development of pk-and pbpk-basedmodeling tools for...

Author's personal copy

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 773–788

jo ur n al hom ep age : www.int l .e lsev ierhea l th .com/ journa ls /cmpb

Development of PK- and PBPK-based modeling tools forderivation of biomonitoring guidance values

M. Bartelsa,∗, D. Ricka, E. Lowea, G. Loizoub, P. Pricea, M. Spendiffb, S. Arnolda,J. Cockerb, N. Ball c

a Toxicology and Environmental Research & Consulting, The Dow Chemical Company, Midland, MI, USAb Mathematical Sciences Unit, Health Improvement Group, Health and Safety Laboratory, Harpur Hill, Buxton, UKc Dow Europe GmbH, Horgen, Switzerland

a r t i c l e i n f o

Article history:

Received 21 May 2011

Received in revised form

14 April 2012

Accepted 27 April 2012

Keywords:

Pharmacokinetic

PBPK

Excel

Micturition

Model

Biomonitoring

QSAR

a b s t r a c t

There are numerous programs ongoing to analyze environmental exposure of humans to

xenobiotic chemicals via biomonitoring measurements (e.g.: EU ESBIO, COPHES; US CDC

NHANES; Canadian Health Measures Survey). The goal of these projects is to determine rel-

ative trends in exposure to chemicals, across time and subpopulations. Due to the lack of

data, there is often little information correlating biomarker concentrations with exposure

levels and durations. As a result, it can be difficult to utilize biomonitoring data to evaluate if

exposures adhere to or exceed hazard/exposure criteria such as the Derived No-Effect Level

values under the EU REACH program, or Reference Dose/Concentration values of the US EPA.

A tiered approach of simple, arithmetic pharmacokinetic (PK) models, as well as more stan-

dardized mean-value, physiologically-based (PBPK) models, have therefore been developed

to estimate exposures from biomonitoring results. Both model types utilize a user-friendly

Excel spreadsheet interface. QSPR estimations of chemical-specific parameters have been

included, as well as accommodation of variations in urine production. Validation of each

model’s structure by simulations of published datasets and the impact of assumptions of

major model parameters will be presented.

© 2012 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

The time–course of drug concentrations in blood or tissuesis often predicted using pharmacokinetic modeling calcula-tions. These calculations are generally based on assumptionsthat the body can be represented simply as a single, homoge-nous compartment (pharmacokinetic (PK) model) [1], or asa more complex system of separate tissues with differingpharmacokinetic properties and incorporating physiologicalparameters such as blood flow and tissue size (physiologi-cally based (PBPK) model) [2]. Measured or estimated rates

∗ Corresponding author. Tel.: +1 989 636 9057; fax: +1 989 638 9305.E-mail address: [email protected] (M. Bartels).

of absorption, distribution, metabolism and excretion (ADME)are utilized in these pharmacokinetic models, which can bevalidated with data from numerous clinical studies.

These models can also be used to interpret biomonitor-ing data, which consists of blood or tissue concentrationsof xenobiotic chemicals or their metabolites in workersor non-occupationally exposed humans. There are numer-ous programs ongoing to analyze environmental exposureof humans to xenobiotic chemicals, in which single time-point samples are analyzed for concentrations of a varietyof compounds in a subset of the population (e.g.: EU ESBIO,COPHES, NewGeneris; US CDC NHANES; Canadian Health

0169-2607/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights reserved.http://dx.doi.org/10.1016/j.cmpb.2012.04.014


774 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 773–788

Measures Survey) [3–7]. While these single-sample assaysmay result in some error of exposure predictions, due tothe known diurnal variability in metabolite levels [8,9], theselarge biomonitoring programs have an overall goal to deter-mine relative trends in exposure to chemicals, across timeand subpopulations. However, these programs do not attemptto calculate actual exposures from the biomarker mea-surements, or compare if biomonitoring-derived exposureestimates are below or exceed acceptable limit values, suchas the REACH Derived No-Effect Level or the US EPA RfC/RfDvalues. As a result, it can be difficult to utilize biomon-itoring data to evaluate acceptable levels of exposure tochemicals.

Some methods, such as the Biomonitoring Equivalentsapproach [10] have been developed to define biomarkerlevels, or biomonitoring guidance values (BGVs) [11], thatare equivalent to reference exposure values. This approachrequires the use of PK or PBPK models to correlate animalor human points of departure (POD) with biomarker levelsat the human equivalent POD. A tiered approach utilizingsimple PK models, as well as more standardized mean-valuePBPK models, would aid researchers in their derivation ofBGVs.

There are numerous examples of how chemical-specificPK or PBPK models have been used to derive biomonitor-ing guidance values [12–14]. Several PBPK models have alsobeen developed directly in the MicrosoftTM Excel application,therefore requiring no user programming [15,16]. However,these models, while well-validated, can require a substantialamount of empirical ADME data and model optimization foraccurate prediction of biomarker levels. There is a paucityof this type of data for most synthetic chemicals. As aresult, for a large percentage of biomonitoring data, onlylimited interpretation has been conducted. The recent Excel-based PBPK model of Jongeneelen et al., is a substantialimprovement in this area, in that it also provides model-derived predictions of tissue distribution, dermal absorp-tion and renal clearance and accommodates modeling ofmetabolites [17].

The purpose of the work described in this manuscript isto develop several user-friendly computer applications thatsupport exposure estimates from biomonitoring data. A multi-tissue compartment PBPK model, as well as an arithmetic,one-compartment PK model requiring minimal ADME data,have been developed. Both applications have been written forthe Microsoft Excel spreadsheet interface and include Quan-titative Structure–Property Relationship (QSPR) estimations ofchemical-specific parameters, such as blood:tissue partitioncoefficients, as in the recent model of Jongeneelen and Berge[17]. The PBPK model includes accommodation of variationsin urine micturition, to allow for direct comparison of urinebiomarker concentrations with timed samples collected frombiomonitoring studies. This model utilizes an Excel interfaceto a compiled acslX model, allowing for full utilization of aclassic modeling tool in a user-friendly manner. Both modelsalso support automatic back-calculation (“reverse dosimetry”)of exposure levels from biomonitoring data. Verification ofeach model’s structure by simulations of published datasetsand the impact of assumptions of major model parameterswill be presented.

Text Fig. 1 – Diagram of one-compartment PK model forreverse dosimetry.

2. Computational methods

An overview of the design of the PK and PBPK models forreverse dosimetry is presented in the following sections. Inaddition, empirical data used for the derivation of the oil:waterpartition QSPR required for volume of distribution (Vd) or tis-sue:plasma partition (Ptp) coefficients is discussed.

2.1. PK model for reverse dosimetry

As shown in Text Fig. 1, a one-compartment PK model isemployed to calculate single- and steady-state exposure levelsfrom plasma (Cp) or urine (Cu) biomarker concentrations. Clas-sic PK equations for single-dose and steady state exposures [1]were incorporated into this model and are fully defined in anappendix section of the Microsoft ExcelTM-compatible work-book file. A first-order uptake of the chemical is employed(kabs), which is most relevant for oral exposures, but is alsoassumed adequate for short-term dermal or inhalation expo-sures. The apparent first-order elimination rate (K) definedby Gibaldi and Perrier is set equal to the urinary excretionrate (kexcr), as it is assumed that any metabolism involvedin production of the relevant biomarker is primarily a first-pass phenomenon, and as such does not need to be separatelyincluded in the model. The assumption of linear kinetics is feltappropriate for non-occupational exposure estimates frombiomonitoring data. Non-linearities in PK parameters mayoccur at higher worker exposure scenarios.

Equations used for the PK model are shown below and alsofully described in the PK model, available as supplementalmaterial to this manuscript:

Single Exposure Amount (from plasma conc)

= Cpt × Vd × (kabs − kexcr)

(kabs × FrxAbs) × (e−kexcr∗T − e−kabs∗T)(1)

Single Exposure Amount (from urine conc)

= CurT × Vur × (kabs − kexcr)

(kexcr × kabs × FrxExcr) × (e−kexcr∗T − e−kabs∗T)(2)

Steady-state Exposure Amount (from plasma conc.)

