comparative assessment of empirical and physiological approaches on predicting human clearances

Comparative Assessment of Empirical and PhysiologicalApproaches on Predicting Human Clearances

SEKIHIRO TAMAKI,1,2 HIROSHI KOMURA,3 MOTOHIRO KOGAYU,3 SHIZUO YAMADA1

1Department of Pharmacokinetics and Pharmacodynamics and Global Center of Excellence (COE) Program, University of Shizuoka,52-1 Yada, Suruga-ku, Shizuoka 422-8526, Japan

2Clinical Pharmacology Team, Clinical Research Planning Department, Pharmaceutical Division, JAPAN TOBACCO INC. 2-1,Toranomon 2-chome, Minato-ku, Tokyo 105-8422, Japan

3Drug Metabolism & Pharmacokinetics Research Laboratories, Central Pharmaceutical Research Institute, JAPAN TOBACCO INC.1-1 Murasaki-cho Takatsuki, Osaka 569-1125, Japan

Received 8 March 2010; revised 7 June 2010; accepted 7 July 2010

Published online 9 September 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/jps.22321

ABSTRACT: The empirical and physiological predictive approaches to human clearance wereevaluated using preclinical in vitro and in vivo data of various datasets to establish a method-ology for the prediction of clearance. Among the examined empirical approaches, an allometricscaling method with the rule of exponent (ROE), based on the exponent in simple allometry,provided better prediction. The effect of lipophilicity (clog P) and clearance on the predictivitywas investigated using the ROE method. High predictivity was found for a low lipophilic com-pound with clog P < 0 and for a compound with moderate or high clearance. As a physiologicalapproach, the in vitro–in vivo scaling method using metabolic stability in liver microsomes andhepatocytes was evaluated, and the predictivity taking the plasma protein binding and the non-specific binding in incubation into consideration was compared with the ROE method. The twomethods appeared to show comparable predictivity, although the in vitro–in vivo scaling wasconducted under limited conditions like the use of physiological scaling factor and lipophilicity-derived nonspecific binding data. The ROE method could be an alternative predictor of thehuman clearance of compounds to which a physiological approach cannot be applied, in addi-tion to low lipophilic compounds, with acceptable accuracy. © 2010 Wiley-Liss, Inc. and theAmerican Pharmacists Association J Pharm Sci 100:1147–1155, 2011Keywords: clearance; metabolism; Log P; preclinical pharmacokinetics; clinical pharmacoki-netics

INTRODUCTION

The prediction of the plasma concentration profile anddose regimen of humans in preclinical pharmacoki-netic studies has given an important position to theselection of a new molecular entity (NME) or to designof first-in-human studies.1 Recently, a more accurateestimation of human clearance (CL) has increasinglybeen required with the advance of the drug develop-

Abbreviations used: BrW, brain weight; CL, clearance;LBF, liver blood flow; MLP, maximum life-span potential; CLint,metabolic intrinsic clearance; MPE, mean of percentage error; CL/F, oral clearance; ROE, rule of exponent; PE, percentage error; SA,simple allometry.

Correspondence to: Hiroshi Komura (Telephone: +81-72-681-3311; Fax: +81-72-681-9859; E-mail: [email protected])Journal of Pharmaceutical Sciences, Vol. 100, 1147–1155 (2011)© 2010 Wiley-Liss, Inc. and the American Pharmacists Association

ment stage.2In vitro systems using liver microsomeshave routinely been utilized to evaluate metabolic in-trinsic clearance (CLint) in preclinical pharmacoki-netic studies.3 To predict the CL in humans basedon in vitro–in vivo extrapolation, the CLint obtainedfrom human liver microsomes has been applied tophysiological models such as the well-stirred model,parallel tube model, or dispersion model.4 There hasbeen reportedly a tendency for the predicted CLto be underestimated when the predicted and ob-served CLs are retrospectively compared.5–8 Some ef-forts have been made to improve the predictivity ofCL by considering unbound fractions under in vitroincubation conditions besides plasma protein bind-ing and some scaling factors of CLint from per mil-ligram microsomal protein to kilogram basis, and/orapplying a hepatocyte system with integrated drug

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 3, MARCH 2011 1147

1148 TAMAKI ET AL.

metabolizing enzymes.4,9–11 Although the predictivityhas improved somewhat, the reason for this underes-timation still remains unclear. It has recently been re-vealed that influx and/or efflux transporters display acrucial role in the pharmacokinetic behavior of drugs,and retrospective evaluations have focused on the roleof hepatic uptake in CL.12,13 The prediction incorpo-rating hepatic uptake in intrinsic CL has not beenconducted intensively using a variety of compounds.

