In Silico Classification of Major Clearance Pathways of ...dmd.aspetjournals.org/content/dmd/early/2010/04/27/dmd.110.032789... · the major clearance pathways of drugs by ... silico

DMD #32789

1

In Silico Classification of Major Clearance Pathways of Drugs with their

Physiochemical Parameters

Makiko Kusama, Kouta Toshimoto, Kazuya Maeda, Yuka Hirai, Satoki Imai, Koji Chiba, Yutaka Akiyama,

and Yuichi Sugiyama

Graduate School of Pharmaceutical Sciences, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo

113-0033, Japan (MK, KM, YH, SI, YS)

Graduate School of Information Science and Engineering, Tokyo Institute of Technology, 2-12-1

Ookayama, Meguro-ku, Tokyo 152-8552, Japan (KT, YA)

Faculty of Pharmacy, Keio University, 1-5-30 Shibakoen, Minato-ku, Tokyo 105-8512, Japan (KC)

DMD Fast Forward. Published on April 27, 2010 as doi:10.1124/dmd.110.032789

Copyright 2010 by the American Society for Pharmacology and Experimental Therapeutics.

This article has not been copyedited and formatted. The final version may differ from this version.DMD Fast Forward. Published on April 27, 2010 as DOI: 10.1124/dmd.110.032789

at ASPE

T Journals on M

ay 8, 2018dm

d.aspetjournals.orgD

ownloaded from

http://dmd.aspetjournals.org/

DMD #32789

2

Running title: Classification system for major clearance pathways of drugs

Corresponding author Yuichi Sugiyama, PhD

Address: Department of Molecular Pharmacokinetics, Graduate School of Pharmaceutical

Sciences, The University of Tokyo

7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan

Tel: +81-3-5841-4770

Fax: +81-3-5841-4766

e-mail: [email protected]

Number of text pages: 16

Number of tables: 4

Number of figures: 5

Number of references: 16

Number of words in the Abstract: 250

Number of words in the Introduction: 705

Number of words in the Discussion: 1480

List of nonstandard abbreviations:

CYP, cytochrome P450; fup, protein unbound fraction in plasma; log D, octanol-water distribution

coefficient (lipophilicity); LOO, leave-one-out, MW, molecular weight; OATP, organic anion transporting

polypeptide


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

3

Abstract

Predicting major clearance pathways of drugs is important in understanding their pharmacokinetic

properties in clinical use, such as drug-drug interactions and genetic polymorphisms, and their subsequent

pharmacological/toxicological effects. Here we established an in silico classification method to predict

the major clearance pathways of drugs by identifying the boundaries of physicochemical parameters in

empirical decisions for each clearance pathway. It requires only 4 physicochemical parameters (charge,

molecular weight (MW), lipophilicity (log D), and protein unbound fraction in plasma (fup)) that were

predicted from their molecular structures without performing any benchwork experiments. The training

dataset consisted of 141 approved drugs, whose major clearance pathways were determined to be

metabolism by CYP3A4, CYP2C9 and CYP2D6, hepatic uptake by OATPs, or renal excretion in an

unchanged form. After grouping by charge, each drug was plotted in a 3-dimensional space according to

three axes of MW, log D, and fup. Then, rectangular boxes for each clearance pathway were drawn

mathematically under the criterion of “maximizing F-value (harmonic mean of precision and recall) with

minimum volume”, yielding to a precision of 88% which was confirmed through two types of validation;

leave-one-out method and validation using a new dataset. With further modification towards multiple

pathways and/or other pathways, not only would this in silico classification system be useful for industrial

scientists at the early stage of drug development, which can lead to the selection of candidate compounds

with optimum pharmacokinetic properties, but also for regulators in evaluating new drugs and giving

regulatory requirements that are pharmacokinetically reasonable.


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

4

Introduction

Once a drug enters the human body, it undergoes detoxification by the complementary functions of a

wide variety of metabolic enzymes and transporters, resulting in numerous clearance pathways such as

urinary elimination through glomerular filtration and renal tubular excretion, passive diffusion into the

liver followed by hepatic metabolism by cytochrome P450 (CYP) enzymes to an inactive metabolite, or

hepatic uptake by transporters followed by excretion into the bile and then into the feces. The major

clearance pathway of drugs is one of the most important pharmacokinetic features relevant for its clinical

use. Understanding the major clearance pathways would enable prediction of the changes in the systemic

exposure of a drug caused by drug-drug interactions or genetic polymorphisms of enzymes and

transporters. Some drugs have been withdrawn from the market owing to fatal drug interactions which

often occurred by drug-drug interactions of their major clearance pathways such as cerivastatin and its

interaction with gemfibrozil (Shitara et al., 2004). The withdrawals might have been avoided by

clarifying their major clearance pathways in the human body and by recognizing the severity of drug

interactions, leading to a better decision making prior to approval or upon post-marketing regulatory

actions. Risk management of drugs can be promoted by individualized dose adjustments in accordance

with its major clearance pathways, such as the dose adjustment recommended for the antitumor agent

irinotecan, which is based on the patients’ genotype for uridine diphosphate glucuronosyltransferase 1A1

(UGT1A1) to avoid severe neutropenia caused by excess exposure of SN-38, the active metabolite of

irinotecan. Genetic polymorphisms of the organic anion transporting polypeptide 1B1 (OATP1B1),


