Approaches to developing alternative and predictive toxicology based on PBPK/PD and QSAR modeling.

Systematic toxicity testing, using conventional toxicology methodologies, of single chemicals and chemical mixtures is highly impractical because of the immense numbers of chemicals and chemical mixtures involved and the limited scientific resources. Therefore, the development of unconventional, efficient, and predictive toxicology methods is imperative. Using carcinogenicity as an end point, we present approaches for developing predictive tools for toxicologic evaluation of chemicals and chemical mixtures relevant to environmental contamination. Central to the approaches presented is the integration of physiologically based pharmacokinetic/pharmacodynamic (PBPK/PD) and quantitative structure--activity relationship (QSAR) modeling with focused mechanistically based experimental toxicology. In this development, molecular and cellular biomarkers critical to the carcinogenesis process are evaluated quantitatively between different chemicals and/or chemical mixtures. Examples presented include the integration of PBPK/PD and QSAR modeling with a time-course medium-term liver foci assay, molecular biology and cell proliferation studies. Fourier transform infrared spectroscopic analyses of DNA changes, and cancer modeling to assess and attempt to predict the carcinogenicity of the series of 12 chlorobenzene isomers. Also presented is an ongoing effort to develop and apply a similar approach to chemical mixtures using in vitro cell culture (Syrian hamster embryo cell transformation assay and human keratinocytes) methodologies and in vivo studies. The promise and pitfalls of these developments are elaborated. When successfully applied, these approaches may greatly reduce animal usage, personnel, resources, and time required to evaluate the carcinogenicity of chemicals and chemical mixtures. ImagesFigure 6

This paper is a presentation of the ongoing development of concepts and approaches toward predictive and alternative toxicology, particularly for chemical mixtures, from our laboratories, with collaboration from U.S. and European scientific colleagues. As a starting point, one asks why such a development is necessary. To answer this question adequately, certain reality checks must be emphasized: There Is No Such Thing as a Single Chemical Exposure This issue has been discussed in detail elsewhere (1)(2)(3). Briefly, a single chemical often ends up with many metabolites in the body. Furthermore, considering the multiple chemicals to which we are commonly exposed in foods, drinks, medicines, cosmetics, and indoor and outdoor pollutants and the fact that even our own body is a chemical mixture, there is really no such thing as a single chemical exposure in anyone's life.