One of the best described techniques for predictingCL, alongside the in vitro–in vivo scaling above, isallometric scaling, in which many physiological pro-cesses (blood flow, creatinine CL, heart rate, glomeru-lar filtration, etc.) and organ sizes are defined bya power–law relationship with the body weight ofspecies.14,15 Extrapolation of CL from animal speciesto humans by allometric scaling has widely been usedbecause of its simplicity.14,15 Several methods havebeen developed; that is, simple allometry (SA), maxi-mum life-span potential (MLP), brain weight (BrW),and rule of exponent (ROE) approaches associatedwith the selection of SA, MLP, or BrW based on theexponent of SA.14,15 Moreover, Ward and Smith devel-oped the liver blood flow (LBF) method in which CLin humans is predicted based on the ratio of humanand animal LBF rate.16 The monkey LBF methodprovided the best accuracy among the examined ap-proaches including allometric scalings.16,17

A major drawback of allometric scaling is its empir-ical nature, but this can, in some cases, be regardedas an advantage. Allometric scaling does not allow forclarification of physiological mechanisms underlyingthe elimination from the body and understanding ofspecies differences in the elimination mechanism.18

In the early phase of drug development, it is diffi-cult to clarify the elimination mechanisms by otherprocesses than typical hepatic metabolism processeslike oxidation. In contrast to in vitro–in vivo scalingwhich is preferentially selected for CYP-dependenteliminated compounds, allometric scaling must be anindispensable predictor of human CL for compoundsthat are eliminated via unknown or multiple mecha-nisms from the body.18 Comparative studies of predic-tivity between allometric and in vitro–in vivo scalinggive insight into making a strategy for the predic-tion of human CL. Only a limited number of reportshave been published by several authors.9,19–21 How-ever, there seemed to be some issues in those studies;that is, the ROE method, which is a widely acceptableapproach, was not employed, and/or the comparisonwas performed based on a dataset of limited size.

Since physicochemical properties are known togovern pharmacokinetic features, we first investi-gated the effect of lipophilicity (clog P), in addi-tion to the classification of CL (low, moderate, high),on the predictivity of allometric scaling, to clarifyhow to adapt allometric scaling for the prediction

strategy of human CL. The evaluated accuracy of al-lometric scaling was compared with the data of invitro–in vivo scaling, including plasma protein bind-ing and microsomal binding during the incubationcondition with liver microsomes or hepatocytes. Basedon our findings through the comparative assessment,the simply described decision tree was developedtaking physicochemical characteristics of compoundsinto consideration.

MATERIALS AND METHODS

Data Collection

In the assessment of allometric scaling, 68 compoundsthat were available for intravenous pharmacokineticparameters in at least three animal species were col-lected from the literature. The dataset was composedof compounds that were eliminated from the body viavarious routes such as hepatic metabolism, and uri-nary and/or biliary excretion. In the assessment of invitro–in vivo scaling, in vitro metabolic parameters of54 and 77 compounds for liver microsomal and hepa-tocyte incubation, respectively, together with in vivopharmacokinetic data, were collected. The profiles ofmolecular weight, clog P, and CL of three differentdatasets (for allometric scaling, in vitro–in vivo scal-ing with liver microsomes and hepatocytes) seemedsimilar to those of marketed oral drugs.22,23 The com-pounds used in the present study are listed in theAppendices (see Supplementary Material).

The clog P and pH-dependent measure of lipophilic-ity (clog D) of test compounds were calculated usingPallas for Windows, Version 3.1TM (Infocom Corpora-tion, Tokyo, Japan). The compounds were divided intoacidic, neutral and basic classes based on the follow-ing equation:

�c log D = c log D6.5 − c log D7.4 (1)

where �clog D expresses the difference betweenclog D at two given pH values, pH 6.5 and 7.4.The compounds with positive and negative values of�log D were classified as the acidic and basic form,respectively.24

Four Different Allometric Scaling Methods

Prediction of human CL in allometric scaling was per-formed by the following four different methods.18 Co-efficients and exponents with the SA, MLP, and BrWmethod were obtained by fitting body weight and CL,CL × MLP or CL × MLP on a log–log scale relation-ship. The CL in humans was estimated by the substi-tution of the reported human body weight or 70 kg (ifit was not reported) to each equation to compare withthe observed human CL:

JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 3, MARCH 2011 DOI 10.1002/jps

COMPARISON OF ROE AND IN VITRO–IN VIVO SCALING 1149

(i) SA Method

CL = aWb (2)

where W is body weight, and a and b are the co-efficient and an exponent of the allometric equa-tion, respectively.

(ii) MLP Method

CL × MLP = aWb (3)

MLP can be calculated from the following equa-tion as described by Sacher25:

MLP (years) = 185.4 (BrW)0.636 (W)−0.225 (4)

(iii) BrW Method

CL × BrW = aWb (5)

(iv) ROE Method

ROE method was performed according to the fol-lowing rule reported by Mahmood and Balian.,14,15

Briefly, if the exponent of SA was within 0.55 to 0.70,SA was used; if the exponent of SA was between 0.71and 0.99, the MLP method was employed; and if theexponent of SA was ≥ 1.0, the BrW method was em-ployed. If the exponent of SA was below 0.55 or greaterthan 1.3, prediction using allometric scaling was notconducted.