= CpSS × Vd × kexcr × Tau

FrxAbs(3)


c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 773–788 775

Steady-state Exposure Amount (from urine conc.)

= CurSS × Vur24hr

FrxExcr(4)

where CpT is plasma concentration of biomarker (Cp) at time T,kabs and kexcr are absorption and excretion rates, respectively,T is time since exposure, FrxAbs is fraction of dose absorbed,Vd is the calculated or measured volume of distribution, CurTis “instantaneous” concentration of biomarker in urine (Cu)at time T, FrxExcr is fraction of dose excreted in urine asbiomarker, CpSS is plasma biomarker concentration at steadystate, Tau is interval between steady state exposures (assumed24 h), CurSS is urine biomarker concentration at steady stateand Vur24hr is the daily urine volume.

2.2. PBPK model for reverse dosimetry

The PBPK model for reverse dosimetry describes thetime–course of ADME for a chemical and up to one biomarkermetabolite following a single exposure. Input routes includeoral, dermal and inhalation (Text Fig. 2). Metabolism of theparent compound, if relevant, occurs in the liver and themetabolite formed is input into the venous blood compart-ment of the parallel metabolite PBPK model. A first-orderelimination from the venous blood is used for both parentchemical and metabolite to account for loss via urinary excre-tion (micturition).

This PBPK model is run from a custom Microsoft Excelworkbook file. User-input parameters are transferred from theExcel file automatically to a compiled, acslX model (AEgisTechnologies, Huntsville, Alabama, USA). Once the modelcompletes calculations of time–course blood and tissue con-centrations, this model output is written back to the Excel file.After 11 iterations of model runs across an array of dose levels,an estimate of exposure is calculated from the various dosetime–course datasets.

Concentrations of the test chemical and metabolite in rep-resentative physiological compartments are calculated acrosstime via a series of differential equations, in which partition-ing between blood and tissues is instantaneous (“well-stirred”approach [18]). The general approach for this model designhas been well described previously [2]. A copy of the modelcode is included in the Excel interface workbook, available assupplementary material to this manuscript:

2.3. QSPR for oil:water partition coefficient

The QSPR approaches for predicting Vd or Ptp values forthe PK and PBPK models, respectively, rely on a funda-mental relationship for partitioning of the test chemicalbetween octanol:water (Koct:water) or vegetable oil:water(Koil:water) [19–22]. While Koct:water values are generallyempirically derived for chemicals, the Koil:water parameteris often not available. Previous researchers have utilized lin-ear relationships between these two parameters to estimateKoil:water from Koct:water [23]. However, this linear relation-ship may not be appropriate for fairly lipophilic compounds(log Koct:water > 4) and may result in an overestimate of theKoil:water parameter [24].

To evaluate the relationship between Koil:water andKoct:water, published Koil:water data for 104 compounds withlog Koct:water values of −1.9 to 6.0 [19,23,25] were com-bined with new Koil:water data for eight lipophilic chemicals(log Koct:water 4.3–8.3) as well as the positive control p-ethylphenol (log Koct:water 2.5), described in the followingsection. This combined data was then fit to a sigmoidal doseresponse curve (with variable slope), using the Prism applica-tion (v5.0; GraphPad Software Inc, La Jolla, CA, USA).

Equations used for the PK model are shown below and alsofully described in the PK model, available as supplementalmaterial to this manuscript:

2.3.1. Koil:water determinations3H(G)-benzo(a) pyrene, 7-3H-cholesterol, 3H-retinol, 14C-p-ethylphenol, and 14C-1-octadecanol were purchased fromAmerican Radiolabeled Chemicals (St. Louis, MI). 14C(u)-octachlorostyrene was purchased from Dupont, NEN Products(Boston, MA). Chlorpyrifos (ring-4-3H) was obtained fromAmersham Biosciences (now GE Healthcare, Piscataway, NJ).14C-hexachlorobenzene, 14C-XDE-175J, and 14C-XDE-007 wereobtained and used with permission from Dow AgroSciencesLLC (Indianapolis, IN). A bottle of 100% pure all natural FilipoBerio Olive Oil (Hackensack, NJ) was purchased from a localgrocery store. Each chemical was dissolved in 10 ml of oliveoil in a 24-ml glass vial. An equal volume of Milli-Q waterwas added to each vial. The concentration of each chemicalfortified into the oil was intended to produce a final waterconcentration of the chemical at approximately 1/10 the watersolubility of the chemical if partitioning were to occur accord-ing to the published (or QSPR-derived) Koil:water values. Eachvial contained a glass magnetic stir bar; the two phases weregently stirred for 7 days on a magnetic stir plate at 25 ◦C. Analiquot of oil and an aliquot of water were taken on the seventhday and assayed by radiochemical detection with liquid scin-tillation counting. Each chemical treatment was performed induplicate. Koil:water was calculated as the ratio of total counts(dpm) in the oil phase vs. the total counts (dpm) in the waterphase.

3. Program descriptions


Use of the PK model for reverse dosimetry is fairly straightfor-ward. User input is minimal, requiring only basic test chemicalinformation and selected pharmacokinetic (PK) parameters,on the User Entry page of the Excel workbook (Fig. 1). The testchemical and biomarker names (if metabolite is biomarker),and molecular weights are needed for molar conversions ofbiomarker concentration to exposure amounts within the PKmodel. An estimated time between exposure and biomarkersampling is needed to back-calculate exposure levels. Thebiomarker concentration in plasma (Cp) and/or urine (Cu) fora given individual/time point is entered, in units of mg/L(or possibly mg/g creatinine for urine). Note that wholeblood concentrations can be used instead of plasma lev-els, if the blood/plasma ratio for the biomarker is ∼1.0. ThePK parameters required for this model are absorption and



Text Fig. 2 – Diagram of PBPK model for reverse dosimetry.

elimination rates (kabs, kexcr), fractions absorbed and elimi-nated as biomarker, and volume of distribution (Vd), whichis needed for plasma biomarkers only. Urine hourly produc-tion rate is set at the default for a 70 kg adult male [26], whichcan be adjusted by the user. Hourly creatinine productionrate is based on a population average creatinine productionrate (mg/L) urine [27] for individuals ages 20–59, multipliedby the average hourly urine production rate (L/h). Note thatcorrection of urine biomarker levels for creatinine concen-tration should be done on a compound-specific basis, and isdependent on the renal clearance mechanism and water sol-ubility of the compound [28]. Required PK values for reversedosimetry calculations are flagged with a red asterisk, once abiomarker value is entered (plasma, urine, urine/Cr). A morecomplete description on the use of this model can be foundin the user’s manual, available as supplementary material tothis manuscript.

These PK parameters are generally obtained from con-trolled metabolism/exposure studies in animals or humanvolunteers. Some parameters, such as kabs, fraction absorbedand Vd can be estimated separately with QSPR programssuch as ADME Suite (Advanced Chemistry Development, Inc.

(ACD/Labs), Toronto, On, Canada) or ADMET Predictor (Sim-ulations Plus, Lancaster, CA, USA). Since Vd is a summationof tissue:plasma partition coefficients multiplied by tissuevolumes (see text Eq. (5)), a QSPR to derive this parameterfrom a few physical–chemical properties is included withinthis spreadsheet-based model. This Vd QSPR is based on thecompilation of published tissue:plasma partition coefficientalgorithms of Rodgers and Rowland [29], which utilizes a com-pound’s solubility in octanol or vegetable oil as a surrogate fordistribution into the lipid fraction of tissues. Note that thesepartitioning algorithms have been developed and validatedprimarily for pharmaceutical compounds (log Koct:water < 4),which are often less lipophilic than industrial chemicals in theenvironment.