Conventional Toxicology Studies Cannot Keep Pace with Single Chemicals, Let Alone Chemical Mixares
The National Cancer Institute's Carcinogenesis Bioassay Program was established in 1962 (4). This bioassay program was later transferred to the U.S. National Toxicology Program upon its establishment in 1978. In the bioassay program's combined 36 years of operation, approximately 500 chemicals have been studied for carcinogenicity and other chronic toxicities (5). These studies and the related range-finding and dose-setting studies are extremely expensive, they require large numbers of animals, and the study duration is long (6). Even though these studies are the gold standards, considering the approximately 70,000 to 600,000 chemicals in commerce (7)(8)(9), the number of chemicals for which we currently have adequate toxicology information for risk assessment is minuscule. At the present mode and rate of study of these chemicals, it is doubtful that our society will ever have a thorough toxicologic evaluation on the majority of the chemicals that are used now or may be used in the future. Further consideration of the issue of health effects of This paper is based on a presentation at the Conference on Current Issues on Chemical Mixtures held 1 [1][2][3][4][5][6][7][8][9][10][11][12][13]  chemical mixture exposures (i.e., real-world issues), it is impossible to deal with the problems of combination toxicology of chemical mixtures by adopting the approach of systematic conventional toxicology and carcinogenicity testing (1,10).
Predictive and Altemative Toxicology Must Be Developed During our approximately 15 years of active research in the toxicology of chemical mixtures, we have become resigned to the fact that the conventional animal toxicology testing methods will not work for chemical mixtures, particularly in the evaluation of carcinogenicity (2,3,6). We believe that to deal with chemical mixture issues effectively we must fully utilize and integrate computational technology, mathematical and statistical modeling, mechanistically based short-term toxicology studies, and cellular and molecular biology methodologies. Furthermore, an efficient experimental approach or system for chemical mixtures must at least meet the following critical requirements: It must be relatively simple, short-term, and inexpensive; be based on the best science; incorporate mechanisms of toxicity; have broad applicability; and have predictive capability.
The Inaeasing Application ofComputer Tecnology in Toxicology Is Inevitle Widespread application of computers in risk assessment, particularly the use of linearized multistage and other models for cancer risk assessment, has been a reality for more than two decades. Since the late 1 980s, the advancement of physiologically based pharmacokinetic (PBPK) modeling and the integration of PBPK/pharmacodynamic (PD) modeling into the risk assessment process further enhanced the utilization of computer technology. Thus, computer application in toxicology is already a fact. Looking into the future, it becomes obvious that some type of predictive approach must be developed to handle the huge number of single chemicals in commerce and the nearly infinite number of chemical mixtures in the environment. It is inevitable that computer technology will be heavily involved in any development of predictive toxicology. Prediction of health effects usually involves some type of mathematical modeling, which may range from the classical compartmental pharmacokinetic modeling to the currently advancing PBPK/PD, biologically based dose response (BBDR), and/or quantitative structure-activity relationship (QSAR) modeling. Some successes, including those from our studies, in prediction of pharmacokinetic fate and toxicity of simple chemical mixtures are already evident in the literature (11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22). The toxicologic end points of prediction, for instance, include an interaction threshold (18,19) and acute lethality due to hepatic injuries (20).
For more complex mixtures the integration of PBPK/PD, BBDR, QSAR modeling, and lumping analysis (a modeling tool developed in the petroleum industry) may allow the development of a predictive tool for the health effects (23). In the 1960s the application of lumping analysis rendered it possible to predict gasoline production based on a few lumps (i.e., similar groups of chemicals based on carbon numbers or boiling points) rather than the thousands of component chemicals of the petroleum (24,25). Thus, even though relatively little is known about the complex mixture of petroleum, a predictive tool was developed and applied from modeling. If a relatively crude lumping analysis was able to predict some aspects of catalytic cracking of petroleum 30 years ago, why can't we attempt to predict health effects from chemical mixtures through a much more sophisticated modeling technique such as structure-oriented lumping (26,27)? Accordingly, our laboratory, in conjunction with scientists from other institutions, is actively pursuing this area of research (28).

Long-Term, Low-Levi Exposures Are
Bdlow the Sensitivity ofPresent-Day

Methods ofExperimental Toxicology
Conventional toxicology methods are mainly observational and descriptive. As an example, in carcinogenesis the animals, usually rodents, are dosed with the chemical of interest for 2 years, the surviving animals are sacriflced, and the tissues are examined histopathologically for the presence of tumors. Because the number of animals per group is limited, usually fairly high dose levels must be used to elicit tumorigenic responses. In cancer risk assessment these high-dose experimental results are extrapolated to the very low dose (i.e., environmentally realistic) regions where such conventional experimental protocols would not be able to detect any carcinogenic responses. Even with the newly proposed point of departure approach for cancer risk assessment (29), the lower 95% confidence limit on a dose associated with 5 or 10% extra risk (LED05 or LEDIO, respectively) region is still not easily attainable with conventional protocols. Moreover, these dose levels (i.e., effective dose, ED05 or EDIo) are still orders of magnitude higher than the estimated environmental exposure levels.
Recent advances in molecular biology techniques offer unique opportunities for the development of more efficient and sensitive methods for toxicology. For instance, techniques such as polymerase chain reaction or reverse transcriptase polymerase chain reaction, coupled with mechanistic studies of the cancer process in cell culture systems, may detect early or late genotypic changes related to carcinogenesis at dose levels much below the limits of detection of more conventional toxicology methods. As we propose here, the integration of this type of mechanistic data with PBPK/PD, BBDR, and/or QSAR modeling may formulate an efficient and predictive approach for carcinogenic potentials of chemicals and chemical mixtures.
Because this is a review paper, many experimental details are not provided. Readers are referred to the papers cited for additional information. The following discussion will follow the chronological order of development of these approaches in our laboratories.