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

5

which mediates the major clearance pathway of the antihyperlipidemic agent, simvastatin acid, is reported

to be associated with the occurrence of myopathy (Link et al., 2008). Thus, clarification of the major

clearance pathways could enhance risk management of marketed drugs by individualizing the dose or by

taking regulatory action. Drug regulators are responsible to their citizenry for securing the efficacy and

safety of drugs. Pharmacokinetic features of concern include specific genotypes of drug metabolizing

enzymes and transporters, the frequency of mutation in a particular geographic region, and interactions

with other drugs used in that region. For example, there is a broad regional diversity in the types and

frequencies of genetic mutations of CYP2D6 (Ingelman-Sundberg, 2005). In Caucasians, there are null

mutations (CYP2D6*4) in which allele frequency is 1%, but this type of mutation is not found in Asians.

Alternatively, Asians have a mutation with decreased activity (CYP2D6*10) in which the allele frequency

is 51%, which is not seen in Caucasians. Therefore, one can see that regulators in Japan or other Asian

countries would have a different point of view concerning the pharmacokinetics of drugs metabolized by

CYP2D6. Nevertheless, new drug consultation and evaluation between the industry and regulatory

authority occurs even before clinical development, with limited observational data. In these situations, in

silico prediction systems could assist decisions on the requirements of certain studies, e.g., in vitro studies

that are not documented in guidelines, drug-interaction studies both in vitro and in vivo, genotyping of

subjects in clinical trials, or clinical trials in special populations such as renal- or hepatic-impaired patients.

In vitro experiments can elucidate intrinsic clearance catalyzed by each metabolic enzyme and

transporter, but their functions cannot be quantitatively investigated using in vitro methods alone. In


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

6

silico prediction of pharmacokinetic parameters has been utilized mainly by researchers involving the

initial stage of drug discovery. There are currently few in silico prediction systems commercially

available to predict in vivo major clearance pathways of compounds. By utilizing some in vitro systems

in addition to an in silico prediction, compounds with inappropriate pharmacokinetic properties can be

screened out prior to preclinical studies. There are some simple empirical classification systems that are

generally applied in drug evaluation. For example, the Biopharmaceutics Drug Disposition Classification

System (BDDCS) (Wu and Benet, 2005) divides compounds into 4 classes according to their solubility and

extent of metabolism to predict the their disposition, interaction, and transporter-enzyme interplay.

However, the BDDCS does not tell us which CYP isozymes or transporters determine the overall clearance

of drugs. Therefore, we here established an in silico method to predict the major clearance pathways with

only 4 physicochemical parameters easily obtained or predicted from chemical structures.


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

7

Methods

Original dataset

A dataset was made using the Appendix of Goodman & Gilman’s The Pharmacological Basis of

Therapeutics (9th, 10th, and 11thed) with the addition of major organic anion transporting polypeptide

(OATP) substrates. Information concerning clearance pathways in humans was collected from published

data; for example, Japanese package inserts, Japanese interview forms, Food and Drug Administration

(FDA) labels, and research articles. Pharmacokinetic profiles after intravenous administration and in

vitro data on metabolism were also collected. If human pharmacokinetic studies consisted of only oral

administration data, it was reconsidered with bioavailability.

Training dataset

Drugs that met the criteria for the following 5 clearance pathways were selected for the training

dataset: metabolism by CYP2C9, 2D6, 3A4, renal excretion, or hepatic uptake by OATP transporters.

Clearance pathways, including non-hepatic pathways and non-CYP pathways, were evaluated by at least 2

experienced researchers per drug. In many cases, published data did not directly indicate the clearance

pathways, therefore we had to interpret and integrate several data together to evaluate the type and

contribution of the pathways for each drug. For example, we often saw information such as “this drug is

mainly metabolized by CYP2D6” with experimental data with expressed human microsomes, which does

not always mean that its major clearance pathway is CYP2D6. The amount of renal excretion of the

parent drug, and not the radioactivity in urine after administration of the 14C-labaled drug, and also the


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

8

possibility of direct conjugation were also investigated and integrated to determine which pathway

contributes most. Drugs were categorized into “renal” if more than 50% of the intravenously

administered dose is excreted unchanged into the urine, although for some drugs, it was difficult to

distinguish the urinary excreted metabolites from the parent drug. Drugs whose major clearance pathway

was attributed to CYP3A4, CYP2C9, or CYP2D6 were allocated in each group. Drugs that are not only

substrates of OATP1B1 or OATP1B3, but whose hepatic influx process is thought to be the rate-limiting

step in its hepatic clearance, were included in the OATP group. For example, anionic drugs such as

HMG-CoA reductase inhibitors and angiotensin II receptor inhibitors were included in this group.

Physicochemical parameters

For the drugs that were selected for the training dataset, their two-dimensional structures were

obtained from PubChem Compound Database (http://pubchem.ncbi.nlm.nih.gov/), and their charge,

defined as the charge of the largest fraction at pH = 7.0, was obtained using ADMET Predictor

(Simulations Plus, Lancaster, CA, USA). The drugs were then categorized by charge at neutral pH into

the following 3 groups: “cation or neutral (cation/neutral),” anion, or zwitterion. Then, their lipophilicity

(log D) at pH=7was predicted using SciFinder Scholar 2007 (Chemical Abstracts Service, Columbus, OH,

USA). In addition, the protein unbound fraction in plasma (fup) was predicted using ADME Boxes v4.0

(Pharma Algorithms, Toronto, Canada), and compared with the performance with the in vivo observed

values (if shown as range, median was used), with a view to streamline in silico prediction by

incorporating predicted parameters rather than experimental ones. These parameters were chosen based


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

9

on personal communications with experienced researchers not only in academia, but also in the

pharmaceutical industry.