Volume of distribution (Vd) = Vp

fu+

∑(Vt × Ptp) (5)

where Vp is the volume of plasma in the body, fu is the frac-tion of the chemical or metabolite not bound to protein, Vtis the volume of individual tissues in the body and Ptp is thetissue:plasma partition coefficient for individual tissues.



Fig. 1 – User interface for Excel-based PK model for reverse dosimetry.

Since the exposure amount calculations are direct rear-rangements of PK equations for Cp or Cu at time Tpost-exposure, a predicted exposure amount is immediatelycalculated from any biomarker concentration, applying Eqs.(1) and (2), above. The model output is given in several for-mats. Exposure amounts, in total mg chemical for a nominalhuman of 70 kg body weight, are given for single exposure orsteady-state assumptions, based on entered plasma or urinebiomarker concentrations (Fig. 1). For visualization purposes,time–course graphs of biomarker levels in plasma, urine andcumulative in urine are also shown for the exposure amountcalculated from the inputted biomarker value. Since the accu-racy in exposure amount determinations falls off over time,due to variability in analytical measurements, no exposureamounts are calculated for single-exposure scenarios whenthe time between exposure and sampling is greater than 5times the sum of the absorption and elimination half-lives.This summation of half-lives will also accommodate “flip-flop” model scenarios, where absorption rates are slower thanelimination rates. This phenomenon is seen following der-mal uptake of a chemical or from sustained release drugformulations, where the initial positive slope of the plasmatime–course curve represents the elimination rate and the

subsequent negative slope represents the slow absorption rate[30].


The data required for the PBPK model is similar to that ofthe PK model. Chemical-specific and PK data are requiredfor both the chemical and metabolite (entered in the UserEntry page of the Excel workbook) (Fig. 2). First-order uptakerates are assumed for oral exposures. Dermal exposures areassumed to occur from infinite-volume solutions (mg/L) of thechemical, on a nominal skin surface area of 100 cm2 (15.5 in2).First-order Kp rates for dermal uptake can be user-enteredor derived via a published QSPR, used in the public-accessDermwin Application (v1.43, US EPA) [31]. For inhalationexposures, a continuous air concentration is assumed, for auser-defined period of time in hours. Biomarker concentra-tions are inputted for either parent chemical or metabolite,from blood, plasma, urine or urine corrected for creatinineconcentrations. Metabolism rates for loss of both chemicaland metabolite are assumed to be saturable, therefore Vmax

and Km terms are required for both parent and metabolite.Plasma elimination rates are required for both chemical and



Fig. 2 – Flowchart of PBPK model for reverse dosimetry use, biomarker time–course and exposure calculations.

metabolite. Urine micturition times (up to 20 values) canbe inputted to match concentrations that would be mea-sured from timed urine collections, otherwise a defaultmicturition interval of 4 h is assumed. Tissue-specific Ptpvalues can be user entered. These Ptp values can also bederived within the model from published QSPR methodsthat derive partitioning based on water/lipid compositions

in tissues [19,20], or offline with a more unified algorithm,incorporating distribution between cells, interstitial fluid, ery-throcytes and plasma [32]. Note that the Koil:water parameter,derived from the updated correlation with Koct:water, as dis-cussed in Section 2.3 (above), can be used as an input inthese Ptp QSPR methods. Note that Ptp values are calcu-lated by default for tissue:plasma, unless biomarker values



are entered for whole blood, in which case these valuesare calculated for tissue:blood, as per Poulin and Krishnan[33].

Tissue : blood partition coefficient (Ptp)

= [Kox : water × (Vnlt + 0.3 × Vpht) + (1 × Vwt + 0.7 × Vpht)] × fup

[Kox : water × (Vnlp + 0.3 × Vphp) + (1 × Vwp + 0.7 × Vphp)] × fut

(6)

where Kox:water is Koct:water or Koil:water, Vnlt, Vpht,Vnlp andVphp are the volume of neutral lipids (nl) and phospholipids(ph) in the tissues and plasma, respectively, Vwt and Vwp arethe volume of water in the tissue and plasma, respectively,and fup and fut are the fractions of the chemical or metabolitenot bound to protein in the plasma or tissue, respectively.

Although this modeling tool is designed to interpret humanbiomonitoring data, the algorithm for predicting Ptp valueshas been expanded to facilitate potential modeling in speciesbesides human. Calculated Ptp values for mouse, rat, rab-bit, or dog can be found on the Ptp derivation page of themodeling workbook. In addition to Ptp values, the required,analogous blood:air partition coefficients (Pba) for the PBPKmodel were derived from calculated Kwater:air and Koil:airpartition coefficients (as defined by Kelly et al. [34]), in addi-tion to the derived Koil:water values from the QSPR derived inthis project.

Reverse dosimetry calculations are then made from thisExcel workbook. The program first runs the PBPK model itera-tively, at 11 different exposure concentrations, logarithmicallydistributed across the user-defined range of low and highexposure values. Simulations are conducted for 24 h or up tothe last micturition time, whichever is longer. Once modelruns are complete, a combined set of four graphs are created,showing the time-courses of parent chemical and metabolitelevels in blood and excreted urine (Fig. 3). A reverse dosimetrycalculation [10] of exposure from the input biomarker valueis determined from a “Lookup” of biomarker time-course val-ues from each model run. Exact exposure values are linearlyinterpolated from the two closest predicted biomarker levelsthat bracket the user-input biomarker value. If the exposureamount is outside of the user-defined exposure range, themodel will need to be re-run with new exposure values. Aswith the PK model, exposure amounts are outputted in unitsof mg/total body weight in kg. Note that unlike the PK model’sdefault body weight of 70 kg, the user can input specific bodyweights for each model run, to better match empirical studydata. A more complete description on the use of this modelcan be found in the User’s Manual, available as supplementarymaterial to this manuscript.

4. Sample program output


4.1.1. Biomarker predictionsThe accuracy of the model to predict biomarker concen-trations over time (i.e.: forward predictions) was evaluated.Simulations of predicted biomarker concentrations for

several published datasets were made and the results shownin Fig. 4. The model does a reasonably good job of fit-ting urinary or blood biomarker concentrations for datasetsof chlorpyrifos, diazinon and/or butoxyethanol followingoral, dermal or inhalation exposures. Assuming that single-exposure evaluations via biomonitoring are generally doneduring or shortly after a workshift, refit values at timepointsof 0.5-, 1.0- and 2.0-fold the elimination rate value were calcu-lated and are shown in Table 1.

For oral or dermal administration of chlorpyrifos, modelsimulations were conducted with the published kabs, kexcr,FrxAbs and FrxExcr values reported by Nolan et al. [35]. Thepredicted blood and urine biomarker concentrations of the3,5,6-trichloropyridinol (TCP) biomarker in simulations of oralexposures, were within 25% of actual values [35]. Lower, butstill acceptable predictions were obtained for the dermal data,with all predictions within 70% of actual levels.

Diazinon biomarker concentrations in urine (sum oftotal dialkylphosphates (DAPs)), following oral administra-tion, were predicted using an estimated absorption half-lifeof 0.5 h, as no data was derived by the authors [36]. This shortabsorption half-life was chosen, based on the rapid elimina-tion half-life (2.13 h) and known rapid absorption of the relatedorganophosphate, chlorpyrifos [35]. Predicted biomarker lev-els were all within 30% of actual dose levels. Simulation ofdermal absorption was done using the same elimination t1/2

as oral uptake (2.13 h), and utilizing the reported 9.12 h elimi-nation t1/2 as the actual dermal uptake rate, assuming flip-flopkinetics for this route. The overprediction of DAP levels at0.5 × t1/2 of 4.2-fold measured concentration (Table 1) appearsto be due to a delay in actual micturition, since predicted levelsat 1- and 2-fold the elimination t1/2 were within 10% of actualvalues (Fig. 4).