Experimental Approaches
This portion of the paper follows the evolution of our thinking in the last few years on how to develop predictive approaches for single chemicals and chemical mixtures. To focus our presentation we chose to limit the discussion to carcinogenesis only. Initially, we concentrated our effort on a medium-term (i.e., 8 weeks) in vivo experimental approach and its integration with PBPK/PD modeling. Although we made progress on this front and obtained interesting findings, it was soon obvious that even this shorter term, more efficient in vivo system was too resource intensive to be routinely used for chemical mixture work. Therefore, still more efficient systems must be developed. We investigated in vitro systems with our specific criteria (i.e., relatively simple, shortterm, and inexpensive; based on the best science; understanding mechanisms of toxicity; broad applicability; and predictive capability) in mind. We found that cell culture systems offer unique opportunities, particularly in mechanistic and time-course studies related to carcinogenesis. Thus, as a further development, in vitro mechanistic studies using cell culture systems and their Environmental Health Perspectives * Vol 106, Supplement 6 * December 1998 integration with pharmacodynamic modeling became a major emphasis in our laboratories. Our present thinking is that cell culture systems and in vivo animal studies may work in concert to bring about much-needed mechanistic and pharmacokinetic information for PBPK/PD modeling. When quantitative information on the selected molecular and cellular end points is available on chemicals with structural correlation, QSAR modeling may then be applied to assess their relationship with carcinogenic potentials. In doing so, the positive QSAR relationship between certain end points of a series of chemicals with their respective carcinogenic potentials may be used to develop a predictive approach for other untested chemicals with structural similarities.

The Medium-Term Liver Foci Bioassay
To simplify the detection of carcinogenicity of chemicals and chemical mixtures, Ito and colleagues developed the medium-term liver foci bioassay (30,31). The details of experimental protocol of this bioassay may be found in Ito et al. (30,31); the bioassay is discussed briefly below.
The medium-term liver foci bioassay (30,31) utilizes the placental form of glutathione S-transferase (GST-P) as a marker for rat hepatic preneoplastic and neoplastic lesions (32,33). The medium-term hepatocarcinogenesis bioassay of Ito et al. (30,31), an 8-week experiment, utilizes F344 rats that are given a single dose of diethylnitrosamine (DEN) to initiate carcinogenesis. After a 2-week period the rats are given repeated exposure to a test compound for a dosing period of 6 weeks. At the end of week 3, rats are subjected to partial hepatectomy to maximize opportunities of promotion via a high rate of cell proliferation. All rats are sacrificed at the end of week 8 for evaluation of development of preneoplastic hepatocellular nodules by staining for expression of GST-P (30,31). Extensive testing has demonstrated that the induction of GST-P-positive foci in the medium-term bioassay for liver carcinogens correlates well with the incidence of hepatocellular carcinomas in parallel long-term assays (30,31). Two hundred seventy-seven chemicals have been tested in Ito et al.'s (34) mediumterm liver foci bioassay; it has correctly identified 97% of genotoxic hepatocarcinogens and 84% of nongenotoxic hepatocarcinogens (34). Thus, for rapid screening of large numbers of chemicals and for reduction in the required number of animals, this assay is of great advantage.