LIVER BLOOD FLOW METHOD

The evaluation of LBF method utilized the samedataset as allometric scaling. The prediction by LBFmethod was performed for 29 compounds in mice,68 compounds in rats, 65 compounds in dogs, and40 compounds in monkeys. Assuming that CL is pri-mary hepatic and the blood to plasma ratio (RB) isconstant across species, CL can be expressed in eachof the preclinical animal species as a fraction of LBF,and the human CL was estimated using the followingEq. (6),

16:

CL = CLanimal × LBFhuman

LBFanimal(6)

In Vitro–In Vivo Scaling

The unit of in vitroCLint (:L/min/mg of protein forliver microsomes, :L/min/106 cells for hepatocytes)was converted to the unit of mL/min/kg using scal-ing factors. As scaling factors, 48.8 mg of protein pergram liver and 120 × 106 cells per gram liver werebasically used for liver microsomes and hepatocytes,

respectively, with 25.7 g liver per kilogram of bodyweight.26–28 The CL was calculated based on the well-stirred model (Eq. 7), because the predictability fromthe well-stirred model was reported to be within anacceptable range when compared with the other twomodels.4

CLh = QH × fub × (CLint/fuinc)QH + fub × (CLint/fuinc)

(7)

where QH is hepatic blood flow (20.7 mL/min/kg), fubis unbound fraction of protein binding in blood, fuincis unbound fraction of microsomal and hepatocyteincubation.29 The fub was obtained from the litera-ture or by dividing the reported unbound fraction inplasma protein binding (fup) by RB (= fup/RB). Thefuinc was calculated by substitution of either log D7.4or log P to Eq. (8) for liver microsomes and to Eq. (9)for hepatocytes.30,31

log(

1 − fuinc

fuinc

)= 0.53logP

/D7.4 − 1.42 (8)

log(

1 − fuinc

fuinc

)= 0.40logP

/D7.4 − 1.38 (9)

Statistical Analysis

The percentage error (PE) between the observed andpredicted human CL for each compound was calcu-lated using Eq. (10) for over-prediction and Eq. (11)for under-prediction.

PE (%) =(CLpredicted − CLobserved

)CLobserved

× 100 (10)

PE (%) =(CLobserved − CLpredicted

)CLpredicted

× 100 (11)

The predictivity of human CL was basically evalu-ated based on the mean (MPE) of PE values estimatedusing Eqs. (10 and 11).

RESULTS

Dataset for Allometric and In Vitro–In Vivo ScalingStudies

The distributions of clog P for the dataset in allo-metric scaling and in vitro–in vivo scaling with livermicrosomes and hepatocytes are shown in Figure 1,because lipophilicity is one of the major determinantfactors of the pharmacokinetic behavior of drugs. Thedistribution of the employed human CL is also shown

DOI 10.1002/jps JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 100, NO. 3, MARCH 2011

1150 TAMAKI ET AL.

Table 1. Mean and Range of clog P and CL for Dataset in Allometric and In Vitro–In Vivo Scaling Method

clog P CL (mL/min/kg)

Mean Range Mean Range

Allometric scalingAll 2.10 ± 2.18 −2.14 to 7.82 5.40 ± 5.38 0.057 to 22.0clog P ≤ 0 −0.90 ± 0.54 −2.14 to −0.09 2.54 ± 2.68 0.277 to 8.17clog P > 0 2.95 ± 1.65 0.14 to 7.82 6.22 ± 5.69 0.057 to 22.0In vitro–in vivo scalingLiver microsomes 2.83 ± 1.27 −0.71 to 5.59 9.02 ± 6.61 0.030 to 20.4Hepatocytes 2.64 ± 1.64 −1.82 to 7.35 6.92 ± 5.82 0.070 to 20.4

in Figure 1. The mean and range of clog P, in addi-tion to the observed CL, for each dataset are summa-rized in Table 1. The mean value of clog P was 2.1for allometric scaling, 2.8 and 2.6 for in vitro–in vivoscaling with liver microsomes and hepatocytes, andthe average value in allometric scaling was slightlylower than that in two in vitro–in vivo scalings. TheCL for the dataset in allometric scaling tended to be

lower than that of two in vitro–in vivo scalings. Basi-cally, predictivity from the two in vitro–in vivo scalingmethods needed to be compared with that from theallometric scaling method using compounds metabol-ically cleared. Since a moderate to high lipophiliccompounds (clog P > 0) possibly undergo hepaticmetabolism, the mean value of CL was obtained sep-arately for the dataset with clog P > 0 and clog

Figure 1. Distribution of clog P and observed human CL on the dataset for predictions. Panela–c refer to the distribution of clog P and d–f refer to observed human CL. Panel a and d includethe distribution of clog P and CL on allometric scaling; b and e, in vitro–in vivo scaling frommicrosomal data; c and f, in vitro–in vivo scaling from hepatocytes data.



