21
Bioaccumulation Assessment Using Predictive Approaches John W Nichols, * 7 Mark Bonnell,8 Sabcho D Dimitrov,6 Beate I Escher, I Xing Han,# and Nynke I Kramer77 7US Environmental Protection Agency, Office of Research and Development, National Health and Environmental Effects Research Laboratory, Mid-Continent Ecology Division, 6201 Congdon Boulevard, Duluth, Minnesota 55804 8Environment Canada, Existing Substances Division, Place Vincent Massey, 20th Floor, 351 Saint Joseph Boulevard, Gatineau, Quebec K1A 0H3, Canada 6University ‘‘Professor Assen Zlatarov’’, Laboratory of Mathematical Chemistry, 1 Yakimov Street, 8010 Bourgas, Bulgaria IDepartment of Environmental Toxicology (Utox), Swiss Federal Institute of Aquatic Science and Technology (Eawag), U ¨ berlandstrasse 133, PO Box 611, 8600 Du ¨bendorf, Switzerland #DuPont Haskell Global Centers for Health and Environmental Sciences, 1090 Elkton Road, Newark, Delaware 19714, USA 77Utrecht University, Institute for Risk Assessment Sciences, PO Box 80176, Utrecht, The Netherlands (Received 25 November 2008; Accepted 17 June 2009) EDITOR’S NOTE: This paper represents 1 of 9 papers generated from a SETAC Pellston Workshop entitled ‘‘Science-Based Guidance and Framework for the Evaluation and Identification of PBTs and POPs,’’ (January 2008, Florida, USA). The workshop objectives were to develop guidance and recommendations on the evaluation of substances fulfilling PBT and POP criteria, using scientific information such as experimental and monitoring data, and computer models. ABSTRACT Mandated efforts to assess chemicals for their potential to bioaccumulate within the environment are increasingly moving into the realm of data inadequacy. Consequently, there is an increasing reliance on predictive tools to complete regulatory requirements in a timely and cost-effective manner. The kinetic processes of absorption, distribution, metabolism, and elimination (ADME) determine the extent to which chemicals accumulate in fish and other biota. Current mathematical models of bioaccumulation implicitly or explicitly consider these ADME processes, but there is a lack of data needed to specify critical model input parameters. This is particularly true for compounds that are metabolized, exhibit restricted diffusion across biological membranes, or do not partition simply to tissue lipid. Here we discuss the potential of in vitro test systems to provide needed data for bioaccumulation modeling efforts. Recent studies demonstrate the utility of these systems and provide a ‘‘proof of concept’’ for the prediction models. Computational methods that predict ADME processes from an evaluation of chemical structure are also described. Most regulatory agencies perform bioaccumulation assessments using a weight-of-evidence approach. A strategy is presented for incorporating predictive methods into this approach. To implement this strategy it is important to understand the ‘‘domain of applicability’’ of both in vitro and structure-based approaches, and the context in which they are applied. Keywords: Bioaccumulation assessment QSAR modeling In vitro–in vivo extrapolation Metabolism Absorption INTRODUCTION Chemicals that persist (P) in the environment, bioaccu- mulate (B) in the tissues of biota, and are inherently toxic (T) are of special concern for chemical management. Global and regional policies that address PBT chemicals are described in detail by van Wijk et al. (2009). All of these policies can accommodate the use of predictive approaches to assess these chemical properties. We may expect, however, that reliance on predictive methods will increase as priorities move from well-studied legacy chemicals to chemicals for which there are little or no data. This report summarizes the findings of a workgroup that was organized as part of a SETAC Pellston Workshop entitled ‘‘Science-Based Guidance and Framework for the Evaluation and Identification of PBTs and POPs.’’ The charge to the workgroup was to review current and proposed uses of predictive methods in bioaccumulation assessments, and provide guidance for development of a weight-of-evidence approach that could employ these methods. The 1st section of this paper describes several global and regional policy initiatives that mandate bioaccumulation assessments for large numbers of compounds. These descrip- tions highlight current and proposed uses of predictive methods in the assessment process. The 2nd section reviews mathematical modeling approaches that are used to perform bioaccumulation assessments. Special emphasis is placed on the inadequacy of data needed to model absorption, distribution, metabolism, and elimination (ADME) processes, particularly for compounds that do not ‘‘fit’’ current prediction paradigms. The 3rd section describes the emerging use of in vitro test systems and quantitative structure-based approaches to predict ADME processes. Guidance for applying predictive approaches to bioaccumulation assess- ment is provided in the 4th and 5th sections. The 4th section gives general guidance on the use of predictive methods with an emphasis on emerging techniques, while the 5th section addresses the use of these methods within the context of a weight-of-evidence approach. The final section summarizes the findings of the workgroup and provides suggestions for future research. Throughout this text the reader will note that the discussion centers on bioaccumulation assessment for fish. The same principles apply, however, to bioaccumulation Special Series * To whom correspondence may be addressed: [email protected] Published on the Web 6/24/2009. Integrated Environmental Assessment and Management — Volume 5, Number 4—pp. 577–597 G 2009 SETAC 577

Bioaccumulation Assessment Using Predictive ApproachesBioaccumulation Assessment Using Predictive Approaches John W Nichols,*7 Mark Bonnell,8 Sabcho D Dimitrov,6 Beate I Escher,IXing

  • Upload
    others

  • View
    5

  • Download
    1

Embed Size (px)

Citation preview

Bioaccumulation Assessment Using Predictive ApproachesJohn W Nichols,*7 Mark Bonnell,8 Sabcho D Dimitrov,6 Beate I Escher,I Xing Han,# and Nynke I Kramer77

7US Environmental Protection Agency, Office of Research and Development, National Health and Environmental Effects ResearchLaboratory, Mid-Continent Ecology Division, 6201 Congdon Boulevard, Duluth, Minnesota 558048Environment Canada, Existing Substances Division, Place Vincent Massey, 20th Floor, 351 Saint Joseph Boulevard, Gatineau,Quebec K1A 0H3, Canada

6University ‘‘Professor Assen Zlatarov’’, Laboratory of Mathematical Chemistry, 1 Yakimov Street, 8010 Bourgas, BulgariaIDepartment of Environmental Toxicology (Utox), Swiss Federal Institute of Aquatic Science and Technology (Eawag), Uberlandstrasse133, PO Box 611, 8600 Dubendorf, Switzerland#DuPont Haskell Global Centers for Health and Environmental Sciences, 1090 Elkton Road, Newark, Delaware 19714, USA77Utrecht University, Institute for Risk Assessment Sciences, PO Box 80176, Utrecht, The Netherlands

(Received 25 November 2008; Accepted 17 June 2009)

EDITOR’S NOTE:This paper represents 1 of 9 papers generated from a SETAC Pellston Workshop entitled ‘‘Science-Based Guidance and

Framework for the Evaluation and Identification of PBTs and POPs,’’ (January 2008, Florida, USA). The workshop objectives were to

develop guidance and recommendations on the evaluation of substances fulfilling PBT and POP criteria, using scientific

information such as experimental and monitoring data, and computer models.

ABSTRACTMandated efforts to assess chemicals for their potential to bioaccumulate within the environment are increasingly

moving into the realm of data inadequacy. Consequently, there is an increasing reliance on predictive tools to complete

regulatory requirements in a timely and cost-effective manner. The kinetic processes of absorption, distribution,

metabolism, and elimination (ADME) determine the extent to which chemicals accumulate in fish and other biota. Current

mathematical models of bioaccumulation implicitly or explicitly consider these ADME processes, but there is a lack of data

needed to specify critical model input parameters. This is particularly true for compounds that are metabolized, exhibit

restricted diffusion across biological membranes, or do not partition simply to tissue lipid. Here we discuss the potential of

in vitro test systems to provide needed data for bioaccumulation modeling efforts. Recent studies demonstrate the utility of

these systems and provide a ‘‘proof of concept’’ for the prediction models. Computational methods that predict ADME

processes from an evaluation of chemical structure are also described. Most regulatory agencies perform bioaccumulation

assessments using a weight-of-evidence approach. A strategy is presented for incorporating predictive methods into this

approach. To implement this strategy it is important to understand the ‘‘domain of applicability’’ of both in vitro and

structure-based approaches, and the context in which they are applied.

Keywords: Bioaccumulation assessment QSAR modeling In vitro–in vivo extrapolation Metabolism Absorption

INTRODUCTIONChemicals that persist (P) in the environment, bioaccu-

mulate (B) in the tissues of biota, and are inherently toxic (T)are of special concern for chemical management. Global andregional policies that address PBT chemicals are described indetail by van Wijk et al. (2009). All of these policies canaccommodate the use of predictive approaches to assess thesechemical properties. We may expect, however, that relianceon predictive methods will increase as priorities move fromwell-studied legacy chemicals to chemicals for which thereare little or no data.

This report summarizes the findings of a workgroup thatwas organized as part of a SETAC Pellston Workshop entitled‘‘Science-Based Guidance and Framework for the Evaluationand Identification of PBTs and POPs.’’ The charge to theworkgroup was to review current and proposed uses ofpredictive methods in bioaccumulation assessments, andprovide guidance for development of a weight-of-evidenceapproach that could employ these methods.

The 1st section of this paper describes several global andregional policy initiatives that mandate bioaccumulationassessments for large numbers of compounds. These descrip-tions highlight current and proposed uses of predictivemethods in the assessment process. The 2nd section reviewsmathematical modeling approaches that are used to performbioaccumulation assessments. Special emphasis is placed onthe inadequacy of data needed to model absorption,distribution, metabolism, and elimination (ADME) processes,particularly for compounds that do not ‘‘fit’’ currentprediction paradigms. The 3rd section describes the emerginguse of in vitro test systems and quantitative structure-basedapproaches to predict ADME processes. Guidance forapplying predictive approaches to bioaccumulation assess-ment is provided in the 4th and 5th sections. The 4th sectiongives general guidance on the use of predictive methods withan emphasis on emerging techniques, while the 5th sectionaddresses the use of these methods within the context of aweight-of-evidence approach. The final section summarizesthe findings of the workgroup and provides suggestions forfuture research. Throughout this text the reader will note thatthe discussion centers on bioaccumulation assessment for fish.The same principles apply, however, to bioaccumulation

Sp

ecia

lSerie

s

* To whom correspondence may be addressed: [email protected]

Published on the Web 6/24/2009.

Integrated Environmental Assessment and Management — Volume 5, Number 4—pp. 577–597G 2009 SETAC 577

assessments for other animals, including terrestrial wildlifeand humans.

CURRENT USE OF PREDICTIVE METHODS FORBIOACCUMULATION ASSESSMENT IN SUPPORT OFGLOBAL AND REGIONAL POLICY INITIATIVES

Stockholm Convention

The United Nations Environment Programme (UNEP)global Stockholm Convention (UNEP 2004) addressedpersistent organic pollutants that can distribute within theenvironment by long-range transport. Annex D of theStockholm Convention states that a party wishing to list achemical in Annexes A or B shall provide information on thechemical and, where relevant, its transformation products,relating to specific screening criteria. Annex D does not statethat predictive approaches may be used to assess bioaccumu-lation potential, as it does for evaluating a compound’spotential for long-range transport. However, predictiveapproaches have been discussed by the UNEP PersistentOrganic Pollutant Review Committee as a viable approach forperforming bioaccumulation assessments (D. van Wijk,personal communication).

Categorization and assessment of the Canadian DomesticSubstances List

The Canadian Environmental Protection Act 1999 (CEPA1999) mandated that the Ministers of Environment Canadaand Health Canada provide a categorization (prioritization) ofthe Canadian Domestic Substances List (DSL) by September2006. The completed ecological categorization for nonhumanorganisms used bioaccumulation criteria specified in thePersistence and Bioaccumulation Regulations (Governmentof Canada 2000) and followed guidance provided byEnvironment Canada (EC 2003). While empirical data aregenerally preferred by Environment Canada, few empiricalbioaccumulation data exist for the majority of substances onthe DSL (Arnot and Gobas 2006). Therefore, 97% ofapproximately 12000 discrete organic substances on theDSL were categorized for bioaccumulation using predictivemethods. Environment Canada continues to use predictivemethods for bioaccumulation assessment of priority substanc-es on the DSL and has begun to consider in vitro data in anoverall weight-of-evidence scheme (ECHC 2008). Predic-tive approaches are also used to assess the bioaccumu-lation potential of new substances according to the NewSubstance Notification Regulations (Government of Canada2005).

REACH

In June 2007, REACH (Registration, Evaluation andAuthorisation of Chemicals) came into force in the EuropeanUnion. Annex XII of the legislative proposal for REACHgives the legal mandate for assessing the PBT properties ofchemicals while Annex IX provides for the use of computa-tional prediction methods. Nonvertebrate animal testing isadvocated under REACH, and an Intelligent/IntegratedTesting Strategy for bioaccumulation assessment was devel-oped by the multistakeholder REACH ImplementationProject endpoint group for bioaccumulation (ECHA 2008).A similar testing strategy was proposed by de Wolf et al.(2007). Both strategies advocate the use of predictivemethods at an early stage in a tiered assessment approach.

Programs administered by the United States EnvironmentalProtection Agency

Longstanding concern for PBTs exists within several UnitedStates Environmental Protection Agency (USEPA) ProgramOffices including the Office of Water, Office of Air andRadiation, and Office of Pollution Prevention and ToxicSubstances, as well as the Superfund Program. Managementapproaches used by individual programs differ in detail. All,however, employ predictive methods to deal with datadeficiencies. Here we highlight one such activity—premanu-facture notification (PMN) review of industrial chemicalsunder the Toxic Substances Control Act.

In 1999, USEPA issued a policy statement applicable tonew compounds subject to review and regulation under theToxic Substances Control Act. The policy statement providedguidance criteria for persistence, bioaccumulation, andtoxicity for new chemicals and advised industry aboutUSEPA’s regulatory approach for chemicals that meet thesecriteria (64 FR 60194; November 4, 1999; www.epa.gov/oppt/newchems/pubs/pbtpolcy.htm). However, because testdata on PMN chemical substances are not required, USEPAtypically receives few PMNs that contain sufficient data onhealth or environmental effects, or on the potential to persistor bioaccumulate in the environment. As a result, the agencyoften relies on computer models (e.g., quantitative structure–activity relationships [QSARs]) and data for structural orfunctional analogues to predict the toxicity and environmen-tal fate of PMN chemical substances.

COMPUTATIONAL MODELINGOF BIOACCUMULATION

Several mathematical modeling approaches for predictingchemical bioaccumulation in biota have been developed inthe last 35 years. Detailed reviews of quantitative structure–activity models (Devillers et al. 1998; Pavan et al. 2008),mass-balance models (Barber 2003), and food web bioaccu-mulation models (Burkhard 1998; Gobas and Morrison 2000,Mackay and Fraser 2000) have been provided. Here wesummarize these approaches, noting limitations that mayexist for predicting bioaccumulation across a diverse array ofchemicals, species, and exposure conditions.