Butoxyethanol (BE) biomarker concentrations (totalbutoxyacetic acid (BAA)) following inhalation exposure werecalculated with an absorption t1/2 estimated at 1 h, which wasequivalent to the t1/2 calculated for dermal uptake (below)since steady-state levels were not achieved via inhalation for0.5 h [37]. The elimination t1/2 reported by Kezic et al. [37] of3.4 h was used. Biomarker concentrations following dermalexposure were calculated with an absorption t1/2 estimated at1 h, based on steady blood BE levels achieved within 4 h [37],while the reported elimination t1/2 of 5.8 h was used. The PKmodel predictions at 0.5 – through 2.0-fold the eliminationt1/2 were all within 2.2-fold of actual levels for both routes.Note that the data from Kezic et al. was obtained by digitizingmean biomarker concentrations from Fig. 2A and B of theirmanuscript, with the application DigitizeIt v1.5.5 (DigitalRiver, Köln, Germany). Accuracy of data digitization was lessthan 0.5% error, as determined by comparison of calculatedand reported sample collection interval time-points. Theabsorption half-life of 1.0 h was selected for both the inhala-tion and dermal datasets, based on the reported time tosteady-state plasma butoxyethanol levels of 2–4 h (Fig. 1B ofKezic et al.).

4.1.2. Model assumptionsThe needed pharmacokinetic parameters may not always beavailable for the determination of exposure levels via theBiomonitoring PK model. However, exposure predictions could



A

Sampling Time

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

100 20 30 40

Blo

od

Co

nc (

mg

/L)

Time (hr)

Time Course of Parent Compound inBlood

0

10

20

30

40

50

60

70

80

90

500 100 150 200 250 300

Blo

od

Co

nc (

mg

eq

uiv

/L)

Time (hr)

Time Course of Metabolite in Blood

0

5

10

15

20

25

30

35

40

45

500 100 150

Uri

ne C

on

c (

mg

/L)

Time (hr)

Time Course ofParent Compound inExcreted Urine

0

100

200

300

400

500

600

700

500 100 150 200 250 300

Uri

ne

Co

nc

(m

g e

qu

iv/L

)

Time (hr)

Time Course of Metabolite in ExcretedUrine

B

C

D

Fig. 3 – Example time–course graphs from PBPK model of biomarkers in blood and urine, from low concentration to highconcentration (Note: Stepped output in urine graphs represents incorporation of micturition intervals in model).

still be made within this simple model if values for unknownparameters are estimated. The model that would require thefewest estimates for unknown parameters would be calcu-lation of exposure from a urine biomarker concentration atsteady-state exposure (Eq. (4)). In this case, urine productionrates would be fairly constant, leaving exposure levels depen-dent only on FrxExcr (fraction of dose excreted in urine asbiomarker). This value could be estimated from ADME data inanimals. Alternately, it could be set at a default value of 10%,which would give at most a one-order of magnitude error inexposure levels (assuming FrxExcr is at least 1%). This maxi-mum 10-fold error in exposure estimate, when using a defaultFrxExcr of 10%, is due to the fact that the calculated exposureamount is inversely proportional to FrxExcr, as defined in Eqs.(2) and (4) (Section 2.1, above). Note that for the three valida-tion compounds discussed above, FrxExcr ranged from 66% to70%.

The remaining three scenarios for the Biomonitoring PKmodel (single exposure blood, steady-state blood or urinelevel) require at least three PK parameters that are specificto a given chemical (kabs, kexcr, Vd, FrxAbs and/or FrxExcr).The Vd term can be estimated from QSPR applications, ifnot known. The FrxAbs and FrxExcr terms have only a linearimpact on exposure calculations, so again assuming a valueof 10% for either of these will only afford at most a 10-folderror in exposure calculations. As shown in the publisheddata for the three validation compounds, biomarker mea-surements in urine matrix are generally more common than

blood measurements for non-pharmaceutical compounds. Asa result, the determination of kexcr from human volunteerexposure studies is generally reported more often than kabs,as urine collection frequency is insufficient for calculation ofmost absorption rates. So for the single exposure scenariosor steady-state exposure calculations from blood levels, theremay be a need to estimate kabs for use in this BiomonitoringPK model.

An evaluation of the error in exposure calculations thatcould arise from inaccurate estimation of compound’s kabs(ln(2)/absorption t1/2) is shown in Fig. 6. Three of this study’stest data sets (diazinon-oral urine, chlorpyrifos-dermal urineand butoxyethanol-inhalation urine) were utilized to predictbiomarker concentrations at 0.1, 0.2, 0.5, 1.0, 2.0, 3.0 and5.0 times the reported absorption half-life. As expected, thelargest errors in biomarker concentrations when varying theabsorption t1/2 parameter were found in the dermal absorptiondataset. The highest overall error was found in the early time-range of all datasets. Maximal errors in predicted biomarkervalues, at timepoints out to five times the absorption t1/2,ranged from 4 to 10 times higher or lower than values fromthe simulation with the default absorption half-life.


4.2.1. Biomarker predictionsAs with the PK model, above, the accuracy of the PBPK modelto predict biomarker concentrations was also evaluated with



Chlorpyrifos (0.5 mg/kg,po)

0 50 1000.1

1

10

100

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/L u

rin

e)

Diazinon (0.011 mg/kg,po)

0 5 10 15 200.01

0.1

1

10

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/g c

rea

tin

ine

)

Chlorpyrifos (5 mg/kg, dermal)

0 50 1000.01

0.1

1

10

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/L u

rin

e)

Diazinon (100 mg, dermal)

0 10 20 30 400.001

0.01

0.1

1

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/g c

rea

tin

ine

)

Chlorpyrifos (0.5 mg/kg,po)

0 50 1000.01

0.1

1

10

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/L b

loo

d)

Butoxyethanol (93ppm x 0.5 hr, inhalation)

0 10 20 30 400.1

1

10

100

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/g c

rea

tin

ine

)

Chlorpyrifos (5 mg/kg, dermal)

0 50 1000.001

0.01

0.1

1

Time (hours)

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/L b

loo

d)

Butoxyethanol (8 mL x 4 hr, dermal)

0 10 20 30 401

10

100

1000

10000

Time (hours)

Bio

ma

rke

rC

on

ce

ntr

ati

on

(mg

/g c

rea

tin

ine

)

A

B

C

D

E

F

G

H

Fig. 4 – Comparison of PK and PBPK model fits to empirical biomonitoring data (red line (closed circles) = empirical data,green line = PK model prediction of biomarker time–course, blue line (open triangles) = PBPK model prediction of biomarkertime–course). (For interpretation of the references to color in this figure legend, the reader is referred to the web version ofthis article.)


782 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 773–788Ta

ble

1

–

Acc

ura

cy

of

pre

dic

ted

uri

ne

or

bloo

d

biom

ark

er

con

cen

trat

ion

s

for

chlo

rpyr

ifos

, dia

zin

on

and

buto

xye

than

ol

at

0.5,

1.0

and

2.0

tim

es

the

elim

inat

ion

t 1/2

.

Test

mat

eria

l

Bio

mar

ker

Exp

osu

re

rou

te

Sam

ple

mat

rix

Abs

orp

tion

hal

flif

e

(h)

Elim

inat

ion

hal

flif

e

(h)

Nu

mbe

r

ofvo

lun

teer

sR

efit

of

biom

arke

r

con

cen

trat

ion

a

PK

mod

elPB

PK

mod

el

0.5

×t 1

/21.

0

×t 1

/22.

0

×t 1

/20.

5

×t 1

/21.

0

×t 1

/22.

0

×t 1

/2

Ch

lorp

yrif

osb

TC

PO

ral

Uri

ne

0.5

26.9

681

%10

0%84

%85

%11

5%13

3%C

hlo

rpyr

ifos

TC

PD

erm

alU

rin

e22

.5

26.9

514

7%14

0%14

8%173

%20

6%12

0%C

hlo

rpyr

ifos

TC

PO

ral

Blo

od0.

