Water, Air, & Soil Pollution: Focus

, Volume 8, Issue 1, pp 3–21

A Multiscale Approach for Assessing the Interactions of Environmental and Biological Systems in a Holistic Health Risk Assessment Framework


    • Computational Chemodynamics Laboratory (CCL), Environmental and Occupational Health Sciences Institute (EOHSI)UMDNJ Robert Wood Johnson Medical School and Rutgers University

DOI: 10.1007/s11267-007-9137-7

Cite this article as:
Georgopoulos, P.G. Water Air Soil Pollut: Focus (2008) 8: 3. doi:10.1007/s11267-007-9137-7


Advances in computing processing power and in availability of environmental and biological data have allowed the development and application of comprehensive modeling systems that utilize a holistic, integrated, approach for assessing the interactions of environmental and biological systems across multiple scales of spatiotemporal extent and biological organization. This approach allows mechanism-based environmental health risk assessments in a person-oriented framework, which accounts for simultaneous exposures to contaminants from multiple media, routes, and pathways. The conceptual basis and example applications of the Modeling ENvironment for TOtal Risk (MENTOR), and the DOse–Response Information ANalysis system (DORIAN) are presented.


Comprehensive modeling systemsDORIANMENTOREnvironmental healthRisk assessmentsExposures to mixtures

1 Introduction: The Evolution of Environmental Risk Assessment

Advances in computational technologies (e.g. affordable cluster and grid computing hardware; efficient software for the management and analysis of massive, distributed, multiscale and multidimensional datasets; fast algorithms for numerical simulation and uncertainty analysis) and the development of widely accessible databases of environmental, physicochemical, biological, demographic, and other information are forcing a major “paradigm shift” in the approaches for analyzing and quantifying the complex interactions of human beings (and other biological entities) and their environment. Human activities continuously modify the environment on a variety of scales, from indoor to global, and the resulting changes of environmental properties act as stressors that can affect human and ecological health. Assessing the health risks associated with environmental factors requires one to understand and quantify, in a mechanistic way, the events and processes involved in the “environmental health sequence” (presented schematically in Fig. 1) from “source” (e.g. the release or the formation of a stressor, such as a chemical or a radiological contaminant or a biological agent in an environmental medium) to “outcome” (e.g. development of an environmentally caused disease).
Fig. 1

Schematic representation of the “environmental health sequence” from “source” to “outcome,” involving a series of environmental and biological “steps”

Human health state reflects multiscale body “system dynamics” that involve interactions among multiple levels (scales) of biological organization (genome, transcriptome, proteome, metabolome, cytome, physiome) that are in turn affected by various types of environmental (“extragenomic”) factors. In fact, what are commonly referred to as “Gene-Environment” interactions occur at various steps of the exposure-to-effect sequence, as genes may be controlling processes that affect exposure, dose, toxicokinetics, and toxicodynamics; at the same time, extragenomic factors and development/aging may also control various aspects of genome dynamics (Fig. 2). The statement, “Genetics loads the gun, but environment pulls the trigger” (Judith Stern, Professor of Nutrition and Internal Medicine, University of California at Davis) is often made. This statement is, of course, a “grand simplification,” but a very useful one, that is affecting both popular perception and evolving approaches toward mechanistic environmental health risk assessments. “Timing” and combination of exposures as well as behavior (which can be affected by genetics and can modify the environment and one’s exposures) are also critical for an individual’s risk. The genetic code for any individual can be seen as a “set of initial conditions” for a dynamic biological system that will evolve under the influence of environmental conditions, which depend critically on human behavior. A “person-specific,” anthropocentric approach, which takes into account all the above, is therefore necessary for realistic risk characterization.
Fig. 2

Human health state reflects multiscale body “system dynamics” (genome, transcriptome, proteome, metabolome, cytome, physiome) that are affected by environmental (“extragenomic”) factors