Quantitative structure–activity relationships

When measured bioconcentration factors (BCFs) for fishare plotted against the log of a compound’s n-octanol–waterpartition coefficient (log KOW) a curvilinear relationship isobtained; BCFs tend to increase with log KOW up to a logKOW value of about 6 and then level off or decline at log KOW

values greater than 6 (Connell and Hawker 1988; Meylan etal. 1999). A number of empirical QSAR models have beendeveloped by correlating the BCF and log KOW using linearand nonlinear ‘‘best-fit’’ techniques (Neely et al. 1974; Veithet al. 1979; Mackay 1982; Connell and Hawker 1988). Theuse of log KOW as a predictor variable reflects the fact thathydrophobic organic compounds tend to partition to tissuelipid. In this context, n-octanol has been viewed as a surrogatefor this lipid.

Early QSAR models of bioconcentration tended to lump allcompounds into one fitted equation. More recent modelshave taken specific chemical properties into account (e.g.,ionization potential) as a means of selecting the ‘‘best’’ logKOW-BCF relationship for a particular substance (Meylan etal. 1999). The QSAR modeling approach also has been

578 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

adapted to consider potential limitations on bioconcentrationcaused by metabolism or restricted absorption. This has beenaccomplished by incorporating empirically determined ad-justment factors that operate mathematically against the‘‘baseline’’ level of accumulation expected from simplepartitioning considerations (Dimitrov, Mekenyan et al.2002; Dimitrov, Dimitrova, Parkerton et al. 2005).

Mass-balance models

Mass-balance models predict the kinetics of accumulationwithin one or more body ‘‘compartments’’ by simulating theprocesses of chemical uptake and elimination. Early modelsrepresented these processes using lumped, 1st-order rateconstants (typically denoted k1 and k2; Hamelink et al. 1971;Branson et al. 1975), and methods were developed toestimate these rate constants experimentally. Over time,however, an effort was made to predict uptake andelimination from a mechanistic understanding of the process-es themselves. For example, it was suggested that the rateconstant for chemical uptake from water could be calculatedas a fraction of the water flow required to meet an animal’srespiratory demand for oxygen (Neely 1979). In a similarmanner, dietary uptake was calculated as the product offeeding rate, the chemical concentration in food, andchemical absorption efficiency (Bruggeman et al. 1981).

Figure 1 illustrates the mass-balance concept. The mass-balance model that corresponds to this figure may be writtenas (assuming that chemical exchange across the skin isnegligible and ignoring the contribution of chemical fromsediment-associated porewater; Arnot and Gobas 2004):

dMFISH~ WFISH: k1

: Q:CWð ÞzkD:X

Pi:CD,ið Þ

h in o

{ k2zkEzkMð Þ:MFISH

ð1Þ

where MFISH is the mass (g) of chemical in the fish, WFISH isthe fish weight (kg) at time t, k1 is the clearance rate constant(L/kg/d) for chemical uptake across the gills, Q (unitless) isthe fraction of the total chemical concentration in water thatis bioavailable (i.e., freely dissolved and neutral), CW is thetotal chemical concentration in water (g/L), kD is theclearance rate constant (kg/kg/d) for chemical uptake fromingested food and water, Pi is the fraction of the dietconsisting of prey item i, CD,i is the concentration of chemical(g/kg) in prey item i, k2 is the rate constant (d21) for chemicalelimination via the gills, kE is the rate constant (d21) forchemical excretion to egested feces, and kM is the rateconstant (d21) for biotransformation. Under steady-stateconditions (dMFISH/dt 5 0), Equation 1 simplifies to

CFISH~ k1: Q:CWð ÞzkD

:X

Pi:CD,ið Þ

h i.k2zkEzkMzkGð Þ ð2Þ

where CFISH is the chemical concentration in the fish (g/kg)and kG is a growth rate constant (d21) that accounts forgrowth dilution of accumulated residues.

Food web bioaccumulation models

Mass-balance models can be used to predict chemicalbioconcentration resulting from a water-only exposure.Alternatively, they may be incorporated into larger descrip-tions of aquatic food webs (Thomann et al. 1992; Gobas1993; Arnot and Gobas 2004) to predict chemical bioaccu-mulation resulting from simultaneous waterborne and dietaryexposures (characterized by a bioaccumulation factor, orBAF). Additional mass-balance models for terrestrial ‘‘air-breathers’’ have been developed to predict bioaccumulationin terrestrial food webs (Kelly and Gobas 2001, 2003). Thesemodels may be modified to consider incidental soil ingestionfor species that consume soil-dwelling organisms (Armitageand Gobas 2007).

In general, mass-balance models have been developed andcalibrated for compounds that exhibit ‘‘simple’’ kineticbehaviors such as passive diffusion across biological mem-branes, nonspecific partitioning among tissues, and limitedmetabolism. As with the QSAR modeling approach, it ispossible to identify a baseline condition for each log KOW

value against which metabolism and/or restricted diffusioncould be expected to operate. If either or both of these factorslimit chemical accumulation at more than one trophic level,they may counteract the tendency of compounds to biomag-nify in food webs and cause chemical concentrations todecrease with increasing trophic level. Finally, it should benoted that the distinction made here between QSAR andmass-balance models may become blurred, depending on thetype of model and its application. Virtually all mass-balancemodels used for bioaccumulation assessment contain embed-ded QSAR relationships (e.g., to predict branchial uptake as afunction of log KOW). Indeed, it is possible to formulate amass-balance model so that all major uptake and eliminationprocesses are described by QSARs (Arnot and Gobas 2003).

Need to improve current representation of ADME processes

All models of chemical bioconcentration or bioaccumula-tion explicitly (via parameters like k1, k2, kM, etc.) orimplicitly (as when using KOW to predict chemical partition-ing to tissue lipid) incorporate knowledge of ADMEprocesses. Of these processes, metabolism has long beenrecognized as an important source of uncertainty in BCF/BAFpredictions (Southworth et al. 1980; Oliver and Niimi 1985;Clark et al. 1990; de Wolf et al. 1992). The sensitivity ofmass-balance models to changes in kM can be shown byvarying its value and examining the impact on model outputs(Figure 2). Metabolism has little potential to impact theaccumulation of neutral organic compounds with log KOW

values less than about 3. For these lower log KOW

compounds, chemical flux across the gills is rapid enough toreplace any compound that is cleared by metabolism. Incontrast, metabolism can substantially impact the accumula-tion of compounds with log KOW values greater than 3(Nichols, Fitzsimmons et al. 2007).

Apparent kM values for fish have been estimated by fittingmodel simulations to observed rates of depuration (Sijm et al.1990; de Wolf et al. 1993; Gert-Jan de Maagd et al. 1998; Fisket al. 2000; Tolls et al. 2000; Konwick et al. 2006) andmeasured BCF values (Arnot et al. 2008). Quantitative

Figure 1. Conceptual diagram of chemical uptake and elimination process-es for fish. Kinetic rate constants are described in the text.

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 579

structure–activity relationship models have been used in aconceptually similar manner to estimate probabilities ofmetabolism by specific metabolic pathways (Dimitrov,Dimitrova, Parkerton et al. 2005). These ‘‘top-down’’

approaches can be used to make comparisons within andamong different chemical classes. Arnot et al. (2008)cautioned, however, that metabolism rates determined inthis manner are dependent on the specification of othermodel parameters (e.g., fish size and temperature) and aretherefore context specific. Alternatively, it may be possible topredict kM values for fish from the ‘‘bottom up’’ byextrapolating measured in vitro rates of metabolism to thewhole animal (Nichols et al. 2006).

Model parameters that depend entirely or in part onchemical transport across biological membranes of fish (gills,gut; e.g., k1, k2, kD, kE) represent a 2nd potential source oferror in BCF/BAF predictions. Mass-balance models ofchemical bioaccumulation use KOW-based relationships topredict chemical flux across the gills and gut. Recent modelsalso incorporate knowledge of fish physiology to account forwater flow limitations on chemical uptake at the gills and toadjust for the digestibility of ingested food items (a factorused to estimate the fecal egestion constant, kE; Arnot andGobas 2004). Despite this sophistication, however, there isconsiderable uncertainty in these relationships, particularly asthey apply to high molecular weight and/or extremelyhydrophobic materials.

USE OF IN VITRO SYSTEMS AND STRUCTURE-BASEDAPPROACHES TO PREDICT ADME PROCESSES

Absorption and elimination

Chemical absorption and elimination occur as a result ofseveral membrane transport processes. Transcellular transportoccurs by passive diffusion across biological membranes. Theparacellular pathway refers to the transfer of hydrophilicmolecules through tight junctions between epithelial cells.

Active carrier-mediated transport is relevant for compoundsthat mimic certain biological substrates, and is especiallyimportant as a pathway for elimination. Vesicular transport(exocytosis and endocytosis) is typically relevant for macro-molecular particles but may be important for industrialnanoparticles and compounds sorbed to natural colloids.

The exchange surfaces of fish (gills, skin, gut) differmarkedly in structure and function. Chemical diffusiongradients across these surfaces may also differ substantiallydepending on the nature of an exposure. In most instances,however, passive diffusion is the primary means by whichchemicals cross these surfaces. This is particularly true forvery hydrophobic compounds, including those of concern formany PBT assessments. Current efforts to predict absorptionand elimination generally focus, therefore, on prediction ofpassive diffusion.

Previously, the curvilinear relationship between log BCFand log KOW was explained in part by a hydrophobicity orsize cutoff for chemical partitioning to biological membranes(characterized by the log of the membrane-water partitioningcoefficient, KMW; Opperhuizen et al. 1985; Gobas, Lahitteteet al. 1988). Support for this conclusion was provided bystudies with artificial membranes which showed that logKMW declines with log KOW at high log KOW values (Gobas,Lahittete et al. 1988; Dulfer and Govers 1995; Shimizu et al.2002). More recently, Jonker and van der Heijden (2007)suggested that the leveling off of the relationship between logKMW and log KOW noted in earlier work was due tononequilibrium artifacts, while the decrease of log KMW athigher log KOW values was due to 3rd-phase artifacts. Thesesuggestions have been supported by studies that show a linearcorrelation between log KMW and log KOW up to a log KOW

value of 7.8 (Jabusch and Swackhamer 2005; Jonker and vander Heijden 2007).

While there is no clear evidence of hydrophobicity or sizecutoffs for chemical partitioning to artificial cell membranes,there is good evidence of hydrophobicity or size limitationson chemical uptake across epithelial tissues of the gills and gut(McKim et al. 1985; Gobas, Muir et al. 1988). Using a seriesresistance model, Gobas and coworkers explained theobserved dependence of branchial uptake on log KOW as thenet result of diffusion resistances within the gill membraneand an unstirred aqueous boundary layer (Gobas et al. 1986;Gobas and Mackay 1987). At low (,1) log KOW values themembrane resistance was thought to control the rate ofuptake while at higher (.3) log KOW values diffusion throughan aqueous boundary layer was considered to be the rate-limiting factor. In practice, these rate limitations weremodeled as factors that determine the efficiency of chemicaluptake from inspired water. Using this approach, the value ofk1 may be viewed as the product of ventilation rate (i.e., therate of chemical delivery to the gills) and a log KOW-dependent adjustment factor. A conceptually similar ap-proach was used to model chemical exchange across thegastrointestinal epithelium (Gobas, Muir et al. 1988). Bothdescriptions are consistent with the general model of Flynnand Yalkowsky (1972) for permeation of nonelectrolytesacross membrane-diffusion layer systems.

An alternative description of chemical flux across fish gillswas provided by Erickson and McKim (1990a, 1990b). Thisdescription considered rate limitations imposed by the capac-ities of blood and water flows to the gills to deliver and removechemical. Based on this work it was concluded that blood and

Figure 2. Modeled effects of metabolism on chemical bioaccumulation infish. Simulations were obtained assuming kM values of 0.00 (solid line), 0.01(upper dashed line), 0.1 (dot-dashed line), 1.0 (dot-dot-dashed line), or 10.0/d (lower dashed line). Information sources: PCBs (red squares) and pesticides(green dots) in Lake Ontario lake trout (Oliver and Niimi 1988); PCDDs/PCDF(blue triangles) and non-ortho PCBs (turquoise triangles) in Lake Ontario lake

trout (USEPA 1995); PAHs (purple diamonds) in Lake Michigan lake trout(U.S. EPA, Mid-Continent Ecology Division, unpublished). All BAF values arenormalized for fish lipid content and the freely dissolved chemicalconcentration in water (L. Burkhard, with permission).

580 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

water flows substantially determine the rate of branchial flux,and by themselves result in the observed dependence of uptakeon chemical KOW, although membrane permeability can be anadditional rate-limiting factor. It was also suggested thatreduced uptake of very hydrophobic compounds may becaused by binding to dissolved organic matter, which reducesthe concentration of freely dissolved chemical species. Thepotential for binding to limit chemical uptake across the gillshad previously been noted by Gobas et al. (1989).

The fundamental basis of rate limitations on chemical fluxacross the gastrointestinal epithelium remains poorly under-stood. Grossly, this process behaves as though simplediffusion controls the rate of chemical uptake from food(Gobas et al. 1993; Nichols et al. 2004). The true nature ofthis ‘‘diffusion limitation’’ is unclear, however, and mayinvolve the formation, movement, and processing of lipidmicelles which sequester and transport hydrophobic com-pounds (Kleinow and James 2001). Assimilation efficienciesfor chemical uptake from food tend to decline at log KOW

values greater than about 6 (Gobas, Muir et al. 1988;Thomann 1989). Lacking a better understanding of theprocesses responsible for this pattern, current models ofbioaccumulation simulate this log KOW dependence usingfitted empirical relationships (Arnot and Gobas 2004).

Studies with artificial membranes have demonstrated aclear dependence of membrane flux on molecular size (Xiangand Anderson 1994). This finding is consistent with asubstantial body of literature showing that the coefficient ofdiffusion in a fluid is strongly affected by molecular size.Because the composition of biological membranes changeswith factors such as acclimation temperature and diet, theexact nature of this membrane size dependence will varysomewhat among individual organisms.