Figure 2. Relationship between the observed and predicted CL for 68 dataset from allometricscaling with the SA (a), MLP (b), BrW (c), or ROE method (d). The solid lines are the lines ofidentity and the dashed lines represent a range associated with twofold error.

P ≤ 0. The dataset with clog P > 0 in allometric scal-ing showed a comparable value to the correspondingdata for hepatocytes.

Four Different Allometric Scaling Methods

Human CLs for 68 compounds were predicted us-ing the different allometric scaling methods with SA,MLP, BrW and ROE, and the relationship betweenthe predicted and observed human CLs is shown inFigure 2. The respective MPE values were estimatedto be 419, 297, 463 and 278% (Table 2). In addition,the percentage of compounds within twofold errorbased on the ratio of the predicted to observed CL washigh (58.3%) in the ROE method compared with otherapproaches. The ROE method provided the most ac-curate prediction among the investigated allometricscaling methods. Therefore, to evaluate the effect ofphysicochemical properties or human CL of test com-pounds on the prediction in the following analyses,the ROE method was employed along with the SAmethod.

LBF Method

Human CLs were predicted from the pharmacokineticdata of either mice, rats, dogs, or monkeys based onthe LBF method (Table 2). The accuracy of the pre-dicted CLs from mouse and monkey LBF methods wasbetter than that from rat and dog LBF methods.

Table 2. Mean of Percentage Error Values for Simple Allometry(SA), Maximum Life-Span Potential (MLP), Brain Weight (BrW),Rule of Exponent (ROE), and Liver Blood Flow (LBF) Method

Mean of Percentage Error (%)

SA 419(47.1)a (n = 68)MLP 297(39.7) (n = 68)BrW 463(23.5) (n = 68)ROE 278(58.3) (n = 60)LBF Mice 518(31.0) (n = 29)

Rats 1186(26.4) (n = 68)Dogs 856(50.8) (n = 65)

Monkeys 669(57.5) (n = 40)

aThe number in parenthesis is percentage within twofold error of thepredicted clearance to the observed clearance.

Effect of CL and Physicochemical Properties onPrediction

The effect of CL on the predictivity was investigated.The test compounds with CL ≥ 6.21 mL/min/kg, whichaccounted for EH ≥ 0.3 in the case of metabolic elim-ination by the liver, showed better prediction thanthose with CL < 6.21 mL/min/kg (Table 3). In the in-vestigation of the effect of physicochemical propertieson the predictivity, the test compounds were classifiedinto low (clog P ≤ 0) and moderate to high lipophilic(clog P > 0) compounds, or into acidic, neutral, andbasic compounds. The MPE values are shown inTable 3. The low lipophilic compounds exhibited ac-curately predicted CLs with a MPE of 87% comparedwith the high lipophilic compounds. Furthermore, theestimated CL seemed to be accurate for acidic andneutral rather than basic compounds.

In Vitro–In Vivo Scaling

The human CLs were predicted from CLint valuesfor 54 and 77 test compounds in liver microsomaland hepatocyte incubation, respectively, based onthe well-stirred model including both fub and fuinc.The relationship between the observed and predictedCLs and the MPE value are shown in Figure 3 and

Figure 3. Relationships between the observed and pre-dicted CL from the in vitro–in vivo scaling method withmicrosomal and hepatocyte data incorporating both fub andfuinc; Panel a refers to data with microsomal data and b, hep-atocytes data. Symbols depict different charge of dataset; �,acid; �, neutral; and ◦, base. The solid lines are the lines ofidentity and the dashed lines represent a range associatedwith twofold error.


1152 TAMAKI ET AL.

Table 3. Effect of CL and Physicochemical Properties on Prediction in Simple Allometry and Rule ofExponent Method

Mean of Percentage Error (%)

SA ROE

All 419 (47.1)a (n = 68) 278 (58.3) (n = 60)CL

<6.21 (mL/min/kg) 588 (53.7) (n = 41) 385 (39.4) (n = 33)≥6.21 (mL/min/kg) 163 (37.0) (n = 27) 148 (81.5) (n = 27)

Lipophilicityclog P ≤ 0 107 (80.0) (n = 15) 87 (69.2) (n = 13)clog P > 0 507 (37.7) (n = 53) 331 (55.3) (n = 47)

ChargeAcid 455 (68.2) (n = 22) 223 (52.6) (n = 19)

Neutral 212 (40.0) (n = 15) 147 (69.2) (n = 13)Base 494 (35.5) (n = 31) 377 (57.1) (n = 28)

aThe number in parenthesis is percentage within twofold error of the predicted clearance to the observed clearance.SA, simple allometry; ROE, rule of exponent; CL, clearance.

Table 4. Mean of Percentage Error Values for Rule of ExponentMethod and In Vitro–In Vivo Scaling Method Using LiverMicrosomes and Hepatocytes

Allometricscaling In Vitro–In Vivo scaling

ROE Liver Microsomes Hepatocytes

Mean of 278 (58.3)a 400 (53.7) 315 (45.5)percentage

error %(n = 60) (n = 54) (n = 77)

aThe number in parenthesis is percentage within twofold error of thepredicted clearance to the observed clearance.