5

26.9

612

3%11

5%10

3%84

%88

%98

%C

hlo

rpyr

ifos

TC

PD

erm

alB

lood

22.5

26.9

515

0%14

9%16

6%127

%12

4%94

%D

iazi

non

cD

APs

Ora

lU

rin

e0.

5

2.13

512

6%95

%96

%110

%96

%18

0%D

iazi

non

DA

PsD

erm

alU

rin

e9.

12

2.13

441

9%10

3%11

0%91

%78

%11

1%B

uto

xyet

han

old

BA

AIn

hal

atio

nU

rin

e1.

0

3.4

621

8%19

5%19

0%NA

190%

146%

Bu

toxy

eth

anol

BA

AD

erm

alU

rin

e1.

0

5.8

618

6%12

1%13

2%NA

118%

165%

aC

alcu

late

d

as

the

rati

o

of

pre

dic

ted

biom

arke

r/ac

tual

amou

nt

mea

sure

d

at

tim

e-p

oin

ts: X

-fol

d

the

elim

inat

ion

hal

f-li

fe

of

each

test

mat

eria

l’s

biom

arke

r.

Not

e:

use

d

uri

ne

coll

ecti

on

mid

poi

nts

for

PK

mod

el

com

par

ison

s

and

uri

ne

coll

ecti

on

tim

es

for

PBPK

mod

el

com

par

ison

s.b

Nol

an

et

al. [

35].

cG

arfi

tt

et

al. [

36].

Not

e:

Refi

ts

in

this

tabl

e

and

Fig.

2B

are

for

one

volu

nte

er

of

stu

dy

grou

p, f

or

each

exp

osu

re

rou

te, a

s

coll

ecti

on

tim

es

dif

fere

d

betw

een

volu

nte

ers.

For

oral

exp

osu

re, 0

.5

h

abso

rpti

on

hal

f-li

fe

valu

efr

om

rela

ted

orga

nop

hos

ph

ate

chlo

rpyr

ifos

use

d

for

calc

ula

tion

s

sin

ce

abso

rpti

on

valu

e

not

rep

orte

d

by

auth

ors.

For

der

mal

exp

osu

re,

abso

rpti

on

hal

f-li

fe

of

9.12

h

use

d

for

elim

inat

ion

hal

f-li

feva

lue

in

calc

ula

tion

s,

as

flip

-flop

kin

etic

s

assu

med

.d

Kez

ic

et

al. [

37].

Not

e:

Refi

ts

in

this

tabl

e

are

from

mea

n

valu

es

of

volu

nte

er

dat

a.

-3 -2 -1 1 2 3 4 5 6 7 8 9

-4

-3

-2

-1

1

2

3

4

5

6

7

8 Leo&Hansch 1971 (H-donor )

Leo&Hansch 1971 (H-acceptor )

New compounds

Kanazawa1983

Leo Hansch& QSPR

Combined QSPR

Datasets

log Koctanol:water

log

Ko

il:w

ate

r

LogKoil:water = -4.653 (7.972 / (1+10+ (0.1175-0.2849*Log Kow) ))

Poulin 2001

Fig. 5 – Correlation of log Koct:water and Koil:water valuesfor a series of chemicals, with regression equation topredict Koil:water from log Koct:water.

the test datasets for chlorpyrifos, diazinon and butoxyethanol.As shown in Fig. 4, the biomarker fits from the PBPK modelmatched the experimental data generally well. Note that thePK and PBPK model fits are also comparable to each other.This is most probably due to the chemical similarity of thethree biomarkers (TCP, DAPs, BAA), all fairly hydrophilic, withPtp values for all tissue compartments that ranged between0.5 and 1.5, and the fact that both models utilize first-orderuptake and elimination rates.

For the oral and dermal absorption routes for chlorpyri-fos, the same kabs and kexcr parameters were utilized as inthe simpler PK model. Note that the plasma kexcr for theparent compound was set equal to that of the trichloropy-ridinol metabolite. Alteration of this rate did not impact thebiomarker levels substantially, as the rate of metabolic loss ofchlorpyrifos is quite high, with ∼99% of total blood metabo-lites present as TCP [38]. The oral dose level was adjustedfor percent absorption value of 72% reported by Nolan et al.[35]. The PBPK model assumes an infinite dose volume fordermal exposures. To account for the finite amount of testmaterial solution used by Nolan et al. (8 �L of a 50% chlor-pyrifos solution/cm2 skin), the limited volume of applied doseconcentration of 50% (w/v) was multiplied by 10%, to afforda more realistic, real-world potential infinite dose concen-tration of 50 g/L (5%) solution. As described below, this doseconcentration used in the PBPK model afforded good fits withactual biomarker concentrations following dermal exposure,in both time–course and magnitude of TCP concentrations.The dermal penetration Kp value was taken from Timchalket al. [38]. In the original PBPK model published for this chem-ical by Timchalk et al. [38], the authors define numerous ratesfor the various metabolic conversions. The Vmax and Km termsfor the primary route of metabolism (chlorpyrifos → TCP) wereutilized in the Biomonitoring PBPK model for chemical conver-sion to metabolite. Also, since TCP is not further metabolized[39], the metabolic rates for conversion of this metabolite wereset to zero. Tissue-plasma partition coefficients for the parentchemical were derived from the QSPR within the PBPK model.The Ptp value for TCP was set to 0.5 for all tissues, consistentwith the Vd value of 35 L/total body weight used by Timchalket al. in their PBPK model of this biomarker [38].



0 10 200.0

0.5

1.0

1.5

2.0

Abs t1/2 = 1.5 hr

Abs t1/2 = 0.05 hr

Abs t1/2 = 0.1 hr

Abs t1/2 = 0.25 hr

Abs t1/2 = 0.5 hr

Abs t1/2 = 1.0 hr

Abs t1/2 = 2.5 hr

Time (hours)

Bio

mark

er

Co

ncen

trati

on

(mg

/L o

r m

g/g

cre

ati

nin

e)

0 25 50 75 100 125 1500.0

0.2

0.4

0.6

0.8

1.0

1.2Abs t1/2 = 2.52 hr

Abs t1/2 = 4.5 hr

Abs t1/2 = 11.25 hr

Abs t1/2 = 22.5hr

Abs t1/2 = 45 hr

Abs t1/2 = 67.5 hr

Abs t1/2 = 112.5 hr

Time (hours)

Bio

mark

er

Co

ncen

trati

on

(mg

/L o

r m

g/g

cre

ati

nin

e)

A

B

0 10 20 30 4005

101520253035404550556065

Abs t1/2 = 1.5 hr

Abs t1/2 = 0.2 hr

Abs t1/2 = 0.5 hr

Abs t1/2 = 1.0 hr

Abs t1/2 = 2.0 hr

Abs t1/2 = 3.0 hr

Abs t1/2 = 5.0 hr

Time (hours)

Bio

mark

er

Co

ncen

trati

on

(mg

/L o

r m

g/g

cre

ati

nin

e)

C

Diazinon- oraladministration(maximum error = 4.4 fold)

Chlorpyrifos- dermal administration(maximum error = 9.7 fold)

Butoxyethanol- inhalation exposure(maximum error = 5.2 fold)

Fig. 6 – Impact of estimating absorption rate: variation in predicted biomarker concentration time-courses at 0.1, 0.2, 0.5,1.0, 2.0, 3.0 and 5.0-fold the estimated or published oral absorption half-life values (actual values in green/thick line): (A)DAP metabolite levels following oral administration of diazinon; (B) TCP metabolite levels following dermal exposure tochlorpyrifos; (C) BAA metabolite levels following inhalation exposure to butoxyethanol. Note maximum errors are predictedout to 5 times the absorption half-life. In all cases, maximum errors listed found at 0.25 h (first timepoint of modelprediction).