In the past, assessing the human impact on the environment and the subsequent impact of environmental change on human well-being, was typically performed on a “stressor-by-stressor basis” (e.g. for individual chemical, radiological, or biological agents) and for a given environmental medium (e.g. atmosphere, groundwater, etc.) typically associated with a single exposure route (e.g. inhalation, ingestion, dermal absorption, etc.; see Table 1). This practice has evolved over time to allow consideration of multiple contaminants present in multiple media and associated with different exposure routes. Nevertheless, the focus has remained on contaminant-specific and route-specific risk characterization, either for a hypothetical “maximally exposed” individual or for a general population with simplified distributions of demographic and physiological attributes. In recent years, however, a major shift has been taking place through the development of “person-oriented” (i.e. anthropocentric) “holistic” approaches and models that aim to account for total (“cumulative” and “aggregate”) exposures of individuals (or of populations consisting of such individuals) to co-occurring stressors (e.g. mixtures of contaminants present in the air inhaled and in the water and food digested by the person; Buck et al. 2003; Christensen et al. 2003; NUREG 2002; Price et al. 2003; Price and Chaisson 2005; USEPA 2003b, 2006; WHO 2005). These new holistic, “systems-based” approaches focus on individual persons, real or “virtual,” with well defined physiological, socioeconomic, etc. attributes, and take into account how the detailed activities of these persons in space and time affect their “microenvironments” and their corresponding exposures to stressors, as well as how they affect the physiological processes determining biologically relevant dose (e.g. inhalation rates, metabolic transformation/elimination rates, etc.). Proceeding a step further, person-oriented approaches aim to take advantage of information on genetic and other factors that may determine individual susceptibility to environmental stressors. The current explosion in toxicoinformatic (genomic, transcriptomic, proteomic, metabonomic, etc.) data is expected with time to provide critical understanding of the interindividual variability in responses to various environmental factors and of related disease susceptibility, which will eventually give rise to truly “personalized” risk assessments.
Table 1

The evolving focus of environmental human and ecological health risk analysis

Past: single pathway analysis of risk

Present: multiple pathway analysis of risk

Future: integrated “person-oriented” systems analysis of risk

Single contaminant

Multiple contaminants

Mixtures of contaminants with environmental and biological interactions

Multiple contaminant sources

Multiple contaminant sources

Multiple co-occurring chemical and nonchemical stressors affecting an individual

Single medium environmental fate and transport

Linked fate and transport in different environmental media

Dynamically integrated multimedia fate and transport in the environmental and biological systems

Single exposure route

Multiple exposure routes

Aggregate/cumulative exposure and dose analysis

Phenotype-based toxicity

Phenotype-based toxicity with susceptibility considerations

Mechanistic linkage of phenotype with genotype

Primary human health criteria for individual contaminants

Chemical and exposure-route specific risk for “standard individuals”

Aggregated risk for diverse human populations (with susceptible subpopulations)

Qualitative uncertainty

Quantitative uncertainty

Quantitative uncertainty and variability resolved for specific environmental and biological processes

The following sections present a summary overview of a continuing research effort, funded primarily by the US Environmental Protection Agency (USEPA), to develop, evaluate, and apply mechanistic and diagnostic computational modeling tools intended to support comprehensive analyses of all steps in the “environmental health sequence” of Fig. 1. This effort has resulted in the evolution of two “libraries of software components,” involving both modeling and data modules, developed in C++, Java, and Matlab, and operating, in conjunction with database engines, on customized Linux clusters. These libraries constitute the Modeling ENvironment for TOtal Risk studies (MENTOR) that addresses the “source-to-dose” steps, and the DOse–Response Information ANalysis system (DORIAN) for the biological “dose-to-effect” steps. It should be noted that the analyses supported by MENTOR and DORIAN are in principle “bi-directional,” i.e. allowing not only calculations of dose and biological effect from environmental and exposure information, but also, under certain conditions, the reconstruction of environmental exposure patterns from appropriate biomarker data.

2 The Modeling Environment for Total Risk Studies (MENTOR)

MENTOR is an open computational toolbox that provides modeling and data analysis tools to facilitate the assessment of cumulative (i.e. over time) and aggregate (multipathway) exposures to mixtures of environmental contaminants, by following a mechanistically consistent probabilistic source-to-dose modeling approach. MENTOR utilizes existing models, when available, and also provides new approaches to “fill gaps” in the source-to-dose sequence (Georgopoulos et al. 2005a, b; Georgopoulos and Lioy 2006). Thus, MENTOR links dynamic numerical models for environmental fate/transport and for human exposure and dose calculations; these models are coupled with up-to-date national, regional, and local (currently primarily from the USA) databases of environmental, microenvironmental, biological, physiological, demographic, etc. parameters (see Appendix for a representative list of such databases). A major issue in implementing consistent source-to-dose modeling is sequentially going from global/regional to local, neighborhood, and eventually personal resolution. MENTOR provides “tools” that link regional and local information with microenvironmental conditions and human activities, as shown schematically in Fig. 3, which is adapted from concepts in Georgopoulos et al. (1997) and USEPA (2003a).
Fig. 3