Structure-based modeling of membrane permeability—Earlyattempts to explicitly consider molecular size effects onchemical accumulation tended to focus on molecular weightas a descriptor variable, and several ‘‘cutoff’’ values wereproposed ranging from 700 to 1000 g/mol (reviewed byDimitrov et al. 2003). More recently, characteristics ofmolecular geometry have been modeled using structure-basedapproaches to investigate size restrictions on chemicalaccumulation. Generally, these efforts focus on the definitionof an effective molecular cross-section. Several definitionshave been proposed including (Opperhuizen et al. 1985;Schuurmann 1990):

N Maximum diameter, DMAX: the largest estimated lengthof the molecule such that the respective perpendiculardiameter, i.e., DEFF, is minimized.

N Effective cross section, ECS (or effective diameter, DEFF):the minimized cross-section or diameter of the moleculethat is perpendicular to DMAX.

N Minimum diameter, DMIN: the effective 3rd largestdiameter, perpendicular to both DMAX and DEFF.

These definitions were recently restated by Dimitrov et al.(2003) resulting in the following mutually independentdescriptions of molecular size:

N Maximum diameter, DMAX: the minimum diameter of asphere circumscribing the molecule.

N Effective diameter, DEFF: the minimum diameter of aninfinite cylinder circumscribing the molecule.

N Minimum diameter, DMIN: the minimum distancebetween 2 parallel planes circumscribing the molecule.

Historically, ECS was the 1st geometric characteristic used toexplain the reduced accumulation of lipophilic chemicals byfish. Opperhuizen et al. (1985) proposed an ECS threshold of0.95 nm for membrane permeation based on an analysis ofmeasured BCF values for nonflexible mono- and polychlori-nated benzenes, naphthalenes, and biphenyls. Subsequentstudies of larger chemical data sets have led investigators topropose other geometric thresholds. For example, Dimitrov,Dimitrova et al. (2002) analyzed the effects of molecularweight, ECS, DMAX, and DMIN on the bioconcentration ofapproximately 100 highly lipophilic chemicals. This analysissuggested that DMAX was the best predictor of bioconcentra-tion and that chemicals with a DMAX greater than ,1.5 nmhave a reduced potential for uptake.

An alternative approach to modeling molecular size effectson chemical accumulation in fish assumes that there is asmooth effect of size on membrane permeability (Figure 3).This approach was implemented in the BCF baseline modelgiven by Dimitrov, Dimitrova, Parkerton et al. (2005). Theeffect of molecular size was accounted for by calculating theprobability of a molecule crossing the cell membrane:

FMS~1{1

1ze{c(DMAX,AVG{DMAX,THR)ð3Þ

Here FMS is the ‘‘mitigating factor of molecular size,’’ whichwas used as an input to the BCF calculation. The term FMS isequal to 1 minus the probability of membrane permeation(the quotient term on the right side of this equation) whereDMAX,AVG is the average maximum diameter across energet-ically stable conformers, DMAX,THR is the diameter at whichthe probability has an inflection point, and parameter cdetermines the slope of the probability as a function ofDMAX,AVG. The model was fitted to a dataset consisting ofmore than 700 fish BCF values. After accounting for theeffects of metabolism, ionization, water solubility, andmolecular size, the adjusted value for DMAX,THR was 1.33± 0.10 nm.

In vitro permeability testing—Several in vitro test systemshave been developed to predict oral uptake of drugs by

Figure 3. Predicted effect of molecular size on cell membrane permeation.DMAX,THR is the average maximum diameter across energetically stableconformers at which the probability of membrane permeation (FMS)equals 50%.

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 581

humans. Perhaps the best known of these systems is thehuman intestinal epithelial Caco-2 cell line (Artursson et al.2001). All processes of intestinal uptake are simultaneouslyassessed with this assay, and since the cells express certaintransporters and metabolic enzymes, even some informationon metabolism can be derived (Shah et al. 2006).

The application of Caco-2 cells to compounds of environ-mental concern has resulted in contradictory findings. Oomenet al. (2001) used Caco-2 cells to model the bioavailabilityand uptake of PCBs and lindane from soil in children. Theinsert filters used to grow the Caco-2 cell monolayer had ahigh resistance to chemical permeation and the cells tended toretain the compounds. This study demonstrates some of theproblems that may occur when cell-based membrane assaysare applied to hydrophobic compounds. In contrast, Dulfer etal. (1998) used Caco-2 cells to study gut uptake of PCBs. Theauthors concluded that the resistance of the membrane wasnegligible for all PCBs investigated and that diffusion over theunstirred water layer was overcome by micellar transport.The more hydrophobic PCBs were transported within Caco-2cells by lipoproteins and triacylgylceride particles. For thelower chlorinated PCBs, the aqueous and lipid resistanceswithin the cells were similar.

It is unclear whether information obtained using Caco-2cells can be used to predict chemical uptake in fish. Thetransepithelial resistance of Caco-2 cells (230 V cm2) resem-bles that of both mammalian intestines (20–100 V cm2) andfish intestines (25–50 V cm2) (Kramer 2006; de Wolf et al.2007). However, the paracellular pathways for which thetransepithelial resistance is a relevant indicator do notcontribute substantially to uptake of PBT compounds. Inaddition, there may be important structural (e.g., membranelipid composition) and functional (e.g., the types andactivities of different membrane transport proteins) differ-ences between Caco-2 cells and the intestinal epithelium offish. Fish gills have a relatively high transepithelial resistance(3500 V cm2). It is likely, therefore, that chemical flux acrossthe gills is dominated by passive diffusion, even for relativelyhydrophilic compounds.

An alternative in vitro system that is used increasingly inhigh-throughput screening and drug discovery is the parallelartificial membrane permeability assay (PAMPA; Kansy et al.1998; Avdeef 2003). This assay employs a polymer filter thatis impregnated with a mixture of solvent (typically dodecane)and phospholipids (2–30%) and placed between 2 aqueouscompartments (Figure 4A). Concentration changes in theacceptor and donor compartments are measured over time,and membrane permeability is derived from these measure-ments. Permeability is then related to the uptake rateconstant through the surface area to volume ratio of themembrane, while the elimination rate constant is related tothe uptake rate constant through the membrane–waterpartition coefficient.

Most forms of PAMPA depend on the ability to measurechemical concentrations in the aqueous phase (the same istrue for Caco-2 cells). This presents a problem when usinghydrophobic compounds because aqueous concentrationsmay be too low to measure. A variety of excipients havebeen used to overcome this limitation (e.g., sodium tauro-cholate, 2-hydroxypropyl-b-cyclodextrin, potassium chloride,propylene glycol, 1-methyl-2-pyrrolidone, and polyethyleneglycol 400; Avdeef 2006; Bendels et al. 2006; Avdeef et al.2007). It is unclear whether these solubilizing agents could be

used to study hydrophobic environmental contaminants.Retention of hydrophobic chemicals by the membranesupport may also present problems.

Alternatively, it may be possible to modify PAMPA so thatit can be applied to hydrophobic compounds. For example,Kwon et al. (2006) modified PAMPA by increasing thevolume of the aqueous compartment. The authors used thissystem to measure the effective permeability of 23 com-pounds with log KOW values ranging from 0.9 to 3.5. Bymeasuring the permeability of organic acids as a function ofpH, it was possible to differentiate between membranepermeability and the permeability of the unstirred waterlayer, and to estimate the thickness of the unstirred waterlayer. A tumble stirring technique was used to adjust thethickness of the unstirred water layer until it was close to thatexisting at fish gills (approximately 20 mm). Membranepermeability values determined in this manner were thenused to predict uptake and elimination rate constants for a 1 gfish. For 12 of the 13 most hydrophobic chemicals, predictedelimination rate constants differed from reported values byless than 1.0 log unit, and most were much closer. Uptake rateconstants were less well predicted by the system, althoughgood agreement (within 1 SD of the reported value) wasobtained for 7 of the 15 chemicals tested.

Figure 4. Variations on the parallel artificial membrane permeability assay(PAMPA). (A) Approach used for more hydrophilic and soluble chemicals(Kansy et al. 1998). (B) Modification of PAMPA for hydrophobic chemicals

using a passive dosing/sampling approach for quantification of chemicalsand a polydimethylsiloxane (PDMS) membrane as the membrane model(adapted from Kwon and Escher 2008).

582 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

A 2nd modification of PAMPA was described recently byKwon and Escher (2008). Silicone disks (polydimethylsilox-ane, PDMS) were used as dosing and sampling systems,eliminating the need to measure chemical concentrations inwater (Kwon et al. 2007). The design of the system is shownin Figure 4B. A preloaded PDMS disk was in contact with awater compartment and this donor compartment wasseparated from the acceptor compartment by a PDMSmembrane that mimicked the gill membrane. The thicknessof the unstirred water layer was again adjusted by vigoroustumble stirring to mimic physiological conditions in fish gills(10–20 mm thickness). Chemical flux from the donor to theacceptor compartment was then measured and the perme-ability of the inner membrane was determined.

Using this system, Kwon and Escher (2008) collected datafor compounds with log KOW values ranging from 1.0 to 6.4.Estimated permeability values (PAPP; cm/h) were wellcorrelated with measured elimination rate constants for fish(see Figure 3 in the cited work) as well as elimination rateconstants determined using the classic assay (Figure 5).Generally, however, compounds that are metabolized by fishhad higher in vivo rates of elimination than values predictedby PDMS-PAMPA. Unlike Caco-2 cells, PAMPA provides apure measure of membrane permeation and does not accountfor possible contributions of metabolism to chemical flux. Toadequately predict the membrane flux of highly metabolizedcompounds it would be necessary to pair PDMS-PAMPAwith an in vitro metabolism assay. Additional work is neededto refine and validate in vitro assays of membrane permeationbefore they can be used to derive input parameters forchemical modeling efforts.

Distribution

The distribution of a compound within an organism isdetermined by its physicochemical properties (e.g., pKA,relative hydrophobicity), the composition of different tissuesand organs (e.g., water and lipid content), the biology of theorganism (e.g., blood perfusion rates to different tissues, pHgradients across membranes), and the activities of various

membrane transport proteins (Kleinow et al. 2008). Howev-er, because bioaccumulation is generally evaluated understeady-state conditions, this complexity tends to simplify,resulting in a system that behaves kinetically like one well-stirred compartment. Under these conditions, the chemicalconcentration in each tissue can be related to that in ameasured or ‘‘reference’’ region by a single proportionalityconstant.

In most bioaccumulation modeling efforts with fish, thewhole animal is taken to be the reference region. This is donepartly as a matter of convenience, especially when workingwith small animals, and partly because regulatory endpoints(e.g., the BCF or BAF) are often expressed on this basis. Insome cases, however, it may be advantageous to characterizechemical kinetics by measuring the chemical concentration inanother reference region such as blood plasma or thesurrounding water. This may be true for large fish that aredifficult to sample in their entirety or for chemicals that canbe measured in water but are difficult to analyze once theyhave been absorbed into tissues. The chemical concentrationin blood may also provide a more direct linkage to ADMEprocesses that ultimately determine the extent to which acompound accumulates. This is particularly true when thegoal is to extrapolate in vitro metabolism information to thewhole animal.

The chemical concentration in the whole animal may berelated to that in blood or water using the concept of‘‘apparent volume of distribution.’’ The apparent volume ofdistribution (V(t); L/kg) is defined as

V (t)~A(t)

CP(t)ð4Þ

where A(t) is the amount of chemical in the body (e.g., mg/kg) at time (t) after exposure and CP(t) is the compoundconcentration in a reference region (e.g., mg/L/kg), assumedhere to be the blood plasma (as indicated by the subscript, P).Under steady-state conditions, the rate of chemical uptakeequals the rate of elimination and V(t) reflects the sorptioncapacity of the animal (mg/kg) relative to that of thereference region (blood plasma or water; mg/L). Thisoutcome represents a special case of Equation 4 and theresulting volume of distribution is denoted VSS.

A list of VSS values determined in modeling efforts withfish was given by Kleinow et al. (2008). The listed valuesrange from 0.08 to 31 L/kg (where blood plasma was thereference region). Although these values do not necessarilycorrespond to real physiological volumes, they are ofteninterpreted in the context of specific chemical attributes andthe volumes of different functional body compartments.Thus, for compounds that distribute primarily to body water,VSS (referenced to plasma) may approximate the fractionalvolume of whole-body water (0.7–0.9). If a compounddistributes primarily to water but is unable to cross cellmembranes, VSS may approximate the fractional volume ofextracellular water (0.2–0.4).

Compounds that accumulate in fish and other biota oftenexhibit substantial hydrophobic character. As chemicalhydrophobicity increases, the distribution of a compoundshifts from one that is determined substantially by chemicalaffinity for water to one that reflects chemical affinity forlipid. Chemical affinity for protein may also play a role,particularly for compounds that exhibit specific protein

Figure 5. Log-log plot showing the linear correlation between apparentpermeability values (PAPP) from parallel artificial membrane permeabilityassay with a polydimethylsiloxane membrane (PDMS-PAMPA) and elimina-tion rate constants (kE) from classic PAMPA. Data were obtained from Kwonet al. (2006) and Kwon and Escher (2008).

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 583

binding characteristics, or when the organism in question isrelatively lean.

Algorithms that account for chemical accumulation inlipid, nonlipid organic matter, and water have been developedto predict chemical partitioning from water to fish underequilibrium conditions. The resulting value is a partitioning-based estimate of the bioconcentration factor (BCFP; L/kg).One such algorithm was given by Arnot and Gobas (2004):

BCFP~nLBKOWznNBbKOWznWB ð5Þ

where nLB (unitless) is the fractional lipid content of theorganism, nNB (unitless) is the fractional content of nonlipidorganic matter, nWB (unitless) is the fractional water content, andb (unitless) is a proportionality constant that reflects the sorptioncapacity of nonlipid organic matter relative to that of n-octanol.In previous reports it was recommended that b be set equal to0.035 (Arnot and Gobas 2004). More recent work suggests thatb has a value of about 0.05 (de Bruyn and Gobas 2007).

The same algorithm can be used to predict blood:waterchemical partitioning (PBW):

PBW~nLBLKOWznNBLbKOWznWBL ð6Þ

where nLBL (unitless) nNBL (unitless), and nWBL (unitless) arethe fractional lipid, nonlipid organic matter, and watercontent of blood, respectively. Dividing BCFP by PBW

provides an estimate of VSS, referenced to the chemicalconcentration in blood (Nichols et al. 2006):

VSS,BL~BCFP=PBW ð7Þ

Equation 7 predicts that VSS,BL will increase with chemicallog KOW up to a log KOW value of about 4. For compoundswith log KOW values greater than 4, VSS,BL remains relativelyconstant and is substantially determined by the lipid contentof the whole fish relative to that of blood. For example, if thelipid content of the whole animal is 10% while that of blood is2%, the ‘‘partitioning-based’’ VSS,BL for a hydrophobiccompound would be approximately 5. This prediction isconsistent with experimentally determined VSS,BL values for anumber of hydrophobic compounds in fish (Han et al. 2007;Kleinow et al. 2008).