The simulations of blood TCP levels with the PBPK model,from both exposure routes, were within 30% of actual values,while urine concentration predictions were generally within50% (Table 1). These relatively good fits for TCP with the PBPKmodel show that the use of the enzymatic rate constants forthe primary metabolic pathway of chlorpyrifos → TCP affordedan accurate representation of biomarker formation in human.

Diazinon DAP biomarker levels following oral absorptionwere estimated with the PBPK model using the same kabsand kexcr values as described in Section 4.1. The plasma kexcrvalue for dermal exposure was set equal to the urinary elim-ination rate observed by Garfitt et al. [36] and the dermal Kpvalue from Poet et al. [40]. The oral dose level was adjustedfor the percent absorption of 80%, calculated by Poet et al. As



discussed for chlorpyrifos above, the finite dose volume of20 �L of a 25% diazinon solution/cm2 skin used by Garfitt et al.[36], was adjusted to a lower concentration of the infinite dosevolume required by the PBPK model. The concentration of0.25% (w/v) was found to afford good predictions of urinaryDAP concentrations, and was therefore assumed appropriatefor this limited dose to infinite dose volume extrapolation.As was the case for chlorpyrifos, the Vmax and Km terms forthe primary route of metabolism, hepatic CYP-based deary-lation, affording IMHP and DAP metabolites, were used todescribe metabolic loss of parent compound. The Vmax formetabolism of the DAP metabolites was set to zero, as thesebiomarkers represent a high percentage of administered oraldose (66%) [36] and Timchalk et al. have showed that thesemetabolites are quite metabolically stable in the rat [39].Blood:tissue partition coefficients for the parent compoundand DAP metabolites were derived from the QSPR within thePBPK model.

The predictions of DAP metabolite levels of diazinon by thePBPK model were quite acceptable (Fig. 4). Refits at 0.5 and 1.0times the elimination t1/2, were within 30%, while fits at 2.0×the elimination t1/2 were within 85% of actual biomarker levels(Table 1).

The predictions of total BAA metabolite following inhala-tion or dermal exposure to BE are again shown in Fig. 4 (Gand H). The plasma elimination t1/2 for both parent compoundand BAA metabolite was set equal to the reported BAA uri-nary elimination rates for both routes. The dermal dose wasset equal to 100% solution (neat BE), as an infinite dose vol-ume, which is the default dose volume assumed by the model.The Vmax and Km terms for the primary route of metabolism(BE → BAA) were obtained from Corley et al. [41]. The Vmax

for subsequent metabolism of BAA was set to zero, as BAAis primarily conjugated with glutamine, once formed [42],and this conjugation step is not relevant, as total BAA levels(free + conjugated) were used in the PK and PBPK model simu-lations. [39]. Blood:tissue partition coefficients for the parentcompound and BAA metabolite were derived from the QSPRwithin the PBPK model.

Predictions of total BAA levels with the Biomonitoring PBPKmodel were all within 2-fold of measured concentrations(Fig. 4 and Table 1). No values were calculated at 0.5 × t1/2, asthe first urine samples collected were at 4 h for both exposureroutes.

4.2.2. Model assumptionsThe PBPK model was designed to require a minimal number ofPK parameters for prediction of exposure from blood or urinebiomarker levels. Chemical-specific parameters required forthe simpler PK model were kabs, kexcr, Vd, FrxAbs and/orFrxExcr. The PBPK model requires rates of absorption and elim-ination, as well as metabolic rates for at least the conversionof the parent chemical. Also, tissue:plasma partition coef-ficients are required for several tissue compartments, evenwhen interpreting urine biomarker data only.

As discussed in Section 4.1, the parameters needed to pre-dict biomarker concentrations with the PBPK model may notalways be available for a given chemical. In that case, esti-mates can still be made if assumptions are made for key model

parameters. The impact of these assumptions are shown inFig. 7.

In predicting the effect of an absorption rate for a chemicalup to 0.2–5-fold different than the actual value, a comparativeanalysis shows only a modest impact on predicted biomarkerlevels, with the maximum error in diazinon DAP metabo-lite values less than 4-fold of actual concentrations, whensimulated out to 5 times the elimination half-life (Fig. 7A).These results are quite comparable to the assumptions forthis parameter described for the PK model (above). In con-trast, the errors in predicted biomarker concentrations withvariations in elimination rates are more substantial. This errorranged up to ∼4000-fold when evaluations were carried out to5 times the elimination half-life of 2.1 h (Fig. 7B). This error wasmost pronounced when elimination half-lives were underes-timated. However, as discussed above, the elimination ratefor a biomarker can often be approximated from occupationalexposure studies, in which workers are isolated from chem-ical exposure to measure decay in blood or urine levels thatoccurred prior to the isolation.

The impact of estimating the Michaelis–Menton metabolicterms on predicted biomarker levels is shown in Fig. 7C and D.One-fifth to five-fold variations in either of these parametersafforded only ∼4-fold error in biomarker predictions, acrossthe time range of 0–5 elimination half-lives. This is probablydue to the fact that the low exposure amount (0.011 mg/kgbody weight) is in the linear (first-order) range of CYP-basedmetabolism of diazinon (<Km). With most environmentalexposures quite low, it is likely that first-order chemicalmetabolism could be assumed. Venkatakrishnan et al. [43]have discussed this approach as the “model-independentin vitro half-life approach”, and have mentioned that, forsubstrate concentrations below 1 �M, numerous groups havecalculated metabolic clearance in the liver as a linear processequal to Vmax/Km.

These model assumption results are consistent with alocalized sensitivity analysis, conducted in acslX for thechemical-specific model parameters of the diazinon (oral),chlorpyrifos (dermal) and butoxyethanol (inhalation) PBPKmodels. These sensitivity analysis results were obtained byvarying individual parameter values and determining theimpact on predicted maximal concentration of metabolite inurine. As shown in Table 2, the models were most sensitive tothe elimination rate of the respective metabolites. Other rates(absorption, metabolism of parent compound) were found tohave lower sensitivity coefficients, but still above 0.1 (or 10%of total parameter sensitivity).

4.3. QSPR for oil:water partition coefficient

The use of vegetable oil as a surrogate for tissue lipids has beenrecommended by researchers ever since early studies wereconducted on anesthetic tissue distributions over 75 years ago[44]. The results of the current Koil:water analyses conductedwith olive oil were found to be comparable to prior studies,with the measured log Koil:water value for p-ethylphenol of1.69 quite similar to the previously reported value of 1.79 [23].The new values for eight lipophilic compounds, derived witholive oil (Fig. 5), should also be comparable to the data fromother studies employing vegetable or fish oils [19,23,25], as



Fig. 7 – Impact of model parameter assumptions on diazinon urinary DAP metabolite data following oral administration at0.2, 0.5, 1.0, 2.0 and 5-fold the estimated or published PK values (actual values in green/thick line) across the time interval of5 elimination half-lives (0–11 h): (A) absorption half-life; (B) elimination half-life; (C) Vmax metabolic rate; (D) Km metabolicrate. Note: Stepped output represents incorporation of micturition intervals into model.

Table 2 – Sensitivity of the PBPK model parameters to prediction of maximal urine concentration of metabolite.