MENTOR supports source-to-dose modeling over multiple scales, down to resolutions relevant to exposures of individuals; it links environmental models and databases with exposure/dose models

In order to provide “user-oriented” implementations of MENTOR for specific applications, a number of “customized” versions have been developed, as follows:
  • ▪ MENTOR-1A (Georgopoulos et al. 2005a) employs a “one atmosphere” approach to characterize simultaneous exposures of individuals or populations to multiple, co-occurring atmospheric contaminants (e.g. ozone, fine particles, air toxics, etc.). The “one atmosphere” approach places emphasis on accounting for the dynamic physical and chemical interactions of the contaminants, in both ambient air and in specific residential, occupational, and vehicular microenvironments.

  • ▪ MENTOR-4M (Georgopoulos et al. 2005b, 2006) provides a unified multiroute/multipathway modeling framework for aggregate and cumulative exposure assessments. The focus is on contaminants that are present in multiple environmental media and on mixtures of such contaminants.

  • ▪ MENTOR-3P is a modular, “generalized” physiologically-based pharmacokinetic modeling framework for human populations that allows simultaneous consideration of multiple, co-occurring chemicals in a consistent manner, while incorporating information for attributes of intra- and inter-individual variation and variability, either physiological or biochemical, from various recent databases (CDC 2005; The Lifeline Group 2006).

  • ▪ MENTOR-2E provides computational tools appropriate for characterizing exposures specifically for emergency events, as a means for improving planning for emergency response. Accordingly, emphasis is given to the critical attributes of physicochemical processes across the various temporal and spatial scales that are relevant to the evolution of the emergency event (Georgopoulos et al. 2004; Stenchikov et al. 2006).

  • ▪ MENTOR-DOT provides a wide range of state-of-the-art diagnostic tools, such as the Stochastic Response Surface Method (SRSM) (e.g. Balakrishnan et al. 2003, 2005), the High Dimensional Model Representation (HDMR) (e.g. Wang et al. 2005a), and Bayesian Markov Chain Monte Carlo (MCMC) optimization techniques (e.g. Balakrishnan et al. 2003), which allow comprehensive sensitivity/uncertainty analysis of complex system models, systematic complex model reduction to obtain Fast Equivalent Operational Models (FEOMs), and efficient model/data fusion.

Typically, MENTOR simulations consist of: (1) Estimation of (multimedia) background levels of environmental pollutants (in air, water, and soil) through either (a) multivariate spatiotemporal analysis of monitor data or (b) regional-scale environmental model predictions; (2) Estimation of local multimedia pollutant levels in an appropriate administrative unit (such as a census tract) or a conveniently defined grid through either (a) field study measurements, (b) subgrid “corrections” of regional model estimates, or (c) application of a local scale environmental model; (3) Characterization of essential demographic attributes of populations (geographic density, age, gender, race, occupation, income, etc.); (4) Development of activity event sequences for each member of a sample population representing the population of concern for the exposure period through either (a) existing databases from composites of past studies (for baseline assessment) or (b) study-specific information (special registries); (5) Estimation of multimedia levels and temporal profiles of pollutants in various microenvironments (streets, residences, offices, restaurants, vehicles, etc.) through either (a) field study measurements or (b) microenvironmental mass-balance modeling (air), drinking water distribution modeling (water), and dietary concentration modeling (food); (6) Calculation of appropriate inhalation rates, as well as drinking water and food consumption rates for the members of the sample population, through a combination of the physiological attributes of the study subjects and the activities pursued during the individual exposure events; and (7) Biologically-based target tissue dose (toxicokinetic) modeling. This last step provides a significant advantage over past practices in exposure assessment, as it allows model evaluation against field measurements of biomarkers, when such information is available.