Rectangular method

After dividing these drugs by charge, the drugs were plotted separately in a three-dimensional space

consisting of the MW, log D, and the fup axes. The boundaries of each of the 5 clearance pathways were

determined by the criterion of “maximum F-value (harmonic mean of precision and recall) with minimum

volume” (Fig. 1). Precision is the number of true-positives (i.e., drugs correctly classified to a pathway)

divided by the sum of true positives and false-positives (i.e., drugs correctly or incorrectly classified to a

pathway), and recall is the number of true-positives (i.e., drugs correctly classified to a pathway) divided

by the sum of true-positives and false-negatives (i.e., total number of drugs that actually belong to the

pathway):

FPTP

TPPrecision

+= Eq (1)

FNTP

TPRecall

+= Eq (2)

F-value is calculated as below:

FN)}(TPFP){(TP

TP*2

RecallPrecision

Recall*Precision*2value-F

+++=

+= Eq (3)

where TP, TN, FP, and FN are the number of drugs that were true-positive, true-negative, false-positive,

and false-negative, respectively. Only when a rectangular boundary was equal to the maximum (or

minimum) value in the training dataset, we shifted the rectangular boundary toward a preset upper (or

lower) limit of the corresponding parameter to broaden the versatility to any drug-like compound. We


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

10

determined the preset limits (MW: 70 to 2000, log D: -15 to 10, and fup: 0 to 1) on the basis of empirical

decisions. For example, if the lower boundary of fup for the anionic OATP rectangle were calculated to

be 0.003, and if this were the lowest among all the pathways, the boundary would be extended to the preset

limit of 0. This whole algorithm was named as the “rectangular method.”

Validation

We performed 2 types of validation to confirm this accuracy: cross-validation by the leave-one-out

(LOO) method and validation using a new dataset. LOO is frequently used for cross validation, and in a

dataset with n data, the training is done a total of n times with n-1 training data and 1 test data. Another

validation was performed with a new test dataset consisting of 36 drugs. These drugs were selected from

drugs approved after 2004 and filtered using the same criteria.


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

11

Results

The original dataset consisted of 294 drugs, where 230 drugs were allocated into a single major

clearance pathway, and 39 drugs were found to have multiple pathways. Of the drugs classified to a

single major clearance pathway, 150 drugs met the criteria for the training dataset. As there were only 9

drugs in the “zwitterion” group, the remaining 141 drugs were used as the training dataset (Fig. 2).

The performance of this type of classification system is evaluated by its precision and recall. The

presumed users of this classification system would classify compounds whose actual clearance pathways

were uninvestigated. From the users’ standpoint, the interpretation of precision, which is calculated here

as the proportion of drugs correctly classified to the total number of drugs classified to that pathway, would

be more practical and informative than recall, which is calculated here as the proportion of drugs correctly

classified to the total number of drugs that actually undergo that pathway.

The fup values observed in vivo correlated well with those predicted by ADME Boxes (R = 0.91, Fig.

3). The rectangular boundaries were determined using both the observed and predicted fup, and the

classification performance (precision and recall) was found to be similar (data not shown). Thus, we

decided to eliminate all observational data for this investigation, and the rectangular boundaries were

drawn using only parameters predicted from the two-dimensional molecular structures (Fig. 4). In the

three-dimensional space, there were spaces where 2 clearance pathway rectangles overlapped or spaces

where there were no rectangular boxes. A drug would be classified to either a “single pathway”, “dual

pathway”, or “none” (Fig. 5). In this investigation, the overall precision of the 117 drugs that were


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

12

classified into “single pathway” was 88% (Table 1).

In all the 14 drugs that were classified into “dual pathway,” one of the pathways was the actual major

clearance pathway. Therefore, we interpreted these drugs to be partially correct. In 8 of these 14 drugs

that were classified into “dual pathway,” the other pathway was not supported by any observed data.

However, in the remaining 6 drugs, the other pathway was supported by other information; the actual

minor clearance pathway in 5 drugs, and an experimentally confirmed pathway in 1 drug (Table 2A).

When we focused on another 14 drugs that were classified incorrectly into “single pathway,” 6 drugs

were classified into their actual minor pathway, and 4 drugs were classified into the experimentally

confirmed pathway (Table 2B). Thus, majority of these drugs were not completely incorrectly classified.

The parameters of the drugs not classified into any pathway were beyond the boundaries of all rectangular

boxes and showed no specific tendency (Table 2C). On the whole, of the 141 drugs in the training dataset,

103 drugs were given a single correct answer (73% accuracy). Furthermore, including the drugs

classified in a partially correct manner, 117 drugs were classified into a somewhat “correct” pathway (83%

accuracy (relaxed)). Validation by LOO resulted in precision and recall of 70% and 51%, and validation

using a new test dataset resulted in precision and recall of 89% and 69%, respectively (Table 3). The

classification results of all drugs in the training dataset are shown in Table 4.