Parameter Normalized sensitivity coefficient

Diazinon oral Chlorpyrifos dermal Butoxyethanol inhalation

RatesAbsorption 0.2 0.0 0.0plasma elimination of metabolite 0.3 0.8 0.7Plasma elimination of parent compound 0.0 0.1 0.1Vmax for metabolite 0.0 0.0 0.0Vmax for parent compound 0.2 0.0 0.0Km for metabolite 0.0 0.0 0.0Km for parent compound 0.2 0.0 0.0Partition coefficients for parent compoundBlood:air 0.0 0.0 0.0Fat:blood 0.0 0.0 0.0GI:blood 0.1 0.0 0.0Kidney:blood 0.0 0.0 0.0Liver:blood 0.0 0.0 0.0Lung:blood 0.0 0.0 0.0Muscle:blood 0.0 0.0 0.1Richly perfused:blood 0.0 0.0 0.0Skin:blood 0.0 0.0 0.0Slowly perfused:blood 0.0 0.0 0.0Partition coefficients for metaboliteBlood:air 0.0 0.0 0.0Fat:blood 0.0 0.0 0.0GI:blood 0.0 0.0 0.0Kidney:blood 0.0 0.0 0.0Liver:blood 0.0 0.0 0.0Lung:blood 0.0 0.0 0.0Muscle:blood 0.0 0.0 0.0Richly perfused:blood 0.0 0.0 0.0Skin:blood 0.0 0.0 0.0Slowly perfused:blood 0.0 0.0 0.0



several groups have shown comparable partitioning betweenwater and oil from various vegetable or fish sources [23,25].

Chiou [45] also conducted similar Koil:water derivationsfor a variety of lipophilic compounds. However the synthetictriglyceride triolein (triglyceride of oleic acid) used as the oilin that study afforded log Koil:water values for several com-pounds substantially higher that those obtained with naturaloils (hexachlorobenzene: 4.39 in new data vs. 5.5 in Chiou; p,p-DDT: 2.4 in Kanazawa vs. 5.9 in Chiou). These differences maybe due to triolein’s lack of lower molecular weight, unsatu-rated and phospholipids that can be present in vegetable orfish oils. As a result, the data of Chiou were not included inthe new Koil:water regression analysis.

As shown in Fig. 5, the majority of the compounds eval-uated by Leo and Hansch show a fairly linear correlationbetween log Koct:water and log Koil:water. However, thelipophilicity for these compounds ranged from log Koct:waterof ∼−2 to 4. Since there is a great deal of research ongo-ing to evaluate the PK fate for fairly lipophilic compounds(log Koct:water 3–8), a sigmoidal regression that includedadditional lipophilic test chemicals from Kanazawa, Poulinand the current data was derived. This sigmoidal regressiongave a good fit to the data, where substantial nonlinearity inKoil:water values were observed above log Koct:water valuesof 5. A similar nonlinear trend was observed by Connell [46]in the data of Chiou [45], which Connell attributes to differ-ential solubilities of compounds due to physical properties ofthe octanol or triolein solvents (i.e.: solvent size). The overallcorrelation for the new regression curve, based on the datapresented in Fig. 5, was r2 = 0.90, with an average error of 0.38between predicted and actual log Koil:water values. Predictedlog Koil:water values, obtained from this QSPR equation, areused for calculations of Vd, individual tissue:plasma Ptp as wellas Pba values in the PK and PBPK models of this project.

5. Conclusions

A user-friendly set of computer-based modeling programshave been developed to aid in the estimation of chemicalexposure levels from biomonitoring datasets. The Excel-basedPK model can be used to solve for single- or steady-stateexposures. The simple interface of this model, and the require-ment for minimal chemical-specific parameters, provides theuser a convenient tool for exposure calculations. The PBPKmodel that has been developed also utilized an Excel pro-gram interface interacting with acslX coded models. Thisphysiologically-based model is similar to other generic PBPKmodels [47–49], but includes QSPRs for Kp, Ptp, and incorpo-rates micturition intervals and repeat model runs with reversedosimetry calculations of exposure levels. Both models wereverified using published data for two pesticides and one indus-trial chemical. Additional comparisons to datasets from othercompounds would extend the validation of these models tobroader chemical domains.

These models should be applicable for estimation ofchemical exposure levels, from blood or urine biomarkerconcentrations. These programs will also provide a goodplatform for training, to understand the impact of various

chemical-specific or physiological parameters on chemicalbiomarker levels in humans.

6. Program access

The PK and PBPK biomonitoring programs described in thismanuscript are available as supplemental material online(LINK). The PK model consists of a Microsoft Excel 2003 orhigher-compatible workbook file. The PBPK model also con-sists of an Excel 2007 or higher-compatible workbook file, withthe associated acslX generic PBPK model, in a compiled .dll fileformat. A detailed Users’s Manual for these programs is alsoincluded as supplemental material.

Conflict of interest

The authors at The Dow Chemical acknowledge that two ofthe chemicals used to validate the Biomonitoring PK and PBPKmodels (chlorpyrifos and butoxyethanol) are produced by thiscompany.

Funding source

The authors wish to thank Long-range Research Initiative Pro-gram of the European Chemical Industry Council (Cefic) forfunding this project.

Appendix A. Supplementary data

Supplementary data associated with this arti-cle can be found, in the online version, athttp://dx.doi.org/10.1016/j.cmpb.2012.04.014.

r e f e r e n c e s

[1] M. Gibaldi, D. Perrier, Pharmacokinetics, M. Dekker, NewYork, NY, 1975 (ISBN: 0824762649).

[2] J.C. Ramsey, M.E. Andersen, A physiologically baseddescription of the inhalation pharmacokinetics of styrene inrats and humans, Toxicology and Applied Pharmacology 73(1984) 159–175.

[3] COPHES, Consortium to Perform Human Biomonitoring on aEuropean Scale, 2010.

[4] B. Dumez, K. Van Damme, L. Casteleyn, Research on ethicsin two large human biomonitoring projects ECNIS andnewgeneris: a bottom up approach, Environmental Health: AGlobal Access Science Source 7Suppl1 (2008) S7.

[5] D.A. Haines, T.E. Arbuckle, E. Lye, et al., Reporting results ofhuman biomonitoring of environmental chemicals to studyparticipants: a comparison of approaches followed in twoCanadian studies, Journal of Epidemiology and CommunityHealth (2010).

[6] L.L. Needham, D.B. Barr, A.M. Calafat, Characterizingchildren’s exposures: beyond NHANES, Neurotoxicology 26(2005) 547–553.

[7] M.F. Reis, J. Tedim, P. Aguiar, et al., Online integrated solutionto collect data, generate information and manage events inthe human biomonitoring field, International Journal ofHygiene and Environmental Health 210 (2007) 403–406.



[8] L.L. Aylward, M. Lorber, S.M. Hays, Urinary DEHP, metabolitesand fasting time in NHANES, Journal of Exposure Analysisand Environmental Epidemiology 21 (2011) 615–624.

[9] D.P. Scher, B.H. Alexander, J.L. Adgate, et al., Agreement ofpesticide biomarkers between morning void and 24-h urinesamples from farmers and their children, Journal ofExposure Science & Environmental Epidemiology 17 (2007)350–357.

[10] S.M. Hays, R.A. Becker, H.W. Leung, et al., Biomonitoringequivalents: a screening approach for interpretingbiomonitoring results from a public health risk perspective,Regulatory Toxicology and Pharmacology: RTP 47 (2007)96–109.

[11] J. Cocker, K. Jones, J. Morton, H.J. Mason, Biomonitoring atthe UK Health and Safety Laboratory, International Journalof Hygiene and Environmental Health 210 (2007) 383–386.

[12] H.J. Clewell, Y.M. Tan, J.L. Campbell, M.E. Andersen,Quantitative interpretation of human biomonitoring data,Toxicology and Applied Pharmacology 231 (2008) 122–133.

[13] J.C. Lipscomb, E.V. Ohanian, Toxicokinetics and riskassessment, Informa Healthcare 71 (2006), ISBN: 0849337224.

[14] National Academies NRCCoHBfE, Human Biomonitoring forEnvironmental Chemicals, National Academies Press, 2006,ISBN: 9780309102728.

[15] S. Haddad, M. Pelekis, K. Krishnan, A methodology forsolving physiologically based pharmacokinetic modelswithout the use of simulation softwares, ToxicologicalLetters 85 (1996) 113–126.

[16] D.J. Marino, Physiologically based pharmacokinetic modelingusing microsoft excel and visual basic for applications,Toxicology Mechanisms and Methods 15 (2005) 137–154.