3 The Dose–Response Information Analysis System (DORIAN)

The DORIAN system is being developed by a USEPA-funded research consortium that includes UMDNJ-RW Johnson Medical School; Rutgers University; Princeton University; and the US Food and Drug Administration’s National Center for Toxicological Research, Center for Toxicoinformatics. It is complementing MENTOR by providing components that mechanistically address the biological phenomena taking place in the dose-to-effect part of the “environmental health sequence” of Fig. 1. This is allowing the completion of comprehensive toxicity and risk assessment studies by incorporating toxicogenomic information that is becoming available from various transcriptomic, proteomic, and metabonomic laboratory studies (the complementary structures of MENTOR and DORIAN are shown schematically in Fig. 4). DORIAN includes ebTrack, which expands the features of ArrayTrack (Tong et al. 2003), by incorporating a range of options for the analysis of proteomic and metabonomic datasets, in addition to providing additional analysis options for microarray gene expression data (e.g. Ouyang et al. 2004). To support usage of DORIAN in real-world applications, an “environmental bioinformatics Knowledge Base” (ebKB; see http://www.ebKB.org or http://www.EnvironmentalBioinformatics.org – Fig. 5) has been developed and is being updated regularly to include advancements in this field. A representative selection from the wide spectrum of databases that are linked through ebKB can be found in Tables 2 and 3 in the Appendix.
Fig. 4

DORIAN complements MENTOR by coupling toxicoinformatic databases and biologically-based dose–response models to the components of MENTOR, to support the analysis of the complete environmental “source-to-outcome” health sequence of Fig. 1

Fig. 5

The environmental bioinformatics Knowledge Base (ebKB) is an evolving compendium of computational tools, databases, and other pertinent information that is needed for the systematic analysis of environment–organism interactions that occur over multiple scales

A specific objective of the DORIAN effort is to produce generalized, integrated, physiologically based models for the “coupled” toxicokinetics and toxicodynamics of contaminants of concern and their mixtures (Abdel-Rahman and Kauffman 2004; Danhof et al. 2007). These models are designed so as to describe quantitatively both the processes affecting the contaminant in the organism (absorption, distribution, metabolism, elimination) and the processes resulting to toxicity due to the contaminant. So, the biologically-based toxicodynamic models link target tissue molecular and cellular events (e.g. receptor activation, changes in ion channel functions, etc.) to consequent biochemical and physiological changes.

The DORIAN development effort benefits directly from a wide range of ongoing activities in computational cell biology (e.g. Slepchenko et al. 2002; Takahashi et al. 2003), computational physiology (e.g. Crampin et al. 2004; Strategy for the EuroPhysiome (STEP) Consortium 2006), toxicogenomics (e.g. NRC 2005; UK Environment Agency 2003; USEPA 2004), systems toxicology (e.g. USEPA 2003b; BMBF 2002; Waters et al. 2003) and other related fields, taking place in North America, Europe, Japan, New Zealand, and elsewhere. Outcomes from these research efforts are regularly incorporated in the DORIAN system to ensure its relevance and usefulness.

4 Examples: Selected Case Studies

Various problem-specific implementations of MENTOR have been developed and applied to a wide range of environmental issues including: regional ozone pollution (Foley et al. 2003); urban and local scale inhalation exposures to co-occurring ozone, particulate matter and air toxics (Georgopoulos et al. 2005a; Isukapalli et al. 2005); multimedia and multipathway exposures to copper (Georgopoulos et al. 2006), to mercury and its compounds (Wang et al. 2005b), to arsenic (Georgopoulos et al. 2005b, 2007), to trichloroethylene (Wang et al. 2005a), etc. Special focus has been given in the evaluation of novel methods for systematic simplification of complex models (e.g. Wang et al. 2005a) and for uncertainty analysis and reduction (e.g. Balakrishnan et al. 2003). A summary review of the range of past applications of MENTOR can be found in Georgopoulos and Lioy (2006). Development and application of DORIAN components and assemblage of related data modules commenced in 2005 and up-to-date information on progress and current activities related to DORIAN is posted regularly on the web site www.ebCTC.org. Selected new results are presented here from applications of coupled MENTOR and DORIAN modules (for environmental and biological processes, respectively) to problems involving human exposures to contaminants that are associated with a variety of toxicological endpoints. Figures 6 and 7 show respectively, the compartmental structures of the environmental and biological modeling frameworks employed in MENTOR and DORIAN.
Fig. 6

Schematic depiction of the modular, multiscale, modeling framework of MENTOR for assessing environmental “material flows” and resulting human exposure to mixtures of contaminants present in different media