The VSS can also be viewed as the plasma volume (VP) plusthe sum of each tissue:plasma partition coefficient (PTP)multiplied by its respective tissue volume (VT):

VSS~VPzX

(PTP:VT) ð8Þ

Here the term PTP is the ratio of the unbound fraction inplasma (fU,P) to that in tissue (fU,T):

PTP~fU,P=fU,T ð9Þ

Equation 8 shows that, conceptually, VSS is affected by thebinding properties of a compound in both plasma and tissues.When a compound is highly bound in tissues, VSS will bemuch larger than VP; whereas when binding occurs mainly inplasma, VSS will be close to VP.

Several in vitro methods have been developed to measurechemical binding to plasma proteins (Wright et al. 1996).Standard techniques include equilibrium dialysis, ultrafiltra-tion, microdialysis, gel filtration, and albumin columnanalysis. Similar approaches have been used to obtain binding

measurements for homogenized tissue samples (Pacifici andViani 1992). Because many environmental contaminants arevery hydrophobic, adsorption to the dialysis devices andmembranes used in these techniques may become a problem.This limitation can be overcome using methods that rely onchemical partitioning to a 3rd-phase sampling device such as asolid phase microextraction fiber (Vaes et al. 1996) or silicondisk (Kwon et al. 2009).

Estimates of PTP can be obtained for volatile compounds bymeasuring changes in chemical concentration within thevapor headspace of a closed vial system (Gargas et al. 1989).A conceptually similar approach that employs an immisciblesolvent as the ‘‘donor’’ compartment was used to estimate PTP

values for nonvolatile compounds (Murphy et al. 1995).Artola-Garicano et al. (2000) used a solid phase microextrac-tion method to obtain PTP estimates for several hydrophobicpesticides.

Metabolism

A comprehensive review of chemical biotransformation infish was provided recently by Schlenk et al. (2008). Broadly,it is possible to distinguish 2 types of metabolic activity: PhaseI and Phase II. Phase I reactions add or expose polar atomswithin the compound that is being acted on. The 3 principaltypes of Phase I activity are oxidation, reduction, andhydrolysis. Phase II reactions generally enhance the polarityof a compound by conjugation with a polar endogenousmolecule (e.g., UDP-glucuronic acid, glutathione, or 39-phosphoadenosine-59-phosphosulfate). Alternatively, PhaseII reactions may protect against the bioactivation of acompound by the addition of a functional group (e.g.,methyl, acetyl) that limits the production of reactiveintermediates.

Compounds that are metabolized by Phase II pathwaysmay be presented to fish in a form that can be conjugateddirectly. This is true for many low log KOW compoundsincluding various drugs, antibiotics, pesticides, and plasticiz-ers. In contrast, many high log KOW compounds must bemetabolized by a Phase I pathway before they can be acted onby Phase II pathways. From a bioaccumulation modelingperspective, an important question is whether the overall rateof metabolic clearance is limited by Phase I or II activity. Forexample, a compound may be hydroxylated by a Phase Ipathway and then conjugated by glucuronidation. The finalproduct of this activity is the glucuronide conjugate. If therate of hydroxylation limits the overall rate of metabolicclearance, however, it may be sufficient to characterize theactivity of this Phase I pathway in order to predict the extentof parent chemical accumulation. The following sectionsdescribe ongoing efforts to predict in vivo rates of metabolismin fish from an evaluation of chemical structure or byextrapolating measured rates of activity from controlled invitro studies.

Structure-based modeling of biotransformation—Several struc-ture-based approaches for modeling chemical biotransforma-tion have been reported in the literature (Mekenyan et al.2004, 2006):

N Explicit enzyme models describe the interaction betweenenzymes and their substrates. These models are based oninformation for substrates metabolized by specific cyto-chrome P450 (CYP) isoenzymes and may incorporate a 3-dimensional representation of the protein active site.Models of this type are mechanistically sound but are

584 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

limited to predicting the outcome of a single reactionpathway.

N Group contribution methods are used to model thebiodegradability of heterogeneous chemicals. This ap-proach is based on an assessment of individual structuralfragments presumed to be inherently biodegradable anddoes not provide insight into likely metabolic products.

N Statistical approaches include neural networks, partialleast square discriminant analysis, and genetic algorithms.In general, these models do not provide mechanisticinterpretations of their predictions.

N Rule-based methods predict the metabolism of chemicalsby simulating biotransformation pathways. These ‘‘expertsystems’’ employ libraries of transformations, rules forordering these transformations, and a substructurematching engine which invokes the transformations in aprescribed sequence. The system of hierarchically orderedtransformations and substructure matching engine iscalled a metabolism simulator. In knowledge-basedexpert systems, the rules and their hierarchy are manuallydefined using expert knowledge. In machine-learningexpert systems, the rules and their hierarchy are derivedfrom training sets without human input.

These models have been used to qualitatively predict thelikelihood that a chemical will undergo biotransformation and(for rule-based methods) the identities of likely metabolicproducts. This latter type of information is particularlyimportant for chemicals that are bioactivated by metabolismto form reactive intermediates.

Probabilistic approach—The probabilistic approach is anextension of rule-based methods and is based on the use of 2submodels. The 1st submodel simulates metabolic pathways.The second uses the generated metabolic maps and a set oftransformation probabilities to predict an endpoint of interest(e.g., biodegradation or bioconcentration; Dimitrov, Breton,et al. 2002; Jaworska et al. 2002; Dimitrov, Dimitrova,Parkerton et al. 2005).

Metabolism is simulated using molecular transformationsextracted from a metabolic pathway database. The moleculartransformations consist of a parent submolecular fragment,transformation products, and inhibitory functional groups(masks). If a fragment assigned as a mask is attached to thetarget subfragment, the transformation of the parent chemicalis prevented. The principal transformations are separated intogroups with differing hierarchy. These groups are based onthe type of transformed parent subfragment (e.g., halogenat-ed carbon, ester, amine). A 2nd level of hierarchy isestablished within the groups by ascribing a probability ofoccurrence to each transformation belonging to the group.The hierarchy of the transformations is used to control thepropagation of the metabolic map. This is performed bymatching the parent molecule with the source fragmentassociated with all transformations. When matches areidentified, the transformation products are generated. Theprocedure is repeated for the newly formed products in orderto obtain the next progression of metabolites. The resultingsequence of transformations can be represented as a ‘‘tree’’ ofchemical structures, as shown in Figure 6.

When using this approach, the key question is how to assessthe probability of each transformation. Dimitrov, Dimitrova,Parkerton et al. (2005) approached this question by assumingthat metabolic reactions that occur in the rat liver arequalitatively the same as those in fish. The metabolism

simulator was developed using a database of 367 documentedbiotransformation maps for rats. The resulting library ofmolecular transformations comprises 382 Phase I and 48 PhaseII reactions. The model was applied to fish by assigning anoverall probability of metabolism for each parent structure(PPARENT). Metabolism probabilities were estimated using thebaseline BCF model by simulating observed fish BCF values.These fitted probabilities are represented in the model by termsthat ‘‘mitigate’’ for the effect of metabolism (FMETABOLISM;Dimitrov, Dimitrova, Parkerton et al. 2005):

FMETABOLISM~1{PPARENT ð10Þ

Acting alone or in combination with other mitigating factors (Fi;including factors that account for the effect of molecular size[FMOLECULAR SIZE] and ionization [FIONIZATION]), metabolismoperates against the baseline extent of metabolism predictedfrom simple partitioning considerations (BCFMAX) according tothe relationship:

log BCF~ log (FMETABOLISMBCFMAX)zX

i

log Fi ð11Þ

An important feature of this approach is that metabolismprobabilities are assigned to each reaction within a metabolicpathway. Assignment of an overall metabolism probabilitytherefore translates to a set of probabilities representing allmodeled metabolic products. In principle, the probabilisticapproach can be used to simulate the kinetics of metabolism byexpressing metabolism probabilities as a function of time:

Pi~1{ exp {kitð Þ ð12Þ

where ki (h21 or d21) is the 1st-order kinetic constant for the ithtransformation. This approach has been used to simulate thekinetics of biodegradataion (Dimitrov et al. 2007), but has notyet been applied to bioconcentration models for fish.

In vitro methods for metabolism rate prediction—Metabolismis generally represented in mass-balance models by theparameter kM (h21 or d21). It is useful, however, to viewkM in terms of blood clearance rates and an apparent volumeof distribution. An equation that interrelates these parametersmay be given as

kM~CLHzCLE

VSS,BLð13Þ

where CLH and CLE are in vivo hepatic and extrahepaticclearances, respectively. Clearance is defined here as thevolume of blood that must be acted upon (‘‘cleared’’), andfrom which all compound is removed in a unit of time in orderto account for the rate at which the material is eliminated by anorgan. Thus, CLH reflects the ability of the liver to remove acompound from the blood by way of metabolism while CLE

represents metabolic elimination from organs other than theliver. For many compounds, the liver is the primary organ ofmetabolic clearance and Equation 13 simplifies to

kM~CLH

VSS,BLð14Þ

The value of expressing kM in these terms is 2-fold. First,explicit consideration of VSS,BL allows the user to adjust forchemical-specific differences in binding to blood and tissues.Second, CLH can be estimated from in vitro measurements of

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 585

Figure 6. Simulated metabolism of camphor and corresponding mathematical interpretation; Parent 5 parent chemical; Mn 5 metabolite; Pn 5 probability;indices 1, 2, …, n, n+1 5 level of metabolism.

586 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

metabolic activity using a physiological model of the liver(Houston 1994). Although several liver models have beenproposed (Kwon 2001), the ‘‘well-stirred’’ model is the onethat is used most widely due to its relative simplicity anddemonstrated performance (Ito and Houston 2004). Thismodel may be expressed as

CLH~QH

:fU,B:CLINT,IN VIVO

QHzfU,B:CLINT,IN VIVO

ð15Þ

where QH is hepatic blood flow, fU,B is the compoundunbound fraction in blood, and CLINT,IN VIVO is the in vivointrinsic clearance. The variable CLINT,IN VIVO reflects theliver’s inherent ability to metabolize a substrate and is the rateof clearance that would result in the absence of any bloodflow limitations on CLH.

Researchers often measure chemical concentrations inplasma rather than whole blood (primarily for ease ofanalysis) and assume that the unbound fraction in plasma(fU,P) is a good estimate of fU,B. This approach is generallyaccepted, although exceptions are known to exist (Hinderling1997; Yang et al. 2007). For lipophilic compounds, however,it may be difficult to measure the free fraction of a compoundin blood or plasma. Moreover, small errors in measuredbinding (e.g., 99.9% vs. 99.99%) can translate to large errorsin estimated fU,P values (10-fold in this example). For thisreason, Kratochwil et al. (2004) recommended that theresults of plasma binding studies be reported using proteinbinding constants (or dissociation constants, KD; mmol/L)rather than fU,P values alone. The unbound fraction in plasmais related to KD according to the relationship:

fU,P~1

1z½P�=KDð16Þ

where [P] is the protein concentration in micromoles per liter(Toutain and Bousquet-Melou 2007). The KD values can beestimated with relatively high accuracy by appropriatelydesigned saturation experiments (Goodrich and Kugel 2007).

Houston (1994) proposed a scaling approach to estimateCLINT,IN VIVO from in vitro measurements of metabolicclearance (CLINT,IN VITRO):

CLINT,IN VIVO~SF:CLINT,IN VITRO ð17Þ

where the in vitro–in vivo scaling factor (SF) has units of 106

cells/g liver (for isolated hepatocytes) or milligrams of proteinper gram of liver (for microsomes or S9 fraction). Thisapproach has been used extensively in the pharmaceuticalindustry and a large amount of work has been done tooptimize in vitro assay conditions and evaluate the accuracyof predicted CLINT,IN VIVO values.

Due to the lipophilic nature of many environmentalcontaminants, nonspecific binding to the components of anin vitro system (hepatocytes, microsomes, or S9 proteins) isexpected. To correct for this binding, CLINT,IN VITRO may beexpressed as

CLINT,IN VITRO~CLINT,UB,IN VITRO

fU,INCð18Þ

where CLINT,UB,IN VITRO is unbound CLINT,IN VITRO andfU,INC is the unbound chemical fraction in the in vitroincubation matrix. Log KOW-based empirical equations havebeen developed to predict fU,INC in hepatocytes, microsomes,

or S9 incubations (Austin et al. 2002, 2005; Han et al. 2007,2009).

Two distinct approaches are commonly used for in vitrometabolism rate determinations: 1) metabolite formation,and 2) substrate depletion. The former approach relies on therelationship:

CLINT,UB,IN VITRO~VMAX=KM ð19Þ

where VMAX (mmol/min/mg protein [or 106 cells]) is themaximum rate of metabolism and KM (mmol/L) is theMichaelis constant. This approach requires knowledge ofthe predominant metabolic products and the existence ofanalytical methods to measure these products. In contrast, thesubstrate depletion approach measures the rate of depletionof parent compound. Assuming 1st-order depletion kinetics,CLINT,UB,IN VITRO equals the product kSD ? VINC, where kSD

is the 1st-order elimination rate constant and VINC is thevolume of incubation (Han et al. 2007).

Because it requires no prior knowledge of metabolicpathways, the substrate depletion approach is better suitedto high-throughput analysis of previously untested com-pounds. A successful substrate depletion experiment, how-ever, requires careful selection of experimental conditionssuch as compound concentration (low enough to reduce thechance of saturating the enzymes but high enough to meetanalytical sensitivity requirements), enzyme concentration(low enough to minimize substrate binding to protein buthigh enough to ensure adequate substrate depletion), andincubation time (short enough to ensure the stability ofenzyme activities but long enough to ensure adequate loss ofsubstrate) (Jones and Houston 2004).

Proof of concept application to bioconcentration predictionsfor fish—The use of in vitro data to predict metabolismimpacts on bioaccumulation was advocated at a recentscientific workshop (Nichols, Erhardt et al. 2007). Recentresearch efforts provide a ‘‘proof of concept’’ for thisapproach (Han et al. 2007, 2008, 2009; Cowan-Ellsberry etal. 2008). Although limited to a small number of referencecompounds, these examples demonstrate the practicability of

Figure 7. Log bioconcentration factor (BCF) values of 6 reference com-pounds predicted with (filled circles) and without (open circles) consider-ation of kM and their correlation with empirical log BCF values. Data wereobtained from Han et al. (2007).

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 587

determining CLINT,IN VITRO in trout or carp hepatocytes,trout liver microsomes, and trout liver S9 fractions using thesubstrate depletion approach.