ROE, rule of exponent.

Table 4, respectively. Both the in vitro–in vivo scal-ings yielded high predictivity with a relatively lowMPE value of 400 and 315%. The percentage of com-pounds within the twofold error was 53.7 and 45.5%for the prediction from liver microsomes and hepato-cytes, respectively.

DISCUSSION

We conducted a comparative study of four allomet-ric scaling methods (SA, MLP, BrW, and ROE) usingdatasets with a wide range of physicochemical prop-erties and CL, and the ROE method showed the high-est predictivity among the four scaling methods, asproved by Mahmood and Balian.14,15 Additionally, theaccuracy of the ROE method was higher than that ofthe LBF method in various animal species, which wasdifferent from the results of Ward and Smith that themonkey LBF method gave a better prediction com-pared with the SA method using rats, dogs and/ormonkeys.16 However, when the predictability of theROE method using two animal species such as rats/dogs and rats/monkeys was compared with that of the

monkey LBF method, relatively high accuracy in ROEmethod was also found (data not shown).

The number of animal species used for allometricscaling would reportedly be one of the important fac-tors of predictivity. In addition to the LBF method,Hosea et al. demonstrated that single species allom-etry with rats, dogs, or monkeys was as accurate asor more accurate than multiple species allometry.32

Interspecies prediction study with one or two speciesconducted by Tang et al. implied that the use of threeor more species in allometric scaling was not neces-sary for the prediction of human CL.33 On the con-trary, the usefulness of multiple species scaling forprediction was demonstrated by some literatures,34,35

and in particular, Goteti et al. pointed out that theprediction from more than three species was far moreaccurate than two-species scaling.34 In the presentstudy, the ROE method, multiple species allome-try, exhibited better predictivity than single speciesallometry.

Shinha et al. evaluated four different allometricapproaches, unbound oral clearance (CL/F) and un-bound fraction-corrected intercept method (FCIM),using ratio between rat and human plasma proteinbinding in addition to SA and ROE methods.36 Theanalysis indicated that the unbound CL/F approachcombined with ROE based on exponents of SA isthe most preferred method, and in the case thatthe unbound CL/F method is inapplicable, the FCIMshould be chosen. Noted that the dataset includedcompounds that showed a large species difference inthe plasma protein binding. Hence, to identify themost accurate prediction tool, further assessments inthose approaches using the same dataset would berequired.

Using the ROE method, the effect of three classi-fications (high, moderate, and low) of CL and clog Pon the predictivity was investigated. The predictedCLs for compounds that were moderately to highly



cleared with a CL > 6.21 mL/min/kg exhibited moreaccuracy. The high predictivity rested on the fact thatthe elimination of high CL compounds depends onthe hepatic blood flow rate, which is well definedby the allometry concept. For a low lipophilic com-pound with clog P ≤ 0, better prediction was notedcompared with the moderate to high lipophilic com-pounds in the ROE method. The low lipophilic com-pound is well known to undergo renal excretion of theunchanged form rather than hepatic metabolism, asVarma et al. have shown.37 The interspecies scalingbased on the allometry concept has been successfullyused to predict pharmacokinetic parameters relatedto renal CL, including glomerular filtration, from an-imals to human.38 Additionally, allometric scaling ofthe renal CLs would reportedly be a good predictor fordrugs such as methotrexate and several $-lactam an-tibiotics, which are recently regarded as substratestoward renal efflux transporters.39,40 Regardingfamotidine, an H2 receptor antagonist, Tahara et al.implied that renal and renal tubular secretion CLs inhumans were well predicted by allometric scaling us-ing data from rats, dogs, and monkeys.41 Therefore,scaling is probably a useful predictive tool for somecompounds that are subjected to glomerular filtrationand/or renal tubular secretion. Considering that mostof those substrates to efflux transporters tend to bemildly lipophilic, possibly with clog P ≤ 0, one of ourimportant findings is that clog P ≤ 0 is an indica-tive parameter of preferentially applied allometricanalysis.

The advantage of the physiological approach overthe empirical approach is mechanism-based methodthat can be applied to an advanced physiologi-cally based pharmacokinetic model for describingplasma concentration profile and furthermore, thisconcept is adaptable for the prediction of drug—drug interaction. We investigated the in vitro–in vivoextrapolation using CLint in liver microsomes andhepatocytes. As reported by Riley et al.,42 consider-ing both binding parameters in plasma and incuba-tion condition in the extrapolation improved the pre-dictivity (data not shown), and the MPE value was400% for liver microsomes and 315% for hepatocytes.However, the predicted values still tended to be un-derestimated. For comparison with these MPE valuesin in vitro metabolic data-based prediction, allometricscaling was conducted using compounds with clog P >

0, which would mainly undergo metabolism, and theMPE value was estimated to be 331%. It is worth not-ing that the in vitro–in vivo scaling methods takingthe two binding parameters into consideration andthe ROE method showed similar accuracy. Moreover,similar variation profile in PE values for the ROEand in vitro–in vivo scaling with liver microsome orhepatocyte data was observed, as shown in Figure 4.However, attention should be paid to the analytical

Figure 4. Percentage error of the predicted clearance forthe rule of exponent and in vitro–in vivo scaling methodswith microsome and hepatocyte data. The bar representsthe mean of percentage error values.

background that this outcome was not derived fromthe same dataset.