Fig. 7

The physiologically-based toxicokinetic modules of MENTOR and DORIAN aim to characterize uptake and target tissue dose for cumulative and aggregate exposure; they are designed so as to account for intra-individual and inter-individual variability within a human population, including the effects of development and aging

Figure 8 provides an example of biologically-based calculations of target tissue doses for exposure to mixtures of metals and their compounds, utilizing MENTOR-3P. Figure 9 shows example results of prototype source-to-dose assessments of arsenic exposures and of corresponding target tissue doses, which were performed using the MENTOR-4M and MENTOR-3P platforms, for a sample population in the USA. These studies included physiologically-based calculations of the toxicokinetics of different forms of arsenic for all the individuals of the virtual population considered. Specifically, Fig. 9a shows predicted cumulative arsenic (total and inorganic) exposure distributions from inhalation, food intake, drinking water consumption, and non-dietary routes and Fig. 9b shows predicted cumulative distributions of inorganic arsenic (AsIII + AsV) and organic arsenic species (MMA + DMA) internal dose in kidney and liver, for the population of Franklin County, OH (Georgopoulos et al. 2005b).
Fig. 8

Simulated hepatic concentration profiles of metals and metabolites for a standard reference male ingesting a mixture of metals (specified in legend)

Fig. 9

a Predicted cumulative arsenic (total and inorganic) exposure distributions from inhalation, food intake, drinking water consumption, and non-dietary routes and b predicted cumulative inorganic arsenic (AsIII + AsV) and organic arsenic (MMA + DMA) species internal dose distributions of kidney and liver, for the population of Franklin County, OH. Calculations were performed with the MENTOR-4M and MENTOR-3P models

Figure 10 shows an example outcome of a MENTOR-1A application involving the prediction of PM2.5 concentrations and corresponding high-end population doses for the population of the city of Philadelphia, PA, USA. Figure 10a shows 24-h averaged local outdoor concentrations on July 19, 1999 for 482 urban Philadelphia census tracts, derived from hourly PM2.5 predictions of the Community Multiscale Air Quality (CMAQ) model (Byun and Ching 1999) that were “scaled” at census tract level using the Bayesian Maximum Entropy (BME) method (Christakos et al. 2001), which “fuses” all available information (from both modeling and observations), taking into account spatial and temporal trends simultaneously. Figure 10b shows the corresponding 95th percentiles, for each census tract, of 24-h aggregated total inhalation doses from both outdoor and indoor sources. The color scheme shows quantiles of the concentration and the dose distributions; the blank census tracts represent airports and other areas not considered in the study (Georgopoulos et al. 2005a). Figure 10c shows the cumulative distribution function for total PM2.5 dose from outdoor sources and the corresponding total dose from indoor sources. One of the many factors affecting the biologically relevant doses shown in Fig. 10b and c is human activity; in the simulations shown here, spatiotemporal distributions of human activities were derived from information contained in the Consolidated Human Activity Database (CHAD – Stallings et al. 2002) and the aerosol dosimetry module of MENTOR-3P was used to calculate the respiratory deposition for virtual individuals of the simulation. Figure 10d shows an example of the importance of considering activity levels in dosimetry calculations: experimental human dosimetry data (Jaques and Kim 2000; Daigle et al. 2003) collected under conditions of moderate exercise are compared to predictions from the Multiple Path Particle Dosimetry (MPPD2 – CIIT 2006) and from the International Commission on Radiological Protection (ICRP – Jarvis et al. 1996) models as well as with predictions from MENTOR-3P (it should be noted that the predictions from the three models do not deviate from each other and from the measurements in the case of resting conditions). These results demonstrate the importance of considering human activities and microenvironmental conditions in assessing dosages for individuals and populations.
Fig. 10

a 24-h averaged local outdoor concentrations on July 19, 1999 for 482 urban Philadelphia, PA, USA census tracts, derived from hourly PM2.5 predictions of the Community Multiscale Air Quality (CMAQ) model that were “scaled” at census tract level, using the Bayesian Maximum Entropy (BME) method; b Corresponding 95th percentiles, for each census tract, of 24-h aggregated total inhalation doses from both outdoor and indoor sources. The color scheme shows quantiles of concentrations/dose distribution; the blank census tracts represent airports and other areas not considered in the study. c Cumulative distribution function for total PM2.5 dose from outdoor sources and the corresponding total dose from indoor sources; d Experimental aerosol dosimetry data under moderate exercise conditions (Jaques and Kim 2000; Daigle et al. 2003) compared to predictions of the Multiple Path Particle Dosimetry (MPPD2 – CIIT 2006) and International Commission of Radiological Protection (ICRP – Jarvis et al. 1996) models, and the inhalation dosimetry module of MENTOR-3P (identified as CCL)