Figure 7 shows the impact of metabolism on BCFpredictions for a number of reference compounds in rainbowtrout (Han et al. 2007). The kM values for these compoundswere estimated by extrapolating measured levels of in vitrometabolism to the whole animal. A one-compartment mass-balance model was then used to predict steady-state BCFvalues. By incorporating metabolism into the model, Han etal. (2007) obtained BCF estimates that were much closer tomeasured values than estimates obtained assuming nometabolism.

A new concept proposed by Han et al. (2008, 2009) was toconsider the ratio of fU,B and fU,INC (fU,B/fU,INC) instead oftrying to estimate these 2 terms separately. The importance ofthis ratio can be shown by combining Equations 15, 17, and19, which results in the following relationship:

CLH~QH

:SF:CLINT,UB,IN VITRO:(fU,B=fU,INC)

QHzSF:CLINT,UB,IN VITRO:(fU,B=fU,INC)

ð20Þ

Assuming that fU,B and fU,INC reflect nonspecific binding thatis driven primarily by chemical lipophilicity, it is reasonableto expect cancellation of common errors associated withempirical equations used for their prediction. A plot of fU,B/fU,INC against log KOW shows that this ratio declines as logKOW increases from 1 to 3 (Figure 8). At higher log KOW

values this ratio tends to level off and at log KOW valuesranging from 4 and 7 the ratio changes very little (averaging inthis example about 0.08). This finding requires furthervalidation. If proven, however, it would suggest that a singlecorrection factor could be used for a wide range of lipophiliccompounds to adjust for chemical binding effects withminimal concern for error.

APPLICATION OF PREDICTIVE APPROACHES TOASSESS BIOACCUMULATION

In general, the application of predictive approaches forbioaccumulation assessment is not as well described as the

approaches themselves. Environment Canada (2003) andRobinson et al. (2004) provided qualitative guidance for PBTprioritization within the Canadian regulatory context. Guid-ance on model application within the European Unionregulatory context was given by de Wolf et al. (2007) and aREACH Intelligent Testing Strategy support document(ECHA 2008). A review of bioaccumulation models by theEuropean Chemicals Bureau (Pavan et al. 2008) also describessome quantitative and semiquantitative approaches forregulatory application of these tools. The concepts describedin these sources are summarized in the following section.Additional guidance is given for the use of in vitroinformation in bioaccumulation assessments.

Assessing the reliability of bioaccumulationmodeling predictions

Bioaccumulation models have evolved over the last 35 yearsfrom empirically based mathematical constructs to mechanis-tically based descriptions of the bioaccumulation process itself.As such, current models represent a kind of ‘‘hypothesis’’regarding the physical, chemical, and biological factors thoughtto control this process. In principle, this mechanistic founda-tion provides a rational basis for extrapolating modeled resultsto untested conditions. In practice, however, these modelswork only as well as the knowledge base from which they werebuilt. The assessment of bioaccumulation can work quite wellwhen a queried chemical is within the domain of modelapplicability. If a queried chemical is outside this domain, thereliability of the prediction is reduced. Often, the domain ofapplicability may be described by multiple model subdomains.Potential subdomains include molecular parameters that mayaffect the quality of measured endpoints (e.g., molecularweight, water solubility, or log KOW), a structural subdomain(e.g., chemical structures or chemical fragments in a model’straining set), a subdomain of explanatory variables orinterpolation space (e.g., range of variation of model descrip-tors), a mechanistic subdomain (e.g., binding of perfluorinatedcompounds to protein), and a metabolism subdomain (e.g., alibrary of metabolic transformations and their reliability;Dimitrov, Dimitrova, Pavlov et al. 2005; Dimitrov, Dimitrova,Parkerton et al. 2005). Alternatively, the choice of modelsubdomains may be left to the user (Meylan et al. 1999).

In addition, the quality of a model output will only be asgood as that of the model inputs upon which it is based. This‘‘garbage in–garbage out’’ rule applies to 2 principal areas ofbioaccumulation modeling: model training data and primarymodel input parameters. Several authors have discussedcriteria for assessing the reliability of in vivo fish bioaccumu-lation data (Arnot and Gobas 2006; Weisbrod et al. 2007;Parkerton et al. 2008), and an assessment of data reliabilitycan be useful for determining the overall reliability of abioaccumulation model. In their review of BCF and BAFestimates Arnot and Gobas (2006) calculated the 95%confidence interval associated with acceptable empiricalBAF data and noted that this can span 4 orders of magnitude.The authors pointed out that, in addition to ecosystem-specific characteristics, many of the same sources ofvariability in BCF data (e.g., organism weight, temperature,and lipid content) are also associated with BAF data.

Additional errors can be attributed to inaccuracies in thelog KOW values used as model inputs. This is especially truefor compounds with log KOW values greater than 8. Otherchemicals exhibit unusual properties that complicate mea-

Figure 8. Predicted relationship between fU,B/fU,INC and log KOW for troutliver microsomes. Data were obtained from Han et al. (2008).

588 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

surement of log KOW. For example, many dyes and pigmentslack appreciable solubility in either solvent phase (Anliker etal. 1981).

Ideally, bioaccumulation models would provide an estimateof the potential error associated with model predictions. Thisis a quantitative way for users to understand model reliability.Presently, very few bioaccumulation models are structured toprovide this information. One notable exception is thebaseline BCF model (Dimitrov, Dimitrova, Parkerton et al.2005), which gives confidence intervals for corrected BCFvalues as well as model domain results.

Potential utility of in vitro systems

In vitro test systems offer many advantages for predictingthe ADME properties of chemicals when compared to whole-animal testing. Among these advantages are reduced animaluse, lower cost, and higher chemical throughput. Commercialsources of characterized in vitro systems for fish (e.g., liverS9, microsomes, and isolated hepatocytes) are not yetavailable; however, several providers of mammalian in vitrosystems are moving in this direction. The commercialavailability of characterized in vitro systems, along with thedevelopment of standardized methods, would allow labora-tories around the world to conduct essentially the sameexperiment, thereby facilitating interlaboratory comparisonsand encouraging broad acceptance of data.

Potential limitations on the use of in vitro data in PBTassessments include limitations of the assays themselves, lackof validation for industrial chemicals, and uncertainty on thepart of legislative authorities on how to incorporate thisinformation into current assessment schemes. As indicatedpreviously, hydrophobic compounds can be difficult to workwith due to analytical challenges (e.g., measurement ofchemical concentrations in aqueous solution) as well as theirtendency to adsorb to materials in contact with water. Theseproblems are exacerbated by the conditions found in mostvitro assays including small sample volumes and relativelylarge adsorptive surfaces. Metabolism assays are routinelyconducted using compounds with log KOW values up to about6 (e.g., the aryl hydrocarbon hydroxylase assay, which usesbenzo[a]pyrene as the metabolic substrate). Additional workis required to determine their utility for more hydrophobiccompounds.

Quality of in vitro test data

A large number of factors may influence the outcome of anin vitro experiment. Among these are the quality of sourcematerial (in vitro systems as well as the test compound),temperature, chemical concentration, and the number andtiming of sampling events. In general, assay temperatures andchemical concentrations should be based on anticipatedenvironmental conditions. Method checks may be requiredto ensure that assumptions underlying the in vitro techniquehave been satisfied (e.g., linear kinetics; Han et al. 2008,2009).

These and other issues underscore the need for standard-ization of methods and the development of standardoperating procedures. Metabolism assays should be conduct-ed using material that has been characterized with respect toits activity toward substrates representing the metabolicpathways of interest. One or more of these substrates alsoshould be used as an internal method check each time that theassay is run. Bioavailability is likely to be an issue for many

compounds, due to both protein binding and adsorption tothe experimental system. The interpretation of these studieswill be enhanced if the results are reported on a free chemicalbasis. Additional guidance on the design of in vitrometabolism studies for bioaccumulation assessment is pro-vided by Nichols, Erhardt et al. (2007).

Potential limitations of in vitro metabolism assays

Potential limitations on the use of in vitro metabolism datafor bioaccumulation assessment derive from 2 possiblesources of error in metabolism predictions (Nichols et al.2006; Nichols, Erhardt et al. 2007). The 1st possibility is thatmeasured in vitro activity inadequately represents the activityof the organ from which the in vitro system was derived.Several studies with mammals have been performed tocompare the accuracy of metabolism predictions obtainedusing in vitro liver data. Predictions generated using isolatedhepatocytes, microsomes, and liver S9 tend to be highlycorrelated. In general, however, isolated hepatocytes providethe most accurate predictions of measured in vivo clearance.Limited data suggest that the same general conclusion mayapply to fish (Han et al. 2009), although additional work inthis area is needed.

The 2nd possible source of error is that extrahepaticclearance (CLE; Eqn. 13) is not accounted for by the in vitroassay. Metabolism that occurs in tissues of the gastrointestinaltract may be of special importance for PBT compounds, sincethis is the primary route of exposure in a natural setting. Theability of the fish gastrointestinal tract to metabolizechemicals has been demonstrated in several studies (VanVeld et al. 1988; Kleinow et al. 1998). Operating in serieswith hepatic metabolism, this activity could substantiallyreduce uptake and accumulation of these compounds(Nichols, Fitzsimmons et al. 2007).

To date, most of the successful in vitro–in vivo metabolismextrapolations for mammals have been performed usingcompounds that are metabolized primarily by hepatic CYP-mediated reactions. Similar work with fish is required toestablish the applicability domain of this approach. A needalso exists for validation of in vitro–in vivo metabolismextrapolations using either in vivo data (Han et al. 2007,2009; Cowan-Ellsberry et al. 2008) or direct measures ofhepatic clearance obtained using isolated perfused fish livers(Forlin and Andersson 1981; Andersson et al. 1983).

Metabolism: Is there a useful cutoff value?

It is possible using current bioaccumulation models to solvefor a kM value that, under a specified set of conditions (e.g.,fish size, lipid content, and temperature), will reduce BCF orBAF values for all compounds below a prescribed value(Arnot and Gobas 2003; Nichols, Fitzsimmons et al. 2007;Cowan-Ellsberry et al. 2008). This analysis is useful because itillustrates the potential impact of metabolism on bioaccu-mulation and shows how this impact may vary with chemicallog KOW. The success of this approach has led some, however,to suggest that models could be used to identify ‘‘global’’ kM

cutoffs for use in bioaccumulation assessment. Unfortunately,this suggestion has 2 major pitfalls.

First, as indicated previously, modeled kM values are highlycontext specific, insofar as they depend on the specification ofother model inputs. In principle, it may be possible to accountfor changes in these inputs. One approach would be togenerate a kM value that represents the ‘‘worst-case’’ scenario;

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 589

that is, an assumed set of conditions under which metabolismis likely to have the lowest possible impact on bioaccumu-lation. In practice, however, this approach is likely to beoverly ‘‘conservative’’ insofar as it would underestimatemetabolism impacts for a large percentage of cases.

More importantly, the relationship between kM and actuallevels of metabolism is not a simple one. As indicatedpreviously, kM reflects both the rate at which a chemical iscleared and its apparent volume of distribution (VSS). Whilethe VSS for many compounds can be estimated from simplepartitioning considerations, this is not always the case.Moreover, metabolic clearance depends not only on theintrinsic ability of the enzymes to act on a compound but alsothe rate at which chemicals are presented to the metabolizingtissue in blood (QH, assuming that metabolism occurs in theliver). Failure to account for this fact could lead to anonsensical result. For example, the fitted value of kM thatwould be required to achieve a prescribed reduction inbioaccumulation may not be achievable due to blood flowlimitations on hepatic clearance. Therefore, application of kM

estimates is probably best done on a chemical by chemicalbasis when in vivo bioaccumulation data are available, asdescribed by Arnot et al. (2008).

A stronger case can be made for using an CLINT,IN VITRO

cutoff value in bioaccumulation assessment and prioritysetting. The value of this parameter is still context specific.Unlike a kM cutoff, however, an CLINT,IN VITRO cutoff can berelated directly to the results of an appropriately designed invitro study.

Chemical absorption: Is there a useful cutoff value?

Although existing data suggest that there are hydrophobic-ity and size limitations on chemical uptake and accumulationby fish, these limitations probably do not reflect a trueabsorption ‘‘cutoff,’’ as might be expected if the gill and gutmembranes acted like simple filters. Instead, these limitationsoccur as a combined result of several physicochemical andkinetic processes including reduced bioavailability, restrictedmembrane diffusion, growth dilution, and biotransformation.Apparent cutoff values for chemical absorption (size- or logKOW-based) can be derived from a model-based analysis ofchemical uptake data or measured levels of accumulation.Like the kM cutoff values discussed above, these cutoffs arecontext specific in the sense that their fitted value depends ona specific set of tested conditions. Many of the processes thatdetermine the shape of hydrophobicity/size–bioaccumulationrelationships are dependent on temperature including chem-ical diffusion in biological membranes, ventilation rate, andgrowth. Many relevant processes also scale to a fractionalexponent of fish body weight. Finally, as chemical sizeincreases, particularly for chemicals taken up across thegastrointestinal epithelium, there is likely to be a transitionfrom simple diffusion as the primary transport process tovesicular transport.

By comparison to fitted kM cutoff values, however, model-determined size cutoffs for chemical absorption can beexpected to vary within a relatively narrower range of values,even after taking fish size and temperature effects intoconsideration. This is because the thickness and molecularcomposition of the relevant diffusion barriers (e.g., gill andgut epithelial tissues) remain relatively fixed. Assuming,therefore, that simple diffusion is the principal determinant ofchemical uptake and accumulation (i.e., absent significant

biotransformation, active uptake/elimination, ion trapping ofionizable species, or chemical reactions within the membraneitself), it may be reasonable to use an apparent size cutoff inbioaccumulation assessments for fish.

Hydrophobicity effects on chemical absorption are moredifficult to predict, in part because they are more dependenton factors external to the absorbing membranes (e.g.,partitioning to dissolved organic carbon and processing oflipid micelles with the gastrointestinal tract). As such, theyare probably best dealt with using empirically derived logKOW relationships. Although several such relationships havebeen published, additional work is needed to refine thesepredictions and determine their underlying basis.

Aquatic and terrestrial systems

The predictive approaches described above apply inprinciple to both aquatic and terrestrial systems. For a givensituation, however, there may be some question about whichtype of bioaccumulation assessment to perform. Knowledgeof a compound’s physicochemical properties, environmentalfate and transport, and likely emission patterns (includingmode of entry into the environment) can help in making thisdecision.