A Modified Tlme-Course Medium-Term Liver Foci Bioassay
In the original protocol of Ito et al.'s (30,31) medium-term liver foci bioassay, the animals undergo terminal sacrifice at the end of the 8-week experimental period and the data are for only one time point. To obtain sufficient data for constructing a viable PBPK/PD model ("Integration of In Vivo Studies and PBPK/PD Modeling"), time-course events must be studied. Therefore, we modified the original Ito et al. (30,31) protocol into time-course studies based on anticipated cell proliferation rates during this 8-week experimental period ( Figure 1). With this modified protocol we obtained pharmacokinetic data, cell proliferation rates, selected molecular and biochemical information (i.e., c-jun, cfos, CYP1A2, reduced glutathione (GSH): oxidized glutahione (GSSG) ratios, porphyrin levels) as well as morphometric data on foci at different stages of the foci development following treatment of several individual chlorobenzene isomers. Detailed descriptions of methodologies involved in the measurement of these parameters are discussed in Thomas et al. (35,36) and Thomas (37).

Integation of in Vivo Studies and PBPK/PD Modeling
The PBPK/PD modeling of Ito et al.'s (30,31) medium-term liver foci bioassay presented some unique challenges. Whereas the foundation of this modeling approach was laid by using pentachlorobenzene (37) as a model system, the principle is applicable for other chemicals or chemical mixtures. The evolution of the approach for a 0 Weeks 4 6 PBPK/PD model for tumorigenesis in our laboratories followed roughly three stages. The first stage was the most ambitious stage, in which we formulated a model as shown in Figure 2. Because cell-cycle kinetics was incorporated into this model, our initial effort was devoted to a biologically based mathematical model of the effects of partial hepatectomy on cell cycle kinetics (38) and a comparison of quantitative immunohistochemical markers for cell-cycle specific changes in F344 rats (39). However, it was soon apparent that there were two pitfalls. First, whereas the proliferating cell nuclear antigen staining appeared to work well for the liver cells undergoing rapid proliferation such as after two-thirds partial hepatectomy (39), quantitative morphometric analysis would be prohibitively resource-intensive for liver at resting state, which has a low rate of cell turnover. Second, although a biologically motivated model depicting cell-cycle kinetics a) Abbreviations: PH, partial hepatectomy; LI, labeling index. Each arrow under the time axis depicts the following experimental procedures: 1) sacrifice five rats/group/time point following a 3to 5-day bromodexyuridine exposure; 2) liver tissue for Ll measurement; 3) liver tissue for molecular biology studies; 4) tissues for chemical analyses; 5) liver tissue for oxidative stress studies; and 6) histopathology and morphometric analyses on liver. Environmental Health Perspectives * Vol 106, Supplement 6 * December 1998 was constructed and computer simulations were consistent with available literature data (38), the data requirements for applying such a model to the specific chemicals and bioassay systems that we were studying were beyond the reach of our available resources and personnel. Thus, we were overly ambitious in our initial modeling attempt.
In the next stage we simplified the model greatly. As shown in Figure 3, this model depicts (from the top to the bottom) that the normal cells have a growth function (PN) that is related to the total number of cells, their proliferation rate, and time. These cells may die at a death rate (kdiel) or may be injured (DNA damage) by chemicals at the rate of kinj. The cellular repair processes may fix the damage at the rate of krep. The injured cell may go on to become initiated cells (foci) at the rate of kinit or go on to die at the death rate of kdie2. The initiated cells have their growth function (PI) and death rate (kdie3) as well. This model (Figure 3) was a prototype and we kept it as simple as possible. Using this model, we attempted to simulate the data published by Tatematsu et  Our subsequent modeling effort (stage 3) was directed toward the linking of a PBPK model with the first two cellular states of the two-stage cancer model (the Moolgavkar-Venzon-Knudson model). Part one of the modeling effort involved the PBPK simulation of single-and multiplegavage dose exposures (37)-a reflection of the experimental procedures. Part two took into consideration the physiologic changes resulting from two-thirds partial hepatectomy (37)-a drastic physiologic state. Part three included linking a deterministic model with a stochastic model and the pharmacodynamic processes involved in the formation of foci (37), the initiated cells in the two-stage carcinogenesis model. To date, all these modeling activities were based on using pentachlorobenzene as a model chemical in the modified time-course mediumterm liver foci bioassay system. However, the principles derived from this effort may easily be applicable to chemical mixtures.
Our present and future modeling direction aims at going beyond the initiated cell stage-i.e., to look at the processes involved in the later events of the multistage carcinogenesis process. Relevant discussion follows in "The Integration of Mechanistic Information from in Vitro and in Vivo Studies and Pharmacodynamic Modeling."