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

13

Discussion

Here we have shown an in silico method to classify drugs into their major clearance pathways without

requiring any experimental data. We performed this investigation focusing on the drugs belonging to

these five clearance pathways as the training dataset and assessed the validity within drugs of these five

pathways. We focused on clearance pathways that are important in clinical use, i.e., metabolism by

CYP3A4, CYP2C9, CYP2D6, hepatic uptake by OATP, and renal excretion in an unchanged form. These

5 clearance pathways are assumed to account for more than 60% of marketed drugs (Williams et al., 2004),

and actually accounted for 51% of our original dataset under our criteria. Nevertheless, owing to

insufficient human observational data, there is a possibility that some drugs were under-evaluated and did

not meet our criteria of the major clearance pathway for the training dataset or the new dataset for

validation. CYP2C9, CYP2D6, and OATP are the pathways that are subject to inter-individual variations

due to relatively frequent genetic polymorphisms of enzymes/transporters. CYP3A4 and OATP are the

pathways that are sensitive to drug-drug interactions, whereas renal excretion is sensitive to

pathophysiological changes in patients.

Drugs that were omitted from our current investigation could be categorized into the rectangular areas

when applied to this classification. The drugs that did not fit into these five clearance pathways were

excluded from the training dataset, instead of forming a sixth category, because drugs not belonging to

these five categories would consist of drugs having a variety of clearance pathways and would not have

any uniform tendency of the physicochemical parameters, which is not favorable as a training dataset. In


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

14

addition, we decided not to integrate the other clearance pathways into the training dataset, because we

thought that the numbers of drugs in each category would be insufficient for this rectangular method, and

because drugs belonging to these five categories accounted for the majority of drugs. Furthermore, our

current scope was to establish a classification method with high accuracy for the training dataset that

account for the majority of drugs, and establishing a classification that would fit all drug-like compounds is

a secondary scope, which we will investigate further on.

The accuracy was validated by 2 methods. Generally, the performance of LOO degrades in

comparison to using a full dataset for learning. However in this study, owing to the small number of data

points and the nature of the rectangular method (by the volume minimizing rule, each boundary is set

exactly on a critical data point, not between points), we observed clear shrinkage of rectangles in some

cases during LOO. Therefore, another validation was performed with a new test dataset consisting of 36

drugs. Considering the nature of LOO, we considered that these validations support the accuracy of this

classification system.

The rectangular boundaries classifying each pathway are in accordance with general empirical rules;

for example, drugs that are eliminated unchanged by the kidneys have a small MW, large fup and small log

D (hydrophilic). Although our scope was not to investigate substrate structure-activity relationship for

the CYP enzymes and transporters involved in the pathways investigated, our results were generally

comparable to previous reports. It has been reported that CYP3A4 allows broader range of substrates in

contrast to CYP2C9 and 2D6 (Smith, 1991), which is also observed in our result. Considering molecular


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

15

size, it has been reported that CYP3A4 substrates are rather large compared to CYP2C9 and 2D6 substrate,

which are medium-sized (Lewis, 2000). In our investigation, the “dual pathway” consisted of only 2 pairs

of pathways, CYP3A4 and 2D6, or CYP2C9 and OATP, which are widely known to overlap frequently.

Furthermore, the relationship between CYP3A4 and CYP2D6 rectangles in this classification was in

accordance with empirical principals that CYP3A4 substrates are commonly more lipophilic (Smith and

Jones, 1992). Regarding charge, our results meet the results of other analyses reporting that CYP2C9

substrates are mostly anionizable compounds, and CYP2D6 substrates are cationic compounds (Lewis and

Dickins, 2002; Yamashita et al., 2008). In comparison, substrate structure-activity relationship for OATP

substrates has not been much investigated. Molecular weight threshold for biliary excretion of parent

drugs are reported to be somewhat around 325 to 400 for rats and 475 for humans (Hirom et al., 1972);

(Yang et al., 2009). The lower rectangular boundary of MW for OATP pathway was 360, which was

similar to these reported values. Focusing on the distribution of renal drugs and OATP drugs (Fig. 5),

these two groups seems to split around MW = 400. MW may be a determining factor of these two

pathways.

In order to reflect the real world, we must also account for the drugs that undergo multiple clearance

pathways. In this investigation, 10% of the drugs were classified into the “dual pathway”, and in 43% of

these drugs, both of the dual pathways were supported by observed or confirmed data. This result

indicates that overlapping rectangles may be a clue for multiple pathway prediction. Together with

expansion of the numbers of pathways, the overlapping rectangles may be used to identify the multiple


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

16

pathways, with some quantitative data. We did not expand the numbers of parameters because we

believed that the 4 parameters performed satisfactorily and because complication of this simple

classification system would be contrary to our intention. As one of our future aims is to classify more

pathways with the information provided by overlapping rectangles, we might have to put less weight on

visual intuitiveness and incorporate other methods such as SVM (support vector machine), decision tree, or

neural network. SVM has a broad capacity for classification owing to its ability to find a nonlinear

boundary in multidimensional data, although it is difficult to translate the results visually. By utilizing

SVM, we can develop a classification system using more parameters.