[17] F.J. Jongeneelen, W.F. Berge, A generic, cross-chemicalpredictive PBTK model with multiple entry routes runningas application in MS Excel; design of the model andcomparison of predictions with experimental results, TheAnnals of Occupational Hygiene 55 (2011) 841–864.

[18] K.S. Pang, M. Rowland, Hepatic clearance of drugs. I.Theoretical considerations of a “well-stirred” model and a“parallel tube” model. Influence of hepatic blood flow,plasma and blood cell binding, and the hepatocellularenzymatic activity on hepatic drug clearance, Journal ofPharmacokinetics and Biopharmaceutics 5 (1977) 625–653.

[19] P. Poulin, K. Schoenlein, F.P. Theil, Prediction of adiposetissue: plasma partition coefficients for structurallyunrelated drugs, Journal of Pharmaceutical Sciences 90(2001) 436–447.

[20] P. Poulin, F.P. Theil, A priori prediction of tissue:plasmapartition coefficients of drugs to facilitate the use ofphysiologically-based pharmacokinetic models in drugdiscovery, Journal of Pharmaceutical Sciences 89 (2000)16–35.

[21] T. Rodgers, D. Leahy, M. Rowland, Physiologically basedpharmacokinetic modeling. 1. Predicting the tissuedistribution of moderate-to-strong bases, Journal ofPharmaceutical Sciences 94 (2005) 1259–1276.

[22] T. Rodgers, M. Rowland, Physiologically basedpharmacokinetic modelling. 2. Predicting the tissuedistribution of acids, very weak bases, neutrals andzwitterions, Journal of Pharmaceutical Sciences 95 (2006)1238–1257.

[23] A. Leo, C. Hansch, D. Elkins, Partition coefficients and theiruses, Chemical Reviews 71 (1971) 525–616.

[24] S. Haddad, P. Poulin, K. Krishnan, Relative lipid content asthe sole mechanistic determinant of the adiposetissue:blood partition coefficients of highly lipophilicorganic chemicals, Chemosphere 40 (2000) 839–843.

[25] J. Kanazawa, Measurement of fat-water partition coefficientof pesticides and its evaluation, Seitai Kagaku 6 (1983) 3–7.

[26] B. Davies, T. Morris, Physiological parameters in laboratoryanimals and humans, Pharmaceutical Research 10 (1993)1093–1095.

[27] D.B. Barr, L.C. Wilder, S.P. Caudill, et al., Urinary creatinineconcentrations in the U.S. population: implications forurinary biologic monitoring measurements, EnvironmentalHealth Perspectives 113 (2005) 192–200.

[28] S.M. Hays, L.L. Aylward, J.S. LaKind, et al., Guidelines for thederivation of biomonitoring equivalents: report from thebiomonitoring equivalents expert workshop, RegulatoryToxicology and Pharmacology: RTP 51 (2008) S4–S15.

[29] T. Rodgers, M. Rowland, Mechanistic approaches to volumeof distribution predictions: understanding the processes,Pharmaceutical Research 24 (2007) 918–933.

[30] Y. Dancik, O.G. Jepps, M.S. Roberts, Physiologically basedpharmacokinetics and pharcodynamics of skin, in: M.S.Roberts, K.A. Walters (Eds.), Dermal Absorption and ToxicityAssessment, CRC Press, New York, NY, 2008.

[31] USEPA, Dermal Exposure Assessment: Principles andApplications, EPA/600/8-91-011B, January 1992, InterimReport.

[32] T. Peyret, P. Poulin, K. Krishnan, A unified algorithm forpredicting partition coefficients for PBPK modeling of drugsand environmental chemicals, Toxicology and AppliedPharmacology 249 (2010) 197–207.

[33] P. Poulin, K. Krishnan, A biologically-based algorithm forpredicting human tissue: blood partition coefficients oforganic chemicals, Human & Experimental Toxicology 14(1995) 273–280.

[34] B.C. Kelly, M.G. Ikonomou, J.D. Blair, et al., Food web-specificbiomagnification of persistent organic pollutants, Science(New York, NY) 317 (2007) 236–239.

[35] R.J. Nolan, D.L. Rick, N.L. Freshour, J.H. Saunders,Chlorpyrifos:, pharmacokinetics in human volunteers,Toxicology and Applied Pharmacology 73 (1984) 8–15.

[36] S.J. Garfitt, K. Jones, H.J. Mason, J. Cocker, Exposure to theorganophosphate diazinon: data from a human volunteerstudy with oral and dermal doses, Toxicology Letters 134(2002) 105–113.

[37] S. Kezic, W.J.A. Meuling, I. Jakasa, Free and total urinary2-butoxyacetic acid following dermal and inhalationexposure to 2-butoxyethanol in human volunteers,International Archives of Occupational and EnvironmentalHealth 77 (2004) 580–586.

[38] C. Timchalk, R.J. Nolan, A.L. Mendrala, et al., APhysiologically based pharmacokinetic andpharmacodynamic (PBPK/PD) model for theorganophosphate insecticide chlorpyrifos in rats andhumans, Toxicological Sciences: An Official Journal of theSociety of Toxicology 66 (2002) 34–53.

[39] C. Timchalk, A. Busby, J.A. Campbell, et al., Comparativepharmacokinetics of the organophosphorus insecticidechlorpyrifos and its major metabolites diethylphosphate,diethylthiophosphate and 3,5,6-trichloro-2-pyridinol in therat, Toxicology 237 (2007) 145–157.

[40] T.S. Poet, A.A. Kousba, S.L. Dennison, C. Timchalk,Physiologically based pharmacokinetic/pharmacodynamicmodel for the organophosphorus pesticide diazinon,Neurotoxicology 25 (2004) 1013–1030.

[41] R.A. Corley, G.A. Bormett, B.I. Ghanayem, Physiologicallybased pharmacokinetics of 2-butoxyethanol and its majormetabolite, 2-butoxyacetic acid, in rats and humans,Toxicology and Applied Pharmacology 129 (1994) 61–79.

[42] R.A. Corley, D.A. Markham, C. Banks, et al., Physiologicallybased pharmacokinetics and the dermal absorption of2-butoxyethanol vapor by humans, Fundamental andApplied Toxicology: Official Journal of the Society ofToxicology 39 (1997) 120–130.



[43] K. Venkatakrishnan, L.L. Von Moltke, D.J. Greenblatt, Humandrug metabolism and the cytochromes P450: application andrelevance of in vitro models, The Journal of ClinicalPharmacology 41 (2001) 1149–1179.

[44] J.K. Seydel, M. Wiese, Drug–Membrane Interactions:Analysis, Drug Distribution, Modeling, Wiley-VCH Verlag,GmbH, Weinheim, 2002.

[45] C. Chiou, Partition coefficients of organic compounds inlipid–water systems and correlations with fishbioconcentration factors, Environmental Science&Technology 19 (1985) 7–62.

[46] D.W. Connell, Bioaccumulation of Xenobiotic Compounds,CRC Press, Boca Raton, 1990.

[47] T.M. Cahill, I. Cousins, D. Mackay, Development andapplication of a generalized physiologically basedpharmacokinetic model for multiple environmentalcontaminants, Environmental Toxicology and Chemistry 22(2003) 26–34.

[48] S.S. De Buck, V.K. Sinha, L.A. Fenu, et al., Prediction ofhuman pharmacokinetics using physiologically basedmodeling: a retrospective analysis of 26 clinically testeddrugs, Drug Metabolism and Disposition: The Biological Fateof Chemicals 35 (2007) 1766–1780.

[49] R.H. Luecke, B.A. Pearce, W.D. Wosilait, et al., Windowsbased general PBPK/PD modeling software, Computers inBiology and Medicine 38 (2008) 962–978.

development of pk-and pbpk-basedmodeling tools for...

Documents