data, Eur. J. Pharm. Biopharm. 45 (1998) 199203. [15] W.R. Gillespie, Convolution-based approaches for in vivoin vitro correlation modeling, Adv. Exp. Med. Biol. 423 (1997) 5365. [16] L.C. Kaus, W.R. Gillespie, A.S. Hussain, G.L. Amidon, The effect of in vivo dissolution, gastric emptying rate, and intestinal transit time on the peak concentration and areaunder-the- curve of drugs with different gastrointestinal permeabilities, Pharm. Res. 16 (1999) 272280. [17] G.L. Amidon, H. Lennernas, V.P. Shah, J.R. Crison, A theoretical basis for a biopharmaceutic drug classication: the correlation of in vitro drug product dissolution and in vivo bioavailability, Pharm. Res. 12 (1995) 413420. [18] J.C. Martin, C.A. Dvorak, D.F. Smee, T.R. Matthews, J.P. Verheyden, 9-[(1,3-Dihydroxy-2-propoxy)methyl]guanine: a new potent and selective antiherpes agent, J Med. Chem. 26 (1983) 759761. [19] D.F. Smee, J.C. Martin, J.P. Verheyden, T.R. Matthews, Anti-herpesvirus activity of the acyclic nucleoside 9-(1,3dihydroxy-2- propoxymethyl)guanine, Antimicrob. Agents Chemother. 23 (1983) 676682. [20] R.D. Anderson, K.G. Griffy, D. Jung, A. Dorr, J.D. Hulse, R.B. Smith, Ganciclovir absolute bioavailability and steadystate pharmacokinetics after oral administration of two 3000-

[21]

[22]

[23]

[24]

[25]

[26] [27]

[28]

[29]