This in silico classification method never substitutes for in vitro experiments, but this novel prediction

method can benefit drug developers in predicting pharmacokinetic properties of drug candidates, especially

the contribution ratio of each clearance pathway to the overall clearance of drugs, which is fairly laborious

if investigated with in vitro experiments. Moreover, drug developers can add information on major

clearance pathways to construct “value-added compound libraries.” If they want to avoid developing

drug candidates that are sensitive to genetic polymorphisms, they can refrain from selecting compounds

that are classified into the CYP2C9 or CYP2D6 pathways at the very beginning of drug discovery, without

the need for experimental data. Together with physiologically based pharmacokinetic (PBPK) modeling

like the one with antihyperlipidemic agents (Shitara and Sugiyama, 2006), organ-specific distribution of

compounds could be predicted within the library. Chemists as well as pharmacokinetic scientists can use

this classification system in selecting backbone structures for hit compounds that meet the target molecular


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

17

structure. Our classification system can help selection of candidate compounds by prioritizing the

candidate backbone structures according to their ability to distribute to the target site. Furthermore, the

rectangular boundaries observed using our method can offer clues to chemists on how to optimize leading

compounds which would have a specific clearance pathway.

These types of classification systems can also be used by drug regulators in drug evaluation as well.

Not only the BDDCS mentioned earlier, but also Lipinski’s Rule of Five (Lipinski, 2003) and

Biopharmaceutics Classification System (BCS) (Yu et al., 2002), are empirical classification systems based

on experience and theory that predict categorical data for bioavailability, which is a pharmaceutical

property desired for successful drug development. BCS is used by Food and Drug Administration (FDA)

drug regulators for biowaiver, and their use is currently under consideration by the European Medicines

Agency (EMEA). That is, BCS Class 1 drugs, which have high solubility and high permeability, will be

exempted from bioavailability trails which are required whenever the formulation or manufacturing

process changes during clinical development as well as submission for generic drugs. It is estimated that

30% of drugs are Class 1 drugs (Takagi et al., 2006). Drug reviewers have long made hard decisions

based on observational data, and the application of BCS might be only the beginning for incorporating in

silico predictions into regulators’ assessments.

In conclusion, our in silico classification system well predicted the clearance pathways of drugs

belonging to the 5 clearance pathways. The rectangular boundaries classifying each pathway were

generally in accordance with empirical rules. With further investigation to increase versatility, the


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

18

pharmacokinetic properties related to drug-drug interaction and the inter-individual/interethnic variability

due to genetic polymorphism and/or renal/hepatic dysfunctions can be optimized, leading to the

development and regulatory evaluation towards more effective and less toxic drugs. This website has

access to the online classification system reported in this study, and can be used to classify the major

clearance pathways of compounds with the input of their 4 parameters (charge, MW, log D and fup) (To

reviewers: The URL with the password is currently shown in the cover letter to the journal. The URL will

be documented here upon acceptance of this manuscript.).


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

19

References

Hirom P, Millburn P, Smith R and Williams R (1972) Species variations in the threshold molecular-weight factor

for the biliary excretion of organic anions. Biochem J 129:1071-1077.

Ingelman-Sundberg M (2005) Genetic polymorphisms of cytochrome P450 2D6 (CYP2D6): clinical

consequences, evolutionary aspects and functional diversity. Pharmacogenomics J 5:6-13.

Lewis D (2000) On the recognition of mammalian microsomal cytochrome P450 substrates and their

characteristics: towards the prediction of human p450 substrate specificity and metabolism. Biochem

Pharmacol 60:293-306.

Lewis D and Dickins M (2002) Substrate SARs in human P450s. Drug Discov Today 7:918-925.

Link E, Parish S, Armitage J, Bowman L, Heath S, Matsuda F, Gut I, Lathrop M and Collins R (2008)

SLCO1B1 variants and statin-induced myopathy--a genomewide study. N Engl J Med 359:789-799.

Lipinski C (2003) Chris Lipinski discusses life and chemistry after the Rule of Five. Drug Discov Today

8:12-16.

Shitara Y, Hirano M, Sato H and Sugiyama Y (2004) Gemfibrozil and its glucuronide inhibit the organic anion

transporting polypeptide 2 (OATP2/OATP1B1:SLC21A6)-mediated hepatic uptake and

CYP2C8-mediated metabolism of cerivastatin: analysis of the mechanism of the clinically relevant

drug-drug interaction between cerivastatin and gemfibrozil. J Pharmacol Exp Ther 311:228-236.

Shitara Y and Sugiyama Y (2006) Pharmacokinetic and pharmacodynamic alterations of

3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors: drug-drug interactions and


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

20

interindividual differences in transporter and metabolic enzyme functions. Pharmacol Ther

112:71-105.

Smith D (1991) Species differences in metabolism and pharmacokinetics: are we close to an understanding?

Drug Metab Rev 23:355-373.

Smith D and Jones B (1992) Speculations on the substrate structure-activity relationship (SSAR) of cytochrome

P450 enzymes. Biochem Pharmacol 44:2089-2098.

Takagi T, Ramachandran C, Bermejo M, Yamashita S, Yu L and Amidon G (2006) A provisional

biopharmaceutical classification of the top 200 oral drug products in the United States, Great Britain,

Spain, and Japan. Mol Pharm 3:631-643.