Gobas, Kelly, and coworkers (Gobas et al. 2003; Kelly andGobas 2001, 2003; Kelly et al. 2007) have shown that KOW,which can be a good predictor of chemical accumulation inaquatic systems, may not provide good predictions ofaccumulation in terrestrial vertebrates. For example, sub-stances with very high log KOW values (.9) may exhibita limited ability to accumulate in fish due to low dietaryuptake efficiency. This trend may not apply to terrestrialmammals and birds. Terrestrial vertebrates are exposed tochemicals in their diet and in the air that they respire. It hasbeen suggested, therefore, that the octanol–air partitioncoefficient (KOA) be used to describe a chemical’s tendencyto partition from organism lipid to air, the respiring medium,similarly to how KOW describes lipid–water partitioning forfish.

These researchers have also reported that organic chemicalswith log KOW values between 2 and 5 and a log KOA greaterthan 5 have the potential to biomagnify in terrestrialvertebrates if not substantially metabolized. Such compoundswould not be classified as bioaccumulative under mostcurrent regulatory policies because these policies are basedon aquatic bioaccumulation potential. To illustrate this point,Gobas et al. (2003) used predicted log KOW and log KOA

values for 12000 organic chemicals on the Canadian DSL toshow that 36% of these compounds are potentially mis-categorized for bioaccumulation potential because terrestrialair-breathers were not considered.

Bioaccumulation modeling efforts for aquatic and terres-trial organisms should also consider multimedia fate infor-mation to determine if a chemical is expected to reside in amedium of exposure according to the mode of entry into theenvironment. For example, bioaccumulative, yet extremelyvolatile substances released to air may not be substantiallydeposited to the earth and may only reside in the upperatmosphere (Wania 2006). If the same substance wasreleased to water it may present an exposure to aquaticbiota. The same substance released to soil might associatewith soil particles, requiring that a soil-dwelling biota modelbe considered to assess bioaccumulation in subsurfaceorganisms.

590 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

FORMING A WEIGHT OF EVIDENCE FORBIOACCUMULATION ASSESSMENT USINGPREDICTIVE APPROACHES

The assessment of bioaccumulation potential by regulatoryagencies is typically performed using a weight-of-evidence(WoE) approach (e.g., CEPA 1999; ECHA 2008). Unfortu-nately, few details on what constitutes a WoE approach havebeen given. In a review of WoE approaches, Weed (2005)noted that ‘‘Weight of evidence is a common term in thepublished scientific and policy-making literature, most oftenseen in the context of risk assessment. Its definition, however,is unclear.’’ In the following discussion, the term WoE is usedto describe how multiple lines of evidence can be broughttogether to reach a conclusion.

In general, a WoE can be categorized as either qualitative(i.e., a determination of what is reasonable in view of allavailable information) or quantitative (i.e., a process by whichconclusions are reached through assignment of numericalweights to scientific evidence).

Figure 9 outlines a suggested approach to developing aWoE for bioaccumulation assessment based on in vitro dataand computational prediction methods. It involves bothqualitative and quantitative techniques as well as expertjudgment. The scheme is necessarily simplistic, in partbecause approaches to creating a WoE are context dependent,but also to increase transparency in decision-making. Impor-tantly, the steps suggested here are designed to allow assessorsof chemicals to document how a WoE was formed based onpredictive methods alone, and how pivotal values frombioaccumulation modeling efforts are selected for chemicalprioritization and assessment.

Data gathering stage

Once the chemicals of interest have been identified, the 1ststep in the process is to collect information regarding factorsthat may contribute to or mitigate the potential for bioaccu-mulation. This would include physicochemical property data,chemical structure information, and fate information such ashydrolysis rate (not inherently accounted for by bioaccumula-tion models) and biodegradation potential. Next, all availablebioaccumulation models relevant for chemical prioritization orassessment are run. The models should incorporate ADMEinformation from in vitro and in vivo data sources whenavailable. Output from bioaccumulation models should then beverified for reliability, as described above (i.e., domain ofapplicability, uncertainty bounds). At this point it is useful toexamine the results in the context of the physicochemicalproperty, structure, and fate information to identify compoundsthat may be difficult to model. For example, a solid chemicalwith a high melting point, low solubility in n-octanol or water,and a high cross-sectional diameter might not be expected toobey the equilibrium-based passive diffusion principles uponwhich bioaccumulation models are based. Similarly, a modelthat uses a predicted log KOW for the neutral form of chemicalwill overpredict bioaccumulation potential if the predominantspecies of the chemical is ionized at environmental pH (e.g.,long-chain ionic surfactants). Results determined to be reliablecan then be combined using qualitative or quantitativeweighting schemes in the decision stage.

Decision stage

In most cases it is likely that a qualitative approach will beused to weight decisions based on model output, given the

uncertainties inherent in this information. A simple approachis to take the highest model prediction as a final value to usefor prioritization or assessment. This approach is conservativeand will result in fewer false negatives, but potentially manyfalse positives.

Regulatory decisions for bioaccumulation can also be madeby seeking consensus among models. This approach can, intheory, minimize both false positives and false negatives. Anexception to this rule exists, however, if all models are subjectto the same error (e.g., an inaccurate log KOW value used asinput). When multiple predictions are produced, a decisionmust be made as to which value should be selected. Whenpredictions from all models agree this decision may be easy(e.g., all models predict a BCF ,5000). In this case, modelconsensus provides a strong WoE for the decision. It is verycommon, however, to have a lack of consensus among modelpredictions which make qualitative decisions more difficult. Ifall predictions are reliable but do not agree, which value orvalues should be used for priority setting or assessment?These decisions may involve weighting results based onregulatory data requirements for bioaccumulation and wouldtherefore be context dependent. In Canada, for example,regulatory decisions for bioaccumulation involve a datahierarchy where BAF evidence is preferred over BCFevidence, which is preferred over log KOW (Government ofCanada 2000).

Because much of the DSL categorization for bioaccumu-lation was based on model prediction, Environment Canadaused a ‘‘bagging’’ approach to analyze the results from oneBAF QSAR and 3 BCF QSARs (Robinson et al. 2004).Bagging is a simple way to combine predictions that belong tothe same type by voting or averaging. The maximum numberof models should be used but each model receives equalweight. This approach reduces the variance while leaving thebias unchanged (T.I. Netzeva, personal communication).

Figure 10 illustrates the bagging concept used by Environ-ment Canada as a means of generating a WoE for bioaccu-mulation predictions and shows where BCF model predictionspaces overlapped. Modeling results showing BAF predictions�5000 and at least one of the 3 BCF models predicting BCF�5000 gave this authority greater confidence that a chemicalwas likely to biomagnify in the aquatic environment. Thesecompounds were categorized as bioaccumulative. Figure 10shows that for approximately 2200 chemicals with predictedBAFs �5000, the best model consensus was achieved whereall 3 BCF model domains overlapped (339 chemicals). Valuesin model domain circles not overlapping another modeldomain circle (45, 366, 0) refer to chemicals with a predictedBAF �5000 but only one predicted BCF �5000 (i.e., thelowest model consensus). Zero values in domain circles do notsuggest that a model performed poorly, but rather that modelpredictions (and domains) were very comparable with otherBCF models and that no additional chemical coverage wasbeing added by this model. Three hundred forty-onesubstances had a predicted BAF �5000 but no predictedBCF values �5000. These were categorized as uncertain.

Quantitative weighting methods involve statistical orsemiquantitative approaches to rank the bioaccumulationpotential of one chemical over another (e.g., multidimen-sional statistical analysis, partial order ranking). Thesemethods are best suited to ranking large inventories ofchemicals as part of a prioritization exercise rather thansingle chemical assessment. Statistical ranking methods

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 591

Figure 9. Suggested approach to forming a weight of evidence for bioaccumulation assessment using predictions based on computational modeling.

592 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

require some knowledge of the distribution of the variable(e.g., BCF), while partial order ranking is parameter anddistribution free, requiring a priori information related to ‘‘�’’

as the only mathematical expression. The European Chem-icals Bureau (Pavan et al. 2008) applied these approaches toQSAR results for persistence, bioaccumulation, and toxicity,and noted that such approaches can be automated. Pavan etal. (2008) also described a structure-based approach that canbe used in combination with principal components analysisfor chemical prioritization.

Other factors may be used to select a value from multiplemodel predictions such as model transparency, provision ofuncertainty estimates, or regulatory compatibility acrossagencies. Regardless of the approach used, it is critical todocument the WoE process that was used for decision-makingso that decisions can be recounted. In some cases it may beadvisable to use the range of model predictions rather than asingle pivotal value; however, most regulatory programs arenot designed to accommodate ranges of values for decision-making.

Once a pivotal model prediction has been selected it can becombined with other lines of evidence (e.g., in vivo testresults, field monitoring data) to form an overall WoE forbioaccumulation. An example WoE for bioaccumulationusing multiple sources of data is presented in the REACHImplementation Project endpoint group guidance documentfor bioaccumulation (ECHA 2008).

SUMMARY AND CONCLUSIONS

Current and future use of predictive methods inbioaccumulation assessments

Predictive methods are used extensively within currentregulatory programs to assess the potential for chemicals toaccumulate in fish and other biota, including humans.Quantitative structure–activity relationship models have beenemployed since the 1970s to predict chemical bioconcentra-tion in fish. Mass-balance models have been used since the1980s to predict bioaccumulation in both aquatic and

terrestrial food webs. In this paper we have reviewed ongoingefforts to improve bioaccumulation assessments performedusing these models. Current research is focused on moreexplicitly including ADME processes. Of special concern arelimited absorption due to low bioavailability or restricteddiffusion across biological membranes, rapid clearance bybiotransformation, and distribution patterns that deviate fromthose predicted by simple partitioning to tissue lipid. Givencurrent and proposed restrictions on whole-animal testing aswell as the need to assess and prioritize large chemicalinventories, reliance on predictive methods is likely toincrease.

Recent advances in predicting ADME processes

In vitro systems are used extensively by the pharmaceuticalindustry to predict the oral bioavailability of drug candidates.These must be adapted for use with hydrophobic compoundsif they are to be of utility for bioaccumulation assessments.The PAMPA assay holds particular promise as a means ofestimating membrane permeation of high log KOW com-pounds. This system has been used to predict branchialuptake of chemicals with log KOW values up to about 3.5(Kwon et al. 2006). More recent work suggests that the assaycan be adapted for use with chemicals that have log KOW

values greater than 6 (Kwon and Escher 2008).

Recent studies have provided a ‘‘proof of concept’’ for invitro–in vivo metabolism extrapolations with fish (Han et al.2007, 2008, 2009; Cowan-Ellsberry et al. 2008). This workhas shown that incorporation of extrapolated metabolismvalues into models of chemical bioconcentration substantiallyimproves the accuracy of steady-state BCF predictions. Animportant aspect of this research is the need to view apparentmetabolism rates for fish (i.e., kM values) as quotients thatintegrate information on metabolic clearance (CLH; assumingthat the liver is the primary site of metabolism) and acompound’s distribution within the organism (VSS,BL; refer-enced here to the chemical concentration in blood): kM 5

CLH/VSS,BL. This clearance-volume approach is well estab-lished in mammalian modeling efforts and has been appliedby several authors to kinetic studies with fish. To date,however, it has not been broadly applied to bioaccumulationmodeling efforts. The advantage of this approach is that itprovides a bridge between computational modeling effortsand in vivo observations, thereby improving current practicein bioaccumulation assessment.

Data gaps and research needs

Data are needed to better define limitations on membranepermeation associated with molecular attributes such as size,hydrophobicity, and the presence or absence of specificfunctional groups. Because the diet is the principal route bywhich fish and other vertebrates are exposed to PBTcompounds, work is needed to translate predictions ofmembrane permeation into predictions of chemical assimila-tion from ingested food items. This will require a comple-mentary research effort involving model-based evaluations ofmeasured uptake and accumulation data, in vitro measure-ments of membrane permeation, and targeted collection of invivo data using experimental designs that test specificmechanistic hypotheses.

Additional work is needed to validate procedures used toextrapolate measured in vitro rates of metabolism for fish tothe whole animal. To accelerate this process we recommend

Figure 10. Results of Canadian Domestic Substances List (DSL) categoriza-

tion showing bioconcentration factor (BCF) model consensus for organicchemicals with predicted bioaccumulation factor (BAF) �5000 (modifiedfrom Robinson et al. 2004).

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 593

that initial efforts focus on a core set of test compounds thatare strategically selected for this purpose. Efforts to developsuch a chemical set have been initiated by a workgroupconvened by the International Life Sciences Institute, Healthand Environmental Sciences Institute (Arnot et al. 2007).Critical aspects of this work include the need to define theapplicability domain of in vitro–in vivo extrapolation proce-dures, evaluate the effects of chemical binding in vitro and invivo, and standardize methods. Mechanistic data from studiesof whole animals or isolated perfused organs are needed tointerpret both ‘‘successes’’ and ‘‘failures’’ as a means offurther defining the utility of these procedures.

Progress on in vitro methods development must be linkedto ongoing efforts to predict metabolism from chemicalstructure. Current structure-based methods can predictmetabolic pathways and the identities of likely metabolicproducts. The ultimate goal of these efforts is to obtainstructure-based predictions of metabolic rate. Measured ratesof in vitro metabolism, combined with estimated rates frommodel-based evaluations of in vivo accumulation data, can beused to ‘‘train’’ these computational models. For their part,structure-based models have the potential to inform in vitrotesting efforts by identifying likely metabolic pathways which,based on previous work, are known to be within or outside ofthe domain of assay applicability.

A need also exists to characterize metabolism at lowertrophic levels. Modeled results suggest that even modestlevels of metabolism could, if operating at several trophiclevels, have a substantial impact on chemical accumulation inhigh trophic-level organisms. There is extensive in vitroinformation on chemical metabolism by several invertebratespecies. To our knowledge, however, this information has notbeen used to predict rates of chemical clearance for use inbioaccumulation modeling efforts.

Guidance for predictive bioaccumulation assessment

A strategy is described for incorporating emerging methodsinto a WoE approach for bioaccumulation assessment. Thisstrategy recommends that the reliability of predictions basedon in vitro measurements or computational approaches betaken into account by considering their uncertainty limits (ifavailable) and the domain of applicability of the methodsthemselves. Only reliable predictions should be used forforming a WoE using qualitative or quantitative approaches,noting that practical considerations must be taken intoaccount for prioritization of large chemical inventories.

Acknowledgment—We thank Charles Stephen and ChristinaCowan-Ellsberry for their critical reviews of this manuscript,and Roger LePage for his assistance in preparing the figures.We also thank Jung-Hwan Kwon for his assistance inpreparing for the Pellston Workshop. This document hasbeen subjected to review by the National Health andEnvironmental Effects Research Laboratory and approvedfor publication. Approval does not signify that the contentsreflect the views of the agency, nor does mention of any tradenames or commercial products constitute endorsement orrecommendation for use.