Studies ofMechanisms of Carcinogenesis Using Cdl Cultures following Treatment of Chemicals andlor Chemical Mixtures
Cell culture systems, because of their simplicity and low cost, make it possible to determine the oncogenic potential and mechanisms of transformation of a variety of chemicals and chemical mixtures at the cellular level. Syrian hamster embyro (SHE) cells have the advantage that the carcinogenic potential of many chemicals has been evaluated using morphologic transformation as an index. This has largely been through the effort of scientists at The Procter & Gamble Company (41,42). Epidermal keratinocytes are one of the most well-characterized cell culture systems for studying transformation.
For both SHE cells and human keratinocytes, cytotoxicity and altered cellular growth characteristics and induced malignant transformation (i.e., tumorigenicity) are end points for the evaluation of chemical mixtures of interest. In addition, in keratinocytes these studies may be extended to link observed phenotypic changes with alterations in expression of defined cell cycle regulators and growth factors. These biologic experiments should be carried out in an iterative manner with the development of a biologically based model. The concept of model-directed experimentation should be followed to save resources by avoiding unnecessary experimentation.
The overall idea of studying the carcinogenesis process using human keratinocytes    is summarized in Figure 6. There are many phenotypic and genotypic changes accompanying the transformation of human keratinocytes; some are early events and others are late events. To cover these changes, commercially available primary human keratinocytes may be used to study the early events (43). A number of immortal keratinocyte cell lines such as HaCat (44), RHEK-1 (45), and NM-1 (46) can be used to study the later events. Selected pertinent biomarkers such as transforming growth factor (TGF)-a, TGF-3, c-myc, c-ras, and p53 may be studied and quantitative information obtained. Such quantitative information may be utilized to calibrate and verify BBDR models.

Integraton ofMechanistic Information ffrom in Vtro and in Vivo Studies and Pharmacodynamic Modeling
We summarize our approach on the integration of mechanistic information and modeling for the assessment of carcinogenic potential for chemicals in the algorithm in Figure 7. We believe that because cancer is a cellular process, the carcinogenic potential of chemicals can be identified through the studies of selected key molecular/cellular markers of the cancer processes; also, QSAR and pharmacodynamic modeling of these key biomarkers for cellular process can be effectively utilized to develop a predictive tool for evaluating carcinogenic potential within certain classes of chemicals.
We listed multiple chemicals for chemical mixtures, but this general approach ( Figure 7) is applicable for s as well. The short-term alt may be represented by the ' formation assay and the n course medium-term li (Figure 1); other such test The molecular or cellular b a number of processes le cancer (including loss of g failure to commit apoptosi repair) may be specificall) compound or a class of con time-course studies, critical (e.g., cellular birth, death, rates for primary and imm( of changes of growth fa expression of oncogene at pressor genes) may be gem structing the cancer model(s data on the selected molecu markers for the cancer proce in QSAR modeling for a clk or chemicals with structur Such integration of QSAR, ing, and cancer modelin modeling) will then provide for developing predictive t risk assessment.