Williams J, Hyland R, Jones B, Smith D, Hurst S, Goosen T, Peterkin V, Koup J and Ball S (2004) Drug-drug

interactions for UDP-glucuronosyltransferase substrates: a pharmacokinetic explanation for typically

observed low exposure (AUCi/AUC) ratios. Drug Metab Dispos 32:1201-1208.

Wu C and Benet L (2005) Predicting drug disposition via application of BCS: transport/absorption/ elimination

interplay and development of a biopharmaceutics drug disposition classification system. Pharm Res

22:11-23.

Yamashita F, Hara H, Ito T and Hashida M (2008) Novel hierarchical classification and visualization method for

multiobjective optimization of drug properties: application to structure-activity relationship analysis of

cytochrome P450 metabolism. J Chem Inf Model 48:364-369.

Yang X, Gandhi Y, Duignan D and Morris M (2009) Prediction of biliary excretion in rats and humans using


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

21

molecular weight and quantitative structure-pharmacokinetic relationships. AAPS J 11:511-525.

Yu L, Amidon G, Polli J, Zhao H, Mehta M, Conner D, Shah V, Lesko L, Chen M, Lee V and Hussain A (2002)

Biopharmaceutics classification system: the scientific basis for biowaiver extensions. Pharm Res

19:921-925.


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

22

Footnotes

This work was supported in part by the Grant-in-Aid for Scientific Research (A) (20249008) of the

Ministry of Education, Culture, Sports, Science and Technology (MEXT) (KM and YS); and the Okawa

Foundation in Japan (KT and YA).

This work was previously presented at the 23rd JSSX (The Japanese Society for the Study of Xenobiotics)

Annual Meeting (Classification of Major Clearance Pathways of Drugs Based on Physicochemical

Parameters, Makiko Kusama, Kouta Toshimoto, Kazuya Maeda, Yutaka Akiyama, and Yuichi Sugiyama)

Person to receive reprint requests:

Name: Yuichi Sugiyama, Ph.D

Address: 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-0033, Japan

Department of Molecular Pharmacokinetics,

Graduate School of Pharmaceutical Sciences, the University of Tokyo

e-mail: [email protected]


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

23

Figure legends

Fig. 1. Scheme of the “rectangular method.” Drugs were plotted in a three-dimensional space

according to their molecular weight (MW), lipophilicity (log D), and fraction in plasma (fup), and

rectangular boundaries for each group were determined mathematically under the “rectangular method”.

Fig. 2. Clearance pathways of drugs in the original dataset.

Fig. 3. Correlation between predicted and observed protein bound unbound fraction in plasma

(fup) values (r = 0.91). In 83% of the drugs, the predicted fup fell between 1/3 and 3-fold of observed

values.

Fig. 4. Boundaries determined for each clearance pathway. The boundaries could not be defined if

the number of drugs allocated in a pathway was 1 or 0. The boundaries were extended according to a

preset rule. ND = not determined.

Fig. 5. Distribution of drugs in the three-dimensional area and the rectangular boundaries for

each pathway. Data shown as two-dimensional figures for each charge group (Left: cation/ neutral;

Right: anion).


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

24

Tables

Table 1. Actual and classified clearance pathways. Shaded areas indicates the drugs that were

classified correctly.

Classified

Total

Single

Dual None

3A4 2C9 2D6 Renal OATP Recallb

Recall

(relaxed)c

Act

ual

3A4 46 0 0 2 0 1 3 52 88% 90%

2C9 1 6 0 0 4d 0 1 12 50% 50%

2D6 2 0 2 3 0 7 4 18 11% 50%

Renal 1 0 0 37 0 1 2 41 90% 90%

OATP 0 0 0 1 12 5 0 18 67% 94%

Total 50 6 2 43 16 14 10 141

Precisiona 92% 100% 100% 86% 75% Accuracy

Accuracy

(relaxed) c

Overall precision 88% 73% 82%

a Precision is the proportion of drugs correctly classified to a pathway, to the total number of drugs


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

25

classified to that pathway.

b Recall is the proportion of drugs correctly classified to a pathway, to the total number of drugs that

actually undergo that pathway.

c Recall (relaxed) and Accuracy (relaxed) are used when drugs that were classified in a partially correct

manner were also regarded as correct (see text and Table 2A). For example, the recall and the recall

(relaxed) for CYP3A4 is 46/52 = 88% and (46 + 1)/52 = 90%, respectively.

d Although the major clearance pathways of these four drugs (glimepiride, glipizide, glyburide

(glibenclamide), and irbesartan) were to be CYP2C9 from published observational data, we have

experimentally confirmed that they are substrates of organic anion transporting polypeptide (OATP)

(unpublished data).


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

26

Table 2. Comparison between the actual and classified pathways. Circles indicate the classified

pathway(s). Dark shaded areas show the actual major clearance pathway, light shaded areas show the

minor clearance pathways that have been published in the literature or in documented in drug review

reports, and boxed areas indicate the pathways that have been experimentally confirmed by our group but

not yet published. Panel A: Drugs classified as “dual pathways”; Panel B: Drugs classified incorrectly as

a “single pathway”; Panel C: Drugs not classified as any pathway and the reasons for such classification.

The up-arrow (down-arrow) means that the parameter of the drug is above (below) the boundary of its

actual pathway.

A.