REFERENCESAndersson T, Forlin L, Hansson T. 1983. Biotransformation of 7-ethoxycoumarin

in isolated perfused rainbow trout liver. Drug Metab Disposit 11:494–498.

Anliker R, Clarke EA, Moser P. 1981. Use of the partition coefficient as an indicator

of bioaccumulation tendency of dyestuffs in fish. Chemosphere 10:263–74.

Armitage J, Gobas FAPC. 2007. A terrestrial food-chain bioaccumulation model

for POPs. Environ Sci Technol 41:4019–4025.

Arnot J, Domoradzki J, Dyer S, Eickhoff C, Erhardt S, Escher B, Han X, Halder M,

Hutchinson T, Leonard M, Panter G, Parkerton T, Sahi J, Segner H, Thomas K,

Weisbrod A. 2007. Considerations for chemical selection supporting in vitro

biotransformation and prediction of bioaccumulation in fish. Society of

Environmental Toxicology and Chemistry (SETAC), SETAC Globe, March–April

2007, p 25–26.

Arnot JA, Gobas FAPC. 2003. A generic QSAR for assessing the bioaccumulation

potential of organic chemicals in aquatic food webs. QSAR Comb Sci

22:337–345.

Arnot JA, Gobas FAPC. 2004. A food web bioaccumulation model for organic

chemicals in aquatic ecosystems. Environ Toxicol Chem 23:2343–2355.

Arnot JA, Gobas FAPC. 2006. A review of bioconcentration factor (BCF) and

bioaccumulation factor (BAF) assessments for organic chemicals in fish.

Environ Rev 14:257–297.

Arnot JA, Mackay D, Bonnell M. 2008. Estimating metabolic biotransformation

rates in fish from laboratory data. Environ Toxicol Chem 27:341–351.

Artola-Garicano E, Vaes WHJ, Hermens JLM. 2000. Validation of negligible

depletion solid-phase microextraction as a tool to determine tissue/blood

partition coefficients for semivolatile and nonvolatile organic chemicals.

Toxicol Appl Pharmacol 166:138–144.

Artursson P, Palm K, Luthman K. 2001. Caco-2 monolayers in experimental and

theoretical predictions of drug transport. Advan Drug Delivery Rev 46:27–43.

Austin RP, Barton P, Cockroft SL, Wenlock MC, Riley RJ. 2002. The influence of

nonspecific microsomal binding on apparent intrinsic clearance, and its

prediction from physicochemical properties. Drug Metab Dispos 30:1497–

1503.

Austin RP, Barton P, Mohmed S, Riley RJ. 2005. The binding of drugs to

hepatocytes and its relationship to physicochemical properties. Drug Metab

Disposit 33:419–425.

Avdeef A. 2003. Absorption and drug development: Solubility, permeability, and

charge state. Hoboken (NJ): Wiley. 312 p.

Avdeef A. 2006. High-throughput solubility, permeability, and the MAD PAMPA

model. In: Testa B, Kramer SD, Wunderli-Allenspach H, Folkers G, editors.

Pharmacokinetic profiling in drug research: Biological, physicochemical, and

computational strategies. Zurich (CH): Wiley-VCH. p 221–243.

Avdeef A, Bendels S, Tsinman KL, Kansy M. 2007. Solubility-excipient

classification gradient maps. Pharm Res 24:530–545.

Barber MC. 2003. A review and comparison of models for predicting dynamic

chemical bioconcentration in fish. Environ Toxicol Chem 22:1963–1992.

Bendels S, Tsinman O, Wagner B, Lipp D, Parrilla I, Kansy M, Avdeef A. 2006.

PAMPA-excipient classification gradient map. Pharm Res 23:2525–2535.

Branson DR, Blau GE, Alexander HC, Neely WB. 1975. Bioconcentration of

2,29,4,49-tetrachlorobiphenyl in rainbow trout as measured in an accelerated

test. Trans Am Fish Soc 104:785–792.

Bruggeman WA, Martron LBJM, Kooiman D, Hutzinger O. 1981. Accumulation

and elimination kinetics of di-, tri-, and tetra chlorobiphenyls by goldfish

after dietary and aqueous exposure. Chemosphere 10:811–832.

Burkhard LP. 1998. Comparison of two models for prediction bioaccumulation of

hydrophobic chemicals in Great Lakes foodwebs. Environ Toxicol Chem

17:383–393.

[CEPA] Canadian Environmental Protection Act. 1999. Statutes of Canada 1999,

Chapter 33. Act assented to September 14, 1999. laws.justice.gc.ca/en/C-15.

31/text.html. Accessed 24 September 2008.

Clark KE, Gobas FAPC, Mackay D. 1990. Model of organic chemical uptake and

clearance by fish from food and water. Environ Sci Technol 24:1203–1213.

Connell DW, Hawker DW. 1988. Use of polynomial expressions to describe the

bioconcentration of hydrophobic chemicals by fish. Ecotoxicol Environ Safety

16:242–257.

Cowan-Ellsberry CE, Dyer SD, Erhardt S, Bernhard MJ, Roe AL, Dowty ME,

Weisbrod AV. 2008. Approach for extrapolating in vitro metabolism data to

refine bioconcentration factor estimates. Chemosphere 70:1804–1817.

de Bruyn MH, Gobas FAPC. 2007. The sorptive capacity of animal protein.

Environ Toxicol Chem 26:1803–1808.

Devillers J, Domine D, Bintein S, Karcher W. 1998. Comparison of fish

bioconcentration models. In: Devillers J, editor, Comparative QSAR.

Washington DC: Taylor and Francis. p 1–50.

de Wolf W, Comber M, Douben P, Gimeno S, Holt M, Leonard M, Lillicrap A, Sijm

D, van Egmond R, Weisbrod A, Whale G. 2007. Animal use replacement,

594 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

reduction, and refinement: Development of an integrated testing strategy for

bioconcentration of chemicals in fish. Integr Environ Assess Manag 3:3–17.

de Wolf W, de Bruijn JHM, Seinen W, Hermens J. 1992. Influence of

biotransformation on the relationship between bioconcentration factors

and octanol-water partition coefficients. Environ Sci Technol 26:1197–1201.

de Wolf W, Seinen W, Hermens JLM. 1993. Biotransformation and toxicokinetics

of trichloroanilines in fish in relation to their hydrophobicity. Arch Environ

Contam Toxicol 25:110–117.

Dimitrov S, Breton R, MacDonald D, Walker JD, Mekenyan O. 2002. Quantitative

prediction of biodegradability, metabolite distribution and toxicity of stable

metabolites. SAR QSAR Environ Res 13:445–455.

Dimitrov S, Dimitrova G, Pavlov T, Dimitrova D, Patlewicz G, Niemela J, Mekenyan

O. 2005. A stepwise approach for defining the applicability domain of SAR

and QSAR models. J Chem Inf Model 45:839–849.

Dimitrov S, Dimitrova N, Parkerton T, Comber M, Bonnell M, Mekenyan O. 2005.

Base-line model for identifying the bioaccumulation potential of chemicals.

SAR QSAR Environ Res 16:531–554.

Dimitrov SD, Dimitrova NC, Walker JD, Veith GD, Mekenyan OG. 2002. Predicting

BCFs of highly hydrophobic chemicals: Effects of molecular size. Pure Appl

Chem 74:1823–1830.

Dimitrov SD, Dimitrova NC, Walker JD, Veith GD, Mekenyan OG. 2003.

Bioconcentration potential predictions based on molecular attributes—An

early warning approach for chemicals found in humans, birds, fish and

wildlife. QSAR Comb Sci 22:58–68.

Dimitrov SD, Mekenyan OG, Walker JD. 2002. Non-linear modeling of

bioconcentration using partition coefficients for narcotic chemicals. SAR

QSAR Environ Res 13:177–188.

Dimitrov S, Pavlov T, Nedelcheva D, Reuschenbach P, Silvani M, Bias R, Comber

M, Low L, Lee C, Parkerton T, Mekenyan O. 2007. A kinetic model for

predicting biodegradation. SAR QSAR Environ Res 18:443–457.

Dulfer WJ, Govers HAJ. 1995. Membrane-water partitioning of polychlorinated

biphenyls in small unilamellar vesicles of 4 saturated phosphatidycholines.

Environ Sci Technol 29:1548–2554.

Dulfer W, Govers H, Groten J. 1998. Kinetics and conductivity parameters of

uptake and transport of polychlorinated biphenyls in the Caco-2 intestinal

cell line model. Environ Toxicol Chem 17:493–501.

[EC] Environment Canada. 2003. Guidance manual for the categorization of

organic and inorganic substances on Canada’s Domestic Substances List:

Determining persistence, bioaccumulation potential, and inherent toxicity to

non-human organisms. Existing Substances Branch. www.ec.gc.ca/substances/

ese/eng/dsl/cat_index.cfm. Accessed 24 September 2008.

[ECHA] European Chemicals Agency. 2008. Guidance on information require-

ments and chemical safety assessment. Chapter R.11: PBT Assessment.

guidance.echa.europa.eu/docs/guidance_document/information_requirements_

r11_en.pdf?vers530_07_08. Accessed 24 September 2008.

[ECHC] Environment Canada and Health Canada. 2008. Screening assessment for

the challenge: Peroxide, (1,1,4,4-tetramethyl-1,4-butanediyl) bis [(1,1-

dimethylethyl)]. Chemical abstracts service registry number 78–63–7. www.

ec.gc.ca/substances/ese/eng/challenge/batch1/batch1_78-63-7_en.pdf. Ac-

cessed 11 April 2009.

Erickson RJ, McKim JM. 1990a. A simple flow-limited model for exchange of

organic chemicals at fish gills. Environ Toxicol Chem 9:159–165.

Erickson RJ, McKim JM. 1990b. A model for exchange of organic chemicals at

fish gills: Flow and diffusion limitations. Aquat Toxicol 18:175–198.

Fisk AT, Tomy GT, Cymbalisty CD, Muir DCG. 2000. Dietary accumulation and

quantitative structure-activity relationships for depuration and biotransfor-

mation of short (C10), medium (C14), and long (C18) carbon-chain

polychlorinated alkanes by juvenile rainbow trout (Oncorhynchus mykiss).

Environ Toxicol Chem 19:1508–1516.

Flynn GL, Yalkowsky SH. 1972. Correlation and prediction of mass transport

across membranes. I. Influence of alkyl chain length on flux-determining

properties of barrier and diffusant. J Pharm Sci 61:838–851.

Forlin L, Andersson T. 1981. Effects of Clophen A50 on the metabolism of

paranitroanisole in an in vitro perfused rainbow trout liver. Comp Biochem

Physiol 68C:239–242.

Gargas ML, Burgess RJ, Voisard DE, Cason GH, Andersen ME. 1989. Partition

coefficients of low-molecular-weight volatile chemicals in various liquids and

tissues. Toxicol Appl Pharmacol 98:87–99.

Gert-Jan de Maagd P, de Poorte J, Opperhuizen A, Sijm DTHM. 1998. No

influence after various exposure times on the biotransformation rate

constants of benzo(a)anthracene in fathead minnow (Pimephales promelas).

Aquat Toxicol 40:157–169.

Gobas FAPC. 1993. A model for predicting the bioaccumulation of hydrophobic

organic chemicals in aquatic food-webs: Application to Lake Ontario. Ecol

Model 69:1–17.

Gobas FAPC, Clark KE, Shiu WY, Mackay D. 1989. Bioconcentration of

polybrominated benzenes and biphenyls and related superhydrophobic

chemicals in fish: Role of bioavailability and elimination into the feces.

Environ Toxicol Chem 8:231–245.

Gobas FAPC, Kelly BC, Arnot JA. 2003. Quantitative structure-activity relation-

ships for predicting the bioaccumulation of POPs in terrestrial food webs.

QSAR Comb Sci 22:329–336.

Gobas FAPC, Lahittete JM, Garofalo G, Shiu WY. 1988. A novel method for

measuring membrane-water partition coefficients of hydrophobic organic

chemicals: Comparison with 1-octanol-water partitioning. J Pharm Sci

77:265–272.

Gobas FAPC, Mackay D. 1987. Dynamics of hydrophobic organic chemical

bioconcentration in fish. Environ Toxicol Chem 6:495–504.

Gobas FAPC, McCorquodale JR, Haffner GD. 1993. Intestinal absorption and

biomagnifications of organochlorines. Environ Toxicol Chem 12:567–

576.

Gobas FAPC, Morrison HA. 2000. Bioconcentration and biomagnification in the

aquatic environment. In: Boethling RS, Mackay D, editors. Handbook of

property estimation methods for chemicals. Environmental and health

sciences. Boca Raton (FL): CRC. p 189–231.

Gobas FAPC, Muir DCG, Mackay D. 1988. Dynamics of dietary bioaccumulation

and faecal elimination of hydrophobic organic chemicals in fish. Chemo-

sphere 17:943–962.

Gobas FAPC, Opperhuizen A, Hutzinger O. 1986. Bioconcentration of

hydrophobic chemicals in fish: Relationship with membrane permeation.

Environ Toxicol Chem 5:637–646.

Goodrich JA, Kugel JF. 2007. Binding and kinetics for molecular biologists.

Woodbury (NY): Cold Spring Harbor Laboratory Press. 182 p.

Government of Canada. 2000. Persistence and bioaccumulation regulations

(SOR/2000–107). Canada Gazette, Vol. 134. www.ec.gc.ca/CEPARegistry/

regulations/detailReg.cfm?intReg535. Accessed 24 September 2008.

Government of Canada. 2005. New substances notification regulations

(chemicals and polymers) (SOR/2005–247). www.ec.gc.ca/CEPARegistry/

Regulations/DetailReg.cfm?intReg592. Accessed 24 September 2008.

Hamelink JL, Waybrant RC, Ball RC. 1971. A proposal: Exchange equilibria control

the degree chlorinated hydrocarbons are biologically magnified in lentic

environments. Trans Am Fish Soc 100:207–214.

Han X, Mingoia RT, Nabb DL, Yang CH, Snajdr SI, Hoke RA. 2008. Xenobiotic

intrinsic clearance in freshly isolated hepatocytes from rainbow trout

(Oncorhynchus mykiss): Determination of trout hepatocellularity, optimiza-

tion of cell concentrations and comparison of serum and serum-free

incubations. Aquat Toxicol 89:11–17.

Han X, Nabb DL, Mingoia RT, Yang CH. 2007. Determination of xenobiotic

intrinsic clearance in freshly isolated hepatocytes from rainbow trout

(Oncorhynchus mykiss) and rat and its application in bioaccumulation

assessment. Environ Sci Technol 41:3269–3276.