Extrapolation uses scaling factors based on hepaticmicrosomal recovery or hepatocyte recovery from theentire liver. Ito and Houston reported that the useof an empirical scaling factor rather than a physio-logical scaling factor appears to be the best method.9

However, since the present study aimed to comparethe empirical approach, which is allometric scaling,with a physiological approach, which is in vitro–in vivo scaling, widely used physiological scaling fac-tors were employed.

Obach et al. comparatively evaluated the accuracyof human CLs predicted from in vitro microsomalstudies with the two binding parameters and fromallometric scaling.19 Average error folds expressedby predicted/observed values were within a similarrange (1.95 vs. 1.91). However, SA/MLP, but not theROE method, was employed as allometric scaling. Acomparative study of the two approaches using 16marketed drugs by Mahmood indicated that the ROEmethod provided much greater accuracy than thein vitro–in vivo scaling method with in vitro micro-somal data, but the two binding parameters were notincluded in the in vitro-based scaling method.43 Itoand Houston demonstrated that the prediction fromin vitro data using an empirical scaling factor wouldbe the best method.9 However, the use of a physi-ological scaling factor gave similar accuracy to SAanalysis, which was conducted by using only rat data.Importantly, the present study compared the predic-tivity of the ROE method and the in vitro–in vivoscaling method, including the two binding parame-ters, which would be more reliable in the empiricaland the physiological approach, respectively. A simi-lar accuracy with both approaches would enable us tohave some options for selecting the prediction method.


1154 TAMAKI ET AL.

A similar result was also described in a report ofShiran et al., who evaluated the prediction of humanCLs by in vitro–in vivo extrapolation, possibly withthe two binding parameters and the ROE method,using marketed drugs eliminated via CYP-mediatedmetabolic pathways.21 This result was obtained froma limited number of drugs; however, the present studyshowed comparable accuracy in a relatively largesize of dataset with widely ranged physicochemicalproperties.

Recently, a rapid increase has been recognized inthe number of publications in which the role of hep-atic uptake was investigated in the CL.12,44 In vitrostudy using hepatocytes that retain the integrity ofmany functions, including transporters, revealed thathepatic uptake is a determinant factor to the pharma-cokinetics of a series of statins such as atorvastatin,cerivastatin, and pravastatin. The pharmacokineticbehavior was retrospectively evaluated using compli-cated kinetic models including the CLint for hepaticuptake in addition to metabolism and efflux.45 Useof the kinetic model is highly likely to be difficult toroutinely estimate human CL due to the complexity.Hence, taking the current situation into considera-tion, allometric scaling would also be an applicablemethod to estimate human CL of compounds thatundergo transporter-mediated hepatic uptake untila robust methodology using a kinetic model is widelyutilized.

Based on the data in the present study and the dis-cussion for applicability of reported allometric scal-ings, we described a decision tree for projecting hu-man CLs, as shown in Figure 5, where the key factoris lipophilicity and the relationship between predictedand observed CLs. When clog P is less than 0, al-lometric scaling is preferentially employed. If a goodcorrelation between the predicted and observed CLs isproved, for example, in preclinical studies using rats

Figure 5. Decision tree for the prediction of human clear-ance with considering lipophilicity and in vitro–in vivo cor-relation in animal species.

for a series of analogues, the in vitro–in vivo extrapo-lation can be regarded as the best predictive tool thatis applicable to evaluate drug interaction, and if notestablished due to the involvement of multiple mecha-nisms in the elimination, the employment of the ROEmethod is considered. In addition, if neither the ROEmethod nor the in vitro–in vivo scaling method is ap-plicable, the LBF method using monkey data needs tobe recognized as one way of prediction. The FCIM ap-proach may become an alternative method, althoughfurther comparison study with the LBF method isnecessary.

In conclusion, the ROE method in allometric scal-ing approaches gave better prediction particularlyfor low lipophilic and highly cleared compounds. TheROE method showed predictive accuracy similar tothe in vitro–in vivo extrapolation with two unboundfractions in plasma binding and in vitro incubationcondition. Regarding the compounds for which pre-diction from the in vitro data cannot be applicable,the ROE method is a better predictor for human CL.