Charge Drug name

Classified pathway

3A4 2C9 2D6 Renal OATP

Cation/neutral

Clonazepam ○ ○

Chlorpromazine ○ ○

Diphenhydramine ○ ○

Doxepin ○ ○

Fluoxetine ○ ○

Imipramine ○ ○

Nortriptyline ○ ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

27

Tolterodine ○ ○

Anion

Dicloxacillin ○ ○

Enalapril ○ ○

Pravastatin ○ ○

Rifampin ○ ○

Rosuvastatin ○ ○

Temocaprilat ○ ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

28

B.

Charge Drug name

Classified pathway


Cation/neutral

Granisetron ○

Ifosfamide ○

Celecoxib ○

Flecainide ○

Fluphenazine ○

Metoprolol ○

Risperidone ○

Timolol ○

Chloroquine ○

Anion

Glimepiride ○

Glipizide ○

Glyburide/glibenclamide ○

Irbesartan ○

Enalaprilat ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

29

C.

Charge Drug name Actual pathway log D fup MW

Cation/neutral

Eplerenone 3A4 ↓ ↑ -

Oxycodone 3A4 ↓ ↑ -

Tamsulosin 3A4 ↓ - -

Chlorpheniramine 2D6 - ↑ -

Galantamine 2D6 - ↑ -

Propranolol 2D6 - ↑ -

Venlafaxine 2D6 - ↑ -

Digoxin Renal ↑ - ↑

Topiramate Renal ↑ - -

Anion Montelukast 2C9 ↑ ↓ ↑


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

30

Table 3. Results of Validation. See the table legend of Table 1 for explanation. Panel A: leave-one-out

(LOO) method; Panel B: new dataset of drugs.

A.

Classified

Total

Single

Dual None

3A4 2C9 2D6 Renal OATP Recall

Recall

(relaxed)

Act

ual

3A4 39 0 2 2 0 4 5 52 75% 83%

2C9 1 2 0 0 6 1 2 12 17% 25%

2D6 7 0 0 5 0 4 2 18 0% 22%

Renal 2 0 0 28 1 0 10 41 68% 68%

OATP 0 1 0 4 3 9 1 18 17% 67%

Total 49 3 2 39 10 18 20 141

Precision 80% 67% 0% 72% 30% Accuracy

Accuracy

(relaxed)

Overall precision 70% 51% 64%


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

31

B.

Classified

Total

Single

Dual None

3A4 2C9 2D6 Renal OATP Recall

Recall

(relaxed)

Act

ual

3A4 12 0 0 1 1 0 6 20 60% 60%

2C9 0 1 0 0 0 0 0 1 100% 100%

2D6 - - - - - - - - - -

Renal 1 0 0 12 0 0 2 15 80% 80%

OATP - - - - - - - - - -

Total 13 1 0 13 1 0 8 36

Precision 92% 100% - 92% - Accuracy

Accuracy

(relaxed)

Overall

Precision

89% 69% 69%


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

32

Table 4. The classified and actual major pathway for all drugs investigated. Shaded areas indicate

the actual major clearance and circles indicate the classified pathway(s).