ModIeling for the Develop of Predictive Capability
For the last few years, we 1 great deal of effort towarc predictive toxicology appr homolog series of 12 chloro will use chlorobenzenes a, the same general princip applicable to other classes of

Malignant
The chlorobenzenes are a small group keratinocyte. of congeneric chemicals. There are 12 congeners in total-from monoto hexachlorobenzene. Most of these are solid at ambient temperature and pressure, except 4. for mono-and some dichlorobenzenes.  (47). In a QSAR model, a correlation is developed between one or more quantitative structural, physicochemical, and/or S_ biologic descriptors and a toxicologic end Paptiloma point (48). The toxicologic end point can formaton either be quantitative (e.g., an effect level such as the median lethal concentration) erations in human or qualitative (e.g., mutagenic or nonmutagenic). Hansch et al. (49) provide an interesting overview of the use of QSAR ,ingle chemicals in toxicology. ternative assays Our plan is to build and verify QSAR SHE cell trans-models based on our ongoing studies of nodified timemolecular and cellular biomarkers of prever foci assay neoplastic foci formation from exposures :s are available. to hexa-, penta-, 1,2,4,5-tetra-, and 1,4iomarkers from dichlorobenzenes; to use these verified ading toward QSAR models to predict each of the toxi-,rowth control, cologic responses and PBPK modeling s, loss of DNA parameters studied to date for the remainf targeted by a ing eight chlorobenzene isomers (1,2,3,4npounds. From tetra-, 1,2,3,5-tetra-, 1,2,3,-tri-, 1,2,4,-tri-, rate constants 1,3,5-tri-, 1,2-di-, 1,3-di-, and monoand mutation chlorobenzenes); and to verify the QSAR Drtal cells, rates model predictions by conducting molecuctors, and/or lar biology/biochemical experiments and nd tumor-sup-PBPK/PD modeling studies on selected erated for con-chlorobenzene isomers using in vitro and s). Quantitative in vivo methods. To achieve that we will lar/cellular bio-proceed as follows: ess may be used The array of toxicologic data on the ass of chemicals first four selected chlorobenzenes will be ral similarities. studied using pattern recognition methods PBPK model-(50,51) to determine which data correlate g (i.e., PBPD with the carcinogenic nature of the com-5 us with a tool pounds. The methods that will be used are toxicology and principal component analysis, linear discriminant analysis, cluster analysis, and soft independent modeling of class analogy and QSAR (50,51). All of these methods can be used ?ment to study which biologic descriptors (or combination of descriptors) are correlated have devoted a with or even responsible for the carcino-A developing a genic nature of the active chlorobenzenes. oach using the In the next step the toxicologic end points obenzenes. We that are perceived important for the s a case study; understanding of the carcinogenicity of dles should be these compounds will be studied with chemicals.
To date, the TRs measured in our ongoing project on hexa-, penta-, 1,2,4,5tetra-and 1,4-dichlorobenzenes include GST-P-positive foci (numbers and area); alteration of c-jun and c-fos expression (thus AP-1 related gene expression of GST-P); changes in CYP1A2 expression (thus Ah-receptor related activity); liver porphyrin levels (possible involvement of oxidative stress); alterations in the GSH:GSSG ratio; and induction of DNA damage (via Fourier transform infrared analyses). Furthermore, time-course changes were also obtained for liver cell proliferation rates under the influence of DEN, chlorobenzene treatment, and partial hepatectomy in the modified timecourse medium-term liver foci bioassay ( Figure 1). All these TRs are either quantitative or semiquantitative parameters; thus, they are amenable to QSAR analysis. Of course, judgment on the potency and importance of these TRs must be made with respect to the toxicologic end point, in this case the foci formation.
For QSAR correlation and prediction of PBPK modeling parameters on the 12 isomers of chlorobenzenes, we also intended to apply the following general QSAR model for all relevant PBPK parameters: QSAR correlation and prediction of toxicologic responses on a number of molecular or cellular biomarkers for carcinogenic potentials of the 12 isomers of chlorobenzenes will be carried out in line with the classical Hansch multiple linear regression (MLR) analysis approach (52)(53)(54). The correlation developed will be similar to the following description, with the understanding that actual relationships will be calculated using partial least square (PLS) regression analysis ( The PKs measured or estimated to date in our ongoing project on hexa-, penta-, 1,2,4,5-tetraand 1,4-dichlorobenzenes include tissue partition coefficients for liver, fat, muscle (representing a slowly perfused tissue group), and kidney (representing a rapidly perfused tissue group) and in vivo metabolic rate constants (i.e., Km and Vmax). The priority of selecting PKs for QSAR modeling will be based on sensitivity analyses in PBPK modeling as to their respective influences on computer simulations; the greater the influence, the higher the priority. Environmental Health Perspectives * Vol 106. Supplement 6 * December 1998 The work of de Jongh and co-workers (65) showed that for a large set of diverse chemicals, in both rats and humans, the tissue:blood partition coefficients could be described by a general function of the compound's log Kow of the following form: where Pt,b is the tissue to blood partition coefficient, fW,t is the fraction of water in the tissue, fl,t is the fraction of lipid in the tissue, ft,b is the fraction of water in the blood, flb is the fraction of lipid in the blood, a and b are exponents, and C is an adjustment factor to provide for the effect of protein binding. This methodology will be used for QSAR modeling of tissue partition coefficients. For metabolic rate constants, the Hansch-like approach using PLS as described earlier.