REFERENCES

1. Kola I, Landis J. 2004. Can the pharmaceutical industry re-duce attrition rates? Nat Rev Drug Discov 3:711–715.

2. Frank R, Hargreaves R. 2003. Clinical biomarkers in drugdiscovery and development. Nat Rev Drug Discov 2:566–580.

3. Obach RS. 2001. The prediction of human clearance fromhepatic microsomal metabolism data. Curr Opin Drug DiscovDev 41:36–44.

4. Ito K, Houston JB. 2004. Comparison of the use of liver mod-els for predicting drug clearance using in vitro kinetic datafrom hepatic microsomes and isolated hepatocytes. Pharm Res21:785–792.

5. Obach RS. 1999. Prediction of human clearance of twenty-nine drugs from hepatic microsomal intrinsic clearance data:An examination of in vitro half-life approach and nonspecificbinding to microsomes. Drug Metab Dispos 27:1350–1359.

6. Brown HS, Griffin M, Houston JB. 2007. Evaluation of cryop-reserved human hepatocytes as an alternative in vitro systemto microsomes for the prediction of metabolic clearance. DrugMetab Dispos 35:293–301.

7. De Buck SS, Sinha VK, Fenu LA, Glissen RA, Mackie CE,Nijsen MJ. 2007. The prediction of drug metabolism, tis-sue distribution, and bioavailability of 50 structurally diversecompounds in rat using mechanism-based absorption, distri-bution, and metabolism prediction tools. Drug Metab Dispos35:649–659.

8. Stringer R, Nicklin PK, Houston JB. 2008. Reliability ofhuman cryopreserved hepatocytes and liver microsomes asin vitro systems to predict metabolic clearance. Xenobiotica38:1313–1329.

9. Ito K, Houston JB. 2005. Prediction of human drug clearancefrom in vitro and preclinical data using physiologically basedand empirical approaches. Pharm Res 22:103–112.

10. Hakooz N, Ito K, Rawden H, Gill H, Limners L, Boobis AR,Edwards RJ, Carlile DJ, Lake BG Houston JB. 2006. De-termination of human hepatic microsomal scaling factor forpredicting in vivo drug clearance. Pharm Res 23:533–539.

11. Rawden HC, Carlile DJ, Tindall A, Hallifax D, Galetin A,Ito K, Houston JB. 2005. Microsomal prediction of in vivo



clearance and associated interindividual variability of six ben-zodiazepines in humans. Xenobiotica 35:603–625.

12. Shitara Y, Horie T, Sugiyama Y. 2006. Transporters as adeterminant of drug clearance and tissue distribution. Eur JPharm Sci 27:425–446.

13. Soars MG, Grime K, Sproston JL, Webborn PJH, Riley RJ.2007. Use of hepatocytes to assess the contribution of hep-atic uptake to clearance in vivo. Drug Metab Dispos 35:859–865.

14. Mahmood I, Balian JD. 1996. Interspecies scaling: Predictingpharmacokinetic parameters of antiepileptic drugs in humansfrom animals with special emphasis on clearance. J Pharm Sci85:411–414.

15. Mahmood I, Balian JD. 1996. Interspecies scaling: Predictingclearance of drugs in humans. Three approaches. Xenobiotica26:887–895.

16. Ward KW, Smith BR. 2004. A comprehensive quantitative andqualitative evaluation of extrapolation of intravenous pharma-cokinetic parameters from rat, dog, and monkey to humans. I.Clearance. Drug Metab Dispos 32:603–611.

17. Nagilla R, Ward KW. 2004. A comprehensive analysis ofthe role of correction factors in the allometric predictivity ofclearance from rat, dog and monkey to humans. J Pharm Sci93:2522–2534.

18. Mahmood I. 2007. Application of allometric principles for theprediction of pharmacokinetics in human and veterinary drugdevelopment. Adv Drug Deliv Rev 59:1177–1192.

19. Obach RS, Baxter JG, Liston TE, Silber BM, Jones BC,MacIntyre F, Rance DJ, Wastall P. 1997. The prediction ofhuman pharmacokinetic parameters from preclinical and invitro metabolism data. J Pharmacol Exp Ther 283:46–58.

20. Lave T, Coassolo P, Reigner B. 1999. Prediction of hepaticmetabolic clearance based on interspecies allometric scalingtechniques and in vitro-in vivo correlations. Clin Pharma-cokinet 36:211–231.

21. Shiran MR, Proctor NJ, Howgate EM, Rowland-Yeo K, TuckerGT, Rostami-Hodjegan A. 2006. Prediction of metabolic drugclearance in humans: In vitro-in vivo extrapolation vs allomet-ric scaling. Xenobiotica 36:567–580.

22. Wenlock MC, Austin RP, Barton P, Davis AM, LeesonPD. 2003. A comparison of physiochemical property pro-files of development and marketed oral drugs. J Med Chem46:1250–1256.

23. Vieth M, Siegel MG, Higgs RE, Watson IA, Robertson DH,Savin KA, Durst GL, Hipskind PA. 2004. Characteristic phys-ical properties and structural fragments of marketed oraldrugs. J Med Chem 47:224–232.