Charge Drug name Classified pathway


Cation/neutral Acyclovir ○

Cation/neutral Alprazolam ○

Cation/neutral Amicacin ○

Cation/neutral Amiodarone ○

Cation/neutral Amlodipine ○

Anion Amoxicillin ○

Cation/neutral Aprepitant ○

Cation/neutral Atazanavir ○

Cation/neutral Atenolol ○

Cation/neutral Atomoxetine ○

Anion Atorvastatin ○

Anion Bosentan ○

Cation/neutral Buprenorphine ○

Cation/neutral Buspirone ○

Anion Candesartan ○

Cation/neutral Carbamazepine ○

Anion Cefazolin ○

Anion Cefdinir ○

Anion Cefixime ○

Anion Cefotetan ○

Anion Cefuroxime ○

Cation/neutral Celecoxib ○

Anion Cephalexin ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

33

Anion Cerivastatin ○

Cation/neutral Chloroquine ○

Cation/neutral Chlorpheniramine

Cation/neutral Chlorpromazine ○ ○

Cation/neutral Chlorthalidone ○

Anion Cidofovir ○

Cation/neutral Cimetidine ○

Cation/neutral Clonazepam ○ ○

Cation/neutral Clonidine ○

Anion Daptomycin ○

Cation/neutral Diazepam ○

Anion Dicloxacillin ○ ○

Cation/neutral Didanosine ○

Cation/neutral Digoxin

Cation/neutral Diltiazem ○

Cation/neutral Diphenhydramine ○ ○

Cation/neutral Dofetilide ○

Cation/neutral Donepezil ○

Cation/neutral Doxepin ○ ○

Anion Doxycycline ○

Anion Enalapril ○ ○

Anion Enalaprilat ○

Cation/neutral Eplerenone

Cation/neutral Erythromycin ○

Cation/neutral Ethambutol ○

Cation/neutral Felodipine ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

34

Cation/neutral Finasteride ○

Cation/neutral Flecainide ○

Cation/neutral Fluconazole ○

Cation/neutral Fluoxetine ○ ○

Cation/neutral Fluphenazine ○

Anion Fluvastatin ○

Anion Foscarnet ○

Anion Furosemide ○

Cation/neutral Galantamine

Cation/neutral Ganciclovir ○

Anion Glimepiride ○

Anion Glipizide ○

Anion Glyburide/glibenclamid

e ○

Cation/neutral Granisetron ○

Cation/neutral Hydrochlorothiazide ○

Anion Ibuprofen ○

Cation/neutral Ifosfamide ○

Cation/neutral Imatinib ○

Cation/neutral Imipramine ○ ○

Anion Indomethacin ○

Anion Irbesartan ○

Cation/neutral Itraconazole ○

Cation/neutral Ketoconazole ○

Anion Ketorolac ○

Cation/neutral Lamivudine ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

35

Cation/neutral Lansoprazole ○

Cation/neutral Leflunomide ○

Cation/neutral Levetiracetam ○

Cation/neutral Lopinavir ○

Cation/neutral Lovastatin ○

Anion Meloxicam ○

Cation/neutral Metformin ○

Cation/neutral Methadone ○

Anion Methotrexate ○

Cation/neutral Methylprednisolone ○

Cation/neutral Metoprolol ○

Cation/neutral Midazolam ○

Anion Montelukast

Cation/neutral Nelfinavir ○

Cation/neutral Nevirapine ○

Cation/neutral Nifedipine ○

Cation/neutral Nortriptyline ○ ○

Anion Olmesartan ○

Cation/neutral Ondansetron ○

Cation/neutral Oxybutynin ○

Cation/neutral Oxycodone

Cation/neutral Paroxetine ○

Anion Penicillin G ○

Anion Phenytoin ○

Anion Pitavastatin ○

Cation/neutral Pramipexole ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

36

Anion Pravastatin ○ ○

Cation/neutral Prednisolone ○

Cation/neutral Prednisone ○

Cation/neutral Procainamide ○

Cation/neutral Propranolol

Cation/neutral Quetiapine ○

Cation/neutral Quinidine ○

Cation/neutral Quinine ○

Cation/neutral Ranitidine ○

Anion Repaglinide ○

Anion Rifampin ○ ○

Anion Risedronate ○

Cation/neutral Risperidone ○

Cation/neutral Ritonavir ○

Anion Rosuvastatin ○ ○

Cation/neutral Sildenafil ○

Anion Simvastatin acid ○

Cation/neutral Sirolimus ○

Cation/neutral Tacrolimus ○

Cation/neutral Tadalafil ○

Cation/neutral Tamoxifen ○

Cation/neutral Tamsulosin

Anion Telmisartan ○

Anion Temocapril ○

Anion Temocaprilat ○ ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


DMD #32789

37

Anion Tenofovir ○

Cation/neutral Timolol ○

Anion Tolbutamide ○

Cation/neutral Tolterodine ○ ○

Cation/neutral Topiramate

Cation/neutral Topotecan ○

Cation/neutral Toremifene ○

Cation/neutral Trazodone ○

Cation/neutral Trimethoprim ○

Anion Valsartan ○

Cation/neutral Venlafaxine

Cation/neutral Verapamil ○

Cation/neutral Vincristine ○

Cation/neutral Vinorelbine ○

Anion Warfarin ○

Cation/neutral Zolpidem ○


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


logD

fup

MW

: Renal: CYP3A4: CYP2D6

Renal

CYP3A4

CYP2D6

Figure 1


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


294drugs

Multiple clearance pathways (39)

Others (80)

Unknown clearance pathways (25)

Single major

clearance pathway

(230)

CYP3A4,CYP2C9, CYP2D6,

renal, or

OATP(150)

Anion (47)

CYP2C9

(11)

CYP3A4

(0)

CYP2D6

(0)

Renal

(18)

OATP

(18)

Cation / Neutral (94)

CYP2C9

(1)

CYP3A4

(52)

CYP2D6

(18)

Renal

(23)

OATP

(0)

Zwitterion (9)

CYP2C9

(0)

CYP3A4

(0)

CYP2D6

(1)

Renal

(7)

OATP

(1)

Figure 2


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


0.0001

0.001

0.01

0.1

1

0.0001 0.001 0.01 0.1 1

predicted

observed

cation/ neutralanion

x3

x1/3

Figure 3


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


Cation/ Neutral

fup =0 - 0.4

MW=220-2000

log D =1.2 - 10

fup =0.04 - 0.1

MW=240 - 340

log D =0.4 - 2.9

CYP3A4

CYP2D6

fup=0.2 - 1

MW=70 - 600

log D =-15 - 0.6

renal

Only one dataCYP2C9

OATPND

Anion

MW=70 - 2000

fup =0.01 - 1

log D =-15 - 0.0

fup =0.008 - 0.2

MW=200 - 360

log D =0.1 - 2.5

MW=360 - 840

fup =0 - 0.3

log D =-2.3 - 10

CYP2C9

OATP

ND

renal

NDCYP3A4

CYP2D6

Figure 4


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


CYP2C9RenalOATP

anion

fup0 0.2 0.4 0.6 0.8 1

0

500

1000

1500

2000 CYP3A4CYP2C9CYP2D6Renal

-15

-10

-5

0

5

10

0 0.2 0.4 0.6 0.8 1

cation/neutralm

olec

ular

wei

ght

log

D

fup

Figure 5


at ASPE

T Journals on M

ay 8, 2018dm


ownloaded from


Documents

In Silico Classification of Major Clearance Pathways of ...dmd.aspetjournals.org/content/dmd/early/2010/04/27/dmd.110.032789... · the major clearance pathways of drugs by ... silico