Figure 11 depicts schematically how the environmental health risk assessment process in the DORIAN framework is applied to connect genotypes and phenotypes to assess disease susceptibility with respect to specific environmental agents. This requires data/information integration on the effects of xenobiotics across multiple biological levels (genes, proteins, metabolites) to determine the corresponding impact on bionetwork (signaling, regulatory, metabolic) dynamics. Numerous studies in the emerging field of toxicogenomics aim to relate susceptibility of individuals to environmental agents by identifying genes that are expressed differentially in the presence of these agents. Genetic variability in these genes (e.g., presence of Single Nucleotide Polymorphisms [SNPs]) may be associated with increased (or decreased) susceptibility to the effects of the environmental agents. However, in general, a phenotype is the result of the collective response of a group of genes (gene network). Furthermore, gene regulation is distributed over multiple levels (genes and messenger RNA, proteins, and metabolites) and it is necessary to identify and characterize processes across the entire bionetwork of interactions involved in the expression of the genes of concern. DORIAN brings together computational tools for assessing and analyzing the information arising from new toxicogenomic studies, in conjunction with the information available in numerous existing bionomic databases (see Table 3 in Appendix). Typical components of this approach include: (a) the development of regulatory networks of signal transduction integrated with hormonal and metabolic networks; (b) the analysis of relevant available data (transcriptomic, proteomic, metabolomic, lipidomic, etc.) for the contaminants and endpoints of concern; (c) the identification of transcription factor binding sites in the promoter regions of gene regulatory clusters; and (d) the identification of key network nodes. This approach and the related computational tools thus support the elucidation of biological action mechanisms and allow the development of quantitative models that relate genotypic variation to variability in the signaling, gene-regulatory and metabolic processes involving environmental contaminants, which in turn affect the distribution of responses (health impact) within a human population.
Fig. 11

Environmental health risk assessment process aims to connect genotypes and phenotypes to assess disease susceptibility to environmental agents by integrating data/information across multiple biological levels and determining bionetwork dynamics: the framework that is presented schematically here, as implemented in DORIAN, aims to relate genotypic variation (e.g. SNPs) to altered enzyme function and metabolic processes and finally to variation in disease susceptibility by elucidating the biological mechanisms involved in signaling and gene regulatory processes [the hypothetical gene-protein-metabolite bionetwork of the left side of the figure has been adapted from Brazhnik et al. (2002)]

5 Conclusions

The various applications of MENTOR to a wide range of environmental problems, and the recent applications of combined MENTOR and DORIAN modules that were reported here, demonstrate the feasibility and the potential of holistic, “person-oriented” approaches in studying environmental health issues. The selected references listed in this work demonstrate that there is extensive activity in North America, Europe, Japan, New Zealand, and elsewhere, towards the development and deployment of methods that rely on a systematic integration of exposure and biological processes for understanding the “coupled dynamics” of environmental and human health states (see also Schwartz and Collins 2007). As such methods evolve toward maturity and widespread acceptance, they can be reasonably expected to provide various opportunities for rethinking and reevaluating environmental and public health policy practices, by taking into account the variability among individuals that simultaneously experience multiple, possibly interacting, environmental stressors.


Support for this work has been provided primarily by the USEPA-funded Environmental Bioinformatics and Computational Toxicology Center (ebCTC) under STAR Grant No. GAD R 832721-010, and the USEPA funded Center for Exposure and Risk Modeling (CERM) under Cooperative Agreement no. CR-83162501. This work has not been reviewed by and does not represent the opinions of the funding agency. Appreciation is extended to the research team of CCL, with special thanks to Profs S. Isukapalli and S. W. Wang, as well as to A. Sasso, Y. C. Yang, and P. Shade. Thanks are also due to Prof P.J. Lioy (CERM), Prof W. Welsh (ebCTC), Dr W. Tong (USFDA-NCTR Center for Toxicoinformatics), and to the numerous USEPA and EOHSI collaborators who have contributed to this research.

Copyright information

© Springer Science+Business Media B.V. 2007