Conclusion
In this paper we described the approaches for the possible development of a predictive tool for the carcinogenic potential of chemicals and chemical mixtures. For illustrative purposes we used single chemicals (i.e., chlorobenzenes) to develop the concept and approach for predicting carcinogenic potentials of different isomers of chlorobenzenes. Toxicologic interactions may be at the levels of pharmacokinetics and/or pharmacodynamics. Those toxicologic interactions that are relevant to the end point (in this case carcinogenicity) may be linked for different chemicals in a chemical mixture via PBPK/PD modeling. QSAR modeling may be used to extrapolate to other chemicals that are structurally similar to the components of a chemical mixture under study. In this integrated manner we may develop predictive capabilities for the toxicology of chemical mixtures.
How do we approach the issue of risk assessment once such modeling and the generation of quantitative information are realized? The linchpin of a realistic risk assessment is an accurate estimate of target tissue dosimetry from which a virtually safe or safe exposure dose may be calculated. PBPK/PD or BBDR models are powerful tools to define such target tissue dosimetry. For chemical mixtures, three scenarios can be considered. In the first case where a simple chemical mixture with known mechanism(s) of toxicologic interaction, PBPK/PD or BBDR modeling of the toxicologic interaction can provide estimates on an interaction threshold, which may in turn be used as a benchmark dose (e.g., LEDO0 or LED05 for a chemical mixture) for risk assessment. In the second case, a chemical mixture (most likely a more complex one with structurally similar components such as a hydrocarbon mixture) may provide quantitative changes of many biomarkers for a toxicologic process as well as defined toxicity for a specific end point. Although the mechanism(s) of interaction may not be entirely clear, PBPK/PD or BBDR modeling may be utilized to estimate the target tissue dosimetry of this chemical mixture as a lump (i.e., a pseudosingle chemical) or as an average chemical based on the average characteristics of such a mixture. In the third case a chemical mixture may have completely different composition including metals, aliphatics, and aromatics. The principles and approaches stated previously may still apply as long as the like components are lumped and quantitative toxicologic information can be obtained for each of the lumps.
The role of QSAR modeling in the overall scheme is to provide interpolations and extrapolations for estimated values for TRs and PKs for those chemicals that are not studied experimentally in a class of chemicals (such as chlorobenzenes) or chemical mixtures (such as hydrocarbons, metals, etc.). These gaps may be filled using QSAR based on structural differences and the related physicochemical changes. In doing so we may carry out PBPK/PD or BBDR modeling on these chemicals or chemical mixtures on an a priori basis.
Discussion in this paper consists of past and ongoing studies as well as theoretical deliberations. Much of the development is still at an embryonic stage. We invite criticisms and suggestions for further modification and refinement of our approaches. Through this type of process we will be able to work toward the goal of establishing a scientifically credible predictive and alternative toxicology.