Han X, Nabb DL, Yang CH, Snajdr SI, Mingoia RT. 2009. Liver microsomes and S9

from rainbow trout (Oncorhynchus mykiss): Comparison of basal level

enzyme activities with rat and determination of xenobiotic intrinsic clearance

in support of bioaccumulation assessment. Environ Toxicol Chem 28:481–

488.

Hinderling PH. 1997. Red blood cells: A neglected compartment in pharmaco-

kinetics and pharmacodynamics. Pharmacol Rev 49:279–295.

Houston JB. 1994. Utility of in vitro drug metabolism data in predicting in vivo

metabolic clearance. Biochem Pharmacol 47:1469–1479.

Ito K, Houston JB. 2004. Comparison of the use of liver models for predicting

drug clearance using in vitro kinetic data from hepatic microsomes and

isolated hepatocytes. Pharm Res 21:785–792.

Jabusch TW, Swackhamer DL. 2005. Partitioning of polychlorinated biphenyls in

octanol/water, triolein/water, and membrane/water systems. Chemosphere

60:1270–1278.

Jaworska J, Dimitrov S, Nikolova N, Mekenyan O. 2002. Probabilistic assessment

of biodegradability based on metabolic pathways: CATABOL system. SAR

QSAR Environ Res 13:307–323

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 595

Jones HM, Houston JB. 2004. Substrate depletion approach for determining in

vitro metabolic clearance: Time dependencies in hepatocyte and microsomal

incubations. Drug Metab Dispos 32:973–982.

Jonker MTO, van der Heijden SA. 2007. Bioconcentration factor hydrophobicity

cutoff: An artificial phenomenon reconstructed. Environ Sci Technol

41:7363–7369.

Kansy M, Senner F, Gubernator F. 1998. Physicochemical high throughput

screening: Parallel artificial membrane permeation assay in the description of

passive absorption processes. J Med Chem 41:1007–1009.

Kelly BC, Gobas FAPC. 2001. Bioaccumulation of persistent organic pollutants in

lichen-caribou-wolf food chains of Canada’s central and western arctic.

Environ Sci Technol 35:325–334.

Kelly BC, Gobas FAPC. 2003. An Arctic terrestrial food chain bioaccumulation

model for persistent organic pollutants. Environ Sci Technol 37:2966–2974.

Kelly BC, Ikonomou MG, Blair JD, Morin AE, Gobas FAPC. 2007. Food web-specific

biomagnification of persistent organic pollutants. Science 317:236–238.

Kleinow KM, James MO. 2001. Response of the teleost gastrointestinal system to

xenobiotics. In: Schlenk D, Benson WH, editors. Target organ toxicity in marine

and freshwater teleosts. Vol. 1. London (UK): Taylor and Francis. p 269–362.

Kleinow KM, James MO, Tong Z. 1998. Bioavailability and biotransformation of

benzo(a)pyrene in isolated perfused intestinal preparation from control and

b-naphthoflavone-induced catfish. Environ Health Perspect 106:155–166.

Kleinow KM, Nichols JW, Hayton WL, McKim JM, Barron MG. 2008.

Toxicokinetics in fishes. In: Di Giulio RT, Hinton DE, editors. The toxicology

of fishes. Boca Raton (FL): CRC. p 55–152.

Konwick BJ, Garrison AW, Black MC, Avants JK, Fish AT. 2006. Bioaccumulation,

biotransformation, and metabolite formation of fipronil and chiral legacy

pesticides in rainbow trout. Environ Sci Technol 40:2930–2936.

Kramer S. 2006. Lipid bilayers in ADME: Permeation barriers and distribution

compartments. In: Testa B, Kramer SD, Wunderli-Allenspach H, Folkers G,

editors. Pharmacokinetic profiling in drug research: Biological, physicochem-

ical, and computational strategies. Zurich (CH): Wiley-VCH. p 203–220.

Kratochwil NA, Huber W, Muller F, Kansy M, Gerber PR. 2004. Predicting plasma

protein binding of drugs—Revisited. Curr Opin Drug Discov Dev 7:507–512.

Kwon J-H, Escher BI. 2008. A modified parallel artificial membrane permeability

assay for evaluating bioconcentration of highly hydrophobic chemicals in

fish. Environ Sci Technol 42:1787–1793.

Kwon J-H, Katz LE, Liljestrand HM. 2006. Use of a parallel artificial membrane

system to evaluate passive absorption and elimination in small fish. Environ

Toxicol Chem 25:3083–3092.

Kwon J-H, Wuthrich T, Mayer P, Escher BI. 2007. Dynamic permeation method to

determine partition coefficients of highly hydrophobic chemicals between

poly(dimethylsiloxane) (PDMS) and water. Anal Chem 79:6816–6822.

Kwon J-H, Wuthrich T, Mayer P, Escher BI. 2009. Development of a dynamic

delivery method for in vitro bioassays. Chemosphere 76:83–90.

Kwon Y. 2001. Handbook of essential pharmacokinetics, pharmacodynamics,

and drug metabolism for industrial scientists. New York (NY): Kluwer

Academic/Plenum. 291 p.

Mackay D. 1982. Correlation of bioconcentration factors. Environ Sci Technol

16:274–278.

Mackay D, Fraser A. 2000. Bioaccumulation of persistent organic chemicals:

Mechanisms and models. Environ Pollut 110:375–391.

McKim JM, Schmieder P, Veith G. 1985. Absorption dynamics of organic

chemical transport across trout gills as related to octanol-water partition

coefficient. Toxicol Appl Pharmacol 77:1–10.

Mekenyan O, Dimitrov S, Dimitrova N, Dimitrova G, Pavlov T, Chankov G, Kotov

S, Vasilev K, Vasilev R. 2006. Metabolic activation of chemicals: In-silico

simulation. SAR QSAR Environ Res 17:107–120.

Mekenyan OG, Dimitrov SD, Pavlov TS, Veith GD. 2004. A systematic approach to

simulating metabolism in computational toxicology. I. The TIMES heuristic

modelling framework. Curr Pharm Design 10:1273–1293.

Meylan WM, Howard PH, Boethling RS, Aronson D, Printup H, Gouchie S. 1999.

Improved method for estimating bioconcentration/bioaccumulation factor

from octanol/water partition coefficient. Environ Toxicol Chem 18:664–672.

Murphy JE, Janszen DB, Gargas ML. 1995. An in vitro method for determination

of tissue partition coefficients of non-volatile chemicals such as 2,3,7,8-

tetrachlorodibenzo-p-dioxin and estradiol. J Appl Toxicol 15:147–152.

Neely WB. 1979. Estimating rate constants for uptake and clearance of chemicals

by fish. Environ Sci Technol 13:1506–1510.

Neely WB, Branson DR, Blau GE. 1974. Partition coefficients to measure

bioconcentration potential of organic chemicals in fish. Environ Sci Technol

8:1113–1115.

Nichols J, Erhardt S, Dyer S, James M, Moore M, Plotzke K, Segner H, Schultz I,

Vasiluk L, Weisbroad A. 2007. Workshop report: Use of in vitro absorption,

distribution, metabolism and excretion (ADME) data in bioaccumulation

assessments for fish. Human Ecol Risk Assess 13:1164–1191.

Nichols JW, Fitzsimmons PN, Burkhard LP. 2007. In vitro-in vivo extrapolation of

quantitative hepatic biotransformation data for fish. II. Modeled effects on

chemical bioaccumulation. Environ Toxicol Chem 26:1304–1319.

Nichols JW, Fitzsimmons PN, Whiteman FW. 2004. A physiologically based

toxicokinetic model for dietary uptake of hydrophobic organic compounds

by fish. I. Feeding studies with 2,29,5,59-tetrachlorobiphenyl. Toxicol Sci

77:206–218.

Nichols JW, Schultz IR, Fitzsimmons PN. 2006. In vitro-in vivo extrapolation of

quantitative hepatic biotransformation data for fish. I. A review of methods,

and strategies for incorporating intrinsic clearance estimates into chemical

kinetic models. Aquat Toxicol 78:74–90.

Oliver BG, Niimi AJ. 1985. Bioconcentration factors of some halogenated

organics for rainbow trout: Limitations in their use for prediction of

environmental residues. Environ Sci Technol 19:842–849.

Oliver BG, Niimi AJ. 1988. Trophodynamic analysis of polychlorinated biphenyl

congeners and other chlorinated hydrocarbons in the Lake Ontario

ecosystem. Environ Sci Technol 22:388–397.

Oomen AG, Tolls J, Kruidenier M, Bosgra SSD, Sips A, Groten JP. 2001.

Availability of polychlorinated biphenyls (PCBs) and lindane for uptake by

intestinal Caco-2 cells. Environ Health Perspect 109:731–737.

Opperhuizen A, van der Velde EW, Gobas FAPC, Liem DAK, van der Steen JMD.

1985. Relationship between bioconcentration in fish and steric factors of

hydrophobic chemicals. Chemosphere 14:1871–1896.

Pacifici GM, Viani A. 1992. Methods of determining plasma and tissue binding of

drugs. Clin Pharmacokinet 23:449–468.

Parkerton TF, Arnot JA, Weisbrod AV, Russom C, Hoke RA, Woodburn K, Traas T,

Bonnell M, Burkhard LP, Lampi MA. 2008. Guidance for evaluating in vivo

fish bioaccumulation data. Integr Environ Assess Manag 4:139–155.

Pavan M, Netzeva TI, Worth AP. 2008. Review of literature-based quantitative

structure-activity relationship models for bioconcentration. QSAR Comb Sci

27:21–31.

Robinson P, MacDonald D, Davidson N, Okonski A, Sene A. 2004. Use of

quantitative structure activity relationships (QSARs) in the categorization of

discrete organic substances on Canada’s Domestic Substances List (DSL).

Environ Informat Arch 2:69–78.

Schlenk, D, Celander M, Gallagher EP, George S, James M, Kullman SW, van den

Hurk P, Willett K. 2008. Biotransformation in fishes. In: Di Giulio RT, Hinton

DE, editors. The toxicology of fishes. Boca Raton (FL): CRC. p 153–234.

Schuurmann G. 1990. QSAR analysis of the acute toxicity of organic

phosphorothionates using theoretically derived molecular descriptors.

Environ Toxicol Chem 9:417–428.

Shah P, Jogani V, Bagchi T, Misra A. 2006. Role of Caco-2 cell monolayers in

prediction of intestinal drug absorption. Biotechnol Prog 22:186–198.

Shimizu Y, Takahashi J, Matsubara J, Ikeda K, Matsui S. 2002. Sorption of

polycyclic aromatic hydrocarbons into liposomes (artificial cell membranes)

and effects of dissolved natural organic matter. Lakes Reservoir Res Manage

7:295–299.

Sijm DTHM, Yarechewski AL, Muir DCG, Webster GRB, Seinen W, Opperhuizen A.

1990. Biotransformation and tissue distribution of 1,2,3,7-tetrachlorodiben-

zo-p-dioxin, 1,2,3,4,7-pentachlorodibenzo-p-dioxin and 2,3,4,7,8-penta-

chlorodibenzo-p-dioxin in rainbow trout. Chemosphere 21:845–866.

Southworth GR, Keffer CC, Beauchamp JJ. 1980. Potential and realized

bioconcentration. A comparison of observed and predicted bioconcentration

of azaarenes in the fathead minnow (Pimephales promelas). Environ Sci

Technol 14:1529–1531.

Thomann RV. 1989. Bioaccumulation model of organic chemical distribution in

aquatic food chains. Environ Sci Technol 23:699–707.

Thomann RV, Connolly JP, Parkerton TF. 1992. An equilibrium model of organic

chemical accumulation in aquatic food webs with sediment interaction.

Environ Toxicol Chem 11:615–629.

Tolls J, Lehmann MP, Sijm DTHM. 2000. Quantification of in vivo biotransfor-

mation of the anionic surfactant C12–2-linear alkylbenzene sulfonate in

fathead minnow. Environ Toxicol Chem 19:2394–2400.

596 Integr Environ Assess Manag 5, 2009—JW Nichols et al.

Toutain PL, Bousquet-Melou A. 2007. Free drug faction vs. free drug concentration:

A matter of frequent confusion. J Vet Pharmcol Therap 25:460–463.

[UNEP] United Nations Environment Programme. 2004. Stockholm Convention

on Persistent Organic Pollutants. www.pops.int. Accessed 24 September

2008.

[USEPA] US Environmental Protection Agency. 1995. Great Lakes water quality

initiative technical support document for the procedure to determine

bioaccumulation factors. Washington DC: USEPA, Office of Water. EPA/

820/B-95/005.

Vaes WHJ, Urrestarazu Ramos E, Verhaar HJM, Seinen W, Hermes JLM. 1996.

Measurement of the free concentration using solid-phase microextraction:

Binding to protein. Anal Chem 68:4463–4467.

Van Veld PA, Patton JS, Lee RF. 1988. Effect of preexposure to dietary

benzo[a]pyrene (BP) on the first-pass metabolism of BP by the intestine of

toadfish (Opsanus tau): In vivo studies using portal vein-catheterized fish.

Toxicol Appl Pharmacol 92:255–265.

van Wijk D, Chenier R, Henry T, Hernando MD, Schulte C. 2009. Integrated

approach to PBT and POP prioritization and risk assessment. Integr Environ

Assess Manag 5:697–711.

Veith GD, DeFoe DL, Bergstedt BV. 1979. Measuring and estimating the

bioconcentration factor of chemicals on fish. J Fish Res Bd Can 36:1040–

1048.

Wania F. 2006. Potential of degradable organic chemicals for absolute and

relative enrichment in the Arctic. Environ Sci Technol 40:569–577.

Weed DL. 2005. Weight of evidence: A review of concepts and methods. Risk

Anal 25:1545–1557.

Weisbrod AV, Burkhard LP, Arnot J, Mekenyan O, Howard PH, Russom C,

Boethling R, Sakuratani Y, Traas T, Bridges T, Lutz C, Bonnell M, Woodburn

K, Parkerton T. 2007. Workgroup report: Review of fish bioaccumulation

databases used to identify persistent, bioaccumulative, toxic substances.

Environ Health Perspect 115:255–261.

Wright JD, Boudinot FD, Ujhelyi MR. 1996. Measurement and analysis of

unbound drug concentrations. Clin Pharmocokinet 30:445–462.

Xiang T-X, Anderson BD. 1994. The relationship between permeant size and

permeability in lipid bilayer membranes. J Membrane Biol 140:111–122.

Yang J, Jamei M, Yeo KR, Rostami-Hodjegan A, Tucker GT. 2007. Misuse of the

well-stirred model of hepatic drug clearance. Drug Metab Dispos 35:501–

502.

Prediction-Based Bioaccumulation Assessment—Integr Environ Assess Manag 5, 2009 597