24. Yoshida F, Topliss JG. 2000. QSAR model for drug human oralbioavailability. J Med Chem 43:2575–2585.

25. Sacher G. 1959. Relation of lifespan to brain weight and bodyweight in mammals. In CIBA Foundation colloquia on aging;Wolstenholme GEW, O’Connor M, Eds. London: Churchill. pp115–133.

26. Iwatsubo T, Suzuki H, Sugiyama Y. 1997. Prediction of speciesdifferences (rats, dogs humans) in the in vivo metabolic clear-ance of YM796 by the liver from in vitro data. J PharmacolExp Ther 283:462–469.

27. Soars MG, Burchell B, Riley RJ. 2002. In vitro analysis of hu-man drug glucuronidation and prediction of in vivo metabolicclearance. J Pharmacol Exp Ther 301:382–390.

28. Houston JB. 1994. Utility of in vitro drug metabolism datain predicting in vivo metabolic clearance. Biochem Pharmacol47:1469–1479.

29. Davies B, Morris T. 1993. Physiological parameters in labo-ratory animals and humans. Pharm Res 10:1093–1095.

30. Austin RP, Barton P, Cockroft SL, Wenlock MC, Riley RJ.2002. The influence of nonspecific microsomal binding on ap-parent intrinsic clearance and its prediction from physico-chemical properties. Drug Metab Dispos 30:1497–1503.

31. Austin RP, Barton P, Mohmed S, Riley RJ. 2005. The bindingof drugs to hepatocytes and its relationship to physicochemicalproperties. Drug Metab Dispos 33:419–425.

32. Hosea NA, Collard WT, Cole S, Maurer TS, Fang RX, JonesH, Kakar SM, Nakai Y, Smith BJ, Webster R, Beaumont K.2009. Prediction of human pharmacokinetics from preclinicalinformation: Comparative accuracy of quantitative predictionapproaches. J Clin Pharmacol 49:513–533.

33. Tang H, Hussain A, Leal M, Mayersohn M, Fluhler E. 2007.Interspecies prediction of human drug clearance based on scal-ing data from one or two animal species. Drug Metab Dispos35:1886–1893.

34. Goteti K, Garner CE, Mahmood I. 2010. Prediction of humandrug clearance from two species: A comparison of several allo-metric methods. J Pharm Sci 99:1601–1613.

35. Mahmood I. 2009. Role of fixed coefficients and exponentsin the prediction of human drug clearance: How accurateare the predictions from one or two species? J Pharm Sci98:2472–2493.

36. Shinha VK, De Buck SS, Fenu LA, Smit JW, Nijsen M, GilissenRA, Van Peer A, Lavrijsen K, Mackie CE. 2008. Predicting oralclearance in Humans. How close can we get with allometry?Clin Pharmacokinet 47:35–45.

37. Varma MVS, Feng B, Obach RS, Troutman MD, Chpka J,Miller HR, El–Kattan A. 2009. Physicochemical determinantsof human renal clearance. J Med Chem 52:4844–4852.

38. Lin JH. 1995. Species similarities and differences in pharma-cokinetics. Drug Metab Dispos 23:1008–1021.

39. Dedrick R, Bischoff KB, Zaharko DS. 1970. Interspecies corre-lation of plasma concentration history of methotrexate (NSC-740). Cancer Chemother Rep 54:95–101.

40. Matsushita H, Suzuki H, Sugiyama Y, Sawada Y, Iga T,Hanano M, Kawaguchi Y. 1990. Prediction of the phar-macokinetics of cefodizime and cefotetan in humans frompharmacokinetic parameters in animals. J Pharmacobiodyn13:602–611.

41. Tahara H, Kusuhara H, Chida M, Fuse E, Sugiyama Y.2006. Is the monkey an appropriate animal model to exam-ine drug–drug interactions involving renal clearance? Effectof probenecid on the renal elimination of H2 receptor antago-nists. J Pharmacol Exp Ther 316:1187–1194.

42. Riley RJ, McGinnity DF, Austin RP. 2005. A unified model forpredicting human hepatic, metabolic clearance from in vitrointrinsic clearance data in hepatocytes and microsomes. DrugMetab Dispos 33:1304–1311.

43. Mahmood I. 2002. Prediction of clearance in humans fromin vitro human liver microsomes and allometric scaling. Acomparative study of the two approaches. Drug Metab DisposInteract 19:49–64.

44. Mizuno N, Niwa T, Yotsumoto Y, Sugiyama Y. 2003. Impactof drug transporter studies on drug discovery and drug devel-opment. Pharmacol Rev 55:425–461.

45. Paine SW, Parker AJ, Gardiner P, Webborn PJH, Riley RJ.2008. Prediction of the pharmacokinetics of atorvastatin,cervastatin, and indomethacin using kinetic models appliedto isolated rat hepatocytes. Drug Metab Dispos 36:1365–1374.


comparative assessment of empirical and physiological approaches on predicting human clearances

Documents